Professional Documents
Culture Documents
Session 1 Plenary
Session 1 Plenary
8:30 AM
FORMAL OPENING OF THE CONFERENCE
The Plenary Session starts with welcoming remarks and introduction from the Conference Chair, Eugenio Cantatore, followed by the International Technical
Program Chair, Piet Wambacq, providing an overview of ISSCC 2023.
The Plenary Session will feature four distinguished keynote speakers, who are leaders or pioneers in their domain, covering together a broad spectrum of our
industry. An Awards Ceremony that recognizes major technical and professional accomplishments presented by the IEEE, Solid-State Circuits Society (SSCS),
and ISSCC, will take place after the first two Plenary talks. Plenary presentations will be live streamed at the Conference and broadcast live on various platforms,
including the ISSCC YouTube Channel.
The first plenary talk “Innovation For the Next Decade of Compute Efficiency”, by Lisa Su, Chair and Chief Executive Officer at AMD, explains how to innovate
in high-performance computing, which is becoming an increasingly indispensable part of modern life. Innovation in this field is only possible with an effective,
holistic strategy to improve energy efficiency, which is described in this talk.
The second plenary talk “Shape the World with Mixed-Signal Integrated Circuits, - Past, Present, and Future”, by Akira Matsuzawa, Professor Emeritus of Tokyo
Institute of Technology and CEO of Tech Idea, shows how over the past decades digitalization in many applications has taken benefit of the performance, power
and cost improvements of mixed-signal integrated circuits. Further, an outlook to emerging applications and developments of mixed-signal circuits is given.
The third plenary talk “EU Chips Act Drives Pan-European Full-Stack Innovation Partnerships”, by Jo De Boeck, Executive Vice-President and Chief Strategy
Officer at imec, gives a projection on how the European Chips Act, which targets to increase Europe’s global semiconductor production, can trigger innovation
in the entire semiconductor ecosystem ranging from chip processing to system design, impacting a broad field of application domains such as heterogeneous
cloud and distributed computing, connectivity, automotive and health.
The last plenary talk “5G Drives Exponential Increase in Processing Needs Across all Industries”, by Erik Ekudden, Senior Vice President and Chief Technology
Officer at Ericsson, explains how 5G - and its successor 6G - drives an exponential increase in the collection, processing and transport of data in and by vehicles,
industrial equipment and consumer products, which are becoming ever more intelligent and connected. The talk will address whether the semiconductor industry
is ready to tackle these challenges.
We hope that you will find these presentations informative, inspiring, and motivating. Enjoy!
8:45 AM
1.1 Innovation For the Next Decade of Compute Efficiency
Lisa Su, Chair and Chief Executive Officer, AMD, Austin, TX
Although traditional scaling has slowed over the past decade, we have made tremendous progress as an industry with new
approaches including chiplet-based architectures, domain-specific accelerators, and advanced packaging technologies which have
enabled major milestones including the first exascale supercomputers. As we look into the future, we need to accelerate the pace
of innovation to drive the next decade of advancement in high-performance computing. By far, the largest limiting factor to delivering
continued compounded growth in computation power is energy efficiency. In this paper, we highlight a holistic strategy for
accelerating innovation in energy efficiency required for next-generation high-performance computing and ultimately achieving
zettascale performance. These approaches will be built on continued innovation in process technologies, modular chiplet
architectures, and advanced packaging. Fully meeting the challenge will require new dimensions of improvement through extending
domain-specific architectures to accelerate core algorithms in combination with wide-scale deployment of AI across all aspects of
the system from transistors to software.
1
9:20 AM
1.2 Shape the World with Mixed-Signal Integrated Circuits - Past, Present, and Future
Akira Matsuzawa, Professor Emeritus of Tokyo Institute of Technology and CEO of Tech Idea, Kawasaki, Japan
The past 50 years has been an era in which analog equipment has been replaced by digital counterparts. Audio, TV, video,
camcorder, camera, recording, wired connection, and wireless communication have been subject to digitization. The digitization
of these devices and systems was due to the technological shift from bipolar to CMOS, and to the development of logic and
memory circuits supported by scaling laws. In addition, design innovation in mixed-signal integrated circuits such as ADCs and
DACs has shown to be indispensable. This talk will look back on the digitization of equipment and the mixed-signal integrated
circuit technology that contributed to it. Further, we will look forward to future applications and developments.
9:55 AM
ISSCC, SSCS, IEEE AWARD PRESENTATIONS
10:20 AM
BREAK
10:40 AM
1.3 EU Chips Act Drives Pan-European Full-Stack Innovation Partnerships
Jo De Boeck, Executive Vice President and Chief Strategy Officer, imec & KU Leuven, Leuven, Belgium
In every aspect of our life and society, semiconductors play a major role. The pandemic in conjunction with supply chain hiccups
and geopolitical tensions made all regions realize that they need to revisit their presence in the semiconductor value chain. The
European Commission projected the ambition of achieving a 20% share of the global semiconductor production by 2030.
Europe can leverage existing strengths such as, among others, the unique position of equipment companies and leadership
positions in 300mm semiconductor technology R&D. The Chips-for-Europe initiative will invest in pilot lines and ecosystems
for chip manufacturing, embracing leading-edge and first-of-a-kind technologies. The pilot lines will allow early exploration of
the potential impact of new technology features in advanced chip and system architectures. This will trigger increased demand
and accelerate industrial uptake of the novel technologies. This type of innovation loop is also essential for deep-tech start-ups
building their unique value proposition. The full-stack, networked model of industry collaboration is at the core of the EU Chips
Act ambition and will impact different application domains such as heterogeneous cloud and distributed computing, connectivity,
automotive, and health.
It is crucial for all this innovation potential that we, as an industry, consider that semiconductor manufacturing is resource-
intensive with respect to energy, water, chemicals, and raw materials. Design-technology co-optimization (DTCO) and
System-Technology co-optimization (STCO) methodologies can develop a framework for early sustainability assessments of
logic technologies. Finally, we urgently need to get the message across that climate, health, safety, and human connectedness
all require complex digital backbones, if we want to stand a chance of attracting the right talent.
11:15 AM
1.4 5G Drives Exponential Increase in Processing Needs Across all Industries
Erik Ekudden, Senior Vice President & Chief Technology Officer, Ericsson, Kista, Sweden
Across essentially all industrial sectors, advanced semiconductor technology is the key enabler for innovations in customer
offerings and internal efficiencies. The increase in the value of data and the related push for AI are examples of forces that
increase the demand for compute power, which translates to more complex and powerful silicon. Moore’s Law, supported by
rapidly evolving semiconductor technology and ever more advanced building practices and assembly technologies, has met the
need for decades.
But what is driving 5G today? If we look at the processing requirements, it is the digital front-end, physical layer processing,
and beam forming. Back in 2010, LTE/4G was a 20MHz carrier with two receive and two transmit branches, and there was a
transmission time interval of one millisecond. Fast forward to where we are today on 5G with massive MIMO, we typically have
100MHz carrier bandwidth. That is a factor of five increase. We have 64 transmitter and 64 receiver radios, which is an increase
by a factor of 32, and the transmission time is down to 0.5 milliseconds. In other words, there is only half the time to do 160
times more processing. This is driving an exponential increase in processing needs across the telecom business today, and will
continue to do so as we race towards 6G. This talk will address whether the semiconductor industry is ready to tackle these
challenges.
11:50 AM
PRESENTATION TO PLENARY SPEAKERS
11:55 AM
CONCLUSION
Server 2P SpecIntRate Over Time GPU Single Precision FLOPs Over Time
100X 100,000 GF
rs
yea
ears 2.2
.4 y ery
ery 2 e ev
ev 10,000 GF anc
nce form
rma Per
Perfo 2X
10X 2X
1,000 GF
1X 100 GF
Mar-09 Apr-10 May-11 Jun-12 Jul-13 Aug-14 Sep-15 Oct-16 Dec-17 Jan-19 Feb-20 Mar-21 Apr-22 2005 2007 2009 2011 2013 2015 2017 2019 2021 2023 2025
Figure 1.1.2: Trend in single precision (FP32) floating point operations for top-end industry
Figure 1.1.1: Trend in mainstream x86 2P server SpecIntRate scores over time [2]. GPUs over time [4][5].
100 GF
10
10 GF
1 GF
1 2005 2007 2009 2011 2013 2015 2017 2019 2021 2023 2025
2008 2010 2012 2014 2016 2018 2020 2022 2024
Figure 1.1.3: The same mainstream 2P x86 server SpecIntRate performance trend as Figure
1.1.1, but divided by TDP [2][3]. Figure 1.1.4: The same GPU FP32 FLOP rate trend as Figure 1.1.2, but divided by TDP [4][5].
Figure 1.1.5: Trend in High Performance Linpack Flops of top performing super-computers on Figure 1.1.6: Trend in power efficiency of top performing super-computers on top500.org list
top500.org list [6]. [6].
1
Technology Trends for Logic
Cost Per Yielded mm2 for a 350mm2, 250mm2, 100mm2 Die
2nm 16
3nm
5/4nm
12
7/10nm
16/14nm 8
22/20nm
32nm
4
45nm
0
28nm 20nm 14/16nm 7nm 5nm 3nm
Figure 1.1.7: Logic density and Iso-frequency energy per operation of mid-Vt device by Figure 1.1.8: Estimated cost (at maturity) for yielded die of 3 different sizes versus process
technology node (AMD internal analysis). node (AMD internal estimates).
DRAM layers
Memory layers
Compute
C mpute
Co
Silicon Interposer
Integration Enable
es Higher Bandwidth at Lower Power
2 5D
2. b ps
D Micro-bum
DIMMS 3D Hybrid Bond
(HBM)
pj/bit ~12 ~3.5 ~0.2
Image source: https:////commons.wikimedia.org/w
/wiki/File:SDRAM-Modul.jpg, Creative Commons 4.0.
Figure 1.1.10: Data movement energy for different integration methods, from traditional on-
Figure 1.1.9: Linear and areal bandwidth density for classes of packaging technology. board DIMMs to High Bandwidth Memory to 3D stacked.
50.0X
Power Efficiency [mW/Gbps]
10.00
FOM [pJ/bit/dB]
100
40.0X
1.00
30.0X
10
0.10 20.0X
10.0X
1 0.01
2005 2008 2011 2014 2017 2020 2023 0 10 20 30 40 50 60
Year Channel Loss at Nyquist [dB] 0.0X
off-package copper off-package optical on-package advanced packaging 3D stacked
Figure 1.1.11: Interconnect power efficiency has improved consistently, but much of that is Figure 1.1.12: Reduced energy per bit from advanced packaging and integration will be critical
from reduced channel loss which is reaching the limits of improvement [19]. for efficiency gains.
Joule
FLOPs//J
Exponent Type
Ty Mantissa
35.0x
FP64 52
30.0x
Co-packaged
Optics FP32 23
High-speed Standardized 25.0x
Chip-to-Chip Interface (UCle)
TF32 10 20.0x
Memory
Heterogeneous FP16 10 15.0x
Compute cores
BF 16 7 10.0x
Domain Specific Advanced
Accelerators 2D/2.5D
/2 5D/3D BF 8 2 5.0x
Packaging
FP8 3 0.0x
FP64 FP32 BF16 FP8
Figure 1.1.13: Highly efficient, tightly coupled system-in-package architecture concept built Figure 1.1.14: Bits allocated to exponent and mantissa for various floating point formats and
around UCIe, domain-specific accelerators and memory integration with optical IO. the associated energy efficiency of computing a result.
San Francisco
3.5 hours
26 hours
Austin
Airport
AMD Office
Figure 1.1.15: Similar to driving (legacy physics calculations) vs. flying (AI accelerated
physics) to the same destination, AI surrogate models leverage physics-trained models for Figure 1.1.16: Gains in lower precision AI-specific floating point operations vs. gains in
fast approximation. traditional, general purpose computation over time [2][3][4][5].
Potential Performance Gains from AI Accelerated HPC The Pathway to Zettascale Efficiency
Wall Time
Baseline Workflow
Time Gain
Hybrid
Workflow "
Physics-based simulation
!
AI-based surrogate training
AI-based surrogate inference
!
Figure 1.1.17: Potential improvements in run time for physics calculations by leveraging AI Figure 1.1.18: Efficiency gains must outstrip those of the last decade in order for a zettascale-
surrogate approaches. capable system to fit within a reasonable power level.
Also, low power consumption is extremely important for portable devices, and it is An important technique in this aspect is the development of gain boosting, see Figure
necessary to extend battery life and reduce temperature rise when downsizing. A 10-bit 1.2.13, to realize high op amp gains. The required gain G of the OP amplifier is related
video-rate ADC was required for the imaging signal, but at that time the power to the resolution N as
consumption was larger than 200mW. In addition, CMOS is necessary for cost reduction
and downsizing, but CMOS ADCs were not necessarily low-power at that time. In 1993, G (dB ) > 6N +10 (3)
we developed a 10-bit 20MS/s 30mW CMOS ADC with a power dissipation that was 8
times smaller compared to the state of the art of that moment [16].
Assuming a resolution of 14 bits, a gain of 94 dB or more is required. A high-speed
single-stage configuration is desirable, but its gain is about 50 dB and does not reach
The comparator used in this design is the chopper-inverter proposed by Dingwall [17]
the target. A gain of 100 dB or more can be achieved in a single stage by boosting the
shown in Figure 1.2.9, composed of CMOS-friendly elements and circuits such as an
gain with an amplifier in the cascode circuit [22]. This technique has improved the
inverter, a switch, and a capacitance. The capacitors absorb and cancel the device
performance of pipelined ADCs and became widely used at conversion frequencies of
mismatches, and also realize a S/H function. With these elements a two-step parallel
several hundred MHz with a resolution of about 12 to 16 bits. Development was also
type ADC can be configured. In addition to suppressing the deterioration of accuracy
boosted by CMOS scaling during the period of widespread use, by the speed
due to the miniaturization of the elements, the area of the circuit itself can be reduced
improvement of op amps, and by the availability of high-precision MIM capacitors.
because the circuit is simple. As a result, the parasitic capacitance and power
consumption can be reduced. However, the accuracy was about 8 bit and further
It is first important to note that a CMOS image sensor does not perform a direct reading
ingenuity was required. Figure 1.2.9 shows the upper and lower comparator circuits. In
of the photocurrent of the photodiode, but it reads a voltage at the output of source
the lower comparator, the output of the first stage inverter was first combined in 4 units
follower [23]. Indeed, the charge generated by the photodiode is accumulated in the
by capacitances, and the output of the inverter of the next stage was combined by
depletion layer and transferred to the floating capacitance by the “charge transfer gate”
capacitances in 2 units. By performing interpolation using capacitances in this way, the
TX, as shown in Figure 1.2.14 [24]. The charge is converted to a voltage and read out as
mismatch can be effectively suppressed to 1/4, so that 10-bit resolution can be achieved.
a voltage via the source follower SF. At this time, the reset level is read first, the signal
Furthermore, since the gray inverter can be removed, it was possible to further reduce
the power consumption. Figure 1.2.9 shows the state of the art in energy consumption level is read next, and the CDS operation is performed so that kT/C noise can be canceled.
of 10-bit video-rate ADCs around the nineties. Although at that time a CMOS ADC was Therefore, a small capacitance can be used for high sensitivity. Initially, the pixel signal
not necessarily lower power compared to a bipolar ADC, the energy consumption of this was read out in the analog domain, but as with the CCD, high-speed reading was required
CMOS ADC is 1/8 compared to the other ADCs. A distinguished low energy consumption as the number of pixels increased. This problem was solved by incorporating a parallel
has been attained. read ADC in each column of the pixel matrix as shown in Figure 1.2.14. The initial
proposal used a SAR ADC [25], but nowadays integral ADCs using a TDC (Time to Digital
Later, almost all ADCs used CMOS technology, which benefited from technology scaling, Converter) and counters as shown in Figure 1.2.15 [26] are widely used. An analog-
high integration, high speed, and low power. based CDS is applied using a comparator, and then a digital-based CDS is applied to
perform the signal conversion using a ramp wave. The main reasons for its widespread
Low-power technologies for integrated circuits required for portable devices have use are that the small size of the ADC requires a simple configuration according to the
attracted attention [15] and low-power technologies have become a major technological reducing pixel pitch and monotonicity is guaranteed. Also, a resolution up to about 12
stream that has led to the realization of smartphones. bits is possible, together with a low power consumption and conversion speed according
to the specifications of the image sensor. Since column ADCs operate in parallel, they
5. Imaging do not require high-speed operation like the ADC used in conventional CCD image
In the development of solid-state image sensors, various technologies have been sensors, and they have become rapidly popular thanks to their low area, low power and
developed and image quality has been improved so much, but it seems that the stage of high speed capabilities.
practical use was reached after the development of image sensors that transfer the
accumulated charges generated in a photodiode by a CCD (Charge-Coupled Device). Additionally, recent image sensors have progressed to three-dimensional sensors by
measuring not only two-dimensional images but also distances. As shown in Figure
CCD is an amazing technology [18] that enables noiseless charge transfer by controlling 1.2.16, the measurement accuracy can be further improved by irradiating infrared light
the potential in the bulk, as shown in Figure 1.2.10. The voltage at the gate on the surface and counting the number of clocks by a TDC and using voltage information by ADC.
of the semiconductor is controlled and the charge is transferred along the potential by Figure 1.2.16 shows an ADC and TDC circuit [27] that can measure voltage information
sequentially changing the potential in the bulk. Figure 1.2.10 shows the basic and time information simultaneously using a VCO.
configuration of an image sensor using a CCD [19]. The accumulated charge in the
photodiode that is optically excited, is transferred by the vertical transfer CCD, then With the spread of smartphones and image sensors, it has become possible for
transferred to the horizontal transfer CCD, converted from charge to voltage by an individuals to communicate on a global level using both still and moving images on social
amplifier, and then converted to a digital value by an ADC outside the image sensor. The networks. By photographing 2D barcodes, it has become possible to authenticate various
correlated noise is suppressed by using CDS (Correlated Double Sampling) technology things and easily access the Internet.
[20] that sequentially transfers the reference signal and the pixel signal and takes the
difference. The signal from the CCD image sensor is an analog serial signal, and a high- 6. Recording
speed ADC with a resolution of about 12 to 14 bits is required to cope with the high ADCs have also been used in HDDs, DVDs, and other recording devices and digital
dynamic range of the image signal. CCD image sensors mainly used pipelined ADCs. transmissions. These signals are not natural signals introduced so far, but “1”, “0” digital
Pipelined ADCs are used for wireless communications, especially base stations, but the signals, and can usually be reproduced by threshold processing. However, due to the
first mass-produced application was for CCD image sensors. high density of the recording device and the high speed of the transmission device, it is
no longer possible to expect an improvement in the recording density and transmission
In a pipelined ADC, whose principle is shown in Figure 1.2.11, the input comparator speed by threshold processing.
controls a DAC according to the level of the input signal, the CMOS operational amplifier
(“OP” in Figure 1.2.11) amplifies the subtracted signal. The next stage samples the output Figure 1.2.17 shows the so-called plus signal, the succeeding minus signal, and the
signal of the previous amplifier and on the resulting signal it performs the same operation synthesized signal that is to be detected by the magnetic head. With increasing recording
as the first stage, in a pipelined manner. In this process, the input / output characteristics density it is subject to ISI (Inter Symbol Interference) which attenuates the signal strength
become folding characteristics of the output of each conversion stage, and as the and shifts the timing of the peak. To deal with this challenge, the PRML (Partial-Response
= +( )
+ +
+
+ +
+
+
D
Figure 1.2.1: 16b Monolithic DAC using resistor trimming [1]. Figure 1.2.2: ΔΣ modulator and multistage noise shaping.
! " #
#
$
% &' ( ) * $
+, -
% &' ( ) +, -
Figure 1.2.3: Pulse-width modulation and voltage modulation. Figure 1.2.4: Data-Weighted Averaging and ΔΣ DAC with SC LPF.
6( 1
0 7!
1 (
&
Figure 1.2.7:Two-step flash ADC with interpolation [13]. Figure 1.2.8: Handy camera recorder using digital signal processing [15].
;
&
7$
1 7-
# /:;
#
#
#
#
;
&
7.
7? 98
4
0
>
7:-
Figure 1.2.9: 10b 20 MS/s CMOS ADC with capacitive interpolation. Figure 1.2.10: Principle of CCD and CCD image sensor.
/ 1 7"
#
$%#
, $*
! ! $&
((
" " ' %
'()$*% + '$*%
Figure 1.2.11: Basic 1b pipelined ADC. Figure 1.2.12: Issue of 1b pipelined ADC and solution with 1.5b architecture.
@
"!
!
'
$
#
#$%
Figure 1.2.13: Gain boost amplifier for high gain OP-amp.[22]. Figure 1.2.14: CMOS image sensor with column ADCs.
Figure 1.2.15: Counter-base column ADC [26]. Figure 1.2.16: LiDAR system, and VCO-based ADC and TDC [27].
// /
0" /
1
#
' *
-(
!
.
.' $ '* * ' 1
/ $
' $,
0
'
$
-( #%
&
+*$ '%#
$
' * &
&
'
-
$ ''%
,)
&
&
''
@+(
#" '
(
%
= ( ) + =
+ =
= =
+ = + =
+ = + =
+ =
+ =
Figure 1.2.17: ISI and ML method in magnetic recording. Figure 1.2.18: Digital read-channel in DVD recorder.
μ
(
) + (
)
=
%
&% '
!"
23-0((!'
#
,
= <
$
%
Figure 1.2.19: One-chip mixed signal SoC for DVD systems and it’s impact. Figure 1.2.20: 7b 400 MS/s CMOS flash ADC with gate-weighted interpolation [32].
%
"' $(
)
*
+,-. +
0
11
)
$
SAR
SAR
SAR
SAR
SAR
$
SAR
$% ' $%$
SAR
Figure 1.2.21: Wireline connection with high-speed ADC. Figure 1.2.22: SAR ADC, dynamic comparator, and ultra-high-speed interleaved ADC.
"
0 1 "
Flash
ADC
"
DAC
DAC
"
0
!"
"
6"
0 1
" 0
" /,
%,
Figure 1.2.23: Input signals to ADCs in wireless communication. Figure 1.2.24 : Continuous-Time ΔΣ ADC.
2 3,2
# $
Figure 1.2.25: Advanced ADCs for wireless communication. Figure 1.2.26: Beamforming with digital controlled phase shifter.
/ $
Figure 1.2.27: Mixed signal neuron circuits. Figure 1.2.28: Cryo-CMOS circuits for quantum computing.
% !7!
&
0 &' %'
% ! !!
%
1 ! !. # ,
&
1 !$
, # &' 02 12
0'
#
# #
/
# #
Figure 1.2.29: Direction of ADC developments. Figure 1.2.30: Performance and progress in ADCs [60].
2022 Lewis Winner Award for Outstanding Paper 2022 ISSCC Award for Outstanding Forum Presenter
“Ponte Vecchio: A Multi-Tile 3D Stacked Processor for Exascale Computing” “6G Communications: Vision and Challenges”
Wilfred Gomes1, Altug Koker2, Pat Stover3, Doug Ingerly1, Scott Siers2, Gary Xu
Srikrishnan Venkataraman4, Chris Pelto1, Tejas Shah5, Amreesh Rao2,
Samsung, Plano, TX
Frank O’Mahony1, Eric Karl1, Lance Cheney2, Iqbal Rajwani2, Hemant Jain4,
Ryan Cortez2, Arun Chandrasekhar4, Basavaraj Kanthi4, Raja Koduri6
1
Intel, Portland, OR, 2Intel, Folsom, CA, 3Intel, Chandler, AZ 2022 Demonstration Session Certificate of Recognition
4
Intel, Bengaluru, India 5Intel, Austin, TX, 6Intel, Santa Clara, CA
“BatDrone: A 9.83M-focal-points/s 7.76μs-Latency Ultrasound Imaging System
with On-Chip Per-Voxel RX Beamfocusing for 7m-Range Drone Applications”
2022 Anantha P. Chandrakasan
Distinguished-Technical-Paper Award Liuhao Wu1, Jiaqi Guo1, Rucheng Jiang1, Yande Peng2, Han Wu1, Jiamin Li1,
Yilong Dong1, Miaolin Zhang1, Zhuoyue Li1, Kian Ann Ng3, Chne-Wuen Tsai1,
“A Single-Crystal-Oscillator-Based Clock-Management IC with 18× Start-Up Time Lian Zhang1, Longyang Lin4, Liwei Lin2, Jerald Yoo1,5
Reduction and 0.68ppm/°C Duty-Cycled Machine-Learning-Based RCO 1
National University of Singapore, Singapore, Singapore
Calibration” 2
University of California, Berkeley, CA
3
Digipen Institute of Technology, Singapore, Singapore
Jaehong Jung, Seunghyun Oh, Joomyoung Kim, Gihyeok Ha, Jinhyeon Lee, 4
Southern University of Science and Technology, Shenzhen, China
Seungjin Kim, Euiyoung Park, Jaehoon Lee, Yelim Yoon, Seungyong Bae, 5
The N.1 Institute for Health, Singapore, Singapore
Wonkang Kim, Yong Lim, Kyungsoo Lee, Junho Huh, Jongwoo Lee,
Thomas Byunghak Cho
Samsung Electronics, Hwaseong, Korea 2022 Demonstration Session Certificate of Recognition
2
Pohang University of Science and Technology, Pohang, Korea
3
Sharif University of Technology, Tehran, Iran ISSCC 2023 Silkroad Award
Qiang Fang , Longyang Lin , Yao Zu Wong , Hui Zhang , Massimo Alioto
1 1,2 1 1 1
1
National University of Singapore, Singapore, Singapore ISSCC 2022 Student-Research Preview (SRP) Poster Award
2
Southern University of Science and Technology, Shenzhen, China
“Amber: a 441.2 GOPS/W 16nm Coarse Grained Reconfigurable Array-Based
SoC Accelerator for Image Processing and Computer Vision”
ISSCC 2022 Jack Kilby Award for Outstanding Student Paper
Kathleen Feng
“A 10GS/s 8b 25fJ/c-s 2850μm2 Two-Step Time-Domain ADC Using Delay- Stanford University, Stanford, CA
Tracking Pipelined-SAR TDC with 500fs Time Step in 14nm CMOS Technology”
Taekwang Jang
SSCS Chapter of the Year 2023 IEEE Solid-State Circuits Society Industry Impact Award
IEEE SSCS Switzerland Chapter Samuel Naffziger
Taekwang Jang, Chair
Advanced Micro Devices, Fort Collins, CO
Michel Bron, Vice Chair
Mathieu Coustans, Secretary
Domenico Pepe, Industry Relations Coordinator The IEEE SSCS Industry Impact Award recognizes and honors SSCS members who have had
a seminal impact and made distinctive contributions to the field of solid-state circuits and the
integrated circuits industry. The award recognizes extraordinary accomplishments in
entrepreneurship, leadership and/or technical excellence that led to significant breakthroughs
and paradigm shifts in the IC business, thereby enabling new products and services within the
SSCS Student Branch Chapter of the Year ten year period prior to the application deadline.
IEEE EDS/SSCS University of Niš Student Branch Chapter
IEEE SSCS University of Science and Technology North Carolina Agricultural and Technical State University, Greensboro, NC
of China Student Branch Chapter
The IEEE SSCS New Frontier award recognizes and honors SSCS members in their early career
Muhammad Hunain Memon, Chair who are exploring innovative and visionary technical work within the field of solid-state circuits.
Fujiang Lin, Faculty Advisor The award aims to pioneering developments that are at the frontiers of IC design or possess
an imminent potential to expand the field through new categories of circuit technologies,
system design, and/or emerging applications.
SSCS Chapter with Best Educational Program Award
IEEE SSCS University of Bordeaux Student Branch Chapter
2023 IEEE Brokaw Award for Circuit Elegance
Maxandre Fellmann, Chair
Yann Deval, Faculty Advisor Benjamin Hershberg
Intel, Hillsboro, OR
IEEE SSCS KU Leuven Student Branch Chapter
This Award was created to enhance appreciation and encourage innovation of simple, smart,
Carl D’heer, Chair and elegant circuit design. The award is presented in recognition of a unique, innovative,
Ariane De Vroede, Vice Chair simple, smart, and elegant circuit design, created during the past decade that has demonstrated
Jonas Pelgrims, Treasurer its viability.
Arnaud Van Mieghem, IEEE SB
Bram Veraverbeke, Media & Public Relations
Senne Gielen, Secretary
Berke Güngör, Member
Ingrid Verbauwhede
KU Leuven, Leuven, Belgium
“For pioneering contributions to energy-efficient and high-performance secure integrated circuits and systems.”
The IEEE annually recognizes outstanding contributors worldwide to the art and science of electro- and information technologies
with technical awards in 3 dozen technical specialties. The IEEE Donald O. Pederson Award for Solid-State Circuits for
outstanding contributions in the field of solid-state circuits goes to an individual or team of not more than three. The award
includes a bronze medal, certificate, and cash honorarium which are presented at the ISSCC.
Omer Oralkan
Jason H. Anderson North Carolina State University, Raleigh, NC
University of Toronto
“For contributions to micromachined ultrasonic transducers
“For contributions to high-level synthesis and low-power FPGAs” and integrated microsystems development,
for imaging, therapy, and sensing”
Keith Bowman
Qualcomm, Raleigh, NC Woogeun Rhee
Tsinghua University, Beijing, China
“For contributions to variation-tolerant adaptive processor designs”
“For contributions to phase-locked circuits and systems”
Tetsuo Endoh
Sriram Vangal
Tohoku University, Sendai, Japan
Intel, Portland, OR
“For contributions to nonvolatile memory and spintronic logi”
“For contributions to network-on-chip architectures”
Roberto Gomez-Garcia
Marian Verhelst
University of Alcala Getafe, Madrid Spain
K.U. Leuven, Heverlee, Belgium
“For contributions to planar multi-function microwave filters”
“For contributions to energy-efficient near-sensor processing
and embedded Machine Learning Processors”
Donhee Ham
Harvard University, Cambridge, MA
Hua Wang
“For contributions to semiconductor electronic interfaces Georgia Institute of Technology, Atlanta, GA
with biological systems”
“For contributions to high-efficiency microwave
and millimeter-wave power amplifiers”
Muhammad M. Khellah
Intel, Tigard, OR
Qiangfei Xia
“For contributions to co-optimization of on-die dense memory University of Massachusetts Amherst, Winchester, MA
and fine-grain power-management circuits”
“For contributions to resistive memory arrays
and devices for in-memory computing”
Thomas Mikolajick
NaMLab, Dresden, Germany
“For contributions to nonvolatile memory”
The pilot-line technologies will indeed have to invest in design enablement to allow early The interplay of deep technology insight, sustainability goals and innovation at the
exploration of the potential impact of new features in advanced chip and system application and systems level is at the core of the EU Chips Act ambition. The sections
architectures (see Figure 1.3.4). Virtual prototyping will exploit experimental PDKs for below illustrate this from different application perspectives and are only indicative of the
logic, memory, and other features. Feedback from these explorations will instigate vast number of other opportunities in these and many other sectors.
Figure 1.3.1: Europe can build on existing strengths like the unique position of certain
materials and equipment companies (e.g., ASML, SOITEC, ASM-I, ...), leadership Figure 1.3.2: Pillar one of the EU Chips Act combines design and system integration
positions in R&D on the industrial 300mm wafer standard (imec, CEA/Leti, infrastructure and an upscaling of advanced and leading-edge semiconductor
Fraunhofer), and existing ecosystems for chip manufacturing. technology pilot lines, allowing to demonstrate new system concepts.
1
10 nm
23 nm
Si Si
WSi=23nm
W Si Si TSi=5nm
SiN TiAl
SiO2 HfO2
liner
(a)
(b)
Figure 1.3.8: TEM cross-section image of an NMOS forksheet device [4] (a) with
Figure 1.3.7: The semiconductor roadmap extends along several concurrent tracks bottom dielectric isolation, after replacement metal gate processing; and (b) TEM
that allow to co-optimize the circuit and system performance, power, area, and cost cross-section images of a CFET [24]: (left) bottom PMOS FinFET and (right) top NMOS
through the technology options delivered by dimensional scaling, device and material nanosheet FET in a CFET device. PMOS and NMOS have LG = 20nm, fin pitch = 45nm
innovations. and gate pitch = 90nm.
Figure 1.3.9: Heterogeneous integration, 3-D stacking and a rapidly evolving chiplet Figure 1.3.10: A comparison of the frequency performance of the major technologies
approach allow to build complex systems-on-chip. for the mm-Wave frequencies beyond 100 GHz, the high data-rate of 6G.
Figure 1.3.11: Analysis of the growth of advanced and mature nodes of semiconductor Figure 1.3.12: Top 10 technical fields in patent applications: number of patent
chips in automotive applications. applications filed with the European Patent Office in 2020.
Figure 1.3.13: Distribution of technology nodes for the circuits fabricated through
EUROPRACTICE by European universities and research centers (2021 data).
1. Introduction Personal mobility is another example of an area going through a massive transition,
The extreme development of semiconductor-based compute technologies has not much fueled by advanced connectivity. Vehicles can for instance collect and process
escaped anyone but the fact that mobile communications has followed a similar video feeds for updating a real-time digital twin of a city for use by other traffic, even
trajectory may be less obvious to most. Over five generations of mobile including bicycles [4]. Real-time maps including traffic flow can be constantly updated
communications, data rates, and capacity has improved with more than six orders-of- for navigation and traffic management. Pedestrian flows can be accurately monitored
magnitude, and this trend shows no signs of weakening. Mobile communications to optimize and manage public transportation, the opportunities are endless, and the
equipment (both infrastructure and devices) is among the most complex and advanced sustainability gains considerable.
electronics produced.
All the examples above, and of course many more, rely on advanced sensors, large
Mobile infrastructure goes far beyond being a means of just communications. It is also amounts of data, and advanced computing, driving the needs for evermore powerful
a platform for continuous innovation. The smartphone revolution would not have and energy efficient semiconductors.
happened without 4G LTE, and the industrial digitalization wave is very much fueled by
5G NR, and with the network compute fabric introduced with 6G, the possibilities are
essentially endless. 3. Optimized Semiconductors for Communication Infrastructure
Modern innovation in industry, services, and consumer products is very much driven 3.1 The Semiconductor Evolution
by connectivity, data, and intelligence. This paper intends to demonstrate how the rapid Across essentially all industrial sectors, advanced semiconductor technology is a key
evolution of mobile infrastructure fuels an as dramatic increase in computational needs enabler for innovations in customer offerings and internal efficiency. The increase in
not only in wireless infrastructure but also in applications. the value of data and the related push for AI are examples of forces that further increase
the demand for compute power and storage, which translates directly into more
1.1 The Mobile Network complex and powerful semiconductors. The exponential evolution, essentially following
A mobile communication network comprises of two main domains: the core network what is known as Moore’s Law [5] has provided large and continuous improvements
(CN) and the radio access network (RAN). The core network manages things like traffic in computational performance at constantly reduced cost. At a component level the
control, billing, and the communications contexts of all connected devices, effectively evolution has however not been linear but includes several stepwise changes.
hiding the mobility to any application. The radio access network is the fabric that
connects all cell towers and manages resources such as spectrum in a highly optimized During the 1980s and 1990s, often referred to as the Megahertz era, processor
manner. The design philosophy of modern networks is to offload as much complexity performance increased mainly through an increase of clock rate. However, the
as possible from the devices, for reasons of economy of scale, and reduced device diminishing returns of this linear development pushed for new approaches and around
costs and energy consumption. the turn of the century the industry entered the Multicore era, adding more and more
processing elements to systems. With this paradigm now facing challenges in scaling
With the ever-growing traffic, cellular networks move to higher frequencies to access and even physical distances on chips, we are now entering the Architecture era with
new frequency bands with more bandwidth. Higher frequency bands have shorter range application-specific computing and algorithmic acceleration as central elements.
which calls for more advanced antenna systems and denser deployments of radio base
stations and towers. This constant increase in management systems complexity and Heterogenous integration is an important tool in the Architecture era. Putting multiple
signal processing capacity drives a drastic increase in computing needs. tiles of silicon, each optimized for purpose and functionality will provide means for
further system scaling. This technology is not new but now there are forces pushing a
2. 5G as a Driver for Compute Needs potential multi-vendor chiplet ecosystem. A promising development is the newly started
Universal Chiplet Interconnect Express [6] . Heterogenous integration opens
2.1 Connectivity and Traffic opportunities to build systems in a package by combining silicon tiles designed in
The main challenge of the mobile communications industry is the exponential growth specialized technologies. An important example is combinations of analog and advanced
in traffic. The Ericsson Mobility Report [1] tracks, correlates, and projects the evolution digital tiles in the same package.
of traffic in mobile networks. Comparing the traffic volume with a normalized compute
index it becomes obvious that the collective compute power of a mobile 3.2 Communication Infrastructure Portfolio
communications network has this same exponential development. To add further to Ericsson’s portfolio [7] is focused on all aspects of mobile communications
the challenge, the wireless industry is committed to achieving this at a fixed energy infrastructure with its radio-near parts being the most computationally intense. To keep
budget. its technology leadership, Ericsson has done highly customized silicon for these
products for decades and radio access network processing is for this reason the focus
5G enables much higher data rates, lower latency, and higher reliability than 4G and its of this paper. Figure 1.4.2, shows an overview of the Ericsson network portfolio.
predecessors, but this is achieved through more advanced concepts. The radio-near
processing needs are essentially proportional to the bandwidth and the number of 3.3 Radio-Near Processing
antennas. For 4G LTE an advanced base station has two antennas and a carrier
To optimize the utilization of the limited resource of radio spectrum, radio access
bandwidth of some 20MHz. For 5G the numbers can be 64 and 100MHz, translating to
networks use massive amounts of signal processing to amplify, filter, shape, code, and
an increase in computational needs of roughly 160 times.
decode radio signals. Figure 1.4.3 shows the typical RAN processing functions with the
Deeper into the network, compute requirements grow more proportionally to transferred connection to the core network to the left and the antenna to the right. The differences
user data. Still the increase is significant as 5G-enabled devices offer up to two orders- between the transmitted and received signals drive differences in algorithms,
of-magnitude higher peak data rates together with much lower latency. An overview of processing, and analog implementation close to the antenna.
the distribution and increase of processing in 5G can be seen in Figure 1.4.1.
3.3.1 Ericsson Silicon
2.2 Innovation and Applications Ericsson has developed highly specialized silicon for many product generations but
5G is a key enabler for taking the next step in mobile communications, for humans as significantly increased investments in recent years. Ericsson Silicon [8] is a tailor-made
well as for machines. Catering for reliable, high-speed, and low-latency access to cloud family of advanced chipsets that integrates a mix of compute and signal processing
or other networked resources forms an extremely strong platform for driving innovation elements, high-speed shared and dedicated memory, digital logic accelerators,
[2]. switching fabrics, and key security technologies for both protocol acceleration and
hardware level intrusion protection. It is a modular architecture that scales across
Cloud robotics [3] is one example of an application that is taking its first steps on the Ericsson’s entire portfolio. Ericsson Silicon enables enhanced network performance,
factory floor but has mass market, consumer driven potential. With a reliable and highly shorter time from invention to customer, and highly optimized energy efficiency.
Requirements on cost, size, and energy consumption exclude the use of off-the-shelf
processor chips which has led Ericsson to develop its own massively parallel DSP
processor architecture known as the Ericsson Many-Core Architecture (EMCA) [9]. The
EMCA architecture enables massive parallel processing of signals through a large pool
of DSP cores sharing a common low latency memory and a set of specific hardware
accelerators.
While EMCA is optimized for radio physical layer procession like FFT, FEC coding, and
beamforming. Higher protocol layers are typically handled using more traditional
processing architectures, albeit still optimized for the purpose. Packet processing in the
user plane can be relatively efficiently implemented on network-accelerated general-
purpose processing (GPP) cores. Due to the possibility of exposure of user data, all
digital parts of Ericsson silicon are protected with advanced and hardware-based
intrusion and compromise protection systems.
Going forward, the increased user data throughput, lower latency, and more complex
control structures is the main compute challenge in this domain. Further integration of
optics may support some performance increase as the interconnect part of the
networking solutions are becoming challenging.
4. Conclusion
In a 5G mobile network, more and wider antenna signal streams combined with more
advanced algorithms have significantly increased the compute demand in the mobile
network infrastructure. Advanced ASIC design with custom processor cores and
accelerators in the latest semiconductor technology combined with advanced packaging
has been Ericsson’s very successful recipe.
With 6G on the horizon, there is no doubt that the requirements on bandwidth, reliability,
security, sustainability, and cost will continue to grow exponentially. However, looking
ahead, keeping up with Moore’s Law just gets tougher and tougher.
In Figure 1.4.4 there is a correlation between the mobile network traffic growth and the
aggregated compute demand in mobile network infrastructure. So far there is good
Conclusion
Figure 1.4.4: Mobile network traffic evolution in Exabytes per month, including
projection to 2027 (line) and total processing delivered to radio access networks [+],
scaled to traffic evolution. As can be seen, the RAN processing needs increase with
Figure 1.4.3: Radio access network processing chain. the mobile network traffic.