Nitride Semiconductor Technology

Nitride Semiconductor Technology
Downloaded from https://onlinelibrary.wiley.com/doi/ by Universite De Sherbrooke, Wiley Online Library on [22/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Edited by Fabrizio Roccaforte and Mike Leszczynski
Nitride Semiconductor Technology
Power Electronics and Optoelectronic Devices

Editors All books published by Wiley-VCH
are carefully produced. Nevertheless,
Dr. Fabrizio Roccaforte authors, editors, and publisher do not
Consiglio Nazionale delle Ricerche - warrant the information contained in
Istituto per la Microelettronica e these books, including this book, to
Microsistemi (CNR-IMM) be free of errors. Readers are advised
Strada VIII, n. 5 - Zona Industriale to keep in mind that statements, data,
95121 Catania illustrations, procedural details or other
Italy items may inadvertently be inaccurate.
Prof. Mike Leszczynski Library of Congress Card No.: applied for

Institute of High Pressure Physics -
Polish Academy of Sciences British Library Cataloguing-in-Publication
(Unipress-PAS) Data
Sokolowska 29/37 A catalogue record for this book is
01-142 Warsaw available from the British Library.
Poland
Bibliographic information published by
Cover Image the Deutsche Nationalbibliothek
© FlashMovie/Shutterstock The Deutsche Nationalbibliothek lists
this publication in the Deutsche
Nationalbibliografie; detailed
bibliographic data are available on the
Internet at <http://dnb.d-nb.de>.
© 2020 Wiley-VCH Verlag GmbH &

Co. KGaA, Boschstr. 12, 69469
Weinheim, Germany
All rights reserved (including those of

translation into other languages). No
part of this book may be reproduced in
any form – by photoprinting,
microfilm, or any other means – nor
transmitted or translated into a
machine language without written
permission from the publishers.
Registered names, trademarks, etc. used
in this book, even when not specifically
marked as such, are not to be
considered unprotected by law.
Print ISBN: 978-3-527-34710-0

ePDF ISBN: 978-3-527-82525-7
ePub ISBN: 978-3-527-82527-1
oBook ISBN: 978-3-527-82526-4
Typesetting SPi Global, Chennai, India

Printing and Binding
Printed on acid-free paper
10 9 8 7 6 5 4 3 2 1
v
Contents
Preface xi
Acknowledgments xv
1 Introduction to Gallium Nitride Properties and Applications 1

Fabrizio Roccaforte and Mike Leszczynski
1.1 Historical Background 1
1.2 Basic Properties of Nitrides 4
1.2.1 Microstructure and Related Issues 7
1.2.2 Optical Properties 13
1.2.3 Electrical Properties 16
1.2.4 Two-Dimensional Electron Gas (2DEG) in AlGaN/GaN
Heterostructures 19
1.3 Applications of GaN-Based Materials 23
1.3.1 Optoelectronic Devices 24
1.3.2 Power- and High-Frequency Electronic Devices 26
1.4 Summary 30
Acknowledgments 31
References 31
2 GaN-Based Materials: Substrates, Metalorganic Vapor-Phase

Epitaxy, and Quantum Well Properties 41
Ferdinand Scholz, Michal Bockowski, and Ewa Grzanka
2.1 Introduction 41
2.2 Bulk GaN Growth 42
2.2.1 Hydride Vapor-Phase Epitaxy (HVPE) 43
2.2.2 Sodium Flux Growth Method 45
2.2.3 Ammonothermal Growth 46
2.3 MOVPE Growth 51
2.3.1 Basics About Nitride MOVPE 54
2.3.2 Epitaxy on Foreign Substrates 58
2.3.2.1 Sapphire as a Foreign Substrate 58
2.3.2.2 GaN on SiC and Si 60
2.3.3 Defect Reduction by ELOG, FACELO, etc. 62
2.3.4 In Situ ELOG by SiN Deposition 64
vi Contents
2.3.5 Doping of Nitrides 64

2.3.6 Growth of Other Binary and Ternary Nitrides 67
2.4 InGaN QWs: Growth and Decomposition 72
2.4.1 Growth of InGaN QWs on Polar, Nonpolar, and Semipolar GaN
Substrates 72
2.4.2 Origins of In Fluctuations 75
2.4.3 Homogenization of InGaN QWs 78
2.4.4 Decomposition of the QWs 79
2.5 Summary 82
Acknowledgments 82
References 83
3 GaN-Based HEMTs for Millimeter-wave Applications 99

Kathia Harrouche and Farid Medjdoub
3.1 Introduction 99
3.2 Targeted Applications for GaN Millimeter-wave Devices 99
3.2.1 High-Power Amplification 100
3.2.2 Broadband Amplifiers 102
3.2.3 5G 103
3.2.3.1 GaN for 5G 104
3.2.3.2 GaN Base Station PAs 106
3.2.3.3 Moving Forward to 6G 108
3.3 GaN-based Material Designs for Millimeter-wave Applications 108
3.3.1 Intrinsic Characteristics and Comparison with Other Materials for RF
Devices 108
3.3.2 Specific Material Systems for RF Devices 114
3.4 Device Design and Fabrication of Millimeter-wave GaN Devices 116
3.4.1 Description of Key Processing Steps for Various GaN Device
Designs 116
3.4.1.1 Device Scaling for Millimeter Wave 116
3.4.1.2 T-shaped Gate Design 116
3.4.1.3 Advanced Ohmic Contact Technology 117
3.4.1.4 N-polar GaN HEMTs 118
3.4.1.5 AlN-Based Device Performances 119
3.4.1.6 InAlGaN-Based Device Performances 121
3.4.2 State-of-the-art Millimeter-wave GaN Transistors 122
3.5 Overview of MMIC Power Amplifiers 123
3.5.1 MMIC Technology Using III-N Devices 123
3.5.1.1 III–V Material-Based MMIC Technology 123
3.5.1.2 Power Amplifiers 124
3.5.1.3 Low-Noise Amplifier 125
3.5.2 MMIC Examples from Ka-band to D-band Frequencies 125
3.6 Summary 126
References 127
Contents vii
4 Technologies for Normally-off GaN HEMTs 137

Giuseppe Greco, Patrick Fiorenza, Ferdinando Iucolano, and Fabrizio
Roccaforte
4.1 Introduction 137
4.1.1 Threshold Voltage in AlGaN/GaN HEMTs 138
4.2 GaN HEMT “Cascode” 140
4.3 “True” Normally-off HEMT Technologies 142
4.3.1 Recessed-gate HEMT 142
4.3.2 Fluorinated HEMT 145
4.3.3 Recessed-gate Hybrid MISHEMT 149
4.3.4 p-GaN Gate HEMT 155
4.4 Other Approaches 163
4.5 Summary 164
Acknowledgments 165
References 165
5 Vertical GaN Power Devices 177

Srabanti Chowdhury and Dong Ji
5.2 Vertical GaN Devices for Power Conversion 177
5.3 Vertical GaN Transistors 178
5.3.1 Current Aperture Vertical Electron Transistor (CAVET) 178
5.3.2 Vertical MOSFETs 182
5.4 High-Voltage Diodes in GaN 185
5.5 Avalanche Electroluminescence in GaN P–N Diodes 186
5.6 Impact Ionization Coefficients in GaN 188
5.6.1 Impact of Impact Ionization Studies on Predictive Modeling 193
5.7 Summary 193
Acknowledgments 193
References 194
6 Reliability Issues in GaN Electronic Devices 199

Milan Ťapajna and Christian Koller
6.1.1 Reliability Testing and Failure Analysis of GaN HEMTs 200
6.2 Reliability of GaN HEMTs for RF Applications 204
6.2.1 AlGaN/GaN HEMTs 204
6.2.1.1 Trapping Effects 204
6.2.1.2 Gate-edge Degradation 207
6.2.1.3 Hot Electron Degradation 209
6.2.2 InAlN/GaN HEMTs 211
6.2.2.1 Hot Electron Degradation 212
6.2.2.2 Role of Hot Phonons 214
6.2.3 Thermal Issues in RF GaN HEMTs 215
viii Contents
6.3 Reliability and Robustness of GaN Power Switching Devices 219

6.3.1 Parasitic Effects in the Carbon-Doped GaN Buffer 221
6.3.1.1 Insulation of GaN Buffer by Carbon Doping 221
6.3.1.2 Time-Dependent “Dielectric” Breakdown (TDDB) of the GaN
Buffer 223
6.3.1.3 Dynamic RDS,ON Due to Buffer Trapping 225
6.3.2 Gate Degradation in p-GaN Switching HEMTs 230
6.3.3 V th Instabilities in GaN MISHEMTs 233
6.3.3.1 Studies of PBTI in MISHEMTs 237
6.4 Summary 241
Acknowledgments 241
References 241
7 Light-Emitting Diodes 253

Amit Yadav, Hideki Hirayama, and Edik U. Rafailov
7.2 State-of-the-Art GaN LEDs 254
7.2.1 Blue LEDs 258
7.2.2 Green LEDs 262
7.3 GaN White LEDs: Approaches and Properties 264
7.3.1 Monolithic LEDs 267
7.3.2 Phosphor-Covered LEDs 271
7.4 AlGaN Deep UV LEDs 275
7.4.1 Growth of High-Quality AlN and Increasing in Internal Quantum
Efficiency (IQE) 278
7.4.2 AlGaN-based UVC LEDs 281
7.4.3 Increasing the Light Extraction Efficiency (LEE) 282
7.5 Summary 287
Acknowledgments 288
References 288
8 Laser Diodes Grown by Molecular Beam Epitaxy 301

Greg Muziol, Henryk Turski, Marcin Siekacz, Marta Sawicka, and Czeslaw
Skierbiszewski
8.2 III-N Growth Fundamentals by Plasma-Assisted MBE 303
8.2.1 Role of N-Flux for Efficient InGaN QWs 304
8.3 Wide InGaN QWs – Beyond Quantum-Confined Stark Effect 305
8.4 Long-Living Laser Diodes on Bulk Ammono-GaN 313
8.5 Laser Diodes with Tunnel Junctions 316
8.5.1 Stacks of Vertically Interconnected Laser Diodes 319
8.5.2 Distributed Feedback Laser Diodes 321
8.6 Summary 324
Acknowledgments 324
References 325
Contents ix
9 Edge Emitting Laser Diodes and Superluminescent

Diodes 333
Szymon Stanczyk, Anna Kafar, Dario Schiavon, Stephen Najda, Thomas Slight,
and Piotr Perlin
9.1 Laser Diode: History and Development 333
9.1.1 Optoelectronics Background 333
9.1.2 Gallium Nitride Technology Breakthroughs 335
9.1.3 Development of Nitride Laser Diodes 337
9.2 Distributed Feedback Laser Diodes 342
9.3 Superluminescent Diodes 348
9.3.1 History of Superluminescent Diode Development 348
9.3.2 Basic SLD Properties 351
9.3.3 Challenges for SLD Optimization 353
9.4 Semiconductor Optical Amplifiers 354
9.5 Summary 357
References 358
10 Green and Blue Vertical-Cavity Surface-Emitting Lasers 367

Yang Mei, Rong-Bin Xu, Huan Xu, and Bao-Ping Zhang
10.1.1 Properties and Application of GaN VCSELs 367
10.1.2 Brief History and Current Status of GaN VCSELs 368
10.1.3 GaN VCSELs with Different DBRs 369
10.1.3.1 GaN VCSELs with Hybrid DBR Structure 370
10.1.3.2 GaN VCSELs with Double Dielectric DBR Structure 371
10.2 Efficiency of Heat Dissipation of Different Device Structures 372
10.2.1 Simulation of Heat Profile of the Device 372
10.2.2 Dependence of Rth on Cavity Length 373
10.3 Green VCSELs Based on InGaN QDs 375
10.3.1 Advantages of QDs Compared with QWs 375
10.3.2 Growth and Optical Properties of InGaN QDs 377
10.3.3 Fabrication Process of VCSELs 379
10.3.4 Properties of QD Green VCSELs 379
10.4 Green VCSELs Based on Cavity-Enhanced Emission of Localized
States in Blue Emitting InGaN QWs 380
10.4.1 Cavity Effect 380
10.4.2 Properties of Cavity-Enhanced Green VCSELs 381
10.5 Dual-Wavelength Lasing Based on QD-in-QW Active Structure 384
10.5.1 Characteristics of QD-in-QW Structure 384
10.5.2 Lasing Characteristics of VCSELs 386
10.6 Blue VCSELs with Different Lateral Confinements 386
10.6.1 Design of Index-Guided Structure 386
10.6.2 Emission Properties of VCSELs with Lateral Confinement 388
10.7 Summary 389
References 390
x Contents
11 Integration of 2D Materials with Nitrides for Novel Electronic

and Optoelectronic Applications 397
Filippo Giannazzo, Emanuela Schilirò, Raffaella Lo Nigro, Pawel Prystawko,
and Yvon Cordier
11.2 Fabrication of 2D Material Heterostructures with Nitride
Semiconductors 400
11.2.1 Transfer of 2D Materials Grown on a Foreign Substrate 400
11.2.2 Direct Growth of 2D Materials on Group III-Nitrides 403
11.2.3 2D Materials as Templates for the Growth of Nitride Semiconductor
Films 407
11.3 Electronic Devices Based on 2D Materials/GaN Heterojunctions 413
11.3.1 Band-to-band Tunneling Diodes Based on MoS2 /GaN
Heterojunctions 413
11.3.2 Hot Electron Transistors with Graphene Base and Al(Ga)N/GaN
Emitter 414
11.4 Optoelectronic Devices Based on 2D Material Junctions with
GaN 421
11.4.1 GaN LEDs with Graphene-Transparent Conductive Electrodes 421
11.4.2 MoS2 /GaN Deep UV Photodetectors 427
11.5 Applications of Graphene for Thermal Management in GaN
HEMTs 428
11.6 Summary 431
Acknowledgments 431
References 432
Index 439
xi
Preface
Today, gallium nitride (GaN) and other related materials (ternary AlGaN
and InGaN and quaternary InAlGaN) are widely used for optoelectronics
components. In addition, some of these nitrides are also emerging as promising
semiconductors for energy-efficient power electronic devices. Hence, the rev-
olution expected in modern electronics and optoelectronics is often regarded,
half-jokingly, as “GaNification.”
Our main intention in preparing this book was to give, with the valuable
contributions of leading specialists, a general overview on the state of the art
of GaN-based technologies, covering both fields of power electronics and
optoelectronic devices.
Chapter 1 is a general introduction to the properties and applications of GaN
and related materials. After an historical background on the relevant milestones
of nitrides research, special emphasis will be put on InGaN quantum wells and
AlGaN/GaN heterostructures, which are important systems for light-emitting
diodes (LEDs), laser diodes (LDs), and high electron mobility transistors
(HEMTs). The main applications of nitride materials for both optoelectronic
devices and power- and high-frequency electronics are also described, antici-
pating some of the most critical issues that are illustrated in detail in the rest of
the book.
Typical nitride-based devices are made of multilayered heterostructures, which
require epitaxial growth on appropriate substrates. Hence, Chapter 2 starts by
discussing some recently developed methods to produce GaN wafers. Then, the
most popular epitaxial method for GaN growth, i.e. metalorganic vapor-phase
epitaxy (MOVPE), is described, discussing the role of growth temperature, the
deposition on foreign substrates, and the methods for reducing the high thread-
ing dislocation density, as well as the doping difficulties for achieving p-type con-
ductivity. A part of this chapter is dedicated to InGaN quantum wells, which find
important applications in light-emitting devices but exhibit serious problems of
composition uniformity and thermal stability with respect to decomposition.
Regarding high-frequency electronics, today, the interest for millimeter-wave
(mmW) band (30–300 GHz) is continuously increasing because of its reduced
wavelength and wide-frequency bands, which enable smaller component with
improved performances. Wireless communication systems are extending to
higher frequencies. However, several challenges need to be overcome in order
to use the mmW spectrum successfully. In this context, Chapter 3 focuses on
xii Preface
GaN-based devices for mmW applications. Targeted applications including

high-power amplifiers, broadband amplifiers, and 5G wireless networks are
introduced. Various GaN-based material designs for mmW applications are
described, showing the advantages and limitations for high-frequency appli-
cations. Device design and fabrication of mmW GaN devices are analyzed.
Finally, an overview of monolithic microwave-integrated circuit (MMIC) power
amplifiers is also reported.
GaN is also considered a promising semiconductor for power electronics.
Owing to the presence of the two-dimensional electron gas (2DEG), AlGaN/GaN
HEMTs are inherently normally-on devices. However, in many power electronic
applications, normally-off transistors are desired. Hence, Chapter 4 reviews
the current technologies for normally-off GaN-based HEMTs. First, the HEMT
“cascode” approach is briefly described, highlighting advantages and limitation
of this design. Then, after illustrating the recessed gate HEMT and the fluorinate
gate approach, the focus is put on the recessed gate hybrid metal insulator
semiconductor high electron mobility transistor (MISHEMT) and on the p-GaN
gate HEMT. These are today the most promising and robust approaches for
normally-off GaN HEMTs. The most critical issues of these technologies (e.g.
heterostructure design, gate dielectrics, and metal gates) are discussed in this
chapter.
In power electronics, the vertical device topology would be preferred to the
lateral one, in order to reduce the on-resistance and increase the current capa-
bility. In this context, Chapter 5 is designed to give an overview on the progress
made in vertical devices based on bulk GaN substrates. In GaN technology, ver-
tical devices are now in the forefront showing impact through both two- and
three-terminal devices developed over the past decade. In particular, two differ-
ent types of vertical devices are discussed, namely the current aperture vertical
electron transistor (CAVET) and a regrowth-based trench MOSFET called oxide
gate interlayer FET (OGFET). Moreover, the latest findings on the coefficients of
impact ionization are also reviewed.
The development of GaN-based power amplifiers for RF, microwave, and
mm-wave applications would have not been possible without an intense
research on the stability and reliability issues. Chapter 6 reviews the most
important reliability issues of GaN HEMTs for RF and microwave applications
as well as power switching transistors. Failure modes and mechanisms of RF
AlGaN/GaN and InAlN/GaN HEMTs are reviewed, focusing on gate-edge, hot
electrons, and hot phonons related failure modes and thermal effects. For power
switching devices, the effect of GaN buffer carbon doping on dynamic on-state
resistance and time-dependent dielectric breakdown of the buffer are described.
The gate degradation of normally-off p-GaN gate HEMTs and threshold voltage
instabilities in GaN MISHEMTs are also discussed.
GaN-based semiconductors are great materials for optoelectronic devices
because of their broad emission wavelength range from ultraviolet to the
yellow-green. GaN-based LEDs are discussed in Chapter 7. LEDs have made
tremendous progress in the past 15 years and have reached a point where
Preface xiii
they are reinventing and redefining artificial lighting. Nevertheless, in their

own realm, they suffer from decrease in efficiency at higher currents (droop).
Moreover, the full potential in terms of light quality, i.e. color rendering
index (CRI) and correlated color temperature (CCT) that can be offered by
these devices, can still be improved with existing or alternative schemes and
device configurations and are discussed in this chapter. AlGaN deep ultraviolet
light-emitting diodes (DUV LEDs) have a wide variety of potential applications,
including sterilization, water purification, etc. However, the efficiency of AlGaN
DUV LEDs is still low in comparison to InGaN blue LEDs. In this chapter,
the methods to improve internal quantum efficiency (IQE), electron injection
efficiency (EIE), and light extraction efficiency (LEE) for AlGaN DUV LEDs are
presented.
Chapter 8 presents the recent advancement in the III-nitride LDs grown by
plasma-assisted molecular beam epitaxy (PAMBE). First, the growth fundamen-
tals in PAMBE are discussed. Then, the nature of carrier recombination in wide
InGaN quantum wells is studied. A model of a highly efficient transition path
through excited states is proposed to explain the experimental observations.
Additionally, the reliability of the LDs grown by PAMBE is presented. Finally,
the tunnel junction (TJ) and its applications to devices are discussed.
In Chapter 9, after recalling the history of nitride semiconductors LDs, the
main technical and scientific challenges related to their development and the
perspective for the future are reviewed. In particular, the recently emerged dis-
tributed feedback (DFB) nitride LDs and its possible applications are described.
The less known, but closely related to LDs, superluminescent diodes and optical
amplifiers are also described.
In Chapter 10, the vertical-cavity surface-emitting lasers (VCSELs) are
discussed. They have many advantages, including small footprint, circular low-
divergent output beams, wafer-level testing, densely packed two-dimensional
arrays, good price–performance ratio, and simple optics/alignment for output
coupling. Green and blue emitting VCSELs are especially important for appli-
cations such as visible light communication, full color display, and microlaser
projector. In particular, the current status and the different device design
approaches are discussed, presenting different VCSEL concepts, based on
InGaN quantum dots (QDs) and quantum wells (QWs).
Finally, Chapter 11 illustrates the recent developments in the integration of
2D materials (graphene and MoS2 ) with nitride semiconductors for electron-
ics and optoelectronics. The state-of-the-art approaches for the fabrication of
heterojunctions between these two classes of materials are discussed, consid-
ering advantages and limitations. Examples of electronic devices based on 2D
material/nitride heterostructures are presented, such as the hot electron tran-
sistor (HET) with a Gr base and Al(Ga)N/GaN emitter for THz electronics and
the band-to-band tunneling diode based on p+ -MoS2 /n+ -GaN heterojunctions
for digital electronics with ultralow-power dissipation. Optoelectronic devices
(LEDs and photodetectors) based on the integration of 2D materials with GaN
are also discussed.
on nitride semiconductor technology for power electronics and optoelectronic
We hope you will enjoy reading the book and find more new useful information
Fabrizio Roccaforte and Mike Leszczynski

Catania and Warsaw
March 2020
devices.
Preface
xiv
xv
Acknowledgments
We would like to acknowledge all the authors of the chapters for the excellent
contributions to this book and for the continuous discussion during the prepara-
tion of the entire volume. They are also acknowledged for their endless patience
against our frequent reminders and strict deadlines. Among them, a specific note
of gratitude is for our colleagues and coworkers at the Institute for Microelec-
tronics and Microsystems of the National Research Council of Italy (CNR-IMM)
in Catania and at the Institute of High Pressure Physics of the Polish Academy of
Science (Unipress-PAS) in Warsaw that accepted this hard work with enthusiasm.
In addition, we wish to acknowledge the bilateral project ETNA (Energy
efficiency Through Novel AlGaN/GaN heterostructures) within the CNR-PAS
Cooperation Agreement for the years 2017–2019 and the bilateral project
GaNIMEDE (Gallium Nitride Innovative Micro-Electronics Devices) within the
Executive Programme for Scientific and Technological Cooperation between
The Italian Republic and the Republic of Poland for the years 2019–2020.
Finally, special thanks is for the Editors and Staff at Wiley who continuously
assisted us during the chapters’ preparation and then carried out an accurate
review during the production process.
1
Introduction to Gallium Nitride Properties and Applications

Fabrizio Roccaforte 1 and Mike Leszczynski 2,3
1 Consiglio Nazionale delle Ricerche – Istituto per la Microelettronica e Microsistemi (CNR-IMM), Strada VIII,
n. 5 – Zona Industriale, 95121 Catania, Italy

2
Institute of High Pressure Physics – Polish Academy of Sciences (Unipress-PAS), Sokolowska 29/37, 01-142
Warsaw, Poland
3
TopGaN Sp. z o.o., Sokolowska 29/37, 01-142 Warsaw, Poland
1.1 Historical Background

Since some decades, gallium nitride (GaN) and other related materials (e.g.
ternary AlGaN and InGaN and quaternary InAlGaN) have been used for
optoelectronic components. Moreover, some of these nitrides are also recently
emerging as promising semiconductors for the next generation of power
electronic devices. In fact, the introduction of GaN-based materials in power
electronics can enable a better efficiency of the devices and a reduction of the
electric power consumption. Hence, the overall nitride device market forecasts
for the next years are much brighter compared to those of the other compound
semiconductors.
For those reasons, the revolution expected in modern electronics and optoelec-
tronics is often regarded, half-jokingly, as “GaNification.”
This book gives an overview on the main nitride semiconductor technolo-
gies for power electronic and optoelectronic devices: transistors, diodes,
light-emitting diodes (LEDs), laser diodes (LDs), etc. The way to these devices
has been long and not easy because nitrides are very difficult to be grown and
processed. The present chapter is an introduction to properties and applications
of GaN and related materials.
The history of GaNification can be dated back to the early 1930s and since then
it has been characterized by many important milestones, leading to the creation
of the key technologies for electronic and optoelectronic devices. A summary of
important historical steps of nitrides research is reported below:
• 1932: The first polycrystalline GaN material was synthesized by flowing ammo-
nia (NH3 ) over liquid gallium (Ga) at elevated temperatures [1]. This material
was examined, proving its stability up to 800 ∘ C in hydrogen atmosphere.
• 1938: The crystal structure of GaN has been studied on GaN powders [2].
Nitride Semiconductor Technology: Power Electronics and Optoelectronic Devices,

First Edition. Edited by Fabrizio Roccaforte and Mike Leszczynski.
© 2020 Wiley-VCH Verlag GmbH & Co. KGaA. Published 2020 by Wiley-VCH Verlag GmbH & Co. KGaA.
2 1 Introduction to Gallium Nitride Properties and Applications
• 1969–1971: Thin GaN layers were grown by Maruska and Tietjen [3] using
hydride vapor-phase epitaxy (HVPE) on sapphire substrates. Because of a large
lattice mismatch of these two materials, the layers exhibited a poor crystal-
lographic quality. However, this material was used for confirming the direct
energy band gap (3.39 eV) and for demonstrating the first LED concepts [4].
• 1972: Manasevit et al. [5, 6] grew the first metal–organic vapor-phase epitaxy
(MOVPE) GaN layers. This method became the most popular one in nitride
technology, and today, there are a few thousands of reactors worldwide. How-
ever, these layers were still rough and not transparent (they looked like frosted
glass) because of the lattice mismatch between sapphire and GaN.
• 1984: The thermodynamics studies by Karpinski et al. [7, 8] enabled the authors
to find the way of growing bulk GaN crystals from the nitrogen solution in
gallium (HPSG – high-pressure solution growth). However, the growth con-
ditions necessary to avoid GaN decomposition were extreme: about 1200 ∘ C
and 10 kbar. This work clarified the unfeasibility to grow GaN crystals from
the melt (i.e. Czochralski or Bridgman methods) because tens of kilobars have
to be used to melt GaN at temperatures higher than 2500 ∘ C.
• 1986: Hiroshi Amano from the group of Isamu Akasaki (Nagoya University)
made a real breakthrough in GaN epitaxy by introducing a low-temperature
AlN nucleation layer [9]. Such a nucleation layer enabled Amano and Akasaki
to obtain smooth and transparent GaN with good crystallographic quality.
• 1989: Amano et al. [10] obtained for the first time p-type GaN by activating
Mg dopants by low-energy electron irradiation. This activation was induced by
breaking Mg—H bonds, inherently present in the MOVPE GaN layers (hydro-
gen is always incorporated together with Mg).
• 1990: Matsuoka et al. [11] succeeded in the growth of the first InGaN layers,
offering an access to a very wide spectral range from 0.7 eV (IR) to 3.5 eV (UV),
through all wavelengths of the visible range.
• 1991–1992: Shuji Nakamura (Nichia Chemical Corporation) optimized the
growth conditions of GaN on sapphire using the AlN nucleation layer. This
technology has been used by most of the companies and academic laboratories
for fabricating LEDs and other devices. The dislocation density of such GaN on
sapphire was in the order of 108 –109 cm−2 . Moreover, annealing above 600 ∘ C
in nitrogen atmosphere led to the electrical activation of p-GaN layers [12, 13].
These studies paved the way to all LED and LD technologies.
• 1992: Nichia started offering blue LEDs (450 nm), which were used soon for
illuminating phosphor to get white light. At present, such LEDs are being pro-
duced in millions every day, creating a multibillion market.
• 1993: Using MOVPE, Asif Khan et al. [14] grew the first AlGaN/GaN het-
erojunction. Despite a moderate crystallographic quality, the mobility of the
two-dimensional electron gas (2DEG) at the AlGaN/GaN interface was around
600 cm2 /V s. This achievement can be regarded as a start of quest to nitride
high electron mobility transistor (HEMTs) technology [15].
• 1996: The first violet LD (405 nm) based on InGaN quantum wells (QWs) was
demonstrated by Nakamura et al. [16]. The laser was made on sapphire sub-
strate, although it was clear that the use low defect density bulk GaN substrates
1.1 Historical Background 3
could lead to better performance. At that time, using the only available small
HPSG bulk GaN crystals enabled Nakamura to increase the LD lifetime from
300 to 3000 hours.
• 1997: Bernardini et al. [17] determined the spontaneous and piezoelectric
polarization constants of nitrides, improving the comprehension of the physics
of AlGaN/GaN heterostructures and paving the way to the optimization of
many electronic and optoelectronic devices.
• 1998: The first LEDs epi structures grown on silicon substrates by molecu-
lar beam epitaxy (MBE) were reported by Guha and Bojarczuk [18]. The MBE
method provided some advantages over MOVPE, as the lack of hydrogen dur-
ing growth, or the lower growth temperature. However, until now, most of the
nitride technologies are based on MOVPE. The structures grown by Guha and
Bojarczuk had a bad crystallographic quality but triggered research on epitaxy
technology using silicon wafers as substrates.
• 1999: Ambacher et al. [19] proposed a model to analytically describe the
2DEG properties in AlGaN/GaN heterostructures. This model is today
widely adopted by the GaN community. In the same year, Sheppard et al.
[20] demonstrated high-power microwave HEMTs based on AlGaN/GaN
heterostructures grown on the silicon carbide (SiC) substrate.
• 2000: The nature of the 2DEG was further clarified by Ibbetson et al. [21],
attributing a key role to the surface states present in nitride materials as source
of electrons.
• 2001: Sumitomo Electric bought patent from Tokyo Agriculture University
on the DEEP method (dislocation elimination by the epitaxial-growth with
inverse-pyramidal pits) [22, 23] to grow GaN single crystals on GaAs substrates
using the HVPE method and bowing dislocations in small regions. This tech-
nology enabled Sumitomo to start fabricating bulk GaN substrates with low
dislocation density (105 cm−2 ) at the areas designated for laser stripes to make
a commercial production of LDs by Nichia.
• 2003: Sony introduced Blu-ray DVD format based on 405 nm LDs fabricated
by Nichia.
• 2006: Saito et al. [24] and Cai et al. [25] proposed, respectively, the recessed
gate structure and fluorine implantation to achieve normally-off AlGaN/GaN
HEMTs.
• 2007: Uemoto et al. [26] from Panasonic demonstrated the first normally-off
HEMT based on the p-GaN gate technology. This device was also called gate
injection transistor (GIT) as hole injection from a p-AlGaN layer resulted in
an increase of the drain current because of the conductivity modulation. In
2009, EPC announced the commercialization of the first devices based on this
technology.
• 2008: The first GaN-based vertical cavity surface emitting laser (VCSEL) oper-
ating at low temperature was demonstrated by Tien-Chang Lu and coworkers
at the National Chiao Tung University (Hsinchu, Taiwan) [27].
• 2012: AlGaN/GaN heterostructures on 200 mm Si(111) substrates were
demonstrated by Tripathy et al. [28]. The possibility to grow such
heterostructures on large-area Si wafers opened the way to integrate
GaN HEMT device fabrication in Si CMOS (complementary metal–oxide

semiconductor) fabs.
• 2013: Iveland et al. [29] provided experimental evidence that the “droop
effect” (efficiency drop for high injection current in LEDs) is related to Auger
electrons.
• 2013: High-frequency (HF) performances of GaN-based HEMTs, with ultra-
high cutoff frequency f t exceeding 450 GHz, were achieved through innovative
device scaling technologies [30].
• 2014: The Noble Prize in Physics was assigned to three Japanese scientists
(Isamu Akasaki, Hiroshi Amano, and Shuji Nakamura) for “the invention of
efficient blue light-emitting diodes which has enabled bright and energy-saving
white light sources”(www.nobelprize.org/prizes/physics/2014/summary).
• 2014: Recessed gate Al2 O3 /AlGaN/GaN normally-off MOS-HEMTs with
reduced leakage current, high blocking voltage (825 V), and high threshold
voltage (2.4 V) were demonstrated on 200 mm silicon substrates [31].
• 2015–2017: The first high-voltage (600 V) normally-off GaN HEMT solu-
tion, based on the “cascode” configuration, is released to the market by
Transphorm [32]. Moreover, the progresses in the technology of p-GaN
gate HEMTs resulted in the first fully industrial qualified “true” normally-off
devices from Panasonic and Infineon [33–35].
• 2017: Haller et al. [36] gave a model for point defect formation at high growth
temperature before the InGaN QWs, explaining their detrimental effect on
LED efficiencies.
• 2017: LG Innotek introduced 100 mW UV-C LEDs into the market (www
.powerelectronictips.com/worlds-first-100-mw-deep-uv-led).
• 2017: Mei et al. [37] demonstrated quantum dot VCSELs covering the “green
gap.”
• 2019: Zhang et al. [38] demonstrated 271.8 nm LD operating at room temper-
ature and in the pulse mode.
1.2 Basic Properties of Nitrides

Nitride semiconductors (GaN, AlGaN, InAlN, InGaN, InAlGaN, AlN, etc.)
possess a number of properties, which make them very useful for many opto-
electronic and microelectronic applications. Table 1.1 reports the main physical
properties of GaN and other nitrides in comparison with other semiconductors,
highlighting some implications in material growth, optoelectronics, and power-
and high-frequency electronics.
As can be seen, these materials exhibit several advantages for optoelectronic
and power electronics, mainly related to their direct band gap, or the wide band
gap (WBG) and high critical field. However, there are also a number of features,
which make the material growth and the technology of devices very difficult. All
these aspects will be discussed in more detail in the following sections, as well as
in the other chapters of this book.
Table 1.1 Properties of GaN and other nitrides in comparison with other semiconductors and their implications in material growth, optoelectronics, and
power- and high-frequency electronics [39–48].
Comparison with other

Property Value/range semiconductors
Material growth Density, 𝜌 6.1 g/cm3 (GaN) Because of the high melting point and low
Atomic density 4.37 × 1022 atoms/cm3 decomposition temperature, nitride crystals (bulk and
(GaN) epi) are grown at low temperatures. Therefore, the
crystals (substrates) cannot be grown from the melt as
Melting point 2573 ∘ C at 60 kbar for other semiconductors. Moreover, the epilayers grown
GaN, 1100 ∘ C for InN, at low temperatures have a large number of
2200 ∘ C for AlN imperfections
Decomposition 900 ∘ C for GaN 600 ∘ C for
temperature at 1 bar InN
Small influence of Blue/green GaN-based LEDs and HEMTs can be
dislocations on luminosity fabricated using foreign substrates (Si, sapphire, and
of InGaN QWs and on SiC)
electron scattering at low
currents
High critical 29.8–54.7 GPa Dislocations do not move upon stress or illumination
Peierls–Nabarro shear (no degradation of optoelectronic devices related to
stress for slip systems dislocation motion as it is the case for other III–V
⟨1123⟩{1122} and materials)
⟨1123⟩{1101}
Optoelectronics Direct band gap From 0.7 eV (InN) to Nitrides dominate in the green/blue/UV spectral
6.1 eV (AlN) range. II–VI compounds are comparable, but they are
too fragile to be used in devices. In infrared and red
spectral range, GaAs- and InP-based devices have still
much higher efficiencies
(Continued)
Table 1.1 (Continued)
Comparison with other

Property Value/range semiconductors
Build-in internal electric Up to 2 MV/cm The strong build-in internal electric field increases the
field (InGaN/GaN) spatial separation of electrons and holes, thus
reducing the efficiency of radiative recombination in
optoelectronic devices
Power/high-frequency Wide band gap, Eg From 3.4 eV (GaN) to Possible applications of GaN-based materials in
(HF) electronics 6.1 eV (AlN) high-voltage, high-power, and high-temperature
electronics, in competition with SiC (and possibly in
Critical electric field, ECR 3–3.75 MV/cm (GaN)
future, with Ga2 O3 and diamond). The high defects
Electron affinity, 𝜒 3.1–4.1 eV (GaN) density still hinders the full exploitation of the electric
Dielectric constant, 𝜀r 9.5 (GaN) field strength
Intrinsic carrier ≈10−10 cm−3 (GaN) at Low leakage currents and high operation
concentration, ni room temperature temperatures are possible, if the GaN material quality
is improved
Electron saturation 3 × 107 cm/s (GaN) Enable the fabrication of devices operating at high
velocity, 𝜈 frequencies, in competition with the traditional GaAs
Electron mobility, 𝜇n 1100–2000 cm2 /V s (GaN, technology
AlGaN/GaN)
Thermal conductivity, 𝜅 1.3–2.1 W/cm K (GaN) Comparable to Si but significantly lower than SiC and
diamond, making the heat dissipation a concern for
GaN-based power devices
1.2.1 Microstructure and Related Issues

The specificity of nitride microstructure is dominated by three factors: (i) the
main crystallographic form of GaN, AlN, and InN is the wurtzite, with high
built-in electric fields; (ii) large lattice mismatches between substrates and
epilayers, as well as between layers that form a device epi structure; and (iii) low
growth temperatures.
The thermodynamically stable crystal phase of nitride semiconductors at room
temperature and atmospheric pressure is the hexagonal wurtzite structure. For
GaN, the wurtzite crystal structure is schematically depicted in Figure 1.1, where
the unit cell, highlighted with bold lines, is characterized by the lattice parameters
a0 (3.189 Å) and c0 (5.185 Å) [50, 51]. The Ga and N atoms are arranged in two
interpenetrating hexagonal closely packed lattices, with a shift of 3/8 c0 .
In the hexagonal wurtzite structure, GaN has no inversion symmetry in the
[0001] direction (the so-called c-axis). For that reason, it is possible to distinguish
two different orientations of GaN crystals, i.e. the Ga-face (Figure 1.1a) and the
N-face (Figure 1.1b), depending on whether the material is grown with Ga or N on
top. These two faces have different chemical properties: Ga-face is more chem-
ically inert and may behave differently during growth (e.g. the Ga-face incorpo-
rates easier acceptors, whereas the N-face incorporates easier donors [52]).
The covalent bonds allow each atom to be tetrahedrally bonded to four atoms
of the other type. In addition, an ionic contribution is also present because of
the large difference in electronegativity of Ga and N atoms. Because nitrogen has
a higher electronegativity than gallium, Ga and N atoms will exhibit anionic and
cationic characteristics, respectively, thus resulting in a spontaneous polarization
PSP oriented along the c-axis [53].
The spontaneous polarization exists even in the absence of strain. The strength
of the spontaneous polarization depends on the asymmetry of the crystal. In fact,
PSP increases with decreasing the c0 /a0 ratio. For example, a GaN crystal with a
Ga-face N-face
[0001] [0001]
Psp Psp
c0 c0
a0 a0
(a) Ga N (b) Ga N
Figure 1.1 Hexagonal crystal structure of GaN (wurtzite) for the Ga-face (a) and for the N-face
(b). The bold lines highlight the unit cell, while the dashed lines indicate the Ga—N bonds. The
spontaneous polarization vectors (PSP ) are also drawn for the two cases. Source: Reproduced
with permission of Roccaforte et al. [49]. Copyright © 2018, Società Italiana di Fisica.
Table 1.2 Lattice parameters of AlN, GaN, and InN [50, 54–56].
Parameters AlN GaN InN
a0 (Å) 3.1113 3.1878 3.537

c0 (Å) 4.9814 5.1850 5.703
c0 /a0 ratio of 1.6259 has a lower PSP (−0.029 C/m2 ) with respect to an AlN crystal
(−0.081 C/m2 ) having a c0 /a0 ratio of 1.6010 [17]. The negative values of the polar-
ization indicate that the vector PSP is pointing opposite to the [0001] direction,
as illustrated in Figure 1.1 [17, 50]. It is worth mentioning that the external stress
changing the ideality of the structure and the c0 /a0 ratio can induce an addi-
tional contribution to the polarization called piezoelectric polarization PPE . This
contribution is particularly important in AlGaN/GaN heterostructures for the
generation of 2DEG, as will be discussed in Section 1.2.4.
For the fabrication of power electronic and optoelectronic devices (e.g. diodes,
transistors, LEDs, and LDs), the suitable epitaxial structures must be grown on
appropriate substrates. The lattice parameters are important for determining the
suitability of a material as a substrate for GaN epitaxy. Table 1.2 reports the values
of the lattice parameters of AlN, GaN, and InN.
Here, it must be pointed out that in the literature, it is possible to find different
data for InN and AlN because of the material differences with respect to their
defect concentration. In addition, many papers report the lattice parameters of
nitrides together with the thermal expansion coefficients (TECs), which may lead
to wrong conclusions, as the TECs vary with temperature [56–58].
Most of the research and technologies of nitrides are based on the epitaxy on
foreign substrates [59] because GaN and AlN single crystals are still small and
expensive.
Table 1.3 reports the lattice and thermal mismatch between GaN and the most
popular foreign substrates: sapphire (Al2 O3 ), silicon (Si), and silicon carbide
(SiC). The typical ranges of the dislocation density measured in GaN layers
grown on these substrates, the possible device layout (lateral or vertical), and
application fields of the materials are also reported in the table.
Sapphire is a common substrate used for GaN heteroepitaxy for optoelectronic
applications. However, GaN exhibits large lattice (+16%) and TEC mismatch
(−34%) with respect to Al2 O3 substrate, typically resulting into a high defect
density of the grown GaN epilayers. Hence, a better choice is represented by
the hexagonal silicon carbide (6H/4H–SiC) with a lattice mismatch of only
3.5%. The first GaN-based transistors for power switching applications were
demonstrated on (0001) c-plane sapphire and (0001) silicon carbide (6H–SiC
and 4H–SiC) [20, 39, 62]. However, in spite of the small lattice mismatch, a
high dislocation density (107 –108 cm−2 ) is still present in GaN layers grown on
SiC [63]. Moreover, the cost of SiC substrates has always represented a limiting
factor for the introduction of this technology in consumer electronic products.
At the beginning of the last decade, great efforts have been devoted to
the development of GaN on Si substrates [60]. In fact, Si offers low price in
Table 1.3 Lattice and thermal mismatch between GaN and the most popular foreign
substrates (Al2 O3 , Si, and SiC).
Properties and
applications of GaN
layers on different
substrate GaN-on-Al2 O3 GaN-on-SiC GaN-on-Si GaN-on-GaN
Lattice mismatch 16 3.5 −17 0

(%)
Thermal expansion −34 21.4 53.5 0
coefficient
mismatch, a0 (%)
Dislocation density 107 –108 107 –108 108 –109 103 –106
(cm−2 )
Device layout Lateral Lateral Lateral Lateral and
vertical
Main application Optoelectronics HF electronics, HF and power HF and power
areas optoelectronics electronics electronics,
optoelectronics
Relevant information for device applications of GaN epilayers (dislocation density, possible device
type, and application fields) are also reported [43, 59–61].
comparison with sapphire and SiC, a high crystalline perfection, and the avail-
ability of large-size substrates. Clearly, the lattice mismatch between GaN(0001)
with respect to Si(111) is very large (−17%), and consequently, a high dislocation
density (in the order of 109 cm−2 ) is generated in the material.
The difference in the TECs between GaN and its substrate also plays an impor-
tant role in the epitaxial process and, hence, in the final quality of the active layer
for the device. Sapphire has a higher TEC than GaN, leading to residual com-
pressive stress in the grown GaN layer, while the TECs of SiC and Si are smaller,
resulting in residual tension [64]. The value of the residual stresses depends on
the growth conditions, as at the growth temperature, the GaN layers are not fully
relaxed but can be strained because of grain coalescence [65].
To overcome the problem of residual strain, in the case of GaN-on-Si, sev-
eral “strain management” techniques are adopted to mitigate crack formation in
the GaN epitaxial layers (e.g. appropriate AlN or graded AlGaN buffer layers,
AlN/GaN superlattices, or sophisticated lithographic patterning techniques of
the substrates [39, 66]). In this context, the rapid progresses in epitaxial growth
of GaN materials resulted in the demonstration of electronic grade large-area
GaN-on-Si heterostructures up to 200 mm in diameter, and power devices with
excellent efficiency and compact size have started to be implemented on these
materials [28, 31, 66, 67].
Clearly, bulk GaN would be the ideal substrate for GaN epitaxy and device
fabrication. In fact, a significant reduction of the dislocation density (down to
103 –106 cm−2 ) is possible using bulk GaN [61]. So far, such substrates are used
for LD fabrication. Besides that, the possibility of fabricating vertical devices (not
possible when GaN is grown on foreign substrates) would represent a big advan-
tage for power electronic applications. However, until now, several obstacles have
hindered a widespread implementation of bulk GaN power devices, including the
high cost and the limited diameter of commercially available substrates [61].
The lattice mismatch is a problem not only in heteroepitaxy but also in GaN
homoepitaxy. In fact, it has been reported that free electrons expand the lattice of
GaN [68]. For example, the free electrons of concentration 5 × 1019 cm−3 expand
the lattice of GaN by about 0.3%. Although this lattice expansion may seem not
significant, it can lead to serious problems in the growth of bulk GaN crystals, as
different crystal faces incorporate different impurities (mainly oxygen) and con-
tain different free electron concentrations.
In majority of cases, the growth of nitrides is in the ⟨0001⟩ direction. It is the
most natural one for nitrides to get a good crystallographic quality. Moreover,
p-type doping is more efficient when the growth is in this direction [69]. How-
ever, for other directions, the internal electric fields, separating electrons and
holes in LEDs, are smaller. Therefore, there have been attempts to use GaN sub-
strates or templates oriented along nonpolar directions of ⟨1010⟩ and semipolar
ones of ⟨1122⟩ to increase the efficiency of optoelectronic devices. Despite the big
research efforts done, the advantage of nonpolar or semipolar device epi struc-
tures has not been proven yet, and almost all commercial devices are constructed
on polar wafers.
In epitaxy, the on-axis orientations are seldom used. In most of the cases,
off-cut substrates are used to grow epi layers in step-flow growth mode [70].
The substrate off-cut influences the epilayers under different points of view,
such as the point defects formation, the indium incorporation in InGaN layers,
the lattice distortions, and surface roughness. For example, Suski et al. [71]
showed a higher hole concentration for GaN:Mg grown on the off-cut Ga-face
GaN substrate. Sarzyński et al. [72] observed a decrease of indium content for
InGaN grown on the off-cut Ga-face GaN substrate. On the other hand, triclinic
distortion of the strained epi layers was reported by Krysko et al. [73]. Also, the
surface morphology is influenced by the off-cut angle, as a smaller roughness
was observed in the GaN layer grown on the off-cut N-face GaN substrate [74].
All optoelectronic devices emitting in visible region are based on InGaN QWs,
which are extremely difficult to be grown because of large lattice mismatch to
GaN and of low growth temperatures. In this context, a widely discussed topic is
the occurrence of indium spatial fluctuations in the QW [75–77]. These fluctua-
tions may appear in the nanometer, micrometer, and even millimeter scale.
An example of indium fluctuations in InGaN/GaN QWs, visualized by
high spatial resolution secondary ion mass spectrometry (SIMS), is shown in
Figure 1.2. Their amplitude and dimensions strongly depend on growth condi-
tions, dislocation density, and layer morphology (e.g. local terrace width) [72].
Concerning InGaN, special attention should be devoted to the influence
of hydrogen on indium incorporation. In fact, even a small percentage of
hydrogen significantly decreases the indium incorporation [78–80]. However,
hydrogen smoothens the surfaces, so it can be used during the InGaN or GaN
barrier growth. The smoother the surface is, the smaller the indium fluctuation
should be.
x in InxGa1–xN
200 0.263
0.230
160
0.197
y direction (μm)
0.164
120
0.131
80 0.099
0.066
40
0.033
0 0.000
0 40 80 120 160 200
x direction (μm)
Figure 1.2 SIMS x–y image showing the lateral distribution of indium in a InGaN/GaN QW. The
vertical scale indicates the In-concentration x in Inx Ga1−x N. Source: Adapted with permission
of Michałowski et al. [77]. Copyright © 2019, Royal Society of Chemistry.
InGaN layers and QWs evolve when they are subjected to high temperatures;
for example, when they are overgrown by p-type GaN:Mg. Although at low tem-
perature the InGaN layers may get homogenized, an increase of the temperature
(>900 ∘ C) leads to InGaN decomposition. Both phenomena occur because of
easy indium diffusion, most probably through Ga-vacancies. This topic will be
discussed in Chapter 2.
When the InGaN layers exceed critical values of strain and thickness, they get
relaxed by misfit dislocations [81]. An example of a net of misfit dislocations in
InGaN is shown in Figure 1.3.
It is worth mentioning that the relaxation of the InGaN layers may be not
only a plastic one (via emission of dislocations), but also an elastic one in
three-dimensional nano-objects on the surface (poor morphology). There-
fore, InGaN relaxation should be monitored not only using X-ray diffraction
Figure 1.3 TEM plan view bright-field image with

diffraction vector g11–20 of the MBE-grown 50 nm
InGaN layer, with an In content of 20%. Source:
Courtesy of Johanna Moneta.
1 μm
Figure 1.4 Optical microscopy view

of a 220 nm thick Al0.27 Ga0.73 N layer
grown on a GaN substrate.
10 μm
(XRD) but also using atomic force microscopy (AFM), transmission electron
microscopy (TEM), and defect-selective etching (DES).
Nitride transistors for power- and high-frequency devices, as well as UV emit-
ters, are based on AlGaN layers. The main microstructural problem encountered
with the AlGaN layers is their tendency to be cracked because of the tensile strain
when they are grown on GaN (e.g. in AlGaN/GaN heterostructures for transis-
tors). Figure 1.4 shows, as an example, an optical microscopy image of a thick
AlGaN layer grown on GaN, showing the presence of cracks on the surface.
In order to avoid cracking, lateral patterning [82] or compliance layers [83] are
applied as technological solutions.
Cracks are obviously a serious concern in GaN heteroepitaxy on Si substrates,
as they act as scattering centers that reduce the carrier mobility and, hence,
the performance of the transistors. In this case, several methods have been
reported to eliminate the cracks in AlGaN/GaN heterostructures grown on
Si(111) substrates and to improve the crystal quality, such as the use of graded
Alx Ga1−x N interlayers between an AlN buffer layer and GaN, or the introduction
of AlGaN/GaN superlattices [84].
Although extended defects, as dislocations, can be easily detected and their
density is measured using selective etching, XRD, or TEM, much more difficult
is the detection of point defects. Ga-vacancies (acceptor-like defects) can be mea-
sured using positron annihilation, but only in the case of thick layers (not QWs).
In the case of nitrogen vacancies, there is no direct method to detect them. There-
fore, only theoretical models can be used to explain various properties of nitrides,
such as diffusion of atoms, luminosity, or electrical properties, based on indirect
assumptions of the presence of point defects.
In summary, nitride semiconductors for optoelectronics and power electronics
have a complex microstructure, which is characterized by a large number of
imperfections: dislocations, point defects, poor morphology, non-uniform
strains, non-uniform electric fields, and non-uniform atom distribution. All
these imperfections are not independent of each other and may influence the
optical and electrical properties. Moreover, all those imperfections depend on
the growth parameters (temperature, pressure, gases flows, reactor geometry,
etc.). Hence, the complex microstructure, as well as the large number of growth
parameters, makes the optimization of the crystal quality a very challenging

issue for GaN and related nitrides.
1.2.2 Optical Properties

Nitrides possess direct band gaps from 0.7 eV (InN), through 3.4 eV (GaN) to
6.1 eV (AlN). Hence, they cover the spectral range from infrared (1770 nm)
through the visible range up to far ultraviolet (about 200 nm) [85]. Figure 1.5
reports the values of the energy gap for nitride semiconductors as a function of
the in-plane lattice parameter.
For a ternary nitride compound Ax B1−x N, the energy band gap does not change
linearly with the composition x, but it follows a phenomenological expression:
A B1−x N
Eg x (x) = xEgA + (1 − x)EgB − x(1 − x)b (1.1)
where b is bowing parameter (in units of eV), defined as the coefficient of the
parabolic term. A positive value of b indicates a downward bowing, while a nega-
tive b indicates an upward bowing in the dependence of the energy gap Eg on the
composition x.
Typically, in nitride compounds, most of the measurements of the energy gap
(and hence of the bowing parameter) are performed using emission techniques,
such as photoluminescence (PL) and cathodoluminescence (CL).
The most important materials for optoelectronics are QWs and wires of
InGaN. The optical properties of these QWs have been reported in thousands of
papers, but still they contain many unrevealed secrets. For InGaN, the reported
values of the bowing parameter b range from 1.4 to 2.8 eV [86]. Such a large
span is related to the variability of the InGaN properties (indium fluctuations,
non-homogeneous QW thickness), already mentioned in Section 1.2.1. The
compositional inhomogeneities of the alloy and even inherent hole localization
in random alloys [87] lead to strong Stokes shift and underestimation of the band
gap [88]. Moses and Van de Walle [89] suggested that the bowing parameter of
InGaN is dependent on the composition.
On the other hand, AlGaN alloys are important for both optoelectronic and
power- and high-frequency electronic devices. For AlGaN alloys, the values of
Figure 1.5 Energy band gap of nitride 6 200

semiconductors (AlN, GaN, InN, and AlN
their ternary alloys) as a function on the

5
Energy gap Eg (eV)
in-plane lattice parameter. The

Wavelength (nm)
corresponding wavelength is reported 300

4 AlxGa1–xN
on the right axis.
3 GaN InxAl1–xN 400
2 600
800
InxGa1–xN InN 1000
1 1400
3.0 3.1 3.2 3.3 3.4 3.5 3.6
In-plane lattice parameter (Å)
the bowing parameter b reported in the literature vary from −0.8 to 2.6 eV, most
likely emanating from AlGaN alloys prepared by different techniques with var-
ious qualities and, in some cases, a range of alloy compositions explored to be
narrow. Yun et al. [90] reported a systematic study of the Alx Ga1−x N band gap
as a function of the Al concentration x. By fitting the experimental data using
Eq. (1.1), a bowing parameter b = 1.0 eV was obtained [90].
Optical transmission (OT) or optical absorption (OA) measurements provide
a better estimation of the energy gap of nitrides. However, it is not easy to grow
thick nitride layers (InGaN and AlGaN) necessary for such measurements. More-
over, OT and OA do not take into account energy dispersion of the refractive
index, whose contribution to the band gap should not be neglected. The alter-
native measurement method that can meet the demands for the accurate energy
band gap bowing is spectroscopic ellipsometry [91]. This method has been used
for the estimation of the bowing parameter of nitrides and has the major advan-
tage of determining the complex dielectric function of the investigated material.
In general, the PL and CL spectra measured on nitrides consist of excitonic
part and defect-related part [92]. For example, for bulk GaN, the following spec-
tral features are typically identified: (i) excitonic part: free excitons and bound
excitons; (ii) donor–acceptor pairs (DAPs); and (iii) defect-related luminescence.
An example of a PL spectrum for GaN is shown in Figure 1.6.
As mentioned already, even the simplest compound of GaN possess a large
variety of point and extended defects, which affect the position, full width at half
maximum (FWHM), and intensity of the PL peaks. A good review of optical prop-
erties of GaN was given by Reschikov and Morkoç [93].
In the case of ternary or quaternary compounds, the situation is much more
complicated than for GaN, as one may deal with chemical composition fluctu-
ations. The main information from the PL or CL peak is its wavelength posi-
tion. For InGaN QWs, it depends on several factors: average indium content,
indium fluctuations, QW thickness (for thin QWs, there is quantum effect of the
105
DBE
PL intensity (a.u.)
FE-A
104 104 FE-B
PL intensity (a.u.)
FE-C FE-C, XC – free exciton C

DBE-LOABE
FE-B, XB – free exciton B
103 DBE-2LO
FE-A, XA – free exciton A
DBE, D0 XA – donor bound exciton
102
103 ABE, A0 XA – acceptor bound exciton
3.30 3.35 3.40 3.45 3.50 3.55
DBE-LO, D0 XA – LO – phonon replica
Energy (eV)
DBE-2LO, D0 XA – 2LO – phonon replica
YL YL – yellow luminescence
102
2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6
Energy (eV)
Figure 1.6 Typical photoluminescence (PL) spectrum for a GaN layer grown on sapphire
substrate. Besides the excitonic part of the spectrum, a broad peak is present in the yellow
region (about 550 nm), which can be related to point defects (Ga-vacancies and carbon
impurity). Source: Courtesy of Grzegorz Staszczak.
blue shift, and for thick QWs, there is a separation of electrons and holes by the
electric field resulting in the red shift), QW thickness variations, and presence
of point defects. The position of the PL or CL peaks also depends on excitation
power, which makes the analysis of the data extremely cumbersome. However,
even more troublesome is the analysis of PL/CL peak intensity. In fact, besides
the factors mentioned above, the peak intensity also depends on the absorption
around the QWs and the diffusion length of minority carriers.
Additional information can be obtained by using time-resolved photolumines-
cence (TRPL) or time-resolved cathodoluminescence (TRCL). The decay times of
the luminescence peaks can be interpreted by radiative and non-radiative recom-
bination of electrons and holes. The experimental luminescence decay time (𝜏) is
related to the radiative (𝜏 r ) and non-radiative (𝜏 nr ) decay times:
1 1 1
= + (1.2)
𝜏 𝜏r 𝜏nr
For binary compounds, long decay times can be observed for the nearly perfect
crystals. For the particular case of InGaN QWs, the long decay times can also be
attributed to indium fluctuations [94].
Many interesting PL and CL studies have been done as a function of temper-
ature or pressure. With increasing pressure, the energy band gaps of nitrides
increase, which can push some of the energy levels into the band gap to make
them observable by optical measurements [95].
Frequently, experiments are performed as a function of temperature (from 4.2
to 300 K), from which an internal quantum efficiency 𝜂 (IQE) can be extracted
from the relation:
I(T)
𝜂= (1.3)
I(0)
where I(T) and I(0) are the PL or CL intensities at a given temperature T and at
0 K, respectively.
At the same time, the IQE can be related to radiative non-radiative decay times:
1 𝜏
=1+ r (1.4)
𝜂 𝜏nr
Hence, from the experimental measurements of I(0), I(T), and 𝜏, using
Eqs. (1.2)–(1.4) makes possible to determine the quantum efficiency 𝜂 and the
radiative and the non-radiative decay times (𝜏 r and 𝜏 nr ).
However, the PL and CL data depend on excitation power and on the properties
of the layers close to the active region. Hence, it is possible that the optimization
of the optical active region (e.g. InGaN QWs) with respect to IQE is not reflected
by higher efficiency of LEDs or LDs.
Figure 1.7 schematically depicts the typical layout of the most popular
GaN-based optoelectronic devices, i.e. LEDs and LDs. In the case of LEDs,
when the p–n junction is forward biased, the potential barrier of the junction
is lowered and the recombination of electrons and holes can occur in the
active region (e.g. a multiple quantum well, MQW), thus leading to photon
emission. The challenge is not to have these photons absorbed or scattered in
a wrong direction. For transparent sapphire substrates, it is possible to extract
hν
p-Contact p-Contact
hν p-AlGaN
p-GaN n-Contact
MQW
MQW
n-AlGaN
n-GaN
Bulk GaN
Sapphire substrate
n-Contact
Light emitting diode Laser diode
(a) (b)
Figure 1.7 Simplified schemes of the two most popular GaN-based optoelectronic devices: (a)
light-emitting diode (LED) and (b) edge emitting laser diode (LD). For LEDs, the light emission
can be either through the back or upper surface. In the case of LDs, the light is confined
between the cladding layers and is partially circulating between the mirrors on two edges of
the chip.
light through them. For silicon substrates, the light extraction takes place only
through the surface, which is not easy because of the presence of electrical
contact. In the case of LDs, the light is confined between two AlGaN “cladding
layers” and the light emission takes place through the edge covered with mirrors,
which make photon circulate between two edges. For a low injection current,
the light is emitted incoherently in a similar way to the LEDs. However, above
a certain current threshold, a sufficiently high concentration of carriers is
generated within the active region, leading to the population inversion. In this
condition, electron–hole recombination is assisted by such photon and the
stimulated emission takes place, dominating over the spontaneous one.
More details on LEDs and LDs operation principles and related technologies
will be reported in Chapters 7–10.
1.2.3 Electrical Properties

GaN possess excellent electrical properties, which make it a promising material
for electronic device fabrication.
First, the WBG of the material (Eg = 3.4 eV) induces a high critical electric
field (ECR = 3 − 3.75 MV/cm), which is the maximum field that the material can
withstand without breakdown. As will be described in Section 1.3.2, the high crit-
ical field is a key factor for the realization of electronic devices operating at high
voltages and with low on-resistance and high efficiency.
The temperature effects within the semiconductor must also be considered, as
they have a strong influence on the carrier generation and, hence, on the elec-
trical properties of the devices. In fact, the semiconductor device characteristics
degrade by increasing the temperature, until their functionality for desired circuit
applications is lost. Hence, an accurate control of the carrier (electrons and holes)
concentration is needed. However, the dopant species are not the only source of
carriers in a semiconductor. In fact, even in the absence of an intentional dop-
ing, every semiconductor would contain a certain amount of thermal carriers in
Figure 1.8 Calculated intrinsic carrier

T = 298 K
concentration ni as a function of 1015
reverse of the temperature (1000/T) for
GaN. For comparison, the curves of ni
for Si and SiC are also reported. The 1010
dashed line indicates the room
ni (cm–3)
temperature (T = 298 K). Source: 105
Adapted from Neudeck et al. 2002 [96] Si
and Baliga 2005 [97]. SiC
100
GaN
1× 10–5
10–10
1 2 3 4 5
1000/T (K–1)
the crystal. The amount of these carriers is referred to as the intrinsic carrier
concentration ni , and it is exponentially dependent on temperature T:
√ Eg
ni = NC NV e− 2kT (1.5)
where k is the Boltzmann constant and N C and N V are the density of the states
in the conduction and valence bands, respectively.
Figure 1.8 reports the calculated intrinsic carrier concentration ni of GaN as a
function of the inverse of the temperature [96]. For comparison, the values of ni
for Si and SiC are also reported.
Noteworthy, if the temperature is increased, e.g. above 300 ∘ C, in Si, the
intrinsic carriers would become comparable or even higher than the intentional
dopant concentration. Then, the electrical properties of the material would
become undesirably influenced by the intrinsic carriers rather than by the
designed doping needed for proper electrical operation [96]. From Figure 1.8,
it is clear that SiC and GaN have much lower intrinsic carrier concentrations
than Si. Hence, they do not suffer from intrinsic carrier conductivity issues even
at a temperature of 600 ∘ C. As an example, as can be seen in Figure 1.8, the
intrinsic carrier concentration in GaN at room temperature (T = 298 K) is about
19 orders of magnitude lower than that of Si. Because of such a low value of ni ,
the generation current is extremely low. Hence, GaN electronic devices should
have theoretically a much lower leakage current with respect to Si and give the
possibility to operate at higher temperatures.
However, it is important to point out that these considerations are valid for
perfect crystals. In the reality, because of the large lattice mismatch between GaN
and the substrate, the presence of material defects in GaN epitaxial layers, such as
dislocations, typically provide preferential leakage pathways, which are a major
obstacle for reaching the ideal electrical performances and low leakage in GaN
electronic devices [98].
Figure 1.9 shows the calculated electron velocity of GaN as a function of the
electric field [99]. As can be seen, the peak velocity in GaN is close to 3 × 107 cm/s
and the saturation velocity is about 1.5 × 107 cm/s. These values are considerably
higher than the values for GaAs and Si. Thanks to the high saturation velocity of
3 Figure 1.9 Electron velocity as a function

GaN of the electric field for GaN. For
SiC
comparison, the values of the electron
Electron velocity (×107 cm/s)
Si
GaAs
velocity of GaAs, Si, and SiC are also
reported. Source: Adapted from Jain et al.
2 2000 [99] and Sze and Ng 2007 [100].
0
0 1 2 3
Electric field (×105 V/cm)
the carriers, it is possible to shorten the transit time in GaN electronic devices,
thus allowing them to operate at high frequency, as will be better explained in
Section 1.2.4.
Finally, the thermal conductivity of GaN is expected to vary in a certain range
(1.3 − 2.1 W/cm K) depending on the defect density. These values are lower than
those in SiC. Hence, SiC is better indicated for high-temperature applications
than GaN. In fact, the heat dissipation in GaN devices must be appropriately
managed, as will be discussed in Chapters 6 and 11.
Thanks to the properties mentioned above, significant advantages can be
obtained using GaN-based devices, in terms of high-voltage, high-frequency,
and high-temperature operation. To highlight these advantages, Figure 1.10
summarizes in a “radar chart” the electronic properties of GaN compared with
those of Si and SiC. Here, it is worth mentioning that the reported material
Si Critical electric field

SiC (MV/cm) High
GaN voltage
5
4
3
2 Energy gap (eV)
Saturation velocity
(×107 cm/s) 1
0
High High
frequency temperature
Electron mobility Thermal conductivity

(×103 cm2/V s) (W/cm K)
Figure 1.10 “Radar chart” of physical and electronic properties of GaN compared with those
of Si and SiC.
properties may vary between different references. However, the graphic gives an
idea of the great potential of this material.
1.2.4 Two-Dimensional Electron Gas (2DEG) in AlGaN/GaN

Heterostructures
One of the most interesting features of nitride materials is the possibility to cre-
ate alloys and heterostructures with tailored optical and electrical properties. An
Alx Ga1−x N alloy is a hexagonal crystal that can be obtained by replacing in the
GaN crystal a percentage of Ga-atoms with Al-atoms. As anticipated in Section
1.2.2, an important characteristic of the Alx Ga1−x N alloys is the possibility to tai-
lor the lattice parameter and the energy gap by varying the Al concentration.
In particular, the in-plane lattice parameter aAlGaN
0 of the Alx Ga1−x N alloy and
the Al-concentration x are correlated by the expression [19]:
aAlGaN
0 (x) = xaAlN
0 + (1 − x)aGaN
0 (1.6)
where aGaN
0 and aAlN
0 are the lattice parameters of GaN and AlN, respectively.
On the other hand, the energy gap EgAlGaN (x) of an Alx Ga1−x N alloy can be writ-
ten as a function of the energy gap of GaN EgGaN and AlN EgAlN [101]:
EgAlGaN (x) = xEgAlN + (1 − x)EgGaN − x(1 − x)1.0 eV (1.7)

where b = 1.0 eV is the bowing parameter typically used for AlGaN alloys [90].
The dependences of the in-plane lattice parameter a0 and of the energy gap Eg
on the Al-concentration x in an Alx Ga1−x N alloy are shown in Figure 1.11.
An AlGaN/GaN heterostructure can be formed by growing a thin Alx Ga1−x N
barrier layer onto a GaN substrate along the [0001] crystallographic direction.
Because of the different energy gaps of these two materials, an energy discon-
tinuity will appear in the band diagram. Furthermore, the lattice mismatch
between GaN and Alx Ga1−x N (aGaN0 > aAlGaN
0 ) will induce a tensile strain in
3.20
AlxGa1–xN 6.0 AlxGa1–xN
Lattice parameter a0 (Å)
3.18 5.5
Energy gap Eg (eV)
3.16 5.0
4.5
3.14
4.0
3.12
3.5
3.10 3.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(a) Al concentration x (b) Al concentration x
Figure 1.11 Dependence of the in-plane lattice parameter a0 (a) and of the energy gap E g (b)
on the Al-concentration x in an Alx Ga1−x N alloy. Source: Adapted with permission of
Roccaforte et al. [49]. Copyright © 2018, Società Italiana di Fisica.
a0(GaN) > a0(AlGaN) a0(GaN) = a0(AlGaN) > a0(AlGaN)
AlGaN
(tensile strain)
GaN
(relaxed)
Substrate Substrate
(a) (b)
Figure 1.12 Schematic of isolated AlGaN and GaN crystals (a) and of the formed AlGaN/GaN
heterostructure (b). After the growth of the AlGaN layer on GaN, a tensile strain is induced to
compensate the lattice mismatch between the two materials. Source: Adapted with
permission of Roccaforte et al. [49]. Copyright © 2018, Società Italiana di Fisica.
the Alx Ga1−x N barrier layer to compensate the in-plane lattice mismatch. This
situation is schematically depicted in Figure 1.12, showing the isolated AlGaN
and GaN crystals and the strained AlGaN/GaN heterostructure.
In the strained AlGaN/GaN heterostructure, a piezoelectric polarization PPE
will be induced along the c-axis, given by [19]:
PPE = e33 𝜀z + e31 (𝜀x + 𝜀y ) (1.8)
where e33 and e31 are the piezoelectric coefficients, 𝜀z = (c − c0 )/c0 is the strain
along the c-axis, 𝜀x = 𝜀y = (a − a0 )/a0 are the in-plane strains assumed to be
isotropic, and a0 and c0 are the equilibrium lattice constants.
The piezoelectric polarization along the c-axis can be also expressed as [19]:
( )
a − a0 C13
PPE = 2 e31 − e33 (1.9)
a0 C33
where C 13 and C 33 are the elastic constants of the material.
Because the term [e31 − e33 (C 13 /C 33 )] is negative in the whole Al-concentration
range typically used in Alx Ga1−x N alloys [19], the piezoelectric polarization will
be negative for tensile strain (aAlGaN > aAlGaN
0 ) and positive for compressive strain
(aAlGaN < aAlGaN
0 ). Hence, for a Ga-face AlGaN/GaN heterostructure with the
AlGaN barrier layer under tensile strain, the piezoelectric polarization PPE will
be negative and parallel to the spontaneous polarization PSP (directed toward
the GaN substrate), as indicated in Figure 1.13a.
The polarization gradient existing at the AlGaN/GaN interface deter-
mines a polarization-induced charge density, which in turn depends on the
Al-concentration x:
|𝜎(x)| = |[PSP (Alx Ga1−x N) + PPE (Alx Ga1−x N) − PSP (GaN)]| (1.10)
Hence, to maintain the charge neutrality in the system, free electrons will tend
to compensate the polarization-induced charge density at the AlGaN/GaN inter-
face, generating a 2DEG. The 2DEG is accumulated in the potential well formed
at the AlGaN/GaN interface (Figure 1.13b).
Ibbetson et al. [21] reported that the polarization-induced charges 𝜎(x) in
Eq. (1.10) represent a dipole, whose net contribution to the total charge in the
AlGaN/GaN (Ga-face)
2DEG
PPE(AlGaN) PSP(AlGaN)
ΦB EC
AlGaN ΔEC
GaN EF
PSP(GaN) AlGaN GaN EV
[0001]
Al
Ga
N
d
(a) (b)
Figure 1.13 (a) Schematic of an AlGaN/GaN heterostructure, showing the spontaneous and
piezoelectric polarization vectors and (b) schematic band diagram of an AlGaN/GaN
heterostructure. The presence of 2DEG in the quantum well at the interface is indicated by the
arrow. Source: Adapted with permission of Roccaforte et al. [49]. Copyright © 2018, Società
Italiana di Fisica.
AlGaN/GaN system is zero. Accordingly, as no assumption can be made on the

position of the Fermi level at the free surface of AlGaN, the presence of 2DEG
can be explained by an electron transfer from donor-like surface states 𝜎 surface
placed at an energy ED into empty states in GaN placed at a lower energy [21].
However, the situation occurring in a real device is different because the mate-
rial surface is not free. In fact, in an AlGaN/GaN device, a Schottky metal elec-
trode is formed on the AlGaN surface and the application of a bias is used to
modulate the sheet carrier density of the 2DEG nS . In the presence of a Schottky
metal, the maximum sheet carrier density of the 2DEG can be expressed as [19]:
[ ]
𝜎(x) 𝜀0 𝜀AlGaN (x)
nS (x) = − ⋅ [qΦB (x) + EF (x) − ΔEC (x)] (1.11)
q dAlGaN q2
where dAlGaN is the Alx Ga1−x N barrier layer thickness, 𝜀AlGaN is its permittivity,
qΦB is the Schottky barrier height of the metal contact, EF is the position of the
Fermi level with respect to the GaN conduction band edge energy, and ΔEC is the
conduction band offset at the AlGaN/GaN interface.
Typically, the 2DEG generated in AlGaN/GaN heterostructures is character-
ized by sheet carrier density values in the order of 1013 cm−2 and mobility in the
range of 1000–2000 cm2 /V s.
The HEMT is a device whose working principle is based on the presence of the
2DEG in AlGaN/GaN heterostructures. HEMT technology will be widely dis-
cussed in other chapters of this book.
A schematic cross section of this device is depicted in Figure 1.14a. In par-
ticular, in a conventional AlGaN/GaN HEMT, the current flowing in the 2DEG
channel, between a source and a drain Ohmic electrode, is modulated by the
application of a negative bias to a Schottky contact acting as a gate electrode of
the transistor.
As the 2DEG is naturally present in the AlGaN/GaN heterostructure and
the Fermi level at the interface lies above the conduction band minimum (see
schematic in Figure 1.13b), such a device is “normally-on,” i.e. a current will flow
between source and drain even when the gate bias is zero (V g = 0). An example
of typical output I DS −V DS characteristic is depicted in Figure 1.14b. As can be
AlGAN/GaN HEMT
Output characteristics
300
Source Gate Drain
Drain current IDS (mA/mm)

VG = 0 V
250
AlxGa1–xN
200 VG = –1 V
2DEG GaN VG ≤ 0
150
VG = –2 V
Buffer layer 100
Substrate 50 VG = –3 V
VG = –4 V
High electron mobility transistor 0 VG = Vth
0 2 4 6 8 10
(a) (b) Drain bias VDS (V)
Figure 1.14 (a) Schematic cross section of an AlGaN/GaN HEMT and (b) typical output IDS –V DS
characteristic of a device.
seen, the output current of a HEMT can be modulated by the application of a

negative bias to the gate, until reaching a “threshold voltage” (V th ) when the
Fermi level is pulled down below the conduction band edge of the AlGaN and
the 2DEG channel is depleted.
The threshold voltage V th of an AlGaN/GaN HEMT depends on the het-
erostructure properties (i.e. AlGaN thickness and doping, Al concentration) and
can be written as [102]:
2
qND dAlGaN 𝜎(x)
Vth (x) = ΦB (x) + EF (x) − ΔEC (x) − − d
2𝜀0 𝜀AlGaN (x) 𝜀0 𝜀AlGaN (x) AlGaN
(1.12)
where N D is the doping density of the AlGaN barrier layer expressed in

atoms/cm3 .
In normally-on HEMTs, the typical values of the AlGaN barrier thickness are
of about 20–25 nm, with Al concentration in the range x = 0.25–0.30. In such
AlGaN/GaN heterostructures, the values of the 2DEG sheet carrier density ns are
in the range of 0.7–1 × 1013 cm−2 , which results in negative values of V th around
−4 V. However, in power electronic applications, “normally-off” devices (i.e. with
a V th > 0) are strongly preferred [103–105]. A recent overview of the most com-
mon technologies for achieving normally-off HEMTs is given in Ref. [106]. This
important topic will be discussed in Chapter 4.
The high saturation velocity of the material and the high mobility of the 2DEG
in AlGaN/GaN heterostructures enable fast switching frequency of HEMTs. In
fact, for high-frequency applications, it is necessary to shorten the transit time 𝜏
between the source–drain spacing Lsd of the switching transistor.
Considering the transit time of electron in the distance Lsd , the cutoff frequency
f T , i.e. the frequency at which the current gain becomes unity, can be written as:
1 v
fT = = sat (1.13)
2𝜋𝜏 2𝜋Lsd
which can be also expressed in terms of the transconductance g m and of the gate
capacitance C g as:
g
fT = m (1.14)
2𝜋Cg
Assuming the saturation electron velocity of GaN, from Eq. (1.13), it can be
deduced that submicron gate HEMTs are able to operate at frequencies in the
millimeter-wave (mmW) range [30].
The use of a Schottky contact as gate electrode in GaN-based HEMTs can
be a limitation both for the off-state characteristics of the devices (i.e. leakage
current) and for the maximum gate voltage swing (i.e. the current capability in
the on-state). For that reason, the introduction of an insulating layer below the
gate electrode is a common solution in GaN-based HEMT technology, espe-
cially for high-voltage applications. GaN HEMTs employing the insulated gate
are often called metal–insulator–semiconductor high-electron mobility transis-
tors (MISHEMTs). The problems related to insulated gate HEMTs will be dis-
cussed in the following chapters.
Some final considerations regard the heterostructures used for GaN-based
HEMTs. Nowadays, GaN-based HEMTs for high-power and high-frequency
operation are mostly fabricated using AlGaN/GaN heterostructures. However,
the presence of a tensile stress in the AlGaN barrier layer, and the consequent
relaxation effects, can be a limit of the heterostructure design. This aspect is
particularly important for high-frequency applications, where a thinner barrier
layer is needed to improve the capability of gate modulation, and to increase the
device transconductance. Hence, a promising alternative to AlGaN as a barrier
layer is the ternary alloy Inx Al1−x N, which is lattice-matched to GaN for an InN
mole fraction of around 17% [107]. At this composition, the stress and piezo-
electric polarization are not present, thus potentially improving the stability of
the heterostructure [108]. Even in the absence of piezoelectric polarization, the
2DEG sheet charge density induced by the difference in spontaneous polariza-
tion is typically larger than in the conventional AlGaN/GaN heterostructure.
This should result in a higher output current and power density of the device
[108]. Clearly, as discussed in Chapter 6, the high current densities achieved
in InAlN/GaN HEMTs result in significant self-heating effects, which must be
appropriately faced in order to limit detrimental effects for device reliability.
AlN/GaN heterostructures are another system that is particularly advan-
tageous for high-frequency (mmW) HEMT operation. In fact, the use of an
ultrathin (e.g. ∼3 nm) AlN barrier ensures the submicrometric device scaling
without incurring in the 2DEG degradation [109, 110]. Design considerations
and critical processing steps for high-frequency AlN/GaN HEMTs will be
discussed in Chapter 3.
1.3 Applications of GaN-Based Materials

GaN-based materials and heterostructures have several applications in both
optoelectronic and power- and high-frequency devices. In the following sections,
a short overview of the main applications of nitride devices will be given, also
mentioning some of the most relevant physical and technological open issues.
These concepts will be useful as an introduction to the contents of the other
chapters of the book devoted to devices.
1.3.1 Optoelectronic Devices

The scientific and technological achievements reached in nitride-based optoelec-
tronic devices since the 1980s have led and are leading to creation of new multibil-
lion markets. Figure 1.15 illustrates the examples of consumer applications of the
nitride-based optoelectronic devices, i.e. LEDs and LDs, in several fields (general
lighting, cars, video projections, etc.).
White LEDs have made a revolution in lighting, contributing significantly in
a decrease of energy consumption. However, still white LEDs are not based on
red–green–blue (RGB) emitters because green LEDs have too low efficiency
(green gap). Instead, white LEDs are constructed using blue LEDs illuminating
phosphor to excite light of longer wavelengths. Such white LEDs have become
very popular, as they are used as bulbs or Christmas-tree illumination, but also
in most of the computer screens.
Blue LEDs can be switched on and off much faster than the incandescent bulbs,
thus making possible to use light to transmit information (Li-Fi instead of Wi-Fi).
Even faster modulation (10 times compared to LEDs) can be achieved using blue
LDs. Most likely, nitride-based LEDs and LDs will be widely used in smart cities
to control traffic and other people activities in parallel to illumination.
Blue LDs are mostly recognized as those used in the BluRay for recording and
playing the compact disks, although this technology will probably vanish soon.
Nevertheless, nitride LDs have many other applications, e.g. they can be used for
exciting phosphor to get white light. Such head lamps are produced and installed
in luxury (so far) cars.
Nitrides optoelectronics applications

Light emitting diodes (LEDs) Laser diodes (LDs)
General lighting Blue ray Industry and
medicine
Car lighting Smart cities Projection
Figure 1.15 Examples of applications of nitrides devices (LEDs and LDs) in optoelectronics.
Lighting and Li-Fi are markets of tens-billion-Euro size, but even bigger one will
be the one of RGB screens and movie projectors (blue and green emitters based
on nitrides, red based on arsenides/phosphides). These RGB projectors will vary
in size, the smallest for mobile phones, medium for TV sets, and the largest as
billboards and for cinema theaters.
RGB screens based on LEDs are already installed as outdoor billboards. Each
contains tens of thousands of LEDs. However, for the TV set and computer
screens, not to mention the mobile phones, one needs much smaller pixels. This
technology is almost ready, and RGB micro-LEDs will soon replace the other
solutions.
Using RGB LDs makes it possible to obtain extra-high color resolution projec-
tors, as well as to create three-dimensional images without using any goggles.
The third (after lighting and RGB projectors) huge market for GaN-based LDs
could be “Last mile” Tb/s communication through plastic fibers. Such fibers can
transmit 490 nm light and have a big advantage of being cheap, light, and more
resistant to shocks as compared to glass waveguides. Therefore, they are of the
first choice for battleships and army vehicles. Then, every house, aircraft, car, etc.,
could have such installation of plastic fibers to transmit data. However, it is not
clear if the “Last mile” market will be developing because of the disadvantage of
additional stage of signal processing.
A very interesting application of blue and green LDs of well-defined wave-
lengths is in the Quantum Technologies [111]. The LDs are used for cooling the
atoms down to micro-kelvins and for exciting these atoms in atomic clocks, which
are able to measure time with accuracy of fractions of picoseconds. Such clocks
will enable GPS to measure the position with a high precision and to construct
gravimeters to be used in geology and in detection of buried objects.
Moreover, there are several other niche markets for nitride LDs, e.g. in
medicine, in environmental protection, as well as for welding of copper and gold.
Light in blue and green emitters is created in InGaN QWs, while using AlGaN
enables to obtain UV radiation. Recent achievements of many laboratories in
technology of 260–280 nm LEDs open interesting application perspectives in the
disinfection of water, air, and food [112].
In the case of LDs, while the shortest wavelength demonstrated at the R&D
level till 2019 has been of 340 nm [113], the commercially available devices oper-
ate in the range of 370–380 nm (www.nichia.co.jp/en/product/laser.html). Very
recently, Z. Zhang et al. [38] demonstrated a deep UV device emitting at 271.8 nm
in the pulse mode. The epitaxy of this device was done on bulk AlN crystal and
p-doping was achieved by polarization-induced doping. The parameters (high
voltage and high threshold current) of this deep UV LD were still far behind the
devices emitting at the blue/green range, but this achievement will pave the way
to the UV LDs, which would be used in medicine for sterilization or cancer curing
via the waveguides.
In spite of the spectacular success of “GaNification” in optoelectronics, still
there are many challenges that must be overcome.
The performance of blue and green LEDs is pegged back by three phenomena:
droop (the gradual decrease in efficiency as the drive current increases), green
gap (the gradual decrease in efficiency as the indium content in InGaN QWs
rises), and the electrical potential drop (increase in the voltage to be provided
to increase current). An excellent discussion of those three phenomena is given
by Han et al. [114].
On the other hand, in the case of nitride LDs, there are still many issues to
be resolved. The low crystallographic quality, small size, and high price of GaN
substrates is the main obstacle for a quick development of blue and green LD
technology. As mentioned in Section 1.2.1, most of the nitride research- and-
technology is based on foreign substrates, e.g. sapphire and silicon. The difference
of lattice parameters and thermal expansion between GaN and those substrates
results not only in a very high dislocation density but also in wafer bowing, which
in turn makes lithography very difficult. The bulk GaN substrates are consid-
ered the future for optoelectronic devices, and in particular for LDs with long
lifetime. Nowadays, such substrates are manufactured by number of companies
all over the world (Sciocs, St. Gobain Lumilog, Furukawa, Nanowin, Ammono,
Sumitomo, etc.). However, these crystals contain large densities of dislocations
(104 –107 cm−2 ) and their size is typically limited to 2-inches, although 4- and
6-inches are already being introduced into the market. The prices of the GaN
substrates are about 10 times higher than those of GaAs, or SiC, and even 100
times as compared to sapphire.
Another technical limitation is given by the low wall-plug efficiencies in the
green region related to poor crystallographic quality and poor p-type of InGaN
and GaN grown at low temperatures.
Finally, the lack of stimulated emission in the UV region related to poor p-type
of AlGaN and the presence of point defects are also current bottlenecks in LD
technology.
Many of these open issues will be discussed in Chapters 7–10.
1.3.2 Power- and High-Frequency Electronic Devices

Power electronics is the key enabling technology devoted to the control and
management of electric power. In particular, the primary goal of a power elec-
tronic system is to provide the electric power in the optimal form (i.e. in terms of
current, voltage, frequency, etc.) for the end user load. Hence, power electronic
devices are used on a daily basis in our society in many fields, e.g. power supplies
for computers, industrial motor drives, energy conversion systems in hybrid
electric vehicles (HEVs), and inverters for renewable energies.
For some decades, silicon (Si) has been the semiconductor of choice for power
devices because of its natural abundance, low cost, excellent crystalline quality,
and device processing maturity. However, today, Si power devices have reached
their operational limits, set by the intrinsic properties of the material, and are
affected by significant power losses in practical applications. Hence, the improve-
ment of energy efficiency in power electronics is one of the challenges of our
society to reduce the global energy consumption. In this context, the introduc-
tion of new semiconductors technologies to overcome the current limitations of
Si devices has become mandatory.
Because of their excellent electronic properties, WBG semiconductors are con-
sidered the materials of choice for the future energy efficient power devices [39].
Low voltage Medium voltage High voltage

PFC/power supply PV inverter Motor control Ships Power grid
Wind energy Train

EV/HEV UPS
transportation
Audio amplifier
<200 V 600 V 900 V 1.2 kV 1.7 kV 3.3 kV >6.5 kV

SiC/GaN coexistence
GaN (low/medium voltage) SiC (medium/high voltage)
Figure 1.16 Potential applications of GaN power devices as a function of the voltage. For
comparison, the application area of SiC is also indicated. Source: Adapted from Roccaforte
et al. [106]. Copyright © 2019 by the authors; licensee MDPI, Basel, Switzerland. Reference [106]
is an open access article distributed under the terms and conditions of the Creative Commons
Attribution License.
Among them, while SiC [115] is the most fitted in terms of crystalline quality and
device maturity, GaN and related alloys are very promising but still suffer from
many concerns, which hinder their full exploitation in power electronic applica-
tions.
Figure 1.16 illustrates the potential applications of GaN devices in power elec-
tronics in the low-, medium-, and high-voltage range. For comparison, the typical
range of application of the other WBG semiconductor SiC is also indicated.
According to the current opinion of market analysts, GaN could be better
suited for the low-medium voltage range (200–600 V), which includes a part
of the consumer electronic market (e.g. computer power supplies and audio
amplifiers). In this voltage range, the material is the best candidate to replace
the existing Si devices. The voltage range 600–900 V is strategic, as it covers the
converters for electric vehicles (EVs) and HEV, and the inverters for renewable
energy (e.g. photovoltaic). Here, GaN is expected to be in competition or
to coexist with SiC. At higher voltage (>1.2 kV, e.g. industrial applications,
trains/ships transportation, and electric energy distribution grids), 4H–SiC is
considered as the preferable choice, owing to a better material quality and device
reliability. The future applications of GaN for high-voltage devices will strongly
depend on the improvement of the material quality and the development of
vertical devices based on bulk GaN.
In general, considering the case of unipolar power devices, the specific
on-resistance RON (expressed in Ω cm2 ) can be approximated by the contribution
of the device drift layer [97]:
4B2V
RON ≅ (1.15)
𝜀0 𝜀GaN 𝜇n ECR
3
where BV is the targeted breakdown voltage, 𝜀GaN is the permittivity of GaN, 𝜇n

is the electron mobility, and ECR is the critical electric field of the material.
Normally-on HEM Ts
1000 Dora 2006
Uemoto 2007
it
lim
Shi 2009
lar
Tipirneni 2006
Specific RON (mΩ cm2)
ipo
Ikeda 2008
100
t
mi
un
Bahat-Treidel 2010
r li
Si
la
ipo
Normally-off HEM Ts
un
Kaneko 2009
10
iC
Hilt 2011
–S
t
Chu 2011
mi
4H
Freedsman 2014
r li
ola
Zhang 2016
1 Huang 2018
ip
un
IR (cascode)
N
Transphorm (cascode)
Ga Microgan (cascode)
EPC (p-GaN)
0.1 Panasonic (p-GaN)
100 1 000 10 000 GaNSystem (p-GaN)
Breakdown voltage Bv (V)
Figure 1.17 Trade-off of the specific on-resistance RON as a function of the breakdown voltage
BV for Si, 4H–SiC, and GaN. A collection of experimental literature data for normally-on and
normally-off GaN HEMTs is also reported [45, 49, 116].
The high critical field ECR of GaN enables the fabricated devices to sustain
high-voltage levels with thinner drift layers. Hence, the specific on-resistance RON
can be reduced and a smaller device area will be required to reach a given current.
The low RON translates into a lower power dissipation in switching devices, which
is a key requirement for a better energy efficiency in power electronic systems.
Figure 1.17 reports the theoretical trade-off between the specific on-resistance
RON and the breakdown voltage BV for Si, SiC, and GaN unipolar devices. A col-
lection of literature data for GaN-based HEMTs (normally-on and normally-off )
is also reported.
From this plot, it can be deduced that the theoretical limits are still far to be
reached. The discrepancy between the state-of-the-art data and the theoretical
limits can be associated with the existing issues in material quality and device
processing [45, 49, 116, 117].
Metal/semiconductor contacts are important bricks of any electronic device. In
particular, in GaN devices, the formation of Ohmic contacts with a low specific
contact resistance 𝜌c and Schottky contacts with an adequate barrier and a low
leakage is required to minimize the device power consumption and improve the
reliability [118].
The formation of Ohmic contacts with a low 𝜌c is a challenging issue in GaN and
related alloys. In fact, the values of metal/semiconductor Schottky barrier height
on WBG semiconductors are typically higher than in Si and, hence, low values of
𝜌c (in the order of 10−4 –10−6 Ω cm2 ) are difficult to obtain. Several metallization
schemes have been proposed as Ohmic contacts to n-type GaN, as reported in a
good review by Greco et al. [119].
In general, sequences of several metal layers are employed [120]. These stacks
consist of a low work function metal (Ti) deposited on GaN and covered by an
overlayer (Al). Then, a barrier layer is inserted to limit the interdiffusion between
metals during annealing (Ni, Ti, Pt, Pd, Mo, etc.) and a top cap layer (Au) to pre-
vent oxidation.
Recently, the interest toward GaN-based heterostructures grown on large-area

Si substrates is grown because of the possibility of fabricating GaN devices using
the Si CMOS facilities. In this context, to prevent contaminations and make GaN
technology compatible with the Si CMOS line standards, the implementation of
“Au-free” metallization schemes becomes mandatory.
On p-type GaN-based materials, the formation of Ohmic contacts is even more
challenging because of the high metal/GaN barrier values and the high ionization
energy of the Mg p-type dopant [121]. Also in the case of p-type GaN, various
metal schemes have been investigated to form Ohmic contacts [119]. High work
function metals (Ni, Pt, and Pd) are preferred, as they are expected to form a low
Schottky barrier on the p-type semiconductor. Then, a cap layer of a noble metal
(e.g. Au and Ag) is deposited on the top to prevent the oxidation. The Ni/Au
bilayer is a very common scheme used for Ohmic contacts to p-GaN. Interest-
ingly, Ohmic contact formation on p-type GaN is favored upon annealing in
oxidizing atmosphere [122], with a complex mechanism that is widely debated
in the literature.
The important role of contacts in GaN-based LEDs, LDs, and HEMTs will be
mentioned in some chapters of the book.
Schottky contacts are used as gate metallization in GaN HEMTs to modulate
the 2DEG concentration, Ni, Pt, and Au being the most widely used metals. How-
ever, the Schottky barriers on (Al)GaN often suffer from non-ideality issues and
high leakage current, which limit the gate voltage swing in power transistors.
Hence, dielectric materials can be used to insulate the Schottky gate and reduce
the leakage current [123]. Moreover, dielectrics are very important as surface
passivation to limit the so-called “current collapse” phenomena in GaN HEMTs
[124–127] caused by the trapping of electrons at the device surface and/or in the
buffer. Several dielectrics have been investigated as gate insulators and passivat-
ing layers for GaN HEMTs (SiNx , SiO2 , Al2 O3 , etc.) [123, 128]. In this context, the
reliability is an important concern in power electronic devices. As an example,
a threshold voltage instability in GaN transistors can be detrimental in power
switching applications. The occurrence of charge trapping effects at interfaces
in the buffer layer or in the gate dielectrics are typically responsible for the V th
instability upon bias stress [129]. Typically, positive V th shifts lead to a degrada-
tion of the device on-resistance, while negative V th shifts can lead to the loss of
the normally-off behavior. The reliability aspects associated with GaN electronic
devices will be described in Chapter 6.
As described in Section 1.2.4, the HEMT working principle is based on the
presence of the 2DEG conduction channel, making these devices inherently
normally-on. Hence, the current is modulated by the application of a negative
bias to a Schottky gate electrode. However, in power electronic applications,
normally-off switches are preferred, as they provide fail-safe operation con-
ditions and gate-driver circuitry simplicity [103–106]. Hence, the academic
community and many industrial players working on GaN are spending significant
efforts on the development of reliable normally-off technologies.
Under the physical point of view, to obtain the desired normally-off HEMT
behavior, the region near the gate must be modified by employing appropriate
near-surface processing or band engineering techniques. Several approaches
have been reported in the literature to achieve normally-off GaN-based HEMTs,

the most common being the p-GaN gate and the recessed gate hybrid MISHEMT.
The advantages and disadvantages of all these technologies will be critically
described in Chapter 4.
In terms of operation frequency, the interest for the mmW band is steadily
increasing, owing to the reduced wavelength and wide frequency bands, enabling
smaller components with improved performances. In fact, today, the wireless
communication systems are extending to higher frequencies because of the
need for more bandwidth for several emerging applications. In this context,
GaN-based devices can represent the best choice for high-power amplifiers
(HPAs), broadband amplifiers, and 5G wireless communication networks.
However, also in this case, several technological challenges need to be overcome
in order to achieve the optimal device performance, as will be discussed in
Chapter 3.
Clearly, the high-frequency performances of GaN-based HEMTs are ulti-
mately limited by the lateral scaling of the device channel. In this context,
the integration of two-dimensional (2D) materials with nitrides has recently
opened the way toward alternative devices beyond the HEMTs, targeted for
ultra-high-frequency operation. As an example, the hot electron transistor
(HET) is a device based on the transversal ballistic transport of hot electrons
through an ultrathin base layer. As will be discussed in the Chapter 11, graphene
junctions with Al(Ga)N/GaN heterostructures have been recently employed to
implement vertical HETs potentially able to work in the terahertz frequency
range.
Finally, it is worth mentioning that GaN electronics is dominated by lateral
transistors fabricated on AlGaN/GaN heterostructures grown on foreign
substrates. However, vertical devices based on bulk GaN are strongly desired
in power electronics. In fact, the vertical topology can enable to increase
the breakdown voltage by increasing the thickness of the drift region, while
maintaining the chip size constant. Furthermore, the maximum electric field
in vertical GaN devices is moved far away from the surface into the bulk, thus
enabling to minimize the trapping phenomena and to eliminate the current
collapse. Not least, vertical devices on high-quality bulk GaN crystals can allow
to achieve a higher power density than lateral devices, thanks to the lower RON
and higher current capability and to the higher thermal conductivity of the bulk
material. The status and perspectives of vertical GaN devices will be discussed
in Chapter 5.
1.4 Summary
In summary, this chapter gave an overview of the main physical, optical, and
electrical properties of GaN-based materials. The direct band gap, combined
with the possibility to have the desired wavelength by tailoring the composition,
is the key aspect that allowed use of nitride semiconductors in optoelectronics
successfully. However, the outstanding characteristics, in terms of electric field
References 31
strength and piezoelectric properties of AlGaN/GaN heterostructures, also

make these semiconductors excellent candidates for the next generation of
power- and high-frequency devices.
Although GaN-based devices (LEDs, LDs, and transistors) have already entered
our daily life, there are still many problems related to material growth and device
processing that deserve further intensive research.
Acknowledgments
The authors would like to thank their coworkers and the coauthors of the other
book chapters for the fruitful inputs provided during the preparation of this
chapter. This work is the outcome of a long-standing collaboration between the
Institute for Microelectronics and Microsystems of the National Research Coun-
cil of Italy (CNR-IMM) in Catania and the Institute of High Pressure Physics
of the Polish Academy of Sciences (Unipress-PAS) in Warsaw. In particular, the
authors would like to acknowledge the bilateral project ETNA (Energy efficiency
Through Novel AlGaN/GaN heterostructures) within the CNR-PAS Cooperation
Agreement for the years 2017–2019 and the bilateral project GaNIMEDE
(Gallium Nitride Innovative Micro-Electronics DEvices) within the Executive
Programme for Scientific and Technological Cooperation between The Italian
Republic and the Republic of Poland for the years 2019–2020.
References
1 Johnson, W.C., Parsons, J.B., and Crew, M.C. (1932). Nitrogen compounds of
gallium – III. gallium nitride. J. Phys. Chem. 36: 2651–2654.
2 Juza, R. and Hahn, H. (1938). Über die Kristallstrukturen von Cu3 N,
GaN und InN metallamide und metallnitride. Z. Anorg. Allg. Chem. 239:
282–287.
3 Maruska, H.P. and Tietjen, J.J. (1969). The preparation and properties of
vapor-deposited single-crystalline GaN. Appl. Phys. Lett. 15: 327–329.
4 Maruska, H.P., Stevenson, D.A., and Pankove, J.I. (1973). Violet luminescence
of Mg-doped GaN. Appl. Phys. Lett. 22: 303–305.
5 Manasevit, H.M., Erdmann, F.M., and Simpson, W.I. (1971). The use of met-
alorganics in the preparation of semiconductor materials – IV. The nitrides
of aluminum and gallium. J. Electrochem. Soc. 118: 1864–1868.
6 Manasevit, H.M. (1972). The use of metalorganics in the preparation of
semiconductor materials: growth on insulating substrates. J. Cryst. Growth
13–14: 306–314.
7 Karpinski, J., Jun, J., and Porowski, S. (1984). Equilibrium pressure of N2
over GaN and high-pressure solution growth of GaN. J. Cryst. Growth 66:
1–10.
8 Karpinski, J. and Porowski, S. (1984). High-pressure thermodynamics of
GaN. J. Cryst. Growth 66: 11–20.
9 Amano, H., Sawaki, N., Akasaki, I., and Toyoda, Y. (1986). Metalorganic
vapor phase epitaxial growth of a high quality GaN film using an AlN buffer
layer. Appl. Phys. Lett. 48: 353–355.
10 Amano, H., Kito, M., Hiramatsu, K., and Akasaki, I. (1989). P-type conduc-
tion in Mg-doped GaN treated with low-energy electron beam irradiation
(LEEBI). Jpn. J. Appl. Phys. 28: L2112–L2114.
11 Matsuoka, T., Tanaka, H., Sasaki, T., and Katsui, A. (1990). Gallium arsenide
and related comp. Inst. Phys. Conf. Ser. 106: 141.
12 Nakamura, S., Senoh, M., and Mukai, T. (1991). High-power GaN P–N junc-
tion blue-light-emitting diodes. Jpn. J. Appl. Phys. 30, 1991: L1708–L1711.
13 Nakamura, S., Mukai, T., Senoh, M., and Iwasa, N. (1992). Thermal
annealing effects on P-type Mg-doped GaN films. Jpn. J. Appl. Phys. 31:
L139–L142.
14 Asif Khan, M., Van Hove, J.M., Kuznia, J.N., and Olson, D.T. (1991). High
electron mobility GaN/Alx Ga1−x N heterostructures grown by low-pressure
metalorganic chemical vapor deposition. Appl. Phys. Lett. 58: 2408–2410.
15 Aisf Khan, M., Bhattarai, A.R., Kuznia, J.N., and Olson, D.T. (1993). High
electron mobility transistor based on a GaN/Alx Ga1−x N heterojunction.
Appl. Phys. Lett. 63: 1214–1215.
16 Nakamura, S., Senoh, M., Nagahama, S. et al. (1996). InGaN-based
multi-quantum-well-structure laser diodes. Jpn. J. Appl. Phys. 35: L74–L76.
17 Bernardini, F., Fiorentini, V., and Vanderbilt, D. (1997). Spontaneous polar-
ization and piezoelectric constants of III–V nitrides. Phys. Rev. B 56:
R10024–R10027.
18 Guha, S. and Bojarczuk, N.A. (1998). Ultraviolet and violet GaN light emit-
ting diodes on silicon. Appl. Phys. Lett. 72: 415–417.
19 Ambacher, O., Smart, J., Shealy, J. et al. (1999). Two-dimensional electron
gases induced by spontaneous and piezoelectric polarization charges in N-
and Ga-face AlGaN/GaN heterostructures. J. Appl. Phys. 85: 3222–3233.
20 Sheppard, S.T., Doverspike, K., Pribble, W.L. et al. (1999). High-power
microwave GaN/AlGaN HEMT’s on semi-insulating silicon carbide sub-
strates. IEEE Electron Device Lett. 20: 161–163.
21 Ibbetson, J.P., Fini, P.T., Ness, K.D. et al. (2000). Polarization effects, surface
states, and the source of electrons in AlGaN/GaN heterostructure field effect
transistors. Appl. Phys. Lett. 77: 250–252.
22 Motoki, K., Okahisa, T., Matsumoto, N. et al. (2001). Preparation of large
freestanding GaN substrates by hydride vapor phase epitaxy using GaAs as a
starting substrate. Jpn. J. Appl. Phys. 40: L140–L143.
23 Motoki, K. (2010). Development of gallium nitride substrates. SEI Tech. Rev.
70: 28–35.
24 Saito, W., Takada, Y., Kuraguchi, M. et al. (2006). Recessed-gate structure
approach toward normally-off high-voltage AlGaN/GaN HEMT for power
electronics applications. IEEE Electron Device Lett. 53: 356–362.
25 Cai, Y., Zhou, Y., Lau, K.M., and Chen, K.J. (2006). Control of threshold volt-
age of AlGaN/GaN HEMTs by fluoride-based plasma treatment: from deple-
tion mode to enhancement mode. IEEE Electron Device Lett. 53: 2207–2215.
References 33
26 Uemoto, Y., Hikita, M., Ueno, H. et al. (2007). Gate injection transistor
(GIT) – a normally-off AlGaN/GaN power transistor using conductivity
modulation. IEEE Trans. Electron Devices 54: 3393–3399.
27 Lu, T.-C., Kao, C.-C., Kuo, H.-C. et al. (2008). CW lasing of current injec-
tion blue GaN-based vertical cavity surface emitting laser. Appl. Phys. Lett.
92 (2008): 141102.
28 Tripathy, S., Lin, V.K.X., Tan, J.P.Y. et al. (2012). AlGaN/GaN
two-dimensional-electron gas heterostructures on 200 mm diameter Si(111).
Appl. Phys. Lett. 101: 082110.
29 Iveland, J., Martinelli, L., Peretti, J. et al. (2013). Direct measurement of
Auger electrons emitted from a semiconductor light-emitting diode under
electrical injection: identification of the dominant mechanism for efficiency
droop. Phys. Rev. Lett. 110: 177406.
30 Shinohara, K., Regan, D.C., Tang, Y. et al. (2013). Scaling of GaN HEMTs
and Schottky diodes for submillimeter-wave MMIC applications. IEEE Trans.
Electron Devices 60 (10): 2982–2996.
31 Freedsman, J.J., Egawa, T., Yamaoka, Y. et al. (2014). Normally-OFF
Al2 O3 /AlGaN/GaN MOS-HEMT on 8 in. Si with low leakage current and
high breakdown voltage (825 V). Appl. Phys. Express 7: 041003.
32 Kikkawa, T., Hosoda, T., Shono, K. et al. (2015). Commercialization and
reliability of 600 V GaN power switches. Proceedings of IEEE International
Reliability Physics Symposium (IRPS 2015), Monterey, CA (19–23 April
2015), 6C.1.1.
33 Kaneko, S., Kuroda, M., Yanagihara, M. et al. (2015). Current-collapse-free
operations up to 850 V by GaN-GIT utilizing hole injection from drain.
Proceedings of the 27th International Symposium on Power Semiconductor
Devices & IC’s (ISPSD2015), Kowloon Shangri-La, Hong Kong (10–14 May
2015), pp. 41–44.
34 Tanaka, K., Morita, T., Umeda, H. et al. (2015). Suppression of cur-
rent collapse by hole injection from drain in a normally-off GaN-based
hybrid-drain-embedded gate injection transistor. Appl. Phys. Lett. 107 (16):
163502.
35 Tanaka, K., Morita, T., Ishida, M. et al. (2017). Reliability of
hybrid-drain-embedded gate injection transistor. Proceedings of IEEE Inter-
national Reliability Physics Symposium (IRPS 2017), Monterey, CA (2–6
April 2017), 4B-2.1.
36 Haller, C., Carlin, J.-F., Jacopin, G. et al. (2017). Burying non-radiative
defects in InGaN underlayer to increase InGaN/GaN quantum well effi-
ciency. Appl. Phys. Lett. 111: 262101.
37 Mei, Y., Weng, G.-E., Zhang, B.-P. et al. (2017). Quantum dot vertical-cavity
surface-emitting lasers covering the ‘green gap’. Light Sci. Appl. 6: e16199.
https://doi.org/10.1038/lsa.2016.199.
38 Zhang, Z., Kushimoto, M., Sakai, T. et al. (2019). A 271.8 nm deep ultravio-
let laser diode for room temperature operation. Appl. Phys. Express 12 (12):
124003.
39 Ren, F. and Zolper, J.C. (2003). Wide Band Gap Electronic Devices. Singa-
pore: World Scientific.
40 Pearton, S.J., Abernathy, C.R., and Ren, F. (2006). Gallium Nitride Processing
for Electronics, Sensors and Spintronics. Springer Verlag-London Ltd.
41 Quai, R. (2008). Gallium Nitride Electronics. Berlin Heidelberg:
Springer-Verlag.
42 Meneghini, M., Meneghesso, G., and Zanoni, E. (2017). Power GaN
Devices – Materials, Applications and Reliability. Switzerland: Springer
International Publishing.
43 Roccaforte, F., Giannazzo, F., Iucolano, F. et al. (2010). Surface and inter-
face issues in wide band gap semiconductor electronics. Appl. Surf. Sci. 256:
5727–5735.
44 Kizilyalli, I.C., Edwards, A.P., Nie, H. et al. (2013). High voltage vertical
GaN p–n diodes with avalanche capability. IEEE Trans. Electron Devices 60:
3067–3070.
45 Roccaforte, F., Fiorenza, P., Greco, G. et al. (2014). Challenges for energy
efficient wide band gap semiconductor power devices. Phys. Status Solidi (a)
211: 2063–2071.
46 Grabowski, S.P., Schneider, M., Nienhaus, H. et al. (2001). Electron affinity of
Alx Ga1−x N (0001) surfaces. Appl. Phys. Lett. 78: 2503–2505.
47 Cook, T.E., Fulton, C.C., Mecouch, W.J. et al. (2003). Band offset measure-
ments of the Si3 N4 /GaN (0001) interface. J. Appl. Phys. 94: 3949–3953.
48 Caldas, P.G., Silva, E.M., Prioli, R. et al. (2017). Plasticity and optical prop-
erties of GaN under highly localized nanoindentation stress fields. J. Appl.
Phys. 121: 125105.
49 Roccaforte, F., Fiorenza, P., Lo Nigro, R. et al. (2018). Physics and technology
of gallium nitride materials for power electronics. Riv. Nuovo Cimento 41:
625–681.
50 Leszczyński, M., Teisseyre, H., Suski, T. et al. (1996). Lattice parameters of
gallium nitride. Appl. Phys. Lett. 69: 73–75.
51 Darakchieva, V., Monemar, B., and Usui, A. (2007). On the lattice parame-
ters of GaN. Appl. Phys. Lett. 91: 031911.
52 Arehart, A., Homan, T., Wong, M.H. et al. (2010). Impact of N- and Ga-face
polarity on the incorporation of deep levels in n-type GaN grown by molec-
ular beam epitaxy. Appl. Phys. Lett. 96: 242112.
53 Yu, E.T., Dang, X.Z., Asbeck, P.M. et al. (1999). Spontaneous and piezoelec-
tric polarization effects in III–V nitride heterostructures. J. Vac. Sci. Technol.
B 17: 1742–1749.
54 Taniyasu, Y., Kasu, M., and Kobayashi, N. (2001). Lattice parameters of
wurtzite Alx Si1-x N ternary alloys. Appl. Phys. Lett. 79: 4351–4353.
55 Nilsson, D., Janzén, E., and Kakanakova-Georgieva, A. (2016). Lattice param-
eters of AlN bulk, homoepitaxial and heteroepitaxial material. J. Phys. D.
Appl. Phys. 49: 175108.
56 Paszkowicz, W., Adamczyk, J., Krukowski, S. et al. (1999). Lattice param-
eters, density and thermal expansion of InN microcrystals grown by
the reaction of nitrogen plasma with liquid indium. Philos. Mag. A 79:
1145–1154.
References 35
57 Kröncke, H., Figge, S., Epelbaum, B.M., and Hommel, D. (2008). Determi-
nation of the temperature dependent thermal expansion coefficients of bulk
AlN by HRXRD. Acta Phys. Pol. A 114: 1193–1200.
58 Leszczyński, M., Suski, T., Teisseyre, H. et al. (1994). Thermal expansion of
gallium nitride. J. Appl. Phys. 76: 4909–4911.
59 Liu, L. and Edgar, J.H. (2002). Substrates for gallium nitride epitaxy. Mater.
Sci. Eng. R 37: 61–127.
60 Krost, A. and Dadgar, A. (2002). GaN-based optoelectronics on silicon
substrates. Mater. Sci. Eng. B 93: 77–84.
61 Kizilyalli, I.C., Bui-Quanga, P., Disney, D. et al. (2015). Reliability studies of
vertical GaN devices based on bulk GaN substrates. Microelectron. Reliab.
55: 1654–1661.
62 Zhang, N.Q., Moran, B., DenBaars, S.P., Mishra, U.K., Wang, X.W., Ma, T.P.
(2001). Effects of surface traps on breakdown voltage and switching speed
of GaN power switching HEMTs. Technical Digest – International Electron
Devices Meeting, 2001 (IEDM ’01), Washington, DC (2–5 December 2001)
pp. 589–592.
63 Lee, C.D., Sagar, A., Feenstra, R.M. et al. (2001). Growth of GaN on
SiC(0001) by molecular beam epitaxy. Phys. Status Solidi (a) 188: 595–599.
64 Choi, S., Heller, E., Dorsey, D. et al. (2013). Analysis of the residual stress
distribution in AlGaN/GaN high electron mobility transistor. J. Appl. Phys.
113: 093510.
65 Böttcher, T., Einfeldt, S., Figge, S. et al. (2001). The role of high-temperature
island coalescence in the development of stresses in GaN films. Appl. Phys.
Lett. 78: 1976–1978.
66 Ishida, M., Ueda, T., Tanaka, T., and Ueda, D. (2013). GaN on Si tech-
nologies for power switching devices. IEEE Trans. Electron Devices 60:
3053–3059.
67 Chen, K.J., Häberlen, O., Lidow, A. et al. (2017). GaN-on-Si power technol-
ogy: devices and applications. IEEE Trans. Electron Devices 64: 779–795.
68 Leszczyński, M., Prystawko, P., Suski, T. et al. (1999). Lattice parameters
of GaN single crystals, homoepitaxial layers and heteroepitaxial layers on
sapphire. J. Alloys Compd. 286: 271–275.
69 Prystawko, P., Leszczyński, M., Beaumont, B. et al. (1998). Doping of
homoepitaxial GaN layers. Phys. Status Solidi B 210: 437–443.
70 Sarzynski, M., Leszczyńki, M., Krysko, M. et al. (2012). Influence of GaN
substrate off-cut on properties of InGaN and AlGaN layers. Cryst. Res.
Technol. 47: 321–328.
71 Suski, T., Staszczak, G., Grzanka, S. et al. (2010). Hole carrier concentration
and photoluminescence in magnesium doped InGaN and GaN grown on
sapphire and GaN misoriented substrates. J. Appl. Phys. 108: 023516.
72 Sarzyński, M., Suski, T., Staszczak, G. et al. (2012). Lateral control of indium
content and wavelength of III–nitride diode lasers by means of GaN sub-
strate patterning. Appl. Phys. Express 5: 021001.
73 Krysko, M., Domagala, J.Z., Czernecki, R., and Leszczyński, M. (2013). Tri-
clinic deformation of InGaN layers grown on vicinal surface of GaN (00.1)
substrates. J. Appl. Phys. 114: 113512.
74 Zauner, A., Aret, E., Enckevort, W. et al. (2002). Homo-epitaxial growth on

the N-face of GaN single crystals: the influence of the misorientation on the
surface morphology. J. Cryst. Growth 240: 14–21.
75 Smeeton, T., Kappers, M., Barnard, J. et al. (2003). Electron-beam-induced
strain within InGaN quantum wells: false indium “cluster” detection in the
transmission electron microscope. Appl. Phys. Lett. 83: 5419–5421.
76 Baloch, K.H., Johnston-Peck, A.C., Kisslinger, K. et al. (2013). Revisiting the
“In-clustering” question in InGaN through the use of aberration-corrected
electron microscopy below the knock-on threshold. Appl. Phys. Lett. 102:
191910.
77 Michałowski, P., Grzanka, E., Grzanka, S. et al. (2019). Indium concentra-
tion fluctuations in InGaN/GaN quantum wells. J. Anal. At. Spectrom. 34:
1718–1723.
78 Suihkonen, S., Svensk, O., Lang, T. et al. (2007). The effect of InGaN/GaN
MQw hydrogen treatment and threading dislocation optimization on GaN
LED efficiency. J. Cryst. Growth 298: 740–743.
79 Czernecki, R., Grzanka, E., Smalc-Koziorowska, J. et al. (2015). Effect of
hydrogen during growth of quantum barriers on the properties of InGaN
quantum wells. J. Cryst. Growth 414: 38–41.
80 Czernecki, R., Grzanka, E., Strak, P. et al. (2017). Influence of hydrogen
pre-growth flow on indium incorporation into InGaN layers. J. Cryst. Growth
464: 123–126.
81 Hestroffer, K., Wu, F., Li, H. et al. (2015). Relaxed c-plane InGaN layers for
the growth of strain-reduced InGaN quantum wells. Semicond. Sci. Technol.
30: 105015.
82 Sarzynski, M., Krysko, M., Targowski, G. et al. (2006). Elimination of AlGaN
epilayer cracking by spatially patterned AlN mask. Appl. Phys. Lett. 88:
121124.
83 Cicek, E., McClintock, R., Vashaei, Z. et al. (2013). Crack-free AlGaN for
solar-blind focal plane arrays through reduced area epitaxy. Appl. Phys. Lett.
102: 051102.
84 Arslan, E., Ozturk, M.K., Teke, A. et al. (2008). Buffer optimization for
crack-free GaN epitaxial layers grown on Si(111)substrate by MOCVD. J.
Phys. D. Appl. Phys. 41: 155317.
85 Schubert, E.F. (2006). Light-Emitting Diodes, 2e. New York: Cambridge
University Press.
86 Orsal, G., El Gmili, Y., Fressengeas, N. et al. (2014). Bandgap energy bow-
ing parameter of strained and relaxed InGaN layers. Opt. Mater. Express 4:
1030–1041.
87 Gu, G.-H., Jang, D.-H., Nam, K.-B., and Park, C.-G. (2013). Composition
fluctuation of In and well-width fluctuation in InGaN/GaN multiple quan-
tum wells in light-emitting diode devices. Microsc. Microanal. 19 (S5):
99–104.
88 Ochalski, T.J., Gil, B., Bigenwald, P. et al. (2001). Dual contribution to the
stokes shift in InGaN–GaN quantum wells. Phys. Status Solidi B 228 (1):
111–114.
References 37
89 Moses, P.G. and Van de Walle, C.G. (2010). Band bowing and band align-
ment in InGaN alloys. Appl. Phys. Lett. 96: 021908.
90 Yun, F., Reschikov, M.A., He, L. et al. (2002). Energy band bowing parameter
in Alx Ga1−x N alloys. J. Appl. Phys. 92 (8): 4837–4839.
91 Kazazis, S.A., Papadomanolaki, E., Androulidaki, M. et al. (2018). Optical
properties of InGaN thin films in the entire composition range. J. Appl. Phys.
123: 125101.
92 Paskov, P.P. and Monemar, B. (2018). Optical properties of III-nitride semi-
conductors. Handbook of GaN Semiconductor Materials and Devices. Bi,
W.W., Kuo, H-C., Ku, P-C., Shen, B. Edts. CRC Press/Taylor & Francis
Group, Boca Raton, FL, pag. 83.
93 Reschikov, M.A. and Morkoç, H. (2005). Luminescence properties of defects
in GaN. J. Appl. Phys. 97: 061301.
94 Wang, Y.J., Xu, S.J., Zhao, D.G. et al. (2006). Non-exponential photolumi-
nescence decay dynamics of localized carriers in disordered InGaN/GaN
quantum wells: the role of localization length. Opt. Express 14 (26):
13151–13157.
95 Perlin, P., Suski, T., Teisseyre, H. et al. (1995). Towards the identification of
the dominant donor in GaN. Phys. Rev. Lett. 75: 296–299.
96 Neudeck, P.G., Okojie, R.S., and Chen, L.-Y. (2002). High temperature
electronics – a role for wide bandgap semiconductors? Proc. IEEE 90:
1065–1076.
97 Baliga, B.J. (2005). Silicon Carbide Power Devices. Singapore: World Scientific
Publishing.
98 Arehart, R., Moran, B., Speck, J.S. et al. (2006). Effect of threading disloca-
tion density on Ni/n-GaN Schottky diode I–V characteristics. J. Appl. Phys.
100: 023709.
99 Jain, S.C., Willander, M., Narayan, J., and Van Overstraeten, R. (2000).
III-nitrides: growth, characterization and properties. Appl. Phys. Rev. 87:
965–1006.
100 Sze, M.S. and Ng, K.K. (2007). Physics of Semiconductor Devices, 3e. Hobo-
ken, NJ: Wiley.
101 Brunner, D., Angerer, H., Bustarret, E. et al. (1997). Optical constants of
epitaxial AlGaN films and their temperature dependence. J. Appl. Phys. 82:
5090–5096.
102 Asgari, A. and Kalafi, M. (2006). The control of
two-dimensional-electron-gas density and mobility in AlGaN/GaN het-
erostructures with Schottky gate. Mater. Sci. Eng. C 26: 898–901.
103 Chen, K.J. and Zhou, C. (2011). Enhancement-mode AlGaN/GaN HEMT
and MIS-HEMT technology. Phys. Status Solidi (a) 208: 434–438.
104 Su, M., Chen, C., and Rajan, S. (2013). Prospects for the application of
GaN power devices in hybrid electric vehicle drive systems. Semicond. Sci.
Technol. 28: 074012.
105 Scott, M.J., Fu, L., Zhang, X. et al. (2013). Merits of gallium nitride based
power conversion. Semicond. Sci. Technol. 28: 074013.
106 Roccaforte, F., Greco, G., Fiorenza, P., and Iucolano, F. (2019). An overview
of normally-off GaN-based high electron mobility transistors. Materials 12:
1599.
107 Carlin, J.-F. and Ilegems, M. (2003). High-quality AlInN for high index con-
trast Bragg mirrors lattice matched to GaN. Appl. Phys. Lett. 83: 668–670.
108 Medjdoub, F., Carlin, J.F., Gaquière, C. et al. (2008). Status of the emerging
InAlN/GaN power HEMT technology. The Open Electr. Electron. Eng. J. 2:
1–7.
109 Medjdoub, F., Zegaoui, M., Waldhoff, N. et al. (2011). Above 600 mS/mm
transconductance with 2.3 A/mm drain current density AlN/GaN
high-electron-mobility transistors grown on silicon. Appl. Phys. Express
4: 064106.
110 Harrouche, K., Kabouche, R., Okada, E., and Medjdoub, F. (2019). High
performance and highly robust AlN/GaN HEMTs for millimeter-wave opera-
tion. IEEE J. Electron Devices Soc. 7: 1145–1150.
111 Najda, S.P., Perlin, P., Suski, T. et al. (2017). AlGaInN diode-laser technology
for optical clocks and atom interferometry. J. Phys. Conf. Ser. 810: 012052.
112 Nyangaresi, P.O., Qin, Y., Chen, G. et al. (2018). Effects of single and com-
bined UV-LEDs on inactivation and subsequent reactivation of E. coli in
water disinfection. Water Res. 147: 331–341.
113 Yamashita, Y., Kuwabara, M., Torii, K., and Yoshida, H. (2013). A
340-nm-band ultraviolet laser diode composed of GaN well layers. Opt.
Express 21 (3): 3133–3137.
114 Han, D.P., Kamiyama, S., Takeuchi, T. et al. (2019). Understanding inefficien-
cies in blue and green LEDs. Compd. Semicond. 25 (3): 56–61.
115 Kimoto, T. and Cooper, J.A. (2014). Fundamentals of Silicon Carbide Tech-
nology: Growth, Characterization, Devices and Applications. Singapore:
Wiley.
116 Amano, H., Baines, Y., Borga, M. et al. (2018). The 2018 GaN power elec-
tronics roadmap. J. Phys. D. Appl. Phys. 51: 163001.
117 Roccaforte, F., Fiorenza, P., Greco, G. et al. (2018). Emerging trends in wide
band gap semiconductors (SiC and GaN) technology for power devices.
Microelectron. Eng. 187–188: 66–77.
118 Chung, J.W., Roberts, J.C., Piner, E.L., and Palacios, T. (2008). Effect of gate
leakage in the subthreshold characteristics of AlGaN/GaN HEMTs. IEEE
Electron Device Lett. 29: 1196–1198.
119 Greco, G., Iucolano, F., and Roccaforte, F. (2016). Ohmic contacts to gallium
nitride materials. Appl. Surf. Sci. 383: 324–345.
120 Mohammad, S.N. (2004). Contact mechanisms and design principles for
alloyed Ohmic contacts to n-GaN. J. Appl. Phys. 95: 7940–7953.
121 Roccaforte, F., Frazzetto, A., Greco, G. et al. (2012). Critical issues for
interfaces to p-type SiC and GaN in power devices. Appl. Surf. Sci. 258:
8324–8333.
122 Greco, G., Prystawko, P., Leszczyński, M. et al. (2011). Electro-structural
evolution and Schottky barrier height in annealed Au/Ni contacts onto
p-GaN. J. Appl. Phys. 110: 123703.
References 39
123 Roccaforte, F., Fiorenza, P., Greco, G. et al. (2014). Recent advances on
dielectrics technology for SiC and GaN power devices. Appl. Surf. Sci. 301:
9–18.
124 Daumiller, I., Theron, D., Gaquière, C. et al. (2001). Current instabilities in
GaN-based devices. IEEE Electron Device Lett. 22: 62–64.
125 Binari, S.C., Ikossi, K., Roussos, J.A. et al. (2001). Trapping effects and
microwave power performance in AlGaN/GaN HEMTs. IEEE Trans. Electron
Devices 48: 465–471.
126 Vetury, R., Zhang, N.Q., Keller, S., and Mishra, U.K. (2001). The impact of
surface states on the DC and RF characteristics of AlGaN/GaN HFETs. IEEE
Trans. Electron Devices 48: 560–566.
127 Meneghesso, G., Verzellesi, G., Pierobon, R. et al. (2004). Surface-related
drain current dispersion effects in AlGaN-GaN HEMTs. IEEE Trans. Elec-
tron Devices 51: 1554–1561.
128 Hashizume, T., Nishiguchi, K., Kanekia, S. et al. (2018). State of the art on
gate insulation and surface passivation for GaN-based power HEMTs. Mater.
Sci. Semicond. Process. 78: 85–95.
129 Meneghesso, G., Meneghini, M., De Santi, C. et al. (2018). Positive and neg-
ative threshold voltage instabilities in GaN-based transistors. Microelectron.
Reliab. 80: 257–265.
41
GaN-Based Materials: Substrates, Metalorganic

Vapor-Phase Epitaxy, and Quantum Well Properties
Ferdinand Scholz 1 , Michal Bockowski 2,3 , and Ewa Grzanka 2
1 Ulm University, Institute of Functional Nanosystems, Albert-Einstein-Allee 45, 89081 Ulm, Germany
2
Institute of High Pressure Physics, Polish Academy of Sciences (Unipress-PAS), Sokolowska 29/37, 01-142
Warsaw, Poland
3
Nagoya University, Center for Integrated Research of Future Electronics, Institute of Materials and Systems
for Sustainability, C3-1 Furo-cho, Chikusa-ku, Nagoya 464-8603, Japan
2.1 Introduction
In the fifties of past century, compound semiconductors came into scientific focus
for applications in electronic and optoelectronic devices, the latter mainly driven
by the direct band structure of most of them. Quite soon, one promising feature of
these materials was identified: the possibility to design heterostructures, through
which many specific device-related demands could be fulfilled, e.g. carrier con-
finement and photon confinement. Hence, respective epitaxial methods needed
to be developed. However, in these early years, only simple epitaxial methods
were available, e.g. liquid-phase epitaxy (LPE) and (hydride or chloride-based)
vapor-phase epitaxy (VPE). By these methods, the formation of heterostructures
was very difficult, if not even impossible for many material combinations. In LPE,
the sequential deposition of different materials depends on the thermodynamic
phase diagrams of these materials blocking the epitaxy of some combinations,
as the later deposited top layer would lead to a dissolution of the already grown
layer for some required combinations. In VPE, particularly the metal needed in
III–V compounds was transported to the substrate only after in situ generation
as a metal chloride in the reactor (see Section 2.2.1).
This need was the strong driving force to develop epitaxial methods where
all elements needed to form the semiconductor layers can be provided and
controlled individually and independently, leading to the development of
metalorganic vapor-phase epitaxy (MOVPE), which makes use of specific
compounds for the transport of the required elements, and molecular beam
epitaxy (MBE), where the elements are just evaporated into an ultrahigh vacuum
chamber. Both methods have turned out to be extremely successful for pushing
forward the development of very sophisticated devices such as light-emitting
diodes (LEDs), laser diodes (LDs), and transistors, where the properties of
heterostructures enable excellent device performance. When the nitrides came

42 2 GaN-Based Materials
into scientific focus, it was hence quite obvious that this went hand in hand with
mastering some basic epitaxial problems related to this material class.
One of these problems is the lack of adequate procedures to grow GaN bulk
crystals, which would be needed for the fabrication of substrates for the epitax-
ial growth of device heterostructures. Owing to the thermochemical properties
of GaN – very high melting temperature and extremely large equilibrium vapor
pressure at that temperature, classical methods such as the Czochralski or ver-
tical gradient freeze, which are well established for other semiconductors such
as Si, GaAs, and InP, cannot be applied. That is why this chapter first describes
some methods to circumvent this problem (Section 2.2). In particular, the classi-
cal hydride-based vapor-phase epitaxy (hydride vapor phase epitaxy, HVPE) has
seen a tremendous renaissance after the detection that it can provide huge growth
rates of several 10–100 μm/h [1]. Hence, it can solve this problem by growing very
thick epitaxial GaN layers in reasonable time. Moreover, in this section, some
other methods to produce GaN wafers are discussed, particularly the method of
ammonothermal crystal growth.
Then, we focus on the method of MOVPE, which turned out to have the
highest importance for nitride heterostructures (Section 2.3). We present an
overview over basics of this method including the growth on foreign substrates.
Such growth is blamed with the development of a huge defect density in the
epitaxial layers; hence, specific methods have been developed to minimize this
problem.
A particular importance for light-emitting devices is dedicated to GaInN
quantum wells (QWs), which, however, create extra problems. That is why
Section 2.4 deals exclusively with these heterostructures. Besides addressing
issues such as growth conditions and defects affecting the microstructure
of InGaN/GaN quantum wells, we will focus particularly on problems of
compositional inhomogeneities and decomposition of the quantum wells.
2.2 Bulk GaN Growth

In this section, the current status of bulk GaN crystal growth technology will
be described. Today, three methods are applied for GaN crystallization. Two
of them, sodium flux and ammonothermal, belong to the group of solution
growth. The third one, HVPE (sometimes also called “halide” vapor-phase
epitaxy), involves crystallization from gas phase. Because of a high growth rate,
HVPE is often applied for obtaining bulk GaN. Because HVPE-GaN is mainly
deposited on foreign substrates (sapphire and gallium arsenide), it enables the
growth of layers up to 6 in. in diameter. However, they are of low structural
quality. The crystallographic planes are bent, and wafers obtained from such
materials can be plastically deformed. Therefore, it is impossible to misorient the
substrates uniformly across the entire surface, which is a critical requirement
for devices. This problem can be overcome by using native crystals of very high
crystallographic quality as seeds for HVPE growth. This enabled to deposit GaN
with a low threading dislocation density (TDD ≤ 105 cm−2 ) and absolutely flat
crystallographic planes. It should, however, be remarked that all of the described

methods allow to obtain only a few millimeter thick GaN. Until now, no one
has demonstrated high-quality bulk GaN crystals yielding several tens of wafers
per boule as well as a convenient technology for growing such thick layers. The
reasons for this will be discussed in this section in detail.
2.2.1 Hydride Vapor-Phase Epitaxy (HVPE)

As mentioned, HVPE is a method of crystallization from gas phase. A scheme
of a horizontal HVPE reactor, used at the Institute of High Pressure Physics of
the Polish Academy of Sciences (IHPP PAS), is presented in Figure 2.1. In the
low-temperature zone of the reactor (800–900 ∘ C), hydrochloride (HCl) reacts
with gallium to form gallium chloride (GaCl). GaCl is transported by the carrier
gas (mainly N2 , H2 , or their mixtures) to the high-temperature zone. There, at
1000–1100 ∘ C, GaN is synthesized (crystallized) due to a reaction of GaCl with
ammonia (NH3 ). As is clearly seen in Figure 2.1, NH3 is flown separately and
directly to the crystal growth zone. Supersaturation, the driving force of crystal-
lization, is created by the difference between the equilibrium partial pressure of
GaCl and real input partial pressure of GaCl and ammonia flow at a given tem-
perature of the growth process. Gallium nitride can be crystallized on a native or
foreign substrate. As foreign seeds, MOVPE-GaN/sapphire templates or GaAs
substrates with a low-temperature buffer layer of GaN are used [2, 3]. The main
crystallographic growth direction in HVPE-GaN technology is the [0001] direc-
tion (c-direction).
HVPE-GaN crystallization in the c-direction has two significant advantages
over other methods: high growth rate and high purity of the material [4–6]. The
average rate can be higher than 100 μm/h. The new-crystallized GaN can be of
high purity with unintentional dopants deriving from the reactant gases or solid
HCI + CG
Ga
GaCl + CG
NH3 + CG
Seed on susceptor
Figure 2.1 Scheme of horizontal HVPE reactor used at IHPP PAS; GaCl is transferred to the
reaction zone through shower head-type quartz nozzles; NH3 is supplied by a quartz nozzle
placed on the same level as the susceptor; CG is the carrier gas.
elements of the reactor (e.g. quartz). Traces of oxygen and silicon, concentrations
lower than 1017 cm−3 , are often found in the HVPE-GaN material. Doping pro-
cesses, with silicon or germanium for obtaining highly conductive crystals or with
iron, carbon, or manganese for the semi-insulating ones, are well developed in
HVPE and described in detail in the literature, e.g. Refs. [7–11].
HVPE is the most popular method used for producing GaN substrates. The
main suppliers of HVPE GaN wafers include SCIOCS by Sumitomo Chemical
(www.sciocs.com/english/products/GaN_substrate.html), Mitsubishi Chemical
Corp. (https://www.m-chemical.co.jp/en/products/departments/mcc/nes/
product/12010299004.html), Sumitomo Electric Industry (global-sei.com/
sc/products_e/gan/), and Furukawa Co. (https://www.furukawakk.co.jp/e/
business/others) from Japan, as well as Nanowin from China (http://shopen
.nanowin.com.cn/) and Lumilog by Saint Gobain from France (https://www
.ceramicmaterials.saint-gobain.com/lumilog). GaN substrates are produced
from free-standing (FS) HVPE-GaN crystals, grown before on foreign seeds. As
mentioned, HVPE-GaN deposition on a foreign foundation allows obtaining
large-diameter GaN crystals. Unfortunately, their crystallographic planes are
bent (see Figure 2.2a). This results from huge differences between the lattice
constants and thermal expansion coefficients of the foreign substrate and the
nitride layer. When 2 in. GaN is grown on sapphire (a typical morphology is
presented in Figure 2.2b), the value of bowing radius of crystallographic planes is
often of the order of 5 m. The crystals can also be plastically deformed and have
dislocation bundles [12]. This is the reason why FS HVPE-GaN is not applied
as a seed for further multiplication or bulk GaN growth. There is no possibility
to improve the crystallographic quality. The development of a small radius of
curvature of crystallographic planes is also the reason for the lack of mass pro-
duction of larger than 2 in. FS HVPE-GaN substrates. Bending is so significant
that it hinders processing in the production of electronic or optoelectronic
θ1
θ2
c-Plane
100 μm
(a) (b)
Figure 2.2 (a) Scheme of bowing of crystallographic planes in the HVPE-GaN substrate
prepared from crystal grown on foreign seed; a nonuniform misorientation angle 𝜃 is
presented. (b) Typical morphology of HVPE-GaN grown on foreign seed (MOVPE-GaN/sapphire
template); many hexagonal hillocks; etch pit density (EPD), correlated with TDD, is of the order
of 106 cm−2 .
18
16 C
14
Radius of curvature (m)
12
2D
10
6
3D
4
100 μm
2
0
0 50 100 150 200 250
(a) 3D GaN thickness (μm) (b)
Figure 2.3 (a) Radius of curvature (bending of crystallographic planes) as a function of 3D

GaN layer thickness for free-standing HVPE-GaN (see description in the text). Source: Data
from IHPP PAS are presented. (b) Optical microscopy image of cross section of GaN grown on
MOVPE-GaN/sapphire; the sample was photoetched; in this technique, the etching rate
depends on the free carrier concentration in the material and crystal sectors with different
electrical properties are visualized; 3D and 2D growth modes are shown.
devices. A uniform misorientation of GaN substrates is important for epitaxy

and homogeneous doping of AlInGaN structures. The structural quality of
the epilayer and incorporation of dopants depend on the miscut degree of the
substrate [13–15].
The trick used to avoid the bending of crystallographic planes is to start
the crystallization process in a three-dimensional (3D) growth mode and, by
changing the supersaturation, switch it in time into a two-dimensional (2D)
one. This solution was demonstrated by Geng et al. [16]. The intrinsic strain
is then reduced by the introduction of 3D growth in the initial crystallization
stage. Bending of the wafers is controlled by the thickness of the 3D GaN layer
(Figure 2.3a). Figure 2.3b presents an image of the cross section (m-plane) of
HVPE-GaN deposited on a sapphire template. The 3D growth mode is well
visible. With increasing GaN thickness, the mode changes to 2D. HVPE-GaN is
separated from an MOVPE-GaN/sapphire template with a lift-off process. The
material grown in the 3D growth mode is removed from the crystal by lapping
and polishing during GaN substrate fabrication.
2.2.2 Sodium Flux Growth Method

Figure 2.4 presents a scheme of the sodium flux growth method. Under pressure
of nitrogen ((5–50) × 105 Pa) and at a constant temperature lower than 1000 ∘ C,
gallium (27%) and sodium (73%) are mixed in the crucible. Sodium increases
the solubility of atomic nitrogen in the flux [17]. Similar to the high nitrogen
pressure solution (HNPS) growth method described in detail elsewhere [18],
nitrogen molecules dissociate on the surface of the flux and dissolve into it.
The high pressure of nitrogen over the solution ensures the supersaturation in
the flux. The value of nitrogen pressure is higher than the equilibrium one for
pN2
Ga–Na solution
Figure 2.4 Scheme of the sodium flux growth method; crystallization is carried out in the
c-direction with a rate lower than 50 μm/h. Source: Courtesy of Profs. Y. Mori and M. Imanishi.
considered phases of the flux (Na/Ga–GaN–N2 ). The mass transport is governed

by convection caused by mechanical stirring of the flux [17].
The crystallization process proceeds on a foreign seed (mainly MOVPE-GaN/
sapphire template) placed at the bottom of a crucible. During the past 20 years,
there were a few approaches to the sodium flux method. Today, the point seed
technique is mainly used [19]. An MOVPE-GaN/sapphire template is patterned
into GaN point seeds (up to 1 mm in diameter). At the beginning of a crystalliza-
tion process, all seeds are overgrown in the c-direction. A pyramidal crystalliza-
tion takes place. This corresponds to a 3D mode in the HVPE technology (see
Figure 2.3b). In order to switch from 3D to 2D, the seed is pulled out from the
solution when the GaN is pyramidal. Then, the Na—Ga solution remains only
between the pyramids and the lateral growth is enhanced.
Because there is a small amount of the solution between the pyramids, the
template should be dipped again into the solution. The pulling up and dipping
procedures are repeated until the GaN surface becomes flat. Then, the crystal-
lization run is continued in the 2D growth mode. The scheme of the described
method is presented in Figure 2.5a. Figure 2.5b presents a 2 in. transparent GaN
crystal with bowing radius larger than 30 m, grown by sodium flux and separated
from the seed above the 3D growth mode.
Sodium flux GaN crystals are of high structural quality and purity. In the
undoped material, the free carrier concentration is not higher than 1016 cm−3 .
The main donor is oxygen. Germanium is intentionally incorporated in order to
obtain highly conductive n-type crystals. There are no communications about
semi-insulating material in the literature. It was demonstrated that Na flux GaN
can be applied as a seed for HVPE growth. First results presented by SCIOCS and
Osaka University were very promising. HVPE-GaN of high structural quality,
crystallographically flat, and with etch pit density (EPD) lower than 105 cm−2
was obtained [20].
2.2.3 Ammonothermal Growth

Ammonothermal crystallization of GaN is analogous to hydrothermal growth
of quartz. In the case of ammonothermal method, supercritical ammonia is used
instead of water. The scheme of the process is as follows: GaN, used as a feedstock,
is dissolved in supercritical ammonia in one zone of a high-pressure autoclave.
Ga–Na melt
GaN crystal Sapphire
(a)
(b)
Figure 2.5 (a) Scheme of sodium flux GaN growth method with pulling up and dipping
procedures for changing the growth mode from 3D to 2D analogous to the HVPE technology.
(b) Two inch transparent GaN crystal with bowing radius higher than 30 m and EPD lower than
105 cm−2 ; the crystal was separated from the seed above the initial 3D growth mode; grid line
1 mm. Source: Courtesy of Profs. Y. Mori and M. Imanishi.
The dissolved feedstock is transported to the second zone, where the solution is
supersaturated and crystallization of GaN on native seeds takes place. An appro-
priate temperature gradient, applied between the two zones, enables the convec-
tion mass transport. Mineralizers are added to ammonia in order to accelerate its
dissociation and enhance the solubility of GaN. The growth can proceed under
basic or acidic environment. Its type is obviously determined by the choice of
the mineralizers. In ammonoacidic growth, halide compounds act as mineral-
izers, while in ammonobasic approach, alkali metals or their amides are used
[18]. In the latter, a negative temperature coefficient of solubility is observed [21].
The chemical transport of GaN is directed from the low-temperature solubility
zone (with feedstock) to the high-temperature crystallization part (with seeds).
This is the consequence of the retrograde solubility. Schemes of basic and acidic
ammonothermal growth methods are presented in Figures 2.6a,b, respectively.
There are several companies and research institutes currently working on the
ammonothermal method, such as SixPoint Materials Inc. (USA) [22], University
of California Santa Barbara (USA) [23], Universities of Erlangen and Stuttgart
(Germany) [24], Mitsubishi Chemical Corp. (Japan) [25], Tohoku University
(Japan) [26], and most probably Kyocera (formerly Soraa, Inc., USA/Japan) [27].
Significant contributions to the field have been and are still provided by IHPP
PAS (formerly Ammono S.A., Poland) [28]. Its crystallization method is based
on the basic ammonothermal approach and proceeds in the temperature range
of 500–600 ∘ C at a pressure between 0.3 and 0.4 GPa. It consists of two parts
schematically shown in Figure 2.7:
Part 1: Enlargement of seeds: this is performed by taking advantage of lateral crys-

tallization in nonpolar and/or semipolar directions (red area in Figure 2.7a).
Dissolving zone
NH3 with
Growth zone
mineralizer
GaN
feedstock Seed
crystals
Baffle
Baffle
Seed
Dissolving zone
Growth zone
crystals GaN
feedstock
NH3 with
mineralizer
(a) Temperature (b) Temperature
Figure 2.6 Scheme of ammonothermal method: (a) basic and (b) acidic.
Part 1 Part 2
–c –c
a a
(a) (b)
Figure 2.7 Scheme of two main parts of the ammonothermal crystallization: (a) Part 1 – lateral
growth and (b) Part 2 – seed multiplication; arrows show directions of growth of specific
crystal regions; crystal grown in Part 1 is used as a seed in Part 2.
Part 2: Seed multiplication: this part is concentrated on growth mainly in vertical

direction (along the c-axis, green part in Figure 2.7b). The deposited GaN has
very flat crystallographic planes and an EPD of the order of 5 × 104 cm−2 .
As can be seen in Figure 2.7, the growth is conducted in the a-[1120], m-[1010],
and −c-[0001] directions. Crystallization in the +c-[0001] direction is treated
as parasitic. In order to avoid it, special holders are used for the seeds. As a
result, the (0001) surface is mainly exposed, while the opposite (0001) face is
masked. One of the most important factors limiting the ammonothermal GaN
(Am-GaN) crystallization in the −c-direction (Part 2 in Figure 2.7) is associ-
ated with the anisotropy of the growth and deposition occurring in the lateral
directions at the edges of the crystal. It was shown that kinds and concentra-
tions of impurities incorporated into the nonpolar facets and the (0001) plane
are vastly different [28]. This causes stress and finally leads to plastic deforma-
tion of the material. Some examples of as-grown Am-GaN crystals grown in the
−c-direction are presented in Figure 2.8. Three types of substrates (n-type with
free carrier concentration of 1 × 1019 and 1 × 1018 cm−3 as well as semi-insulating
with resistivity higher than 1 × 108 Ω cm at room temperature) are prepared at
Figure 2.8 Ammonothermally grown GaN crystals; the difference in color results from
different free carrier concentrations; left side: 1 × 1019 cm−3 ; right side: 1 × 1018 cm−3 ; and grid
line: 1 mm.
IHPP PAS from ammonothermally crystallized GaN (https://www.unipress.waw

.pl/growth/index.php/ammono-gan-wafers-sales).
As can be seen in Figure 2.8, the ammonothermal GaN crystals have a round
shape. Changing it from hexagonal to round allowed to deposit thicker layers. If
the growth process takes place only on one crystallographic plane, Am-GaN can
be grown longer without losing its high structural quality. As already mentioned,
appearing of the side facets causes the formation of stress on the edges of the
crystal because of different incorporation of impurities on different facets. The
same phenomenon can be observed in the HVPE technology. Fujito et al. [29]
demonstrated the formation of a dodecagon shape of an HVPE-GaN crystal if
the growth started on a round substrate. The inclined facets (1011) and (1022)
on m-plane sides and (1122) facets on a-plane sides were formed. The forma-
tion of this shape through the collapse of the growth facet reduced the size of
the c-plane. Sochacki et al. [30] reported lateral overgrowth of sidewalls during
HVPE crystallization. It was demonstrated that the oxygen concentration on the
laterally overgrown facets and on the c-plane is different by more than 2 orders
of magnitude. The high content of dopants can lead to an increase of the lattice
constants. It was shown that both a and c lattice parameters increased in later-
ally grown GaN [31]. The strain in HVPE-GaN deposited in the c-direction on a
native seed was studied by Raman spectroscopy [32]. A stress of 200 MPa in the
laterally overgrown area was determined. It was concluded that the lateral growth
of GaN leads to the formation of large stress in the crystal, close to its edges.
This stress is much more significant than that generated by the lattice mismatch
between the seed and the deposited layer (close to 2 MPa). For many years, it was
thought that if the foreign seed is replaced by a high-quality native one, it would
be possible to grow truly bulk HVPE-GaN. However, it turned out to be extremely
difficult. HVPE-GaN crystallization with high growth rate, up to 350 μm/h, was
reported on 1 in. Am-GaN substrates [33]. Up to 2 mm thick HVPE-GaN crys-
tals were grown. Few or just one hillock was observed on the entire surface (see
Figure 2.9a). The almost perfect structural quality of the ammonothermal seed
was reflected by the HVPE layer. Obtaining thicker boules was, however, lim-
ited by the anisotropy of the growth and crystallization occurring in the lateral
directions (see Figure 2.9b).
It seems that in the case of HVPE, only one growth facet (the c-facet) should be
stabilized and grown out for an arbitrary period of time. According to Prof. Sitar
[34], this condition may be achieved by controlling the thermal field around the
crystal. It has to reach its final shape by adapting to the thermal field rather than
taking its equilibrium hexagonal habit. This was demonstrated for an aluminum
nitride (AlN) growth process by physical vapor transport (PVT) and presented on
the HexaTech web page (http://www.hexatechinc.com/company.html). An AlN
crystal takes the round shape from the seed and expands as forced by the designed
thermal field. Once the crystals reach a constant thermal field, they resume their
equilibrium shape as dictated by the surface energetics, forming various facets.
This clearly shows that the equilibrium shape can be overpowered by a proper
thermal field design. In this case, the crystal will follow the thermal field and grow
in a direction perpendicular to the isotherms. Obviously, there is a big difference
in the formation of supersaturation in PVT and HVPE methods. The supersatu-
ration is the difference of thermodynamic potentials at the interface between a
2.3 MOVPE Growth 51
(a) (b)
Figure 2.9 (a) Typical morphology of HVPE-GaN grown on Am-GaN seed; a few hillocks on the
entire growing surface of the crystal; grid line 1 mm. (b) Fluorescence image of m-plane cross
section with edge of HVPE-GaN crystal; the laterally overgrown part is well visible; this part
creates large stress in the growing crystal.
crystal and its environment. In the case of PVT, it is almost unambiguous with the
temperature distribution on the growing surface. In the case of HVPE, reactions
of all vapor species should be considered.
2.3 MOVPE Growth

As mentioned in the introduction of this chapter, in the sixties of the past century,
some epitaxial method was lacking, which would give full freedom of designing
layer sequences in compound semiconductor heterostructures. Although VPE
was already strongly considered, it could not solve this problem as more or less
all required elements of the third and fifth column of the periodic system are per
se no gases but in many cases unvolatile solids at or around room temperature
prohibiting their individual and independent control as a gas stream. One possi-
ble solution was to heat all elements to evaporation temperatures, which led to
the development of the method of MBE (see Chapter 8). However, this method
requires a very complicated and sensitive machinery. That is why scientists went
on considering conventional vapor-phase deposition methods. For the group V
elements, the problem was already solved by introducing their gaseous hydrides
(PH3 , AsH3 , and NH3 ) into the epitaxial process. However, for the required group
III metals (Al, Ga, In, etc.), no stable gaseous compounds are available.
It needed Harold Manasevit’s ingenious idea to take metalorganic1 compounds
[35] which at reasonable ambient temperatures are still not gaseous, but liquid or
solid, providing however a fairly high vapor pressure of typically several hundred
pascals (Table 2.1). Typical examples are tri-methyl gallium (TMGa, (CH3 )3 Ga)
1 Indeed, according to chemical nomenclature, the correct term is “organo-metallic” compound.

Table 2.1 Physical properties of some metalorganics mainly used in nitride MOVPE.
Metalorganic Short Chemical Melting Constant Constant Vapor pressure

compound name formula point (∘ C) A (K) B (Pa) at temperature References
Trimethylgallium TMGa Ga(CH3 )3 −16 3921 1.56 × 1010 9030 Pa at 0 ∘ C [36]

Triethylgallium TEGa Ga(C2 H5 )3 −82 5826 1.95 × 1011 368 Pa at 17 ∘ C [37]
Trimethylindium TMIn In(CH3 )3 88 7474 1.64 × 1013 105 Pa at 17 ∘ C [38]
Trimethylaluminum TMAl Al(CH3 )3 15.4 4914 2.23 × 1010 976 Pa at 17 ∘ C [36]
Bis-cyclopentadienyl- Cp2 Mg (C5 H5 )2 Mg 176 8110 3.77 × 1012 2.7 Pa at 17 ∘ C [39]
magnesium
The vapor pressure can be obtained from the constants A and B by the equation p = B⋅e−A/T . Notice the decreasing vapor pressure with increasing molecular weight.
The vapor pressure of TMAl is exceptionally small because this compound vaporizes as a dimeric molecule [Al(CH3 )3 ]2 [40].
2.3 MOVPE Growth 53
Figure 2.10 Schematic structure of a modern MOVPE system. Just one channel for
metalorganics and hydrides are shown for simplicity, whereas a real system may comprise
5–10 metalorganic channels and also several hydride channels. Doping channels are basically
similar but may contain a so-called double-dilution configuration (see Section 2.3.5).
and its In- and Al-containing counterparts. After fixing their vapor pressure by
stabilizing their temperature, this vapor can be transported to the epitaxial reac-
tor by an inert carrier gas – in most cases, hydrogen, but in some cases also
nitrogen or other gases, see Section 2.3.6 – bubbling through the liquid (see the
lower left part in Figure 2.10). Precursors such as TMIn or Cp2 Mg, which are
solid at reasonable bubbler temperatures, need special attention: they should be
kept in the bubbler in powder form; moreover, specific bubbler designs have been
developed for such precursors (see, e.g. Ref. [41]).
Now indeed, this novel method of MOVPE2 satisfied the above-mentioned
demand: each element needed to form complex III–V compounds and het-
erostructures could be supplied to the growth chamber as a single, independently
controllable gas stream (Figure 2.10). The interested reader may find more details
about MOVPE basics in Refs. [40, 42–44]. Here, we concentrate on the spe-
cific peculiarities of MOVPE when applied to GaN and related compound
semiconductors.
2 Often, particularly outside of Europe, synonymously called “metalorganic chemical vapor

deposition”, MOCVD, which does not recognize the epitaxial growth character of this method,
when applied to semiconductors.
2.3.1 Basics About Nitride MOVPE

GaN with its large bandgap has since long been a promising candidate for
achieving short-wavelength light emitters. That is why already Manasevit
applied his newly developed method “MOVPE” to grow such material [45],
although not very successful. However, these early experiments clearly demon-
strated that GaN can be grown by MOVPE in a similar way as GaAs and other
arsenides and phosphides. Basic differences concerning the growth process
(besides some other issues discussed in Sections 2.3.2–2.3.6) are mainly caused
by the strong binding forces of the nitrides and the chemical stability of the
simplest nitrogen precursor NH3 (ammonia),3 both directly a consequence of
the chemical properties of nitrogen, which, as an extremely small atom, forms
very stable chemical compounds.
On the one hand, only at temperatures above 900 ∘ C, significant fractions of
NH3 are thermally cracked [46]. However, it dissociates predominantly into inac-
tive N2 and H2 . At typical growth temperatures, only a small fraction of NH3 may
reach the substrate surface where it can contribute to the GaN growth by dis-
sociative adsorption [47], obviously catalytically enhanced by the GaN surface
[46]. Anyway, GaN MOVPE requires typically a very large V–III ratio (ratio of the
molar flows of group V and group III precursors) beyond 1000, with even larger
values needed for nitride compounds which require lower growth temperatures
(see Section 2.3.6).
On the other hand, the large thermodynamic stability of GaN and its related
compounds also requires very high growth temperatures, as the atoms arriving
on the growing surface need a reasonable mobility to migrate on the surface to a
well-suited site for incorporation, forming a structure with a minimum of crys-
talline defects.
Both requirements lead to a typical growth temperature for GaN of at least
about 1000 ∘ C, 200–300 ∘ C higher than required for arsenides and phosphides.
Unfortunately, such high temperatures enforce another typical MOVPE prob-
lem: the occurrence of parasitic gas-phase prereactions between group III and
group V precursors (see, e.g. Ref. [48]), leading to substantial precursor losses and
negative impact on the epilayer quality. This is one reason why GaN is typically
grown at low pressure (in most cases around 100 hPa). Then, such parasitic reac-
tions are effectively suppressed. However, TMAl – when growing Al-containing
layers – is much more susceptible for prereactions with ammonia. Hence, even
lower pressures are required for AlGaN with larger Al content, particularly when
grown at higher temperatures (see also Section 2.3.6).
As introduced in Chapter 1, another major difference of nitrides as compared
to arsenides and phosphides is the substrate problem: for any epitaxial process, it
is recommended to use substrates that have about the same crystalline properties
(crystalline symmetry and lattice constant) as the growing layer in order to get no
defects generated at the substrate–layer interface. For arsenides and phosphides,
the typical substrate materials are therefore GaAs and InP. These materials can
be fairly easily grown as bulk substrates of reasonable size (e.g. with a diameter
3 Nitrogen gas N2 is chemically much too stable to be used as a nitrogen precursor in MOVPE.
2.3 MOVPE Growth 55
of 6–8 in. and a length of several 10 cm), from which wafers can be prepared by
slicing and polishing. The epitaxial layers should then be chosen to have the same
lattice constant as the substrate. Hence, for GaN, the natural substrate choice is
GaN itself. However, owing to its extremely high melting point above 2500 ∘ C
and, even worse, its very high equilibrium vapor pressure at this temperature of
at least 3 GPa, it is very difficult, if not impossible, to grow GaN bulk substrates
of reasonable size following conventional crystal growth techniques. Various
methods are currently under investigation to solve this problem as discussed
in more detail in Section 2.2. However, such GaN wafers are still extremely
expensive, making them a bad choice for most simple device applications such
as LEDs or transistors. This is why much cheaper so-called “foreign” substrates
are used for these applications. This and respective problems and challenges are
discussed in Section 2.3.2.
The excellent results discussed below, which have been achieved over the recent
years on foreign substrates, became only possible because of a steady evolution of
the MOVPE equipment and the respective processes. A significant contribution
must be credited to the development of in situ characterization tools, which are
now standard tools at all modern MOVPE reactors. Different to MBE where in
situ analysis mostly relies on material beams interacting with the growing surface
like “reflection of high energy electron diffraction” (RHEED) – easily possible
because of the ultrahigh vacuum conditions – optical methods turned out to be
most successful in MOVPE (check Ref. [49] for a nice overview). For GaN growth,
mainly three parameters can be measured with good precision (see Figure 2.11):
• The flatness of the growing layer by analyzing the intensity of the reflected
probe light,
• the growth rate by analyzing the interference pattern of the signals reflected at
the surface and the layer-substrate interface;
• the wafer surface temperature by analyzing the infrared emission spectrum of
the growing surface,
• and the strain of the epitaxial structure by measuring the reflection angle of
two or three probe laser beams and hence the wafer bow.
These measurements have significantly contributed to the optimization of GaN
heterostructures on foreign substrates.
Figure 2.11 Setup for measuring layer data in pyro
situ during an MOVPE process (schematically):
flatness and growth rate determined from the
Las
R
reflected signal (“R”); wafer or susceptor
EpiTT
er
era
surface temperature by pyrometrically

cam D
CC
analyzing their thermal emission (“pyro”), and

curvature by analyzing the deviation of
several reflected laser beams (laser and
charge-coupled device [CCD] camera). Source: Reactor lid
Maaßdorf et al. 2013 [50]. Reproduced with window
the permission of AIP Publishing.
Wafer
Satellite
®
Figure 2.12 AIXTRON PLANETARY REACTOR (schematically). The wafers posted on the
satellite disks rotate individually induced by “gas foil rotation” [52], while the main susceptor
also rotates (driven mechanically). Source: Courtesy of AIXTRON.
For the MOVPE growth of group III nitrides, basically, the same MOVPE
equipment can be used as for other III–V compounds. A major difference is
the need for very high growth temperatures of 1000–1100 ∘ C for GaN or even
higher for AlN (see Section 2.3.6). This is achieved in many cases by radio
frequency induction heating, while some reactors also use resistive heating
of the susceptor. Owing to the needs of industry of large-area deposition,
multiwafer reactors dominate the market nowadays. From the many reactor
variations found over the world, let us briefly discuss just three concepts here
(see also Ref. [51]):
• Planetary reactor (Figure 2.12, see, e.g. Refs. [52–54]): this is conceptually a
horizontal reactor, as the reactive gases flow horizontally over the wafers. Sev-
eral wafers are arranged on a circular susceptor around a central gas inlet.
The gases flow radially over the wafers. To compensate the reactant gas con-
sumption and improve the growth rate and composition homogeneity, each
wafer, being placed on a so-called satellite disk, is rotated during growth; more-
over, the whole main susceptor rotates. The number of wafers varies between
5 and several 10 for wafer sizes between 2 and 8 in. On such large areas, excel-
lent uniformity of all layer properties has been achieved by careful process
optimization supported by the already mentioned optical in situ character-
ization tools enabling to some extent the measurements on each individual
wafer.
• Close-coupled showerhead reactor (Figure 2.13, see, e.g. Refs. [54, 56, 57]): in
this case, the main gas stream flows vertically from a showerhead-like inlet
on top of the reaction chamber to the wafers below. A single wafer of large
diameter (8 in. or more) or many wafers can be placed on a rotating suscep-
tor. By adjusting the gas flows in different radial areas of the showerhead, the
layer uniformity can be nicely controlled, again further supported by in situ
characterization tools [54, 57]. Similarly, the temperature uniformity on the
wafer(s) can be optimized by adjusting several radial resistive heater zones
below the susceptor. Moreover, the distance between the showerhead and the
2.3 MOVPE Growth 57
Showerhead
Wafer Resistive heater

on rotating susceptor
(a)
Optical probes
Showerhead
water cooling
Top plenum
chamber
Double O-ring seal
Gas inlet bottom Susceptor

plenum chamber Water cooled chamber
Three zone heater Quartz liner
Themocouple
Exhaust
(b)
Figure 2.13 Principle (a) and sketch (b) of a showerhead reactor. Source: (a) Courtesy of
AIXTRON and (b) Gibart 2004 [55]. Reproduced with permission of IOP Publishing.
wafers plays an important role. In modern reactors, it can be easily adjusted

even during an epitaxial run.
• Turbo disk reactor (see, e.g. Ref. [58, 59]): this concept is also based on a verti-
cal gas flow from top to bottom. However, here, excellent layer properties are
obtained by a rotation of the susceptor with high speed (several 100 to more
than 3000 rounds per minute). This fast rotation leads to the formation of a
very uniform gas layer over the substrate surface, enhancing the layer unifor-
mity significantly. Hence, the susceptor rotation speed is a further variable to
optimize the epitaxial growth [60].
In such reactors, a particular focus is given to introduce the metalorganic group

III precursor and the ammonia flows physically separate into the growth chamber
(see Figure 2.12), which further helps in suppressing the above briefly discussed
parasitic prereactions [48].
2.3.2 Epitaxy on Foreign Substrates

As mentioned above, a substrate with the same crystalline properties as GaN
would be the best choice. However, other boundary conditions reduce the choice
further:
• Wafer stability at GaN deposition conditions, i.e. mainly high-temperature sta-
bility;
• High-crystalline perfection and homogeneity over larger areas;
• Possibility to prepare a clean wafer surface (i.e. no or minimal change of pre-
pared surface by oxidation);
• Reasonable price, i.e. manufacturability of respective bulk substrates of rea-
sonable size.
These demands disqualify a lot of potential materials having – on a first
glance – well-suited crystalline properties. A detailed discussion about this
issue may be found in Ref. [61]. Over the recent years, only few materials have
proven their applicability as wafers for GaN epitaxy, including sapphire, SiC, and
Si. Here, we discuss briefly critical issues to be taken into account when using
these foreign substrates. Their most important crystalline properties are listed
in Table 2.2.
2.3.2.1 Sapphire as a Foreign Substrate

The most prominent foreign substrate material is definitely still sapphire (Al2 O3 ),
which has already been used in the pioneering work of Manasevit et al. [45]. Sim-
ilar to GaN, it has a hexagonal symmetry. However, its lattice constants differ
strongly from those of GaN. In most applications, the c-plane of sapphire is taken
as the wafer surface. Looking into respective data sheets, we find a mismatch of
the in-plane a lattice constants of more than 30% – for epitaxial heterostructures
seemingly out of any reasonable range. Luckily, when growing GaN on sapphire,
we observe a rotation of the epitaxial lattice by 30∘ , which reduces the in-plane
lattice mismatch to about 15%, as now every second lattice point of GaN is fairly
close to a sapphire lattice point [65]. However, this rotation makes it impossi-
ble to establish an excellent vertical facet by cleaving of a sapphire–GaN struc-
ture, which would be needed as mirrors for GaN-based Fabry–Pérot laser diodes.
Table 2.2 Crystalline data of sapphire, SiC, and Si which are important for epitaxial growth.
In-plane thermal
Crystalline Typical surface In-plane lattice expansion
Material structure orientation constant (nm) coefficient (K−1 )
GaN Hexagonal c-Plane 0.3189 [62] 5.6 × 10−6

Sapphire Rhombohedral c-Plane 0.476 [62] 7.3 × 10−6 [61]
SiC Hexagonal c-Plane 0.308 [62] 4.5 × 10−6 [61]
Si Cubic {111} 0.384 3.6 × 10−6
The data for the thermal expansion coefficients represent averaged values as this property typically
changes significantly with temperature.
Source: Data for sapphire in [63] and Si in [64].
2.3 MOVPE Growth 59
Moreover, this still enormous lattice mismatch causes the generation of so-called
“threading dislocations” (TDs) at the substrate–layer interface, typically running
perpendicular to the growing surface with areal densities above 1012 cm−2 , mak-
ing any reasonable device application impossible. Therefore, other tricky recipes
are needed to grow low defect-density GaN on sapphire.
Here, we touch a first great invention made by Akasaki and coworkers more
than 30 years ago pushing the GaN quality a very significant step forward and
qualifying these guys for the Noble Prize in 2014: similar as proposed for the
growth of GaAs on Si [66], Amano et al. introduced a low-temperature nucleation
layer made of AlN [67].
Few years later, Nakamura achieved similarly good results by using a GaN
low-temperature nucleation layer [68]. Such nucleation layers are typically
deposited at temperatures of 400–800 ∘ C (with lower temperatures for GaN
and higher ones for AlN nucleation) with a thickness of 10–20 nm. They
have a fairly polycrystalline or even amorphous texture, providing a kind of
decoupling between the substrate and the subsequently grown GaN layer. After
appropriate annealing of the nucleation layers [69, 70], the latter, grown at
the above-mentioned high temperatures of about 1000 ∘ C, quickly improves
in quality, leading to TDDs in the mid 109 cm−2 range after a few hundred
nanometer thickness [65]. Even better data can be achieved by slight oxygen
doping of the AlN nucleation layer [71–73].
Although similar quality can be achieved by either nucleation layer type, the
GaN layer grown on top typically provides a different strain state, mainly depend-
ing on the details of the nucleation layer deposition process. It can be measured
by X-ray diffraction (XRD) by analyzing the change of the c lattice constant of the
main GaN layer. For this kind of biaxial strain, it is related to the in-plane lattice
constant a by the respective components of the elastic tensor (Hooke’s tensor) C:
Δc C Δa C
= 𝜀⟂ = −2 𝟏𝟑 = −2 𝟏𝟑 𝜀∥ (2.1)
c C𝟑𝟑 a C𝟑𝟑
Moreover, the bandgap of GaN and hence the position of the donor-bound
excitonic peak (D0 X), the dominating signal of fairly pure GaN at low tempera-
ture (Figure 2.14), changes by strain, offering another simple method for strain
measurement.
This strain (at room or low temperature) is a result of mainly two sources:
(i) strain generated during the deposition of the GaN main layer and (ii) strain
induced during cooling down after epitaxy by the difference of the thermal
expansion coefficients of GaN and sapphire. Both lead to biaxial strain, i.e. the
strain-generating stress acts at the substrate–layer interface. For the typical
temperature difference between 1000 ∘ C and room temperature, the latter can
result in a compressive strain contribution of about 𝜀 = 1.3 × 10−3 , leading to
a shift of the (D0 X) peak to higher energies of up to 25 meV from the value of
3.47 eV for unstrained GaN. Hence, if the main GaN layer grows unstrained,
such strain in the sample after cooling down can cause a (D0 X) peak position at
low temperature (close to liquid He temperature) of about 3.495 eV.
A general trend seems to be that a GaN nucleation layer leads to less com-
pressive strain, indicating that the main layer grows under tensile strain, which
λvac (nm)
360 358 356 354
105
T = 10 K
D0X
104
PL intensity (a.u.)
XA
103 A0X
TES
2
XB
10
3.44 3.46 3.48 3.50 3.52

Energy (eV)
Figure 2.14 Low-temperature (10 K) photoluminescence spectrum of an about 2 μm thick

MOVPE-grown GaN layer on an AlN:O nucleation layer and an in situ deposited SiN nanomask
layer (c.f. Section 2.3.4). Notice the very narrow peak width of only 870 μeV of the donor-
bound excitonic peak (D0 X) at 3.486 eV. Source: Hertkorn et al. 2007 [72]. Reprinted with
permission of Elsevier.
then gets compensated during cooling down [74]. A possible explanation is that
in the beginning of the growth process, GaN islands nucleate on the substrate.
When these islands coalesce, the initial contact is followed by a subsequent sur-
face roughness reduction, leading to tensile stress formation [75]. The amount of
strain then depends on the island size [76].
Certainly, one of the most impressive features of GaN is that even with such
huge dislocation densities, reasonable LED performance could be achieved,
mainly driven by Nakamura et al. in the early nineties of past century [77] as
discussed in more detail in other chapters of this book. Nevertheless, since then,
a major goal was to reduce the TDD on foreign substrates even further. We will
discuss some successful methods later in this chapter.
2.3.2.2 GaN on SiC and Si

As mentioned above, a bunch of other materials have been discussed as prospec-
tive substrates for GaN epitaxy (see, e.g. Ref. [61] and references therein). How-
ever, many of them could not meet their expectations. Besides sapphire, only
SiC and Si have proven fairly well suitable for the successful growth of GaN het-
erostructures and devices; hence, they will be briefly discussed here.
The expectations about SiC stem from its material properties that are not
too far from those of GaN: it typically crystallizes in hexagonal symmetry, its
lattice constant is only about 3% smaller than that of GaN, and it is a thermally
and chemically very stable crystal with a bandgap of about 3 eV. In contrast to
sapphire, SiC can be regarded as a semiconductor; in particular, its electrical
conductivity can be varied over a large range by doping. The latter is particularly
2.3 MOVPE Growth 61
useful for optoelectronic devices where a conductive substrate enables a

back-side n-contact as well established for arsenide and phosphide LEDs and
lasers. The in-plane crystalline orientation of GaN grown on the c-plane of SiC
is the same as that of the substrate making cleaving for laser mirrors a fairly
simple task. On the other hand, SiC can be made semi-insulating excellently,
qualifying it as a substrate for GaN-based field effect transistors. Moreover, its
great thermal conductivity of about 5 W/cm K makes it particularly suitable for
high-power devices. The most dominant disadvantage of SiC is still its huge price
being at least about 20 times higher than that of equivalent sapphire wafers.
The latter is the main driving argument for Si to be used as a substrate for GaN:
this material is easily available in excellent quality and large wafer size for an
extremely low price owing to the decades of development for conventional elec-
tronic applications. Moreover, such large wafer size (8–12 in.) could simplify any
device processing after epitaxial growth; the possibility to use (even older gener-
ation) sophisticated Si processing equipment adds to this supporting statement.
On the other hand, the basic properties of Si seem to be quite unfavorable
for GaN epitaxy: the lattice mismatch is as large as about 20%, further down-
graded by a large mismatch of the thermal expansion coefficients (Table 2.2). Its
low bandgap may impose problems for optoelectronic GaN-based devices as the
generated light is absorbed in the substrate. Indeed, Si forms a cubic structure.
Fortunately, its {111}-plane is fairly compatible with the c-plane of GaN.
Both materials have in common silicon as a major (or the only) component that
involves two basic problems for the epitaxial growth of GaN: first, Si easily reacts
with the nitrogen precursor to form Six Ny , a dielectric material that may block
further GaN growth (see also Section 2.3.4). Second, at higher temperatures, Ga
and Si form an alloy that also leads to the suppression of GaN growth with some
detrimental formation of Ga–Si conglomerations by even consumption of deeper
Si of the substrate, also well known as “melt-back etching” problem, and leads to
polycrystalline GaN growth.4
Hence, any epitaxial growth on these substrates should start by covering the
substrate surface with a layer that suppresses these parasitic reactions. Indeed,
AlN turned out to be a good choice: besides being Ga-free, an Al—N bond forms
much easier than a Si—N bond, thus preventing the formation of a SiN interlayer
[78].
On SiC, a single AlN nucleation layer deposited at fairly high temperatures has
proven to enable subsequent growth of GaN with high quality (see, e.g. Refs. [79,
80] and many others) as needed for GaN-based field effect transistors. With thick-
nesses ranging typically between 40 and 200 nm, it is significantly thicker than
an AlN nucleation layer on sapphire. However, owing to its large bandgap, such a
layer is electrically insulating, thus preventing the formation of an electrical back-
side contact as would be needed for optoelectronic devices. Therefore, Alx Ga1−x N
has also been studied as a buffer layer. Surprisingly, even for small amounts of Al
(x > 8%), the Ga-related melt-back etching does not occur and a good GaN layer
and device properties have been achieved [81, 82].
4 Notice that such problems do not occur when growing GaAs on Si at substantially lower
temperatures as compared to GaN [66].
The epitaxial growth of GaN on Si is even more challenging. Also here, growth
typically starts with an AlN layer, which solves the melt-back etching problem.
Then, the major remaining problem is the tensile strain as a consequence of the
mismatch of the thermal expansion coefficients (see Table 2.2), leading – if not
compensated – to severe cracking of the top GaN layer even at thicknesses of a
few hundred nanometers. Over the recent years, scientists have typically man-
aged to cope with this problem by introducing multiple Alx Ga1−x N–Aly Ga1−y N
heterostructures and superlattices including specific variations of the growth
temperature of the single layers. For example, Lin et al. could optimize the
strain situation with an 80 period AlN–GaN superlattice by adjusting the GaN
layer thickness [83]. An excellent overview over many such approaches has
been published recently by Zhang and Liu [78]. As another example, we note
that Dadgar et al. have achieved a crack-free total epitaxial layer thickness
of 14.3 μm on a 6 in. Si wafer with seven intermediate thin AlN layers where
the last pure GaN layer amounts to 4.5 μm [84]. These very good results often
need further optimization steps, which will be discussed in more detail in
Section 2.3.4. They enabled to achieve outstanding GaN-based device properties
on Si substrates, particularly for field effect transistors (high electron mobility
transistors, HEMTs) [85]. Consequently, the commercialization of such devices
is progressing rapidly.
2.3.3 Defect Reduction by ELOG, FACELO, etc.

Owing to the huge lattice mismatch to all these foreign substrates mentioned
above, gallium nitride epitaxial layers are typically blamed by a large dislocation
density, as mentioned above. Hence, methods to decrease such dislocation den-
sity have found significant research focus over the recent years. A basic idea of
most of these methods is to stop the further development of threading disloca-
tions by some intermediate blocking layer.
As these defects are particularly harmful to laser diodes, Nakamura et al.
applied the method “epitaxial lateral overgrowth” (ELOG) (Figure 2.15) to
the MOVPE growth of the GaN buffer layer for their laser diodes [86], which
was then already developed for GaAs VPE [87] and LPE [88] and GaN HVPE
[89]. Nakamura et al. achieved a significant reduction of the TDD and hence a
strong improvement of their laser diode lifetime. This method works as follows:
typically, a GaN buffer layer is covered with a dielectric mask (e.g. SiO2 ), which
is then structured into a stripe geometry by conventional optical lithography. In
the subsequent epitaxial process, GaN only nucleates on the GaN sublayer areas
in the mask openings. After growing out of the openings, it expands laterally over
the mask and eventually coalesces to a defect-poor layer [90–93]. Dislocation
densities below 106 cm−2 have been reported on the defect-poor regions [94].
ELOG provides good-quality GaN over the masked areas, leaving still fairly
broad stripes of high dislocation density over the mask openings. Hence, sensitive
devices (e.g. laser diodes) should be positioned over the defect-poor regions. This
problem can be overcome by a slight variation, the method of “facet-assisted epi-
taxial lateral overgrowth” (FACELO) [95]. In this case, the process parameters of
the second GaN growth are chosen so that triangular stripes grow out of the mask
2.3 MOVPE Growth 63
Figure 2.15 Epitaxial lateral overgrowth: after depositing a stripe mask on a GaN buffer layer,
GaN grows out of the stripe openings (left) and coalesces by lateral growth (right). The
threading dislocations continue to develop vertically over the stripe openings, whereas the
overgrown regions remain virtually defect-free.
Figure 2.16 Facet-assisted epitaxial lateral overgrowth: after depositing a stripe mask on a
GaN buffer layer, GaN grows out of the stripe openings forming triangular stripes (left). Then,
the growth mode is changed to favor lateral growth, which leads to a bending of the
dislocations (right).
openings (Figure 2.16) by pushing mainly the vertical growth. After full devel-
opment of these triangles, the growth parameters are changed to favor lateral
growth. This enforces originally vertically developing dislocations to bend and
grow laterally over the masked area. Hence, they do no longer thread to the final
surface, and very defect-poor surfaces can be obtained. The density of defects
over the entire surface is reported to be below 2 × 107 cm−2 , decreasing down to
5 × 106 cm−2 in the regions between the coalescence boundaries [55]. Only a thin
defective stripe develops where the lateral growth fronts of two openings meet.
Such defective stripes can be suppressed by designing a FACELO mask with a
hexagonal honeycomb-like grid leading to a 2-D lateral overgrowth and eventu-
ally to a very uniform surface with dislocation densities below 106 cm−2 [96].
Some groups have etched down the ELOG stripe pattern into the GaN buffer
calling this method “lateral overgrowth from trenches” (LOFTs). The growth now
nucleates at the vertical GaN sidewalls exposed by the etching process. Chen et al.
achieved by this method a dislocation density of 6 × 107 cm−2 [97]. When etched
down even further into the foreign substrate, e.g. sapphire, leaving GaN stripes on
sapphire ridges, this method was named “Pendeo-Epitaxy” [98]. Now, the dielec-
tric mask is no longer needed for the subsequent steps. The second GaN growth
nucleates on the vertical sidewalls of the formed trenches, also leading to a very
defect-poor material as it potentially grows without contact to the substrate.
Over the recent years, many more variations of these methods have been
investigated. For the interested reader, an excellent review has been published by
P. Gibart [55]. These procedures are nowadays also applied successfully to even
more challenging materials such as AlN (see, e.g. Ref. [99]).
2.3.4 In Situ ELOG by SiN Deposition

The methods discussed in Section 2.3.3 all have in common the problem that
a quite complicated multistep processing including even two epitaxial steps is
required. Hence, they are not well suited for industrial production processes. This
triggered research activities to deposit a suitable mask layer in situ during the
MOVPE process (see, e.g. Refs. [100–102]), leading to a fairly simple procedure
then called “in situ ELOG” [103].
It is briefly described as follows: after the optimized growth of a thin GaN buffer
layer (on a foreign substrate, e.g. sapphire), the TMGa supply is interrupted.
Instead, SiH4 is introduced into the reactor, a gas, which is commonly available
in MOVPE machines as it is needed for n-type doping (see Section 2.3.5).
Together with ammonia, this leads to the growth of a thin SiN masking layer.
Its thickness should be less than a monolayer, i.e. a fairly porous SiN layer is
developing. Then, GaN growth is resumed. It is assumed that this layer only
nucleates in the openings of the porous SiN layer; hence, large parts of the GaN
sublayer are masked, leading to a blocking of the threading dislocations. Initially,
the second GaN layer shows a fairly rough morphology, as measured by in situ
roughness analysis (see Section 2.3.1) but flattens after few hundred nanometer
growth. For layer thicknesses of about 1–2 μm, excellent GaN layer quality (c.f.
Figure 2.14) with dislocation densities of less than 2 × 108 cm−2 can be achieved
after careful process optimization [104]. In these studies, we found an optimum
position of the SiN nanomask layer about 100–300 nm above the substrate, while
others obtained also very dislocation-poor layers when depositing the SiN in
situ nanomask layer directly on the sapphire wafer surface before they start the
GaN nucleation growth [105].
2.3.5 Doping of Nitrides

A big advantage of MOVPE as compared to most of the other epitaxial methods
is that doping of the epilayers can be achieved easily and well controlled if an
adequate precursor exists. Similar to the main components, metalorganic com-
pounds are used for metallic dopants, whereas for many n-type dopants, hydrides
are available.
For n-type doping of GaN and its ternaries, the most common dopant is sili-
con (Si), incorporated on a metal site in the crystal. The mostly used respective
precursors are gaseous SiH4 and Si2 H6 . Under normal MOVPE conditions, the
Si incorporation efficiency is very high. Therefore, these hydrides are typically
provided as strongly diluted gas mixture, e.g. 0.1% in N2 or H2 or even lower.
To increase the dynamics of the doping channel enabling a doping concentration
range between low 1017 cm−3 and mid 1019 cm−3 or even higher, the respective
doping channel is in most cases constructed as a double-dilution channel: the
gas flowing out of the precursor container (see, e.g. Figure 2.10) is mixed with a
carrier gas stream. Only a fraction of this mixture is fed into the reactor, whereas
2.3 MOVPE Growth 65
the other part goes directly into the exhaust, all well controlled by mass flow and
pressure controllers.
Consequently, Si doping of GaN can then be achieved quite easily. Its amount
can be excellently controlled, and abrupt doping profiles can be realized, as Si
diffusion or other detrimental effects are negligible. Therefore, highly Si-doped
layers can be used as marker layers to monitor the growth procedure when struc-
tured sublayers are overgrown. We have used this method [106] to optimize the
above-discussed FACELO procedure (Section 2.3.3).
However, Si leads to tensile strain in GaN [107], which may even cause cracks
in the epilayer. This is obviously caused by some interaction of the dopant
atoms and the threading dislocations [108, 109] as defect-poor GaN develops
significantly less strain. Anyway, this problem triggered to study germanium
(Ge) as an alternative n-type dopant where no significant strain developed
at similar or higher doping concentrations [110], enabling Ge concentrations
significantly above 1020 cm−3 [111]. Typical Ge precursors are germane (GeH4 )
or metalorganic Ge compounds such as isobutylgermane. However, the Ge
incorporation efficiency is significantly lower than that of Si [111, 112]. There-
fore, for not too high doping concentrations as required for most optoelectronic
and electronic devices, still Si is the mostly preferred dopant, as it seems to be
easier in handling.
On the contrary, p-type doping of GaN is still regarded as a big challenge. Until
the late eighties of past century, this could not be achieved at all, although all
prospective dopant elements such as Be, Cd, Zn, or Mg have been investigated
[113, 114]. In 1989, Akasaki and coworkers contributed with another great inven-
tion by detecting that Mg can be “activated” to act as an acceptor in GaN. They
originally achieved such activation by a process then called “LEEBI” (low-energy
electron beam irradiation) performed in a secondary electron microscope [115].
Few years later, Nakamura et al. [116] demonstrated that such activation can
also be achieved by thermally annealing GaN:Mg in a hydrogen-free atmosphere
(Figure 2.17a), while a subsequent annealing in a hydrogen-containing atmo-
sphere leads again to semi-insulating properties of the layers (Figure 2.17b).
They then found out that Mg incorporated in a hydrogen-containing atmosphere
forms a complex with a hydrogen atom in the crystal disabling its acceptor
state [117]. Such passivation is regarded as unavoidable in MOVPE, particularly
owing to the use of NH3 as a nitrogen precursor, and a subsequent thermal
annealing step (or LEEBI) is required, which enforces the hydrogen to diffuse
out of the crystal changing GaN:Mg to p-type conductivity. Similar effects have
been observed in other III–V compounds [118]. More recent studies have shown
that hydrogen seems to act beneficially by increasing the Mg incorporation
efficiency at least up to the mid 1019 cm−3 range [119]. Therefore, the passivation
is accepted, as it can be fairly easily alleviated after epitaxial growth.
Until today, Mg is the only element that leads to reasonable p-type conduc-
tivity in GaN, as it has the smallest ionization energy as compared to those
other prospective elements mentioned above (see, e.g. Ref. [120]). As this
energy still amounts to about 180–220 meV (literature is not yet completely
conclusive about this value), even in perfectly depassivated GaN:Mg, only a
small fraction of the acceptor atoms is ionized at room temperature. Then,
Figure 2.17 Mg depassivation by annealing in hydrogen-free atmosphere. (a) Specific

resistivity of GaN:Mg as a function of annealing temperature. Source: From Nakamura et al.
[116]. Copyright 1992, The Physical Society of Japan and The Japan Society of Applied Physics.
(b) If conductive layers are annealed in NH3 , then their resistivity increases again as a
consequence of in-diffusing hydrogen. Source: From Nakamura et al. [117]. Copyright 1992,
The Physical Society of Japan and The Japan Society of Applied Physics.
typically, a hole concentration of end 1017 cm−3 can be found in GaN layers
with a Mg concentration of mid 1019 cm−3 , which have a mobility of about
10–15 cm2 /V s, leading to a specific conductance of not more than about 2 S/cm.
Higher Mg concentrations do not further contribute to p-type conductivity
[121] (Figure 2.18) but may give rise for the development of inversion domains
(see, e.g. Ref. [122]) degrading heavily the crystalline quality of such layers.
Typically, Mg is provided as bis-cyclo-pentadienyl-magnesium (Cp2 Mg), a met-
alorganic precursor that is solid at room temperature and has a fairly low vapor
pressure (see Table 2.1), making its use somewhat less handy. Other Mg metalor-
ganics have an even lower vapor pressure.
Another inherent problem of Mg doping is the lack of achieving abrupt dop-
ing profiles. The Mg concentration only changes gradually when switching the
precursor flow on and off. Moreover, strong Mg doping can be found in layers
grown intentionally undoped after a GaN:Mg growth run, an effect called “mem-
ory effect.” The latter points to one commonly accepted explanation for these
problems: the Mg precursor tends to stick on the walls of the gas tubing and of
the reactor [123, 124]. Hence, these walls need to get saturated before the full
precursor flow reaches the substrate after switching on, leading to some doping
delay, and they act as a further doping supply after switching off. These effects
can be minimized by optimizing the geometry and temperature of the respective
reactor parts [125] but seemingly cannot be completely avoided. Additionally,
Mg diffuses from the later grown p-top layers back into the active region of LEDs
[126, 127]. The profile at the upper side is further degraded by Mg segregation
during growth. Particularly, the on-switching – in most p–n-devices, the relevant
2.3 MOVPE Growth 67
Figure 2.18 Hole concentrations in GaN:Mg as measured by the room temperature Hall effect.
The three curves correspond to the calculated hole densities in an impurity compensation
model for three residual donor densities as shown in the legend. Source: Kaufmann et al. 2000
[121]. Reprinted with permission of American Physical Society.
gradient – can be improved by reducing the temperature of the p-layer growth

and thorough timing of the switching moment [126].
2.3.6 Growth of Other Binary and Ternary Nitrides

Until now, we have mainly described the MOVPE growth of GaN as the most
important and exemplary representative of the nitride compound semiconduc-
tors. In this section, we will discuss briefly some growth-related peculiarities
when growing the binaries AlN and InN and – more important with respect to
this book – ternaries formed by all these binaries, i.e. Alx Ga1−x N, Inx Ga1−x N, and
Al1−x Inx N.
The MOVPE growth of AlN is performed at similar conditions as GaN. A major
difference comes from the stronger binding forces of Al in the crystal. Therefore,
Al having arrived on the growing surface is less mobile than Ga. Such thermally
induced surface migration mobility is an important prerequisite for the atoms
to find their best incorporation site on the surface and hence for the growth of
defect-poor epitaxial layers. Consequently, AlN needs typically a higher growth
temperature than GaN [128]. However, at too high temperatures, other problems
arise [129], particularly parasitic gas-phase reactions between TMAl and NH3 .
Although still many studies continue to find best growth conditions, tempera-
tures about 100 ∘ C higher than for GaN seem to be a good compromise. Although
similar procedures as discussed in Section 2.3.3 are heavily under investigation
for the growth of AlN, the resulting layers still have substantially inferior quality,
i.e. higher dislocation density. Recently, a postgrowth high-temperature anneal-
ing step of an AlN buffer layer at more than 1600 ∘ C has helped to get excellent
AlN layers even on foreign substrates [130, 131], which are mainly needed as a
buffer for deep UV LED structures.
Mixing GaN and AlN leads to Alx Ga1−x N, a very interesting ternary semi-
conductor whose bandgap can be varied between that of the two binary end
members, i.e. between 3.4 and 6 eV. Hence, it finds applications not only as bar-
rier material in many optoelectronic and electronic device structures but also as
active material in UV light-emitting devices. The composition of such layers can
be controlled fairly easily in MOVPE by the respective metalorganic precursor
flows. At least for low Al compositions x, where similar growth conditions as
for GaN can be accepted, the molar ratio of these precursor flows is about the
same as the Al/Ga ratio in the solid (taking into account the dimeric character of
TMAl, see Table 2.1). For higher desired Al content, a compromise between best
GaN and best AlN growth conditions should be made; in particular, a higher
growth temperature is recommended. However, several problems arise:
• TMAl is more reactive than TMGa; hence, parasitic gas-phase reactions with
NH3 in the MOVPE chamber become important as a loss mechanism for Al
incorporation (see, e.g. Refs. [132, 133]). This can be circumvented to some
extent by reducing the reactor pressure significantly below 100 hPa. Also, short
distances between precursor inlet into the MOVPE reactor and susceptor can
help making the close-coupled showerhead reactor (see Section 2.3.1) a good
choice [134].
• The higher growth temperature may give rise to a significant decomposition
of the GaN fraction (see Figure 2.19) in the ternary layer, leading to Al-rich
layers grown with reduced growth rate. As observed by Lundin et al. [133], the
respective reactions may have complex dependencies on the precursor flows
and temperature.
• As a compromise regarding the growth temperature must be found, the
reduced surface migration mobility of Al becomes more and more an issue.
Nowadays, high-quality AlGaN layers more or less over the full range of com-
position can be grown by MOVPE, mainly limited by the quality of the buffer
layer (GaN for low x values and AlN for large values). What comes in addition
Figure 2.19 Temperature

dependence of the
decomposition rate of InN
[135], GaN [136], and AlN
layers [137] in flowing H2 :
open symbols: metal polarity,
solid symbols: N-polarity.
Source: From Togashi et al.
[135]. Copyright © 2009,
John Wiley & Sons.
2.3 MOVPE Growth 69
are “intrinsic” material problems such as the challenge to achieve good n- and
p-type conductivity for high-bandgap materials, but these problems are beyond
our MOVPE-related chapter.
InN, the binary nitride with the lowest bandgap of about 0.67 eV [138] still
seems to be the most difficult concerning MOVPE growth. Opposite to AlN,
extremely low growth temperatures (550–600 ∘ C) are required because of the
very low InN dissociation temperature (Figure 2.19) [135, 139]. Owing to the
minor dissociation of NH3 at these temperatures [46], only low growth rates can
be established at a high V–III ratio [140]. To make things worse, hydrogen – even
that released from NH3 –leads to significant etching of InN [141, 142]. Moreover,
surface migration length of In is very small owing to the low growth tempera-
ture, making it difficult to grow high-quality material. Hence, a major concern
of InN growth investigations is how to increase the growth temperature without
enforcing too much the other problems mentioned above.
As a consequence of the bandgaps of the end members GaN and InN,
Inx Ga1−x N potentially covers a huge spectral range from 360 to 1850 nm,
particularly the full visible spectral range. Consequently, it is heavily used as
an active material in shorter wavelength LEDs and laser diodes (mainly green
to blue, whereas longer wavelengths are covered by phosphide and arsenide
compounds). Owing to its huge importance, Section 2.4 is dedicated to InGaN
quantum wells. Here, we briefly discuss some challenges directly related to its
epitaxial growth:
• Owing to the big difference of the lattice constants of GaN and InN of about
11%, InGaN with reasonable In content has a large lattice mismatch to GaN
of at least 1–1.5% (for an In content x ∼ 0.1–0.15). Such mismatch implies that
only fairly thin layers, i.e. quantum wells can be grown coherently strained on
GaN.
• As the nitrides form hexagonal crystals without inversion symmetry, they have
strong piezoelectric properties. They even show spontaneous polarization
because of their crystal structure. Consequently, coherently strained InGaN
quantum wells grown on the c-plane of GaN develop giant internal electric
fields of few million electronvolts per centimeter (indeed mainly caused
by strain-related piezoelectricity, whereas the spontaneous polarization at
GaN–InGaN interfaces is quite small). Particularly in quantum wells, this
leads to the quantum-confined Stark effect (QCSE, see Chapters 1 and 8).
This causes a separation of the electron and hole wave functions in a quantum
well reducing the radiative transition probability [143]. The latter may be one
reason why the efficiency of green light-emitting LEDs is still inferior than
that of their blue counterparts.
• The large material differences between GaN and InN also lead to the prob-
lem that Inx Ga1−x N shows a miscibility gap at intermediate compositions x
[144]. Hence, not all potentially available wavelength ranges can be adequately
reached.
• As discussed for AlGaN above, the growth of InGaN is also a compromise
between the best conditions for GaN and for InN. Owing to the tendency
of InN to decompose, InGaN layers with significant In content can only be
1.0 Figure 2.20 Solid composition of

Indium mole fraction X InxGa1–xN Inx Ga1−x N vs. molar gas-phase
0.8 composition in MOVPE. Source: Matsuoka
et al. 1992 [145]. Reprinted with
0.6 500 °C permission of Springer Nature.
0.4 700 °C
800 °C
0.2
0 0.2 0.4 0.6 0.8 1.0

TMI
TMI + TEG
achieved at comparably low temperatures, at least 200–300 ∘ C below those of

the best GaN MOVPE conditions. Even then, the In—Ga ratio in the gas phase
must be chosen much larger than its value in the solid (Figure 2.20). On the
other hand, as discussed above, higher temperatures would be optimum for
providing enough surface migration mobility to the metals.
• This decomposition tendency is particularly strong when hydrogen is present
[146] (Figure 2.21). Obviously, volatile InH is formed under such condi-
tions [141]. That is why InGaN quantum wells need to be grown in an as
hydrogen-free atmosphere as possible requiring nitrogen as carrier gas, while
the hydrogen released from ammonia still must be accepted. Moreover, the In
incorporation efficiency depends on the growth rate [147]. Hence, a careful
optimization regarding temperature, growth rate, V–III ratio, and other
important MOVPE parameters is required to obtain best-quality defect-poor
layers.
• The finite thermal stability of already grown InGaN layers also sets some limits
to the growth temperature of later grown layers such as the (Al)GaN barriers
needed in optoelectronic devices.
As a consequence of the required low growth temperatures and hence the
reduced surface migration mobility, low growth rates (only few 100 nm/h)
must be chosen for InGaN. Therefore, only small molar Ga precursor flows are
allowed, also demanded by the need of a small Ga/In ratio in the gas phase. Such
low Ga flows cannot be easily established with TMGa because of its huge vapor
pressure (see Table 2.1). That is why typically TEGa is used as a Ga precursor for
the growth of InGaN.
All these problems are discussed in much more detail in Section 2.4.
The third possible combination of these binaries leads to Al1−x Inx N, which is of
great scientific and technological importance, as it may enable lattice-matched
heterostructures with substantial difference of bandgap: for x ≈ 0.18, it has the
same lattice constant as GaN but a larger bandgap. Moreover, as the polarization
properties can also be tuned by the composition, it provides the potential
to polarization-matched structures. Notice that the lattice-matched het-
erostructure still shows some internal electric field because of the spontaneous
2.3 MOVPE Growth 71
50
(a) 710 °C
40 (a) (b) 730 °C
(c) 750 °C
(d) 780 °C
InN% in InGaN
30
(b)
20
(c)
10 (d)
0
0 20 40 60 80 100
Column III hydrogen flow (sccm)
Figure 2.21 InN percent in InGaN as determined by 𝜃 − 2𝜃 XRD as a function of the hydrogen
flow at the growth temperature (a) 710, (b) 730, (c) 750, and (d) 780 ∘ C. Source: Piner et al. 1997
[146]. Reprinted with the permission of AIP Publishing.
polarization differences to GaN, while the piezoelectric field is of course zero.

This makes it a good candidate as an unstrained barrier material in GaN-based
HEMTs (see Chapters 3 and 6).
However, this material is the most difficult ternary for epitaxial growth, as it
is composed of the two extreme binary end members InN and AlN: while low
temperatures are needed for efficient deposition of the InN fraction, very high
temperatures are required for the AlN part. Similarly, other growth parameters
follow opposing trends. Additionally, its miscibility gap is estimated to be even
more severe than that for InGaN [148–150]. That is why the quality of epitaxially
grown AlInN layers is fairly low, making a precise determination of many prop-
erties very difficult. Even the dependence of the bandgap on the composition x,
i.e. the bowing parameter, is still discussed (see, e.g. [151]).
Owing to the desorption problems of InN discussed above, AlInN MOVPE is
performed at similar, albeit slightly higher temperatures as InGaN [152], also with
N2 as the carrier gas [153], i.e. some compromise between InGaN and AlGaN
growth conditions, details depending on the desired AlInN composition.
Besides these basic growth problems, some groups have found significant
amounts of Ga in their AlInN layers, which may stem from depositions in the
MOVPE reactor from preceding growth runs (see, e.g. [154, 155]). This parasitic
effect seems to occur particularly in close-coupled showerhead reactors.
Combining InGaN and AlInN leads to the quaternary material Alx Ga1−x−y Iny N.
Having now one compositional degree of freedom more, it opens possibilities to
adjust lattice mismatch and bandgap or polarization independently. Applications
in LEDs and HEMTs have been investigated (see, e.g. Refs. [156, 157]). However,
the above discussed growth problems increase further [158], making it very dif-
ficult to controllably obtain such attractive properties.
2.4 InGaN QWs: Growth and Decomposition

2.4.1 Growth of InGaN QWs on Polar, Nonpolar, and Semipolar GaN
Substrates
InGaN-based multiquantum wells (MQWs) are widely used as active parts in
green, blue, and near-ultraviolet LEDs and LDs [159]. Changes of the wavelength
are achieved by varying the indium composition in QWs (see Figure 2.22).
The wavelength and efficiency of the emitted light depend on indium composi-
tion, presence of defects, width of the QWs and quantum barriers (QBs), device
architecture, and homogeneity of the indium distribution in the well [161–164].
Indium inhomogeneity in the wells will be described in Section 2.4.2. The indium
composition is very sensitive to the growth conditions [165] as growth rate and
growth temperature of the wells (for the latter, see Figure 2.23). The growth rate
QW 2.5 nm – PL (325 nm at 50 mW/cm2) Figure 2.22 Dependence of

525 emission wavelength on
indium composition in QWs.
500 Calculation made using the
software SiLENSe [160].
475
Wavelength (nm)
450
425
400
375
350
5 10 15 20 25 30
Indium content (%)
MOVPE, CCS AIXTRON reactor Figure 2.23 Dependence of

30 indium composition in QWs
p = 400 mbar, Vgr = 3 nm/min on growth temperature.
Source: Data provided by
25
Robert Czernecki.
Indium content (%)
20
15
10
5
730 740 750 760 770 780
Temperature (°C)
that also influences the indium incorporation is strongly dependent on other

parameters as growth temperature, carrier gas, V/III ratio, and growth pressure
[166]. These parameters are dependent on each other, making the MOVPE
growth of InGaN QWs a nontrivial problem.
The indium content in the InGaN QWs also depends on the miscut of the
GaN substrate [167, 168]. Higher miscut leads to a lower indium content because
of slower atomic step flow. This phenomenon explains partly the indium inho-
mogeneity because the surface of InGaN is always rough. Another interesting
approach to increase the indium incorporation in QWs is the growth on InGaN
pseudosubstrates that are partially or fully relaxed (see, e.g. Ref. [169]). Their
larger lattice parameter results in a higher indium incorporation efficiency in the
subsequent growth of InGaN MQWs.
Many groups found that the indium composition increases with increasing
NH3 flow rate (see, e.g. Refs. [170–172]). Unfortunately, depending on the pres-
sure during growth, dissociation of NH3 into N2 and H2 may occur and cause
decreasing indium content in the well because of higher concentration of H2 . The
influence of H2 on InGaN QWs growth is widely discussed, see for example Refs.
[173, 174]. On the one hand, hydrogen improves the morphology of the growing
layers (both InGaN and GaN) but on the other hand lowers the indium content
in the wells (see Figure 2.21 and respective discussion there) even if it is used only
during the GaN QB growth [173].
A popular approach for optimizing an InGaN/GaN QW/QB-active region is
growing the QBs at higher temperature than the InGaN QWs. This reduces point
and extended defects [175] and also influences the morphology of the interface
between well and barrier and/or indium distribution in the InGaN QWs [176].
Oliver et al. [175] showed that an increased growth temperature of the QBs causes
some width fluctuations of the wells having a more significant effect on the inter-
nal quantum efficiency than an improvement of the quality of the GaN QBs. Effi-
ciency improvement due to width fluctuations is also discussed in Section 2.4.2.
Many research groups (see, e.g. Refs. [176, 177]) indicate that the microstruc-
ture of InGaN/GaN QWs is dominated by V-pits and/or by trench defects. V-pits
or V defects are inverted hexagonal pyramids which nucleate on threading dislo-
cations crossing the InGaN QW with sidewalls formed by {1011} planes. Shio-
jiri et al. described their formation mechanism taking into account the growth
kinetics of GaN and segregation of In atoms in the strain field around the cores
of TDs [178]. At lower growth temperature as in the case of QB growth, GaN
exhibits {1011} facets that are terminated by N atoms. The N surface is stabi-
lized at the lower temperature and the high V/III ratio typical for the MOVPE
growth of these layers. These {1011} facets become unstable at higher growth
temperature because N surface atoms are not stabilized. It means that the growth
rate of the {1011} N surfaces decreases and that of (0001)-Ga surfaces increases
with decreasing reactor temperature. Then, if a masking occurs disturbing the
(0001) layer growth, the termination would occur on the six {1011} planes, which
become the sidewalls of the V-pits. Such masking, which is actually the nucleus
of the V-pit, may be caused by indium atoms, which segregate in the strain field
around the core of a TD. It can hinder Ga atoms from migrating on the {1011}
planes to make a smooth monolayer causing the termination of the InGaN/GaN

growth on the {1011} planes, which become the sidewalls of the V-pit. As the
growth rate on the {1011} facets is lower compared to that on the (0001) plane,
the InGaN/GaN MQWs grow with less thickness on the sidewalls. The V-pits can
be buried during growth of the subsequent GaN capping layers and then the TD
incorporated into the V defect is taken over into the upper layer. Besides indium,
also contaminants such as segregated oxygen may act as such masking material
(= nucleus of V-pits) [179].
Trench defects consist of basal plane stacking faults (BSFs) with ⟨1100⟩
oriented stacking mismatch boundaries (SMBs), which terminate at the apex of
V-shaped trenches [180]. They are formed during the low-temperature transition
from the InGaN QW to the GaN QB growth [181–183]. Sahonta et al. [184]
reported that such trenches may have raised, the same level or lowered centers
compared to the surrounding surface and also exhibit some red shift of the
respective local cathodoluminescence (CL) peak in some cases. Their formation
seems to modify the indium diffusion or incorporation kinetics, resulting in
higher growth rate of the InGaN and hence thicker QWs within the trench
defects, but does not affect the GaN QB growth rate. The higher growth rate
of InGaN QWs can explain both red shift and raised center within the trench
defect. In many publications, trench defects are named “indium inclusions
embedded within V-defects” because of the stronger red shift within the trench
(see, e.g. Ref. [181]).
Addition of H2 during QB growth even at a relatively low temperature of 800 ∘ C
prevents degradation of the surface/interface morphology. It was also demon-
strated that an elevated QB growth temperature of 900 ∘ C even in the absence of
H2 prevents the formation of inclusions and V-defects [183]. Cheong et al. [185]
also showed that a growth interruption affects the optical and structural proper-
ties of InGaN QWs. During the growth interruption, the group-III metalorganic
source was closed before and after InGaN well growth. With increasing time of
the interruptions, surface segregation and well thickness decrease, but the size
and density of V-pits do not change.
However, despite the remarkable progress in device performance, many prob-
lems are associated with the growth of QWs with indium content close to and
above 20%, which is necessary to reach the green spectral range. The drop of the
external quantum efficiency (EQE) with increasing indium content in quantum
wells is known as “green gap” [186, 187]. Its origin is still under discussion, but
as the main factors, large internal piezoelectric fields as a consequence of the lat-
tice mismatch-induced strain (see Section 2.3.6), Auger recombination, and point
defects are considered. Additionally, increasing indium concentration leads to
surface roughening and to a deterioration of the InGaN/GaN interface [188]. To
minimize the “green gap,” different approaches have been developed. Tian et al.
[188] suggest to optimize the surface morphology by switching from 2D and/or
3D island growth mode to a step flow mode, which significantly improved the
internal quantum efficiency. This can be achieved by using GaN substrates with
a miscut of 0.5∘ . Also, Shmidt et al. [189] showed a strong correlation between
surface morphology, random indium fluctuation, and optical properties of green
light-emitting InGaN QWs. Another approach to minimize the “green gap” is
the use of two or three staggered InGaN QWs, i.e. wells grown with a gradient
temperature profile. Such design improves the electron–hole wave function over-
lap, which leads to an enhancement of the radiative recombination rate [190, 191].
Huge effort of many groups has been put into InGaN QW growth optimization
on nonpolar and semipolar substrates (see, e.g. Ref. [192]) in order to avoid lattice
mismatch-induced strong piezoelectric fields along the c-direction and hence the
QCSE (see Chapters 1 and 8). Wernicke et al. [193] conducted a very systematic
study of the InGaN QW growth on various planes: nonpolar (1010) and semipolar
(1012), (1122), (1011), and (2021). Their main conclusions are as follows:
• Indium incorporation strongly depends on the plane orientation: highest In
content is found in QWs grown on the (1011) plane, slightly lower on the (1122)
plane and further decreased on the (2021) plane. The indium content on (1010)
and (1012) planes is similar to that on the (2021) plane in the whole range
of growth temperatures (670–780 ∘ C). A comparison of these results to the
growth on the (0001) c-plane revealed that below 710 ∘ C, the indium incorpo-
ration is the lowest on this plane among all investigated planes, but above this
temperature, it is similar or slightly higher than on the (2021) plane.
• The lowest emission energy in the case of QWs grown on (1011)and (1122)
planes can be easily explained by higher indium incorporation. When explain-
ing the differences in emission energy for the other investigated planes, the
influence of strain and QCSE must be considered.
Interesting results concerning the growth of InGaN QWs on differently ori-
ented planes, particularly on planes with opposite polarities, have been presented
by Zhao et al. [194]. They observed that indium incorporation on Ga-polar sur-
faces (2021) and (3031) is higher than on N-polar surfaces (2021) and (3031).
This is fully consistent with the reports indicating that Ga- and N-polar c-planes
exhibit different indium incorporations [195].
2.4.2 Origins of In Fluctuations

Indium compositional inhomogeneity in InGaN QWs and its influence on the
optical properties of LEDs and LD has been widely discussed for more than
20 years [196–199]. The regions with high indium content have a lower band-gap
energy than the average; therefore, excitons and carriers are expected to be
spatially localized in those regions [199, 200].
Several mechanism may cause indium fluctuation: strain-induced formation
[201, 202], spinodal decomposition [144], differences in local miscut, or compo-
sition pulling effect [203]. Indium fluctuation has been observed both within the
growth plane (lateral direction) and along the growth (vertical) direction. Ther-
modynamic calculations revealed that a large difference in the In—N and Ga—N
interatomic distance produces a wide gap in immiscibility up to a critical tem-
perature of at least 1250 ∘ C [144]. Hence, the growth of InGaN QWs is unstable
at growth temperatures required to achieve significant In incorporation, leading
to local segregation and even to the formation of “dot-like” structures inside the
QWs [163, 164, 204] in lateral direction. Along the growth direction, a composi-
tion pulling effect has been observed that involves the rejection of indium atoms
from the InGaN lattice to the surface in response to mismatch strain because of
the growth on GaN [205]. This effect also has a strong influence on the fluctua-
tions of the QW width [206]. Indium fluctuations originating at the surface of the
growing InGaN are caused by (i) relatively week InN bonds [176] or in extreme
cases by (ii) insufficient ammonia supply that can lead to the formation of indium
droplets on the InGaN surface [206, 207]. Additionally, dislocations and stacking
faults can lead to the formation of In-rich precipitates [176, 181, 208, 209].
Many techniques are used to investigate such indium fluctuations, including
indirect methods such as X-ray photoemission spectroscopy (XPS) [210],
high-resolution X-ray diffraction (HR-XRD) [211], photoluminescence (PL)
and time-resolved photoluminescence [212], RHEED [213], CL [176, 181],
atomic force microscopy (AFM) [176, 181], and direct methods such as sec-
ondary ion mass spectrometry (SIMS) [214], transmission electron microscopy
(TEM) [215], and three-dimensional atomic probe tomography (3DAPT) [216].
Depending on the interpretation of the respective results, several models of In
fluctuations have been established on the nanometer (Figure 2.24) and even
micrometer scale.
The most important information on indium compositional fluctuations in
InGaN QWs comes from electron microscopy. Using this technique, combined
with highly spatially resolved X-ray analysis, Duxbury et al. [215] showed
a direct evidence for In segregation in InGaN/GaN QWs. Moreover, they
observed that In fluctuations were increasingly pronounced in the vicinity of
dislocations. Fluctuations were found to be between 10 and 20 nm in size and
up to 20% of the average composition in the samples grown using MOVPE.
By high-angle annular dark-field (scanning) microscopy (HAADF-STEM) and
Z-contrast imaging, Jinschek et al. [217] revealed the presence of 1–3 nm wide
indium-rich clusters in QWs with nominal In content in the range of 30–40%,
narrowing the bandgap locally to energies as small as 2.65 eV. Potin et al. [218]
compared the In distribution in InGaN QWs grown by MBE and by MOVPE
using high-resolution transmission electron microscopy (HRTEM) and a digital
analysis of lattice images (DALIs) method based on the HRTEM lattice fringe
images. They found similar lateral indium fluctuations in both kinds of samples.
The indium concentration varied on a small scale – In-rich clusters with a lateral
Figure 2.24 Indium fluctuations in InGaN MQWs seen on a nanometer scale – scanning
transmission electron microscopy (STEM) image. Source: Courtesy to Artur Lachowski.
size of 3 nm (MBE) and 4 nm (MOVPE) – and on a larger scale of a few tens

of nanometers, which was attributed to phase separation. It was also shown
that the In distribution in growth direction differs significantly in the MBE and
MOVPE samples.
An enhancement of indium segregation in size and density in In0.31 Ga0.69 N
QWs or the generation of structural defects such as dislocations and stacking
faults with increasing number of wells was described by Cho et al. [208]. Using
HRTEM combined with AFM and energy-dispersive X-ray spectroscopy (EDX)
and XRD, they showed the dependence of composition fluctuations on the
indium content in InGaN QWs. Disk-like indium-rich clusters were found to be
confined in QWs with relatively low indium content. Additionally, it was shown
that In-rich precipitates with a diameter of 5–12 nm were formed near V-shaped
defects.
Some dependence of lateral In fluctuations on the substrate type was described
by Pantzas et al. [219]. The composition was found to be laterally and vertically
homogenous within 1% and 1.5%, respectively, on semibulk substrates in contrast
to wells grown on sapphire: the HAADF-STEM results revealed that the lateral
composition can vary from 16% to 34% over a few nanometers.
However, such electron microscopy studies of indium fluctuations must be
regarded with great care because InGaN is extremely sensitive to an accelerated
electron beam. This problem has been extensively described in a number of pub-
lications, for example, in Refs. [209, 216, 220]. Humphreys [216] showed that
In-rich clusters are produced in InGaN QWs by the electron beam in a TEM.
Small In fluctuation, however, cannot be ruled out. Therefore, additional direct
methods to characterize compositional indium fluctuation on the atomic scale
are required [209, 216]. Atom probe tomography (APT) allows to analyze the
individual atoms and provides not only compositional but also structural infor-
mation on a subnanometer scale, which is more important because the carrier
localization can originate not only from In fluctuations but also from width fluc-
tuations of the InGaN wells [209, 216, 221, 222], see also Section 2.4.1. The local-
ization energy provided by a monolayer well width fluctuation of an InGaN QW
is about 60 meV, sufficient to localize excitons at room temperature [216]. There-
fore, the roughness of the upper interface of the InGaN QW is crucial and can
affect the efficiency of InGaN-based devices [209]. Past and current literature
provides quite different information concerning the presence of indium fluctu-
ations. Using APT analysis, Humphreys [216] showed that indium atoms are
uniformly distributed inside the InGaN QWs without any In clustering. In con-
trary, also using APT analysis, Yang et al. [223] observed randomly distributed In
fluctuations on a nanometer scale, whereas Wu et al. [224] have even found some
periodicity in such fluctuations.
The exact relations between growth conditions and the development of indium
fluctuations in InGaN are still not clear. Lateral and vertical In fluctuations have
been independently reported by a variety of groups and using different growth
conditions. Obviously, it is difficult to relate them to specific growth parame-
ters. Karpov and coworkers [225] showed by model calculations that the indium
composition profile is poorly sensitive to such MOVPE growth parameters of
InGaN QWs as pressure and growth rate. Therefore, more elaborated methods
are required to minimize the negative impact of In fluctuations on the operation

especially of green light emitters [225, 226]. For example, Zhen Deng et al. [227]
obtained a uniform spatial indium distribution by using indium predeposition
(IP) before InGaN well growth.
As expressed by Yang et al. [223], many simulation studies of MQW LEDs have
been performed under the assumption of uniform indium composition. However,
such simulations reveal large deviations from experimental data. Wu et al. [224]
and later Yang et al. [223] showed if random indium fluctuations measured by
APT were put into LED QW simulations, good agreement with observed emis-
sion spectra could be obtained. It was also found that the simulation results of
electrical and optical properties in LEDs such as carrier transport, radiative and
Auger recombination, and efficiency droop are greatly improved by considering
nanoscale indium fluctuations.
In MOVPE, the InGaN QWs must be grown at temperatures of 700–800 ∘ C
(see Section 2.3.6), whereas good p-type layers doped with Mg are obtained at
much higher temperatures in the range of above 900 ∘ C. At lower temperatures,
Mg is passivated by oxygen and nitrogen vacancies [228]. Low hole concentra-
tion causes not only higher resistivity but also poor injection of holes into the
quantum wells and hence low efficiency of light emitters. Therefore, it is desir-
able to grow the p-type barrier of LEDs and laser diodes at high temperature, but
this will influence the quantum wells. This issue will be discussed in the next two
sections.
2.4.3 Homogenization of InGaN QWs

Indium fluctuation can change during thermal treatment. Chuo et al. [229]
reported that after a short time (up to 15 minutes) of annealing of the InGaN
QWs at 900 ∘ C with In content ranging from 23% to 40%, the photoluminescence
peak red-shifted followed by a blue shift with increasing annealing time. They
explained the experimental data by out-diffusion of In atoms from the In-rich
region into the surrounding matrix. TEM analysis confirmed that the density of
“dot-like” structures was reduced and revealed that at 900 ∘ C, a decomposition
of the InGaN QWs did not occur yet. This is a strong indication that before
decomposition, some homogenization process of the In atoms inside the QWs
occurs.
Such a homogenization process was also confirmed in our research. InGaN
QWs with an indium content of 18% were overgrown in different epitaxial
processes with p-type layers at 830, 880, 900, and 930 ∘ C. Electroluminescence
(EL) measurements at room temperature showed a decreasing wavelength with
increasing temperature of p-type layer growth up to 900 ∘ C (Figure 2.25a).
Changes of the wavelength were accompanied by a decreasing full width of half
maximum (FWHM) of the electroluminescence peaks (Figure 2.25b). HR-XRD
measurements confirmed good-quality QWs with lower average In content in
the sample with the p-type layer grown at 900 ∘ C – in good agreement with the
decreasing wavelength of the emitted spectra.
Also, measurements of the temperature-dependent photoluminescence
revealed an “s-shape” of the spectral position of the QW line – indicating the
Figure 2.25 Electrolumine-

scence measurements of
MQWs overgrown with
p-type layers at different
temperatures:
(a) wavelength and
(b) FWHM.
presence of indium fluctuations in QWs – in the sample where the p-type layer
was grown at 830 ∘ C, whereas the sample with the p-type layer grown at 900 ∘ C
showed no “s-shape” behavior. These results confirm a reduction of the In-rich
fluctuations, i.e. homogenization inside the InGaN QWs, similar to observations
in [229]. In the sample with p-type layer growth at 930 ∘ C, decomposition of the
QWs was observed.
2.4.4 Decomposition of the QWs

As mentioned in Section 2.4.3, InGaN quantum wells get decomposed when
overgrown with p-type GaN at high temperatures. Such thermal degradation
was observed for various orientations of InGaN/GaN QWs, i.e. polar (0001)
plane [230] and semipolar plane (2021) [231].
This phenomenon can be observed using various analytical methods such as
PL, EL, and CL and also HR-XRD and TEM at various scales: nanometers (TEM),
few micrometers (CL), and hundred micrometers (PL, EL, and CL), which makes
the experiments extremely laborious.
On the micrometer scale, thermal degradation results in dark areas visible, for
example, in cathodoluminescence measurements [232]. At the nanometer scale,
the thermally degraded areas consist of extended voids (of 50 nm diameter on
average) typically surrounded by thin layers of high In content material and hav-
ing metallic inclusions inside [230, 232].
The models of the decomposition of InGaN QWs proposed in the literature
are not consistent. One approach focuses on the presence of In-rich clusters or
fluctuations, which are supposed to initiate the structural decomposition when
exposed to high temperatures [230, 233]. Another proposed mechanism is based
on the phase separation of InGaN QWs into InN and GaN, followed by decompo-
sition of InN areas to metallic In and molecular nitrogen, which evaporates out
of the structure and leaves some voids behind [232]. However, this mechanism is
also not likely to happen because a phase separation should not occur in the case
of strained InGaN layers with In content below 25% [234]. Taking into account the
results presented in Refs. [235, 236], the diffusion of point defects from the layers
below the QWs and laterally within the InGaN QWs are the factors affecting the
decomposition of the QWs.
A strong correlation exists between indium content and decomposition tem-
perature. Figure 2.26 shows a comparison of two samples, which have the same
QW thickness of 2.7 nm, but different In-contents: 15% and 20%. The samples
were annealed at 930 ∘ C for half an hour. It can be seen that the sample with the
lower In content was not degraded at that temperature.
The literature suggests several methods to suppress the InGaN QW decompo-
sition, for example:
• lowering the growth temperature of p-type layers [237].
Measurements of optical (EL) and structural (HR-XRD and TEM) properties
revealed good quality and homogeneous QWs in structures with p-type layers
grown at 900 ∘ C. Also, Oh et al. [237] showed good-quality LED structures
emitting at 525 nm with the p-type layer grown at 900 ∘ C.
• decrease of InGaN QW thickness [238].
In InGaN/GaN MQWs with the same In content above 20%, but different
QW thickness of 3.2 nm (a) and 1.2 nm (b), we observed that after 30 minutes
of thermal stress at 930 ∘ C, the thinner QWs remain nondecomposed
(Figure 2.27). Unfortunately, thinner QWs emit light of shorter wavelength as
a consequence of stronger quantization and less influence of the QCSE [239].
Therefore, QW narrowing cannot be a direct way to prevent decomposition
in the case of green emitters.
• lowering the growth rate of InGaN layers [240, 241],
• growth on substrates with lower density of TDD [242],
• growth of a GaN cap layer between QW and GaN QB layer with H2 added
during the growth of the barrier [243],
• temperature ramping in QB growth [244].
It should be emphasized that the issues discussed above, besides lowering of the
p-type growth temperature, are correlated with a possible increase of the indium
inhomogeneity inside and/or at the surface of the QWs. An increase in the well
thickness can increase the inhomogeneity of the indium distribution and also
50 mm 50 mm
(a) (b)
Figure 2.26 TEM topographs of InGaN MQWs, (a) emitting in blue (In-content of about 15%);
(b) emitting in green (In-content above 20%) after thermal stress of 930 ∘ C for half an hour.
20 mm 20 mm
(a) (b)
Figure 2.27 TEM topographs of InGaN MQWs with In content slightly above 20%. (a) QW
thickness 3.2 nm (emitting in green). (b) QW thickness 1.2 nm (emitting in blue), both after
thermal stress of 930 ∘ C for half an hour.
increase the stress distribution resulting in weaker thermal stability of thicker

QWs. Decreasing the growth rate helps to suppress In-rich cluster formation and
in consequence decomposition of the QWs. As described in Section 2.4.2, indium
compositional fluctuation are less pronounced on bulk GaN substrates because of
the lower density of dislocations, which also result in a suppression of the decom-
position. Regarding the latter two points, growth of a GaN cap above the QWs
using H2 or temperature ramping during the QB growth reduces indium segre-
gation on the surface of the QWs and also the V-pit density. Hence, it directly
improves thermal robustness of the InGaN QWs.
The most interesting feature (see Figures 2.26 and 2.27) is that the decom-
position of InGaN MQWs starts usually from the first quantum well. In order
to explain this phenomenon, we have done our best to detect differences
between the first and the subsequent quantum wells, but all our experimental
results indicated that the first InGaN quantum well was not different from the
subsequent ones. It strongly supports the earlier mentioned thesis that diffusion
of point defects from the layers below the QWs and laterally within the InGaN
QWs should be taken into account as the factor affecting the decomposition of
the QWs.
2.5 Summary
The methods of HVPE, sodium flux, and ammonothermal GaN crystallization
were presented and briefly discussed. All methods seem to be very promising for
GaN growth and fabrication of high-quality GaN substrates, which can be used
for further applications. Because the future will belong to GaN-on-GaN technol-
ogy, crystallization of this nitride becomes a crucial problem for further develop-
ment of high-power–high-frequency electronic as well as optoelectronic devices.
The main goal is to design a process that will overcome the equilibrium crystal
shape in GaN growth. This will allow to demonstrate true bulk GaN crystals. The
three described technologies are today ready for this breakthrough achievement.
Moreover, we have discussed MOVPE as the most important method to grow
GaN-based heterostructures. Besides some basics, we have mainly concentrated
on GaN-specific problems and how to overcome them including the defect
reduction when growing on foreign substrates, p-type doping, and the growth
of ternary and quaternary materials.
In the last part, we have focused our attention on the growth and microstruc-
ture of InGaN QWs, as they are of particular interest for optoelectronic devices,
but exhibit very specific problems during the epitaxial growth process as
compared to GaN and AlGaN. The origin of indium compositional fluctuations
and its influence on the optical properties as well as their change during
high-temperature treatment were discussed. Moreover, we pointed out the
problem and possible reasons of the decomposition of the quantum wells.
Acknowledgments
We are grateful to M. Heuken (AIXTRON AG) for providing some figures.
The research presented in Section 2.2 was supported by the Department of
the Navy, Office of Naval Research (ONRG-NICOP-N62909-17-12004), by the
Polish National Science Center through projects No. 2017/25/B/ST5/02897 and
2018/29/B/ST5/00338, as well as by TEAM TECH program of the Foundation for
Polish Science cofinanced by the European Union under the European Regional
Development Fund (POIR.04.04.00-00-5CEB/17-00). The research presented
in Section 2.4 was supported by the TEAM program of the Foundation for
References 83
Polish Sciences cofinanced by the European Union under the European Regional
Development Fund (POIR.04.04.00-00-3C81/16), by Polish National Science
Center through projects No. 2018/29/B/ST5/00623, as well as National Centre
for Research and Development projects No. WPC/20/DefeGaN/2018.
References
1 Grüter, K., Deschler, M., Jürgensen, H. et al. (1989). Deposition of high
quality GaAs films at fast rates in the LP-CVD system. J. Cryst. Growth 94:
607–612.
2 Oshima, Y., Yoshida, T., Eri, T. et al. (2010). Freestanding GaN wafers by
hydride vapor phase epitaxy using void-assisted separation technology.
In: Technology of Gallium Nitride Crystal Growth (eds. D. Ehrentraut, E.
Meissner and M. Bockowski), 79–96. Berlin, Heidelberg: Springer-Verlag.
3 Motoki, K. (2010). Development of GaN substrates. SEI Tech. Rev. 70: 28–35.
4 Fujikura, H., Konno, T., Yoshida, T., and Horikiri, F. (2017).
Hydride-vapor-phase epitaxial growth of highly pure GaN layers with
smooth as-grown surfaces on freestanding GaN substrates. Jpn. J. Appl.
Phys. 56 (8): 0855031.
5 Yoshida, T., Oshima, Y., Watanabe, K. et al. (2011). Ultrahigh-speed growth
of GaN by hydride vapor phase epitaxy. Phys. Status Solidi C 8 (7–8):
2110–2112.
6 Sochacki, T., Amilusik, M., Fijalkowski, M. et al. (2014). Examination of
growth rate during hydride vapor phase epitaxy of GaN on ammonothermal
GaN seeds. J. Cryst. Growth 407: 52–57.
7 Oshima, Y., Yoshida, T., Watanabe, K., and Mishima, T. (2010). Properties of
Ge-doped, high-quality bulk GaN crystals fabricated by hydride vapor phase
epitaxy. J. Cryst. Growth 312 (24): 3569–3573.
8 Richter, E., Gridneva, E., Weyers, M., and Tränkle, G. (2016). Fe-doping in
hydride vapor-phase epitaxy for semi-insulating gallium nitride. J. Cryst.
Growth 456: 97–100.
9 Xu, K., Wang, J.F., and Ren, G.Q. (2015). Progress in bulk GaN growth.
Chin. Phys. B 24 (6): 066105.
10 Iwinska, M., Piotrzkowski, R., Litwin-Staszewska, E. et al. (2016). Highly
resistive C-doped hydride vapor phase epitaxy-GaN grown on ammonother-
mally crystallized GaN seeds. Appl. Phys. Express 10 (1): 011003.
11 Iwinska, M., Zajac, M., Lucznik, B. et al. (2019). Iron and manganese as
dopants used in the crystallization of highly resistive HVPE-GaN on native
seeds. Jpn. J. Appl. Phys. 58 (SC): SC1047.
12 Kirste, L., Danilewsky, A.N., Sochacki, T. et al. (2015). Synchrotron
white-beam X-ray topography analysis of the defect structure of HVPE-GaN
substrates. ECS J. Solid State Sci. Technol. 4 (8): P324–P330.
13 Keller, S., Suh, C.S., Fichtenbaum, N.A. et al. (2008). Influence of the
substrate misorientation on the properties of N-polar InGaN/GaN and
AlGaN/GaN heterostructures. Appl. Phys. Lett. 104: 093510.
14 Suski, T., Litwin-Staszewska, E., Piotrzkowski, R. et al. (2008). Substrate mis-

orientation induced strong increase in the hole concentration in Mg doped
GaN grown by metalorganic vapor phase epitaxy. Appl. Phys. Lett. 93 (17):
172117.
15 Liu, Z., Nitta, S., Robin, Y. et al. (2019). Corrigendum to “morphological
study of InGaN on GaN substrate by supersaturation” [J. Cryst. Growth 508
(2019) 58–65]. J. Cryst. Growth 514: 13.
16 Geng, H., Sunakawa, H., Sumi, N. et al. (2012). Growth and strain charac-
terization of high quality GaN crystal by HVPE. J. Cryst. Growth 350 (1):
44–49.
17 Mori, Y., Imade, M., Maruyama, M. et al. (2015). Growth of bulk nitrides
from a Na flux. In: Handbook of Crystal Growth, Chapter 13, 2e (ed. P.
Rudolph), 505–533. Boston, MA: Elsevier.
18 Ehrentraut, D. and Bockowski, M. (2015). High-pressure, high-temperature
solution growth and ammonothermal synthesis of gallium nitride crystals.
In: Handbook of Crystal Growth, Chapter 15, 2e (ed. P. Rudolph), 577–619.
Boston, MA: Elsevier.
19 Imanishi, M., Murakami, K., Yamada, T. et al. (2019). Promotion of lateral
growth of GaN crystals on point seeds by extraction of substrates from melt
in the Na-flux method. Appl. Phys. Express 12 (4): 045508.
20 Imanishi, M., Yoshida, T., Kitamura, T. et al. (2017). Homoepitaxial hydride
vapor phase epitaxy growth on GaN wafers manufactured by the Na-flux
method. Cryst. Growth Des. 17 (7): 3806–3811.
21 Doradziński, R., Dwiliński, R., Garczyński, J. et al. (2010). Ammonothermal
growth of GaN under ammono-basic conditions. In: Technology of Gallium
Nitride Crystal Growth (eds. D. Ehrentraut, E. Meissner and M. Bockowski),
137–160. Berlin, Heidelberg: Springer-Verlag.
22 Hashimoto, T., Letts, E.R., Key, D., and Jordan, B. (2019). Two inch GaN
substrates fabricated by the near equilibrium ammonothermal (NEAT)
method. Jpn. J. Appl. Phys. 58 (SC): SC1005.
23 Pimputkar, S., Kawabata, S., Speck, J.S., and Nakamura, S. (2014). Improved
growth rates and purity of basic ammonothermal GaN. J. Cryst. Growth 403:
7–17.
24 Schimmel, S., Duchstein, P., Steigerwald, T.G. et al. (2018). In situ X-ray
monitoring of transport and chemistry of Ga-containing intermediates under
ammonothermal growth conditions of GaN. J. Cryst. Growth 498: 214–223.
25 Mikawa, Y., Ishinabe, T., Kawabata, S. et al. (2015). Ammonothermal growth
of polar and non-polar bulk GaN crystal. In: Gallium Nitride Materials
and Devices X (eds. J.I. Chyi, H. Fujioka and H. Morkoç), 1–6. International
Society for Optics and Photonics, SPIE.
26 Bao, Q., Hashimoto, T., Sato, F. et al. (2013). Acidic ammonothermal
growth of GaN crystals using GaN powder as a nutrient. CrystEngComm
15: 5382–5386.
27 Ehrentraut, D., Pakalapati, R.T., Kamber, D.S. et al. (2013). High quality,
low cost ammonothermal bulk GaN substrates. Jpn. J. Appl. Phys. 52 (8S):
08JA01.
References 85
28 Zajac, M., Kucharski, R., Grabianska, K. et al. (2018). Basic ammonothermal

growth of gallium nitride – state of the art, challenges, perspectives. Prog.
Cryst. Growth Charact. Mater. 64 (3): 63–74.
29 Fujito, K., Kubo, S., Nagaoka, H. et al. (2009). Bulk GaN crystals grown by
HVPE. J. Cryst. Growth 311: 3011–3014.
30 Sochacki, T., Amilusik, M., Fijalkowski, M. et al. (2015). Examination
of defects and the seed’s critical thickness in HVPE-GaN growth on
ammonothermal GaN seed. Phys. Status Solidi B 252 (5): 1172–1179.
31 Domagala, J.Z., Smalc-Koziorowska, J., Iwinska, M. et al. (2016). Influence
of edge-grown HVPE GaN on the structural quality of c-plane oriented
HVPE-GaN grown on ammonothermal GaN substrates. J. Cryst. Growth 456:
80–85.
32 Amilusik, M., Wlodarczyk, D., Suchocki, A., and Bockowski, M. (2019).
Micro-Raman studies of strain in bulk GaN crystals grown by hydride vapor
phase epitaxy on ammonothermal GaN seeds. Jpn. J. Appl. Phys. 58 (SC):
SCCB32.
33 Bockowski, M., Iwinska, M., Amilusik, M. et al. (2016). Challenges and
future perspectives in HVPE-GaN growth on ammonothermal GaN seeds.
Semicond. Sci. Technol. 31 (9): 093002.
34 Sitar, Z. (2017). Private Communication.
35 Manasevit, H.M. (1968). Single-crystal gallium arsenide on insulating sub-
strates. Appl. Phys. Lett. 12 (4): 156–159.
36 Zilko, J.L. (2001). Metal organic chemical vapor deposition: technology and
equipment. In: Handbook of Thin Film Deposition Processes and Techniques,
Chapter 4, 2e (ed. K. Seshan), 151–203. Norwich, NY: William Andrew
Publishing.
37 Kayser, O., Heinecke, H., Brauers, A. et al. (1988). Vapour pressures of
MOCVD precursors. Chemtronics 3: 90–93.
38 Shenai, D.V., Timmons, M.L., DiCarlo, R.L. et al. (2003). Correlation of vapor
pressure equation and film properties with trimethylindium purity for the
MOVPE grown III–V compounds. J. Cryst. Growth 248: 91–98.
39 Lewis, C.R., Dietze, W.T., and Ludowise, M.J. (1983). The growth of
magnesium-doped GaAs by the OM-VPE process. J. Electron. Mater. 12
(3): 507–524.
40 Stringfellow, G.B. (1999). Organometallic Vapor Phase Epitaxy, 2e. San
Diego, CA: Academic Press.
41 Andre, C., El-Zein, N., and Tran, N. (2007). Bubbler for constant vapor
delivery of a solid chemical. J. Cryst. Growth 298: 168–171.
42 Scholz, F. (2017). Compound Semiconductors: Physics, Technology and Device
Concepts. Singapore: Pan Stanford Publishing.
43 Scholz, F. (2020). MOVPE of group-III heterostructures for optoelectronic
applications. Cryst. Res. Technol. 55: 1900027. DOI: https://doi.org/10.1002/
crat.201900027.
44 Irvine, S. and Capper, P. (eds.) (2019). Metalorganic Vapor Phase Epitaxy
(MOVPE): Growth, Materials Properties, and Applications. Boston, MA:
Wiley.
45 Manasevit, H.M., Erdmann, F.M., and Simpson, W.I. (1971). The use of met-
alorganics in the preparation of semiconductor materials: IV. The nitrides of
aluminum and gallium. J. Electrochem. Soc. 118 (11): 1864–1868.
46 Liu, S.S. and Stevenson, D.A. (1978). Growth kinetics and catalytic effects
in the vapor phase epitaxy of gallium nitride. J. Electrochem. Soc. 125 (7):
1161–1169.
47 Beaumont, B., Gibart, P., and Faurie, J.P. (1995). Nitrogen precursors in
metalorganic vapor phase epitaxy of (Al,Ga)N. J. Cryst. Growth 156 (3):
140–146.
48 Watson, I.M. (2013). Metal organic vapour phase epitaxy of AlN, GaN, InN
and their alloys: a key chemical technology for advanced device applications.
Coord. Chem. Rev. 257 (13): 2120–2141.
49 Brown, A.S. and Losurdo, M. (2015). In situ characterization of epitaxy. In:
Handbook of Crystal Growth, Chapter 29, 2e (ed. T.F. Kuech), 1169–1209.
Boston, MA: North-Holland.
50 Maaßdorf, A., Zeimer, U., Grenzer, J., and Weyers, M. (2013). Linear
thermal expansion coefficient determination using in situ curvature and tem-
perature dependent X-ray diffraction measurements applied to metalorganic
vapor phase epitaxy-grown AlGaAs. J. Appl. Phys. 114 (3): 033501.
51 Yang, F. (2014). Modern metal-organic chemical vapor deposition (MOCVD)
reactors and growing nitride-based materials. In: Nitride Semiconductor
Light-Emitting Diodes (LEDs), Chapter 2 (eds. J. Huang, H.C. Kuo and S.C.
Shen), 27–65. Sawston, Cambridge, UK: Woodhead Publishing.
52 Frijlink, P. (1988). A new versatile, large size MOVPE reactor. J. Cryst.
Growth 93 (1): 207–215.
53 Christiansen, K., Luenenbuerger, M., Schineller, B. et al. (2002). Advances
in MOCVD technology for research, development and mass production of
compound semiconductor devices. Opto-Electron. Rev. 10 (4): 237–242.
54 Dauelsberg, M., Martin, C., Protzmann, H. et al. (2007). Modeling and pro-
cess design of III-nitride MOVPE at near-atmospheric pressure in close
coupled showerhead and planetary reactors. J. Cryst. Growth 298: 418–424.
55 Gibart, P. (2004). Metal organic vapour phase epitaxy of GaN and lateral
overgrowth. Rep. Prog. Phys. 67 (5): 667–715.
56 Van der Stricht, W., Moerman, I., Demeester, P. et al. (1997). Study of
GaN and InGaN films grown by metalorganic chemical vapour deposition.
J. Cryst. Growth 170 (1): 344–348.
57 Boyd, A.R., Degroote, S., Leys, M. et al. (2009). Growth of GaN/AlGaN on
200 mm diameter silicon (111) wafers by MOCVD. Phys. Status Solidi C 6
(S2): S1045–S1048.
58 Tompa, G., McKee, M., Beckham, C. et al. (1988). A parametric investigation
of GaAs epitaxial growth uniformity in a high speed, rotating-disk MOCVD
reactor. J. Cryst. Growth 93 (1): 220–227.
59 Tompa, G.S., Zawadzki, P.A., Mckee, M. et al. (1993). Development and
implementation of large area, economical rotating disk reactor technology
for metalorganic chemical vapor deposition. MRS Proc. 335: 241.
60 Vigdorovich, E.N. (2016). Improving the functional characteristics of gallium
nitride during vapor phase epitaxy. Semiconductors 50 (13): 1697–1701.
References 87
61 Kukushkin, S.A., Osipov, A.V., Bessolov, V.N. et al. (2008). Substrates for
epitaxy of GaN: new materials and techniques. Rev. Adv. Mater. Sci. 17:
1–32.
62 Liu, L. and Edgar, J.H. (2002). Substrates for gallium nitride epitaxy.
Mater. Sci. Eng., R 37 (3): 61–127.
63 Miyagawa, C., Kobayashi, T., Taishi, T., and Hoshikawa, K. (2013). Demon-
stration of crack-free c-axis sapphire crystal growth using the vertical
Bridgman method. J. Cryst. Growth 372: 95–99.
64 Watanabe, H., Yamada, N., and Okaji, M. (2004). Linear thermal expansion
coefficient of silicon from 293 to 1000 K. Int. J. Thermophys. 25 (1): 221–236.
65 Akasaki, I., Amano, H., Koide, Y. et al. (1989). Effects of AlN buffer layer on
crystallographic structure and on electrical and optical properties of GaN
and Ga1−x Alx N (0 < x ≤ 0.4) films grown on sapphire substrate by MOVPE.
J. Cryst. Growth 98 (1): 209–219.
66 Akiyama, M., Kawarada, Y., and Kaminishi, K. (1984). Growth of GaAs on Si
by MOVCD. J. Cryst. Growth 68: 21–26.
layer. Appl. Phys. Lett. 48 (5): 353–355.
68 Nakamura, S. (1991). GaN growth using GaN buffer layer. Jpn. J. Appl. Phys.
30 (10A): L1705–L1707.
69 Koleske, D., Coltrin, M., Cross, K. et al. (2004). Understanding GaN nucle-
ation layer evolution on sapphire. J. Cryst. Growth 273 (1): 86–99.
70 Lorenz, K., Gonsalves, M., Kim, W. et al. (2000). Comparative study of
GaN and AlN nucleation layers and their role in growth of GaN on sap-
phire by metalorganic chemical vapor deposition. Appl. Phys. Lett. 77 (21):
3391–3393.
71 Kuhn, B. and Scholz, F. (2001). An oxygen doped nucleation layer for the
growth of high optical quality GaN on sapphire. Phys. Status Solidi A 188
(2): 629–633.
72 Hertkorn, J., Brückner, P., Thapa, S.B. et al. (2007). Optimization of nucle-
ation and buffer layer growth for improved GaN quality. J. Cryst. Growth 308
(1): 30–36.
73 Bläsing, J., Krost, A., Hertkorn, J. et al. (2009). Oxygen induced strain field
homogenization in AlN nucleation layers and its impact on GaN grown by
metal organic vapor phase epitaxy on sapphire: an X-ray diffraction study.
J. Appl. Phys. 105: 033504.
74 Hearne, S., Chason, E., Han, J. et al. (1999). Stress evolution during metalor-
ganic chemical vapor deposition of GaN. Appl. Phys. Lett. 74 (3): 356–358.
75 Raghavan, S., Acord, J., and Redwing, J.M. (2005). In situ observation of
coalescence-related tensile stresses during metalorganic chemical vapor
deposition of GaN on sapphire. Appl. Phys. Lett. 86 (26): 261907.
76 Böttcher, T., Einfeldt, S., Figge, S. et al. (2001). The role of high-temperature
island coalescence in the development of stresses in GaN films. Appl. Phys.
Lett. 78 (14): 1976–1978.
77 Nakamura, S., Mukai, T., and Senoh, M. (1994). Candela-class

high-brightness InGaN/AlGaN double-heterostructure blue-light-emitting
diodes. Appl. Phys. Lett. 64 (13): 1687–1689.
78 Zhang, B. and Liu, Y. (2014). A review of GaN-based optoelectronic devices
on silicon substrate. Chin. Sci. Bull. 59 (12): 1251–1275.
79 Warren Weeks, T., Bremser, M.D., Ailey, K.S. et al. (1995). GaN thin films
deposited via organometallic vapor phase epitaxy on a (6H)–SiC(0001) using
high-temperature monocrystalline AlN buffer layers. Appl. Phys. Lett. 67 (3):
401–403.
80 Bassim, N., Twigg, M., Eddy, C. Jr., et al. (2005). Lowered dislocation densi-
ties in uniform GaN layers grown on step-free (0001) 4H-SiC mesa surfaces.
Appl. Phys. Lett. 86 (2): 021902.
81 Boeykens, S., Leys, M., Germain, M. et al. (2004). Influence of AlGaN nucle-
ation layers on structural and electrical properties of GaN on 4H–SiC.
J. Cryst. Growth 272 (1): 312–317.
82 Cui, S., Zhang, Y., Huang, Z. et al. (2017). The property optimization of
n-GaN films grown on n-SiC substrates by incorporating a SiNx interlayer.
J. Mater. Sci. - Mater. Electron. 28 (8): 6008–6014.
83 Lin, P.J., Huang, S.Y., Wang, W.K. et al. (2016). Controlling the stress of
growing GaN on 150-mm Si (111) in an AlN/GaN strained layer superlat-
tice. Appl. Surf. Sci. 362: 434–440.
84 Dadgar, A., Hempel, T., Bläsing, J. et al. (2011). Improving GaN-on-silicon
properties for GaN device epitaxy. Phys. Status Solidi C 8 (5): 1503–1508.
85 Zhang, Y., Dadgar, A., and Palacios, T. (2018). Gallium nitride vertical power
devices on foreign substrates: a review and outlook. J. Phys. D: Appl. Phys.
51 (27): 273001.
86 Nakamura, S., Senoh, M., Nagahama, S. et al. (1997).
InGaN/GaN/AlGaN-based laser diodes with modulation-doped
strained-layer superlattices. Jpn. J. Appl. Phys. 36 (12A): L1568–L1571.
87 McClelland, R.W., Bozler, C.O., and Fan, J.C.C. (1980). A technique for
producing epitaxial films on reuseable substrates. Appl. Phys. Lett. 37 (6):
560–562.
88 Ujiie, Y. and Nishinaga, T. (1989). Epitaxial lateral overgrowth of GaAs on a
Si substrate. Jpn. J. Appl. Phys. 28 (Part 2, No. 3): L337–L339.
89 Usui, A., Sunakawa, H., Sakai, A., and Yamaguchi, A.A. (1997). Thick GaN
epitaxial growth with low dislocation density by hydride vapor phase epitaxy.
Jpn. J. Appl. Phys. 36 (Part 2, No. 7B): L899–L902.
90 Craven, M.D., Lim, S.H., Wu, F. et al. (2002). Threading dislocation reduc-
tion via laterally overgrown nonpolar (11¯ 22) a-plane GaN. Appl. Phys. Lett.
81 (7): 1201–1203.
91 Johnston, C.F., Kappers, M.J., Moram, M.A. et al. (2009). Assessment of
defect reduction methods for nonpolar a-plane GaN grown on r-plane
sapphire. J. Cryst. Growth 311: 3295–3299.
92 Chen, C., Yang, J., Wang, H. et al. (2003). Lateral epitaxial overgrowth of
fully coalesced a-plane GaN on r-plane sapphire. Jpn. J. Appl. Phys. 42 (6B):
L640–L642.
References 89
93 Ni, X., Özgür, Ü., Morkoç, H. et al. (2007). Epitaxial lateral overgrowth of
a-plane GaN by metalorganic chemical vapor deposition. J. Appl. Phys. 102
(5): 053506.
94 Zheleva, T.S., Nam, O.H., Ashmawi, W.M. et al. (2001). Lateral epitaxy and
dislocation density reduction in selectively grown GaN structures. J. Cryst.
Growth 222 (4): 706–718.
95 Hiramatsu, K., Nishiyama, K., Onishi, M. et al. (2000). Fabrication and char-
acterization of low defect density GaN using facet-controlled epitaxial lateral
overgrowth (FACELO). J. Cryst. Growth 221: 316–326.
96 Jazi, M.A., Meisch, T., Klein, M., and Scholz, F. (2015). Defect reduction in
GaN regrown on hexagonal mask structure by facet assisted lateral over-
growth. J. Cryst. Growth 429: 13–18.
97 Chen, Y., Schneider, R., Wang, S.Y. et al. (1999). Dislocation reduction in
GaN thin films via lateral overgrowth from trenches. Appl. Phys. Lett. 75
(14): 2062–2063.
98 Zheleva, T.S., Smith, S.A., Thomson, D.B. et al. (1999). Pendeo-epitaxy: a
new approach for lateral growth of gallium nitride films. J. Electron. Mater.
28 (4): L5–L8.
99 Kueller, V., Knauer, A., Brunner, F. et al. (2011). Growth of AlGaN and AlN
on patterned AlN/sapphire templates. J. Cryst. Growth 315 (1): 200–203.
100 Vennéguès, P., Beaumont, B., Haffouz, S. et al. (1998). Influence of in situ
sapphire surface preparation and carrier gas on the growth mode of GaN in
MOVPE. J. Cryst. Growth 187 (2): 167–177.
101 Tanaka, S., Takeuchi, M., and Aoyagi, Y. (2000). Anti-surfactant in III-nitride
epitaxy–quantum dot formation and dislocation termination. Jpn. J. Appl.
Phys. 39 (Part 2, No. 8B): L831–L834.
102 Contreras, O., Ponce, F.A., Christen, J. et al. (2002). Dislocation annihila-
tion by silicon delta-doping in GaN epitaxy on Si. Appl. Phys. Lett. 81 (25):
4712–4714.
103 Datta, R. and Humphreys, C.J. (2006). Mechanisms of bending of threading
dislocations in MOVPE-grown GaN on (0001) sapphire. Phys. Status Solidi
C 3 (6): 1750–1753.
104 Hertkorn, J., Lipski, F., Brückner, P. et al. (2008). Process optimization for
the effective reduction of threading dislocations in MOVPE grown GaN
using in situ deposited SiNx masks. J. Cryst. Growth 310 (23): 4867–4870.
105 Sakai, S., Wang, T., Morishima, Y., and Naoi, Y. (2000). A new method of
reducing dislocation density in GaN layer grown on sapphire substrate by
MOVPE. J. Cryst. Growth 221: 334–337.
106 Habel, F., Brückner, P., and Scholz, F. (2004). Marker layers for the devel-
opment of a multistep GaN FACELO process. J. Cryst. Growth 272 (1–4):
515–519.
107 Romano, L.T., Van de Walle, C.G., Ager, J.W. et al. (2000). Effect of Si dop-
ing on strain, cracking, and microstructure in GaN thin films grown by
metalorganic chemical vapor deposition. J. Appl. Phys. 87 (11): 7745–7752.
108 Forghani, K., Schade, L., Schwarz, U.T. et al. (2012). Strain and defects in
Si-doped AlGaN epitaxial layers. J. Appl. Phys. 112: 093102.
109 Weinrich, J., Mogilatenko, A., Brunner, F. et al. (2019). Extra half-plane
shortening of dislocations as an origin of tensile strain in Si-doped (Al)GaN.
J. Appl. Phys. 126 (8): 085701.
110 Dadgar, A., Bläsing, J., Diez, A., and Krost, A. (2011). Crack-free, highly con-
ducting GaN layers on Si substrates by Ge doping. Appl. Phys. Express 4 (1):
011001.
111 Fritze, S., Dadgar, A., Witte, H. et al. (2012). High Si and Ge n-type dop-
ing of GaN doping - limits and impact on stress. Appl. Phys. Lett. 100 (12):
122104.
112 Nakamura, S., Mukai, T., and Senoh, M. (1992). Si- and Ge-doped GaN
films grown with GaN buffer layers. Jpn. J. Appl. Phys. 31 (Part 1, No. 9A):
2883–2888.
113 Ilegems, M., Dingle, R., and Logan, R.A. (1972). Luminescence of Zn- and
Cd-doped GaN. J. Appl. Phys. 43 (9): 3797–3800.
114 Ilegems, M. and Dingle, R. (1973). Luminescence of Be- and Mg-doped GaN.
J. Appl. Phys. 44 (9): 4234–4235.
(LEEBI). Jpn. J. Appl. Phys. 28: L2112–L2114.
116 Nakamura, S., Mukai, T., Senoh, M., and Iwasa, N. (1992). Thermal anneal-
ing effects of p-type Mg-doped GaN films. Jpn. J. Appl. Phys. 31: L139–L142.
117 Nakamura, S., Iwasa, N., Senoh, M., and Mukai, T. (1992). Hole compensa-
tion mechanism of p-type GaN films. Jpn. J. Appl. Phys. 31 (Part 1, No. 5A):
1258–1266.
118 Antell, G.R., Briggs, A.T.R., Butler, B.R. et al. (1988). Passivation of zinc
acceptors in InP by atomic hydrogen coming from arsine during metalor-
ganic vapor phase epitaxy. Appl. Phys. Lett. 53 (9): 758–760.
119 Castiglia, A., Carlin, J.F., and Grandjean, N. (2011). Role of stable and
metastable Mg–H complexes in p-type GaN for cw blue laser diodes. Appl.
Phys. Lett. 98 (21): 213505.
120 Fischer, S., Wetzel, C., Haller, E.E., and Meyer, B.K. (1995). On p-type dop-
ing in GaN–acceptor binding energies. Appl. Phys. Lett. 67 (9): 1298–1300.
121 Kaufmann, U., Schlotter, P., Obloh, H. et al. (2000). Hole conductivity and
compensation in epitaxial GaN:Mg layers. Phys. Rev. B 62 (16): 10867–10872.
122 Martínez-Criado, G., Cros, A., Cantarero, A. et al. (2003). Study of inver-
sion domain pyramids formed during the GaN:Mg growth. Semicond. Sci.
Technol. 47 (3): 565–568.
123 Ohba, Y. and Hatano, A. (1994). A study on strong memory effects for Mg
doping in GaN metalorganic chemical vapor deposition. J. Cryst. Growth 145
(1): 214–218.
124 Kuech, T., Wang, P.J., Tischler, M. et al. (1988). The control and modeling
of doping profiles and transients in MOVPE growth. J. Cryst. Growth 93 (1):
624–630.
125 Köhler, K., Gutt, R., Müller, S. et al. (2011). Reactor dependent starting tran-
sients of doping profiles in MOVPE grown GaN. J. Cryst. Growth 321 (1):
15–18.
References 91
126 Köhler, K., Stephan, T., Perona, A. et al. (2005). Control of the Mg doping
profile in III-N light-emitting diodes and its effect on the electrolumines-
cence efficiency. J. Appl. Phys. 97: 104914.
127 Köhler, K., Gutt, R., Wiegert, J., and Kirste, L. (2013). Diffusion of Mg
dopant in metal-organic vapor-phase epitaxy grown GaN and Alx Ga1–x N.
J. Appl. Phys. 113 (7): 073514.
128 Imura, M., Nakano, K., Fujimoto, N. et al. (2006). High-temperature
metal-organic vapor phase epitaxial growth of AlN on sapphire by multi
transition growth mode method varying V/III ratio. Jpn. J. Appl. Phys. 45
(11): 8639–8643.
129 Kakanakova-Georgieva, A., Nilsson, D., and Janzén, E. (2012). High-quality
AlN layers grown by hot-wall MOCVD at reduced temperatures. J. Cryst.
Growth 338 (1): 52–56.
130 Miyake, H., Nishio, G., Suzuki, S. et al. (2016). Annealing of an AlN
buffer layer in N2 –CO for growth of a high-quality AlN film on sapphire.
Appl. Phys. Express 9 (2): 025501.
131 Itokazu, Y., Kuwaba, S., Jo, M. et al. (2019). Influence of the nucleation con-
ditions on the quality of AlN layers with high-temperature annealing and
regrowth processes. Jpn. J. Appl. Phys. 58 (SC): SC1056.
132 Coltrin, M.E., Creighton, J.R., and Mitchell, C.C. (2006). Modeling the par-
asitic chemical reactions of AlGaN organometallic vapor-phase epitaxy.
J. Cryst. Growth 287 (2): 566–571.
133 Lundin, W., Nikolaev, A., Rozhavskaya, M. et al. (2013). Fast AlGaN growth
in a whole composition range in planetary reactor. J. Cryst. Growth 370:
7–11.
134 Stellmach, J., Pristovsek, M., Savaş, Ö. et al. (2011). High aluminium content
and high growth rates of AlGaN in a close-coupled showerhead MOVPE
reactor. J. Cryst. Growth 315 (1): 229–232.
135 Togashi, R., Kamoshita, T., Adachi, H. et al. (2009). Investigation of polarity
dependent InN{0001} decomposition in N2 and H2 ambient. Phys. Status
Solidi C 6 (S2): S372–S375.
136 Mayumi, M., Satoh, F., Kumagai, Y. et al. (2002). Influence of lattice polar-
ity on wurtzite GaN{0001} decomposition as studied by in situ gravimetric
monitoring method. J. Cryst. Growth 237–239: 1143–1147.
137 Kumagai, Y., Akiyama, K., Togashi, R. et al. (2007). Polarity dependence of
AlN 0001 decomposition in flowing H2 . J. Cryst. Growth 305 (2): 366–371.
138 Walukiewicz, W., Ager, J.W., Yu, K.M. et al. (2006). Structure and electronic
properties of InN and In-rich group III-nitride alloys. J. Phys. D: Appl. Phys.
39 (5): R83–R99.
139 Matsuoka, T., Sasaki, T., and Katsui, A. (1990). Growth and properties of a
wide-gap semiconductor InGaN. Optoelectron.-Devices Technol. 5 (1): 53–64.
140 Drago, M., Vogt, P., and Richter, W. (2006). MOVPE growth of InN with
ammonia on sapphire. Phys. Status Solidi A 203 (1): 116–126.
141 Koukitu, A., Taki, T., Takahashi, N., and Seki, H. (1999). Thermodynamic
study on the role of hydrogen during the MOVPE growth of group III
nitrides. J. Cryst. Growth 197 (1): 99–105.
142 Maleyre, B., Briot, O., and Ruffenach, S. (2004). MOVPE growth of InN films
and quantum dots. J. Cryst. Growth 269 (1): 15–21.
143 Im, J.S., Kollmer, H., Off, J. et al. (1998). Reduction of oscillator strength due
to piezoelectric fields in GaN/Alx Ga1−x N quantum wells. Phys. Rev. B 57
(16): R9435–R9438.
144 Ho, I. and Stringfellow, G.B. (1996). Solid phase immiscibility in GaInN.
Appl. Phys. Lett. 69 (18): 2701–2703.
145 Matsuoka, T., Yoshimoto, N., Sasaki, T., and Katsui, A. (1992). Wide-gap
semiconductor InGaN and InGaAlN grown by MOVPE. J. Electron. Mater.
21: 157–163.
146 Piner, E.L., Behbehani, M.K., El-Masry, N.A. et al. (1997). Effect of hydrogen
on the indium incorporation in InGaN epitaxial films. Appl. Phys. Lett. 70:
461–463.
147 Keller, S., Keller, B.P., Kapolnek, D. et al. (1996). Growth and characteri-
zation of bulk InGaN films and quantum wells. Appl. Phys. Lett. 68 (22):
3147–3149.
148 Ferhat, M. and Bechstedt, F. (2002). First-principles calculations of gap
bowing in Inx Ga1–x N and Inx Al1–x N alloys: relation to structural and ther-
modynamic properties. Phys. Rev. B 65: 075213.
149 Deibuk, V.G. and Voznyi, A.V. (2005). Thermodynamic stability and redistri-
bution of charges in ternary AlGaN, InGaN, and InAlN alloys. Semiconduc-
tors 39 (6): 623–628.
150 Zhao, G., Xu, X., Li, H. et al. (2016). The immiscibility of InAlN ternary
alloy. Sci. Rep. 6: 26600.
151 Schulz, S., Caro, M.A., Tan, L.T. et al. (2013). Composition-dependent band
gap and band-edge bowing in AlInN: a combined theoretical and experimen-
tal study. Appl. Phys. Express 6 (12): 121001.
152 Takeuchi, T., Kamiyama, S., Iwaya, M., and Akasaki, I. (2018). GaN-based
vertical-cavity surface-emitting lasers with AlInN/GaN distributed Bragg
reflectors. Rep. Prog. Phys. 82 (1): 012502.
153 Sadler, T.C., Kappers, M.J., and Oliver, R.A. (2011). The impact of hydro-
gen on indium incorporation and surface accumulation in InAlN epitaxy. J.
Cryst. Growth 331 (1): 4–7.
154 Smith, M.D., Taylor, E., Sadler, T.C. et al. (2014). Determination of Ga
auto-incorporation in nominal InAlN epilayers grown by MOCVD. J. Mater.
Chem. C 2: 5787–5792.
155 Choi, S., Kim, H.J., Lochner, Z. et al. (2014). Origins of unintentional incor-
poration of gallium in AlInN layers during epitaxial growth, part I: Growth
of AlInN on AlN and effects of prior coating. J. Cryst. Growth 388: 137–142.
156 Ahl, J.P., Hertkorn, J., Koch, H. et al. (2014). Morphology, growth mode and
indium incorporation of MOVPE grown InGaN and AlInGaN: a comparison.
J. Cryst. Growth 398: 33–39.
157 Hahn, H., Reuters, B., Wille, A. et al. (2012). First polarization-engineered
compressively strained AlInGaN barrier enhancement-mode MISH-FET.
158 Loganathan, R., Balaji, M., Prabakaran, K. et al. (2015). The effect of growth
temperature on structural quality of AlInGaN/AlN/GaN heterostructures
grown by MOCVD. J. Mater. Sci. - Mater. Electron. 26 (7): 5373–5380.
References 93
159 Nakamura, S., Pearton, S., and Fasol, G. (2000). The Blue Laser Diode - The
Complete Story. Berlin: Springer Nature.
160 SiLENSe Laser Edition version 5.4, build date: Feb7, 2013 (2013). http://www
.str-soft.com (accessed 13 March 2020).
161 Chichibu, S.F., Abare, A.C., Mack, M.P. et al. (1999). Optical properties of
InGaN quantum wells. Mater. Sci. Eng., B 59 (1–3): 298–306.
162 Nakamura, S. (1998). The roles of structural imperfections in InGaN-based
blue light-emitting diodes and laser diodes. Science 281 (5379): 956–961.
163 Ponce, F.A. and Bour, D.P. (1997). Nitride-based semiconductors for blue
and green light-emitting devices. Nature 386 (6623): 351–359.
164 O’Donnell, K.P. (2001). Mystery wrapped in an enigma: optical properties of
InGaN alloys. Phys. Status Solidi A 183 (1): 117–120.
165 Dupuis, R.D. (1997). Epitaxial growth of III-V nitride semiconductors by
metalorganic chemical vapor deposition. J. Cryst. Growth 178 (1–2): 56–73.
166 Dworzak, M., Periera, T.S., Bügler, M. et al. (2007). Gain mechanisms in
field-free InGaN layers grown on sapphire and bulk GaN substrate. Phys.
Status Solidi RRL 1 (4): 141–143.
167 Sarzyński, M., Suski, T., Staszczak, G. et al. (2012). Lateral control of indium
content and wavelength of III-nitride diode lasers by means of GaN sub-
strate patterning. Appl. Phys. Express 5 (2): 021001.
168 Drózḋ z,̇ P.A., Korona, K.P., Sarzyński, M. et al. (2016). Photoluminescence of
InGaN/GaN quantum wells grown on c-plane substrates with locally variable
miscut. Phys. Status Solidi B 253 (2): 284–291.
169 Even, A., Laval, G., Ledoux, O. et al. (2017). Enhanced In incorporation in
full InGaN heterostructure grown on relaxed InGaN pseudo-substrate. Appl.
Phys. Lett. 110 (26): 262103.
170 Yang, J., Zhao, D.G., Jiang, D.S. et al. (2017). Increasing the indium incorpo-
ration efficiency during InGaN layer growth by suppressing the dissociation
of NH3 . Superlattices Microstruct. 102: 35–39.
171 Li, J., Liu, Z., Liu, Z. et al. (2016). Advances and prospects in nitrides based
light-emitting-diodes. J. Semicond. 37 (6): 61001.
172 Kim, S., Lee, K., Lee, H. et al. (2003). The influence of ammonia pre-heating
to InGaN films grown by TPIS-MOCVD. J. Cryst. Growth 247 (1–2): 55–61.
173 Czernecki, R., Grzanka, E., Smalc-Koziorowska, J. et al. (2015). Effect of
hydrogen during growth of quantum barriers on the properties of InGaN
quantum wells. J. Cryst. Growth 414: 38–41.
174 Suihkonen, S., Lang, T., Svensk, O. et al. (2007). Control of the morphol-
ogy of InGaN/GaN quantum wells grown by metalorganic chemical vapor
deposition. J. Cryst. Growth 300 (2): 324–329.
175 Oliver, R.A., Massabuau, F.C.-P., Kappers, M.J. et al. (2013). The impact of
gross well width fluctuations on the efficiency of GaN-based light emitting
diodes. Appl. Phys. Lett. 103 (14): 141114.
176 Ting, S.M., Ramer, J.C., Florescu, D.I. et al. (2003). Morphological evolu-
tion of InGaN/GaN quantum-well heterostructures grown by metalorganic
chemical vapor deposition. J. Appl. Phys. 94 (3): 1461–1467.
177 Massabuau, F.C., Davies, M.J., Oehler, F. et al. (2014). The impact of trench
defects in InGaN/GaN light emitting diodes and implications for the “green
gap” problem. Appl. Phys. Lett. 105 (11): 112110.
178 Shiojiri, M., Chuo, C.C., Hsu, J.T. et al. (2006). Structure and formation
mechanism of V defects in multiple InGaN/GaN quantum well layers.
J. Appl. Phys. 99 (7): 073505.
179 Liliental-Weber, Z., Chen, Y., Ruvimov, S., and Washburn, J. (1997). Forma-
tion mechanism of nanotubes in GaN. Phys. Rev. Lett. 79 (15): 2835–2838.
180 Massabuau, F.C.P., Sahonta, S.L., Trinh-Xuan, L. et al. (2012). Morphological,
structural, and emission characterization of trench defects in InGaN/GaN
quantum well structures. Appl. Phys. Lett. 101 (21): 212107.
181 Florescu, D.I., Ting, S.M., Ramer, J.C. et al. (2003). Investigation of V-defects
and embedded inclusions in InGaN/GaN multiple quantum wells grown by
metal organic chemical vapor deposition on (0001) sapphire. Appl. Phys.
Lett. 83 (1): 33–35.
182 Kumar, M.S., Lee, Y.S., Park, J.Y. et al. (2009). Surface morphological studies
of green InGaN/GaN multi-quantum wells grown by using MOCVD. Mater.
Chem. Phys. 113 (1): 192–195.
183 Smalc-Koziorowska, J., Grzanka, E., Czernecki, R. et al. (2015). Elimination
of trench defects and V-pits from InGaN/GaN structures. Appl. Phys. Lett.
106 (10): 101905.
184 Sahonta, S.L., Kappers, M.J., Zhu, D. et al. (2013). Properties of trench
defects in InGaN/GaN quantum well structures. Phys. Status Solidi A 210
(1): 195–198.
185 Cheong, M.G., Suh, E.K., and Lee, H.J. (2002). Properties of InGaN/GaN
quantum wells and blue light emitting diodes. J. Lumin. 99 (3): 265–272.
186 Alhassan, A.I., Young, N.G., Farrell, R.M. et al. (2018). Development of high
performance green c-plane III-nitride light-emitting diodes. Opt. Express 26
(5): 5591–5601.
187 Nippert, F., Karpov, S.Y., Callsen, G. et al. (2016). Temperature-dependent
recombination coefficients in InGaN light-emitting diodes: hole localization,
Auger processes, and the green gap. Appl. Phys. Lett. 109 (16): 161103.
188 Tian, A., Liu, J., Zhang, L. et al. (2017). Significant increase of quantum effi-
ciency of green InGaN quantum well by realizing step-flow growth. Appl.
Phys. Lett. 111 (11): 112102. https://doi.org/10.1063/1.5001185.
189 Shmidt, N.M., Chernyakov, A.E., Tal’nishnih, N.A. et al. (2019). The
impact of the surface morphology on optical features of the green emitting
InGaN/GaN multiple quantum wells. J. Cryst. Growth 520: 82–84.
190 Zhao, H., Liu, G., Li, X. et al. (2009). Growths of staggered InGaN quan-
tum wells light-emitting diodes emitting at 520–525 nm employing
graded-temperature profile. Opt. InfoBase Conf. Pap. 95 (6): 61104.
191 Jönen, H., Rossow, U., Bremers, H. et al. (2011). Indium incorporation in
GaInN/GaN quantum well structures on polar and nonpolar surfaces. Phys.
Status Solidi B 248 (3): 600–604.
192 Scholz, F. (2012). Semipolar GaN grown on foreign substrates: a review.
Semicond. Sci. Technol. 27: 024002.
193 Wernicke, T., Schade, L., Netzel, C. et al. (2012). Indium incorporation and
emission wavelength of polar, nonpolar and semipolar InGaN quantum
wells. Semicond. Sci. Technol. 27 (2): 024014.
References 95
194 Zhao, Y., Yan, Q., Huang, C.-Y. et al. (2012). Indium incorporation and emis-
sion properties of nonpolar and semipolar InGaN quantum wells. Appl.
Phys. Lett. 100 (20): 201108.
195 Keller, S., Fichtenbaum, N.A., Furukawa, M. et al. (2007). Growth and char-
acterization of N-polar InGaN-GaN multiquantum wells. Appl. Phys. Lett. 90
(19): 191908.
196 Kwon, Y.H., Gainer, G.H., Bidnyk, S. et al. (1999). Structural and optical
characteristics of Inx Ga1-x N/GaN multiple quantum wells with different In
compositions. Appl. Phys. Lett. 75 (17): 2545–2547.
197 Jiang, H., Minsky, M., Keller, S. et al. (1999). Photoluminescence and
photoluminescence excitation spectra of In0.2 Ga0.8 N-GaN quantum wells:
comparison between experimental and theoretical studies. IEEE J. Quantum
Electron. 35 (10): 1483–1490.
198 Chichibu, S., Sota, T., Wada, K., and Nakamura, S. (1998). Exciton local-
ization in InGaN quantum well devices. J. Vac. Sci. Technol., B 16 (4):
2204–2214.
199 Ruterana, P., Kret, S., Vivet, A. et al. (2002). Composition fluctuation in
InGaN quantum wells made from molecular beam or metalorganic vapor
phase epitaxial layers. J. Appl. Phys. 91 (11): 8979–8985.
200 Cheng, Y.C., Lin, E.-C., Wu, C.-M. et al. (2004). Nanostructures and carrier
localization behaviors of green-luminescence InGaN/GaN quantum-well
structures of various silicon-doping conditions. Appl. Phys. Lett. 84 (14):
2506–2508.
201 Tachibana, K., Someya, T., and Arakawa, Y. (1999). Nanometer-scale InGaN
self-assembled quantum dots grown by metalorganic chemical vapor deposi-
tion. Appl. Phys. Lett. 74 (3): 383–385.
202 Tersoff, J. (1998). Enhanced nucleation and enrichment of strained-alloy
quantum dots. Phys. Rev. Lett. 81 (15): 3183–3186.
203 Hiramatsu, K., Kawaguchi, Y., Shimizu, M. et al. (1997). The composition
pulling effect in MOVPE grown InGaN on GaN and AlGaN and its TEM
characterization. MRS Internet J. Nitride Semicond. Res. 2: e6.
204 Narukawa, Y., Kawakami, Y., Funato, M. et al. (1997). Role of self-formed
InGaN quantum dots for exciton localization in the purple laser diode
emitting at 420 nm. Appl. Phys. Lett. 70 (8): 981.
205 Shimizu, M., Kawaguchi, Y., Hiramatsu, K., and Sawaki, N. (1997). Metalor-
ganic vapor phase epitaxy of thick InGaN on sapphire substrate. Jpn. J. Appl.
Phys. 36 (Part 1, No. 6A): 3381–3384.
206 Dussaigne, A., Damilano, B., Grandjean, N., and Massies, J. (2002). Indium
surface segregation in InGaN/GaN quantum wells. Int. Conf. Mol. Beam
Epitaxy 251 (1): 151–152.
207 Ou, J., Chen, W.K., Lin, H.C. et al. (1998). An elucidation of solid incorpora-
tion of InGaN grown by metalorganic vapor phase epitaxy. Jpn. J. Appl. Phys.
37 (Part 2, No. 6A): L633–L636.
208 Cho, H.K., Lee, J.Y., Kim, C.S., and Yang, G.M. (2002). Influence of strain
relaxation on structural and optical characteristics of InGaN/GaN mul-
tiple quantum wells with high indium composition. J. Appl. Phys. 91 (3):
1166–1170.
209 Gu, G.H., Jang, D.H., Nam, K.B., and Park, C.G. (2013). Composition fluc-
tuation of In and well-width fluctuation in InGaN/GaN multiple quantum
wells in light-emitting diode devices. Microsc. Microanal. 19 (Suppl. 5):
99–104.
210 Moison, J.M., Houzay, F., Barthe, F. et al. (1991). Surface segregation in III-V
alloys. J. Cryst. Growth 111 (1–4): 141–150.
211 Krysko, M. and Leszczynski, M. (2007). Quantification of In clustering in
InGaN-GaN multi-quantum-wells by analysis of X-ray diffraction data. Appl.
Phys. Lett. 91 (6): 061915.
212 Pophristic, M., Long, F.H., Tran, C. et al. (1998). Time-resolved photolumi-
nescence measurements of InGaN light-emitting diodes. Appl. Phys. Lett. 73
(24): 3550–3552.
213 Gérard, J.M. and D’Anterroches, C. (1995). Growth of InGaAs/GaAs het-
erostructures with abrupt interfaces on the monolayer scale. J. Cryst. Growth
150: 467–472.
214 Michałowski, P.P., Grzanka, E., Grzanka, S. et al. (2019). Indium concentra-
tion fluctuations in InGaN/GaN quantum wells. J. Anal. At. Spectrom. 34 (8):
1718–1723.
215 Duxbury, N., Bangert, U., Dawson, P. et al. (2000). Indium segregation in
InGaN quantum-well structures. Appl. Phys. Lett. 76 (12): 1600–1602.
216 Humphreys, C.J. (2007). Does In form In-rich clusters in InGaN quantum
wells? Philos. Mag. 87 (13): 1971–1982.
217 Jinschek, J., Erni, R., Gardner, N. et al. (2006). Local indium segregation
and band gap variations in high efficiency green light emitting InGaN/GaN
diodes. Solid State Commun. 137 (4): 230–234.
218 Potin, V., Hahn, E., Rosenauer, A. et al. (2004). Comparison of the In distri-
bution in InGaN/GaN quantum well structures grown by molecular beam
epitaxy and metalorganic vapor phase epitaxy. J. Cryst. Growth 262 (1–4):
145–150.
219 Pantzas, K., Patriarche, G., Troadec, D. et al. (2015). Role of compositional
fluctuations and their suppression on the strain and luminescence of InGaN
alloys. J. Appl. Phys. 117 (5): 055705.
220 ̇
Kret, S., Dłuzewski, P., Szczepańska, A. et al. (2007). Homogenous indium
distribution in InGaN/GaN laser active structure grown by LP-MOCVD on
bulk GaN crystal revealed by transmission electron microscopy and X-ray
diffraction. Nanotechnology 18 (46): 465707.
221 Grandjean, N., Damilano, B., and Massies, J. (2001). Group-III nitride quan-
tum heterostructures grown by molecular beam epitaxy. J. Phys. Condens.
Matter 13 (32): 6945–6960.
222 Brandt, O., Waltereit, P., Jahn, U. et al. (2002). Impact of In bulk and surface
segregation on the optical properties of (In,Ga)N/GaN multiple quantum
wells. Phys. Status Solidi A 192 (1): 5–13.
223 Yang, T.J., Shivaraman, R., Speck, J.S., and Wu, Y.R. (2014). The influence of
random indium alloy fluctuations in indium gallium nitride quantum wells
on the device behavior. J. Appl. Phys. 116 (11): 113104.
References 97
224 Wu, Y.R., Shivaraman, R., Wang, K.C., and Speck, J.S. (2012). Analyzing the
physical properties of InGaN multiple quantum well light emitting diodes
from nano scale structure. Appl. Phys. Lett. 101 (8): 083505.
225 Talalaev, R.A., Karpov, S.Y., Evstratov, I.Y., and Makarov, Y.N. (2002). Indium
segregation in MOVPE grown InGaN-based heterostructures. Phys. Status
Solidi C 192 (1): 311–314.
226 Karpov, S.Y. (2017). Carrier localization in InGaN by composition fluctua-
tions: implication to the “green gap”. Photonics Res. 5 (2): A7–A12.
227 Deng, Z., Jiang, Y., Wang, W. et al. (2014). Indium segregation measured in
InGaN quantum well layer. Sci. Rep. 4: 6734.
228 Liu, N.X., Wang, H.B., Liu, J.P. et al. (2006). Growth of p-GaN at low tem-
perature and its properties as light emitting diodes. Wuli Xuebao/Acta Phys.
Sin. 55 (3): 1424–1429.
229 Chuo, C.C., Chang, M.N., Pan, F.M. et al. (2002). Effect of composition inho-
mogeneity on the photoluminescence of InGaN/GaN multiple quantum wells
upon thermal annealing. Appl. Phys. Lett. 80 (7): 1138–1140.
230 Li, Z., Liu, J., Feng, M. et al. (2013). Suppression of thermal degradation
of InGaN/GaN quantum wells in green laser diode structures during the
epitaxial growth. Appl. Phys. Lett. 103 (15): 152109.
231 Hardy, M.T., Wu, F., Huang, C.Y. et al. (2014). Impact of p-GaN thermal
damage and barrier composition on semipolar green laser diodes. IEEE
Photonics Technol. Lett. 26 (1): 43–46.
232 Hoffmann, V., Mogilatenko, A., Zeimer, U. et al. (2015). In-situ observation
of InGaN quantum well decomposition during growth of laser diodes. Cryst.
Res. Technol. 50 (6): 499–503.
233 Liu, J., Liang, H., Zheng, X. et al. (2017). Degradation mechanism of crys-
talline quality and luminescence in In0.42 Ga0.58 N/GaN double heterostruc-
tures with porous InGaN layer. J. Phys. Chem. C 121 (33): 18095–18101.
234 Tessarek, C., Figge, S., Aschenbrenner, T. et al. (2011). Strong phase separa-
tion of strained Inx Ga1-x N layers due to spinodal and binodal decomposition:
formation of stable quantum dots. Phys. Rev. B 83 (11): 115316.
235 Haller, C., Carlin, J.F., Jacopin, G. et al. (2018). GaN surface as the source of
non-radiative defects in InGaN/GaN quantum wells. Appl. Phys. Lett. 113
(11): 111106.
236 Haller, C., Carlin, J.F., Jacopin, G. et al. (2017). Burying non-radiative defects
in InGaN underlayer to increase InGaN/GaN quantum well efficiency. Appl.
Phys. Lett. 111 (26): 262101.
237 Oh, M.S., Kwon, M.K., Park, I.K. et al. (2006). Improvement of green LED
by growing p-GaN on In0.25 GaN/GaN MQWs at low temperature. J. Cryst.
Growth 289 (1): 107–112.
238 Moon, Y.T., Kim, D.J., Song, K.M. et al. (2001). Effects of thermal and hydro-
gen treatment on indium segregation in InGaN/GaN multiple quantum
wells. J. Appl. Phys. 89 (11): 6514–6518.
239 Ryou, J.H., Lee, W., Limb, J. et al. (2008). Control of quantum-confined Stark
effect in InGaNGaN multiple quantum well active region by p-type layer
for III-nitride-based visible light emitting diodes. Appl. Phys. Lett. 92 (10):
101113.
240 Hikosaka, T., Shioda, T., Harada, Y. et al. (2011). Impact of InGaN growth
conditions on structural stability under high temperature process in
InGaN/GaN multiple quantum wells. Phys. Status Solidi C 8 (7–8):
2016–2018.
241 Zhao, Y., Wu, F., Huang, C.Y. et al. (2013). Suppressing void defects in
long wavelength semipolar (2021) InGaN quantum wells by growth rate
optimization. Appl. Phys. Lett. 102 (9): 091905.
242 Yang, J., Zhao, D.G., Jiang, D.S. et al. (2018). Improvement of thermal stabil-
ity of InGaN/GaN multiple-quantum-well by reducing the density of thread-
ing dislocations. Opt. Mater. 85: 14–17.
243 Ishikawa, H., Nakada, N., Mori, M. et al. (2001). Suppression of GaInN/GaN
multi-quantum-well decomposition during growth of light-emitting-diode
structure. Jpn. J. Appl. Phys. Lett. 40 (11 A): L1170–L1172.
244 Zhou, K., Ikeda, M., Liu, J. et al. (2017). Thermal degradation of InGaN/GaN
quantum wells in blue laser diode structure during the epitaxial growth. In:
International Conference on Optoelectronics and Microelectronics Technol-
ogy and Application, vol. 10244 (eds. Y. Su, C. Xie, S. Yu, et al.), 102441X.
International Society for Optics and Photonics, SPIE.
99
GaN-Based HEMTs for Millimeter-wave Applications

Kathia Harrouche and Farid Medjdoub
Institut d’électronique de microélectronique et de nanotechnologie (IEMN-CNRS), Avenue Poincaré,
CS 60069-59652 Villeneuve d’Ascq, France
3.1 Introduction
Today, the interest for millimeter-wave (mmW) band (30–300 GHz) is
continuously increasing because of its reduced wavelength and wide fre-
quency bands, which enable smaller components with improved performances.
Wireless communication systems are extending to higher frequencies as system
designers need high bandwidth for many emerging applications. However,
several challenges need to be overcome in order to use the mmW spectrum
successfully. Monolithic microwave integrated circuits (MMICs) based on
III–V semiconductors are the key to meet the requirements of mmW range.
The targeted performances required for high-frequency MMICs include the
combination of high power/high efficiency, compact size, and low cost. Gallium
nitride (GaN) is one of the most promising semiconductors in this frame.
In this context, this chapter focuses on GaN-based devices for mmW applica-
tions. Targeted applications including high-power amplifiers (PAs), broadband
amplifiers, and fifth-generation (5G) wireless networks are introduced. Various
GaN-based material designs for mmW applications are described, showing the
advantages and limitations for high-frequency applications. The device design
and fabrication of mmW GaN devices are analyzed. Finally, an overview of MMIC
power amplifiers is reported.
3.2 Targeted Applications for GaN Millimeter-wave

Devices
Recent improvements in GaN HEMT (high electron mobility transistor) devices
have allowed the demonstration of a variety of next-generation mmW circuits.
As introduced in Chapter 1, with its wide bandgap, high saturated electron veloc-
ity, and higher breakdown voltage, GaN technology is a very promising candidate
for realizing circuits with high power, high efficiency, and wide operational band-
width. These advantages are attractive not only for compact solid-state PAs but

100 3 GaN-Based HEMTs for Millimeter-wave Applications
Base station Today

2.0 SATCOM
Millitary
Investment (billion $)
1.5
1.0
Arrival
of 5G
0.5
2016 2017 2018 2019 2020 2021 2022
Figure 3.1 Provisional investments on GaN up to 2022 for several types of applications.
also for the future transmitter design of 5G cellular communications. Figure 3.1
represents the provisional investments on GaN until 2022 for several types of
applications. GaN technology investment will continue to grow over time until
reaching US$2.5 billion for the year 2022. Such growth is explained by the fact
that GaN will play a major role in many applications and is considered as a strate-
gic technology.
3.2.1 High-Power Amplification

Recent and ongoing progress in semiconductor device fabrication and pro-
cessing technology has pushed the limits of MMICs for mmW frequencies.
MMIC components operating at these frequencies will be used to improve
the sensitivity and performance of radiometers, receivers for communication
systems, and transceivers for radar instruments. The first MMIC was reported
in 1976 [1], and progress in output power, gain, and operating frequency are
continually being achieved afterward. Recent literature has demonstrated that
silicon (Si)-based MMICs, such as complementary metal oxide semiconductor
(CMOS) Si-on-insulator (SOI)-stacked MMICs, and power-combining Si ger-
manium (SiGe) MMICs can achieve relatively high power at high frequencies.
Consequently, they can be applied to high-power applications requiring a
few watts below Ka-band. However, at higher frequencies, Si-based MMICs
are limited to deliver the necessary output power compared to GaN-based
devices. High-power amplifiers have been largely developed in the mmW range
using other technologies based on III–V materials such as gallium arsenide
(GaAs) [2–4] and indium phosphide (InP) [5–7] HEMTs. Despite the excellent
achievements, InP- and GaAs-based mmW power amplifiers are also limited
in saturated power levels because of the low breakdown voltage and related
drain bias operation. Advanced GaN HEMTs have demonstrated high power
in the mmW range surpassing any other technologies by a factor of 5–10 [8].
The superior mechanical robustness and the capability to operate at higher
temperatures as compared to Si, GaAs, and InP materials are additional benefits
of GaN [8]. GaN MMICs revolutionize the field of mmW solid-state power
amplifiers (SSPAs) and enable new applications that were previously not practical
because of the limited output power of SSPAs and large size of traveling wave
tube amplifiers (TWTAs).
PAs based on GaN HEMT technology operating in the mmW frequency
range have been demonstrated up to the W-band. Figure 3.2 shows the main
semiconductor technologies and their limits in terms of output power and
operating frequency. The State-of-the-art power-level amplifiers have been
reported with about 10 and 3 W at Ka-band [9–11] and W-band [8, 12–20],
respectively, under continuous-wave (CW) conditions. As expected, the output
power decreases at higher frequencies because of device scaling. Indeed, the
gate length and the lateral device dimensions (gate–drain distance) are key
parameters for high-frequency operation. Furthermore, in order to reach high
output power density, a high drain voltage is mandatory, which is typically
inversely proportional to the device downscaling. Nevertheless, spatial power
combining enabled 5.2 W output power W-band GaN MMIC [21]. GaN is well
known for its high output impedance and low output capacitance as well as
high breakdown voltage. These features lead to high efficiency and high-power
MMIC performances.
mmW MMIC amplifiers

101
Output power (W)
100
10–1
GaN HEMTs
GaAs HEMTs
InP HEMTs
10–2
20 30 40 50 60 70 80 90 100 110
Frequency (GHz)
Figure 3.2 Output power of MMIC amplifiers under CW operation based on various
semiconductor transistor technologies.
3.2.2 Broadband Amplifiers

In recent years, the application of GaN-based mmW MMIC has pushed the
state-of-the-art performance of power amplifiers to the next level. For many
applications such as instrumentation or communication systems that require
the integration of several services with reduced number of components and
size, radio frequency (RF) power amplifiers covering a wide bandwidth would
be highly beneficial. To date, systems covering a wide frequency range require
multiple narrowband power amplifiers. These amplifiers are connected by
means of switches or triplexers. In either case, losses are introduced by the
additional circuitry, and therefore, such a system is not favorable. To reduce
costs and system complexity, it is necessary to replace multiple amplifiers by a
single-broadband power amplifier covering multiple bands. Wideband power
amplifier MMICs based on GaN are thus key components that are employed in
numerous applications such as military and wireless communication. For 5G,
GaN MMIC PAs are expected to be widely deployed in cellular base stations in
order to reduce size and improve system integration. Consequently, it is essential
for MMIC implementation to develop low loss and compact circuits in order to
ensure high efficiency and broad bandwidth.
The performance evaluation of power amplifiers is based on five key parame-
ters: efficiency, output power, linearity, gain, and bandwidth. The improvement
of power amplifiers has generated significant research and development in recent
years. The optimum operation class of power amplifiers depends on the targeted
linearity, efficiency, and output power. In the conventional operation classes (A,
B, AB, and C), the power amplifier operates like a voltage-controlled current
source. Nevertheless, another aspect that has recently attracted a lot of attention
is the enhancement of the amplifier efficiency using the so-called switch-mode
concepts such as class D, E, and F with the aim of reaching drain efficiencies of
70% and beyond [22].
The state-of-the-art broadband high-power mmW GaN MMIC is summarized
in Figure 3.3. These results indicate the great potential of GaN-based PAs to
increase solid-state power levels while maintaining a wide bandwidth. Broadband
GaN MMICs have been reported up to 140 GHz with an output power ranging
from 10 W over 32–38 GHz [10] to 47 mW over 90–140 GHz [27]. Although these
MMICs have produced attractive power levels, the highest power level is associ-
ated with a much narrower bandwidth as expected. However, by using on-chip
traveling-wave power combiner circuit techniques, several watts output power
using GaN MMICs have been reported by Quinstar with a bandwidth approach-
ing the entire W-band [28]. Another reported technology enabling wide band-
width amplifiers with high gain is the nonuniform distributed power amplifiers
(NDPAs). In this case, the amplifier uses dual-gate HEMTs in the driver stage of
an NDPA, which boost the gain of the overall amplifier at a wide bandwidth. IAF
reported NDPA MMICs covering a frequency range from 6 to 37 GHz and from 8
to 42 GHz with an output power of 1 W and 500 mW with corresponding power
gains above 11 and 8 dB, respectively [29, 30].
Broadband MMIC amplifiers

103
IAF 2012 Schellenberg et al. [25]
IAF 2014 Darwish et al. [26]
Thome et al. [23] Quinstar 2015
Neininger et al. [24] Qorvo 2016
102
Raytheon 2018
Output power (W)
Gain = 27 dB
101
Gain = 22 dB
Gain = 13 dB Gain = 17.5 dB
Gain = 11 dB Gain = 14.3 dB

100
Gain = 8 dB
Gain = 19.4 dB
10–1
Gain = 10 dB
10–2
0 20 40 60 80 100 120 140
Frequency (GHz)
Figure 3.3 Output power of broadband MMIC amplifiers [10, 23–30].
3.2.3 5G
The ever-increasing data rate and the number of connections for mobile com-
munication offer exciting user experiences in our everyday lives. Currently, the
wireless communication frontier is shifting from the current fourth generation
(4G) to the forthcoming 5G. Major international communication companies and
manufacturers are all competing to demonstrate 5G capabilities and features,
while simultaneously paving the way for mmW technology. The broadband radio
access and wireless networks underline the pioneering aspects of 5G, not only
for the telecommunication industry but also for a wide range of vertical sec-
tors, including robotics, automotive, factory automation, health care, and edu-
cation. Although the expected features and use cases for 5G are extensive and
diverse, the start of 5G deployment will likely address only a few of the high-
lighted use cases through three scenarios: ultrareliable low-latency communica-
tions (uRLLC), enhanced mobile broadband (eMBB), and massive machine-type
communications (mMTC) as illustrated in Figure 3.4. Under the 5G umbrella,
these scenarios have quite different system-level performance requirements such
as latency, mobility, number of users, and data rate.
The key performance indicators (KPIs) for evaluating 5G wireless networks
include peak data rate, user-experienced data rate, latency, mobility, connec-
tion density, energy and spectrum efficiency, and area traffic capacity [31].
Table 3.1 summarizes the network features of human-oriented 4G and internet
Enhanced mobile broadband
Gigabytes in a second
3D video, UHD screens
Work and play in the cloud

Smart home/building
Augmented reality
Industry automation
Voice
Self-driving car
Smart city Mission critical
applications,
e.g. e-health
Massive machine-type Ultrareliable and low latency

communications communications
Figure 3.4 Some usage scenarios proposed by international mobile telecommunications

(IMT)-2020 [31].
of everything (IoE)-oriented 5G. The targets include the following:

– A peak data rate of at least 20 Gb/s, which is 10 times that of 4G for some
special scenarios such as mmW network backhaul.
– A user-experienced data rate of 0.1 Gb/s, which is 10 times that of 4G. In
hotspot cases, the user-experienced data rate is expected to reach higher val-
ues (e.g. 1 Gb/s indoor).
– An energy efficiency of three times and a spectrum efficiency of 10–100 times
higher compared to 4G.
– An over-the-air latency of 1 ms and high mobility up to 500 km/h. This will
provide acceptable quality of service (QoS) for uRLLC scenarios such as auto-
mated driving.
– Ten times the connection density of 4G. This will reach up to 106 devices/km2 ,
for example, mMTC scenarios and area traffic capacity of up to 10 Mb/s/m2 .
3.2.3.1 GaN for 5G

Compared with the current 4G LTE (long term evolution) network, the future 5G
network for eMBB targets a 20 Gb/s peak data rate, which represents a 10 times
improvement. New waveforms, along with multiple-input–multiple-output
(MIMO), beamforming, and mmW technologies, are considered as key features
for 5G to enable dramatic network performance in terms of both energy and
spectrum efficiency. MIMO at mmW frequencies presents major challenges on
the hardware design because of the requirements of mmW front-end modules
in order to favor beamforming functionality. This poses news challenges in
the development of high-performance PAs for 5G network. Advanced PA
Table 3.1 The network features of 4G and 5G.

4G 5G
Usage scenarios • MBB • eMBB
• uRLLC
• mMTC
Applications • High-definition videos • VR/AR/360∘
• Voice videos
• Mobile TV • UHD videos
• Mobile internet • V2X
• Mobile pay • IoT
• Smart city/
factory/home
• Telemedicine
• Wearable devices
KPI Peak data rate 100 Mb/s 20 Gb/s
Experienced data rate 10 Mb/s 0.1 Gb/s
Spectrum efficiency 1× 3× that of 4G
Network energy efficiency 1× 10–100× that of 4G
Area traffic capacity (Mb/s/m2 ) 0.1 10
Connection density (devices/km2 ) 105 106
Latency (ms) 10 1
Mobility (km/h) 350 500
Technologies • OFDM • mm-Wave
• MIMO communications
• Turbo code • Massive MIMO
• Carrier aggregation • LDPC and polar
• HetNet codes
• D2D communications • Flexible frame
• Unlicensed spectrum structure
• Ultradense
networks
• Cloud/fog/
edge computing
• SDN/NFV/
network slicing
VR, virtual reality; AR, augmented reality; UHD, ultrahigh definition; V2X, vehicle to everything;
IoT, internet of thing; OFDM, orthogonal frequency division multiplexing; MIMO,
multiple-input–multiple-output; HetNet, heterogeneous network; D2D, device to device; LDPC,
low-density parity check codes; SDN, software-defined networking; and NFV, network function
virtualization.
architectures have evolved with the development of GaN technology to meet

ever-greater system-level requirements especially in terms of efficiency, power
levels, and modulation bandwidth. GaN will surpass conventional semicon-
ductor materials for 5G network applications, requiring higher frequencies,
tight integration, and minimal implementation cost while operating under
harsh environments. These features result in high-efficiency and high-power PA
performance over a large frequency range, leading to low-cost, large bandwidth,
and small-size base station system [32].
3.2.3.2 GaN Base Station PAs

To meet the complex requirements of PAs for 5G applications, the device
technology selection and the circuit configuration are critical. Several key
directions for 5G have emerged. In turn, silicon still wins in the sub 6 GHz,
but at higher frequencies, GaN is increasingly attractive. On the other hand,
the critical allocation of spectrum will dictate the design and implementation
of transceiver hardware. However, the available spectrum for high data rates is
limited to the microwave range. That is why, mmW frequencies are necessary to
extend the current 4G frequency bands. Recently, various GaN PAs have been
reported in this frame. Modulated signals with higher bandwidth and a more
complex scheme are used to realize greater data rate. Moreover, the demand
is increasing for efficiency under multiband, multimode operation. Various
approaches such as Doherty amplifiers, envelope tracking (ET) amplifiers, and
a digital transmitter based on GaN have been demonstrated [33–35]. While
mmW 5G is being developed, it will be first implemented on the sub-6 GHz
5G systems using the same MIMO beamforming techniques but at lower, more
technologically accessible frequencies [36]. A 5G communication network is
designed not only for spectrum bands below 6 GHz but also for high bands
above 24 GHz (mmW). A number of sub-6 GHz 5G MIMO systems have been
demonstrated at 3.3, 4.2, and 2.14 GHz [36, 37]. In the mmW, several GaN PAs
have been proposed. Table 3.2 summarizes some performance results of PAs at
different frequencies especially in Ka-band for 5G applications.
With the increasing need of data rates, modern communication networks use
high bandwidth modulation schemes with high power average ratio (PAR) but at
the expense of PA efficiency. At high PAR, the PA must be operated with signif-
icant back-off from its saturation point in order to achieve the required linearity
at the PA output. Furthermore, the PA maximum output power level must be
reduced so that the entire signal is within the linear region, thus decreasing the
efficiency [38]. A solution is the use of advanced PA architectures in order to
enhance the PA efficiency while maintaining high linearity. Examples would be
Doherty, ET, outphasing, and switch mode PAs based on class D and class S
amplifiers.
The Doherty PA technology, which uses active load modulation, is a promising
architecture to enhance efficiency and linearity at back-off operation. Doherty
amplifiers covering multiband frequency range using GaN technology have been
demonstrated with high efficiency up to Ka band [36, 38, 42, 43]. In addition, ET
PAs based on dynamic modulation are also competitive to increase PA efficiency.
In this case, to maintain high efficiency in the back-off region, the envelope
amplifiers dynamically modulate the supply voltage of the RF PA. Both advanced
PA architectures will continue to dominate RF and mmW applications requiring
multiband and high efficiency for 5G applications. Nevertheless, bandwidth and
linearity are still one of the most important limitations in the Doherty and ET PA
techniques. To overcome these issues, several techniques have been proposed,
including digital predistortion (DPD) technique [38, 44]. The other recent
progress in this frame is the linear amplification using nonlinear component
(LINC) outphasing PA architecture, which is attractive for 5G systems because of
its strong potential to provide high efficiency with a large peak to PAR signal [45].
Table 3.2 Performance comparison of different PAs for 5G applications [33, 36, 38–42].
References Type Size f (GHz) POUT (dBm) PAE (%) PAR (dBm) Gain (dB)
[33] PA 2.9 mm × 1.7 mm 26.5–29.5 36.9–38 17.9–23 NA NA

[36] Doherty PA 1.8 mm × 1.7 mm 28 36 51 NA 30
[39] Switching mode PA NA 28–39 24.3 59 NA 8.2
[40] PA 3.8 mm × 6.2 mm 26–28 43.3–41.6 19.8–13.2 NA NA
[41] HPA 3.4 mm × 3.3 mm 26.5–29 39 25 NA 21.1–24
[38] Doherty 3.4 mm × 2 mm 23 36.9 27 29.4 15.4
[42] Doherty PA 2.7 mm × 1.6 mm 27.5–29.5 35.6 25.5 NA NA
Despite the proposed approaches, there is a large room for improvement of

PA performances in order to satisfy such practical requirements of 5G wireless
communication network, such as mmW, high linearity, high output power,
large bandwidth, and compact size. GaN HEMTs are among the most suitable
devices for power amplifiers and will certainly play a major role as a broadband
technology for 5G wireless communication.
3.2.3.3 Moving Forward to 6G

The 5G network system has been defined as the key for IoE application. Research
efforts invested in mmW wireless communications and the success of 5G prelimi-
nary tests ensure that the commercialization and first deployments of 5G wireless
networks are expected to ramp up during 2020 [31]. The promised groundbreak-
ing 5G system operating essentially at a high frequency between 24 and 71 GHz
will solve the spectrum deficit in 4G cellular communication systems currently
limited to sub-6 GHz frequencies. Although the 5G systems that are being mar-
keted will support basic IoE application, the increasing number of new applica-
tions such as wireless backhaul, virtual reality (VR)/augmented reality (AR), and
space travel make it questionable whether they can satisfy emerging services and
newer applications that have not been conceived yet. This creates a motivation
toward sixth-generation (6G) networks.
Recent investigations have identified the key enabling technologies that might
define the 6G. Future 6G wireless communication will be implemented in about
10 years (2030) [46] with devices operating up to the terahertz range. The key
figure of merits for the evaluation of 6G wireless networks include a peak data
rate of 1 Tb/s, which is 100 times that of 5G, a latency of 10–100 μs, and an energy
efficiency of 10–100 times better than 5G. The level of 6G maturity reachable in
10 years by sub-terahertz can make this technology a powerful enabler.
3.3 GaN-based Material Designs for Millimeter-wave

Applications
Historically, advances of semiconductor technologies based on MMICs have led
to continuous performance improvements. GaN represents a leap ahead as com-
pared to other technologies for many RF applications. In early 1990s, GaN has
been recognized as a potential solution to satisfy the combination of high-power
and high-frequency requirements [47].
3.3.1 Intrinsic Characteristics and Comparison with Other Materials

for RF Devices
Figure 3.5 shows the material properties of widely used mmW semiconductors,
demonstrating advantages of GaN-based material system for high-frequency and
high-power applications. Because of its excellent properties, GaN technology is
recognized as a revolutionary material with a wider energy gap of 3.4 eV that
exceed three times those of InP, GaAs, and Si, thus allowing higher breakdown
Critical electric field MV/cm

GaN
GaAs
InP
Si
SiC
EGAP Saturation velocity

(eV) (107 cm2/s)
Electron mobility (103 cm2/V s)
Figure 3.5 Material properties for millimeter-wave semiconductors.
voltage and higher operating voltage. The high-saturated electronic velocity of

2.5 × 107 cm/s is another attractive characteristic of GaN. The electron velocity
is related to the current density; thus, GaN is able to produce high current at
high voltage. Because power is a function of voltage and current, the wide energy
gap and the high electron velocity allow for ideal high power devices. Moreover,
as described in Chapter 1, GaN-based heterostructures deliver a high electron
mobility of 1–2 × 103 cm2 /V s, which allows low on resistance. Therefore, high
power-added efficiency (PAE) can be achieved at high frequencies. GaN thermal
conductivity (in the range of 1.3 to 2.1 W/cm K) is much higher than GaAs and
InP. The thermal conductivity is a key factor directly related to the power dissipa-
tion from the device. The comparison of the various materials using the Johnson
figure of merit (JFoM) clearly demonstrates the superiority of GaN HEMTs over
its counterparts [48, 49].
In addition to the device performances, the associated reliability is crucial. For
short devices delivering high performances at high drain bias, the main limita-
tion is the device reliability due to a significant electric field peak and subsequent
high junction temperatures. It can be pointed out that the so-called field plates
have been successfully implemented in the microwave range to ensure high
device reliability by spreading the electric field peak. However, this approach
cannot really be used for mmW devices because of highly induced parasitic
capacitances degrading the gain. Assessing the reliability at an early stage of a
technology development is of prime importance. Tests are typically carried out
under specific operating and environmental conditions (such as temperature,
voltage, current, etc.) in a given time duration. The reliability and performance
under high-temperature conditions have been studied [50]. GaN mmW HEMTs
can provide stable device performance up to high junction temperature [51–56].
30 Room temperature
tbp = 50 °C tbp = 80 °C
60 1000 After 40 GHz RF stress at Tbp = 50 °C
tbp = 65 °C tbp = 120 °C
After 40 GHz RF stress at Tbp = 65 °C
28 tbp = 120 °C tbp = 140 °C 100 After 40 GHz RF stress at Tbp = 80 °C
Output power (dBm)
50 After 40 GHz RF stress at Tbp = 100 °C

ID (A/mm)
1
PAE (%)
40
PAE
POUT
100m
24
30 10m
22 0.11 × 100 µm2

1m
VDS = 20 V 20
PAE matching GD = 1.5 µm
Lg = 110 nm
100µ
40 GHz VDS = 10 V
20 10 10µ
0 5 10 15 20 25 –6 –4 –2 0 2
(a) Time (h) (b) VGS (V)
Figure 3.6 (a) Output power and PAE monitoring for 24 hours up to140 ∘ C and (b) transfer
characteristics before and after the stress. Source: Harrouche et al. 2019 [54]. https://ieeexplore
.ieee.org/stamp/stamp.jsp?tp=&arnumber=8894405. Licensed under CC-BY 4.0.
0.25 mm pHEMT
PD: ~650 mW/mm 0.15 mm pHEMT
PD: ~800 mW/mm 0.15 mm GaN HEMT
PD: ~2800 mW/mm
2.6 × 0.9 mm
82% size reduction
3.0 × 2.9 mm at 4× power density
4.3 × 3.0 mm
2014 GaN
2005 GaAs
2002 GaAs
Figure 3.7 MMICs comparison with different technologies.
Figure 3.6 shows an example of RF monitoring performed at 40 GHz on a

promising AlN/GaN HEMT structure with a gate length Lg = 110 nm and gate
to drain distance Lgd = 1.5 μm. The devices have been tested during 24 hours up
to a base-plate temperature of 140 ∘ C (corresponding to a junction temperature
above 250 ∘ C) by steps of 4 hours at V DS = 20 V. A high PAE (50%) with a POUT
around 3 W/mm at 40 GHz remains stable (Figure 3.6a) with no increase of gate
and drain leakage currents (Figure 3.6b).
GaN is well suited not only for high power and high frequency but also lead to
smaller and cheaper chip size. The GaN MMICs reported so far have more than
five times higher power density with smaller size than GaAs MMICs [15, 52, 57].
As illustrated in Figure 3.7, GaN MMICs can be reduced by 82% as compared to
GaAs pHEMT (pseudomorphic high electron mobility transistor) MMIC while
providing more than four times power density. As a result, GaN MMICs can
deliver higher efficiency because of the reduced on-chip combining losses both
at the MMIC and module levels. That is why, GaN MMICs will revolutionize the
field of mmW SSPAs and enable new applications that were previously not prac-
tical because of limited power of SSPAs or large size and high cost of TWTAs.
However, GaN HEMTs are still plagued by two important phenomena,

especially when reducing the device dimension: the trapping effects and the
self-heating, which can directly cause current collapse and kink effect, thus
reducing the device performances [58, 59]. Trapping effects that occur at dif-
ferent locations of the devices are mostly related to the crystalline imperfection
induced during growth and device processing as shown in Figure 3.8a. Surface
[60, 61] or buffer trapping [62] is generally dependent on electric field. Several
techniques are used to assess the trapping effects such as deep level transient
spectroscopy measurements [63, 64], temperature-dependent threshold voltage
analysis [65], or pulsed measurements [54, 63, 66]. As shown in Figure 3.8b,
pulsed I–V characteristics performed with quiescent drain voltages up to 25 V
at V GS = +2 V of AlN/GaN HEMT devices show rather strong trapping effects
as seen from the gate and drain lag because of the residual surface and buffer
traps.
The trapping/detrapping mechanisms induce electrical parasitic effects such
as current collapse and kink effect are shown in Figure 3.9a. Several investiga-
tions have demonstrated that current collapse effects are related to the presence
of traps and hot electron injection into the buffer layer under high electric field
[59]. It was also shown that the current collapse is attributed to trapping under
the gate and in the gate–drain access region using phototransient measurements
[67]. Another electrical parasitic effect due to the trapping mechanism is the kink
effect that increases the drain current, resulting in a shift of pinch-off voltage
toward more negative voltages. Several explanations have been suggested [68]:
the impact ionization and subsequent hole accumulation causing the change of
surface or channel/substrate interface, field-dependent trapping/detrapping in
deep levels [69], and a combined effect of impact ionization and deep levels that
induce a modification of surface states, buffer, or channel/substrate interface deep
levels by the generated holes [70]. Other studies have reported that kink effects
1.2 VGS = 2V
Passivation Trapped electrons
1.0
S G D
ID (A/mm)
Barrier 0.8
Depleted
0.6
GaN channel 2DEG
DC
Trapped 0.4 [0;0]
GaN buffer electrons t° = 23 °C Gate lag
0.2 Pulse width = 1 µs Drain lag 10 V
Drain lag 20 V
Substrate Duty cycle = 1%
Drain lag 25 V
0.0
0 2 4 6 8 10
(a) (b) VDS (V)
Figure 3.8 Schematic cross section of AlN/GaN HEMT, showing electron-trapping location
(a), pulsed ID –V DS characteristics with various quiescent bias points: cold point: V DS0 = 0 V,
V GS0 = 0 V, gate lag: V DS0 = 0 V, V GS0 = −6 V; and drain lag: V DS0 = 10–25 V, V GS0 = −6 V (b).
Source: Harrouche et al. 2019 [54]. https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&
arnumber=8894405. Licensed under CC-BY 4.0.
1.2 40 80
VGS from –4 V to +2 V GD 0.5 Pulsed mode
35 40 GHz CW mode
Step 1 V PAE 70
1.0
pOUT (W/mm), gain (dB)

30 60
Drain current (A/mm)
0.8
25 50
PAE (%)
0.6 Kink effect 20 40
Gain
15 30
0.4
Current collapse 10 20
0.2 pOUT
5 10
0.0 PAE Matching

0 0
0 5 10 10 15 20
(a) Drain voltage (V) (b) VDS (V)
Figure 3.9 I–V characteristics showing the current collapse and kink effects due to electron
trapping (a) and CW (circle) and pulsed (square) output power density, PAE, and small signal
gain as a function of V DS at 40 GHz of AlN/GaN HEMTs (b). Source: Harrouche et al. 2019 [54].
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8894405. Licensed under
CC-BY 4.0.
in GaN HEMTs are related to both impact ionization coupled with the presence
of slow traps in the epitaxial layers under the gate, possibly into the GaN buffer
[71, 72].
Figure 3.9b shows a comparison between large signal CW and pulsed mode
at 40 GHz of output power density, PAE, and small signal gain as a function of
V DS of AlN/GaN HEMTs. The gap in terms of performances between the CW
and pulsed mode, especially for the PAE, confirms the presence of traps within
these devices. That is why, the optimization of material quality and related process
technology are necessary in order to minimize the trapping effect phenomena.
Many efforts have been carried out to minimize the parasitic effects because of
electron trapping such as the following:
– The use of silicon nitride passivation (Si3 N4 ) to improve the gate lag
[58, 63, 73],
– The optimization of epitaxial growth conditions in order to suppress deep level
traps into the buffer layers [74],
– The use of gate field plates technology to spread the electric field in the vicinity
of the gate [75] or the use of an in situ SiN passivation reducing drastically the
surface states are key parameters to improve RF performances.
High-frequency operation requires aggressive device scaling to increase the
gain and the frequency performances of GaN HEMTs. Figure 3.10 shows a bench-
mark of PAE and POUT of GaN HEMTs as a function of frequency. Very high
efficiency up to 60% and POUT up to 8 W/mm at Q-band have been reported [54].
However, at higher frequency, the efficiency of GaN HEMTs is still limited mainly
because of an insufficient gain. At W-band, the highest PAE reported to date is
27.8% [76]. As shown in Figure 3.10a, the PAE decreases with the frequency, while
POUT remains above 8 W/mm (Figure 3.10b). The major current challenge for
mmW GaN-based devices is to maintain high PAE at high frequencies together
with strong robustness.
70 mmW GaN HEMTs 14 mmW GaN HEMTs
60 12
PA
E de
50 gra 10
d
POUT (W/mm)
atio
nw
PAE (%)
ith High POUT up to W-band

40 fre 8
qu
en
30 cy 6
20 4
10 2
0 0
0 10 20 30 40 50 60 70 80 90 100 110 0 10 20 30 40 50 60 70 80 90 100 110
(a) Frequency (GHz) (b) Frequency (GHz)
Figure 3.10 PAE (a) and associated output power density (POUT ) (b) of GaN HEMTs vs. the
frequency of operation.
Gate-to-drain distance = 0.5 µm

Gate-to-drain distance = 1.5 µm
3-Terminal breakdown voltage (V)
150 Gate-to-drain distance = 2.5 µm
100
50
0
0.004 0.006 0.008 0.010
1 / Lg (nm–1)
Figure 3.11 3-Terminal breakdown voltage of AlN/GaN HEMTs as a function of gate length
and gate-to-drain distance (0.5 and 1.5 μm).
On top of trapping enhancement and limited PAE, the device scaling can
directly affect the breakdown voltage (V BK ) because of the reduction of the
gate-to-drain distance [77]. Figure 3.11 shows the three-terminal breakdown
voltage as a function of Lg for different Lgd . The plot demonstrates that Lgd has
more impact on the breakdown voltage than Lg . Therefore, a high breakdown
voltage can still be maintained with short devices by using proper device designs.
Today, GaN HEMTs are the most attractive electronic devices for high-power
mmW applications owing to their intrinsic properties. High-power and
high-frequency GaN HEMT performances are continuously improving in the
mmW range. Nevertheless, the robustness and subsequent reliability remain
under investigation as both scaled materials and devices need to demonstrate
high stability, reproducibility, and uniformity.
3.3.2 Specific Material Systems for RF Devices

GaN HEMT epitaxial heterostructures based on III-nitride materials consist of
a wide bandgap barrier layer (AlN, InAlN, InAlGaN, or AlGaN), a GaN channel,
and GaN-based buffer layers grown on top of a substrate using metal organic
chemical vapor deposition (MOCVD) or molecular beam epitaxy (MBE).
GaN-based HEMTs are typically grown as follows from bottom to top:
– Nucleation layer: Typically based on AlN, which is essential in order to accom-
modate the lattice mismatch between the buffer and the substrate, enabling
high-quality GaN heteroepitaxy.
– GaN buffer layers: Should be of high quality to avoid deep trap levels and lead
to a high electron confinement into the channel by using a back barrier or by
incorporating acceptor-type dopants such as carbon (C) or iron (Fe) in order
to increase the resistivity.
– GaN channel: Generally undoped to allow high electron transport quality, the
channel thickness is part of a trade-off between the electron confinement and
the trapping effects.
– Barrier layer: AlGaN-based microwave devices and preferably ultrathin
Al-rich materials for mmW range in order to avoid gate recess, which can
compromise the device reliability. The thickness and alloy composition are
key parameters to mechanical strain and piezoelectric polarization, also
defining the two-dimensional gas (2DEG) density [78].
– SiN cap layer: Enables to prevent the stress relaxation of the heterointerface
Al-rich barrier/GaN channel and passivate the surface states in order to reduce
the DC (direct current) to RF dispersion [79, 80].
Figure 3.12a shows an example of an AlN/GaN HEMT cross section consist-
ing in 10 nm SiN cap layer, 3 nm AlN ultrathin barrier, 100 nm GaN channel, a
carbon-doped GaN buffer, and an AlN nucleation layer grown on SiC substrate.
Figure 3.12b reveals the transmission electron microscopy (TEM) of this struc-
ture by MOCVD grown.
The high spontaneous polarization between GaN and AlN leads to a significant
sheet carrier density [81], which combined to a gradient in alloy composition
at the barrier/channel interface yields an increasing amount of free carrier into
the barrier layer. This results in a strong electron scattering and a reduction
in the 2DEG mobility [82]. Figure 3.13 shows the electrical properties of GaN
HEMT structures based on different barrier layers grown on SiC substrate.
Room temperature Hall measurements show that the high sheet charge den-
sity of 1 × 1013 cm−2 obtained using an AlGaN barrier layer can be increased
to 1.9 × 1013 cm−2 for an AlN barrier layer. Moreover, the 2DEG mobility
decreases as expected with the increase of the sheet carrier density from more
than 2000 cm2 /V s for rather low Al content (<25%) AlGaN barriers to about
1000 cm2 /V s for AlN barrier layer.
As introduced in Chapters 1 and 2, an important aspect of GaN technology is
the substrate material. The choice of substrate depends on the available size, cost,
thermal conductivity, coefficient of thermal expansion (CTE), lattice mismatch,
and targeted applications. Because bulk GaN substrates are still unavailable on
10 nm SiN
S G G
3 nm AIN
100 nm GaN
2DEG
GaN
C-doped GaN
AlN NL
AlN
SiC
SiC
500 nm
(a) (b)
Figure 3.12 Cross section (a) and transmission electron microscopy (TEM) of the MOCVD
grown AlN/GaN/SiC with 3 nm AlN barrier (b).
Ns ≈ 1.9 × 1013 cm–2

µ ≈ 960 cm2/V s
Ns ≈ (1.2 – 1.7) × 1013 cm–2
µ ≈ 1300 cm2/V s
AIN
Ns ≈ (1.2 – 1.5) × 1013 cm–2
InAIN
µ ≈ 1900 cm2/V s
InAIGaN
Ns ≈ 1 × 1013 cm–2
µ ≈ 2200 cm2/V s
AIGaN
Figure 3.13 2DEG properties for different barrier layers.
large wafer diameter, GaN HEMTs are typically grown on SiC, Si, and sapphire.
However, sapphire has a low thermal conductivity with CTE and lattice constant,
showing a significant mismatch with GaN. Si substrates present many advan-
tages such as compatible processing with advanced CMOS, availability of large
wafer diameter, and an acceptable thermal conductivity. Nevertheless, it also suf-
fers from a large lattice mismatch with GaN as reflected by dislocations. Recent
reported data confirm that SiC is the most attractive substrate for GaN mmW
power devices because of its low lattice mismatch with GaN and high thermal
conductivity, enabling superior power operations that are not reachable by any
other materials. Figure 3.14 shows the TEM image of GaN grown on SiC, showing
a rather low dislocation density.
Figure 3.14 TEM image of

GaN grown on SiC.
GaN
AIN
0.5 µm
SiC
3.4 Device Design and Fabrication of Millimeter-wave

GaN Devices
3.4.1 Description of Key Processing Steps for Various GaN Device
Designs
3.4.1.1 Device Scaling for Millimeter Wave
GaN HEMT device scaling is required in order to enhance the high-frequency
performances. These optimizations are needed not only for the epitaxial structure
but also for some processing steps such as Ohmic contacts and gate module. The
electron transit time is reduced by using shorter gate lengths; however, a high
aspect ratio Lg /a (gate length/gate-to-channel distance) above 15 combined with
a high carrier density N s should be maintained [83]. This allows preventing short
channel effects while improving the F t /F max ratio that is defined by the following
equations:
gm Ft
Ft = ,F =
2π(Cgs + Cgd ) max 2 (Rg + Rds )1∕2
where g m , C gs , C gd , Rg , and Rds are transconductance, gate-to-source capaci-
tance, gate-to-drain capacitance, gate resistance, and drain-to-source resistance,
respectively.
To increase F max , each parameter including F t and parasitic elements such as Rg
and C gd needs to be optimized. Following many efforts from the research com-
munity, the best F t and F max reported to date are about 450 GHz (Figure 3.15).
These performances have been achieved through innovative device technologies
such as T-shaped gate [84], n+ -GaN Ohmic contact regrowth [85], self-aligned
gate process [86], and vertically scaled epitaxy [84].
3.4.1.2 T-shaped Gate Design

T-shaped gates are widely used for achieving both low gate resistance and para-
sitic capacitances when reducing the gate length [87]. Parasitic elements become
of paramount importance with shorter devices. As a matter of fact, the frequency
performances are currently mainly limited by the short channel effects and par-
asitic elements. Figure 3.16a shows the schematic cross section of a GaN HEMT,
600 W-band applications

Ka and Q-band applications
500
400
Fmax (GHz)
300
200
100
0
0 50 100 150 200 250 300 350 400 450 500
Ft (GHz)
Figure 3.15 Maximum oscillation frequency (F max ) as a function of cutoff frequency (F t ) of

GaN HEMTs.
illustrating the parasitic elements and the benefits of the T-gate with reduced gate
length, which basically enabled to achieve high RF performances [84]. Moreover,
the reduction of the gate length led to a significant increase of F t up to 450 GHz
with a gate length of 40 nm as shown in Figure 3.16b.
3.4.1.3 Advanced Ohmic Contact Technology
For mmW applications, low contact resistances are mandatory to minimize
total parasitic resistances and enhance the device performances. Nevertheless,
a low contact resistance is difficult to achieve because of the wide bandgap of
III-nitride HEMT heterostructures. Moreover, the high potential of barrier layer,
especially when using high Al content, may lead to high contact resistances [88].
Conventional planar Ohmic contacts are the simplest and low-cost methods,
which are formed by alloying a complex metal stacks followed by annealing
at an optimum temperature. However, high-temperature annealing generally
results in severe lateral Ohmic metal diffusion as well as rough metal surface.
Typical contact resistances are about 0.25–0.5 Ω mm, leading to high parasitic
resistances proportional to the device scaling. Several approaches have been
studied in order to successfully reduce the contact resistance such as recessed
Ohmic contacts enabling lower temperature and thus better contact defini-
tion [89, 90], ion implantation before the metallization [91], face treatment
[92], and regrown Ohmic contacts by the MBE technique [84]. Among these
approaches, regrown n+ -GaN Ohmic contacts is the most promising technology
to minimize parasitic access resistances. The Ohmic contact regrowth approach
enables direct contact between n+ -GaN and 2DEG layers that results in low
interface resistance [86]. Very low contact resistances are demonstrated [93, 94].
As expected, the combination of high electron mobility and high sheet carrier
density reduce the access resistances where the gate-to-source distance needs to
be reduced, leading to excellent device performances. This makes such a design
attractive for mmW applications.
High Hgf > 100 nm

T-shaped gate High Fmax
Reduction of
parasitic resistance Longer Lgd
and capacitance G
High breakdown
voltage
Hgf
Cp Rg Cp
S D
a Lg Lgd
Rgs Rgd
High Lg /a > 15 Short gate length

Reduction of short
High Ft
channel effects
(a)
500
W-band applications
450
Ka and Q-band applications
400
350
300
Ft (GHz)
250
200
150
100
50
0
0.00 0.01 0.02 0.03 0.04 0.05
(b) 1 / Lg (nm–1)
Figure 3.16 Cross section of T-gate GaN HEMT (a) and scaling of cutoff frequency (F t ) of GaN
HEMTs with the gate length (Lg ) (b).
3.4.1.4 N-polar GaN HEMTs

The conventional GaN HEMTs are generally developed with a Ga-polar orienta-
tion where the 2DEG is formed at the barrier layer/channel interface while the
polarization is inverted in N-polar heterostructures as shown in Figure 3.17. The
N-polar structure presents several advantages such as low contact resistances
because of the 2DEG induced above the heterointerface and an excellent electron
confinement owing to the back barrier under the channel.
Development of N-polar GaN devices and recent results reported by Univer-
sity of California, Santa Barbara (UCSB), show large potential for improvements
in the mmW power applications. Wienecke et al. reported an output power
density of 6.7 W/mm with a PAE of 14.4% at 94 GHz using N-polar GaN cap
Ga-polar N-polar
S G S G
SiN SiN
AlGaN cap
AlGaN top barrier
2DEG
GaN channel
GaN channel
AlGaN back barrier
Buffer GaN buffer
Substrate Substrate
Figure 3.17 Device schematic cross section of Ga-polar and N-polar HEMTs.
MISHEMT (metal insulating high electron mobility transistor) with a deep

gate recess in order to control the dispersion [95]. Using a self-aligned gate
to recess, Romanczyk et al. demonstrated a high breakdown voltage and high
power density of 8 W/mm associated with a PAE of 28% at 94 GHz [96]. It can
be noticed that the device robustness under harsh conditions still need to be
demonstrated with this configuration system.
3.4.1.5 AlN-Based Device Performances

Figure 3.18 shows the epitaxial structure and the schematic cross section of
Hughes Research Laboratory (HRL) 4G asymmetric self-aligned T-gate GaN
HEMTs. The gate length was 20 nm and the Ohmic contact regrowth tech-
nique is used, which allows direct contact of n+ -GaN to the 2DEG, enabling
extremely low interface resistance of 0.026 Ω mm. An ultrathin sub-5 nm AlN
barrier is used, which delivers high electron density nS , while maintaining a
high aspect ratio Lg /a. The asymmetric self-aligned gate with Lgs = 30 nm and
Lgd = 80 nm allows high speed with a breakdown voltage of 17 V. This results in
a high RF performance reflected by an F t of 454 GHz and an F max of 444 GHz
[97]. The same device with a gate length of 40 nm achieved high large signal
performances at W-band. The output power was 1.37 W associated with a
PAE of 27% at 83 GHz [19]. Furthermore, graded channel AlGaN/GaN HEMT
demonstrated a large potential to operate at mmW range. Thirty gigahertz large
signal measurements showed an excellent PAE of 72% with an output power of
2 W/mm [98].
Figure 3.19a represents a schematic cross section of an AlN/GaN HEMT
structure based on a 3 nm AlN barrier grown on GaN/SiC substrate. The HEMT
was capped with 10 nm in situ SiN layer used both as early passivation and to
reduce trapping effects. The T-gate length shown in Figure 3.19b was 110 nm and
G
SiN
2.5 nm GaN cap

Lgd Lgd
3.5 nm AIN top barrier S D
20 nm GaN channel n+ -GaN
Al0.08Ga0.92N back barrier
SiC substrate
(a) (b)
Figure 3.18 Epitaxial structure (a) and schematic cross section of the AlN/GaN HEMT HRL
Gen-IV (b).
tu
D
SiN ex si
G SiN in situ
S
barrier
3 nm AIN
l
N channe
100 nm Ga Lgs Lgd
ffer
C-dop ed GaN bu
eation layer
AIN nucl
rate
SiC subst
(b)
(a)
Figure 3.19 Schematic cross section of an AlN/GaN HEMT (a) and focused ion beam (FIB) view
of a 110 nm T-gate (b).
three different Lgd of 0.5, 1.5, and 2.5 μm yielding a high breakdown voltage of 50,
100, and 130 V, respectively. F t /F max of 63/300 GHz are achieved at V ds = 20 V
for Lg = 110 nm and Lgd = 0.5 μm. The F t /F max ratio close to 5 is attributed to
the highly favorable aspect ratio: gate length/gate-to-channel distance (>25).
Large signal characterizations have been carried out in the Q-band and W-band.
Figure 3.20a shows typical pulsed power performances at 40 GHz of a 2 × 50 μm
transistor with Lgd = 1.5 μm, a saturated output power of 5.3 W/mm with a
peak PAE above 60% at V ds = 30 V, and around 50% with a saturated output
power of 8.3 W/mm at V ds = 40 V. Despite the residual trapping effects, a high
PAE around 50% up to V ds = 40 V is obtained in the CW mode as shown in
Figure 3.20b. Furthermore, CW large signal characterization at 94 GHz have
then been performed on the same devices. A high output power of 4 W/mm is
obtained with a PAE of 14.3% at V ds = 20 V (Figure 3.21) [54].
30 60
50 Pulsed mode 60
40 GHz 25 CW mode 40 GHz
40 VDS = 10 V 50
POUT (dBm), gain (dB)

50 20 Gain
VDS = 20 V
30 15
VDS = 30 V 40
40 10
VDS = 40 V*Gain
PAE (%)
PAE (%)
20 POUT
5 30
30
10 0
0 20 –5 VDS = 10 V 20
POUT –10 VDS = 20 V
–10 10 10
–15 VDS = 30 V
PAE
–20 PAE matching –20 PAE PAE matching
0 0
–10 0 10 20 –10 0 10 20
(a) Pinj (dBm) (b)
Pinj (dBm)
Figure 3.20 Pulsed (a) and CW (b) power performances of a 2 × 50 μm AlN/GaN HEMT
(Lg = 110 nm, GD = 1.5 μm) at V DS = 10, 20, 30, and 40 V*. * denotes power matching only.
25 20
CW mode 4 W/mm
20 94 GHz
VDS = 20 V
15
14.3% 15
10 Gain
PAE (%)
pOUT 10
0
–5
–10 5
–15
–20 PAE
0
–10 0 10 20
Pinj (dBm)
Figure 3.21 CW power performances of an AlN/GaN HEMT 2 × 25 μm with Lg = 110 nm for

GD = 0.5 μm at V DS = 20 V.
3.4.1.6 InAlGaN-Based Device Performances

Figure 3.22 shows the cross section of Fujitsu’s 80 nm gate length InAlGaN/GaN
HEMT technology. InAlGaN barrier was used to avoid the formation of surface
pits, thus reducing the gate leakage current (I g ) as compared to the ternary InAlN,
which is required to improve the breakdown voltage and device reliability [53]. In
addition, InAlGaN barrier is well known for its high 2DEG density combined to a
high electron mobility. A double SiN passivation layer has been used to eliminate
the current collapse. The offset overhanging gate was adopted to decrease the
electric field without degrading high-frequency performances. Load-pull mea-
surements at 96 GHz show a high output power of 3.0 W/mm at V ds = 20 V under
pulsed mode operation. Besides, an improved 80 nm gate length InAlGaN/GaN
HEMT demonstrated high power operation in the wide frequency range from
G Figure 3.22 Device cross section of an

Double-layer SiN InAlGaN/GaN HEMT.
S D
InAlGaN barrier
AlN spacer
GaN channel
Buffer
SiC substrate
S- to W-band, using regrown n+ -GaN contact layer, an InGaN back barrier to

reduce off-state drain leakage current, and an AlGaN spacer layer to realize high
2DEG close to 2 × 1013 cm−2 . Moreover, a diamond heat spreader was introduced
in order to decrease the thermal resistance and further improve the output power
density. The maximum pulsed output power density of this InAlGaN/GaN MMIC
amplifier was 4.5 W/mm at 94 GHz [99]. It is intended that this power amplifier
can be used in wireless backhaul communications.
3.4.2 State-of-the-art Millimeter-wave GaN Transistors

Figure 3.23 shows a benchmark of PAE as a function of POUT for GaN HEMTs
in the mmW range up to W-band. In 2005, T. Palacios et al. reported a record
Q-band POUT of 10.5 W/mm with a PAE of 33% at 40 GHz using AlGaN/GaN
HEMT technology with Lg = 160 nm, where the PAE was limited mainly because
of the linear gain of 6 dB [100]. J.S. Moon et al. reported gate-recessed and
field-plated AlGaN/GaN HEMT at 30 GHz with a POUT of 5.7 W/mm associated
with a PAE of 45% [101]. Also at 40 GHz, F. Medjdoub et al. demonstrated in
pulsed mode AlN/GaN HEMT an outstanding PAE of 65% at V DS = 10 V with
a high POUT of 8.3 W/mm at V DS = 40 V [54]. Hence, compared to other tech-
nologies, much higher efficiency and POUT can be achieved with GaN HEMTs
up to W-band and certainly beyond. Many efforts have been made in order to
satisfy W-band requirements such as a short gate length (deep sub-100 nm),
high F t /F max , and low contact resistances. Micovic et al. from HRL reported the
first W-band GaN transistors with high performance in 2006 (14% efficiency
and 2.1 W/mm POUT ) [102]. Mikiyama et al. from Fujitsu labs developed 80 nm
Schottky-gate InAlGaN/GaN HEMT, which provides more than 3 W/mm high
power density at 96 GHz [53]. Recently, F. Medjdoub et al. demonstrated a
state-of-the-art POUT of 4 W/mm at 94 GHz associated with a PAE of 14% using
AlN/GaN HEMT technology [54]. Although these conventional GaN HEMT
100
mmW GaN HEMTs (CW mode)
90
Cu
80 rren Ga
tG N
aN tar
lim get
70 it
60
PAE (%)
50
40
30
20
10 Ka and Q-band
W-band
0
0 2 4 6 8 10 12 14
POUT (W/mm)
Figure 3.23 Benchmark of PAE as a function of POUT for mmW GaN HEMTs.
structures are developed on Ga-polar heterostructures, recent research from

UCSB has shown the emergence of N-polar heterostructures, providing 28%
efficiency with a POUT of 8 W/mm at 94 GHz [96]. It can be pointed out that the
optimum technology choice is based not only on performances but also on the
device reliability under harsh conditions. This will determine in the near future,
where the GaN-based structure will be best suited for mmW applications.
3.5 Overview of MMIC Power Amplifiers

3.5.1 MMIC Technology Using III-N Devices
3.5.1.1 III–V Material-Based MMIC Technology
MMICs, based on III–V semiconductor technology, are one of the key building
blocks to meeting the requirements for mmW applications. Monolithic means
that the full fabricated circuit is built on a single piece of a semiconductor mate-
rial such as GaAs, InP, SiGe, or GaN, resulting in highly integrated and com-
pact devices. Although high-performance mmW GaAs- and InP-based MMICs
have been reported so far, the increasing bandwidth and power requirements for
mmW applications are limited with these traditional semiconductor technolo-
gies. GaN is a potential solution to satisfy the RF mmW transmitter require-
ments for communications and SATCOM applications [47]. For instance, GaN
devices are being used in harsh environment receiver applications owing to its
low-noise performances and robustness. Therefore, GaN-based devices can cover
any MMIC functions such as mixing, low-noise amplification, power amplifi-
cation, and high-frequency switching. Figure 3.24 shows an example of typical
TX Power
Up-converter TX filter
amplifier
Antenna
RX Down- Low-noise
RX filter
converter amplifier
Figure 3.24 Schematic example of RF transmit/receive in communication base station.
elements in the RF transmit and receive modules in wireless communications.

There are two key components inside the transmit/receive module, which are the
PAs and low-noise amplifiers (LNAs). In addition to these components, there are
different transmit/receive functions integrated in the mmW modules, including
the oscillators, filters, linear mixers, converters, and RF switches [103].
Millimeter-wave MMIC systems need to deliver not only high perfor-
mances but also low cost and compact size. Along with active devices,
standard passive components such as metal–insulator–metal (MIM) capacitors,
metal-evaporated thin film resistors, bulk resistors, and via holes are used for
the circuit realization [104]. A number of foundries use specific techniques for
the MMIC implementation such as grounded coplanar waveguide (GCPW),
microstrip transmission line technology, and Cu damascene multilayer process
for 3D interconnection. Microstrip transmission lines are mostly used for
MMICs and become more critical at higher frequency [105]. Cu damascene
multilayer process for 3D interconnection provides high power, high current,
and low loss passive component [106]. The GCPW allows for low-inductive
ground connection [107]. Moreover, coplanar waveguide (CPW) enables lower
dispersion than microstrip technology for broadband applications as well as
simple and cheaper fabrication because of the absence of holes and backside
processing, which is beneficial for mmW operation.
3.5.1.2 Power Amplifiers

Power amplifiers and circuit design are extensively investigated for mmW appli-
cations. The main challenges for next-generation power MMICs are the achieve-
ment of a high output power up to at least 100 GHz combined with high PAE in
order to reduce the power consumption.
Class A, A/B, and C for Power Amplifier GaN MMICs Power amplifiers are classified
as linear class A and nonlinear A/B and C for analog design. Each class has its
advantages and drawbacks as they typically generate trade-offs [103]. Class A
operation is the main class to provide high power density with better linearity
but with a lower efficiency. Indeed, in class A, the quiescent drain current is fixed
to half of the maximum drain current, leading to high power density and therefore
significant power consumption [108]. Class A/B operation is quite popular. The
quiescent drain current is set to the optimum value corresponding to a trade-off
between the linearity and efficiency. Finally, the class C operation is also limited
by the linearity, and it is used typically for narrowband applications, in driver

amplifier [109], and in rectifier mode to enhance efficiency [110].
Class D, E, and F for Switch Amplifier GaN MMICs Class D, E, and F operations are
used in switch mode GaN amplifiers in order to improve the efficiency. Class D
operation is typically used for amplifiers operating at low frequencies. Class E
amplifiers are designed to reduce the power loss during RF switching. The ideal
class E amplifier is a switched-mode circuit that combines zero dissipated power
with zero voltage switching and zero derivative of voltage switching [111]. Class
E corresponds to a highly efficient tuned switching power amplifier. In class F
amplifiers, a harmonic tune technique was developed to further improve the per-
formance beyond those of class A and A/B designs [112, 113]. Therefore, class F
amplifiers can operate at higher frequencies but are limited by the complexity of
the circuit and the tuning requirements.
3.5.1.3 Low-Noise Amplifier

The noise performance of the first amplifier stage affects the receiver noise
figure (NF) and thus its sensitivity. The aim of an LNA is to amplify the desired
small powered signal from the antenna while minimizing the degradation of the
signal-to-noise ratio. An LNA is located at the first stage in a RF receiver module
and is a key element as its noise performance dominates in a cascaded RF receiver
system. The figure of merits of an LNA includes high gain (typically >20 dB)
and a low NF (<2 dB) for higher linearity. GaN is also a promising candidate for
reaching this overall figure of merits. They can provide a good sensitivity at low
power input and a reasonable gain, thus a good linearity at high power levels.
Several LNAs based on GaN operating at various frequency have been reported.
For Ka-band (27–40 GHz), an NF as low as 1 dB measured at 37 GHz with
24 dB gain using GaN HEMT T4-A generation of HRL technology [114]. In the
frequency band (100 MHz to 45 GHz), a low-noise gate termination-less cascode
distributed amplifier GaN MMICs demonstrated 10 dB gain and 1.6–3 dB of NF
from 130 MHz to 26 GHz [115]. At high frequency, a highly linear LNA MMIC
using AlGaN/GaN HEMT technology reported 25 dB gain and 5.6 NF at 84 GHz
[116]. The required NFs of the LNAs are in the range of 1–5 dB. GaN-based
LNA designs reported in the literature deliver lower NF in comparison to other
technologies based on III–V materials, which proves that GaN is promising for
next-generation robust receiver designs.
3.5.2 MMIC Examples from Ka-band to D-band Frequencies

Figure 3.25 provides examples of state-of-the art mmW GaN MMICs with dif-
ferent designs. Figure 3.25a shows an image of three-stage Ka-band design [11]
with a chip size of 1.74 × 3.24 mm2 . The MMIC technology is based on a 0.15 μm
gate length AlGaN/GaN HEMTs fabricated on SiC. The measured results under
CW operation for the balanced three stages demonstrate 9.5–11 W output
power and 26–30% associated PAE between 28 and 31 GHz. For a single-ended
amplifier type, the measurement results show an output power of 5.8–6.4 W and
28–34% of PAE for a similar frequency range. Figure 3.25b illustrates a Ka-band
(a) (b)
(c) (d)
Figure 3.25 Examples of millimeter-wave MMICs at different frequencies: (a, b) Ka-band,

(c) W-band, and (d) D-band.
0.15 μm AlGaN/GaN HEMTs MMIC amplifier. The gate pitch was optimally
designed to obtain the maximum output power. The designed amplifier is a
two-stage single-ended amplifier and the size is 3.8 × 6.2 mm2 . The achieved
output power under CW operation is 20 W, and the PAE is 19% at 26.5 GHz. In
the frequency range of 26–28 GHz, the output power is greater than 15 W with
an associated PAE of 13% [40]. Figure 3.25c shows a fabricated W-band GaN
PA MMICs using 80 nm InAlGaN/GaN HEMTs [13]. The MMIC consists of
two-stage cascade units with two transistors each and identical gate lengths in
order to provide high gain and low loss matching circuit. The size of the MMIC
was 2 × 1.8 mm2 . CW large signal characteristics were carried out at 86 GHz.
The maximum output power density was 3.6 W/mm with a PAE of 12.3% at
V ds = 20 V. Figure 3.25d depicts a D-band PA MMICs using 100 nm T-gate
AlGaN/GaN HEMTs. The S parameters of the active devices demonstrated an
F t of 100 GHz and an F max around 300 GHz. The MMIC consists of four actively
matched cascode stages, and the MMIC size was 3.75 × 2.0 mm2 . Large signal
measurements show a maximum output power density of 1.4 W/mm at 120 GHz
at V ds = 15 V with an associated PAE of 11.5% [107].
3.6 Summary
Nowadays, the GaN-based HEMT is becoming a critical technology for mmW
applications. Because of the outstanding material properties, a recent advanced
References 127
GaN device design provided a sharp increase in performances including high

power, high efficiency, reliability, and compact size. These capabilities are ideally
suited for numerous mmW power applications such as beamforming technol-
ogy for both 5G, 6G wireless networks and PAs MMIC for W-band frequency
and beyond. Therefore, GaN will play a strong role in advanced RF and mmW
applications including 5G and satellite communications. Thermal management,
especially when using small component sizes, together with full control of surface
and bulk traps, will be the main challenge to ensure the required device reliability.
References
1 Pengelly, R.S. and Turner, J.A. (1976). Monolithic broadband GaAs F.E.T.
amplifiers. Electron. Lett. 12: 10.
2 Aust, M.V., Sharma, A.K., Fordham, O. et al. (2006). A 2.8-W Q-band
high-efficiency power amplifier. IEEE J. Solid State Circuits 41 (10):
2241–2247.
3 Kong, K.K., Nguyen, B., Nayak, S., and Kao, M. (2005). Ka-band MMIC high
power amplifier (4 W at 30 GHz) with record compact size. IEEE Com-
pound Semiconductor Integrated Circuit Symposium, Palm Springs, CA,
USA (30 October-2 November 2005), pp. 232–235.
4 Campbell, C.F., Dumka, D.C., Kao, M., and Fanning, D.M. (2011). High effi-
ciency Ka-band power amplifier MMIC utilizing a high voltage dual field
plate GaAs pHEMT process. IEEE Compound Semiconductor Integrated
Circuit Symposium, pp. 1–4.
5 Liu, S.M.J., Tang, O.S.A., Kong, W., et al. (1999). High efficiency monolithic
InP HEMT V-band power amplifier. IEEE Gallium Arsenide Integrated Cir-
cuit Symposium, Monterey, CA, USA (17–20 October 1999), pp. 145–147.
6 Chen, Y.C., Ingram, D.L., Lai, R. et al. (1998). A 95-GHz InP HEMT MMIC
amplifier with 427-mW power output. IEEE Microw. Guid. Wave Lett. 8 (11):
399–401.
7 Chen, Y.C., Ingram, D.L., Yamauchi, D., et al. (1999). A single chip 1-W
InP HEMT V-band module. IEEE Gallium Arsenide Integrated Circuit
Symposium, Monterey, CA, USA (17–20 October 1999) pp. 149–152.
8 Fung, A., Ward, J., Chattopadhyay, G. et al. (2011). Power combined gallium
nitride amplifier with 3 watt output power at 87 GHz. IEEE J. Solid State
Circuits 41.
9 Chavarkar, P.M. and Parikb, P. (2003). 3.5-watt AlGaNlGaN HEMTs and
amplifiers at 35 GHz. IEEE International Electron Devices Meeting, Wash-
ington, DC, USA (8–10 December 2003), pp. 579–582.
10 Chen, S., Nayak, S., Campbell, C., and Reese, E. (2016). High efficiency
5W/10W 32–38 GHz power amplifier MMICs utilizing advanced 0.15 μm
GaN HEMT technology. IEEE Compound Semiconductor Integrated Circuit
Symposium, Austin, TX, USA (23–26 October 2016), pp. 1–4.
11 Campbell, C.F., Liu, Y., Kao, M., and Nayak, S. (2013). High efficiency
Ka-band gallium nitride power amplifier MMICs. IEEE International Confer-
ence on Microwaves, Communication Antennas and Electronic Systems, Tel
Aviv, Israel (21–23 October 2013), pp. 1–5.
12 Masuda, S., Ohki, T., Makiyama, K., et al. (2009). GaN MMIC amplifiers
for W-band transceivers. European Microwave Conference, Rome, Italy
(29 September–1 October 2009), pp. 1796–1799.
13 Niida, Y., Kamada, Y., Ohki, T., et al. (2016). 3.6 W/mm high power density
W-band InAlGaN/GaN HEMT MMIC power amplifier. 2016 IEEE Topi-
cal Conference on Power Amplifiers for Wireless and Radio Applications
(PAWR), Austin, TX, USA (24–27 January 2016), pp. 24–26.
14 Micovic, M., Kurdoghlian, A., Shinohara, K., et al. (2010). W-band GaN
MMIC with 842 mW output power at 88 GHz. IEEE MTT-S International
Microwave Symposium, Anaheim, CA, USA (23–28 May 2010), pp. 237–239.
15 Brown, A., Brown, K., Chen, J., et al. (2011). W-band GaN power amplifier
MMICs. IEEE MTT-S International Microwave Symposium, Baltimore, MD,
USA (5–10 June 2011), pp. 1–4.
16 Brown, D.F., Williams, A., Shinohara, K., et al. (2011). W-band power per-
formance of AlGaN/GaN DHFETs with regrown n+ -GaN Ohmic contacts by
MBE. IEEE International Electron Devices Meeting, Washington, DC, USA
(5–7 December 2011), pp. 461–464.
17 Schwantuschke, D., Godejohann, B.J., Brückner, P., et al. (2018). mm-Wave
operation of AlN/GaN-devices and MMICs at V- & W-band. 2018 22nd
International Microwave and Radar Conference, Poznan, Poland (14–17 May
2018), pp. 238–241.
18 Brown, A., Brown, K., Chen, J., et al. (2012). High power, high efficiency
E-band GaN amplifier MMICs. IEEE International Conference on Wireless
Information Technology and systems, Maui, HI, USA (11–16 November
2012), pp. 1–4.
19 Margomenos, A., Kurdoghlian, A., Micovic, M., et al. (2014). GaN technol-
ogy for E, W and G-band applications. IEEE Compound Semiconductor
Integrated Circuit Symposium, La Jolla, CA, USA (19–22 October 2014),
pp. 1–4.
20 Micovic, M., Kurdoghlian, A., Moyer, H.P., et al. (2008). GaN MMIC PAs for
E-band (71 GHz–95 GHz) radio. IEEE Compound Semiconductor Integrated
Circuit Symposium, Monterey, CA, USA (12–15 October 2008), pp. 1–4.
21 Schellenberg, J., Watkins, E., Micovic, M., et al. (2010). W-band, 5 W
solid-state power amplifier/combiner. IEEE MTT-S International Microwave
Symposium, Anaheim, CA, USA (23–28 May 2010), pp. 240–243.
22 Carrubba, V., Akmal, M., Quay, R. et al. (2012). The continuous inverse
class-F mode with resistive second-harmonic impedance. IEEE Trans.
Microwave Theory Tech. 60 (6): 1928–1936.
23 Thome, F., Leuther, A., Schlechtweg, M., and Ambacher, O. (2018). Broad-
band high-power W-band amplifier MMICs based on stacked-HEMT unit
cells. IEEE Trans. Microw. Theory Tech. 66 (3): 1312–1318.
24 Neininger, P., John, L., Br, P., and Friesicke, C. (2019). Design, analysis and
evaluation of a broadband high-power amplifier for Ka-band frequencies. In:
References 129
IEEE MTT-S International Microwave Symposium, Boston, MA (2–7 June

2019), 564–567.
25 Schellenberg, J., Kim, B., and Phan, T. (2013). W-band, broadband 2W GaN
MMIC. IEEE MTT-S International Microwave Symposium, Seattle, WA (2-7
June 2013).
26 Darwish, A.M., Boutros, K., Luo, B., Huebschman, B., et al. (2006). 4-watt
Ka-B and AlGaN / GaN power amplifier MMIC. IEEE MTT-S Interna-
tional Microwave Symposium Digest, San Francisco, CA (11–16 June 2006),
pp. 730–733.
27 Fung, A., Samoska, L., Kangaslahti, P., et al. (2018). Gallium nitride ampli-
fiers beyond W-band. IEEE Radio and Wireless Symposium, Anaheim, CA,
USA (15–18 January 2018), pp. 150–153.
28 Schellenberg, J., Tran, A., Bui, L., et al. (2016). 37 W, 75–100 GHz GaN
power amplifier. IEEE MTT-S International Microwave Symposium, San
Francisco, CA, USA (22–27 May 2016), pp. 81–84.
29 Dennler, P., Br, P., Schlechtweg, M., and Ambacher, O. (2014). Watt-level
non-uniform distributed 6–37 GHz power amplifier MMIC with dual-gate
driver stage in GaN technology. IEEE Topical Conference on Power Ampli-
fiers for Wireless and Radio Applications, Newport Beach, CA, USA (19–23
January 2014), pp. 37–39.
30 Dennler, P., Schwantuschke, D., and Ambacher, O. (2012). 8–42 GHz GaN
non-uniform distributed power amplifier MMICs in microstrip technology.
IEEE MTT-S International Microwave Symposium Digest, Montreal, QC,
Canada (17–22 June 2012).
31 IMT Vision–Framework and overall objectives of the future development of
IMT for 2020 and beyond. Online. https://www.itu.int/dms_pubrec/itu-r/
rec/m/R-REC-M.2083-0-201509-I!!PDF-E.pdf.
32 5G Semiconductor Solutions – Infrastructure and Fixed Wireless Access.
June 2018. Online. https://www.qorvo.com/resources/d/5g-semiconductor-
solutions-infrastructure-and-fixed-wireless-access-ebook.
33 Nakatani, K., Yamaguchi, Y., Komatsuzaki, Y., and Shinjo, S. (2019).
Millimeter-wave GaN power amplifier MMICs for 5G application. IEEE
International Symposium on Circuits and Systems, Sapporo, Japan (26–29
May 2019), pp. 6–9.
34 Lie, D.Y.C., Mayeda, J.C., and Lopez, J. (2017). Highly efficient 5G linear
power amplifiers (PA) design challenges. International Symposium on VLSI
Design, Automation and Test,Hsinchu, Taiwan (24–27 April 2017), pp. 1–3.
35 Popovic, Z. (2017). Amping up the PA for 5G: efficient GaN power ampli-
fiers with dynamic supplies. IEEE Microw. Mag. 18 (3): 137–149.
36 Yuk, K., Branner, G.R., and Cui, C. (2017). Future directions for GaN in 5G
and satellite communications. IEEE 60th International Midwest Symposium
on Circuits and Systems, Boston, MA, USA (6–9 August 2017), pp. 803–806.
37 Pelk, M.J., Neo, W.C.E., Gajadharsing, J.R. et al. (2008). A high-efficiency
100-W GaN three-way doherty amplifier for base-station applications. IEEE
Trans. Microwave Theory Tech. 56 (7): 1582–1591.
38 Campbell, C.F., Tran, K., Kao, M., and Nayak, S. (2012). A K-band 5W
Doherty amplifier MMIC utilizing 0.15 μm GaN on SiC HEMT technology.
2012 IEEE Compound Semiconductor Integrated Circuit Symposium, La

Jolla, CA, USA (14–17 October 2012), pp. 1–4.
39 Micovic, M., Brown, D.F., Regan, D., et al. (2016). High frequency GaN
HEMTs for RF MMIC applications. IEEE International Electron Devices
Meeting, San Francisco, CA, USA (3–7 December 2016), pp. 3.3.1–3.3.4.
40 Yamaguchi, Y., Kamioka, J., Hangai, M., et al. (2017). A CW 20W Ka-band
GaN high power MMIC amplifier with a gate pitch designed by using
one-finger large signal models. IEEE Compound Semiconductor Inte-
grated Circuit Symposium CSIC’05, Miami, FL, USA (22–25 October 2017),
pp. 5–8.
41 Noh, Y., Choi, Y., and Yom, I. (2015). Ka-band GaN power amplifier
MMIC chipset for satellite and 5G cellular communications. IEEE Fourth
Asia-Pacific Conference on Antennas and Propagation, Kuta, Indonesia (30
June–3 July 2015), pp. 453–456.
42 Nakatani, K., Yamaguchi, Y., Komatsuzaki, Y., and Sakata, S. (2018).
A Ka-band high efficiency Doherty power amplifier MMIC using
GaN-HEMT for 5G application. IEEE MTT-S International Microwave
Workshop Series on 5G Hardware and System Technologies, Dublin, Ireland
(30–31 August 2018), pp. 1–3.
43 Coffey, M., Momenroodaki, P., Zai, A., and Popovi, Z. (2015). A 4.2-W
10-GHz GaN MMIC Doherty power amplifier. IEEE Compound Semi-
conductor Integrated Circuit Symposium, New Orleans, LA, USA (11–14
October 2015).
44 Cheng, Q., Zhu, S., and Wu, H. (2013). Investigating the global trend of
RF power amplifiers with the arrival of 5G. IEEE International Wireless
Symposium (IWS 2015), Shenzhen, China (30 March–1 April 2015), pp. 1–4.
45 Tmoya, K., Kazumi, S., and Kunihiro, K. (2014). GaN HEMT high efficiency
power amplifiers for 4G/5G mobile communication base stations. IEEE
Asia-Pacific Microwave Conference, Sendai, Japan (4–7 November 2014),
pp. 6–9.
46 Zhang, Z., Xiao, Y., Ma, Z. et al. (2019). 6G wireless networks: vision,
requirements, architecture, and key technologies. IEEE Veh. Technol. Mag.
14 (3): 28–41.
47 Runton, D.W., Trabert, B., Shealy, J.B., and Vetury, R. (2013). History of
GaN. IEEE Microw. Mag. 14 (3): 82–93.
48 Arulkumaran, S. and Vicknesh, S. (2013). Enhanced breakdown voltage with
high Johnson’s HEMTs on silicon by (NH4 )2 Sx treatment. IEEE Electron
Device Lett. 34 (11): 1364–1366.
49 Huang, S., Wei, K., Liu, G. et al. (2014). High-fmax High Johnson’s
figure-of-merit 0.2-μm gate AlGaN/GaN HEMTs on silicon substrate with
AlN/SiNx passivation. IEEE Electron Device Lett. 35 (3): 315–317.
50 Zanoni, E. (2017). GaN HEMT reliability research – a white paper. Univer-
sity of Padua. pp. 1–61.
51 Nayak, S., Kao, M., Chen, H., et al. (2015). 0.15 μm GaN MMIC manufactur-
ing technology for 2–50 GHz power applications. CS ManTech Conference,
Scottsdale, AZ, USA (18–21 May 2015), pp. 43–46.
References 131
52 Whelan, C.S., Kolias, N.J., Brierley, S., et al. (2012). GaN technology for
radars. CS ManTech Conference, Boston, MA, USA (23–26 April 2012).
53 Makiyama, K., Ozaki, S., Ohki, T., et al. (2015). Collapse-free high power
InAlGaN/GaN-HEMT with 3 W/mm at 96 GHz. IEEE International Electron
Devices Meeting, pp. 9.1.1–9.1.4.
54 Harrouche, K., Kabouche, R., Okada, E., and Medjdoub, F. (2019). High
performance and highly robust AlN/GaN HEMTs for millimeter-wave opera-
tion. IEEE J. Electron Devices Soc. 7: 1145–1150.
55 Meneghesso, G., Verzellesi, G., Danesin, F. et al. (2008). Reliability of GaN
high-electron-mobility transistors: state of the art and perspectives. IEEE
Trans. Device Mater. Reliab. 8 (2): 332–343.
56 Meneghesso, G., Meneghini, M., Tazzoli, A. et al. (2010). Reliability issues of
gallium nitride high electron mobility transistors. Int. J. Microw. Wirel. Tech-
nol. 2 (1): 39–50.
57 Ono, N., Senju, T., and Takagi, K. (2018). 53% PAE 32W miniaturized
X-band GaN HEMT power amplifier MMICs. Asia-Pacific Microwave
Conference, Kyoto, Japan (6–9 November 2018), pp. 557–559.
58 Binari, S.C., Ikossi, K., Roussos, J.A. et al. (2001). Trapping effects and
microwave power performance. IEEE Trans. Electron Devices 48 (3):
465–471.
59 Binari, S.C., Klein, P.B., and Kazior, T.E. (2002). Trapping effects in GaN and
SiC microwave FETs. Proc. IEEE 90 (6): 1048–1058.
Trans. Electron Devices 48 (3): 560–566.
61 Mitrofanov, O. and Manfra, M. (2003). Mechanisms of gate lag in
GaN/AlGaN/GaN high electron mobility transistors. Superlattice. Microst.
34: 33–53.
62 Faqir, M., Verzellesi, G., Chini, A. et al. (2008). Mechanisms of RF current
collapse in AlGaN–GaN high electron mobility transistors. IEEE Trans.
Device Mater. Reliab. 8 (2): 240–247.
63 Zhang, A.P., Rowland, L.B., Kaminsky, E.B. et al. (2003). Correlation of
device performance and defects in AlGaN/GaN high-electron mobility
transistors. J. Electron. Mater. 32 (5): 388–394.
64 Kim, H., Vertiatchikh, A., Thompson, R.M. et al. (2003). Hot electron
induced degradation of undoped AlGaN/GaN HFETs. Microelectron. Reliab.
43 (6): 823–827.
65 Kordoš, P., Donoval, D., Florovič, M. et al. (2008). Investigation of trap
effects in AlGaN/GaN field-effect transistors by temperature dependent
threshold voltage analysis. Appl. Phys. Lett. 92 (5): 1–4.
66 Kabouche, R., Pecheux, R., Harrouche, K. et al. (2019). High efficiency
AlN/GaN HEMTs for Q-band applications with an improved thermal dissi-
pation. Int. J. High Speed Electron. Syst. 28: 1–2.
67 Meneghesso, G., Chini, A., Zanoni, E., et al. (2000). Diagnosis of trapping
phenomena in GaN MESFETs. International Electron Devices Meeting, San
Francisco, CA, USA (10–13 December 2000), pp. 389–392.
68 Haruyama, J., Negishi, H., Nishimura, Y., and Nashimoto, Y. (1997).

Substrate-related kink effects with a strong light-sensitivity in
AlGaAs/InGaAs pHEMT. IEEE Trans. Electron Devices 44 (1): 25–33.
69 Wang, M. and Chen, K.J. (2011). Kink effect in AlGaN/GaN HEMTs induced
by drain and gate pumping. IEEE Electron Device Lett. 32 (4): 482–484.
70 Mazzanti, A., Verzellesi, G., Canali, C. et al. (2002). Physics-based explana-
tion of kink dynamics in AlGaAs/GaAs HFETs. IEEE Electron Device Lett. 23
(7): 383–385.
71 Sun, H.F. and Bolognesi, C.R. (2007). Anomalous behavior of heterostruc-
ture field-effect transistors at cryogenic temperatures: from current collapse
to current enhancement with cooling anomalous behavior of AlGaN/GaN
heterostructure field-effect. Appl. Phys. Lett. 90 (12): 123505.
72 Lin, C., Wang, W., Lin, P. et al. (2005). Transient pulsed analysis on GaN
HEMTs at cryogenic temperatures. IEEE Electron Device Lett. 26 (10):
710–712.
73 Green, B.M., Chu, K.K., Chumbes, E.M. et al. (2000). The effect of sur-
face passivation on the microwave characteristics of undoped AlGaN/GaN
HEMT’s. IEEE Electron Device Lett. 21 (6): 268–270.
74 Fujimoto, H., Saito, W., Yoshioka, A., et al. (2008). Wafer quality target
for current-collapse-free GaN-HEMTs in high voltage applications. CS
MANTECH Conference, Chicago, IL, USA (14–17 April 2008).
75 Wu, Y., Saxler, A., Moore, M. et al. (2004). 30-W/mm GaN HEMTs by field
plate optimization. IEEE Electron Device Lett. 25 (3): 117–119.
76 Romanczyk, B., Guidry, M., Wienecke, S., et al. (2016). W-band N-polar
GaN MISHEMTs with high power and record 27.8% efficiency at 94 GHz.
IEEE International Electron Devices Meeting, San Francisco, CA, USA (3–7
December 2016), pp. 67–70.
77 Hickman, A., Chaudhuri, R., Bader, S.J. et al. (2019). High breakdown volt-
age in RF AlN/GaN/AlN quantum well HEMTs. IEEE Electron Device Lett.
40 (8): 1293–1296.
78 Smorchkova, I.P., Chen, L., Mates, T. et al. (2001). AlN /GaN and (Al,
Ga) N/AlN/GaN two-dimensional electron gas structures grown by
plasma-assisted molecular-beam epitaxy. J. Appl. Phys. 90 (10): 5196–5201.
79 Cheng, K., Leys, M., Derluyn, J. et al. (2007). AlGaN/GaN HEMT grown on
large size silicon substrates by MOVPE capped with in-situ deposited Si3 N4 .
J. Cryst. Growth 298: 822–825.
80 Tadjer, M.J., Anderson, T.J., Hobart, K.D. et al. (2010). Electrical and optical
characterization of AlGaN/GaN HEMTs with in situ and ex situ deposited
SiNx layers. J. Electron. Mater. 39 (11): 2452–2458.
81 Eastman, L.F., Tilak, V., Smart, J. et al. (2001). Undoped AlGaN/GaN HEMTs
for microwave power amplification. IEEE Trans. Electron Devices 48 (3):
479–485.
82 Ambacher, O., Smart, J., Shealy, J.R. et al. (1999). Two-dimensional elec-
tron gases induced by spontaneous and piezoelectric polarization charges
in N- and Ga-face AlGaN/GaN heterostructures. J. Appl. Phys. 85 (6):
3222–3233.
References 133
83 Jessen, G.H., Fitch, R.C., Gillespie, J.K. et al. (2007). Short-channel effect
limitations on high-frequency operation of AlGaN/GaN HEMTs for T-gate
devices. IEEE Trans. Electron Devices 54 (10): 2589–2597.
84 Shinohara, K., Regan, D.C., Tang, Y. et al. (2013). Scaling of GaN HEMTs
and Schottky diodes for submillimeter-wave MMIC applications. IEEE Trans.
85 Milosavljevic, I., Shinohara, K., Regan, D. et al. (2010). Vertically scaled
GaN/AlN DH-HEMTs with regrown n+ -GaN Ohmic contacts by MBE.
Device Res. Conf. - Conf. Dig. DRC. 54: 159–160.
86 Shinohara, K., Regan, D., Corrion, A., et al. (2012). Self-aligned-gate
GaN-HEMTs with heavily-doped n+ GaN Ohmic contacts to 2DEG. IEEE
International Electron Devices Meeting, San Francisco, CA, USA (10–13
December 2012), pp. 27.2.1–27.2.4.
87 Chung, J.W., Hoke, W.E., Chumbes, E.M. et al. (2010). AlGaN/GaN HEMT
with 300-GHz f max . IEEE Electron Device Lett. 31 (3): 195–197.
88 Shinohara, K., Corrion, A., Regan, D., et al. (2010). 220 GHz f T and 400
GHz f max in 40-nm GaN DH-HEMTs with re-grown Ohmic. IEEE Interna-
tional Electron Devices Meeting, San Francisco, CA, USA (6–8 December
2010), pp. 30.1.1–30.1.4.
89 Lin, Y.K., Bergsten, J., Leong, H. et al. (2018). A versatile low-resistance
Ohmic contact process with Ohmic recess and low-temperature annealing
for GaN HEMTs. Semicond. Sci. Technol. 33 (9): 095019.
90 Buttari, D., Chini, A., Palacios, T. et al. (2003). Origin of etch delay time
in Cl2 dry etching of AlGaN/GaN structures. Appl. Phys. Lett. 83 (23):
4779–4781.
91 Wong, M.H., Pei, Y., Palacios, T. et al. (2007). Low nonalloyed Ohmic con-
tact resistance to nitride high electron mobility transistors using N-face
growth. Appl. Phys. Lett. 91 (23): 1–4.
92 Selvanathan, D., Zhou, L., Kumar, V., and Adesida, I. (2002). Low resistance
Ti/Al/Mo/Au Ohmic contacts for AlGaN/GaN heterostructure field effect
transistors. Phys. Status Solidi Appl. Res. 194 (2): 583–586.
93 Guo, J., Li, G., Faria, F. et al. (2012). MBE-regrown Ohmics in InAlN
HEMTs with a regrowth interface resistance of 0.05 Ωmm. IEEE Electron
Device Lett. 33 (4): 525–527.
94 Yue, Y., Hu, Z., Guo, J. et al. (2012). InAlN/AlN/GaN HEMTs with regrown
Ohmic contacts and f T of 370 GHz. IEEE Electron Device Lett. 33 (7):
988–990.
95 Wienecke, S., Romanczyk, B., Guidry, M. et al. (2017). N-polar GaN cap
MISHEMT with record power density exceeding 6.5 W/mm at 94 GHz.
IEEE Electron Device Lett. 38 (3): 359–362.
96 Romanczyk, B., Wienecke, S., Guidry, M. et al. (2018). Demonstration of
constant 8 W/mm power density at 10, 30, and 94 GHz in state-of-the-art
millimeter-wave N-polar GaN MISHEMTs. IEEE Trans. Electron Devices 65
(1): 45–50.
97 Tang, Y., Shinohara, K., Regan, D. et al. (2015). Ultrahigh-speed GaN
high-electron-mobility transistors with f T /f max of 454/444 GHz. IEEE Elec-
tron Device Lett. 36 (6): 549–551.
98 Moon, J., Wong, J., Grabar, B., et al. (2019). High-speed linear GaN tech-
nology with a record efficiency in Ka-band. The European Microwave
Conference, Paris, France (30 September–1 October 2019), pp. 57–59.
99 Kotani, J., Yamada, A., Ohki, T., et al. (2018). Recent advancement of GaN
HEMT with InAlGaN barrier layer and future prospects of AlN-based elec-
tron devices. IEEE International Electron Devices Meeting, San Francisco,
CA, USA (1–5 December 2018), pp. 30.4.1–30.4.4.
100 Palacios, T., Member, S., Chakraborty, A. et al. (2005). High-power
AlGaN/GaN HEMTs for Ka-band applications. IEEE Electron Device Lett.
26 (11): 781–783.
101 Moon, J.S., Wu, S., Wong, D. et al. (2005). Gate-recessed AlGaN – GaN
HEMTs for high-performance millimeter-wave applications. IEEE Electron
Device Lett. 26 (6): 348–350.
102 Micovic, M., Kurdoghlian, A., Hashimoto, P., et al. (2006). GaN HFET for
W-band power applications. IEEE International Electron Devices Meeting,
San Francisco, CA, USA (11–13 December 2006), pp. 5–7.
103 Quay, R. (2014). Group III-Nitride Monolithically Microwave Integrated
Circuits (MMICs) (ed. F. Medjdoub), 372. CRC Press, 19 December 2017.
104 Pengelly, R.S., Wood, S.M., Milligan, J.W. et al. (2012). A review of GaN
on SiC high electron-mobility power transistors and MMICs. IEEE Trans.
Microwave Theory Tech. 60 (2): 1764–1783.
105 Kolias, N.J., Whelan, C.S., Kazior, T.E., et al. (2010). GaN technology
for microwave and millimeter wave applications. IEEE MTT-S Interna-
tional Microwave Symposium, Anaheim, CA, USA (23–28 May 2010),
pp. 1222–1225.
106 Margomenos, A., Micovic, M., Butler, C., et al. (2013). Low loss, Cu dama-
scene interconnects and passives compatible with GaN MMIC. IEEE MTT-S
International Microwave Symposium Digest, Seattle, WA, USA (2–7 June
2013), pp. 1–4.
107 Cwiklinski, M., Brückner, P., Leone, S., et al. (2019). D-band and G-band
high-performance. IEEE Transactions on Microwave Theory and Techniques
(December 2019), pp. 1–10.
108 Inoue, K., Sano, S., Tateno, Y., et al. (2010). Development of gallium nitride
high electron mobility transistors for cellular base stations. SEI Technical
Review, pp. 88–93.
109 Santhakumar, R., Thibeault, B., Member, S. et al. (2011). Two-stage
high-gain high-power distributed amplifier using dual-gate GaN HEMTs.
IEEE Trans. Microwave Theory Tech. 59 (8): 2059–2063.
110 Litchfield, M., Schafer, S., Reveyrand, T., and Popovic, Z. (2014).
High-efficiency X-band MMIC GaN power amplifiers operating as recti-
fiers. IEEE MTT-S International Microwave Symposium, Tampa, FL, USA
(1–6 June 2014), pp. 1–4.
111 Kee, S.D., Aoki, I., Hajimiri, A., and Rutledge, D. (2003). The class-E/F family
of ZVS switching amplifiers. IEEE Trans. Microwave Theory Tech. 51 (6):
1677–1690.
References 135
112 Senju, T., Takagi, K., and Kimura, H. (2018). A 2 W 45% PAE X-band GaN
HEMT class-F MMIC power amplifier. Asia-Pacific Microwave Conference,
Kyoto, Japan (6–9 November 2018), pp. 956–958.
113 Gao, S., Xu, H., Mishra, U.K., and Barbara, S. (2006). MMIC class-F power
amplifiers using field-plated AlGaN/GaN HEMTs. IEEE Compound Semi-
conductor Integrated Circuit Symposium, San Antonio, TX, USA (12–15
November 2006), pp. 81–84.
114 Micovic, M., Brown, D., Regan, D., et al. (2016). Ka-band LNA MMIC’s
realized in f max > 580 GHz GaN HEMT technology. IEEE Compound Semi-
conductor Integrated Circuit Symposium (CSICS), Austin, TX, USA (23–26
Oct. 2016), pp. 1–4.
115 Kobayashi, K.W., Denninghoff, D., and Miller, D. (2015). A novel 100
MHz–45 GHz GaN HEMT low noise non-gate-terminated distributed
amplifier based on a 6-inch 0.15 m GaN–SiC mm-wave process technology.
2015 IEEE Compound Semiconductor Integrated Circuit Symposium, New
Orleans, LA, USA (11–14 October 2015), pp. 1–4.
116 Masslerl, H., Wagnerl, S., and Brucknerl, P. (2011). A highly linear 84 GHz
low noise amplifier MMIC In AIGaN/GaN HEMT technology 1. IEEE
MTT-S International Microwave Workshop Series on Millimeter Wave Inte-
gration Technologies, Sitges, Spain (15–16 September 2011), pp. 144–147.
137
Technologies for Normally-off GaN HEMTs

Giuseppe Greco 1 , Patrick Fiorenza 1 , Ferdinando Iucolano 2 , and Fabrizio
Roccaforte 1
1
Consiglio Nazionale delle Ricerche – Istituto per la Microelettronica e Microsistemi (CNR-IMM), Strada VIII,
2 STMicroelectronics, Stradale Primosole 50, 95121 Catania, Italy
4.1 Introduction
One of the most interesting features of GaN-based materials is the presence of
a two-dimensional electron gas (2DEG) in heterostructures (e.g. AlGaN/GaN,
AlN/GaN, and InAlN/GaN), which gives the possibility to fabricate high electron
mobility transistors (HEMTs) [1]. Because of the presence of a 2DEG with a high
charge density and mobility, AlGaN/GaN HEMTs are inherently normally-on
devices. These devices are excellent components for high-frequency applications.
In fact, today, the 5G technology is the driving force for large semiconductor
companies to develop GaN-on-Si HEMTs for power amplifiers, high-efficiency
devices in RF, microwave, and millimeter-wave (mmW) applications [2].
However, GaN is also a suitable material for power electronics. In particular,
for power switching applications, the use of normally-off transistors is highly
preferred, not only to simplify the gate drive configuration but also for safety rea-
sons [3–5]. In fact, if the gate driver fails and the gate bias goes to zero voltage,
a normally-off HEMT switches to the off-state, avoiding circuit burnout prob-
lems. Obviously, this is a more safe condition than having a switch remaining in
the on-state. Hence, in the past 10 years, many efforts have been made to explore
the viable routes toward the development of reliable normally-off GaN HEMT
technologies.
From the physical point of view, to achieve the normally-off operation, the
threshold voltage (V th ) must be shifted in the positive bias direction. Practically,
to achieve this goal, the region below the gate contact in GaN-based heterostruc-
tures must be properly modified.
This chapter is focused on the most common and feasible approaches to obtain
normally-off HEMTs on GaN-based heterostructures.
Although the “cascode” configuration, combining a Si MOSFET with a
normally-on GaN HEMT, can enable to realize a chip with normally-off
operation, “true” normally-off GaN HEMTs are requested in many applications.

138 4 Technologies for Normally-off GaN HEMTs
The first approaches toward the achievement of “true” normally-off HEMTs

consisted in the thinning of the AlGaN barrier layer (recessed gate) or the intro-
duction of fluorine ions in the gate region. Although both approaches resulted
in a positive shift of the threshold voltage, they also exhibited some limitations,
which pushed the technology to move toward more robust solutions.
The natural evolution of these device concepts was the fabrication of
recessed-gate metal insulator semiconductor high electron mobility transistors
(MISHEMTs). These devices are continuously object of scientific interest, but
a number of issues, related to the control of threshold voltage and dielectric
reliability, have hindered their commercialization.
The real breakthrough in normally-off HEMT technology was the introduc-
tion of the p-GaN gate HEMT, which allows to obtain a stable positive threshold
voltage and is not affected by the instability issues of the insulated gate devices
(https://epc-co.com/epc). As a matter of fact, today, the p-GaN gate HEMT is the
only solution available on the market together with the HEMT cascode (https://
www.infineon.com and https://www.transphormusa.com).
Other “unconventional” solutions have been proposed in the literature to
achieve normally-off GaN HEMT operation. Although these approaches are
very far from a real application, they will be mentioned at the end of this chapter
to provide additional inputs for the comprehension of the physics behind the
GaN HEMT technology.
4.1.1 Threshold Voltage in AlGaN/GaN HEMTs

To obtain a normally-off AlGaN/GaN HEMT, an accurate control of the
2DEG sheet charge density (ns ) is required. Figure 4.1 schematically shows the
ΦB
EC
ΔEC
EF
EF(X)
Schottky dAIGaN
gate
AIxGa1–xN GaN
Figure 4.1 Schematic conduction band diagram of an AlGaN/GaN heterostructure in the

Schottky gate region of a HEMT.
conduction band diagram of an AlGaN/GaN heterostructure in the Schottky

gate region of a conventional HEMT. The device threshold voltage (V th ) is
defined as the bias needed to deplete the 2DEG under the gate region.
In AlGaN/GaN HEMTs, the threshold voltage V th depends on the different
characteristics of the heterostructure (residual doping, polarization charges, bar-
rier layer thickness, etc.) and has been often described by the following analytical
expression [6]:
𝜎(x) qND
Vth (x) = ΦB (x) + EF (x) − ΔEC (x) − d − (d )2
𝜀0 𝜀AlGaN (x) AlGaN 2𝜀0 𝜀AlGaN (x) AlGaN
(4.1)
where ΦB is the metal/AlGaN Schottky barrier height, EF is the energy differ-
ence between the intrinsic Fermi level and the GaN conduction band edge (and
depends on the sheet charge density of the 2DEG [7]), 𝜎 is the polarization charge
that uniquely depends on the Al-concentration x [7], and dAlGaN , N D , and 𝜀AlGaN
are the barrier layer thickness, doping, and permittivity, respectively.
For a typical AlGaN/GaN heterostructure used for normally-on HEMTs, i.e.
with dAlGaN = 20 nm, x = 25%, and N D ∼ 1 × 1017 cm−3 , considering a barrier
height ΦB = 1 eV, a threshold voltage V th = −4.6 V can be predicted using
Eq. (4.1).
Clearly, Eq. (4.1) gives an indication on the main parameters that can be tailored
to move the V th toward positive values.
An increase of the metal/AlGaN Schottky barrier height ΦB results into a
positive shift of the threshold voltage V th . However, the typical values of the
metal/AlGaN barrier height are limited to the range of 0.8–1.2 eV [8–10]. Hence,
a positive V th shift of only 0.4 eV can be obtained by maximizing the barrier
height of the metal gate. Based on these considerations, it is clear that acting on
the Schottky barrier height of the metal gate is not an efficient solution to obtain
the normally-off operation.
Changing the Al-concentration x and the thickness of the AlGaN barrier layer
dAlGaN produces more significant variations of the threshold voltage V th . As an
example, considering again an AlGaN/GaN heterostructure with dAlGaN = 20 nm,
a reduction of the Al-concentration x from 25% to 10% leads to a shift of V th
from −4.6 to −1.2 V. Moreover, for an Al-concentration x = 25%, a shift of V th
from −4.6 to −1.9 V is expected by reducing the AlGaN thickness from 20 to
10 nm. However, the sheet carrier density in the 2DEG ns decreases by reducing
the AlGaN barrier layer thickness dAlGaN (see Eq. (1.11)). Hence, dAlGaN cannot
be reduced below a critical thickness of few nanometers [11]. The recessed-gate
HEMT, described in Section 4.3.1, is based on the reduction of dAlGaN and has the
peculiarity that the AlGaN thinning is done only below the gate region, without
pauperizing the 2DEG in the access region of the device.
In Eq. (4.1), the value of EF depends on the 2DEG sheet charge density and
typically increases by increasing the sheet charge density [7].
The conduction band offset ΔEC at the AlGaN/GaN interface depends on the
band gap of the Alx Ga1−x N alloy used as a barrier layer. As an example, a decrease
of the Al-concentration from 30% to 20% translates into a change of the band
offset ΔEC from 0.42 to 0.27 eV (see Eq. (4.1)) [12, 13]. For an AlGaN thickness
of 20 nm, it will result in a V th shift of 2.5 V.
As already seen in Chapter 1, the polarization charge (𝜎) depends on both the
spontaneous (PSP ) and piezoelectric (PPE ) polarization, and these parameters in
the AlGaN barrier layer are determined by the Al-concentration x [7].
The AlGaN doping concentration (N D ) also has an impact on V th , as it
determines the conduction band bending in the barrier layer [14]. Moreover,
an increase of the AlGaN donor concentration N D causes a polarization field
screening, which deepens the potential well and increases the 2DEG concen-
tration [15]. In this context, the introduction of an additional negative charge
(e.g. fluorine ions) in the AlGaN barrier layer to deplete the 2DEG is another
possibility to positively shift V th , as will be discussed in Section 4.3.2.
Although the use of a high metal/AlGaN barrier height can produce only a
small positive V th shift, the use of a p-type GaN gate layer on the top of the
AlGaN/GaN heterostructure can raise the AlGaN conduction band of an amount
comparable with the GaN band gap (3.4 eV). Hence, the p-GaN gate HEMT, dis-
cussed in Section 4.3.4, is one of the most efficient routes to obtain a positive
threshold voltage in GaN-based heterostructures.
Finally, it is worth mentioning that the presence of other charges, e.g. trapped
in the surface and in the buffer, can also have a strong impact on the experimental
values of threshold voltage V th in GaN HEMTs. These coulombic effects are not
explicitly considered in Eq. (4.1). For that reason, it is always extremely difficult
to well reproduce the experimental data using the analytical expression of V th .
4.2 GaN HEMT “Cascode”

Driven by commercial market requirements, some companies (IR, Sharp,
and Transphorm) have overpassed the difficulties in achieving a single-chip
normally-off GaN HEMT by using the so-called “cascode” structure. Such
configuration has been originally proposed by Baliga to obtain a stable pos-
itive threshold voltage in 4H–SiC transistors (JFETs) [16]. However, the first
high-voltage (HV, 600 V) normally-off GaN HEMT cascode was released into
the market by Transphorm in 2015 [17].
In a cascode structure, a HV normally-on GaN HEMT is connected in series
with a low-voltage (LV) normally-off Si MOSFET. Figure 4.2 schematically shows
the equivalent circuit of the normally-off GaN HEMT cascode and the two
devices connected in the package.
Using this configuration, the gate-to-source voltage (V GS ) in GaN HEMT is
equivalent to source-to-drain voltage (V DS ) in Si MOSFET. Then, the switch
between off- and on-states of the GaN HEMT can be easily controlled by the
normally-off Si MOSFET. The two devices share the same channel current in the
on-state, while in the off-state, the blocking voltage is distributed between them.
The application of a positive gate bias to the Si MOSFET above its thresh-
old voltage will turn-on this device. In this condition, the gate voltage of the
GaN HEMT is close to zero and, then, this normally-on transistor is turned-on.
4.2 GaN HEMT “Cascode” 141
HV normally on
LV normally off GaN HEMT
Si MOSFET
S D GaN
D Si D S
G
G G S G D
S G S D
(a) (b)
Figure 4.2 (a) Equivalent circuit of a normally-off GaN HEMT cascode, connecting a
high-voltage (HV) normally-on GaN HEMT with a low-voltage (LV) normally-off Si MOSFET.
(b) Schematic representation of the two devices connected in a package.
Because the two devices are connected in series, in the on-state, the same channel
current will flow through the two devices.
On the other hand, when the Si MOSFET is turned-off by removing the gate
voltage, an applied bias to the drain terminal creates a negative gate-to-source
voltage on the GaN HEMT, contributing to deplete the 2DEG in the HEMT chan-
nel. Then, the device will be in the off-state and any further increase of the drain
voltage will drop on the GaN HEMT. Hence, because of the high critical field
of GaN, a high breakdown voltage can be reached with the cascode configura-
tion. Practically, in a 600 V GaN HEMT cascode, most of the bias drops on the
GaN HEMT, while the resistance added by the low-voltage Si MOSFET is very
low (about 3% of the total resistance). On the other hand, if the rated voltage is
reduced to 50 V, the on-resistance of the GaN HEMT decreases. Consequently,
the percentage contribution of the on-resistance from the Si MOSFET increases.
This concept is illustrated in Figure 4.3, showing the percentage on-resistance
contribution from the Si MOSFET as a function of rated voltage in a commercial
GaN HEMT cascode [18]. For this reason, the GaN HEMTs cascode are preferred
for the voltage classes of 600 and 900 V (https://www.infineon.com and https://
www.transphormusa.com).
An advantage of the GaN HEMT cascode is the protection offered by the Si
MOSFET to avoid GaN gate breakdown. As an example, for a 600 V normally-on
GaN HEMT, the gate breakdown voltage is around −35 V. Therefore, using a 30 V
Si MOSFET enables to turn-off the GaN HEMT, while leaving some secure mar-
gin between turn-off and V th of the GaN HEMT.
Moreover, from the reliability point of view, the cascode configuration is a
more robust solution with respect to other approaches for normally-off GaN
(e.g. p-GaN HEMT and recessed-gate hybrid MISHEMT). Indeed, because the
cascode configuration uses a normally-on HEMT, the electric field on the gate
in the on-state is low because of the fact that the gate voltage is zero. Different
is the case of a “true” normally-off HEMT, where a positive gate bias has to be
applied to turn-on the device, leading to many reliability issues [19]. This finally
Figure 4.3 Percentage of on-resistance from Si

Cascode configuration
40 HV GaN HEMT + LV Si MOSFET MOSFET in a normally-off GaN HEMT cascode.
from Si-MOSFET (%) Source: Adapted with permission of Lidow et al.
2014 [18]. Copyright © 2014, Wiley VCH.
Percentage RON
30
20
10
0
0 200 400 600
Rated breakdown voltage (V)
automatically solves any criticism for one of the important tests of the standard
qualification procedure (JEDEC, AEC-Q100/101), i.e. the high-temperature gate
bias (HTGB) [20].
Finally, the fabrication of normally-on HEMTs for the cascode configuration
is simpler than that of a “true” normally-off GaN HEMT, as will be seen in
Section 4.3.
Although the “cascode” configuration has the advantage to be driven by
conventional Si MOSFET drivers, this solution exhibits some drawbacks. As
an example, the series connection of the two devices increases the package
complexity and introduces parasitic inductances that affect the switching
performance of the system. In fact, the two dies are typically connected inside
the package with wire bonds or in a planar architecture (see schematic in
Figure 4.2b). Then, the switching performance of the HEMT cascode relies
heavily on the parasitic inductances in the package, especially between the two
dies and also on how well the junction capacitances of the two are matched.
If the inductances are too high, or the capacitances are not matched well, the
switching losses can increase significantly [21].
Moreover, the advantage of the high-temperature operation offered by a
“pure GaN” device is lost in the HEMT cascode because of the presence of a Si
device [22].
For those reasons, the power electronics market requires the adoption of “true”
normally-off HEMT solutions, and several semiconductor companies all over the
word are working to develop such a technology.
4.3 “True” Normally-off HEMT Technologies

4.3.1 Recessed-gate HEMT
The use of recessed-gate contacts in GaN-based HEMT technology has always
been an attractive topic for the scientific community working on AlGaN/GaN
heterostructures. At the early stages, recessed-gate HEMTs have been fabricated
to minimize the parasitic resistance and achieve a high transconductance
[23–26]. As a side approach, the recessed gate, obtained by a plasma etch below
Recessed
Nonrecessed
Source Drain ΦB
Gate ΔEC
EC
AlGaN dAlGaN-recessed
2DEG
Schottky dAlGaN
GaN gate
AlGaN GaN
(a) (b)
Figure 4.4 (a) Schematic cross section of a recessed-gate normally-off AlGaN/GaN HEMT.
(b) Conduction band diagram of the AlGaN/GaN heterostructure below the gate region before
(dashed line) and after the recession (continuous line) of the barrier layer.
the gate electrode, has been considered to obtain a positive shift of the threshold
voltage V th [27, 28]. In fact, as can be deduced from Eq. (4.1), a reduction of
the AlGaN barrier layer thickness dAlGaN leads to a decrease of the 2DEG sheet
carrier concentration and to a positive shift of V th .
Figure 4.4 schematically depicts a recessed-gate AlGaN/GaN HEMT, in which
a partial thinning of the AlGaN barrier has been obtained below the gate contact
by a plasma etch process. Clearly, in this configuration while the 2DEG channel is
depleted below the gate, in the device access regions (source-gate and gate-drain),
the high sheet charge density and low resistance are preserved.
Saito et al. [29] introduced the recessed-gate concept, demonstrating the
possibility to achieve a considerable positive V th shift (i.e. from −4 to −0.14 V)
by reducing the AlGaN thickness from 30 to 9.5 nm. Many other works
[16, 17, 30–35] reported on positive V th shifts in the range of 0.1–0.5 V, main-
taining an AlGaN thickness below the gate contact between 6.5 and 13 nm.
Table 4.1 reports a collection of literature data obtained in recessed-gate
GaN-based HEMTs. Clearly, a direct correlation of the V th values with the resid-
ual AlGaN thickness is not simple, as those experiments have been carried out on
heterostructures with different Al-concentration x. Moreover, the metal/AlGaN
interface quality is strongly influenced by the plasma etch condition used for the
recession, whose effect on the final V th is difficult to quantify. However, from
Table 4.1, it is evident that the values of V th in the recessed-gate HEMT are
always close to zero (in the range 0.05–0.5 V).
Although Saito et al. [29] did not obtain a positive threshold voltage, they
observed a linear correlation between the measured V th and the thickness of
the residual AlGaN layer below the gate contact. From this correlation, shown
in Figure 4.5 for the case with an Al-concentration of 25%, it was possible to
extrapolate the normally-off condition for a residual AlGaN thickness smaller
than 8.2 nm [29]. A similar correlation between the residual AlGaN thickness
after plasma etch and the experimental V th has also been observed in other
works [31, 34].
Table 4.1 Collection of literature data obtained in recessed-gate AlGaN/GaN HEMTs.
Residual AlGaN Threshold

AlGaN Al- Threshold thickness after voltage after
thickness concentration voltage gate recession gate recession
dAlGaN (nm) x (%) V th (V) (nm) V th (V) References
10 10 – – +0.05 [27]
18 25 – – +0.075 [28]
30 25 −4.42 9.5 −0.14 [29]
25 33 −4.2 13 +0.1 [30]
22 27 −2.2 7 +0.47 [31]
38 35 −0.6 10 +0.51 [32]
22 22 −2.7 6.5 +0.1 [33]
25 25 −1.5 3 −0.1 [34]
25 25 −4 10 +0.5 [35]
1 Figure 4.5 Experimental values of the threshold

voltage V th as a function of the residual AlGaN
Normally-off
0 barrier layer thickness in recessed-gate
Threshold voltage (V)
Normally-on AlGaN/GaN HEMTs. Source: The data are taken

from Refs. [29, 31, 34].
–1
–2
–3
Saito et al. 2008
Hao et al. 2008
–4 Anderson et al. 2010
0 5 10 15 20 25 30
Residual AlGaN thickness (nm)
Inductively coupled plasma (ICP) or reactive ion etching (RIE) based on the
chemistry of Cl2 or BCl3 is typically used to etch the GaN or AlGaN layers and
are widely used processes in recessed-gate HEMT technology [36]. However, the
nanometric control of the recessed AlGaN layer thickness and the achievement of
the desired V th is not simple. In fact, plasma-induced damage of the recessed-gate
region represents a serious concern in this technology because of the increase of
the gate leakage current or the device on-resistance [30–33, 37].
A possible solution to overcome the drawbacks of the plasma etch in the
recessed-gate region is the use of a selective area growth (SAG) of the AlGaN
barrier layer [38]. This technology starts from an AlGaN/GaN heterostructure
with a very thin AlGaN barrier layer. On this material, an additional SAG of
AlGaN is carried out on the access regions of the device. In this way, the exposure
of the gate region to the plasma etch process is avoided. He et al. [35] compared
the recessed-gate normally-off HEMTs fabricated by standard ICP with those
fabricated with the SAG of AlGaN layer in the access regions. Although the
value of the threshold voltage obtained with the SAG approach (V th = +0.4 V)
was comparable to that of a standard ICP recession (V th = +0.5 V), avoiding
the use of plasma below the gate electrode allowed to obtain a reduction of
the gate leakage current and higher values of saturation current I Dmax and
transconductance g m .
4.3.2 Fluorinated HEMT

The incorporation of negative charges in the gate region represents another pos-
sible approach to positively shift V th and obtain a normally-off HEMT.
This concept was first proposed by Cai et al. [39] that controlled the thresh-
old voltage V th through the introduction of fluorine (F) ions in the AlGaN/GaN
heterostructure using a CF4 plasma treatment. The presence of fluorine in the
sample was verified by means of secondary ion mass spectrometry (SIMS) [40].
Figure 4.6 reports a schematic in cross section of a fluorine-implanted HEMT,
together with a conduction band diagram of this system to explain the physics
behind. When a certain amount of immobile negative charges is introduced (e.g.
by plasma or ion implantation) inside the AlGaN barrier layer under the gate
electrode (Figure 4.6a), these charges will electrostatically deplete the 2DEG in
the channel region.
Fluorine (F) is the most electronegative element of the periodic table. When F
is introduced in interstitial sites in the AlGaN matrix, close to the neighboring
atoms (Al, Ga, or N), it tends to capture a free electron, becoming a negative fixed
charge. The immobile negative F− -ions incorporated in the AlGaN barrier layer
will raise the potential of the AlGaN of an additional quantity ΦF , thus causing the
depletion of the 2DEG. Consequently, the threshold voltage V th will be positively
shifted and a normally-off condition can be achieved. This situation is illustrated
in Figure 4.6b, which compares the conduction band diagram of an AlGaN/GaN
heterostructure before and after a fluorine treatment.
E Before F treatment
After F treatment
ΦF –
– –
–
–
–
Source Gate Drain
F– F– F– F–
ΦB
AlGaN ΔEC EC
–
F– F F– F–
2DEG
EF
GaN Schottky AlGaN
gate GaN
(a) (b)
Figure 4.6 (a) Schematic cross section of a normally-off fluorinated AlGaN/GaN HEMT.
(b) Conduction band diagram of a fluorinated HEMT structure (continuous line). The band
diagram of the AlGaN/GaN heterostructure before the F treatment is also reported for
comparison (dashed line).
2 Figure 4.7 Calculation of the threshold

voltage V th in fluorinated GaN HEMT as a
1 function of incorporated fluorine
concentration. The following parameters
Normally-off have been assumed for this calculation: an

0
Normally-on Al-concentration x = 30%, an AlGaN
thickness dAlGaN = 20 nm, and a barrier height
–1 ΦB = 1 eV. Under these conditions, the
normally-off behavior occurs for a fluorine
–2 concentration NF = 1.6 × 1019 cm−3 .
–3
–4
0 1 2 3
Fluorine concentration (×1019) (cm–3)
To consider the effect of fluorine incorporation on the threshold voltage

V th , the last term of Eq. (4.1) can be rewritten in an approximate form as
q(ND −NF )
2𝜀AlGaN
(dAlGaN )2 , i.e. including the electrostatic effect caused by a uniform
distribution of negative F− -ions N F [39]. Accordingly, it is possible to estimate
the threshold voltage of a fluorine-implanted HEMT for different concentrations
N F in the AlGaN layer, as reported in Figure 4.7. An AlGaN layer with an
Al-concentration x = 30%, an AlGaN thickness dAlGaN = 20 nm, and a barrier
height ΦB = 1 eV have been assumed for such calculation.
Under these conditions, the transition from a normally-on to a normally-off
behavior occurs for a fluorine concentration N F = 1.6 × 1019 cm−3 . This value
corresponds to a sheet carrier density of 3.2 × 1013 cm−2 within the 20 nm thick
barrier layer.
A similar result has been reported by Cai et al. [39] that estimated a F − sheet
concentration of about 3 × 1013 cm−2 to compensate both the AlGaN doping N D
and the polarization-induced charges 𝜎.
In the fluorine-implanted gate technology, the plasma process conditions are
extremely important, as they influence the final performances of the fabricated
device. Cai et al. [39] investigated different CF4 plasma conditions of power
(90–200 W) and duration (20–180 seconds), demonstrating the possibility to
obtain threshold voltage values V th = +0.9 [39, 41]. Figure 4.8 shows the thresh-
old voltage V th as a function of the process duration for different F-based plasma
power conditions, i.e. 90 and 150 W, on two different heterostructures [39, 41].
The threshold voltage V th of the untreated heterostructures is different in the
two cases, depending on their characteristics. However, in both cases, V th
increased with the plasma process duration because of the increasing amount of
incorporated F-ions. On the other hand, an increase of the plasma power enables
the F-ions to penetrate deeper inside the AlGaN layer. Clearly, the negative
immobile charges must be localized as closest as possible to the 2DEG channel,
in order to efficiently deplete the 2DEG and maximize the V th shift. Moreover,
the V th value changes much faster when the plasma power is increased.
An interesting effect observed in fluorine-implanted HEMTs was a 4 orders
of magnitude reduction of the gate current (from 10−2 to 10−6 A/mm) [39]. This
Figure 4.8 Threshold voltage V th as a 2

function of plasma process duration for 90 W (Chen et al. 2011)
F-based plasma treatments performed on 1 150 W (Cai et al. 2006)
two different heterostructures. Source: The Normally-off

data are taken from [39, 41]. 0
Normally-on
–1
–2
–3
–4
0 50 100 150 200

Plasma process duration (s)
latter can be explained with the electrostatic effect of the immobile negative
charges, leading to an upward bending of the AlGaN conduction band and an
additional barrier ΦF to the gate diode current flow in both forward and reverse
bias (see Figure 4.6b).
However, the effect of fluorine treatments on Schottky contacts is controver-
sial. As an example, Chu et al. [42] explained the reduction of the gate leakage
current after CF4 plasma treatment with the formation of an insulating surface
layer, which passivates the surface states and reduces the recombination current
as well as the tunneling current [43]. In addition, the possibility that nonvolatile
F-based compounds are created during the plasma process, and act as an insulat-
ing surface layer blocking the leakage current, was also considered [44]. Indeed,
TEM micrographs of fluorine-treated AlGaN/GaN heterostructures show a con-
siderable amount of crystalline defects [45], thus suggesting the presence of a
highly resistive defect-rich region.
In this context, the effect of fluorine-based plasma process on the electrical
properties of AlGaN/GaN heterostructures has been monitored by means of
local current measurements carried out by conductive atomic force microscopy
(C-AFM) [45]. In this experiment, schematically illustrated in Figure 4.9a–d,
the surface of an AlGaN/GaN heterostructure has been first patterned by a
photoresist hard mask to selectively expose some regions to the effect of a CHF3
plasma and then scanned by means of C-AFM after the photoresist hard mask
removal. As can be seen in Figure 4.9e, the current maps acquired on the sample
surface revealed a significant reduction of the current in the regions exposed
to the CHF3 plasma process. This evidence is consistent with the reduction of
the leakage current observed by other authors and can be related either to a
screening of the 2DEG or to an increase of the local resistance of the AlGaN
because of plasma-induced damage.
The thermal stability is another important concern for the practical implemen-
tation of the fluorine plasma approach in normally-off GaN HEMTs.
In fact, common to all fluorinated GaN HEMTs is a drain–current degradation,
correlated with a deterioration of the 2DEG channel mobility. In this context,
thermal annealing processes at moderate temperatures (e.g. 400 ∘ C, 10 minutes)
Lithography
Photoresist
250 pA Current map
AlGaN AlGaN
2DEG 2DEG
GaN GaN F-treated

region
(a) (b)
Untreated
F-based plasma region
F– F– F– F
–
F
–
F– F– F– AlGaN 20 μm
2DEG
AlGaN
2DEG
GaN
GaN (e)
(d)
(c)
Figure 4.9 (a–d) Schematic of the process sequence used to localize fluorine ions on selective
regions of the AlGaN/GaN heterostructure surface. (e) C-AFM current map of the sample,
showing a different electrical behavior between the fluorine-treated regions and the
untreated ones. Source: Adapted from Greco et al. [45]. The Ref. [45] is an Open Access article
distributed under the terms of the Creative Commons Attribution License. Copyright © 2011
by the authors; licensee Springer.
showed the possibility to repair the plasma-induced damage and obtain a partial
recover (about 76%) of the drain current [39].
The physical mechanism of fluorine incorporation and its stability in
AlGaN/GaN heterostructures is still an open issue because of the complexity of
the scenario [46]. In fact, the thermal annealing processes not only contribute to
the plasma-induced damage recovery but also influence the fluorine distribution
in the heterostructure. In particular, the fluorine diffusion in AlGaN or GaN is
a process assisted by Ga-vacancies [47]. Accordingly, during annealing, fluorine
tends to diffuse in regions with higher concentration of Ga-vacancies, i.e. toward
the AlGaN surface where a large number of defects have been created by the
plasma process [48]. On the other hand, by increasing annealing time, a large
number of Ga-vacancies are annihilated, so that fluorine diffusion stops and the
immobile negative charges are stabilized in the AlGaN layer [48]. Of course,
extended defects, such as dislocations, can affect the diffusion process, making a
precise control of threshold voltage in fluorinated GaN HEMTs difficult [49].
19 +
F -ion implantation, at energies in the order of 10–50 keV and a dose of
1 × 1012 cm−2 , has also been used as an alternative to the plasma processes to
introduce fluorine into the AlGaN/GaN heterostructures [50]. In particular,
a beam energy of 50 keV results in a F-concentration peak located at a depth
of about 64 nm. Lower implantation energies move the F-concentration peak
closer to the AlGaN/GaN interface. Noteworthy, using F-ion implantation
improved the V th stability with respect to the plasma treatment because of the
deeper F-profile reached by implantation [50]. On the other hand, the higher
fluorine concentration obtained by plasma treatments results in a much efficient
depletion of the 2DEG and, then, in a larger positive threshold voltage shift.
Recently, Shen et al. [51] used AlF3 as diffusion source to introduce fluorine in
the gate region in the AlGaN barrier layer. In particular, an AlF3 layer has been
evaporated on the AlGaN surface and annealed in N2 . After annealing, the AlF3
layer was removed by wet etching. In this way, fluorine could reach a diffusion
depth of about 20 nm, thus leading to a considerable positive V th shift from −2.5
to 1.8 V.
The natural evolution of the fluorine implantation technology to obtain
normally-off HEMTs was the combination of the fluorination with a partial
AlGaN gate recession. Liu et al. [52] used a controlled CF4 plasma etch to reduce
the thickness of the AlGaN layer below the gate and simultaneously introduce
immobile negative charge in the gate region. With this approach, it was possible
to obtain a threshold voltage V th = +0.6 V and a reasonable thermal stability of
the normally-off behavior from room temperature to 200 ∘ C.
Clearly, the gate leakage current can represent a serious concern in GaN-based
HEMTs. Hence, the use of an insulator below the metal gate is a common method
to reduce the leakage current, forming the so-called MISHEMTs [53]. However,
in general, the introduction of a gate insulator in MISHEMTs produces a varia-
tion of the threshold voltage with respect to a standard HEMT with a Schottky
gate [54]. Roberts et al. [55] introduced the fluorine incorporation technology
in AlGaN/GaN MISHEMTs, using in situ fluorine doping of an Al2 O3 insulating
layer grown by atomic layer deposition (ALD). In particular, using the in situ flu-
orination of Al2 O3 during the ALD in combination with a fluorine implantation
of the gate region enabled the authors to obtain a V th = +2.35 V.
On the other hand, Zhang et al. [56] demonstrated that a modulation of the V th
is possible by a precise control of the F-plasma AlGaN etching and the thickness
of a gate insulator (Al2 O3 ). In fact, in this approach, the negative charges intro-
duced by the F-treatment compensate the positive charges present in the gate
insulator.
All these issues are crucial in the recessed-gate hybrid MISHEMT technology
and will be better clarified in Section 4.3.3.
4.3.3 Recessed-gate Hybrid MISHEMT

In a recessed-gate hybrid MISHEMT, the AlGaN barrier layer under the gate elec-
trode is removed by means of a plasma etch process, and this recessed region
is subsequently covered by an insulator and a gate electrode. A schematic of
this device is depicted in Figure 4.10, together with the conduction band dia-
gram in the recessed-gate region. The device is often called “hybrid” as it can be
described as the combination of a transistor controlled by a metal insulator semi-
conductor (MIS) gate (i.e. where the AlGaN barrier layer and the 2DEG have been
removed) connected in series with two access regions in which the AlGaN/GaN
heterostructure remains unaltered. Because of the fact that the 2DEG is inter-
rupted in the recessed-gate region, such a transistor is a normally-off structure.
Hence, a positive bias must be applied to the MIS gate in order to accumulate a
negative charge in the recessed region and enable the conduction between the
source and drain electrodes.
Source Drain Dielectric

Gate
EC
AlGaN
Dielectric
2DEG 2DEG
EF
GaN Metal
GaN
gate
(a) (b)
Figure 4.10 (a) Schematic cross section of a normally-off recessed-gate hybrid GaN MISHEMT.
(b) Conduction band diagram in the recessed-gate region.
The first studies highlighting the potentialities and limitations of this device
for power switching applications were reported in the years 2008–2010,
demonstrating recessed-gate hybrid MISHEMTs with V th up to +2 V and
RON < 10 mΩ cm2 using SiO2 as a gate insulator [57–62]. Afterward, several
progresses have been done until Freedsman et al. [63] demonstrated the feasi-
bility of fabricating recessed-gate hybrid MISHEMT on large-area (200 mm) Si
substrates with high breakdown voltage and low leakage current, using an Al2 O3
layer as the gate insulator.
As mentioned before, the insulated gate channel is the main block of
a recessed-gate hybrid MISHEMT. The total on-resistance of the device
RON_(MISHEMT) is given by the sum different contributions:
RON (MISHEMT) = 2RC + RSG (2DEG) + RGD (2DEG) + Rch (4.2)
where RC is the contact resistance of the source/drain electrodes, RSG_(2DEG) and
RGD_(2DEG) are the access resistance contributions, and Rch is the resistance of the
recessed channel region, i.e. where the 2DEG has been removed.
The channel resistance Rch , expressed in Ω mm, can be written as:
Lg
Rch = (4.3)
qnch 𝜇ch
where Lg is the recessed-gate length, nch and 𝜇ch are the accumulated sheet charge
density and the carrier mobility in the recessed channel region, respectively, and
q is the elementary charge. Hence, this resistive contribution depends on the
channel length Lg , and on the sheet charge density and mobility of the electrons
moving in the recessed channel.
Clearly, the removal of the 2DEG in the gate region has a significant impact on
the total device RON_(MISHEMT) . To make this concept clear, it is useful to quantify
the sheet charge density nch accumulated in the recessed-gate channel of the
device. This charge density can be expressed as nch = C ox × (V gs − V th ), where
C ox is the accumulation capacitance of the gate insulator. Considering, as an
example, a 50 nm thick SiO2 gate insulator (with 𝜀SiO2 = 3.9), having an oxide
capacitance C ox = 6.9 × 10−8 F/cm2 , for a gate bias well above the threshold
100 100
Lg = 1 μm
Lg = 2 μm
RON_(MiSHEMT) (Ω mm)
RON_(MiSHEMT) (Ω mm)
80 80
Lg = 3 μm
Lg = 4 μm
60 60
40 40
µch = 100 cm2/(V s)
20 µch = 150 cm2/(V s) 20
µch = 200 cm2/(V s)
µch = 250 cm2/(V s)
0 0
0 2 4 6 8 10 0 100 200 300 400
(a) Gate length, Lg (μm) (b) Channel Mobility, µch (cm2/(V s))
Figure 4.11 Total on-resistance RON_(MISHEMT) of a recessed-gate hybrid MISHEMT calculated (a)
as a function of the gate length Lg for different channel mobility values and (b) as a function of
the channel mobility 𝜇ch for different gate lengths.
voltage (i.e. V gs − V th = 4–6 V), the accumulated sheet charge density nch will be
in the range of 2.7–4.1 × 1012 cm−2 . These values are evidently much lower with
respect to the sheet charge density available in a 2DEG. Hence, the recessed-gate
hybrid MISHEMT is limited in terms of on-resistance and saturation drain
current. Moreover, as described later on, also the mobility of the electrons in
the recessed-gate channel of a MISHEMT is typically lower with respect to the
mobility of the 2DEG.
Hence, for a fixed transistor gate length Lg , a recessed-gate hybrid MISHEMT
is expected to have a higher on-resistance with respect to other technological
solutions not implying the 2DEG removal.
Figure 4.11 reports the total on-resistance of a recessed-gate hybrid MISHEMT
RON_(MISHEMT) calculated as a function of the gate length Lg for different chan-
nel mobility values (a) and as a function of the channel mobility 𝜇ch for differ-
ent gate lengths (b). For this calculation, we assumed a source–drain distance of
10 μm, a contact resistance RC = 0.5 Ω mm, and a semiconductor sheet resistance
RSH = 400 Ω/sq.
As can be seen in Figure 4.11a, for a given channel mobility 𝜇ch , the
RON_(MISHEMT) increases linearly with the gate length Lg . Moreover, for a given
geometry (fixed Lg ) (see Figure 4.11b), the RON_(MISHEMT) becomes lower with
the increase of the channel mobility 𝜇ch , until a saturation is reached for high
electron channel mobility values, i.e. when the recessed channel resistance Rch
becomes less important with respect to the other contributions (RC , RSG_(2DEG) ,
and RGD_(2DEG) ).
The physical and electronic properties of the recessed channel (i.e. roughness
and electronic quality of the insulator/GaN interface) have a significant influence
on the channel mobility 𝜇ch and, hence, on the total on-resistance RON_(MISHEMT)
of the device.
In the literature, several materials have been used as gate insulators in
recessed-gate hybrid MISHEMTs (SiO2 , SiN, Al2 O3 , AlN/SiN, etc.) [53, 64, 65].
250 Figure 4.12 Field effect mobility 𝜇FE as

Al2O3 (Im et al. 2010) a function of V GS –V th for recessed-gate
AlN/SiN (Greco et al. 2017)
SiN (Oka and Nozawa 2008)
hybrid GaN MISHEMTs using SiO2 ,
200 SiO2 (Fiorenza et al. 2017) AlN/SiN, SiN, or Al2 O3 gate insulators.
Field effect mobility
Source: The data are taken from Refs.

µFE (cm2/(V s))
[71–74].
150
100
50
0
–2 –1 0 1 2 3 4
Vg – Vth (V)
The values of the specific on-resistance RON_(MISHEMT) for transistors with gate
lengths in the order of 1–2 μm, extracted in the operation conditions, range
typically between 7.2 and 22 Ω mm [66–70].
Regarding the mobility, the parameter typically used to describe the channel
properties in recessed-gate hybrid MISHEMTs is the field effect mobility 𝜇FE
defined as:
Lg
𝜇FE = g
W Cox VDS m
where Lg and W are the gate length and gate width, respectively, C ox is the specific
capacitance of the gate oxide, V DS is the applied drain-to-source bias, and g m is
the device transconductance.
The values of the maximum field effect mobility reported in the literature for
recessed-gate hybrid GaN MISHEMTs vary in a wide range (∼30–250 cm2 /(V s))
with values of threshold voltage V th between 1 and 2 V [70].
Figure 4.12 reports the field effect channel mobility 𝜇FE of recessed-gate hybrid
GaN MISHEMTs fabricated using SiO2 , AlN/SiN, SiN, and Al2 O3 as gate insu-
lators. As can be seen, the field effect mobility curves 𝜇FE increase with the gate
bias V GS up to a maximum 𝜇FE(peak) and then decrease at high electric field. It can
be seen that the use of Al2 O3 allowed a higher mobility peak (∼225 cm2 /V s) [71],
with respect to AlN/SiN (∼180 cm2 /V s) [72], SiN (∼110 cm2 /V s) [73], or SiO2
(∼110 cm2 /V s) [74]. However, the mobility curves of Al2 O3 /GaN and SiN/GaN
MISHEMTs drop with a steeper slope at higher V gs –V th values. The mobility
behavior with the electric field is typically associated with the occurrence of dif-
ferent scattering mechanisms limiting the current transport at the insulator/GaN
recessed interface.
In this context, monitoring the dependence of the channel mobility on the
temperature can give useful information on the physical mechanisms limiting
the current transport in the recessed-gate MISHEMT. Recently, the field effect
mobility in a recessed-gate hybrid MISHEMTs employing SiO2 as a gate insu-
lator has been studied by Fiorenza et al. [74] starting from the model proposed
by Pérez-Tomás et al. [75–77] for GaN MOSFETs and normally-on MISHEMTs.
Figure 4.13 Peak values of the field 2000

Experimental data
effect mobility 𝜇FE (peak) (experimental Bulk µB
data) as a function of the temperature Surface roughness µSR
Acoustic phonon µac
for a recessed-gate hybrid SiO2 /GaN
µFE (peak) (cm2/ (V s))

µB Coulomb µC
MISHEMT. The experimental data are 1000 Total µTOT
fitted considering all the contributions µSR
of bulk 𝜇B , roughness 𝜇SR , phonon
scattering 𝜇AC , and Colombian
300
scattering 𝜇C . Source: Adapted with µac
permission of Fiorenza et al. [74]. 200 µC
Copyright © 2017, IEEE.
100
µTot
0
300 350 400 450
Temperature (K)
In particular, the different temperature-dependent contributions to the mobility

have been adapted to the specific case of the normally-off recessed-gate hybrid
MISHEMT, in which electrons are induced in the channel when the gate elec-
trode is in accumulation condition [70].
The peak mobility 𝜇FE(peak) as a function of the measurement temperature for
a recessed SiO2 /GaN MISHEMT is reported in Figure 4.13. As can be noticed,
the experimental values of 𝜇FE(peak) slightly decrease with increasing measure-
ment temperature. In analogy to a standard MOSFET, the channel mobility was
expressed including in the Matthiessen’s rule different scattering contributions,
i.e. the bulk mobility factor (𝜇B ), the acoustic phonon scattering (𝜇AC ), the sur-
face roughness scattering (𝜇SR ), and the Coulomb scattering (𝜇C ) due to inter-
face charges [76]. Using the experimental values of interface-trapped charges
(Qtrap = 1.35 × 1012 cm−2 ) and root mean square roughness of the recessed GaN
surface (RMS = 0.15 nm) enabled the authors to obtain a good fit of the experi-
mental temperature dependence of the total mobility 𝜇TOT .
From this analysis, it can be argued that the main limiting factors for the carrier
transport in the channel are the surface roughness scattering (𝜇SR ), the acoustic
phonon scattering (𝜇AC ), and the Coulomb scattering (𝜇C ) contributions. Hence,
the optimization of the dielectric/GaN interface in the recessed channel, i.e. in
terms of roughness and the interface trap density, is a fundamental issue to opti-
mize the mobility and consequently the RON of the device.
The quality of the gate insulator is a crucial issue in recessed-gate MISHEMT
technology also for the stability of the device characteristics. In fact, the presence
of trapping states located in different positions at the insulator/GaN interface
and/or with different energy positions in the bang gap can result into instabilities
of the threshold voltage V th , when the device is subjected to positive or negative
bias stresses. In particular, positive and/or negative stress may vary the amount
of the occupied slow traps [79–81] located at the insulator/GaN interface and/or
in the bulk of the gate insulator [78, 82].
Figure 4.14a shows, as an example, the transfer characteristics of a
recessed-gate hybrid GaN MISHEMT, employing AlN/Al2 O3 as gate insu-
lator, before and after a positive gate bias stress at V G = 12 V for 10 μs [83].
10–2
Before stress T = 150 °C
After stress
EC
10–3
Drain current (A)
EF
VG-stress EV
10–4 EF
metal Al2O3/AlN GaN
10–5
VDS = 0.5 V
VG-stress = 12 V
10–6
0 1 2 3 4
(a) Gate voltage (V) (b)
Figure 4.14 (a) Transfer characteristics IDS –V GS of recessed-gate hybrid Al2 O3 /AlN/GaN
MISHEMTs before and after a positive gate bias stress at 12 V for 10 μs. (b) Schematic
conduction band diagram illustrating the trapping effect occurring under positive bias stress
in the device. Source: Adapted with permission of Acurio et al. [83]. Copyright © 2017,
Elsevier Ltd.
As can be seen, the transfer characteristics of the device measured after the
overvoltage stress of +12 V are notably shifted in the positive direction with
respect to the original ones, with a threshold voltage shift ΔV th = 2.13 V.
This positive shift indicates an accumulation of negative charges (electrons)
inside pre-existing traps in the gate insulator (see schematic representation in
Figure 4.14b). A similar effect was observed when using SiO2 as a gate insulator
[84]. In this case, the stress-induced ΔV th can be fully recovered by applying a
negative voltage, which causes the electron detrapping [84].
These existing trapping states are associated with defects in the insulator, which
can be located in different positions with respect to the GaN interface and/or
can have different energy positions in the band gap. Their characteristic trap-
ping/detrapping time constant 𝜏 depends on the physical distance of the trap
from the allowed states in the semiconductor substrate. A trap located at a dis-
tance x from the insulator–semiconductor interface can be emitted with an inver-
sion electron tunneling process with time constant 𝜏 ≈ 𝜏 0 exp(x/𝜆), where 𝜏 0 is the
characteristic time for the insulator and 𝜆 is the attenuation length that depends
on the insulator/GaN barrier height [85]. Deeper traps (e.g. >1 nm from the semi-
conductor interface) can be charged during prolonged gate stress and may affect
significantly the V th stability during the transistor switching operation.
Interestingly, more pronounced and faster charge trapping processes have been
observed in recessed-gate hybrid MISHEMTs with bilayer gate insulators [72, 83]
with respect to single-layer insulators [78, 84].
Finally, a further progress of the recessed-gate MISHEMT technology was the
implementation of fluorine treatments below the gate region. In fact, as discussed
in Section 4.3.2, F-implantation introduces negative charges that give the possi-
bility to induce a positive shift of the threshold voltage.
In particular, a partial recession of the AlGaN barrier layer by means of a
CF4 -plasma etch in a Al2 O3 -insulated MISHEMT enabled Zhang et al. [56] to
compensate an intrinsic positive charge of the Al2 O3 gate insulator and obtain
V th up to +4 V without affecting the maximum drive current [56].
In conclusion, the recessed-gate hybrid MISHEMT is a very promising
approach to obtain the normally-off operation in GaN transistors. This
approach is attracting great interest, also because of the similarity with the
well-consolidated MOSFET technology. However, in spite of the large number
of research studies reported in the literature, recessed-gate hybrid MISHEMTs
based on GaN have not reached yet an adequate level of maturity to be
introduced into the market.
4.3.4 p-GaN Gate HEMT

In spite of some preliminary work appeared in 2000 [86], the concept of
normally-off AlGaN/GaN HEMT controlled by a p-type GaN gate was clearly
demonstrated by Uemoto et al. [87] only in 2007. Today, the p-GaN gate HEMT
represents the only “true” normally-off solution available on the market. After
the introduction in the market of the low-voltage (100–200 V) products by EPC
(https://epc-co.com/epc), the p-GaN gate technology has attracted the interest
of other important companies (e.g. Panasonic and Infineon), which in some
years were able to fully qualify and commercialize their products up to 650 V,
introducing some important technological breakthroughs [88, 89].
To understand the operation principle of a normally-off p-GaN HEMT, it
is useful to consider the band structure. Figure 4.15a,b reports the schematic
cross section of a p-GaN gate HEMT (a) and the conduction band diagram of a
p-GaN/AlGaN/GaN heterostructure (b).
As can be seen, the introduction of a p-GaN layer on top of the AlGaN/GaN
heterostructure raises the conduction band of the AlGaN above the Fermi level
of an amount of energy comparable with the GaN band gap (3.4 eV), with the
consequent depletion of the 2DEG. Hence, using a p-GaN/AlGaN/GaN system
can enable to achieve a normally-off condition.
Gate
Source p-GaN Drain
ΔEC
AlGaN EC
p-GaN
2DEG
EF
GaN Metal
gate AlGaN GaN
(a) (b)
Figure 4.15 (a) Schematic cross section of a p-GaN gate HEMT and (b) conduction band
diagram of a p-GaN/AlGaN/GaN heterostructure.
In such a structure, the p-GaN gate is defined by a selective plasma etch process
that removes the p-GaN layer from the access regions, until reaching the under-
lying AlGaN barrier.
Although the operation principle of the p-GaN HEMT may appear simple,
introducing a p-GaN layer on top of an AlGaN/GaN heterostructure is not suf-
ficient to ensure the normally-off behavior. Rather, the fabrication of functional
devices depends on several parameters such as the heterostructure properties,
p-GaN etching and doping, gate contact, and thermal annealing [90].
First, an appropriate choice of the heterostructure is essential to efficiently
deplete the 2DEG at the interface. Hence, the acceptor concentration of p-GaN
(N A ), the thickness (dAlGaN ) and Al-concentration (x) of the AlGaN barrier layer,
and the residual donor concentration in both AlGaN and GaN layers must be
appropriately selected to obtain a positive threshold voltage of the device.
Regarding the p-GaN layer, it is well known that the achievement of the p-type
conductivity represented a long-standing problem in GaN technology [91]. Mag-
nesium (Mg) is the reference p-type dopant for GaN, when it replaces Ga in
the nitride lattice and acts as an acceptor. However, because the high ioniza-
tion energy of the acceptors (150–200 meV), it is quite difficult to obtain high
hole concentrations in the p-GaN layer [92, 93]. The common understanding is
that a high doping level of p-GaN (>1018 cm−3 ) is required to obtain an efficient
depletion of the 2DEG.
The effect of the p-GaN doping level on the band structure of a p-GaN/GaN/
AlGaN heterostructure has been recently studied by Efthymiou et al. [94].
In particular, TCAD simulations showed that the threshold voltage V th first
increases for doping levels of p-GaN in the range 1017 –1018 cm−3 and then
slightly decreases for N A > 6 × 1018 cm−3 [94]. In fact, while the positive voltage
needed to modulate the AlGaN/GaN conduction band initially increases with
the p-GaN doping, at sufficiently high doping levels, the depletion region at
the metal/p-GaN interface becomes very narrow, thus enhancing the tunneling
of holes through the barrier. Hence, a further increase of the gate bias does
not result in an additional widening of the depletion region but leads only to
a shift of the AlGaN/GaN conduction band toward the Fermi level. Typically,
p-GaN layers with thickness in the range of 50–100 nm and doping level
N A ∼ 1018 –1019 cm−3 are used in p-GaN gate HEMTs.
The choice of the AlGaN barrier layer thickness (dAlGaN ) and Al-concentration
(x) has a great influence on the electrical behavior of the p-GaN/AlGaN/GaN
heterostructures [90]. To highlight this aspect, Figure 4.16 reports the conduc-
tion band diagrams for some p-GaN/AlGaN/GaN heterostructures, simulated
using a one-dimensional Poisson–Schrödinger solver. For all the simulated cases,
a p-GaN thickness of 50 nm with N A = 6 × 1018 cm−3 was used, with a n-type dop-
ing level of the AlGaN and GaN layers of 1 × 1016 and 1 × 1015 cm−3 , respectively.
Figure 4.16a shows the case of p-GaN/AlGaN/GaN heterostructures with
an Al-concentration x = 20% in the AlGaN layer for two thicknesses of 10
and 20 nm. As can be seen, only in the case of the thinner AlGaN barrier
layer (dAlGaN = 10 nm), the conduction band is raised above the Fermi level
(normally-off condition), whereas for the thicker layer (20 nm), the 2DEG is
still present at the interface. On the other hand, Figure 4.16b shows the case
5 5
EC dAlGaN = 10 nm EC XAlGaN = 0.1
EC dAlGaN = 20 nm EC XAlGaN = 0.2
4 EF
4 EF
3 3
Energy (eV)
Energy (eV)
2 2 u-GaN
u-GaN
1 1
p-GaN p-GaN
0 0
Thinner AlGaN
–1 Thicker AlGaN –1
AlGaN
–2 –2
0 40 80 120 0 40 80 120
(a) Depth (nm) (b) Depth (nm)
Figure 4.16 Simulated conduction band diagrams for p-GaN/AlGaN/GaN heterostructures for
(a) two different values of the AlGaN thickness (10 and 20 nm) or (b) two different values of the
Al-concentration (10% and 20%).
of a heterostructure with an AlGaN barrier layer thickness of 20 nm for two

different values of the Al-concentration x (10% and 20%). In this case, only for
the lower Al-concentration (x = 10%), the conduction band is raised above the
Fermi level (normally-off condition), whereas the 2DEG is still present for an
Al-concentration x = 20%.
An interesting physical effect observed by Uemoto et al. [87] is the injection of
holes from the p-AlGaN gate toward the AlGaN/GaN interface, for sufficiently
high gate bias values, which led to the definition of the device as gate injection
transistor (GIT).
Figure 4.17a schematically explains the operation principle of a normally-off
p-AlGaN/n-AlGaN/GaN GIT. In this device, because of the presence of the
p-AlGaN layer, the 2DEG channel is interrupted below the gate contact. Hence,
a positive gate bias above the threshold voltage V th is necessary to restore the
2DEG at the AlGaN/GaN interface, enabling the conduction in the channel. By
further increasing the gate bias above the built-in voltage of the p-GaN/AlGaN
junction (V F ), holes are injected from p-AlGaN layer toward the interface,
where the 2DEG has been restored. There, to maintain the charge neutrality in
the material, the injected holes are balanced by an equal number of electrons
that reach the 2DEG at the interface. At this point, while electrons move toward
the drain electrode driven by the applied bias, holes remain in the gate region
because of their lower mobility (about 2 orders of magnitude lower with respect
to that of electrons). Such conductivity modulation results in a further increase
of the drain current I D when hole injection starts, while the gate current remains
low. As can be seen, the occurrence of hole injection manifests itself in the
transconductance curve g m by the appearance of a second peak (Figure 4.17b)
[87], which is typically not observed in standard HEMTs.
A drawback encountered in the switching operation of GaN-based GITs is the
increase of the dynamic on-resistance (“current collapse”) at high drain voltages
(>600 V) [95, 96]. In order to overcome this problem, the hybrid drain-embedded
Gate VG = 0 V
250 200
Source Drain
p-AIGaN
i-AIGaN
Drain current (mA/mm)

200
I
i-GaN
gm (mS/mm)
150
I II III
Gate Vth < VG<VF 100
Source Drain
p-AIGaN 100
i-AIGaN
II
i-GaN 50
Gate V G > VF 0 0
Source
p-AIGaN
Drain –2 0 2 4 6
i-AIGaN (b) Gate voltage (V)
III
i-GaN
(a)
Figure 4.17 (a) Schematic explanation of the operation principle of a normally-off

p-AlGaN/n-AlGaN/GaN in gate injection transistor (GIT). (b) Corresponding drain current and
transconductance (gm ) of the devices. Source: Adapted with permission of Greco et al. [90].
Copyright © 2018, Elsevier Ltd.
gate injection transistor (HD-GIT) [97, 98] has been developed. In the HD-GIT,
an additional p-GaN region between the gate and the drain is introduced to
induce a hole injection in the off-state. Noteworthy, in this device, a thicker
AlGaN barrier layer is used in order to maintain the 2DEG below the p-GaN
drain region, while the normally-off operation is ensured by the recession of
the AlGaN barrier layer below the p-GaN gate. This technology allows a hole
injection from the p-GaN drain that can effectively release the trapped electrons
during the switching process [97, 98]. In this way, the dynamic on-resistance
remains low during fast switch also at a high drain voltage (>600 V). More details
on the reliability aspects of the HD-GIT are discussed in Chapter 6.
One of the most important building blocks of the p-GaN gate HEMT is surely
the metal gate. In fact, the electrical properties of the metal/p-GaN barrier can
have a strong impact on the value of the threshold voltage V th and on the device
leakage current [99–101].
In general, the first question is whether a Ohmic or a Schottky contact must
be used as gate electrode on the p-GaN layer. In the original work of Uemoto
et al. [87], a Pd-based Ohmic contact was used as a gate electrode to improve
hole injection and conductivity modulation. However, a clear description of
the metal/p-GaN electrode formation has not been disclosed in that work. As
a matter of fact, obtaining good Ohmic contacts on p-GaN layers is a complex
task, requiring specific annealing processes in oxidizing atmosphere [102],
which in turn can be detrimental for the underlying 2DEG [103, 104]. Hence,
the formation of good Ohmic contacts on p-GaN gate HEMTs is difficult
to implement in the HEMT fabrication flow. For that reason, normally-off
HEMTs with a p-GaN gate typically employ a Schottky metal as a gate electrode.
In this context, TCAD simulations predicted that low work function metals on
p-GaN/AlGaN/GaN systems, which give higher Schottky barrier height ΦB ,
should enable to obtain a higher threshold voltage V th and a lower gate leakage
current [94, 99]. In real cases, the situation is more complex. As an example, the
large threshold voltage difference (ΔV th = 1.8 V) observed between devices using
W-gate (Φm(W) = 4.6 eV, V th = 3.1 V) and Ni-gate (Φm(Ni) = 5.2 eV; V th = 1.3 V)
could not be explained simply by the energy difference between the two work
function values ΔΦm = 0.6 eV [99]. In fact, the thickness of the depletion region
generated at the metal/p-GaN interface increases with decreasing metal work
function (i.e. increasing the barrier height). Then, for a metal gate with a high
barrier height, the positive gate bias will partially drop on the wider depletion
region, thus leading to an increase of V th . On the other hand, for a metal gate
with a lower barrier height, the voltage drop on the thinner depletion region will
be negligible, similarly to what happens in an Ohmic contact, thus leading to a
lower V th . A direct comparison between Schottky (WSiN-based) and Ohmic
(Ni/Au) metal gate has been done by Meneghini et al. [105]. In particular, using
the WSiN-based Schottky gate resulted in an increase of the transistor gate
voltage swing, giving the possibility to reduce the gate leakage current by about
4 orders of magnitude in the on-state. Hence, the use of Schottky gate contact
is typically preferred on p-GaN normally-off HEMTs respect to Ohmic contact
solution.
Table 4.2 reports a collection of literature data on normally-off p-GaN gate
HEMTs. As can be seen, several metals have been investigated as gate contacts for
p-GaN HEMTs [87, 99–101, 105–114]. However, a clear dependence of V th on the
metal work function cannot be deduced because many parameters (semiconduc-
tor surface defects, metal deposition technique, surface preparation, annealing
conditions, etc.) influence the value of the metal/p-GaN barrier height.
Lee et al. [100] showed that metals with lower work function lead to a higher
V th but reduce the output current. TiN is often used as gate contact for p-GaN
because it is widely diffused in silicon technology and has a good thermal sta-
bility in contact with GaN. In particular, TiN can be adopted in a “self-aligned”
approach. In this case, TiN is first patterned and used as a hard mask to remove
the p-GaN layer from the access regions and is then left as gate electrode for the
device. A self-aligned process has also been demonstrated by Lükens et al. [106],
using Mo metal gate encapsulated on SiOx cap layer and AlOx sidewall space as
protection during annealing or dry etching process.
As mentioned before, the definition of the p-GaN gate is obtained by means of
selective plasma etching, i.e. the p-GaN must be removed in the access regions
and remain only under the gate. In a “self-aligned” process, the metal gate is
deposited at the beginning of the fabrication flow (“gate first”), is used as a
hard mask for the p-GaN definition, and is then left as a metal electrode on the
p-GaN. This approach clearly simplifies the device fabrication flow. However,
in this case, the source and drain Ohmic contacts must be fabricated later, thus
typically requiring high annealing temperatures (>800 ∘ C) [115]. Such thermal
budget can lead to thermal reactions between the metal gate and the p-GaN,
which worsen the electrical quality of the barrier. Greco et al. [101] compared a
“gate first” approach, where the Ti/Al metal gate contact is used as a hard mask
Table 4.2 Collection of literature data obtained on normally-off p-GaN HEMTs.
p-GaN p-GaN Threshold Maximum

Metal thickness doping voltage gate bias
gate (nm) (cm−3 ) V th (V) (V) References
Pd 100 (Al15 GaN) N A = 1 × 1018 1 6 [87]

Ni 100 N A = 2 × 1019 1.23 7 [99]
W 100 N A = 1 × 1019 3.0 10 [99]
Mo/Ti/Au 60 N A ∼ 2 × 1018 1.9 10 [100]
18
Ni/Au 60 N A ∼ 2 × 10 1.8 10 [100]
Ti/Au 60 N A ∼ 2 × 1018 1.7 6 [100]
Ti/Al 50 N A = 3 × 1019 1.5 10 [101]
WSiN N.A. N A not given 1.87 11 [105]
19
Mo/Ni 80 N A = 3 × 10 1.08 10 [106]
Ni/Au 50 N A = 3 × 1018 0.48 8 [107]
Pd/Au 70 N A not given 1.0 5 [108]
Ti/Au 70 N A not given 1.2 6 [109]
18
TiN 70 N A = 1 × 10 1.6 10 [110]
TiN 70 N A not given 2.1 9.5 [111]
Ni/Au 95 N A = 3 × 1019 1.5 9 [112]
Ni/Au 70 p = 3 × 1019 1.75 10 [113]
Ni/Au 70 p = 1 × 1018 1.75 8 [114]
for the dry etch of p-GaN and then subjected to annealing for Ohmic contacts
(800 ∘ C), with a “gate last” approach, where the Ti/Al gate contact is defined at
the end of the flowchart [101]. As can be seen from the transfer characteristics
of the devices, shown in Figure 4.18, a V th = +1.5 V has been obtained using
a nonannealed Ti/Al metal gate (“gate last” approach). On the other hand, the
annealing of the metal gate at 800 ∘ C (“gate first” approach) resulted in a negative
V th shift and an increase of the leakage current because of the structural change
of the Al/Ti/p-GaN systems occurring upon annealing. In particular, in the
Al/Ti/p-GaN system, a reduction of the barrier height from 2.08 to 1.60 eV has
been observed after 800 ∘ C annealing [116].
An important figure of merit of p-GaN gate HEMTs is the maximum bias that
can be applied to the gate before breakdown. In fact, under positive gate bias, the
metal/p-GaN junction is reversed biased, thus leading to a further extension of
the depletion region. In these conditions, the electrons coming from the 2DEG
channel through the AlGaN can be injected in the p-GaN layer, where they will
be further accelerated by the high electric filed of the reverse-biased depletion
region. When the electrons moving across the depletion region acquire sufficient
energy from the electric field, the avalanche breakdown occurs. Wu et al. [110]
confirmed this effect, observing that the gate breakdown voltage increases with
Figure 4.18 Transfer characteristics (IDS vs. 104

V GS ) at V DS = +10 V of normally-off p-GaN “Gate last” (not annealed)
gate HEMTs fabricated with “gate last “Gate first” (annealed)
102
approach” (nonannealed) and “gate first
Drain current (mA/mm)

approach” (annealed at 800 ∘ C). Source:
Adapted with permission of Roccaforte et al. 100
[22]. Copyright © 2019 by the authors;
10–2 Vth shift
licensee MDPI, Basel, Switzerland. Ref. [22] is
an open access article distributed under the
terms and conditions of the Creative 10–4
Commons Attribution License. Leakage
10–6 increase
VDS = 10 V
–2 –1 0 1 2 3 4
Gate voltage (V)
12 Uemoto et al. 2007

Hwang et al. 2013
Lee et al. 2016
Greco et al. 2016
10
Maximum gate bias (V)
Meneghini et al. 2017

Lukens et al. 2018
Chiu et al. 2018
Chang et al. 2015
8 Xu et al. 2018
Wu et al. 2015
Tallarico et al. 2017
Tapajna et al. 2016
6 Hao et al. 2016
Jiang et al. 2019
2
0 1 2 3 4
Figure 4.19 Maximum gate bias vs. threshold voltage for different p-GaN gate HEMTs. Source:
The data are taken from Table 4.2.
the temperature. In fact, a positive temperature coefficient of the gate breakdown

is typically observed in avalanche breakdown. Figure 4.19 shows a survey of
literature results for maximum applied gate bias vs. threshold voltage for p-GaN
gate HEMTs.
In normally-off p-GaN gate HEMTs, the removal of the p-GaN layer is typically
obtained by ICP or RIE processes. However, a smooth surface morphology, a low
damage, and a controlled etch rate are required. For that reason, this process can
represent a critical step on the fabrication of normally-off p-GaN gate HEMTs.
Hence, some alternative approaches have been proposed in the literature to avoid
plasma etch of the p-GaN. Figure 4.20 schematically illustrates the standard solu-
tion (p-GaN etch) (a) and two alternative solutions, i.e. selective growth of p-GaN
(b) and H-plasma treatment (c).
The “ideal” solution to avoid plasma etch would be the selective epitaxial
growth of p-GaN in the gate region (Figure 4.20b). Indeed, SAG is mostly used
Standard p-Gan Standard p-Gan Figure 4.20 Schematic cross section

dry etching Gate dry etching of normally-off p-GaN HEMTs
fabricated with (a) standard p-GaN
Source Drain dry etching process, (b) selective area
p-Gan
growth (SAG) of p-GaN, and (c)
AlGaN hydrogen plasma surface treatment
for local Mg deactivation. Source:
GaN Reproduced with permission of Greco
et al. [90]. Copyright © 2018, Elsevier.
(a)
Selective p-Gan
growth
Gate
Source Drain
Hard mask p-Gan Hard mask
AlGaN
GaN
(b)
Hydrogen plasma Hydrogen plasma
treatment treatment
Gate
Source Drain
p-Gan
AlGaN
GaN
(c)
for other technologies, such as the fabrication of nonalloyed Ohmic contacts

[117, 118], the recessed-gate MISHEMT structures [68, 119], or for optoelec-
tronics nanostructures (light emission diodes and laser diodes) [120, 121].
However, this approach presents several limitations, such as the nature and
the geometry of the hard mask [122, 123], the growth temperature [124, 125],
and the reactor pressure [126]. Nevertheless, Yuliang et al. [127] obtained a
positive V th shift from −3.95 to −0.35 V, using the selective MOCVD growth of a
p-GaN layer defined with a SiO2 hard mask. Despite the smooth morphology,
a nonuniform growth along the hard mask sidewall was observed [128]. Clearly,
a further optimization of this process is necessary to obtain lager V th shift
and stable normally-off operation, as well as to achieve thickness and dopant
uniformity in the grown layer.
Another very interesting approach consists in the possibility to replace p-GaN
etch process with a hydrogen treatment of the p-GaN surface (Figure 4.20c). In
fact, hydrogen atoms have the capability to compensate p-type doping in GaN by
the formation of Mg–H neutral complexes, which passivate the Mg-acceptors
in the p-type GaN layer [113]. In this condition, the conduction band of the
p-GaN layer can be selectively pulled below the Fermi level, restoring the 2DEG
at AlGaN/GaN interface in the access regions. The devices fabricated with this
approach exhibited a threshold voltage V th = +1.75 V and a current density of
188 mA/mm at the maximum gate voltage swing of +6 V.
4.4 Other Approaches 163
4.4 Other Approaches

The technological approaches described in Sections 4.2 and 4.3 are the most
common and feasible ones to achieve normally-off operation in GaN-based
HEMTs. However, other routes to obtain normally-off HEMT operation have
been attempted in the literature. Although the obtained results were often far
from a practical application, interesting physical insights can be gained from
those studies.
One of these methods consists in the oxidation of the AlGaN barrier layer in
AlGaN/GaN heterostructures, which has been investigated for several scopes,
such as surface passivation [129, 130] or device isolation [131]. As an example,
Roccaforte et al. [103, 104] demonstrated that a local AlGaN surface thermal
oxidation at 900 ∘ C for some hours can be used for 2DEG isolation, even if the
formed oxidized layer is not thick enough to reach the AlGaN/GaN interface
depth. Clearly, a more accurate control of the localized surface oxidation was
required to tailor the properties of the 2DEG, e.g. by using a rapid oxidation pro-
cesses. Figure 4.21 shows a TEM image of a AlGaN layer subjected to a rapid ther-
mal annealing in O2 at 900 ∘ C for 10 minutes [132]. As can be seen, this process
leads to the formation of a uniform 1.5 nm thick crystalline oxide layer. Litera-
ture data obtained on AlGaN alloys with a similar Al-concentration suggest that
the oxidized layer is composed of Al2 O3 and Ga2 O3 [133–135]. The application of
such a controlled thermal oxidation process to GaN HEMTs resulted in a positive
shift of the threshold voltage V th , although a significant reduction of the device
current prevented its practical used in real devices [132].
Alternatively, Chang et al. [136] demonstrated the possibility to use local
plasma oxidation in the gate region to fabricate normally-off AlN/GaN HEMTs.
In this case, an oxygen plasma treatment with a small RF power range (20–40 W)
was used to convert the surface of the AlN layer into aluminum oxide, reducing
the thickness of the underlying AlN below the critical thickness, similarly to
Figure 4.21 High-resolution TEM

images of an AlGaN layer subjected
to a rapid thermal oxidation at
900 ∘ C for 10 minutes. The
formation of a thin (1.5 nm) oxidized
surface layer is visible. Source:
Reproduced with permission of
Greco et al. [132]. Copyright © 2014, an
ed AlG
IOP Publishing. Oxidiz
AlGan
2 nm
Lithography
O2 plasma
Photoresist
AlN AlN AlN
2DEG 2DEG 2DEG
GaN GaN GaN
(a) (b) (c)
Source Drain Source Gate Drain

Al2O3 AlN Al2O3 AlN Al2O3 AlN
2DEG 2DEG 2DEG
GaN GaN GaN
(d) (e) (f)
Figure 4.22 (a–f ) Fabrication process to obtain normally-off AlN/GaN HEMTs by a local oxygen
plasma treatment.
what happens in a recessed HEMT. This process is schematically illustrated in

Figure 4.22. Clearly, long oxidation treatments lead to a thicker oxide layer in the
gate area and to a further depletion of the 2DEG (more positive V th ). However,
the increase of the channel resistance, caused by the decrease of the 2DEG
density, results in a lower saturation drain current of the devices.
Mizutani et al. [137] used a p-In0.23 Ga0.77 N cap layer, instead of p-GaN, to raise
the conduction band at the AlGaN/GaN interface and to deplete the 2DEG in the
gate region. In this approach, the polarization-induced field in the InGaN cap and
the negative charge in the p-InGaN allowed to achieve a V th = +1.2 V. Moreover, a
reduction of the leakage current is expected in such devices because of the higher
barrier height observed in heterostructures using the InGaN cap layer.
Finally, another interesting possibility to obtain normally-off HEMT is using a
nickel oxide (NiOx ) layer onto AlGaN/GaN heterostructures. Crystalline nickel
oxide has been successfully used as an insulator to reduce the leakage current
in Schottky contacts on HEMTs structures [138, 139]. However, under certain
growth and post-annealing conditions, NiOx layers can exhibit a p-type semicon-
ductor behavior. Hence, this material could be a candidate to replace the p-GaN
in normally-off HEMTs. In particular, Kaneko et al. [140] demonstrated a GaN
HEMT with a V th = +0.8 V by the combination of NiO/Ni/Au gate with a recessed
AlGaN gate region. However, the postannealing condition to obtain p-type con-
ductivity in NiO films is difficult to control and can produce detrimental effects
for the 2DEG channel.
4.5 Summary
In this chapter, the main technological approaches to obtain normally-off GaN
HEMTs were presented. The development of a reliable normally-off HEMT
References 165
technology is a key step toward a massive introduction of GaN transistors in

power semiconductor market.
Normally-on HEMTs can be used in “cascode” with Si-MOSFET to obtain
normally-off chips. However, this configuration presents some drawback related
to the application voltage and the parasitic effects caused by the package
complexity.
The technologies to achieve “true” normally-off HEMTs are based on surface
modification of the AlGaN layer. The recessed-gate approach and the introduc-
tion of fluorine ions in the AlGaN barrier were the first concepts that triggered
the entire research on normally-off HEMTs. These studies clearly showed that
the increase of the threshold voltage and reduction of leakage current required
an additional improvement of the device layout.
Nowadays, the technology is oriented toward two main device concepts, i.e. the
p-GaN gate and the recessed-gate hybrid MISHEMT. The first one is currently the
only “true” normally-off GaN HEMT available on the market, whereas the second
one is intensively studied at R&D level. These approaches exhibit sill several open
issues that must be addressed to optimize the reliability and manufacturability.
Hence, the long-term choice of the normally-off technology will probably depend
on the targeted application, i.e. the p-GaN gate being more suitable for the low-
and medium-voltage applications and the recessed-gate hybrid MISHEMT for
higher voltage applications.
Acknowledgments
The authors would like to thank all the colleagues that contributed to the
research activity on normally-off GaN HEMT in these years: F. Giannazzo, R. Lo
Nigro, S. Di Franco, E. Schilirò, M. Spera (CNR-IMM, Italy), A. Patti, S. Reina
and A. Parisi (STMicroelectronics, Italy), M. Leszczyński, P. Prystawko and P.
Kruszewski (Unipress-PAS, Poland).
This work was partially supported by the Italian Ministry for Education,
University and Research (MIUR) in the framework of the National Project PON
EleGaNTe (Electronics on GaN-based Technologies), ARS01_01007.
References
1 Aisf Khan, M., Bhattarai, A.R., Kuznia, J.N., and Olson, D.T. (1993). High
electron mobility transistor based on a GaN/Alx Ga1−x N heterojunction.
Appl. Phys. Lett. 63: 1214–1215.
2 Iucolano, F. and Boles, T. (2019). GaN-on-Si HEMTs for wireless base sta-
tions. Mater. Sci. Semicond. Process. 98: 100–105.
3 Chen, K.J. and Zhou, C. (2011). Enhancement-mode AlGaN/GaN HEMT
and MIS-HEMT technology. Phys. Status Solidi A 208: 434.
4 Su, M., Chen, C., and Rajan, S. (2013). Prospects for the application of
GaN power devices in hybrid electric vehicle drive systems. Semicond. Sci.
Technol. 28: 074012.
5 Scott, M.J., Fu, L., Zhang, X. et al. (2013). Merits of gallium nitride based
power conversion. Semicond. Sci. Technol. 28: 074013.
6 Asgari, A. and Kalafi, M. (2006). The control of
two-dimensional-electron-gas density and mobility in AlGaN/GaN het-
erostructures with Schottky gate. Mater. Sci. Eng., C 26: 898–901.
7 Ambacher, O., Smart, J., Shealy, J.R. et al. (1999). Two-dimensional elec-
tron gases induced by spontaneous and piezoelectric polarization charges in
N- and Ga-face AlGaN/GaN heterostructures. J. Appl. Phys. 85: 3222–3233.
8 Yu, L.S., Xing, Q.J., Qiao, D. et al. (1998). Ni and Ti Schottky barriers on
n-AlGaN grown on SiC substrates. Appl. Phys. Lett. 73: 3917.
9 Qiao, D., Yu, L.S., Lau, S.S. et al. (2000). Dependence of Ni/AlGaN Schottky
barrier height on Al mole fraction. J. Appl. Phys. 87: 801.
10 Roccaforte, F., Giannazzo, F., Iucolano, F. et al. (2010). Surface and inter-
face issues in wide band gap semiconductor electronics. Appl.Surf. Sci. 256:
5727–5735.
transistors. Appl. Phys. Lett. 77: 250.
12 Vurgaftman, I., Meyer, J.R., and Ram-Mohan, L.R. (2001). Band parame-
ters for III–V compound semiconductors and their alloys. J. Appl. Phys. 89:
5815–5875.
13 Martin, G., Strite, S., Botchkaev, A. et al. (1994). Valence-band discontinuity
between GaN and AlN measured by X-ray photoemission spectroscopy.
14 Chu, R.M., Zhou, Y.G., Zheng, Y.D. et al. (2001). Influence of doping on the
two-dimensional electron gas distribution in AlGaN/GaN heterostructure
transistors. Appl. Phys. Lett. 79: 2270.
15 Di Carlo, A., Sala, F.D., Lugli, P. et al. (2000). Doping screening of polariza-
tion fields in nitride heterostructures. Appl. Phys. Lett. 76: 3950.
16 Baliga, B.J. (2005). Silicon Carbide Power Devices. Singapore: World Scientific
Publising.
17 Kikkawa, T., Hosoda, T., Shono, K. et al. (2015). Commercialization and
reliability of 600 V GaN power switches. Proceedings of IEEE International
Reliability Physics Symposium (IRPS2015), Monterey, CA (19–23 April
2015), pp. 6C.1.1.
18 Lidow, A., Strydom, J., de Rooij, M., and Reutsch, D. (2014). GaN Transis-
tors for Efficient Power Conversion. Chichester, UK: Wiley.
19 Dalcanale, S., Meneghini, M., Tajalli, A. et al. (2017). GaN-based
MIS-HEMTs: impact of cascode-mode high temperature source current
stress on NBTI shift. Proceedings of IEEE International Reliability Physics
Symposium Monterey, CA (2–6 April 2017), pp. 4B.1.1–5.
20 McPherson, J.W. (2018). Brief history of JEDEC qualification standards for
silicon technology and their applicability (?) to WBG semiconductors. IEEE
International Reliability Physics Symposium (IRPS) Burlingame, CA (11–15
March 2018), pp. 3B.1–1.
21 Jones, E.A., Wang, F.F., and Costinett, D. (2016). Review of commercial GaN
power devices and GaN-based converter design challenges. IEEE J. Emerging
References 167
Sel Top. Power Electron. 4: 707–719. https://doi.org/10.1109/JESTPE.2016

.2582685.
of normally-off GaN-based high electron mobility transistors. Materials 12:
1599.
23 Chen, C.H., Keller, S., Haberer, E.D. et al. (1999). Cl reactive ion etching for
gate recessing of AlGaN/GaN field-effect transistor. J. Vac. Sci. Technol., B
17: 2755–2758.
24 Egawa, T., Zhao, G.Y., Ishikawa, H. et al. (2001). Characterizations of
recessed gate AlGaN/GaN HEMTs on sapphire. IEEE Trans. Electron Devices
48: 603–608.
25 Binari, S.C. (1997). GaN FET’s for microwave and high-temperature applica-
tions. Solid State Electron. 41: 177–180.
26 Burm, J., Schaff, W.J., Martin, G.H. et al. (1997). Recessed gate GaN MOD-
FETs. Solid State Electron. 41: 247–250.
27 Aisf Khan, M., Chen, Q., Sun, C.J. et al. (1996). Enhancement and depletion
mode GaN/AlGaN heterostructure field effect transistors. Appl. Phys. Lett.
68: 514–516.
28 Kumar, V., Kuliev, A., Tanaka, T. et al. (2003). High transconductance
enhancement-mode AlGaN/GaN HEMTs on SiC substrate. Electron. Lett.
39: 1758.
approach toward normally off high-voltage AlGaN/GaN HEMT for power
electronics applications. IEEE Electron Device Lett. 53: 356.
30 Palacios, T., Suh, C.S., Chakraborty, A. et al. (2006). High-performance
E-mode AlGaN/GaN HEMTs. IEEE Electron Device Lett. 27: 428.
31 Hao, Y., Chong, W., Ni, J.Y. et al. (2008). Development and characteris-
tic analysis of enhancement-mode recessed-gate AlGaN/GaN HEMT. Sci.
China, Ser. E: Technol. Sci. 51: 784.
32 Adachi, T., Deguchi, T., Nakagawa, A. et al. (2008). High-performance
E-mode AlGaN/GaN HEMTs with LT-GaN cap layer using gate recess tech-
niques. Proceedings of IEEE Device Research Conference, Santa Barbara,
CA, (23–25 June 2008), p. 129.
33 Maroldt, S., Haupt, C., Pletschen, W. et al. (2009). Gate-recessed
AlGaN/GaN based enhancement mode high electron mobility transistors
for high frequency operation. Jpn. J. Appl. Phys. 48: 04C083.
34 Anderson, T.J., Tadjer, M.J., Mastro, M.A. et al. (2010). Characterization of
recessed-gate AlGaN/GaN HEMTs as a function of etch depth. J. Electron.
Mater. 39: 478–481.
35 He, Z., Li, J., Wen, Y. et al. (2012). Comparison of two types of recessed-gate
normally-off AlGaN/GaN heterostructure field effect transistors. Jpn. J. Appl.
Phys. 51: 054103.
36 Pearton, S.J., Shul, R.J., and Ren, F. (2000). A review of dry etching of GaN
and related materials. MRS Internet J. Nitride Semicond. Res. 5: 11.
37 Micovic, M., Tsen, T., Hu, M. et al. (2005). GaN
enhancement/depletion-mode FET logic for mixed signal applications.
Electron. Lett. 41: 1081.
38 Wen, Y.H., He, Z.Y., Li, J.L. et al. (2011). Enhancement-mode AlGaN/GaN
heterostructure field effect transistors fabricated by selective area growth
technique. Appl. Phys. Lett. 98: 072108.
39 Cai, Y., Cheng, Z., Tang, W.C.W. et al. (2006). Control of threshold
voltage of AlGaN/GaNHEMTs by fluoride-based plasma treatment: from
depletion mode to enhancement mode. IEEE Trans. Electron Devices 53:
2207.
40 Cai, Y., Zhou, Y., Chen, K.J., and Lau, K.M. (2005). High-performance
enhancement-mode AlGaN/GaN HEMTs using fluoride-based plasma
treatment. IEEE Electron Device Lett. 26: 435–437.
41 Chen, K.J., Yuan, L., Wang, M.J. et al. (2011). Physics of fluorine plasma
ion implantation for GaN normally-off HEMT technology. IEEE
IEDM Technical Digest, Washington, DC (5–7 December 2011),
pp. 19.4.1–19.4.4.
42 Chu, R., Shen, L., Fichtenbaum, N. et al. (2008). Plasma treatment for leak-
age reduction in AlGaN/GaN and GaN Schottky contacts. IEEE Electron
Device Lett. 29: 297–299.
43 Stringfellow, G.B. (1976). Effect of surface treatment on surface recombina-
tion velocity and diode leakage current in GaP. J. Vac. Sci. Technol., A 13:
908–913.
44 Chu, R., Suh, C.S., Wong, M.H. et al. (2007). Impact of CF4 plasma treat-
ment on GaN. IEEE Electron Device Lett. 28: 781.
45 Greco, G., Giannazzo, F., Frazzetto, A. et al. (2011). Near-surface process-
ing on AlGaN/GaN heterostructures: a nanoscale electrical and structural
characterization. Nanoscale Res. Lett. 6: 132.
46 Yuan, L., Wang, M.J., and Chen, K.J. (2008). Defect formation and annealing
behaviors of fluorine-implanted GaN layers revealed by positron annihilation
spectroscopy. J. Appl. Phys. 104: 116106.
47 Yi, C.W., Wang, R.N., Huang, W. et al. (2007). Reliability of enhancement-
mode AlGaN/GaN HEMTs fabricated by fluorine plasma treatment.
IEEE IEDM Technical Digest, Washington, DC (10–12 December 2007),
pp. 389–392.
48 Wang, M.J., Yuan, L., Chen, K.J. et al. (2009). Diffusion mechanism and the
thermal stability of fluorine ions in GaN after ion implantation. J. Appl. Phys.
105: 083519.
49 Lorenz, A., Derluyn, J., Das, J. et al. (2009). Influence of thermal anneal
steps on the current collapse of fluorine treated enhancement mode
SiN/AlGaN/GaN HEMTs. Phys. Status Solidi C 6: S996–S998.
50 Tadjer, M.J., Horcajo, S.M., Anderson, T.J. et al. (2011). Temperature and
time dependent threshold voltage characterization of AlGaN/GaN high
electron mobility transistors. Phys. Status Solidi C 8: 2233.
51 Shen, F., Hao, R., Song, L. et al. (2019). Enhancement mode AlGaN/GaN
HEMTs by fluorine ion thermal diffusion with high V th stability. Appl. Phys.
Express 12: 066501.
References 169
52 Liu, C., Yang, S., Liu, S.H. et al. (2015). Thermally stable enhancement-mode
GaN metal-isolator-semiconductor high-electron-mobility transistor with
partially recessed fluorine-implanted barrier. IEEE Electron Device Lett. 36:
318–323.
53 Roccaforte, F., Fiorenza, P., Greco, G. et al. (2014). Recent advances on
dielectrics technology for SiC and GaN power devices. Appl. Surf. Sci. 301:
9–18.
54 Roccaforte, F., Fiorenza, P., Greco, G. et al. (2014). Challenges for energy effi-
cient wide band gap semiconductor power devices. Phys. Status Solidi A 211:
2063–2071.
55 Roberts, J., Chalker, P., Lee, K. et al. (2016). Control of threshold voltage
in E-mode and D-mode GaN-on-Si metal-insulator-semiconductor het-
erostructure field effect transistors by in-situ fluorine doping of atomic layer
deposition Al2 O3 gate dielectrics. Appl. Phys. Lett. 108: 072901.
56 Zhang, Y., Sun, M., Joglekar, S.J. et al. (2013). Threshold voltage control
by gate oxide thickness in fluorinated GaN metal-oxide-semiconductor
high-electron-mobility transistors. Appl. Phys. Lett. 103: 033524.
57 Huang, W., Li, Z., Chow, T.P. et al. (2008). Enahancement-mode GaN hybrid
MOS-HEMT with Ron,sp of 20 mΩ cm2 . Proceedings of International Sympo-
sium on Power Semiconductor Devices & ICs, Orlando, USA (18–22 May
2008), pp. 295–298.
58 Tang, K., Li, Z., Chow, T.P. et al. (2009). Enhancement-mode GaN hybrid
MOS-HEMTs with breakdown voltage of 1300 V. Proceedings of Interna-
tional Symposium on Power Semiconductor Devices & ICs, Orlando, USA
(18–22 May 2008), pp. 279–282.
59 Kambayashi, H., Satoh, Y., Ootomo, S. et al. (2010). Over 100 A opera-
tion normally-off AlGaN/GaN hybrid MOS-HFET on Si substrate with
high-breakdown voltage. Solid State Electron. 54: 660–664.
60 Lu, B., Saadat, O.I., and Palacios, T. (2010). High-performance integrated
dual-gate AlGaN/GaN enhancement-mode transistor. IEEE Electron Device
Lett. 31: 990–992.
61 Li, Z. and Chow, T.P. (2011). Channel scaling of hybrid GaN MOS-HEMTs.
Solid State Electron. 56: 111–115.
62 Ikeda, N., Tamura, R., Kokawa, T. et al. (2011). Over 1.7 kV normally-off
GaN hybrid MOS-HFETs with a lower on-resistance on a Si substrate.
Proceedings of the 23rd International Symposium on Power Semicon-
ductor Devices and IC’s (ISPSD2011), San Diego, CA (23–26 May 2011),
pp. 284–287.
63 Freedsman, J.J., Egawa, T., Yamaoka, Y. et al. (2014). Normally-OFF
Al2 O3 /AlGaN/GaN MOS-HEMT on 8 in. Si with low leakage current and
high breakdown voltage (825 V). Appl. Phys. Express 7: 041003.
64 Yatabe, Z., Asubar, J.T., and Hashizume, T. (2016). Insulated gate and surface
passivation structures for GaN-based power transistors. J. Phys. D: Appl.
Phys. 49: 393001.
65 Hashizume, T., Nishiguchi, K., Kaneki, S. et al. (2018). State of the art on
gate insulation and surface passivation for GaN-based power HEMTs. Mater.
Sci. Semicond. Process. 78: 85.
66 Wang, Y., Wang, M., Xie, B. et al. (2013). High-performance normally-off
MOSFET using a wet etching-based gate recess technique. IEEE Electron
Device Lett. 34: 1370.
67 Wang, M., Wang, Y., Zhang, C. et al. (2014). 900 V/1.6 mΩ cm2 normally off
Al2 O3 /GaN MOSFET on silicon substrate. IEEE Trans. Electron Devices 61:
2035.
68 Yao, Y., He, Z., Yang, F. et al. (2014). Normally-off GaN recessed-gate MOS-
FET fabricated by selective area growth technique. Appl. Phys. Express 7:
016502.
69 Hua, M., Zhang, Z., Wei, J. et al. (2016) Integration of LPCVD-SiNx gate
dielectric with recessed-gate E-mode GaN MIS-FETs: toward high perfor-
mance, high stability and long TDDB lifetime. Proceedings of International
Electron Device Meeting 2016, San Francisco, CA (3–7 December 2016),
pp. 260–263.
70 Roccaforte, F., Fiorenza, P., Greco, G. et al. (2018). Emerging trends in wide
band gap semiconductors (SiC and GaN) technology for power devices.
Microelectron. Eng. 66: 187–188.
71 Im, K.-S., Ha, J.-B., Kim, K.-W. et al. (2010). Normally off GaN MOSFET
based on AlGaN/GaN heterostructure with extremely high 2DEG density
grown on silicon substrate. IEEE Electron Device Lett. 31: 192.
72 Greco, G., Fiorenza, P., Iucolano, F. et al. (2017). Conduction mechanisms
at interface of AlN/SiN dielectric stacks with AlGaN/GaN heterostruc-
tures for normally-off high electron mobility transistors: correlating device
behavior with nanoscale interfaces properties. ACS Appl. Mater. Interfaces 9:
35383–35390.
73 Oka, T. and Nozawa, T. (2008). AlGaN/GaN recessed MIS-gate HFET with
high-threshold-voltage normally-off operation for power electronics applica-
tions. IEEE Electron Device Lett. 29: 668–670.
74 Fiorenza, P., Greco, G., Iucolano, F. et al. (2017). Channel mobility in GaN
hybrid MOS-HEMT using SiO2 as gate insulator. IEEE Trans. Electron
Devices 64: 2893.
75 Peréz-Tomàs, A., Placidi, M., Perpiñà, X. et al. (2009). GaN
metal-oxide-semiconductor field-effect transistor inversion channel mobility
modeling. J. Appl. Phys. 105: 114510.
76 Pérez-Tomás, A., Placidi, M., Baron, N. et al. (2009). GaN transistor charac-
teristics at elevated temperatures. J. Appl. Phys. 106: 074519.
77 Pérez-Tomás, A. and Fontserè, A. (2011). AlGaN/GaN hybrid MOS-HEMT
analytical mobility model. Solid State Electron. 56: 201.
78 Fiorenza, P., Greco, G., Giannazzo, F. et al. (2017). Effects of interface states
and near interface traps on the threshold voltage stability of GaN and SiC
transistors employing SiO2 as gate dielectric. J. Vac. Sci. Technol., B 35:
01A101.
References 171
79 Fiorenza, P., Greco, G., Iucolano, F. et al. (2015). Slow and fast traps in
metal-oxide-semiconductor capacitors fabricated on recessed AlGaN/GaN
heterostructures. Appl. Phys. Lett. 106: 142903.
80 Yuan, Y., Wang, L., Yu, B. et al. (2011). A distributed model for border traps
in Al2 O3 InGaAs MOS devices. IEEE Electron Device Lett. 32: 485.
81 Bisi, D., Chan, S.H., Liu, X. et al. (2016). On trapping mechanisms at
oxide-traps in Al2 O3 /GaN metal-oxide-semiconductor capacitors. Appl.
Phys. Lett. 108: 112104.
82 Meneghesso, G., Meneghini, M., De Santi, C. et al. (2018). Positive and neg-
ative threshold voltage instabilities in GaN-based transistors. Microelectron.
Reliab. 80: 257.
83 Acurio, E., Crupi, F., Magnone, P. et al. (2017). Impact of AlN layer sand-
wiched between the GaN and the Al2 O3 layers on the performance and
reliability of recessed AlGaN/GaN MOS-HEMTs. Microelectron. Eng. 178:
42–47.
84 Acurio, E., Crupi, F., Magnone, P. et al. (2017). On recoverable behavior of
PBTI in AlGaN/GaN MOS-HEMT. Solid State Electron. 132: 49.
85 Fiorenza, P., Greco, G., Fisichella, G. et al. (2013). High permittivity cerium
oxide thin films on AlGaN/GaN heterostructures. Appl. Phys. Lett. 103:
112905.
86 Hu, X., Simin, G., Yang, J. et al. (2000). Enhancement mode AlGaN/GaN
HFET with selectively grown pn junction gate. Electron. Lett. 36:
753–754.
(GIT) – a normally-off AlGaN/GaN power transistor using conductivity
modulation. IEEE Trans. Electron Devices 54: 3393–3399.
operations up to 850 V by GaN-GIT utilizing Hole Injection from drain.
Proceedings of the 27th International Symposium on Power Semiconductor
Devices and IC’s (ISPSD2015), Kowloon Shangri-La, Hong Kong (10–14 May
2015), pp. 41–44.
89 Tanaka, K., Morita, T., Ishida, M. et al. (2017). Reliability of
hybrid-drain-embedded gate injection transistor. Proceedings of IEEE Inter-
national Reliability Physics Symposium (IRPS2017), Monterey, CA (2–6 April
2017), pp. 4B-2.1.
90 Greco, G., Iucolano, F., and Roccaforte, F. (2018). Review of technology for
normally-off HEMTs with p-GaN gate. Mater. Sci. Semicond. Process. 78:
96–106.
91 Nakamura, S., Iwata, N., Senoh, M., and Mukai, T. (1992). Hole compensa-
tion mechanism of P-type GaN films. Jpn. J. Appl. Phys. 31: 1258.
92 Kozodoy, P., Xing, H., DenBaars, S.P. et al. (2000). Heavy doping effects in
Mg-doped GaN. J. Appl. Phys. 87: 1832.
93 Roccaforte, F., Frazzetto, A., Greco, G. et al. (2012). Critical issues for
interfaces to p-type SiC and GaN in power devices. Appl. Surf. Sci. 258:
8324–8333.
94 Efthymiou, L., Longobardi, G., Camuso, G. et al. (2017). On the physical

operation and optimization of the p-GaN gate in normally-off GaN HEMT
devices. Appl. Phys. Lett. 110: 123502.
95 Moens, P., Liu, C. Banerjee, A. et al. (2014). An industrial process for 650 V
rated GaN-on-Si power devices using in-situ SiN as a gate dielectric. Pro-
ceedings of IEEE 26th International Symposium on Power Semiconductor
Devices and IC’s (ISPSD), p. 314.
96 Saito, W., Nitta, T., Kakiuchi, Y. et al. (2010). Influence of electric field upon
current collapse phenomena and reliability in high voltage GaN-HEMTs.
Proceedings of International Symposium on Power Semiconductor Devices
and ICs 2010, Hiroshima, Japan (6–10 June 2010), pp. 339–342.
97 Tanaka, K., Morita, T., Umeda, H. et al. (2015). Suppression of cur-
rent collapse by hole injection from drain in a normally-off GaN based
hybrid-drain-embedded gate injection transistor. Appl. Phys. Lett. 107:
163502.
operations up to 850 V by GaN-GIT utilizing hole injection from drain. Pro-
ceedings of International Symposium on Power Semiconductor Devices and
ICs (ISPSD’15), Kowloon Shangri-La, Hong Kong (10–14 May 2015), p. 41
99 Hwang, I., Kim, J., Choi, H.S. et al. (2013). p-GaN gate HEMTs with tung-
sten gate metal for high threshold voltage and low gate current. IEEE
100 Lee, F., Su, L.Y., Wang, C.H. et al. (2015). Impact of gate metal on the per-
formance of p-GaN/AlGaN high electron mobility transistors. IEEE Electron
Device Lett. 36: 232–234.
101 Greco, G., Iucolano, F., Di Franco, S. et al. (2016). Effects of annealing treat-
ments on the properties of Al/Ti/p-GaN interfaces for normally OFF p-GaN
HEMTs. IEEE Trans. Electron Devices 63: 2735–2741.
102 Greco, G., Prystawko, P., Leszczynski, M. et al. (2011). Electro-structural
evolution and Schottky barrier height in annealed Au/Ni contacts onto
p-GaN. J. Appl. Phys. 110: 123703.
103 Roccaforte, F., Giannazzo, F., Iucolano, F. et al. (2008). Two-dimensional
electron gas insulation by local surface thin thermal oxidation in
104 Roccaforte, F., Giannazzo, F., Iucolano, F. et al. (2009). Electrical behavior of
AlGaN/GaN heterostuctures upon high-temperature selective oxidation. J.
Appl. Phys. 106: 023703.
105 Meneghini, M., Hilt, O., Würfl, J., and Meneghesso, G. (2017). Technology
and reliability of normally-off GaN HEMTs with p-type gate. Energies 10:
153.
106 Lükens, G., Hanhn, H., Kalisch, H., and Vescan, A. (2018). Self-aligned pro-
cess for selectively etched p-GaN-gated AlGaN/GaN-on-Si HFETs. IEEE
Trans. Electron Devices 65: 3732–3738.
107 Chiu, H.C., Chang, Y.S., Li, B.H. et al. (2018). High-performance normally
off p-GaN gate HEMT with composite AlN/Al0.17 Ga0.83 N/Al0.3 Ga0.7 N barrier
layers design. J. Electron Devices Soc. 6: 201–206.
References 173
108 Chang, T.F., Hsiao, T.C., Huang, C.F. et al. (2015). Phenomenon of drain
current instability on p-GaN gate AlGaN/GaN HEMTs. IEEE Trans. Electron
Devices 62: 339–345.
109 Xu, N., Hao, R., Chen, F. et al. (2018). Gate leakage mechanisms in normally
off p-GaN/AlGaN/GaN high electron mobility transistors. Appl. Phys. Lett.
113: 152104.
110 Wu, T.L., Marcon, D., You, S. et al. (2015). Forward bias gate breakdown
mechanism in enhancement-mode p-GaN gate AlGaN/GaN high-electron
mobility transistors. IEEE Electron Device Lett. 36: 1001–1003.
111 Tallarico, A.N., Stoffels, S., Magnone, P. et al. (2017). Investigation of the
p-GaN gate breakdown in forward-biased GaN-based power HEMTs. IEEE
112 Ťapajna, M., Hilt, O., Bahat-Treidel, E. et al. (2016). Gate reliability investi-
gation in normally-off p-type-GaN Cap/AlGaN/GaN HEMTs under forward
bias stress. IEEE Electron Device Lett. 37: 385–388.
113 Hao, R., Fu, K., Yu, G. et al. (2017). Normally-off p-GaN/AlGaN/GaN high
electron mobility transistors using hydrogen plasma treatment. Appl. Phys.
Lett. 109: 152106.
114 Jiang, H., Zhu, R., Lyu, Q., and Lau, K.M. (2019). High-voltage p-GaN
HEMTs with OFF-state blocking capability after gate breakdown. IEEE
115 Greco, G., Iucolano, F., and Roccaforte, F. (2016). Ohmic contacts to gallium
nitride materials. Appl. Surf. Sci. 383: 324–345.
116 Roccaforte, F., Vivona, M., Greco, G. et al. (2017). Ti/Al-based contacts to
p-type SiC and GaN for power device applications. Phys. Status Solidi A
214: 1600357.
117 Heikman, S., Keller, S., DenBaars, S., and Mishra, U. (2001). Mass transport
regrowth of GaN for Ohmic contacts to AlGaN/GaN. Appl. Phys. Lett. 78:
2876–2878.
118 Joglekar, S., Azize, M., Beeler, M. et al. (2016). Impact of recess etching and
surface treatments on Ohmic contacts regrown by molecular-beam epitaxy
for AlGaN/GaN high electron mobility transistors. Appl. Phys. Lett. 109:
041602.
119 Zheng, Y., Yang, F., He, L. et al. (2016). Al2 O3 /𝛽-Ga2 O3 (−201) interface
improvement through piranha pretreatment and postdeposition annealing.
IEEE Electron Device Lett. 37: 1193.
120 Yang, J.W., Lunev, A., Simin, G. et al. (2000). Selective area deposited blue
GaN–InGaN multiple-quantum well light emitting diodes over silicon sub-
strates. Appl. Phys. Lett. 76: 273.
121 Puybaret, R., Patriarche, G., Jordan, M.B. et al. (2016). Nanoselective area
growth of GaN by metalorganic vapor phase epitaxy on 4H–SiC using
epitaxial graphene as a mask. Appl. Phys. Lett. 108: 103105.
122 Kato, Y., Kitamura, S., Hiramatsu, K., and Sawaki, N. (1994). Selective
growth of wurtzite GaN and Alx Ga1−x N on GaN/sapphire substrates by
metalorganic vapor phase epitaxy. J. Cryst. Growth 144: 133.
123 Nam, O.H., Bremser, M.D., Zheleva, T.S., and Davis, R.F. (1997). Lateral
epitaxy of low defect density GaN layers via organometallic vapor phase
epitaxy. Appl. Phys. Lett. 71: 2638.
124 Kitamura, S., Hiramatsu, K., and Sawaki, N. (1995). Fabrication of GaN
hexagonal pyramids on dot-patterned GaN/sapphire substrates via selective
metalorganic vapor phase epitaxy. Jpn. J. Appl. Phys. 34: L1184.
125 Marchand, H., Ibbetson, J.P., Fini, P.T. et al. (1998). Fast lateral epitaxial
overgrowth of gallium nitride by metalorganic chemical vapor deposition
using a two-step process. MRS Proc. 537: G4.5.
126 Hiramatsu, K., Nishiyama, K., Motogaito, A. et al. (1999). Recent progress in
selective area growth and epitaxial lateral overgrowth of III-nitrides: effects
of reactor pressure in MOVPE growth. Phys. Status Solidi A 176: 535–543.
127 Yuliang, H., Lian, Z., Zhe, C. et al. (2016). AlGaN/GaN high electron mobil-
ity transistors with selective area grown p-GaN gates. J. Semicond. 37:
114002.
128 Heikman, S., Keller, S., Denbaars, S.P. et al. (2003). Non-planar selective
area growth and characterization of GaN and AlGaN. Jpn. J. Appl. Phys. 42:
6276–6628.
129 Taking, S., MacFarlane, D., and Wasige, E. (2011). AlN/GaN MOS-HEMTs
with thermally grown Al2 O3 passivation. IEEE Trans. Electron Devices 58:
1418.
130 Tajima, M., Kotani, J., and Hashizume, T. (2009). Effects of surface oxidation
of AlGaN on DC characteristics of AlGaN/GaN high-electron-mobility tran-
sistors. Jpn. J. Appl. Phys. 48: 020203.
131 Masato, H., Ikeda, Y., Matsuno, T. et al. (2000). Novel high drain break-
down voltage AlGaN/GaN HFETs using selective thermal oxidation process.
Proceedings of International Electron Devices Meeting, San Francisco, CA
(10–13 December 2000), p. 377
132 Greco, G., Fiorenza, P., Giannazzo, F. et al. (2014). Nanoscale electrical and
structural modification induced by rapid thermal oxidation of AlGaN/GaN
heterostructures. Nanotechnology 25: 025201.
133 Higashiwaki, M., Chowdhury, S., Swenson, B.L., and Mishra, U.K. (2010).
Effects of oxidation on surface chemical states and barrier height of
134 Jang, H.W., Lee, M.K., Shin, H.J., and Lee, J.L. (2003). Investigation of oxygen
incorporation in AlGaN/GaN heterostructures. Phys. Status Solidi C: 2456.
135 Chiu, H.-C., Yang, C.-W., Chen, C.-H. et al. (2011). Characterization of
enhancement-mode AlGaN/GaN high electron mobility transistor using N2 O
plasma oxidation technology. Appl. Phys. Lett. 99: 153508.
136 Chang, C.Y., Pearton, S.J., Lo, C.F. et al. (2009). Development of enhance-
ment mode AlN/GaN high electron mobility transistors. Appl. Phys. Lett. 94:
263505.
137 Mizutani, T., Yamada, H., Kishimoto, S., and Nakamura, F. (2013). Normally
off AlGaN/GaN high electron mobility transistors with p-InGaNcap layer. J.
Appl. Phys. 113: 034502.
138 Roccaforte, F., Greco, G., Fiorenza, P. et al. (2012). Epitaxial NiO gate dielec-
tric on AlGaN/GaN heterostructures. Appl. Phys. Lett. 100: 063511.
References 175
139 Fiorenza, P., Greco, G., Giannazzo, G. et al. (2012). Poole–Frenkel emission
in epitaxial nickel oxide on AlGaN/GaN heterostructures. Appl. Phys. Lett.
101: 172901.
140 Kaneko, N., Machida, O., Yanagihara, M. et al. (2009). Normally-off
AlGaN/GaN HFET using NiOx gate with recess. Proceedings of Interna-
tional Symposium on Power Semiconductor Devices and ICs, Barcelona,
Spain (14–18 June 2009), pp. 25–28.
177
Vertical GaN Power Devices

Srabanti Chowdhury and Dong Ji
Stanford University, Department of Electrical Engineering, 330 Jane Stanford Way, Stanford, CA 94305, USA
5.1 Introduction
This chapter will give the readers a view of the current status of vertical
GaN-based device technology. Vertical devices in GaN have gained a remarkable
momentum with the maturing of single crystalline bulk GaN substrate, often
referred to as “bulk GaN.” In this context, under the material point of view,
the crystallization of bulk GaN material is the most challenging process. As dis-
cussed in detail in Chapter 2, the free-standing GaN substrate quality has shown
dramatic improvement over the past decade delivering a very rich platform
to grow the material up to 6 in., build device, and evaluate their performances.
Two different types of devices will be discussed in this chapter, namely CAVET
and MOSFET (metal oxide semiconductor field effect transistor), with a specific
focus on the regrowth-based trench MOSFET called oxide gate interlayer field
effect transistor (OGFET). Moreover, some of the latest findings on coefficients
of impact ionization that form a very useful addition to GaN’s database will be
reviewed. These fundamental studies are made possible by novel device designs
and the high-quality material to build devices.
5.2 Vertical GaN Devices for Power Conversion

Power conversion is a topic that directly impacts the society with energy sav-
ings and the economy with a staggering US$ 10 billion market (targeted in next
15 years) in power electronics and relies on solid-state devices featuring diodes
and transistors as their basic building blocks.
GaN technology is an ever-expanding topic for R&D, proving its potential to
solve several challenges in power conversion that cannot be addressed by Si.
Medium voltage (650–900 V) devices using the high electron mobility transistor
(HEMT) configuration have been able to reduce form factor at the system level
by driving circuits at higher frequencies (100 kHz–1 MHz) and eliminating heat
sinks or reducing cooling requirements. This alone sparked the interest in GaN

178 5 Vertical GaN Power Devices
device research to address power conversion needs. However, in power conver-

sion, the demand of high current (50 A and higher) from a single chip for a rated
voltage (1 kV and higher) is a standard requirement. Particularly, when the mar-
ket is favorable toward electrification of cars and other means of transportations,
GaN must expand its scope to provide high-power solutions with higher power
density compared to Si and even SiC.
Vertical devices have been the choice of power device engineers for economic
use of the material and maximum use of its physical properties (which allow high-
est possible blocking electric field, field mobility, etc.). Our work has focused on
different flavors of vertical devices such as current aperture vertical electron tran-
sistor (CAVET) [1–6], static induction transistors (SITs) [7, 8], and MOSFETs
with regrown GaN interlayers in the channel [9, 10]. One device cannot fit all
applications primarily because of their maximum current carrying and voltage
blocking limitations. However, in each voltage class, GaN has a promising perfor-
mance metric to offer. Some of the remarkable results we achieved in our group
over the past five years on vertical GaN devices include dispersionless vertical
device in the form of CAVETs to highest channel mobility (185 cm2 /V s) [10] in
MOSFET with 1.4 kV blocking and less than 2.8 mΩ cm2 of on-resistance. On a
separate study, we have demonstrated SIT with triode-like output characteris-
tics that can operate without any p-type GaN or dielectric material, but yet to
understand the voltage range that it can best cover [7, 8].
All these device topologies together make a strong case for GaN in power
conversion and sets a path for other WBG materials such as gallium oxide,
diamond, and aluminum nitride. Last but not the least, the role of fundamental
studies will be discussed. Fundamental studies on materials and devices, starting
from the first discussion on the observation of avalanche electrolumines-
cence (EL) in GaN to the experimental determination of impact ionization
coefficients of electrons and holes in GaN using homojunction structures,
helps set the design windows for high-voltage switches, as well as for other
areas of research [11–14]. In one of our recent study, by using P–N diodes
(PNDs) with avalanche capabilities based on homogeneously grown GaN
structures, impact ionization coefficients of GaN have been experimentally
determined. The extracted hole impact ionization coefficient can be written as
𝛽(E) = 4.39 × 106 exp(−1.8 × 107 /E) cm−1 , and the electron impact ionization
coefficient can be written as 𝛼(E) = 2.11 × 109 exp(−3.689 × 107 /E) cm−1 [12].
5.3 Vertical GaN Transistors

In this section, two distinct types of vertical transistors based on bulk GaN will be
discussed, namely the CAVET and the OGFET – a variant of a MOSFET based
on a trench and regrowth technology.
5.3.1 Current Aperture Vertical Electron Transistor (CAVET)

CAVETs are vertical GaN devices that take full advantage of the polarization-
based two-dimensional electron gas (2DEG) in its design [15]. Composed
of a 2DEG induced channel to carry the current, a CAVET needs a thick
homoepitaxially grown drift region to hold the blocking voltage. There have
been two types of approaches in fabricating CAVETs: (i) planar CAVET [1, 2, 4]
and (ii) trench CAVET [3, 6]. The planar CAVET has an AlGaN/GaN 2DEG
channel in the c-plane, which has high electron density and high mobility. The
trench CAVET adopts the semipolar plane as the channel, the density of the
electrons in the channel depends on the slope of trench sidewalls, which in other
words suggests threshold voltage to be a function of the slope of the trench.
In 2008, the first power GaN CAVET based on bulk GaN substrate was
reported with the current blocking layer (CBL) formed by Mg-ion implantation
[1]. The device process comprises a metal organic chemical vapor deposition
(MOCVD)-grown n-GaN on conductive bulk GaN substrate, followed by a
selective area Mg-ion implantation to form the CBL to block current from the
aperture region other than any other paths. The top AlGaN/GaN structure
in the particular example was regrown by molecular beam epitaxy. Since the
first high-voltage CAVET with switching characteristics with 300 V breakdown
voltage and a Ron,sp of 2.2 mΩ cm2 demonstration by Chowdhury et al. [2], a sig-
nificant progress in the device development was realized with the demonstration
of 1.5 kV junction field effect transistor (JFET) (a variant of CAVET) on bulk
substrate by H. Nie et al. in 2014 from Avogy Inc. [16]. One distinctive feature
of a CAVET is the achievement of dispersionless I–V characteristics because of
buried peak electric field as illustrated in Figure 5.1a,b.
The conventional GaN CAVET typically uses Mg-implanted p-GaN for the
CBL. A major fabrication challenge is posed by the Mg out-diffusion during
the regrowth process. First, the channel resistance can go very high because
of the diffusion of Mg species into the channel, thereby depleting 2DEG. To
prevent the diffusion of Mg into the regrown channel, the regrowth temperature
of the channel in a CAVET needs to be restricted, thereby making it rely
on low-temperature growth processes. Low-temperature MOCVD regrowth
process does not typically result into a high-quality material. MBE regrowth
although has proven to be successful in arresting Mg diffusion, and there are
concerns on the formation of vertical highly conductive paths under metal-rich
VGS = 0 V VGS = 0 V 80 μs VGS = –6 to 0 V, Step = 1 V

G EF 4.0 DC VGS = 0 to –6 V, Step = –1 V
Si3N4
S AlGaN S EC 3.0
ID (kA/cm2)
2°
p-GaN 2.0
2°
n-GaN 1.0
n+-GaN 0.0
D
0 20 40 60 80
(a)
(b) VDS (V)
Figure 5.1 (a) Cross section of an on-state GaN CAVET and the conduction band diagram.
(b) Dispersionless drain characteristics establishing the switching capability of a GaN vertical
device (CAVET) for the first time. Source: Reproduced with permission from Chowdhury et al.
[2]. Copyright 2012, IEEE.
Figure 5.2 Schematic illustrations of the

Si3N4 G trench CAVET.
S AlGaN S
GaN
p-GaN
n-GaN
n+-GaN
D
growth conduction, causing shorts in the device. An MBE-regrown channel

requires an exposure of the sample to the atmosphere for prolonged interval, and
process limitations make it challenging to remove environmental contaminants
from the regrowth interface. In 2007, Chowdhury [17] first noticed the presence
of Si at the regrowth interface. The presence of Si at the interface was carefully
studied and a surprisingly high concentration of Si (∼1013 cm−2 ) was observed
that was removed by UV ozone exposure (to oxidize the elemental Si at surface),
followed by a hydrofluoric acid dip. The presence of Si at the regrown interface
can lead to parasitic channels that are detrimental for the device performance.
Low-temperature flow modulation epitaxy by MOCVD may be considered an
alternative, but it is not yet understood whether the material quality produced in
such a manner that can support high-current and high-voltage devices.
A solution to the aforementioned problems imposed by implanted Mg was pro-
posed by introducing a trench gate structure into the conventional CAVET, as
shown in Figure 5.2 [3, 6]. Instead of using Mg-implanted p-GaN for the CBL,
the trench CAVET adopts MOCVD-grown Mg-doped p-GaN as the CBL mate-
rial. The trench CAVET utilizes the regrown AlGaN/GaN on the trench sidewall
as the channel. The trench sidewall angle determines the polarization scale of
the channel: a 90∘ angle indicates a nonpolar plane, and a 45∘ angle indicates a
semipolar plane. The threshold voltage in a trench CAVET can be adjusted by the
trench sidewall angle. The functioning of a trench CAVET is similar to a trench
MOSFET, except for the AlGaN/GaN channel. In a conventional trench MOS-
FET, the channel is formed by inversion layer between the oxide and the p-type
semiconductor. The electron mobility in the inversion layer is typically below
50 cm2 /V s, limited by the interface roughness scattering. However, in a trench
CAVET, the channel is formed by the 2DEG, which takes the advantage of the
high mobility in the HEMT structure. Ideally, the channel mobility in a trench
CAVET was measured as high as 1690 cm2 /V s [18].
In GaN, a robust oxide technology is not available yet. However, thanks to
the development of GaN HEMT in the past decade, the in situ-grown Si3 N4 on
AlGaN/GaN structure has low interface traps, which is essential to the device
reliability. A trench CAVET also takes the advantage of the matured HEMT gate
dielectric technology to suppress gate leakage.
The first metal–insulator–semiconductor (MIS) gate trench CAVET was
reported by Ji et al. in 2016 on bulk GaN substrates [3]. The scanning electron
microscope (SEM) image as shown in Figure 5.3 shows a buried P–N junction
inside a trench CAVET (Figure 5.3). The device had a breakdown voltage of
225 V limited by the gate-to-drain breakdown. By using a thicker gate dielectric
Figure 5.3 Side profile of the MIS-gate trench

CAVET from SEM. Source: Reproduced with
permission from Ji et al. [3]. Copyright 2017, IEEE.
Gate metal
p-GaN
Regrown layers
p–n diode
n-GaN drift region

4.0kV 12.0mm ×25.0 k SE (U) 200 μm
Figure 5.4 Double-swept ID –V GS transfer 104

characteristics of the fabricated trench CAVET.
Source: Reproduced with permission from Ji ID
101
ID, IG (A/cm2)
et al. [6]. Copyright 2018, IEEE.
Ion/Ioff = 2 × 108
10–2 Vth = –21 V
Hysteresis = 0.125 V
10–5
IG
10–8
–25 –20 –15 –10 –5 0
VDS (V)
Figure 5.5 (a) ID –V DS 2000

characteristics of the fabricated VGS = 0 to –22 V 24
trench CAVET and (b) analysis of Trench
Resistance (Ω mm)
1500
contact resistance. Source: S 22
ID (A/cm2)
Reproduced with permission

Slope = Rsh/Z
from Ji et al. [6]. Copyright 2018, 1000 20
IEEE. 22 μm = 0.27 Ω mm
18 2RC = 18.88 Ω mm
500
16
0
0 1 2 3 4 5 0 5 10 15 20 25
(a) VDS (V) (b) d (μm)
layer and a low damage RIE trench etch, a high breakdown voltage of 880 V was
obtained in 2018 [6]. A low hysteresis of ∼0.12 V in the transfer characteristics
was achieved by using high-quality in situ MOCVD-grown Si3 N4 gate dielectric.
Figures 5.4–5.6 depict the trench CAVET performance.
The p-GaN gate structure is widely used in normally-off lateral GaN HEMTs.
Because of the high electron density of 2DEG induced by the polarization charge,
the threshold voltage of p-GaN-gated HEMT is typically less than 2 V. However, a
more positive threshold voltage can be realized in a p-GaN-gated trench CAVET
with the channel formed at the semipolar plane while blocking over 1.7KV. This
was first demonstrated by Shibata et al. [18]. Based on a P–N epitaxial struc-
ture on the bulk GaN substrate, a “V”-shaped trench was formed using induc-
tively coupled plasma (ICP) etching, the p-GaN/AlGaN/GaN triple layers were
102 Figure 5.6 Off-state characteristics

of the fabricated trench CAVET.
Source: Reproduced with permission
from Ji et al. [6]. Copyright 2018, IEEE.
ID (mA/cm2)
101
G-D breakdown
100
0 200 400 600 800 1000

VDS (V)
Figure 5.7 Trench CAVET with a p-GaN

p-GaN G gate layer.
S S
GaN
p-GaN
n-GaN
n+-GaN
regrown over the trench by MOCVD as shown in Figure 5.7. Because the chan-
nel locates at the semipolar plane instead of the c-plane, the threshold voltage
shifted toward the positive side by 1.5 V. A high positive threshold voltage of 2.5 V
was demonstrated with 1700 V blocking capability on a 13 μm-thick drift region
offering a low specific on-resistance of 1 mΩ cm2 .
5.3.2 Vertical MOSFETs

Vertical GaN MOSFETs are the other branch of devices, showing a promising
performance by offering a normally-off solution, thereby overcoming a significant
drawback of any GaN HEMT-based design.
There are three types of vertical MOSFETs reported so far: (i) nonregrowth-
based trench MOSFET [19–23]; (ii) regrowth-based trench MOSFET
[9, 10, 24–27]; and (iii) fin MOSFET [28–30]. In this chapter, a regrowth-based
trench MOSFET termed as OGFET is discussed.
Figure 5.8 shows the structure of a vertical GaN MOSFET. A key feature of
these devices, the P–N junction between source and drain, is formed by p-base
and n-drift regions. The device breakdown voltage is determined by the reverse
characteristics of the main P–N junction. A n+ source region is created partially
on top of the p-base region, while the junction between the n+ source region and
the p-base region is connected to the source contact to improve the breakdown
voltage by eliminating the N–P–N open base effect. The channel located at the
etched sidewall is formed by the inversion layer of the MOS structure.
Figure 5.8 Structure of GaN

MOSFET (a) and an S Oxide S S
Oxide
S
G G
OGFET (b).
p-GaN p-GaN
n-GaN Regrown GaN layer

n-GaN
n+-GaN n+-GaN
D D
(a) (b)
Compared to the CAVET structure, there are two basic advantages of the
MOSFET: (i) the MOSFET is a reliable normally-off device with a high threshold
voltage over 2 V and (ii) regrowth can be avoided, which makes the process less
challenging, reducing the cost and turnaround time. Such advantages of the
MOSFET make it an attractive design for vertical GaN transistors. However, for
GaN MOSFET, one of the critical challenges lies in the channel electron mobility
of the device. During on-state, the electrons flow through the inversion layer
of the sidewall MOS structure, the channel electron mobility is limited by the
surface roughness and impurity scattering. Another issue along with the poor
channel property is the device reliability. The GaN MOSFET cannot be widely
recognized without a strong reliability.
GaN OGFET is a modified structure based on the conventional trench MOS-
FET. Compared to the conventional trench MOSFET, the OGFET has two fea-
tures: (i) an unintentional doped (UID) GaN interlayer is used as the channel
region, which enhances the channel electron mobility to reduce the Coulomb
scattering by the dopants and (ii) the oxide is in situ grown by MOCVD, which
reduces the interface states and improves the gate oxide reliability. The novelty
of OGFET is enhancing the channel electron mobility without sacrificing the
normally-off behavior.
In 2016, Gupta et al. reported the first OGFET results based on the sapphire
substrates; the device showed a 60% Ron,sp reduction while maintaining the a
threshold voltage >2 V [24]. In 2017, Gupta et al. demonstrated an OGFET
on bulk GaN substrates with a breakdown voltage of 990 V and a low Ron,sp of
2.6 mΩ cm2 [25]. In the same year, Ji et al. demonstrated a high-performance
OGFET with a breakdown voltage over 1.43 kV with a low Ron,sp of 2.2 mΩ cm2
[10]. The I–V characteristics of the fabricated OGFET are shown in Figure 5.9.
Figure 5.9 I–V characteristics of fabricated OGFET 1200

with a saturation current density of 850 A/cm2 and VGS = 0 – 18 V, Vstep = 3 V
Ron,sp of 2.2 mΩ cm2 . Source: Reproduced with 900 Ron = 2.2 mΩ cm2
ID (A/cm2)
permission from Ji et al. [10]. Copyright 2017, IEEE.

600
300
0
0 2 4 6 8
VDS (V)
Figure 5.10 Transfer characteristics of the fabricated

ID (A/cm2), IG (A/cm2) 101 Vth,up = 4.7 V
Vth,down = 5.0 V OGFET (black curves) and gate leakage (red curves).
ΔVth = 0.3 V
ID
10–1 The threshold voltages of sweep up and down were
4.7 and 5 V (defined at a current level of 10−4 A/cm2 ).
10–3 10–4 A/cm2 The subthreshold slope was 283 mV/decade
measured from 10−5 to 10−2 A/cm2 . Source:
10–5 IG Reproduced with permission from Ji et al. [10].
–7 Copyright 2017, IEEE.
10
0 2 4 6 8 10 12
VGS (V)
G 100
S S
50 mA/cm2
Mesa Mesa
ID (mA/cm2)
10
n– -GaN VBR = 700 V

1
n+-GaN
Bulk GaN substrate
Drain 0.1
0 350 700
No field plate VDS (V)
(a) (b)
100
G 50 mA/cm2
S S
ID (mA/cm2)
FP FP 10
VBR = 1015 V
n– -GaN 1
n+-GaN
Bulk GaN substrate 0.1
Drain 0 400 800 1200
Single field plate VDS (V)
(c) (d)
100
FP2
SOG S n+
+ FP1 50 mA/cm2
Gate p GaN
(mA/cm2)
10
VBR = 1435 V
n– -GaN
1
ID
n+-GaN
Bulk GaN substrate

Drain 0.1
0 500 1000 1500
Double field plates VDS (V)
(e) (f)
Figure 5.11 (a) The device structure of the fabricated OGFET with no field plate (mesa etch
only); (b) the off-state characteristic of (a); (c) the device structure of the OGFET with a single
field plate; (d) the off-state characteristic of (c); (e) the device structure of the OGFET with
double field plates; and (f ) the off-state characteristic of (e). With the help of the double field
plate structure, a 1.4 kV breakdown voltage was achieved. Source: Adapted with permission
from Li et al. [10]. Copyright 2017, IEEE.
5.4 High-Voltage Diodes in GaN 185
Figure 5.10 shows the transfer I D –V GS characteristics and the gate leakage. The
threshold voltage, V th , defined at a current level of 10−4 A/cm2 (I on /I off = 106 ),
obtained was 4.7 V (when V GS sweeps up). A clockwise hysteresis of ΔV th of
0.3 V was observed. Low subthreshold slope of 283 mV/decade was measured
from I D = 10−5 to 10−2 A/cm2 . The off-state characteristics of the fabricated
devices are shown in Figure 5.11. The enhancement in the breakdown voltage
was obtained by the use of a field plate structure that mitigated the electric field
at the corner of the etched pillars. The peak electric field was gradually lowered
with addition of field plates as depicted in Figure 5.11.
5.4 High-Voltage Diodes in GaN

SiC high-voltage diodes in the form of Schottky barrier diodes (SBD), junction
bipolar Schottky (JBS), and PiN diodes are highly implemented beyond Si in
power applications with a growing market in discrete parts.
GaN diodes present similar potential to serve the future power electronics mar-
ket, particularly in integrated power modules. The Schottky barrier in GaN is
consistently reported to be between 0.9 and 1 V because of Fermi-level pinning,
which allows Schottky diodes with turn-on voltage between 1 and 2 V based on
a particular design. P–i–N diodes in GaN are of high interest because of their
well-behaved turn-on voltage (around 3 V) and good forward current density.
In 2013, Kizilyalli et al. reported vertical P–N diodes fabricated on pseudobulk
gallium nitride (GaN) substrates [31].
The blocking voltage in a P–N diode is determined and designed by the
n-type drift layer doping and the thickness. Typical drift layer thicknesses for
600 to 1.7 kV diodes are between 6 and 20 μm with targeted doping densities
of ND ≈ (1–3) × 1016 cm−3 . The p+ region is achieved by in situ growth of
Mg-doped ((5–10) × 1019 cm−3 ) GaN epitaxial layer on top of the n-type GaN
epitaxial drift region. Hole concentration of (5–10) × 1017 cm−3 and a hole
mobility of between 10 and 50 cm2 /V s at 25 ∘ C are normally expected from Hall
effect measurements. Appropriate edge termination techniques are required for
high-voltage performance.
In their work, Kizilyalli et al. measured devices demonstrating breakdown volt-
ages of 2600 V with a differential specific on-resistance of 2 mΩ cm2 . This perfor-
mance placed GaN diodes beyond the SiC theoretical limit on the power device
figure of merit chart as shown in Figure 5.12a adopted from Kizilyalli et al.’s work.
Moreover, these diodes demonstrated avalanche capability contrary to common
belief. That GaN devices do not avalanche was a common belief developed largely
by the lack of avalanching in unipolar GaN HEMTs. The temperature coefficient
of the breakdown voltage was reported positive, showing that the breakdown
is indeed because of impact ionization and avalanche. In their work, the GaN
diodes were driven into avalanche breakdown using 30 mS and 15 mA current
pulses demonstrating avalanche energy capability of 1000 mJ. The reverse recov-
ery time of the vertical GaN P–N diode was not noticeable, as the authors sug-
gests “because it was limited by capacitance rather than minority carrier storage,
1000
it
lim
Si
100
RDS,ON (mΩ cm2)
it
lim
SiC
10
it
lim
N
1 Ga
0.1
100 1000 10 000
Breakdown voltage (V)
(a) (b)
Figure 5.12 (a) Power device figure-of-merit comparison of P–i–N diodes. (b) Double-pulse
testing of vertical GaN P–N diode and a high speed 1200 V rated Si diode. Source: Reproduced
with permission from Kizilyalli et al. [31]. Copyright 2013, IEEE.
and because of this, its switching performance exceeds the highest speed sili-
con diode.” A snapshot of the GaN diode characteristics reported in by the same
team is shown in Figure 5.12b. Very recently, Ohta et al. [32] demonstrated 4.9 kV
breakdown voltage on GaN-on-GaN vertical P–i–N diodes.
5.5 Avalanche Electroluminescence in GaN P–N Diodes

Harnessing the full potential of GaN in the power electronics domain necessi-
tates the achievement of near-ideal breakdown operation, with limitations solely
imposed by intrinsic material properties. Devices must be fabricated such that all
sources of premature breakdown – dislocations, impurities, and interfaces – are
mitigated. This requires advancements in epitaxial growth targeted toward dis-
location minimization, control of ionized and nonionized impurities, and device
fabrication that requires engineering of electric field and maintenance of pristine
interfaces. Recent availability of bulk GaN substrates with low dislocation densi-
ties has made it possible to reach closer to the GaN breakdown field limit. These
high-quality substrates have been used to demonstrate avalanche breakdown in
GaN P–N junctions – which marks an intrinsic material junction breakdown
instead of a premature surface breakdown.
In the work described below, a simple low-power etch technique is used to
achieve avalanche breakdown in GaN P–N diodes using curved shaping of the
mesa edge profile. Following a P–N junction grown on bulk GaN substrate, and
post-Mg activation anneal, an optimum etch recipe was developed and used to
achieve the curved mesa profile as shown in Figure 5.13.
The curved edge prevented premature spike of electric field because of
the absence of any sharp corners and allowed the device to reach intrinsic
breakdown field and result in impact ionization. To verify the nature of break-
down, temperature-dependent reverse IV were measured, which showed an
increase of breakdown voltage with temperature – a trait synonymous with
impact ionization-induced carrier multiplication. The peak electric field from
5.5 Avalanche Electroluminescence in GaN P–N Diodes 187
20
Concentration (atoms/cm3)
10
19
200 nm p-GaN (Mg: 5 × 10 cm–3)
18
1.2 μm n-GaN (2 × 1017 cm–3) 10
1016
1 μm n+-GaN buffer (1 × 1018 cm–3)
Si
14
10 Mg
+ 18 –3
n -GaN substrate (2 × 10 cm )
10
12 500 nm
0 400 800 1200 1600 2000
(a) (b) Depth (nm) (c)
Figure 5.13 GaN P–N diode with curved mesa edge: (a) the epitaxial layers of the device
grown by MOCVD; (b) doping concentration obtained by secondary ion mass spectroscope
(SIMS); and (c) the SEM image of the etched mesa edge. Source: Reproduced with permission
from Mandal et al. [11]. Copyright 2018, IOP Publishing.
16 000
14 000 Reverse current = 100 μA
12 000 Reverse current = 1 mA
Intensity (a.u.)
Reverse current = 10 mA
10 000
8000
6000
4000
2000
0
3.6 3.2 2.8 2.4 2.0 1.6 1.2
(a) (b) Emission energy (eV)
Figure 5.14 (a) Reverse breakdown electroluminescence of the fabricated GaN P–N diode and
(b) measured spectrum of the electroluminescence. Source: Reproduced with permission from
Mandal et al. [11]. Copyright 2018, IOP Publishing.
triangular field approximation was estimated to be 3 MV/cm, which is close to

the theoretical breakdown limit of GaN.
One of the key characteristics of the P–N diodes is electroluminescence
(EL) during avalanche breakdown. At the onset of avalanche, an emission was
observed along the mesa edge, as shown in Figure 5.14a. The emission continued
as the diode was biased into avalanche. To measure the EL emission spectrum,
a pristine device was biased by using a voltage sweep and a constant current
compliance. The constant current method ensured that the diode does not burn
off because of Joule heating, which is a threat during an uncontrolled avalanche
breakdown event. The EL spectra of the diode under the constant current are
shown in Figure 5.14b. Three different current compliances – 100 μA, 1 mA,
and 10 mA – were used for the measurement. Two distinct spectral wavelength
ranges were detected. Multiple high-intensity peaks at and below the bandgap of
GaN (Figure 5.14b) were observed. The intensity of the peaks increased with the
reverse current flowing through the diode.
Photonic emission during avalanche breakdown has been reported earlier in
Si and GaAs material systems. The source of photons in those systems has usu-
ally been attributed to (i) band to band recombination (BTBR) of electron–hole
pairs, (ii) bremsstrahlung or braking radiation, and (iii) indirect interband emis-
sion [33]. It has also been shown that, of these mechanisms, only the radiation
due to BTBR scaled with reverse current. Similar scaling is observed in the GaN
avalanche diodes as shown in Figure 5.14b, which can be explained as follows.
As the reverse-bias current limit is increased, the field in the depletion region
increases, thereby generating more carriers by impact ionization. The increased
carrier concentration results in higher number of recombination processes in the
depletion region. Hence, the intensity of BTBR increases as the reverse current
is increased from 100 μA to 1 and 10 mA.
The broad-range wavelength emissions between 1.9 and 2.7 eV (Figure 5.14b)
occurred because of sub bandgap recombination of some of the electrons travel-
ing in the drift layer. Broad-range emissions in p-type GaN have been attributed
to carbon impurity levels (CN ) [34], gallium vacancies (VGa ), or nitrogen vacancies
(VN ) [35]. Hence, it can be inferred that the emission between 1.9 and 2.7 eV can
occur via multiple trap levels and is not limited to any particular type of defect. It
has also been observed that plasma etch damage during RIE induces VN centers
in GaN [36]. The VN center has an emission peak of around 2.18 eV [35], which
is very close to the emission peak observed here. Therefore, etch-induced VN in
the mesa sidewalls could also have allowed sub-bandgap transitions, resulting in
a broad-range wavelength spectrum. The broad-range emission is therefore vis-
ible along the mesa edge, as shown in Figure 5.14a. This was further verified by
measuring devices with small metal contact area and large metal-to-mesa-edge
distance. The electroluminescence was primarily observed from the mesa periph-
ery, confirming the emission from the etched sidewall region. As the reverse
current is increased, the number of carriers generated increases, thereby leading
to more recombination along the mesa edge. This increases the emission intensity
with the increase of reverse current.
5.6 Impact Ionization Coefficients in GaN

Ji et al. using the structures depicted in Figure 5.15 to extract the impact ioniza-
tion coefficients of electrons and holes in GaN experimentally [12].
The approach is summarized below:
(i) Two approximations were made: (i) 𝛽 ≫ 𝛼 (when electric field <2.5 MV/cm,
according to Monte Carlo simulated data [37]) and (ii) the electric field in
the N–P diode (NPD) structure is uniformly distributed because of the low
doping density in the NPD.
(ii) NPD was used to obtain 𝛽 under assumption stated in point (i). The impact
ionization integration equation can be simplified as
W
1 x
1− = 𝛽e− ∫0 𝛽dx′
dx (5.1)
Mp ∫ 0
The 𝛽 can be written as
ln Mp
𝛽= (5.2)
W
(iii) PND’s photocurrent–dark current would have impact on both coefficients
(𝛽 and 𝛼).
Figure 5.15 Device

structures for experimental UV UV
determination of impact illumination illumination
Cathode
ionization coefficients in Anode
GaN: (a) the N–P diode +-GaN p+-GaN
n
(NPD) and (b) the P–N diode Anode n- GaN
(PND). Source: Reproduced i- GaN ([Si]:2e17 cm–3)
with permission from Ji et al.
[12]. Copyright 2019, AIP n+-GaN
p+-GaN
Publishing. GaN substrate
GaN substrate Cathode
(a) (b)
(iv) Because 𝛽 was obtained in point (ii), the electron’s impact ionization coeffi-
cient 𝛼 can be written as a function of both 𝛽 and Mn .
qND 1 dMn
𝛼= − (Mn − 1)𝛽(Em ) (5.3)
𝜀0 𝜀s Mn dEm
(v) With the obtained experimental data, 𝛽 and 𝛼 can be projected over the
whole electric field using linear extrapolation. This fits to the Chynoweth’s
law [38], which is an important step. The projected 𝛽 and 𝛼 can be written as
7 1
𝛽(E) = 4.39 × 106 e−1.8×10 E cm−1 (5.4)
7 1
α(E) = 2.11 × 109 e−3.689×10 E cm−1 (5.5)
(vi) These 𝛼 and 𝛽 are fed to obtain the photocurrent over all voltages. The
full multiplication equations (see Eqs. (5.6) and (5.7)) in Matlab carries no
assumption (stated in point (i)) using the extrapolated 𝛼 and 𝛽 as obtained
in point (v).
W
1 x
𝛽e− ∫0 (𝛽−𝛼)dx dx
′
1− = (5.6)
Mp ∫ 0
W
1 W
1− = 𝛼e− ∫x (𝛼−𝛽)dx′
dx (5.7)
Mn ∫0
(vii) Multiplication factor is thus obtained as a function of electric field and
thus voltage. Now using this multiplication factor, the photocurrent can be
reproduced, which matches the experimental data very well, proving that
the assumption led to accurate 𝛼 and 𝛽 values.
To measure the hole-initiated impact ionization multiplication, a UV laser
with a wavelength of 350 nm was used to generate electron–hole pairs in the
top n+ -GaN layer in the NPD. For the UVL with a wavelength of 350 nm, the
absorption coefficient in GaN is about 8 × 104 cm−1 [39]. The thicknesses of
top n+ -GaN and p+ -GaN layers in NPD and PND are 200 nm, ensuring that
the UVL is absorbed in the top layers. When the NPD was reverse-biased,
only UVL-generated holes were swept into the space charge region, while the
UVL-generated electrons were collected by the cathode [40]. The UVL-generated
holes passed through the high electric field region and gained enough kinetic
energy to knock a bound electron out of the valence band to the conduction
band, creating an electron–hole pair by the impact ionization process. By
measuring the current under high reverse bias voltage close to the breakdown
voltage, the multiplication of the hole was observed, and the hole’s impact
ionization coefficient was calculated.
The PND was used to measure the electron-initiated impact ionization
multiplication and coefficient. Under the UV illumination, UVL-generated holes
were collected by the anode, and the UVL-generated electrons were swept into
the space charge region to initiate the impact ionization process. The electron’s
impact ionization coefficient was obtained by measuring the reverse current due
to the avalanche.
Figure 5.16a shows the reverse characteristics of the NPD under dark condition
and UVL, respectively. Under a high electric field, the hole-initiated multipli-
cation (Mp ) can be obtained by (I UV − I dark )/(I UV,Init ), where I UV,Init is the initial
photocurrent. The hole-initiated multiplication as a function of the electric field
is shown in Figure 5.16b; when the electric field is over 1.5 MV/cm, carrier mul-
tiplication occurs.
The reverse characteristics of the fabricated PND under dark condition and UV
illumination are shown in Figure 5.16c. The initial current was determined by
the linear extrapolation method to consider the increasing depletion layer under
reverse bias [41]. The electron-initiated multiplication is shown in Figure 5.16d.
10–3 102
Photocurrent
10–4
Initial unmultiplied current
10–5
I (A)
Mp
101
10–6
10–7 Dark current
10–8 100
100 200 300 1 × 106 2 × 106 3 × 106
Electric field (V/cm)
(a) Voltage (V) (b)
10–3 102
10–4
Current (A)
Photocurrent
10–5 101
Mn
Initial
10–6 unmultiplied
current
Dark current
10–7 100
50 100 150 200 3.0 × 106 3.5 × 106 4.0 × 106
(c) Voltage (V) (d) Em (V/cm)
Figure 5.16 Measured multiplication by UVL-assisted method: (a) I–V characteristics of the
NPD measured under UV illumination and dark, respectively; (b) the measured hole-initiated
multiplication as a function of the electric field; (c) I–V characteristics of the PND measured
under UV illumination and dark condition, respectively; (d) the measured electron-initiated
multiplication as a function of the electric field. Source: Reproduced with permission from Ji
et al. [12]. Copyright 2019, AIP Publishing.
104 105
103 300 K
α (cm–1)
β (cm–1)
375 K
102 425 K
104
101 300 K
375 K
100 425 K
10–1 103
0.4 0.5 0.6 0.30 0.35 0.40
(a) 1/E (10–6 cm/V) (b) 1/E (10–6 cm/V)
Figure 5.17 Experimentally determined impact ionization coefficients in GaN: (a) hole impact
ionization coefficient in GaN and (b) electron impact ionization coefficient. The symbols
represent the experimentally determined data, and the solid curves represented the fitted
data using Chynoweth’s law. Source: Reproduced with permission from Ji et al. [12]. Copyright
2019, AIP Publishing.
In the NPD, because the doping concentration is about 1016 cm−3 , the electric
field can be regarded as uniform within an error of less than 10%. The hole impact
ionization coefficient can be obtained using Eq. (5.2).
The calculated hole’s impact ionization coefficient as a function of the reverse
electric field is shown in Figure 5.17a. The exponential fit to hole impact ioniza-
7 1
tion coefficient using Chynoweth’s law yields 𝛽(E) = 4.39 × 106 e−1.8×10 E cm−1 .
In the PND, because the doping density in the drift region is 2.2 × 1017 cm−3 ,
which creates a triangular electric field distribution over the drift region. The
electron impact ionization coefficient can be obtained by Eq. (5.3).
The electron impact ionization coefficient can be written as 𝛼(E) =
7 1
2.11 × 109 e−3.689×10 E cm−1 . The measured electron impact ionization coefficient
is shown in Figure 5.17b.
To validate the analysis, the multiplication factor and photocurrent in the NPD
and PND structures were calculated by solving the impact ionization integral
equations without any assumptions (Eqs. (5.6) and (5.7)) using the extrapolated
𝛼 and 𝛽. Mn and Mp obtained from the solution were then used to calculate the
photocurrent, as shown in Figure 5.18a,b.
The measured and calculated photocurrents in both NPD and PND have been
overlaid for comparison purposes. The close agreement between the calculated
and measured photocurrent in the two device structures proves the accuracy of
the present study. It is important to note that using the most general form of
the multiplication equations (Eqs. (5.6) and (5.7)), we can accurately reproduce
the photocurrent. The validation process in essence proves the correctness of the
initial assumption of 𝛽 > 𝛼 under low electric field.
A comparison of impact ionization coefficients of GaN [12], Si [42], and SiC
[43, 44] is shown in Figure 5.19. It is clear that GaN has the lowest impact ion-
ization coefficients. By solving the impact ionization integral equations in a 1-D
nonpunch through P–N junction, the critical electric field for breakdown as a
function of doping density in the drift region can be obtained, which is shown
in Figure 5.20, indicating that GaN has the highest critical electric field. By com-
bining the high electron mobility and the high critical electric field, GaN shows
a promising roadmap for the next-generation power devices.
Photocurrent (A)
10–2 10–3
Photocurrent (A)
Measured Calculated
10–3 Calculated 10–4
Measured
10–4 10–5
300 K
10–6 300 K
10–5
150 200 250 300 350 50 100 150 200
(a) Voltage (V) (b) Voltage (V)
Figure 5.18 (a) Comparison of the measured photocurrent in the NPD structure with the
calculated photocurrent using obtained impact ionization coefficients and (b) comparison of
the measured photocurrent in the PND structure with the calculated photocurrent using
obtained impact ionization coefficients. The symbols represent the experimentally
determined data, and the solid curves represent the calculated data by using 𝛼 and 𝛽 values
given by Eqs. (5.2) and (5.4). Source: Reproduced with permission from Ji et al. [12]. Copyright
2019, AIP Publishing.
Impact ionization coefficient (cm–1)
104 Figure 5.19 Comparison of impact

αn (Si) ionization coefficients among GaN, Si, and
αp (SiC) αn (SiC)
SiC.
103
αp (GaN)
102
αp (Si) αn (GaN)
101
105 106
Electric field (V/cm)
Critical electric field for breakdown (V/cm)
Figure 5.20 Comparison of the critical

4M electric field for breakdown of GaN, SiC,
and Si.
GaN
3M
2M
SiC
1M Si
0
1014 1015 1016 1017
Doping concentration (cm–3)
Acknowledgments 193
Figure 5.21 Comparison of the

4M
projected critical electric field using the
Critical electric field (V/cm)

extracted impact ionization coefficients Projected data [22]
[22] and the reported experimental data
in GaN pn diodes with avalanche 3M
capability [32, 45–49].
2M Stanford 2019 [49]

Avogy 2014 [45]
Cornell 2016 [46]
Nagoya 2018 [47]
1M Nagoya 2019 [48]
Hosei 2019 [32]
1015 1016 1017

Doping concentration (cm–3)
5.6.1 Impact of Impact Ionization Studies on Predictive Modeling

One immediate outcome of the impact ionization study was realized when the
experimentally determined impact ionization coefficients yielded an accurate
breakdown voltage estimation through TCAD models of recently reported
data. Figure 5.21 shows the plot of the predicted breakdown voltage estimation
[22] and the experimental results [32, 45–49]. The authors anticipate a very
significant impact of these studies through their use in TCAD models to do
predictive analysis.
5.7 Summary
In this chapter, we discussed the key device structures of vertical power GaN
devices, CAVETs and OGFETs (a variant of MOSFET), and the physics of
avalanche breakdown in GaN-on-GaN devices.
The two devices (CAVET and OGFET) have so far demonstrated encourag-
ing performance in two different voltage ranges. While CAVETs are probably
best suited for voltages up to 900 V or so, OGFETs definitely push the limits
to 1.2 kV and more. One key achievement that has been discussed here is the
ability to achieve a channel mobility in GaN OGFET up to 185 cm2 /V s, using a
regrowth-based technology. Higher channel mobility enabled by the creative use
of AlGaN/GaN heterostructures is also a key strength in a CAVET. The method
of determining the impact ionization coefficient in GaN, experimentally, and its
role in modeling were discussed.
Acknowledgments
The authors would like to thank Dr. Saptarshi Mandal and Dr. Matthew Laurent
for their contribution on avalanche electroluminescence of P–N diodes.
References
1 Chowdhury, S., Swenson, B.L., and Mishra, U.K. (2008). Enhancement and
depletion mode AlGaN/GaN CAVET with Mg-ion-implanted GaN as current
blocking layer. IEEE Electron Device Lett. 29 (6): 543–545.
2 Chowdhury, S., Wong, M.H., Swenson, B.L., and Mishra, U.K. (2012). CAVET
on bulk GaN substrates achieved with MBE-regrown AlGaN/GaN layers to
suppress dispersion. IEEE Electron Device Lett. 33 (1): 41–43.
3 Ji, D., Laurent, M.A., Agarwal, A. et al. (2017). Normally OFF trench CAVET
with active Mg-doped GaN as current blocking layer. IEEE Trans. Electron
Devices 64 (3): 805–808.
4 Ji, D., Agarwal, A., Li, W. et al. (2018). Demonstration of GaN current
aperture vertical electron transistors with aperture region formed by ion
implantation. IEEE Trans. Electron Devices 65 (2): 483–487.
5 Mandal, S., Agarwal, A., Ahmadi, E. et al. (2017). Dispersion-free 450 V p
GaN-gated CAVETs with Mg-ion implanted blocking layer. IEEE Electron
Device Lett. 38 (7): 933–936.
6 Ji, D., Agarwal, A., Li, H. et al. (2018). 880V/2.7 mΩ cm2 MIS gate trench
CAVET on bulk GaN substrates. IEEE Electron Device Lett. 39 (6): 863–865.
7 Li, W., Ji, D., Tanaka, R. et al. (2017). Demonstration of GaN static induction
transistor (SIT) using self-aligned process. IEEE J. Electron Device Society 5
(6): 485–490.
8 Chun, J., Li, W., Agarwal, A., and Chowdhury, S. (2019). Schottky junction
vertical channel GaN static induction transistor with sub-micrometer fin. Adv.
Electron. Mater. 5 (1): 1800689.
9 Ji, D., Gupta, C., Agarwal, A. et al. (2018). Large-area in-situ oxide, GaN
interlayer-based vertical trench MOSFET (OG-FET). IEEE Electron Device
Lett. 39 (5): 711–714.
10 Ji, D., Gupta, C., Chan, S.H. et al. (2017). Demonstrating >1.4 kV OG-FET
performance with a novel double field-plated geometry and the successful
scaling of large-area devices. In: Proceedings of IEEE Electron Devices Meeting
(IEDM), 223–226. San Francisco, CA: IEEE.
11 Mandal, S., Kanathila, M., Pynn, C. et al. (2018). Observation and discussion
of avalanche electroluminescence in GaN pn diodes offering a breakdown
electric field of 3 MV/cm. Semicond. Sci. Technol. 33 (6): 065013.
12 Ji, D., Ercan, B., and Chowdhury, S. (2019). Experimental determination of
impact ionization coefficients of electrons and holes in gallium nitride using
homojunction structures. Appl. Phys. Lett. 115 (7): 073503.
hole impact ionization coefficient and saturation velocity in GaN. In: Proceed-
ings of Compound Semiconductor Week, 1–2. Nara, Japan: IEEE.
velocity-field characteristic of holes in GaN. IEEE Electron Device Lett. 40
(12): 1–4.
15 Ben-Yaacov, I., Seck, Y.K., Mishra, U.K., and DenBaars, S.P. (2004).
AlGaN/GaN current aperture vertical electron transistors with regrown
channels. J. Appl. Phys. 95 (4): 2073.
References 195
16 Nie, H., Diduck, Q., Alvarez, B. et al. (2014). 1.5-kV and 2.2-mΩ-cm2 vertical
GaN transistors on bulk GaN substrates. IEEE Electron Device Lett. 35 (9):
939–941.
17 Chowdhury, S. (2010). AlGaN/GaN CAVETs for high power switching appli-
cation. Doctoral Dissertation. University of California, Santa Barbara.
18 Shibata, D., Kajitani, R., Ogawa, M. et al. (2016). 1.7 kV/1.0 mΩ⋅cm2
normally-off vertical GaN transistor on GaN substrate with regrown
p-GaN/AlGaN/GaN semipolar gate structure. In: Proceedings of IEEE Electron
Devices Meeting (IEDM), 248–251. San Francisco, CA: IEEE.
19 Otake, H., Egami, S., Ohta, H. et al. (2007). GaN-based trench gate metal
oxide semiconductor field effect transistors with over 100 cm2 /Vs channel
mobility. Jpn. J. Appl. Phys. 46 (25): L599–L601.
20 Otake, H., Chikamatsu, K., Yamaguchi, A. et al. (2008). Vertical GaN-based
trench gate metal oxide semiconductor field-effect transistors on GaN bulk
substrates. Appl. Phys. Express 1 (1): 011105.
21 Oka, T., Ueno, Y., Ina, T., and Hasegawa, K. (2014). Vertical GaN-based
trench metal oxide semiconductor field-effect transistors on a free-standing
GaN substrate with blocking voltage of 1.6 kV. Appl. Phys. Express 7 (2):
021002.
22 Oka, T., Ina, T., Ueno, Y., and Nishii, J. (2015). 1.8 mΩ⋅cm2 vertical GaN
based trench metal oxide semiconductor field effect transistors on a free-
standing GaN substrate for 1.2-kV-class operation. Appl. Phys. Express 8 (5):
05401.
23 Li, R., Cao, Y., Chen, M., and Chu, R. (2016). 600 V/1.7 Ω normally-off GaN
vertical trench metal-oxide-semiconductor field-effect transistor. IEEE Elec-
tron Device Lett. 37 (11): 1466–1469.
24 Gupta, C., Chan, S.H., Enatsu, Y. et al. (2016). OG-FET: an in-situ oxide, GaN
interlayer based vertical trench MOSFET. IEEE Electron Device Lett. 37 (12):
1601–1604.
25 Gupta, C., Lund, C., Chan, S.H. et al. (2017). In-situ oxide, GaN interlayer
based vertical trench MOSFET (OG-FET) on bulk GaN substrates. IEEE
Electron Device Lett. 38 (3): 353–355.
26 Ji, D., Gupta, C., Agarwal, A. et al. (2017). First report of scaling a normally-
off in-situ oxide, GaN interlayer based vertical trench MOSFET (OG-FET).
In: Proceedings of Device Research Conference, 1–2. South Bend, IN: IEEE.
27 Li, W., Nomoto, K., Lee, K. et al. (2018). Development of GaN vertical
trench-MOSFET with MBE regrown channel. IEEE Trans. Electron Devices
65 (6): 2558–2564.
28 Li, W. and Chowdhury, S. (2016). Design and fabrication of a 1.2 kV
GaN-based MOS vertical transistor for single chip normally off operation.
Phys. Status Solidi A 213 (10): 2714–2720.
29 Sun, M., Zhang, Y., Gao, X., and Palacios, T. (2017). High-performance GaN
vertical fin power transistors on bulk GaN substrates. IEEE Electron Device
Lett. 38 (4): 509–512.
30 Zhang, Y., Sun, M., Piedra, D. et al. (2017). 1200 V GaN vertical fin power
field-effect transistors. In: IEEE International Electron Devices Meeting,
215–218. San Francisco, CA: IEEE.
31 Kizilyalli, I.C., Edwards, A., Nie, H. et al. (2013). High voltage vertical GaN
p-n diodes with avalanche capability. IEEE Trans. Electron Devices 60 (10):
3067–3070.
32 Ohta, H., Asai, N., Horikiri, F. et al. (2019). 4.9 kV breakdown voltage vertical
GaN p-n junction diodes with high avalanche capability. Jpn. J. Appl. Phys. 58
(SC): SCCD03.
33 Lahbabi, M. and Ahaitouf, A. (2004). Analysis of electroluminescence spectra
of silicon and gallium arsenide p-n junctions in avalanche breakdown. J. Appl.
Phys. 95 (4): 1822.
34 Lyons, J.L., Janotti, A., and Van de Walle, C.G. (2010). Carbon impurities and
the yellow luminescence in GaN. Appl. Phys. Lett. 97 (15): 152108.
35 Yan, Q., Janotti, A., Scheffler, M., and Van de Walle, C.G. (2012). Role of
nitrogen vacancies in the luminescence of Mg-doped GaN. Appl. Phys. Lett.
100 (14): 142110.
36 Cao, X.A., Pearton, S.J., Dang, G.T. et al. (2000). GaN n- and p-type Schot-
tky diodes: effect of dry etch damage. IEEE Trans. Electron Devices 47 (7):
1320–1324.
37 Oguzman, I.H., Bellotti, E., and Brennan, K.F. (1997). Theory of hole initiated
impact ionization in bulk zincblende and wurtzite GaN. J. Appl. Phys. 81 (12):
7827.
38 Chynoweth, A.G. (1958). Ionization rates for electrons and holes in silicon.
Phys. Rev. 109 (3): 1537–1545.
39 Muth, J.F., Lee, J.H., Shmagin, I.K., and Kolbas, R.M. (1997). Absorption coef-
ficient, energy gap, exciton binding energy, and recombination lifetime of
GaN obtained from transmission measurements. Appl. Phys. Lett. 71 (18):
2572.
40 Sze, S.M. and Ng, K.K. (2006). Physics of Semiconductor Devices, 3e, 105.
Hoboken), chapter 2: Wiley.
41 Niwa, H., Suda, J., and Kimoto, T. (2015). Impact ionization coefficients in
4H-SiC toward ultrahigh-voltage power devices. IEEE Trans. Electron Devices
62 (10): 3326–3333.
42 Baliga, B.J. (2008). Fundamentals of Power Semiconductor Devices. New York:
Springer-Science. Section 2.1.5.
43 Konstantinov, A.O., Wahab, Q., Nordell, N., and Lindefelt, U. (1997). Ion-
ization rates and critical electric fields in 4H-SiC. Appl. Phys. Lett. 71 (1):
90.
44 Konstantinov, A.O., Wahab, Q., Nordell, N., and Lindefelt, U. (1998). Study of
avalanche breakdown and impact ionization in 4H silicon carbide. J. Electron.
Mater. 27 (4): 335–341.
45 Kizilyalli, I.C., Edwards, A.P., Aktas, O. et al. (2014). Vertical power
p-n diodes based on bulk GaN. IEEE Trans. Electron Devices 62 (2):
414–422.
46 Nomoto, K., Song, B., Hu, Z. et al. (2016). 1.7-kV and 0.55-mΩ⋅cm2 GaN
p-n diodes on bulk GaN substrates with avalanche capability. IEEE Electron
Device Lett. 37 (2): 161–164.
47 Maeda, T., Narita, T., Ueda, H. et al. (2018). Parallel-plane breakdown
fields of 2.8–3.5 MV/cm in GaN-on-GaN p-n junction diodes with
References 197
double-side-depleted shallow bevel termination. In: Proceedings of IEEE

Electron Devices Meeting (IEDM), 689–693. San Francisco, CA: IEEE.
48 Fukushima, H., Usami, S., Ogura, M. et al. (2019). Deeply and vertically
etched butte structure of vertical GaN p–n diode with avalanche capability.
Jpn. J. Appl. Phys. 12 (SC): SCCD25.
49 Ji, D., Li, S., Ercan, B. et al. (2020). Design and fabrication of ion-implanted
moat etch termination resulting in 0.7 mΩ⋅cm2 /1500 V GaN diodes. IEEE
199
Reliability Issues in GaN Electronic Devices

Milan Ťapajna 1 and Christian Koller 2
1
Institute of Electrical Engineering, Slovak Academy of Sciences (IEE–SAS), Department of III-V Semiconductors,
Dubravska cesta 9, 841 04 Bratislava, Slovakia
2
KAI – Kompetenzzentrum Automobil- und Industrieelektronik GmbH, Europastrasse 8, 9524 Villach, Austria
6.1 Introduction
GaN-based electronic devices have brought unprecedented performance in
terms of power, frequency, and efficiency with respect to Si- and GaAs-based
devices. The GaN high electron mobility transistors (HEMTs) offer high cutoff
frequency and high breakdown (BD) voltage, allowing the design of efficient
power amplifiers for wireless communication, satellite, and radar systems
commercially available already for a decade. GaN HEMTs are also excellent
candidates for switching devices for power converter systems. Great effort in
the past decade led to the development of compact GaN-based high-efficiency
power converters that are currently emerging into the market [1]. Apart from
being cost-effective, any new technology offering improved performance must
also ensure reliable operation for successful commercialization. Although GaN
HEMTs have proved to represent a reliable technology for RF applications,
device technology for millimeter-wave (see Chapter 3) and normally-off power
switching applications (see Chapter 4) are currently entering this stage. To
exploit the full potential of GaN technology, a key task is to identify and gain a
detail understanding of the degradation mechanisms occurring in the devices
with application-specific design. This represents a complex issue because
of the unique properties of GaN-based materials (wide-bandgap nature and
piezoelectricity) and high electric field and power dissipation existing in the
operating devices. Moreover, because GaN heterostructures are grown on
foreign substrates, the active area of the device typically contains relatively high
density of various defects. These promote new failure mechanisms, different
from those observed in GaAs- and InP-based compound semiconductors.
The reliability of RF GaN HEMTs and monolithic microwave-integrated
circuits (MMICs) has significantly improved in the past decade. It was achieved
through great research effort, bringing more details for the understanding of
parasitic dispersion effects, new failure modes related to electric field concen-
tration, and generated heat distribution in GaN HEMTs. Furthermore, tailored

200 6 Reliability Issues in GaN Electronic Devices
characterization techniques to study these effects have been introduced. More

recently, similar effort has been devoted to reliability studies of GaN power
switching HEMTs. These devices are predominantly grown on Si substrates,
which require semi-insulating GaN buffer commonly achieved by carbon dop-
ing. However, carbon doping in GaN represents a rather complex behavior and
has been connected to dynamic on-resistance effects during switching. Sophis-
ticated device design needs to be applied to mitigate this parasitic behavior. It
has also been recognized that precise lifetime extrapolation requires dynamic
test sequences similar to the actual operating conditions for a given application.
In addition, breakdown mechanisms governed by percolative process have been
identified to limit the blocking capabilities of GaN switching HEMTs as well as
gate robustness under forward bias in the case of p-GaN HEMTs. Transistors
with insulated gate have been observed to suffer from ample threshold voltage
instabilities.
6.1.1 Reliability Testing and Failure Analysis of GaN HEMTs

The study of electron device reliability represents interdisciplinary research and
development with the aim to find the operation conditions at which the device
can operate without compromising its performance for a given period of time
(typically >20 years). The reliability study essentially relies on the monitoring
of the device failure rate under application-specific conditions and environ-
ment. The device failure typically occurs by two modes: (i) gradual change of
electrical parameters with time and (ii) sudden catastrophic failure or burnout.
Although electronic device degradation is a complex process, the failure rate
time dependence often follows a “bathtub” curve depicted in Figure 6.1a. In
the initial “burn-in” period, the failure rate is high because of the pre-existing
imperfections in the device population. Then, the failure rate drops to a constant
value defining a useful operation life and finally increases again as the wear
out occurs. Although device failures during the burn-in process can be readily
inspected, monitoring of large device populations until wear out is very difficult.
Therefore, accelerated lifetime tests are performed where certain operating or
environmental parameters are elevated to accelerate failure of the device.
Most commonly, elevated temperatures are used as an accelerating factor, and
the lifetime prediction is based on an assumption that the device median time to
failure (MTTF), i.e. 50% failure rate, decreases with junction temperature (T j ) and
follows an Arrhenius dependence, MTTF ∼ exp(Ea /k B T j ), where Ea is the activa-
tion energy and k B is the Boltzmann constant. In the standard high-temperature
operating life (HTOL) test defined by Joint Electron Device Engineering Coun-
cil (JEDEC) [3, 4], a number of devices are stressed in the “direct-current” (DC),
dynamic, or RF conditions at several elevated temperatures. The electrical param-
eters of the device such as drain (I D ) or gate (I G ) current, transconductance (g m ),
or RF output power are periodically tested against the failure criteria. MTTF is
then determined statistically for each temperature and plotted as a function of
1/k B T j , where the slope gives Ea . Such a test is shown in Figure 6.1b, comparing
the HTOL tests of RF AlGaN/GaN HEMTs performed in DC and RF conditions.
Junction temperature
256 °C 225 °C 175 °C
1E + 8
Infant Useful life Wear out 1E + 7

mortality
1E + 6
Failure rate
MTTF (h)
1E + 5
Ea = 1.92eV
Ea = 1.82eV
Quali
1E + 4
s
re
ilu
Observed failures
fa
ty fail
1E + 3
ut
ro
Random failures
ures
ea
W
1E + 2
1E + 1
Time 18 19 20 21 22 23 24 25 26 27 28
(a) (b) 1/KT (1/eV)
Figure 6.1 (a) Typical dependence of device failure rate on time known as “bathtub curve” in
the reliability community. (b) DC high-temperature operating life (HTOL) test data obtained at
V DS = 50 V and ID = 50 mA and junction temperatures from 270 to 310 ∘ C (red triangle
symbols, 5–10 tested devices per point) and RF HTOL results (blue diamond symbols, 4 tested
devices per point). Extrapolations and E a determination using the Arrhenius equation are
shown using T j determined from a finite element thermal model validated using Raman
thermography measurements. Source: Panel (b) is reprinted with permission of Pomeroy et al.
[2]. Copyright © 2015, Elsevier.
Extracted Ea and fictitious lifetime at infinite temperatures (intercepts) can then

be used to extrapolate the device wear out at actual maximum operating T j .
The Arrhenius model for lifetime prediction correlates well with the degrada-
tion modes known for Si devices (see Table 6.1) and the diffusion-driven modes
for GaN HEMTs such as Schottky gate metal interdiffusion [7] and Ohmic metal-
lization degradation [8]. There are, however, limitations of the Arrhenius model
for lifetime predictions of GaN HEMTs. Several new failure mechanisms with
various T-dependence have been reported for GaN HEMTs (Sections 6.2 and
6.3), and dominant mechanisms at elevated temperatures and stress conditions
are not necessarily the same as those at operating conditions. In addition, this
approach critically depends on precise determination of T j . This is particularly
important in GaN HEMTs where thermal gradients as high as 100 ∘ C/μm can
exist close to the channel [9], questioning the application of commonly used tech-
niques such as IR thermography for channel temperature determination in RF
GaN HEMTs (see Section 2.3).
To ensure precise reliability testing and lifetime prediction of electronic
devices, several steps are required. First, key failure mechanisms have to be iden-
tified. The left side of Table 6.1 shows seven key failure mechanisms occurring
in the Si electronic devices. The right side shows the next step which is to find a
MTTF model that describes the failure times for every failure mode accurately.
Based on the knowledge gained over the past 50 years, qualification standards
such as JEDEC and Automotive Electronics Council (AEC) have been developed,
defining the test stress conditions, durations, and sample sizes for qualification
of Si power switching devices (see exemplary test list in Table 6.2). These
qualification standards comprise the three categories: electrostatic discharge
Table 6.1 MTTF models for various failure mechanisms in Si.
Failure mechanism Lifetime model
Electromigration MTTF ∝ I −N exp(Ea /(k B T))

Stress migration MTTF ∝ (T 0 − T)−N exp(Ea /(k B T))
Corrosion MTTF ∝ (%RH)−N exp(Ea /(k B T))
TDDB (time-dependent MTTF ∝ exp(−𝛾 Eox ) exp(Ea /(k B T))
dielectric breakdown), E-model
Fatigue MTTF ∝ (ΔT − T 0 )−N
−1
Mobile ions MTTF ∝ Jion exp(Ea ∕(kB T))
Hot carrier injection MTTF ∝ (I sub /W )−N exp(Ea /(k B T))
Source: Adapted from McPherson 2018 [5] and McDonald 2018 [6].
Table 6.2 Standard reliability tests for Si power devices.
Abbr. Test/standard Conditions Duration Sample size
TC Temperature cycling T = −55/150 ∘ C 1 000× 3 lots × 77 pcs

JESD22 A104
HTRB High-temperature reverse bias V = 600 V 1 000 h 3 lots × 77 pcs
JESD22 A108 T = 150 ∘ C
HTS High-temperature storage life T = 150 ∘ C 1 000 h 3 lots × 77 pcs
JESD22 A103
HTGS(+) Positive high-temperature I = 50 mA 1 000 h 3 lots × 77 pcs
gate stress
JESD22 A108
HTGS(−) Negative high-temperature V = −10 V 1 000 h 3 lots × 77 pcs
gate stress
JESD22 A108
H3TRB Temperature and V = −10 V 1 000 h 3 lots × 77 pcs
humidity bias Humidity = 85%
JESD22 A101 T = 150 ∘ C
IOL Intermittent operating life test ΔT = 100 ∘ C 15 000× 3 lots × 77 pcs
JESD22 A105
ESD ESD-HBM and CDM 1 000 h 1 lot × 3 pcs
JS-001 and JS-002
CDM, charge device model; HBM, human body model.

Source: Data from McPherson 2018 [5].
(ESD), package, and device. Although the first two are similar for both GaN
and Si device technology, the device category is very different for GaN devices
because of the different transistor form and material properties. Thus, in this
chapter, we only focus on the device category. The schematic cross section of a
Si superjunction device in Figure 6.2 illustrates four regions of interest in such a
device and the qualification test to examine their reliability. Although some of
Figure 6.2 Schematic cross section of S G 4 Tested part Test

a superjunction MOSFET with n+ 3
identification of several failure 2 1 pn-body diode HTRB
p+
mechanisms and the regions in which 2 Gate oxide HTGB
they are relevant [6]. p 1 n 3 Passivation H3TRB
Aluminum,
4 TC
wire bonds
D
these failures can also occur in a similar form in GaN power devices, the device
form and materials are very different, for example, failure modes number (1)
and (2) in Figure 6.2, which indicate that the respective failure of the p–n body
diode and the gate oxide cannot happen in GaN as such structures do not exist
in many GaN HEMTs. Instead, GaN devices have other weak points, where less
long-term knowledge exists due to the young history of GaN power devices.
Piezoelectricity and pyroelectricity of GaN-based materials couple electrical,
thermal, and mechanical stresses. Power devices operated at high current
densities and electric fields can be expected to reveal new degradation mech-
anisms. Indeed, electric field approaching several megavolts per centimeter
concentrating at the drain side of the gate at pinch-off was found to trigger
physical gate-edge degradation originating from the combination of mechanical
and thermal stresses. In the on-state, extremely high-energy hot electrons can
generate new trapping states, deteriorating the device performance and large
power dissipation in localized areas, which can cause the appearance of hot
spots and thermal stresses leading to possible mechanical degradation of the
device structure. These issues will be briefly reviewed in Section 6.2, focusing
on reliability and thermal issues of AlGaN/GaN and InAlN/GaN HEMTs for
RF applications. For the failure analysis, the degraded devices are commonly
inspected by optical, scanning electron microscopy (SEM) or in some cases
transmission electron microscopy (TEM). In GaN HEMTs, electroluminescence
(EL) microscopy can be applied to visualize local degradation of devices. Special
attention is therefore given to recent discussions on the origin of EL in GaN
HEMTs.
Several groups [10, 11] have reported that the breakdown mechanisms in
power GaN HEMTs show characteristics similar to time-dependent dielectric
breakdown (TDDB) rather than impact ionization typical for most semicon-
ductors. This mechanism is frequently described by a percolation model,
i.e. stress-induced generation of defects forming the leakage path across the
structure once a sufficient concentration of defects is available (see Section 3.1).
To characterize this process, the time to breakdown (t BD ) is measured for a
number of identical devices for different biases and temperatures. Distributions
of t BD are then described by Weibull statistics, where the cumulative failure
probability, F(t BD ), is related to t BD as F(t BD ) = 1 − exp(−t BD /𝜂)𝛽 [12]. Here, 𝜂 is
the scale factor or 63.2% of the value of the distribution and 𝛽 is the shape factor
or Weibull slope, proportional to the spread in t BD . By plotting ln[−ln(1–F(t BD ))],
also known as Weibit, as a function of ln(t BD ), both 𝛽 and 𝜂 can be extracted by
fitting the experimental data to a linear function. Using the extracted 𝜂 values, it
is possible to estimate the maximum reliable bias safe operation by formulating

the reliability specification, i.e. failure rate percentile, gate area, time period,
bias, and temperature. Weibull distribution is often bimodal, with lower shape
factor for shorter t BD originated from extrinsic defects and higher shape factor
for longer t BD characteristics for intrinsic defects in the material.
Section 6.3 summarizes current understanding of the degradation modes of
GaN power switching devices and their characterization, focusing on investi-
gations of carbon doping in the GaN buffer, time-dependent buffer breakdown,
and trapping in the GaN buffer, thereby affecting dynamic on-state resistance
(RDS,ON ) of power devices. Finally, gate degradation of normally-off p-GaN
HEMTs and physics and characterization of threshold voltage (V th ) instabilities
in GaN MISHEMTs (metal–insulator–semiconductor HEMT) will be reviewed
in this section.
6.2 Reliability of GaN HEMTs for RF Applications

RF GaN HEMTs are characterized by short gate length inducing very high
electric field under the gate with a peak located at the drain side of the gate.
In order to spread out the peaked electric field, design approaches using T-
and G-shaped gates and source-connected field plates have been applied.
Degradation modes related to material defects (e.g. III-N surface and bulk traps
[BTs]) and processing imperfections can be suppressed by appropriate surface
passivation and growth optimization. In contrast, those originating from intrin-
sic material properties and operating device conditions need to be tackled by the
field mitigation approaches (hot electrons and self-heating) or by changing the
operating conditions (soft and hard breakdown). Several new degradation modes
have been identified in GaN HEMTs and some of them have been reviewed in
the literature [13–17]. We will therefore briefly describe the most important
reliability issues reported for AlGaN/GaN HEMTs in Section 6.2.1, focusing
on frequency dispersion, gate-edge physical degradation, hot-electron-related
degradation, and thermal effects. In Section 6.2.2, an overview of InAlN/GaN
HEMTs reliability studies will be reviewed, concentrating on hot electron and
hot phonon effects.
6.2.1 AlGaN/GaN HEMTs

6.2.1.1 Trapping Effects
Because of the initial immaturity of growth technology, frequency dispersion was
realized as a main reliability concern in AlGaN/GaN HEMTs. The frequency dis-
persion effects represent the temporal change of the electrical parameters (I D ,
g m , and I G ) of the device caused by capture and emission of carriers by elec-
trically active defects in various regions of an operating device. In the capture
process, a free carrier loses its energy and becomes localized by the defect site
described by its density, energy level in the semiconductor bandgap, and capture
cross section. In the emission process, the carrier must gain enough energy to
overcome the energy barrier given by the defect level and conduction/valence
band minimum/maximum. Therefore, the emission process and thus the emis-
sion time constant are typically much longer than the capture process, as the
latter depends mostly on the availability of free carriers.
Despite great improvements in the growth quality, III-N heterostructures pos-
sess relatively high density of bulk defects and surface/interface states. In par-
ticular, surface states are considered to play a vital role in defining the surface
potential of an ungated barrier surface by pinning the Fermi level at the trap
level with a high density [17]. Although this notion has been recently revisited
(see Section 3.4), many studies suggest surface-state density well in the range of
1012 eV−1 cm−2 , even for the passivated III-N surface [18–20]. Trapping caused
by surface states spatially localized in the drain-side access region is depicted
in Figure 6.3a. As originally proposed by Vetury et al. [22], when the device is
biased into off-state, high gate-to-drain voltage induces injection of electrons into
surface states from the gate edge. High density of surface traps allows electron
redistribution by hopping toward the drain so that negative charge is extending
several tens or hundreds of nanometers [23] from the gate edge until steady-state
condition is reached due to electrostatic feedback of the piled negative charge
[15]. These electrons, forming the well-known “virtual gate” effect, extend the
depletion region of the two-dimensional electron gas (2DEG) channel. When gate
bias is switched to on-state, the electrons cannot be removed immediately from
the traps and the extended virtual gate part of the channel remains depleted for
a period given by the characteristic emission time constant. Dynamic emission
of electrons from surface/interface states results in I D transient known as current
collapse or gate-lag effect, resulting in dynamic change of drain resistance and
peak g m .
Bulk traps in the barrier layer can result in more complex behavior. Those
located in the gate-to-drain region can capture electrons injected from the gate
[15] (see Figure 6.3a) and high-energy electrons from the channel [24], affecting
the lateral electric field distribution and thus RDS,ON and g m , similar to surface
traps. In contrast, bulk traps located in the barrier under the gate can capture
electrons injected from the gate in the off-state (negative gate bias, V G ), leading
to negative charge build-up and shift in V th (see Figure 6.3a) [25]. When V G is
biased to on-state, emission of these traps gives rise to dynamic V th shift without
impact on the RDS,ON and g m . Similarly, bulk traps in the GaN buffer, distributed
in the whole device active area, can give rise to a complex trapping behavior.
In off-state, channel electrons are repelled into buffer and get trapped by bulk
traps [26]. When the device is biased back to on-state, trapped electrons act as a
“back gate,” decreasing 2DEG population as compared to its equilibrium density,
causing V th shift under the gate as well as dynamic RDS,ON and g m change in the
gate-to-drain region.
Trapping behavior of GaN-based HEMTs can be quickly evaluated by pulsed
I D –V DS and I D –V GS characterization also called gate-lag measurements, where
I–V characteristics measured in gate-pulsed mode are compared to DC charac-
teristics. An example is given in Figure 6.3b,c showing output and transfer char-
acteristics of an RF AlGaN/GaN HEMT before and after off-state stressing [21].
While stress-induced positive V th shift was observed from DC measurements
G
– – – –
S – D
– – – –
AIGaN –
ΔVTH
ΔRD
Δgm
GaN – – – –
(a)
1.2 1.2 100

Vgs = 0 V Vds = 20 V
Fresh
1.0 1.0
Fresh
0.8 0.8 10–1

After stress
Id (A/mm)
Id (A/mm)
Id (A/mm)
After stress
After stress
0.6 (Vth -corr.) 0.6
30
Gate-lag (%)
0.4 After stress 0.4 10–2

20
10
0.2 Fresh 0.2
After stress
0
0 Vds (V) 20 (Vth -corr.)
0.0 0.0 10–3
0 5 10 15 20 –6 –4 –2 0
Vds (V) Vgs (V)
(b) (c)
Figure 6.3 (a) Schematic representation of surface and bulk traps (in the AlGaN barrier and
GaN buffer layers) affecting V th , RD , and gm after applying negative V G . DC (b) output and
(c) transfer characteristics of AlGaN/GaN HEMT before and after 10-hour off-state stressing at
V DS = 30 V and V GS = −5 V. After correction of I–V characteristics to V th shift, notable
degradation of RDS,ON can be inferred in the linear part of the ID –V DS characteristic. Much
stronger degradation of dynamic RDS,ON can be observed from the gate-lag measured before
and after OFF-state stress (inset). Source: Panels (b) and (c) are reprinted with permission of
Ťapajna et al. [21]. Copyright © 2010, IEEE.
(Figure 6.3c) indicating trap generation under the gate, pulsed gate-lag measure-
ment (inset of the left panel) also clearly shows dynamic RDS,ON degradation,
suggesting trap generation in the gate-to-drain access region. A more compre-
hensive way to explore trapping effects employs a dual-channel pulsing system
with different sets of quiescent bias conditions, also called drain-lag measure-
ment. Despite the effectiveness of pulsed I–V characterization, these techniques
do not provide any information on the properties of traps responsible for device
degradation subjected to stressing. For this aim, Joh and del Alamo [27] pro-
posed I D -transient technique, capable to extract trap-level energy and capture
cross section, based on the principle used in isothermal deep-level transient spec-
troscopy (DLTS). Here, I D transients are measured in the logarithmic timescale
at different temperatures after trap prefilling. The measured I D transients are

analyzed by various approaches [21, 27, 28] to visualize the characteristic time
constants of trap levels. The I D -transient technique has been successfully applied
by many research groups to monitor trap generation of short- and long-term
stressed devices [16, 28]. DLTS is also commonly used to analyze defects in bulk
GaN material [29], characterization of traps [24], and reliability studies for mon-
itoring stress-induced trap generation in AlGaN/GaN HEMTs [28].
6.2.1.2 Gate-edge Degradation

When an AlGaN/GaN HEMT is biased into off-state, a high electric field at the
gate edge was found to cause a specific degradation mode. As first observed by
Joh and del Alamo [30], application of reverse gate bias (V DS = 0 V) step stress
can lead to sudden and irreversible increase in the off-state I G , accompanied
with degradation of maximum I D and source and drain resistance (RS , RD ) at V G
referred to as critical voltage (V crit ), see Figure 6.4a. Qualitatively same degrada-
tion with varying V crit was also reported for on-state stressed devices [34] and
was correlated with worsening of the current collapse and trap generation with
energy level of ∼0.5–0.6 eV, as deduced from I D -transient analysis [34] and cur-
rent DLTS [35]. Sudden degradation of I D was investigated by Chowdhury et al.
[32], where RF AlGaN/GaN HEMTs submitted to HTOL test (250–320 ∘ C) were
extensively analyzed by TEM. As exemplified in Figure 6.4b, material degrada-
tion at the drain side of the gate was observed, containing pit-like defects and
in some cases a crack developed under the pit. Elsewhere, oxygen was found in
the pits [36]. Gate-edge degradation was then confirmed by other studies, where
I G degradation upon reverse V G or off-state step stress was correlated with evo-
lution of localized hot spots observed from EL microscopy [37]. Bajo et al. [33]
evidenced a one-to-one correlation between evolution of the EL hot spots and
pit conductivity measured by conductive atomic force microscopy (AFM) after
passivation removal shown in Figure 6.4c,d.
Increase of I G together with trap generation upon reverse V G -stressed
AlGaN/GaN HEMTs has been explained by a new degradation mechanism
identified as converse piezoelectric effect [31]. The model considers the piezo-
electric nature of GaN-based materials and high electric field across the AlGaN
layer, subjected to in-plane tensile stress, originating from lattice mismatch
between AlGaN and GaN. In off-state (for V DS > 0), the vertical component of
the electric field increases particularly at the drain side of the gate edge, which
enhances the tensile strain in the AlGaN layer. When a certain level of strain or
stored elastic energy in the system reaches the critical level, a crystallographic
defect in the AlGaN barrier forms at the place with highest electric field and
initial stress. This defect then gives rise to enhanced electron injection from the
gate and trap generation responsible for worsening of the trapping effects. The
degradation related to converse piezoelectric effect is expected to be electric
field driven. However, the situation is more complicated when the AlGaN/GaN
HEMT is operated in on-state, where also thermal stress, because of a thermal
gradient of heat generation, is present in GaN [38]. In fact, detailed fully coupled
electroelastic 2D simulations of AlGaN/GaN HEMT performed by Ancona et al.
[39] suggested that the converse piezoelectric effect adds about 17% of principal
1.8 1.E–02
IGoff S (c)
1.6
IDmax/IDmax(0), R/R(0)
G
1.4 1.E–03 7 56 2 43 1
Rs
|IGoff| (A/mm)
1.2 D
RD 1.E–04
1
0.8 1.E–05
0.6 IDmax 20 μm
Vcrit
0.4 1.E–06
10 20 30 40 50
(a) VDGstress (V) 6
EL integrated intensity
300 5
4
3
(arb. units)
2
200 1
60
Ig (μA)
100 40
50 nm 20
0
0
400 500 600 700
(b) (d) Time (s)
Figure 6.4 (a) Change in IDmax , RS , RD , and IGoff during V GS step-stress experiment (V DS = 0). V GS
was stepped from −10 to −50 V in 1 V steps (1 minute/step). Source: Reprinted with permission
of Joh et al. [31]. Copyright © 2007, IEEE. (b) Cross-sectional high resolution transmission
electron microscopy (HREM) of stressed devices showing the crack formation. The trapezoidal
shape defines the gate metal; right side is toward the drain. Source: Reprinted with permission
of Chowdhury et al. [32]. Copyright © 2008, IEEE. (c) EL image overlaid on a white light image
from a device stressed at V GS = 15 V and V DS = 30 V for 760 seconds (only upper finger was
biased). The numbers indicate the order of appearance of the EL spots. (d) EL-integrated
intensity as a function of stress time from each of the EL spots of (c) together with sum of the
EL-integrated intensities of all spots together with IG as a function of stress time. Source: Panels
(c) and (d) are reprinted with permission of Bajo et al. [33]. Copyright © 2012, AIP Publishing.
stress at the gate edge compared to elsewhere in AlGaN. The authors argued that
such stress elevation can cause crack propagation, while triggering of pit/crack
formation may originate from strong electron injection from the gate. They
also pointed out strong impact of intrinsic stress in the SiN passivation layer to
maximum stress in AlGaN. All these aspects may be responsible for the observed
impact of temperature on the pit size of the degraded devices [40], absence of pit
or crack formation in the off-state stressed devices even under high electric field
[41], and some contradiction between the correlation of I G and I D degradation
in the devices with apparent crystallographic defects present in AlGaN [36].
Other reliability studies of AlGaN/GaN HEMTs also showed appearance of
other gate-edge degradation mechanisms with characteristics different from
that of converse piezoelectric effect. Ťapajna et al. [21] studied early stage
degradation of GaN HEMTs subjected to both on- and off-state stress using
a combination of EL, I–V measurements, and I D -transient characteristics.
Off-state stress-induced degradation was attributed to trap generation at the
drain side of the gate edge; however, the trap generation was found to follow a
diffusion process. These traps were later proposed to relate to oxygen diffusion
into AlGaN layer at the electric field peak [42]. An effective method to study
structural damage of tested device was proposed by Makaram et al. [43], where
the surface of the GaN-cap/AlGaN layer is studied by AFM and SEM after
removal of SiN passivation and gate metallization after electrical stressing. For
off-state stressed device, formation of linear grooves for gate-drain voltages
V GD < V crit while deep pits for devices stressed beyond V crit accompanied with I G
and I D degradation was observed at T = 150 ∘ C. A lower density of pits appeared
for stressing at T = 25 ∘ C. This electric-field-induced diffusion was attributed to
electrochemical oxidation or gate metal interdiffusion phenomenon. Gao et el.
[44] also reported electric-field-oxidation at the surface, where strings and pits
containing oxygen were observed by Auger spectroscopy in reverse-V G stressed
AlGaN/GaN HEMTs with unpassivated and Al2 O3 passivated surface. The pit
formation was greatly reduced in devices stressed in vacuum. It is therefore likely
that gate-edge degradation represents a multistep failure process as proposed by
Li et al. [40]: groove formation initiated by electrochemical reaction eventually
results in the creation of surface pits providing increased I G . Subsequently,
thermal effects in combination with converse piezoelectric effect generate cracks
and expand the defective area, increasing dispersion effects and I D degradation.
Several authors [10, 11] pointed out that I G degradation in AlGaN/GaN RF
HEMTs can appear even well below V crit if constant reverse V G is applied
for sufficiently long time. Marcon et al. [10] reported comprehensive V G
step-stress experiments, demonstrating voltage and temperature dependence of
time-dependent gate breakdown occurring below V crit . It was shown that t BD
follows the Weibull distribution and exhibit strong voltage-accelerated kinetics
and weak temperature dependence. This means that V crit is only relevant for
specific step-stress conditions, and more importantly, I G degradation represents
a time-dependent phenomenon similar to dielectric breakdown [12]. I G degrada-
tion under off-state was attributed to percolative model, where leakage paths are
correlated with EL hot spots without significant I D degradation and structural
damage appearance [36]. In contrast, on-state stressing of the same devices at
elevated T resulted in oxygen-containing pit/crack formation, underlining a vital
effect of thermal stress for gate-edge degradation.
6.2.1.3 Hot Electron Degradation

When GaN HEMTs are operated in on-state at high V DS , electrons in the chan-
nel, accelerated by high electric field, gain a high energy. These hot electrons can
interact with lattice defects and hot phonons in the channel, causing irreversible
degradation of I D . Further, hot electrons can be transferred into the barrier and
passivation layer, where they become trapped and enhance trapping effects. A
common way to characterize hot electron effects in GaN HEMTs is to measure
EL emission. As shown in Figure 6.5a, EL intensity shows a nonmonotonic
dependence on V GS for constant V DS [13]. With increasing V GS above V th , the
EL intensity first increases because of hot electron population in the channel
and then drops as the electric field (V GD ) decreases at positive V GS , suggesting
that light emission is proportional to the product of electron density and electric
field in the channel. Another feature linking EL intensity to hot electrons is
the linear dependence of the EL intensity on exp(−1/(V DS −V DSAT )) known
as Chynoweth’s law for GaAs field effect transistors (FETs) [46]. EL intensity
imaging can therefore be used to envisage electric field distributions, allowing
100
EL intensity (a.u.)
“T1” device Tel 3450 K
VDS from 106 Tel 2400 K
LG = 1 μm 107 Tel 2850 K 105
80 20 V
104
EL intensity (a.u.)
106 MB
Intensity (a.u)
103
60 1.0 1.5 2.0 2.5
105 Photon energy (eV)
40
104 Corrected data
Exponential equation
20 103 Bremsstrahlung equation
Maxwell–Boltzmann
0 102
–5.5 –3.5 –1.5 0.5 2.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Photon energy (eV)
(a) VG (V) (b)
Figure 6.5 (a) EL intensity in AlGaN/GaN HEMT as a function of V GS from pinch-off (−5.5 V) to
2.5 V at various V DS (step 2.4 V). Source: Reprinted with permission of Meneghesso et al. [13].
Copyright © 2008, IEEE. (b) Measured EL spectrum of AlGaN/GaN HEMT at V DS = 20 V and
V GS = 0 V after correction using a Fabry–Perot etalon transmittance for the SiNx passivation
(black), fitted using the simplified exponential equation (red dotted curve), and the full
analytical expression for bremsstrahlung (green dashed curve). The Maxwell–Boltzmann
distribution (MB) is also shown for comparison (blue dash-dotted curve). The inset shows
fitting of the experimental data with simple exponential dependence without correction for
interference fringes in the higher energy tail of the spectrum. Source: Reprinted with
permission of Brazzini et al. [45]. Copyright © 2016, IOP Publishing.
monitoring of stress-induced changes in the electric field profile resulting from

trap generation [21, 47]. Moreover, Meneghini et al. [48] found a dependence
between the degradation rate of on-state stressed devices and the number of hot
electrons in the transistor channel deduced from the measured EL intensity.
Spectra of EL emission in operated GaN HEMTs have been studied by several
groups [49, 50]. The shape of the EL spectra in GaN devices shows only sub-band
emission with the EL intensity roughly proportional to exp(−C Eph ), where C is a
constant and Eph is the energy of emitted photons. As high energy tail of this spec-
tra follows Maxwell–Boltzmann distribution of carriers f (E) ∼ exp(−E/k B T e ), its
slope is commonly used to extract the hot electron temperature T e [21, 47]. It
has been found that T e increases approximately linearly with V DS (for constant
V GS ) and spatially peaks at the drain side of the gate where electric field is highest
[21]. A combination of T e and the number of hot electrons (proportional to EL
intensity and being highest for semi-on-state) then allows to evaluate the severity
of the “hot-electron-stress” applied to the device under test.
Despite intensive application of EL microscopy and spectroscopy for evalua-
tion of hot-electron-related degradation, there is a still ongoing debate on the
origin of the EL emission. This is mostly due to the fact that the EL spectrum
is affected by interference fringes related to multilayer heterostructure of GaN
devices. Gütle et al. [51] measured the EL spectra with Eph ranging from 1.2 to
3.2 eV and based on the observed features associated the dominant EL mech-
anism with the hot-electron-radiative intervalley transition. Recently, Brazzini
et al. [45] applied correction of the measured spectral distribution to the inter-
ference fringes in the Eph range of 0.8–2.8 eV and obtained almost featureless
spectral response as shown in Figure 6.5b. Based on the derived analytical expres-
sion, the EL mechanism was ascribed to bremsstrahlung, i.e. deceleration of hot
electrons by charged scattering centers (e.g. point defects, charged dislocations,

and interface roughness). This interpretation was further supported by measure-
ments with different wafers (thus different densities of defect centers) and light
polarization, showing preferential scattering direction of hot electrons [45].
Hot electrons have been known to provide sufficient energy to convert
pre-existing defects into a metastable configuration in GaAs (well-known EL2
defect) and release H atom from the passivated dangling bond at the SiO2 /Si
interface. Hydrogen release of hydrogenated point defects by hot electrons has
been systematically investigated as a possible degradation mechanism using
first-principle density functional calculation studies [52–54]. Puzyrev et al. [54]
proposed that release of H from hydrogenated Ga vacancies (VGa Hx ) and nitro-
gen antisites (NGa Hx , where x = 1, 2, and 3) can account for hot-electron-related
degradation of I D in GaN HEMTs. Calculated formation energies suggest
that hydrogenated vacancies and antisites are favored over the bare defects
during growth in ammonia-rich MBE (molecular beam epitaxy) or MOCVD
(metal-organic chemical vapor deposition), where H is readably available during
growth. In typical n-type channel GaN HEMTs, VGa H3 and NGa H2 are neutral
defects with no empty level in the bandgap and therefore benign as they cannot
trap carriers and do not act as Coulomb scattering centers. However, when
hydrogen is released by hot electrons, these defects become electrically active
defects, and stripping of each H atom results in an increase or decrease in the
number of acceptors for VGa H3 and NGa H2 , respectively. Increased number of
charged trapping centers can then lead to I D degradation while the interplay
between a number of acceptors affects the V th shift of the device. The energy
needed to remove a hydrogen atom from the VGaH3 complex and place it to the
H2 molecule was calculated to be 2.2 eV [52–54].
6.2.2 InAlN/GaN HEMTs

InAlN/GaN HEMTs offer two main advantages over AlGaN/GaN devices. First,
InAlN can be grown lattice matched to GaN for an In composition of 17% [55].
The absence of lattice strain naturally held a promise of enhanced reliability as
lower converse piezoelectric effect can be expected in these devices compared
to AlGaN/GaN counterparts. Indeed, any significant I G or physical degradation
of the gate-edge region has not been reported yet for stress conditions similar
or even more severe than in the case of AlGaN/GaN devices. Second, large
difference in the spontaneous polarization between InAlN and GaN results
in a high 2DEG density (up to 2.5 × 1013 cm−2 ), which translates into higher
current density and therefore higher output power of InAlN/GaN HEMTs
compared to AlGaN/GaN devices. On the other hand, high current densities
result in considerable self-heating effects, increasing the channel temperature.
In addition, because of higher vertical electric field component that adds up
to the lateral component, considerably higher energy of hot electrons (and
consequently hot phonons) has been predicted to exist in InAlN/GaN HEMTs
as compared to AlGaN/GaN devices [56, 57]. Indeed, dominant hot electron
and hot-phonon-related degradation mechanisms have been reported by
several groups [57–60]. Therefore, advanced strategies for lateral electric-field
mitigation and thermal management are imperative for reliable InAlN/GaN RF

HEMTs.
6.2.2.1 Hot Electron Degradation

Among the first reliability studies, Kuzmik et al. [61] investigated early
degradation of lattice-matched InAlN/GaN HEMTs subjected to off-state,
semi-on-state, and reverse gate bias stress. Only reversible changes were
observed for V GD < 38 V for all stressing conditions attributed to capture and
emission of pre-existing traps. For higher V GD , only temporal and negligi-
ble changes in I G were reported under reverse gate bias stressing. However,
strong degradation of intrinsic channel resistance (an order of magnitude)
accompanied by a decrease in I D was observed for off-state and semi-on-state
stressing, which was attributed to formation of new or ionizing pre-existing
defects by hot electrons in the GaN buffer and the InAlN/GaN interface. The
impact of hot electrons was later studied using hydrodynamic calculations of
hot electron distributions in AlGaN/GaN and InAlN/GaN HEMTs [57]. Devices
with simple GaN buffer structure (referred therein to as single-quantum well
(QW)) and GaN/Al0.04 Ga0.96 N back-barrier structure, referred to as double-QW
heterostructure were analysed. Calculated results for a single-QW HEMT are
shown in Figure 6.6, comparing the distribution of hot electron density and their
T e along the channel of AlGaN/GaN and InAlN/GaN HEMTs located 11 nm
below the QW. The thickness of the barrier layer (22 nm for Al0.22 Ga0.78 N and
13 nm for In0.17 Al0.83 N) was chosen so that both devices show the same V th .
The calculations indicate that the maximum energy of injected electrons in the
buffer can be expected to reach T e of about 20 000 K for InAlN/GaN HEMTs
at pinch-off (V DS = 20 V, V GS = −8 V), while considerably lower maximum T e
of about 7000 K was observed for AlGaN/GaN HEMTs. It was argued that the
higher T e for InAlN/GaN HEMTs originates from a higher vertical electric field
(resulting from higher polarization-induced 2DEG density and thinner barrier)
that deflects the channel electrons into the buffer. The observed threshold
(V GD ∼ 38 V) for intrinsic channel resistance degradation [61] was therefore
linked to the onset of dehydrogenation of defects by hot electrons available
at relatively high density in the buffer, as can be inferred from Figure 6.6b.
In another study [57], much smaller hot-electron-related degradation was
observed for double-QW HEMTs compared to single-QW devices subjected
to short-term off-state stress (V DS = 20 V, V GS = −8 V, and 1 hour) [57]. The
improved reliability of double-QW HEMTs over those with single-QW was
attributed to effective blocking of hot electron injection into the buffer by a back
barrier. This was indicated by the calculations, even though hot electrons have
similar energy in both types of structures.
Higher T e in InAlN/GaN RF HEMTs compared to AlGaN/GaN devices was
also confirmed experimentally based on EL spectroscopy measurements [58].
Figure 6.7a compares the measured EL spectra on InAlN/GaN and AlGaN/GaN
HEMTs with a similar geometry. From the simple exponential fit to the experi-
mental data, the T e of ∼2600 K was determined for AlGaN/GaN HEMTs, while
almost a twofold higher T e of 4700 K was found for InAlN/GaN devices. The lat-
ter was also confirmed by hydrodynamic simulations, where considerably higher
30000 1020 30000 1020

Gate Gate
Electron concentration (cm–3)

Electron concentration (cm–3)
25 000 25 000
Electron temperature (K)
Electron temperature (K)

1015 1015
20 000 20 000
15 000 1010 15 000 1010
10000 10000
105 105
5000 5000
0 100 0 100
11.6 11.7 11.8 11.9 12.0 12.1 11.6 11.7 11.8 11.9 12.0 12.1
(a) Distance x (μm) (b) Distance x (μm)
Figure 6.6 Cross sections of the electron concentration and temperature in the buffer of (a)
Al0.22 Ga0.78 N (22 nm)/GaN and (b) In0.17 Al0.83 N (14 nm)/GaN at V GS = −8 and V DS = 20 V along
the GaN channel at a distance of 11 nm from the interface. Devices have gate lengths of
0.25 μm, and source-to-gate and gate-to-drain distances are 1.5 μm. Source: Reprinted with
permission of Kuzmik et al. [57]. Copyright © 2012, The Japan Society of Applied Physics.
80 1.2
Vds = 30 V VDS = 5V, Pristine
InAIN/GaN VGS = 0
Vgs = 0 V
EL Intensity (a.u.)
1000
60 0.9
ΔID.max (%)
AIGaN/GaN
ID (A/mm)
100 0.6 Stressed
40
Fit
0.3
10 20
VGS = 0
InAIN/GaN Change in ID Resonance
Curve fit 0.0 Off-resonance
1.2 1.4 1.6 1.8 2.0 2.2 2.4 0
8 10 12 14 16 0 4 8 12 16 20
(a) Energy (eV) (b)
2DEG density during
(c) VDS (V)
stress (1012cm–2)
Figure 6.7 (a) EL spectra of InAlN/GaN and reference AlGaN/GaN HEMTs operated at
V DS = 30 V and V GS = 0 V. The oscillations in the spectra are related to Fabry–Perot interference
fringes and therefore artifacts. Inset shows the overlap of optical and EL image of InAlN/GaN
HEMT. Source: Reprinted with permission of Ťapajna et al. [58]. Copyright © 2014, IEEE. (b)
Stress-induced change in ID vs. 2DEG density during the stress and the Lorentzian fit to the
data. (c) Corresponding ID –V DS characteristics measured at V GS = 0 V for InAlN/GaN HEMTs
before and after stress performed at V DS = 20 V and V GS biased to the hot phonon–plasmon
resonance (closed diamonds) and at the off-resonance (open circles). Source: Panels (b) and (c)
are reprinted with permission of Kayis et al. [59]. Copyright © 2011, AIP Publishing.
calculated values of maximum T e (27 000–29 000 K) were shown to fit well to the
experimental data (4700–8000 K) when a volumetric effect of the light collecting
efficiency is considered. InAlN/GaN devices subjected to harsh off (V GS = −7 V)
and semi-on (V GS = −3 V) state stressing (V GD = 50, 75, and 100 V) showed
only negligible degradation of I G for all bias conditions. However, the strongest
degradation of I D , intrinsic channel resistance, and pulsed I–V characteristics
was observed for devices stressed in the semi-on-state. Such degradation modes
clearly point to hot-electron-related degradation. The degradation mechanism
was attributed to dehydrogenation of pre-existing hydrogenated point defects
(e.g. Ga vacancy and divacancy), as the determined hot electron energies (>2 eV)
are well beyond the energies predicted for dehydrogenation of these defects (2.2
and 2.0 eV, respectively) [52–54].
Recently, Downey et al. [62] studied the reliability of RF SiNx /InAlN/GaN
metal–insulator–semiconductor (MIS) HEMT with 120 nm gate length sub-
jected to RF stress at 40 GHz biased in class AB (V DS = 20 V). Although
MISHEMTs with a 3- and 6-nm thick SiN gate dielectric showed only a minor
degradation of the output power (∼1 dB), devices with Schottky gate and 1-nm
thick SiN dielectric degraded quickly within the first 10 hours of stressing. The
output power degradation was associated with the observed temporal (recov-
erable upon UV light exposure) decrease in the maximum I D and g m at positive
V GS , while negligible degradation of I G was observed. Based on the simulations
indicating that the gate-edge lateral field increases from 3.3 to 5.2 MV/cm as
the SiN thickness is decreased from 6 to 1 nm, the degradation was ascribed to
hot-electron-induced trapping in the gate-to-drain access region. Importantly,
several devices were also stressed in DC operation in the semi-on-state with
bias conditions similar to RF stressing (V DS = 20 V and I D = 200 mA/mm). A
similar degradation in I D and g m was observed for both conditions, illustrating
the applicability of DC HTOL testing for RF devices.
6.2.2.2 Role of Hot Phonons

An alternative mechanism for hot-electron-related degradation is connected to
hot phonons localized in the channel of a GaN HEMT operating at high power
[63]. Generation of hot phonons originates from strong electron–phonon cou-
pling in GaN, where hot electrons tend to lose their energy mainly via emission
of longitudinal optical (LO) phonons. However, because of low group velocity
of LO phonons, they remain localized in the channel and their energy dissipa-
tion is limited by conversion to other modes with higher group velocity such as
longitudinal acoustic (LA) phonons that subsequently diffuse into a remote heat
sink. Because emission of LO phonons by hot electrons is faster than their decay
into LA modes, nonequilibrium population of LO phonons (hot phonons) builds
up in the channel and is commonly treated in terms of LO phonon lifetime. If
LO phonon lifetime is long enough, the accumulation of hot phonons occurs and
results in enhanced carrier scattering and device performance degradation [56].
The LO phonon lifetime was found to depend on the electron density in the
GaN bulk as well as 2DEG [63], which is explained by a plasmon–LO–phonon
coupling model [64]. For a 2DEG channel, the shortest lifetime was observed at
an electron density of 6.5 × 1012 cm−2 at low field [63] and 9.3 × 1012 cm−2 at high
fields [65], when the plasmon and LO phonon frequencies reach a resonance.
The resonant 2DEG density provides strong enhancement in LO phonon decay
and can in turn be expected to minimize the channel degradation. Based on
this notion, Leach et al. [66] studied the dependence of transistor degradation
rate on 2DEG density of lattice-matched InAlN/GaN HEMTs. Devices were
subjected to high-field stress (V DS = 20 V) at V GS ranging from −3 to −7 V
for stressing time corresponding to 1500 mA h/mm passing through the drain
and the decrease in maximum I D was monitored as a prominent feature of
degradation. The degradation rate exhibited a clear minimum at 2DEG density
around 1013 cm−2 , similar to that shown in Figure 6.7b, despite the fact that
devices stressed at lower 2DEG densities were subjected to lower power densities
and thus lower T j . This behavior was attributed to hot phonon build-up with the
least degradation at phonon–plasmon resonance where the shortest lifetime of
hot phonons is expected. Similar results were presented by Kayis et al. [59], who
also analyzed degradation rate in InAlN/GaN HEMTs as a function of 2DEG
density. As shown in Figure 6.7c, the minimum degradation of I D was observed
at a 2DEG density of 9.2 × 1012 cm−2 (Figure 6.7b). The low-frequency phase
noise of the degraded devices was found to be fully consistent with maximum
I D degradation behavior. Although relatively small stress-induced noise increase
(12 dB/Hz) at the resonance conditions was ascribed to mild trap generation in
the channel, much stronger noise increase for off-resonance stressed devices
was attributed to higher trap generation rate in the barrier and channel region.
In both studies, negligible I G degradation was observed. Finally, Zhu et al. [60]
investigated the degradation of In0.16 AlN0.84 /GaN HEMTs subjected to four
different stress conditions. Negligible I D and noise spectra density degradation
took place for on-state low-field stress performed at 200 ∘ C and reverse gate
bias stress. Although off-state high-field stress led to a notable degradation of
electrical parameters, on-state high-field stress condition was found to be the
most deteriorating, causing strong degradation of maximum I D and g m and
increase in noise-spectral density, which was ascribed to increased intrinsic
channel resistance. These results clearly point to dominant channel degradation
caused by hot electron and hot phonon effects.
6.2.3 Thermal Issues in RF GaN HEMTs

RF GaN HEMTs are designed to operate at high power densities up to 10 W/mm.
A product of high current density and electric field in the on-state results in
Joule self-heating, generating temperature rise in the channel. In GaN HEMTs,
the peak channel temperature (T j ) therefore occurs close to the gate edge at the
drain side, where electric field is the highest. Design strategies ensuring effective
waste heat extraction must be adopted to remove the excessive heat generated.
One of the most widely employed approaches is the use of a substrate with
sufficiently high thermal conductivity (𝜅). This is why high-performance RF GaN
HEMTs are typically manufactured on SiC substrates. Yet heat transport across
interfaces between the GaN channel and the substrate as well as that of the die
attaches and the carrier in packaged devices needs to be carefully optimized.
Although T j can be calculated using thermal or thermo–electro–mechanical
simulations (see [39]), some critical parameters such as thermal resistance (RTH )
of GaN/SiC transition (nucleation) layer depends on the (often proprietary)
epitaxial growth conditions. These parameters are usually unknown, which leads
to erroneous T j predictions. In addition, relatively high 𝜅 of materials com-
prising typical GaN HEMT (𝜅 GaN ∼ 1.6 W/cm K and 𝜅 SiC ∼ 4 W/cm K at room
temperature) results in very localized heat generation in the form of a hot spot
and high thermal gradients surrounding it, making experimental T j extraction
challenging. In accordance with the current standard [3], T j is often determined
using a combination of IR imaging of the device-active area and finite element
(FEs) thermal simulations. However, IR imaging is a thermography technique
based on the collection of thermal radiation emitted from the device surface (and
volume if IR-transparent substrate such as SiC) and its spatial resolution limit
is on the order of the used wavelengths (∼5.5 μm at T = 250 ∘ C). Temperature
measured by IR thermography is therefore averaged over a very large volume
as compared to the hot spot volume. Even with the aid of thermal simulations,
extracted T j can be greatly underestimated, leading to uncertainties in the device
lifetime predictions. In this section, we will therefore briefly introduce optical
and electrical techniques typically adopted for T j determination and highlight
some technological aspects affecting thermal properties with consequences to
reliability testing of GaN HEMTs.
Among optical methods, micro-Raman thermography is the most widely
used technique for thermal effect characterization in GaN HEMTs [9]. In
typical configuration, source–gate and gate–drain openings of a device are
irradiated by a sub-bandgap energy laser, and the temperature is derived
from temperature-dependent shift in the phonon frequency of GaN E2 (high)
and/or A1 (LO) phonon modes. Because the phonon shift also depends on
the built-in mechanical and on-state thermoelastic stress present in the GaN
layer (E2 (high) being more stress sensitive than A1 (LO)), calibration against
elevated base-plate temperature (T b ) must be performed for the investigated
heterostructure. Impact of converse piezoelectric effect in the operated device
can also be corrected for by subtracting the phonon shift measured under
pinch-off condition, or in more advanced analysis, temperature and biaxial stress
can be measured independently by simultaneous analysis of E2 and A1 phonon
modes [38]. Although submicrometer (0.5–0.7 μm) lateral resolution can be
obtained, the measured temperature is averaged through the entire GaN layer
thickness. Surface temperature (several tens of nanometers) can be measured by
using an above bandgap laser [67]. T j is then obtained from thermal simulation
by fitting the temperature averaged through the GaN layer to the experimental
data. Other optical methods probing the operated device temperature include
thermal reflectivity mapping, studying the temperature-dependent change in
the amplitude [68] or phase shift of the reflected light (also known as transient
interferometric mapping, TIM) [69], and microphotoluminescence exploiting
the temperature dependence of the band-edge emission [70].
Electrical methods for T j determination are noninvasive, fast, and can be per-
formed on widely available characterization setups. Most of them rely on the
relation between self-heating and the change of saturated I D (ΔI D,sat ) [71–73].
Depending on the technique, linear dependence between dissipated power and
ΔI D,sat or RDS,ON (measured in DC [71] or pulsed mode [73]) is used to determine
T j based on the calibrated changes in I D,sat measured at different T b . Kuzmík et al.
[72] derived a simple formula relating ΔI D,sat = −g m (I D,sat ΔRS + V th ). Using cal-
ibrated temperature dependences of RS , V th , and g m (measured at small V DS at
different T B ), T j can be calculated iteratively from DC or pulsed ΔI D,sat measure-
ments. All these techniques average T j in the entire active region of the device.
The latter method has been recently refined by Florovič et al. [74], where channel
potential variation along the device width was derived, allowing T j profile deter-
mination. Different principle is exploited in gate resistance thermometry method
proposed by Pavlidis et al. [75]. Here, the temperature coefficient of the gate
metallization resistance was used for T j measurement using specially designed

structure connected to one of the gate fingers. In contrast to ΔI D,sat methods,
this technique averages the gate metal temperature along the device width.
For a given dissipated power density, T j depends mainly on the device layout
(gate width and gate pitch) and on the efficiency of generated heat removal, which
into large extent occurs through the substrate for lateral GaN devices. The latter
requires minimization of thermal resistances of the epitaxial layers constituting
the die (GaN channel, the substrate, and the thermal boundary resistance (TBR)
between these layers) but also the thermal resistance of the die attach (solder) and
chip carrier for power devices with large gate periphery. Because GaN is typically
grown on foreign substrates, it contains a large number of threading dislocations
that were shown to have a negative impact on GaN layer thermal conductiv-
ity for densities higher that 107 cm2 [76]. Therefore, both quality of the GaN
epitaxy for a particular substrate (depending on lattice and thermal expansion
coefficient mismatch) and the thermal conductivity of the substrate itself must be
considered in thermal management. These parameters together with typical GaN
HEMT thermal resistances measured by Raman thermography are summarized
in Table 6.3. This illustrates the main reason why SiC is the substrate of choice
for high-performance power RF applications, while Si is attractive for low-cost
applications. Special attention needs to be paid to thermal properties of the nucle-
ation layer when GaN is grown on SiC. Manoi et al. [79] analyzed effective TBR
(TBReff ) for various AlGaN/GaN/SiC wafers using Raman thermography. It was
found that apart from true TBR caused by the phonon mismatch between two
different materials, variation in the thermal conductivity of AlN nucleation layer
was the main contributor to TBReff (up to the factor of 4) and can account for
as much as 30% of the total RTH of GaN-on-SiC HEMTs. TBReff was found to be
temperature dependent, increasing with interface temperature [79].
A similar effect of degraded heat removal properties was reported for power
switching p-GaN/AlGaN/GaN HEMTs incorporating Al0.05 Ga0.95 N buffer
layer grown on Ar-implanted SiC substrate [80]. AlGaN:uid buffer instead of
Fe-doped GaN was proposed to enhance the breakdown voltage of switching
devices without negative impact on dynamic RDS,ON while Ar implantation into
Table 6.3 Reported lattice mismatch, thermal expansion coefficient mismatch for
GaN grown on various substrates, substrate thermal conductivity/temperature
dependences, and GaN HEMT thermal resistances.
Thermal Thermal
expansion conductivity AlGaN/GaN
Lattice coefficient at 25 ∘ C/T- HEMT thermal
Substrate mismatch mismatch dependence resistance
material (%)a (%) [77] (W/m K) [9] (∘ C mm/W)
GaN 0 0 160 × (300/T)1.4 15 [78]

Sapphire 14 34 24 43 [9]
4H–SiC 3.8 25 420 × (300/T)1.4 15 [9]
Si −17 56 130 × (300/T)1.5 25 [9]
250 225
10 fingers 20 fingers 50 fingers
B 4.4 W/mm 200
200 Die
175
Temperature (°C)
GaN
N.L
150 SiC
150 Ar-implantation
ΔT (°C)
125 Brazing
and package
A 4.7 W/mm
100 Die attach
100 Carrier
75
AIGaN Buffer Test fixure
50 50 TIM
C 4.7 W/mm Test fixture
D 4.5 W/mm
25
0 0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 10 20 30
Pdiss (W)
(a) Time (μs) (b)
Figure 6.8 (a) Extracted and simulated transient temperature rises for structures with
AlGaN:uid (A, B) and Fe-doped GaN (C, D) buffer. Arrows highlight the changes in self-heating
by different technological designs. Power dissipation level is labeled at each transient. Source:
Reprinted with permission of Kuzmík et al. [80]. Copyright © 2014, IEEE. (b) Simulated
distribution of temperature drops across layers comprising packaged AlGaN/GaN HEMTs of
various sizes (10, 20, and 50 finger devices) attached to the test fixture, showing contributions
from the GaN layer, nucleation layer (NL), SiC substrate, die attach, chip carrier, and test fixture.
Total dissipated power density was chosen to give the same peak T j for the devices compared.
Source: Courtesy of M. Kuball [81].
cheaper n-type SiC (as compared to semi-insulating SiC substrate) prevented

vertical breakdown. Figure 6.8a shows average T j rise measured under transient
operation for dissipated power density range of 4.4–4.7 W/mm, determined
by electrical technique described above [72]. Clearly, application of AlGaN
buffer as well as Ar implantation significantly influence the self-heating effect
with the highest temperature rise for the devices combining AlGaN buffer and
Ar-implanted SiC. While the former effect can be attributed to well-known sharp
decrease in thermal conductivity of AlGaN with low Al composition [82], the
negative impact of Ar-implantation into SiC on T j rise was linked to variation in
Al incorporation during the growth and/or decrease in SiC thermal conductivity
because of enhanced implantation-induced defect phonon scattering [80].
In small two- to four-finger devices often used in the research studies, the
amount of heat flow is low, and owing to the small heated volume, most of the
temperature drop occurs across the die, while the thermal resistance of the die
attaches and the test fixture is negligible. However, in power bars with large gate
periphery used in real applications, device layout (gate width, gate pitch) and
packaging materials have significant impact on resulting RTH . The amount of heat
flow through all the constituting layers to the heat sink is significantly higher,
yielding higher temperature drops not only across the die but also across the die
attach, chip carrier, and test fixture (if present). Also, temperature dependence
of thermal conductivity of used materials further increases RTH . The influence
of the device size and the corresponding amount of the heat flow are illustrated
in Figure 6.8b, showing simulation of the different contributions to peak T j in
each layer in a packaged device as a function of dissipated power for various
numbers of gate fingers [81]. It clearly shows that the thermal resistance of the
die attach and the chip carrier material can represent almost 50% contribution
to T j in multifinger devices. It has recently been shown that in comparison to
commonly used AuSn solder (𝜅 ∼ 57 W/m K), AuSi- or Ag-sintered epoxies can
decrease the T j rise by up to 4% for investigated 10-finger GaN/SiC devices. In
addition, replacement of CuW or CuMo alloys typically used as chip carriers by
Ag-diamond composites can decrease T j rise by 12% in these devices [83]. These
effects underline the necessity of experimental determination of T j for device
reliability testing.
In the lifetime prediction of RF GaN HEMTs, HTOL should be ideally per-
formed on a number of devices in the actual RF conditions. However, because of
the complexity of RF-HTOL tests, lifetime prediction of large number of devices
is typically carried out in the DC regime using bias conditions resulting in simi-
lar average power dissipation as in the RF operation. An obvious question arises
whether both tests provide the same lifetime prediction. An example is given in
Figure 6.1b, where RF- and DC-HTOL tests gave similar Ea ; however, apparent
offset between Arrhenius slopes can be inferred for the two conditions, result-
ing in a longer projected lifetime of the RF-HTOL compared to the DC-HTOL
test. Pomeroy et al. [2] investigated the possible impact of different tempera-
ture distributions in AlGaN/GaN/SiC HEMTs during RF and DC HTOL test
on the observed Arrhenius data offset. Using electrical and thermal simulations
calibrated by Raman thermography experiments, the averaged channel tempera-
ture for class-B load line operation was compared to channel temperature profile
at DC bias point, matching the same power dissipation. The simulations sug-
gest marginal differences in lateral temperature profiles for both conditions at
V DS = 30 V, and only slight decrease in maximum T j (∼9%) for load line opera-
tion compared to DC operation for V DS = 100 V. The difference originates from
pinch-off of the field-plate electrode (occurring for V DS > 60 V) in the DC mode,
spreading the electric field and thus Joule heating toward the drain contact. In
contrast, majority of the heating contribution is averaged around the center of
the load line (V DS ∼ 50 V), i.e. below the field-plate electrode pinch-off voltage.
In any case, such marginal differences in T j profile between DC- and RF-HTOL
testing cannot explain the observed offset in the Arrhenius plot, and this effect
is unlikely to result from thermal effects. Instead, the offset shown in Figure 6.1b
has been attributed to different failure criteria chosen for the two tests [2].
6.3 Reliability and Robustness of GaN Power Switching

Devices
In order to match the different applications of power devices compared to RF
devices, bias sequences and designs are significantly different, which also cause
different failure modes. To give an overview and classification of possible failure
and degradation modes in GaN power HEMTs, five different aspects have to be
considered, with detailed discussion in the subsequent sections:
1. Stress condition: The majority of the time, a transistor is in off-state or
on-state, which can easily be reproduced for qualification. The transition
Semi-on Figure 6.9 Typical V GS , V DS , and ID waveforms

Off On Off On for a hard switching application. Besides
off-state and on-state, the semi-on-state
during turn-on with simultaneous high V DS
VGS and ID is indicated.
VDS
ID Time
between these two states, i.e. switching, gives added complexity. The exact
switching loci of the turn-on and turn-off crucially depend on the application
and cannot be tested by a general test. Figure 6.9 shows a hard switching
application with simultaneous presence of high current and high voltage
(semi-on) causes a much more severe stress than for soft switching applica-
tions without this semi-on-state. The reliability of GaN HEMTs under certain
operation modes is commonly tested by application-relevant stress tests,
i.e. HTOL.
2. Breakdown vs. trapping: Figure 6.10a illustrates that on the one hand, various
parts of the GaN HEMT can permanently degrade and lead to an increase of
the leakage current, which can eventually cause destructive breakdown [84].
On the other hand as shown in (b), charges can be trapped, which is usually a
recoverable process; however, this can also lead to critical parameters such as
the on-resistance not meeting the specifications anymore [85]. Complicating
things, both processes also interact with each other. For example, charge trap-
ping can lead to an increase in the on-resistance, which can lead to thermal
destruction, or degradation can create defects that can then capture charges.
3. Location of degradation: Figure 6.10 demonstrates regions in the HEMT,
which are prone for (a) degradation and breakdown as well as for (b)
trapping. One of the most important regions is the buffer, here defined
as carbon-doped GaN (GaN:C) + transition layer, which determines the
vertical drain–substrate breakdown [84, 86] and the lateral source–drain
breakdown [87] as well as trapping-related issues such as dynamic RDS,ON ,
current collapse [28], and might lead to a V th drift [88]. The second important
region is the gate stack that can degrade and cause breakdown [89–91], but
in combination with its bottom interface, it can also cause trapping-related
issues such as V th drift, negative bias temperature instability (NBTI) [92],
and positive bias temperature instability (PBTI) [88]. As these regions are
the most important, Sections 6.3.1–6.3.3 will cover them in more detail.
There are additional failure modes such as lateral gate–drain breakdown [16]
Source Gate Lateral G–D Drain

breakdown
AlGaN
GaN:uid 2DEG
Gate
Vertical D-sub
GaN:C Lateral S-D
breakdown
breakdown breakdown
Transition layer
Si substrate
(a)
Source Gate Drain
e– e–
AlGaN e– e –
GaN:uid 2DEG
e– e–
e– e–
GaN:C dyn. Rds,on
V th drift, Current
NBTI, PBTI collapse
Transition layer
Si substrate
(b)
Figure 6.10 Schematic cross section of a GaN-on-Si HEMT indicating (a) relevant breakdown
paths as well as (b) regions prone to charge trapping, which affects the device performance.
and trapping at the AlGaN surface [93] or in the passivation [94], which go
beyond the scope of this section.
4. Device design: Although this book covers only HEMTs, the designs still vary
significantly especially concerning the gate stack. Although it can be rather
simple for normally-off devices, normally-on devices require more complex
structures. Common designs are stacks incorporating a p-doped GaN layer as
in gate injection transistors (GITs) in Figure 6.11a or MIS structures as in (b).
5. Environment: The majority of physical processes show to different extent a
dependence on the temperature, but other environmental conditions such as
extended humidity can also have an effect on the reliability of HEMTs.
6.3.1 Parasitic Effects in the Carbon-Doped GaN Buffer

6.3.1.1 Insulation of GaN Buffer by Carbon Doping
GaN-based power devices are predominantly grown on rather conductive Si sub-
strates that are grounded together with the source. A bad insulating GaN buffer
would allow large vertical leakage currents between drain and substrate as well
Source Gate Drain
AlGaN pGaN
GaN:uid 2DEG
GaN:C
Transition layer
Si substrate
(a)
Source Gate Drain

Gate
insulator
AlGaN
GaN:uid 2DEG
GaN:C
Transition layer
Si substrate
(b)
Figure 6.11 Schematic cross sections of GaN-on-Si normally-on HEMTs with the two most
common designs of the gate stack: (a) p-GaN gate and (b) MISHEMT.
as laterally between drain and source as shown in Figure 6.10a. Thus, the GaN
buffer has to be well insulating. The major problem therefore is the high con-
centration of unavoidable background donors, mainly originating from Si impu-
rities [95]. Figure 6.12a shows that these donors are energetically close to the
conduction band and make unintentionally doped GaN (GaN:uid) rather con-
ductive (n-type) by causing a high concentration of free electrons in the conduc-
tion band. However, these background donors can effectively be compensated by
adding carbon dopants, which pin the Fermi level approximately 0.7 eV above
the valence band [96]. This deep Fermi level results in the absence of free elec-
trons and holes in respective conduction and valence bands and makes GaN:C
so-called semi-insulating.
Carbon dopants are predominantly built in the GaN crystal, substituted at Ga
sites (CGa ) or at N sites (CN ). Schematic band diagrams in Figure 6.12b,c illus-
trate that CGa is reported to form donors energetically close to the conduction
band [97]. CN on the other hand forms acceptors with energy levels in the range
of 0.5–1.1 eV above the valence band [98]. In general, defects are occupied when
sufficiently below the Fermi level and unoccupied when above. Hence, donors
like CGa are charge neutral below and positively charged above the Fermi level,
while acceptors like CN are charge neutral above and negatively charged below
the Fermi level. In thermal equilibrium, charge neutrality must apply, which is
Conduction
band Occupied
3
Unoccupied
Energy (eV)
2 Free electrons
Fermi level
1 Background donors
C-donors CGa
0 C-acceptors CN
Valence band
(a) GaN:uid (b) Auto- (c) Dominant
compensation acceptor
Figure 6.12 Schematic band diagrams show the energy levels, donor to acceptor ratios rd/a ,
and from these parameters derived occupancies of defects and conduction and valence bands
as well as the calculated Fermi levels. These calculations are compared for (a) GaN:uid,
considering only shallow background donors; and for GaN:C with rd/a values of 1 (b), leading to
autocompensation and (c) for 0.5, i.e. a dominant acceptor. While the ratio of donors to
acceptors are shown proportional for the sake of visualization, the part of background donors
in the overall sum of donors is displayed exaggeratedly.
achieved for a given doping profile only at one energy level. Figure 6.12b shows
if there are exactly as many acceptors and donors, all dopants are ionized, i.e.
donors positively charged and acceptors negatively charged. The Fermi level pins
exactly between their energy levels, and this is referred to as autocompensation.
However, this is in contradiction with experimental results such as the observed
Fermi level pinning at 0.7 eV [96]. Instead, the acceptor concentration is consid-
ered to be significantly higher than the donor concentration, which is called the
dominant acceptor model. Figure 6.12c illustrates the case for an acceptor con-
centration being twice as high as the donor concentration. In order to establish
charge neutrality, acceptors are half occupied while donors are fully unoccupied,
which is achieved only when the Fermi level pins exactly at the acceptor level, fit-
ting well to experimental results [96]. Although the ratio of donors to acceptors
must be significantly less than 1, the exact ratio is still debated [99] with suggested
rates between 0.4 and 1.0.
6.3.1.2 Time-Dependent “Dielectric” Breakdown (TDDB) of the GaN Buffer

Although carbon doping makes the buffer in general insulating, at high biases,
there is still a certain leakage current, and eventually, at certain conditions,
GaN can still experience breakdown. The nature of the buffer leakage current is
controversially discussed and assigned mainly to space–charge-limited current
or Poole–Frenkel conduction [100]. The most severe phase for buffer failure is
the off-state at enhanced temperatures, which can be tested by high-temperature
reverse bias (HTRB) similar to that in Si power devices, although the failure
mode is different. Figure 6.13a compares typical I D –V DS characteristics for Si
superjunction metal–oxide–semiconductor field effect transistors (MOSFETs)
and GaN HEMTs with voltage ratings of 650 and 600 V, respectively [6].
Exceeding the rated voltage in the Si device by only a few volts leads to avalanche
breakdown, where because of the large current increase, the temperature rises,
eventually causing thermal destruction. In contrast to this avalanche breakdown,
GaN devices exhibit a TDDB [86] that is well known from gate oxides such as
SiO2 [12].
TDDB
Avalanche BD
VDS=970 V
GaN
0.8
ID (mA) 10–3
ID (A)
Si
0.6
VBD criterion
0.4 VBD criterion
0.2 I = 250 μA ID = 250 μA
0
D 10–4
0 200 400 600 800 1000 100 101 102 103 104 105
(a) VDS (V) (b) Time (s)
Figure 6.13 (a) Comparison of ID –V DS for a 650 V-rated Si SJ (super junction) MOSFET and a
600 V-rated GaN HEMT at 25 ∘ C. Source: Data from McDonald 2018 [6]. (b) Schematic shows
that for the GaN device even at 970 V, which is well below the breakdown voltage in (a) and
marked by an arrow, breakdown occurs after a certain amount of time.
Figure 6.14 Weibull plot with simulated data points for demonstration purposes; every data
point represents one fail. (a) Identification of extrinsic and intrinsic failures and (b) voltage
acceleration for intrinsic failures, which allows prediction of the intrinsic lifetime under use
condition of 480 V. The gray star indicates that under use condition after 500 hours, the failure
rate is below 100 ppm.
Figure 6.13a shows that for a 650 V-rated GaN HEMT, the current in the
I D –V DS characteristic increases only after 800 V and destructive failure can occur
for fast sweeps well above 1000 V [6]. However, the schematic in Figure 6.13b
shows that even for voltages well below the breakdown voltage in (a) after a
certain time, the device also fails.
In order to predict the lifetime of a device, the temperature and voltage
dependencies of the failure time are tested for a certain sample size as shown
in Figure 6.14 for a simulated sample set. Although experimental data can be
found [84, 86], because of the higher complexity that is not fully understood,
dependencies and inconsistencies come from derived parameters. Therefore,
the discussion within this book is based on a simplified simulated data set with
qualitatively similar behavior. Although Si devices exhibit negligible voltage
acceleration, GaN devices show an exponential decrease of the lifetime with
increasing voltage, i.e. the Eyring model. As in Si, the temperature acceleration
obeys Arrhenius law but with a small activation energy less than 0.25 eV [86].
The steep slope in the Weibull plot, i.e. large 𝛽-values, indicates wear out failure
with an increasing failure rate with time, whereby the 𝛽-values are observed to
vary significantly with temperature [86]. Large 𝛽-values are in general considered
beneficial as they indicate low failure rates during the lifetime of a device.
The physical origin of the failure is speculated to lie in the polar nature of
the Ga—N bond [86]. The Ga—N bond is analogous to the Si—O bond in SiO2 ,
Figure 6.15 Schematic of the

transient formation of a percolation
path in an oxide film, with the final
breakdown due to conduction via Fresh Breakdown time
this path.
Pre-existing defects
Generated defects
Defects form percolation path
while in Si, the unpolar Si—Si bond prevents this form of time-dependent break-
down [5]. In an electric field, the polar bond leads to asymmetrical lattice dis-
tortion that strains the chemical bond, which eventually breaks after a certain
time. Figure 6.15 demonstrates that with increasing time, more and more bonds
break, thereby forming a conductive percolation path that leads to destructive
breakdown [84].
6.3.1.3 Dynamic RDS,ON Due to Buffer Trapping

Despite their great ability to make GaN insulating, the carbon atoms are also the
main source for the dynamic on-resistance (dynamic RDS,ON ) of GaN HEMTs. Its
principles can be best understood by discussion of the two phases of its most
common characterization technique, often referred to as current transient spec-
troscopy in Figure 6.16. The test sequence in Figure 6.16a consists of a stress
and recovery phase, which are discussed in more detail in Figure 6.16b,c, respec-
tively. During the stress phase, the device is usually in off-state and a drain bias
establishes an electric field within the GaN buffer (Figure 6.16b). This electric
field can not only lead to TDDB but also lead to trapping of electrons in carbon
acceptors in the buffer [28]. The problem of these trapped electrons is that after
switching from off- to on-state, the electrons are still trapped in the acceptors,
which decreases the 2DEG density and therefore increases RDS,ON . As illustrated
in Figure 6.16b,c, it takes a certain amount of time until the electrons disap-
pear and RDS,ON recovers. In HEMTs employing a p-GaN gate, i.e. GITs, trapped
electrons vanish faster as they recombine with injected holes from the gate in
Test sequence: Stress condition: Transient recovery

0V 0V 500 V
G
Normalized Rds,on (–)
Stress Recovery S AlGaN D 1.5

GaN:uid 1.4
VS = 0
500 V GaN:C Captured– – 1.3
electrons – – –
0.1 V 1.2
E-field
VD 1.1
Transition layer T
0V 4V 1.0
VG
Si substrate 10–1 101 102 104
Recovery time (s)
(a) (b) (c)
Figure 6.16 (a) Typical test sequence for current transient spectroscopy with indicated stress
and recovery phases. (b) Schematic cross section of a HEMT under stress condition with
indicated electric field and the resulting electron trapping in the GaN:C buffer. (c) Schematic
transient change of RDS,ON (i.e. recovery) in the recovery phase due to detrapping for different
temperatures.
1 Figure 6.17 Schematic output characteristic

Normalized Id Before trapping Current before and after trapping, illustrating the
dyn. collapse common definition of the current collapse and
Rds,on
After trapping dynamic RDS,ON .
0
0 2 4 6 8 10
Vds (V)
on-state [101]. However, detrapping time constants can vary between microsec-
onds and many hours, depending on the exact design, gate voltages, and temper-
atures [102].
The terms dynamic RDS,ON and current collapse are often used interchange-
ably to describe this phenomenon. Figure 6.17 shows a more precise distinction:
dynamic RDS,ON describes the decrease of the slope in the output characteristic,
while current collapse refers to the decrease of the saturation current. For rea-
sons of simplicity and as both phenomena are linked to similar trapping events,
within this book, we refer to dynamic RDS,ON , although the same applies to current
collapse.
Uren et al. argue that for discharging, the hole transport process determines
the time constants and therefore their temperature dependence, which indirectly
give information on the energy level of carbon acceptors [103]. In general, that
cannot be excluded, and Figure 6.16c demonstrates that with increasing tem-
perature, detrapping becomes faster. In most studies that employ current tran-
sient spectroscopy, temperature-dependent time constants are extracted in order
to derive activation energies with values in the range 0.5–1.1 eV [28, 98]. The
physical background for conventional trapping and detrapping is illustrated in
Figure 6.18a: on the right half, acceptors are charged, i.e. holes are emitted from
acceptors to the valence band. For this process, an energy barrier of Ea has to be
overcome, requiring the time 𝜏em,VB
h
. From the temperature dependence of 𝜏em,VB
h
,
Ea can be extracted. On the other hand during recovery, acceptors are discharged
as shown in the left half of Figure 6.18a. In this case, no significant energy bar-
rier has to be overcome. Newer studies [98, 103] point out that the temperature
dependence of the recovery as in Figure 6.16c cannot be directly related to the
energy levels of traps. Instead, in the left half of Figure 6.18a, the recovery time
constant is determined by the time it takes a charge to propagate from the 2DEG
to the location of the defect, i.e. the transport process (𝜏 trans ). The capture pro-
h
cess itself is much faster (𝜏cap,VB ) and can be neglected. Uren et al. [103] suggest
transport of holes in the valence band as in Figure 6.18a.
To get more insight into this transport process, Koller et al. [98] performed
detrapping and conductivity measurements on special test structures that allow
characterization of single GaN:C layers [96]. Figure 6.19a shows extracted time
constants for acceptor charging and discharging as well as the current in this
single GaN:C layer in a wide temperature range between 20 and 560 K. All the
three processes reveal the same non-Arrhenius-like temperature dependence.
Furthermore, at cryogenic temperatures as low as 20 K, trapping rates are still
high. Both are in clear contradiction to the conventional trapping model shown
in Figure 6.18a [98]. In Figure 6.19b, the acceptor discharging time constants are
Conventional model Defect band model

Conduction band
Acceptor Acceptor Acceptor Acceptor
discharging charging discharging charging
Carbon defect τtrans,DB τtrans,DB

EA τhcap,VB τhem,VB Defect band
τtrans τtrans
Valence band
τcharge = τtrans,VB + τhem,VB τcharge = τtrans,DB
(a) τdischarge = τtrans,VB + τcap,VB
h
(b) τdischarge = τtrans,DB
Figure 6.18 Schematic band diagrams with the processes that discharge (left half of (a) and
(b)) and charge (right half ) carbon acceptors considering (a) conventional charge transport in
the valence band [103] and (b) transport in a defect band [104]. The time constants with the
arrows indicate the physical processes required for acceptor charging and discharging
[98, 104].
106 10–3 14 ) –1
Acceptor discharging (aT K
105 Acceptor charging 10–4 12 exp 0.03
13
Itest structure
10–5 a= 5e
V
104 IHEMT@800V 10 12 E A =
0.0
In (τT2) (–)
10
10–6
I (A/cm2)
103 eV
1/τ (s−1)
8 = 0.13 11 150–200 K
9 EA 58 77
102 10–7
6 8 300–350 K
101 10–8 33 38
T) –1 V
(a K 10–9 4 5 2e
100 p
x 3 = 0.9
∼e 0.0 2 4
EA
10–1 = 10–10
a 3 530–560 K
10–2 10–11 0 21 22
0 100 200 200 400 500 600 20 30 40 50 60 70 80 90
(a) T (K) (b) 1/(kT) (eV−1)
Figure 6.19 (a) Comparison of acceptor charging and discharging rates (1/𝜏) and leakage
current through a single GaN:C layer within a special test structure (Itest structure ) and vertically
between drain and substrate in a HEMT test structure at 800 V (IHEMT@800V ). (b) Charging rate
from (a) in an Arrhenius plot in the large temperature range and for smaller temperature
ranges in the insets. The green area represents the temperature range in which all
conventional investigations were performed. Source: Data from Koller et al. 2017 [98].
shown in an Arrhenius plot emphasizing the non-Arrhenius dependence. How-

ever, the insets in the figure also show that within small temperature ranges, this
non-Arrhenius dependence is not obvious. As all previous literature studies on
GaN:C were performed within the green area, discrepancies of the conventional
model have not been reported on HEMTs yet. Koller et al. [96, 98, 104] sug-
gested that instead of charge transport in the valence band, charges propagate
through a defect band within the GaN bandgap as shown in Figure 6.18b. This
explains the same time constants for acceptor charging and discharging and the
same temperature dependence of trapping and leakage current.
Source Gate Drain Figure 6.20 Schematic cross

p–GaN section of GaN-on-Si HD-GIT
drain with the distinctive p-GaN
AlGaN pGaN drain in red.
GaN:uid 2DEG
GaN:C
Transition layer
Si substrate
Dynamic RDS,ON and current collapse are problematic in that the increased
RDS,ON decreases the efficiency, which can become a problem itself. However, the
loss of efficiency indicates an increase in thermal losses, which leads to increased
temperature and thus to even further increased RDS,ON . As a result, the device
might be robust in off- and on-state but not for switching operation.
This problem can be solved by slightly modifying the simple HEMT structure,
e.g. by the insertion of a p-doped region next to the drain metal as shown in
Figure 6.20 [105, 106]. In these devices called hybrid-drain-embedded gate injec-
tion transistors (HD-GITs), during high drain biases, holes are injected from the
p-doped drain into the buffer where they recombine with trapped electrons that
would cause the dynamic RDS,ON . Electrical characterization as in Figure 6.16
indeed reveals no significant dynamic RDS,ON after off-state stresses, even at biases
close to the breakdown voltage [102, 106, 107].
High drain bias results in high lateral and vertical electric fields. In combination
with high current, this forms hot electrons that can be captured, e.g. in the buffer,
and cause significant increase of the dynamic RDS,ON [102, 107]. As in HD-GITs
holes are introduced into the buffer during the switching event, hot carrier degra-
dation is reduced compared to structures without the hybrid drain.
In order to guarantee safe operation of GaN devices over their lifespan, the
form of switching has to be considered, and the devices must be tested in stress
sequences that are close to actual applications. Commonly known as dynamic
high-temperature operating life (D-HTOL) tests, transistors are stressed in
a boost converter configuration or by inductive load switching. Reference
[85] gives an overview of techniques used by different manufacturers. From
performing these measurements at different biases, currents, and temperatures,
a lifetime model such as the following [108] can be developed
LSW = A exp(−(𝛽V V + 𝛽C IDP )) (6.1)
with LSW as the switching lifetime, A as a constant, and I DP as the peak current;
the values for the voltage and current acceleration factors 𝛽 V and 𝛽 C , respectively,
can be found in [108]. Depending on the device design, the end of the lifetime can
be reached when the device fails physically, for example, by thermal destruction.
However, the lifetime can also be reached when RDS,ON no longer meets specifi-
cations, e.g. it can be defined as the time when RDS,ON increases by 50% [85]. On
basis of this lifetime model, a switching safe operating area (SSOA) can be defined
Figure 6.21 Schematic switching safe 60 (1) Rds,on (2) Id,sat VBD
operating area (SSOA) for a GaN HEMT with (3
the limiting parameter indicated in the four )s
40 wi
ID (A)
tch
different regimes. ing
SSOA
20
0
0 200 400 600
VDS (V)
10–2
T = 25°C
Vg = 9 V 10–3
Vg = 8 V
Gate current (A)
10–3 Ig,init (A)
Vgstress =
10–4
10–4 7V
8V
Vg = 7 V 9V
10–5 10–5
100 101 102 103 104 105 106 10–1 101 103 105
(a) Stressing time (s) (b) tBD (s)
2
10 years
Vg = 9 V 8V 7V 108
63
1
.2%
106
In(–In(1–F))
fai
0
1%
lu
tBD (s)
re
fai
104
lu
–1
re
–2 102
T = 25°C
–3 100
100 101 102 103 104 105 106 3 4 5 6 7 8 9
(c) tBD (s) (d) Vg (V)
Figure 6.22 (a) FBS tests performed at V G = 7, 8, and 9 V and T = 25 ∘ C and (b) correlation of
initial IG (IG,init ) at the beginning of stressing and tBD . (c) Weibull distribution of tBD as
determined from (a) together with linear fit to the data. (d) Extrapolation of safe forward V G
using linear fit to the log(𝜂) as a function of V G and scaled to 1% failure rate for a period of
10 years, yielding a maximum V G of 3.7 V. Source: Reprinted with permission of Ťapajna et al.
[90]. Copyright © 2016, IEEE.
as shown schematically in Figure 6.21. As in Si, the indicated phases (1), (2), and
(4) are limited by RDS,ON , saturation current, and breakdown voltage, respectively.
In Si, phase (3) is limited by thermal issues, while for GaN, it is limited by the
hard-switching. Therefore, exact values depend not only on the temperature but
also on the exact form of switching and the device design.
6.3.2 Gate Degradation in p-GaN Switching HEMTs

Besides excellent device metrics in terms of RDS,ON and V BD , p-(Al)GaN gate
technology also offers better V TH controllability and stability compared to other
designs for normally-off GaN switching devices [109]. In this device concept
schematically depicted in Figure 6.20, p-GaN layer inserted between the gate
metal and AlGaN/GaN heterostructure provides the normally-off operation.
The value of V th , reaching ∼1 to 2.5 V, depends on the p-GaN layer thickness,
Mg concentration, and resulting hole density. On the other hand, normally-off
p-GaN HEMTs are sensitive to gate forward bias stress (FBS), where progressive
I G degradation similar to TDDB mechanism has been reported by several groups
[89, 90, 110, 111]. In general, it seems that V G at which this irreversible gate
degradation occurs is related to the gate turn-on voltage. However, it is not
fully clear how the latter depends on the metal/p-GaN Schottky barrier height
(SBH), and in addition, some controversy exists between the dependence of
metal work function and the resulting SBH. We will therefore describe the
characteristics of FBS-induced gate degradation and then discuss possible
degradation mechanisms based on the current understanding of the p-GaN
HEMTs gate operation.
Degradation of normally-off p-GaN/AlGaN/GaN HEMTs upon FBS was ini-
tially reported by Ťapajna et al. [89, 90]. In these studies, devices with GaN-on-Si
heterostructures, 95-nm thick Mg-doped p-GaN layer (Mg concentration of
2 × 1019 cm−3 and hole density of 2 × 1017 cm−3 ) and Ni/Au gate metallization
were subjected to FBS at various gate voltages (V GS = 7, 8, and 9 V; V DS = 0 V)
and temperatures (25, 75, and 125 ∘ C) while time-dependent I G degradation
was monitored. As can be inferred from Figure 6.22a, I G transients show initial
continuous and recoverable change of I G related to trapping effects, followed
by strong increase in the I G noise and/or sudden I G steps, eventually leading to
breakdown of the gate. Such behavior has been attributed to percolative process
of defect generation in the gate structure similar to TDDB. To investigate its
dynamics, the elapsed time to sudden increase in I G (referred to as soft BD) was
recorded for each transient and defined as t BD . Initial I G (I G,init ) at the beginning
of the stressing was found to correlate with t BD (Figure 6.22b), which is a typical
feature of percolation mechanism. This is because the devices with higher I G,init
(being related to the defect density) are expected to show shorter t BD because
of faster overlapping of the defects forming the leakage path [12]. Alternatively,
higher I G,init gives higher level of charge injection into the gate structure that can
accelerate the degradation, providing that the defect generation rate depends
on the charge injection level. Based on this interpretation, t BD data obtained for
stressing at V G = 7, 8, and 9 V at T = 25 ∘ C were analyzed using the Weibull plot
shown in Figure 6.22c. Similar shape factors (0.7–0.8) extracted for stressing at
different V G indicate a similar degradation mechanism. Using the extracted 𝜂
values (63.2% failure) and assuming their linear dependence on V G (the so-called
V model), the maximum positive V G up to 3.7 V was estimated for safe operation
of the analyzed p-GaN HEMTs with W G = 0.25 mm, 1% failure rate for a useful
life of 10 years at 25 ∘ C (Figure 6.22d). For larger W G , maximum operating V G
needs to be lowered using gate periphery area scaling 𝜂 1 = 𝜂 2 (W G1 /W G2 )1/𝛽 [12].
From analysis of t BD measured at different T B , degradation process was found to

be weakly temperature dependent with an Ea of 0.1 eV [90].
Similar results in terms of FBS-induced degradation mechanism were observed
by Rossetto et al. [110]. In this study, p-GaN/AlGaN/GaN HEMTs with a 70 nm
Mg-doped p-GaN gate layer (W G = 0.1 mm and V th ∼ 2 V) also revealed TDDB
behavior upon FBS performed at V G up to 10 V. In contrast to devices analyzed in
[90] showing I G onset at V G = 1 V (in semilog plot), devices analyzed here showed
I G onset at V G = 8 V. The t BD data was described by Weibull distribution with
shape factor 𝛽 > 1, resulting in estimated 20-year operation at a maximum V G of
7.2–7.5 V. The degradation was found to be enhanced by temperature with an Ea
of 0.5 eV. Further, localized leakage spots were identified by EL microscopy in the
degraded devices, and the emission was attributed to yellow luminescence and/or
bremsstrahlung radiation based on their EL spectra as shown in Figure 6.23c.
Recently, time-dependent degradation of normally-off HEMTs with p-GaN gates
have also been observed by Masin et al. [111] who studied large power devices
(I D ∼ 60 A) with a W G of 130 mm. In contrast to previous studies, t BD data were
found to correlate with charge to breakdown rather than I G,init , suggesting that
TDDB occurs once a critical charge is injected into the gate. In addition, positive
dependence of t BD on temperature was evaluated with a negative Ea of −331 meV.
To understand the degradation mechanism of p-GaN/AlGaN/GaN HEMTs
upon FBS, it is important to discuss the character of the metal/p-GaN junction.
As already reported in Chapter 4, different metal gates have been proposed
for p-GaN/AlGaN/GaN HEMTs. Because the energy level of the GaN valence
band edge is deeper than the work function of typical metals, metal/p-GaN
junction dominantly forms a Schottky contact, with SBH theoretically given by
the difference between the metal work function and the p-GaN valence band
edge. The lower the metal work function is, the higher the SBH. As indicated
300 Fresh 100 6.0m 6.0

Mild degradation Spot 1 Spot 2
Spectral intensity (a.u.)
Severe degradation 10–1 Sum of the spots

Id 5.0m
Capacitance (nF/cm2)
250
10–2 IG = 10 mA/mm
5.5
4.0m
Gate voltage (V)
10–3
Id, Ig (A/mm)
200
Fresh 10–4 3.0m
150 10–5 5.0
10–6 2.0m
100 Ig 10–7 1.0m
4.5
50 10–8
0.0
10–9
0 10–10 –1.0m 4.0
–3 –2 –1 0 1 2 3 4 –4 –2 0 2 4 400 450 500 550 600 650 700
(a) Gate voltage (V) (b) VGS (V) (c) Wavelength (nm)
Figure 6.23 (a) C G –V G hysteresis measured on a typical gate diode before and after
FBS-induced degradation. (b) Transfer and IG –V GS characteristics of p-GaN/AlGaN/GaN HEMTs
after mild and severe gate degradation plotted in the semilog scale. Prolonged stressing
resulted in a hump formation in the subthreshold region. Source: Panels (a) and (b) are
reprinted with permission of Ťapajna et al. [89]. Copyright © 2015, AIP Publishing. (c) Spectra
of the light emitted from two leakage spots observed after FBS-induced gate degradation of
p-GaN/AlGaN/GaN HEMTs measured at IG = 10 mA/mm. Source: Panel (c) is reprinted with
permission of Rossetto et al. [110]. Copyright © 2016, Elsevier.
by the simulations [112] and experiments [113], increased SBH increases V th

and the gate turn-on voltage by preventing hole injection from the metal
into the p-GaN layer. For metals with higher work function forming low SBH
junctions (Ohmic-like contact), holes are effectively injected into the p-GaN
layer under positive bias and accumulate at the p-GaN/AlGaN interface. The
injected holes then pull down the bands and enhance electron injection from
the transistor channel over the AlGaN conduction band edge [114]. Difference
in the metal/p-GaN SBH can therefore explain slightly higher V th as well as
remarkably higher gate turn-on voltage in p-GaN HEMTs studied by Rossetto
et al. [110] (∼8 V) compared to those reported by Ťapajna et al. [90] (∼1.5 V).
Owing to observed correlation between I G,init (or charge to breakdown) and t BD ,
devices with higher initial I G degrade faster upon FBS as a result of higher level
of charge injection into the gate structure.
Figure 6.23a compares two-terminal C G –V G curves (V DS = 0) and
three-terminal I–V transfer characteristics (b) of p-GaN/AlGaN/GaN HEMTs
[89]. Fresh and degraded devices with different levels of duration upon FBS after
appearance of soft BD (labeled as mild and sever degradation) are compared.
For fresh devices, C G –V G shows a maximum value close to V th . The normalized
C G is lower than the barrier capacitance of the AlGaN layer, suggesting partial
depletion of the p-GaN layer. This is consistent with the recently proposed
circuit model of the gate structure, consisting an antiparallel connection of
Schottky metal/p-GaN diode and the AlGaN capacitor [115]. This means
that even for Ohmic-like gate metallization giving a low I G turn-on voltage, a
low-barrier height Schottky contact is likely to be formed at the metal/p-GaN
interface. Degraded devices showed an increased capacitance at positive V G ,
indicating shrinking of the Schottky depletion region and therefore increased
electric field in this layer. For more severe degraded devices, a hump appears
in the subthreshold I DS –V GS transfer characteristic and C G –V G curve, while it
is not present in the I G –V G characteristic (see Figure 6.23a,b). This hump can
therefore be attributed to increased source–drain I D leakage, possibly because of
the V th spatial inhomogeneities originated from larger gate leakage spots after
severe degradation.
Based on the data and discussion above, a two-step model describing
p-GaN/AlGaN/GaN gate degradation mechanism can be proposed. Upon FBS,
the meta/p-GaN Schottky barrier is reverse biased so that the barrier becomes
thinner and the hole current through the junction increases (I G is larger for
devices with higher metal gate work function [113]). Locally enhanced hole cur-
rent due to SBH fluctuations leads to generation of defects in the p-GaN Schottky
barrier and formation of localized leakage path via percolative process, which
is consistent with C G increasing observed for degraded gates (cf. Figure 6.23a).
Although electric field in the AlGaN layer decreases with increasing forward
bias [89, 113, 115], the injected holes accumulate at the p-GaN/AlGaN interface
and in turn enhances electron injection from the channel [114]. High carrier
injection across the AlGaN layer can then result in localized defect generation in
this layer and eventually forms the leakage path across the whole gate stack. In
other words, the gate degradation seems to be initiated in the depletion region
of the reverse-biased metal/p-GaN junction, while the failure is accomplished
by the local leakage path formation through the AlGaN layer. This can explain
some similarities between time-dependent p-GaN HEMTs gate degradation
subjected to FBS [90] and the reverse-biased Schottky gates of normally-on
HEMTs ascribed to AlGaN layer degradation [36]. Further, this model would also
be consistent with the EL spectra shown in Figure 6.23c. Although the carrier
injection (electrons and/or holes) through less-evolved leakage spots can give
rise to observed bremsstrahlung in the p-GaN layer, higher gate current through
the well-developed leakage spots can result in dominant yellow luminescence
originating from traps located in the GaN channel.
Other possible mechanisms for p-GaN gate degradation subjected to FBS have
also been proposed in the literature. Meneghini et al. [116] pointed out that the
observed percolative degradation of the p-GaN gates may originate from TDDB
of the SiN passivation layer close to the gate edge, which is exposed to a high elec-
tric field under FBS. Wu et al. [91] proposed avalanche breakdown mechanism
for the p-GaN gate degradation. In this work, the gate breakdown at forward bias
was found to increase with temperature. The positive temperature dependence
of the gate breakdown voltage has been associated with avalanche mechanism,
i.e. electron/hole multiplication in the depletion region of the reverse-biased
metal/p-GaN Schottky junction. This model was further supported by the
observed light generation at high positive V G before the catastrophic gate BD,
ascribed to recombination of the generated electron-hole pairs.
6.3.3 V th Instabilities in GaN MISHEMTs

An effective way to suppress relatively high gate leakage current of Schottky-
barrier-gated HEMTs is to apply the gate dielectric, forming a metal–insulator–
semiconductor (MIS)/metal–oxide–semiconductor (MOS) gate structure.
MISHEMTs with several orders of magnitude reduction in I G as compared to
Schottky-gated devices has been reported by many groups using various gate
dielectrics including Al2 O3 , Si3 N4 , SiO2 , AlN, HfO2 , and others (see [117]).
By suppressing I G in particular at forward bias, MISHEMTs with improved
gate controllability and stability under DC as well as RF operation have been
reported [62]. In addition, MIS gate structure offer several approaches to achieve
normally-off operation of GaN HEMTs desired for switching devices. Proposed
concepts are based on increased gate capacitance providing depletion of 2DEG
under the gate by employing partial [118] or full barrier recessing [119], growth
of ultrathin barrier layer [120], or introduction of sufficiently high negative
charge at the dielectric/barrier interface [121] or in the dielectric layer itself
[122]. On the other hand, MIS gate structures inevitably incorporate relatively
high density of traps [123–125]. Depending on the distribution of these traps,
MISHEMTs suffer from forward bias-induced V th drift, resulting in commonly
observed PBTI [88] or, in some cases, also NBTI [92, 126]. Although quite similar
PBTI behavior can be found in the literature, there is still ongoing discussion
on the origin of the observed effects. In this part, we will therefore summarize
models for trap distribution in the MISHEMT gate structures and their impact
on V th instabilities and review some recent studies on PBTI in GaN MISHEMTs.
Because V th instabilities are dominantly affected by dynamics of various

trapping states present in the gate stack, it is useful to summarize current under-
standing of the traps’ nature and their spatial distribution in GaN MISHEMT
structures. Dielectric/III-N semiconductor interface traps (IT), inevitably
forming at the interface of crystalline semiconductor and (amorphous, poly-
or nanocrystalline) dielectric, have been characterized by many researchers.
Commonly, IT were assumed to form a continuum of states, Dit (E), in the
bandgap with acceptor/donor traps in the upper/lower part of the bandgap
divided by the charge neutrality level (ECNL ) [123, 127]. Because of the absence
of an efficient IT passivation scheme similar to that used for SiO2 /Si interface,
relatively high Dit in the range of 1012 –1013 eV−1 cm−2 has been consistently
reported for GaN MISHEMT structures using various methods, including CV
[117], CV –T [128], and photoassisted C–V [124], C transients [92], DLTS [129],
and AC admittance (C–𝜔 and G–𝜔) techniques [130, 131]. It is however worth
mentioning that application of admittance techniques for Dit evaluation has
to be taken with caution because of gate bias and temperature dependence of
the barrier conductivity affecting the C and G frequency response [132]. Matys
et al. [133] proposed the presence of disorder-induced gap states (DIGSs) with
U-shaped energy distribution exponentially decay from the dielectric/III-N
interface toward the dielectric bulk. In addition, Bakeroot et al. [134] suggested
formation of “border” traps at the transition between thin nanocrystalline and
amorphous SiN layers. “Border” trapping was also identified by Zhu et al. [135]
in Al2 O3 /AlGaN/GaN MOSHEMTs analyzing the CV hysteresis with increased
maximum V G . Further, the presence of traps distributed in the dielectric BT,
affecting both NBTI and PBTI, has been earlier proposed by Ťapajna et al. [92].
Recently, the impact of dielectric bulk traps (BTs) on PBTI was clearly identified
in Al2 O3 /GaN MOS capacitors [136] as well as fully recessed MISHEMTs with
both SiN and Al2 O3 gate dielectric [137]. Although some variation in trap
density can be anticipated, it seems to be clear that both insulator/barrier IT and
BT with possible complex distribution in the dielectric should be considered in
the analysis of V th instabilities in GaN MISHEMTs.
Processes responsible for V th instabilities in MISHEMTs are schematically
depicted in Figure 6.24, showing band diagram across the gate structure in
thermal equilibrium (a) and under applied V G (b,c). In this simplified picture,
dielectric/barrier IT and dielectric BT are considered, while any trapping states
in the III-N epitaxial layer are neglected. This assumption has been shown
to be reasonably well satisfied owing to almost negligible V th instabilities in
Schottky-compared to MOS-gated HEMTs [92, 138]. Under positive V G applied,
free electrons spillover from the channel into the barrier and are captured
by empty IT and BT (Figure 6.24b). This shifts the V th into positive direction
regardless of the traps’ nature (i.e. donor or acceptor). Noteworthy, besides the
availability of free electrons in the barrier and IT properties, the capture rate
also depends on tunneling processes in the case of BT. In typical stress–recovery
measurement [88, 125] (discussed in Section 6.3.3.1), zero gate bias is then
applied and the recovery process is monitored via I D measurement, where
electron emission from IT and BT typically results in negative V th shift toward
8 8
Vg = 0 V G 4 Vg = 3 V Injection Vg = –5 V
S D
6 Dielectric 6 into BT
E–EFGaN (eV)
AIGaN
2 Hopping
4 GaN channel: uid
Spillover 4 Tunneling and
2 BT 0
therlam
Capture 2 emission
0 –2
–2 IT 0
–4
0 20 40 60 0 20 40 60 0 20 40 60
(a) (b) x (nm) (c)
Figure 6.24 (a) Schematic cross-sectional band diagrams of MISHEMT under the gate (inset of
(a)) showing an example of dielectric/barrier interface traps (IT) and bulk traps (BT) distributed
in the dielectric layer in thermal equilibrium (a) and under positive (b) and negative (c) V G
applied and possible capture and emission processes inducing different V th shifts (b–c).
equilibrium. In other techniques such as capacitance-transient technique, nega-

tive V G is applied after trap filling process (Figure 6.24c), so that field-enhanced
trap emission is monitored via capacitance measurement [126]. The monitoring
window of both techniques is limited by the reaction time of the measurement
systems. As depicted in Figure 6.24c, another mechanism for V th drift can
take place at negative V G . If the BT level coincides with the metal Fermi level,
electrons from the metal can be injected into BT, giving rise to a positive V th
shift, which counteracts the effect of electron emission from IT and BT localized
close to the interface [92]. Electrostatically, this effect has much weaker impact
on V th change than IT capture/emission. Yet coaction of processes depicted in
Figure 6.24c can result in negligible apparent CV hysteresis [126].
An important aspect directly affecting V th instabilities in GaN MISHEMTs is
the origin of a charge compensating the barrier (or any Ga- or N-face III-N sur-
face [139]) surface polarization charge. This situation is depicted in Figure 6.25a,
showing charge distribution with corresponding band diagram of a Ga-face
AlGaN/GaN MOS-HEMT structure under the gate. As dictated by the charge
neutrality, AlGaN surface polarization charge (PS ) must be compensated by
some charge of similar density and opposite polarity. Otherwise, uncompensated
PS would deplete the channel and result in an increase of V th with increasing gate
oxide thickness (red dashed line in Figure 6.25b), in clear contrast to commonly
observed behavior in GaN MISHEMTs shown by the solid blue line and typical
experimental data [126]. It is commonly accepted that PS is compensated by
surface donor (SD) states with a single level and density (N DS ) well above the
1013 cm−2 level, also providing free electrons into 2DEG, as originally proposed
by Ibbetson et al. [17]. Ibbetson’s model was later revisited by Higashiwaki et al.
[18]. The authors suggested distributed SD states at 1–2 eV below AlGaN CB with
a density of 4–6 × 1012 cm−2 . In a MISHEMT structure, these SD should act as
interface states. However, the lack of correlation between evaluated Dit and N DS
necessary for PS compensation [20, 134, 140] led to introduction of new models
for origin of the PS compensating charge. These models assume fixed charge
formed by the ionized donor states between CB bottom of the barrier and the
dielectric layer [141, 142], or the aforementioned “border” traps spatially located
in the dielectric with a high density (3.7 × 1013 cm−2 ) of energy states laying
Oxide AIGaN GaN

4 AI2O3/AI0.24Ga0.76N/GaN
Charge q(NDS+Nit) PQW
2 NDS = 0
PS qnS
6 0
Vth (V)
4
–2
E–EF (eV)
2 qΦint Mos
t of
0 –4 exp
erim
–2 ents
–6 qNDS = –Ps
–4
–6 –8
0 10 20 30 40 50 0 5 10 15 20
(a) Distance (nm) (b) tox (nm)
Figure 6.25 (a) Schematic charge distribution (top panel) and energy band diagram (bottom
panel) of a metal/Al2 O3 /AlGaN/GaN MISHEMT gate structure calculated using Poisson
equation. (b) Calculated dependence of V th on the dielectric thickness using analytical model
from [127]. Red dashed lines represent the situation for uncompensated surface barrier
polarization charge PS and blue solid lines represent the situation when PS is fully
compensated by “surface donors” (NDS = −PS /q). Source: (b) Adapted from Ťapajna et al. 2014
[126] and Ťapajna and Kuzmík 2012 [127].
2.4 eV below SiN CB edge [134]. Recently, Ber et al. [138] also concluded that Dit
is insufficient to account for PS compensation and proposed surface polarization
self-compensation mechanism. The authors speculated that displacement of
less rigid surface ions may be responsible for PS self-compensation [138]. The
notion of fixed charge compensation or polarization self-compensation seem to
be relevant in the light of recent data obtained from high-quality GaN-on-GaN
MOS structures, showing nearly zero flat-band voltage (i.e. compensated PS )
and Dit on the level of 1011 eV−1 cm−2 (evaluated under UV illumination) [143].
Apparent unavailability of sufficiently high Dit seems to exclude IT being
responsible for a full compensation of PS . This is also why we formally separated
“surface donor” charge from the interface-trapped charge (N it ) in the charge
diagram shown in Figure 6.25a, as the latter may greatly vary for different gate
dielectric growth technologies and barrier surface treatment.
For the sake of comparison between various MISHEMT technologies, it
is useful to define the net dielectric/barrier charge, Qint = PS + q(N DS + N it ),
where N DS represents the apparent fixed charge and N it represents the apparent
time-variable charge. As described in [20], in the case of dominant sheet charges
in the gate structure, Qint can be determined experimentally from nominally
identical MISHEMTs with varying dielectric thickness (t ox ). From the slope of
V th dependence on t ox , Qint = 𝜀ox (dV th /dt ox ) − PQW , where 𝜀ox is the dielectric
permittivity and PQW is the polarization charge in the QW. Qint does not
depend on the separation between interface fixed and variable charge (being
measurement technique dependent), and it directly determines barrier interface
potential Φint depicted in Figure 6.25a. The interface potential then defines
the trap occupancy level of IT and has critical impact on the evaluated PBTI.
Positive Qint (in the level of 1012 cm−2 ) results in a low Φint and availability of
shallower IT with fast response during recovery after FBS. In contrast, negative
250 250
4 Measurements
tox = 21 nm Simulations
Capacitance (nF/cm2)
w/o PDA 200 200
2 w/ PDA w/ PDA
w/o PDA
E–EF (eV)
150 w/o PDA 150

Vg,max=
0
0V
1V
w/ PDA 100 100
2V
–2 3V
4V
50 5V 50
–4 6V tox = 21 nm
7V
0 0
0 20 40 60 –10 –5 0 5 –10 –5 0 5
(a) x (nm) (b) Gate voltage (V) (c) Gate voltage (V)
Figure 6.26 (a). Band diagrams of MISHEMT structures calculated using Poisson equation and
assuming Qint /q of −1 × 1013 (w/o PDA) and 1 × 1012 cm−2 (w/ PDA). Experimental (b) and
simulated (c) CV hysteresis of MOSHEMT structures without and with PDA measured using
different maximum V G ranging from 0 to 7 V. Source: Reprinted with permission of Ťapajna
et al. [144] . Copyright © 2019, Elsevier.
Qint of similar density results in a higher Φint and thus availability of deeper IT
with much longer recovery response upon FBS.
Effect of Qint on the measured CV hysteresis was recently analyzed by our
group [144]. In this study, Ni/Al2 O3 /AlGaN/GaN MOSHEMTs with nominally
same heterostructures but different Qint /q of −1 × 1013 and 1 × 1012 cm−2
resulting from different oxide postdeposition annealing (PDA) were prepared.
Corresponding band diagrams for both structures (Figure 6.26a) suggest only
shallower empty IT for MOSHEMT with applied PDA, while much deeper IT are
available for the structure without PDA. Despite relatively high Dit (on the level
of 1012 eV−1 cm−2 in the upper part of the bandgap) determined experimentally
for both structures, MOSHEMT with PDA showed negligible CV hysteresis
with increasing maximum positive V G during CV measurement (Figure 6.26b).
This is because electrons captured by shallower IT at forward sweep are quickly
re-emitted during the backward sweep, resulting in similar IT population at
V G ∼ V th for forward and backward sweep. In contrast, MOSHEMTs without
PDA (high density of negative Qint ) showed enhanced CV hysteresis with
increasing maximum positive V G during CV measurement. Here, electrons
are captured by deep empty IT at positive V G , followed by slower re-emission
during the backward sweep, resulting in higher negative charge trapped by IT as
compared to thermal equilibrium. This behavior was also confirmed by transient
simulations and illustrates that MISHEMTs with a similar interface quality but
different Qint can show different CV hysteresis as well as PBTI characteristics.
It also indicates that normally-off GaN MISHEMTs featuring high density
of negative Qint (see [121]) can be expected to be more susceptible to PBTI
compared to normally-on counterparts.
6.3.3.1 Studies of PBTI in MISHEMTs

PBTI has been studied by many groups in GaN MISHEMTs with various
dielectrics. In the following, we will highlight only selected research works
aiming deeper understanding of the PBTI physics and the impact of dielectric
material as well as III-N heterostructure on the observed V th instabilities. We
will not discuss recent technological advances toward improvements in the V th
stability of GaN MISHEMTs.
An effective way to study V th instability in transistors is the measurement–
stress–measurement (MSM) technique employed for GaN MISHEMTs by Lag-
ger et al. [125]. The MSM sequence is schematically depicted in Figure 6.27a
[145]. Device under test is first stressed at forward bias V G,stress for a period t stress
(V DS = 0 V). Then, V G is stepped to zero (V DS = 0 V) and the device recovery
is monitored via transient response of V D measured at certain recovery time
intervals (t rec ) and small V D,meas . V th drift (ΔV th ) is evaluated from the transient
response of V D in respect to the virgin device characteristics. Measurements
are performed in sequences of constant V G,stress and logarithmically increased
t stress . MSM technique allows evaluation of V th drift transients for a wide range
of stressing and recovery times. Figure 6.27b,d exemplifies recovery V th drifts
after FBS at different voltages and times, suggesting the presence of trap states
with a wide range of capture and emission time constants. For the data inter-
pretation, the authors considered any defects in the active energy region (empty
traps between Φint and EF in thermal equilibrium) of the gate stack capable to
exchange the charge with 2DEG, without explicit discrimination between IT and
BT. The measured ΔV th are related to change of interface states concentration
(ΔN it ) via relation ΔN it = −C ox (ΔV th /q). Instead of calculating distribution of
Dit or density of BT, ΔN it is interpreted using the concept of capture emission
time (CET) map, as originally proposed for analysis of NBTI in Si MOSFETs
[146]. Figure 6.27c shows the CET map calculated from the recovery data exem-
plified in Figure 6.27b,d. Here, all trap states are described by their ΔV th (ΔN it )
per decade in three-dimensional space of corresponding capture and emission
time constants, for a given V G,stress and temperature [145].
In the case of first-order processes, where capture/emission of an electron
from/to CB of the semiconductor with corresponding lattice relaxation takes
place, the CET map should comprise only positive entries. However, the CET
map shown in Figure 6.27c also includes negative values that correspond
to the change in curvature of the recovery curves for increasing t stress (cf.
Figure 6.26b,d). This effect can also be inferred from the CET map for a fixed
recovery time of 1 ms, where ΔN it first increases with t stress and starts to
decrease for t stress > 100 ms. Such behavior was explained by the appearance
of the second-order defect kinetics, originating from multistate defects, where
charge transformation between different defect states occurs for longer t stress or
Coulomb effect of the charged defect states [145].
MSM technique in conjunction with CET map data analysis has been used to
study MISHEMTs with various dielectric materials [147]. Although different gate
materials gave different ΔV th for given t stress and t rec , the authors argued that also
gate electrostatics, affecting the availability of free electrons for capture, has to
be considered in the comparison. Indeed, it was shown that dependence of ΔN it
on the gate displacement charge (QD = C dielectric × V G ) for all dielectric materi-
als converged toward the same dependence, reaching its upper limit [147]. This
Threshold voltage drift ΔVTh (V)

0.5
VG,stress Stress bias 1 V, 2 V, …, 7V
Time 0.4
0V Stress
VG,meas 0.3
Recovery Stress time 100 ns
VD,meas Measure
Time 0.2
0V
(a) tstress trecovery 0.1
0.0
ΔVth per decade capture/emission (mV dec–2) 1μ 10 μ 100 μ 1 m 10 m 100 m 1 10 100
(b) Recovery time (s)
104
Threshold voltage drift ΔVTh (V)

103
Capture time constant (s)
102 20 3.5 Stress bias 1 V, 2 V, …, 7V

101 3.0
10 0 10 Stress time 100 ns
2.5
10–1 0
10–2 2.0
10–3 –10 1.5
10–4
10–5 1.0
–20
10–6 0.5
10–7 –30
0.0
10–6 10–5 10–4 10–3 10–2 10–1 100 101 102 103 104 1 μ 10 μ 100 μ 1 m 10 m100 m 1 10 100 1000
Emission time constant (s) Recovery time (s)
(c) (d)
Figure 6.27 (a) Pulse pattern used for stress–recovery cycle of MSM technique. To measure
the transient response of V D , the biases are pulsed to V G = V G,meas and V D = V D,meas . (c, d)
Recovery transients for stress times of 100 ns (c) and 100 s (d) using varying V G,stress from 1 to
7 V. (b) Capture emission time (CET) map extracted from the recovery transients exemplified in
(c–d). Source: Reprinted with permission of Lagger et al. [145]. Copyright © 2014, IEEE.
observation was then used for definition of practical MISHEMT lifetime require-
ment, defined by two common criteria for normally-off devices: (i) a sufficiently
low ΔV th assuring maximum specified RDS,ON and the minimum specified I D at
the end-of-life must be fulfilled at operating voltage and (ii) the stability of the gate
dielectric must be assured at maximum V G at end of life [88]. Using the experi-
mental values of critical electric field, maximum ΔN it was estimated to be in the
range of 4–8 × 1012 cm−2 for a typical sheet channel density of 0.5–1 × 1012 cm−2 .
Gate dielectric growth technology has also strong impact on the MISHEMTs
stability. For example, Meneghesso et al. [148] reported comprehensive reliabil-
ity investigation of partially recessed barrier MISHEMTs with SiN gate dielectric
grown by rapid thermal CVD (RTCVD) and plasma-enhanced atomic layer depo-
sition (PEALD). As deduced from MSM measurement, MISHEMTs with PEALD
dielectric showed much lower V th drift with resulting ΔN it of ∼3.5 × 1011 cm−2
(t stress = 1000 seconds) as compared to devices with RTCVD dielectric revealing
ΔN it of ∼2 × 1012 cm−2 . Interestingly, the recovery times for both devices well
exceeded 1000 seconds for V G,stress = 2.5 V. Similar to results reported by Bisi et al.
[136], V th drift was found to correlate with forward I G , indicating trapping pro-
cesses dominated by dielectric BT. In addition, the gate robustness under forward
bias was examined using step stress and TDDB measurements [148]. Improved
gate robustness of PEALD-grown SiN gate dielectric as compared to RTCVD SiN
has been attributed to lower density of BT in this dielectric because of lower
probability of the percolation leakage path formation. These results underline
the importance of the gate dielectric growth technology, as nonoptimal growth
G
S D
AI2O3
n+ GaN 6 nm 0.6 0.6
InGaN 3 nm
AlGaN 5 nm
GaN
0.5 0.5
ΔVTH (V)
ΔVTH (V)
Sapphire
(a)
6
AI2O3 InGaN AlGaN GaN
0.4 0.4
4
Energy (eV)
2 Stress time
hv
hv
0 0.1 s 0.5 s
0.3 1s 5s Recovery time 50 s 0.3
–2 10 s 50 s
–4 1
0.01 0.1 10 100 0.1 1 10 100
–6 tSTRESS (s) tREC (s)
0 10 20 30
(b) Distance (nm) (c)
Figure 6.28 (a) Schematic cross section of Ni/Al2 O3 /InGaN/AlGaN/GaN MOSHEMT and (b)
corresponding band diagram under the gate area with the proposed model for V th shift upon
stressing/recovery measurement. (c) Successive MSM stress (left)–recovery (right) cycles,
starting with the shortest tstress (0.1 seconds) and continuing with the increased tstress . Dashed
lines depict ΔV th at the beginning of each measurement cycle. Source: Adapted from
Pohorelec et al. 2019 [149].
conditions leading to formation of higher density of BT can result in inferior

device reliability in terms of V th stability and gate robustness.
So far, we have only discussed an expected PBTI behavior, i.e. positive V th
shift under FBS followed by negative V th shift during recovery. However, also
opposite V th drift can take place in normally-off devices with special design of
the gate structure. In our recent study [149], PBTI in Al2 O3 /InGaN/AlGaN/GaN
MOSHEMTs (Figure 6.28a) was investigated using MSM experiments. As
depicted in the band diagram shown in Figure 6.28b, a high negative polarization
charge at the InGaN/AlGaN interface raises the bands, which leads to formation
of two-dimensional hole gas (2DHG) at this interface and provides positive V th
[121]. Typical MSM measurements are shown in Figure 6.28c for both stress (left,
t stress = 100 ms to 50 s) and recovery (right, t rec = 50 s) period. Interestingly, our
devices show negative V th drift during stress period and positive V th drift during
recovery. Yet the final V th shift ends up at higher value in respect to the beginning
of the stressing period. This peculiar behavior can be explained by a model
assuming 2DHG present in the InGaN layer in thermal equilibrium. When posi-
tive V G,stress is applied to the gate, electrons populating 2DEG at the AlGaN/GaN
interface are injected into optically active InGaN layer. These electrons start to
recombine with holes in the InGaN, emitting photons with an energy of around
3 eV, which is the bandgap energy of the InGaN. Photons are absorbed by the
Al2 O3 /InGaN interface states and deep levels in the AlGaN layer. Electrons
released from these states either tunnel through the oxide toward the gate (in the
case of interface states) or fall back into the 2DEG channel (in the case of AlGaN
states). The emptied states build up positive charge, which shifts V th to negative
values. During recovery, the structure tends to return toward thermal equilib-
rium state; however, some holes are depleted from the 2DHG. Therefore, result-
ing V th is shifted to more positive values as compared to prestress condition.
References 241
6.4 Summary
In this chapter, a comprehensive review of reliability issues and degradation
modes in power GaN HEMTs is presented. Extensive research of AlGaN/GaN
and InAlN/GaN HEMTs for microwave applications has led to deeper under-
standing of main failure modes related to gate-edge physical degradation and
hot-electron- and hot-phonon-related degradation. Design approaches includ-
ing barrier GaN capping, in situ SiN passivation, and electric field mitigation
have been developed to provide reliable operation of these devices. However,
reliability tests commonly applied for GaAs-based technology need to be
modified for the degradation modes specific for GaN HEMTs. Yet the dominant
long-term thermally accelerated mechanism needs to be identified. Special
attention must be paid on a detailed thermal characterization of tested GaN
devices to validate HTOL test data.
A major obstacle for GaN power switching devices entering mass market
originates from trapping effects in unavoidable point defects. By employing
sophisticated design of epitaxial structures and device concepts, these issues can
be mitigated. Because of the relatively short history, not all failure modes are
known in detail. Only elaborate application of relevant stress testing (D-HTOL)
as well as enhanced safety margins can therefore ensure reliable operation over
the entire lifetime. Although gate robustness under forward bias in p-GaN
HEMTs seems not to be a major reliability concern, further understanding of
this failure mode is clearly necessary. For GaN MISHEMTs, PBTI currently
represents a limiting reliability issue, and more dramatic technology improve-
ments are required for stable operation of these devices. Nevertheless, ongoing
research effort is devoted toward development of technological approaches for
high-quality dielectric/III-N interface processing as well as deeper understanding
of the PBTI physics.
Acknowledgments
One of the authors (MT) would like to thank to Dr. F. Gucmann for helpful dis-
cussions on preparation of Section 6.1.
References
1 Rosina, M. (2018). GaN and SiC power device: market overview. Semicon
Europa, Munich, Germany (13–16 November 2018).
2 Pomeroy, J.W., Uren, M.J., Lambert, B., and Kuball, M. (2015). Operating
channel temperature in GaN HEMTs: DC versus RF accelerated life testing.
Microelectron. Reliab. 55 (12): 2505–2510.
3 JESD22-A108 (2017). Temperature, bias, and operating life. Joint Electron
Device Engineering Council (JEDEC).
4 JEP118A (2018). Guidelines for GaAs MMIC PHEMT/MESFET and HBT

reliability accelerated life testing. Joint Electron Device Engineering Council
(JEDEC).
5 McPherson, J.W. (2018). Brief history of JEDEC qualification standards for
silicon technology and their applicability(?) to WBG semiconductors. 2018
IEEE International Reliability Physics Symposium (IRPS), Burlingame, CA,
USA (11–15 March 2018), pp. 3B.1-1–3B.1-8.
6 McDonald, T. (2018). Reliability and qualification of CoolGaN TM. Infineon
Technologies AG, White Paper.
7 Marcon, D., Kang, X., Viaene, J. et al. (2011). GaN-based HEMTs tested
under high temperature storage test. Microelectron. Reliab. 51 (9–11):
1717–1720.
8 Dammann, M., Baeumler, M., Brückner, P. et al. (2015). Degradation of
0.25 μm GaN HEMTs under high temperature stress test. Microelectron.
Reliab. 55 (9–10): 1667–1671.
9 Kuball, M. and Pomeroy, J.W. (2016). A review of Raman thermography for
electronic and opto-electronic device measurement with sub-micron spatial
and nanosecond temporal resolution. IEEE Trans. Device Mater. Reliab. 16
(4): 667–684.
10 Marcon, D., Kauerauf, T., Medjdoub, F. et al. (2010). A comprehensive reli-
ability investigation of the voltage-, temperature- and device geometry-
dependence of the gate degradation on state-of-the-art GaN-on-Si HEMTs.
2010 International Electron Devices Meeting, San Francisco, CA, USA (6–8
December 2010), pp. 20.3.1–20.3.4.
11 Meneghini, M., Stocco, A., Bertin, M. et al. (2012). Time-dependent degra-
dation of AlGaN/GaN high electron mobility transistors under reverse bias.
Appl. Phys. Lett. 100 (3): 033505-1–033505-3.
12 Degraeve, R., Kaczer, B., and Groeseneken, G. (1999). Degradation and
breakdown in thin oxide layers: mechanisms, models and reliability predic-
tion. Microelectron. Reliab. 39 (10): 1445–1460.
13 Meneghesso, G., Verzellesi, G., Danesin, F. et al. (2008). Reliability of GaN
high-electron-mobility transistors: state of the art and perspectives. IEEE
Trans. Device Mater. Reliab. 8 (2): 332–343.
14 del Alamo, J.A. and Joh, J. (2009). GaN HEMT reliability. Microelectron.
Reliab. 49 (9–11): 1200–1206.
15 Trew, R., Green, D., and Shealy, J. (2009). AlGaN/GaN HFET reliability.
IEEE Microwave Mag. 10 (4): 116–127.
16 Zanoni, E., Meneghini, M., Chini, A. et al. (2013). AlGaN/GaN-based
HEMTs failure physics and reliability: mechanisms affecting gate edge and
Schottky junction. IEEE Trans. Electron Devices 60 (10): 3119–3131.
transistors. Appl. Phys. Lett. 77 (2): 250–252.
18 Higashiwaki, M., Chowdhury, S., Miao, M.-S. et al. (2010). Distribution of
donor states on etched surface of AlGaN/GaN heterostructures. J. Appl.
Phys. 108 (6): 063719.
References 243
19 Hua, M., Wei, J., Tang, G. et al. (2017). Normally-off LPCVD-SiNx /GaN
MIS-FET with crystalline oxidation interlayer. IEEE Electron Device Lett. 38
(7): 929–932.
20 Ťapajna, M., Stoklas, R., Gregušová, D. et al. (2017). Investigation of ‘surface
donors’ in Al2 O3 /AlGaN/GaN metal-oxide-semiconductor heterostructures:
correlation of electrical, structural, and chemical properties. Appl. Surf. Sci.
426 (12): 656–661.
21 Ťapajna, M., Simms, R.J.T., Pei, Y. et al. (2010). Integrated optical and elec-
trical analysis: identifying location and properties of traps in AlGaN/GaN
HEMTs during electrical stress. IEEE Electron Device Lett. 31 (7): 662–664.
23 Faramehr, S., Kalna, K., and Igić, P. (2014). Drift-diffusion and hydrodynamic
modelling of current collapse in GaN HEMTs for RF power application.
Semicond. Sci. Technol. 29 (2): 025007-1–025007-11.
24 Meneghesso, G., Meneghini, M., Bisi, D. et al. (2013). Trapping phenomena
in AlGaN/GaN HEMTs: a study based on pulsed and transient measure-
ments. Semicond. Sci. Technol. 28 (7): 074021-1–074021-8.
25 Ťapajna, M., Jimenez, J.L., and Kuball, M. (2012). On the discrimination
between bulk and surface traps in AlGaN/GaN HEMTs from trapping
characteristics. Phys. Status Solidi A 209 (2): 386–389.
26 Faqir, M., Verzellesi, G., Fantini, F. et al. (2007). Characterization and analy-
sis of trap-related effects in AlGaN–GaN HEMTs. Microelectron. Reliab. 47
(9–11): 1639–1642.
27 Joh, J. and del Alamo, J.A. (2008). Impact of electrical degradation on trap-
ping characteristics of GaN high electron mobility transistors. 2008 IEEE
International Electron Devices Meeting, San Francisco, CA, USA (15–17
December 2008), pp. 1–4.
28 Bisi, D., Meneghini, M., de Santi, C. et al. (2013). Deep-level characterization
in GaN HEMTs-Part I: advantages and limitations of drain current transient
measurements. IEEE Trans. Electron Devices 60 (10): 3166–3175.
29 Look, D.C., Fang, Z.-Q., and Claflin, B. (2005). Identification of donors,
acceptors, and traps in bulk-like HVPE GaN. J. Cryst. Growth 281 (1):
143–150.
30 Joh, J. and del Alamo, J.A. (2006). Mechanisms for electrical degradation of
GaN high-electron mobility transistors. 2006 International Electron Devices
Meeting, San Francisco, CA, USA (11–13 December 2006), pp. 1–4.
31 Joh, J., Xia, L., and del Alamo, J.A. (2007). Gate current degradation mech-
anisms of GaN high electron mobility transistors. 2007 IEEE International
Electron Devices Meeting, Washington, DC, USA, pp. 385–388.
32 Chowdhury, U., Jimenez, J.L., Lee, C. et al. (2008). TEM observation of
crack- and pit-shaped defects in electrically degraded GaN HEMTs. IEEE
33 Bajo, M.M., Hodges, C., Uren, M.J., and Kuball, M. (2012). On the link
between electroluminescence, gate current leakage, and surface defects in
AlGaN/GaN high electron mobility transistors upon off state stress. Appl.

Phys. Lett. 101 (3): 033508-1–033508-4.
34 Joh, J. and del Alamo, J.A. (2008). Critical voltage for electrical degradation
of GaN high-electron mobility transistors. IEEE Electron Device Lett. 29 (4):
287–289.
35 Chini, A., Esposto, M., Meneghesso, G., and Zanoni, E. (2008). Evaluation
of GaN HEMT degradation by means of pulsed I-V , leakage and DLTS
measurements. Electron. Lett. 45 (8): 426–427.
36 Marcon, D., Viaene, J., Favia, P. et al. (2012). Reliability of AlGaN/GaN
HEMTs: permanent leakage current increase and output current drop.
Microelectron. Reliab. 52 (9–10): 2188–2193.
37 Zanoni, E., Danesin, F., Meneghini, M. et al. (2009). Localized damage in
AlGaN/GaN HEMTs induced by reverse-bias testing. IEEE Electron Device
Lett. 30 (5): 427–429.
38 Batten, T., Pomeroy, J.W., Uren, M.J. et al. (2009). Simultaneous measure-
ment of temperature and thermal stress in AlGaN/GaN high electron
mobility transistors using Raman scattering spectroscopy. J. Appl. Phys.
106 (9): 094509-1–094509-4.
39 Ancona, M.G., Binari, S.C., and Meyer, D. (2011). Fully-coupled electrome-
chanical analysis of stress-related failure in GaN HEMTs. Phys. Status Solidi
C 8 (7–8): 2276–2278.
40 Li, L., Joh, J., del Alamo, J.A., and Thompson, C.V. (2012). Spatial
distribution of structural degradation under high-power stress in
AlGaN/GaN high electron mobility transistors. Appl. Phys. Lett. 100 (17):
172109-1–172109-3.
41 Christiansen, B.D., Coutu, R.A., Heller, E.R. et al. (2011). Reliability test-
ing of AlGaN/GaN HEMTs under multiple stressors. 2011 International
Reliability Physics Symposium, Monterey, CA, USA,break pp. CD.2.1–CD.2.5.
42 Ťapajna, M., Mishra, U.K., and Kuball, M. (2010). Importance of impu-
rity diffusion for early stage degradation in AlGaN/GaN high elec-
tron mobility transistors upon electrical stress. Appl. Phys. Lett. 97 (2):
023503-1–023503-3.
43 Makaram, P., Joh, J., del Alamo, J.A. et al. (2010). Evolution of structural
defects associated with electrical degradation in AlGaN/GaN high electron
mobility transistors. Appl. Phys. Lett. 96 (23): 233509-1–233509-3.
44 Gao, F., Lu, B., Li, L. et al. (2011). Role of oxygen in the OFF-state degrada-
tion of AlGaN/GaN high electron mobility transistors. Appl. Phys. Lett. 99
(22): 223506-1–223506-3.
45 Brazzini, T., Sun, H., Sarti, F. et al. (2016). Mechanism of hot electron elec-
troluminescence in GaN-based transistors. J. Phys. D: Appl. Phys. 49 (43):
435101-1–435101-6.
46 Hui, K., Hu, C., George, P., and Ko, P.K. (1990). Impact ionization in GaAs
MESFETs. IEEE Electron Device Lett. 11 (3): 113–115.
47 Simms, R.J.T., Pomeroy, J.W., Uren, M.J. et al. (2010). Electric field distri-
bution in AlGaN/GaN high electron mobility transistors investigated by
electroluminescence. Appl. Phys. Lett. 97 (3): 033502-1–033502-3.
References 245
48 Meneghini, M., Stocco, A., Silvestri, R. et al. (2012). Degradation of

AlGaN/GaN high electron mobility transistors related to hot electrons.
Appl. Phys. Lett. 100 (16): 12–15.
49 Shigekawa, N., Shiojima, K., and Suemitsu, T. (2001). Electroluminescence
characterization of AlGaN/GaN high electron mobility transistors. Appl.
Phys. Lett. 79 (8): 1196–1198.
50 Nakao, T., Ohno, Y., Akita, M. et al. (2002). Electroluminescence in
AlGaN/GaN high electron mobility transistors under high bias voltage.
Jpn. J. Appl. Phys. 41 (4A): 1990–1991.
51 Gütle, F., Polyakov, V.M., Baeumler, M. et al. (2012). Radiative inter-valley
transitions as a dominant emission mechanism in AlGaN/GaN high electron
mobility transistors. Semicond. Sci. Technol. 27 (12): 125003-1–125003-7.
52 Van de Walle, C.G. (1997). Interactions of hydrogen with native defects in
GaN. Phys. Rev. B 56 (15–16): R10020–R10023.
53 Wright, A.F. (2001). Interaction of hydrogen with gallium vacancies in
wurtzite GaN. J. Appl. Phys. 90 (3): 1164–1169.
54 Puzyrev, Y.S., Roy, T., Beck, M. et al. (2011). Dehydrogenation of defects and
hot-electron degradation in GaN high-electron-mobility transistors. J. Appl.
Phys. 109 (3): 034501-1–034501-8.
55 Kuzmik, J. (2001). Power electronics on InAlN/(In)GaN: prospect for a
record performance. IEEE Electron Device Lett. 22 (11): 510–512.
56 Matulionis, A., Liberis, J., Matulionienė, I. et al. (2003). Hot-phonon temper-
ature and lifetime in a biased Alx Ga1−x N/GaN channel estimated from noise
analysis. Phys. Rev. B 68 (3): 035338-1–035338-7.
57 Kuzmik, J., Vitanov, S., Dua, C. et al. (2012). Buffer-related degradation
aspects of single and double-heterostructure quantum well InAlN/GaN
high-electron-mobility transistors. Jpn. J. Appl. Phys. 51 (5R):
054102-1–054102-5.
58 Ťapajna, M., Killat, N., Palankovski, V. et al. (2014). Hot-electron-related
degradation in InAlN/GaN high-electron-mobility transistors. IEEE Trans.
59 Kayis, C., Ferreyra, R.A., Wu, M. et al. (2011). Degradation in
InAlN/AlN/GaN heterostructure field-effect transistors as monitored by
low-frequency noise measurements: hot phonon effects. Appl. Phys. Lett. 99
(6): 063505-1–063505-3.
60 Zhu, C.Y., Wu, M., Kayis, C. et al. (2012). Degradation and phase noise of
InAlN/AlN/GaN heterojunction field effect transistors: implications for hot
electron/phonon effects. Appl. Phys. Lett. 101 (10): 103502-1–103502-4.
61 Kuzmik, J., Pozzovivo, G., Ostermaier, C. et al. (2009). Analysis of degra-
dation mechanisms in lattice-matched InAlN/GaN high-electron-mobility
transistors. J. Appl. Phys. 106 (12): 124503-1–124503-7.
62 Downey, B.P., Meyer, D.J., Roussos, J.A. et al. (2015). Effect of gate insu-
lator thickness on RF power gain degradation of vertically scaled GaN
MIS-HEMTs at 40 GHz. IEEE Trans. Devise Mater. Reliab. 15 (4): 474–477.
63 Matulionis, A., Liberis, J., Matulionienė, I. et al. (2009). Plasmon-enhanced
heat dissipation in GaN based two-dimensional channels. Appl. Phys. Lett.
95 (19): 192102-1–192102-3.
64 Dyson, A. and Ridley, B.K. (2008). Phonon-plasmon coupled-mode lifetime

in semiconductors. J. Appl. Phys. 103 (11): 114507-1–114507-4.
65 Leach, J.H., Zhu, C.Y., Wu, M. et al. (2010). Effect of hot phonon lifetime on
electron velocity in InAlN/AlN/GaN heterostructure field effect transistors
on bulk GaN substrates. Appl. Phys. Lett. 96 (13): 133505-1–133505-3.
66 Leach, J.H., Zhu, C.Y., Wu, M. et al. (2009). Degradation in
InAlN/GaN-based heterostructure field effect transistors: role of hot
phonons. Appl. Phys. Lett. 95(22): 223504 (3pp).
67 Nazari, M., Hancock, B.L., Piner, E.L., and Holtz, M.W. (2015). Self-heating
profile in an AlGaN/GaN heterojunction field-effect transistor studied by
ultraviolet and visible micro-Raman spectroscopy. IEEE Trans. Electron
Devices 62 (5): 1467–1472.
68 Michaud, J., Del Vecchio, P., Bechou, L. et al. (2015). Precise facet tempera-
ture distribution of high-power laser diodes: unpumped window effect. IEEE
69 Kuzmík, J., Bychikhin, S., Neuburger, M. et al. (2005). Transient thermal
characterization of AlGaN/GaN HEMTs grown on silicon. IEEE Trans.
70 Batten, T., Manoi, A., Uren, M.J. et al. (2010). Temperature analysis of
AlGaN/GaN based devices using photoluminescence spectroscopy: chal-
lenges and comparison to Raman thermography. J. Appl. Phys. 107 (7):
074502-1–074502-5.
71 McAlister, S.P., Bardwell, J.A., Haffouz, S., and Tang, H. (2006). Self-heating
and the temperature dependence of the dc characteristics of GaN het-
erostructure field effect transistors. J. Vac. Sci. Technol., A 24 (3): 624–628.
72 Kuzmík, J., Javora, P., Alam, A. et al. (2002). Determination of channel tem-
perature in AlGaN/GaN HEMTs grown on sapphire and silicon substrates
using dc characterization method. IEEE Trans. Electron Devices 48 (8):
1496–1498.
73 Joh, J., del Alamo, J.A., Chowdhury, U. et al. (2009). Measurement of chan-
nel temperature in GaN high-electron mobility transistors. IEEE Trans.
74 Florovič, M., Szobolovszký, R., Kováč, J. Jr., et al. (2019). Rigorous channel
temperature analysis verified for InAlN/AlN/GaN HEMT. Semicond. Sci.
Technol. 34 (6): 065021-1–065021-1.
75 Pavlidis, G., Pavlidis, S., Heller, E.R. et al. (2017). Characterization of
AlGaN/GaN HEMTs using gate resistance thermometry. IEEE Trans. Elec-
tron Devices 64 (1): 78–83.
76 Mion, C., Muth, J.F., Preble, E.A., and Hanser, D. (2006). Accurate depen-
dence of gallium nitride thermal conductivity on dislocation density. Appl.
Phys. Lett. 89 (9): 092123-1–092123-3.
77 Pearton, S.J. (ed.) (2000). GaN and Related Materials II, (692 p). CRC Press.
ISBN: 9789056996864.
78 Killat, N., Montes, M., Pomeroy, J.W. et al. (2012). Thermal properties of
AlGaN/GaN HFETs on bulk GaN substrates. IEEE Electron Device Lett. 33
(3): 366–368.
References 247
79 Manoi, A., Pomeroy, J.W., Killat, N., and Kuball, M. (2010). Benchmarking
of thermal boundary resistance in AlGaN/GaN HEMTs on SiC substrates:
implications of the nucleation layer microstructure. IEEE Electron Device
Lett. 31 (12): 1395–1397.
80 Kuzmík, J., Ťapajna, M., Válik, L. et al. (2014). Self-heating in GaN tran-
sistors designed for high-power operation. IEEE Trans. Electron Devices 61
(10): 3429–3434.
81 Kuball, M., Pomeroy, J.W., Gucmann, F., and Oner, B. (2019). Thermal anal-
ysis of semiconductor devices and materials - why should I not trust a
thermal simulation?. BCICTS 2019 IEEE BiCMOS and Compound Semi-
conductor Integrated Circuits and Technology Symposium, Nashville, USA,
2019.
82 Liu, W. and Balandin, A.A. (2005). Thermal conduction in Alx Ga1−x N alloys
and thin films. J. Appl. Phys. 97 (7): 073710-1–073710-6.
83 Gucmann, F., Pomeroy, J.W., Sarua, A., and Kuball, M. (2019). Channel tem-
perature determination for GaN HEMT lifetime testing – impact of package
and device layout. 2019 International Conference on Compound Semiconduc-
tor Manufacturing Technology (CS MANTECH), Minneapolis, USA, 2019.
84 Meneghini, M., Rossetto, I., Hurkx, F. et al. (2015). Extensive investigation of
time-dependent breakdown of GaN-HEMTs submitted to OFF-state stress.
IEEE Trans. Electron Devices 62 (8): 2549–2554.
85 Lin, M. (2019). New circuit topology for system-level reliability of GaN.
2019 31st International Symposium on Power Semiconductor Devices and ICs
(ISPSD), Shanghai, China, 2019, pp. 299–302.
86 Borga, M., Meneghini, M., Rossetto, I. et al. (2017). Evidence of
time-dependent vertical breakdown in GaN-on-Si HEMTs. IEEE Trans.
87 Zagni, N., Puglisi, F., Pavan, P. et al. (2019). Insights into the off-state break-
down mechanisms in power GaN HEMTs. Microelectron. Reliab.: 100–101.
88 Ostermaier, C., Lagger, P., Reiner, M., and Pogany, D. (2018). Review of
bias-temperature instabilities at the III-N/dielectric interface. Microelectron.
Reliab. 82: 62–83.
89 Ťapajna, M., Hilt, O., Bahat-Treidel, E. et al. (2015). Investigation of
gate-diode degradation in normally-off p-GaN/AlGaN/GaN high-electron-
mobility transistors. Appl. Phys. Lett. 107 (19): 193506-1–193506-4.
90 Ťapajna, M., Hilt, O., Würfl, J., and Kuzmík, J. (2016). Gate reliability inves-
tigation in normally-off p-type-GaN cap/AlGaN/GaN HEMTs under forward
bias stress. IEEE Electron Device Lett. 37 (4): 385–388.
91 Wu, T.-L., Marcon, D., You, S. et al. (2015). Forward bias gate breakdown
mechanism in enhancement-mode p-GaN gate AlGaN/GaN high-electron
mobility transistors. IEEE Electron Device Lett. 36 (10): 1001–1003.
92 Ťapajna, M., Jurkovič, M., Válik, L. et al. (2013). Bulk and interface trap-
ping in the gate dielectric of GaN based metal-oxide semiconductor
high-electron-mobility transistors. Appl. Phys. Lett. 102 (24): 243509.
93 Hashizume, T., Ootomo, S., Inagaki, T., and Hasegawa, H. (2003). Surface
passivation of GaN and GaN/AlGaN heterostructures by dielectric films
and its application to insulated-gate heterostructure transistors. J. Vac. Sci.

Technol., B 21 (4): 1828–1838.
94 Rossetto, I., Meneghini, M., Pandey, S. et al. (2017). Field-related failure of
GaN-on-Si HEMTs: dependence on device geometry and passivation. IEEE
95 Parish, G., Keller, S., Denbaars, S.P., and Mishra, U.K. (2000). SIMS investi-
gations into the effect of growth conditions on residual impurity and silicon
incorporation in GaN and Alx Ga1-x N. J. Electron. Mater. 29 (1): 15–20.
96 Koller, C., Pobegen, G., Ostermaier, C. et al. (2017). The interplay of block-
ing properties with charge and potential redistribution in thin carbon-doped
GaN on n-doped GaN layers. Appl. Phys. Lett. 111 (3): 032106-1–032106-5.
97 Seager, C.H., Wright, A.F., Yu, J., and Götz, W. (2002). Role of carbon in
GaN. J. Appl. Phys. 92 (11): 6553–6560.
98 Koller, C., Pobegen, G., Ostermaier, C., and Pogany, D. (2017). Evidence of
defect band in carbon-doped GaN controlling leakage current and trapping
dynamics. 2017 IEEE International Electron Devices Meeting (IEDM), San
Francisco, CA, USA, 2017, pp. 33.4.1–33.4.4.
99 Rackauskas, B., Uren, M.J., Stoffels, S. et al. (2018). Determination of the
self-compensation ratio of carbon in AlGaN for HEMTs. IEEE Trans. Elec-
tron Devices 65 (5): 1838–1842.
100 Uren, M.J., Cäsar, M., Gajda, M.A., and Kuball, M. (2014). Buffer transport
mechanisms in intentionally carbon doped GaN heterojunction field effect
transistors. Appl. Phys. Lett. 104 (26): 263505.
(GIT): a normally-off AlGaN/GaN power transistor using conductivity
modulation. IEEE Trans. Electron Devices 54 (12): 3393–3399.
102 Padovan, V., Koller, C., Pobegen, G. et al. (2019). Stress and recovery
dynamics of drain current in GaN HD-GITs submitted to DC semi-ON
stress. Microelectron. Reliab. 100–101: 113482.
103 Uren, M., Karboyan, S., Chatterjee, I. et al. (2017). “Leaky dielectric” model
for the suppression of dynamic RON in carbon-doped AlGaN/GaN HEMTs.
IEEE Trans. Electron Devices 64 (7): 2826–2834.
104 Koller, C., Pobegen, G., Ostermaier, C., and Pogany, D. (2018). Effect of
carbon doping on charging/discharging dynamics and leakage behavior of
carbon-doped GaN. IEEE Trans. Electron Devices 65 (12): 5314–5321.
operations up to 850 V by GaN-GIT utilizing hole injection from drain.
IEEE 27th International Symposium on Power Semiconductor Devices & IC’s
(ISPSD), Hong Kong, 2015, pp. 41–44.
106 Tanaka, K., Morita, T., Umeda, H. et al. (2015). Suppression of current
collapse by hole injection from drain in a normally-off GaN-based hybrid-
drain-embedded gate injection transistor. Appl. Phys. Lett. 107 (16):
163502-1–163502-4.
107 Fabris, E., Meneghini, M., De Santi, C. et al. (2019). Hot-electron trapping
and hole-induced detrapping in GaN-based GITs and HD-GITs. IEEE Trans.
References 249
108 Ikoshi, A., Toki, M., Yamagiwa, H. et al. (2018). Lifetime evaluation for
Hybrid-Drain-embedded Gate Injection Transistor (HD-GIT) under practical
switching operations. 2018 IEEE International Reliability Physics Symposium
(IRPS), Burlingame, CA, USA, 2018, pp. 4E.2-1–4E.2-7.
of normally-off GaN-based high electron mobility transistors. Materials 12
(10): 1599-1–1599-18.
110 Rossetto, I., Meneghini, M., Rizzato, V. et al. (2016). Study of the stability of
e-mode GaN HEMTs with p-GaN gate based on combined DC and optical
analysis. Microelectron. Reliab. 64: 547–551.
111 Masin, F., Meneghini, M., Canato, E. et al. (2019). Positive temperature
dependence of time-dependent breakdown of GaN-on-Si E-mode HEMTs
under positive gate stress. Appl. Phys. Lett. 115 (5): 052103-1–052103-4.
112 Efthymiou, L., Longobardi, G., Camuso, G. et al. (2017). On the physical
operation and optimization of the p-GaN gate in normally-off GaN HEMT
devices. Appl. Phys. Lett. 110 (12): 123502-1–123502-5.
113 Hwang, I., Kim, J., Choi, H.S. et al. (2013). p-GaN gate HEMTs with tung-
sten gate metal for high threshold voltage and low gate current. IEEE
114 Lee, F., Su, L.-Y., Wang, C.-H. et al. (2015). Impact of gate metal on the
performance of p-GaN/AlGaN/GaN high electron mobility transistors. IEEE
115 Wu, T.-L., Bakeroot, B., Lian, H. et al. (2017). Analysis of the gate
capacitance–voltage characteristics in p-GaN/AlGaN/GaN heterostructures.
IEEE Electron Device Lett. 38 (12): 1696–1699.
116 Meneghini, M., Rossetto, I., Rizzato, V. et al. (2016). Gate stability of
GaN-based HEMTs with p-type gate. Electronics 5 (4): 14-1–14-8.
117 Yatabe, Z., Asubar, J.T., and Hashizume, T. (2016). Insulated gate and surface
passivation structures for GaN-based power transistors. J. Phys. D: Appl.
Phys. 49 (39): 393001-1–393001-19.
approach toward normally off high-voltage AlGaN/GaN HEMT for power
electronics applications. IEEE Trans. Electron Devices 53 (2): 356–362.
119 Capriotti, M., Fleury, C., Bethge, O. et al. (2015). E-mode AlGaN/GaN
True-MOS, with high-k ZrO2 gate insulator. 45th European Solid-State
Device Research Conference (ESSDERC), Graz, Austria, 2015, pp. 60–63.
120 Gregušová, D., Jurkovič, M., Haščík, Š. et al. (2014). Adjustment of threshold
voltage in AlN/AlGaN/GaN high-electron mobility transistors by plasma
oxidation and Al2 O3 atomic layer deposition overgrowth. Appl. Phys. Lett.
104 (1): 013506-1–013506-4.
121 Blaho, M., Gregušová, D., Haščík, Š. et al. (2015). Self-aligned normally-off
metal-oxide-semiconductor n++ GaN/InAlN/GaN high-electron mobility
transistors. Phys. Status Solidi A 112 (5): 1086–1090.
122 Zhang, Y., Sun, M., Joglekar, S.J. et al. (2013). Threshold voltage control
by gate oxide thickness in fluorinated GaN metal-oxide-semiconductor
high-electron-mobility transistors. Appl. Phys. Lett. 103 (3):
033524-1–033524-5.
123 Miczek, M., Mizue, C., Hashizume, T., and Adamowicz, B. (2008).
Effects of interface states and temperature on the C-V behavior of
metal/insulator/AlGaN/GaN heterostructure capacitors. J. Appl. Phys. 103
(10): 104510-1–104510-11.
124 Matys, M., Stoklas, R., Kuzmik, J. et al. (2016). Characterization of capture
cross sections of interface states in dielectric/III-nitride heterojunction
structures. J. Appl. Phys. 119 (20): 205304-1–205304-7.
125 Lagger, P., Ostermaier, C., Pobegen, G., and Pogany, D. (2012). Towards
understanding the origin of threshold voltage instability of AlGaN/GaN
MIS-HEMTs. 2012 International Electron Devices Meeting, San Francisco,
CA, USA, 2012, pp. 13.1.1–13.1.4.
126 Ťapajna, M., Jurkovič, M., Válik, L. et al. (2014). Impact of GaN cap
on charges in Al2 O3 /(GaN/)AlGaN/GaN metal-oxide-semiconductor
heterostructures analyzed by means of capacitance measurements and
simulations. J. Appl. Phys. 116 (10): 104501-1–104501-7.
127 Ťapajna, M. and Kuzmík, J. (2012). A comprehensive analytical model for
threshold voltage calculation in GaN based metal-oxide-semiconductor
high-electron-mobility transistors. Appl. Phys. Lett. 100 (11):
113509-1–113509-4.
128 Shih, H.-A., Kudo, M., and Suzuki, T. (2014). Gate-control efficiency and
interface state density evaluated from capacitance-frequency-temperature
mapping for GaN-based metal-insulator-semiconductor devices. J. Appl.
Phys. 116 (18): 184507-1–184507-9.
129 Jackson, C.M., Arehart, A.R., Cinkilic, E. et al. (2013). Interface trap
characterization of atomic layer deposition Al2 O3 /GaN metal-insulator-
semiconductor capacitors using optically and thermally based deep level
spectroscopies. J. Appl. Phys. 113 (20): 204505-1–204505-6.
130 Hori, Y., Yatabe, Z., and Hashizume, T. (2013). Characterization of inter-
face states in Al2 O3 /AlGaN/GaN structures for improved performance of
high-electron-mobility transistors. J. Appl. Phys. 114 (1): 244503-1–244503-8.
131 Yang, S., Tang, Z., Wong, K.-Y. et al. (2013). Mapping of interface traps in
high-performance Al2 O3 /AlGaN/GaN MIS-heterostructures using frequency-
and temperature-dependent C-V techniques. 2013 IEEE International Elec-
tron Devices Meeting, Washington, DC, USA, 2013, pp. 6.3.1–6.3.4.
132 Capriotti, M., Lagger, P., Fleury, C. et al. (2015). Modeling small-signal
response of GaN-based metal-insulator- semiconductor high electron mobil-
ity transistor gate stack in spill-over regime: effect of barrier resistance and
interface states. J. Appl. Phys. 117 (2): 024506-1–024506-7.
133 Matys, M., Kaneki, S., Nishiguchi, K. et al. (2017). Disorder induced gap
states as a cause of threshold voltage instabilities in Al2 O3 /AlGaN/GaN
metal-oxide-semiconductor high-electron-mobility transistors. J. Appl. Phys.
122 (22): 224504-1–224504-7.
134 Bakeroot, B., You, S., Wu, T.-L. et al. (2014). On the origin of the
two-dimensional electron gas at AlGaN/GaN heterojunctions and its
References 251
influence on recessed-gate metal-insulator-semiconductor high electron

mobility transistors. J. Appl. Phys. 116 (13): 134506-1–134506-10.
135 Zhu, J., Hou, B., Chen, L. et al. (2018). Threshold voltage shift and inter-
face/border trapping mechanism in Al2 O3 /AlGaN/GaN MOS-HEMTs. 2018
IEEE International Reliability Physics Symposium (IRPS), Burlingame, CA,
USA, 2018, pp. P-WB.1-1–P-WB.1-4.
136 Bisi, D., Chan, S.H., Liu, X. et al. (2016). On trapping mechanisms at
oxide-traps in Al2 O3 /GaN metal-oxide-semiconductor capacitors. Appl.
Phys. Lett. 108 (11): 112104-1–112104-5.
137 Wu, T.-L., Franco, J., Marcon, D. et al. (2016). Toward understanding posi-
tive bias temperature instability in fully recessed-gate GaN MISFETs. IEEE
138 Ber, E., Osman, B., and Ritter, D. (2019). Measurement of the variable sur-
face charge concentration in Gallium Nitride and implications on device
modeling and physics. IEEE Trans. Electron Devices 66 (5): 2100–2105.
139 Yang, J., Eller, B.S., and Nemanich, R.J. (2014). Surface band bending and
band alignment of plasma enhanced atomic layer deposited dielectrics on
Ga- and N-face gallium nitride. J. Appl. Phys. 116 (12): 123702-1–123702-12.
140 Matys, M., Stoklas, R., Blaho, M., and Adamowicz, B. (2017). Origin of pos-
itive fixed charge at insulator/AlGaN interfaces and its control by AlGaN
composition. Appl. Phys. Lett. 110 (24): 243505-1–243505-5.
141 Esposto, M., Krishnamoorthy, S., Nathan, D.N. et al. (2011). Electrical prop-
erties of atomic layer deposited aluminum oxide on gallium nitride. Appl.
Phys. Lett. 99 (13): 133503-1–133503-3.
142 Ganguly, S., Verma, J., Li, G. et al. (2011). Presence and origin of interface
charges at atomic-layer deposited Al2 O3 /III-nitride heterojunctions. Appl.
Phys. Lett. 99 (19): 193504-1–193504-3.
143 Hashizume, T., Kaneki, S., Oyobiki, T. et al. (2018). Effects of postmetalliza-
tion annealing on interface properties of Al2 O3 /GaN structures. Appl. Phys.
Express 11 (12): 124102-1–124102-4.
144 Ťapajna, M., Drobný, J., Gucmann, F. et al. (2019). Impact of
oxide/barrier charge on threshold voltage instabilities in AlGaN/GaN
metal-oxide-semiconductor heterostructures. Mater. Sci. Semicond. Process.
91: 356–361.
145 Lagger, P., Reiner, M., Pogany, D., and Ostermaier, C. (2014). Comprehensive
study of the complex dynamics of forward bias-induced threshold voltage
drifts in GaN based MIS-HEMTs by stress/recovery experiments. IEEE
146 Reisinger, H., Grasser, T., Gustin, W., and Schluandnder, C. (2010). The sta-
tistical analysis of individual defects constituting NBTI and its implications
for modeling DC- and AC stress. 2010 IEEE International Reliability Physics
Symposium (IRPS), Anaheim, CA, USA, 2010.
147 Lagger, P., Steinschifter, P., Reiner, M. et al. (2014). Role of the dielectric
for the charging dynamics of the dielectric/barrier interface in AlGaN/GaN
based metal-insulator-semiconductor structures under forward gate bias

stress. Appl. Phys. Lett. 105 (3): 033512-1–033512-5.
148 Meneghesso, G., Meneghini, M., Bisi, D. et al. (2016). Trapping and reli-
ability issues in GaN-based MIS HEMTs with partially recessed gate.
Microelectron. Reliab. 58: 151–157.
149 Pohorelec, O., Ťapajna, M., Gregušová, D. et al. (2019). Investigation of
threshold voltage instabilities in MOS-gated InGaN/AlGaN/GaN HEMTs.
Proceedings of Advances in Electronic and Photonic Devices, High Tatras,
Slovakia (24–27 June 2019).
253
Light-Emitting Diodes
Amit Yadav 1 , Hideki Hirayama 2 , and Edik U. Rafailov 1
1 Optoelectronics and Biomedical Photonics Group, Aston Institute of Photonic Technologies (AIPT),
Aston University, Birmingham, B4 7ET, UK

2
RIKEN, Cluster for Pioneering Research (CPR), Quantum Optodevice Laboratory, W514 2-1 Hirosawa, Wako,
Saitama, 351-0198, Japan
7.1 Introduction
Light-emitting diodes (LEDs) have made tremendous progress in the past
15 years and have reached to a point where they are reinventing and redefining
artificial lighting. The efficiency and better control over light-quality parameters
have been the key attributes of LEDs that make them better than the existing
lighting solutions. Nevertheless, in their own realm, they suffer from a decrease
in efficiency at higher currents, i.e. the “efficiency droop” phenomenon. Thus, a
better understanding of the mechanisms leading to droop is of utmost impor-
tance. Moreover, the full potential in terms of light quality, i.e. color rendering
index (CRI) and correlated color temperature (CCT) that can be offered by these
devices, can be further improved with the existing or alternative schemes and
device configurations.
Inorganic LEDs are currently the most efficient light sources that are used
for indoor and outdoor lighting applications. These devices are fundamentally
monochromatic light emitters based on two material systems, InGaN/GaN
and AlGaInP. Emission across the whole visible spectrum from blue (based
on InGaN) to red (based on AlGaInP) can be achieved depending on the
composition of indium (In) and aluminum (Al) in the respective alloy. The
monochromatic (emission spectrum width of about 20–50 nm) nature of
emission of LEDs has made targeted lighting efficient and versatile with better
control on light parameters. On the other hand, to generate white light, either a
broadband source emitting across the visible spectral region or a combination of
two or more monochromatic sources of appropriate wavelength is needed. Given
the narrow emission width of the LEDs, white light emission is achieved using
few different approaches. The approaches of interest are (i) monolithic and (ii)
phosphor-converted light-emitting diode (pc-LED). The ability of these sources
to render true colors of an object is of critical importance for an adequate and
appropriate visual appearance.

254 7 Light-Emitting Diodes
Apart from lighting, LEDs are used in automotive, indoor nonwhite lighting,
signage, displays, and many more. The devices used in these applications, irre-
spective of the emission wavelength, internal quantum efficiency (IQE), are one
of the most critical parameters. The highly efficient InGaN/GaN devices emit in
the blue spectral. However, they suffer from the phenomenon of efficiency droop,
i.e. reduction of efficiency with increasing current density. The physical process
leading to this droop is still under investigation. The Auger recombination, carrier
leakage, and carrier delocalization along with device self-heating are the different
mechanisms that have been proposed. A similar decrease in efficiency as a func-
tion of current is also associated with AlGaInP-based LEDs. In this case, because
of different material properties, carrier leakage is considered to be the dominant
mechanism leading to reduction of efficiency. Nonetheless, further experimental
studies accompanied by theoretical estimations are needed to better understand
the physical mechanisms for efficiency droop in blue LEDs and decrease of effi-
ciency in red LEDs.
On the other hand, AlGaN deep ultraviolet light-emitting diodes (DUV LEDs)
have a wide variety of potential applications, including sterilization, water purifi-
cation, UV curing, and in the medical and biochemistry fields. However, the
efficiency of AlGaN DUV LEDs remains below values in comparison with that of
InGaN blue LEDs. Crystal growth techniques for wide-bandgap AlN and AlGaN
have been developed, and using these techniques, 220–350 nm band DUV LEDs
were demonstrated. Considerable increases in the IQE of AlGaN quantum wells
(QW) were achieved by developing low-threading dislocation density (TDD)
AlN grown. The electron injection efficiency (EIE) was substantially increased
by optimizing an electron-blocking layer (EBL). The light extraction efficiency
(LEE) was also improved by using a transparent p-AlGaN contact layer, a highly
reflective (HR) p-type electrode, and an AlN template fabricated on a patterned
sapphire substrate (PSS). Further improvements were made by implementing
a reflective photonic crystal (PhC) p-contact layer. A record external quantum
efficiency (EQE) of 20.3% was demonstrated for a 275 nm AlGaN UVC-LED.
7.2 State-of-the-Art GaN LEDs

One of the key features of the LEDs is the promise of high energy efficiency,
leading to reduced energy consumption and hence contributing toward climate
change. In an ideal scenario, 100% efficiency is achieved when each injected elec-
tron results in emission of a photon. This would translate into a linear dependence
of optical output with respect to injected current, which would indicate that
dependence of EQE on current (linear scale) would be nonvarying and constant.
However, in real world, seldom such ideal conditions could exist especially when
energy transformation is the subject. In that respect, LEDs are no different, and
one of the crucial challenges to be addressed to achieve their full potential is the
well-known phenomenon of “Efficiency Droop” [1]. This gradual decrease in effi-
ciency, after certain current density, is a serious issue with both InGaN LEDs. It
can be calculated using the expression:

EQEpeak − EQEI
EQEdroop = (7.1)
EQEpeak
The Efficiency Droop phenomenon in III-nitride LEDs has been observed under
both photoluminescence (PL) and electroluminescence (EL) and is under intense
investigation and debate in the research community. Several different physical
processes have been proposed over the past decade as the probable cause while
a consensus on one or group of these processes still evades researchers. A brief
introduction about some of the proposed mechanisms is presented below.
Auger Recombination
Being a three-particle process, Auger recombination is the most debated
mechanism concerning droop phenomenon for GaN LEDs. Direct Auger
recombination is a nonradiative process in which when an electron and hole
combines, the energy is not released as a photon and is instead transferred to
the third carrier, an electron or a hole, exciting it to another energy level, i.e.
higher in conduction band for electron and deeper or lower in the valence band
for holes. It has a cubic dependence on carrier concentration and hence should
contribute more toward the droop with increasing current. It is accounted
by the coefficient C in the ABC-model. The debate on the contribution of
Auger recombination to efficiency droop is effectively centered around different
experimental values reported for Auger coefficient (C). Shen et al. [2] reported
values of the order of 10−30 cm6 /s by performing resonant optical excitation
on double-heterostructure (DH) InGaN layer, whereas values in the range of
10−27 –10−24 cm6 /s were reported by Ryu et al. [3]. A number of experimental
studies to extract the value of C using the ABC model, neglecting carrier
escape or leakage, have been performed on a single and multiple QW and are
reviewed by Piprek [1]. On the other hand, much lower theoretical values for
Auger coefficient of the order of 10−34 cm6 /s [4] have been calculated. The
low theoretical values (based on ABC model) are in contrast to the exper-
imental values and are less likely to provide a good fit to the experimental
data. Toward this, to account for the difference between the theoretical and
experimental values, additional Auger recombination mechanisms have been
proposed. Phonon-assisted indirect Auger recombination requiring a phonon
for a third carrier to make a transition has been accounted along with direct
Auger recombination by Kioupakis et al. [5] yielding Auger coefficient values
in the range of (0.5–2) × 10−31 cm6 /s. However, this range is only enough to
account for the lowest experimentally reported values of 1.8 × 10−31 cm6 /s [6]
and is not enough to account for most experimentally observed values in the
range of ∼1.4 × 10−30 to 1 × 10−24 cm6 /s [2, 3, 7–9]. On the other hand, Delaney
and coworkers calculated, from first principles using density functional and
many-body-perturbation theory, the values of Auger coefficient in the range
of 1 × 10−34 to 5 × 10−28 cm6 /s. They attributed this agreement of the values
to the interband Auger recombination by accounting for another band above

the conduction band for the wide-bandgap semiconductors, thus estimating a
coefficient of 2 × 10−30 cm6 /s in complete agreement with Shen et al. [2].
Carrier Delocalization
Most GaN devices are grown on sapphire substrate with a large lattice constant
mismatch, leading to a large number of defects, mainly point defects and thread-
ing dislocations, with a density of ∼109 cm−2 . Despite that, III-nitride devices
exhibit high IQE [10–14]. This has been explained by localization of carrier in an
in-plane local potential minimum before the carriers can reach and recombine at
dislocations. Several mechanisms have been proposed for such localization to be
plausible and they are (i) fluctuations in width of QW [15–18], (ii) In clustering
[19, 20], (iii) random alloy fluctuations [21, 22], and (iv) hexagonal V-shaped pits
or funnels caused by threading dislocations [23]. The first three mechanisms indi-
vidually or collectively are responsible for localized potential minima, whereas
the V-pits around the threading dislocations are antilocalization. It has been sug-
gested that the very narrow QWs at the sides of the hexagonal pits creates a high
potential barrier around dislocation sites and thus keeping them physically dis-
tant from these threading dislocations.
Thus, it is reasonable to assume that local fluctuations in potential keeps the
carriers and nonradiative recombination centers physically separated. Therefore,
when the active region (AR) is injected with low current densities, these poten-
tial minima are able to confine the carriers locally. However, at higher current
densities, these local minima are overfilled with carriers and they start to escape,
i.e. become delocalized only to be captured by dislocations and lost to nonradia-
tive recombinations and thus lowering the efficiency [10, 11, 13, 24, 25]. These
carriers can further lead to increased junction temperature because of heating
caused by parasitic tunneling current facilitated by threading dislocations [26].
This would only increase with increased carrier injection promoting acoustic
phonon-assisted tunneling [26].
Haider et al. have in their studies proposed density-activated defect recombi-
nation (DADR) as the possible mechanism for efficiency droop. They argued that
the potential minima caused either by In composition fluctuations or QW width
fluctuations shield the carriers from defects at low carrier density; however, at
higher densities, the carriers escape to rest of the QW, thus recombining easily
at defect recombination centers. They modeled the nonradiative loss because of
carrier delocalization and demonstrated a good fit with the experimental data for
LEDs emitting at 410 and 530 nm.
More evidence toward delocalization was presented by Wang et al. [13]. They
studied the efficiency droop behavior of two LEDs grown with different under-
layers (GaN and InGaN). Based on the PL studies, they argued that the LED with
InGaN underlayer had higher degree of localization. For analysis, EQE(I) is ana-
lyzed in three parts. They attributed higher peak IQE at lower current densities,
i.e. part (1) of EQE(I) and rapid drop at increased current densities, i.e. part (2) of
EQE(I), to carrier delocalization. They further suggested that for current densi-
ties >24 A/cm2 , i.e. part (3) of EQE(I), efficiency droop is more likely to be caused
by carrier leakage. Similar observation regarding carrier delocalization has been

made by Hammersley et al. [27] in their temperature-dependent PL studies on
InGaN QW LEDs.
On the other hand, contrary to all these observations, Schubert et al. [28] based
on recombination rate equation analysis of two InGaN LEDs with different dis-
location densities, 5.3 × 108 and 5.7 × 109 cm−2 , suggest carrier leakage at high
current densities to be the dominant mechanism for efficiency droop. The experi-
mental results presented in their paper clearly demonstrate higher peak efficiency
and larger droop from low dislocation density LED.
Moreover, indium-rich islands or In clustering due to segregation of InGaN
is brought into doubt by Smeeton et al. [29]. They described that observation
of such structures in transmission electron microscopy (TEM) measurements
[20] is due to the damage done by the electron beam rather than the compo-
sitional fluctuations. Galtrey et al. [30] later confirmed these observations with
3D atom probe measurements and describing observed indium distribution as
random alloy distribution.
Electron Leakage
Electron leakage is the term used to account for all the injected electrons that
escape the QWs in the active region and thus not contributing toward photon
emission irrespective of the physical mechanism responsible for loss of electrons
to nonradiative recombination. These electrons then recombine in the p-GaN or
p-type electrode. To restrict the leakage of carriers to p-side of the device, AlGaN
EBL is implemented adjacent to p-GaN in the device structure. The idea is to cre-
ate a high enough potential barrier for electrons to be reflected back toward the
QWs. However, an emission beyond EBL in the p-doped region of the device
has been observed [31, 32]. This observation on the one hand affirms the inef-
fectiveness of EBL; on the other hand, it also relates electron leakage directly to
efficiency droop. The key observations made by the author relating droop and
leakage are (i) spontaneous emission increases with current, (ii) lower peak EQE,
and (iii) onset of droop shifting to higher currents. The ineffectiveness of EBL
has been mainly attributed to polarization fields [33] for polar GaN LEDs grown
on c-plane sapphire substrate. The positive sheet charge accumulated at the GaN
barrier and AlGaN EBL interface, due to the difference in the degree of spon-
taneous and piezoelectric polarization between the two, attracts electrons. This
skews the energy band diagram of EBL rather negatively, thus reducing its effec-
tive height in terms of energy. One way to increase EBL barrier height would be to
increase the Al content of the AlGaN EBL layer; however, the achieved increase
in barrier height would be ineffective because of a larger increment in the con-
duction band offset between GaN spacer and AlGaN EBL [34].
Among other mechanisms, poor hole injection [35, 36] and asymmetric carrier
concentration and mobility are considered as the cause of electron leakage [37].
Both these mechanisms have common root in not being able to achieve high hole
concentration because of self-compensation of Mg dopant, limiting the p-type
doping of GaN. This results in lesser number of holes as compared to electrons
obtained at n-GaN with Si doping due to low ionization energy of ∼17 meV [38]
compared to 170 meV ionization energy needed for Mg acceptors in GaN [39].
Along with this, EBL, which is implemented to restrict electron escape, poses a
hindrance to hole movement because of valence band offset at the AlGaN and
GaN spacer, thus leading to poor hole injection into the active region. This has
been demonstrated to be a factor contributing to droop by implementation of
a three-terminal diode, where two anodes were used to improve hole injection
[40]. On the other hand, the carrier concentration asymmetry due to difference
in electron and hole concentration, for the above-mentioned reasons, along with
lower hole mobility due to its higher effective mass has been together identified
as mechanisms contributing to electron leakage and hence to droop [41].
7.2.1 Blue LEDs

InGaN-based blue LEDs are at the core of the present-day phosphor-covered
white LEDs. These devices have seen significant improvements in their quantum
efficiencies over the years with improved chip designs, growth, and fabrication
technologies. Although the IQE for such devices has improved to more than 80%
[42], they still suffer from drop in efficiency with increased operating current,
i.e. the efficiency droop phenomenon. This presents as a fundamental impedi-
ment for InGaN-based high-power devices designed to emit more optical power
per unit area of the device because they operate under current/current densities
different from the peak efficiencies.
Efficiency droop in InGaN LEDs has been studied rigorously, and various
mechanisms contributing to it have been reported. Among these processes,
Auger recombination [2, 43, 44] is considered to be the major contributor as
has been demonstrated experimentally [45–47]. On the other hand, carrier
leakage outside the active region due to asymmetric p–n junction [41, 48]
and/or polarization effects [34, 49], saturated radiative recombination [50],
poor hole injection [51], and defect-assisted tunneling are among the suggested
theories for efficiency droop. Nevertheless, a general consensus still evades the
researchers. This however has not impeded the progress in addressing the issue.
Among the LEDs grown on c-plane sapphire, engineering of QW and EBL has
been done to reduce or overcome carrier leakage [13, 52–56], polarization effects
[34, 57–59], and QW carrier density [60, 61]. Nevertheless, understanding the
processes involved in efficiency droop is critical for further improvements to
EQE of InGaN-based EQE.
External quantum efficiency is inherently a product of internal quantum
efficiency (𝜂 int ) and light extraction efficiency (𝜂 ext ) (assuming negligible carrier
leakage) and provides information on radiative and nonradiative recombination
processes. Thus, separate evaluation of IQE and LEE would be useful to deter-
mine a correlation between device structure and recombination processes. IQE
and LEE are most commonly determined theoretically by finite difference time
domain (FDTD) modeling [62] in the case of LEE and temperature-dependent
variable excitation PL [63, 64] and temperature-dependent electroluminescence
(TDEL) [65, 66] for IQE. Although FDTD is resource hungry and time consum-
ing, PL and TDEL are based on intuitive assumptions that temperature has no
affect in LEE, and at cryogenic temperatures, nonradiative recombination can be
ignored. In the latter case, this leads to nearly 100% IQE, which according to the
data presented in Refs. [67, 68], indicating a peak value of IQE under a certain
range of current, seems improbable even at low temperatures, and would require
further experiments to confirm the assumption. The former assumption of LEE
not being affected by temperature would be rendered incorrect, taking into
account that light absorption, whether within the die or in the contact layers due
to free carriers, affects LEE because of thermally activated donors and acceptors.
Recently, more practical approaches requiring minimum or no computation
based on ABC-model have been used to determine IQE [69, 70]. In this section,
the efficiency evolution and spectral behavior of blue LED over the broad range of
temperatures 13–440 K is presented. To do so, a stepwise procedure as described
in Ref. [67] based on ABC model is used.
Current–voltage (I–V ) characteristics of the blue LED are shown in Figure 7.1.
The I–V characteristics of the LED at all temperatures are similar to those of a
typical diode, thus indicating good electrical quality of the p–n junction and the
other layers of the chip.
For all temperatures, the measured I–V curve can be analyzed for low and high
driving currents. Typically, the section of the I–V plot with forward voltage less
than ∼2.2 V is associated with low driving currents. On the other hand, for high
currents, the typical forward voltage is more than ∼2.7 V. Different mechanisms
or processes can be attributed to these voltage levels.
Figure 7.2 depicts the experimentally measured EQE as a function of current
at various temperatures. It can be seen from Figure 7.2 that for practically all
temperatures, the minimum current at which EQE could be measured is well
above the low current (>10−8 A, for temperature 13 and 50 K) region of the I–V
curves corresponding to tunneling of carriers. For higher temperature, the low
and high current sections of the I–V curves merge together for current values,
less than or equal to the values of current for which EQE can be measured. This
100
10–1
10–2
10–3
Current (A)
10–4
13 K
10–5 50 K
100 K
10–6 150 K
200 K
10–7 250 K
300 K
10–8 400 K
440 K
10–9
1 2 3 4 5 6 7 8 9 10
Forward voltage (V)
Figure 7.1 Temperature-dependent current–voltage characteristics of blue LED. Source:

Adapted from Titkov et al. 2014 [67], with the permission of IEEE.
80
70
70 60
50
60
40
50 0.01 0.1 1
EQE (%)
40
30
13 K
20 100 K
200 K
10 300 K
440 K
0
10–7 10–6 10–5 10–4 10–3 10–2 10–1 100 101 102 103
Current (A)
Figure 7.2 Temperature-dependent EQE of blue LED. (Inset shows that the temperature-
dependent EQE does not intersect even at high currents.) Source: Adapted from Titkov et al.
2014 [67], with the permission of IEEE.
indicates that low current carrier loss is not affecting the EQE measurements and
thus allowing us to process the EQE data with the above-mentioned procedure.
On the other hand, carrier injection in the active region dominates at a high
current section of the I–V curves irrespective of operating temperature. For
this part of the I–V curve, the slope changes with temperature. This variation
is accounted for by the current dependent p–n junction resistance and a diode
series resistance. The LED series resistance is estimated across the complete
temperature spectrum from 13 to 440 K by accounting for the series resistance in
the adapted Shockley’s diode equation. Fitting the experimental curves with the
adapted diode current equations reveals a linearly decreasing series resistance
from 7.1 to 6.0 Ω with increasing temperature from 13 to 440 K, respectively.
Furthermore, a very little change in LED series resistance establishes the
electrical stability of the Ohmic contacts for both the n-type and p-type contact
pads. It also indicates that carrier freezing affect is absent from the n- and p-type
contact layers even at temperatures as low as 13 K.
EQE curves for all temperatures exhibit the characteristic dome like shape as
reported in the literature for InGaN/GaN LEDs. The well-known efficiency droop
can be observed at all temperatures. It is crucial to note that even at high currents
up to 800 mA, the temperature EQE curves do not intersect. This observation is
quiet opposite to the previous results reported on blue LEDs [68], where EQE
curves at lower temperatures intersect with the high-temperature curves. Unlike
the results presented here in Ref. [68], the onset of droop shifts to lower currents
with decreasing temperature, thus resulting in the intersection of curves.
The maximum value of EQE along with the EQE value at an operating current of
350 mA for each temperature is shown in Figure 7.3. The maximum value of EQE
decreases with temperature from ∼74% at 13 K to ∼45% at 440 K. On the other
hand, the operating current EQE shows a very weak dependence on temperature
as it is reduced only by ∼9% from 13 to 440 K.
100
90 EQEmax
EQE (350 mA)
80
70
EQE (%)
60
50
40
30
20
10
0 100 200 300 400
Temperature (K)
Figure 7.3 Temperature-dependent behavior EQE (350 mA) and EQEmax of blue LED. Source:
Adapted from Titkov et al. 2014 [67], with the permission of IEEE.
Spectral Analysis
The EL spectra measured at 3 mA for the blue LED is presented in Figure 7.4.
A sharp peak around ∼452 nm is observed at all temperatures from 13 to 300 K.
This peak corresponds to the band-to-band excitonic emission from the QWs.
Also, the presence of two phonon replicas, denoted as 1LO and 2LO, where LO
13 K
50 K 457
Wavelength (nm)
80 K
100 K 456
150 K
200 K 455
EL intensity (a.u.)
250 K
300 K 454
0 100 200 300
Temperature (K)
1st LO
2nd LO
420 440 460 480 500

Wavelength (nm)
Figure 7.4 Emission spectrum for blue LED and distinctly visible LO phonon (first and second)
at temperatures ≤150 K. Inset: S-shaped dependence of dominant wavelength. Source:
Adapted From Titkov et al. 2014 [67], with the permission of IEEE.
indicates longitudinal optical phonon, for temperatures ≤150 K is an indicator of

good-quality active region on the device.
From Figure 7.4, it is observed that at higher temperatures, the EL spectra
exhibit wings on both sides of the main emission peak around ∼452 nm. The high
energy wings are attributed to carrier evolution in the energy bands of the active
region with increasing temperature, whereas the long-wavelength wings are mere
merging of LO phonon replicas with the peak.
Under EL excitations with increasing temperature, the dominant emission
wavelength undergoes a red shift, followed by a blue shift and then again
gets red-shifted, thus exhibiting a S-shaped temperature dependence. Such
a dependence can be explained by changing carrier dynamics due to carrier
localization and inhomogeneities in the InGaN/GaN QW structures. A rather
small shift of ∼3 nm over the temperature range of 13–300 K is observed, thus
demonstrating a weak temperature dependence and stable emission behavior.
7.2.2 Green LEDs

Green LEDs with high efficiency are of significance for solid-state lighting. The
EQE of these LEDs have made considerable improvements over the last decade.
This improvement is made possible because of improved carrier transport and
chip designs along with strain engineering of multiple quantum well (MQW)
active regions. Currently, green LEDs exhibit more than 30% efficiency, which
can be further improved to >50% with internal down conversion [71]. The EQE
for green LEDs with all the improvements is still lagging, by a long distance, when
compared to more developed blue LEDs with >70% EQE [71]. This huge gulf in
the EQE of green LEDs has been primarily attributed to inbuilt polarization fields
in the active region, the In composition fluctuation in the QWs leading to car-
rier localization and nonoptimal crystal quality at high In composition to achieve
emission in the green LED. This drop in EQE for InGaN-based LEDs emitting at
longer wavelengths, moving from blue to orange, is called “green-gap.” To better
understand the “green-gap” problem, the contributions of the different mecha-
nisms mentioned above needs to investigated. Toward this, dependence of EQE
on temperature can be a very useful tool. In contrast to blue LEDs, the depen-
dence of EQE of green LEDs with respect to current is asymmetric [7]. Various
mechanisms such as carrier delocalization [72], electron and hole imbalance in
QWs [73], suppressed nonradiative recombination in InGaN alloys [74], and elec-
tron leakage [75] have been cited for this asymmetric EQE dependence. In this
section, spectral and efficiency behaviors of the green LEDs are presented.
Spectral Analysis
The emission spectrum of the commercial green LEDs is measured over a long
range of temperature from 13 to 300 K. This investigation over a wide range
of operating currents reveals the presence of two emission peaks that are not
distinct. Such a behavior is clearly observed for temperatures up to 200 K.
For higher temperatures, the two peaks are not distinctly distinguishable. The
emission peaks are ∼10 nm apart at 535 and 545 nm. The shorter wavelength
dominates at lower output powers, while longer wavelength dominates at higher
powers, see Figure 7.5. This transition of intensity between two peak occurs in
Figure 7.5 Emission spectrum Output

of the green LED at cryogenic power (μW)
temperature depicting
3.4 18.1
two-peak behavior and
9.0 21.5
transition from short
EL intensity (a.u.)
14.9 25.2
wavelength to long
wavelength with respect to
output powers. Source:
Adapted from Titkov et al. T = 13 K
2017 [76].
450 480 510 540 570 600 630 660

Wavelength (nm)
a narrow range of optical power from 10−6 to 10−5 W. Such an observation of

emission peak transition from ∼535 to ∼ 545 nm is made by normalizing EL
intensity as explained in Ref. [76]. For temperatures up to 200 K, a similar behav-
ior of normalized emission spectra can be observed. At higher temperatures, the
distinction between two peaks and hence the transition is indistinguishable. This
switching of peak emission can be attributed to nonuniformity of active region
either in vertical or in lateral direction.
Emission Efficiency
To evaluate EQE for the green LED, output power and emission wavelength at a
wide range of operating currents are utilized. Figure 7.6 presents the EQE evo-
lution at a wide current range at different temperatures from 13 to 300 K. Unlike
60
50
40
EQE (%)
30
20
Temperature (K)
13 300
10 75 250
150 200
0
10–8 10–7 10–6 10–5 10–4 10–3 10–2 10–1 100
Current (A)
Figure 7.6 EQE of green LED at various temperatures in the range of 13–300 K over a wide
range of currents. Source: Adapted from Titkov et al. 2017 [76].
70 Figure 7.7 Simple ABC mode

fit (magenta line) for the
experimental points (cyan
60 squares). Source: Adapted
from Titkov et al. 2017 [76].
50
EQE (%)
40
T = 13 K
30
20
10–9 10–8 10–7 10–6 10–5 10–4 10–3 10–2 10–1

Output power (W)
a symmetrical dome-like EQE dependence on current for blue LEDs, a nonflat

two-peak-shaped behavior of EQE for green LEDs is clearly observed for tem-
peratures up to 75 K. A similar observation can be made when EQE is plotted as
a function of output power, see Figure 7.7. The two peaks are located unambigu-
ously at different current/output power values. For temperatures >75 K, the peak
at lower current/output power starts to diminish until it merges with the peak at
high current/output power at temperatures >250 K. To calculate EQE, centroid
wavelength is used because of switching of peak wavelength (as discussed above).
It is also evident from Figure 7.7 that the simple ABC model does not pro-
vide the best fit for the twin peak EQE experimental data, and therefore, accurate
evaluation of IQE is difficult. To mitigate the deficiencies of simple ABC model,
a modified model must be used; one such approach is described in Ref. [76].
The model described in Ref. [76] accounts for nonuniform properties of QWs,
which is the case here. Using such model, a better fitting of experimental data
accounting for both the peaks and high current/output power can be achieved
and is presented in Figure 7.8. From this representation, it can further be ascer-
tained that the low power and high power EQE can be attributed to two spectral
emissions at 535 and 545 nm, respectively. The model in Ref. [76] accounts for
nonuniformity in the active region (AR) by assuming different subregions within
the AR emitting at different wavelengths under varying conditions. This nonuni-
formity could be vertical or in-plane or both.
7.3 GaN White LEDs: Approaches and Properties

Human eyes have three types of cones that are most sensitive to specific
wavelength range. Three primary wavelength or color ranges are 420–440 nm,
530–540 nm, and 600–630 nm. Each of these ranges translates into blue, green,
and red color regions, respectively. When mixed in an appropriate spectral ratio,
other colors including white can be generated from these primary wavelengths.
Figure 7.8 Experimental 70

data for EQE (cyan squares)
are well described by the
modified ABC model 60
accounting of nonuniformity
(magenta line). Source: 50
Adapted from Titkov et al.
EQE (%)
2017 [76].
40
T = 13 K
30
20
10
10–9 10–8 10–7 10–6 10–5 10–4 10–3 10–2 10–1
Output power (W)
On the other hand, chromaticity coordinates for white light are located in the
center at the equal intensity point of the chromaticity diagram. Thus, a number
of optical spectra combinations can provide the needed white light emis-
sion. Toward this, LED-based dichromatic, trichromatic, and tetra-chromatic
approaches with two, three, and four emission peaks, respectively, can be
exploited to generate white light. These approaches can be realized in more than
one configuration. The most common and commercially available configuration
is phosphor-covered (pc-) LED. Most of these sources use InGaN/GaN-based
blue LEDs to pump broadband yellow phosphor (YAG:Ce3+ ) and produce white
light. These light sources have shown good values of CRI around 70–80, CCT
in the range of 4000–8000 K [77], and luminous efficacy of 160 lm/W [78].
The lack of red component has been the primary reason for low CRI for this
configuration. Thus, a multiphosphor approach is adopted where pumping LED
can be UV or blue, but the photon recycling coating of phosphor is a mixture of
two or three phosphors based on yttrium aluminum garnet (YAG; 540–560 nm),
lutetium aluminum garnet (520–540 nm), oxynitrides (500–650 nm), and
nitrides (615–660 nm), and LEDs with CRI ∼95 are made commercially available
[79, 80]. Although high CRI sources can be realized using many phosphors, it
has some inherent issues. First, the well-known Stoke’s losses increase with the
addition of another phosphor, thus affecting the overall efficiency of the source.
Second, the angular uniformity of emission is dependent on the scattering
of light due to particle size, and efficiency of phosphor is also a function of
particle size. Thus, use of multiphosphor poses optimization issue in achieving
reasonable efficiency and uniformity of emission. Despite this, the advantages
of broad emission spectrum and absorption strength, stable emission spectrum,
and efficiency at a broad range of temperatures, along with saturation tolerance
at stronger flux levels, are driving the development and use of pc-LED.
The phosphor-free approaches to white light generation are multichip and
monolithic LEDs. A multichip source is a combination of three or more LEDs
(thus tri- or tetra-chromatic) emitting at different wavelengths in the red,

green, and blue region of the spectrum. Although addition of more LEDs will
improve the CRI, the luminous efficacy of these sources is limited primarily
because of “green gap” problem, i.e. low EQE in the wavelength region of eye
sensitivity curve. Furthermore, because the two different material systems (as
discussed before) are responsible for emission in high- and low-energy part of
the spectrum, they exhibit different optimal operating parameters, leading to a
requirement of complex circuitry for operation. Strong temperature dependence
of AlGaInP LEDs also affects their operation at elevated temperature, leading to
nonoptimal efficiency performance along with color variation [81].
Because InGaN/GaN, in principle, can emit across the visible spectrum
by varying the In composition, it has inspired researchers to develop sin-
gle LED emitting at multiple wavelengths. This monolithic approach is also
under development and investigation. In 2001, Damilano et al. proposed
InGaN/GaN MQW monolithic white LEDs [82]. Since then, these LEDs
are under development for improvement of efficiency and color parame-
ters. Under this approach, following the basic principle of mixing different
wavelengths (colors) in appropriate proportion (intensity), multiple InGaN
quantum wells (QW) emitting at different wavelengths are used to achieve
white light. MQWs are sandwiched between p- and n-type GaN layers and
are intrinsic (or undoped) themselves. To achieve emission at short and long
wavelengths, indium (In) incorporation in the stacked QWs is varied and
is predefined during the growth process. As no phosphor is used in this
approach, the efficiency losses associated with phosphor are nonexistent along
with the reduction of one process step (of phosphor deposition) in the LED
fabrication. Although this approach follows the same principle as a multichip
approach of color mixing, the fundamental difference is that a single chip
emits different wavelengths against the assembly of several monochromatic
chips using separate driver circuits and a feedback mechanism to main-
tain appropriate color mixing. Also, as this approach uses a single material
system, the different aging times in the multichip approach is addressed.
Hence, monolithic LEDs show promising improvements and can be the way
forward for solid state lighting (SSL) in the near future. Apart from QWs,
nanopyramid GaN [83] and ZnO nanowires on GaN heterostructures [84]
are also under investigation. Other approaches such as QW light converters
[85] and distributed Bragg reflector (DBR) resonant cavity have also been
investigated [86].
The LED-based lighting solutions with their superior efficiency are rapidly
replacing the traditional sources for indoor and outdoor applications alike. The
ability of these sources to render true colors of an object is of critical importance
for an adequate and appropriate visual appearance at place such as art galleries,
museums, hospitals, and exhibitions. Figure 7.9 depicts an example of color
perception using Munsell color pallets under light sources with different CRIs.
Source with a CRI of 99 (see Figure 7.9) better differentiate the colors, thus
placing emphasis on an LED source with ultimate color rendition along with
maximum efficacy.
Figure 7.9 Munsell samples under

illumination with different (99 and
56) CRI sources.
CRI = 99 CRI = 56
7.3.1 Monolithic LEDs

White LEDs based on InGaN are used in many applications from backlighting
to mobile displays and general illumination [77, 87] 1 . The requirements on
characteristic and quality of white light emission differ with applications. While
phosphor-covered blue LEDs with improved CRI and luminous efficacy of
radiation (LER) are replacing traditional light sources1 at offices, museum,
and similar application areas, high CRI is irrelevant for indicator and signage.
For outdoor street lighting, industrial use and parking spaces sources with
CRI ≥ 60 and CCT ≤ 8000 K are considered adequate [88, 89].1 Hence, white
light source with tunable CRI and CCT are desirable. Toward this, a multichip
approach with vertical or lateral combination of blue, green, and red LEDs can
be used for white light generation. However, this approach requires complex
fabrication procedures, driving circuits, and device design, thereby affecting
the reliability and increasing production cost [77]. White light emission has
also been achieved with CdSe/ZnS nanocrystals and by doping of InGaN QWs
with Si and Zn [79, 80]. These approaches along with the shortcomings of the
multichip approach have an additional disadvantage of nontunability. Mean-
while, the theoretical possibility of emission from ∼0.7 to ∼3.5 eV by indium (In)
variation in InGaN QWs is also being explored [82, 90–92]. This semiconductor
monolithic approach holds the potential of efficient color-tunable sources with
high CRI.
Monolithic approaches involving nanostructure engineering have been
explored previously for white light emission. A dual-wavelength, 5000–20 000 K
color-tunable, multifacet QW LED was demonstrated by Funato et al. [93].
Nguyen et al. have reported color tunability and CRI values more than 90
for “dot in a wire” core shell LEDs on silicon [94]. Nevertheless, this approach
requires a complex growth and fabrication process with precise control over wire
diameter and dot size, which is not ideal for mass production. Also, luminous
efficacy of these devices is still far from phosphor-covered LEDs [94]. Li et al.
demonstrated dual-wavelength MQW LED with 46% indium (In) content for
red emission peak at 2.12 eV, thus achieving emission color from red to yellow
1 Inc., C. Cree First to Break 300 Lumens-Per-Watt Barrier.

to white with CRI of 85.6 [95]. However, large lattice mismatch between higher
In-incorporated InGaN red QW on GaN substrate results in issues related
to charge separation and increased defect density [96]. A simpler fabrication
approach of vertically stacked QWs emitting at two distinct wavelengths has
been reported. Active region of such devices can consist of either (i) longer
wavelength emitting passive QWs pumped by active blue QW [97] or (ii) all
electrically pumped active QWs [82, 91, 98]. The first approach is similar to
phosphor-covered LEDs, and the total emission spectrum is dependent on
passive QWs in active region. Also, because the passive QWs are designed to
operate on the green gap spectral range, they being less efficient than phosphor
for that range and their sensitivity to active QW emission wavelength is a
disadvantage [99]. The best CRI for such devices reported is 41 [97]. White
light emission from all electrically pumped QWs with CCT ∼ 6000 K has been
reported previously [100].
In this section, an all electrically pumped phosphor-free, color-tunable mono-
lithic white LED is presented. The device is designed for dichromatic emission
at 450 and 550 nm. The emission wavelengths are chosen such that the line join-
ing the corresponding x–y coordinates on CIE 1931 chromaticity diagram passes
through the white region.
CRI and Spectral Analysis

An EL behavior of monolithic white LEDs is studied at room temperature (RT)
in continuous-wave (CW) regime. Figure 7.10 shows EL spectra for these devices
with increasing currents up to 500 mA. Two distinct peaks in Figure 7.10 indi-
cate that both shallow and deep In concentration QWs are operating in their
respective blue and green spectral regions. From Figure 7.10, it is seen that at
lower currents less than 80 mA, the green peak dominates the emission spectrum
with blue emission getting stronger with increasing carrier concentration. This is
attributed to nonuniform distribution of injected holes. Primarily because of their
4 50 mA
80 mA
100 mA
200 mA
3 400 mA
Intensity (a.u.)
500 mA
0
400 450 500 550 600 650
Wavelength (nm)
Figure 7.10 Electroluminescence (EL) spectra of monolithic white LED pumped under CW
current from 50 to 500 mA. Source: Adapted from Yadav et al. 2018 [101].
Correlated color temperature (K)

12 500 CCT
10 000
7500
5000
0.1 0.2 0.3 0.4 0.5

Current (A)
Figure 7.11 CCT tuning as a function of current. Source: Adapted from Yadav et al. 2018 [101].
lower mobility and higher effective mass, holes will radiatively recombine in the
QW closer to the p-side. However, with further increase in current, more holes
travel through the barrier layer and are available for recombination in blue QWs.
On further increasing the current up to 100 mA, the emission from green QW
is clamped and the radiative recombination in the blue QW is enhanced. For very
high CW current injection, i.e. >300 mA, increased radiative recombination in
green QW and saturated blue emission peak is observed (see Figure 7.11). This
is understood to be caused by the carrier redistribution because of band-filling
of blue QW leading to electron overflow, thus making more carriers, i.e. electron
available for recombination in green QW.
The tuning of CCT and G/B ratio with varying current is shown in Figure 7.11.
From Figures 7.11 and 7.12, it can be seen that the highest CCT obtained from
this device and the minima of green/blue (G/B)-integrated intensity ratio occurs
at the almost same values of current. This indicates the dominance of blue peak in
this region of operation. For injection current between 100 and 350 mA, G/B ratio
Figure 7.12 Variation 3

green/blue (G/B) ratio with G/B ratio
CW current. Source: Adapted
from Yadav et al. 2018 [101].
Green/blue (G/B) ratio
0.0 0.1 0.2 0.3 0.4 0.5

Current (A)
is <1 (see Figure 7.12) and blue peak dominates, resulting in cool white emission.
For other regions, increase in green intensity results in tuning of emission toward
warmer color temperature.
Commission International de I’Elairage (CIE 1931) chromaticity coordinates
with associated CCT at different currents is shown Figure 7.13. The coordinates
(0.4172, 0.4375) at 40 mA moves to (0.2686, 0.2716) at 240 mA, and CCT
increases from 3600 to 13 240 K (Figure 7.13a). With further increase in current,
a movement toward warmer CCT values can be seen with 4775 K at (0.3607,
0.4278) for 500 mA (Figure 7.13b). This distinctly depicts an excellent warm
to cool white CCT tunable simple monolithic LED. The CRI for device under
test is under 40 for less than 70 mA; however, with increasing carrier density,
the spectral broadening of the green peak is asymmetric and spectral emission
0.6
40–240 mA
0.5
40 mA
CIE y coordinate
50 mA
0.4 3600 K
70 mA 3978 K
100 mA
4819 K
150 mA 6220 K
0.3
8956 K
240 mA
13 241 K
0.2
0.25 0.30 0.35 0.40 0.45

CIE x coordinate
(a)
500 mA
260–500 mA
4775 K
0.4
440 mA
CIE y coordinate
5411 K
370 mA
6557 K
320 mA
0.3
8242 K
260 mA 12 431 K
0.26 0.28 0.30 0.32 0.34 0.36

CIE x coordinate
(b)
Figure 7.13 CIE chromaticity coordinates and corresponding CCT at various injection currents
(a) I from 40 to 240 mA and (b) I from 260 to 500 mA. Source: Adapted from Yadav et al. 2018
[101].
Figure 7.14 Evolution of CRI 80

with increasing current. (Inset: CRI ≥ 67
EL spectrum for CRI 67 with a 70
maximum of 67.3 at 335 mA.) 60
Source: Adapted from Yadav
CRI (Ra)
et al. 2018 [101]. 50
310 mA
40 20
Intensity (a.u.)
330 mA
15
30 10
5
20
400 500 600
10 Wavelength (nm)
0
0.1 1
Current (A)
contribution at longer wavelengths (Figure 7.10) >600 nm improves the visible

spectral region coverage, thus improving the CRI values to >60 with a maximum
of 67.3 at 335 mA ever reported for such devices (Figure 7.14).
Although further increase in current broadens the green peak, a blue shift
of 4 nm in peak wavelength is observed for the current is increased to 500 mA
from 400 mA. Also, this broadening is asymmetric with increased spectral
emission at shorter wavelengths. This shift can be attributed to band-filling and
screening of quantum cascade stark effect at higher currents. A red shift of 7 nm
in peak wavelength for blue emission is also observed for current >300 mA up
to 500 mA, which otherwise remains constant at 469 nm, which is generally
indicative of increase in junction temperature. The decrease in CRI for current
>350 mA (Figure 7.14) can be explained by the change in the green/blue (G/B)
spectral power density ratio and shift in the peak emission wavelength for both
blue and green QWs. This indicates that apart from broadened spectra, CRI in
such devices can be sensitive to G/B ratio because maximum CRI is obtained
for G/B ratio between 0.84 and 1. The spectrum of this region is shown in
Figure 7.14 (inset).
7.3.2 Phosphor-Covered LEDs

The most common and popular approach for white light generation is of yellow
phosphor (YAG:Ce3+ )-coated blue LED. These devices have achieved LER of
>250 lm/W [102]; however, in general, their CRI < 75 [103–105]. The low CRI
is attributed to missing red component in their emission spectrum, which also
keeps CCT ≥ 6000 K [103–105]. A theoretical study suggest that CRI can be
improved up to ∼96–98 by optimizing the emission spectra with the use of mul-
tiple, broadband phosphors while maintaining LER between 234 and 285 lm/W
[106]. Based on the discussed approach, Fukui et al. [107] demonstrated a white
light source with a CRI of 99.1 and luminous efficacy of 59 lm/W with a near-UV
(405 nm) source. They used a multilayered approach by stacking layer of red,
green, and blue phosphors to suppress cascade excitation, i.e. the overlapping of
excitation spectra of one phosphor with the PL spectra of the other one.
Nevertheless, the use of UV LEDs as a pump source requires three phosphors

to obtain a quasi-uniform spectral emission, and it is important that the number
of phosphors used are kept at a minimum, firstly because of increase in losses
due to Stokes shift with each added phosphor and second to suppress cascade
excitation. Furthermore, the difference in degradation time of each type of phos-
phor and temperature dependence of quantum efficiencies of phosphors suggests
minimizing the number of phosphors used. Moreover, using blue LED as a pump
source would improve the luminous efficacy because unlike UV, the emission
spectra of the LED is within the photopic eye sensitivity spectrum along with
the benefit of higher quantum efficiency. Thus, more research is needed to pro-
duce an excellent color rendering source by minimizing the use of phosphors and
improved efficacy [89].
Toward this, a recent theoretically study suggests an intermediate approach
of a phosphor (YAG:Ce3+ )-covered dichromatic LED to further improve CRI
from 78 with blue (𝜆 = 475 nm) LED to 91 with a dual-wavelength (𝜆1 = 475 nm,
𝜆2 = 490 nm) LED [108]. This approach has been tested experimentally by Stauss
et al. [109] with an improvement in CRI to 76 from 67 while not sacrificing
on efficiency. In this section, the approach is further exploited with the use of
only two phosphors (red and green) and dichromatic blue-cyan LED for better
light-quality parameters [110].
Spectral Analysis
The evolution of centroid wavelength and full width at half maximum (FWHM)
with current and normalized optical power for the monolithic blue-cyan (MBC)
LED is presented Figure 7.15. It can be seen from the figure that with increasing
current, the centroid wavelength shifts to the higher emission energy, i.e. a
blue shift is observed with practically negligible change in the emission spectral
Current (mA)
0.06 0.3 2 30
70
456
60
Centroid wavelength (nm)
454 50
FWHM (nm)
40
452
30
450
20
448 10
10–2 10–1 100 101
Normalized optical power (p)
Figure 7.15 Centroid wavelength and spectral FWHM of monolithic blue-cyan LED as a
function of normalized optical power (p) and current (mA). Source: Adapted from Titkov et al.
2016 [110], with the permission of John Wiley & Sons.
Figure 7.16 Emission

spectrum for monolithic I (mA)
Spectral power density (a.u.)

blue-cyan LED. Source:
190
Adapted from Titkov et al.
2016 [110], with the 150
permission of John Wiley & 110
Sons. 70
30
400 450 500 550

Wavelength (nm)
width for currents less than ∼30 mA. This observation is attributed to the
polarization-induced electric fields within the QWs [111]. With further increase
of current (see Figure 7.15), the centroid emission wavelength starts to shift
toward the lower energy side of the visible spectrum, i.e. the onset of red shift,
which becomes strong for currents greater than ∼70 mA. A pronounced increase
in spectral width must also be noticed under these operating conditions.
Investigating the spectrum under these operating conditions, as can be seen in
Figure 7.16, a few observations are made.
(a) For I < 30 mA, the blue emission peak dominates with a blue shift of peak
wavelength for both blue and cyan emissions.
(b) For 30 mA < I < 130 mA, the blue emission peak gradually saturates, whereas
the cyan emission is steadily increasing with negligible red shift of peak wave-
length compared to at I ∼ 30 mA.
(c) For I > 130 mA, spectral power decrease in blue peak along with noticeable
red shift of peak wavelength. On the other hand, although the peak emission
wavelength of cyan emission exhibits no red shift while the spectral power
keeps increasing with current.
Therefore, the simultaneous increase of spectral width and red shift of centroid
wavelength are understood in terms of increased intensity of the cyan emission
peak. Moreover, because the peak emission wavelength for both emission peaks
exhibits a negligible red shift, device self-heating can be neglected as the probable
cause for efficiency decrease. Thus, it can be concluded that cyan QW, being in
close proximity of the “green gap” region, has lower efficiency, and thus, increased
cyan intensity with current for I > 30 mA leads to the discrepancy between exper-
imental and predicted EQE.
Characteristics of pc-LED
To achieve white light emission from the MBC LED, it was covered with a
phosphor mixture. The spectral evolution of such a pc-LED as a function of
current is depicted in Figure 7.17. Comparing the emission spectrum of MBC
LED (Figure 7.16) and pc-LED (Figure 7.17), it can be seen that the blue emission
from the MBC LED gets strongly absorbed by the phosphor mixture with
Figure 7.17 Emission

Spectral power density (a.u.) I (mA) spectrum for phosphor-
covered MBC LED; black-body
150
radiation spectrum at 3400 K
110 shown as the dashed line.
70 Source: Adapted from Titkov
et al. 2016 [110], with the
30 permission of John Wiley &
Sons.
400 500 600 700 800

Wavelength (nm)
increasing current even though the blue emission is around ∼435 nm instead
of 450 nm intended for the phosphor used, thus indicating that the pump
wavelength used is in close proximity of the peak of the excitation spectra of the
phosphor mixtures. On the other hand, the cyan emission (see Figure 7.17) peak
gets stronger with current, thus not absorbed as strongly as the blue emission.
The observed selective absorption allows the total emission from the pc-LED
to cover almost whole of the visible emission spectrum and is advantageous for
achieving utmost color rendition from such devices. Figure 7.17 also depicts a
truncated black-body radiation spectrum at a color temperature of 3400 K and
the pc-LED emission spectra for similar CCT (for 70 mA < I < 190 mA). It is
evident, and as observed, that the emission spectrum of pc-LED must have a
prominent cyan emission peak with spectral magnitude greater than the blue
peak to achieve such color temperatures.
Next, the evolution of color parameters, CCT, CRI, and the chromaticity coor-
dinates of the pc-LED as a function of current are depicted in Figures 7.18 and
7.19, respectively. A variation of CCT from ∼3500 to ∼3300 K in the current range
of 1–200 mA is quiet stable. Figure 7.19 depicts a reasonably stable evolution of
chromaticity coordinates with increasing current in the range of 30–150 mA. Fur-
thermore, the LER for the device varies in the range of 282–262 lm/W with a
maximum value of 282 lm/W at 10 mA. The LER values are within the expected
range of truncated solar spectrum [112].
The CRI is determined with the basic eight (Ra(8)) and extended (Ra(14))
Munsell reference color samples. Evolution of CRI, both Ra(8) and Ra(14),
with current is depicted in Figure 7.18. Both Ra(8) and Ra(14) bear qualitative
resemblance as a function of current. The general CRI Ra(8) is around ∼96 at
low currents, which is as claimed by the vendor, because of predominantly blue
emission under these operating currents. From Figure 7.18, it can be seen that
both Ra(8) and Ra(14) increases with current, which corresponds well with the
increasing cyan emission. A maximum Ra(8) = 98.6 is achieved at 80 mA. Further
increase in current leads to steep decrease in Ra(8). Spectral broadening and red
shift in the MBC spectrum is also observed for similar current values. This indi-
cates that color rendition for such devices is tunable with current. It is important
for the white LEDs to perform with best color rendition and efficiency along
100
3500
Correlated color temperature (K)

99 CCT
Color rendering index
3450
98
97 3400
Ra(8)
96
3350
95
Ra(14)
3300
94
10–3 10–2 10–1

Current (A)
Figure 7.18 CRI and CCT determined for standard 8 Munsell samples [Ra (8)] and 14 (8
standard +6 extended) Munsell samples [Ra (14)]. Source: Adapted from Titkov et al. 2016
[110], with the permission of John Wiley & Sons.
Figure 7.19 x, y CIE 1931 0.414

chromaticity coordinates of
phosphor-covered MBC LED 0.412
as evolving with injection 10 mA
CIE chromacity coordinate – y
current. Source: Adapted 0.410

from Titkov et al. 2016 [110],
with the permission of John 0.408
Wiley & Sons.
0.406
0.404
0.402
0.400
150 mA
0.398
0.415 0.416 0.417 0.418
CIE chromacity coordinate – x
with right color temperature. A brief summary of key characteristics pertinent

to white light emission from different approaches is presented in (Table 7.1).
7.4 AlGaN Deep UV LEDs

The development of semiconductor light sources operating in the deep ultra-
violet (DUV) region, such as DUV LEDs and laser diodes (LDs), is an impor-
tant subject because these devices are required for a wide variety of applications.
Figure 7.20 gives an overview of these applications. Potential applications for
UVC and UVB lights are in sterilization, water purification, medicine and bio-
chemistry, agriculture, and as light sources for high-density optical memory. The
Table 7.1 Summarizes the CRI, luminous efficacy, CCT, and achievable luminous efficacy (at
90% LEE) for monolithic and hybrid approaches to achieve white light emission.
Summary table: Monolithic and hybrid approaches

LED-type and Max. LE Max. LE (lm/W)
technological CRI (lm/W) Max. WPE (%) {achievable with
approach Ra (8) {meas} CCT (K) {meas} LEE = 90%} —
Monolithic 67.3 19 3000–12 000 4.5 65 [101]

blue-green
MBC LED+ R/G 98.6 33 3300 13 161 [110]
phosphor (hybrid)
Blue LED+ Y 85 198.9 4000 70 203 —
phosphor (hybrid)
LE, luminous efficacy; WPE, wall plug efficiency; LEE, light extraction efficiency; MBC, monolithic
blue-cyan; R/G, red/green; CRI, color rendering index; CCT, correlated color temperature; meas,
measured.
Optical storage UV curing

DUV-DVD Regines, UV adhesives, 3D printers
Sterilization
water purification
surface disinfection
DUV light
Medical, agriculture Printing, painting

Power
Skin cure (Narrow band UVB) Ink jet printers, UV coatings

Plant disease preventation
Refrigerator Air conditioner
Water purifier
UVC UVB UVA

240 260 280 300 320 340 360
Wavelength (nm)
Figure 7.20 Potential applications of DUV LEDs and LDs.
peak wavelength at 265 nm in UVC waveband is known as sterilization effects

curve, which well matches with the absorption spectrum of DNA (deoxyribonu-
cleic acid). The wavelength between 260 and 280 nm is effective for sterilization,
water purification, and surface disinfection. UVA together with UVB and UVC
lights also have potential for curing, adhesives, printing, and coating [113, 114].
The wavelength range covered by AlGaN LEDs is from UVA to UVC.
The direct transition energy range of AlGaN covers the region from 6.2 (AlN)
to 3.4 eV [114]. AlGaN is a direct transition semiconductor having an emission
wavelength range from 200 to 360 nm. Therefore, AlGaN is considered to be the
most appropriate semiconductor to develop a DUV LED [114].
Several research groups have started the research on AlGaN-based UV
LEDs with wavelength below 360 nm, between 1996 and 1999 [115–117]. The
sub-300 nm DUV LEDs were achieved by a group at the University of South
Carolina between 2002 and 2006 [118–120]. The shortest wavelength (210 nm)
LED using an AlN emitting layer was reported in 2006 [121]. We started research
into AlGaN DUV LEDs in 1997 and reported the first efficient DUV (230 nm) PL
from AlGaN/AlN QWs [122] and a 330 nm band AlGaN-QW UV LED in 1999
[116]. We have also developed highly efficient UV LEDs by incorporating In
into AlGaN [113, 123, 124]. We demonstrated several mW power 340–350 nm
InAlGaN-QW UV LEDs on both GaN single-crystal substrates [125] and
sapphire substrates [126].
The development of 260–280 nm AlGaN DUV LEDs performed in 2005–2010
was an important step in the progress toward sterilization applications. High
IQEs in AlGaN and quaternary InAlGaN QWs were achieved in 2007 [127–129],
by developing a low-TDD AlN buffer layers on sapphire substrates utilizing a
pulse-flow growth method. EIE was significantly increased by introducing a mul-
tiquantum barrier (MQB) [130]. Wide-range emission from 222 to 351 nm was
demonstrated in AlGaN and InAlGaN LEDs [129–133]. We began to improve
LEE of UVC LEDs by introducing a transparent p-AlGaN contact layer and a
reflective p-type electrode [134–136]. We also developed commercially available
DUV LED modules to be used for sterilization in 2014 [137, 138].
Sensor electronic technology (SET) developed the first commercially available
LEDs with wavelengths ranging between 240 and 360 nm [139, 140]. They
reported a maximum EQE of 11% for a 278 nm LED in 2012 [140]. They also
conducted detailed investigations into the properties of AlGaN epilayers and
UVC LED devices [141–143].
Since 2010, many companies have started developing UVC LEDs aiming at
sterilization applications. Nikkiso has developed highly efficient UVC LEDs
[144–146] and reported EQEs of over 10% [144]. They improved the LED
properties by introducing an encapsulating resin that does not deteriorate
under UVC radiation [146]. Crystal IS developed efficient 265 nm LEDs on bulk
AlN substrates fabricated by a sublimation method [147, 148], and Tokuyama
developed UVC LEDs on a thick transparent AlN layer grown, also on bulk
AlN substrates, by hydride vapor-phase epitaxy (HVPE) [149–151]. Nichia has
developed high wall-plug efficiency (WPE) UVC LEDs [152, 153] using a lens
bonding technique [153]. Also, M. Kneissl’s group in the Technical University of
Berlin carried out a series of studies on the properties of AlGaN epilayers and
AlGaN and InAlGaN UV LEDs [114, 154–157]. The reported EQEs for AlGaN
and InAlGaN UVA–UVC LEDs are summarized in Refs. [114, 157].
In spite of continuous efforts to develop an AlGaN DUV LED, its WPE is still
as low as 3%, which is much lower than that of InGaN blue LEDs. The limited
efficiency of DUV LED is mainly due to the following three factors:
1. An IQE of AlGaN is sensitive to TDD.

2. Hole concentration of p-AlGaN is low, resulting in low carrier injection effi-

ciency (IE) and high operating voltage.
3. The LEE is quite low by the light absorption of p-GaN contact layer.
An IQE value at around 50–60% is standard after developing the low-TDD
(5 × 108 cm−2 ) AlN templates on sapphire. To obtain higher IQE more than 80%,
bulk AlN substrates have advantages. The hole concentration of p-AlGaN used
in UVC LEDs is as low as 1 × 1014 cm−3 owing to its deep acceptor levels, i.e. 240
(GaN)–590 meV (AlN). Electron overflow to the p-side layers results in the reduc-
tion of IE for UVC LEDs. Because the hole density of p-type AlGaN is not very
high, we use p-GaN for the contact layer. This results in a significant reduction in
LEE, typically to below 8%, owing to the strong absorption of DUV light.
The typical value of EQE of 280 nm DUV LEDs is approximately 5%, which is
determined by 60% IQE, 80% EIE, and 10% LEE. Further improvements in EQE
are expected. Techniques for increasing each of these efficiencies are described
in the following sections.
7.4.1 Growth of High-Quality AlN and Increasing in Internal Quantum

Efficiency (IQE)
In order to obtain low-TDD, crack-free AlN buffer layer with atomically flat sur-
face on sapphire, we introduced an “ammonia (NH3 ) pulsed-flow multilayer (ML)
growth method” [127]. Figure 7.21 shows a schematic view of the growth con-
trol method and a typical gas flow sequence using pulsed and continuous growth
modes.
First, an AlN nucleation layer and a “buried” AlN layer were deposited,
both by NH3 pulsed-flow growth. The pulsed-flow mode is effective for initial
high-quality AlN growth on sapphire because of the increased migration of the
precursor. After the growth of the first layers, we introduced a continuous-flow
mode AlN growth to reduce the surface roughness. By repeating the pulsed-
NH3 pulsed-flow growth

AIN
Migration-enhanced epitaxy
AI-rich condition = stable AI (+c) polarity
TMAI
NH3 AIN
5s 3s 5s 3s 5s
AIN
AIN
Sapphire Sapphire Sapphire Sapphire
1. Growth of 2. Burying growth with 3. Reduction of surface 4. Repeat 2 and 3

nucleation AIN layer lateral enhancement roughness with
(NH3 pulsed-flow) growth mode high-speed growth
(NH3 pulsed-flow) (Continuous flow)
Reduction of threading Crack-free thick AIN buffer

dislocation density (TDD) with atomically flat surface
Figure 7.21 Gas flow sequence and schematic view of the growth control method used for
the NH3 pulsed-flow multilayer (ML)-AlN growth technique.
LED layers
Al0.76Ga0.24N:Si
2.45 μm
Al0.88Ga0.12N:Si
5-Step
multilayer
AIN buffer
3.8 μm
Sapphire
1 μm
Figure 7.22 Cross-sectional TEM image of an AlGaN/AlN template including a five-step

ML-AlN buffer layer grown on a sapphire substrate.
and continuous-flow modes, we can obtain crack-free, thick AlN layers with
atomically flat surfaces. By maintaining Al-rich growth conditions, we can obtain
stable Al (+c) polarity, which is necessary for suppressing polarity inversion
from Al to N. The detailed growth conditions are described in Refs. [127, 131].
The FWHM of X-ray diffraction (10–12) 𝜔-scan rocking curve (XRC(10–12))
was dramatically reduced by executing the pulsed-flow mode.
Figure 7.22 shows a cross-sectional TEM image of an AlGaN/AlN template
including ML-AlN buffer layer grown on a sapphire substrate. The typical
FWHMs of the (10–12) and (0002) XRCs of the ML-AlN were approximately
330′′ and 180′′ , respectively. This was grown in a 3 in. × 2 in. MOCVD reactor
[137]. The minimum edge- and screw-type dislocation densities were below
5 × 108 and 4 × 107 cm−2 , respectively, as observed in the TEM image.
We observed a considerable increase in the DUV emission from AlGaN-QWs
by fabricating them on low TDD AlN templates [128, 129]. Figure 7.23 shows
(a) the intensity of the PL peak at 255-nm measured at RT as a function of the
FWHM of the XRC (10–12), and (b) the relationship between IQE and TDD in
DUV AlGaN-QWs investigated in Refs. [139, 159]. The PL intensity significantly
increases as the FWHM of XRC was reduced. We can see that the PL intensity
was increased by a factor of about 80 by reducing the FWHM from 1400′′ to 500′′ .
Regarding the estimation of IQE, we have obtained a reliable value by examin-
ing the excitation power density dependence of PL emission at low temperature
(4 K) and room temperature [160]. Figure 7.24 shows the edge emission inten-
sity of an AlGaN QW emitting at 270 nm as a function of the excitation power
100
λ = 255 nm
80
PL intensity (a.u.)
AIGaN-QW
on ML-AIN
IQE (%)
60
40
20
Measured at RT
0 500 1000 1500 0

107 108 109 1010
FWHM of XRC (10–12) ω –scan
TDDs (cm–2)
(a) (b)
Figure 7.23 (a) Intensity of the PL emission peak at 255 nm measured at RT as a function of the
FWHM of the XRC (10–12) and (b) the relationship between IQE and TDD in DUV AlGaN-QWs.
Source: Adapted from Shatalov et al. 2010 [139] and Yun and Hirayama 2017 [158].
LT
Figure 7.24 Edge emission intensity
V181024 of an AlGaN QW emitting at 270 nm
100 K
200 K as a function of the excitation power
100 RT density measured at 4, 100, 200, and
300 K.
IQE (%)
50
0 54%
10–3 10–2 10–1 100 101 102 103

Excitation power density (kW/cm2)
density measured at 4, 100, 200, and 300 K. As can be seen in Figure 7.24, the
emission efficiency is high when the excitation is relatively weak at low tempera-
ture (4 K). In this condition, we can assume that the nonradiative recombination
is small, and the radiative recombination emission is dominant [160]. It can also
be seen that a high efficiency can be obtained at a relatively strong excitation
power density at room temperature. The IQE at room temperature is calculated
assuming that the luminous efficiency of low-temperature (4 K) weak excitation
power density is almost 100% [160].
From Figure 7.24, the highest IQE at room temperature was estimated to be
54%. From Figure 7.23b, this IQE is a value corresponding to a TDD of about
5 × 108 cm−2 . Because the edge-type TDD of an AlN template produced in this
study is in the region of 2 × 108 to 1 × 109 cm−2 , it is considered possible to achieve

about 70% as the maximum IQE.
The quaternary alloy InAlGaN is also a strong candidate as a material for real-
izing DUV LEDs because the inclusion of In leads to efficient UV emission as
well as higher hole concentration. Just a few percent of In in AlGaN is needed
to obtain high IQE, where the increases in efficiency are due to the In segrega-
tion effect, an effect previously investigated for the ternary alloy, InGaN. We have
described the advantages of the use of InAlGaN in Refs. [113, 123, 124, 126, 129].
7.4.2 AlGaN-based UVC LEDs

We fabricated AlGaN-based DUV LEDs on low-TDD AlN templates [127–138].
Figure 7.25 shows a schematic diagram of the structure of an AlGaN-based DUV
LED fabricated on a sapphire substrate. Large compositions of Al in AlGaN were
used to obtain DUV LEDs operating at short wavelengths. The detail layer struc-
tures and device geometries of the LEDs are described in Refs. [127, 129]. The
output power was measured using a Si photodetector located behind the LED
sample, which was calibrated by measuring the luminous flux from a flip-chip
(FC) LED. The forward voltages (V F ) of the bare wafer and the FC samples were
approximately 15 and 7 V, respectively, with an injection current of 20 mA.
Figure 7.26 shows an EL spectra of the AlGaN and InAlGaN MQW LEDs mea-
sured at RT. We obtained single-peak operations for LED samples with emission
wavelength from 222 to 351 nm.
RIKEN and Panasonic have developed commercially available UVC LED mod-
ules for use in sterilization in 2014 [137, 138]. To develop commercially available
devices with constant high EQEs and long device lifetimes, the reproducibility
and uniformity of the AlN template and the AlGaN LED layer structure need to
Figure 7.25 Schematic GaN:Mg(60 nm)

structure of a typical Ni/Au electrode contact layer
AlGaN-based DUV LED
fabricated on a sapphire AI0.77Ga0.23N:Mg
substrate. (25 nm)
Ni/Au
AI0.95Ga0.5N:Mg/
AI0.77Ga0.23N:Mg
6-layer multiquantum
barrier (MQB)
AI0.62Ga0.38N(1.5 nm)/
AI0.77Ga0.23N(6 nm)
Multilayer (ML) 3-layer MQW
AIN buffer Emitting layer
Sapphire substrate n-AI0.77Ga0.23N:Si
UV output
AIGaN-QW
DUV LEDs
222 nm Pulsed
227 nm Pulsed
Normalized intensity
234 nm CW
240 nm CW
248 nm CW
255 nm CW
261 nm CW
InAIGaN-QW
DUV LED
282 nm CW
342 nm CW
351 nm CW
Measured
at RT
200 250 300 350 400 450
Wavelength (nm)
Figure 7.26 Electroluminescence (EL) spectra of fabricated AlGaN and InAlGaN MQW LEDs
with emission wavelengths between 222 and 351 nm, all measured at room temperature (RT)
with injection currents of around 50 mA.
be maintained. We achieved highly uniform ML-AlN templates on sapphire in a

3 × 2 in. MOCVD reactor. We consistently obtained FWHMs of the XRC (10–12)
of 340′′ for these templates. A 270 nm 10 mW DUV LED module was developed
for applications to sterilization. Lifetimes longer than 10 000 hours have been
achieved for devices with EQEs of 2–3% [137, 138].
7.4.3 Increasing the Light Extraction Efficiency (LEE)

Improving the LEE is particularly important for the development of AlGaN DUV
LEDs. However, increasing LEE is not so easy because of the scarcity of suitable
transparent, conducting p-type contact layers and transparent p-type electrodes,
and also the lack of highly reflective p-type electrodes applicable to UVB–UVC
range.
Figure 7.27 shows schematic diagrams of several structures designed to
improve LEE and the approximate values of LEE calculated for them [136]. In
a conventional DUV LED, the light going upward from the QWs is completely
absorbed by the p-GaN contact layer. The light going downward is reflected at
the sapphire/air interface by total internal reflection. As a result, the LEE is less
than 8%. Although we have used photonic nanostructures on the surface of the
sapphire substrate, the improvement in LEE is not sufficiently high. To improve
LEE, an introduction of a transparent contact layer and a highly reflective
p-type electrode is effective. If we use a transparent p-AlGaN contact layer and
an electrode with a reflectivity of 80%, LEEs more than 20% can be obtained.
Further improvements can be made from light-scattering effects obtained by
using an AlN buffer layer grown on a PSS. Yet more improvements can be made
by having a vertical LED with a back-surface photonic structure, which can be
realized by removing the sapphire substrate. LEEs of >70% are expected for such
LEDs, as analyzed in Ref. [158].
p-GaN (absorb) + Transparent p-AIGaN + HR electrode +

PhC backside extraction
HR electrode
P-GaN
Transparent contact layer
Emitter
PSS
AIGaN/AIN
buffer layer
Sapphire
AIN pillar
LEE = 4–8% LEE ~ 12% LEE ~ 20% LEE ~ 35% LEE > 70%
Figure 7.27 Schematic illustrations of structures designed to improve the LEE of a DUV LED
and rough estimates of the values of LEE for each structure.
We demonstrated a DUV LED with a transparent p-AlGaN contact layer and

a reflective p-type electrode [136]. We replaced the conventional Ni (20 nm)/Au
(100 nm) p-type electrode with a highly reflective Ni (1 nm)/Al (200 nm) elec-
trode [161]. Using the highly reflective Ni (1 nm)/Al (200 nm) in place of the
conventional Ni (20 nm)/Au (100 nm) electrode increases the EQE from 5% to
9% owing to the increase in LEE [161]. We also confirmed that Ni (1 nm)/Mg and
rhodium (Rh) p-electrodes are effective to increase the LEE of UVC LED [162].
Increases in LEE were demonstrated for a 275 nm UVC FC LED by using a
transparent p-type AlGaN contact layer, a Rh mirror electrode, an AlN buffer
layer grown on PSS, and an encapsulating resin. The effects of each of these
were systematically investigated [163]. Conventional and LEE enhanced type
LED structures were fabricated to investigate the effects of the aforementioned
features on LEE. Schematics of these are shown in Figure 7.28a,b, respectively.
Figure 7.29 shows a photograph of the FC LED sample [163]. An Al-coated Si
submount was used for the FC LED. The chip was encapsulated in hemispherical
lens-like by silicon resin. We confirmed almost perfect transparency for the
p-AlGaN contact layer, even the Mg doping concentration was as high as
8 × 1019 cm−3 as measured by secondary ion mass spectrometry (SIMS).
Figure 7.30a,b shows the current–light output (I–L) and current–EQE (I–EQE)
characteristics for the conventional and LEE enhanced type UVC LEDs under RT
CW operation [163]. The inset in Figure 7.30a shows the EL spectra at 20 mA.
The output power for both samples has good linearity, and the EQEs are almost
constant up to 50 mA. The output power was increased from 3.9 to 18.3 mW
at 20 mA and from 9.3 to 44.2 mW at 50 mA, both by a factor of 5, by intro-
ducing LEE-enhanced structure. These values correspond to EQEs of 4.3% and
20.3%, respectively. Thus, the EQE was substantially improved by including a
transparent p-AlGaN contact layer, a Rh mirror electrode, a PSS, and the lens-like
encapsulation [163].
To clarify the individual effects on the EQE, each structure for LEE enhance-
ment was introduced step-by-step. Table 7.2 summarizes the device structures
Ni/Au electrode Rh mirror electrode
p-GaN p-AIGaN
EBL EBL
n-electrode AIGaN MQW n-electrode AIGaN MQW
n-AIGaN n-AIGaN
AIN template AIN template
Saapire substrate Patterned sapphire substrate
(a) (b)
Figure 7.28 Schematics of (a) conventional and (b) LEE-enhanced UV–LED structures. In the
LEE-enhanced UV–LED structure, we introduced a transparent p-type AlGaN:Mg contact layer,
a Rh mirror electrode, a PSS, and an encapsulating resin.
Figure 7.29 Photograph of the flip-chip (FC)

LED mounted on a Si submount with an Al
Resin
coating. The chip size is 0.5 × 0.5 mm2 and it
Si submount
is encapsulated in resin with a hemispherical
shape. The inset shows a side view of the
encapsulating resin.
LED chip
Dummy
Resin electrode
LED chip
50 30
EL intensity (a.u.)
Novel
Conventional
Output power (mW)
40
Novel Maximum EQE : 20.3%
20
EQE (%)
30 Novel
250 300 350 400
Wavelength (nm)
20
10
Conventional
10 Conventional
0 0
0 10 20 30 40 50 0 10 20 30 40 50
Current (mA) Current (mA)
(a) (b)
Figure 7.30 (a) Current–output power (I–L) and (b) current–EQE (𝜂 ext ) characteristics for the
conventional and LEE-enhanced UVC-LED. The inset in (a) shows the EL spectra of the LEDs at a
direct current of 20 mA.
Table 7.2 Summary of the device structures and their LED characteristics.
Device structures LED characteristics

p-Type
Sample contact Output power at Maximum
no. Substrate layer p electrode Geometry 20 mA (mW) EQE (%)
1 Flat GaN:Mg NiAu FC only 3.9 4.3

2 Flat AlGaN:Mg Rh FC only 11.6 12.7
3 PSS AlGaN:Mg Rh FC only 14.5 16.1
4 PSS AlGaN:Mg Rh FC + resin coating 18.3 20.3
Sample nos. 1 and 4 correspond to the conventional and novel UV–LED structures, respectively.
Sample nos. 2 and 3 demonstrate the effects of including the AlGaN:Mg/Rh layers and the PSS,
respectively.
and the LED characteristics. We found that the enhancement factors for intro-
ducing a transparent p-AlGaN contact layer and Rh electrode, PSS, and lens-like
encapsulation were approximately 3, 1.5, and 1.5, respectively [163].
The driving voltage of the LED was increased from 9 to 16 V at 20 mA by intro-
ducing a p-AlGaN contact layer. The main reason for the increase in driving
voltage is the increase of contact resistance by introducing the p-AlGaN contact
layer. Improving the conductivity of the p-AlGaN contact layer is an important
issue in future for obtaining high WPE.
To improve the LEE of UVB and UVC LEDs, the introduction of a transparent
contact layer and a highly reflective electrode is important as indicated previ-
ously. A p-AlGaN layer with high Al composition (50–70%) is used for the trans-
parent p-contact layer for UVC LEDs; however, the low hole concentration of this
layer leads to an increase in the contact resistance, resulting in a higher operating
voltage.
In order to realize both high LEE and low voltage operation in DUV LEDs,
we proposed using a highly reflective photonic crystal (HR-PhC) [164–167]. It
is possible to reflect UV light efficiently by using a 2-dimensional (2D) PhC on
the surface of the p-GaN top contact layer. We can obtain low contact resistance
because the top p-GaN layer has a high hole concentration. Therefore, a HR-PhC
fabricated on p-GaN contact layer makes it possible to achieve not only high LEE
but also high WPE in DUV LEDs [164].
Figure 7.31 shows (a) the schematic cross-sectional structures, (b)
electronic-field (E-field) mappings, and (c) calculated LEE values as a func-
tion of distance from HR-PhC and QW, calculated by using a FDTD method for
280 nm UVC LEDs. To obtain high reflectivity of UV radiation from the QW
emitting region, we set the distance between the bottom of PhC air-rod and the
QW to be approximately 60 nm [164]. We can see in Figure 7.31b that radiation
from the QWs does not penetrate into the PhC, resulting in realizing a high
reflection of radiated light. From the FDTD analysis, we found that the LEE is
increased by factors of approximately 2.8 and 1.8 at maximum by introducing
the HR-PhC into the p-GaN and p-AlGaN contact layers, respectively.
Ni/Au
100 nm
p-GaN p-AIGaN
AIGaN AIGaN
MQW MQB-EBL
n-AIGaN
Sapphire
(a) (b)
p-AIGaN contact layer

p-GaN contact layer
30
20
LEE (%)
p-AIGaN w/o PhC

10
p-GaN w/o PhC
0
50 60 70 80
Distance from QW to PhC D (nm)
(c)
Figure 7.31 (a) Schematic cross-sectional structures, (b) electronic field (E-field) mappings,
and (c) LEE values as a function of distance from HR-PhC and quantum well (QW), calculated
by using FDTD method for 280 nm UVC LEDs.
Based on these designs, we fabricated DUV LEDs with HR-PhCs on the

p-AlGaN contact layer. We used nanoimprinting and inductively coupled
plasma (ICP) dry etching to fabricate a low-damage PhC. Figure 7.32 shows
(a) a cross-sectional TEM image of the air hole of HR-PhC and (b) the I–EQE
characteristics of 283 nm AlGaN DUV LEDs with Ni/Mg electrodes (reflectivity
of >80%) with and without a HR-PhC on the transparent p-AlGaN contact layer.
The period, diameter, and depth of the air holes were 252, 100, and 64 nm,
respectively. A Ni/Mg p-type electrode was deposited via a tilted evaporation
method. The maximum EQEs of the LEDs with and without the HR-PhC were
10% and 7.9%, respectively. The introduction of the PhC increased the EQE by a
factor of 1.23, which is almost the same as obtained by FDTD simulation [164].
The air hole diameter of the PhC used in the experiment was not optimized value.
If we use a larger value of R/a, i.e. R/a = 0.4, where R and a are the radius and the
lattice constant of air hole, we would expect to obtain a significantly higher LEE.
The LEE can be further increased by adopting FC technology and encapsulation.
7.5 Summary 287
PhC air hole 10

with PhC
w/o PhC
8
63.77 nm
EQE (%)
88.16 nm 6
Intensity (a.u.)
283 nm
60 nm
4 20 mA
RT
2
260 280 300 320 340
Wavelength (nm)
2-MQB–EBL 0
50 nm 0 10 20
3-MQW
Current (mA)
(a) (b)
Figure 7.32 (a) Cross-sectional TEM image of the air hole of HR-PhC and (b) the I–EQE
characteristics of 283 nm AlGaN DUV LEDs with Ni/Mg electrodes (reflectivity of >80%) with
and without a HR-PhC on the transparent p-AlGaN contact layer.
By introducing a PhC into the contact layer and reducing the operating voltage,
it is expected that LEDs with higher WPE can be obtained.
7.5 Summary
In this chapter, we present updated investigations on the InGaN-based LEDs. A
dichromatic monolithic white LED based on InGaN/GaN MQW is presented.
The MQW-active region consists of vertically stacked two blue QWs emitting
at ∼450 nm and one green QW emitting at ∼550 nm. A CRI of 67 is achieved,
which is the highest value demonstrated till date to the best of the knowledge of
the author for such devices, i.e. phosphor-free, monolithic dichromatic MQW
LED emitting in blue and green spectral region. To improve the CRI even fur-
ther and to restrict the CCT in the warmer region of white light emission, a red
phosphor with an absorption spectrum in the blue region can be implemented, or
augmenting them with AlGaInP red LEDs will also allow to achieve much warmer
emission without compromising too much on CCT tunability.
In search for better CRI values, a novel hybrid approach, comprising of
dual-wavelength LED and two phosphors, to generate warm white light is
presented. Using this approach, a warm white light at CCT of 3400 K with an
utmost color rendering with Ra of 98.6 is demonstrated. Moreover, the color
characteristics can be tuned with operating current by adjusting the amplitude
of blue and cyan bands of emission from the epistructure.
Next, a commercial high-brightness blue LED is investigated under a wide
range of variable temperatures from 13 to 440 K. The evolution of both electrical
and optical properties under these temperatures is presented. Maximum EQE as
determined from the experimental data decreases from ∼74% at 13 K to ∼45%
at 440 K. However, for an operating current of 350 mA, EQE exhibits a weak
dependence on temperature with a drop of only 9% over the whole range of
temperatures.
Finally, commercial green LED investigated over a wide range of temperatures

and operating current and is presented. The study reveals a two-peak emission
spectrum for optical power range 10−6 –10−5 W. Beyond these power levels,
the devices emit either at ∼535 or at ∼545 nm. This two-peak behavior is also
observed for the EQE as a function of current/optical power. A simple ABC
model does not accurately fit such an EQE behavior and needs modifications, as
described in Ref. [76], to account for nonuniform properties of QWs.
We also demonstrate the technologies to develop high-efficiency AlGaN-based
DUV LEDs from the view point of increasing IQE, IE, and LEE. Significant
increases in the IQE of DUV emission have been achieved for AlGaN-QWs
by developing a low-TDD AlN layer grown on sapphire substrate. Using
this technology, 222–351 nm DUV LEDs were made. We also demonstrated
improvements in LEE by using a transparent p-AlGaN contact layer, a highly
reflective p-electrode, an AlN buffer layer on a PSS, and an encapsulating resin.
The maximum EQE obtained was 20.3% for a 275 nm UVC LED, which is the
highest EQE reported so far. We also demonstrated that an HR-PhC fabricated
on the p-contact layer increases the efficiency.
Acknowledgments
The authors, AY and ER, would like to acknowledge the financial support from
EU Seventh Framework Program (NEWLED project) with Grant No. 318388 and
EPSRC (Grant No. EP/R024898/1) for this work.
References
1 Piprek, J. (2010). Efficiency droop in nitride-based light-emitting diodes.
2 Shen, Y.C., Mueller, G.O., Watanabe, S. et al. (2007). Auger recombination in
InGaN measured by photoluminescence. Appl. Phys. Lett. 91 (14): 141101.
3 Ryu, H.Y., Kim, H.S., and Shim, J.I. (2009). Rate equation analysis of effi-
ciency droop in InGaN light-emitting diodes. Appl. Phys. Lett. 95 (2009):
081114.
4 Hader, J., Moloney, J.V., Pasenow, B. et al. (2008). On the importance of
radiative and Auger losses in GaN-based quantum wells. Appl. Phys. Lett. 92
(26): 261103.
5 Kioupakis, E., Rinke, P., Delaney, K.T., and Van de Walle, C.G. (2011).
Indirect Auger recombination as a cause of efficiency droop in nitride
light-emitting diodes. Appl. Phys. Lett. 98 (16): 161107.
6 Brendel, M., Kruse, A., Jönen, H. et al. (2011). Auger recombination in
GaInN/GaN quantum well laser structures. Appl. Phys. Lett. 99 (3): 031106.
7 Dai, Q., Shan, Q., Cho, J. et al. (2011). On the symmetry of
efficiency-versus-carrier-concentration curves in GaInN/GaN light-emitting
diodes and relation to droop-causing mechanisms. Appl. Phys. Lett. 98 (3):
033506.
References 289
8 David, A. and Grundmann, M.J. (2010). Droop in InGaN light-emitting

diodes: a differential carrier lifetime analysis. Appl. Phys. Lett. 96 (10):
103504-1–103504-3.
9 Zhang, M., Bhattacharya, P., Singh, J., and Hinckley, J. (2009). Direct mea-
surement of Auger recombination in In0.1 Ga0.9 N/GaN quantum wells and
its impact on the efficiency of In0.1 Ga0.9 N/GaN multiple quantum well light
emitting diodes. Appl. Phys. Lett. 95 (20): 201108.
10 Cao, X.A., Yang, Y., and Guo, H. (2008). On the origin of efficiency roll-off
in InGaN-based light-emitting diodes. J. Appl. Phys. 104 (9): 093108.
11 Shatalov, M., Yang, J., Sun, W. et al. (2009). Efficiency of light emission
in high aluminum content AlGaN quantum wells. J. Appl. Phys. 105 (7):
073103.
12 Sugahara, T., Sato, H., Hao, M. et al. (1998). Direct evidence that disloca-
tions are non-radiative recombination centers in GaN. Jpn. J. Appl. Phys. 37
(4): L398–L400.
13 Wang, C.H., Chang, S.P., Chang, W.T. et al. (2010). Efficiency droop alle-
viation in InGaN/GaN light-emitting diodes by graded-thickness multiple
quantum wells. Appl. Phys. Lett. 97 (18): 181101.
14 Watson-Parris, D., Godfrey, M.J., Dawson, P. et al. (2011). Carrier localiza-
tion mechanisms in Inx Ga1−x N/GaN quantum wells. Phys. Rev. B: Condens.
Matter Mater. Phys. 83 (11): 1–7.
15 Dhar, S., Jahn, U., Brandt, O. et al. (2002). Effect of exciton localization
on the quantum efficiency of GaN/(In,Ga)N multiple quantum wells. Phys.
Status Solidi A 192 (1): 85–90.
16 Graham, D.M., Soltani-Vala, A., Dawson, P. et al. (2005). Optical and
microstructural studies of InGaN/GaN single-quantum-well structures.
J. Appl. Phys. 97 (10): 103508.
17 Grandjean, N., Damilano, B., and Massies, J. (2001). Group-III nitride quan-
tum heterostructures grown by molecular beam epitaxy. J. Phys. Condens.
Matter 13 (32): 6945–6960.
18 Narayan, J., Wang, H., Ye, J. et al. (2002). Effect of thickness variation in
high-efficiency InGaN/GaN light-emitting diodes. Appl. Phys. Lett. 81 (5):
841–843.
19 Cheng, Y.C., Lin, E.C., Wu, C.M. et al. (2004). Nanostructures and carrier
localization behaviors of green-luminescence InGaN/GaN quantum-well
structures of various silicon-doping conditions. Appl. Phys. Lett. 84 (14):
2506–2508.
20 Gerthsen, D., Hahn, E., Neubauer, B. et al. (2000). Composition fluctuations
in InGaN analyzed by transmission electron microscopy. Phys. Status Solidi
A 177 (1): 145–155.
21 Bellaiche, L., Mattila, T., Wang, L.W. et al. (1999). Resonant hole localization
and anomalous optical bowing in InGaN alloys. Appl. Phys. Lett. 74 (13):
1842–1844.
22 Nguyen, D.P., Regnault, N., Ferreira, R., and Bastard, G. (2004). Alloy
effects in Ga1−x Inx N/GaN heterostructures. Solid State Commun. 130 (11):
751–754.
23 Hangleiter, A., Netzel, C., Fuhrmann, D. et al. (2007). Anti-localization sup-

presses non-radiative recombination in GaInN/GaN quantum wells. Philos.
Mag. 87 (13): 2041–2065.
24 Hader, J., Moloney, J.V., and Koch, S.W. (2010). Density-activated defect
recombination as a possible explanation for the efficiency droop in
GaN-based diodes. Appl. Phys. Lett. 96 (22): 221106.
25 Kaneta, A., Funato, M., and Kawakami, Y. (2008). Nanoscopic recombination
processes in InGaN/GaN quantum wells emitting violet, blue, and green
spectra. Phys. Rev. B 78 (12): 125317.
26 Monemar, B. and Sernelius, B.E. (2007). Defect related issues in the "cur-
rent roll-off" in InGaN based light emitting diodes. Appl. Phys. Lett. 91 (18):
181101–181103.
27 Hammersley, S., Badcock, T.J., Watson-Parris, D. et al. (2011). Study of
efficiency droop and carrier localisation in an InGaN/GaN quantum well
structure. Phys. Status Solidi C 8 (7–8): 2194–2196.
28 Schubert, M.F., Chhajed, S., Kim, J.K. et al. (2007). Effect of dislocation den-
sity on efficiency droop in GaInN/GaN light-emitting diodes. Appl. Phys.
Lett. 91 (23): 231114.
29 Smeeton, T.M., Kappers, M.J., Barnard, J.S. et al. (2003). Analysis of
InGaN/GaN single quantum wells by X-ray scattering and transmission
electron microscopy. Phys. Status Solidi B 240 (2): 297–300.
30 Galtrey, M.J., Oliver, R.A., Kappers, M.J. et al. (2007). Three-dimensional
atom probe studies of an Inx Ga1−x N/GaN multiple quantum well structure:
assessment of possible indium clustering. Appl. Phys. Lett. 90 (6): 061903.
31 Chang, L.B., Lai, M.J., Lin, R.M., and Huang, C.H. (2011). Effect of electron
leakage on efficiency droop in wide-well InGaN-based light-emitting diodes.
32 Vampola, K.J., Iza, M., Keller, S. et al. (2009). Measurement of electron over-
flow in 450 nm InGaN light-emitting diode structures. Appl. Phys. Lett. 94
(6): 061116.
33 Piprek, J., Farrell, R., DenBaars, S., and Nakamura, S. (2006). Effects of
built-in polarization on InGaN-GaN vertical-cavity surface-emitting lasers.
IEEE Photonics Technol. Lett. 18 (1): 7–9.
34 Kim, M.H., Schubert, M.F., Dai, Q. et al. (2007). Origin of efficiency droop in
GaN-based light-emitting diodes. Appl. Phys. Lett. 91 (18): 183507.
35 Liu, J.P., Ryou, J.H., Dupuis, R.D. et al. (2008). Barrier effect on hole trans-
port and carrier distribution in InGaN/GaN multiple quantum well visible
36 Xie, J., Ni, X., Fan, Q. et al. (2008). On the efficiency droop in InGaN multi-
ple quantum well blue light emitting diodes and its reduction with p-doped
quantum well barriers. Appl. Phys. Lett. 93 (12): 121107.
37 Verzellesi, G., Saguatti, D., Meneghini, M. et al. (2013). Efficiency droop in
InGaN/GaN blue light-emitting diodes: physical mechanisms and remedies.
J. Appl. Phys. 114 (7): 071101.
38 Götz, W., Johnson, N.M., Chen, C. et al. (1996). Activation energies of Si
donors in GaN. Appl. Phys. Lett. 68 (22): 3144–3146.
References 291
39 Nam, K.B., Nakarmi, M.L., Li, J. et al. (2003). Mg acceptor level in AlN
probed by deep ultraviolet photoluminescence. Appl. Phys. Lett. 83 (5):
878–880.
40 Hwang, S., Jin Ha, W., Kyu Kim, J. et al. (2011). Promotion of hole injection
enabled by GaInN/GaN light-emitting triodes and its effect on the efficiency
droop. Appl. Phys. Lett. 99 (18): 181115.
41 Meyaard, D.S., Shan, Q., Dai, Q. et al. (2011). On the temperature
dependence of electron leakage from the active region of GaInN/GaN
42 Sano, T., Doi, T., Inada, S.A. et al. (2013). High internal quantum efficiency
blue-green light-emitting diode with small efficiency droop fabricated on low
dislocation density GaN substrate. Jpn. J. Appl. Phys. 52 (8S): 08JK09.
43 Bulashevich, K.A. and Karpov, S.Y. (2008). Is Auger recombination respon-
sible for the efficiency rollover in III-nitride light-emitting diodes? Phys.
Status Solidi C 5 (6): 2066–2069.
44 Laubsch, A., Sabathil, M., Bergbauer, W. et al. (2009). On the origin of
IQE-’droop’ in InGaN LEDs. Phys. Status Solidi C 6 (S2): S913–S916.
45 Binder, M., Nirschl, A., Zeisel, R. et al. (2013). Identification of nnp and npp
Auger recombination as significant contributor to the efficiency droop in
(GaIn)N quantum wells by visualization of hot carriers in photolumines-
cence. Appl. Phys. Lett. 103 (7): 071108.
46 Galler, B., Lugauer, H.J., Binder, M. et al. (2013). Experimental determina-
tion of the dominant type of Auger recombination in InGaN quantum wells.
droop. Phys. Rev. Lett. 110 (17): 177406.
48 Meyaard, D.S., Lin, G.B., Shan, Q. et al. (2011). Asymmetry of carrier trans-
port leading to efficiency droop in GaInN based light-emitting diodes. Appl.
Phys. Lett. 99 (25): 251115.
49 Xu, J., Schubert, M.F., Noemaun, A.N. et al. (2009). Reduction in effi-
ciency droop, forward voltage, ideality factor, and wavelength shift in
polarization-matched GaInN/GaInN multi-quantum-well light-emitting
50 Jong-In, S., Hyungsung, K., Dong-Soo, S., and Han-Youl, Y. (2011). An expla-
nation of efficiency droop in InGaN-based light emitting diodes: saturated
radiative recombination rate at randomly distributed in-rich active areas.
J. Korean Phys. Soc. 58 (3): 503.
51 Rozhansky, I.V. and Zakheim, D.A. (2006). Analysis of the causes
of the decrease in the electroluminescence efficiency of AlGaInN
light-emitting-diode heterostructures at high pumping density. Semicon-
ductors 40 (7): 839–845.
52 Han, S.H., Lee, D.Y., Shim, H.W. et al. (2010). Improvement of efficiency
droop in InGaN/GaN multiple quantum well light-emitting diodes with
trapezoidal wells. J. Phys. D: Appl. Phys. 43 (35): 354004.
53 Kuo, Y.K., Tsai, M.C., Yen, S.H. et al. (2010). Effect of p-type last barrier
on efficiency droop of blue InGaN light-emitting diodes. IEEE J. Quantum
Electron. 46 (8): 1214–1220.
54 Tu, P.M., Chang, C.Y., Huang, S.C. et al. (2011). Investigation of efficiency
droop for InGaN-based UV light-emitting diodes with InAlGaN barrier.
Appl. Phys. Lett. 98 (21): 211107.
55 Yen, S.H., Tsai, M.C., Tsai, M.L. et al. (2009). Effect of n-type AlGaN layer
on carrier transportation and efficiency droop of blue InGaN light-emitting
diodes. IEEE Photonics Technol. Lett. 21 (14): 975–977.
56 Zhang, Y.Y. and Yao, G.R. (2011). Performance enhancement of blue
light-emitting diodes with AlGaN barriers and a special designed
electron-blocking layer. J. Appl. Phys. 110 (9): 093104.
57 Kuo, Y.K., Chang, J.Y., and Tsai, M.C. (2010). Enhancement in hole-injection
efficiency of blue InGaN light-emitting diodes from reduced polarization
by some specific designs for the electron blocking layer. Opt. Lett. 35 (19):
3285.
58 Schubert, M.F. and Schubert, E.F. (2010). Effect of heterointerface polariza-
tion charges and well width upon capture and dwell time for electrons and
holes above GaInN/GaN quantum wells. Appl. Phys. Lett. 96 (13): 131102.
59 Xu, J., Schubert, M.F., Zhu, D. et al. (2011). Effects of polarization-field
tuning in GaInN light-emitting diodes. Appl. Phys. Lett. 99 (4): 041105.
60 Maier, M., Köhler, K., Kunzer, M. et al. (2009). Reduced nonthermal rollover
of wide-well GaInN light-emitting diodes. Appl. Phys. Lett. 94 (4): 041103.
61 Zakheim, D.A., Pavluchenko, A.S., and Bauman, D.A. (2011). Blue
LEDs – way to overcome efficiency droop. Phys. Status Solidi C 8 (7–8):
2340–2344.
62 Zhmakin, A. (2011). Enhancement of light extraction from light emitting
diodes. Phys. Rep. 498 (4–5): 189–241.
63 Hangleiter, A., Fuhrmann, D., Grewe, M. et al. (2004). Towards understand-
ing the emission efficiency of nitride quantum wells. Phys. Status Solidi A
201 (12): 2808–2813.
64 Watanabe, S., Yamada, N., Nagashima, M. et al. (2003). Internal quantum
efficiency of highly-efficient Inx Ga1−x N-based near-ultraviolet light-emitting
65 Chen, G., Craven, M., Kim, A. et al. (2008). Performance of high-power
III-nitride light emitting diodes. Phys. Status Solidi A 205 (5): 1086–1092.
66 Peter, M., Laubsch, A., Bergbauer, W. et al. (2009). New developments in
green LEDs. Phys. Status Solidi A 206 (6): 1125–1129.
67 Titkov, I.E., Karpov, S.Y., Yadav, A. et al. (2014). Temperature-dependent
internal quantum efficiency of blue high-brightness light-emitting diodes.
IEEE J. Quantum Electron. 50 (11): 911–920.
68 Fujiwara, K., Jimi, H., and Kaneda, K. (2009). Temperature-dependent droop
of electroluminescence efficiency in blue (In,Ga)N quantum-well diodes.
Phys. Status Solidi C 6 (S2): S814–S817.
69 Kivisaari, P., Riuttanen, L., Oksanen, J. et al. (2012). Electrical measure-
ment of internal quantum efficiency and extraction efficiency of III-N
References 293
70 Lin, G.B., Shan, Q., Birkel, A.J. et al. (2012). Method for determining
the radiative efficiency of GaInN quantum wells based on the width of
efficiency-versus-carrier-concentration curve. Appl. Phys. Lett. 101 (24):
241104.
71 Broell, M., Sundgren, P., Rudolph, A. et al. (2014). New developments on
high-efficiency infrared and InGaAlP light-emitting diodes at OSRAM Opto
Semiconductors. Proceedings of SPIE 9003, Light-Emitting Diodes: Materials,
Devices, and Applications for Solid State Lighting XVIII, San Francisco, CA,
90030L (27 February 2014), https://doi.org/10.1117/12.2039078.
72 Wang, J., Wang, L., Wang, L. et al. (2012). An improved carrier rate model
to evaluate internal quantum efficiency and analyze efficiency droop origin
of InGaN based light-emitting diodes. J. Appl. Phys. 112 (2): 023107.
73 Kisin, M.V. and El-Ghoroury, H.S. (2015). Inhomogeneous injection in
III-nitride light emitters with deep multiple quantum wells. J. Comput.
Electron. 14 (2): 432–443.
74 Karpov, S.Y. (2010). Effect of localized states on internal quantum efficiency
of III-nitride LEDs. Phys. Status Solidi RRL 4 (11): 320–322.
75 Lin, G.B., Meyaard, D., Cho, J. et al. (2012). Analytic model for the efficiency
droop in semiconductors with asymmetric carrier-transport properties based
on drift-induced reduction of injection efficiency. Appl. Phys. Lett. 100 (16):
161106.
76 Titkov, I., Karpov, S., Yadav, A. et al. (2017). Efficiency of true-green light
emitting diodes: non-uniformity and temperature effects. Materials 10 (11):
1323.
77 Crawford, M. (2009). LEDs for solid-state lighting: performance challenges
and recent advances. IEEE J. Sel. Top. Quantum Electron. 15 (4): 1028–1040.
78 Lundin, W.V., Nikolaev, A.E., Sakharov, A.V. et al. (2010). High-efficiency
InGaN/GaN/AlGaN light-emitting diodes with short-period InGaN/GaN
superlattice for 530–560 nm range. Tech. Phys. Lett. 36 (11): 1066–1068.
79 Chang, S., Wu, L., Su, Y. et al. (2003). Si and Zn co-doped InGaN-GaN
white light-emitting diodes. IEEE Trans. Electron Devices 50 (2): 519–521.
80 Nizamoglu, S., Ozel, T., Sari, E., and Demir, H.V. (2007). White light gener-
ation using CdSe/ZnS core–shell nanocrystals hybridized with InGaN/GaN
light emitting diodes. Nanotechnology 18 (6): 065709.
81 Chhajed, S., Xi, Y., Gessmann, T. et al. (2005). Junction temperature in
light-emitting diodes assessed by different methods. In: Proceedings Volume
5739, Light-Emitting Diodes: Research, Manufacturing, and Applications IX,
16–24. San Jose, CA, United States: SPIE.
82 Damilano, B., Grandjean, N., Pernot, C., and Massies, J. (2001). Monolithic
white light emitting diodes based on InGaN/GaN multiple-quantum wells.
Jpn. J. Appl. Phys. 40 (Part 2, No. 9A/B): L918–L920.
83 Kim, T., Kim, J., Yang, M. et al. (2013). Polychromatic white LED using GaN
nano pyramid structure. In: Proceedings Volume 8641, Light-Emitting Diodes:
Materials, Devices, and Applications for Solid State Lighting XVII; 86410E
(eds. K.P. Streubel, H. Jeon, L.W. Tu and M. Strassburg). San Francisco, CA,
USA: SPIE.
84 Sadaf, J.R., Israr, M.Q.M., Kishwar, S. et al. (2010). White electrolumines-

cence using ZnO nanotubes/GaN heterostructure light-emitting diode.
Nanoscale Res. Lett. 5 (6): 957–960.
85 Damilano, B., Trad, N., Brault, J. et al. (2012). Color control in monolithic
white light emitting diodes using a (Ga,In)N/GaN multiple quantum well
light converter. Phys. Status Solidi A 209 (3): 465–468.
86 Yu, C., Lirong, H., and Shanshan, Z. (2009). Monolithic white LED based on
Alx Ga1−x N/Iny Ga1−y N DBR resonant-cavity. J. Semicond. 30 (1): 014005.
87 Schubert, E.F. (2005). Solid-state light sources getting smart. Science 308
(5726): 1274–1278.
88 Schubert, E.F. (2006). Light-Emitting Diodes, 2e. Cambridge: Cambridge Uni-
versity Press.
89 Tsao, J.Y., Crawford, M.H., Coltrin, M.E. et al. (2014). Toward smart and
ultra-efficient solid-state lighting. Adv. Opt. Mater. 2 (9): 809–836.
90 Lu, C.-F., Huang, C.-F., Chen, Y.-S. et al. (2009). Phosphor-free monolithic
white-light LED. IEEE J. Sel. Top. Quantum Electron. 15 (4): 1210–1217.
91 Tsatsulnikov, A.F., Lundin, W.V., Sakharov, A.V. et al. (2012). Effect of stim-
ulated phase separation on properties of blue, green and monolithic white
LEDs. Phys. Status Solidi C 9 (3–4): 774–777.
92 Yamada, M., Narukawa, Y., and Mukai, T. (2002). Phosphor free
high-luminous-efficiency white light-emitting diodes composed of InGaN
multi-quantum well. Jpn. J. Appl. Phys. 41 (Part 2, No. 3A): L246–L248.
93 Funato, M., Hayashi, K., Ueda, M. et al. (2008). Emission color tunable
light-emitting diodes composed of InGaN multifacet quantum wells. Appl.
Phys. Lett. 93 (2): 19–22.
94 Nguyen, H.P.T., Zhang, S., Cui, K. et al. (2011). p-Type modulation doped
InGaN/GaN dot-in-a-wire white-light-emitting diodes monolithically grown
on Si(111). Nano Lett. 11 (5): 1919–1924.
95 Li, H., Li, P., Kang, J. et al. (2013). Phosphor-free, color-tunable monolithic
InGaN light-emitting diodes. Appl. Phys. Express 6 (10): 102103.
96 Tan, C.K. and Tansu, N. (2015). Nanostructured lasers: electrons and holes
get closer. Nat. Nanotechnol. 10 (2): 107–109.
97 Damilano, B., Demolon, P., Brault, J. et al. (2010). Blue-green and white color
tuning of monolithic light emitting diodes. J. Appl. Phys. 108 (7): 073115.
98 Titkov, I.E., Yadav, A., Zerova, V.L. et al. (2014). Internal quantum efficiency
and tunable colour temperature in monolithic white InGaN/GaN LED. In:
Proceedings Volume 8986, Gallium Nitride Materials and Devices IX; 89862A
SPIE OPTO (eds. J.I. Chyi, Y. Nanishi, H. Morkoç, et al.). International
Society for Optics and Photonics.
99 Karpov, S.Y., Cherkashin, N.A., Lundin, W.V. et al. (2016). Multi-color
monolithic III-nitride light-emitting diodes: factors controlling emission
spectra and efficiency. Phys. Status Solidi A 213 (1): 19–29.
100 Tsatsulnikov, A.F., Lundin, W.V., Sakharov, A.V. et al. (2010). A mono-
lithic white LED with an active region based on InGaN QWs separated by
short-period InGaN/GaN superlattices. Semiconductors 44 (6): 808–811.
References 295
101 Yadav, A., Titkov, I., Sakharov, A. et al. (2018). Di-chromatic InGaN based
color tuneable monolithic LED with high color rendering index. Appl. Sci. 8
(7): 1158.
102 Narukawa, Y., Ichikawa, M., Sanga, D. et al. (2010). White light emitting
diodes with super-high luminous efficacy. J. Phys. D: Appl. Phys. 43 (35):
354002.
103 Krames, M.R., Shchekin, O.B., Mueller-Mach, R. et al. (2007). Status and
future of high-power light-emitting diodes for solid-state lighting. J. Disp.
Technol. 3 (2): 160–175.
104 Nakamura, S. (1997). Present performance of InGaN-based
blue/green/yellow LEDs. In: Proceedings Volume 3002, Light-Emitting Diodes:
Research, Manufacturing, and Applications (ed. E.F. Schubert), 26–35.
105 Setlur, A. (2009). Phosphors for LED-based solid-state lighting. Electrochem.
Soc. Interface 16 (4): 32.
106 Žukauskas, A., Vaicekauskas, R., Ivanauskas, F. et al. (2008). Špectral opti-
mization of phosphor-conversion light-emitting diodes for ultimate color
rendering. Appl. Phys. Lett. 93 (5): 051115.
107 Fukui, T., Kamon, K., Takeshita, J. et al. (2009). Superior illuminant charac-
teristics of color rendering and luminous efficacy in multilayered phosphor
conversion white light sources excited by near-ultraviolet light-emitting
diodes. Jpn. J. Appl. Phys. 48 (11): 112101.
108 Mirhosseini, R., Schubert, M.F., Chhajed, S. et al. (2009). Improved color
rendering and luminous efficacy in phosphor-converted white light-emitting
diodes by use of dual-blue emitting active regions. Opt. Express 17 (13):
10806–10813.
109 Stauss, P., Mandl, M., Rode, P. et al. (2011). Monolitically grown dual
wavelength InGaN LEDs for improved CRI. Phys. Status Solidi C 8 (7–8):
2396–2398.
110 Titkov, I.E., Yadav, A., Karpov, S.Y. et al. (2016). Superior color rendering
with a phosphor-converted blue-cyan monolithic light-emitting diode. Laser
Photonics Rev. 10 (6): 1031–1038.
111 Young, N.G., Farrell, R.M., Oh, S. et al. (2016). Polarization field screening
in thick (0001) InGaN/GaN single quantum well light-emitting diodes. Appl.
Phys. Lett. 108 (6): 1–6.
112 Murphy, T.W. (2012). Maximum spectral luminous efficacy of white light.
J. Appl. Phys. 111 (10): 104909.
113 Hirayama, H. (2005). Quaternary InAlGaN-based high-efficiency ultraviolet
light-emitting diodes. J. Appl. Phys. 97 (9): 091101.
114 Kneissl, M. and Rass, J. (eds.) (2016). III-Nitride Ultraviolet Emitters,
Springer Series in Materials Science, vol. 227. Cham: Springer International
Publishing.
115 Han, J., Crawford, M.H., Shul, R.J. et al. (1998). AlGaN/GaN quantum well
ultraviolet light emitting diodes. Appl. Phys. Lett. 73 (12): 1688–1690.
116 Kinoshita, A., Hirayama, H., Ainoya, M. et al. (2000). Room-temperature
operation at 333 nm of Al0.03 Ga0.97 N/Al0.25 Ga0.75 N quantum-well
light-emitting diodes with Mg-doped superlattice layers. Appl. Phys. Lett.
77 (2): 175–177.
117 Nishida, T., Saito, H., and Kobayashi, N. (2001). Efficient and high-power
AlGaN-based ultraviolet light-emitting diode grown on bulk GaN. Appl.
Phys. Lett. 79 (6): 711–712.
118 Sun, W., Adivarahan, V., Shatalov, M. et al. (2004). Continuous wave mil-
liwatt power AlGaN light emitting diodes at 280 nm. Jpn. J. Appl. Phys. 43
(No. 11A): L1419–L1421.
119 Adivarahan, V., Wu, S., Zhang, J.P. et al. (2004). High-efficiency 269 nm
emission deep ultraviolet light-emitting diodes. Appl. Phys. Lett. 84 (23):
4762–4764.
120 Adivarahan, V., Sun, W.H., Chitnis, A. et al. (2004). 250 nm AlGaN
light-emitting diodes. Appl. Phys. Lett. 85 (12): 2175–2177.
121 Taniyasu, Y., Kasu, M., and Makimoto, T. (2006). An aluminium nitride
light-emitting diode with a wavelength of 210 nanometres. Nature 441
(7091): 325–328.
122 Hirayama, H., Enomoto, Y., Kinoshita, A. et al. (2002). Efficient 230–280 nm
emission from high-Al-content AlGaN-based multiquantum wells. Appl.
Phys. Lett. 80 (1): 37–39.
123 Hirayama, H., Kinoshita, A., Yamabi, T. et al. (2002). Marked enhancement
of 320–360 nm ultraviolet emission in quaternary Inx Aly Ga1−x−y N with
In-segregation effect. Appl. Phys. Lett. 80 (2): 207–209.
124 Hirayama, H., Enomoto, Y., Kinoshita, A. et al. (2002). Room-temperature
intense 320 nm band ultraviolet emission from quaternary InAlGaN-based
multiple-quantum wells. Appl. Phys. Lett. 80 (9): 1589–1591.
125 Hirayama, H., Akita, K., Kyono, T. et al. (2004). High-efficiency 352 nm qua-
ternary InAlGaN-based ultraviolet light-emitting diodes grown on GaN sub-
strates. Jpn. J. Appl. Phys. 43 (No. 10A): L1241–L1243.
126 Fujikawa, S., Takano, T., Kondo, Y., and Hirayama, H. (2008). Realization of
340-nm-band high-output-power (>7 mW) InAlGaN quantum well ultra-
violet light-emitting diode with p-type InAlGaN. Jpn. J. Appl. Phys. 47 (4):
2941–2944.
127 Hirayama, H., Yatabe, T., Noguchi, N. et al. (2007). 231–261 nm AlGaN
deep-ultraviolet light-emitting diodes fabricated on AlN multilayer buffers
grown by ammonia pulse-flow method on sapphire. Appl. Phys. Lett. 91 (7):
071901.
128 Hirayama, H., Yatabe, T., Ohashi, T., and Kamata, N. (2008). Remarkable
enhancement of 254–280 nm deep ultraviolet emission from AlGaN quan-
tum wells by using high-quality AlN buffer on sapphire. Phys. Status Solidi C
5 (6): 2283–2285.
129 Hirayama, H., Fujikawa, S., Noguchi, N. et al. (2009). 222–282 nm AlGaN
and InAlGaN-based deep-UV LEDs fabricated on high-quality AlN on
sapphire. Phys. Status Solidi A 206 (6): 1176–1182.
130 Hirayama, H., Tsukada, Y., Maeda, T., and Kamata, N. (2010). Marked
enhancement in the efficiency of deep-ultraviolet AlGaN light-emitting
diodes by using a multiquantum-barrier electron blocking layer. Appl. Phys.
Express 3 (3): 031002.
131 Hirayama, H., Noguchi, N., Yatabe, T., and Kamata, N. (2008). 227 nm
AlGaN light-emitting diode with 0.15 mW output power realized using
References 297
a thin quantum well and AlN buffer with reduced threading dislocation
density. Appl. Phys. Express 1: 051101.
132 Hirayama, H., Noguchi, N., and Kamata, N. (2010). 222 nm deep-ultraviolet
AlGaN quantum well light-emitting diode with vertical emission properties.
133 Fujikawa, S., Hirayama, H., and Maeda, N. (2012). High-efficiency AlGaN
deep-UV LEDs fabricated on a- and m-axis oriented c-plane sapphire sub-
strates. Phys. Status Solidi C 9 (3–4): 790–793.
134 Maeda, N. and Hirayama, H. (2013). Realization of high-efficiency deep-UV
LEDs using transparent p-AlGaN contact layer. Phys. Status Solidi C 10 (11):
1521–1524.
135 Hirayama, H., Maeda, N., Fujikawa, S. et al. (2014). Recent progress
and future prospects of AlGaN-based high-efficiency deep-ultraviolet
light-emitting diodes. Jpn. J. Appl. Phys. 53 (10): 100209.
136 Hirayama, H., Maeda, N., Fujikawa, S. et al. (2014). Development of AlGaN
deep-UV LEDs with high light-extraction efficiency by introducing transpar-
ent layer structure. Optronics 33: 58.
137 Mino, T., Hirayama, H., Takano, T. et al. (2012). Highly-uniform
260 nm-band AlGaN-based deep-ultraviolet light-emitting diodes developed
by 2-inch×3 MOVPE system. Phys. Status Solidi C 9 (3–4): 749–752.
138 Mino, T., Hirayama, H., Takano, T. et al. (2013). Development of 260 nm
band deep-ultraviolet light emitting diodes on Si substrates. In: Proceedings
Volume 8625, Gallium Nitride Materials and Devices VIII; 86251Q, SPIE
OPTO (eds. J.I. Chyi, Y. Nanishi, H. Morkoç, et al.). San Francisco, CA,
USA: SPIE.
139 Shatalov, M., Sun, W., Bilenko, Y. et al. (2010). Large chip high power deep
ultraviolet light-emitting diodes. Appl. Phys. Express 3 (6): 062101.
140 Shatalov, M., Sun, W., Lunev, A. et al. (2012). AlGaN deep-ultraviolet
light-emitting diodes with external quantum efficiency above 10%. Appl.
Phys. Express 5 (8): 082101.
141 Mickevicius, J., Tamulaitis, G., Shur, M. et al. (2012). Internal quantum effi-
ciency in AlGaN with strong carrier localization. Appl. Phys. Lett. 101 (21):
211902.
142 Moe, C.G., Garrett, G.A., Rotella, P. et al. (2012). Impact of
temperature-dependent hole injection on low-temperature electrolumi-
nescence collapse in ultraviolet light-emitting diodes. Appl. Phys. Lett. 101
(25): 253512.
143 Mickevicius, J., Tamulaitis, G., Shur, M. et al. (2013). Correlation between
carrier localization and efficiency droop in AlGaN epilayers. Appl. Phys. Lett.
103 (1): 011906.
144 Pernot, C., Kim, M., Fukahori, S. et al. (2010). Improved efficiency of
255–280 nm AlGaN-based light-emitting diodes. Appl. Phys. Express 3 (6):
061004.
145 Inazu, T., Fukahori, S., Pernot, C. et al. (2011). Improvement of light extrac-
tion efficiency for AlGaN-based deep ultraviolet light-emitting diodes. Jpn. J.
Appl. Phys. 50: 122101.
146 Yamada, K., Furusawa, Y., Nagai, S. et al. (2015). Development of underfill-
ing and encapsulation for deep-ultraviolet LEDs. Appl. Phys. Express 8 (1):
012101.
147 Grandusky, J.R., Gibb, S.R., Mendrick, M.C. et al. (2011). High output power
from 260 nm pseudomorphic ultraviolet light-emitting diodes with improved
thermal performance. Appl. Phys. Express 4 (8): 082101.
148 Grandusky, J.R., Chen, J., Gibb, S.R. et al. (2013). 270 nm pseudomorphic
ultraviolet light-emitting diodes with over 60 mW continuous wave output
power. Appl. Phys. Express 6 (3): 032101.
149 Kinoshita, T., Hironaka, K., Obata, T. et al. (2012). Deep-ultraviolet
light-emitting diodes fabricated on AlN substrates prepared by hydride
vapor phase epitaxy. Appl. Phys. Express 5 (12): 122101.
150 Kinoshita, T., Obata, T., Nagashima, T. et al. (2013). Performance and relia-
bility of deep-ultraviolet light-emitting diodes fabricated on AlN substrates
prepared by hydride vapor phase epitaxy. Appl. Phys. Express 6 (9): 092103.
151 Kinoshita, T., Obata, T., Yanagi, H., and Inoue, S.I. (2013). High p-type
conduction in high-Al content Mg-doped AlGaN. Appl. Phys. Lett. 102 (1):
012105.
152 Fujioka, A., Misaki, T., Murayama, T. et al. (2010). Improvement in output
power of 280-nm deep ultraviolet light-emitting diode by using AlGaN multi
quantum wells. Appl. Phys. Express 3 (4): 041001.
153 Ichikawa, M., Fujioka, A., Kosugi, T. et al. (2016). High-output-power deep
ultraviolet light-emitting diode assembly using direct bonding. Appl. Phys.
Express 9 (7): 072101.
154 Li, X.H., Detchprohm, T., Kao, T.T. et al. (2014). Low-threshold stimulated
emission at 249 nm and 256 nm from AlGaN-based multiple-quantum-well
lasers grown on sapphire substrates. Appl. Phys. Lett. 105 (14): 141106.
155 Mehnke, F., Kuhn, C., Stellmach, J. et al. (2015). Effect of heterostructure
design on carrier injection and emission characteristics of 295 nm light
emitting diodes. J. Appl. Phys. 117 (19): 195704.
156 Susilo, N., Hagedorn, S., Jaeger, D. et al. (2018). AlGaN-based deep UV
LEDs grown on sputtered and high temperature annealed AlN/sapphire.
Appl. Phys. Lett. 112 (4): 041110.
157 Kneissl, M., Seong, T.Y., Han, J., and Amano, H. (2019). The emergence
and prospects of deep-ultraviolet light-emitting diode technologies. Nat.
Photonics 13 (4): 233–244.
158 Yun, J. and Hirayama, H. (2017). Investigation of the light-extraction effi-
ciency in 280 nm AlGaN-based light-emitting diodes having a highly
transparent p-AlGaN layer. J. Appl. Phys. 121 (1): 013105.
159 Ban, K., Yamamoto, J.I., Takeda, K. et al. (2011). Internal quantum efficiency
of whole-composition-range AlGaN multiquantum wells. Appl. Phys. Express
4 (5): 052101.
160 Kohno, T., Sudo, Y., Yamauchi, M. et al. (2012). Internal quantum effi-
ciency and nonradiative recombination rate in InGaN-based near-ultraviolet
light-emitting diodes. Jpn. J. Appl. Phys. 51: 072102.
References 299
161 Maeda, N., Jo, M., and Hirayama, H. (2018). Improving the efficiency of
AlGaN deep-UV LEDs by using highly reflective Ni/Al p-type electrodes.
Phys. Status Solidi A 215 (8): 1700435.
162 Maeda, N., Yun, J., Jo, M., and Hirayama, H. (2018). Enhancing the
light-extraction efficiency of AlGaN deep-ultraviolet light-emitting diodes
using highly reflective Ni/Mg and Rh as p-type electrodes. Jpn. J. Appl. Phys.
57 (4S): 04FH08.
163 Takano, T., Mino, T., Sakai, J. et al. (2017). Deep-ultraviolet light-emitting
diodes with external quantum efficiency higher than 20% at 275 nm achieved
by improving light-extraction efficiency. Appl. Phys. Express 10 (3): 031002.
164 Kashima, Y., Maeda, N., Matsuura, E. et al. (2018). High external quantum
efficiency (10%) AlGaN-based deep-ultraviolet light-emitting diodes achieved
by using highly reflective photonic crystal on p-AlGaN contact layer. Appl.
Phys. Express 11 (1): 012101.
165 Kashima, Y., Matsuura, E., Kokubo, M. et al. (2015). Deep-UV LED device
and its fabrication method, Patent 5757512, Japan.
and its fabrication method. Patent 5999800, Japan.
and its fabrication method. Patent 6156898, Japan.
301
Laser Diodes Grown by Molecular Beam Epitaxy

Greg Muziol, Henryk Turski, Marcin Siekacz, Marta Sawicka, and Czeslaw
Skierbiszewski
Institute of High Pressure Physics – Polish Academy of Sciences (Unipress-PAS), Sokolowska 29/37 Warsaw
01-142, Poland
8.1 Introduction
Successful operation of laser diodes (LDs) based on the III-nitride material sys-
tem has been demonstrated in the deep-UV [1], near-UV [2], violet [3], blue
[4], and green [5, 6] spectral regions. The high efficiency [7, 8] and high relia-
bility [9, 10] have allowed for commercialization of these devices within a decade
after the first demonstration. Recent progress has led to development of LDs
with a wall plug efficiency of over 40% [11], lifetime above 20 000 hours [12], and a
maximum output power of over 7 W [13] from a single device. Thanks to the
high performance, the LDs based on III-nitrides found many applications such
as projection, displays, spot illumination, medical equipment, and spectroscopy.
Further development will lead to new applications. LDs are even considered as a
possible candidate for general lighting as a successor of light-emitting diodes
(LEDs) [14].
The vast majority of the optoelectronic devices, including LDs, are grown
by metalorganic chemical vapor deposition (MOCVD) [15, 16]. The growth
of high-quality GaN in MOCVD is conducted at a temperature of 1000–1050 ∘ C
with a V/III ratio in the range of 1000–5000. Use of ammonia and metalorganics
as precursors causes a drawback, which is related to the presence of hydrogen
during the MOCVD growth. Hydrogen is incorporated during the growth of lay-
ers doped with Mg, which is added to obtain p-type conductivity, and causes
compensation of this acceptor. Postgrowth electron irradiation [17] or thermal
annealing [18] in nitrogen atmosphere had been proposed to activate the Mg
acceptors. It was shown that hydrogen diffuses out of the grown layers at ele-
vated temperature [19]. However, the activation process is limited to structures
with uncapped Mg-doped layers [20] because hydrogen atoms do not diffuse
in n-type material [21]. This restricts the design of MOCVD-grown devices
to those without buried p-type layers.
This drawback can be addressed by using an alternative technique – molecular
beam epitaxy (MBE). In MBE, there is no hydrogen incorporation, and Mg

302 8 Laser Diodes Grown by Molecular Beam Epitaxy
acceptors provide p-type conductivity without postgrowth activation. Interest-

ingly, layers grown by MBE with a comparable quality to MOCVD are achieved
under metal-rich conditions [22, 23] with a growth temperature of 700–750 ∘ C
and a V/III ratio in the order of 0.5. It has been shown theoretically that a thin
metal adlayer opens an efficient diffusion channel for lateral adatom transport
[24, 25], leading to a smooth growth morphology. The low growth temperature
is beneficial for long-wavelength emitters because high-In-content InGaN quan-
tum well (QW) decomposes at temperatures normally used during the growth
of subsequent layers in MOCVD [26]. In MBE, two types of the nitrogen
precursors are used: ammonia or molecular nitrogen. Development of both
MBE techniques eventually led to demonstration of LDs grown by ammonia
MBE [27] and RF-plasma-assisted molecular beam epitaxy (PAMBE) [28].
Figure 8.1a presents a standard structure schematics of the LDs grown
by PAMBE. It consists of standard AlGaN claddings, AlGaN:Mg electron
blocking layer (EBL), and InGaN waveguides with an InGaN-active region [29].
The advantage of the PAMBE technique is the ability to grow thick high-indium
content InGaN waveguides. A 160 nm thick In0.08 Ga0.92 N waveguide that is
fully strained to GaN substrate has been achieved [30]. This waveguide not only
Figure 8.1 (a) Structure of the LDs grown by PAMBE with thick high indium content
waveguides. (b) LIV characteristics of a high-power LD grown by PAMBE measured at various
temperatures. (c) The lasing spectra of devices grown by PAMBE emitting from near-UV
to cyan. Source: Obtained at Institute of High Pressure Physics PAS.
8.2 III-N Growth Fundamentals by Plasma-Assisted MBE 303
enhances the optical confinement factor [31] but also can be designed to fully
eliminate the leakage of optical mode to GaN substrate [32]. The leakage of light
to GaN substrate is a detrimental effect deteriorating the optical beam quality
[33]. Figure 8.1b presents a light–current–voltage (L–I–V ) characteristics of a
wide-ridge blue LD operating at 𝜆 = 450 nm. The maximum output power was
0.5 W [34]. The spectra of LDs operating in continuous-wave (CW) mode grown
by PAMBE at the Institute of High Pressure Physics PAS are shown in Figure 8.1c
and range from near-UV [35] up to cyan [36]. Emission in green spectral region
is achieved only under pulsed operation.
This chapter will focus on characteristics of LDs grown by PAMBE. First,
in Section 8.2, an outlook on the growth mechanism of InGaN and its implication
to optical properties of QWs and LDs will be discussed. In Section 8.3, LDs
operating on excited states will be demonstrated. It will be shown that, in a
material with high piezoelectricity, the excited states in a QW can have a much
higher oscillator strength than the ground states. In Section 8.4, lifetime studies
of LDs grown by PAMBE will be presented. The type of defects responsible
for the degradation will be identified. In Section 8.5, the LDs with tunnel
junctions (TJs) will be presented. The application of TJs to stacks of devices
and DFB (distributed feedback) LDs will be demonstrated.
8.2 III-N Growth Fundamentals by Plasma-Assisted MBE

Unlike for other semiconductors, in PAMBE of III-N materials, the metal-rich
growth conditions are typically employed. The main reason for this is a rather
high decomposition rate of nitrides in vacuum [37]. This limits available
growth temperatures to a regime in which adatoms have a low mobility under
nitrogen-rich growth conditions [24]. Fortunately, mobility of adatoms can be
enhanced by formation of an efficient diffusion channel under indium or gallium
wetting layer at the crystal surface [25]. However, such an approach creates a
very restrictive growth window, and it induces the need for a precise control
of metal flux. Using insufficient metal flux results in the appearance of rough
areas, where no wetting layer was present. On the other hand, excessive metal
flux leads to formation of droplets that result in thickness and alloy composition
inhomogeneities. For other growth techniques, such as MOCVD or ammonia
MBE, higher growth temperature is used. Under such conditions, high ammonia
overpressure is important to oppose the crystal decomposition.
In PAMBE, the active nitrogen is obtained by exciting nitrogen molecules
injected into a cavity by radio frequency electromagnetic field. The big advan-
tages of such an approach are (i) lack of any foreign atoms during the growth
(e.g. hydrogen generated from ammonia) and (ii) the independence of nitrogen
flux on growth temperature (contrary to techniques using ammonia as a source
of nitrogen in which the ammonia decomposition depends strongly on growth
temperature). The disadvantage is a rather low efficiency of excitation for
nitrogen molecules. It is estimated that only about 1% of N2 molecules injected
into the growth chamber take part in the growth process. Such huge background
pressure, even though caused by inactive N2 molecules, is a significant obstacle

for ballistic transport of materials from effusion cells. This is why, for high
growth rates in PAMBE, high pumping speed capabilities are important. Despite
the general low efficiency problem with plasma sources, growth rates of more
than 8 μm/h (133 nm/min) for GaN using PAMBE have been reported [38, 39].
Recently, it has also been shown that smooth surface can be obtained in PAMBE
under nitrogen-rich conditions, however only on N-polar (000-1) GaN substrates
with a miscut angle of 2∘ or higher [40]. For the same growth conditions and
miscut angles on a Ga-polar substrate, a rough three-dimensional growth was
obtained. Similarly, nitrogen-rich growth conditions were also employed to grow
InGaN QWs on N-polar substrates [41]. A much brighter emission than that
obtained under indium-rich growth conditions was achieved [41]. Despite the
clear improvement in N-polar structures when the nitrogen-rich growth condi-
tions were used, the efficiency of such structures still fall behind structures grown
in metal-rich regime on Ga-polar substrates. This is why in the remaining part of
this chapter, only Ga-polar devices will be analyzed.
8.2.1 Role of N-Flux for Efficient InGaN QWs

It is well known that in vacuum, InN decomposes more rapidly than GaN [37,
42, 43]. Therefore, in order to incorporate indium into GaN lattice, and main-
tain the high crystal quality, different growth conditions need to be used. It has
been previously shown that indium content in InGaN layers can be increased by
two ways. The first and most commonly used approach is to decrease the growth
temperature [44, 45] to minimize crystal decomposition and enhance indium
incorporation. It is illustrated in Figure 8.2a in which a dependence of In con-
tent on the growth temperature is presented. The advantage of this approach
is that the entire InGaN composition range can be obtained if sufficiently low
growth temperature and gallium flux are used. On the other hand, for blue-green
emitters, in which only up to 20–25% indium content is needed, a significant
decrease in the optical quality of InGaN QWs with decreasing growth tempera-
ture was observed [47]. Alternatively, higher nitrogen flux can be used to increase
Figure 8.2 Indium content in InGaN layers as a function of (a) growth temperature and (b)
excess of nitrogen flux over gallium flux. Experimentally obtained compositions are
accompanied by values obtained using phenomenological equation presented in [46]. Source:
Reprinted with permission from Turski et al. [46]. Copyright 2013, Elsevier.
Figure 8.3 (a) Full width at half maximum (FWHM) for room temperature PL emission for
quantum wells grown using indicated nitrogen fluxes. (b) Semi-logarithmic dependence of
lasing wavelength for optically pumped laser structures (empty circles) and full laser diodes
(filled circles) as a function of active nitrogen flux used for the growth of its active region.
indium incorporation by contracting crystal decomposition (Figure 8.2b) [44, 46].

This approach is much more complicated from the technical point of view. As it
was already mentioned, higher active nitrogen flux needs much higher pumping
capabilities.
At the Institute of High Pressure Physics PAS, we equipped a PAMBE reac-
tor with three CTI CT10 cryogenic pumps, each characterized by a pumping
speed of 3000 l/s. This setup enabled a significant progress in accessible nitro-
gen fluxes and allows to achieve high indium content without decreasing the
growth temperature. Importantly, the high nitrogen flux improves the quality of
InGaN layers. Figure 8.3a presents the dependence of full width at half maximum
(FWHM) on emission wavelength for QWs grown with various nitrogen fluxes.
QWs with the same emission wavelength grown using higher nitrogen flux have
lower FWHM [36]. Improvement of optical quality of QWs grown with increas-
ing nitrogen flux was further confirmed by obtaining optically pumped lasers. As
presented in Figure 8.3b by empty circles, the increase of active nitrogen flux from
4 to 10 nm/min led to achievement of lasing above 500 nm [47]. Because electri-
cally pumped structures suffer from extra optical losses due to light absorption in
p-type layers, even higher active nitrogen flux was needed to push LD operation
into cyan wavelengths [36]. Filled circles in Figure 8.3b indicate the longest laser
action wavelengths obtained for electrically driven LDs as a function of the active
nitrogen flux used for the growth of their QWs.
Further development of nitride structures presented in this chapter was done
based on the QWs grown using high active nitrogen flux presented above.
8.3 Wide InGaN QWs – Beyond Quantum-Confined

Stark Effect
In this section, we will describe the properties of wide QWs grown on Ga-polar
(0001) GaN. We will demonstrate experimental data that are counterintuitive and
show that in spite of high piezoelectric fields, wide QWs can be a viable alternative
for efficient active region of LDs.
A model based on recombination through excited states will be presented to

describe the high efficiency of wide QWs [48]. Additionally, it will be shown that
application of wide InGaN QWs to LDs can increase their performance [49].
There are two reasons to use wide QWs: (i) enhancement in the optical con-
finement factor (Γ) and (ii) increase of the QW efficiency. The increase of Γ is
important because in general, it is low in III-nitride LDs and it is hard to increase
Γ via increase in the number of QWs. The reason lays in a nonuniform hole distri-
bution among QWs [50], which results in uneven optical gain [51]. The increase in
efficiency by use of wide QWs can be counterintuitive in LDs based on III-nitrides
and will be discussed more broadly below. The internal quantum efficiency (IQE)
is given by [52]:
Γeh Bn2
IQE = (8.1)
Γeh An + Γeh Bn2 + Γeh Cn3
where Γeh is the overlap between electron and hole wave functions, n is the car-
rier density, and A, B, and C are the Shockley–Read–Hall, radiative, and Auger
recombination coefficients, respectively. All of the recombination processes are
known to depend on the wave function overlap [53, 54]. Because of the fact that,
for wide QWs, the changes in wave function overlap will be on the order of mag-
nitudes, it is important to include this in this model. The phase-space filling also
influences the B and C parameters [55]; however, in this model, this effect is omit-
ted. The IQEs for In0.17 Ga0.83 N (blue color regime) and In0.30 Ga0.70 N (green color
regime) have been calculated for the ABC parameters reported in Ref. [56] and
plotted in Figure 8.4. The maximum efficiency is observed at a relatively low car-
rier density; afterward, a “droop” of efficiency occurs because of the nonradiative
Auger process becoming a significant recombination path. The LDs are normally
operated in the high carrier density regime (n > 2 × 1019 cm−3 ) [57] – far in the
“droop” regime. Therefore, any improvement in IQE and mitigation of “droop”
will decrease the threshold current of the LDs.
For conventional III-V semiconductors, the increase in the QW thickness
would decrease the carrier density, at a given current density, and thus would
decrease the part of carriers recombining through the Auger process. However,
the III-nitrides have extremely large spontaneous and piezoelectric constants
Figure 8.4 Dependence of

the internal quantum
efficiency on carrier density
for In0.17 Ga0.83 N and
In0.30 Ga0.70 N QWs. It is
important to stress that it is
independent on the
thickness of the QW. Source:
Reprinted with permission
from Muziol et al. [48].
Copyright 2019, American
Chemical Society.
[58–61]. They cause appearance of built-in electric field in heterostructures

due to lattice mismatch, which is detrimental to the optoelectronic devices.
There are two major effects of the built-in field: (i) quantum-confined Stark
effect (QCSE), which causes a red shift of the emission spectra and (ii) spatial
separation of electron and hole wave functions, which reduces the wave function
overlap [60, 62–71]. The reduction of wave function overlap leads to an increase
of the carrier density because of a lower probability of carrier recombination.
This in turn causes a larger part of the carriers to recombine through the
nonradiative Auger process causing the reduction of the quantum efficiency
[72–76]. This seems to limit the viable QW thickness in the devices. On the
other hand, the carrier density could be decreased by an increase in the number
of QWs. However, as mentioned before, the carrier population among QWs
is inhomogeneous, which will diminish the effect of carrier density reduction
[50]. The problem gets even more pronounced at longer wavelengths because of
higher In content in the QW and thus higher polarization field. This is believed
to be the primary reason for the loss of efficiency of the LEDs in the green
spectral range commonly called the “green gap” problem [77, 78].
There are a few reports in the literature studying the thick InGaN QWs on the
polar c-plane orientation [63, 66, 71, 79, 80], and although the calculations predict
a severe decrease of the transition probability [63, 71], some of the results show
an increase in the efficiency [80]. Additionally, the current state-of-the-art violet
LDs contain wide (6.6 nm thick) InGaN QWs [13]. There is clearly a discrepancy
between the theoretical understanding and the experimental results.
To explain this behavior, we need to consider what happens with the band
structure of the QW when its thickness is increased. Figure 8.5 presents the
calculated band structures of a thin (2.6 nm) and a wide (10.4 nm) QW (a) without
Figure 8.5 Calculated band structure of a thin (2.6 nm) and a wide (10.4 nm) 17% InGaN QW
(a) without and (b) with excitation. Source: Reprinted with permission from Muziol et al. [49].
Copyright 2019, The Japan Society of Applied Physics. (c) Dependence of wave function
overlap on current density. Source: Reprinted with permission from Muziol et al. [48].
Copyright 2019, American Chemical Society.
excitation and (b) with excitation - a current density of j = 1.6 kA/cm2 . The
excitation can be either optical or electrical, and it does not change the general
picture. In case of a thin (2.6 nm) QW, there is an increase of the wave function
overlap upon excitation from <e1|h1> = 0.19 up to 0.39 because of partial
screening of the piezoelectric field. It is a substantial increase; however, the
qualitative behavior of the transition does not change. On the other hand, in
the wide (10.4 nm) QW, there is a fundamental difference in the nature of the
carrier recombination. Without excitation, the wide QW is triangular just as
the thin QW. The electron and hole wave functions are separated, and their
overlap is extremely small <e1|h1> = 10−13 . However, the almost zero overlap
leads to a build-up of the carrier density upon excitation because the carriers
cannot recombine. The ongoing increase in the carrier density will stop only if a
recombination path appears. This path emerges when the piezoelectric polariza-
tion is close to be fully screened. Surprisingly, the transition path is not through
the ground states because their wave function overlap of <e1|h1> = 0.0006 is
too low to support efficient recombination. Instead, there is a highly efficient
recombination path through the excited states [48]. The overlap between the
excited states is <e2|h2> = 0.56. This is to our knowledge the first demonstration
of a peculiar QW system with a zero-probability transition between the ground
states and an extremely high one through the excited states. The reason for the
huge difference in the overlaps between the ground and excited states lays in
their localization. The ground states, despite the screening of the piezoelectric
field, are still localized in the triangular part of the QW as can be seen in
Figure 8.5b. On the other hand, the excited states have higher energies and fill
the entire width of the QW. The excited states can be thought of as exhibiting
almost a rectangular QW with only a small perturbation at the interfaces. Thus,
the overlap is almost equal to unity as for a rectangular QW. The full evolution of
wave function overlaps with current density is presented in Figure 8.5c. At very
high current densities j > 1.6 kA/cm2 , the <e2|h1> starts to have an even higher
value than <e2|h2>. This can be counterintuitive at first because it is a forbidden
transition in the case of a rectangular QW. However, in QWs with piezoelectric
sheet charges, the symmetry is broken and such transitions are allowed.
Let us now have a look at the carrier and current density required to fill the
excited states and take advantage of their high oscillator strength. The sheet
charges induced by strain at both interfaces of a 17% InGaN QW are equal to
1.7 × 1013 cm−2 . To screen them, a matching number of electron and holes need
to be introduced to the QW. The corresponding carrier density would be equal
to 5 × 1019 cm−3 , which at first looks extremely high. Such high carrier densities
are achieved only at very high current densities such as in LDs at threshold [57].
However, this is true only in the case of thin QWs. A calculation of an LED band
structure has been carried out with SiLENSe 5.4 package [81] and reveals the
interplay between the accumulation and loss of carriers. Counterintuitively, the
screening of the piezoelectric field in wide QWs occurs at relatively low current
densities. It can be easily explained using Figure 8.5c. Above a surprisingly low
current density of 16 A/cm2 , the <e2|h2> wave function overlap in the wide QW
is higher than the <e1|h1> in the thin QW. This current density is relatively low
and comparable to the operating regime of standard LEDs. The LEDs with wide
QWs can therefore take advantage of the decreased carrier density and higher
“droop” onset. Operating at higher current densities can additionally result in a
decrease of the footprint of the devices.
The spectral region in which wide QWs start to reveal their true potential cor-
responds to long wavelength at which the “green gap” problem is observed. The
standard thin QWs exhibit a large decrease of the wave function overlap even at
high excitation as presented in Figure 8.6a. On the other hand, the highest wave
function overlap for a 10.4 nm wide QW equals <e2|h2> = 0.63 and is achieved
exactly at an InGaN composition of 30% (see Figure 8.6b), which corresponds to
green emission. Surprisingly, such a value of oscillator strength is much higher
than in thin QWs with a much lower piezoelectric field. Counterintuitively, the
reason for the high oscillator strength lays in the high value of the piezoelectric
polarization itself. The higher the sheet charges generated at the interfaces, the
more rectangular the QW appears after screening. Therefore, the excited states
have a higher wave function overlap.
To show how the high wave function overlap between excited states influences
IQE, the current density needs to be linked with the carrier density through the
relation:
j = qdQW (Γeh An + Γeh Bn2 + Γeh Cn3 ) (8.2)
where q is the elementary charge and dQW is the QW thickness. It can be
derived from Eq. (8.2) that, for a given current density, the carrier density
can be decreased by an increase of QW thickness and/or the wave function
overlap. A series of four QWs are examined to show the impact of thickness
and composition on IQE: (i) 2.6 nm In0.17 Ga0.83 N, (ii) 10.4 nm In0.17 Ga0.83 N,
(iii) 2.6 nm In0.30 Ga0.70 N, and (iv) 10.4 nm In0.30 Ga0.70 N. For each QW, only
the transition with the highest value wave function overlap has been taken
into account, indicating that for the 2.6 nm In0.17 Ga0.83 N, the carrier density is
Figure 8.6 Dependence of wave function overlap at high excitation (j = 2 kA/cm2 ) on the
composition of the (a) thin 2.6 nm and (b) wide 10.4 nm QW. The ⟨e1|h1⟩ transition probability
drops with indium content in both cases, whereas the ⟨e2|h2⟩ highly increases for the wide
QW. Inset to (a) presents the band profile and carrier wave functions of a 30% InGaN QW. The
energetic distance between the h1 and h2 levels is large enough to prevent occupation of the
h2 level. Source: Reprinted with permission from Muziol et al. [48]. Copyright 2019, American
Chemical Society.
calculated assuming only the <e1|h1> transition. For the 10.4 nm In0.17 Ga0.83 N,
initially only the <e2|h2> transition is taken and is changed to the <e2|h1>
transition above j = 1.6 kA/cm2 (please compare Figure 8.5c to see that <e2|h1>
surpasses <e2|h2>). This approximation is very good in the case of the 2.6 nm
QW. In the case of the wide QW, both transitions should be present, which
would lead to a further decrease of the carrier density and an increase of the IQE.
Therefore, the calculated carrier densities would be overestimated for the wide
QW. Transitions for 2.6 and 10.4 nm In0.30 Ga0.70 N are <e1|h1> and <e2|h2> in
the whole current density range, respectively.
The calculated dependences of carrier density on current density are presented
in Figure 8.7a. When the In content is changed from 17% to 30% in the case of
the thin QW, the carrier density increases for a given current density. This will
translate into a higher part of the carriers recombining through the nonradiative
Auger process and thus a lower efficiency. In the case of the 10.4 nm thick QW,
the carrier density is much lower and is comparable for both compositions. The
calculated dependence of IQE on current density is presented in Figure 8.7b. The
results show a well-known decrease of the IQE with the current density for all
QWs (commonly referred to as “droop”) and a decrease of the IQE with In con-
tent of the QW (commonly referred to as “green gap”). Additionally, it can be seen
that the wide QWs provide a much higher IQE in the high current regime for both
cases of blue (In0.17 Ga0.83 N) and green (In0.30 Ga0.70 N) emitting LEDs. The reason
is the decreased carrier density in wide QWs. At lower carrier density, the IQE is
higher, as can be seen in Figure 8.4, because of a decrease in the part of carriers
lost to recombination through the nonradiative Auger process. The increase in
IQE coming from the use of wide QW at j = 1 kA/cm2 is 40% and 70% for the
In0.17 Ga0.83 N and the In0.30 Ga0.70 N QW, respectively. It is surprising and counter-
intuitive that the increase in IQE is higher for the higher indium content QW.
Figure 8.7 (a) Dependence of carrier density on current density for four kinds of QWs
calculated taking into account the change in active region volume and wave function overlap.
(b) Dependence of internal quantum efficiency on current density. Only one transition with
the highest wave function overlap is taken into account. In case of 10.4 nm In0.17 Ga0.83 N QW,
the dominant transition changes from <e2|h2> to <e2|h1>, which is indicated in (a) and (b).
Source: Reprinted with permission from Muziol et al. [48]. Copyright 2019, American Chemical
Society.
To test the idea of efficient radiative transition through excited states, we have
grown a set of samples with varying thickness of a 17% InGaN QW (sample
structure is presented in the inset of Figure 8.8a). Figure 8.8a presents the mea-
sured intensities of photoluminescence (PL) under CW excitation with a He–Cd
laser operated at 325 nm. A transmission electron microscopy image of one of
the wide QWs is presented in Figure 8.8b to show the sharp bottom and top
interfaces between the QW and quantum barrier. For high excitation, we have
indeed observed an increase of the PL intensity with QW width, which supports
the claim of efficient transition in wide QWs. However, an opposite trend can be
observed if the excitation power density is low. This is exactly what one would
expect if a small carrier loss mechanism is added to the model. In the case of the
wide QW, the carriers do not initially recombine because a high carrier concen-
tration in the ground state is needed to screen the piezoelectric sheet charges. In
the process of accumulation of carriers in the ground states, a part of the carri-
ers will be thermionically emitted to barriers surrounding the QW. This loss of
carriers prevents the increase in their density. We attribute the low PL intensity
from the wide QWs at low excitation to this mechanism.
Figure 8.8 (a) Dependence of photoluminescence intensity of the QW thickness for two
excitation powers. The inset shows the structure of samples. (b) TEM image of the 15.6 nm QW
showing sharp bottom and top interfaces. (c) Dependence of the full width at half maximum
of photoluminescence measured at an excitation power of 60 W/cm2 . Source: Reprinted with
permission from Muziol et al. [49]. Copyright 2019, The Japan Society of Applied Physics. (d)
Dependence of photoluminescence decay time on the QW width. The legend gives our
interpretation of the nature of observed transition. The dashed line is a guide to the eye,
showing the behavior predicted by a drop of wave function overlap with QW thickness. Source:
Reprinted with permission from Muziol et al. [48]. Copyright 2019, American Chemical Society.
Apart from the increased wave function overlap, the wide QWs also benefit
from robustness to sharpness of QW interfaces. Fluctuations in QW width have
a high impact on the FWHM of the light emitted from thin QWs. Fluctuations on
the level of one monolayer are inevitable and cause unwanted broadening. This
has a bad influence especially on LDs in which optical gain should be as narrow as
possible. Figure 8.8c presents the dependence of the measured FWHM of the PL
on the QW width. We have observed strong narrowing of the PL spectra going
from FWHM = 23 nm for a 2.6 nm thin QW down to FWHM = 14 nm above a
thickness of 15 nm.
Another proof showing the qualitative difference between the thin and wide
QW in the nature of the carrier recombination can be observed in carrier dynam-
ics. Figure 8.8d presents the decay time measured using time-resolved PL. The
initial increase in the thickness of the QW causes a giant increase in the decay
time of PL. Values as high as 3 μs were observed for the 5.2 nm QW. Such a strong
increase in PL decay time is known in systems with a high dependence of oscilla-
tor strength on the thickness of the QW. The calculated wave function overlap for
the 5.2 nm thick QW is as small as <e1|h1 > = 0.0006. However, after a certain
thickness is reached, the decay time starts to drop. In this regime, the excited
states start to play a major role in the recombination. QWs of thickness above
10 nm have a lower PL decay time than the commonly used thin, 2.6 nm thick
QWs. This is not due to an increase in the nonradiative recombination because
the intensity of emission is higher as shown in Figure 8.8a, but it is due to the
superior overlap between the excited states.
We have fabricated LDs operating in blue (𝜆 = 450 nm) and cyan (𝜆 = 490 nm)
to illustrate the influence of the thickness of the active region. In one case, the
LDs had three QWs with a thickness of 2.6–3.0 nm, and in the other case, the
active region was composed of a single 10.4 nm wide QW. The optical gain was
measured with the Hakki–Paoli method. An exemplary optical gain spectra of the
LD with a 10.4 nm thick In0.24 Ga0.76 N QW is shown in Figure 8.9a. The maximum
Figure 8.9 (a) Optical gain spectra collected for current densities ranging from 0.33 to 5.33 in
steps of 0.33 kA/cm2 . The green diamonds are maximum optical gain values used to show the
dependence of optical gain on current density. (b) Measured maximal optical gain of four LDs
with different active region design given in the legend. Solid and dashed lines are used to
extract the differential gain. The arrows indicate the increment achieved due to the use of the
wide QW. Source: Reprinted with permission from Muziol et al. [48]. Copyright 2019, American
Chemical Society.
gain at each current density is collected and plotted as a function of current den-
sity in Figure 8.9b. The increase of the optical gain with current density depends
on the efficiency of the QW and the optical confinement factor. Both of these are
enhanced by the use of wide QWs in the active region. The rates at which the opti-
cal gain increases with current density in the blue LDs are 10.4 and 12.7 cm/kA
for the active regions composed of three 2.6 nm QWs and a single 10.4 nm wide
QW, respectively. The increase in differential gain in this case is 22%. On the other
hand, for the LDs with the higher In content in the active region, the differen-
tial gain is 2.2 and 6.5 cm/kA for three 3.0 nm thick QWs and the single 10.4 nm
QW, respectively. The increase is much more pronounced and equal to 195%.
This result proves that there is a higher increase in efficiency for long-wavelength
devices. We hope this result will help to mitigate the “green gap” problem.
8.4 Long-Living Laser Diodes on Bulk Ammono-GaN

The lifetime of the first LDs grown by PAMBE was in the order of hours [82].
The advancement in epitaxial growth and processing of the devices led to an
increase of the lifetime up to 2000 hours [83]. A recent study by Bojarska et al.
compared degradation of LDs grown by MOCVD and PAMBE and showed a
different activation energy of the degradation rate [84]. This finding is of great
importance because it points out that the limits of the reliability of devices grown
by MOCVD and PAMBE can be different. In this section, we will discuss the
mechanism of degradation present in the LDs grown by PAMBE. The under-
standing of this mechanism led to demonstration of LDs with an estimated life-
time of 100 000 hours [85].
In this study, high-quality GaN substrates grown by ammonothermal method
[86] with a low threading dislocation density (TDD in the order of 104 –105 cm−2 )
were used. The use of low TDD substrate allows us to investigate the influence of
defects generated during epitaxy on the degradation of devices. Three LDs will
be compared with a different behavior under stress current. The LD structure
consisted of an n-type 700 nm AlGaN cladding followed by 100 nm GaN and
InGaN waveguide with multiquantum wells (MQWs) inside. At the end of the
InGaN waveguide, the 20 nm EBL is placed. It is followed by 100 nm GaN:Mg
and 400 nm AlGaN:Mg cladding. The MQW region consists of three 2.6 nm
In0.17 Ga0.83 N QWs separated with 8 nm quantum barriers (QB). The indium
content in the QB is kept the same as in the whole InGaN waveguide and is
equal to 8%. The Al0.14 Ga0.86 N:Mg EBL is placed behind the MQW to form
an energetic barrier in the conduction band to prevent the electron overflow.
Below, three types of LDs with different EBL design will be compared. The
EBLs in these LDs are composed of (i) Laser A – In0.01 Al0.14 Ga0.85 N:Mg (Mg:
5 × 1019 cm−3 ), (ii) Laser B – Al0.14 Ga0.86 N:Mg (Mg: 5 × 1019 cm−3 ), and (iii)
Laser C – Al0.14 Ga0.86 N:Mg (Mg: 1 × 1019 cm−3 ). The EBLs were grown at the
same temperature as the InGaN waveguide (T g = 650 ∘ C). Indium was supplied
during growth in all three cases but acted only as a surfactant in Laser B and
C. It was not incorporated because the sum of gallium and aluminum fluxes
were set to be higher than the nitrogen flux, and these species are preferentially
incorporated before indium. However, indium acts as a surfactant and allows
for the growth in the preferential step-flow regime [25]. After the growth of the
EBL, the temperature was ramped up for the growth of GaN:Mg. During the
ramping of the temperature, the excess indium (in the case of Laser A, B, and C)
and gallium (in case of Laser B and C) was desorbed. Laser B and C differ only
by the amount of supplied and thus incorporated magnesium.
The use of low TDD GaN substrates allows to identify the density of defects
generated during epitaxy that produce dislocations through a simple tech-
nique of defect-selective etching (DSE) [87, 88]. In this technique, the molten
NaOH–KOH eutectic with addition of MgO powder (3 : 2 : 1) is used as the
etchant. The samples were etched at a temperature of 450 ∘ C for 30 minutes for
Laser A and B and 60 minutes for sample C. Figure 8.10a–c present the scanning
electron microscope (SEM) images of the surface after DSE of Laser A, B, and
C, respectively. Please note the 100 times larger area in Figure 8.10c used to
show the revealed defects. In the case of Laser C, two kinds of dislocations are
observed. The first kind showing up as big hexagonal craters are the same defects
as in Laser A and B. However, because of longer etching time, they etched down
Figure 8.10 (a–c) The SEM images of three LD structures with EBL grown at different
conditions after defect-selective etching. Please note the different scale in (c). The values give
the density of defects revealed by etching. The red arrows in (c) depict the threading
dislocations originating from the substrate. (d) Dependence of the change of threshold
current density during the reliability tests. (e) Dependence of LD lifetime on defect density.
Dashed line is a guide to the eye. Source: Reprinted with permission from Muziol et al. [85].
Copyright 2019, Elsevier.
to the EBL and started etching in the horizontal direction forming a symmetrical
structure. The other kind of defect is marked in Figure 8.10c with red arrows.
Those are the threading dislocations originating from the substrate. There is a
tremendously high span in the densities of the exposed dislocations between
the samples. The densities of the dislocations originating from epitaxy are
5 × 108 cm−2 , 2 × 107 cm−2 , and 1 × 105 cm−2 for Laser A, B, and C, respectively.
The current stress tests reveal that the defects exposed in DSE are responsible
for the degradation of the devices. The reliability tests had been performed on
LDs mounted in TO-56 cans. The test condition was to keep a constant output
power of 15 mW – during the degradation, the operating current was adjusted.
The tests were conducted with the temperature stabilized at 22 ∘ C. Figure 8.10d
presents the measured change of threshold current density. The lifetime of the
devices is defined as the time after which the jth increases by 50%. Throughout the
tests, the jth of only one device, Laser A, degraded above 50%. In the case of Laser
B and C, the lifetime is approximated assuming a linear degradation rate. The life-
times of Laser B and C are 15 000 and 100 000 hours, respectively. As can be seen
in Figure 8.10d, Laser C showed almost no degradation during the 4000 hours
of test, which gives the extremely high estimated lifetime. The lifetime of Laser
C is as good as the lifetimes reported for LDs grown by MOCVD [12, 14] – if
not higher. The dependence of the extracted lifetime values on the defect den-
sity exposed in DSE is presented in Figure 8.10e. The devices with lower defect
density show a lower degradation rate and thus have a higher lifetime.
We have used the high-angle annular dark-field scanning transmission electron
microscopy (HAADF STEM) to gain insight into the origin of the dislocations
exposed in DSE. Figure 8.11a,b present the images taken along [11–20] zone axis
of Laser A and B, respectively. In the case of Laser A, threading dislocations were
found to originate in the EBL. The threading dislocations observed in Figure 8.11a
start from I 1 basal stacking fault (BSF) domains [89]. When such a BSF is intro-
duced, then the domain is closed with the introduction of the second I 1 BSF
(a) (b)
Figure 8.11 (a) Large-area HAADF STEM images of LD with: (a) InAlGaN EBL and (b) AlGaN EBL.
The black arrow in (a) depicts the threading dislocations generated in EBL via the BSF domain.
Source: Reprinted with permission from Muziol et al. [85]. Copyright 2019, Elsevier.
above. Depending on the number of wurtzite (0002) planes introduced between

the two SFs, the BSF domain may be surrounded by prismatic stacking fault or by
Shockley 1/3 <1–100> partial dislocations. Interaction between the partial dislo-
cations led to the creation of perfect 1/3 <11–20> dislocation half-loops, which
may then propagate to the surface during the growth of the layer. Then, these
dislocations can be observed with DSE. The quaternary InAlGaN layers exhibit
a surprisingly large number of BSFs even if the indium content is low. Similar
effects have been observed in InAlGaN layers grown by MOCVD [89, 90].
In the case of Laser B, no such defect was observed in transmission electron
microscope (TEM) specimens. However, this is only because the density of
the defects is low and the probability to observe them in TEM is low. It has
been shown that in the Mg-doped layers grown by MOCVD, the formation of
BSFs occurs near the Mg atoms [91]. Therefore, although we have not identified
directly the origin of the dislocations exposed with DSE in Laser B and C, we
conclude that these defects arise due to BSFs. The mechanism responsible for
the formation of BSF is likely to be similar in both cases – the addition of indium
or magnesium to AlGaN layers. Indium and magnesium atoms that incorporate
into the crystal change locally the lattice arrangement and may lead to formation
of BSFs because these local lattice deformations are experienced by the atoms,
which are subsequently incorporated. On the other hand, aluminum atoms
solely seem not to cause generation of BSFs, at least up to the concentration of
15%. This has been confirmed by our previous report showing no influence of
the presence of aluminum on degradation of LDs [30].
In summary, we have studied the influence of the EBL growth parameters on
the formation of defects and their influence on degradation of LDs grown by
PAMBE. It was found that an enormous amount of defects can be formed in the
EBL region because of the use of unfavorable growth conditions. The quaternary
InAlGaN:Mg layer was found to generate TDD of 5 × 108 cm−2 even for indium
content as low as 1%. In the case of ternary AlGaN:Mg EBL, TDD was decreased
to 2 × 107 cm−2 . It was further reduced to 1 × 105 cm−2 by optimizing the Mg dop-
ing level. Additionally, it has been found through TEM studies that the threading
dislocations originate at domains formed by two I 1 BSFs.
Reliability test of LDs had shown a clear correlation with the defect density.
The lifetime of the devices has been increased from 2000 up to 100 000 hours
when defect density was decreased from 5 × 108 cm−2 to 1 × 105 cm−2 . The life-
time of 100 000 hours is at least as good as the lifetimes reported for devices
grown by MOCVD. Therefore, combined with the novel device designs enabled
by the incorporation of TJs, which will be discussed in Section 8.5, the importance
of PAMBE technique in the field of III-nitride optoelectronics is increasing.
8.5 Laser Diodes with Tunnel Junctions

The most challenging issues to be addressed in devices based on III-nitrides are
relatively poor p-type conductivity and difficulties with low-resistance Ohmic
p-type contact processing. Recently, increased attention has been dedicated to
the interband TJs [92] for the efficient conductivity conversion from p-type to
n-type in III-nitride devices [20, 93–96]. Application of TJs creates more free-
dom in device design – e.g. it eliminates the need for p-type contact deposition
[95, 97–99]. However, the utilization of TJ in wide-band semiconductors is a
counterintuitive approach. It is well known that the carrier tunneling through
p–n junction in reverse direction increases exponentially with the energy gap.
The additional complication that slowed down the progress of nitride TJ devel-
opment was caused by the p-type doping procedure used MOCVD, the dominant
technology for the nitride optoelectronic devices fabrication. In MOCVD, acti-
vation of the p-type conductivity in the (In)GaN:Mg layers requires breaking the
Mg-H complexes and hydrogen removal. For the case of (In)GaN:Mg layers that
are buried below n-type layers, p-type activation is difficult because the diffu-
sion of hydrogen is completely blocked through n-type layers [21]. For GaAs and
Si semiconductors it has been also observed that the migration of hydrogen in
p-type is easier than in n-type material [100].
The issue with the activation of the Mg-doped p-type layers is not present
for the hydrogen-free PAMBE technology. For PAMBE process, no postgrowth
annealing is required. Therefore, PAMBE seems to be better suited than
MOCVD for practical realization of the vertical devices with buried p-type
layers [101]. Recently, making use of PAMBE, it was shown that TJ resistance for
wide-bandgap semiconductors can be significantly reduced by making use of the
piezoelectric fields in the region of the junction [93, 96]. The use of piezoelectric
fields and heavy p- and n-type doping levels allowed to reduce the resistivity
of TJs grown by PAMBE to a level appropriate for demonstration of the CW
operation of nitride LDs [101].
TJs enable making a stack of LDs by interconnecting them vertically and can
be a cheap and viable alternative to LD bars. High-power pulsed LDs are very
attractive for many applications such as gas sensing, printing, and environmental
pollution control. Recently, one of the widely growing fields is the light detection
and ranging (LIDAR) in cartography, automotive, and industrial systems [102].
The LIDAR systems require high optical power (10–100 W) and very short light
pulses – for the safety reasons. The coupling of the light coming from stack of LDs
with external optics is much easier than from arrays of LDs because the spatial
separation between devices can be smaller by 2 orders of magnitude. The simul-
taneous operation of a cascade of n LDs increases the slope efficiency (SE) of
the full device n-times, which makes high power lasing conditions accessible for
smaller currents. In addition, the level of catastrophic optical damage (COD) is n
times higher in comparison with a single LD. In spite of the increasing interest in
such device designs, there is only one report on a stack of two III-N LDs grown by
MOCVD (probably because of the hydrogen passivation issue discussed above),
which shows very weak evidence of simultaneous laser action from both active
regions [103]. In particular, no extra peak in lasing spectrum nor improvement
in SE was observed for that device. This is probably due to difficulties with Mg
acceptor activation in buried p-type layers.
In this section, we will describe the properties of nitride TJ and give an insight
on how the TJ can be used to change the LD design. In particular, we will demon-
strate a LD stack and DFB LDs grown by PAMBE.
The proof of concept that TJ can be applied for nitride LDs is shown below.
We compare two LDs grown by PAMBE. One is a standard LDs with a p-type
Figure 8.12 (a) Schematic image of a standard and a TJ LDs. (b) The LIV characteristics of the
standard LD (dashed lines) and the TJ LD (solid lines). Both LDs are operating in CW mode at
450 nm. Source: Reprinted with permission from Skierbiszewski et al. [101]. Copyright 2018,
The Japan Society of Applied Physics.
contact. In the second one, on the top of the p-type cladding, the TJ was grown
(see Figure 8.12a).
The application of the TJ above claddings of LDs (see Figure 8.12a) should not
influence basic laser parameters such as internal losses, SE and the threshold cur-
rent. We can expect only a slight increase of the voltage drop across the LD device
related to the resistance of the TJ. In Figure 8.12b, we present LIV characteris-
tics for standard (dashed line) and TJ LDs (solid line) emitting at 450 nm in the
CW mode. For both LDs, the threshold current density is around 3 kA/cm2 and
the SE is 0.5–0.6 W/A. It can be concluded from the similar SE that internal loss
remained unchanged. Indeed, when the TJ is located far away from the active
region, there is no substantial overlap of the optical modes with the heavily doped
TJ region. The application of the TJ does not influence optical performance of the
LD; however, it slightly increases the operating voltage of the LD. In the provided
example, the TJ adds about 0.6 V to the turn-on voltage and a series resistance,
both of which adds up to an additional 0.8 V at the laser threshold.
To increase the tunneling current through the TJ, Krishnamoorthy et al.
[96] postulated insertion of InGaN QW, which reduces the depletion width
of TJ, thanks to the piezoelectric fields. We have found that additional doping
inside this QW further enhances tunneling [104]. However, heavy doping of
the TJ region creates some challenges. The most critical for the device stacking
application is the deterioration of the surface morphology related to the growth
of extremely highly doped layers. In Figure 8.13a, we present the atomic force
microscopy (AFM) image of the surface of the TJ LDs. We observe that high Si
doping (above 5 × 1020 cm−3 ) on the one hand reduces the resistance of the TJ,
but on the other hand, it causes surface morphology roughening – which does
not allow for stacking of the devices. Additionally, it is well known that for the
p-type doping, the maximum level of magnesium atoms acting as acceptors is
limited to (2–7) × 1019 cm−3 [105]. Above these values (depending on growth
details), the autocompensation related with creation of Mg double donors
Figure 8.13 AFM image showing the surface morphology of (a) a single TJ LDs with Si doping
in TJ above 5 × 1020 cm−3 and (b) a stack of two LDs grown by PAMBE with Si doping in TJ of
2 × 1020 cm−3 .
takes place. The other drawback with the Mg doping of nitrides is the polarity
inversion observed for high Mg concentration.
The way to decrease the TJ resistivity and keep the high quality of layers is to
understand the interplay between the QW parameters used to create TJ, i.e. (i)
QW width, (ii) In content of QW, and (iii) doping level of QW. For sufficiently
thin layers with intermediate doping levels of Mg and Si, we were able to achieve
smooth morphology with atomic steps – as shown in Figure 8.13b. The atomically
flat surface after the growth of TJ enables the epitaxy of subsequent devices and
realization of stacks of interconnected devices.
8.5.1 Stacks of Vertically Interconnected Laser Diodes

The stack of two LDs interconnected with TJs is shown schematically in
Figure 8.14a. We achieved smooth surface morphology after the epitaxy of
the structure of two LDs interconnected by TJ as shown in the AFM image
presented in Figure 8.13b. In Figure 8.14b, we present the band diagram of such
LDs stack. The electrons are injected into LD1 and recombine in the active
region with holes injected from the opposite side. The holes are generated by
tunneling of the electrons out of the valence band of LD1 into the conduction
band of LD2 – which take place in the TJ between LD1 and LD2. These electrons
are then injected into the active region of LD2, recombine with holes and so on,
which makes the whole structure act as a cascade.
Details of epitaxial structure are presented in Figure 8.14c, where the sequence
of layers is marked on the STEM image obtained for the studied LD stack. In
this structure, we have combined the idea of Al-free LD [30] and profit from
improved performance of wide single quantum well (SQW) [48]. The structures
of waveguides of both LDs as well as the TJs are schematically depicted in
Figure 8.14d–f. The bottom LD (LD1) contains Al0.05 GaN claddings, a 220 nm
In0.04 Ga0.96 N waveguide, and a 25 nm wide In0.17 Ga0.83 N QW [48]. The top LD
(LD2) is an Al-cladding free to reduce the tensile strain in the structure [30, 106].
It has GaN claddings, a 120 nm In0.08 Ga0.92 N waveguide, and a 25 nm wide
In0.18 Ga0.82 N QW. We intentionally designed a slightly different In content in the
Figure 8.14 (a) Schematics and (b) band diagram of the stack of two III-nitride laser diodes
interconnected with tunnel junctions. (c) STEM image of the stack of two LDs grown by PAMBE
with the layer sequence. Details of (d) top LD2 active region lasing at 459 nm, (e) tunnel
junctions used to interconnect the LDs and grown on top of the LD stack, and (f ) bottom LD1
active region lasing at 456 nm. Source: Reprinted with permission from Siekacz et al. [104].
Copyright 2019, The Optical Society.
SQWs of both LDs to be able to verify their lasing via observation of two peaks
in the lasing spectra.
The TJ region consists of 60 nm In0.02 GaN:Mg, followed by a 10 nm In0.17 GaN
QW and 20 nm In0.02 GaN:Si as presented in Figure 8.14e. The layers on the
sides of the QW are doped with Mg and Si at the levels of 5 × 1019 cm−3 and
4 × 1019 cm−3 , respectively. The first 5 nm of the QW is heavily doped with Mg at
a level of 1 × 1020 cm−3 , while the following 5 nm of the QW is n-type doped at a
level of 1.8 × 1020 cm−3 . The Mg and Si doping profiles in TJs were optimized to
achieve atomically flat surface without defects, which is essential for the growth
of the subsequent devices on top of the stack.
The LDs were operated with 200 ns long pulses and a repetition rate of 1 kHz.
The light–current (L–I) characteristics of the cascade of two LDs is shown in
Figure 8.15a. Two lasing thresholds had been observed. The first one at a cur-
rent density of 2.8 kA/cm2 with an SE of 0.7 W/A. The second one occurred at
4.4 kA/cm2 and the observed SE increased up to 1.4 W/A. The doubling of the SE
indicates that the same electrons (holes) are used twice to generate light in both
LDs. Obtained SE exceeds the theoretical limit for a single LD, which is a proof
that we observe lasing from two LDs. The maximum value of SE for a wavelength
of 460 nm is equal to 2.7 W/A, assuming internal losses and injection efficiency
equal to 0 cm−1 and 100%, respectively. Our devices are prepared without facet
Figure 8.15 (a) Light–current characteristics of the stack of two LDs grown by PAMBE. Two
lasing thresholds are observed. The slope efficiency is doubled after the second LD starts to
lase. Lasing spectra of the stack of two LDs obtained for (b) 3.7 kA/cm2 and (c) 5.3 kA/cm2 .
Insets show the collected near-field patterns. Source: Reprinted with permission from Siekacz
et al. [104]. Copyright 2019, The Optical Society.
coatings. Therefore, the light is emitted half through the front and half through
the back facet, leading to a limit of SE of 1.35 W/A for one facet. Furthermore,
to verify the observation of lasing from both LDs, the emission spectra were col-
lected at 3.7 and 5.3 kA/cm2 and are presented in Figure 8.15b,c, respectively.
Insets to Figure 8.15b,c show near-field patterns collected at these current den-
sities using a Gaussian telescope setup [107]. Strong filamentation is observed as
expected for wide-ridge LDs [108]. At j = 3.7 kA/cm2 , there is only one peak in
the spectrum at 𝜆 = 459 nm and a single near-field pattern visible. Above the sec-
ond threshold, a second peak 𝜆 = 456 nm in the spectrum and a second near-field
pattern appear. Observation of one peak in the spectra before the doubling of SE
and two peaks after the doubling is a clear indication that both LDs operate simul-
taneously for current densities above 4.4 kA/cm2 . The maximum optical power
obtained for the studied structure was 2.2 W per laser facet and can be further
increased by the use of dielectric coatings. Application of this design for n-LDs
interconnected by (n−1) TJ will allow to increase SE n-times. This construction
paves a way to achieving III-nitride high power pulse LD stacks for LIDAR appli-
cations.
8.5.2 Distributed Feedback Laser Diodes

The DFB LDs are designed to offer single mode operation with a high spectral
purity, which makes them ideal for applications such as communication, detec-
tion of particles, interferometry, and atomic clocks. Nowadays, the DFB LDs are
commercially available in a broad spectral range of 760 nm up to 16 μm [109].
There is a lack of DFB LDs based on III-nitrides. The first attempts to manu-
facture the DFB LDs were conducted shortly after the initial demonstration of
InGaN-based LDs [110, 111]. A technology, known from III-phosphides, con-
sisting of growth interruption after the active region, patterning, and regrowth
of a layer with a changed refractive index was used. However, because of a large
mismatch and low refractive index contrast between the AlGaN/GaN alloys, the
obtained side mode suppression ratio (SMSR) was low. It was not until 2016 that
new approaches had been proposed to solve this material problem. A postgrowth
approach to DFB LDs was proposed with the grating placed on the sides of the
ridge, which makes the grating laterally coupled to the optical mode [112, 113].
In these approaches, there is a high refractive index contrast between the GaN
waveguide and air or SiN passivation. However, because of the low overlap of
the optical mode with the grating, the coupling is weak and deposition of met-
allization is challenging. Only pulsed operation was achieved and the resulting
SMSR were in the order of 20 dB. Recent advancement in this technique resulted
in obtaining CW operation and SMSR of 35 dB [114]. Further details on this tech-
nology can be found in Section 9.2. Even more recently, new approaches have
been proposed such as electron beam lithography of DFB grating on: (i) the sides
of ridge of a fully processed devices [115] and (ii) indium tin oxide cladding layer
on top of the laser ridge [116].
We propose a different solution, which exploits the idea to move the metalliza-
tion to the sides of the laser ridge [95]. This is possible thanks to the application
of a TJ to change the conductivity of the top layer to n-type and in consequence
leaving the ridge exposed to air. The DFB grating is placed on top of the ridge as
shown in Figure 8.16a. The high refractive index contrast between air and GaN
together with the high overlap of the optical mode to the grating due to transver-
sal coupling should result in superior mode selectivity. To realize this concept, the
epitaxy of a standard LD is finished with a TJ. Afterward, the mesa structure is
Figure 8.16 (a) Bird’s-eye view and (b) side view of DFB LD schematics. (c) Top-view SEM of the
etched fifth order grating. (d) Bird’s-eye view of the device showing the metalization on the
sides of the ridge.
etched. Importantly, the depth needs to be sufficient to remove the TJ outside of

the ridge. Then, a regrowth of highly conductive GaN:Si is performed and metal-
lization is placed outside of the ridge. Such a configuration enabled us to constrain
the current flow from the metallization horizontally through the GaN:Si and then
into the TJ, where a change from n-type to p-type occurs and then to the active
region, as shown with black arrows in Figure 8.16b.
We have designed a DFB LD operating at 𝜆 = 450 nm with a fifth order grat-
ing on top of the ridge. The top view of the grating on the ridge is shown in
Figure 8.16c. The period of the grating was Λ = 456 nm, and the dimensions of
the top (unetched) and bottom (etched) parts were the same and equal to 228 nm.
The etching depth was 100 nm. The calculated 𝜅L parameter was equal to 0.6. It
is important to stress that the placement of the grating on top of the ridge ensures
a large coupling, even for a very small depth of etching. Figure 8.16d presents the
bird’s-eye view of the processed device with the patterned ridge, cleaved mirrors,
and metallization on the sides. The resonator length was 700 μm, the mesa width
was 2 μm, and the facets were left uncoated.
The DFB LDs were mounted in TO-56 cans and operated in quasi-CW mode.
The pulse length was 1 ms and the duty cycle was up to 20%, above that the
devices stopped lasing. A comparison of the LIV characteristics between a regu-
lar Fabry–Pérot LD and DFB LD are presented in Figure 8.17a,b, respectively. The
threshold current densities of the standard LD and DFB LD are 3.3 and 7 kA/cm2 ,
respectively. This is a more than twofold increase, and it shows how much room
Figure 8.17 LIV characteristics of (a) regular Fabry–Pérot LD and (b) DFB LD. Lasing spectra of
(c) regular Fabry–Pérot LD and (d) DFB LD. The Fabry–Pérot LD was operated in CW mode
while the DFB LD was operated in quasi-CW mode.
for improvement is for the DFB LDs. Additionally, the SE decreased from 0.5 to
0.17 W/A. Both changes in jth and SE point to an increase in the internal losses
of the DFB LD. It is possible that the grating is not homogenous throughout the
whole length of the ridge – this would lead to an increase in the internal losses.
A decrease of the threshold current density is necessary to obtain the CW oper-
ation.
The spectra of the Fabry–Pérot LD and DFB LD are presented in Figure 8.17c,d,
respectively. Thanks to the operation of the Fabry–Pérot LD in CW mode, clear
longitudinal modes can be seen. In contrast, the DFB LD has only one peak. The
FWHM is only 40 pm. However, the lasing mode is highly asymmetrical. This is
due to the quasi-CW operation mode. During the pulse, the DFB LD heats up and
its emission wavelength red shifts. The spectrum is collected through the whole
pulse width; therefore, the resulting peak is unnaturally broadened. It is expected
that after the CW operation is obtained, the peak will be symmetrical and much
narrower. Nevertheless, the achieved SMSR was above 35 dB, which is exception-
ally high for a pulse-operated device. Such a high SMSR is a consequence of the
high overlap of the optical mode with the grating placed on top of the ridge.
8.6 Summary
In this chapter, we have presented the characteristics of III-nitride LDs grown
by plasma-assisted MBE. We have discussed the growth mechanism of InGaN
in PAMBE and showed the need of using high active nitrogen flux. The design
of the InGaN active region was examined. It was shown that a highly efficient
transition path through the excited states is present in wide QWs. Additionally,
impact of the thickness of the QWs on LDs was analyzed. The thorough reliability
studies were presented. The type of defects responsible for degradation of devices
was identified. This allowed to greatly reduce the defect density, which led to a
substantial increase in the lifetime of the LDs up to 100 000 hours. Most impor-
tantly, the strongest attribute of the PAMBE technique was discussed – the lack
of unintentional incorporation of hydrogen, a shortcoming unavoidable with the
commonly used MOCVD. The utilization of buried TJs to novel devices such as
stacks of LDs for high-power applications and DFB LDs for high spectra purity
were presented.
Even though the conventional LDs grown by MOCVD have still better effi-
ciency, there can be niche applications for devices grown by PAMBE such as
applications requiring extremely long lifetime, high optical power from single
chip achievable only by stacks of devices, or DFB LDs.
Acknowledgments
This work has been partially supported by TEAM-TECH POIR.04.04.00-00-210C
/16-00 and HOMING POIR.04.04.00-00-5D5B/18-00 and POWROTY
POIR.04.04.00-00-4463/17-00 projects of the Foundation for Polish Sci-
ence co-financed by the European Union under the European Regional
References 325
Development Fund the National Centre for Research and Development grants
LIDER/29/0185/L-7/15/NCBR/2016 and LIDER/35/0127/L-9/17/NCBR/2018.
References
1 Zhang, Z., Kushimoto, M., Sakai, T. et al. (2019). A 271.8 nm
deep-ultraviolet laser diode for room temperature operation. Appl. Phys.
Express 12 (12): 124003.
2 Nagahama, S.-i., Yanamoto, T., Sano, M., and Mukai, T. (2001). Ultraviolet
GaN single quantum well laser diodes. Jpn. J. Appl. Phys. 40(part 2, no. 8A):
L785–L787.
3 Nakamura, S., Senoh, M., Nagahama, S.-i. et al. (1996). InGaN-based
multi-quantum-well-structure laser diodes. Jpn. J. Appl. Phys. 35 (1B):
L74–L76.
4 Nakamura, S., Senoh, M., Nagahama, S.-i. et al. (2000). Blue InGaN-based
laser diodes with an emission wavelength of 450 nm. Appl. Phys. Lett. 76 (1):
22–24.
5 Miyoshi, T., Masui, S., Okada, T. et al. (2009). 510–515 nm InGaN-based
green laser diodes on c-plane GaN substrate. Appl. Phys. Express 2: 062201.
6 Enya, Y., Yoshizumi, Y., Kyono, T. et al. (2009). 531 nm green lasing of
InGaN based laser diodes on semi-polar {20\bar21} free-standing GaN
substrates. Appl. Phys. Express 2: 082101.
7 Kuramoto, M., Sasaoka, C., Futagawa, N. et al. (2002). Reduction of inter-
nal loss and threshold current in a laser diode with a ridge by selective
re-growth (RiS-LD). Phys. Status Solidi A 192 (2): 329–334.
8 Uchida, S., Takeya, M., Ikeda, S. et al. (2003). Recent progress in high-power
blue-violet lasers. IEEE J. Selec. Top. Quantum Electron. 9 (5): 1252–1259.
9 Nakamura, S., Senoh, M., Nagahama, S.-i. et al. (1998). High-power,
long-lifetime InGaN/GaN/AlGaN-based laser diodes grown on pure GaN
substrates. Jpn. J. Appl. Phys. 37(part 2, no. 3B): L309–L312.
10 Nagahama, S.-i., Iwasa, N., Senoh, M. et al. (2000). High-power and
long-lifetime InGaN multi-quantum-well laser diodes grown on
low-dislocation-density GaN substrates. Jpn. J. Appl. Phys. 39(part 2, no.
7A): L647–L650.
11 Strauss, U., Somers, A., Heine, U. et al. (2017). GaInN laser diodes from
440 to 530nm: a performance study on single-mode and multi-mode R&D
designs. In: Novel In-Plane Semiconductor Lasers XVI, vol. 10123, 101230A.
International Society for Optics and Photonics.
12 Murayama, M., Nakayama, Y., Yamazaki, K. et al. (2018). Watt-class green
(530 nm) and blue (465 nm) laser diodes. Phys. Status Solidi A 215 (10):
1700513.
13 Kawaguchi, M., Imafuji, O., Nozaki, S. et al. Optical-loss suppressed InGaN
laser diodes using undoped thick waveguide structure. In: Gallium Nitride
Materials and Devices XI, (Proceedings SPIE, 2016), 974818. San Francisco,
California, USA.
14 Wierer, J.J., Tsao, J.Y., and Sizov, D.S. (2013). Comparison between blue
lasers and light-emitting diodes for future solid-state lighting. Laser Photon-
ics Rev. 7 (6): 963–993.
15 Nakamura, S., Harada, Y., and Seno, M. (1991). Novel metalorganic chem-
ical vapor deposition system for GaN growth. Appl. Phys. Lett. 58 (18):
2021–2023.
(LEEBI). Jpn. J. Appl. Phys 28(part 2, no. 12):: L2112–L2114.
18 Nakamura, S., Mukai, T., Senoh, M., and Iwasa, N. (1992). Thermal anneal-
ing effects on P-type Mg-doped GaN films. Jpn. J. Appl. Phys. 31(part 2, no.
2B):: L139–L142.
tion mechanism of P-type GaN films. Jpn. J. Appl. Phys. 31(part 1, no. 5A):
1258–1266.
20 Kuwano, Y., Funato, M., Morita, T. et al. (2013). Lateral hydrogen diffusion
at p-GaN layers in nitride-based light emitting diodes with tunnel junctions.
Jpn. J. Appl. Phys. 52 (8S): 08JK12.
21 Czernecki, R., Grzanka, E., Jakiela, R. et al. (2018). Hydrogen diffusion in
GaN:Mg and GaN:Si. J. Alloys Compd. 747: 354–358.
22 Heying, B., Averbeck, R., Chen, L.F. et al. (2000). Control of GaN surface
morphologies using plasma-assisted molecular beam epitaxy. J. Appl. Phys.
88 (4): 1855–1860.
23 Tarsa, E.J., Heying, B., Wu, X.H. et al. (1997). Homoepitaxial growth of GaN
under Ga-stable and N-stable conditions by plasma-assisted molecular beam
epitaxy. J. Appl. Phys. 82 (11): 5472–5479.
24 Zywietz, T., Neugebauer, J., and Scheffler, M. (1998). Adatom diffusion at
GaN (0001) and (0001) surfaces). Appl. Phys. Lett. 73 (4): 487–489.
25 Neugebauer, J., Zywietz, T.K., Scheffler, M. et al. (2003). Adatom kinetics on
and below the surface: the existence of a new diffusion channel. Phys. Rev.
Lett. 90 (5): 056101.
26 Oh, M.-S., Kwon, M.-K., Park, I.-K. et al. (2006). Improvement of green LED
by growing p-GaN on In0.25GaN/GaN MQWs at low temperature. J. Cryst.
Growth 289 (1): 107–112.
27 Hooper, S.E., Kauer, M., Bousquet, V. et al. (2004). InGaN multiple quantum
well laser diodes grown by molecular beam epitaxy. Electron. Lett. 40 (1):
33–34.
28 Skierbiszewski, C., Wasilewski, Z.R., Siekacz, M. et al. (2005). Blue-violet
InGaN laser diodes grown on bulk GaN substrates by plasma-assisted
molecular-beam epitaxy. Appl. Phys. Lett. 86 (1): 011114.
29 Skierbiszewski, C., Turski, H., Muziol, G. et al. (2014). Nitride-based laser
diodes grown by plasma-assisted molecular beam epitaxy. J. Phys. D: Appl.
Phys. 47 (7): 073001.
References 327
30 Muziol, G., Turski, H., Siekacz, M. et al. (2017). Aluminum-free nitride laser
diodes: waveguiding, electrical and degradation properties. Opt. Express 25
(26): 33113–33121.
31 Muziol, G., Turski, H., Siekacz, M. et al. (2015). Enhancement of optical
confinement factor by InGaN waveguide in blue laser diodes grown by
plasma-assisted molecular beam epitaxy. Appl. Phys. Express 8 (3): 032103.
32 Muziol, G., Turski, H., Siekacz, M. et al. (2016). Elimination of leakage of
optical modes to GaN substrate in nitride laser diodes using a thick InGaN
waveguide. Appl. Phys. Express 9 (9): 092103.
33 Nakamura, S. (1997). RT-CW operation of InGaN multi-quantum-well struc-
ture laser diodes. Mater. Sci. Eng., B 50 (1): 277–284.
34 Muziol, G., Siekacz, M., Turski, H. et al. (2015). High power nitride laser
diodes grown by plasma assisted molecular beam epitaxy. J. Cryst. Growth
425: 398–400.
35 Sawicka, M., Muziol, G., Turski, H. et al. (2013). Ultraviolet laser diodes
grown on semipolar (2021) GaN substrates by plasma-assisted molecular
beam epitaxy. Appl. Phys. Lett. 102 (25): 251101.
36 Turski, H., Muziol, G., Wolny, P. et al. (2014). Cyan laser diode grown by
plasma-assisted molecular beam epitaxy. Appl. Phys. Lett. 104 (2): 023503.
37 Ambacher, O., Brandt, M.S., Dimitrov, R. et al. (1996). Thermal stability and
desorption of group III nitrides prepared by metal organic chemical vapor
deposition. J. Vac. Sci. Technol., B 14 (6): 3532–3542.
38 McSkimming, B.M., Chaix, C., and Speck, J.S. (2015). High active nitrogen
flux growth of GaN by plasma assisted molecular beam epitaxy. J. Vac. Sci.
Technol. A 33 (5): 05e128.
39 Gunning, B.P., Clinton, E.A., Merola, J.J. et al. (2015). Control of ion content
and nitrogen species using a mixed chemistry plasma for GaN grown at
extremely high growth rates >9 μm/h by plasma-assisted molecular beam
epitaxy. J. Appl. Phys. 118 (15): 155302.
40 ̇
Turski, H., Krzyzewski, ̇
F., Feduniewicz-Zmuda, A. et al. (2019). Unusual step
meandering due to Ehrlich-Schwoebel barrier in GaN epitaxy on the N-polar
surface. Appl. Surf. Sci. 484: 771–780.
41 ̇
Turski, H., Feduniewicz-Zmuda, A., Sawicka, M. et al. (2019). Nitrogen-rich
growth for device quality N-polar InGaN/GaN quantum wells by
plasma-assisted MBE. J. Cryst. Growth 512: 208–212.
42 Gallinat, C.S., Koblmüller, G., Brown, J.S., and Speck, J.S. (2007). A growth
diagram for plasma-assisted molecular beam epitaxy of in-face InN. J. Appl.
Phys. 102 (6): 064907.
43 Grandjean, N., Massies, J., Semond, F. et al. (1999). GaN evaporation in
molecular-beam epitaxy environment. Appl. Phys. Lett. 74 (13): 1854–1856.
44 Averbeck, R. and Riechert, H. (1999). Quantitative model for the
MBE-growth of ternary nitrides. Phys. Status Solidi A 176 (1): 301–305.
45 Fabien, C.A.M., Gunning, B.P., Alan Doolittle, W. et al. (2015).
Low-temperature growth of InGaN films over the entire composition range
by MBE. J. Cryst. Growth 425: 115–118.
46 Turski, H., Siekacz, M., Wasilewski, Z.R. et al. (2013). Nonequivalent atomic
step edges-role of gallium and nitrogen atoms in the growth of InGaN
layers. J. Cryst. Growth 367: 115–121.
47 Siekacz, M., Sawicka, M., Turski, H. et al. (2011). Optically pumped 500 nm
InGaN green lasers grown by plasma-assisted molecular beam epitaxy. J.
Appl. Phys. 110 (6): 063110.
48 Muziol, G., Turski, H., Siekacz, M. et al. (2019). Beyond quantum efficiency
limitations originating from the piezoelectric polarization in light-emitting
devices. ACS Photonics 6 (8): 1963–1971.
49 Muziol, G., Hajdel, M., Siekacz, M. et al. (2019). Optical properties of
III-nitride laser diodes with wide InGaN quantum wells. Appl. Phys Express
12 (7): 072003.
50 David, A., Grundmann, M.J., Kaeding, J.F. et al. (2008). Carrier distribution
in (0001)InGaN/GaN multiple quantum well light-emitting diodes. Appl.
Phys. Lett. 92 (5): 053502.
51 Scheibenzuber, W.G. and Schwarz, U.T. (2012). Unequal pumping of quan-
tum wells in GaN-based laser diodes. Appl. Phys. Express 5 (4): 042103.
52 Coldren, L.A., Corzine, S.W., and Mashanovitch, M.L. (2012). Diode Lasers
and Photonic Integrated Circuits. New York, NY: Wiley.
53 David, A., Young, N.G., Hurni, C.A., and Craven, M.D. (2017). All-optical
measurements of carrier dynamics in bulk-GaN LEDs: beyond the ABC
approximation. Appl. Phys. Lett. 110 (25): 253504.
54 David, A., Hurni, C.A., Young, N.G., and Craven, M.D. (2017). Field-assisted
Shockley-read-hall recombinations in III-nitride quantum wells. Appl. Phys.
Lett. 111 (23): 233501.
55 David, A. and Grundmann, M.J. (2010). Droop in InGaN light-emitting
diodes: a differential carrier lifetime analysis. Appl. Phys. Lett. 96 (10):
103504.
56 Schiavon, D., Binder, M., Peter, M. et al. (2013). Wavelength-dependent
determination of the recombination rate coefficients in single-quantum-well
GaInN/GaN light emitting diodes. Phys. Status Solidi B 250 (2): 283–290.
57 Scheibenzuber, W.G., Schwarz, U.T., Sulmoni, L. et al. (2011). Recombination
coefficients of GaN-based laser diodes. J. Appl. Phys. 109 (9): 093106.
58 Bernardini, F., Fiorentini, V., and Vanderbilt, D. (1997). Spontaneous polar-
ization and piezoelectric constants of III–V nitrides. Phys. Rev. B 56 (16):
R10024–R10027.
59 Takeuchi, T., Wetzel, C., Yamaguchi, S. et al. (1998). Determination of piezo-
electric fields in strained GaInN quantum wells using the quantum-confined
stark effect. Appl. Phys. Lett. 73 (12): 1691–1693.
60 Langer, R., Simon, J., Ortiz, V. et al. (1999). Giant electric fields in
unstrained GaN single quantum wells. Appl. Phys. Lett. 74 (25): 3827–3829.
61 Ambacher, O., Majewski, J., Miskys, C. et al. (2002). Pyroelectric properties
of Al(In)GaN/GaN hetero- and quantum well structures. J. Phys. Condens.
Matter 14 (13): 3399–3434.
62 Fiorentini, V., Bernardini, F., Della Sala, F. et al. (1999). Effects of macro-
scopic polarization in III–V nitride multiple quantum wells. Phys. Rev. B 60
(12): 8849–8858.
References 329
63 Della Sala, F., Di Carlo, A., Lugli, P. et al. (1999). Free-carrier screening of
polarization fields in wurtzite GaN/InGaN laser structures. Appl. Phys. Lett.
74 (14): 2002–2004.
64 Chichibu, S.F., Abare, A.C., Minsky, M.S. et al. (1998). Effective band gap
inhomogeneity and piezoelectric field in InGaN/GaN multiquantum well
structures. Appl. Phys. Lett. 73 (14): 2006–2008.
65 Grandjean, N., Damilano, B., Dalmasso, S. et al. (1999). Built-in electric-field
effects in wurtzite AlGaN/GaN quantum wells. J. Appl. Phys. 86 (7):
3714–3720.
66 Lefebvre, P., Morel, A., Gallart, M. et al. (2001). High internal electric field
in a graded-width InGaN/GaN quantum well: accurate determination by
time-resolved photoluminescence spectroscopy. Appl. Phys. Lett. 78 (9):
1252–1254.
67 Leroux, M., Grandjean, N., Laügt, M. et al. (1998). Quantum confined Stark
effect due to built-in internal polarization fields in (Al, Ga)N/GaN quantum
wells. Phys. Rev. B 58 (20): R13371–R13374.
68 Seo Im, J., Kollmer, H., Off, J. et al. (1998). Reduction of oscillator strength
due to piezoelectric fields in GaN/Alx Ga1−x N quantum wells. Phys. Rev. B 57
(16): R9435–R9438.
69 Takeuchi, T., Sota, S., Katsuragawa, M. et al. (1997). Quantum-confined
Stark effect due to piezoelectric fields in GaInN strained quantum wells. Jpn.
J. Appl. Phys. 36 (4A): L382–L385.
70 Chichibu, S., Azuhata, T., Sota, T., and Nakamura, S. (1996). Spontaneous
emission of localized excitons in InGaN single and multiquantum well
71 Young, N.G., Farrell, R.M., Oh, S. et al. (2016). Polarization field screening
in thick (0001) InGaN/GaN single quantum well light-emitting diodes. Appl.
Phys. Lett. 108 (6): 061105.
72 Kim, M.-H., Schubert, M.F., Dai, Q. et al. (2007). Origin of efficiency droop
in GaN-based light-emitting diodes. Appl. Phys. Lett. 91 (18): 183507.
73 Piprek, J. (2010). Efficiency droop in nitride-based light-emitting diodes.
74 Kioupakis, E., Rinke, P., Delaney, K.T., and Van de Walle, C.G. (2011).
Indirect auger recombination as a cause of efficiency droop in nitride
droop. Phys. Rev. Lett. 110 (17): 177406.
76 Shen, Y.C., Mueller, G.O., Watanabe, S. et al. (2007). Auger recombination in
InGaN measured by photoluminescence. Appl. Phys. Lett. 91 (14): 141101.
77 Krames, M.R., Shchekin, O.B., Mueller-Mach, R. et al. (2007). Status and
future of high-power light-emitting diodes for solid-state lighting. J. Disp.
Technol. 3 (2): 160–175.
78 Auf der Maur, M., Pecchia, A., Penazzi, G. et al. (2016). Efficiency drop in
green InGaN/GaN light emitting diodes: the role of random alloy fluctua-
tions. Phys. Rev. Lett. 116 (2): 027401.
79 Li, Y.L., Huang, Y.R., and Lai, Y.H. (2007). Efficiency droop behaviors of
InGaN/GaN multiple-quantum-well light-emitting diodes with varying
quantum well thickness. Appl. Phys. Lett. 91 (18): 181113.
80 Gardner, N.F., Müller, G.O., Shen, Y.C. et al. (2007). Blue-emitting InGaN–
GaN double-heterostructure light-emitting diodes reaching maximum quan-
tum efficiency above 200 A/cm2 . Appl. Phys. Lett. 91 (24): 243506.
81 STR Group. (2008). SiLENSe 5.4 package. http://www.str-soft.com/products/
SiLENSe (accessed 30 March 2020).
82 Skierbiszewski, C., Wiśniewski, P., Siekacz, M. et al. (2006). 60 mW
continuous-wave operation of InGaN laser diodes made by plasma-assisted
molecular-beam epitaxy. Appl. Phys. Lett. 88: 221108.
83 Skierbiszewski, C., Siekacz, M., Turski, H. et al. (2012). True-blue nitride
laser diodes grown by plasma-assisted molecular beam epitaxy. Appl. Phys.
Express 5 (11): 112103.
84 Bojarska, A., Muzioł, G., Skierbiszewski, C. et al. (2017). Influence of the
growth method on degradation of InGaN laser diodes. Appl. Phys. Express
10 (9): 091001.
85 Muziol, G., Siekacz, M., Nowakowski-Szkudlarek, K. et al. (2019). Extremely
long lifetime of III-nitride laser diodes grown by plasma assisted molecular
beam epitaxy. Mater. Sci. Semicond. Process. 91: 387–391.
86 Dwiliński, R., Doradziński, R., Garczyński, J. et al. (2009). Bulk ammonother-
mal GaN. J. Cryst. Growth 311 (10): 3015–3018.
87 Weyher, J.L., Brown, P.D., Rouvière, J.L. et al. (2000). Recent advances in
defect-selective etching of GaN. J. Cryst. Growth 210 (1): 151–156.
88 Kamler, G., Weyher, J.L., Grzegory, I. et al. (2002). Defect-selective etching
of GaN in a modified molten bases system. J. Cryst. Growth 246 (1): 21–24.
89 Smalc-Koziorowska, J., Bazioti, C., Albrecht, M., and Dimitrakopulos, G.P.
(2016). Stacking fault domains as sources of a-type threading dislocations in
III-nitride heterostructures. Appl. Phys. Lett. 108 (5): 051901.
90 Meng, F.Y., Rao, M., Newman, N. et al. (2008). Stacking faults in quaternary
Inx Aly Ga1−x−y N layers. Acta Mater. 56 (15): 4036–4045.
91 Khromov, S., Hemmingsson, C.G., Amano, H. et al. (2011). Luminescence
related to high density of mg-induced stacking faults in homoepitaxially
grown GaN. Phys. Rev. B 84 (7): 075324.
92 Esaki, L. (1958). New phenomenon in narrow germanium p−n junctions.
Phys. Rev. 109 (2): 603–604.
93 Krishnamoorthy, S., Akyol, F., and Rajan, S. (2014). InGaN/GaN tunnel junc-
tions for hole injection in GaN light emitting diodes. Appl. Phys. Lett. 105
(14): 141104.
94 Leonard, J.T., Young, E.C., Yonkee, B.P. et al. (2015). Demonstration of a
III-nitride vertical-cavity surface-emitting laser with a III-nitride tunnel
junction intracavity contact. Appl. Phys. Lett. 107 (9): 091105.
95 Malinverni, M., Tardy, C., Rossetti, M. et al. (2016). InGaN laser diode with
metal-free laser ridge using n+ -GaN contact layers. Appl. Phys. Express 9 (6):
061004.
References 331
96 Krishnamoorthy, S., Nath, D.N., Akyol, F. et al. (2010). Polarization-

engineered GaN/InGaN/GaN tunnel diodes. Appl. Phys. Lett. 97 (20):
203502.
97 Malinverni, M., Martin, D., and Grandjean, N. (2015). InGaN based micro
light emitting diodes featuring a buried GaN tunnel junction. Appl. Phys.
Lett. 107 (5): 051107.
98 Diagne, M., He, Y., Zhou, H. et al. (2001). Vertical cavity violet light emitting
diode incorporating an aluminum gallium nitride distributed Bragg mirror
and a tunnel junction. Appl. Phys. Lett. 79 (22): 3720–3722.
99 Kurokawa, H., Kaga, M., Goda, T. et al. (2014). Multijunction GaInN-based
solar cells using a tunnel junction. Appl. Phys. Express 7 (3): 034104.
100 Pearton, S.J., Corbett, J.W., and Borenstein, J.T. (1991). Hydrogen diffusion in
crystalline semiconductors. Physica B Condens. Matter 170 (1-4): 85–97.
101 Skierbiszewski, C., Muziol, G., Nowakowski-Szkudlarek, K. et al. (2018).
True-blue laser diodes with tunnel junctions grown monolithically by
plasma-assisted molecular beam epitaxy. Appl. Phys. Express 11 (3): 034103.
102 Schwarz, B. (2010). Mapping the world in 3D. Nat. Photonics 4 (7): 429–430.
103 Okawara, S., Aoki, Y., Kuwabara, M. et al. (2018). Nitride-based stacked laser
diodes with a tunnel junction. Appl. Phys. Express 11 (1): 012701.
104 Siekacz, M., Muziol, G., Hajdel, M. et al. (2019). Stack of two III-nitride laser
diodes interconnected by a tunnel junction. Opt. Express 27 (4): 5784–5791.
105 Obloh, H., Bachem, K.H., Kaufmann, U. et al. (1998). Self-compensation
in mg doped p-type GaN grown by MOCVD. J. Cryst. Growth 195 (1):
270–273.
106 Skierbiszewski, C., Siekacz, M., Turski, H. et al. (2012). AlGaN-free laser
diodes by plasma-assisted molecular beam epitaxy. Appl. Phys. Express 5 (2):
022104.
107 Rogowsky, S., Braun, H., Schwarz, U.T. et al. (2009). Multidimensional near-
and far-field measurements of broad ridge (Al, In)GaN laser diodes. Phys.
Status Solidi C 6 (S2): S852–S855.
108 Scholz, D., Braun, H., Schwarz, U.T. et al. (2008). Measurement and sim-
ulation of filamentation in (Al, In)GaN laser diodes. Opt. Express 16 (10):
6846–6859.
109 Zeller, W., Naehle, L., Fuchs, P. et al. (2010). DFB lasers between 760 nm
and 16 μm for sensing applications. Sensors 10 (4): 2492.
110 Hofstetter, D., Thornton, R.L., Romano, L.T. et al. (1998). Room-temperature
pulsed operation of an electrically injected InGaN/GaN multi-quantum well
distributed feedback laser. Appl. Phys. Lett. 73 (15): 2158–2160.
111 Masui, S., Tsukayama, K., Yanamoto, T. et al. (2006). First-order AlInGaN
405 nm distributed feedback laser diodes by current injection. Jpn. J. Appl.
Phys. 45 (29): L749–L751.
112 Slight, T.J., Odedina, O., Meredith, W. et al. (2016). InGaN/GaN distributed
feedback laser diodes with deeply etched sidewall gratings. IEEE Photonics
Technol. Lett. 28 (24): 2886–2888.
113 Kang, J.H., Martens, M., Wenzel, H. et al. (2017). Optically pumped DFB
lasers based on GaN using 10th-order laterally coupled surface gratings.
IEEE Photonics Technol. Lett. 29 (1): 138–141.
114 Slight, T.J., Stanczyk, S., Watson, S. et al. (2018). Continuous-wave operation
of (Al, In)GaN distributed-feedback laser diodes with high-order notched
gratings. Appl. Phys. Express 11 (11): 112701.
115 Holguín-Lerma, J.A., Ng, T.K., and Ooi, B.S. (2019). Narrow-line InGaN/
GaN green laser diode with high-order distributed-feedback surface grating.
116 Zhang, H., Cohen, D.A., Chan, P. et al. (2019). Continuous-wave operation of
a semipolar InGaN distributed-feedback blue laser diode with a first-order
indium tin oxide surface grating. Opt. Lett. 44 (12): 3106–3109.
333
Edge Emitting Laser Diodes and Superluminescent Diodes

Szymon Stanczyk 1,2 , Anna Kafar 1,3 , Dario Schiavon 1,2 , Stephen Najda 2 ,
Thomas Slight 4 , and Piotr Perlin 1,2
1
Institute of High Pressure Physics – Polish Academy of Sciences (Unipress-PAS), Optoelectronic Devices
Laboratory, Al. Prymasa Tysiaclecia 98, 01-424 Warsaw, Poland
2 TopGaN Sp. z o.o., Sokolowska 29/37, 01-142 Warsaw, Poland
3
Kyoto University, Department of Electronic Science and Engineering, Kyotodaigakukatsura, Nishikyo Ward,
Kyoto, 615-8246, Japan
4
Compound Semiconductor Technologies Global Ltd., 6 Stanley Boulevard, G72 0BN, Glasgow, United
Kingdom of Great Britain and Northern Ireland
9.1 Laser Diode: History and Development

9.1.1 Optoelectronics Background
Data are the lifeblood of the modern economy. Data need to be transferred,
stored, and displayed in an efficient and economical way. During the past three
decades, it has become obvious that almost always light does this work most effi-
ciently. The preferred source of light is semiconductor laser diodes (LDs) because
of their outstanding beam quality, leading to easy coupling to optical fibers, small
form factor, high reliability, and low production cost. Presently, laser diodes are
indispensable in telecommunication, optical data storage (CD, DVD, and Blu-
Ray), digital printing, and notably in emerging applications – RGB light sources
for modern display systems. Infrared VCSELs are the solution of choice for lidar
imaging systems for autonomous cars. Conventional infrared laser diodes are
still used as pumps for solid-state lasers and particularly for fiber lasers. Visi-
ble diodes are to be used for specialized metal welding, such as microwelding
for next-generation battery applications.
The history of semiconductor light sources can be traced back to 1961. Bob
Biard and Gary Pittman were working at Texas Instruments to develop gallium
arsenide (GaAs) p–n diodes. Using an infrared microscope, they found that these
devices emitted significant light in the infrared region. Texas Instruments soon
obtained a patent for the world’s first infrared light-emitting diode (LED) devices
[1]. In 1962, Texas Instruments launched production of early GaAs LEDs. A
bit earlier, in 1960, Theodore Maiman from Hughes Industries demonstrated
the operation of a visible laser based on the emission from a ruby crystal
[2]. The idea of using a semiconductor p–n junction as a laser medium was
almost immediately took up by scientists (e.g. Basov et al. [3]). Consequently,

334 9 Edge Emitting Laser Diodes and Superluminescent Diodes
in 1962, almost at the same time, four American groups announced the emission
of coherent light from a semiconductor p–n junction [4–7]. The devices were
based on a GaAs p–n homojunction with notable exception of a GaAsP device,
which emitted red coherent light. The idea of using a ternary GaAsP semicon-
ductor was a result of a vision of Nick Holonyak [6] who believed that the new
emitters should operate in the visible light to have more commercial impact. The
introduction of a GaAsP [6] device was also the first example of the usefulness
of ternary semiconductor compounds and paved the way for the present energy
gap engineering so crucial for modern semiconductor optoelectronics. The first
GaAs laser diodes were small semiconductor chips with parallel facets forming
a Fabry–Pérot resonator and a p–n junction fabricated by zinc diffusion. These
devices were characterized by a large threshold current, required cryogenic
cooling, and were inefficient compared to gas- and solid-state lasers, hence
lacked potential for real applications. At this time, many researchers predicted
no commercial future for these devices. However, this way of thinking was
questioned by a German researcher Herbert Kroemer. He realized that the use
of semiconductor heterostructures may lead to a significant improvement
in the performance of semiconductor laser diodes [8, 9]. This idea was initially
overlooked (the paper was rejected by a major journal) and had to wait until
further advancements in epitaxy growth in 1970 before a room temperature
GaAs/GaAlAs heterostructure laser diode could be realized [10, 11]. Kroemer
and Alferov won the Nobel Prize for this work.
These achievements soon led to new practical applications: optical data record-
ing introduced by Sony and Phillips (1977) and fiber optic telecommunication. In
April 1977, General Telephone and Electronics tested and deployed the world’s
first live telephone traffic through a fiber-optic system running at 6 Mbps, in Long
Beach, California. They were soon followed by Bell in May 1977, with an opti-
cal telephone communication system installed in downtown Chicago, covering a
distance of 1.5 mi. A further development of the semiconductor laser diode was
achieved with the introduction of quantum wells into the device structure. Such
devices were first realized by J.P. van der Ziel in 1975 [12] and Russel Dupuis
in 1978 [13]. Single and multiquantum well laser diodes have gradually become
a standard solution for the optoelectronic industry.
Most of the semiconductor laser diodes are based on III–V semiconductors,
in particular on the GaInAsP family. However, the big disadvantage of these
materials is the inability to obtain wide bandgap semiconductors that are able
to emit yellow, green, or blue light. The GaAs semiconductor when alloyed
with Al or P increases its bandgap; however, at a certain composition, they
unavoidably become indirect semiconductors ceasing light emission. For
instance, the GaAs1−x Px system becomes indirect for x = 0.46 [14], which
corresponds to a bandgap of 2.09 eV (593 nm). Similarly, for the Ga1−x Alx As
system [15], the direct–indirect transition occurs roughly at a composition
of x = 0.45, which corresponds to a wavelength of 1.98 eV (625 nm). Obviously,
the short-wavelength spectral range of devices based on classical III–V semi-
conductors is limited to red/yellow. This very basic physical limit has prohibited
shorter wavelength laser diodes from being developed. Researchers have looked
to more novel wide bandgap materials, in particular II–VI zinc-containing
Figure 9.1 Lattice constant end energy

AlN
gap of all three binary nitrides. 6
Energy gap (eV)

Visible spectrum
4
GaN
2
InN
0
3.0 3.1 3.2 3.3 3.4 3.5 3.6
Lattice constant, a (Å)
compounds such as ZnMgCdSeS. These devices emit in the blue-green spectral

region and show interesting lasing parameters but lacked reliability [16] – limited
to a few hundred hours [17]. The very short device lifetime was attributed to low
cohesion energy of the II–VI compound lattice, relatively high photon energy,
high operating voltage, and lack of an ideal substrate. Most of the mentioned
issues will appear again in the case of nitride laser diodes. AlGaInN provides
a very interesting III–V semiconductor material system for wide bandgap
optoelectronic applications in the UV and visible light region (see Figure 9.1).
However, the development of nitride system for the optoelectronics application
has been a long and complicated process.
9.1.2 Gallium Nitride Technology Breakthroughs

Gallium nitride (GaN), a key member of the group three nitrides family, was syn-
thesized almost 70 years ago (1938) by Juza and Hahn [18] by passing ammonia
over hot gallium. In the mid-1960s, already after the discovery of the first semi-
conductor light emitters, researchers in Radio Corporation of America (RCA)
started looking for a material suitable for blue light emission needed for the con-
struction of LED-based flat TV sets (it is surprising how close it was to the mod-
ern concept of laser TV). James Tietjen, a director in RCA, approached a young
coworker Herbert Maruska proposing a new development of gallium nitride crys-
tal growth by a halide method (now more commonly known as hydride vapor
phase epitaxy [HVPE]). Maruska accepted the challenge and succeeded in devel-
oping the HVPE method, one of the most important technologies of gallium
nitride growth. The HVPE method remains at present the main technology for
producing thick layers and free-standing crystals of GaN. However, at the time
of Maruska’s early studies, the quality of the film was low and the doping con-
trol was very limited. In 1970, Jacques Pankove returned from a sabbatical year
in University of California Berkeley and immediately joined the group of Tietjen
and Maruska. Together, they demonstrated the first blue GaN LED [19], but they
fail to obtain the p-type material. In 1973 and 1974, the RCA GaN effort stopped
because of a shift in business policy and GaN drifted into obscurity.
The revival of gallium nitride technology is due in great extend to the talent and
persistency of Isamu Akasaki who has been working for years trying to realize
gallium nitride based blue light-emitting diodes. Akasaki together with Hiroshi
Amano explored the possibilities of metalorganic vapor phase epitaxy (MOVPE)
for the growth of gallium nitride layers on sapphire substrates. They invented a
new growth strategy consisting of initiating the growth at very low temperature
(using a low-temperature buffer layer). This low-temperature buffer layer (orig-
inally made out of AlN) [20] allowed for a smooth, two-dimensional growth of
GaN and removed the first barrier for obtaining device-grade layers. This ini-
tial important result was followed by fabrication of p-type gallium nitride by Mg
doping and postgrowth activation by electron irradiation [21] by the same group
in 1989. The path toward construction of the first practical devices was opened.
(It is worth mentioning that the first patent application for practical GaN-made
low-temperature buffer layer was filed by Theodore Moustakas from Boston Uni-
versity in 1991.)
The opportunity was not overlooked by a researcher of a Japanese company
Nichia Chemical – Shuji Nakamura. Nakamura and coworkers introduced a new,
more practical p-type GaN activation method by thermal treatment [22]. Soon
after, they demonstrated the first efficient blue light-emitting diode [23], making
the dream of the old RCA team of Tietjen, Maruska, and Pankove a reality. Blue
and shortly later green InGaN LEDs of Nakamura [24] turned out to be a turning
point in the history of optoelectronics. They allowed for the construction of full
color displays and most importantly white LEDs [25]. White LEDs, being a com-
bination of blue light-emitting diode and yellow phosphor [26], are nowadays
a crucial element of TV and computer monitor backlighting systems and also
important in general lighting applications. Blue/green/white LEDs correspond
to the majority of nitride device production.
A GaN laser diode was the next natural step and an obvious technical chal-
lenge for Shuji Nakamura to develop. The commercial motivation was to fab-
ricate a laser suitable for use in new systems such as optical data storage and
high-density video disk needed for high-resolution TV. Although the concept of
using a short-wavelength laser diode had been around for some time, the first
definition of a new standard called BluRay was revealed by Sony and Pioneer at
the CEATEC electronics show in Chiba Japan in 2000. However, in the beginning
of the 1990s, realization of a nitride laser diode was proving a difficult task even
though blue LEDs were already produced in large numbers. The main reason was
the relatively low quality of nitride layers grown on sapphire substrates. The lat-
tice mismatch between gallium nitride and sapphire is close to 16% and results
in a high density of misfit dislocations. Typically, the density of dislocations in
GaN layers grown on sapphire substrates is of the order of 108 –1010 cm−2 . Unex-
pectedly, this high density of defects did not prevent InGaN LEDs from efficient
emission of light. This very surprising observation was explained in a famous
paper of Chichibu et al. [27] by carrier localization on indium fluctuations, which
prevents photogenerated carriers from diffusing to nonradiative recombination
centers. Again, in the beginning of nitride laser diode history, it was not clear if
it is possible to manufacture laser diodes on sapphire substrates. Additionally, as
sapphire does not conduct current, it was necessary to fabricate n- and p-type
electrodes on top of the structure which added to the complication of the laser
diode design.
9.1.3 Development of Nitride Laser Diodes

In the fall of 1995, Nakamura and his team [28] demonstrated the first
nitride-based laser diode and the shortest wavelength semiconductor laser
diode ever realized at 417 nm. The structure of this early nitride laser was
complicated; for instance, the active area was composed of 26 InGaN quantum
wells. Lasing was achieved at a very high voltage close to 30 V, which clearly
prohibited continuous-wave (CW) operation. The laser design was improved by
introducing a ridge geometry structure and reducing the number of quantum
wells to five [29], giving an improved voltage of 20 V and a threshold current
density of 3 kA/cm2 . By the end of 1996, Nakamura et al. demonstrated the
first CW operation of a violet InGaN laser diode [30]. However, the lifetime
of the device was only one second, but the threshold voltage of the device was
further reduced to 8 V. By the end of 1996, the lifetime of CW-operated devices
was extended to 27 hours [31], the threshold voltage reduced to 5.5 V, and the
threshold current density kept at the level of 3.6 kA/cm2 . During 1997, the
laser lifetime was further extended to 300 hours [32]. At this point, the laser
performance was one step away from a commercially useful device. However, the
reliability was unsatisfactory because of the low crystalline quality of the nitride
structure grown on sapphire. In 1997, it was clear for Nakamura and researchers
from other groups that a substantial improvement in the crystalline quality of
the laser structures was required; otherwise, a commercially viable device would
not be achieved. Because high-quality GaN substrates were unavailable at this
time, researchers turned to develop a dislocation filtering method called ELOG
(epitaxial lateral overgrowth), also known under other acronyms such as ELO
and LEO (see Figure 9.2). This technology had been used in the past for the
growth of GaAs layers on silicon and was readapted for nitrides by the University
of Santa Barbara group [33]. This method consists of interrupting the GaN
layer growth by the deposition of an oxide or metal mask in the form of stripes
(e.g. 5 μm wide stripe with a periodicity of 13 μm [33]). After this initial step, the
wafer returns to the MOVPE reactor where the mask is overgrown by a relatively
thick layer of GaN (e.g. 12–20 μm [33]). The regions above the mask (wings) are
characterized by a much lower dislocation density than the areas not masked by
the oxide. A structure prepared in this way is called a “ELOG substrate” and is a
starting point for laser diode manufacturing.
The other challenge existing in gallium nitride laser diode technology was
related to the growth of Alx Ga1−x N cladding layers, which confine the light
(in the transversal direction) around the active layer of the device. This solution
SiO2
GaN
Sapphire
(a) (b) (c)
Figure 9.2 Schematic representation of three stages of epitaxial lateral overgrowth:

GaN with SiO2 mask (a), GaN overgrowth with gradual coalescence (b), and totally coalesced
GaN layer (c).
was well known from GaAs/AlGaAs devices; however, the substantial strain
existing between AlN and GaN (around 2.5%) caused problems with strain
relaxation. The AlGaN cladding being under tensile strain has the tendency to
crack, making the structure unusable for device fabrication. One method of
improving the quality of the laser structure was to replace thick AlGaN layers by
superlattices [34]. Such superlattices were grown from thin layers of GaN and
Alx GA1−x N of thickness around 25 Å and Al composition x around 0.16 [34].
The introduction of superlattice structures helped a great deal in the elimination
of cracks induced by the presence of large tensile strain in the cladding layers
of the laser diode. The above improvements (realized between 1997 and 1998)
helped in demonstrating a laser diode lifetime reaching 10 000 hours [34].
Meanwhile, other Japanese and US groups, including Toshiba [35], Fujitsu [36],
Xerox [37], Cree, and North Carolina State University (1998) [38], succeeded in
the fabrication of pulsed operated InGaN lasers.
However, the advancement in ELOG technology did not remove all the
obstacles hampering the proliferation of nitride laser diode technology. The
nitride ELOG structure fabricated on the sapphire was thick enough to cause the
wafer bowing, which made the laser diode processing very difficult. Also, mirror
fabrication by dry etching could not produce sufficiently good-quality mirrors
(at least at this time) equal to those cleaved in standard materials. Consequently,
the need for true GaN substrates for laser diode epitaxy was already at that time
very clear.
The Nichia team [34] was probably the first to prepare free-standing gallium
nitride substrates for laser diode fabrication. These wafers were fabricated first
as ELOG substrates, but then, they were overgrown by a thick (e.g. 80 μm)
layer of GaN, and eventually, the sapphire substrate was polished off leaving a
free-standing crystal of GaN.
However, the mass production of nitride-based laser diodes had to wait for
a new GaN substrate fabrication technology. Surprisingly, this technology was
developed from an earlier HVPE method by Maruska and modified by the Tokyo
University of Agriculture and Technology and Sumitomo [39]. To enhance the
crystalline quality of this material, a method of dislocation reduction by means of
hexagonal pits was used. This dislocation elimination method was called DEEP.
In parallel to the Sumitomo method, other companies developed various types
of GaN wafers grown by HVPE, including Japanese Furukawa and Mitsubishi
Chemicals, French Saint Gobain-Lumilog, and Polish Ammono. At present, it
is possible to find on the market 2–4 in. GaN substrate crystals, with a density
of dislocations in the order between 104 cm−2 and 107 cm−2 . However, the price
of these substrates remains a limiting factor in the development of nitride laser
diode.
With the advent of high-quality GaN substrates, ridge waveguide GaN laser
diodes can be fabricated. In Figure 9.3, you can observe a typical view of a laser
diode made on GaN substrate with InGaN QWs forming the active layer.
Laser diode development has been driven further by market needs including
high-power diodes for display and lighting applications, green lasers for full color
displays, UV semiconductor light sources for lithography and chemical sensing,
and finally the development of components for visible telecommunication. The
Dielectric
passivation
Top contact
p - type waveguide
Top cladding
InGaN quantum wells
p - AlGaN
and quantum barriers
Electron
blocking layer
n - type waveguide Bottom cladding
n - AlGaN
n - GaN substrate
Bottom contact
Figure 9.3 The schematic view of a typical nitride semiconductor-based laser diode structure.
last goal requires development of true single mode lasers such as distributed feed-
back (DFB) lasers and corresponding semiconductor optical amplifiers (SOAs).
Rapid development of high-power devices has resulted in great improvement
in the slope efficiency of nitride laser devices. Sony demonstrated almost 1 W of
optical power, with a slope efficiency exceeding 1.1 W/A from a 10 μm wide stripe
[40].
Similar devices were commercialized later by Nichia Chemical (2010). High
optical output powers in the CW regime of 2.5 W from multiemitter nitride sys-
tems (three-stripe miniarray) were also demonstrated by the group from TopGaN
[41]. A very high pulse power of 8 W was demonstrated by Osram [42]. An opti-
cal power of 300 W in picosecond pulses was achieved by the Tohoku University
group and Sony using a mode-locked InGaN laser with the optical semiconduc-
tor amplifier [43]. Recent laser diodes (2019) are characterized by a high injec-
tion efficiency reaching 90% and low internal losses of the order between 2 and
10 cm−1 (see reported data in [44]). These technical achievements have allowed
the availability of single stripe devices emitting 3.5 W of optical power at blue
wavelength (Nichia Corporation and Osram OS) (www.nichia.co.jp/en/product/
laser.html) [45], fully realizing the potential of the exceptional properties of the
GaN material system.
The final challenge for nitride laser diode technology was to take full advantage
of the extremely broad tunability of the (InAlGa)N energy gap (0.7–6.2 eV). No
other semiconductor system is, in principle, able to cover such a broad range of
wavelengths. The choice of target lasing wavelength was to a great extent defined
by the needs of potential applications. Thus, the 405 nm wavelength was chosen
for a new generation of DVD, 450 nm became a component of RGB laser dis-
plays, and 530 nm devices are expected to be a green component of RGB displays.
Fabrication of true blue and green nitride lasers was difficult because of compli-
cations in growth of high indium content quantum wells, including large strain,
high piezoelectric fields, a natural tendency of InGaN to phase separate, and,
last but not least, difficulty in designing a transversal waveguide for long wave-
length emission. The first blue (450 nm) InGaN laser diode was realized by Nichia
Chemical in 2000 [46]. It took much longer to demonstrate a green light-emitting
nitride laser diode. However, in 2009/2010, the first devices emitting in the green
region have been demonstrated by a couple of groups, most notably were the
Nichia demonstrating lasing at 515 nm [47], Sumitomo demonstrating lasing at
531 nm [48], and Osram demonstrating laser at 524 nm [49].
Laser emission in the UV region has also attracted considerable attention for
fluoroscopy and chemical sensing applications; however, the issue of pushing the
emitting wavelength down to deeper UV has turned out to be very complicated
too. The reason was strain management, low conductivity of AlGaN alloys, and
low radiative efficiency of In-free active layers. For a long time, the shortest wave-
length laser emission record (343 nm) was held by Hamamatsu [50].
A deep UV nitride laser has been a more significant challenge. This task has
been pursued by many groups including the team of Sitar (North Carolina State
University, USA) [51], Hirayama (Riken, Japan) [52], and Wunderer (PARC, USA)
[53]. Although, all three groups showed excellent optically pumped deep UV
laser diodes, none of them succeeded in demonstrating an electrically injected
laser because of poor hole injection, which is somehow expected for a very wide
bandgap semiconductor such as AlGaN having an Al content above 40%. The
first successful demonstration of an electrically injected deep UV nitride laser at
271.8 nm was by the Nagoya University group – Ziyi Zhang et al. [54]. In this
work, it is worth noticing that the p-type doping was achieved via polarization
doping not involving the Mg acceptor. A breakthrough in the classical p-type dop-
ing of high Al content AlGaN was demonstrated by Wang et al. [55]. He argued
that p-type Mg doping in nanorods is much more efficient than in “bulk” material
used in conventional laser diodes.
An interesting possibility, especially for long-wavelength nitride laser diodes,
is the InGaN quantum dot devices. In principle, for green and red nitride lasers,
InGaN quantum dots could be ideal because of better strain management, low
defect concentration, and limited quantum-confined Stark effect. However,
the problem is the uniformity of InGaN QDs, their stability during the final
overgrowth, and high enough density. In 2011, the group of Pallab Bhattacharya
demonstrated a green light-emitting (524 nm) InGaN QD laser [56], with
low threshold current density (≈1 kA/cm2 ); in 2013, the same group [57]
demonstrated a QD red laser operating at 630 nm, also with excellent threshold
parameters (J th = 2.5 kA/cm2 ). These outstanding results are of great potential
for modern optoelectronics but unfortunately have not yet been reproduced so
far by any other group.
Another important point in the developing of high wall plug efficiency laser
diodes is the introduction of an indium tin oxide (ITO) layer as a replacement
for the top AlGaN:Mg cladding. The schematic view of such a solution is given in
Figure 9.4 [58].
The advantage of introducing an ITO layer is that it has much better electrical
conductivity than p-type AlGaN and at the same time provides high refractive
index and low optical losses. The authors of Ref. [58] report optical losses below
1 cm−1 , while a typical resonator loss of AlGaN p-cladding laser diode is of the
order of 5–15 cm−1 .
The epitaxial growth of nitride laser structures has historically always been per-
formed on a c-plane GaN substrate. Although it is a natural growth direction
for MOVPE epitaxy, the polar nature of the (Al, In)GaN heterostructures leads
Figure 9.4 The structure of ITO

cladding laser grown on Bonding metal
semipolar GaN substrate.
ITO
Source: Murayama et al. 2018
[58]. Reproduced with
permission of John Wiley & Sons. SiO2 P-GaN
P-InGaN waveguide
P-InGaN waveguide
GaN substrate and buffer
to the appearance of large built-in electric fields in the InGaN quantum wells,
which manifest through the quantum-confined Stark effect (QCSE). The influ-
ence of the QCSE on laser performance is not straightforward. From one point
of view, the QCSE leads to optical gain reduction, as the transition matrix ele-
ment depends on the wave function overlap. On the other hand, the long carrier
lifetime promotes fast pumping and reaching transparency conditions at a rela-
tively low current. Nevertheless, especially for high indium content QWs (green
and red regions), the elimination of the QCSE seems to be very beneficial. As the
piezoelectric effect depends on the position of the QW plane in relation to crys-
tallographic direction, one can choose the proper growth direction to reduce the
internal fields. This method of reducing the internal electric field was proposed
among others by Northrup [59] and pursued by group from the University of
California at Santa Barbara and companies like Sumitomo and Sony.
The growth of nitride-based optoelectronic devices on semipolar and nonpolar
directions of GaN has been attracting quite a lot of attention during the past
15 years since InGaN quantum wells grown on nonpolar and semipolar planes
are subjected to much smaller electric field caused by piezoelectricity and
pyroelectricity. The reduction of the electric field leads to enhanced radiative
recombination, larger optical gain, and the possibility of growing thicker
quantum wells. Additionally, high indium content InGaN layers (quantum
wells), grown at relatively low temperature, tend to demonstrate better quality
compared to those grown on c-plane substrates. The major breakthrough in
the development of optoelectronic devices based on semipolar plane GaN was
achieved by Sumitomo Electrics [60] who demonstrated green lasing (520 nm) of
an InGaN laser diode grown on (20–21) plane GaN. Thanks to the large research
effort of primarily University of California, Santa Barbara and Sumitomo groups,
the large potential of semipolar structures for the light emitters’ applications was
eventually revealed [60–62].
Although good parameters of blue and green laser diodes fabricated on semipo-
lar and nonpolar planes of GaN were demonstrated, in most of the cases, the elec-
trical properties of these devices remained an issue. For instance, Kelchner et al.
[61] demonstrated a blue light-emitting laser diode on the nonpolar m-plane of
gallium nitride with a reasonable threshold current density of around 10 kA/cm2
but had a threshold voltage close to 32 V, which excluded CW operation. Tyagi
et al. [62] demonstrated a green laser diode on (2021) semipolar plane of GaN.
Again, the threshold voltage was 16 V, too high to obtain CW operation. It took
a relatively long time to achieve a high-performance green laser diode grown on
semipolar GaN with an operating voltage of 4 V and a maximum optical power
of 2 W at 2.5 A [58].
9.2 Distributed Feedback Laser Diodes

Another interesting type of optoelectronic nitride-based devices that provides
unique properties is the DFB laser. The working principle of the DFB-type laser
diode was first explained in 1971 by Kogelnik and Shank [63, 64] by the use of cou-
pled wave theory. The unique property of such a laser diode is the lack of necessity
for the existence of Fabry–Pérot cavity mirrors because the positive feedback is
provided via backward Bragg scattering by many weak reflection planes periodi-
cally distributed through the gain medium. Therefore, in contrast to Fabry–Pérot
cavity-based laser diodes, in which traveling waves do not interfere with them-
selves until they reach the mirrors, in DFB lasers, the traveling waves couple to
each other during propagation through the waveguide. The reflection planes are
made by the introduction of a periodical change of the effective refractive index,
which can be achieved by several different methods. The scheme showing the
difference between the Fabry–Pérot and DFB laser diode is shown in Figure 9.5.
In contrast to standard Fabry–Pérot laser diodes, the basic DFB type – both
with and without phase shift – assumes that the laser has no additional reflection
from the facets (mirrorless). As it can be seen in Figure 9.5, the difference between
the Fabry–Pérot-based laser diode and DFB lies in the introduction of a periodic
change of the refractive index for the DFB, where the optical mode overlaps with
this change and propagates perpendicularly to the refractive index corrugation.
In such a waveguide, there will be N reflected wavelets from N refractive index
changes, and if the phase difference between those wavelets is a multiple of 2𝜋,
the wavelets will interfere constructively, satisfying the Bragg condition (9.1):
𝜆m
Λ= (9.1)
2neff
where Λ is the grating period, 𝜆 is the emission wavelength, neff is the effective
refractive index of the laser diode structure, and m is the order of the grating.
One of the most important parameters of DFB laser diodes is the interaction
strength between the propagating optical mode and the grating as well as the
amount of the light feedback (the fraction of the light intensity reflected back at
each period), which is determined by the coupling coefficient 𝜅 [65], which can
Standard laser diode QW

Mirror
Mirror
R1 T2
T1
Optical mode R2
(a) L
Conventional DFB (center mirror-antisymmetry)
Λ n2 > n1 QW
AR coating AR coating
T1 T2
R1 R2 Optical mode R1 R2
(b) L
DFB with phase shift
Λ (ϕ/2π)Λ n2 > n1
QW AR coating
AR coating
T1 T2
(c) L
Fabricated conventional DFB
AR coating Λ n2 > n1 QW
AR coating
T1 T2
L DFF
DBF
(d)
Figure 9.5 The comparison between (a) Fabry–Pérot-type laser diode, (b) AR/AR-coated DFB
laser – often called conventional DFB designed as center mirror-antisymmetric device
(L/Λ = integer), (c) AR/AR-coated DFB laser with phase shift in the middle of the device, and (d)
more realistic “as-cleaved” DFB scheme, which includes the phase relation between the
grating and the front (DFF ) and back (DBF ) facet. Because there are very different approaches of
achieving the DFB laser by the different placement of the grating, the full structure of the laser
diode is not shown on purpose. The factor Λ represents the period of the change of the
refractive index, (𝜑/2𝜋)Λ represents the position of the phase shift, blue arrows show the
reflection of the light, and n1,2 are the effective modal indices.
be approximated by the following Eq. (9.2):

Γgrating (n22 − n21 ) ( )
w
𝜅= sin m 𝜋 (9.2)
𝜆B neff Λ
where Γgrating is the optical confinement factor of the grating, n1,2 are the effective
modal indices (as shown in Figure 9.5), w/Λ is the duty cycle, and 𝜆B is the Bragg
wavelength. It is worth pointing out that for a (Al, In)GaN-based laterally coupled
DFB, emitting, for example, at 400 nm, the first-order grating dimension would
be ∼40 nm, which is very challenging for fabrication.
Therefore, as it will be shown later, in (Al, In)GaN-based DFB laser diodes,
usually higher order gratings are implemented.
Figure 9.6 shows the spectral properties of the Fabry–Pérot and different types
of the DFB laser diodes, which were introduced in Figure 9.5.
The spectrum of a conventional DFB laser diode (Figure 9.6b) shows the degen-
eracy of the modes, which are symmetrical with respect to the Bragg wavelength.
The space between those two modes is called the stop band. The existence of the
stop band is a consequence of high reflectance of the light at each period of the
refractive index, which is highest at the Bragg wavelength. Therefore, the photon
penetration at the Bragg wavelength is the shortest.
Because the mode sees both the grating and the gain region, the Bragg wave-
length sees the least net gain, and for high 𝜅, it will not reach threshold gain. How-
ever, for wavelengths, which have zero reflectivity – from both facets and grat-
ing – the transmission is higher, but because of zero reflectivity and even though
the net gain is the highest, a resonance cannot be established. Therefore, lasing
modes will appear for wavelengths partially reflected by the grating – enough to
establish a resonance – and simultaneously enough gain to support the mode.
As it can be deduced, the lasing will occur for the wavelengths that correspond
to the edges of the stop band.
Stop band
λB
0
Intensity (dB)
Intensity (dB)
–10
–20
Fabry-Pérot Conventional DFB

–30
429.5 429.6 429.7 429.8 429.9 λ (nm)
(a) λ (nm) (b)
λB 0
Fabricated
Intensity (dB)
uncoated DFB
Intensity (dB)
Phase-shifted DFB w/o phase-shift

Stop band –10
–20
–30
418.4 418.5 418.6 418.7 418.8
(c) λ (nm) (d) λ (nm)
Figure 9.6 The comparison of the ASE spectrum (below the threshold) between (a)
Fabry–Pérot-type, uncoated laser diode (measured), (b) AR/AR-coated conventional DFB laser
(scheme), (c) AR/AR-coated DFB laser with phase shift (scheme), and (d) typical, III-N-based,
uncoated DFB laser diode (measured). Red dotted line shows the Bragg wavelength of the
grating and blue arrows show the edges of the stop band.
The dual-mode operation in a conventional refractive-index-coupled DFB

comes from the fact that in such a cavity, two equal standing wave patterns
are existing, which differ in the direction of propagation – one from back to
front and one from front to back of the device. Both share the same gain and
have the same mirror/cavity losses. However, they differ in the position of the
peak intensity in the gating structure: one of the waves has a peak intensity
aligned with the high refractive index region, whereas second wave has a peak
intensity aligned with low refractive index region. Therefore, as it can be seen in
Figure 9.6b, the effective refractive index of one of them will be higher and will
lase at longer wavelength, whereas the effective refractive index of the second
wave will be lower and will lase at a shorter wavelength. Nevertheless, such
symmetry of double degeneracy can be easily destroyed by, for example, a slight
increase of the reflectivity of one of the facets or by the uneven shape of the
active region gain spectrum. However, with this configuration, lasing will tend
to occur at one wavelength with a low side mode suppression ratio (SMSR).
The conventional grating and degeneracy in the spectrum of a DFB laser diode
is like the bandgap in the electron energy spectrum in the crystals, while here the
perturbation of the refractive index is created in 1D photonic crystal.
Therefore, to obtain a single mode at the Bragg wavelength, the periodicity of
the refractive index of the photonic crystal must be interrupted, which can be
done by the introduction of an extra period in the photonic lattice. To obtain
lasing exactly at the Bragg wavelength (Figure 9.6c), this additional feature must
be equal to half of the period of the photonic lattice – which corresponds to a
quarter-wavelength phase shift (Figure 9.5c). However, from the optical power
point of view emitted from the device, this approach is limited. On one side, the
SMSR will be highest if the 𝜆/4 shift position will be in the middle of the device
[66]. On the other side, the device will emit the same amount of optical power
from both facets, without the possibility to increase or redirect the light from the
back facet to the front – like in Fabry–Pérot type-laser diodes, where it can be
done by HR coating the device on the back facet.
Another approach, chosen by every reported (Al, In)GaN DFB laser diode, is to
fabricate the device with a standard grating, without a 𝜆/4 phase shift, and to HR
coat the back facet. With this approach, it is accepted that it is almost impossible
to cleave to the required tolerance with respect to the phase of the grating as the
corrugation period for a first-order grating for 400 nm is ∼40 nm, whereas the
cleaving precision is a few micrometers.
Such a situation is shown in Figure 9.5d, where DBF and DFF represent the back
and front facet misalignment with respect to the grating period. The phase rela-
tion of the grating corrugation and facets can be defined as:
DBF,FF ⋅ 360∘
𝜑DBF,FF = (9.3)
Λ
where DBF and DFF are the thicknesses of the misalignment of the back and front
facets with respect to the grating (as shown in Figure 9.5d). To increase the optical
power, the back facet can be covered with HR coating. However, because such
devices are sensitive to the value of the end facet phases and because those facet
phases are distributed randomly from bar to bar, the yield of the DFB laser diodes
with desirable parameters is limited.
From a historical point of view, the first DFB nitride-based laser diode was
realized by Hofmann et al. [67] in 1996 and was optically pumped based on
a second-order grating and lased at a wavelength of 390 nm. The grating was
fabricated on the top of the ridge by electron beam lithography and reactive
ion etching. After three years, the first electrically injected pulse-operating DFB
GaN laser diode was reported [68]. In this case, the grating was holographically
defined and dry-etched ridge as a third-order grating with the period of 240 nm
and with as low as 1.1. A threshold current lasing is at 402.5 nm. This time,
the realization of the grating was not implemented in the form of a tooth-like
structure on top of the ridge, but by overgrowth. After growth of a thick GaN
buffer layer, bottom cladding, bottom waveguide, QWs, and upper waveguide,
the epitaxy was stopped to fabricate a “rectangular tooth geometry with rounded
tops” grating on top of the last layer. After fabrication and matching between the
grating resonance and peak of the gain spectrum by optical pumping, the wafer
was put into the reactor to grow the rest of the layers: p-doped cladding layer
and p-doped GaN subcontact layer. It is worth mentioning that the laser diode
structure was grown on sapphire, which causes a problem with the formation
of high-quality mirrors and complicates the fabrication of low-threshold and
high-efficiency laser diodes. Therefore, one of the motivations of the develop-
ment of the DFB type of the laser was to avoid the mirror formation problem.
Because, as mentioned, in contrast to Fabry–Pérot-type lasers, the DFB can lase
in a mirrorless system.
The breakthrough came in 2006, when Masui et al. [69] from Nichia reported
a DFB GaN laser diode operating CW at room temperature with outstanding
parameters. In contrast to their previous work [70], where the grating was
formed on top of the p-side waveguide and the devices were working only under
pulse current operation, the device had first-order grating fabricated in n-type
cladding.
Moreover, this time, the authors had grown the epitaxial structure on bulk
GaN, instead of sapphire. Even though the DFB was fabricated by one of the
most complicated methods, by overgrowth, they reported obtaining “a fine
tooth-shaped first-order grating” by electron beam lithography and ICP dry
etching on the surface of the n-type cladding layer and regrowth in the rest of the
structure – including QWs. The location of the grating was not the only difficulty
in the fabricated DFB. The lowest, first order of the grating is also challenging
from the point of view of the quality and resolution of the electron beam
lithography and etching. For a 404 nm DFB GaN device of 300 μm cavity length,
a 2 μm wide ridge with an 80 nm period and 100 nm depth grating, the output
single-mode power was 60 mW at 63 mA, a threshold current of only 22 mA
(J th = 3.7 kA/cm2 ), and a slope efficiency of 1.44 W/A was obtained. However,
despite the good results and the reported lifetime (which was estimated to be
approximately 4000 hours), the DFBs were not commercialized, and for the next
decade, no significant work concerning nitride-based DFBs was published. The
reason for that could be the increase of availability of bulk GaN substrates, which
allows to switch the growth of the laser structures from sapphire. By that, the
motivation of avoiding of the need to form high-quality mirrors was no longer

driving DFB laser development. Moreover, there were not many significant
applications where the DFB laser diodes were needed and could not be replaced
by the external cavity laser diode.
Another breakthrough was made in 2018 and 2019, when many scientific
groups had reported their results of the fabrication and properties of the
nitride-based DFBs. The reason of such sudden and wide interest in (Al, In)GaN
DFB laser diodes, especially emitting at 399, 422, 457, and 461 nm, was due to the
rapid development of atomic optical clocks – especially those based on stron-
tium neutral atom (for which, the optical cooling wavelength is 𝜆 = 460.9 nm)
and ion (for which, the optical cooling wavelength is 𝜆 = 421.7 nm) [71]. Also,
nitride-based SOAs have been reported a few years earlier [72, 73], which could
have also stimulated the scientists to return to development of the DFB LD.
The most interesting aspect of those studies lies in the different approaches of
fabricating the Bragg grating. For example, Kang et al. [74] reported an electrically
pulse operating, 404.6 nm emitting, DFB based on a 10th order laterally coupled
surface grating with V-shaped grooves obtained by i-line stepper lithography and
ICP. Another interesting result was the pulse operation of an MBE-grown DFB
laser diode made by Muziol et al. [75], fabricated with a fifth order Bragg grating
on top of the ridge (see Section 8.6). In this approach, the realization of this DFB
was done, thanks to the introduction of a tunnel junction in the p-side layers,
which allows n-type overgrowth of the rest of the top structure. This thick n-type
layer at the top works as a current spreader and makes it possible to avoid the gold
electrical contact on the top of the ridge and its vicinity and instead fabricate the
grating.
A similar type of DFB grating – on top of the ridge – was realized by Zhang
et al. [76]. However, this device was grown on semipolar substrate without a tun-
nel junction. Instead of the tunnel junction, ITO was employed and processed to
obtain a first-order grating on top of it and the device lased CW at 442.6 nm with
a threshold current of 445 mA.
TopGaN and CST Global managed to obtain CW operation of a DFB
laser diode with a 39th order grating, with a relatively low threshold current
(I th = 130 mA) and 408.55 nm single-mode emission at room temperature [77].
It was based on the formation of the grating along the sidewalls of the ridge
waveguide, which is one of the simplest ways to obtain single-mode emission.
This approach has some advantages over the buried grating approach because it
does not require difficult overgrowth steps, which could introduce epi-defects.
Moreover, in contrast to a surface grating, the sidewall grating does not com-
promise the p-type top contact [78]. Also, when compared with the grating
formed by a tunnel junction or ITO, the sidewall grating approach does not
need any special technology beside electron beam lithography and can be done
on both MBE- and MOVPE-grown wafers. However, it needs to be noted that
the combination of the tunnel junction approach and sidewall grating could be
interesting from the point of view of thermal management. The introduction
of the tunnel junction on p-side allows to replace the highly resistive p-AlGaN
cladding layer by a highly conductive n-AlGaN cladding layer [79] and hence
reduces the Joule heating. It leads to smaller self-heating of the device, which
reduces the increase of the QW temperature in the device with the current, and
by that, it decreases the red shift of the gain spectrum (gain detuning). Because
the red shift of the gain spectrum is usually ∼3 times larger than the red shift of
the stop band [67, 76], it can decrease the overlap between them at a point where
fundamental Fabry–Pérot modes can start to lase. Therefore, the decrease of the
device self-heating should allow to increase the single-mode current operating
range. The situation, where a reduction of the device self-heating could improve
the maximum output optical power in single-mode DFB operation, can be seen
in Figure 9.7.
As it can be seen in Figure 9.7b, below the threshold current, the device experi-
ences a blue shift caused by the QCSE. After the threshold current is reached, the
quasi-Fermi levels are fixed, further built-in electric field screening is suspended,
and only thermal red shift remains.
Figure 9.7c shows the shift of the DFB single mode and the peak of the electro-
luminescence spectrum with the current. As the current increases, the overlap of
the gain spectrum and stop band decreases to a point where the DFB mode dis-
appears. This point is the limit of maximum optical power of the device working
as a DFB.
The decrease of the overlap is mainly caused by the increase of the device tem-
perature, which leads to the electroluminescence red shift. The DFB mode shifts
more than two times slower and, as can be seen in Figure 9.7d and Figure 9.8, the
position of the stop band and DFB mode is from the start at the short-wavelength
side of the spectrum. As a result, it is relatively easy to suppress the DFB mode
by a temperature increase. If the stop band was on the long-wavelength side of
the gain spectrum at threshold or the thermal resistance of the device would be
lower, the maximum optical power from the device would be larger.
9.3 Superluminescent Diodes

9.3.1 History of Superluminescent Diode Development
Visible light emitters based on the InAlGaN material system include not only
light-emitting diodes and laser diodes but also devices called superluminescent
diodes and SOAs. The mentioned device – superluminescent diode (SLD) – is an
emitter, which combines the features of laser diodes and light-emitting diodes.
SLD emitters utilize the stimulated emission, which means that these devices
operate at current densities similar to those of laser diodes. The main difference
between LDs and SLDs is that in the latter case, we design the device waveguide
in a special way preventing the formation of a standing wave and lasing. Still,
the presence of the waveguide ensures the emission of a high-quality light beam
with high spatial coherence of the light, but the light is characterized by low time
coherence at the same time. SLDs were first demonstrated early in 1970s based
on arsenide semiconductors [80, 81], as a device being easily coupled to optical
fiber, in contrast to standard LEDs, but at the same time showing a broad spec-
trum with small time coherence. The fundamental work of Alphonse et al. [82]
7 16
Ll 80 mA
30 Vl 15
6
75 mA
Optical power (mW)
14
5
Single-mode
Voltage (V)
73 mA
operating range 13
20 4
72 mA
12
3
70 mA
11
10 2
60 mA
Ith = 45 mA 10
1
Intensity (a.u.)
50 mA
9
0 0
0 25 50 75 100 45 mA
(a) 8
Current (mA)
44 mA
7
Peak of the EL
428
Peak of the DFB 40 mA
6
30 mA
5
426
20 mA
λ (nm)
FP lasing 4
10 mA
424 3
dλ/mA 5 mA
= 0.009 nm/mA
2
422 3 mA
dλ/mA = 0.00377 nm/mA 1
1 mA
0
420 415 420 425 430 435 440
0 10 20 30 40 50 60 70 80
(c) I (mA) (b) λ (mA)
0
I = 72 mA I = 75 mA
I = 45 mA I = 60 mA
–10
Intensity (dB)
–20
–30
–40
420 422 424 420 422 424 420 422 424 420 422 424
(d) λ (nm) λ (nm) λ (nm) λ (nm)
Figure 9.7 Optoelectrical parameters of third-order, 1000 μm long, CW working in RT DFB

laser diode mounted in TO 5.6 mm: (a) L–I–V characteristic. The green arrow shows the range
of the DFB single mode operation; (b) electroluminescence spectra for different driving
currents; (c) shift of the electroluminescence peak, and the DFB mode with current. The green
arrow shows the current at which the device starts lasing on the fundamental Fabry–Pérot
modes; (d) lasing spectra in a narrow wavelength range, and decibel scale for several pumping
currents, showing the switch from the DFB singe mode lasing into the multilongitudinal
Fabry–Pérot modes.
15 Figure 9.8 Gain spectra of the DFB

1 mA 3 mA laser diode presented in previous
10 5 mA 10 mA
figure. The red dotted line shows the
20 mA 30 mA
5 40 mA 44 mA wavelength position of the stop band.
0
Gain (cm–1)
–5
–10
–15
–20
–25
–30
420 425 430 435 440
λ (nm)
was a milestone marking the maturity of the SLD technology. The high perfor-
mance of these devices was reached because of a unique design of these emitters
utilizing a tilted-waveguide geometry with a precisely chosen inclination angle,
which effectively lowers the facet reflectivity to the values below 10−4 . The opti-
mization of these devices enabled the demonstration of 100 mW optical power
[83, 84]. The important aspect of the development was also an optimization of
spectral quality to meet the demands of applications including optical coherence
tomography (OCT) and fiber optic gyroscopes (FOGs), which benefit from the
low temporal coherence of SLDs – wide emission spectrum – and the high quality
of the light beam emitted by SLDs. In both cases, the devices should be optimized
to reach as wide and smooth emission spectrum as possible.
The first III-N-based SLD was reported in 2009 by Feltin et al. [85]. From this
time onward, these devices have undergone rapid development. The studies per-
formed by many groups were focused on both the scientific understanding of
the physical aspects of operation of those devices and the fabrication processes
[86–104]. Many of the early devices suffered from thermal issues, which led to a
strong reduction of the optical power under CW operation when compared with
pulsed mode [85–88]. When compared to laser diodes, SLDs are more suscep-
tible to thermal problems, as they require a continuous increase in the carrier
density in the whole operation regime. The first high-power CW-operated SLD
was reported in 2011 by Ohno et al. and reached 200 mW of optical power with
emission around 400 nm [89]. This was followed by more high-power reports in
the following years [92–94, 96]; a comparison of probably all literature reports is
presented in Figure 9.9a. In 2019, a record high-output power of 2000 mW under
pulsed operation was reported by Cahill et al. using a novel surface-emitting SLD
geometry [104].
Many groups achieved this result because of using a double-pass bent-
waveguide geometry, described in the next paragraph. With the achievement of
optical power above 100 mW, scientists concentrated more on the application
aspects of SLDs. This was reflected in the spectral quality of the devices.
120
2000
100
Optical power (mW)
500
80
λ2 / Δλ (μm)
400
60
300
200 40
100 20
0
0
2010 2013 2016 2019 2010 2013 2016 2019
(a) Year of publication (b) Year of publication
CW operation: Pulsed operation:

Central emission wavelength (nm): < 430 < 430
430 – 490 430 – 490
> 490 > 490
Figure 9.9 Comparison of the parameters of nitride superluminescent diodes as a function of

the year of publication: (a) optical power and (b) quality parameter for OCT. Graphs were
prepared based on the device parameters reported in [85–104].
Figure 9.9b compares the quality parameter 𝜆2 /Δ𝜆, central emission wavelength
divided by the full width at half maximum of the spectrum, which is often used
for estimation of OCT in-depth resolution (smaller parameter corresponds to
better resolution). We can see that after year 2016, the value of 𝜆2 /Δ𝜆 stopped
its clear increase, although not all the points from Figure 9.9a are presented here
because of the lack of information. In terms of the applications, extensive work
was done on visible light communication, with reports on above 1 Gbit/s data
transmission [100]. However, there are also studies concentrated on OCT appli-
cations, in the form of spectral optimization [98] and, what is most important,
the examination of direct use nitride SLD light source in an OCT system [99].
9.3.2 Basic SLD Properties

There are many possible approaches to suppress the light oscillation between
the ends of the waveguide, creating an SLD. Usually, this is done by significantly
increasing the losses of one of the waveguide. Currently, the most successful
designs of nitride SLD are bent, curved, or tilted waveguide geometries as well
as tilted facet geometries, whereas in all cases, the front end of the waveguide
meets the device facet in an inclined way, as shown in Figure 9.10. The inclined
waveguide suppresses the reflection of light from the facet to the waveguide by
directing it outside to the lossy unpumped area of the device chip. Because of
the small waveguide size, its end acts similar to a slit causing interference effects
and leading to a complicated dependence of facet reflectivity on the bend angle
of the waveguide (details of the device properties are described in [92] or [105]).
The right choice of the inclination angle enables reaching extremely low effective
reflectivity values down to the level of below 10−4 [82].
ϕ Figure 9.10 Waveguide

geometry of typical nitride
ϕ SLDs: (a) bent waveguide and
(b) tilted facet.
(a) (b)
Figure 9.11 Example of the evolution with

3 Current current of nitride superluminescent diode
(mA): emission spectra.
50
Intensity (a.u.)
2 200
500
0
410 420 430 440
Wavelength (nm)
With efficient cavity suppression, SLDs show smooth and relatively wide emis-
sion spectra. With increasing operation current, the spectrum narrows along
with stronger light amplification and often Fabry–Pérot cavity modulations
appear in the spectra. This is because the complete cavity suppression is very
difficult. The spectra of high-power devices usually have between 3 and 6 nm of
full width at half maximum. An example of the spectral evolution is presented in
Figure 9.11.
Figure 9.12 shows the comparison of the light–current dependences of a laser
diode and bent waveguide SLD fabricated on one wafer with the same total area.
We can see the characteristic exponential shape of the SLD L–I curve. This shape
is related to the exponential character of the amplitude growth in the single-pass
SLD with increasing optical gain. This character is well explained by the classical
SLD equation [106]:
Popt = Psp (𝜆){exp[(Γg(𝜆) − 𝛼i )L] − 1} (9.4)
where Popt is the optical power, Γ is the confinement factor, g is the gain coeffi-
cient, 𝛼 i are the internal losses of the waveguide, and L is the chip length. The “−1”
Figure 9.12 The light–current characteristics

60 Laser of a violet light-emitting nitride SLD compared
Optical power (mW)
SLD with the LD fabricated on the same wafer.
40
20
0
0 100 200 300 400
Current (mA)
Figure 9.13 Measured emission peak position 436

of laser diode and SLD. The additional curve Laser
SLD
shows the SLD emission wavelength corrected
Wavelength (nm)
SLD – l (dλ/dT)
for self-heating effects. 432
428
424
0 100 200 300

Current (mA)
term can often be omitted because of the large value of the exponent. Psp corre-
sponds to the guided component of spontaneous emission and can be expressed
as:
Δ𝜈 ⋅ nsp hc
Psp (𝜆) = (9.5)
𝜂 λ
where Δ𝜈 is the bandwidth of the emission, h is the Plank constant, 𝜈 is the optical
frequency, 𝜂 is the quantum efficiency, and nsp is the spontaneous emission factor.
While the initial part of the L–I curve is not surprising in terms of theoretical
predictions of Eq. (9.4), the surprising part lies in the high current part of the
behavior shown in Figure 9.12. At high current, L–I dependence assumes a linear
form, reaching the differential efficiency like a counterpart laser diode. The only
explanation of this behavior is that the optical gain in an SLD is being saturated,
causing the exponential part of Eq. (9.4) to be fixed. It is important to notice that
linearization of L–I curve is not associated with the lasing action of SLD, the
spectrum remains broad (see Ref. [93] for more details).
The characteristic feature of SLDs is the lack of Fermi pinning or in other words
the lack of electron concentration stabilization. This is visible in Figure 9.13,
which shows the emission peak position as a function of a drive current. We can
see that in the case of an LD, the emission peak wavelength shifts with increasing
current toward shorter values, marking the carrier screening of the QCSE.
After reaching the laser threshold, the emission peak starts moving toward the
red part of the spectrum because the QCSE is fixed at a certain level, and a
self-heating effect starts shrinking the semiconductor bandgap. In the case of
SLDs, we observe a steady shift of the emission line (when corrected for thermal
effects), meaning no Fermi level pinning in these devices [107]. The spectral blue
shift of the device emission may also be caused by a band-filling effect. However,
in the case of SLDs, the operating current densities are much higher than the
reported band-filling saturation regime [108].
9.3.3 Challenges for SLD Optimization

In order to reach high output power from an SLD, one needs to take care to fully
utilize the lack of Fermi-level pinning, that is to prevent the device from reach-
ing gain saturation and L–I linearization. The onset of saturation depends on the
details of the epitaxial structure design and processing. From a structure point of
view, one should ensure as high carrier injection efficiency as possible. It is also
important to keep in mind that the devices are susceptible to electron escape or
overflow, which increases with the operation current because of the increasing
temperature. If saturation is related purely to the carrier injection problem, an
elegant solution is to increase the waveguide width, which decreases the carrier
density, preferably only on the front part of the waveguide where the carrier con-
sumption by the stimulated emission process is the fastest. Currently, there have
been many reports of high optical power nitride SLDs, reaching hundreds of mW
in the blue-violet region [92–94, 96, 100, 101, 103, 104] and several to tens of mW
in the green region [101]. It seems that at this stage, the blue and violet devices
are ready to be used as light sources in projection systems.
Another challenge of SLD development, directly related to the high carrier den-
sities, is achieving long lifetime. Many studies of nitride laser diodes suggest that
aging speeds up with higher carrier densities. Unfortunately, SLDs by definition
work at higher current and carrier density, which means that the lifetime may
be expected to be shorter than that of the laser counterpart. Still, the currently
reported extremely long lifetimes of laser diodes allow being optimistic that even
with a fraction of the LD lifetime, SLDs will still be attractive from an application
point of view. Until now, Castiglia et al. demonstrated an SLD with a 5000 hour
lifetime, thanks to the optimization of the p-type doping profile [94].
In order to meet the expectations of typical SLD applications such as OCT
or FOG, a very wide emission spectrum is required. In the arsenide and phos-
phide material systems, many solutions were proposed to artificially increase the
spectral width of emission and improve the quality of the devices. The reported
approaches include fabrication of quantum wells having different widths, differ-
ent compositions, or quantum dot-active region [109, 110]. Such optimization is
not popular at this stage in the nitride SLD community, but probably will become
in the future, in order to expand the application range. One feature, related to the
nitride material system, which may be hindering this direction of development,
is the low mobility of the holes in the nitride system. This often leads to uneven
pumping of the quantum wells. Kafar et al. [95, 98] proposed an alternative solu-
tion, which utilizes the relation between the substrate misorientation and indium
incorporation to fabricate an indium content profile along the waveguide. Fur-
thermore, the first nitride SLDs with a quantum dot-active region were reported
in 2019 [102] and may mark the beginning of a new development direction.
9.4 Semiconductor Optical Amplifiers

In the previous section, we discussed the properties of SLDs, devices that do not
lase, but they utilize the mechanism of stimulated emission to amplify their own
radiation. If we inject laser light to a well-designed SLD, we can use the stimulated
emission to amplify the input light preserving its optical mode and spectral qual-
ity. Such an optical element is called a semiconductor optical amplifier (SOA) and
if combined with a seed laser is denominated an MOPA (master oscillator power
amplifier). A scheme of such a system, not monolithic, is presented in Figure 9.14.
9.4 Semiconductor Optical Amplifiers 355
Pumping laser Optical amplifier
Propagation direction
Pumping laser Optical amplifier

Input signal
Laser Active region

mirrors Output signal
Figure 9.14 Scheme of the light generation and amplification in a MOPA system.
The first electrically pumped nitride-based SOAs (intended for the ampli-
fication of an external optical signal) were realized in 2010 by Koda et al.
[111]. At that time, the main reason for the development of such a system
in the blue-violet region was to use it in a two-photon 3D optical storage
application [112]. The main advantage of an (Al, In)GaN-based MOPA system
in such application is compactness, when compared with the bulky Nd:vanadate
recording system solution. This first SOA had the same structure, including the
number and thickness of the QWs, as the mode-locked master oscillator laser.
Beside the structure of the device, one of the most important parameters of the
semiconductor-based optical amplifier is the ridge geometry, which defines the
distribution of the light and carriers in lateral and longitudinal direction in the
device. In this case, the amplifier was based on a flared geometry, where the
input and output width of the ridge was 1.4 and 15 μm, respectively. The length
of the device was 2 mm with a flared angle of 5∘ . To suppress the longitudinal
modes inside the amplifier, the whole ridge was bend with respect to the facets.
In such a configuration, and under DC operation at 5 ∘ C, the authors managed
to amplify a 3 ps optical pulse with 3 W of mode-locked master oscillator peak
power up to 103–119 W, which gives 15–16 dB of gain.
A typical direction for optoelectronic device optimization is to increase the
optical power. In case of every SOA, there is a saturation output power that marks
the highest achievable value. The saturation output power is defined as the output
optical power at which its single-pass gain decreases by half (3 dB). This phe-
nomenon is caused by the fact that the carrier injection rate, which is defined
by the current applied to the amplifier, is not fast enough when compared to the
stimulated emission rate. The current supplier sets a limit to the amount of the
electron–hole pairs, which can possibly recombine in a unit of time. By that, it
limits the rate of generated photons. In other words, under constant current oper-
ation, the amplifier provides a limited number of photons, which can be extracted
from it. Therefore, for some specific input optical power, because of the men-
tioned limit, the optical amplifier cannot provide an exponential increase of the
number of injected photons through the whole length of the device. Saturation
output power can be calculated by Eq. (9.6):
A
Psat = h𝜐 (9.6)
Γa𝜏
where A is the cross section of the active region (the width of the ridge times
thickness of the quantum well or wells), Γ is the confinement factor, a is the dif-
ferential gain, and 𝜏 is the carrier lifetime. It is clearly visible that the geometry
of the ridge is a very important parameter directly influencing the device power
limit. The saturation output power can be increased by increasing the active area
cross section, for example, by the use of a tapered of flared-shape waveguide – like
Koda et al. [111]. Moreover, this parameter can also be increased by decreasing
the confinement factor (thicker epilayers of the waveguide), differential gain, or
carrier lifetime (increase of the overlap of the electron and hole wave functions).
It is worth to note that the saturation output power does not depend on the length
of the device. It is also worth to note that the optical amplifier can even saturate
itself in the case of efficient light generation, when SOA works as an SLD.
Two years later (2012), the same authors [43] managed to increase the peak
pulse of an improved mode-locked laser diode up to 8 W (1.6 ps pulse duration;
1 GHz repetition). Next, the light was amplified by the improved SOA up to
308 W (1.9 ps pulse duration). The improvement was achieved, thanks to the
optimization of the structure and shape of the ridge of both, the mode-locked
laser diode and the optical amplifier. The improvement of the amplifier consisted
in a decrease of the optical confinement factor and an increase of the amplifier
length (from 2 to 2.5 mm). Even though the amplifier was driven in CW oper-
ation at 5 ∘ C of ambient temperature – similar to conditions reported for the
previous device – and the total output peak power increased almost three times,
the gain remained similar to the previously reported (15.9 dB). It is worth to note
that this amplifier, when working as a SLD, without the external optical signal,
achieved an optical power of around 0.5 W at a current of 2 A (𝜆 = 405 nm). This
makes this device one of the most powerful CW working SLDs up to the date of
this publication.
The latest MOPA system reported by Koda et al. achieved 1.1 kW (3 nJ) of
uncompressed optical peak power. The SOA was driven in pulse operation mode
with a current of 8 A (at a repetition rate of 100 kHz and a duty ratio of 20%) with
the amplification of an input optical pulse of 17.6 dB [72]. An improvement of the
peak power was obtained – once more – by the decrease of the confinement fac-
tor by 20%, by the increase of the amplifier length (from 2.5 to 3 mm), and by the
increase of the active area volume. The active area volume was increased through
changing the ridge shape from narrow 15 μm ridge to a flared 117 μm wide output
aperture.
The recent development of visible light communication and optical atomic
clocks has also stimulated interest in other groups. For example, Shen et al.
[113] had shown the (Al, In)GaN QW-based dual-section device consisting
of a laser diode and integrated amplifier, which was designed for visible light
communication applications. The laser diode section was 1190 μm long, the SOA
section was 300 μm long, and the ridge waveguide was 2 μm wide. The obtained
9.5 Summary 357
Table 9.1 The most relevant transition wavelengths and linewidths of

atoms and ions considered for optical clocks.
Wavelength of Wavelength of clock

cooling/detection transition and
Atom and linewidth linewidth
88
Sr 460.9 nm, 32 MHz 698.4 nm, 1 mHz
88
Sr+ 421.7 nm, 20.2 MHz 674.0 nm, 0.4 Hz
40
Ca+ 396.8 nm, 23.4 MHz 729.1 nm, 140 mHz
24
Mg 285.2 nm, 79 MHz 457.1 nm, 36 Hz
171
Yb 398.9 nm, 29 MHz 578.4 nm, 10 mHz
171
Yb+ 369.5 nm, 19.6 MHz 435.5 nm, 3.1 Hz
(quadrupole transition)
Bolded data show the wavelengths that are in the wavelength range of (Al,
In)GaN-based modern laser diodes.
Source: The data for Mg are taken from Goncharov et al. 2014 [114] and for the
rest of the atoms/ions are taken from Ludlow et al. 2015 [71].
increase of optical power was from 8.2 to 30.5 mW, giving an amplification
of 5.7 dB for 404.3 nm. Moreover, a Gbps rate of data transmission was also
achieved by this device. The reported results strongly support the concept
of using an LD–SOA-integrated device in future visible light communication
systems.
Another important application of nitride SOAs (together with the
nitride-based DFB LDs) are for optical atomic clocks. The interest lies mostly in
atomic clocks based on neutral atoms (and their isotopes) and their ions such
as 87 Sr, 88 Sr, 88 Sr+ , 40 Ca+ , 24 Mg, 171 Yb, and 171 Yb+ . The most relevant transition
wavelengths and linewidths are shown in Table 9.1. Many of the cooling or
clock transitions lie in the wavelength range, which can be obtained by (Al,
In)GaN-based devices. By that, it creates the possibility of replacing solid-state
laser light sources or frequency-doubled GaAs- or InP-based laser diodes, by
more compact III-N systems. With this motivation in mind, our group has
reported the development of the blue optical amplifier with a double “j-shape”
waveguide for MOPA systems [73]. The amplifier was 2.5 mm long, with three
QWs emitting at 450 nm. It was CW operated at room temperature. Our results
obtained 29 dB amplification, with a saturation output power higher than
20 dBm and high stability, showing that there is a big potential for nitride-based
MOPA optical systems in modern and compact optical atomic clocks.
9.5 Summary
The revolution that was brought the advent of III-N semiconductors extends
much beyond LED technologies and lighting. The possibility of fabricating laser
diodes emitting from deep UV to red based on a single material system is very
unique and opens many new applications and improves the existing ones. The
story of nitride lasers started more than 20 years ago, motivated by the improve-
ment of classical optical data storage. Today, nitride laser diodes are needed for
car headlights, micro and movie projectors, printing technologies, quantum tech-
nologies, and many others. Together with better lasers, we have developed other
optical components such as optical amplifiers and SLDs and maybe in near future
also photonic integrated circuits. The future of nitride laser optoelectronics looks
indeed bright!
References
1 Biard, J.R. and Pittman, G.E. (1966). Semiconductor radiant diode. US Patent
3,293,513, filed 8 August 1962 and issued 20 December 1966.
2 Maiman, T.H. (1960). Optical and microwave-optical experiments in ruby.
Phys. Rev. Lett. 4 (11): 564–566.
3 Basov, N.G., Krokhin, N., and Popov, Y.M. (1961). Possibility of using indi-
rect transitions to obtain negative temperatures in semiconductors. J. Exp.
Theor. Phys. 12 (5): 1033.
4 Quist, T.M., Rediker, R.H., Keyes, R.J. et al. (1962). Semiconductor maser of
GaAs. Appl. Phys. Lett. 1 (4): 91–92.
5 Hall, R.N., Fenner, G.E., Kingsley, J.D. et al. (1962). Coherent light emission
from GaAs junctions. Phys. Rev. Lett. 9 (9): 366–368.
6 Holonyak, N. and Bevacqua, S.F. (1962). Coherent (visible) light emission
from Ga(As1−x Px ) junctions. Appl. Phys. Lett. 1 (4): 82–83.
7 Nathan, M.I., Dumke, W.P., Burns, G. et al. (1962). Stimulated emission of
radiation from GaAs p–n junctions. Appl. Phys. Lett. 1 (3): 62–64.
8 Kroemer, H. (1963). A proposed class of hetero-junction injection lasers.
Proc. IEEE 51 (12): 1782–1783.
9 Kroemer, H. (1967). Solid state radiation emitters. US Patent 3,309,553, filed
16 August 1963 and issued 14 March 1967.
10 Alferov, Z.I., Andreev, V.M., Garbuzov, D.Z. et al. (1971). Investigations of
the influence of the AlAs-GaAs heterostructure parameters on the laser
threshold current and the realization of continuous emission at room tem-
perature. Sov. Phys. 4 (9): 1573–1575.
11 Hayashi, I., Panish, M.B., Foy, P.W., and Sumski, S. (1970). Junction lasers
which operate continuously at room temperature. Appl. Phys. Lett. 17 (3):
109–111.
12 van der Ziel, J.P., Dingle, R., Miller, R.C. et al. (1975). Laser oscillation from
quantum states in very thin GaAs−Al0.2 Ga0.8 as multilayer structures. Appl.
Phys. Lett. 26 (8): 463–465.
13 Dupuis, R.D., Dapkus, P.D., Holonyak, N. et al. (1978). Room-temperature
laser operation of quantum-well Ga(1−x) Alx as-GaAs laser diodes grown by
metalorganic chemical vapor deposition. Appl. Phys. Lett. 32 (5): 295–297.
14 Onton, A. and Foster, L.M. (1972). Indirect, Γ8 v -X 1 c , band gap in
GaAs1−x Px . J. Appl. Phys. 43 (12): 5084–5090. https://doi.org/10.1063/1
.1661076.
References 359
15 Adachi, S. (1992). Physical Properties of III-V Semiconductor Compounds:

InP, InAs, GaAs, GaP, InGaAs, and InGaAsP. Chichester: Wiley.
16 Gunshor, R.L. and Nurmikko, A.V. (eds.) (1997). II-VI Blue/Green Light
Emitters: Device Physics and Epitaxial Growth. San Diego: Academic Press.
17 Adachi, M., Yukitake, H., Watanabe, M. et al. (2002). Mechanism of
slow-mode degradation in II–VI wide bandgap compound based blue-green
laser diodes. Phys Status Solidi (b) 229 (2): 1049–1053.
18 Juza, R. and Hahn, H. (1938). Über die Kristallstrukturen von Cu 3 N, GaN
und InN Metallamide und Metallnitride. Z. Anorg. Allg. Chem. 239 (3):
282–287.
19 Pankove, J.I., Miller, E.A., and Berkeyheiser, J.E. (1972). GaN blue
light-emitting diodes. J. Lumin. 5 (1): 84–86.
(LEEBI). Jpn. J. Appl. Phys. 28 (part 2, no. 12): L2112–L2114.
tion mechanism of P-type GaN films. Jpn. J. Appl. Phys. 31 (part 1, no. 5A):
1258–1266.
23 Nakamura, S., Senoh, M., and Mukai, T. (1993). P-GaN/N-InGaN/N-GaN
double-heterostructure blue-light-emitting diodes. Jpn. J. Appl. Phys. 32 (part
2, no.1A/B): L8–L11.
24 Nakamura, S., Senoh, M., Iwasa, N. et al. (1995). Superbright green InGaN
single-quantum-well-structure light-emitting diodes. Jpn. J. Appl. Phys. 34
(part 2, no. 10B): L1332–L1335.
25 Narukawa, Y., Niki, I., Izuno, K. et al. (2002). Phosphor-conversion white
light emitting diode using InGaN near-ultraviolet chip. Jpn. J. Appl. Phys. 41
(part 2, no. 4A): L371–L373.
26 Narukawa, Y., Narita, J., Sakamoto, T. et al. (2007). Recent progress of high
efficiency white LEDs. Phys. Status Solidi (a) 204 (6): 2087–2093.
27 Chichibu, S., Azuhata, T., Sota, T., and Nakamura, S. (1996). Spontaneous
emission of localized excitons in InGaN single and multiquantum well
28 Nakamura, S., Senoh, M., Nagahama, S. et al. (1996). InGaN-based
multi-quantum-well-structure laser diodes. Jpn. J. Appl. Phys. 35
(part 2, no. 1B): L74–L76.
29 Nakamura, S., Senoh, M., Nagahama, S. et al. (1996). Ridge-geometry
InGaN multi-quantum-well-structure laser diodes. Appl. Phys. Lett. 69
(10): 1477–1479.
30 Nakamura, S., Senoh, M., Nagahama, S. et al. (1996). Room-temperature
continuous-wave operation of InGaN multi-quantum-well structure laser
31 Nakamura, S., Senoh, M., Nagahama, S. et al. (1997). Room-temperature
continuous-wave operation of InGaN multi-quantum-well structure laser
diodes with a lifetime of 27 hours. Appl. Phys. Lett. 70 (11): 1417–1419.
32 Nakamura, S., Senoh, M., Nagahama, S. et al. (1997). High-power,

long-lifetime InGaN multi-quantum-well-structure laser diodes. Jpn. J. Appl.
Phys. 36 (part 2, no. 8B): L1059–L1061.
33 Kapolnek, D., Keller, S., Vetury, R. et al. (1997). Anisotropic epitaxial lateral
growth in GaN selective area epitaxy. Appl. Phys. Lett. 71 (9): 1204–1206.
34 Nakamura, S. (1999). InGaN/GaN/AlGaN-based laser diodes grown on
epitaxially laterally overgrown GaN. J. Mater. Res. 14 (7): 2716–2731.
35 Itaya, K., Onomura, M., Nishio, J. et al. (1996). Room temperature pulsed
operation of nitride based multi-quantum-well laser diodes with cleaved
facets on conventional C-face sapphire substrates. Jpn. J. Appl. Phys. 35
(part 2, no. 10B), L1315: –L1317.
36 Kuramata, A., Domen, K., Soejima, R. et al. (1997). InGaN laser diode grown
on 6H-SiC substrate using lo-pressure metal organic vapor phase epitaxy.
Jpn. J. Appl. Phys. 36 (part 2, no. 9A/B: L1130–L1132.
37 Kneissl, M., Bour, D.P., Johnson, N.M. et al. (1998). Characterization of
AlGaInN diode lasers with mirrors from chemically assisted ion beam
etching. Appl. Phys. Lett. 72 (13): 1539–1541.
38 Brown, J.D., Swindell, J.T., Johnson, A.L. et al. (1998).
Nitride-semiconductors-symposium. Mater. Res. Soc. 482: 1179–1184.
39 Motoki, K., Okahisa, T., Matsumoto, N. et al. (2001). Preparation of large
freestanding GaN substrates by hydride vapor phase epitaxy using GaAs as a
starting substrate. Jpn. J. Appl. Phys. 40 (part 2, no. 2B): L140–L143.
40 Goto, S., Ohta, M., Yabuki, Y. et al. (2003). Super high-power
AlGaInN-based laser diodes with a single broad-area stripe emitter fabri-
cated on a GaN substrate. Phys. Status Solidi (a) 200 (1): 122–125.
41 Perlin, P., Marona, L., Holc, K. et al. (2011). InGaN laser diode mini-arrays.
42 Brüninghoff, S., Eichler, C., Tautz, S. et al. (2009). 8 W single-emitter InGaN
laser in pulsed operation. Phys. Status Solidi (a) 206 (6): 1149–1152.
43 Koda, R., Oki, T., Kono, S. et al. (2012). 300 W peak power picosecond
optical pulse generation by blue-violet GaInN mode-locked laser diode and
semiconductor optical amplifier. Appl. Phys. Express 5 (2): 022702.
44 Kawaguchi, M., Imafuji, O., Nozaki, S. et al. (2016). Optical-loss suppressed
InGaN laser diodes using undoped thick waveguide structure. In: Proceed-
ings Volume 9748, Gallium Nitride Materials and Devices XI, 974818. San
Francisco, CA.
45 Strauss, U., Somers, A., Heine, U. et al. (2017). GaInN laser diodes from
440 to 530 nm: a performance study on single-mode and multi-mode R&D
designs. In: Proceedings SPIE 10123, Novel In-plane Semiconductor Lasers
XVI, 101230A. San Francisco, California (20 February 2017).
46 Nakamura, S., Senoh, M., Nagahama, S. et al. (2000). Blue InGaN-based
laser diodes with an emission wavelength of 450 nm. Appl. Phys. Lett. 76 (1):
22–24.
47 Miyoshi, T., Masui, S., Okada, T. et al. (2009). 510–515 nm InGaN-based
green laser diodes on c-plane GaN substrate. Appl. Phys. Express 2: 1,
062201–3.
References 361
48 Enya, Y., Yoshizumi, Y., Kyono, T. et al. (2009). 531 nm green lasing of
InGaN based laser diodes on semi-polar {20-21} free-standing GaN sub-
strates. Appl. Phys. Express 2: 1, 082101–3.
49 Avramescu, A., Lermer, T., Müller, J. et al. (2010). True green laser diodes at
524 nm with 50 mW continuous wave output power on c-plane GaN. Appl.
Phys. Express 3 (6): 061003.
50 Yoshida, H., Yamashita, Y., Kuwabara, M., and Kan, H. (2008). A 342-nm
ultraviolet AlGaN multiple-quantum-well laser diode. Nat. Photonics 2 (9):
551–554.
51 Kirste, R., Guo, Q., Dycus, J.H. et al. (2018). 6 kW/cm2 UVC laser threshold
in optically pumped lasers achieved by controlling point defect formation.
52 Hirayama, H., Yatabe, T., Noguchi, N., and Kamata, N. (2010). Development
of 230-270 nm AlGaN-based deep-UV LEDs. Electron. Commun. Jpn. 93 (3):
24–33.
53 Wunderer, T., Jeschke, J., Yang, Z. et al. (2017). Resonator-length depen-
dence of electron-beam-pumped UV-A GaN-based lasers. IEEE Photonics
Technol. Lett. 29 (16): 1344–1347.
54 Zhang, Z., Kushimoto, M., Sakai, T. et al. (2019). A 271.8 nm
deep-ultraviolet laser diode for room temperature operation. Appl. Phys.
Express 12 (12): 124003.
55 Wang, Q., Nguyen, H.P.T., Cui, K., and Mi, Z. (2012). High efficiency ultravi-
olet emission from Alx Ga1−x N core-shell nanowire heterostructures grown
on Si (111) by molecular beam epitaxy. Appl. Phys. Lett. 101 (4): 043115.
56 Zhang, M., Banerjee, A., Lee, C.-S. et al. (2011). A InGaN/GaN quantum dot
green (𝜆 = 524 nm) laser. Appl. Phys. Lett. 98 (22): 221104.
57 Frost, T., Banerjee, A., Sun, K. et al. (2013). InGaN/GaN quantum dot red 𝜆
= 630 nm laser. IEEE J. Quantum Electron. 49 (11): 923–931.
58 Murayama, M., Nakayama, Y., Yamazaki, K. et al. (2018). Watt-class green
(530 nm) and blue (465 nm) laser diodes. Physica Status Solidi (a) 215 (10):
1700513.
59 Northrup, J.E. (2009). GaN and InGaN(1122) surfaces: group-III adlayers and
indium incorporation. Appl. Phys. Lett. 95 (13): 133107.
60 Yoshizumi, Y., Adachi, M., Enya, Y. et al. (2009). Continuous-wave opera-
tion of 520 nm green InGaN-based laser diodes on semi-polar {20-21} GaN
substrates. Appl. Phys. Express 2 (9): 092101.
61 Kelchner, K.M., Lin, Y.-D., Hardy, M.T. et al. (2009). Nonpolar
AlGaN-cladding-free blue laser diodes with InGaN waveguiding. Appl.
Phys. Express 2: 1, 071003–3.
62 Tyagi, A., Farrell, R.M., Kelchner, K.M. et al. (2010). AlGaN-cladding free
green semipolar GaN based laser diode with a lasing wavelength of 506.4
nm. Appl. Phys. Express 3 (1): 011002.
63 Kogelnik, H. and Shank, C.V. (1972). Coupled-wave theory of distributed
feedback lasers. J. Appl. Phys. 43 (5): 2327–2335.
64 Kogelnik, H. and Shank, C.V. (1971). Stimulated emission in a periodic struc-
ture. Appl. Phys. Lett. 18 (4): 152–154.
65 Venghaus, H. and Grote, N. (eds.) (2017). Fibre Optic Communication: Key

Devices. Cham: Springer International Publishing.
66 Coldren, L.A., Corzine, S.W., and Mashanovitch, M. (2012). Diode Lasers
and Photonic Integrated Circuits. Hoboken, NJ: Wiley.
67 Hofmann, R., Gauggel, H.-P., Griesinger, U.A. et al. (1996). Realization of
optically pumped second-order GaInN-distributed-feedback lasers. Appl.
Phys. Lett. 69 (14): 2068–2070.
68 Hofstetter, D., Thornton, R.L., Romano, L.T. et al. (1999). Characterization
of InGaN/GaN-based multi-quantum well distributed feedback lasers. MRS
Internet J. Nitride Semicond. Res. 4 (S1): 69–74.
69 Masui, S., Tsukayama, K., Yanamoto, T. et al. (2006). First-order AlInGaN
405 nm distributed feedback laser diodes by current injection. Jpn. J. Appl.
Phys. 45 (29): L749–L751.
70 Masui, S., Tsukayama, K., Yanamoto, T. et al. (2006). CW operation of the
first-order AlInGaN 405 nm distributed feedback laser diodes. Jpn. J. Appl.
Phys. 45 (46): L1223–L1225.
71 Ludlow, A.D., Boyd, M.M., Ye, J. et al. (2015). Optical atomic clocks. Rev.
Mod. Phys. 87 (2): 637–701.
72 Kono, S., Koda, R., Kawanishi, H., and Narui, H. (2017). 9-kW peak power
and 150-fs duration blue-violet optical pulses generated by GaInN master
oscillator power amplifier. Opt. Express 25 (13): 14926.
73 Stanczyk, S., Kafar, A., Grzanka, S. et al. (2018). 450 nm (Al, In)GaN opti-
cal amplifier with double ‘j-shape’ waveguide for master oscillator power
amplifier systems. Opt. Express 26 (6): 7351.
74 Kang, J.H., Wenzel, H., Hoffmann, V. et al. (2018). DFB laser diodes based
on GaN using 10th order laterally coupled surface gratings. IEEE Photonics
Technol. Lett. 30 (3): 231–234.
75 Muziol, G., Turski, H. , Hajdel, M. et al. (2019). III-N tunnel junctions
as an enabling technology for efficient distributed feedback laser diodes.
13th International Conference on Nitride Semiconductors 2019 (ICNS-13)
(7–12 July 2019), Bellevue, Washington.
76 Zhang, H., Cohen, D.A., Chan, P. et al. (2019). Continuous-wave operation of
a semipolar InGaN distributed-feedback blue laser diode with a first-order
indium tin oxide surface grating. Opt. Lett. 44 (12): 3106.
77 Slight, T.J., Stanczyk, S., Watson, S. et al. (2018). Continuous-wave operation
of (Al, In)GaN distributed-feedback laser diodes with high-order notched
gratings. Appl. Phys. Express 11 (11): 112701.
78 Pearton, S.J., Lee, J.W., MacKenzie, J.D. et al. (1995). Dry etch damage in
InN, InGaN, and InAlN. Appl. Phys. Lett. 67 (16): 2329–2331.
79 Piprek, J. (2017). Internal power loss in GaN-based lasers: mechanisms and
remedies. J. Opt. Quant. Electron. 49 (10): 329.
80 Kurbatov, L.N., Shakhidzhanov, S.S., Bystrova, L.V. et al. (1971). Investigation
of superluminescence emitted by a gallium arsenide diode. Sov. Phys. 4 (11):
1739.
81 Lee, T.-P., Burrus, C., and Miller, B. (1973). A stripe-geometry
double-heterostructure amplified-spontaneous-emission (superluminescent)
diode. IEEE J. Quantum Electron. 9 (8): 820–828.
References 363
82 Alphonse, G.A., Gilbert, D.B., Harvey, M.G., and Ettenberg, M. (1988).

High-power superluminescent diodes. IEEE J. Quantum Electron. 24 (12):
2454–2457.
83 Zang, Z., Minato, T., Navaretti, P. et al. (2010). High-power (>110 mW)
superluminescent diodes by using active multimode interferometer. IEEE
84 Zang, Z., Mukai, K., Navaretti, P. et al. (2012). Thermal resistance reduc-
tion in high power superluminescent diodes by using active multi-mode
interferometer. Appl. Phys. Lett. 100 (3): 031108.
85 Feltin, E., Castiglia, A., Cosendey, G. et al. (2009). Broadband blue super-
luminescent light-emitting diodes based on GaN. Appl. Phys. Lett. 95 (8):
081107.
86 Hardy, M.T., Kelchner, K.M., Lin, Y.-D. et al. (2009). m-plane GaN-based
blue superluminescent diodes fabricated using selective chemical wet etch-
ing. Appl. Phys. Express 2 (12): 121004.
87 Rossetti, M., Dorsaz, J., Rezzonico, R. et al. (2010). High power blue-violet
superluminescent light emitting diodes with InGaN quantum wells. Appl.
Phys. Express 3 (6): 061002.
88 Holc, K., Marona, Ł., Czernecki, R. et al. (2010). Temperature dependence
of superluminescence in InGaN-based superluminescent light emitting diode
structures. J. Appl. Phys. 108 (1): 013110.
89 Ohno, H., Orita, K., Kawaguchi, M. et al. (2011). 200 mW GaN-based super-
luminescent diode with a novel waveguide structure. IEEE Photonic Society
24th Annual Meeting, Arlington, VA, USA (9–13 October 2011), 505–506.
90 Kafar, A., Stańczyk, S., Grzanka, S. et al. (2012). Cavity suppression in
nitride based superluminescent diodes. J. Appl. Phys. 111 (8): 083106.
91 Kopp, F., Lermer, T., Eichler, C., and Strauss, U. (2012). Cyan superlumi-
nescent light-emitting diode based on InGaN quantum wells. Appl. Phys.
Express 5 (8): 082105.
92 Kopp, F., Eichler, C., Lell, A. et al. (2013). Blue superluminescent
light-emitting diodes with output power above 100 mW for Picoprojection.
Jpn. J. Appl. Phys. 52 (8S): 08JH07.
93 Kafar, A., Stanczyk, S., Targowski, G. et al. (2013). High-optical-power
InGaN superluminescent diodes with “j-shape” waveguide. Appl. Phys.
Express 6 (9): 092102.
94 Castiglia, A., Rossetti, M., Matuschek, N. et al. (2016). GaN-based super-
luminescent diodes with long lifetime. In: Proceedings SPIE 9748, gallium
nitride materials and devices XI, 97481V . San Francisco, California,
(26 February 2016).
95 Kafar, A., Stanczyk, S., Sarzynski, M. et al. (2016). Nitride superlumines-
cent diodes with broadened emission spectrum fabricated using laterally
patterned substrate. Opt. Express 24 (9): 9673.
96 Shen, C., Ng, T.K., Leonard, J.T. et al. (2016). High-brightness semipolar
(2021) blue InGaN/GaN superluminescent diodes for droop-free solid-state
lighting and visible-light communications. Opt. Lett. 41 (11): 2608.
97 Shen, C., Lee, C., Ng, T.K. et al. (2016). High-speed 405-nm superlumines-
cent diode (SLD) with 807-MHz modulation bandwidth. Opt. Express 24
(18): 20281.
98 Kafar, A., Stanczyk, S., Sarzynski, M. et al. (2018). InAlGaN superlumines-
cent diodes fabricated on patterned substrates: an alternative semiconductor
broadband emitter: publisher’s note. Photonics Res. 6 (6): 652–652.
99 Goldberg, G.R., Boldin, A., Andersson, S.M.L. et al. (2017). Gallium nitride
superluminescent light emitting diodes for optical coherence tomography
applications. IEEE J. Sel. Top. Quantum Electron. 23 (6): 1–11.
100 Alatawi, A.A., Holguin-Lerma, J.A., Kang, C.H. et al. (2018). High-power
blue superluminescent diode for high CRI lighting and high-speed visible
light communication. Opt. Express 26 (20): 26355–26364.
101 Rossetti, M., Castiglia, A., Malinverni, M. et al. (2018). 3-5: RGB superlu-
minescent diodes for AR micro-displays. SID Symp. Dig. Tech. Pap. 49 (1):
17–20.
102 Wang, L., Wang, L., Yu, J. et al. (2019). Abnormal Stranski–Krastanov mode
growth of green InGaN quantum dots: morphology, optical properties, and
applications in light-emitting devices. ACS Appl. Mater. Interfaces 11 (1):
1228–1238.
103 Shen, C., Holguin-Lerma, J.A., Alatawi, A.A. et al. (2019). Group-III-nitride
superluminescent diodes for solid-state lighting and high-speed visible light
communications. IEEE J. Sel. Top. Quantum Electron. 25 (6): 1–10.
104 Cahill, R., Maaskant, P.P., Akhter, M., and Corbett, B. (2019). High power
surface emitting InGaN superluminescent light-emitting diodes. Appl. Phys.
Lett. 115 (17): 171102.
105 Kafar, A., Stanczyk, S., Schiavon, D. et al. (2020). Review—review on opti-
mization and current status of (Al, In)GaN superluminescent diodes. ECS J.
Solid State Sci. Tech. 9 (1): 015010.
106 Henry, C. (1986). Theory of spontaneous emission noise in open resonators
and its application to lasers and optical amplifiers. J. Lightwave Technol. 4
(3): 288–297.
107 Kafar, A., Stanczyk, S., Grzanka, S. et al. (2019). Screening of
quantum-confined stark effect in nitride laser diodes and superluminescent
diodes. Appl. Phys. Express 12 (4): 044001.
108 Ji, Y., Liu, W., Erdem, T. et al. (2014). Comparative study of field-dependent
carrier dynamics and emission kinetics of InGaN/GaN light-emitting diodes
grown on (1122) semipolar versus (0001) polar planes. Appl. Phys. Lett. 104
(14): 143506.
109 Lin, C.-F. and Lee, B.-L. (1997). Extremely broadband AlGaAs/GaAs superlu-
minescent diodes. Appl. Phys. Lett. 71 (12): 1598–1600.
110 Li, L.H., Rossetti, M., Fiore, A. et al. (2005). Wide emission spectrum from
superluminescent diodes with chirped quantum dot multilayers. Electron.
Lett. 41 (1): 41–43.
111 Koda, R., Oki, T., Miyajima, T. et al. (2010). 100 W peak-power 1 GHz rep-
etition picoseconds optical pulse generation using blue-violet GaInN diode
laser mode-locked oscillator and optical amplifier. Appl. Phys. Lett. 97 (2):
021101.
References 365
112 Walker, E., Dvornikov, A., Coblentz, K., and Rentzepis, P. (2008). Terabyte
recorded in two-photon 3D disk. Appl. Opt. 47 (22): 4133.
113 Shen, C., Ng, T.K., Lee, C. et al. (2018). Semipolar InGaN quantum-well
laser diode with integrated amplifier for visible light communications. Opt.
Express 26 (6): A219.
114 Goncharov, A.N., Bonert, A.E., Brazhnikov, D.V. et al. (2014). Precision spec-
troscopy of Mg atoms in a magneto-optical trap. Quantum Electron. 44 (6):
521–526.
367
10
Green and Blue Vertical-Cavity Surface-Emitting Lasers

Yang Mei, Rong-Bin Xu, Huan Xu, and Bao-Ping Zhang
Xiamen University, School of Electronic Science and Engineering, Department of Electronic Engineering,
Laboratory of Micro/Nano-Optoelectronics, 422 Siming South Road, Xiamen, Fujian 361005, China
10.1 Introduction
10.1.1 Properties and Application of GaN VCSELs
Vertical-cavity surface-emitting lasers (VCSELs) are a type of semiconductor
lasers with their laser beams propagating vertically to the layer surface/interface,
in contrast with traditional edge-emitting lasers (EELs) that emit light from the
edge of a chip. This property provides VCSELs distinct advantages including
circular low-divergent output beams, wafer-level testing, and densely packed
two-dimensional arrays [1]. Benefiting from the short cavity length and small size
of active region, VCSELs could also guarantee low-threshold single-longitudinal
mode lasing with better temperature stability on the lasing wavelength and a
much higher modulation speed than EELs [2]. Different from well-established
GaAs and InP-based VCSELs that emit in the infrared regime [3], GaN-based
VCSELs can provide available emission wavelengths in the ultraviolet to visible
regime so that more applications could benefit from such a light source. One
important application of GaN VCSELs is the solid-state lighting (SSL) [4]
because the lasers can provide peak efficiencies at much higher current densities
and may overcome the efficiency droop limitations in the commercial blue
LEDs [5–7]. GaN VCSELs can offer circular-symmetric and directional beam,
which can be more readily captured and focused, leading to the possibility
of new and more compact systems. High-speed visible light communication
(VLC) [8] is another important application for GaN VCSELs because of the
much higher modulation speed than LEDs. GaN VCSELs are also essential
for next-generation full-color displays [9], pico projectors [10], and near-eye
displays [11] because of their tunable wavelengths from ultraviolet to visible
and the low power consumption. In addition, GaN VCSELs also have promising
applications in high-density optical storage [12], biosensors [13], atomic clocks
[14], and medical treatment [15].

368 10 Green and Blue Vertical-Cavity Surface-Emitting Lasers
10.1.2 Brief History and Current Status of GaN VCSELs

The concept of vertical-cavity surface-emitting lasers (VCSELs) was firstly
proposed by Prof. Kenichi Iga and his colleagues in 1977 [16]. The first VCSEL
device was demonstrated in 1979 for GaAs material system [17] and then was
soon commercialized in 1994. However, the development of GaN VCSELs is
much slower because the growth of the GaN film was still very difficult at that
time until Nakamura [18] developed the two-step growth method and solved
the problems of p-doping in GaN in the 1990s. The earliest research related to
GaN-based VCSEL dates back to 1995. Honda et al. [19] theoretically studied
the threshold characteristics of UV GaN VCSELs. In 1996, Redwing et al. [20]
reported the first optically pumped GaN VCSEL with a fully epitaxial VCSEL
structure consisting of a 10 μm GaN bulk layer active region sandwiched between
30 period Al0.12 Ga0.88 N/Al0.4 Ga0.6 N-based DBRs. The relatively low reflectivity
of the DBRs and the employment of a thick bulk GaN gain layer resulted in
the high-threshold pumping energy of ∼2.0 MW/cm2 . Then, in the following
10 years, optically pumped GaN VCSELs with an InGaN quantum well (QW)
active layer and different DBR structures were demonstrated by researchers
in different groups including Tokyo University [21–23], Russian Academy of
Sciences (RAS) [24], Brown University [25], NTT Corporation [26], Seoul
National University (SNU) [27], University of California Santa Barbara (UCSB)
[28], National Chiao Tung University (NCTU) [29–33], École Polytechnique
Fédérale de Lausanne (EPFL) [34], Xiamen University (XMU) [35–37], etc. After
the realization of optically pumped GaN VCSELs, more attention has been paid
to the electrically pumped GaN VCSELs for practical applications. The first
electrically injected GaN-based VCSEL was reported by the NCTU in 2008 [38].
It had one bottom epitaxial AlN/GaN DBR, one top dielectric DBR, and a 240 nm
indium tin oxide (ITO) layer as a current-spreading layer. Continuous-wave
(CW) lasing was realized under 77 K at a wavelength of 462.8 nm. In the same
year, Nichia Corporation [39] demonstrated room temperature (RT) CW lasing
of GaN VCSEL with double dielectric DBRs fabricated by laser lift off (LLO) and
wafer-bonding technique. Lasing action was realized at 414 nm with a threshold
current density of 13.9 kA/cm2 , and the output power was 0.14 mW. Since then,
GaN VCSELs entered a rapid development stage, and lasing under electrically
injection has been demonstrated by many groups including both companies
such as Nichia [40, 41], Panasonic [42], SONY [43–45], Stanley [46–48], and
academic research groups including NCTU [49, 50], UCSB [51–57], EPFL [34],
XMU [58–61], Meijo [62–66], National Taiwan University (NTU) [67], and
University of New Mexico [68]. To date, the device performance has improved
greatly when compared with the first demonstration of electrically pumped
VCSEL. In 2015, SONY [43] reported GaN VCSEL with double dielectric DBRs
fabricated by the epitaxial lateral overgrowth (ELO) method. The emission
wavelength of the device was 454 nm, and the output power exceeded 1 mW for
the first time. In 2018, Stanley and Meijo [46] reported GaN VCSEL emitting
at 441 nm with an output power exceeding 6 mW. The lasing action could be
maintained under 110 ∘ C. The device had an epitaxial AlInN/GaN bottom DBR
and a dielectric top DBR, and a buried SiO2 guide layer was utilized to confine
both current and lateral optical fields. Later in the same year, by increasing the
cavity length to improve thermal dissipation, the output power was further
increased to 16 mW [47]. In 2019, Stanley [48] reported the 16×16 blue VCSEL
array with a total output power of 1.19 W, taking GaN VCSELs one step near
to the practical application stage. For the GaN VCSELs emitting at a longer
wavelength, XMU [60] reported InGaN quantum dot (QD)-based VCSELs that
pushed the emission of GaN VCSELs to green. The emission wavelength of the
device extended from 491.8 to 565.7 nm, covering the “green gap.”
10.1.3 GaN VCSELs with Different DBRs

The one main reason that GaAs VCSELs could be commercialized rapidly is that
the high reflective and electrically conductive semiconductor-based DBRs could
be easily grown because AlAs and GaAs are almost lattice matched. Therefore,
the entire VCSEL layer structure could be sufficiently formed by a single step
of epitaxial growth. However, it is challenging in GaN VCSELs to incorporate
the cavity with high reflective DBRs, particularly semiconductor-based DBRs.
So far, the two mainstream methods to fabricate GaN VCSELs with high reflec-
tive DBRs have been developed. One is using an epitaxial nitride semiconductor
DBR as the bottom DBR and a dielectric top DBR, the so-called hybrid DBR
structure [34, 38, 40, 41, 49], as shown in Figure 10.1a, and the other is using an
existing dielectric (e.g. SiO2 /ZrO2 ) DBR with a high reflectivity of 99.9% at both
the top and the bottom of GaN-based VCSELs, the so-called double dielectric
DBRs structure [37, 39, 41, 42, 51], as shown in Figure 10.1b–d, but both meth-
ods have their challenges. For GaN VCSELs with hybrid DBR structure, the use of
the epitaxial bottom DBR makes the fabrication processes simple and can accu-
rately control the cavity length, but the growth of nitride DBR with high crystal
quality and high reflectivity remains challenging. For GaN VCSELs with double
dielectric DBRs structure, it is easy to realize very high reflectivity because of
the large refractive index difference between the dielectric layers. However, to
form the DBR on the substrate side, an additional complicated fabrication pro-
cess, such as bonding a supporting substrate on the top side and removing the
original substrate for epitaxial growth [39, 41, 42, 51], is usually required to com-
plete the vertical structure. In this section, the development and current status
of GaN VCSELs with these two kinds of structures are discussed.
Dielectric DBR Electrode

ITO SiO2 n-GaN p-GaN
p-GaN p-GaN
p-GaN
n-GaN n-GaN n-GaN
Epitaxial DBR Bonding
GaN subtrate
Sapphire subtrate Si subtrate GaN subtrate
(a) (b) (c) (d)
Figure 10.1 Structures of GaN VCSELs with (a) hybrid DBRs. (b) Double dielectric DBRs
fabricated by substrate removal and transfer. (c) Double dielectric DBRs fabricated by ELO.
(d) Double dielectric DBRs containing a curved bottom mirror on the back of substrate and
planner top mirror.
10.1.3.1 GaN VCSELs with Hybrid DBR Structure

Similar to AlGaAs/GaAs DBRs in infrared VCSELs, AlGaN/GaN DBRs have
been developed for GaN VCSELs [3]. However, there is a large lattice mismatch
up to 2.4% and a large difference in thermal expansion coefficient between AlN
and GaN. This will induce significant tensile strain in epitaxial layers during
growth and cooling and then leads to the formation of cracks in AlGaN/GaN
DBRs with high Al mole fraction. Decreasing Al mole fraction in the AlGaN
layer can help to reduce the strain accumulation during DBR growth and to
suppress the crack formation, but the refractive index contrast will also be
decreased and more DBR pairs are needed to maintain the high reflectivity.
Thus, the peak reflectivity of such DBR is limited to 99% as previously reported
[69–73]. Actually, the first reported optically pumped GaN VCSEL by Redwing
et al. [20] in 1996 was featured with Al0.12 Ga0.88 N/Al0.4 Ga0.6 N-based DBRs, not
only bottom but also the top DBR. However, the relatively low reflectivity of
the DBRs induced a very high threshold pumping energy of ∼2.0 MW/cm2 .
In 1999, Arakawa and coworkers [23] reported optically pumped hybrid DBR
structure GaN VCSEL with a 43-pair Al0.34 GaN/GaN bottom DBR and a 15-pair
ZrO2 /SiO2 top DBR. InGaN MQWs were utilized as an active region and the
threshold energy was 43 nJ. Next year, in 2000 [25], the optically pumped hybrid
DBR VCSEL with a 30-pair Al0.25 Ga0.75 N/GaN bottom DBR and a HfO2 /SiO2
top DBR were reported by researchers in Brown University. In 2005 [29], NCTU
reported an optically pumped hybrid DBR VCSEL with the nitride bottom DBR
consisting of 25-pair AlN/GaN layers. To obtain a crack-free and high reflectivity
AlN/GaN DBR, AlN/GaN superlattices were inserted into the AlN/GaN DBR
structure during the epitaxial growth to reduce the biaxial tensile strain. The first
electrically pumped GaN VCSEL reported by Wang and coworkers [38] in 2008
also have a similar structure. On the other hand, AlInN/GaN DBR is another kind
of nitride DBR developed for GaN VCSELs with hybrid DBR structure. AlInN is
lattice-matched with GaN when the InN mole fraction is of about 18%, so that
the huge tensile strain could be avoided during the growth of Al0.82 In0.18 N/GaN
DBR. In 2007 [74], researchers from the University of Southampton reported an
optically pumped hybrid DBR VCSEL with a 39.5-pair Al0.82 In0.18 N/GaN bottom
DBR, and the reflectivity of the DBR exceeded 99%. Then, in 2012, researchers
from EPFL [34] reported an electrically pumped hybrid DBR VCSEL with a
41.5-pair AlIn0.2 N/GaN bottom DBR. Although AlInN/GaN DBR is promising
for GaN VCSELs, the epitaxial growth of AlInN is an issue because of the
markedly different MOVPE growth parameters for AlN [75] and InN [76]. It is
very difficult to choose optimum growth parameters for the epitaxial growth of
the AlInN alloy and accurately control of the InN mole fraction. In addition, the
growth rate of AlInN crystal is also very low (∼0.2 μm/h) [77]. In recent years,
researchers from Meijo University and Stanley Corporation have paid much
efforts in AlInN/GaN DBR-based GaN VCSELs and considerable progress has
been made. By optimizing growth parameters, Al0.82 In0.18 N/GaN DBRs with
high crystal quality and high reflectivity exceeding 99.9% was successfully grown
with a relative high growth rate of ∼0.5 μm/h [63]. In the years from 2016 to
2019, they reported a series of results about Al0.82 In0.18 N/GaN DBR-based hybrid
structure GaN VCSELs including hybrid GaN VCSELs with lateral guiding [46],
with conductive Al0.82 In0.18 N/GaN bottom DBR [66], with high output power
[47], and hybrid GaN VCSEL array with total output power exceeding 1 W [48].
Recently, hybrid structure GaN VCSELs with air-gap [78] or porous [68] GaN
bottom DBR have been reported. The air-gap or porous GaN was used as a low
refractive index layer so as to provide a large contrast in the refractive index
owing to the low refractive index of air. However, the fabrication of such kind
of DBR is difficult to control. The higher thermal resistance due to air is also
expected to deteriorate the temperature characteristics of VCSELs.
10.1.3.2 GaN VCSELs with Double Dielectric DBR Structure

Benefiting from the higher refractive index contrast in dielectric material sys-
tems, it is easy to fabricate DBRs with a high reflectivity and a broad stopband
without having to deposit too many pairs. However, to realize the double dielec-
tric DBR configuration in GaN VCSELs, the device usually needs to be flip-chip
bonded on a suspending substrate and the original substrate is then removed to
allow the deposition of the DBR on the substrate side. The removal of original sub-
strate could be achieved by LLO [38, 39], chemical mechanical polishing (CMP)
[40, 41], and photoelectron chemical etching (PEC) of a sacrificial layer [79]. In
the years between 1996 and 2009, researchers from Brown University [25], Seoul
National University [27], NTT Corporation [26], NCTU [26, 29–33], and XMU
[35–37] have reported optically pumped double dielectric DBR VCSELs with dif-
ferent dielectric materials including SiO2 /HfO2 , SiO2 /ZrO2 , and SiO2 /TiO2 . The
first electrically pumped GaN VCSEL with double dielectric DBR structure was
reported by Nichia Corporation in 2008 [39]. The laser structure was epitaxially
grown on a sapphire substrate. To fabricate the double dielectric structure, the
wafer was bonded on a Si suspending substrate and the sapphire was removed by
LLO. In the following years, Panasonic [39], XMU [58–61], and UCSB [52–55]
also reported electrically pumped GaN VCSELs with a similar structure. In 2015,
SONY reported a GaN VCSEL with double dielectric DBRs fabricated by ELO,
through which the complicated substrate transfer process is no longer needed
[44]. In this case, the bottom dielectric DBR was at first deposited and patterned
on a GaN substrate, and then, the substrate was put into the metal–organic chem-
ical vapor deposition (MOCVD) furnace to grow the following epitaxial laser
structure, and the bottom dielectric DBR was buried in the n-GaN layer. Cav-
ity length could be accurately controlled in this structure because all the cavity
layers are finished by MOCVD growth. The schematic structure of the electrically
pumped double dielectric DBR VCSEL fabricated through ELO is illustrated in
Figure 10.1c. In 2018, they further proposed double dielectric DBR-based GaN
VCSEL with a new structure that also do not need the substrate removal and
transfer processes [45]. The device contains an atomically smooth curved dielec-
tric mirror that is monolithically formed on a thinned GaN substrate surface and
a planar dielectric mirror on the device top surface, as shown in Figure 10.1d.
Resonators cladded with curved and planar mirrors are known to form stable
cavities without diffraction loss, giving the beam waist just on the planar mirror,
so that lasing was realized although the device has a relatively long cavity length
of ∼30 μm.
10.2 Efficiency of Heat Dissipation of Different Device

Structures
Because of the small volume and high current density, self-heating in the active
region is inevitable. Thermal effect will cause degradations of both materials and
device properties, including the gain of active medium, threshold current, maxi-
mum output power, and spectrum linewidth [80, 81]. More seriously, it makes
GaN VCSELs unstable and shortens the device lifetime, which impedes them
from practical applications. It is essential for GaN VCSELs to have a good thermal
dissipation ability so that the heat could be extracted out of the cavity effec-
tively. In this section, the thermal characteristics of GaN VCSELs with three typ-
ical structures were studied by finite element method (FEM) using a steady-state
quasi three-dimensional (3D) cylindrical heat dissipation model. Thermal prop-
erties including heat flux, thermal resistance, and internal temperature distri-
bution were analyzed in these structures. Several methods to improve thermal
dissipation of GaN VCSELs were proposed.
10.2.1 Simulation of Heat Profile of the Device

Thermal characteristics of GaN VCSELs with hybrid DBR structure (defined
as structure A), with double dielectric DBR structure fabricated by substrate
removal and transfer technique (defined as structure B), and with double
dielectric DBR structure fabricated by ELO (defined as structure C) are studied.
In the simulation model, heat energy is assumed to be extracted out of the
device mainly by conduction through the substrate, and the interfaces between
the device and air were assumed to be thermally insulated. The GaN VCSELs
discussed here are featured with intracavity contact and the current does not
flow through the DBRs with large resistances. Hence, all the heat in the device is
assumed to be produced in the active region by nonradiative recombination and
the free carrier absorption of photons, as well as in the p-GaN by Joule heating.
Because the thickness of p-GaN layer is very small when compared with the
thickness of n-GaN and substrate, we lump all heat sources into a single source
placed at the active region during simulation. The steady-state quasi-3D heat
dissipation model is given by [82, 83].
−∇(k∇T) = Q (10.1)
k∇T = h(Tinf − T) (10.2)
where Q, k, T, h, and T inf are the heat source density, thermal conductivity,
temperature in device, heat transfer coefficient, and heat sink temperature
(293 K), respectively. Assuming a uniform current flow in the active region, the
heat source density Q at the threshold could be obtained by Q = U th ⋅ I th /V ,
where U th , I th , and V are threshold voltage, threshold current, and volume of
active region defined by the current confinement aperture, respectively. Here, we
assume that all the electrical power injected to the device was converted to heat
energy because the wall plug efficiency of GaN VCSELs is still low at present. In
this study, U th was set to be 4.5 V and I th was set to be 8 mA. The cavity length
10.2 Efficiency of Heat Dissipation of Different Device Structures 373
in the three structures was the same, and other detailed information about the
devices including the thickness and thermal conductivity of each layer could be
found in Ref. 84.
For GaN VCSELs with structure A, AlN/GaN DBR and AlInN/GaN DBR are
two kinds of epitaxial bottom nitride DBRs most commonly used. However,
these two different kinds of DBRs will affect the thermal characteristics of the
devices very much because the thermal conductivities of these materials are dif-
ferent (200 and 4.5 W/mK for AlN and Al0.82 In0.18 N, respectively). Figure 10.2a,b
shows the heat flux profile of structure A with AlN/GaN and AlInN/GaN bottom
DBRs, respectively, which are zoomed in to present the temperature distribution
near-active region more clearly. The temperature rise ΔT inside the active region
is 14.3 and 29 K for these two DBRs, respectively, corresponding to a thermal
resistance (Rth ) of 397 and 806 K/W calculated by Rth = ΔT/(U th ⋅I th ). Device with
Al0.82 In0.18 N/GaN DBR is featured with an Rth nearly twice of that with AlN/GaN
DBR, which is mainly due to the low thermal conductivity of Al0.82 In0.18 N/GaN
DBR in vertical direction. The heat flux profile of GaN VCSELs with structures
B and C are shown in Figure 10.2c,d, respectively. The temperature rises are
39 and 37 K, corresponding to an Rth of 1083 and 1027 K/W, respectively. For
structure B, heat transfer in vertical direction, which is the main path for heat to
be conducted to the substrate, is impeded by both dielectric bottom DBR and
SiO2 current confinement layer because of their low thermal conductivities, as
illustrated by the heat flux in Figure 10.2c. Similarly, structure C has a dielectric
bottom DBR, and heat transfer directly through DBR to substrate is also difficult.
However, there is no SiO2 current confinement layer beneath the active region
in structure C, allowing the heat energy to spread in lateral direction and then
be conducted to the substrate bypassing the bottom DBR, as illustrated in
Figure 10.2d. Generally, GaN VCSELs of double dielectric DBR structure are
featured with a larger Rth than that of the hybrid DBR structure because of the
low thermal conductivity of the dielectric bottom DBR.
10.2.2 Dependence of Rth on Cavity Length

Cavity length is an important parameter for GaN VCSELs and its influence on
Rth is discussed in this section. Thermal characteristics of GaN VCSELs with dif-
ferent cavity lengths were studied, and the thickness of p-GaN in all devices was
fixed to be 120 nm, and only the thickness of n-GaN was changed to obtain dif-
ferent cavity lengths in the simulation. Rth of GaN VCSELs with structures A, B,
and C as a function of n-GaN thickness is shown in Figure 10.3. The Rth of all
devices turned out to be decreased with a longer cavity. The Rth of structure A
with Al0.82 In0.18 N/GaN DBR is decreased by ∼43% from 825 to 468 K/W when
the thickness of n-GaN is increased from 1 to 6 μm. In structures B and C, the Rth
is decreased by ∼45% from 1033 to 565 K/W and by ∼44% from 988 to 549 K/W,
respectively. The heat fluxes of structure A with AlIn0.18 N/GaN DBR, structure B,
and structure C are illustrated in Figure 10.4a–f. A thicker n-GaN layer improves
the thermal dissipation in all structures because GaN has large thermal conduc-
tivity that benefits the out-diffusion of thermal energy from the active region.
This effect is more pronounced in structures where the thermal conductivity of
307 322
p-contact TOP DBR p-contact 321.95
TOP DBR 306.95
305.52 318.98
Heat source SiO2 Heat source 316.01
304.09
302.66 316.04
301.23 310.07
AlN/GaN Bottom DBR AllnN/GaN Bottom DBR
299.8 307.1
298.36 304.13
GaN substrate 296.93 GaN substrate 301.16
2 μm 295.5 298.2
294.07 295.23
293 293
(a) (b) 293
293
331 329
p-contact
Top DBR
Top DBR 329.43
328.9 325.69
SiO2 325.02 Heat source SiO2 321.95
321.14 318.22
Bottom DBR p-contact 317.26 Bottom DBR 314.48
313.38 310.75
309.5 307.01
Si substrate 305.61
GaN substrate 303.27
301.73
299.54
297.85
295.8
293 293
293 293
(c) (d)
Figure 10.2 Heat flux in GaN VCSELs with (a) structure A with AlN/GaN DBR. (b) Structure A with AlInN/GaN DBR. (c) Structure B. (d) Structure C.
1100
Structure A-AlN/GaN DBR
1000 Structure A-AllnN/GaN DBR
Structure B
Thermal resistance (K/W)
900 Structure C
800
700
600
500
400
1 2 3 4 5 6
n-GaN thickness (μm)
Figure 10.3 Thermal resistance as a function of the thickness of n-GaN for GaN VCSELs of
structures A, B, and C.
the bottom DBR is low. In structure C, thermal energy can be transferred by

passing the bottom DBR much easier with a thicker n-GaN, enhancing the ther-
mal dissipation. According to the simulation results, increasing cavity length is
an effective method to optimize thermal dissipation of GaN VCSELs. By utilizing
this method, Stanley Corporation and Meijo University have doubled the maxi-
mum output power to 16 mW in their AlInN DBR-based VCSELs [47].
10.3 Green VCSELs Based on InGaN QDs

10.3.1 Advantages of QDs Compared with QWs
InGaN QWs have been successfully used in blue LEDs and laser diodes. The
405–450 nm laser diodes containing InGaN QWs as active layers have already
been commercialized. However, in green region, the QW structure usually suf-
fers from low emission efficiency, which is also known as the “green gap.” Higher
indium content in the InGaN QW layer is necessary to extend the emission wave-
length longer to green. To enhance the indium incorporation, the QWs usually
need to be grown at low temperature, at which the crystal quality is not very
high. Meanwhile, because of the large lattice constant mismatch between GaN
and InN, higher indium content will lead to more defects and a larger strain
induced built-in electric field. Defects are nonradiative recombination centers
and built-in electric field causes quantum-confined Stark effect (QCSE), which
pulls apart electrons and holes to different sides of the quantum well, reducing
their recombination efficiency [85]. In addition, the large effective mass of the
carriers in the GaN-based material system leads to a higher transparent carrier
density, which is another impediment to produce a high-performance green laser
with InGaN QWs.
322 329 328
Top DBR 321.95 329.25 327.7
Top DBR Top DBR
1 μm 318.98
1 μm
325.53 324.14
316.01 321.81 1 μm 320.58
AlInN/GaN
313.04 Bottom DBR 318.09 Bottom DBR 317.02
Bottom DBR
310.07 314.38 313.46
307.1 310.66 309.9
304.13 306.94 306.35
303.22 302.79
301.16
299.51 299.23
298.2
295.79 295.67
295.23 293 293
293 293 293
293
(a) (b) (c)
309 313
Top DBR Top DBR
313
Top DBR
309.46 312.85 312.74
307.77 310.82 310.71
306.08 308.78 308.69
6 μm 304.39 6 μm 306.74 6 μm 306.66
302.71 304.71 304.64
301.02 302.67 302.61
299.33 300.64 300.59
AlInN/GaN 297.64 298.6 Bottom DBR 298.57
295.95 Bottom DBR 296.56 296.54
Bottom DBR
294.27 294.53 296.52
293 293 293
293 293 293
(d) (e) (f)
Figure 10.4 (a–c) Heat flux in GaN VCSEL of structures A, B, and C when the thickness of n-GaN is 1 μm, respectively. (d–f ) Heat flux in the GaN VCSEL of
structures A, B, and C when the thickness of n-GaN is 6 μm, respectively.
For QD structures, the situation is different. It is well known that using QDs
as an active region could effectively reduce the threshold current in laser devices
[86]. GaN-based QDs are zero-dimensional nanoscale crystal in which electrons
and holes are confined in a small volume, inducing a unique atomic-like dis-
crete states with a 𝛿-function like density of states. Very high gain peak could
be expected in QD structures because of such kind of density states. Moreover,
if the QD is small enough, the carrier population in higher sub-band could be
ignored and the emission property of QDs is similar to the two atomic-level sys-
tem. When QDs are used as the active media of semiconductor lasers, the lasing
property is finally independent of the effective mass of the carriers. Hence, the
use of QDs is more effective in GaN-based lasers to achieve low threshold current
[87]. Apart from this, the localization of carriers in QDs can impede them from
being captured by defects and dislocations. The QDs used as the active region
of light-emitting devices are usually grown by Stranski–Krastanov (SK) mode,
which is driven by the strain, so that the strain remaining in the QD can be signif-
icantly reduced when compared with the case of two-dimensional QW epitaxial
layer. The strain relaxation enables the reduction of QCSE that pegs back radia-
tive recombination. These properties indicate the potential to fabricate VCSELs
emitting in the “green gap” by employing InGaN QDs as the active region.
10.3.2 Growth and Optical Properties of InGaN QDs

The InGaN self-assembled QD wafer used in this work was epitaxially grown
on a (0001)-oriented sapphire substrate by a MOCVD system [88]. The epi-
layers include, in order, undoped GaN, n-type GaN, two layers of InGaN/GaN
QDs, a p-type AlGaN electron-blocking layer, and p-type GaN. The GaN cap
layers on QDs were deposited using a two-step method: firstly, a 2-nm-thick
low-temperature grown GaN matrix layer was deposited at the same growth
temperature (670 ∘ C) of QDs to protect them from decomposition during
subsequent temperature ramping process. Then, the temperature was ramped
up to 850 ∘ C, and finally, a 8-nm-thick GaN barrier layer was grown. The indium
content of the QDs is about 27%. Other detailed growth procedures are available
in Ref. [88]. Figure 10.5a shows the atomic force microscopy (AFM) image of the
uncapped InGaN QDs. The QDs with a density of ∼1.5 × 1010 cm−2 are aligned
along the step edge, and the diameter of the QDs ranges from 20 to 60 nm, while
the average height of QDs is about 2.5 nm. The QD wafer shows a broad spon-
taneous emission spectrum between 450 and 600 nm, as shown in Figure 10.5b.
The large linewidth of the PL spectrum is attributed to the inhomogeneity in
the QD size and the fluctuation of the indium content. On the other hand, the
broad gain spectrum of the QDs is expected to allow VCSELs lasing at different
wavelengths by adjusting cavity length. The cavity length of VCSELs determines
the wavelength of cavity longitudinal modes and could also shift the optical field
in the cavity and thereby modulate the coupling strength between the active
layer and the optical fields of the cavity modes. Low-temperature (4 K) spatially
resolved cathode-luminescence (CL) property of a single-layer capped InGaN
QD sample was studied, as shown in Figure 10.5c. The single-layer capped QD
sample was grown under the same growth condition as the QD wafer used to
Height 3 nm
PL
Room temperature
Intensity (a.u.)
450 475 500 525 550 575 600
0
(a) 2 μm (b) Wavelength (nm)
Energy (eV)
2.44 2.42 2.40 2.38 2.36 2.34 2.32 2.30
1.6
2.5 mW
Normalized CL intensity (a.u.)
Normalized PL intensity (a.u.)

0.9 W 1.0
1.2 W 4.7 mW
4K 11.7 mW
1.2 21.9 mW
0.8
43.3 mW
0.8 0.6
0.4
0.4
0.2
0.0
0.0
505 510 515 520 525 530 535 540 495 510 525 540 555 570
(c) Wavelength (nm) (d) Wavelength (nm)
Figure 10.5 (a) AFM image of the uncapped InGaN QDs. (b) Room-temperature PL spectrum
of the QD wafer. (c) Low-temperature CL spectra of a single-layer capped InGaN QD sample
under different excitation power. (d) Low-temperature normalized power-dependent PL
spectra of the QD wafer.
fabricate VCSELs. The diameter of the focus spot of the CL measurement was
200 nm. Sharp emission peaks with a linewidth of less than 3 meV were observed
in the CL spectra at 2.3653 and 2.3811 eV, and the higher energy peak becomes
dominant with an increased excitation power. These sharp peaks are believed
to come from the exciton and biexciton emission of a single QD because of the
𝛿-like state density. Figure 10.5d shows the normalized PL spectra of the QD
wafer under different excitation power at 5 K. With increasing pump level, the
PL spectra exhibit a slight blue shift. However, it is worth to note that the shift of
the peak wavelength is mainly caused by the broadening of spectra at the shorter
wavelength side, while the blue shift of the longer wavelength side is negligi-
ble. This phenomenon indicates that the QCSE in the QD wafer is weak, and
the broadening of the spectrum is mainly caused by the band-filling effect. If
the QCSE plays a role, the screening effect of the QCSE with increasing exci-
tation power will induce the blue shift of the whole PL spectra at both short-
and long-wavelength sides [89]. The weak QCSE in the QD sample is mainly
attributed to the stress releasing during the QD growth, which enhances the over-
lap of electron and hole wave functions and allows a higher internal quantum
efficiency. Temperature-dependent PL measurement shows that the QD wafer
used in this work has a large localization energy of 105.9 meV and a high inter-
nal quantum efficiency of 41.1% [90]. The strong carrier localization suppresses
nonradiative recombination by impeding carriers from being captured by defects

outside the QDs.
10.3.3 Fabrication Process of VCSELs

The InGaN QD-based GaN VCSEL is featured with the double dielectric DBR
structure. We first deposited a 100-nm-thick SiO2 current insulation layer on the
surface of the epitaxial wafer and a 10-μm diameter current injection aperture
was opened. Then, a 30-nm-thick indium tin oxide (ITO) Ohmic contact and
current-spreading layer was deposited, followed by the formation of a Cr/Au
p-electrode and a 12.5-pair TiO2 /SiO2 bottom dielectric DBR. The DBR was
selectively deposited on the current confinement aperture. After that, the sample
was flip-chip bonded on a copper plate through the metal-bonding technique,
and the sapphire substrate was removed by LLO process. An inductively coupled
plasma (ICP) etching and CMP were adopted to remove the undoped GaN layer
and then the n-GaN layer. At last, an n-contact metal and an 11.5-pair TiO2 /SiO2
top dielectric DBR was deposited. A schematic structure of the green InGaN
QD VCSEL is shown in Figure 10.6.
10.3.4 Properties of QD Green VCSELs

Figure 10.7a,b shows the RT CW lasing spectra of one QD VCSEL, as well as
the corresponding light output–current–voltage (L–I–V ) curves. A single longi-
tudinal mode lasing at 560.4 nm reaching to yellow-green region was observed.
The emission intensity increases dramatically above the threshold and the L–I
curve shows a clear kink at 0.61 mA, corresponding to a threshold current den-
sity of 0.78 kA/cm2 . Benefits from the broad emission spectrum of the QD wafer,
lasing at different wavelengths could be obtained in devices with different cavity
lengths. Figure 10.7c shows the lasing spectra obtained from different VCSELs
made from the same quantum dot wafer. Lasing wavelength extends from 480 to
565 nm, and a substantial part of the “green gap” has been covered. Figure 10.7d
shows the threshold current as a function of wavelength of GaN VCSELs reported
by different groups. The InGaN QD VCSELs shown in this work are featured with
not only longer emission in green but also a much lower threshold current. We
mainly attribute the low threshold lasing in green to the utilization of QDs as the
active layer, as well as the low optical loss in the cavity.
Figure 10.6 Schematic structure of QD VCSEL.
Top DBR
n-Electrode
SiO2
ITO
Active layer
Bottom DBR
Copper substrate
(p-Electrode)
10
7
Intensity (a.u.)
0.16 nm 6
8
Output power (μW)

Intensity (a.u.)
Voltage (V)
6 4
559.8 560.4 561.0
Wavelength (nm) 1.96 Ith 3
4
Near Ith 2
0.82 Ith Ith = 0.61 mA 1

2
0
500 520 540 560 580 600 0.0 0.3 0.6 0.9 1.2 1.5 1.8
(a) Wavelength (nm) (b) Current (mA)
Threshold current density (kA/cm2)

560 nm565 nm
479 nm 507 nm Nichia
Normalized intensity (a.u.)
546 nm 100 NCTU

Panasonic
UCSB
492 nm
EPFL
Sony
Meijo
10 XMU
Stanley
UNM
QD VCSELs
480 500 520 540 560 580 400 420 440 460 480 500 520 540 560
(c) Wavelength (nm) (d) Wavelength (nm)
Figure 10.7 (a) Lasing spectra of the QD VCSEL emitting at 560.4 nm. (b) L–I–V curve of the QD
VCSEL. (c) Lasing spectra obtained from different VCSELs made from the same QD wafer.
(d) Threshold current as a function of wavelength of GaN VCSELs reported by different groups.
10.4 Green VCSELs Based on Cavity-Enhanced Emission

of Localized States in Blue Emitting InGaN QWs
10.4.1 Cavity Effect
In a micro-optical cavity, light is confined in a small volume. When the
light-emitting medium is placed inside a small resonant cavity, the spontaneous
emission rate at cavity modes will be boosted, which is known as the optical
cavity effect (Purcell effect). Devices based on micro-optical cavities are already
indispensable for a wide range of applications and studies [91]. The Purcell
factor, that is, the magnitude of the spontaneous emission enhancement, can be
given by:
( )
3 𝜆 3Q
P= 2
(10.3)
4𝜋 n V
where 𝜆/n is the wavelength within the cavity material of refractive index n and
Q and V are the quality factor and mode volume of the cavity, respectively. It is
well known that the probability for spontaneous emission (SE) is proportional
to the matrix element of the initial and final electron state and proportional to
the optical mode density [92]. The optical mode density in a Fabry–Pérot res-
onator is strongly enhanced for on-resonance wavelengths. As a consequence,
10.4 Green VCSELs 381
on-resonance transitions of the microcavity devices are enhanced. In addition,

owing to the cavity effect, on-resonance optical transitions have a much shorter
lifetime when compared with SE without a cavity. The influence of the cavity on
the emission dynamics of InGaN/GaN QWs has been studied. The emission life-
times are around 155, ∼100, and <25 ps (the system limit) for wafer only, wafer in
half-cavity, and wafer in full-cavity structures, respectively [93]. Note that the car-
riers recombine much faster when the spontaneous lifetime is small, and the non-
radiative Shockley–Read processes will have less impact on the emission property
of the QWs. In case of a silicon-vacancy center in diamond embedded within a
monolithic optical cavity, 10-fold lifetime reduction and 42-fold enhancement in
emission intensity were observed when the cavity is tuned into resonance with the
optical transition of the vacancy [94]. Blue LEDs based on InGaN/GaN MQWs
have recently shown high brightness characteristics in spite of the tremendous
density of dislocations. The widely accepted viewpoint is that the phase separa-
tion during QW growth leads to the formation of indium-rich localization centers
[95]. These localization centers are quantum dot-like structures that can localize
carriers and impeded them from being captured by nonradiative recombination
centers. Thus, high external quantum efficiency could be obtained in InGaN/GaN
LEDs [96]. In this section, a green VCSEL with the In0.18 Ga0.82 N/GaN QW active
region is presented. Lasing in green region is realized, although the main emission
peak of In0.18 GaN QWs locates in the blue region. This is believed to be induced
by the indium-rich localization centers with combination of the cavity effect.
10.4.2 Properties of Cavity-Enhanced Green VCSELs

The device is featured with a double dielectric DBR structure, which is similar
to the InGaN QD green VCSEL presented in Section 10.3.3. Figure 10.8 shows
the simulated reflection spectrum of the cavity and the measured EL spectrum
of the device without top DBR. Because the EL spectrum is approximately equal
Reflection spectrum
EL spectrum
100
80
Reflectivity (%)
Intensity (a.u.)
60
40
20 Emission from In-rich

localization centers
0
390 420 450 480 510
Wavelength (nm)
Figure 10.8 Calculated reflection spectrum of the microcavity and measured EL spectrum of
the device without top DBR. EL spectrum was plotted in logarithmic scale on the y axis.
to the gain distribution, several cavity modes are covered by the broad emission
band, satisfying the gain-cavity alignment. The emission spectrum is modulated
by the cavity formed between the bottom DBR and the top GaN/air interface.
For the blue emission in the spectrum, it originates from two-dimensional QWs.
On the other hand, the green emission originates from indium-rich localization
centers caused by the condensation of indium in the InGaN/GaN QWs during
growth, which are similar to zero-dimensional structures such as quantum dots
or color centers [97].
Figure 10.9a shows the EL spectra measured from a device with full cavity under
different injected currents. With a resonant cavity, the emission of cavity modes
with on-resonance wavelengths covering both blue and green bands is strongly
enhanced. Generally, the carriers in the QW can freely move in the well plane and
be easily captured by the localization centers. At relatively small currents, the blue
emission is stronger than the green. This is similar to the case with no cavity and
can be understood when considering the low density of localized centers. With
increasing current, much more carriers can be captured and recombine through
the localization centers, and the emission has a much faster growth rate. Eventu-
ally, the intensity of the emission peaks in near green shows superliner increase
with injection current, and the lasing action is achieved. Near-field patterns of the
device at 20 and 40 mA are shown in Figure 10.9b,c, respectively. The green light
emitted from current aperture is strongly enhanced when the threshold condi-
tion is reached. On the other hand, the leakage light from the circular DBR edge,
which is due to SE without the cavity effect, is always blue. This phenomenon
demonstrates unambiguously that the green emission originates from the cavity
effect. Figure 10.10a shows the L–I–V curves of the device. The device exhibits an
output power of ∼178 μW and a threshold current (I th ) of ∼32 mA. The degree of
(a)
(b) 20 mA
50 mA
35 mA
20 mA
Intensity (a.u.)
10 mA ITO aperture
15 μm
(c) 40 mA
400 425 450 475 500

Wavelength (nm)
Figure 10.9 (a) EL spectra of the device under different currents. Near-field patterns at
(b) 20 mA and (c) 40 mA. The images were taken under low-gain settings to avoid the charge
coupled device (CCD) saturation. In the photographs, the luminous circles from inside to
outside are ITO aperture, the interior boundary of n-type electrode, and the top DBR edge,
respectively.
10.4 Green VCSELs 383
12
180 2000
10
150
Output power (μW)

1600
8
Intensity (a.u.)
120
Voltage (V)
6 1200
90
4
60 P = 71 %
800
2
Ith = 32 mA 30 Experiment
400 Fitting
0 0
0 10 20 30 40 50 0 30 60 90 120 150 180
(a) Current (mA) (b) Polarizer angle (°)
Figure 10.10 (a) L–I–V characteristics of the VCSEL with a 15-μm-diameter current aperture
under CW operation at 300 K. (b) Polarization characteristics of the laser emission at an
injection current of 1.09 Ith .
In(Normalized intensity) (a.u.)

0
In(Normalized intensity) (a.u.)
440 nm 440 nm
1000 ps
510 nm 510 nm
2000 ps
4000 ps –1
Intensity (a.u.)
–1 τ-initial stage
–2
200 400 600 800 1000 1200
Time (ps)
–2
τ - final stage
400 425 450 475 500 525 550 575 600 0 1000 2000 3000 4000 5000 6000
(a) Wavelength (nm) (b) Time (ps)
Figure 10.11 (a) Emission spectra at different times after the excitation pulse. (b) TRPL curves
of blue and green emission regions.
polarization of the emission spectrum is of around 71% under 1.09 I th , as shown

in Figure 10.10b.
Time-resolved photoluminescence (TRPL) of the devices without top DBR
was measured at RT. The emission spectra at different times after the excitation
pulse and the decay curves of blue and green emission regions are shown in
Figure 10.11. For the decay curves, the normalized PL intensity of the blue and
green region are plotted on a logarithm scale, as shown in Figure 10.11b. There
is a short time range at the initial stage, and a relatively long time range, the
final stage, where the curve can be fitted with a single exponential function.
The 𝜏-initial stage and the 𝜏-final stage represent the lifetime at the initial and
the final stage, respectively. In Kim’s previous studies [98, 99], the radiative
recombination carrier lifetime 𝜏 r could be calculated by the lifetime of the initial
and final stage, and the relationship can be written as:
2𝜏initial × 𝜏final
𝜏r = (10.4)
𝜏final − 𝜏initial
Here, 𝜏 initial and 𝜏 final are fitted to be 124 and 7075 ps for the decay curve
of green emission and 252 and 4296 ps for the decay curve of blue emission,
respectively. Then, 𝜏 r values of the green emission and blue emission are
calculated to be 252 and 535 ps, respectively. The much smaller 𝜏 r in green
region means a higher radiative recombination efficiency in indium-rich
localization centers. This benefits from their QD-like carrier localization and
confinement. The cavity effect also helps to reduce the carrier lifetime and
increase the emission rate. However, the emission intensity of the green region
is still smaller than the blue region. This is probably caused by the low density
of the localization centers. After the fabrication of the full cavity structure, the
cavity effect is more pronounced and the emission lifetime of excitons in the
deep localized luminescence centers is expected to be significantly shortened
when the emission coincides with the cavity mode. This effect can compensate
the low density of emission centers, resulting in stronger emission intensity, and
eventually, lasing in green region can be realized under high current injection.
Moreover, the simulation result shows that the active region and the antinode
of the optical field for the 493-nm cavity mode show a good spatial alignment.
A better spatial overlap or coupling between the cavity mode and the emitting
center means a larger gain enhancement factor, which can effectively decrease
the threshold gain of the VCSEL. The gain enhancement factor, defined as the
averaged optical density in the active layer normalized to that in the cavity, is
given by [100]
L∫d |E(z)|2 dz
a
Γr = (10.5)
da ∫L |E(z)|2 dz
where L, da , and E(z) are the cavity length, thickness of the active region, and
optical field standing-wave pattern, respectively. Γr is estimated to be 1.82 for the
493 nm mode inside the cavity. Such a good coupling between the cavity mode
and the emitting center has a crucial effect on the lasing characteristics of the
device.
10.5 Dual-Wavelength Lasing Based on QD-in-QW

Active Structure
10.5.1 Characteristics of QD-in-QW Structure
QDs hold tremendous promise for reducing the threshold current density in
semiconductor lasers [101]. However, if the density of the QDs is not very high,
the coverage of QDs on the sample surface is usually very small and efficient
carrier capture is hampered. On the other hand, the hybrid structure of QDs and
QWs, in other words, the quantum dot in quantum well (QD-in-QW) structure,
has been widely used in GaAs-based light emitters. This hybrid structure can
improve the efficiency of carrier capture and thus lead to a lower threshold
current as well as higher emission efficiency [102, 103]. Additionally, a wider gain
spectrum can be obtained under high injection currents. This section presents
the first attempt at achieving dual-wavelength lasing GaN VCSELs by inserting
the dot material in a strained InGaN QW.
10.5 Dual-Wavelength Lasing Based on QD-in-QW Active Structure 385
425 nm 0.1 mA
0.4 mA
2 mA
8 mA
Intensity (a.u.)
525 nm
400 450 500 550 600

Wavelength (nm)
Figure 10.12 EL spectra of the device without top DBR measured at different currents.
The epitaxial layer of the device was grown on the c-plane (0001) sapphire
substrate by MOCVD. The active region consisted of two pairs of In0.27 Ga0.73 N
(2.5 nm) QD layers embedded in In0.1 Ga0.9 N (2 nm) QWs capped by 8 nm-thick
GaN layer. Such a thin active layer can prevent a couple of phenomena related to
carrier injections, especially in the nitride-based materials. The effective mass of
hole is much larger and their mobility is low in GaN, so the injection of holes into
the QWs near the n-GaN side is not effective if the active layer is too thick, induc-
ing large absorption loss. AFM scans indicate a dot density of ∼1.5 × 1010 cm−2 .
We measured the spontaneous emission spectra of the fabricated device without
top DBR under different currents at RT, as shown in Figure 10.12. Under the injec-
tion current of 100 μA, only a single broad peak at 525 nm can be observed. The
broad spectrum is attributed to the inhomogeneity of the QDs in size and alloy
composition. As the current increases, the emission peak at 525 nm exhibits a fur-
ther broadening at the shorter wavelength side instead of the longer wavelength
side, which can be explained by the band-filling effect in QDs [90]. With the fur-
ther increase of current, another sharp peak emitted from the QWs at 425 nm
becomes stronger. The energy separation between the two peaks is of ∼556 meV.
A model combining the energy states in both QD and QW can be used to
explain the spectra evolution under different injection currents, as shown in
Figure 10.13. Under low currents, the relaxation of carriers from high energy of
the QW to the lowest energy state of QDs is dominant, resulting in the emission
peak at 525 nm. With the increase of current, the band-filling effect in QDs
becomes dominant, which accounts for the broadening of the peak. At the same
time, some localized carriers thermally escape from the QDs to QWs, the carrier
concentration increases in QWs, and the peak at 425 nm emerges. At even
higher injection currents, the QDs become saturated and the recombination of
carriers has mostly taken place in the QWs, thus the emission peak at 425 nm
becomes pronounced.
Cap layer Cap layer

Relaxation Relaxation
Escape QW
QW
Energy Energy
QD QD
(a) Space (b) Space
Figure 10.13 Schematic diagrams indicating the possible processes of carrier transport and
relaxation in QD and QW. The carrier distributions at low and high currents are shown in
(a) and (b), respectively.
10.5.2 Lasing Characteristics of VCSELs

The device has a double dielectric DBR structure, and Figure 10.14a shows the
lasing spectra under RT CW condition. The two lasing peaks at 545 nm (defined
as M1) and 430 nm (defined as M2) were clearly observed. It can be seen that
under the current of 0.03 mA, only M1 exists. M2 appears at a higher current.
Then, the two lasing modes coexist simultaneously, with M1 being stronger
than M2. The intensities of the two lasing peaks are almost equal when the
current is 10 mA and then M2 becomes dominant when the current is further
increased. The energy difference between the two modes is 609 meV. According
to the EL spectra in Figure 10.12 and the energy model in Figure 10.13, the two
lasing modes observed in Figure 10.14 comes from the emission of QDs and
QWs, respectively. Figure 10.14b,c show the L–I characteristics of the two lasing
modes. The transition from spontaneous emission to stimulated emission can
be observed clearly from the nonlinear curves. The threshold currents were 2 μA
for M1 and 5 mA for M2. The threshold current of M1 is much smaller than that
of M2, suggesting that the carrier competition occurs in the two energy states.
Note that the threshold current of M1 is only 2 μA. To the best of our knowl-
edge, this is the lowest value reported to date for GaN-based VCSELs. At low
injection current levels, the distribution of current was not uniform because the
emitting dots act as narrow current paths. The density of carriers flowing through
the activated QDs is much higher than the surrounding material. In fact, the true
threshold current density may be relatively high in QDs, and it should be inaccu-
rate to describe the threshold in “density” for the device.
10.6 Blue VCSELs with Different Lateral Confinements

10.6.1 Design of Index-Guided Structure
In the early works on electrically injected GaN VCSELs, the main focus on cur-
rent confinement aperture was lateral confinement of current, without paying too
much attention to its impact on optical properties. However, the typical intra-
cavity contact structure used in GaN VCSELs will form a concave structure in
the mesa center, resulting in antiguiding [104]. It was shown that the degree of
10.6 Blue VCSELs with Different Lateral Confinements 387
427 nm 0.03 mA
0.8 mA
545 nm 6 mA
M2 10 mA
15 mA
Intensity (a.u.)
M1
400 450 500 550 600

(a) Wavelength (nm)
Experiment Experiment
fitting fitting
M1 M2
Intensity (a.u.)
Intensity (a.u.)
Ith
Ith
0 2 4 6 8 10 0 4000 8000 12 000 16 000

Current (μA) Current (μA)
(b) (c)
Figure 10.14 (a) Laser emission spectra at various injection currents measured at RT. Intensity
of (b) M1 and (c) M2 as a function of current.
the depression directly decides whether laser cavity could work as an effective
waveguide in lateral direction. An optimum design of the optical cavity to have a
positive waveguide effect is of great importance to achieve a good lateral overlap
between the optical mode and carriers and to minimize lateral diffraction and
radiation loss.
An effective index model can be applied to calculate the lateral effective index
profile [105]. The relative refractive index difference Δn/n can be estimated from
the local resonance wavelength shift between the center and peripheral region of
the VCSEL mesa, as described by the following equation:
𝜆 c − 𝜆p Δ𝜆 Δn
= = (10.6)
𝜆c 𝜆c n
DBR
Cr/Au
100
N-type GaN 457 nm
Reflectivity (%)
InGaN active layer 80
SiO2 SiO2 469 nm
DBR 60 Center area

ITO
Surrounding area
P-type GaN
40
Cu plate (P-type contact) 471

456 459 462 465 468
(a) (b) Wavelength (nm)
Figure 10.15 (a) Cross-sectional schematic of the SiO2 -buried structure VCSEL. (b) Calculated
reflectivity spectra of the VCSEL cavity with LOC at center and peripheral areas.
where 𝜆c and 𝜆p are the resonance wavelengths of the center and peripheral areas,
respectively. When 𝜆c – 𝜆p is positive, the optical guiding is obtained. 𝜆c and 𝜆p
can be obtained from the calculated reflectivity spectrum of an entire VCSEL
cavity.
As illustrated in Figure 10.15a, the VCSEL with lateral optical confinement
(LOC) here employed a dual-dielectric DBR design and SiO2 aperture/ITO
electrode structure. The SiO2 -buried structure was fabricated by etching a
patterned groove of the p-GaN layer to a depth of 200 nm using ICP etching,
and a 200 nm SiO2 film was deposited by magnetron sputtering. We calculated
resonance wavelengths of the center and peripheral areas to estimate Δn/n in the
VCSELs with LOC structure, as shown in Figure 10.15b. The calculation shows
that the resonance wavelengths of the center and peripheral areas are 469.3
and 457.5 nm, respectively. Based on Eq. (10.1), Δn/n of SiO2 -buried structure
VCSEL is 0.026, showing that positive optical guiding effect is realized.
10.6.2 Emission Properties of VCSELs with Lateral Confinement

The VCSELs with a 15 μm aperture diameter were tested under CW operation
at RT. Figure 10.16a shows the emission spectra of the device with SiO2 -buried
structure under different current levels. Multitransverse modes due to the lat-
eral confinement of the proposed shallow etched mesa structure can be observed
associated with each single longitudinal mode, and the fundamental mode at low-
est energy is always dominant. The transverse modes become more clearly after
the lasing condition was reached.
The I–L curves of the VCSELs with and without the LOC structure are shown
in Figure 10.16b. We applied linear fits to the I–L data so that the threshold
currents could be extracted. As shown in Figure 10.16b, the VCSELs with
SiO2 -buried structure has lower threshold current and higher maximum output
power when compared with those of the VCSELs without such LOC structure.
The maximum output power of both devices is limited by thermal rollover. The
improvements of device performance after utilizing the LOC structure could
be attributed to the reduction of the intracavity losses. It has been reported
10.7 Summary 389
Standard structure
5 mA 200 SiO2-buried structure
15 mA
25 mA
160
Intensity (a.u.)
Power (μW)
ηi = 1.3%
120
80 Ith = ~16 mA
ηi = 0.36%
40 Ith = ~31 mA
440 450 460 470 480 490 500 10 20 30 40 50

(a) Wavelength (nm) (b) Current (mA)
Figure 10.16 (a) The emission spectra of the SiO2 -buried structure VCSEL at different injection
currents. (b) The light output of two VCSELs of different structures as a function of injection
current at 300 K and CW operation.
that the internal loss caused by lateral optical leakage is very high and occupies
more than two-third of the total internal loss in GaN VCSELs without LOC
structure [106]. The internal loss of the VCSEL incorporating a LOC structure
was significantly reduced because of the decrease of transverse radiation loss. It
should be mentioned that VCSELs with a much higher output power of 16 mW
have already been demonstrated [46] by introducing a SiO2 -buried structure in
GaN-based VCSEL with hybrid epitaxial/dielectric DBR.
10.7 Summary
We have presented an overview of the current state of the art in the development
of GaN-based VCSELs. The technical approaches, fabrication processes, and key
performance characteristics of GaN VCSELs with different structures includ-
ing hybrid DBR VCSELs and double dielectric DBR VCSELs are introduced in
this chapter. The realization of electrically injected GaN VCSELs is challenging,
but the progress in recent years is encouraging. The output power has exceeded
16 mW for single GaN VCSEL and 1 W for the GaN VCSEL array. We highlighted
our work on blue and green double dielectric DBR VCSELs. By using InGaN QD
as the active region, RT, CW lasing of GaN VCSELs in green gap was demon-
strated. The QD VCSELs are featured with low threshold current density and las-
ing at different wavelengths from 491.8 (blue-green) to 565.7 nm (yellow-green),
covering most of the “green gap.” Green lasing was also realized in GaN VCSELs
with QW active region benefiting from the cavity effect and indium-rich energy
localization centers. Green VCSELs with a dot in well active region were also
successfully demonstrated. These results open up opportunities to design and
fabricate green emitting VCSELs with excellent performance that may lead to
wide-gamut compact displays and projectors.
References
1 Guenter, J.K., Lei, C., and Tatum, J.A. (2014). Evolution of VCSELs. In:
Vertical-Cavity Surface-Emitting Lasers XVIII. San Francisco, California:
International Society for Optics and Photonics.
2 Feezell, D.F. (2015). Status and future of GaN-based vertical-cavity
surface-emitting lasers. In: Gallium Nitride Materials and Devices X. San
Francisco, California: International Society for Optics and Photonics.
3 Yu, H.c., Zheng, Z.w., Mei, Y. et al. (2018). Progress and prospects of
GaN-based VCSEL from near UV to green emission. Prog. Quantum Elec-
tron. 57: 1–19.
4 Panajotov, K., Sciamanna, M., Valle, A. et al. (2016). Progress and challenges
in electrically pumped GaN-based VCSELs. In: Semiconductor Lasers and
Laser Dynamics VII. Brussels: International Society for Optics and Photon-
ics.
5 Ryu, H.Y. and Shim, J.I. (2011). Effect of current spreading on the efficiency
droop of InGaN light-emitting diodes. Opt. Express 19 (4): 2886–2894.
6 Kim, M.H., Schubert, M.F., Dai, Q. et al. (2007). Origin of efficiency droop in
GaN-based light-emitting diodes. Appl. Phys. Lett. 91 (18): 183507.
7 Shin, D.S., Han, D.P., Oh, J.Y., and Shim, J.I. (2012). Study of droop
phenomena in InGaN-based blue and green light-emitting diodes by
temperature-dependent electroluminescence. Appl. Phys. Lett. 100 (15):
153506.
8 Habib, B. and Baz, B. (2016). Hardware MIMO channel simulator for
cooperative and heterogeneous 5G networks with VLC signals. In: Interna-
tional Conference on Wired/Wireless Internet Communication. Thessaloniki:
Springer.
9 Huang, S.J. and Yen, S.T. (2007). Improvement in threshold of InGaN/GaN
quantum-well lasers by p-type modulation doping. J. Appl. Phys. 102 (11):
113103.
10 Wagner, T., Werner, C.F.B., Miyamoto, K. et al. (2011). A high-density
multi-point LAPS set-up using a VCSEL array and FPGA control. Procedia
Chem. 154 (2): 124–128.
11 Hainich, R.R. and Bimber, O. (2016). Displays: Fundamentals & Applica-
tions. New York: AK Peters/CRC Press.
12 Shinada, S., Koyama, F., Nishiyama, N. et al. (1999). Fabrication of
micro-aperture surface emitting laser for near field optical data storage.
Jpn. J. Appl. Phys. 38 (11B): L1327.
13 Birkbeck, A.L., Flynn, R.A., Ozkan, M. et al. (2003). VCSEL arrays as micro-
manipulators in chip-based biosystems. Biomed. Microdevices 5 (1): 47–54.
14 Miah, M.J., Al-Samaneh, A., Kern, A. et al. (2013). Fabrication and char-
acterization of low-threshold polarization-stable VCSELs for Cs-based
miniaturized atomic clocks. IEEE J. Sel. Top. Quantum Electron. 19 (4):
1701410–1701410.
15 Mahadevan-Jansen, A., Hibbs-Brenner, M.K., Jansen, E.D. et al. (2009).
VCSEL technology for medical diagnostics and therapeutics. In: Photons
References 391
and Neurons. San Jose, California: International Society for Optics and
Photonics.
16 Iga, K. (2000). Surface-emitting laser-its birth and generation of new opto-
electronics field. IEEE J. Sel. Top. Quantum Electron. 6 (6): 1201–1215.
17 Soda, H., Iga, K., Kitahara, C., and Suematsu, Y. (1979). GaInAsP/InP surface
emitting injection lasers. Jpn. J. Appl. Phys. 18 (12): 2329.
18 Nakamura, S. (1991). GaN growth using GaN buffer layer. Jpn. J. Appl. Phys.
30 (10A): L1705.
19 Honda, T., Katsube, A., Sakaguchi, T. et al. (1995). Threshold estimation of
GaN-based surface emitting lasers operating in ultraviolet spectral region.
Jpn. J. Appl. Phys. 34 (7R): 3527.
20 Redwing, J.M., Loeber, D.A., Anderson, N.G. et al. (1996). An optically
pumped GaN–AlGaN vertical cavity surface emitting laser. Appl. Phys. Lett.
69 (1): 1–3.
21 Chen, S.Q., Okano, M., Zhang, B.P. et al. (2012). Blue 6-ps short-pulse gen-
eration in gain-switched InGaN vertical-cavity surface-emitting lasers via
impulsive optical pumping. Appl. Phys. Lett. 101 (19): 191108.
22 Someya, T., Tachibana, K., Lee, J. et al. (1998). Lasing emission from an
In0.1Ga0.9N vertical cavity surface emitting laser. Jpn. J. Appl. Phys. 37
(12A): L1424.
23 Someya, T., Werner, R., Forchel, A. et al. (1999). Room temperature lasing
at blue wavelengths in gallium nitride microcavities. Science 285 (5435):
1905–1906.
24 Krestnikov, I.L., Lundin, W.V., Sakharov, A.V. et al. (1999).
Room-temperature photopumped InGaN/GaN/AlGaN vertical-cavity
surface-emitting laser. Appl. Phys. Lett. 75 (9): 1192–1194.
25 Song, Y.-K., Zhou, H., Diagne, M. et al. (2000). A quasicontinuous wave,
optically pumped violet vertical cavity surface emitting laser. Appl. Phys.
Lett. 76 (13): 1662–1664.
26 Tawara, T., Gotoh, H., Akasaka, T. et al. (2003). Low-threshold lasing of
InGaN vertical-cavity surface-emitting lasers with dielectric distributed
Bragg reflectors. Appl. Phys. Lett. 83 (5): 830–832.
27 Park, S.H., Kim, J., Jeon, H. et al. (2003). Room-temperature GaN
vertical-cavity surface-emitting laser operation in an extended cavity scheme.
Appl. Phys. Lett. 83 (11): 2121–2123.
28 Geske, J., Gan, K.G., Okuno, Y.L. et al. (2004). Vertical-cavity
surface-emitting laser active regions for enhanced performance with optical
pumping. IEEE J. Quantum Electron. 40 (9): 1155–1162.
29 Kao, C.-C., Peng, Y.C., Yao, H.H. et al. (2005). Fabrication and performance
of blue GaN-based vertical-cavity surface emitting laser employing AlN/GaN
and Ta2 O5 /SiO2 distributed Bragg reflector. Appl. Phys. Lett. 87 (8): 081105.
30 Chu, J.T., Lu, T.c., Yao, H.H. et al. (2006). Room-temperature operation of
optically pumped blue-violet GaN-based vertical-cavity surface-emitting
lasers fabricated by laser lift-off. Jpn. J. Appl. Phys. 45 (4A): 2556–2560.
31 Chu, J.T., Lu, T.C., You, M. et al. (2006). Emission characteristics of optically
pumped GaN-based vertical-cavity surface-emitting lasers. Appl. Phys. Lett.
89 (12): 121112.
32 Chih-Chiang, K., Lu, T.C., Huang, H.W. et al. (2006). The lasing characteris-
tics of GaN-based vertical-cavity surface-emitting laser with AlN-GaN and
Ta2 /O5 /SiO2 distributed Bragg reflectors. IEEE Photonics Technol. Lett. 18
(7): 877–879.
33 Lu, T., Kao, C., Huang, G. et al. (2007). Optically and electrically pumped
GaN-based VCSELs. In: Conference on Lasers and Electro-Optics/Pacific Rim.
Seoul: Optical Society of America.
34 Cosendey, G., Castiglia, A., Rossbach, G. et al. (2012). Blue monolithic
AlInN-based vertical cavity surface emitting laser diode on free-standing
GaN substrate. Appl. Phys. Lett. 101 (15): 151113.
35 Cai, L.E., Zhang, J.Y., Zhang, B.P. et al. (2008). Blue-green optically pumped
GaN-based vertical cavity surface emitting laser. Electron. Lett. 44 (16):
972–974.
36 Zhang, J.Y., Cai, L.E., Zhang, B.P. et al. (2008). Low threshold lasing of
GaN-based vertical cavity surface emitting lasers with an asymmetric
coupled quantum well active region. Appl. Phys. Lett. 93 (19): 191118.
37 Liu, W.J., Chen, S.Q., Hu, X.L. et al. (2013). Low threshold lasing of
GaN-based VCSELs with sub-nanometer roughness polishing. IEEE Pho-
tonics Technol. Lett. 25 (20): 2014–2017.
38 Lu, T.C., Kao, C.C., Kuo, H.C. et al. (2008). CW lasing of current injection
blue GaN-based vertical cavity surface emitting laser. Appl. Phys. Lett. 92
(14): 141102.
39 Higuchi, Y., Omae, K., Matsumura, H., and Mukai, T. (2008).
Room-temperature CW lasing of a GaN-based vertical-cavity
surface-emitting laser by current injection. Appl. Phys Express 1: 121102.
40 Omae, K., Higuchi, Y., Nakagawa, K. et al. (2009). Improvement in lasing
characteristics of GaN-based vertical-cavity surface-emitting lasers fabricated
using a GaN substrate. Appl. Phys. Express 2: 052101.
41 Kasahara, D., Morita, D., Kosugi, T. et al. (2011). Demonstration of blue and
green GaN-based vertical-cavity surface-emitting lasers by current injection
at room temperature. Appl. Phys. Express 4 (7): 072103.
42 Onishi, T., Imafuji, O., Nagamatsu, K. et al. (2012). Continuous wave oper-
ation of GaN vertical cavity surface emitting lasers at room temperature.
IEEE J. Quantum Electron. 48 (9): 1107–1112.
43 Izumi, S., Fuutagawa, N., Hamaguchi, T. et al. (2015). Room-temperature
continuous-wave operation of GaN-based vertical-cavity surface-emitting
lasers fabricated using epitaxial lateral overgrowth. Appl. Phys. Express 8 (6):
062702.
44 Hamaguchi, T., Fuutagawa, N., Izumi, S. et al. (2016). Milliwatt-class
GaN-based blue vertical-cavity surface-emitting lasers fabricated by epitaxial
lateral overgrowth. Phys. Status Solidi A 213 (5): 1170–1176.
45 Nakajima, H., Hamaguchi, T., Tanaka, M. et al. (2019). Single transverse
mode operation of GaN-based vertical-cavity surface-emitting laser with
monolithically incorporated curved mirror. Appl. Phys. Express 12 (8):
084003.
46 Kuramoto, M., Kobayashi, S., Akagi, T. et al. (2018). Enhancement of slope
efficiency and output power in GaN-based vertical-cavity surface-emitting
References 393
lasers with a SiO2 -buried lateral index guide. Appl. Phys. Lett. 112 (11):
111104.
47 Kuramoto, M., Kobayashi, S., Akagi, T. et al. (2018). High-output-power
and high-temperature operation of blue GaN-based vertical-cavity
surface-emitting laser. Appl. Phys. Express 11 (11): 112101.
48 Kuramoto, M., Kobayashi, S., Akagi, T. et al. (2019). Watt-class blue
vertical-cavity surface-emitting laser arrays. Appl. Phys. Express 12 (9):
091004.
49 Lu, T.C., Chen, S.W., Wu, T.T. et al. (2010). Continuous wave operation
of current injected GaN vertical cavity surface emitting lasers at room
temperature. Appl. Phys. Lett. 97 (7): 071114.
50 Chang, T.C., Kuo, S.Y., Lian, J.T. et al. (2017). High-temperature operation
of GaN-based vertical-cavity surface-emitting lasers. Appl. Phys. Express 10
(11): 112101.
51 Holder, C., Speck, J.S., DenBaars, S.P. et al. (2012). Demonstration of nonpo-
lar GaN-based vertical-cavity surface-emitting lasers. Appl. Phys. Express 5
(9): 092104.
52 Holder, C., Leonard, J., Farrell, R. et al. (2014). Nonpolar III-nitride
vertical-cavity surface emitting lasers with a polarization ratio of 100%
fabricated using photoelectrochemical etching. Appl. Phys. Lett. 105 (3):
031111.
53 Leonard, J.T., Cohen, D.A., Yonkee, B.P. et al. (2015). Nonpolar III-nitride
vertical-cavity surface-emitting lasers incorporating an ion implanted aper-
ture. Appl. Phys. Lett. 107 (1): 011102.
54 Leonard, J.T., Young, E.C., Yonkee, B.P. et al. (2015). Demonstration of a
III-nitride vertical-cavity surface-emitting laser with a III-nitride tunnel
55 Leonard, J.T., Yonkee, B.P., Cohen, D.A. et al. (2016). Nonpolar III-nitride
vertical-cavity surface-emitting laser with a photoelectrochemically etched
air-gap aperture. Appl. Phys. Lett. 108 (3): 031111.
56 Forman, C.A., Lee, S., Young, E.C. et al. (2018). Continuous-wave operation
of m-plane GaN-based vertical-cavity surface-emitting lasers with a tunnel
57 Forman, C.A., Lee, S., Young, E.C. et al. (2018). Continuous-wave operation
of nonpolar GaN-based vertical-cavity surface-emitting lasers. In: Gallium
Nitride Materials and Devices XIII. San Francisco, California: International
Society for Optics and Photonics.
58 Liu, W.J., Hu, X.L., Ying, L.Y. et al. (2014). Room temperature continuous
wave lasing of electrically injected GaN-based vertical cavity surface emitting
lasers. Appl. Phys. Lett. 104 (25): 251116.
59 Weng, G.E., Mei, Y., Liu, J.P. et al. (2016). Low threshold continuous-wave
lasing of yellow-green InGaN-QD vertical-cavity surface-emitting lasers. Opt.
Express 24 (14): 15546–15553.
60 Xu, R., Mei, Y., Zhang, B. et al. (2017). Simultaneous blue and green lasing
of GaN-based vertical-cavity surface-emitting lasers. Semicond. Sci. Technol.
32 (10): 105012.
61 Mei, Y., Weng, G.E., Zhang, B.P. et al. (2017). Quantum dot vertical-cavity
surface-emitting lasers covering the ’green gap’. Light Sci. Appl. 6 (1):
e16199.
62 Furuta, T., Matsui, K., Kozuka, Y. et al. (2016). 1.7-mW nitride-based
vertical-cavity surface-emitting lasers using AlInN/GaN bottom DBRs.
In: 2016 International Semiconductor Laser Conference (ISLC). Kobe: IEEE.
63 Ikeyama, K., Kozuka, Y., Matsui, K. et al. (2016). Room-temperature
continuous-wave operation of GaN-based vertical-cavity surface-emitting
lasers with n-type conducting AlInN/GaN distributed Bragg reflectors. Appl.
Phys. Express 9 (10): 102101.
64 Furuta, T., Matsui, K., Horikawa, K. et al. (2016). Room-temperature CW
operation of a nitride-based vertical-cavity surface-emitting laser using thick
GaInN quantum wells. Jpn. J. Appl. Phys. 55 (5S): 05FJ11.
65 Matsui, K., Kozuka, Y., Ikeyama, K. et al. (2016). GaN-based vertical cavity
surface emitting lasers with periodic gain structures. Jpn. J. Appl. Phys. 55
(5S): 05FJ08.
66 Takeuchi, T., Kamiyama, S., Iwaya, M., and Akasaki, I. (2019). GaN-based
vertical-cavity surface-emitting lasers with AlInN/GaN distributed Bragg
reflectors. Rep. Prog. Phys. 82 (1): 012502.
67 Yeh, P.S., Chang, C.C., Chen, Y.T. et al. (2016). GaN-based vertical-cavity
surface emitting lasers with sub-milliamp threshold and small divergence
angle. Appl. Phys. Lett. 109 (24): 241103.
68 Mishkat-Ul-Masabih, S.M., Aragon, A.A., Monavarian, M. et al. (2019).
Electrically injected nonpolar GaN-based VCSELs with lattice-matched
nanoporous distributed Bragg reflector mirrors. Appl. Phys. Express 12 (3):
036504.
69 Yagi, K., Kaga, M., Yamashita, K. et al. (2012). Crack-free AlN/GaN dis-
tributed Bragg reflectors on AlN templates. Jpn. J. Appl. Phys. 51 (5R):
051001.
70 Huang, G.S., Lu, T.C., Yao, H.H. et al. (2006). Crack-free GaN/AlN dis-
tributed Bragg reflectors incorporated with GaN/AlN superlattices grown by
metalorganic chemical vapor deposition. Appl. Phys. Lett. 88 (6): 061904.
71 Ng, H.M., Moustakas, T.D., and Chu, S.N.G. (2000). High reflectivity
and broad bandwidth AlN/GaN distributed Bragg reflectors grown by
molecular-beam epitaxy. Appl. Phys. Lett. 76 (20): 2818–2820.
72 Someya, T. and Arakawa, Y. (1998). Highly reflective GaN/Al0.34Ga0.66N
quarter-wave reflectors grown by metal organic chemical vapor deposition.
Appl. Phys. Lett. 73 (25): 3653–3655.
73 Waldrip, K.E., Han, J., Figiel, J.J. et al. (2001). Stress engineering during met-
alorganic chemical vapor deposition of AlGaN/GaN distributed Bragg reflec-
tors. Appl. Phys. Lett. 78 (21): 3205–3207.
74 Feltin, E., Christmann, G., Dorsaz, J. et al. (2007). Blue lasing at room
temperature in an optically pumped lattice-matched AlInN/GaN VCSEL
structure. Electron. Lett. 43 (17): 924–926.
75 Imura, M., Nakano, K., Fujimoto, N. et al. (2006). High-temperature
metal-organic vapor phase epitaxial growth of AlN on sapphire by multi
References 395
transition growth mode method varying V/III ratio. Jpn. J. Appl. Phys. 45
(11): 8639–8643.
76 Yamamoto, A., Murakami, Y., Koide, K. et al. (2001). Growth temperature
dependences of MOVPE InN on sapphire substrates. Phys. Status Solidi B
228 (1): 5–8.
77 Lobanova, A.V., Segal, A.S., Yakovlev, E.V., and Talalaev, R.A. (2012). AlInN
MOVPE: growth chemistry and analysis of trends. J. Cryst. Growth 352 (1):
199–202.
78 Wang, J., Tsou, C.W., Jeong, H. et al. (2019). III-Nitride vertical resonant
cavity light-emitting diodes with hybrid air-gap/AlGaN-dielectric distributed
Bragg reflectors. In: Gallium Nitride Materials and Devices XIV . San Fran-
cisco, California: International Society for Optics and Photonics.
79 Youtsey, C., McCarthy, R., Reddy, R. et al. (2017). Wafer-scale epitaxial
lift-off of GaN using bandgap-selective photoenhanced wet etching. Phys.
Status Solidi B 254 (8): 1600774.
80 Chen, G. (1995). A comparative study on the thermal characteristics of
vertical-cavity surface-emitting lasers. J. Appl. Phys. 77 (9): 4251–4258.
81 Osinski, M. and Nakwaski, W. (1995). Thermal analysis of closely-packed
two-dimensional etched-well surface-emitting laser arrays. IEEE J. Sel. Top.
Quantum Electron. 1 (2): 681–696.
82 Lee, H.K. and Yu, J.S. (2010). Thermal analysis of InGaN/GaN multiple
quantum well light emitting diodes with different mesa sizes. Jpn. J. Appl.
Phys. 49 (4): 04DG11.
83 Wang, J.H., Savidis, I., and Friedman, E.G. (2011). Thermal analysis of
oxide-confined VCSEL arrays. Microelectron. J. 42 (5): 820–825.
84 Mei, Y., Xu, R.B., Xu, H. et al. (2018). A comparative study of thermal char-
acteristics of GaN-based VCSELs with three different typical structures.
85 Waltereit, P., Brandt, O., Trampert, A. et al. (2000). Nitride semiconductors
free of electrostatic fields for efficient white light-emitting diodes. Nature
406 (6798): 865.
86 Tao, R. and Arakawa, Y. (2019). Impact of quantum dots on III-nitride
lasers: a theoretical calculation of threshold current densities. Jpn. J. Appl.
Phys. 58 (SC): SCCC31.
87 Arakawa, Y. (2002). Progress in GaN-based quantum dots for optoelectronics
applications. IEEE J. Sel. Top. Quantum Electron. 8: 823–832.
88 Li, Z.C., Liu, J.P., Feng, M.X. et al. (2013). Effects of matrix layer compo-
sition on the structural and optical properties of self-organized InGaN
quantum dots. J. Appl. Phys. 114 (9): 093105.
89 Chtanov, A., Baars, T., and Gal, M. (1996). Excitation-intensity-dependent
photoluminescence in semiconductor quantum wells due to internal electric
fields. Phys. Rev. B 53: 4704.
90 Weng, G.E., Zhao, W.R., Chen, S.Q. et al. (2015). Strong localization effect
and carrier relaxation dynamics in self-assembled InGaN quantum dots
emitting in the green. Nanoscale Res. Lett. 10 (1): 31.
91 Purcell, E.M., Torrey, H.C., and Pound, R.V. (1946). Resonance absorption by
nuclear magnetic moments in a solid. Phys. Rev. 69 (1–2): 37–38.
92 Schubert, E.F., Wang, Y.H., Cho, A.Y. et al. (1992). Resonant cavity
light-emitting diode. Appl. Phys. Lett. 60 (8): 921–923.
93 Liu, L., Wang, L., Liu, N.Y. et al. (2012). Investigation of the light
emission properties and carrier dynamics in dual-wavelength InGaN/GaN
multiple-quantum well light emitting diodes. J. Appl. Phys. 112 (8): 083101.
94 Zhang, J.L., Sun, S., Burek, M.J. et al. (2018). Strongly cavity-enhanced spon-
taneous emission from silicon-vacancy centers in diamond. Nano Lett. 18
(2): 1360–1365.
95 De, S., Layek, A., Raja, A. et al. (2011). Two distinct origins of highly local-
ized luminescent centers within InGaN/GaN quantum-well light-emitting
diodes. Adv. Funct. Mater. 21 (20): 3828–3835.
96 Huh, C., Schaff, W.J., Eastman, L.F., and Park, S.J. (2004). Temperature
dependence of performance of InGaN/GaN MQW LEDs with different
indium compositions. IEEE Electron Device Lett. 25 (2): 61–63.
97 Chichibu, S.F., Uedono, A., Onuma, T. et al. (2006). Origin of
defect-insensitive emission probability in In-containing (Al,In,Ga)N alloy
semiconductors. Nat. Mater. 5 (10): 810.
98 Kim, H., Han, D.P., Oh, J.Y. et al. (2012). Estimate of the nonradiative carrier
lifetime in InGaN/GaN quantum well structures by using time-resolved pho-
toluminescence. J. Korean Phys. Soc. 60 (11): 1934–1938.
99 Kim, H., Shin, D.S., Ryu, H.Y., and Shim, J.I. (2010). Analysis of
time-resolved photoluminescence of InGaN quantum wells using the carrier
rate equation. Jpn. J. Appl. Phys. 49 (11R): 112402.
100 Rhodes, W.T. (2003). Fundamentals, Technology and Applications of
Vertical-Cavity Surface-Emitting Lasers. Heidelberg, New York, Dordrecht,
Londres: Springer Series in Optical Sciences.
101 Lester, L., Stintz, A., Li, H. et al. (1999). Optical characteristics of 1.24-μm
InAs quantum-dot laser diodes. IEEE Photonics Technol. Lett. 11 (8):
931–933.
102 Liu, G.T., Stintz, A., Li, H. et al. (1999). Extremely low room-temperature
threshold current density diode lasers using InAs dots in In0.15Ga0.85As
quantum well. Electron. Lett. 35 (14): 1163–1165.
103 Ustinov, V.M., Maleev, N.A., Zhukov, A.E. et al. (1999). InAs/InGaAs quan-
tum dot structures on GaAs substrates emitting at 1.3 μm. Appl. Phys. Lett.
74 (19): 2815–2817.
104 Hashemi, E., Gustavsson, J., Bengtsson, J. et al. (2013). Engineering the lat-
eral optical guiding in gallium nitride-based vertical-cavity surface-emitting
laser cavities to reach the lowest threshold gain. Jpn. J. Appl. Phys. 52 (8S):
08JG04.
105 Hadley, G.R. (1995). Effective index model for vertical-cavity
surface-emitting lasers. Opt. Lett. 20 (13): 1483–1485.
106 Hashemi, E., Bengtsson, J., Gustavsson, J. et al. (2014). Analysis of struc-
turally sensitive loss in GaN-based VCSEL cavities and its effect on modal
discrimination. Opt. Express 22 (1): 411–426.
397
11
Integration of 2D Materials with Nitrides for Novel

Electronic and Optoelectronic Applications
Filippo Giannazzo 1 , Emanuela Schilirò 1 , Raffaella Lo Nigro 1 , Pawel Prystawko 2 ,
and Yvon Cordier 3
1
Consiglio Nazionale delle Ricerche – Istituto per la Microelettronica e Microsistemi (CNR-IMM), Strada VIII,
2 Institute of High Pressure Physics – Polish Academy of Sciences (Unipress–PAS), Sokolowska 29/37, 01-142
Warsaw, Poland
3
Universitè Cote d’Azur, Centre National de la Recherche Scientifique - Centre de Recherche pour
l’Hétéro-Epitaxie et ses Applications (CNRS-CRHEA), Rue Bernard Gregory, 06560 Valbonne, France
11.1 Introduction
In 2004, the isolation of graphene (Gr) [1] initiated the research on an entire class
of materials, the two-dimensional (2D) materials [2], which currently represent
one of the hot topics in condensed matter physics. Graphene is a 2D layer of
C atoms with sp2 hybridization displaced in a hexagonal lattice (as illustrated
in Figure 11.1a). From the electronic point of view, it exhibits a semimetallic
behavior, with the valence and conduction energy bands merging in a singularity
point of the reciprocal space (i.e. the Dirac point) and a linear dispersion relation
between energy and wave vector [3]. This peculiar energy band structure is the
origin of many of the interesting electronic transport and optical properties of
Gr, including the large electron mean-free path [4–6], the high carrier mobility
(from 103 to 105 cm2 /V s, depending on the substrate) [7, 8], the field effect tun-
able ambipolar carrier transport, and the high optical transparency (≈97.7% in a
wide wavelength range, from UV to near-IR) [9, 10]. The excellent electron mobil-
ity has been exploited for the demonstration of Gr field effect transistors (GFETs)
operating at high frequencies (cutoff frequencies up to 300 GHz) [11, 12]. How-
ever, the lack of a bandgap results in a low on/off current ratio (typically <10) in
the transfer characteristics of GFETs, making these devices unsuitable for logic
applications.
Besides semimetallic Gr, semiconductor transition metal dichalcogenides
(TMDs) have been the object of intense investigation in the past years [13].
A single layer of TMD, generally indicated by the chemical formula MX2 , is
composed by a layer of transition metal atoms M (such as Mo, W, Te, etc.)
embedded (covalently bonded) between chalcogen atoms X (such as S, Se, etc.),
as schematically illustrated in Figure 11.1b. Individual TMD layers are free of
dangling bonds and can stack on top of each other forming multilayers, which

398 11 Integration of 2D Materials with Nitrides
Transition metal dichalcogenides

MX2, with X = S, Se, Te,… and M = Mo, W,…
Graphene X
C M
X
(a) (b)
Figure 11.1 Schematic representation of the lattice structure of (a) a monolayer of graphene
and (b) a monolayer of transition metal dichalcogenide (side and top view).
Table 11.1 Main electronic and thermal properties of graphene and MoS2 .
Electron
Electron Saturation mobility Critical Thermal
Bandgap effective velocity (cm2 /V s) electric conductivity
7
(eV) mass (10 cm/s) at 300 K field (V/cm) (W /m K)
Graphene (1L) 0 ∼0 5.5 [17] 103 –105 [7] 105 [18] 5000 [19]
MoS2
1L 1.8 (direct) 0.41m0 1–200 [14] 34.5 [21]
[13] [20]
Multilayer 1.2 (indirect) 0.57m0 0.28 [18] ∼100 [15] 1.15 × 105 ∼50 [23]
[13] [22] [18]
are bound by van der Waals (vdW) interaction. To date, molybdenum disulfide
(MoS2 ) has been the most widely investigated member of the TMD family. In
particular, its layer-number-dependent semiconducting behavior (with a direct
bandgap of 1.8 eV for a monolayer and an indirect bandgap of 1.2 eV for few
layers or bulk MoS2 ), combined with a good chemical/structural stability under
ambient conditions, makes this material interesting for field effect transistors
[14–16] and optoelectronic applications [13].
A summary of the main electronic and thermal properties of Gr and MoS2 is
reported in Table 11.1.
Recently, advanced or novel electronic/optoelectronic devices have been
demonstrated by the combination of different 2D materials into vertical van
der Waals heterostructures [2] or by 2D material integration with bulk (3D)
semiconductors [24]. In particular, this latter approach presents the advantage of
combining the functional properties of 2D materials with the well-assessed elec-
tronic quality of 3D substrates, and it currently represents the most viable root
toward the exploitation of 2D materials in electronics/optoelectronics [25]. To
date, several efforts have been done to integrate 2D materials with silicon, which
still represents the dominant platform for digital and low-power electronics. The
integration of 2D materials with group III-nitride semiconductors (III-N) has

also been intensively explored by different research groups worldwide [26–28],
with the aim to improve the performances of existing GaN-based devices, as
well as to demonstrate novel device concepts.
Because of its excellent electrical conductivity and the high transparency, Gr
has been considered as a transparent conductive electrode (TCE) for GaN-LEDs
[29–31] in replacement to currently used indium–tin-oxide (ITO). Single or
few layers of Gr have also been employed as compliant interlayers to reduce the
dislocation density in GaN grown by metalorganic chemical vapor deposition
(MOCVD) on sapphire, by accommodating the strain due to the lattice and
thermal expansion coefficient mismatch between the GaN epilayer and the
sapphire substrate [32]. Furthermore, Gr interlayers have also been used to
grow good-quality GaN on (100)-oriented Si substrates (commonly used in the
fabrication of silicon electronic devices), which is an important step toward
monolithic integration of GaN with Si CMOS (complementary metal oxide
semiconductor) technology [33]. Thanks to its excellent thermal conductivity
(up to 5000 W/m K, the highest of all known materials) [19], Gr has also been
proposed as a suitable candidate to address self-heating problems in high-power
high electron mobility transistor (HEMT) devices based on AlGaN/GaN
heterostructures [34], as well as in high-power solid-state optoelectronic
devices [35]. Finally, the ultimate single atomic thickness of a Gr electrode
(allowing ballistic electronic transit in its transversal direction), combined with
the high-quality rectifying properties of the Gr junction with Al(Ga)N/GaN
heterostructures, has recently been exploited to implement vertical hot electron
transistors (HETs) for ultrahigh-frequency (THz) applications [28, 36].
The integration of TMDs, in particular, MoS2 with GaN, has also been the
object of increasing interest in the past years. In particular, the very low lattice
mismatch (<1%) between the two hexagonal crystals allows high-quality epitaxial
growth of MoS2 on GaN by chemical vapor deposition (CVD) [37]. MoS2 /GaN
heterostructures have been proposed for interesting applications in electronics
and optoelectronics. As an example, the heterojunction between p+ -MoS2 and
n+ -GaN has been recently employed to implement the band-to-band tunnel
diode [38], a device potentially capable of fast switching with very low power
consumption [39]. Furthermore, high responsivity and self-powered deep
UV photodetectors based on n-MoS2 /p-GaN heterojunctions have also been
demonstrated [40], taking benefit of the wide bandgap of GaN, combined to the
proper band alignment with MoS2 .
This chapter will present an overview of the recent developments in this
expanding research field.
Section 11.2 will discuss the main approaches for the integration of 2D mate-
rials with nitride semiconductors, including the transfer of Gr or MoS2 layers
grown on foreign substrates, and the direct growth (by CVD techniques) of these
layers on top of the AlN or GaN. Recent progress in the deposition of thin nitride
semiconductor films onto Gr and MoS2 will also be reported.
Section 11.3 will present examples of novel electronic devices based on
2D materials/III-N heterostructures, such as the HET with a Gr base and a
Al(Ga)N/GaN emitter for THz electronics and the band-to-band tunneling
diode based on p+ -MoS2 /n+ -GaN heterojunctions for digital electronics with
ultralow-power dissipation.
Section 11.4 will present examples of optoelectronic devices based on 2D mate-
rial junctions with GaN, such as GaN LEDs with Gr-transparent conductive elec-
trodes and deep UV photodetectors based on MoS2 /GaN junction.
Finally, Section 11.5 addresses the use of Gr heat spreaders for thermal man-
agement in high-power AlGaN/GaN HEMTs.
11.2 Fabrication of 2D Material Heterostructures with

Nitride Semiconductors
The availability of large-area heterostructures of 2D materials with nitride semi-
conductors is the prerequisite for industrial applications of these material sys-
tems. The two fabrication approaches adopted so far are (i) the transfer of 2D
layers grown on foreign substrates and (ii) the direct growth of 2D materials on
GaN. Although the second approach would be highly desirable, the direct growth
of Gr on GaN by scalable techniques such as CVD poses severe challenges in
terms of compatibility of the deposition conditions (temperature and precur-
sors) with GaN structural and compositional stability and, more generally, with
device process integration. For this reason, most of the Gr/III-N heterostructures
reported so far for basic studies or proof-of-concept devices have been fabricated
by the transfer approach. On the other hand, although the direct deposition of
MoS2 on GaN by a CVD approach at relatively low temperatures (700–900 ∘ C)
has been demonstrated, further studies are still required about process integra-
tion. These aspects are discussed in detail in the next two sections.
11.2.1 Transfer of 2D Materials Grown on a Foreign Substrate

In the case of Gr, the CVD growth using gas-phase carbon precursors (such as
CH4 , …) is commonly carried out on catalytic substrates (typically Ni or Cu in
the form of polycrystalline thin films or foils), which lower the energy barrier
for precursor dissociation, allowing Gr formation at temperatures in the order of
1000 ∘ C [41, 42]. The postgrowth transfer procedure typically requires the use of
a protective polymeric layer onto Gr to enable handling of this ultrathin mem-
brane. Detachment of Gr from the native substrate is obtained either by complete
chemical etching of the metal or by Gr delamination using electrolytic methods
[43]. The transfer of the polymer/Gr stack on the target substrate is carried out,
either by fishing, printing, or roll-to-roll [43, 44]. Finally, removal of the polymeric
carrier layer from Gr surface is performed by using proper solvents, eventually
followed by thermal treatments to eliminate polymer residuals.
Although transfer is a versatile and widely used method for Gr integration with
arbitrary substrates, it can suffer from some drawbacks related to Gr damage
(cracks, wrinkles, and folding) during handling and from undesired contamina-
tions, including metal contaminations [45] originating from the catalytic metal
substrate, that are only partially reduced by Gr delamination without substrate
11.2 Fabrication of 2D Material Heterostructures with Nitride Semiconductors 401
etching [43]. Furthermore, the final device structure can suffer from a lack of
robustness because of adhesion problems between transferred Gr and the sub-
strate.
In the past years, optimized transfer approaches for Gr onto GaN or
Alx Ga1−x N/GaN heterostructures have been demonstrated [28, 46]. Figure 11.2
schematically illustrates the sequence of steps employed in the transfer procedure
adopted in Ref. [46].
Spin-coated poly-methyl-methacrylate (PMMA) onto Gr/Cu was used as a
protective layer for the Gr membrane during manipulation. Furthermore, a ther-
mal release tape (TRT) laminated onto PMMA worked as a carrier layer to allow
PMMA/Gr handling after detachment from Cu. The Cu substrate was com-
pletely etched by prolonged immersion in an ammonium persulfide (NH4 )2 SO4
water solution. After cleaning in deionized water, the TRT/PMMA/Gr stack
was transferred to the target substrate by thermocompression printing, with the
TRT released during the heating ramp of this process [43]. Finally, the PMMA
carrier layer was removed in acetone. A subsequent annealing at 400 ∘ C in Ar
ambient was performed to eliminate the nanometric polymer residues, which
remained on the Gr surface even after solvent cleaning. Adhesion of Gr on the
target substrate was found to be critically dependent on the water contact angle
of the substrate, i.e. on its surface energy [46]. Highly hydrophilic surfaces with
very low contact angle are typically unsuitable for Gr adhesion because of the
hydrophobic character of this material. On the other hand, the presence of water
on the substrate can be beneficial to reduce mechanical stress experienced by
the Gr membrane during the initial stages of the thermocompression printing.
An intermediate value of the contact angle (around 40∘ –45∘ ) was found to be the
optimal trade-off to achieve adhesion of Gr on the target substrate. In the case
of the AlGaN surfaces, the initial water contact angle is around 80∘ , and a soft
PMMA
Thermal release
PMMA tape
CVD graphene on Spin coating TRT Iamination on Cu etching in

Cu foil PMMA on PMMA/graphene/Cu (NH4)2SO4
graphene
Heat/Pressure
Final substrate
Graphene on final PMMA removal in Tape thermal release TRT/PMMA/graphene

substrate acetone transfer by thermal
printing
Figure 11.2 Schematic representation of an optimized procedure to transfer CVD graphene

from the primary Cu substrate to the final AlGaN substrate. Source: Adapted with permission
from Giannazzo et al. [46]. Copyright 2017, Wiley.
O2 plasma treatment was found to reduce this value to ∼40∘ , with a beneficial
effect on the transferred Gr morphology.
Figure 11.3a shows a typical atomic force microscopy (AFM) image of an
Al0.22 Ga0.78 N/GaN heterostructure grown on silicon (with dAlGaN = 21 nm), used
as the substrate for Gr transfer [46]. The sample exhibits a smooth morphology
with a root mean square (RMS) roughness of 0.45 nm. Small pits present on
the sample surface can be associated with threading dislocations with a surface
density lower than 2 × 109 cm−2 . The AFM morphology of Gr transferred onto
AlGaN is reported in Figure 11.3b, showing a uniform coverage by the Gr mem-
brane without pinholes and cracks. The higher RMS roughness with respect to
bare AlGaN is mainly related to the presence of wrinkles, i.e. nanometer height
corrugations of the Gr membrane. Some of these features are present in Gr
starting from the CVD growth [47], whereas part of them are introduced during
the transfer procedure. Raman spectroscopy analyses indicated a high structural
quality, i.e. a very low defect density, of the transferred Gr. Furthermore, Raman
and electrical analyses revealed a peculiar high n-type doping (1.1 × 1013 cm−2 )
for Gr in contact with AlGaN, which was explained by the combined effect of
Fermi level pinning by AlGaN surface states and charge transfer [36].
The carrier injection across the Gr/AlGaN/GaN heterojunction has been inves-
tigated at nanoscale by conductive atomic force microscopy (CAFM) [48, 49], as
schematically illustrated in Figure 11.4a. A typical current–voltage (I–V tip ) char-
acteristic measured in this configuration is reported in Figure 11.4b, showing a
rectifying behavior with negligible current at negative bias values and current
onset at positive ones. Figure 11.4c,d shows a typical morphology and the corre-
sponding vertical current map measured with the tip scanned on the Gr mem-
brane. Uniform current injection can be deduced from Figure 11.4d, except for a
local reduction of current on the wrinkles. Such effect can be ascribed to a local
reduction of doping induced by the AlGaN substrate in these corrugations of the
Gr membrane.
AIGaN Gr transferred onto AIGaN

2 5
RMS = 0.45 nm RMS = 1.5 nm
Height (nm)
Height (nm)
0 0
1 μm 2 μm
(a) (b)
Figure 11.3 AFM morphology of as-grown AlGaN/GaN heterostructure on Si (a) and after
transfer of a monolayer Gr transfer on the AlGaN surface (b). Source: Adapted with permission
from Giannazzo et al. [36]. Copyright 2019, American Chemical Society.
nA
400
Gr 300
I (nA)
AIGaN 200
2DEG 100
GaN
0
–3 –2 –1 0 1 2 3
(a) (b) Vtip (V)
5 500
Current (nA)
Height (nm)
2.5 nm
0 0
Vtip = 2.5 V
0.5 μm 0.5 μm
(c) (d)
Figure 11.4 (a) Schematic illustration of the setup for vertical current measurements with
CAFM. (b) Typical current–voltage (I–V tip ) characteristic measured in the vertical configuration,
showing a rectifying behavior, with negligible current at negative bias values and current
onset at positive ones. (c) Morphology and (d) vertical current map measured with the tip
scanned on the Gr membrane. A line scan showing the height of a Gr wrinkle is shown in the
inset of panel (c). Source: Adapted with permission from Giannazzo et al. [36]. Copyright 2019,
American Chemical Society.
11.2.2 Direct Growth of 2D Materials on Group III-Nitrides

Although transfer currently represents the most widely used approach for Gr
integration with nitride semiconductors, some investigations on the direct CVD
growth of Gr from carbon precursors on III-N substrates/templates have been
reported in the literature [50, 51]. Gr deposition on these noncatalytic surfaces
represents a challenging task, as it requires significantly higher temperatures as
compared to conventional deposition on metals. The first experimental works
addressing this issue showed the possibility of depositing few layers of Gr both
on bulk AlN (Al and N face) and on AlN templates grown on different substrates,
such as Si(111) and SiC, at temperatures >1250 ∘ C using propane (C3 H8 ) as
the carbon source, without significantly degrading the morphology of AlN
substrates/templates [50, 51]. Figure 11.5a schematically illustrates the CVD
growth conditions of Gr on the surface of an AlN/SiC template, i.e. the C3 H8
and N2 gas fluxes, the pressure p = 800 mbar, and the temperature T = 1350 ∘ C.
Figure 11.5b shows Raman spectra collected at different positions of the AlN
3.0х104 (c)
Gr CVD growth N2(10slm),
C3H8(17 sccm), p = 800 mbar, 2.5х104 G
G
T = 1350 °C, t = 6 min D’
Intensity (a.u.)
2.0х104
D
Few layer Gr 1.5х104 1500 1600 1700
–1
2D
Raman shift (cm )
AIN 1.0х104
SiC(0001)
5.0х103 D + D´
400 nm Z = 15 nm
0.0
1250 1500 1750 2500 2750 3000 RMS = 3.48 nm
(a) (b) Raman shift (cm–1) (c)
Figure 11.5 (a) Schematic illustration of the direct (noncatalytic) CVD growth of Gr at high
temperature (T = 1350 ∘ C) on the surface of an AlN/SiC template using propane (C3 H8 ) as the
carbon precursor. (b) Raman spectra collected at two different positions of the AlN surface
after Gr deposition. (c) Surface morphology of AlN with deposited Gr. Wrinkles on the Gr
surface are highlighted by white arrows. Source: Adapted from Dagher et al. [51]. Copyright
2017, Wiley.
surface after Gr deposition, demonstrating the presence of few layers of Gr with

domain size in the order of 30 nm. Figure 11.5c illustrates the typical surface
morphology of AlN with deposited Gr. Wrinkles, i.e. typical corrugations of the
Gr membrane, are highlighted by white arrows.
In spite of these interesting results about direct CVD growth of Gr on AlN,
further work will be required to evaluate the feasibility and the effects of similar
processes onto AlN/GaN or AlGaN/GaN heterostructures. Moreover, the inte-
gration of these high-temperature processes in the fabrication flow of GaN-based
devices remains a major obstacle. In this context, plasma-enhanced chemical
vapor deposition (PECVD) of Gr can help to lower the process temperature.
Different from Gr, the direct CVD growth of TMDs, in particular, MoS2 , on
noncatalytic insulating or semiconducting substrates (including SiO2 [52], sap-
phire [53], and GaN [37]) has been recently shown to be possible at relatively low
temperatures, in the range from 700 to 900 ∘ C. Furthermore, alternative chemical
deposition methods, e.g. atomic layer deposition (ALD) [54], or physical deposi-
tions, e.g. molecular beam epitaxy (MBE) [55] and pulsed laser deposition (PLD)
[56–58], are currently explored to achieve layer-by-layer growth of MoS2 on large
area, up to wafer scale.
Figure 11.6 illustrates the energy bandgap as a function of the in-plane lattice
parameter for nitride semiconductors (GaN, AlN, and InN) and for the most
common TMDs (MoS2 , WS2 , MoSe2 , and WSe2 ). From this plot, it can be
deduced how epitaxial growth of MoS2 on the basal plane of GaN is especially
favored by the low in-plane lattice mismatch (<1%) between the two hexagonal
crystals. Furthermore, the small difference in the thermal expansion coefficients
between the two materials is expected to result in a reduced strain during cooling
down from the growth temperature of MoS2 to room temperature [59, 60].
Recently, Ruzmetov et al. [37] reported the nucleation and growth of epitaxi-
ally oriented MoS2 islands on the GaN basal plane, using CVD with vapors from
solid sources of sulfur (S) and molybdenum oxide (MoO3 ). Figure 11.7a shows a
schematic illustration of the CVD setup, which consisted of a single-zone furnace,
Figure 11.6 Bandgap vs. in-plane lattice

6 AIN
parameter for the main nitride
semiconductors (AlN, GaN, and InN) and
the most common TMDs. The lattice 5
Energy bandgap (eV)

mismatch of MoS2 and WS2 with respect to
GaN is 0.8% and 1.0%, respectively. 4
GaN
3
ws2
2 WSe2
MoS2
1
MoSe2
InN
0
3.0 3.1 3.2 3.3 3.4 3.5 3.6
In-plane lattice parameter (Å)
where the MoO3 crucible was placed just below the GaN sample at a temper-
ature of 800 ∘ C, the S powder (at a temperature 130 ∘ C) was placed upstream,
and Ar was employed as the carrier gas. Figure 11.7b reports a scanning electron
microscopy (SEM) image of the as-grown MoS2 on the GaN surface, consisting of
triangular domains of monolayer MoS2 (as confirmed by Raman and photolumi-
nescence [PL] analyses) with a typical size of ∼1 μm. The sides of these triangles
were perfectly aligned with the m-plane (1–100) of the wurtzite GaN substrate,
indicating the in-plane epitaxial alignment of the GaN and MoS2 lattices. Thanks
to this rotational order, no evidence of grain boundaries was observed in larger
size monolayer MoS2 islands formed by the coalescence of these small domains.
This is a major advantage of MoS2 grown on GaN with respect to the more com-
monly used CVD MoS2 on amorphous SiO2 , which is a polycrystalline material
with a large density of grain boundaries. In fact, grain boundaries have been
shown to be one of the main sources of mobility degradation in MoS2 [61, 62].
Finally, X-ray photoelectron spectroscopy (XPS) analyses confirmed the stoichio-
metric composition of deposited MoS2 (S/Mo ratio 2.05 ± 0.1) and allowed to
exclude sulfurization of the GaN surface during the deposition process.
The electrical properties of the CVD MoS2 /GaN heterojunction (specifically,
the vertical current flow across the heterointerface and the surface potential) have
been investigated by CAFM and Kelvin Probe Force Microscopy (KPFM), respec-
tively [37]. Figure 11.8a schematically illustrates the CAFM setup for local current
measurements, and Figure 11.8b reports a current vs. tip bias characteristic on
an individual monolayer (1L) MoS2 domain, showing a rectifying behavior of the
tip/MoS2 /GaN junction. A KPFM surface potential map on a GaN region par-
tially covered by 1L MoS2 domain is reported in Figure 11.8c, while Figure 11.8d
shows a line scan of the surface potential along the red dashed line in the map,
showing a surface potential difference ≈360 meV between the unintentionally
n-type-doped MoS2 and n-GaN [37]. Finally, Figure 11.8e shows an illustrative
energy band diagram showing the type I energy band alignment between 1L MoS2
and GaN, as deduced from the surface potential map. As it will be discussed in
Sections 11.3 and 11.4 of this chapter, this peculiar energy band alignment can
be exploited in various device applications of MoS2 /GaN heterojunction.
Monolayer
2nd layer MoS2 triangles
800 °C GaN
130 °C
Ar
GaN
S Si
MoO3
1 μm
(a) (b)
Figure 11.7 (a) Schematic illustration of the CVD system for MoS2 growth on GaN. (b) SEM image of triangular domains of epitaxial monolayer MoS2
rotationally aligned with the m-plane (1−100) of the hexagonal GaN substrate. Source: Adapted with permission from Ruzmetov et al. [37]. Copyright 2016,
MoS2 6 I (nA)
4
nGaN
n+GaN 2
–10 –5 5 10
Sapphire
Bias (V)
(a) (b)
Potential (V) 0.0 MoS2

V 360 mV
–0.2
0.0 –0.4 GaN GaN
(d) 0 500 1000 1500 2000

–0.2 Distance (nm)
1L MoS2
Eg = 1.8 eV
Eg = 3.4 eV
–0.4
GaN
500 nm
(c) (e)
Figure 11.8 (a) Schematic of the CAFM setup for local I–V analyses on the MoS2 /GaN junction.
(b) Current–voltage characteristic on an individual MoS2 domain. (c) Surface potential map
measured by KPFM on a monolayer (1L) MoS2 domain on GaN and (d) potential line scan along
the dashed line in the map, showing a 360 mV surface potential difference between 1L MoS2
and GaN. (e) Energy band alignment at the 1L MoS2 /GaN interface, as deduced from the
surface potential map. Source: Adapted with permission from Ruzmetov et al. [37]. Copyright
2016, American Chemical Society.
11.2.3 2D Materials as Templates for the Growth of Nitride

Semiconductor Films
Because of its hexagonal symmetry, Gr has also been considered as an interme-
diate layer for the growth of high-quality thin films of nitride semiconductors
onto arbitrary substrates. Araki et al. [33] employed Gr as a compliant layer to
improve the quality of GaN grown on Si(100) substrates, with the aim to achieve
monolithic integration between the mature Si-based electronic devices on Si(100)
and GaN-based optoelectronic devices. Gr grown by CVD on Cu was transferred
onto Si(100) and GaN deposition was carried out by radio frequency molecular
beam epitaxy (RF-MBE), exploiting the sp3 bonds induced in Gr by the plasma as
the nucleation sites for GaN growth. The comparison between thick (∼400 nm)
GaN films grown on Si(100) without and with the Gr interlayer showed in both
cases a columnar structure, but with significantly larger grains in the presence of
Gr/Si(100) substrate. Furthermore, X-ray diffraction measurements showed that
the GaN columns grown on Gr/Si(100) have a c-axis-oriented hexagonal wurtzite
structure.
More recently, Chen et al. [32] employed Gr films directly grown by CVD
on sapphire substrates as templates for the MOCVD growth of high-quality
GaN with low stress and low dislocation density for high-brightness blue LEDs.
Figure 11.9 schematically illustrates the sequence of process steps for the growth
of the GaN layers.
Figure 11.10a shows an SEM image of the CVD-grown Gr on sapphire (using
CH4 /H2 precursors at 1050 ∘ C), indicating a uniform coverage with a large den-
sity of wrinkles in the Gr membrane. Afterward, the Gr surface was subjected
to a N2 plasma treatment to increase the density of sp3 bonds in the Gr lattice,
as demonstrated by the larger D/G intensity ratio in the Raman spectrum of Gr
after N2 plasma (Figure 11.10b). The direct growth of GaN on the plasma-treated
Gr resulted in a rough and irregular morphology because of the low adsorption
energy of Ga atoms on Gr. For this reason, an AlN thin film was first deposited at
high temperature (1200 ∘ C) because of the higher adsorption energy of Al atoms
on Gr, followed by GaN growth at lower temperature (1045 ∘ C). Figure 11.10c
reports an AFM image of the as-grown GaN surface, showing a smooth mor-
phology, with a surface roughness of 0.67 nm. Finally, Figure 11.10d and (e) shows
the (0002) and (10–12) X-ray rocking curves of the GaN epilayer grown on the
sapphire substrate without and with the Gr interlayer. A reduction in the den-
sities of the screw (edge) dislocations from 6.33 × 108 cm−2 (1.07 × 1010 cm−2 ) to
9.46 × 107 cm−2 (5.07 × 109 cm−2 ) was achieved thanks to the use of the Gr inter-
layer. These results indicated that the Gr film could greatly relax the compressive
stress and reduce the dislocation density for improving the quality of GaN epi-
layer.
Kovács et al. [63] recently demonstrated the MOCVD growth of GaN on
few-layer graphene (FLG) obtained by high-temperature decomposition of
6H-SiC(0001). Before MOCVD growth, the FLG was prepatterned by lithog-
raphy and Ar/O2 plasma etching, in order to leave micrometer-wide SiC areas
uncovered by Gr. The nucleation of an AlN buffer layer on SiC started from these
bare SiC regions, followed by lateral overgrowth on the FLG-covered areas. After
the buffer layer formation, an intermediate Al0.2 Ga0.8 N layer and the thicker
GaN layer were grown. Figure 11.11a,b shows a schematic illustration and a
bright-field cross-sectional transmission electron microscopy (TEM) image of
the final heterostructure grown on the patterned FLG. The vertical arrows mark
the regions where the Gr layers were partially etched away in 1 μm wide stripes.
The GaN layer contains semicircular polycrystalline regions above the Gr layers.
The vertical dark lines are inversion domains, which travel straight to the surface
from regions where the AlN/GaN grows directly on the SiC. A dislocation
density of ∼3 × 109 cm−2 was evaluated in this sample, which was very similar
to that obtained on a reference sample without Gr layers grown using identical
parameters. This indicates the possibility of introducing a patterned epitaxial
graphene interlayer between the SiC substrate and GaN, which can be exploited
to improve heat dissipation in high-power GaN HEMTs on SiC [63].
In all the above discussed growth approaches, a modification of Gr structure
(plasma, prepatterning, functionalization, etc.) was required before III-N film
deposition. Recently, Kim et al. [64] explored the possibility of performing the
direct van der Waals epitaxy (vdWE) of high-quality single-crystalline GaN films
on Gr. Epitaxial Gr grown on SiC(0001) was used as a template because it retains
a unique orientation over the entire substrate.
Direct growth of N2 plasma treatment
graphene films graphene films
Gr
Sapph Gr
Sapph ire
ire Sapph
(a) (b) (c) ire
GaN film growth AIN thin film growth
GaN
AIN
AIN Gr
Sap Gr Sap
ph phire
ire
(e) (d)
Figure 11.9 Schematic illustration of the key steps involved in the growth of high-quality GaN films on Gr-buffered sapphire substrates. (a) Sapphire substrate.
(b) Direct growth of Gr films on sapphire substrates by using CVD method. (c) Gr films are treated in N2 plasma. (d) Growth of high-temperature AlN films on
Gr/sapphire substrates by MOCVD. (e) Growth of GaN films on AlN/Gr/sapphire. Source: Adapted with permission from Chen et al. [32]. Copyright 2018, Wiley.
After N2 plasma
Before N2 plasma
Intensity (a.u.)
Sapphire
Gr
2 μm 1500 2000 2500

(a) (b) Raman shift (cm–1)
RMS = 0.67 nm
0002 Without Gr 1012 Without Gr
10 nm
With Gr With Gr
Intensity (a.u.)
Intensity (a.u.)
–5 nm
2 μm
–1500 –750 0 750 1500 –1500 –750 0 750 1500
(c) (d) ω (arcsec) (e) ω (arcsec)
Figure 11.10 (a) Scanning electron microscopy (SEM) of the CVD-deposited Gr film on the sapphire substrate. (b) Raman spectra of Gr on sapphire before
(black) and after the plasma N2 treatment (red). (c) Atomic force microscopy (AFM) image of as-grown GaN films on AlN/Gr/sapphire. X-ray rocking curves of
(0002) (d) and (10–12) GaN (e) films grown on the sapphire substrate without (black line) and with (red line) the Gr interlayer. Source: Adapted with permission
from Chen et al. [32]. Copyright 2018, Wiley.
Figure 11.11 (a) Scheme of the

deposited layer sequence to grow GaN
GaN on patterned epitaxial Gr on SiC. AI0.2Ga0.8N
(b) Bright-field TEM image of the AIN
heterostructure. Vertical arrows mark FLG FLG FLG
the regions where the Gr layers were
etched away. Source: Adapted with SiC
(a)
permission from Kovács et al. [63].
Copyright 2015, Wiley. Pt/C
GaN
AI0.2Ga0.8N
AIN
Graphene 1 μm
SiC
(b)
Figure 11.12a shows the AFM surface morphology of as-grown epitaxial Gr

on SiC. It exhibits parallel and nearly equally spaced steps originating from the
SiC step bunching phenomenon occurring during high-temperature graphitiza-
tion. Step edges were demonstrated to play a key role to obtain a uniform and
single-crystalline GaN layer on Gr using MOCVD. Such a result was obtained by
an optimized two-step deposition process, consisting in a lower temperature step
(at 1100 ∘ C) permitting the preferential GaN nucleation along the periodic step
edges, followed by a higher temperature step (at 1250 ∘ C), allowing the lateral
(2D) growth and coalescence of GaN nuclei. The first deposition step at 1100 ∘ C
resulted in the formation of continuous GaN stripes aligned along the SiC vici-
nal steps (Figure 11.12b). The two-step deposition resulted in the formation of
continuous and smooth GaN films (Figure 11.12c). This was attributed to the
faster lateral growth at an increased growth temperature of 1250 ∘ C from the GaN
nuclei along the periodic terrace edges formed at 1100 ∘ C. The final thickness of
the GaN film grown under these optimal conditions was ∼2.5 μm. Its morphol-
ogy, obtained by high-resolution AFM, is reported in Figure 11.12d to illustrate
the low roughness (RMS ∼ 0.3 nm) of GaN surface. The threading dislocation
density for this film was ∼4 × 108 cm−2 . These results indicated that, even without
using any buffer layer, the GaN crystalline quality comparable with that typically
obtained via conventional AlN buffer-assisted GaN epitaxy on SiC or sapphire
substrates can be obtained on Gr.
Besides Gr, some TMDs, specifically MoS2 and WS2 , have also been recently
considered as templates for GaN growth [65, 66] because of the very low lattice
mismatch with respect to GaN (see Figure 11.6). Recently, Gupta et al. [65]
reported detailed studies of the growth of GaN on mechanically exfoliated flakes
of WS2 and MoS2 by metalorganic vapor-phase epitaxy. Structural and optical
characterization showed that strain-free, single-crystal islands of GaN were
obtained on the underlying chalcogenide flakes. Tangi et al. [66] reported the
MBE growth of GaN on single-layer MoS2 deposited by CVD on sapphire.
According to the diagram in Figure 11.6, MoS2 is expected to be lattice matched
with the In0.15 Al0.85 N alloy, which is well known to present the same lattice param-
eter of GaN. Recently, Tangi et al. [67] demonstrated both the lattice-matched
CVD growth of large-area monolayer MoS2 onto MBE-grown In0.15 Al0.85 N films
As-grown epitaxial Gr on SiC(0001) 2 steps: T1 = 1100 °C + T2 = 1250 °C
1 step: T1 = 1100 °C
15
15
RMS=0.3 nm
Height (nm)
Height (nm)
1 μm
10 μm 10 μm 10 μm
(a) (b) (c) (d)
Figure 11.12 (a) AFM surface morphology of as-grown epitaxial Gr on SiC (0001); (b) plan-view SEM image of GaN grown on Gr by 1-step deposition at
1100 ∘ C; (c) two-step deposition with nucleation at high temperature 1100 ∘ C and growth at 1250 ∘ C; (d) high-resolution AFM morphology of GaN grown
under the optimal conditions. Source: Adapted with permission from Kim et al. [64]. Copyright 2014, Nature Publishing Group.
and vice versa the MBE growth of In0.15 Al0.85 N on MoS2 . A type-I band align-
ment between the two lattice-matched semiconductors was also demonstrated,
with the conduction band edge of MoS2 lying 0.6 eV below that of In0.15 Al0.85 N.
Finally, a MoS2 quantum well embedded between two In0.15 Al0.85 N films was real-
ized [67]. All these studies open new perspectives in MoS2 /nitride heterojunction
engineering.
11.3 Electronic Devices Based on 2D Materials/GaN

Heterojunctions
11.3.1 Band-to-band Tunneling Diodes Based on MoS2 /GaN
Heterojunctions
Heterojunction tunnel diodes are a class of devices potentially capable of fast
switching with very low power consumption, based on the band-to-band tunnel-
ing across the heterojunction of p+ and n+ degenerately doped semiconductors
[39]. Several attempts of implementing this device concept have been made in
the past using conventional semiconductor systems (Si, Ge, and III-V), but with
limited success [39]. A key aspect to obtain efficient tunnel diodes is the realiza-
tion of an ideally abrupt semiconductor heterojunction without interface states.
In this context, the integration of a dangling bond-free MoS2 with GaN can be a
viable approach to realize ultrasharp 2D/3D semiconductor heterojunctions on
large area to implement an heterojunction tunnel diode.
Recently, Krishnamoorthy et al. [38] reported a p+ -MoS2 /n+ -GaN band-to-
band tunnel diode fabricated by transferring a p+ -doped MoS2 film onto a
highly n+ -doped GaN substrate grown by MBE. The p+ -type doped MoS2 with
a very large hole density of 7.6 × 1013 cm−2 was produced by high-temperature
(1100 ∘ C) sulfurization of a sputtered Mo/Nb/Mo stack on sapphire. Here,
Nb atoms incorporated in substitutional positions in the MoS2 lattice work
as acceptors. Figure 11.13a shows a schematic of the heterojunction diode
structure, while Figure 11.13b reports the current density–bias (J–V ) char-
acteristics of the device measured at room temperature. A very high current
density was measured under reverse bias, even for low negative voltage values.
Under forward bias, negative differential resistance (a characteristic feature of
band-to-band tunneling in Esaki diodes) was observed, with a peak current
density J p = 446 A/cm2 at 0.8 V and a valley current density of J v = 368 A/cm2
at 1.2 V. To illustrate the tunnel diode’s electrical behavior, schematic band
diagrams of the multilayer p+ -MoS2 /n+ -GaN junction under forward and
reverse bias are also reported in Figure 11.13c,d. Under forward bias, electrons
in the conduction band of GaN are able to tunnel into empty states (holes)
available in the valence band of multilayer MoS2 . With increasing forward bias,
the tunneling current increases up to the peak value J p (corresponding to perfect
alignment of the GaN conduction band edge with the MoS2 valence band edge),
followed by a decrease down to J p because of band misalignment for larger bias.
This current decrease results in a negative differential resistance. For further
increase of forward bias, the tunneling current contribution becomes negligible,
600 Figure 11.13 (a) Schematic

400 of p+ -MoS2 /n+ -GaN
Ni/Au/Ni
200 heterojunction diode. (b)
0 Current density–voltage
J (A/cm2)
P+ MoS2 (J–V) characteristics of the
–200
device measured at room
–400 400
temperature, showing
20 nm n+ GaN –600 negative differential
[Si] = 5 x 1019 cm–3 –800 resistance, i.e. the Esaki
1.0
GaN/Sapphire –1000 diode behavior, under
–1.0 –0.5 0.0 0.5 1.0 1.5
(a) (b) Voltage (V) forward polarization (see
inset). Schematic band
Reverse bias diagrams under forward
bias (c) and reverse bias (d)
Forward bias
illustrating the interband
Ec tunneling process. Source:
Ec Adapted with permission
Ef,p
Ef,n from Krishnamoorthy et al.
Ef,p Ef,n
[38]. Copyright 2016, AIP.
Ev Ev
(c) p+ MoS2 n+ GaN (d) p+ MoS2 n+ GaN
and the diffusion current is the dominant conduction mechanism across the
p+ –n+ junction. Under reverse bias (Figure 11.13d), the electrons in the valence
band of MoS2 are able to tunnel into empty states in the conduction band of
GaN, and this mechanism is referred to as Zener tunneling.
11.3.2 Hot Electron Transistors with Graphene Base and Al(Ga)N/GaN

Emitter
One of the main challenges in modern electronics is the development of
transistors able to operate at frequencies in the terahertz (THz) range, i.e. the
electromagnetic spectrum range separating millimeter wave electronics from
photonics, which is strategic for application areas such as communications,
medical diagnostics, and security. Currently, HEMTs based on the field effect
modulation of the lateral current transport in III–V semiconductor heterostruc-
tures are the main devices for such applications [68]. In particular, the key
advantage of AlGaN/GaN HEMTs over those made with lower bandgap com-
pound semiconductors (GaAs, InP, etc.) are the superior on-state current and
the lower off-state power consumption. To date, cutoff frequencies >400 GHz
have been achieved in III-N-based HEMTs with properly optimized channel
geometry and low resistance source–drain contacts [69]. However, further
improvements in the high-frequency performances of these devices will be
limited by the technological and physical issues related to lateral scaling of the
channel. In this context, the HET is an alternative device concept, based on the
transversal ballistic transport of hot electrons through an ultrathin base layer,
which has been proposed from a long time as the potential candidate to operate
in the THz frequency range [70, 71].
Figure 11.14 shows a schematic cross section of a HET (a) and the corre-
sponding band diagram illustrating the device working principle in the on-state
(b). The HET is a unipolar majority carrier vertical device consisting of three
terminals (emitter, base, and collector) separated by an emitter–base and
base–collector barriers. Hot electrons (i.e. electrons with energy larger than the
Fermi energy of carriers in the base) are injected from the emitter to the base
terminal under forward base-to-emitter polarization. For a base thickness lower
than the electron mean-free path, most of these hot carriers transit through the
base without energy loss and can reach the collector terminal after overcoming
the base–collector filtering barrier modulated by the collector bias.
For a long time, one of the main obstacles to efficiently implement this
device concept has been the fabrication of the ultrathin highly conductive base
electrode. Recently, the appearance of 2D materials provided new solutions for
the implementation of high-performance HETs [72–74]. In particular, Gr has
been proposed as an ideal base material, as it combines monoatomic thickness,
enabling ballistic electron transit in the transversal direction, with excellent
in-plane transport properties [4–6]. Theoretical studies have predicted excel-
lent high-frequency performances, with a cutoff frequency (f T ) up to several
terahertz, for Gr base HETs (GBHET) [75–77].
Besides the ultrathin base, an emitter–base barrier allowing efficient hot elec-
tron injection (either by thermionic emission over the barrier or by tunneling
through the barrier) is a key element for the implementation of the HET device.
Recently, the possibility of implementing GBHETs with high on-state current by
the integration of Gr with group III-nitride semiconductors has been consid-
ered [26–28]. In particular, thin films of AlN or Alx Ga1−x N, epitaxially grown
on GaN by MOCVD or MBE, proved to be excellent emitter–base barriers for
GBHETs [28, 78]. Further advantages of these material systems are the pres-
ence of the high-density (1013 cm−2 ) two-dimensional electron gas (2DEG) at the
Alx Ga1−x N/GaN interface, working as the hot electron emitter, as well as the pos-
sibility of tailoring the conduction band discontinuity between Alx Ga1−x N and
GaN by the Al content.
Very efficient current injection by Fowler–Nordheim (FN) tunneling mecha-
nism has been recently demonstrated in the case of Gr junctions with thin barri-
ers of AlN (3 nm) [28] or Al-rich Al0.65 Ga0.35 N (4.7 nm) [78] grown on n+ -doped
GaN. High-quality bulk GaN substrates with dislocation density <105 cm−2 have
been used as substrates to grow these very thin barrier layers with a sufficient
quality to avoid leakage current through defects.
Figure 11.15a shows a schematic representation of a diode structure fabricated
by Gr transfer onto the Al0.66 Ga0.34 N barrier layer on bulk GaN. Two representa-
tive current–voltage characteristics measured at room temperature (25 ∘ C) and
at 75 ∘ C on this diode are reported in Figure 11.15b. Notably, a very small depen-
dence on the temperature can be observed from these measurements, both in
the forward and in the reverse bias polarization. This is a first indication that cur-
rent injection from the graphene contact through the barrier layer is ruled by a
tunneling mechanism. The Fowler–Nordheim (FN) plot ln(J/E2 ) vs. 1/E is also
reported in Figure 11.15c, where J is the forward current density and E = V BE /d
is the electric field across the barrier layer. From this plot, a linear behavior with
Figure 11.14 (a) Cross-sectional schematic of a hot electron transistor (HET) and (b) energy band diagram of the device in the on-state.
VCB > 0
C
B
E
VBE > 0
(b)
VBE
VB = 0
VCB
B-C barrier
E-B barrier
Collector
Emitter
Base
(a)
10–5 –31
VB VB T = 25 °C
–6 T = 75 °C –32
10
Ni/Au Ni/Au
Graphene –33
AI2O3 AI2O3 10–7 R = 0.998
–34
Current (A)
AI0.66Ga0.34N, 4.7 nm
10–8 In (J/E2) –35
n+-GaN –36
10–9
n2+-GaN –37
10–10
–38
T = 25 °C
Ti/AI/Ni/Au 10–11 –39
VE –2 –1 0 1 2 0.2 0.4 0.6 0.8 1.0
(a) (b) VBE (V) (c) 1/E (cm/MV)
Figure 11.15 (a) Schematic cross section of a Gr/Al0.66 Ga0.34 N/GaN diode on bulk GaN. (b) Current–voltage characteristics of the diode measured at 25 and
75 ∘ C, showing very small dependence on the temperature. (c) Fowler–Nordheim plot from the forward bias characteristics. Source: Adapted with permission
from Prystawko et al. [78]. Copyright 2019, Elsevier.
excellent linear correlation coefficient (R = 0.998) can be observed for high elec-
tric fields, indicating FN tunneling across the Al-rich ternary barrier as the dom-
inant transport mechanism in this high bias regime.
Figure 11.16a illustrates a schematic cross section of a Gr/AlGaN/GaN
diode, fabricated by transfer of monolayer Gr onto an optimized quality
Alx Ga1−x N/GaN heterostructure (with x = 0.22, t AlGaN = 21 nm) on Si(111)
[36]. Figure 11.16b shows a typical current–voltage characteristic measured at
a temperature T = 25 ∘ C. This curve exhibits an excellent rectifying behavior,
with a very low current under reverse (negative) bias and a linear increase of
the current (in the semilog scale) in a range of eight decades under forward
(positive) bias. The mechanisms of current injection at this heterojunction were
investigated performing a temperature-dependent I–V characterization in the
range from 25 to 175 ∘ C, as reported in Figure 11.16c. The strong dependence
of the current on the temperature indicates thermionic emission as the main
current injection mechanism. In order to evaluate the Schottky barrier height
ΦB of the Gr/AlGaN interface, a linear fit of the I–V curves in Figure 11.16c has
been performed in the low bias region. The intercept on the current axis of this
fit is the saturation current term I s = AA* T 2 exp(−qΦB /kT) of the thermionic
emission equation I = I s exp(qV /nkT), where A is the Schottky diode area,
A* is the Richardson constant, k is the Boltzmann constant, q is the electron
charge, T is the temperature, and n is the ideality factor. Figure 11.16d shows
the semilog-scale plot of I s /T 2 vs. 1000/T. The Gr/AlGaN Schottky barrier
height value (ΦB = 0.62 ± 0.03 eV) is obtained as the slope of the linear fit of
these data. It is worth noting that this barrier height value is much lower than
the one expected according the Schottky–Mott theory for an ideal Gr/AlGaN
Schottky barrier, i.e. ΦB = W Gr − 𝜒 AlGaN = 1.9 eV, with W Gr = 4.5 eV being the
work function of neutral graphene and 𝜒 AlGaN = 2.6 eV being the electron affinity
for Al0.22 Ga0.78 N [79]. This large discrepancy can be ascribed to a Fermi level
pinning at the interface between Gr and AlGaN [36, 64].
The Gr/AlGaN/GaN Schottky junction was used as the key building block for
a GBHET device. A schematic cross section of the complete device is illustrated
in Figure 11.17a. With respect to the diode structure shown in Figure 11.16a,
it includes a thin Al2 O3 film (10 nm) grown on Gr by an optimized ALD pro-
cess [80, 81]. This insulating layer worked as the base–collector barrier of the
GBHET. Figure 11.17b reports a top-view optical microscopy of the device, where
the Ni/Au collector contact (C) deposited on the thin Al2 O3 film, the Ni/Au pads
contacting the Gr base (B), and the alloyed Ti/Al/Ni/Au Ohmic contacts on the
AlGaN/GaN emitter (E) are indicated. The device active area (100 μm × 100 μm),
i.e. the region where the emitter, base, and collector are overlapped, is delimited
by a red dashed line.
Figure 11.17c shows the emitter (J E ) and collector (J C ) current densities
measured as a function of the emitter–base bias V BE in the common base
configuration (V B = 0 V) and for a fixed collector bias (V CB = 2 V). The injected
current measured at the emitter terminal (J E ) exhibits an exponential depen-
dence on V BE , consistently with the behavior of the Gr/AlGaN Schottky diode.
The J C –V BE characteristic exhibits a turn-on voltage ∼1.3 V, with a low off-state
current density J C,OFF ≈1 μA/cm2 (for V BE < 1.3 V) associated with the leakage
10–4
Alloyed
Ti/AI/Ni/Au Ni/Au 10–6
Thick Graphene Thick I (A)
oxide oxide
10–8
AI0.22Ga0.78N, 21 nm
GaN 10–10
T = 25 °C
(a)
10–12
–3 –2 –1 0 1 2 3
(b) V (V)
10–3
10–4
10–5 10–18
10–6
10–7 10–19
Is /T2 (A/K2)
25 °C
10–8
I (A)
50 °C
10–9 75 °C
100 °C 10–20
10–10
125 °C
10–11
150 °C
10–12 175 °C 10–21 ФB = 0.62 ± 0.03 eV
10–13
0.0 0.5 1.0 1.5 2.0 2.5 3.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4
(c) Forward bias, V (Volt) (d) 1000/T (K–1)
Figure 11.16 (a) Schematic cross section of a Gr/AlGaN/GaN diode. (b) Current–voltage (I–V) characteristic measured on this diode at a temperature of 25 ∘ C,
under forward and reverse polarization. (c) Sequence of forward bias I–V curves measured at different temperatures, in the range from 25 to 175 ∘ C. For each
curve, a linear fit in the low bias region has been carried out to extract the saturation current value IS . (d) Semilog-scale plot of Is /T 2 vs. 1000/T and linear fit of
the data, from which the Gr/AlGaN Schottky barrier height value (ΦB = 0.62 ± 0.03 eV) is obtained. Source: Adapted with permission from Giannazzo et al. [36].
c
VB VB
VE VC VE
B B
Thick Thick E E
oxide
AI2O3, 10 nm oxide
AI0.22Ga0.78N, 21 nm B B
GaN c
(a) (b) 100 μm
101 100
JE
Current density, J (A/cm2)
100 JC
10–1
10–1
α = JC/JE
10–2
10–3
10–2
10–4
10–5
VCB = 2 V
10–6 10–3
0.0 0.5 1.0 1.5 2.0 2.5 3.0 1.5 2.0 2.5 3.0
(c) VBE (V) (d) VBE (V)
Figure 11.17 (a) Schematic cross section of the hot electron transistor structure and (b)
top-view optical microscopy image of the HET device. (c) Emitter (JE ) and collector (JC ) current
densities measured in the common base configuration (V B = 0 V) as a function of the
emitter–base bias (V BE from 0 to 3 V) and for a fixed collector bias V CB = 2 V. (d) Common base
current gain of the transistor. Source: Adapted with permission from Giannazzo et al. [36].
current of cold electrons through the Al2 O3 barrier. The exponentially increasing
J C for V BE > 1.3 V is associated with the current of hot electrons injected from
the emitter into the Gr base, which are able to reach the collector. Thanks
to the efficient hot electron injection at the Gr/AlGaN/GaN heterojunction,
an ON/OFF current density ratio J C,ON /J C,OFF ≈ 106 with a J C,ON = 1 A/cm2 is
achieved. Figure 11.17d shows the common base current gain of the transistor,
i.e. the ratio 𝛼 = J C /J E , which reaches values from 0.1 to 0.15 at V BE > 2 V. The
on-state current and the gain of this device were shown to be limited by the
high-energy barrier between the Gr base and the Al2 O3 material. Large space
of improvement in the device performances is expected by the development
of alternative base–collector barrier layers on Gr with more favorable band
alignment and suitable structural quality. As an example, further progresses in
the van der Waals epitaxy of thin GaN or InGaN layers on Gr [33] should meet
these requirements.
11.4 Optoelectronic Devices Based on 2D Material Junctions with GaN 421
11.4 Optoelectronic Devices Based on 2D Material

Junctions with GaN
The integration of Gr and other 2D materials with GaN for optoelectronics appli-
cations has been the object of several investigations in the past years. Many efforts
have been dedicated to the development of Gr-based TCE for GaN LEDs, in order
to improve current injection and light extraction [29]. Finally, the MoS2 /GaN
heterojunction has been investigated to demonstrate high-responsivity photode-
tectors in the deep UV wavelength range [40]. Examples of these optoelectronic
devices are discussed in the next sections.
11.4.1 GaN LEDs with Graphene-Transparent Conductive Electrodes

The first requirement for a material to be used as a TCE is an optimal “trade-off”
between the optical transmittance (Tr) and its sheet resistance (Rs ). The most
commonly employed TCEs for optoelectronic devices are currently represented
by the conductive oxides, such as the ITO [82, 83]. However, the increasing cost
of ITO due to the scarcity of In supply makes the search of an alternative solution
urgent. Among these, hybrid systems made by semiconducting oxides and thin
metal films (such as ZnO/Ag/ZnO [81] or TiO2 /Ag/TiO2 [70] stacks), networks
of single-walled carbon nanotubes (SWCNT) [84], or metal nanowires [82] have
been considered. Figure 11.18a shows the measured transmittance of a mono-
layer (1L) of Gr compared with that of ITO and the above-mentioned alternative
TCE materials [10]. Noteworthy, 1L Gr exhibits and nearly constant Tr ≈ 97.7% in
a wide range of wavelengths (energies) from 200 nm (6.2 eV) to 900 nm (1.38 eV).
In the UV frequency range, Gr presents clear advantages with respect to the more
commonly used ITO and ZnO/Ag/ZnO electrodes, for which a strong decrease
of the transmittance is observed.
The constant value of monolayer and few-layer Gr transmittance as a func-
tion of radiation wavelength has also been predicted theoretically [9], and it was
found to depend on the number of Gr layers (N Gr ) and on a fundamental phys-
ical constants, i.e. the fine-structure constant 𝛼 = q2 /(4𝜋𝜀0 ℏc) ≈ 1/137 (q being
the electron charge, ℏ the reduced Planck constant, 𝜀0 the absolute permittivity,
and c the velocity of light), according to the relation:
( )−2
𝜋
Tr = 1 + 𝛼NGr ≈ 1 − 𝜋𝛼NGr (11.1)
2
On the other hand, the sheet resistance Rs of few layers of Gr has been found to
depend on the carrier density n, on the mobility 𝜇, and on the number of layers
as follows:
1
Rs = (11.2)
qn𝜇NGr
It should be noticed that the Gr mobility depends on several factors, including
the defect density, the doping, and the dielectric properties of the substrate [6]. In
particular, for the typical doping levels of Gr TCEs, 𝜇 decreases with increasing n.
Figure 11.18b shows the comparison of literature results for the measured
transmittance vs. sheet resistance for ITO, a silver nanowire mesh, a SWCNT
100 100
n = 3.4 x 1012 cm–2
μ = 2 x 104 cm2/V s
80
Transmittance (%)
Transmittance (%)
80
n = 1013 cm–2
60
μ = 2 x 103cm2/V s
40 Graphene
ITO
ITO 60
SWNTs
ZnO/Ag/ZnO
Graphene CVD
20 TiO2/Ag/TiO2
Ag nanowire mesh
Arc discharge SWNTs
Graphene calculated
40
200 400 600 800 1 10 100
(a) Wavelength (nm) (b) Sheet resistance (Ω/◽)
Figure 11.18 (a) Transmittance spectrum as a function of radiation wavelength for a monolayer (1L) of Gr, ITO, ZnO/Ag/ZnO, TiO2 /Ag/TiO2 , and single-walled
carbon nanotubes (SWCNTs). (b) Experimental data of the “trade-off” between transmittance and sheet resistance for different TCEs: CVD Gr produced with a
roll-to-roll method; ITO, a mesh of silver nanowires and SWCNTs. The calculated Tr-Rsh trade-off for an ideal Gr TCE is also reported. Source: Adapted with
permission from Bonaccorso et al. [10]. Copyright 2010, Nature Publishing Group.
layer, and the state-of-the-art Gr TCE [44]. In this plot, the transmittance and
the sheet resistance for each transparent conductor was changed by varying the
film thickness. The Gr TCE was produced by CVD growth onto Cu foils and
wet-chemical doping, followed by roll-to-roll transfer [44]. The Gr film thickness
was increased from N Gr = 1 to 4 by subsequent transfers. The theoretical Tr vs.
Rs trade-off for multilayer Gr, calculated by combining Eqs. (11.1, 11.2), has also
been reported in the plot. The cases of lowly doped Gr (n = 3.4 × 1012 cm−2 ) with
high mobility (𝜇 = 2 × 104 cm2 /V s) and of highly doped Gr (n = 1013 cm−2 ) with
a low mobility (𝜇 = 2 × 103 cm2 /V s) have been considered as the two limiting
cases for the calculation.
This comparison shows that TCE based on high-quality monolayer or few-layer
Gr can already outperform currently used ITO and other transparent conductors.
Motivated by these promising electrical and optical properties, several research
groups explored the possibility of using Gr as a TCE for GaN-based LEDs, with
the aim to improve current spreading and light emission [29–31]. Figure 11.19a
reports a schematic illustration of a GaN LED structure with the Gr TCE in
contact with the topmost p-GaN layer. The optimal trade-off between sheet resis-
tance and optical transmittance in the case of Gr can be exploited to optimize the
lateral current spreading in the TCE and the light extraction from the LED, as
illustrated in Figure 11.19b. However, a crucial role for vertical current injection
is played by the contact resistance between Gr and p-GaN, which is ultimately
related to Schottky barrier height at the interface of the two materials.
Figure 11.20 illustrates the energy band alignment for a monolayer of Gr and
p-GaN before the contact formation (a) and under equilibrium conditions after
the contact formation (b). For simplicity, the Gr is assumed to be initially neu-
tral, i.e. with the Fermi level coincident with the Dirac point (EF,Gr = ED ). After
the contact formation, because of charge transfer at the interface, the Gr Fermi
level is downward-shifted with respect to the Dirac point (ED − EF = 0.3–0.4 eV),
resulting in an increased Gr work function [85–87]. From the band diagram of
Figure 11.20b, a theoretical value of the Gr/p-GaN Schottky barrier height of
2.5–2.6 eV can be estimated according to the relation:
ΦB = [WGr + (ED − EF )] − [𝜒GaN + Eg,GaN − (EF − EV )] (11.3)
Gr TCE
p-GaN
MQWs
n-GaN
Gr
u-GaN
p GaN
Sapphire
(a) (b)
Figure 11.19 (a) Schematic cross section of a GaN-based LED with a Gr-transparent
conductive electrode (TCE). (b) Illustration of the main physical mechanisms involving the Gr
TCE, i.e. the lateral current spreading in the TCE, the vertical current injection at the Gr/p-GaN
interface, and the transmission of light from the LED-active region.
Evac Evac
WGr GaN Evac
Ec Ec
EF,Gr ED
p-GaN Eg,GaN p-GaN
EF,GaN ED
EF EF
Ev Ev
ΦB
(a) (b)
Figure 11.20 (a) Energy band diagrams for isolated monolayer and p-GaN and (b) after the
junction formation.
Gr/p-GaN contact
0.04
Nonannealed 8 x 10–1
RTA at 500 °C ΦB = 1.68 eV
0.02 6 x 10–1
Current (mA)
ρc (ohm-cm2)
5 x 10–1
0.00
3 x 10–1
–0.02
2 x 10–1
–0.04 1 x 10–2
–3 –2 –1 0 1 2 3 200 250 300 350 400 450
Voltage (V) Temperature (K)
(a) (b)
Figure 11.21 (a) Typical I−V curves measured on a circular-transmission line model (cTLM)
test structure for the as-transferred CVD–Gr contact to p-GaN and after a rapid thermal
annealing (RTA) process at 550 ∘ C. (b) Temperature dependence of the specific contact
resistance 𝜌c for the annealed Gr/p-GaN contact and fit of experimental data with the TFE
model. Source: Adapted with permission from Chandramohan et al. [30]. Copyright 2013,
where W Gr = 4.5 eV is the work function for neutral Gr, 𝜒 GaN = 4.1 eV is the GaN
electron affinity, Eg,GaN is the GaN energy bandgap, and EF − EV ≈ 0.1 eV for GaN
p-type doping in the order of 1018 cm−3 .
Recently, Chandramohan et al. [30] investigated the current injection mecha-
nisms at the junction between transferred CVD-Gr and MOCVD grown p-GaN
on sapphire with a hole concentration p ≈ 6.9 × 1017 cm−3 . Figure 11.21a shows
two typical I–V curves measured on a circular-transmission line model (cTLM)
test structure for the as-transferred CVD-Gr contact to p-GaN and after a
rapid thermal annealing (RTA) process at 550 ∘ C in Ar ambient. The nonlinear
behavior of the I–V characteristic before the RTA process was ascribed to
the poor adhesion of wet-transferred Gr onto p-GaN. The annealing process

results in a nearly linear behavior with significantly high current injection.
Such improvement was mainly explained in terms of an enhanced Gr adhesion
with p-GaN, as structural/chemical modifications of the Gr contact and/or
the substrate were not observed [30]. A room-temperature-specific contact
resistance value 𝜌c ≈ 0.5 Ω cm2 was evaluated for this annealed Gr/p-GaN
contact. Figure 11.21b reports the temperature dependence of the specific
contact resistance (𝜌c ) on the annealed Gr/p-GaN contact. The 𝜌c decrease with
increasing measurement temperature was fitted by the thermionic field emission
(TFE) current injection mechanism and a Schottky barrier height ΦB ≈ 1.68 eV
was estimated [30].
Noteworthy, the contact resistance values for Gr contacts on p-GaN are 2 or 3
orders of magnitude higher than the 𝜌c typically measured for the commonly used
ITO or Ni/Au contacts [29]. This typically results in a higher turn-on voltage of
GaN-LEDs with a Gr TCE electrode as compared to the same LED structures with
an ITO TCE [88, 89]. On the other hand, the superior transmittance of Gr TCE
allows an improved light output power (LOP) with respect to a corresponding
LED with an ITO TCE [89].
Clearly, the formation of a low contact resistance Ohmic contact between
Gr and p-GaN is a major challenge to achieve optimal current injection for
GaN-LEDs. For this reasons, different solutions to reduce 𝜌c of the Gr TCE on
p-GaN have been explored in the past years. A common approach is the insertion
of an additional transparent and conductive interlayer between Gr and p-GaN
[30]. Obviously, such interlayer should have low 𝜌c with both p-GaN and Gr. To
date, several interlayer materials suitable to this purpose have been considered,
including very thin films of Ni [90] or NiOx [91], nanorods of ZnO [92] or of Ag
[93], and nanoparticles of Ag [94] or Au [95]. The interlayer generally reduces
𝜌c and the turn-on voltage. As an example, the use of an Au interlayer between
Gr and p-GaN resulted in a 𝜌c ≈ 0.1–0.2 Ω cm2 . However, the presence of this
interlayer often reduces the transmittance, as well, resulting in a decrease of the
light output power.
As an example, Figure 11.22a shows the comparison between the forward
current–voltage characteristics measured on InGaN/GaN blue LEDs with
different TCEs: i.e. a Ni/Au thin film, a Gr layer, a NiOx /Gr hybrid electrode
(with the NiOx interlayer between Gr and p-GaN), and a NiOx layer only. Here,
the NiOx interlayer (2 nm) was obtained by RTA of a Ni thin film deposited
on p-GaN in oxygen ambient. It is evident how the Gr layer only exhibits the
highest turn-on voltage (V on = 4.5 V at 20 mA) among the considered TCEs,
which is significantly reduced (down to 3.16 V) in the case of the NiOx /Gr hybrid
electrode and is very close to the value (3.04 V) for a reference device with Ni/Au
electrode.
Figure 11.22b illustrates the LOP as a function of injection current measured
on LEDs with the Ni/Au, Gr, and NiOx /Gr electrodes. It is evident how, in spite of
large contact resistance, the LED with the Gr electrode offered a LOP comparable
to that of the reference LED with the Ni/Au electrode. However, the large differ-
ence in power consumption (P = VI = 61 and 90 mW, respectively, for the LEDs
with Ni/Au and Gr electrodes) still results in a lower power efficiency (defined as
100 50
Ni/Au Gr
c
80 40
a b a: Ni/Au b
Light output power (a.u.)
b: GR a
Current (mA)
60 C: NiOX/GR 30
d d: NiOX
c NiOX
40 20 NiOX/Gr
20 10 a: Ni/Au
b: GR
C: NiOX/GR
0 0
0 2 4 6 8 10 0 20 40 60 80 100
(a) Voltage (V) (b) Current (mA) (c)
Figure 11.22 (a) Current–voltage characteristics of GaN LEDs with different transparent conductive electrodes: Ni/Au, Gr, NiOx /Gr, and NiOx . (b) Light output
power as a function of injection current measured on LEDs with the Ni/Au, Gr, and NiOx /Gr electrodes. (c) Optical images of light emission form the LEDs with
the different TCEs, showing the current spreading ability of the different electrodes. Source: Adapted with permission from Chandramohan et al. [30].
𝜂 = LOP/VI, where LOP is the optical output power) [96] of the Gr/p-GaN LED
with respect to the reference LED with the Ni/Au electrode. In the considered
case, the power efficiency at an injection current of 20 mA is ∼30% lower in the
case of the Gr/p-GaN LED.
The lower LOP of the GaN-LED with the NiOx /Gr hybrid electrode was par-
tially ascribed to the postannealing-induced structural modifications and associ-
ated transmittance degradation in the NiOx /Gr stack [30].
Finally, Figure 11.22c shows the comparison between the optical images of light
emission by the considered blue LEDs (captured at the same injection current of
200 μA), showing the current spreading ability of the different electrodes [30].
Uniform emission from the whole device area was observed for the devices with
the reference Ni/Au electrode and with the Gr and NiOx /Gr electrodes, thus con-
firming the excellent lateral current spreading of Gr. On the other hand, the thin
(2 nm) NiOx contact only is not able to work as a current spreading electrode.
11.4.2 MoS2 /GaN Deep UV Photodetectors

Recently, photodetectors based on the MoS2 /GaN junction showing a high
responsivity to deep UV radiation have been demonstrated [40]. Figure 11.23a
reports a schematic illustration of a n-MoS2 /p-GaN heterojunction device,
fabricated by transfer of few-layer MoS2 film (∼3.67 nm thick) onto a p-GaN
substrate, followed by deposition of contacts to MoS2 and p-GaN.
Typical current–voltage characteristics of the device measured in dark condi-
tion and under deep UV light illumination (with 265 nm wavelength) are reported
in Figure 11.23b. A very good rectifying behavior (with a rectification ratio over
104 within ±5 V) was observed in dark condition, and an excellent photoresponse
was obtained under deep UV light. Notably, the current at 0 V is greatly increased
from 3 × 10−12 A in the dark to 2.7 × 10−7 A under illumination, which is indica-
tive of a remarkable photovoltaic behavior of the heterojunction. This is better
elucidated in Figure 11.23c, showing the measured current as a function of bias
for the device illuminated by the 265 nm wavelength radiation with an intensity
of 2.4 mW/cm2 . An open-circuit voltage V OC = 1.3 V and short-circuit current
I SC = 0.27 mA can be observed. Hence, such an n-MoS2 /p-GaN heterojunction
device can serve as a self-powered deep UV photodetector.
This device showed a very high responsivity (defined as the ratio between the
photocurrent and the incident light intensity) of 187 mA/W and fast response
speed to pulsed light. These performances are superior to the ones reported for
self-powered UV photodetectors based on alternative wide-bandgap semicon-
ductor heterojunctions, including Gr/ZnO [97] and Ga2 O3 /ZnO [98].
The photoresponse properties of the n-MoS2 /p-GaN heterojunction can be
understood from the energy band diagram illustrated in Figure 11.23d. As previ-
ously discussed in this chapter, MoS2 and GaN exhibit a type-I band alignment,
with the MoS2 conduction band lying 0.3–0.4 eV below the conduction band
of GaN. When the n-type doped MoS2 and the p-type doped GaN come into
contact, because of the Fermi level difference between the two semiconductors,
electrons in the MoS2 film move to the GaN side, whereas holes in GaN move to
the MoS2 film. At the equilibrium, the MoS2 and GaN Fermi levels are aligned,
10–5
–6
10
Ni/Au
10–7
Current (A)
Ni/Au
10–8
AI2O3 n-MoS2
10–9
p-GaN 10–10
10
–11 Dark
(a) 265 nm
–12
10
–5 –4 –3 –2 –1 0 1 2 3 4 5
(b) Voltage (V)
0.00 Ec
–0.05
UV
Current (μA)
–0.10
–0.15 p-GaN
Voc = 1.3 v Ec
–0.20 Isc = 0.27 μA EF
–0.25 n-MoS2
265 nm Ev
–0.30 2.4 mW/cm
2
Ev
–0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
(c) Voltage (V) (d)
Figure 11.23 (a) Schematic of a deep UV photodetector based on an n-MoS2 /p-GaN

heterojunction. (b) Current–voltage characteristics of the heterojunction device measured in
dark condition and under illumination with deep UV light (wavelength of 265 nm). (c)
Photovoltaic effect of the MoS2 /GaN heterojunction device. (d) Energy band diagram for the
heterojunction, illustrating current photogeneration and charge collection in the device.
Source: Adapted with permission from Zhuo et al. [40]. Copyright 2018, Royal Society of
Chemistry.
and the energy bands of GaN near the interface are bent downward, whereas
the MoS2 energy bands are bent upward. As a result, a built-in electric field will
appear near the MoS2 /GaN interface, as depicted in Figure 11.23d. Under UV
light illumination, absorption of the incident light results in the generation of
electron–hole pairs, which are quickly separated by this built-in electric field and
collected by the electrodes, giving rise to photocurrent even at zero bias voltage.
Besides the built-in electric field, the fast response speed of the MoS2 photode-
tector is due to the dangling-bond-free nature of MoS2 , leading to reduced charge
trapping at the MoS2 /GaN interface.
11.5 Applications of Graphene for Thermal

Management in GaN HEMTs
A large number of methods have been used to improve heat removal from
GaN devices. Conventional sapphire substrates with low thermal conductivity
𝜅 = 30 W/m K at room temperature (RT) have been replaced with SiC substrates
with a higher thermal conductivity 𝜅 = 100–350 W/m K at RT. However, even in
GaN transistors on a SiC substrate, self-heating can lead to temperature rises ΔT
above 180 ∘ C. This is due to the high electric field induced at the gate edge while
11.5 Applications of Graphene for Thermal Management in GaN HEMTs 429
increasing the drain bias of the transistors. In this context, solutions for the local
thermal management of high-power density devices, specifically targeting the
hotspots at nanometer and micrometer scale, are highly desirable.
Yan et al. [34] showed that the local thermal management of AlGaN/GaN tran-
sistors can be substantially improved via introduction of additional heat-escaping
channels, represented by top-surface heat spreaders made of FLG. FLG films
present many advantages as heat spreaders with respect to ordinary metal films.
In fact, heat conduction in FLG films is ruled by phonon transport, and it is
preserved even reducing film thickness down to a single layer of Gr. On the con-
trary, the thermal conductivity of metals is dominated by electron transport, and
it becomes significantly lower than the bulk value for thin films with thickness
comparable to the electron mean-free path [99]. Beside the thermal conductivity,
another important parameter for heat spreaders is the thermal boundary resis-
tance (TBR) at the interface with other materials. Noteworthy, the TBR at the
interface between Gr or graphite and various substrates is relatively small, in the
order of ∼10−8 m2 K/W at RT, and does not strongly depend on the interfacing
material [100–102].
In the proof-of-concept experiment performed by Yan et al. [34], FLG films
were exfoliated from highly oriented pyrolytic graphite (HOPG) and trans-
ferred in contact with the drain of AlGaN/GaN devices on a semi-insulating
4H–SiC substrate. Figure 11.24a illustrates the structure of a tested Al0.2 Ga0.8 N
(30 nm)/GaN (0.5 μm) transistor with FLG flakes transferred on top of it as
heat spreaders. Figure 11.24b shows a direct comparison between the output
characteristics (I D −V D ) of the device without (solid lines) and with FLG heat
spreaders (dashed lines). A significant increase in the output current I D (12% at
V G = 2 V and 8% at V G = 0 V) can be observed as a result of better heat removal
with the local FLG heat spreaders.
The temperature rise ΔT due to self-heating in the channel region of operating
(i.e. biased) AlGaN/GaN transistors was also monitored in situ by micro-Raman
spectroscopy. The laser probe was focused between the gate and the drain (closer
to the gate), where ΔT is expected to be the highest and ΔT was evaluated from
the shift of the position of the characteristic Raman peak at 567 cm−1 associated
with the E2 mode of GaN [103]. As an example, at a power density of 12.8 W/mm,
ΔT for the AlGaN/GaN transistor with and without Gr heat spreaders was 92
and 118 ∘ C, respectively. Those experiments presented a direct evidence of the
improvement in the device performance with the top-surface FLG heat spreaders.
Although these proof-of-concept experiments were performed using FLG
exfoliated from graphite, practical applications of Gr heat spreaders could be
enabled by the progress of Gr CVD growth directly on GaN and nitride semi-
conductors. Because the FLG quality for heat spreaders does not need to be as
high as that for the electronic applications, probably this could represent the first
application of CVD grown Gr on nitrides. Furthermore, recent demonstration of
direct low-temperature growth of synthetic diamond on GaN [104] can lead to
the development of heterogeneous FLG diamond lateral heat spreaders, where
the diamond layers provide electrical insulation and additional heat spreading
[105].
Drain current density (A mm–1)
VG = 2 V
Heat sink FLG heat spreader 0.8
Drain Gate
Source
SiO2 0.6
VG = 0 V
0.4
AIGaN
GaN
0.2 VG = –2 V
SiC substrate
0.0
0 5 10 15
(a) (b) VDS (V)
Figure 11.24 (a) Scheme of an AlGaN/GaN heterostructure transistor with FLG flakes transferred on top of it as surface heat spreaders. (b) Comparison of the
I–V characteristics of the transistor without (solid lines) and with (dashed lines) Gr heat spreaders. Source: Adapted with permission from Yan et al. [34].
Copyright 2012, Nature Publishing Group.
Acknowledgments 431
11.6 Summary
In summary, this chapter presented an overview of recent developments in the
integration of 2D materials (specifically Gr and MoS2 ) with nitride semiconduc-
tors for electronics and optoelectronics.
The state-of-the art approaches for the fabrication of heterojunctions between
these two classes of materials have been discussed, considering advantages
and limitations. Although the CVD growth of Gr on AlN at high temperatures
(>1250 ∘ C) have been recently reported, the transfer of Gr grown on catalytic
metals still remains the main approach to integrate Gr with GaN materials. On
the other hand, many progresses have been reported in the lattice-matched CVD
growth of MoS2 monolayer on GaN, as well as in the use of MoS2 as templates
for strain-free GaN deposition. Examples of post-CMOS electronic devices
based on 2D material/nitride heterostructures have been presented, such as the
HET with a Gr base and Al(Ga)N/GaN emitter for THz electronics, and the
band-to-band tunneling diode based on p+ -MoS2 /n+ -GaN heterojunctions for
digital electronics with ultralow-power dissipation. Furthermore, progresses
and limitations in the use of Gr or Gr-based transparent conductive electrodes
for GaN blue LEDs have been illustrated, with a focus on the issues related to the
Gr/p-GaN Schottky barrier. A recently reported example of self-powered deep
UV photodetector based on n-MoS2 /p-GaN heterojunction has been discussed.
Finally, some results on the use of Gr heat spreaders for thermal management in
high-power AlGaN/GaN HEMTs have been presented.
All these results hold great promises for the future developments of a 2D mate-
rial/nitride hybrid technology for next-generation electronic and optoelectronic
devices.
Acknowledgments
The authors would like to acknowledge the following colleagues for useful dis-
cussions: F. Roccaforte, P. Fiorenza, G. Greco, S. Di Franco, I. Deretzis, A. La
Magna, G. Nicotra (CNR-IMM, Catania, Italy), F. Iucolano, and S. Ravesi (STMi-
croelectronics, Catania, Italy), A. Michon (CNRS-CRHEA, Valbonne, France), P.
Kruszewski and M. Leszczynski (IHPP-PAS UNIPRESS, Warsaw, Poland), B. Pecz
(MFA, Budapest, Hungary), A. Kakanakova, and R. Yakimova (Linköping Univer-
sity, Sweden). This work has been supported, in part, by the Flag-ERA JTC 2015
project “GraNitE: Graphene heterostructures with Nitrides for high frequency
Electronics” (MIUR Grant No. 0001411), by the Flag-ERA JTC 2019 project “ET-
MOS: Epitaxial Transition Metal dichalcogenides Onto wide bandgap hexagonal
Semiconductors for advanced electronics”, and by the National Project PON “El-
eGaNTe: Electronics on GaN-based Technologies” (ARS01_01007).
References
1 Novoselov, K.S., Geim, A.K., Morozov, S.V. et al. (2004). Electric field effect
in atomically thin carbon films. Science 306: 666.
2 Geim, A.K. and Grigorieva, I.V. (2013). Van der Waals heterostructures.
Nature 499: 419–425.
3 Giannazzo, F. and Raineri, V. (2012). Graphene: synthesis and nanoscale
characterization of electronic properties. Rivista del Nuovo Cimento 35:
267–304.
4 Mayorov, A.S., Gorbachev, R.V., Morozov, S.V. et al. (2011).
Micrometer-scale ballistic transport in encapsulated graphene at room
temperature. Nano Lett. 11: 2396–2399.
5 Sonde, S., Giannazzo, F., Vecchio, C. et al. (2010). Role of graphene/substrate
interface on the local transport properties of the two-dimensional electron
gas. Appl. Phys. Lett. 97: 132101.
6 Giannazzo, F., Sonde, S., Lo Nigro, R. et al. (2011). Mapping the density of
scattering centers limiting the electron mean free path in graphene. Nano
Lett. 11: 4612–4618.
7 Bolotin, K.I., Sikes, K.J., Hone, J.H. et al. (2008). Temperature-dependent
transport in suspended graphene. Phys. Rev. Lett. 101: 096802.
8 Dean, C.R., Young, A.F., Meric, I. et al. (2010). Boron nitride substrates for
high-quality graphene electronics. Nat. Nanotechnol. 5: 722–726.
9 Nair, R.R., Blake, P., Grigorenko, A.N. et al. (2008). Fine structure constant
defines transparency of graphene. Science 320: 1308–1308.
10 Bonaccorso, F., Sun, Z., Hasan, T., and Ferrari, A.C. (2010). Graphene pho-
tonics and optoelectronics. Nat. Photonics 4: 611–622.
11 Lin, Y.-M., Dimitrakopoulos, C., Jenkins, K.A. et al. (2010). 100-GHz transis-
tors from wafer-scale epitaxial graphene. Science 327: 662.
12 Wu, Y., Jenkins, K.A., Valdes-Garcia, A. et al. (2012). State-of-the-art
graphene high-frequency electronics. Nano Lett. 12: 3062–3067.
13 Wang, Q.H., Zadeh, K.K., Kis, A. et al. (2012). Electronics and optoelectron-
ics of two-dimensional transition metal dichalcogenides. Nat. Nanotechnol.
7: 699–712.
14 Radisavljevic, B., Radenovic, A., Brivio, J. et al. (2011). Single-layer MoS2
transistors. Nat. Nanotechnol. 6: 147–150.
15 Kim, S., Konar, A., Hwang, W.S. et al. (2012). High-mobility and low-power
thin-film transistors based on multilayer MoS2 crystals. Nat. Commun. 3:
1011.
16 Giannazzo, F., Fisichella, G., Greco, G. et al. (2017). Ambipolar MoS2 tran-
sistors by nanoscale tailoring of Schottky barrier using oxygen plasma
functionalization. ACS Appl. Mater. Interfaces 9: 23164–23174.
17 Meric, I., Han, M.Y., Young, A.F. et al. (2008). Current saturation in
zero-bandgap, top-gated graphene field-effect transistors. Nat. Nanotechnol.
3: 654.
18 Fiori, G., Szafranek, B.N., Iannaccone, G., and Neumaier, D. (2013). Velocity
saturation in few-layer MoS2 transistor. Appl. Phys. Lett. 103: 233509.
References 433
19 Balandin, A.A. (2011). Thermal properties of graphene and nanostructured

carbon materials. Nat. Mater. 10: 569.
20 Yoon, Y., Ganapathi, K., and Salahuddin, S. (2011). How good can monolayer
MoS2 transistors be? Nano Lett. 11: 3768.
21 Yan, R., Simpson, J.R., Bertolazzi, S. et al. (2014). Thermal conductivity of
monolayer molybdenum disulfide obtained from temperature-dependent
Raman spectroscopy. ACS Nano 8: 986.
22 Tosun, M., Fu, D., Desai, S.B. et al. (2015). MoS2 heterojunctions by thick-
ness modulation. Sci. Rep. 5: 10990.
23 Jo, I., Pettes, M.T., Ou, E. et al. (2014). Basal-plane thermal conductivity of
few-layer molybdenum disulfide. Appl. Phys. Lett. 104: 201902.
24 Giannazzo, F., Greco, G., Roccaforte, F., and Sonde, S.S. (2018). Vertical
transistors based on 2D materials: status and prospects. Crystals 8: 70.
25 Jariwala, D., Marks, T.J., and Hersam, M.C. (2017). Mixed-dimensional van
der Waals heterostructures. Nat. Mater. 16: 170–181.
26 Fisichella, G., Greco, G., Roccaforte, F., and Giannazzo, F. (2014). Current
transport in graphene/AlGaN/GaN vertical heterostructures probed at
nanoscale. Nanoscale 6: 8671–8680.
27 Giannazzo, F., Fisichella, G., Greco, G. et al. (2017). Graphene integration
with nitride semiconductors for high power and high frequency electronics.
Phys. Status Solidi A 214: 1600460.
28 Zubair, A., Nourbakhsh, A., Hong, J.-Y. et al. (2017). Hot electron transistor
with van der Waals base-collector heterojunction and high-performance
GaN emitter. Nano Lett. 17: 3089–3096.
29 Wang, L., Liu, W., Zhang, Y. et al. (2015). Graphene-based transparent
conductive electrodes for GaN-based light emitting diodes: challenges and
countermeasures. Nano Energy 12: 419–436.
30 Chandramohan, S., Kang, J.H., Ryu, B.D. et al. (2013). Impact of interlayer
processing conditions on the performance of GaN light-emitting diode with
specific NiOx /graphene electrode. ACS Appl. Mater. Interfaces 5: 958–964.
31 Wang, L., Zhang, Y., Li, X. et al. (2013). Improved transport properties of
graphene/GaN junctions in GaN-based vertical light emitting diodes by acid
doping. RSC Adv. 3: 3359.
32 Chen, Z., Zhang, X., Dou, Z. et al. (2018). High-brightness blue
light-emitting diodes enabled by a directly grown graphene buffer layer.
Adv. Mater. 30: 1801608.
33 Araki, T., Uchimura, S., Sakaguchi, J. et al. (2014). Radio-frequency
plasma-excited molecular beam epitaxy growth of GaN on graphene/Si(100)
substrates. Appl. Phys Express 7: 071001.
34 Yan, Z., Liu, G., Khan, J.M., and Balandin, A.A. (2012). Graphene quilts for
thermal management of high-power GaN transistors. Nat. Commun. 3: 827.
35 Han, N., Cuong, T.V., Han, M. et al. (2013). Improved heat dissipation in
gallium nitride light-emitting diodes with embedded graphene oxide pattern.
Nat. Commun. 4: 1452.
36 Giannazzo, F., Greco, G., Schilirò, E. et al. (2019). High-performance
graphene/AlGaN/GaN Schottky junctions for hot electron transistors. ACS
Appl. Electron. Mater. 1: 2342–2354.
37 Ruzmetov, D., Zhang, K., Stan, G. et al. (2016). Vertical 2D/3D semiconduc-
tor heterostructures based on epitaxial molybdenum disulfide and gallium
nitride. ACS Nano 10: 3580–3588.
38 Krishnamoorthy, S., Lee, E.W., Hee Lee, C. et al. (2016). High current den-
sity 2D/3D MoS2 /GaN Esaki tunnel diodes. Appl. Phys. Lett. 109: 183505.
39 Ionescu, A.M. and Riel, H. (2011). Tunnel field-effect transistors as
energy-efficient electronic switches. Nature 479: 329–337.
40 Zhuo, R., Wang, Y., Wu, D. et al. (2018). High-performance self-powered
deep ultraviolet photodetector based on MoS2 /GaN p–n heterojunction. J.
Mater. Chem. C 6: 299.
41 Reina, A., Jia, X., Ho, J. et al. (2009). Large area few-layer graphene films on
arbitrary substrates by chemical vapor deposition. Nano Lett. 9: 30.
42 Li, X., Cai, W., An, J. et al. (2009). Large-area synthesis of high-quality and
uniform graphene films on copper foils. Science 324: 1312–1314.
43 Fisichella, G., Di Franco, S., Roccaforte, F. et al. (2014). Microscopic mech-
anisms of graphene electrolytic delamination from metal substrates. Appl.
Phys. Lett. 104: 233105.
44 Bae, S., Kim, H., Lee, Y. et al. (2010). Roll-to-roll production of 30-inch
graphene films for transparent electrodes. Nature Nanotech. 5 (8): 574–578.
45 Lupina, G., Kitzmann, J., Costina, I. et al. (2015). Residual metallic con-
tamination of transferred chemical vapor deposited graphene. ACS Nano 9:
4776–4785.
46 Giannazzo, F., Fisichella, G., Greco, G. et al. (2017). Fabrication and charac-
terization of graphene heterostructures with nitride semiconductors for high
frequency vertical transistors. Phys. Status Solidi A: 1700653.
47 Fisichella, G., Di Franco, S., Fiorenza, P. et al. (2013). Micro- and nanoscale
electrical characterization of large-area graphene transferred to functional
substrates. Beilstein J. Nanotechnol. 4: 234.
48 Giannazzo, F., Fisichella, G., Greco, G. et al. (2017). Conductive atomic
force microscopy of two-dimensional electron systems: from AlGaN/GaN
heterostructures to graphene and MoS2 . In: Conductive Atomic Force
Microscopy: Applications in Nanomaterials, 1–28. Wiley-VCH.
49 Sonde, S., Giannazzo, F., Raineri, V. et al. (2009). Electrical properties of the
graphene/4H-SiC (0001) interface probed by scanning current spectroscopy.
Physiol. Rev. B 80: 241406(R).
50 Michon, A., Tiberj, A., Vezian, S. et al. (2014). Graphene growth on AlN
templates on silicon using propane-hydrogen chemical vapor deposition.
51 Dagher, R., Matta, S., Parret, R. et al. (2017). High temperature annealing
and CVD growth of few-layer graphene on bulk AlN and AlN templates.
Phys. Status Solidi A 214: 1600436.
52 Lee, Y.-H., Zhang, X.Q., Zhang, W. et al. (2012). Synthesis of large-area
MoS2 atomic layers with chemical vapor deposition. Adv. Mater. 24:
2320–2325.
References 435
53 Dumcenco, D., Ovchinnikov, D., Marinov, K. et al. (2015). Large-area epitax-

ial monolayer MoS2 . ACS Nano 9: 4611–4620.
54 Tan, L.K., Liu, B., Teng, J.H. et al. (2014). Atomic layer deposition of a MoS2
film. Nanoscale 6: 10584.
55 Barton, A.T., Yue, R., Anwar, S. et al. (2015). Transition metal dichalcogenide
and hexagonal boron nitride heterostructures grown by molecular beam epi-
taxy. Microelectron. Eng. 147: 306–309.
56 Serrao, C.R., Diamond, A.M., Hsu, S.-L. et al. (2015). Highly crystalline
MoS2 thin films grown by pulsed laser deposition. Appl. Phys. Lett. 106:
052101.
57 Serna, M.I., Yoo, S.H., Moreno, S. et al. (2016). Large-area deposition of
MoS2 by pulsed laser deposition with in situ thickness control. ACS Nano
10: 6054–6061.
58 Chromik, S., Sojková, M., Vretenár, V. et al. (2017). Influence of
GaN/AlGaN/GaN (0001) and Si (100) substrates on structural properties
of extremely thin MoS2 films grown by pulsed laser deposition. Appl. Surf.
Sci. 395: 232–236.
59 Huang, L.F., Gong, P.L., and Zeng, Z. (2014). Correlation between structure,
phonon spectra, thermal expansion, and thermomechanics of single-layer
MoS2 . Phys. Rev. B 90: 045409.
60 Leszczynski, M., Suski, T., Teisseyre, H. et al. (1994). Thermal expansion of
gallium nitride. J. Appl. Phys. 76: 4909–4911.
61 Ly, T.H., Perello, D.J., Zhao, J. et al. (2016). Misorientation-angle-dependent
electrical transport across molybdenum disulfide grain boundaries. Nat.
Commun. 7: 10426.
62 Giannazzo, F., Bosi, M., Fabbri, F. et al. (2020). Direct probing of grain
boundary resistance in chemical vapor deposition-grown monolayer MoS2
by conductive atomic force microscopy. Phys. Status Solidi RRL 14: 1900393.
63 Kovács, A., Duchamp, M., Dunin-Borkowski, R.E. et al. (2015). Graphoepi-
taxy of high-quality GaN layers on graphene/6H–SiC. Adv. Mater. Interfaces
2: 1400230.
64 Kim, J., Bayram, C., Park, H. et al. (2014). Principle of direct van der Waals
epitaxy of single-crystalline films on epitaxial graphene. Nat. Commun. 5:
4836.
65 Gupta, P., Rahman, A.A., Subramanian, S. et al. (2016). Layered transition
metal dichalcogenides: promising near lattice-matched substrates for GaN
growth. Sci. Rep. 6: 23708.
66 Tangi, M., Mishra, P., Ng, T.K. et al. (2016). Determination of band offsets at
GaN/single-layer MoS2 heterojunction. Appl. Phys. Lett. 109: 032104.
67 Tangi, M., Mishra, P., Li, M.-Y. et al. (2017). Type-I band alignment at
MoS2 /In0.15 Al0.85 N lattice matched heterojunction and realization of MoS2
quantum well. Appl. Phys. Lett. 111: 092104.
68 Rode, J.C., Chiang, H.-W., Choudhary, P. et al. (2015). Indium phosphide
heterobipolar transistor technology beyond 1-THz bandwidth. IEEE Trans.
69 Tang, Y. et al. (2015). Ultrahigh-speed GaN high-electron-mobility transis-

tors with fT/ fmax of 454/444 GHz. IEEE Electron Device Lett. 36: 549–551.
70 Mead, C.A. (1961). Operation of tunnel-emission devices. J. Appl. Phys. 32:
646–652.
71 Atalla, M.M. and Soshea, R.W. (1963). Hot-carrier triodes with thin-film
metal base. Solid-State Electron. 6: 245–250.
72 Vaziri, S., Lupina, G., Henkel, C. et al. (2013). A graphene-based hot electron
transistor. Nano Lett. 13: 1435.
73 Zeng, C., Song, E.B., Wang, M. et al. (2013). Vertical graphene-base
hot-electron transistor. Nano Lett. 13: 2370.
74 Giannazzo, F., Greco, G., Roccaforte, F. et al. (2018). Hot electron transis-
tors with graphene base for THz electronics. In: Low Power Semiconductor
Devices and Processes for Emerging Applications in Communications, Com-
puting, and Sensing (ed. S. Walia), 95–115. CRC Press.
75 Mehr, W., Dabrowski, J., Scheytt, J.C. et al. (2012). Vertical graphene base
transistor. IEEE Electron Device Lett. 33: 691–693.
76 Kong, B.D., Jin, Z., and Kim, K.W. (2014). Hot-electron transistors for ter-
ahertz operation based on two-dimensional crystal heterostructures. Phys.
Rev. Appl. 2: 054006.
77 Di Lecce, V., Grassi, R., Gnudi, A. et al. (2013). Graphene-base heterojunc-
tion transistor: an attractive device for terahertz operation. IEEE Trans.
Electron Devices 60: 4263–4268.
78 Prystawko, P., Giannazzo, F., Krysko, M. et al. (2019). Growth and charac-
terization of thin Al-rich AlGaN on bulk GaN as an emitter-base barrier for
hot electron transistor. Mater. Sci. Semicond. Process. 93: 153–157.
79 Grabowski, S.P., Schneider, M., Nienhaus, H. et al. (2001). Electron affinity of
Alx Ga1−x N(0001) surfaces. Appl. Phys. Lett. 78: 2503–2505.
80 Schilirò, E., Lo Nigro, R., Roccaforte, F., and Giannazzo, F. (2019). Recent
advances in seeded and seed-layer-free atomic layer deposition of high-K
dielectrics on graphene for electronics. C – J. Carbon Res. 5: 53.
81 Fisichella, G., Schilirò, E., Di Franco, S. et al. (2017). Interface electrical
properties of Al2 O3 thin films on graphene obtained by atomic layer deposi-
tion with an in situ seedlike layer. ACS Appl. Mater. Interfaces 9: 7761–7771.
82 Minami, T. (2005). Transparent conducting oxide semiconductors for trans-
parent electrodes. Semicond. Sci. Technol. 20: S35–S44.
83 Lee, J.Y., Connor, S.T., Cui, Y., and Peumans, P. (2008). Solution-processed
metal nanowire mesh transparent electrodes. Nano Lett. 8: 689–692.
84 Geng, H.Z., Kim, K.K., So, K.P. et al. (2007). Effect of acid treatment on car-
bon nanotube-based flexible transparent conducting films. J. Am. Chem. Soc.
129: 7758–7759.
85 Tongay, S., Lemaitre, M., Miao, X. et al. (2012). Rectification at
graphene-semiconductor interfaces: zero-gap semiconductor-based diodes.
Phys. Rev. X 2: 011002.
86 Tongay, S., Lemaitre, M., Schumann, T. et al. (2011). Graphene/GaN Schot-
tky diodes: stability at elevated temperatures. Appl. Phys. Lett. 99: 102102.
87 Zhong, H., Liu, Z., Xu, G. et al. (2012). Self-adaptive electronic contact
between graphene and semiconductor. Appl. Phys. Lett. 100: 122108.
References 437
88 Jo, G., Choe, M., Cho, C.Y. et al. (2010). Large-scale patterned multi-layer
graphene films as transparent conducting electrodes for GaN light-emitting
diodes. Nanotechnology 21 (17): 175201.
89 Seo, T.H., Oh, T.S., Chae, S.J. et al. (2011). Enhanced light output power of
GaN light-emitting diodes with graphene film as a transparent conducting
electrode. Jpn. J. Appl. Phys. 50: 125103.
90 Shim, J.-P., Hoon Seo, T., Min, J.-H. et al. (2013). Thin Ni film on graphene
current spreading layer for GaN-based blue and ultra-violet light-emitting
91 Zhang, Y., Li, X., Wang, L. et al. (2012). Enhanced light emission of
GaN-based diodes with a NiOx /graphene hybrid electrode. Nanoscale 4
(19): 5852–5855.
92 Min Lee, J., Yi, J., Woo Lee, W. et al. (2012). ZnO nanorods-graphene hybrid
structures for enhanced current spreading and light extraction in GaN-based
light emitting diodes. Appl. Phys. Lett. 100 (6): 061107.
93 Seo, T.H., Park, A.H., Lee, G.H. et al. (2014). Efficiency enhancement of
nanorod green light emitting diodes employing silver nanowire-decorated
graphene electrode as current spreading layer. J. Phys. D: Appl. Phys. 47 (31):
315102.
94 Shim, J.P., Kim, D., Choe, M. et al. (2012). A self-assembled Ag nanoparticle
agglomeration process on graphene for enhanced light output in GaN-based
LEDs. Nanotechnology 23 (25): 255201.
95 Choe, M., Cho, C.-Y., Shim, J.-P. et al. (2012). Au nanoparticle-decorated
graphene electrodes for GaN-based optoelectronic devices. Appl. Phys. Lett.
101 (3): 031115.
96 Schubert, E.F. (2006). Light-Emitting Diodes, 87. Cambridge: Cambridge Uni-
versity Press.
97 Duan, L., He, F., Tian, Y. et al. (2017). Fabrication of self-powered
fast-response ultraviolet photodetectors based on graphene/ZnO:Al
nanorod-array-film structure with stable Schottky barrier. ACS Appl. Mater.
Interfaces 9: 8161–8168.
98 Zhao, B., Wang, F., Chen, H. et al. (2017). Adv. Funct. Mater. 27: 1700264.
99 Chen, G. and Hui, P. (1999). Thermal conductivities of evaporated gold films
on silicon and glass. Appl. Phys. Lett. 74: 2942–2944.
100 Mak, K.F., Liu, C.H., and Heinz, T.F. (2010). Thermal conductance at the
graphene-SiO2 interface measured by optical pump-probe spectroscopy.
101 Koh, Y.K., Bae, M.-H., Cahill, D.G., and Pop, E. (2010). Heat conduction
across monolayer and few-layer graphenes. Nano Lett. 10: 4363–4368.
102 Schmidt, A.J., Collins, K.C., Minnich, A.J., and Chen, G. (2010). Thermal
conductance and phonon transmissivity of metal-graphite interfaces. J. Appl.
Phys. 107: 104907.
103 Liu, M.S., Bursill, L.A., Prawer, S. et al. (1999). Temperature dependence
of Raman scattering in single crystal GaN films. Appl. Phys. Lett. 74:
3125–3127.
104 Goyal, V., Sumant, A.V., Teweldebrhan, D., and Balandin, A.A. (2012). Direct
low-temperature integration of nanocrystalline diamond with GaN substrates
for improved thermal management of high-power electronics. Adv. Funct.

Mater. 22: 1525–1530.
105 Tadjer, M.J., Anderson, T.J., Hobart, K.D. et al. (2012). Reduced self-heating
in AlGaN/GaN HEMTs using nanocrystalline diamond heat-spreading films.
IEEE Elec. Dev. Lett. 33: 23–25.
439
Index
a blue emitting InGaN QWs

advanced Ohmic contacts technology cavity effect 380–381
117 cavity-enhanced green VCSELs
AlGaN deep UV LEDs 381–384
high-quality AlN growth blue LEDs 24, 254, 258–263, 265, 267,
278–281 271, 272, 276, 277, 287, 336, 367,
internal quantum efficiency 375, 381, 408, 425, 427, 431
278–281 blue VCSELs
light extraction efficiency (LEE) emission properties 388–389
282–287 index-guided structure, design of
potential applications 276 386–388
UVC LEDs 281–282 broadband amplifiers 30, 99, 102
AlGaN/GaN heterostructures buffer trapping 111, 225–229
Schottky gate 138 bulk GaN growth
transistor with FLG flakes 430 ammonothermal growth 46–51
two-dimensional electron gas (2DEG) hydride vapor-phase epitaxy (HVPE)
19–23 43–45
AlGaN/GaN HEMTs 138–140 sodium flux growth method 45–46
gate-edge degradation 207–209
hot electron degradation 209–211 c
threshold voltage in 22, 138–140 carbon dopants 222
trapping effects 204–207 carbon-doped GaN buffer
Arrhenius model 201 insulation 221–223
Auger recombination 74, 78, 254–256, dynamic RDS,ON due to buffer
258, 306 trapping 225–229
avalanche electroluminescence 178, time-dependent “dielectric”
186–188 breakdown (TDDB) 223–225
carrier delocalization 254, 256–257,
b 262
barrier layer 19–23, 28, 114, 115, 117, cascade 125, 126, 137, 138, 140–142,
118, 138–140, 143–146, 149, 154, 165
156–158, 163, 205, 212, 233, 269, cavity-enhanced green VCSELs
377, 415, 420 381–384
bis-cyclo-pentadienyl-magnesium close-coupled showerhead reactor
(Cp2 Mg) 66 56–57, 68, 71

440 Index
color rendering index (CRI) 253, f

267 fabrication, of 2D materials
current aperture vertical electron heterostructures
transistor (CAVET) 178–182 direct growth 403–407
foreign substrate, transfer of
d 400–403
deep ultraviolet (DUV) nitride semiconductors films
AlGaN-based UVC LEDs 281–282 407–413
high-quality AlN growth 278–281 facet-assisted epitaxial lateral
internal quantum efficiency overgrowth (FACELO) 62–64
278–281 5G network system 103, 108
light extraction efficiency 282–287 GaN base station Pas 106–108
potential applications 276 GaN for 104–105
digital predistortion (PDP) technique 6G 108
106 few-layer graphene (FLG) 408
19 F+ -ion implantation 148
distributed feedback laser diodes (DFB
LDs) 321–324, 342–348 fluorine (F) 138, 140, 145–149, 154,
Doherty PA technology 106 165
double dielectric DBR structure fluorinated HEMT 145–149
371–373, 379, 381, 386
dual-wavelength lasing 384–386 g
gallium nitride (GaN) 1, 99
application
e
optoelectronic devices 24–26
efficiency droop phenomenon 253,
power-and high-frequency
258
electronic devices 26–30
electroluminescence (EL) 78, 178,
avalanche electroluminescence
187, 188, 203, 255, 268, 282, 348,
186–188
349
base station Pas 106–108
electronic devices
buffer layers 114
reliability testing and failure analysis
channel 114
200–204 electronic devices see electronic
of 2D materials heterostructures devices
band-to-band tunneling diodes 5G for 104–105
413–414 HEMT “cascade,” 140–142
hot electron transistors 414–420 high-voltage diodes 185–186
MoS2 /GaN heterojunctions history 1–4
413–414 impact ionization coefficients in
electron leakage 188–193
blue LEDs 258–262 P–N diodes 186–188
green LEDs 262–264 properties 4–6
enhanced mobile broadband (eMBB) electrical 16–19
103 microstructure and related issues
epitaxial lateral overgrowth (ELOG) 7–13
62, 63, 337, 368 optical 13–16
external quantum efficiency (EQE) 74, two-dimensional electron gas
254, 258, 278 19–23
Index 441
reliability testing and failure analysis threshold voltage 138

200–204 high-power amplification 100–101
RF applications high-temperature operating life (HTOL)
AlGaN/GaN HEMTs 204–211 test 200, 201, 228
InAlN/GaN HEMTs 211–215 high-voltage diodes 185–186
thermal issues 215–219 hot electron degradation
on SiC and Si 60–62 AlGaN/GaN HEMTs 209–211
V th instabilities in 233–240 InAlN/GaN HEMTs 212–214
GaNification 1, 25 hot electron transistor (HET) 30, 399,
gate degradation 204, 230–233 414–420
gate-edge degradation 203, 207–209 hot phonons role 214–215
gate injection transistor (GIT) 3, 157, hybrid DBR structure 370–371
158, 221, 228 hydride-based vapor phase epitaxy
graphene (Gr) 397 (HVPE) 42
few-layer graphene (FLG) 408 hydride vapor phase epitaxy (HVPE)
monolayer of 398 2, 42–45
for thermal management 428–430
graphene-transparent conductive i
electrodes 421–427 indium compositional inhomogeneity
green gap 4, 24, 25, 74, 262, 266, 268, 75
273, 307, 309, 310, 313, 369, 375, indium inclusions embedded within
377, 379, 389 v-defects 74
green LEDs 24, 25, 262–264, 288 inductively coupled plasma (ICP) 144,
green vertical-cavity surface-emitting 181, 286, 379
laser (VCSELs) InAlN/GaN HEMTs
advantages 375–377 hot electron degradation 212–214
cavity effect 380–381 hot phonons role 214–215
cavity properties 381 InGaN QDs
fabrication process 379 green VCSELs on 375–377
InGaN QDs growth and optical growth and optical properties of
properties 377–379 377–379
properties of cavity-enhanced InGaN QWs
381–384 blue emitting 380–384
decomposition 79–82
h fluctuations 75–78
halide 42, 48, 335 homogenization 78–79
heterojunction tunnel diodes 413 polar, nonpolar, and semipolar GaN
hexagonal symmetry 58, 60, 407 substrates growth 72–75
high electron mobility transistors internal quantum efficiency (IQE) 15,
(HEMTs) 21, 137, 199 73, 74, 254, 278–281, 306,
cascode 140–142 378
normally-off technology
fluorinated 145–149 l
p-GaN gate HEMTs 155–162 laser diodes (LDs) 303
recessed-gate HEMTs 142–145 bulk ammono-GaN 313–316
recessed-gate hybrid MISHEMT distributed feedback 321–324,
149–155 342–348
442 Index
laser diodes (LDs) (contd.) low-noise amplifier (LNA) 124, 125

history and development luminous efficacy (LER) 265–267,
gallium nitride technology 271, 272, 276
335–336
nitride laser diodes 337–342 m
optoelectronics 333–335 measurement-stress-measurement
plasma-assisted MBE 303–305 (MSM) technique 238
semiconductor optical amplifiers median time to failure (MTTF) models
354–357 200–202
superluminescent diodes (SLD) memory effect 66
development 348–351 metal–insulator–semiconductor high
optimization 353–354 electron mobility transistors
properties 351–353 (MISHEMTs) 23, 138,
with tunnel junctions 233–240
316–324 metal organic chemical vapor deposition
vertically interconnected stacks (MOCVD) 114, 211, 371
319–321 metal organic vapor-phase epitaxy
wide InGaN QWs 305–313 (MOVPE) 2, 41, 42, 53
lateral overgrowth from trenches basics about nitride 54–57
(LOFT) 63 binary and ternary nitrides growth
light-emitting diodes (LEDs) 67–71
253 defect reduction 62–64
AlGaN deep UV epitaxy on foreign substrates 58–62
high-quality AlN growth in situ ELOG by SiN deposition 64
278–281 nitrides doping 64–67
internal quantum efficiency physical properties of 52
278–281 metal–oxide–semiconductor field effect
light extraction efficiency (LEE) transistors (MOSFETs) 137,
282–287 140–142, 152, 153, 155, 165, 177,
potential applications 276 178, 180, 182, 183, 192, 203, 223,
UVC LEDs 281–282 224, 238
Auger recombination millimeter-wave (mmW)
255–256 applications
blue 258 broadband amplifiers 102–103
carrier delocalization 256–257 high-power amplification
inorganic 253 100–101
electron leakage 257–264 5G 103–108
green 262 device design and fabrication
white advanced Ohmic contacts
monolithic LEDs 267–268 technology 117–118
phosphor-covered LEDs AlN-based devices performances
271–275 119–121
light extraction efficiency (LEE) 254, device scaling 116
258, 276, 282–287 InAlGaN-based device
liquid phase epitaxy (LPE) 41 performances 121–122
low-energy electron beam irradiation intrinsic characteristics 108–113
(LEEBI) 65 N-polar GaN HEMTs 118–119
Index 443
RF devices 114–116 MoS2 /GaN deep-UV photodetectors

state-of-the-art mm-Wave GaN 427–428
transistors 122–123 oxide gate interlayer field effect
T-shaped gates 116–117 transistor (OGFET) 177
material designs 108–116
molecular beam epitaxy (MBE) 114 p
III-N growth fundamentals Pendeo-epitaxy 63
303–304 phosphor-covered LEDs 271–275
quantum-confined Stark effect p-GaN gate HEMT 155–162
305–313 p-GaN switching HEMTs 230–233
monolithic LEDs 265–268, 270 phosphor-converted light-emitting
monolithic microwave integrated diode (pc-LED) 253, 265,
circuits (MMICs) 99, 125 273–274
Ka-band to D-band frequencies piezoelectricity 69, 199, 203, 303, 341
125–126 piezoelectric polarization 3, 8, 20, 21,
power amplifier III-N devices 23, 114, 257, 308, 309
III–V materials 123–124 planetary reactor 56
low noise amplifiers 125 P–N diode (PND) 178, 185–190, 193,
333
multi quantum wells (MQWs) 72, 262,
positive bias temperature instability
313
(PBTI) 220, 233, 234, 236–238,
240, 241
n
power-added efficiency (PAE) 109
nitride laser diodes 335–342,
power and high frequency electronic
354, 358
devices 13, 26–30
nonuniform distributed power
power amplifier GaN MMICs
amplifiers (NDPA) 102
Class A, A/B, and C for 124–125
normally-off HEMT technology
Class D, E, and F for 125
fluorinated 145–149 power average ratio (PAR) 106
p-GaN gate 155–162 power switching devices, reliability and
recessed-gate HEMT 142–145 robustness
recessed-gate hybrid MISHEMT breakdown vs. trapping 220
149–155 degradation location 220–221
N–P diode (NPD) 188, 189 device design 221
nucleation layer 2, 59–61, 114, 215, environment 221
217, 218, 278 stress condition 219–220
predictive modeling 193
o pyroelectricity 203, 341
Ohmic contacts 28, 29, 116, 117, 119,
158–160, 162, 260, 379, q
418, 425 QD-in-QW active structure 384–386
optical absorption (OA) 14 quantum barriers (QBs) 72, 311, 313
optical transmission (OT) 14 quantum-confined Stark effect (QCSE)
optoelectronic devices, of 2D materials 69, 305–313, 340, 341, 375
heterostructures quantum dots (QDs) 384
graphene-transparent conductive green VCSELs properties 379
electrodes 421–427 vs. quantum wells 375–377
444 Index
quantum dots (QDs) (contd.) semiconductor optical amplifiers (SOA)

quantum wells (QWs) 13, 72 339, 354–357
decomposition 79 sensor electronic technology (SET)
in fluctuations 75 277
homogenization 78 SiN cap layer 114
vs. quantum dots 375–377 Sixth generation (6G) networks 108,
polar, nonpolar and semipolar GaN 127
substrates 72 sodium flux growth method 45–46
static induction transistors (SITs) 178
r strain management 9, 340
recessed-gate HEMT 139, 142–145 superluminescent diodes (SLD)
recessed-gate hybrid MISHEMT 141, development 348–351
149–155, 165 optimization 353–354
reflection of high energy electron properties 351–353
diffraction (RHEED) 55 switch amplifier GaN MMICs 125
reliability 199
GaN HEMTs for RF applications t
204–219 thermal expansion coefficients (TECs)
power switching devices 219 8, 9, 44, 58, 59, 61, 62, 217, 370,
breakdown vs. trapping 220 399, 404
carbon-doped GaN buffer, parasitic thermal issues 203, 215–219, 229, 350
effects 221–229 threshold voltage (V th ) 4, 22, 29,
degradation location 220–221 137–140, 143–149, 152–154,
device design 221 156, 157, 159–163, 165, 179–185,
environment 221 200, 204, 337, 342, 372
gate degradation 230–233 time-dependent dielectric breakdown
stress condition 219 (TDDB) 203, 223–225
V th instabilities 233–240 transition metal dichalcogenides
testing and failure analysis 200–204 (TMDs) 397, 398, 431
RF applications trapping effects 29, 111, 112, 114, 119,
AlGaN/GaN HEMTs 204–211 120, 154, 204–207, 209, 230, 241
InAlN/GaN HEMTs 211–215 T-shaped gate design 116–117
thermal issues 215–219 tunnel junctions
RF GaN HEMTs 215–219 distributed feedback laser diodes
321–324
s stacks of vertically interconnected
sapphire 2, 5, 8, 9, 14, 15, 26, 42–46, laser diodes 319–321
58–61, 63, 64, 77, 115, 183, 254, turbo disk reactor 57
256–258, 277–279, 281, 282, 288, two-dimensional electron gas (2DEG)
336–338, 346, 371, 377, 379, 385, 19–23, 137, 178, 205, 415
399, 404, 407–411, 413, 424, 428
Schottky contacts 21, 23, 28, 29, 147, v
158, 164, 231, 232 vapor phase epitaxy (VPE) 41
Schottky gate 29, 122, 138, 139, 149, vertical MOSFETs 182–185
159, 201, 214, 233 vertical power devices
secondary ion mass spectrometry avalanche electroluminescence
(SIMS) 10, 76, 145, 283 186–188
Index 445
current aperture vertical electron growth and optical properties

transistor (CAVET) 178–182 377–379
ionization coefficients impact 188 properties of cavity-enhanced
MOSFETs 182–185 381–384
power conversion 177–178 history and current status 368–369
vertical-cavity surface-emitting lasers lasing characteristics 386
(VCSELs) 367 properties and application 367
with different DBR structure vertically interconnected laser diodes
with double dielectric 371 319–321
with hybrid 370–371
efficiency of heat dissipation w
cavity length 373–375 white LEDs
heat profile 372–373 lighting solutions 266
green monolithic 267–268
advantages 375–377 phosphor-covered 271–275
cavity effect 380–381 phosphor free approaches 265
fabrication process 379 wide band gap (WBG) 4, 6, 26

Nitride Semiconductor Technology - 2020 - Roccaforte

Uploaded by

Copyright:

Available Formats

You might also like

Nitride Semiconductor Technology - 2020 - Roccaforte

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nitride Semiconductor Technology - 2020 - Roccaforte

Uploaded by

Copyright:

Available Formats

Power Electronics and Optoelectronic Devices

Prof. Mike Leszczynski Library of Congress Card No.: applied for

© 2020 Wiley-VCH Verlag GmbH &

All rights reserved (including those of

Print ISBN: 978-3-527-34710-0

Typesetting SPi Global, Chennai, India

Printed on acid-free paper

1 Introduction to Gallium Nitride Properties and Applications 1

2 GaN-Based Materials: Substrates, Metalorganic Vapor-Phase

2.3.5 Doping of Nitrides 64

3 GaN-Based HEMTs for Millimeter-wave Applications 99

4 Technologies for Normally-oﬀ GaN HEMTs 137

5 Vertical GaN Power Devices 177

6 Reliability Issues in GaN Electronic Devices 199

6.3 Reliability and Robustness of GaN Power Switching Devices 219

7 Light-Emitting Diodes 253

8 Laser Diodes Grown by Molecular Beam Epitaxy 301

9 Edge Emitting Laser Diodes and Superluminescent

10 Green and Blue Vertical-Cavity Surface-Emitting Lasers 367

11 Integration of 2D Materials with Nitrides for Novel Electronic

GaN-based devices for mmW applications. Targeted applications including

they are reinventing and redeﬁning artiﬁcial lighting. Nevertheless, in their

Fabrizio Roccaforte and Mike Leszczynski

Introduction to Gallium Nitride Properties and Applications

n. 5 – Zona Industriale, 95121 Catania, Italy

1.1 Historical Background

Nitride Semiconductor Technology: Power Electronics and Optoelectronic Devices,

GaN HEMT device fabrication in Si CMOS (complementary metal–oxide

1.2 Basic Properties of Nitrides

Comparison with other

Comparison with other

1.2.1 Microstructure and Related Issues

Parameters AlN GaN InN

a0 (Å) 3.1113 3.1878 3.537

Lattice mismatch 16 3.5 −17 0

Figure 1.3 TEM plan view bright-ﬁeld image with

Figure 1.4 Optical microscopy view

parameters, makes the optimization of the crystal quality a very challenging

1.2.2 Optical Properties

Figure 1.5 Energy band gap of nitride 6 200

their ternary alloys) as a function on the

in-plane lattice parameter. The

corresponding wavelength is reported 300

FE-C FE-C, XC – free exciton C

1.2.3 Electrical Properties

Figure 1.8 Calculated intrinsic carrier

3 Figure 1.9 Electron velocity as a function

Si Critical electric field

Electron mobility Thermal conductivity

1.2.4 Two-Dimensional Electron Gas (2DEG) in AlGaN/GaN

EgAlGaN (x) = xEgAlN + (1 − x)EgGaN − x(1 − x)1.0 eV (1.7)

a0(GaN) > a0(AlGaN) a0(GaN) = a0(AlGaN) > a0(AlGaN)

AlGaN/GaN system is zero. Accordingly, as no assumption can be made on the

Drain current IDS (mA/mm)

seen, the output current of a HEMT can be modulated by the application of a

where N D is the doping density of the AlGaN barrier layer expressed in

1.3 Applications of GaN-Based Materials

1.3.1 Optoelectronic Devices