Professional Documents
Culture Documents
Sustainable Interactive Wireless Stickers Phd ARORA-DISSERTATION-2023
Sustainable Interactive Wireless Stickers Phd ARORA-DISSERTATION-2023
TO DEVICES TO APPLICATIONS
A Dissertation
Presented to
The Academic Faculty
By
Nivedita Arora
In Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy in the
School of Interactive Computing
College of Computing
May 2023
Thesis committee:
I would first like to say thanks to the cosmos or the higher power, whatever you have to
call it, for giving me the opportunity and strength to work on the COSMOS project vision.
I am incredibly grateful to both my advisors, Gregory Abowd and Thad Starner, for
believing in me and being my rock-solid support structures all throughout these years.
Thank you for being my parents and family away from home. Thank you for teaching me
awesome research values and showing me what great mentorship and leadership look like.
I hope I can pay it forward to my future students. I would also like to thank both Meghan
and Tavenner for opening their homes for me and showering me with so much love. And a
special shout out to Tavenner for reading all my papers and this thesis and patiently giving
comments.
I am very thankful to my thesis committee members – Josiah Hester, Joe Paradiso,
Hyunjoo Oh, and Sauvik Das, for their guidance, encouragement, critical feedback, and
their own research work that has inspired me over the years.
I would like to thank my absolutely incredible academic “kids” without whom I could
never imagine this thesis even existing – Ali Mirzazadeh, Injoo Moon, Zhihan Zhang,
Shirley Xue, Diego Osorio, Peter McAugen, Dhruva Bansal, Jin Yu, Daniela C. Rodriguez,
Harsh Varma, Michelle Ma, and Yunzhi. I am grateful to all my student collaborator/friends
for their help in my research Steven Zhang, Jung Wook Park, Tingyu Cheng, Mohit Gupta,
Yuhui Zhao, Youngwook Do, Dingtian Zhang, and Charles Ramey. All of your contri-
butions and hard work are greatly appreciated and have amplified the work done in this
thesis.
I am blessed to have collaborators at different departments at Georgia Tech and beyond,
who have supported me in the formation of ideas, and research prototypes – Z.L. Wang,
Canek Hernandez, Anna M. Österholm, John Reynolds, Shannon Yee, Michael Filler, Eric
Vogel. I would like to thank Cisco, especially Ramana Kompella, for funding a major part
v
of the research done in this thesis.
I am very grateful to my amazing GT Ubicomp and Abowd family, brothers and sisters,
who have taught me so much and championed me through thick and thin.
Big shout out to Anirudh Sharma, who taught me how to be fearless and collaborate
extensively to tackle interdisciplinary research problems even before I set foot at GT. I
would also like to thank Abhishek Kumar, who helped me discover my first interest in
building things. I would also like to thank some amazing female mentors I was lucky
enough to meet and work with, who have instilled in me the confidence that it’s all possible
– Lama Nachman, Vaishnavi Ranganathan, Asta Roseway, and Bichlein Nguyen.
In no particular order, I would also like to thank my absolutely amazing set of best
friends (basically sisters and brothers!) without whom I could not have thrived on this crazy
roller coaster Ph.D. journey – Bashima Islam, Sai Ganesh Swaminathan, Ceara Byrne, Jin
Yu, Sonal Mahendru, Pragya Agarwal, Harshit Jain, Shanjit Singh Jajmaan, Nidhi Sharma,
Shruti Goutham, Ruth Vinisha, Rushil Khurana, Priyanka Sarawgi, Rohit Jangid, and Chin-
tan Kaur. I am very blessed and grateful to have each one of you in my life. Thank you for
being my safe space. Special shout out to Mohit Virmani for helping me ‘orbit shift’ and
teaching me how to balance ‘inner/outer’ momentum. Also, thanks to Rebecca Thomas
Hitt for her compassionate care :)
Finally but not least, I would like to thank my ‘real’ family, who have been a constant
source of love and care. Thank you for teaching me important life lessons of hard work,
empathy, and never giving up. Special shout out to my nephew Avyan for always bringing
a smile to my face! Even though my grandmothers and grandfathers are no longer in this
world today, their sacrifices and blessings have laid the foundation for the existence of
this thesis, making me the first woman in my entire family tree with a Ph.D. degree! I
am extremely blessed to have you all in my life. Thanks would not even come close to
describing what or how I feel.
vi
TABLE OF CONTENTS
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xxvii
Chapter 1: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
vii
2.4 Enabling Technologies for Sustainable Interactive Wireless Stickers . . . . 21
2.6.1 Touch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.6.2 Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
viii
3.5 SATURN Device Design Optimization . . . . . . . . . . . . . . . . . . . . 43
3.6.3 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.6.5 Flexibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.9 Exploring the Application Space for the Thin Flexible SATURN Microphone 66
3.10.3 Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
ix
3.12 Potential Applications for Swipe and Touch Sensor . . . . . . . . . . . . . 76
3.14 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
x
4.10 Contributions of MARS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.15 Comparative Case Study: Amazon Dash button, RF-band-aid, and MARS . 120
xi
4.16.4 Placement of Tag . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Chapter 5: Feedback: Low Start-up Voltage and Low Power Display . . . . . . 130
xii
5.8.1 Thermoelectric Based Touch Interaction: Clear to Onset Mode . . . 149
6.2 Stepping into the Era of Sustainable Computational Materials and Objects . 161
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
xiii
Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
xiv
LIST OF TABLES
3.1 Relation between acoustic sensitivity and youngś modulus of flexibility for
full back support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2 SNR(dB) at 1kHz, 5kHz and power consumed for different mic . . . . . . . 63
4.1 Power consumption and backscatter signal SNR for different ESCO with
drain output configuration (PD- Photodiode) . . . . . . . . . . . . . . . . . 104
4.2 Power consumption and signal SNR for different MCO gate output config-
uration (∗ connected in series, PD- Photodiode) . . . . . . . . . . . . . . . . 106
xv
LIST OF FIGURES
1.1 Traditional sticky note next to the prototype example of a sustainable inter-
active wireless sticker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.9 VENUS display a) Device design and working b) Low-power VENUS dis-
play can be powered by heat from a human thumb through a thermoelectric
generator or ambient light on a few photodiodes . . . . . . . . . . . . . . . 11
xvi
2.3 Process of building sustainable interactive wireless stickers is an iterative
one, requiring going back and forth between device fabrication, low-power
circuits and system, and design . . . . . . . . . . . . . . . . . . . . . . . . 16
3.6 Factors effecting potential difference generation : σef f ective surface charge
density and def f ective separation distance between the two plates . . . . . . 44
3.7 Modal analysis was done using ANSYS Inc. software: a. 3D Finite El-
ement model of paper with meshed tetrahedral structure with zoomed-in
central fixed support point. b. Modal shapes of paper at different natural
frequencies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
xvii
3.10 Introduction of holes enhances vibration: a. Mini-resonant air cavities
formed between paper and PTFE which result in a reduction of air-dampening
b. Example of the modal analysis of paper with and without holes showing
the introduction of holes increases the dmax and def f . . . . . . . . . . . . 49
3.11 Selection of dimensions for hole on paper: a. 0.4 mm diameter hole had
the best acoustic sensitivity amongst different diameter sizes tested. b. Fre-
quency response for different hole spacing shows that 0.2 mm spacing per-
forms the best with 0.4mm diameter. c. Circular and grid hole pattern
performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.12 Different attachment positions: a. Paper glued to PTFE across all edges b.
Paper glued to PTFE at 9 points . . . . . . . . . . . . . . . . . . . . . . . 52
3.15 Circular shape performs better than the square of the same area: a. Experi-
mental analysis of the effect of geometry on Frequency response b. Modal
analysis for circle and square for mode 3 showing circle has more max.
amplitude of separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.18 Effect of different support structures on the signal quality: a. Frames with
full back support and unsupported with hole b. Frequency response com-
parison between supported and unsupported c. Modal analysis of com-
bined PTFE/Foam structure with full back support at 2887 Hz and 6962
Hz, which explains the sudden rise in sensitivity around 6-8 kHz . . . . . . 57
3.19 Frequency response for different frame materials for full back support . . . 57
xviii
3.20 Effect on orientation on acoustic sensitivity: a. Schematic diagram of the
experimental setup showing the SATURN microphone in vertical and hor-
izontal positions with 94 dB sound incident b. Polar patterns representing
SATURN Microphones’ directionality . . . . . . . . . . . . . . . . . . . . 58
3.27 SATURN as power harvester: Voltage and power generated for different
working frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.30 Comparative spectrogram for ground truth recorded using a iPhone 6 and
SATURN mic simultaneously . . . . . . . . . . . . . . . . . . . . . . . . . 68
xix
3.31 Dropping remotely interrogable stickers to monitor tanks and ordinance in a
war zone: Loud sound acoustic sounds like tanks passing can be potentially
used to flip bits in SRAM bits that consume approximately 10-100 pW of
power which can later be interrogated . . . . . . . . . . . . . . . . . . . . 69
3.32 Recording a loud acoustic event using power generated from a SATURN
microphone. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.34 SATURN embedded in the face mask acts as a human breath-based power
harvester . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.41 Low-power real-time local sensing and recognition system design: schematic
diagram of self-powered analog sensor like SATURN microphone con-
nected to analog recognition SoC like Field Programmable Analog Arrays
[61] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
xx
4.2 Theory of operation: Analog Backscatter Communication a) Backscatter
explained with light rays b) Typical backscatter communication system c)
Equivalent tag circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.7 Minimum current requirement Igs vs Vgs for the transistor to act like a VCR
(blue). Current and voltage produced by SATURN (orange). . . . . . . . . 91
4.13 MARS tags: a. Prototype b. The three major components of the MARS tag
are wireless communication, power harvester, and sensor. . . . . . . . . . . 98
xxi
4.15 Steps of building a Modified Clapp Oscillator (MCO): a. Start with the con-
ventional Colpitts oscillator configuration b. Replace the MOSFET with a
zero-Vth MOSFET and replace the Colpitts current source with a degen-
erative current source (an inductor) to build the Enhanced Swing Colpitts
Oscillator (ESCO) c. Complete ESCO drain-output RF-backscatter circuit . 102
4.16 Steps of building MCO (Continued): a. ESCO with gate output with the
addition of Radjust b. Modified Collpits Oscillator (MCO) with frequency
dependent on L1, C1, C2, C0 c. MCO is a type of Clapp oscillator d. MCO
with all three terminals oscillating. The gate oscillates to create a negative
feedback loop. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.17 Characteristics of the RF-switch JFET: Voltage versus current and resis-
tance graph. The region marked in green is the ideal operation region for
JFET where the power consumption is low and impedance changes are linear.108
4.20 FM transceiver pipeline for audio: The three signal processing steps dis-
played here create the pipeline which is used for: a. Receiving the backscat-
tered signals b. Shifting backscatter signals to 0 Hz where they can then be
c. Demodulated into human perceptible audio data . . . . . . . . . . . . . . 113
4.21 MARS facilitated wireless speech transmission: a. Audio signals from the
microphone are utilized to change the capacitance of a varactor which mod-
ulates the oscillation frequency of the backscattered signal. b. Conference
program insert MARS wireless microphone, demonstration of the MARS
wireless microphone in a conference environment, and close up of the pow-
ered prototype c. MARS wireless microphone attached to a foam pad for
affixing to home appliances. Demonstration of utilizing the MARS wire-
less microphone, powered by a touch of a finger, to extend the range of a
smart home device. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
xxii
4.22 MARS facilitated capacitive dimmer demo: a. MARS swipe direction sen-
sor device prototype. b. The capacitive touchpad acts as a variable capaci-
tor to modify the oscillating frequency of the MARS tag. c. The touchpad
sensor is constructed out of copper tape and an adhesive-backed polyamide
sheet (Kapton). Various-size pads create varying amounts of electrical ca-
pacitance. d. Three peaks from the touchpad are detectable on the GRC-
based transceiver’s spectrogram E. The swipe output of the MARS tag was
interfaced with a Phillips Hue light to act as a dimmer. . . . . . . . . . . . 116
4.23 Capacitive swipe sensor processing: a. The capacitance of each slider pad
varies slightly between participants. b. The signal processing steps for a
left swipe: the spectrogram is generated, the spectrogram is filtered into
a 1D time series, and a linear regression is found between the maximum
and minimum points of the signal. c. The same signal processing pipeline
is demonstrated for a right swipe. Notably, the regression line’s slope is
opposite of the left swipe. . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.24 MARS facilitated capacitive menu demo: a. The menu contains a MARS
tag and a corresponding capacitive ID sensor for each menu item. When the
user swipes an ID sensor on the menu, the signal is transmitted to a remote
application for servers to manage the orders. b. The ID sensors act as
variable capacitors, which modify the oscillating frequency of the MARS
tag when they are touched. c. Pictured is an ID sensor with its capacitive
pad values labeled along with corresponding spectrograms, which illustrate
the frequency changes in the backscatter signal produced by touching the
ID sensor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.25 Pong game controller demo: a. Paper game controller attached to a flexible
MARS tag b. Touch-sensitive buttons switch the MARS oscillator’s fre-
quency by shorting a pair of inductors c. The spectrogram shows the two
controller carrier signals (at 289kHz and 349kHz) and their correspond-
ing button press frequency shifts (303 and 314kHz for controller one and
366 and 379kHz for controller two) d. The backscatter button signals are
processed in GRC and are fed as control inputs to Pong. . . . . . . . . . . . 119
xxiii
5.2 Theory of operation: An example of a conjugate polymer, ECP-Magenta,
changes its color by applying +/- V due to a redox reaction. . . . . . . . . . 133
5.5 Oxidation and reduction states of VENUS display with nano-ITO . . . . . 137
5.7 Device design crosssection with CE, WE, electrolyte, and separator . . . . . 140
5.11 Shift of VENUS display from clear mode to onset mode with application
+/- V square pulse voltage . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.12 Shift of VENUS display from clear mode to off mode with application +/0
Volt square pulse voltage . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
5.14 Different shapes can be etched out in the working electrode . . . . . . . . . 145
5.17 Touching TEG module with hand on side 1 produces negative potential
and on side 2 produces positive potential. It is similar to the square pulse
produced during to testing (Figure 5.11) to control VENUS display to go
from onset mode to clear mode. . . . . . . . . . . . . . . . . . . . . . . . . 148
xxiv
5.19 Box with conductive frame window: When the box is open the VENUS
display is in a clear state with a green indicator. In the closed state, it turns
to the default state due to short-circuiting and discharge of charge by the
window frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
5.20 Hand touch powered interactive cards a. TEG connected to VENUS display
embedded in card b.VENUS display in onset and clear mode c. Interactions
that change the state of VENUS display with respect to card . . . . . . . . 151
6.4 SATURN device design that considers the whole lifecycle including man-
ufacturing and disposal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.6 Playful paper elephant with ears as sensors as a computational object . . . 169
A.2 Gate output AC analysis of Modified Clapp Oscillator (MCO) based backscat-
ter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
xxv
LIST OF ACRONYMS
xxvi
SUMMARY
Today’s Internet of things (IoT) devices are bulky, expensive, require battery main-
tenance, and involve costly installation. In contrast, the interactive stickers introduced
in this dissertation are low maintenance, inexpensive, and easy to deploy. Focusing on
power, form factor, and cost as system design parameters, I create stickers that have simple
circuitry and can sustain themselves while wirelessly communicating and responding to
various human interactions.
This work will introduce four projects. SATURN is a self-powered flexible microphone
and vibration sensor based on a triboelectric generator made from inexpensive everyday
materials like paper and plastic. Next, ZEUSSS stickers extend a single SATURN micro-
phone to have wireless communication capability leveraging extremely simple passive cir-
cuitry. MARS stickers improve ZEUSSS by allowing simultaneous multiple-channel com-
munication capability for speech, swipe, and touch interactions in sub-microwatt power.
Finally, VENUS adds feedback to the stickers in the form of a low-voltage display pow-
ered by the heat of a human finger or ambient room light.
The material device, circuit, and system innovations in this dissertation pave the way
forward for a world where interfaces can be sustainably instrumented onto everyday phys-
ical objects and surfaces.
xxvii
CHAPTER 1
INTRODUCTION
1.1 Motivation
The exponential technological progress from the mainframes to the Internet of Things (IoT)
generation in the past 70 years has witnessed computing devices become faster, smaller,
cheaper, and more proliferated. However, a quick look around any home, office, or indus-
try environment would demonstrate that the current Internet of ‘Things’ are cornered near
a power outlet with wires hanging from them and look and feel significantly different from
real things of daily use. We are yet to achieve everyday objects and surfaces, e.g., walls,
tables, pamphlets, and face masks, with computational functionalities. Further, augmenting
computation onto the existing legacy infrastructure and objects is not possible without mak-
ing them clunky with electronics and requiring constant upkeep of batteries. Alternatively,
imagine a thin material with computational properties (sensing, communication, feedback)
that needs no additional power maintenance once deployed and can be instrumented onto
any surface easily, enabling it to become smart. To explore this idea in this dissertation, I
lead with the example of a thin paper-like sticker.
Figure 1.1: Traditional sticky note next to the prototype example of a sustainable interactive wire-
less sticker
1
Sticky notes have been traditionally designed to be easily attached to everyday objects
and surfaces to augment them with information. However, sticky notes are passive objects
with no sensing, communication, or feedback capability. In this dissertation, I ask, “What if
a sticky note could enhance the surfaces it is attached to and provide additional interactive
functions?” Imagine stickers that can sense human interactions like speech and touch and
provide feedback as they wirelessly communicate with smart assistants, yet still preserve
the maintenance-free, thin, and inexpensive properties of a regular sticky note. Figure 1.1
is an example of an interactive sticky note that can listen to human speech commands and,
while it is actively communicating, turn on its green indicator and switch it off when done.
Such sustainable interactive wireless sticky notes can enable the following applications in
indoor environments:
Figure 1.2: Creation of object-contextualized in-situ tasks a) Add grocery with an interactive sticker
on the refrigerator b) Order food items with the interactive food menu
.
• Creation of in-situ object-contextualized tasks: Sustainable, inexpensive interac-
tive stickers can be used as an extended microphone for a smart home equipped with
a smart voice assistant. Figure 1.2a shows interactive stickers on a refrigerator that
can record groceries to be ordered. Swiping on the interactive sticker attached nearby
a plant can keep tabs on daily home chores like watering plants. Just like post-it®
notes, these interaction stickers can be made into many different shapes and sizes to
attach to traditionally passive objects and give them computational properties. For
example, a rectangular interaction sticker attached under each item on a printed food
2
menu can allow a customer to order the item by simply sliding their finger across it
(Figure 1.2b).
Figure 1.3: Use objects/surfaces as easily accessible smart controllers a) A sticky note in a confer-
ence brochure can act as an extended microphone in an auditorium b) Light control interface that
can be retrofitted onto everyday objects
.
3
in a classroom for interactive polling. These stickers can replace familiar interactive
panels in public restrooms to allow more granular feedback (Figure 1.4b).
Figure 1.4: a) Use of touch button controlled interface as a game controller b) Stickers installed in
public restroom booth to get feedback
4
passive objects, like books and greeting cards, can be augmented to provide inter-
active feedback capability without the cost of maintenance associated with batteries
(Figure 1.5).
Figure 1.6: System design parameters for sustainable interactive wireless stickers
1. Power: Like an everyday sticky note, the interactive stickers should be self-sustainable
in operation and should not require regular power maintenance. One way to do that
is to remove batteries from the system and instead use ambient power for operation.
Since we are targeting interactive stickers for indoor home or office use, the power
harvesting needs should be satisfied in illumination ∼ 500 lux [70], temperatures in
the range of 68-76° F, and humidity controlled to be in the range of 20%-60% as
provided by OSHA guidelines [134].
2. Cost: The stickers should be low-cost (e.g., under $1) such that they can be used
and discarded (even recycled) when not needed. Removing batteries, often the most
expensive component in low-power electronics significantly lowers the component
5
cost and minimizes the harvester size. Reducing circuit part count and complexity
further lowers assembly costs. A careful selection of cheap everyday materials and
simple fabrication methods for functional devices further reduces the cost of stickers.
3. Form Factor: An interactive sticker should be flat and small enough to be easily ap-
plied and unobtrusive while large enough to be seen, understood, and accommodate
sensing interactions. They should be comparable to the 3-inch by 3-inch Post-it®
note. To achieve this, all sub-parts of the tags (sensing, communication, feedback,
and power harvesting) need to conform to the form factor constraints. The required
components for sensing, communication, and feedback should be thin or show the
future possibility of being printable.
This dissertation demonstrates prototypes for wireless interactive stickers that are self-
sustaining, thin, and inexpensive. My thesis statement is as follows –
I break each aspect of this thesis statement into a research question, which will provide
evidence for my claim. I also describe the contributions of each of the research questions
in terms of system parameters and the functionalities they enable within the constraints of
sustainable interactive wireless stickers.
RQ1: Sensing: What kind of interaction sensors are suitable for a thin, self-
sustaining, and inexpensive sensing system?
Figure 1.7 demonstrates different sensor device designs built for touch, direction, identi-
6
Figure 1.7: Different interaction sensors developed in this dissertation
fication, and speech sensing. The form factor of sensors is the property of the materials
used. Everyday materials are leveraged to build sensors – paper, thin plastics (FEP, PTFE,
Kapton), and conductive materials like silver ink and copper foil. All the sensors built are
analog; they show a change in voltage, capacitance, inductance, or resistance with interac-
tion. Further to optimize power, Chapter 3 introduces the concept of self-powered sensing,
where the phenomenon being sensed provides power to do the sensing task as well.
Contributions
7
hundreds of MHz of the frequency range is the most power-consuming functional block.
Since a few milliwatts to hundreds of mW of power is required for radio operation, making
such wireless nodes self-sustainable with respect to power is difficult. Backscatter com-
munication has been demonstrated as an alternative for low-power communication, where
power-intensive tasks are shifted from the tag to infrastructure. This division of labor leaves
the backscatter communication tag only with the responsibility of modulating the incident
high-frequency radio waves and reflecting or back-scattering them back. In addition to the
backscatter communication, I advocate for fully analog tag circuitry rather than digital to
avoid high power-consuming blocks, e.g., analog-to-digital converters and digital micro-
controllers. I use low-voltage threshold transistors to build the interactive sticker circuitry.
Figure 1.8: Two prototypes of wireless analog backscatter communication tags – ZEUSSS [11] and
MARS [13]
8
enables amplitude modulation of the incident high-frequency radio waves corresponding to
the changes in the voltage from a self-powered sensor. Further, MARS stickers introduce an
LC oscillator that leverages a zero-threshold transistor. This oscillator consumes hundreds
of nanowatts to enable frequency shifting and modulation of the incident radio waves based
on the analog sensor interfaced with it (RQ1).
Contributions
ZEUSSS
Both techniques have advantages and disadvantages. While MARS supports the simul-
taneous operation of multiple tags, ZEUSSS is limited to single-tag communication and is
more prone to noise. A ZEUSSS tag does not require any power at the node for operation.
While MARS does require power for operation, it is minimal and can be easily harvested
in the ambient office environment. Broadly, MARS is an improvement over ZEUSSS, as it
enables multiple tags and longer-range communication.
Both ZEUSSS and MARS have simple circuitry with few components. It makes the
cost of both about $1 in bulk. Simple communication circuitry, small power harvesters,
and thin form factor of sensors allow for achieving a sticky note-like form factor. It enables
9
easy mounting on a thin-flexible substrate that can be augmented onto various objects and
surfaces.
Contributions
MARS
RQ3 : Feedback: What type of feedback can be provided to the user in a thin
self-sustaining inexpensive wireless sensing system?
An imperative need from a usability and privacy standpoint in interactive wireless sys-
tems is feedback for the users about when the tag is working and communicating. Most
display technology solutions consume 10's of milli-Watts of power, which is orders of mag-
nitude high for a sustainable operation. In this thesis work, I explore an electrochromic
ink-based display feedback solution. I introduce a novel fabrication technique for VENUS
display that lowers the power to < 10uW for an indicator of 8mm2 -25mm2 with an ac-
tivation voltage of ≈ 350mV -≈ 500mV . This brings display power to a range that is
feasible to be part of a self-sustaining system. Low-power VENUS display also facili-
tates the on/off operation with ambient power harvesters of limited surface area. Further,
I design interactions for controlling different modes of operation of the VENUS display
specific to power harvesters (ambient light and body heat). Finally, I integrate the VENUS
10
Figure 1.9: VENUS display a) Device design and working b) Low-power VENUS display can be
powered by heat from a human thumb through a thermoelectric generator or ambient light on a few
photodiodes
display into my MARS tags as well as demonstrate greeting cards that use the self-powered
feedback mechanism.
Contributions
VENUS
11
1.4 Dissertation Overview
Figure 1.10 shows the dissertation overview. Chapter 2 re-thinks the computing stack
with respect to the design parameters power, form factor, and cost. It also discusses related
work that covers the background of sustainable interactive wireless stickers from the per-
spective of design, system, and material science communities. Chapter 3 introduces device
design, fabrication, optimization, and evaluation of sensors for different types of natu-
ral human interactions like speech, direction, and touch. Chapter 4 covers the low-power
backscatter communication adopted in the thesis and the novel circuits built for its two vari-
ants – amplitude modulation and frequency modulation. In Chapter 5, I discuss the design
choices and fabrication for the VENUS display. I characterize its power, suggest power
harvesters that would work in indoor settings, and design interactions for them. In parallel,
I also focus on applications and specific circuit modifications required to enable them in all
the chapters. Finally, in Chapter 6, I discuss the learnings and observations while building
sustainable wireless interactive stickers and some emerging research themes. I conclude
the dissertation in Chapter 7.
12
1.5 Acknowledgment of Collaborators
This thesis presents multiple projects that I have led during my Ph.D. program. However,
this work would not have been possible without the support of my advisers, mentors, and
numerous internal and external collaborators. I recognized their contributions in the Ac-
knowledgments section of this dissertation. However, I use only the first person singular
for the remainder of this dissertation.
13
CHAPTER 2
RESEARCH APPROACH AND BACKGROUND
In this chapter, I first discuss my research approach toward building sustainable wireless
interactive stickers. Next, I situate the work done in this dissertation with respect to the ex-
isting literature. Further, I provide technological milestones over decades that have pushed
the larger vision of interactive computational materials and objects (section 2.3). I will
also describe a set of enabling technologies leveraged in this dissertation (section 2.4). Fi-
nally, I will elucidate existing developments in object and surface-based interaction sensing
(section 2.5) and methods of accomplishing it in a battery-free way (section 2.6).
Figure 2.1: Comparing the Internet of Things (IoT) devices to sustainable interactive wireless
stickers
There is a fundamental chasm between how the Internet of Things (IoT) devices are
built and what we want to achieve with sustainable interactive wireless stickers. They
diverge on the system design parameters of power, cost, and form factor (Figure 2.1). It is
14
particularly because of the following practices adopted at the device and circuit level while
building a general wireless interactive IoT system (Figure 2.2a):
15
Figure 2.2: a. Building interactive everyday surfaces require re-thinking the computing stack from
the materials level b. Units of an interactive wireless system
maintenance-free.
Figure 2.3: Process of building sustainable interactive wireless stickers is an iterative one, requiring
going back and forth between device fabrication, low-power circuits and system, and design
My research approach iterates between three phases (Figure 2.3): (1) Device fabrica-
tion at the level of the material with specific computational, power, and material properties;
(2) Low-power systems that creatively combine the material devices fabricated into func-
tional circuits that can handle low and sporadic power harvested from the surroundings;
16
and (3) Design of application scenarios that allow for opportunistic ways of power har-
vesting, are meaningful and have societal impact. This research cuts across the boundaries
of different disciplines requiring collaboration with researchers from different fields like
materials science, chemical engineering, mechanical, circuits, communication, signal pro-
cessing, machine learning, HCI, privacy, and industrial design. I exlain this in detail in
subsection 6.2.2.
17
this dissertation (SATURN, ZEUSSS, MARS, and VENUS) with respect to the background
literature in terms of power, form factor, and functionality. The Y-axis is the magnitude of
power, and the X-axis is the level of computational integration with the material. The
target aspirational region area for indoor maintenance-free inexpensive wireless interactive
surfaces is the green region.
There are smart IoT-based solutions or tangible user interfaces that are mostly embed-
ded inside the object’s body and are wired (dark green) [83]. Flexible hybrid electronic
systems focus more on the form factor, where they optimize all system units to be thin, but
they do not necessarily optimize for power [96]. They are mostly dependent on batteries.
Computational RFIDs [156] consume ≈1mW of power; RF powering is only practical in
the near field; otherwise, they need large solar harvesters or batteries. They also contain
hundreds of components, making it hard for them to be printable. There is some work
in the literature [176] that is self-sustaining using amplitude modulation backscatter, but
it is not optimized for the form factor. I build self-powered sensors and wireless sensing
systems (ZEUSSS, MARS) that consider power and form factors together. In addition to
wireless sensing, some works would allow for a programmable display (e.g., a single red
pixel) [68]. Inspired by that, I fabricated VENUS, which allows for wireless sensing and
self-powered display.
The vision of building interactive computational materials is not a new one and has gone
through several iterations in the last several decades (Figure 2.5).
Even before the first transistor was built [21], in the 1940s, the research was focused on
defining the role of interfaces as a way for humans to communicate with computational sys-
tems. General cybernetic inquiries of control and communication laid down the foundation
for the concept of interfaces as modern communicative media between humans and com-
puters [203]. Interfaces were further expounded in 1960 's, by visionaries like Licklider
18
Figure 2.5: History of interactive computational materials and objects
[110], who projected “cooperative interaction or symbioses between man and electronic
computer” and Ivan Sutherland, who envisioned a room where “computers can control the
existence of matter” [175]. After the 90s, there was a gradual push for computational in-
terfaces to be in the everyday physical environment for many versatile applications. Mark
Wieser sparked this revolution with his famous article “Computer of the 21st Century”
where he discussed the applications of computing when ‘woven’ into everyday physical en-
vironments [199]. This vision was soon followed by Ishiis'Tangible Bits, which provided a
physical form for digital information [83]. The vision of blurring the digital-physical divide
was further developed by Roy Want with electronic tags [196]. These works focused the re-
search community on the need for the versatility of form factors for computational systems
and interfaces. Gershenfeld wrote a book, “Things That Think” [62], which sparked the in-
terest of HCI researchers and designers, resulting in the addition of computational devices
in everyday materials like textiles [145]. The burgeoning of such computational objects
has led to design theories related to embodied interaction [40]. Designers have looked at
ways of intermingling the computing matter with physical matter, leading to concepts like
‘Transitive Materials’ [30], ‘Computational Composites’ [182] and ‘Sensate Media’ [138].
With the progress in embedding computation in different form factors, the need to
focus on power has become more prominent. There have been thrusts towards building
extremely low-power systems with improvements in power harvesting and management
19
technologies. In 1945, there were passive communication devices based on carefully de-
signed LC resonant cavities like the ‘Great Seal Bug’ that could communicate audio by
amplitude-modulated backscattered high-frequency radio waves passively, i.e., no need for
power at the remote sensor [22]. Simultaneous device miniaturization owning to Moore's
Law [123, 158] eventually led to the point in the late 90s where the power available for
harvesting started to match the power required for wireless communication, leading to spur
of devices. Researchers at Berkeley envisioned a wireless sensing node of the size of a
small dust particle called ‘smart dust’ in 1997 [91]. Concurrently, a wearable technology
revolution started at MIT, where systems cognizant of both power and form factors were
being built [171]. While these efforts emphasized miniaturization and low power, it is
only now, after ≈ 25 years more of work, that we see reasonably thin, flexible, lightweight
computational materials owing to the rise of flexible hybrid electronics [126].
Such technological improvements in power and form factor have affected human-computer
interaction design theories [49] and eventually resulted in ‘Entanglement HCI’ [56], and
‘Materiality of Interaction’ [202]. These concepts are not just valid for human-computer
interactions but also extend to computational objects and human interactions in the phys-
ical world. The vision of instrumenting objects and surfaces to enhance their computa-
tional affordance has been referred to in different fields by different names over the years:
Smart fabric [146], Material-driven-design (MDD) [93], Computational material [2], Flexi-
ble electronics [129], trans-material [23], Smart skin [79], SPIME [173] and Sensate media
[138]. It has found applications in human health, structural health monitoring, sustainable
sensing, robotics, industrial monitoring, biodiversity, and smart homes.
Much of the advancements in form factor, power, and performance were due to the
materials science community realizing in the 1960's that there was “plenty of room at the
bottom.” Properties of the matter could be changed at the nanoscale to affect properties
at the macro level [51]. These discoveries also eventually led to improvements in power
harvesting technologies like solar cells, where P and N-type semiconductors in a diode
20
were appropriately doped with the help of nanotechnology. By the 90's, Tofoli envisioned
the ability of matter to change its physical properties (shape, density, conductivity, optical
properties, etc.) in a programmable fashion based on user input or autonomous sensing
[180]. The material science and chemical engineering community has also pushed hard to
adopt manufacturing techniques that are scalable and inexpensive [52].
Finally, in the last five years, there has been a slow and gradual shift in the research
community to go beyond just the functionality and performance as system design param-
eters and look at a more holistic life cycle of the computational objects. Gregory Abowd
introduced the vision of the Internet of Materials in 2020, which stressed optimizing power,
form factor, and cost together [2]. Efforts are being made to be cognizant of the environ-
mental effect of computing. Computational objects are built with different types of mate-
rials to create different time scales, so they are easier to decompose or reuse at the end of
their life cycle. Novel methods are being invented for low-cost democratized manufactur-
ing. Inspired by all the technological breakthroughs and visions in this dissertation, I have
pushed the envelope of sustainable computational materials and built sustainable wireless
interaction stickers as its realized example.
I discuss specific enabling technologies at the device and circuits level in the computing
stack for sensing, communication, feedback, and power harvesting that take into consider-
ation one or more system parameters together.
To build low (or no) power and versatile form factor electronics, I start from individual
device components, which are also low power. An example of one such device component
is a self-powered vibration sensor. Energy harvesters can be redesigned to have high sen-
sitivity for the phenomenon they generate energy from to create sensors that do not require
21
power to be operated. A new breed of self-powered sensors based on the triboelectric effect
works exactly on this concept, and one recent example is the Triboelectric Nanogenerator
(TENG) [191, 192, 193]. The design of these sensors is based on the principles of tribo-
electrification (or contact electrification) and electrostatic induction. It converts any kind
of mechanical energy to a highly correlated electrical response. With their lightweight,
low-cost, and high efficiency even at low frequency, TENGs have been shown as passive or
self-powered sensors for detecting mechanical motion such as pressure [112, 211], touch
[39, 74], vibrations [27, 206], linear displacement [114, 216], speed [207], rotation [111],
and acceleration [214]. Beyond saving power, TENG-based sensors are flexible, fulfilling
the requirements for a sensing unit for skin-like devices. In this dissertation, I am inspired
by the promise of TENGs and demonstrate how a simpler manufacturing process that does
not rely on any sophisticated nano-manufacturing can be used to develop a self-powered
sensor for capturing vibrations from the sound.
There have been previous attempts to build triboelectric-based acoustic energy har-
vesters. Yang et al. [210] designed a resonant air cavity using a TENG structure based on
aluminum and polyvinylidene fluoride (PVDF), which converted loud sounds (e.g., a clap)
to an electrical signal. This approach was redesigned to be flexible by Fan et al. [47] using
a PTFE-nanowire and paper-based device structure. As an energy harvester, the experi-
ments were focused on maximizing electrical response at the natural resonant frequency.
Thus, only low-range frequencies for loud sounds at high sound pressure were tested. Re-
cently, other nano-structure designs based on a ferroelectric nanogenerator (FENG) [109]
and a piezoelectric nanogenerator (PENG) [194] have also been explored to make flexible
acoustic sensors, actuators, and energy harvesters.
The triboelectric effect is not new to researchers wanting to leverage it for interactive
purposes. Karagozler et al. used the triboelectric effect to support simple interactions in a
children’s storybook [92]. Paradiso’s ‘Interactive Balloon’ used the piezoelectric proper-
ties of PVDF mounted on a mylar balloon [141]. It demonstrated a common object turned
22
into an actuator and input device (e.g., speaker and microphone) together. These exam-
ples, as well as the insights from the above-mentioned materials science work, inspire my
development of the SATURN microphone and its applications.
Circuit operation at a reduced supply voltage is a common practice adopted to reduce over-
all power consumption. The main constraint for operating circuits at low supply voltage,
like 300mV, is the MOSFET 's noise level and threshold voltage (Vth ) [209]. Reduction
in Vth is dependent on the device 's manufacturing technology. Higher Vth gives better
noise immunity, and the lower Vth reduces the noise margin to result in poor SNR. Sev-
eral advancements in transistor device fabrication techniques and circuit design have been
accomplished to allow lower the threshold voltage Vth without affecting the device perfor-
mance, e.g., Bulk-driven MOSFETs have been designed that operate in the sub-threshold
region because of a self-cascade structure or a floating gate design approach [151, 209].
For MARS stickers, I explore one such low-voltage threshold device, the Zero-Vth MOS-
FET by Advanced Linear Devices [36], for its applicability in building a low-voltage, low-
power oscillator circuit (subsection 4.11.1). Lowering Vth allows lowering of the circuit’s
startup voltage that, in turn, enables impedance matching with common low-voltage DC
harvesters (e.g., a photo-diode harvesting indoor room lighting or a thermal electric gen-
erator harvesting the heat differential when touched by a human finger) without complex
power management. Thus, for this dissertation, low voltage circuits support both our aims
of low power and circuit simplicity.
Since the invention of the “Great Seal Bug” [22], the field of low-power wireless communi-
cation has progressed a great deal. The advent of transistors, along with Moore’s Law, has
pushed for the optimization of functionality (transmit distance, SNR, bit rate) with other
23
system parameters like power, cost, and form factor. In this subsection, I will first look
at wireless communication from a power and cost (area, complexity) perspective, where I
discuss three types of technologies – digital, hybrid (A/D), and fully analog. Next, I will
look at advances in flexible wireless communication technology from a form factor and
functionality perspective.
With the availability of low-power computation modules like the MSP430 [81] and STM
ARM cortex [119] processors and the ability to produce application-specific ICs, small
devices that can sense, compute and transmit data have been developed. Examples of
such devices are the vibration telemetry device [179] and the early work on accelerometers
powered by RF energy harvesting on WISP [156]. These devices have A/D converters that
digitize the sensed signals and a state machine to either perform computation or commu-
nicate the data out as packets. They perform these tasks at an impressive power budget
as low as 1 mW to 2 mW. With the ability to store harvested energy in super-capacitors
and batteries, tasks that require up to several 10 's of mW can be handled by these devices
by duty-cycling based on power availability (about 25% duty-cycling for the WISP [127,
156]). While these devices come with superior capabilities in terms of sensing and commu-
nication, they have more components, consume more power, and cost more than an analog
design. Thus, I will not be using digital wireless solutions in this dissertation.
Talla et al. proposed a hybrid solution that switches between pure analog operation and
digital operation [176]. This work combines the use of a microphone sensor with an RFID
activation sequence on a WISP. Other recent work on hybrid devices has led to the develop-
ment of a battery-free phone, which uses hybrid technology to establish a connection using
RFID-based digital communication. One of the methods to harvest more power suggested
24
here is to harvest from multiple bands of frequencies. However, in practical scenarios,
the available power is an order of magnitude insufficient for the continuous operation of
the existing digital processing, supporting the need to develop analog systems. Thus, the
hybrid/analog approach, though lower power than digital, is hard to adapt for continuous
operation, low power, and low-cost systems.
Fully-Analog
After the initial surge in analog electronics and wireless projects like the Great Seal Bug,
there was a wave of digital wireless communication for a few decades with the advent of
transistors. More recently, researchers have started realizing the value of cleverly designed
fully-analog circuitry as an opportunity for sufficiently sophisticated circuits at low cost
and printable form factors. Chipless antenna technology has been used to detect and mea-
sure different physical parameters such as cracks and mechanical strain or identity [75].
RF BandAid [152] showed physiological data as a wearable sensor. Analog backscatter-
based soil moisture sensing for agricultural monitoring has been exhibited [33]. These
employed commercial low-power oscillators, which still consumed power in the range of
hundreds of microwatts. More recently, Varshney et al. have demonstrated the use of tun-
nel diodes to build custom high-frequency low-power oscillators to create backscatter tags
[184, 185]. Inspired by these projects and low-voltage circuits, I designed circuitry wireless
FM backscatter communication, MARS, in this dissertation.
Passive displays do not emit visible radiation; rather, they control the passage of externally
generated light through the display. This property makes the passive non-emissive dis-
plays with a power budget of < 100mW/cm2 a good candidate for low-power sustainable
systems.
Liquid Crystal (LCD), Electrochromic ECD, and Electrophoretic (EPD) displays are
25
three examples of passive non-emissive display technologies that are able to provide dy-
namic visual content. A liquid-crystal display (LCD) is a flat-panel display that uses the
light-modulating properties of liquid crystals combined with polarizers. Electronic ink (e-
ink) is an electrophoretic (EPD) substance made of microcapsules, within which smaller
particles of different colors and charges are suspended. When an electric field is applied,
the particles within the move to the top or bottom, acting as programmable pigments. Elec-
trochromism (EC) is defined as the reversible color change induced by an electrochemical
redox reaction.
While LCDs continuously consume power during use, EPDs have the advantage of
being fully bistable, consuming power only to switch the display. ECDs fall somewhere
in between, as they are not fully bistable but have optical memories ranging from 30 min
to days (depending on the nature of the chromophore and the colored state it is in) and,
therefore, only require energy during switching and refreshing of the display (short refresh
pulse, < 1 mJ/cm2 ). A commercial ECD display vendor, R-dot [153], compared the
energy requirements for LCD, ECD, and e-ink displays of comparable size for switching
on and off. An LCD continuously consumes about 6µW/cm2 . The energy required for
a full switch for an E-Ink display is about 4mJ/cm2 . The corresponding energy number
for the R-dot ECD display is about 1mJ/cm2 with the addition of 0.25 mJ/cm2 every 15
minutes (Figure 2.6).
Figure 2.6: Comparison of different passive non-emissive display with respect to their power and
switching energy [153]
26
Electrochromic displays specifically have an advantage over e-ink displays in the fol-
lowing ways :
• Simpler fabrication: ECDs are much easier and cheaper to fabricate with techniques
like roll-to-roll printing. Their design is customizable, and they are thin and flexi-
ble in form factor. They have the possibility of being fabricated in a do-it-yourself
manner, and the ability to be transparent provides advantages that are unmatched by
e-ink [88].
• Semi-bi-stable display works well for interaction applications: E-inks are com-
pletely bi-stable, and electrochromic displays are semi-bistable but with a lot lower
activation energy. Since the interactions I am designing are generally short-lived,
producing short bursts of energy, ECD fits my requirements better than an e-ink dis-
play.
Inspired by the properties of ECD, in this dissertation, I will be creating a type of ECD
called VENUS and exploring it as part of a sustainable interaction sensing system.
Over the years, many different methods have been invented for converting different forms
of ambient energy. For example, mechanical (wind, human walking), light, heat, and elec-
tromagnetic energy have been converted into electrical power. In the indoor environment,
27
power density is low, and thus the opportunities for energy harvesting are limited. For
example, in normal room lighting of 300-500 lux, a solar cell harvests 5µW/cm2 , and
ambient RF power harvesting is generally limited to 0.1 µW/cm2 . Human body heat at
room temperate generates 10-100µW/cm2 from the touch of a palm transferring heat to
a surface. Such energy numbers make it hard for a digital microcontroller to cold start
and leave operating continuously. In addition, since the voltages generated by the power
harvesters are low, harvesting power requires employing complex and costly power man-
agement circuitry, thus increasing the overall cost of sustainable devices. Inspired by these
issues, this dissertation looks at creating low-voltage circuits that do not require any power
management circuitry. They have simple part complexity and thus are inexpensive. These
circuits can still provide sophisticated functionality with power harvested from the ambient
environment.
In the last decade, the democratization of fabrication tools and techniques like inkjet and 3D
printing has opened the opportunity for wireless components such as antennas, inductors,
and capacitors to be created in versatile form factor [121, 155]. Several different types
of conductive inks like graphene, silver, graphite, PEDOT:PSS on different substrates of
plastic, wood, walls, and textiles have been demonstrated to create these components. It
has opened avenues for sustainable wireless interaction stickers to be produced in thin,
flexible form factors and even for the possibility of being embedded in different types of
materials in the future.
With the larger vision of building sustainable interactive objects and surfaces, there have
been research thrusts that have taken form factor and functionality as the major design goal.
More recently, there have been works that have also taken into account power and cost.
28
In the subsection below, I segregate these works in terms of the focus on system design
parameters. I also further take into account specific interactions: speech, ID, direction, and
touch.
In Mark Wieser’s vision of ubiquitous computing, computing could be brought into the
physical world through tabs, pads, and boards [200]. As this technology improved, touch
and speech became topics of great interest in the HCI community. Acoustic sensors (mi-
crophone [73, 100], capacitive [132, 157], light [26, 204], piezoelectric [133], and resistive
[76]) have been used to recognize touch as well as fine-grained gestures.
Yet these systems typically depend on complicated circuitry in the form of a micro-
controller, analog to digital converter (ADC), and power management that limit their suit-
ability for scaling to a large number of low-cost interactive objects [84]. In addition, the
batteries typically required to support active sensors can render the form factor unsuitable
for being incorporated into or onto everyday objects. Conductive paint [217, 219] and flex-
ible circuits [189] have been proposed to enable touch interactions at scale. While these
proposed sensing parts are flexible, complete systems often have shortcomings in power,
the complexity of the circuit, and the form factor.
Speech interactions with smart home assistants [34, 208] have been a focus of com-
mercial and academic research. However, such assistants are often limited to microphones
in a single location where the smart hub is located. To enable users to be able to mount
a remote microphone where ever needed, I created a flexible microphone called SATURN
that can be used to sense human speech and environmental sounds [12, 14].
To create a battery-free wireless sensing system, the power is harvested from the environ-
ment, or human interaction needs to match the system power budget. Often satisfying this
requirement results in a trade-off of functionality, power, form factor, and circuit complex-
ity. One of the most power-consuming components of a wireless sensing system is the ac-
29
tive generation of radio waves to communicate the sensed information; often, 10-100 mW
is required depending on the frequency and distance desired. To realize a self-sustainable
battery-free system with active radio, the power source needs to be intense (e.g., constant
motion/vibration of an equipment or human, large temperature difference to generate ther-
mal energy, close proximity to RF source [140]) and often bulky in form factor with com-
plicated power management (e.g., a sheet of solar panel, electromagnetic coils). Sometimes
a balance can be achieved between power and functionality with clever hardware design.
For example, EnOcean’s ECO200 battery-free wireless switches [45] send an ID powered
by the mechanical movement of pushing a button. Sozu [218] can detect presence and
ID by harvesting power in everyday environments. These results are encouraging yet are
still limited for sensing interactions like audio and gestures that are power-intensive due to
requiring a higher sampling rate and bandwidth.
Another common battery-free communication technique is backscatter [205]. In this
scenario, an active transmitter transmits a carrier radio wave in the vicinity of the tag,
while the tag backscatters the carrier wave with modulation (e.g., amplitude, frequency,
and phase) based on the sensed information.
Both analog and digital techniques can be utilized to modulate the incident radio waves
in backscatter communication devices. While digital backscatter communication (e.g.,
RFID [212]) is generally considered more robust, analog communication requires an or-
der of magnitude less power, and simpler circuitry than digital circuits [11, 152, 215].
While amplitude modulation backscatter is noisy and does not support simultaneous op-
eration of tags [22], frequency-shifted/modulated-backscatter communication allows for
wireless communication from different types of analog sensor tags on different channels.
Here, I exploit the interaction of the human body with the tag’s tank circuit to re-tune the
FM resonator for simultaneously sensing and transmitting the interaction, simplifying the
circuit and lowering the required power. Next, I discuss how chipless RFID [148], UHF
RFID [107, 108], and analog backscatter (amplitude [178, 11], and frequency [215, 152]
30
modulation) technologies support the battery-free wireless sensing of different interactions.
2.6.1 Touch
In the early 2000s, Paradiso et al. demonstrated electromagnetic tagging technology, where
an ensemble of passively tagged objects is identified and tracked in real-time when placed
in the vicinity of a reader [139] for applications like the creation of musical notes [53].
With digital chips becoming low in price point, ID sense [107], PaperID [108], and RapID
[170] have demonstrated extending RFID-based single-touch interaction capabilities to ob-
jects by detecting changes in backscattered signals. These methods often require separate
tags for each touchpoint sensor. LiveTag [60] showed multiple touchpoint detection by
chipless RFID antennas but was limited by the number of tags (every two meters) that
must reside within the five-meter operable distance of the transmitter. Tip-Tap is an RFID
tag-based wearable input technique for 2-dimensional discrete touch events but requires
wearing gloves for interaction [94]. Sustainable wireless interactive stickers allow simulta-
neous detection of multiple discrete touch-points on separate channels in < 1µW of power
with a range of 15 m.
2.6.2 Identity
Many of the RFID-based wireless touch sensors mentioned previously can be considered
wireless tags that communicate identity [107, 108]. For example, human-powered RFID
buttons harvest the power from hard push-button interactions to activate their radio to com-
municate a unique ID to a central server [137]. These events are discrete whereas, with the
MARS work in this dissertation, more continuous analog states can be transmitted. For ex-
ample, the dashes and dots of Morse code keying can be conveyed, or the rhythm of a song.
Papergenerators [92] communicates ID through an IR LED powered by the prolonged and
continuous rubbing of a triboelectric generator. With reduced power needs, sustainable
wireless interactive stickers can communicate a touch by the user simply resting their fin-
31
ger on the thermoelectric part of the tag.
Individual RFID tags connected in a circular or linear pattern can detect swipe gestures
[108]. RIO [147] detects gestures by recognizing phase changes caused by finger con-
tact. 3D printed objects which have rotatory parts, e.g., pill bottles, prosthetic hands, and
anemometers can manipulate the impedance of a 3d printed antenna with their movement
for direction and speed sensing [85, 86]. Self-sustainable active communication powered
by high-energy actions (e.g., a door closing) has been proposed for direction sensing, but
such techniques do not support interactions such as speech or simple touches as they are
too low in power [218]. Recently, Ubiquitouch has shown FM backscatter-based touch and
swipe sensing but requires a power budget of about 35 µW and is powered by a solar cell
of 50 cm2 [186]. Such systems cannot be easily converted to a post-it-like form factor or
have the simplicity of MARS as presented in this dissertation (<10 components), whose
simplicity may lead to it being printable (subsection 4.16.2).
Battery-less wireless sensing of sound is specifically tricky since it requires a high sampling
rate and communication bandwidth, both of which come at the price of power and increase
the complexity of the circuit. Often the power is high (mW), such that only batteries can
satisfy the power requirement, thus affecting both form factor and cost. A hybrid analog-
digital modulation scheme has been proposed to sense and transmit acoustic data, which
leverages the Wireless Identification and Sensing Platform (WISP) [176]. Later, the RF
BandAid [152] opened the possibility for more robust purely analog frequency-modulated
backscatter communication of audio range signals for physiological monitoring in 35-150
µW up to 1Mhz. Pre-encoded audio has been backscattered in ≈ 10 µW through FM mod-
ulation leveraging a simulated capacitor and a cross-coupled NMOS relaxation oscillator
32
circuit [187]. Inspired by analog-based FM-backscatter, I built MARS to enable sensing
different types of human interactions in indoor settings, in < 1µW power, < 10 compo-
nents, and a part cost of ≈$1 in orders of 10,000 units (section 4.15).
Given the enabling technologies introduced in this chapter, in the next chapters, I will
build projects that leverage this knowledge and bring closer to reality the vision of sustain-
able interactive stickers.
33
CHAPTER 3
SENSING: THIN, FLEXIBLE, AND SELF-POWERED ACOUSTIC VIBRATION,
SWIPE, AND TOUCH SENSORS
3.1 Introduction
By being thin and flexible, wireless interaction stickers afford instrumentation on everyday
objects and surfaces by simply placing the stickers over the object, enabling the object with
sensing and control capability. This chapter introduces sensors for different types of natural
human interactions like speech, direction, and touch. These sensors are designed to follow
the original systems design parameters of power, cost, and form factor and are described
briefly below:
Figure 3.1: Different types of self-powered and passive interaction sensors explored in this disser-
tation
34
based on both simulation and empirical evaluation.
2. Swipe: I demonstrate the design and fabrication of a thin, flexible capacitative swipe
gesture sensor that can be used for direction and identity sensing. The sensor has in-
formation encoded as a pattern in its material design and leverages that humans have
high capacitance. Thus, swiping a finger across the sensor changes its capacitance in
a unique signature that helps detect the direction of the swipe or the identity of the
object on which the sensor is placed.
3. Touch: Generally, inductive sensors have the property of being more robust to noise
than capacitance. I use traditional off-the-shelf inductors to design and fabricate a
button-push switch mechanism that changes its overall inductance when pressed.
In the further sections, I explore the design, fabrication, evaluation, and potential appli-
cations of each of these sensors in detail.
Sound is a critical source of information for understanding and controlling the environ-
ment. Sound is sensed using a microphone, an acoustic-electric transducer, which outputs
an electrical signal that represents the sound pressure variations that it senses. In addi-
tion to the obvious applications of recording events for playback and automated speech
recognition, the increased availability of low-cost embedded microphones has resulted in
a variety of other interesting applications. These applications include occupancy detection
[71, 98], control [106, 142, 166], human behavior studies [29], structural maintenance [65],
health monitoring[102, 197], hearing aids [31], activity recognition [101, 130, 183, 220]
and sound source localization [144, 181].
As ubiquitous as microphones may seem today, there is still room for progress. A mi-
crophone’s design balances several important features, including recording quality, form
35
factor (e.g., size, weight, flexibility, thickness), and power consumption. Electronic de-
vices most commonly use electret [161], or condenser [143] microphones based on MEMS
technology [55, 198]. CMOS–MEMS acoustic devices allow miniaturization and on-chip
electronics but are active, that is, they require power for operation and sound amplification
[57, 43]. Commercially available passive (or self-powered) microphones do not consume
power but are bulky (e.g., a moving coil dynamic microphone [41]) or use PVDF films,
which either results in a low sensitivity contact microphone [128, 5]. These microphones
can also be complex to manufacture and costly to scale in size [97, 160]. The challenge lies
in designing a microphone that is passive and has sound quality comparable to its active
counterparts (acoustic sensitivity > −25 dBSP L ) while still preserving a lightweight and
versatile form factor [104]. Recent advances in materials science have demonstrated the
possibility of such self-powered, easy-to-manufacture sensors that take advantage of the
triboelectric nanogenerator (TENG) to convert mechanical vibrations into electrical energy
[191, 193, 195]. The TENG relies on the triboelectrification and electrostatic induction
effects, which converts a tiny mechanical triggering into electric signal output without ap-
plying external power.
Figure 3.2: SATURN Microphone in variety of configurations a. Device detail b. Soda bottle with
a flexible microphone to enable interactions c. Sensing table that detects the location of people
around it
I present the design, fabrication, and evaluation of a flexible and self-powered micro-
phone that is made up of a thin and inexpensive PTFE/paper/copper structure. My solution
36
results in a Self-powered Audio Triboelectric Ultra-thin Rollable Nanogenerator (SAT-
URN) microphone. This chapter provides three research contributions for the SATURN
microphone:
• Device fabrication: Inspired by the thin, flexible design of a sound energy harvester
based on the triboelectric nanogenerator (TENG) [47], I have developed a simpler
and cheaper fabrication technique for the SATURN Microphone.
What sets SATURN apart from previous work is a rigorous exploration of the design
37
features that make triboelectric-based acoustic sensing practical for a variety of interesting
application scenarios. These design features include
• Wider acoustic range: Low-frequency ranges (< 1000 Hz) and high sound pressure
levels are sufficient to demonstrate a device as a sound energy harvester, working at
some resonant frequency, but not to demonstrate the device as a sound sensor. For
example, in telephony, the usable frequency band range for voice is approximately
300 Hz to 3400 Hz. I seek a flexible thin sound-sensing material that is sensitive
across this wider band of the human audible range. I also want sensitivity beyond the
voice band, including up 4-6 kHz for sound clarity and definition.
SATURN hopes to bridge the gap between material science and ubiquitous computing
communities for self-powered, flexible acoustic sensors. I explore a simpler fabrication
technique for the SATURN microphone versus the techniques for the original TENG de-
vices. I provide a detailed and reproducible guide to constructing SATURN microphones
that can be followed and improved upon by others. I evaluate the SATURN microphone’s
audio characteristics and compare it with state-of-the-art microphones. In Section 3.9, I
demonstrate two implemented applications of the SATURN microphone to show its use in
practice. Those implemented applications, however, only exhibit the passive microphone
38
capabilities of SATURN and still require traditional powered components to receive and
process the audio signal.
The operation of the SATURN microphone is based on the principle of two coupled phenomena—
electrostatic induction and contact electrification. Electrostatic induction is the generation
of opposite charges on two different materials, while contact electrification, or triboelectri-
fication, is charge transfer between two surfaces in contact. The fundamental theory of tri-
boelectrification lies in Maxwell's displacement current, and change in surface polarization
[190]. By introducing a thin conducting electrification layer, the charge difference between
the two polarized surfaces generated due to triboelectricity can be measured. This com-
bined structure is called the triboelectric nanogenerator (TENG). The polarity and strength
of the charges produced are dependent on many material variables such as surface rough-
ness, temperature, and dielectric constant.
39
the vibration effect. Paper is used because of its flexibility, lightweight structure, low cost,
and ease of cutting holes.
Figure 3.3: Structural design of SATURN microphone consisting of copper coated paper and PTFE
TENGs also belong to the family of electrostatic kinetic energy harvesters (e-KEH).
The core device design of the SATURN is inspired by an electret microphone sans the
transistor that acts as the pre-amplifier and traditional encasing. It allows the SATURN
microphone to explore the advantages of form factor to create an acoustically sensitive and
flexible microphone.
The SATURN microphone works on the principle of vibration-induced contact and charge
generation due to triboelectrification and electrostatic induction. This process is explained
in detail in Figure 3.4. Propagation of the sound through the air causes compression and
rarefaction corresponding to the frequencies present in it.
40
Figure 3.4: Cycle of electricity generation process under external acoustic excitation
becomes negatively charged, whereas the copper layer on the paper becomes positively
charged. When subsequent rarefaction separates the paper and the PTFE (Figure 3.4b),
it induces a potential difference across the two copper electrodes, causing current to flow
from paper towards PTFE if the device is connected to an external load. This flow of current
reverses the polarity (Figure 3.4c) of charges on the two copper electrodes (i.e., now the
copper on PTFE has more positive charge than the copper layer on the paper). The next
compression results in the paper moving towards the PTFE again, resulting in a reversed
direction of current flow (Figure 3.4d), completing the cycle of electricity generation.
Some electret-based energy harvesters use a similar working process [8, 19, 78]. In
this dissertation, I have broadly referred to SATURN as a “triboelectric” and not an electret
microphone because of the charge redistribution between the copper layer on paper and the
PTFE due to contact and separation, which is similar to traditional TENG design [112].
The SATURN microphone consists of two attached layers—paper and PTFE, both with
deposition of copper ( Figure 3.5). The fabrication steps are explained in detail below and
depicted in Figure 3.5.
41
Figure 3.5: Fabrication process : (1) Preparation of micro-hole paper (2) Deposition of copper layer
(3) Attaching copper tape as electrodes (4) Stacking paper and PTFE (5) Gluing paper and PTFE.
All dimensions are in mm.
2. Deposition of copper layer: The paper sheet with micro-holes (step 1) and a PTFE
sheet (6cm x 6cm with 0.05 mm thickness) are coated on one side with a thin layer of
copper that act as electrodes. The copper is applied using a standard sputtering tech-
nique inside a PVD chamber (a Leskar PVD75) with a chamber pressure of 6x10−6
torr. The deposition time is set to 45 minutes, resulting in a 0.15 µm copper layer
thickness on the paper and the PTFE. It allows the copper to deposit on the paper but
not obstruct the micro-holes. The border of the perforated paper is also coated with
42
copper.
4. Stacking paper and PTFE: The paper and PTFE are placed on top of each other
such that the copper layer of paper is on top of the non-coated side of PTFE, which
is non-conducting. To avoid a short circuit, I ensure that the copper tape attached
to the copper-coated side of the paper does not touch the copper-laminated side of
the PTFE and vice versa. Finally, the copper tape from the paper is attached to the
second copper tape on the uncoated side of PTFE.
5. Gluing paper and PTFE: The paper layer (copper side facing PTFE) is glued to the
uncoated PTFE side using glue dots at nine anchor points.
Structural parameter values mentioned above, like hole size, hole spacing, and attach-
ment points for the paper to PTFE, were determined by performing experiments and simu-
lation, as discussed in detail in section 3.5 on device design optimization. The final thick-
ness of the constructed SATURN microphone patch is measured to be 150 µm, which is
comparable to that of standard copier paper. This sensor can be attached to objects using
glue on the edges of the copper side of the PTFE layer.
The main aim of device design optimization for the SATURN microphone is to increase
electrical response across a wider range of frequencies in order to achieve better acoustic
sensitivity. It allows sounds of lower decibel levels to be detected, thus opening doors for
a wide variety of applications.
43
3.5.1 Factors Affecting Device Performance
The SATURN microphone consists of paper and PTFE, which are both flexible and vibrate
to act as two plates of a capacitor to produce an electrical response. In previous descriptions
of the behavior of a Triboelectric Nanogenerator ([190, 191]), it was assumed that the two
layers are rigid. In that case, the open circuit potential difference generated by the device
as a function of time(t) is given by the equation [131]:
σx(t)
Voc = (3.1)
ϵ0
where x(t) is the physical separation distance between the PTFE and paper, σ is the
charge density generated on the surface, and ϵ0 is the relative permittivity of the dielec-
tric. This mathematical model is overly simplified and would not work for the SATURN
microphone because neither the paper nor the PTFE layer is rigid. In practice, the PTFE
layer will be attached to a surface, so I will continue to assume it is rigid. The paper layer,
however, behaves more like a flexible membrane and will vibrate. Keeping the vibration of
paper in mind, I have derived a modified formula below.
Figure 3.6: Factors effecting potential difference generation : σef f ective surface charge density and
def f ective separation distance between the two plates
The separation distance between the paper and PTFE layers changes over time when
I assume that the paper layer is vibrating, as shown in Figure 3.6. When the membrane
is placed in the Y-Z plane of a right-handed reference coordinate frame, an infinitesimal
element with area dy.dz at a location (0, y, z) with respect to the origin has a potential
44
difference given by a slight modification of Equation 3.1:
σef f
Voc = x(y, z, t), (3.2)
ϵ0
where σef f is the effective surface charge density, and the separation distance x is a function
of y, z, and time, which varies along y and z for the flexible membrane. Hence, the open
circuit potential difference for a flexible paper layer can be written as:
ZZ
σef f
Voc = x(y, z, t) dy dz. (3.3)
Aϵ0 A
To increase VOC for the SATURN microphone there are two parts in this equation that
can be optimized:
1. σef f : Effective surface charge density, which is dependent on the roughness of the
PTFE surface; and
RR
2. def f : Effective separation distance during flexural vibrations, which is A
x(y, z, t) dy dz/A
and is mainly dependent on the ability of the paper to vibrate while being anchored
to the PTFE sheet.
To guide our structural device design I use a combination of two evaluation techniques:
45
Structural Modal Analysis
Modal analysis is the method to identify the natural frequencies of vibrations of a material
and the mode shapes of a structure. The deformed shape of the structure at a specific
natural frequency of vibration is termed its mode shape of vibration. A thin membrane-like
structure such as that of the SATURN microphone has infinite modal frequencies and mode
shapes. The material response for a given input load is a linear combination of these mode
shapes. I used a 3-D finite element (FE) model mesh using tetrahedral elements to perform
the modal analysis using ANSYS.1 The glued attachments points of the paper with PTFE
(described in Step 5 of the fabrication process in Figure 3.5), were meshed separately, to
be assigned as fixed support (see Figure 3.7a).
(a) (b)
Figure 3.7: Modal analysis was done using ANSYS Inc. software: a. 3D Finite Element model
of paper with meshed tetrahedral structure with zoomed-in central fixed support point. b. Modal
shapes of paper at different natural frequencies.
I use lower modes of vibrations, i.e., mode shapes with lower natural frequencies, as
they are easy to visualize with a number of peaks and crests as compared to the mode
associated with a higher natural frequency. Figure 3.7b shows an example of different
modes of vibration for paper anchored around the edges.2 The transverse deformation, or
vertical deflection, is plotted with the help of colored contours. For each normal mode
shape, the contours are plotted and arranged from blue to red such that blue represents zero
1
http://www.ansys.com/
2
I used values of 3 GPa Young's modulus of elasticity, 0.04 mm thickness, 47mm x 44mm dimensions,
and 1.2 g/cm3 density for the vibrating paper layer.
46
or negligible vertical deflection, whereas red represents the location of maximum vertical
deflection possible.
The value of the vertical deflection of a point, obtained from the modal analysis, is
referred to as the amplitude of separation of a point, x(y,z,t) in Equation 3.3. The integral
RR
of the amplitude of separation at each of these points is def f or A x(y, z, t) dy dz/A. I will
choose the structural design which has a higher def f during modal analysis, as it will have a
higher electrical response. For structural designs which have the same boundary condition
(points of fixation), I can also use dmax , the maximum amplitude of separation of the same
mode shape to compare. It is done because calculating dmax is simpler than determining
def f , which involves integral.
The analyses for dmax is done under the assumption that the def f is produced with PTFE
and Paper having anchor points such that dmax is only a few micrometers. The paper and
PTFE do not move to a large distance that the signal itself stops getting produced. Thus
effectively, we assume the acoustic sensitivity is proportional also to the effective relative
displacement def f .
(a) (b)
Figure 3.8: Acoustic characterization: a. Chirp sound input: linearly increasing 20Hz-20000 Hz
frequency sweep in 20 seconds b. Experimental setup for recording the electrical response of SAT-
URN microphone.
47
I use a frequency sweep or a chirp as the input sound recording to observe the electrical
response of the SATURN microphone to characterize its quality. A chirp is a sine wave
linearly increasing in frequency (20Hz-20kHz) in a particular time period (Figure 3.8). I
standardize the 1000 Hz frequency to a sound pressure level of 94 dBSP L or 1 Pa pressure.
The power in dB (mV/Pa) at 1000 Hz frequency is defined as the acoustic sensitivity of the
microphone. It is used as a representative of the sound quality of a microphone. A JBL
Flip 2 speaker (100 Hz-20kHz flat frequency response) is used as a sound input device (fre-
quency sweep, tone) for our experiments. The sound loudness was measured using a Sound
Pressure Level (SPL) meter by Extech Instruments. The electrical response generated by
the SATURN microphone is measured as voltage using an Analog Discovery oscilloscope
which has 1 MΩ resistance. The SATURN microphone and the speaker (Figure 3.8) are
placed in the center of a large hall structured room, with walls approximately 5m apart
from the experimental setup.
Figure 3.9 shows an example of the electrical response for the SATURN microphone
for the chirp sound input. I use voltage as a measure of the electrical response, as the
current produced by the SATURN device is extremely small (nano-amps in magnitude),
which makes it susceptible to background noise. The maximum voltage is achieved at the
resonant frequency of approximately 275 Hz. At 1000 Hz, the acoustic sensitivity is -26.63
dB (mV/Pa). I will be using both a frequency response curve and acoustic sensitivity to
select optimal structural design parameters.
The paper layer is a critical component to be designed precisely in the SATURN micro-
phone structure to maximize def f the separation distance between the two layers and hence
the output from the device as explained in subsection 3.5.1. To do so, I introduce various
structural changes to the paper layer, described next.
48
(a) (b) (c)
Figure 3.9: Example of an electrical response from the SATURN microphone given a chirp as
input: a. Voltage time series output b. Spectrogram with linearly increasing frequency c. Acoustic
sensitivity variation across the audio frequency band (20Hz-20kHz).
(a) (b)
Figure 3.10: Introduction of holes enhances vibration: a. Mini-resonant air cavities formed between
paper and PTFE which result in a reduction of air-dampening b. Example of the modal analysis of
paper with and without holes showing the introduction of holes increases the dmax and def f
49
µm which is approximately 40 µm less.
• def f : Paper with no holes has an effective separation distance of 40 µm , while the
one with holes is 45.3 µm.
Both parameters, dmax and def f , suggest that the holes on the paper can help in the
optimization of separation distance. I can explain it with the following reasons:
• Holes allow the air to flow between the two layers, allowing air to pass through the
holes and forcing the two layers to move away from each other.
• Perforated paper has less stiffness. The bending stiffness of a structure refers to the
resistance to transverse deformation. Thus, the perforated paper layer would result
in a larger separation distance as compared to the paper layer without holes.
While introducing holes increases the separation between layers, it also reduces the
contact surface area, and thus effective charge density σ is decreased, which would decrease
the voltage. Thus, there is a trade-off between the contact area and the hole parameters of
size, distribution, and pattern. Modeling vibrations of the system to respect air pressure
changes and weight is beyond the scope of this paper. Instead, I provide empirical results
to determine good hole dimensions.
Figure 3.11: Selection of dimensions for hole on paper: a. 0.4 mm diameter hole had the best
acoustic sensitivity amongst different diameter sizes tested. b. Frequency response for different
hole spacing shows that 0.2 mm spacing performs the best with 0.4mm diameter. c. Circular and
grid hole pattern performance.
50
Keeping the distance between holes fixed at 0.2 mm, I fabricated 4x4 cm2 SATURN
microphone patches with increasing hole diameters (Figure 3.11). 0.4 mm diameter holes
achieved the best acoustic sensitivity of -26 dB among our samples. Next, keeping the
hole diameter at 0.4 mm, I increased the spacing between the holes to be 0.2 mm, 0.3 mm,
and 0.4 mm, corresponding to a hole area percentage of 35%, 26%, and 20%, respectively.
A SATURN microphone with 0.3 mm spacing gives the best acoustic sensitivity, but the
device with 0.2 mm spacing performs slightly better within the first 2 kHz frequency band,
where more than 70% of the sound information is present [46]. Next, I experimented with
two different patterns of holes – concentric and grid – to determine their effect on the
frequency response. The grid pattern of holes performed better than the concentric pattern.
I determined previously in our hole distribution study that 35% hole area performs better
than the 25% hole area. Since the grid has a 35% hole area, it performs better than the
concentric circle, which is 25%. Thus, for all our future experiments and modal analysis, I
will choose to use 0.2 mm hole spacing and 0.4 mm hole diameter in a grid pattern.
By controlling the locations where paper is attached to the PTFE in the SATURN micro-
phone structure, I can generate a motion with a higher amplitude of motion with the same
sound input. It is done by choosing attachment points that are coincident with the nodes of
zero movements for the majority of the mode shapes possible (Figure 3.12). I tested two
cases for the perforated paper model: glued across all the edges; and glued at 9 points like
a grid.
Figure 3.13 shows the modal shapes for mode 3 for paper in the two cases discussed.
Just by visual observation, I can notice that the mode shape for paper pasted at edges (case
1) is out of phase, creating minimum and maximum contact with PTFE at the same time,
whereas for the nine glue points (case 2), they are in phase. In addition, as shown the
legend, dmax for case 2 is 7µm (194µm − 187µm) more than case 1. This trend is also
51
(a) (b)
Figure 3.12: Different attachment positions: a. Paper glued to PTFE across all edges b. Paper glued
to PTFE at 9 points
supported by def f as case 2 (52.6 µm) is 8µm more than case 1 (45.3µm). By simulations,
I can conclude that 9 points glue grid has more separation distance and, thus, consequently,
should generate more electrical signal.
(a) (b)
Figure 3.13: Modal analysis demonstrates that grid-glue pattern attachment performs better than
glue on all edges: a. Mode 3 of paper glued at all edges showing maximum. separation distance to
be 187µm b. Mode 3 of paper glued at 9 points like a grid with a maximum separation distance of
194µm.
52
amplitude of separation than even the central nodes. By choosing appropriate boundary
conditions for the fixed points, the maximum vibration amplitude can be obtained under
certain load conditions to increase the voltage.
(a) (b)
Figure 3.14: Impact of gluing pattern: a. Increase of approximately 20 dB with 9-point grid glue
pattern (red line). b. Modal analysis of perforated square paper with holes pasted at 9 points
demonstrates the involvement of edges which increases the voltage response.
Changing the geometry of the surface on which sound is incident while maintaining the
same area changes the air pressure which is applied at different points. It changes the mode
shapes and affects the way the structure will exchange kinetic energy and strain energy at its
nodes and anti-nodes. Thus, geometry impacts the amplitude of vibrations which directly
impacts the separation distance and the output voltage.
I tested two different geometries with the same 1600 mm2 area: a square (4x4 cm2 ); and
a circle (22.57mm diameter). Experimentally, up to 1000 Hz the circle and square perform
nearly the same. For higher frequencies, the circle performs almost 10 dB better than the
square (Figure 3.15a). Thus, choosing a circular shape for the SATURN microphone would
improve sound quality. A modal analysis confirms our empirical results. Just by visual
inspection of the dmax , the amplitude of maximum separation reached by a circle is almost
16µm more ( Figure 3.15b) than the square. I use Equation 3.3 to calculate def f , effective
separation amplitude, which was found to be 67.2 µm for the circle and 52.6 µm for the
square, which is approximately 15 µm more. After optimizing device structural parameters
53
both theoretically and empirically for the separation distance, I am able to reach the best
acoustic sensitivity of -25.63 dB (mV/Pa) at 1000 Hz with a circular shape of 16cm2 area
with a grid pattern of holes of 0.4mm diameter and 0.2mm spacing glued at 8 equally
distant points around the edges and the center to the PTFE.
(a) (b)
Figure 3.15: Circular shape performs better than the square of the same area: a. Experimental
analysis of the effect of geometry on Frequency response b. Modal analysis for circle and square
for mode 3 showing circle has more max. amplitude of separation
Equation 3.3 shows that the voltage response of the SATURN microphone is dependent
on the surface charge density, which is the measure of electric charge per unit length. Fan
et al. [47] used PTFE polymer nanowires to increase the charge density σ by increasing
the effective surface area of the dielectric surface in contact. Growing nano-wires is a
very expensive process that is currently unsuitable for large-scale, low-cost manufacturing.
To understand the effect of such nano-structures on PTFE for the SATURN microphone’s
performance, I use a simpler, though still relatively expensive, a method called plasma
treatment.
The PTFE was etched by O2 plasma produced by a PE-100 Plasma System (from
Plasma Etch Inc.) O2 was uniformly distributed in the reactor throughout the etching pro-
cess. Figure 3.16a shows a schematic diagram of the O2 etching process where a blast of
high-speed stream of glow discharge is shot at PTFE. The RF power input was 300 W using
a 13.56 MHz RF generator with RF auto-matching network and the plasma treatment time
54
was 20 min. Figure 3.16b shows the scanning electron microscope (SEM) image of PTFE
after plasma treatment. This PTFE was later coated with copper using as general fabrication
process explained in section 3.4. Due to increased roughness on the PTFE surface result-
ing in increased contact electrification the performance with plasma treatment improves by
approximately 10 dB across the entire frequency response as shown in Figure 3.16c.
Even though there is an increase in the signal quality, with acoustic sensitivity of -16.28
dB (mV/Pa), the cost addition does may not justify the performance increase for SATURN
microphone fabrication in many scenarios. Therefore for our further experiments, I have
tried to focus more on regular PTFE rather than one treated with plasma.
Figure 3.16: Effect of plasma treatment on SATURN microphone sensitivity: a. Schematic dia-
gram of O2 plasma etching b. Scanning electron microscope (SEM) image of plasma treated PTFE
used for SATURN mic manufactured at 500nm scale c. Experimental comparison between the per-
formance of SATURN microphone fabricated with PTFE treated with and without plasma
55
Figure 3.17: Intra-device performance of SATURN microphone patch represented as a deviation in
acoustic sensitivity (n=6).
Figure 3.17 shows the standard deviation of acoustic sensitivity (1000 Hz tone at 94
dBSP L ) performance. All devices have > -35 dB acoustic sensitivity, with the best perfor-
mance at -26.44 dB and mean as -29.5 dB. The intra-device reproducibility has a tolerance
of 10 dB for our fabrication process.
Just as a pattern of holes allows for greater vibration of the paper layer, so can the support
structure used to mount the PTFE layer affect its vibration and thus the electrical response
of the SATURN microphone.
To understand the effect of back support, I experimented with two kinds of frames (Fig-
ure 3.18a): a full solid back support; and a hollow frame, in which I cut a hole to allow
more vibration. The material used for the frame was foam board to which a non-plasma
treated 4x4 cm2 SATURN patch with optimal separation distance parameters was attached.
56
(a) (b) (c)
Figure 3.18: Effect of different support structures on the signal quality: a. Frames with full back
support and unsupported with hole b. Frequency response comparison between supported and un-
supported c. Modal analysis of combined PTFE/Foam structure with full back support at 2887 Hz
and 6962 Hz, which explains the sudden rise in sensitivity around 6-8 kHz
the free movement of PTFE (Figure 3.18b). Interestingly, having back support showed an
increase in sensitivity around 6-8kHz, while its framed counterpart just has a decreasing
trend. I studied this further by doing a modal analysis combined system of PTFE (440 MPa
Young's modulus of elasticity, 0.05 mm thickness, 2.2 g/cm3 density) pasted to foam-board
(Young’s modulus 0.1 GPa, 0.5mm thickness, 0.3 g/cm3 density). The def f , the amplitude
of effective separation for natural frequencies of the system at 2887 Hz is 28 mm while
6900 Hz is about 34 mm. This behavior may be attributed to the coherence of the natural
frequency of movement for PTFE and back support. Further consideration of different solid
back supports is warranted, as they might produce sensitivity peaks at different frequency
bands. I will address this next.
Figure 3.19: Frequency response for different frame materials for full back support
57
Table 3.1: Relation between acoustic sensitivity and youngś modulus of flexibility for full back
support
Even though the framed, unsupported structure is better, there might be situations when
a SATURN microphone needs to be embedded directly into an object. I embedded the
device on different material surfaces (Figure 3.19) to determine their frequency response.
Table 3.1 shows the acoustic sensitivity I recorded from the experiment versus Youngś
modulus of the material. There is a correlation between the flexibility of the material and
the sensitivity of a SATURN microphone mounted on that material. For example, metal,
with the highest stiffness amongst all materials tested, reduces the vibration more than
plastic or wood. Thus the back support material properties are a factor in determining the
signal quality.
3.6.3 Orientation
(a) (b)
Figure 3.20: Effect on orientation on acoustic sensitivity: a. Schematic diagram of the experimental
setup showing the SATURN microphone in vertical and horizontal positions with 94 dB sound
incident b. Polar patterns representing SATURN Microphones’ directionality
Changing the orientation of the SATURN microphone changes the sound field incident
58
on it and, consequently, electrical response produced due to the vibrations. For example, a
SATURN microphone placed on a table would receive different amplitudes of voltage for
its horizontal and vertical orientations.
To understand this better, I performed an experiment with a SATURN microphone
(4x4cm2 optimal structural parameters, non-plasma PTFE, attached to the foam-board
frame with unsupported back) placed horizontally and vertically as shown in Figure 3.20.
A sound source of 1000 Hz frequency tone at 1 Pa was rotated from 0 to 360 degrees
around the microphone to plot the directivity pattern. The SATURN microphone is om-
nidirectional in both horizontal and vertical orientations, which is useful for applications
such as gathering context in the environment. However, when SATURN is embedded in
objects with full back support, I would obtain unidirectional directivity of the microphone,
i.e., a semi-circle instead of a circle.
In addition, when tested for different orientations, there is a drop in acoustic sensitivity
by 10 dB when SATURN is placed horizontally in front of the sound source as opposed
to vertically. Even though the vertical orientation is preferred, there may be applications
where horizontal placement is required, as such the experimental comparison is important.
For a traditional microphone, the size of the diaphragm affects the microphone's, sound
pressure level handling, sensitivity, dynamic range, and internal noise level. The SATURN
microphone is similar to the diaphragm of the traditional microphone, so it is reasonable to
consider how its size impacts performance.
I did a preliminary test of 3 different-sized SATURN patches (8x8 cm2 ,4x4 cm2 , 2x2
cm2 ), each using the structural parameters that produce the best results for the 4x4 cm2
patch described earlier (section 3.5) and placed on a vertical frame with sound chirp input.
The acoustic sensitivity is -18 dB, -25 dB, and -40dB, respectively, suggesting improved
performance for a larger SATURN microphone. These preliminary results show a favorable
59
trend, but a more in-depth analysis is needed, and each patch size should be separately
optimized, in the same way, I optimized for the 4x4 cm2 size.
(a) (b)
Figure 3.21: Effect of SATURN microphone size on the signal quality: a. Different patch sizes
used for the experiment b. Acoustic sensitivity comparison plot for different sizes
3.6.5 Flexibility
The SATURN microphone has a thin structure which gives it the affordance to be bent. It
is important to understand how the bending impacts the signal quality. The bending of any
object adds strain to the object, which leads to increased potential energy, and loss of ability
to vibrate or gain kinetic energy. I study the effect of bending using a 4x4 cm2 SATURN
microphone patch with non-plasma PTFE with optimal hole size and fixing point. The
patch is bent to 7 different radii of curvature corresponding to the central angle theta (5◦ ,
15◦ , 30◦ , 45◦ , 60◦ , 75◦ , 100◦ ) as shown in Figure 3.22a. I performed modal analysis for
the curved 3D models of paper with holes for 4 different angles: 5◦ , 15◦ , 30◦ and 45◦ .
Figure 3.22b shows the effective amplitude of separation, def f of the 4 models for mode
nearest to 1000 Hz, the frequency where acoustic sensitivity is defined. Figure 3.22c is
the snapshot of the corresponding to def f . I can see that with an increase in bending def f
reduces almost linearly.
Next, I performed the experiment to determine the change in acoustic sensitivity with
increased bending for a fabricated 4x4 cm2 patch. The SATURN patch was embedded in
cardboard as shown in Figure 3.23a and bent at 7 different radii of curvature successively
60
making sure that the cardboard follows the lines of curvature. Figure 3.23b shows a succes-
sive drop in acoustic sensitivity with smaller radii of curvature, thus following the expected
trend from modal analyses. Having a SATURN microphone structure and bending results
in increased bending stiffness, which affects the ability of the SATURN microphone to vi-
brate and reduces the voltage it can achieve. A flat SATURN mic patch used can achieve a
sensitivity of -27.3 dB which reduces by 8 dB for 45 degrees bent. For the next 45 degrees
of bending the drop is much more enhanced as the stiffness increases.
Figure 3.22: Effect of flexibility on effective separation: a. 7 different radii of curvature were
drawn on paper corresponding to central angle theta which 4x4 cm2 patch would make when bent
b. Plot demonstrating a decrease in effective separation distance with increased bending c. Modal
analysis of a 3D meshed model of paper at when subjected to 1 Pa pressure at 1000 Hz
As a result of the design and the fabrication process, the SATURN patch reliability
in flexible scenarios may be strongly affected by material stress on the glue points, and
tension in the material when transitioning from flat to bent layouts. Some scenarios were
tested during the flexibility analysis, however, a more extensive analysis of the fabrication
process for flexible scenarios is required in order to find a solution that preserves the sound
sensitivity and adaptability to different shapes.
There are factors in microphone device design—signal quality, power, form factor (size and
flexibility), and cost/ease of manufacture—which dictate whether a given acoustic sensor
61
(a) (b)
Figure 3.23: Experimental results for the effect of flexibility: a. SATURN microphone was bent
accurately to follow the curved curvature line e.g. 30◦ b. Change in acoustic sensitivity (1000Hz @
94 dBSP L obtained for different radii of curvature
meets the requirements of any given acoustic application. Device designers often face a
trade-off between the sensor signal quality and the other design parameters. For example,
condenser microphones are used in the recording studio have very high sound quality but
consume power in tens or hundreds of mW and are relatively expensive, all of which are
appropriate for that niche application. Electret microphones, commonly found in consumer
electronic devices, consume 200-500 µW to achieve sensitivity greater than -25 dB. They
mostly use power as biasing voltage for the MOSFET or for amplification.
62
Figure 3.24: Plotting different types of microphone with respect to sensitivity, power, and form
factor
Table 3.2: SNR(dB) at 1kHz, 5kHz and power consumed for different mic
63
10 dB better than the plasma-treated PTFE SATURN microphone. It was expected, given
the iPhone’s microphone has both hardware and software amplifiers to improve the qual-
ity of the signal. The SNR curve of the self-powered SATURN microphone with plasma
treatment is comparable to Sparkfun ADMP 401 and Foil electret condenser with 1.5 V
bias, up to 5000 kHz. The passive SATURN microphone is competitive with some active
microphones in terms of signal quality to 5000 Hz. In addition to being self-powered, the
SATURN microphone’s flexibility lends it the ease of being embedded in different physical
objects.
A MEMS microphone is another interesting comparison point. These are also small
microphones that can be embedded in objects. They offer low power consumption [105]
of approximately 40 µW at -20 dB sensitivity. While 40 µW may seem like a small power
budget, it would not be appropriate for scenarios requiring a large number of microphones
or a very long life. It would be better to use that power budget for other local computing
tasks [61] or communication [178]. MEMS is also a fairly complex micro-fabrication
technique compared to our SATURN design.
(a) (b)
Figure 3.26: Determination of load resistance: a. Peak voltage b. Peak power of 4x4 cm2 SATURN
microphone as functions of the external load resistance at resonance frequency
I performed an experiment to determine the peak voltage and peak power of the SAT-
64
URN microphone as functions of the external load resistance at its resonance frequency.
The resonant frequency of the SATURN microphone can vary slightly with each fabri-
cated patch. Thus, I first perform a frequency sweep to determine the system’s maximum
Vopen−circuit of 0.9 Volts at 255 Hz with a 105 dBSP L sound source. The measurements
for the experiment were done using a capacitative oscilloscope (Keithley 6514). Next, the
external loads were changed successively using a variable resistance box (Elenco electron-
ics) and the corresponding Vpeak−peak was recorded. As shown in Figure 3.26a, the output
Vpeak−peak increases quickly as the resistance increases from 0.1 Ω to 2 M Ω and approaches
an asymptote at 8 MΩ resistance. When I wish to use the SATURN microphone as a sen-
sor, a load resistance of 8 MΩ would give the best result. Such high resistance, however,
is not ideal for general electronic circuits. I suggest using a load resistance of 2 MΩ when
connecting the SATURN mic as an audio sensor. Figure 3.26b shows the power vs load
resistance curve, where power is calculated as Vpp2 /Rload . Maximum power of 6.9 micro
Watt can be generated from the SATURN microphone at a load impedance of 0.9 MΩ when
excited by a 255 Hz tone at 105 dBSP L .
Going further, I analyzed the 4x4 cm2 SATURN non-plasma microphone patch as a
power harvester. The power curve and Vpp with a load of 0.9 M Ω at different frequencies
is shown in Figure 3.27. The voltage is approximately 0.5 Vpp at 150 Hz and rises to 2.5
Vpp, and then comes back down again at 350 Hz. The same behavior is shown in the power
curve, with a maximum of 6499 nW.
Figure 3.27: SATURN as power harvester: Voltage and power generated for different working
frequencies
65
Table 3.3: Summary table of power generated by SATURN at 105 dBSP L
3.9 Exploring the Application Space for the Thin Flexible SATURN Microphone
I explore different application scenarios where the SATURN microphone can be embedded
in everyday settings. The applications mentioned are exploratory in nature and are shown
to demonstrate that the quality of the audio signal recovered by the SATURN microphone
in different configurations is good enough to support a variety of interesting situations. The
first two applications take advantage of SATURN as a passive microphone with a thin and
flexible form factor. However, the full application still requires signal acquisition and pro-
cessing from traditional computing devices. The third example, while not implemented,
demonstrates the potential for using SATURN and its power harvesting capability to pro-
vide more end-to-end service.
The SATURN microphone is a skin-like sensor that can be placed on different flat or curved
surfaces in a room, like a curtain, wall, or tabletop, to gather context. Multiple SATURN
microphone patches can easily be placed on the surface of a table and, in combination, can
be used to localize a speaker. As people speak, the location can simply be determined by
comparing the voltage output of multiple SATURN microphone patches. The patch placed
near the speaker will pick up more signal than one placed further such that even a simple
algorithm of threshold amplitude comparison can detect which speaker is actively talking.
Figure 3.28 shows a simple example of such localization. When speaker A (yellow) speaks,
66
the closer SATURN microphone (1) has a higher amplitude than the other microphone (2).
Similarly, when speaker B (blue) starts speaking, the closer microphone (2) has a higher
amplitude than the other microphone (1). Such infrastructure can be expanded to multiple
parts of the table, given the number of speakers. I could even imagine placing the SATURN
microphones on the walls/ceilings/floors in order to localize speakers within the entire room
using more sophisticated processing of the combined signals.
Figure 3.28: Localization of speakers around a tabletop: Multiple SATURN microphones are
placed on the table for localization of speaker. The SATURN microphone placed nearer to the
speaker has more voltage output e.g., microphone 1 has higher electrical output than microphone 2
when speaker A speaks and vice versa.
Figure 3.29: A Sound-sensitive bottle: a. SATURN microphone embedded in a soda bottle b. User
talking to soda bottle c. Recovered audio signal (d) Spectrogram
The SATURN microphone’s flexibility and form factor allow the possibility of attach-
ing a microphone to everyday objects, such as a bottle (Figure 3.29a). A 4x4 cm2 SATURN
patch was placed on a soda bottle to enable interactions. A person may give voice com-
mands like - “let’s share a coca cola,” which can enable control like actuating a display
67
(Figure 3.29b). The time series graph of live speech is shown in the time series volt-
age (Figure 3.29c) with corresponding spectrogram (Figure 3.29d), which shows sufficient
detail to do spectral feature extraction. Further, Figure 3.30 shows a comparative spectro-
gram for ground truth recorded using a iPhone 6 and SATURN mic simultaneously. With
appropriate storage, computation, and communication, we can imagine a wide variety of
interactive voice-activated capabilities.
Figure 3.30: Comparative spectrogram for ground truth recorded using a iPhone 6 and SATURN
mic simultaneously
68
ance, such as in construction zones, mines, music venues, power stations, airports, space-
ports, and military environments. Similarly, SATURN-based sensors might be used for
monitoring events such as landslides, avalanches, polar ice breaking, mine cave-ins, and
mine gas explosions. In a more futuristic application, United Nations could drop SATURN-
based sensors from an airplane into a conflict zone. The sensors would monitor the acous-
tic environment for the movement of tanks Figure 3.31, mortars, or exploding ordinance.
Later, an officer with a reading device might sweep the field to interrogate the sensors. In a
more extreme scenario, a low-flying helicopter might sweep a strong RF signal over the re-
gion and record which sensors report hearing an event. The pattern of reporting sensors can
reveal the direction of travel and point to possible hiding areas for that equipment which
could possibly prevent destruction and loss of lives.
Figure 3.31: Dropping remotely interrogable stickers to monitor tanks and ordinance in a war zone:
Loud sound acoustic sounds like tanks passing can be potentially used to flip bits in SRAM bits that
consume approximately 10-100 pW of power which can later be interrogated
Figure 3.32: Recording a loud acoustic event using power generated from a SATURN microphone.
As a motivating example, imagine an airport that would like to monitor its acoustic
environment so as not to exceed safe noise levels for its employees or to keep aircraft
69
noise footprints within airport boundaries as shown in Figure 3.32. A The SATURN-based
system can be tiled on various buildings and at various distances on the runway. As planes
take off, they generate loud sounds due to gears, fans, and air turbulence. The peak in the
sound spectrum generated by aircraft is near the 200-300 Hz band (Figure 45 in khorrami
[95]) with decibel levels reaching > 105 dbSP L at 5m3 . These values are consistent with
the resonant frequency of the SATURN patch and would result in the generation of power
> 6.9 µW accumulated over different frequency bands. Considering the maximum power
transfer theorem (Jacobi’s law), the usable power I can obtain from such a phenomenon is
approximately 50%. Thus, I might harvest up to 3.4 µW. The energy required to program
a “1” in a NAND flash memory is 2µJ [122]. Given that the sounds I am expecting will
probably last for several seconds, there is more than enough power to record the acoustic
event. Going further, SRAM bits can be flipped at approximately 10-100 pW of power [64,
150], suggesting that rudimentary computation might be performed to determine if the flash
memory bit should be written. A worker can then visit each SATURN site, interrogating
the system using a passive RFID mechanism. When the worker places the active RFID
reader above each SATURN system, it reads the state of the recorded bit and resets the
system so that it is ready to catch the next episode.
Going further, after detecting a loud sound, the SATURN system might use its harvested
power to power an RF transmitter to announce the event. For example, Talla et al. [177]
have recently demonstrated a 915MHz analog LoRa backscatter communications device
that can communicate at greater than 11 bits/sec while hundreds of meters away from its RF
source and receiving antenna. While their system currently uses a battery, their theoretical
IC design consumes only 9.25 micro-watts of power. With sound events lasting on the order
of seconds, one can imagine a SATURN-based system storing power until it has enough to
enable a 915 Mhz backscatter transmission to the receiving antenna, announcing the event.
As long as the event continues to occur, the SATURN system can transmit alerts every few
3
NASA jet engine spectra : https://www.grc.nasa.gov/WWW/Acoustics/Education/InternArchive/Schifer.
htm
70
Figure 3.33: SATURN as surface vibration harvester: a. SATURN placed on surface of a blender
b. Voltage produced c. SATURN generates 5 µW of power due to vibrations from the blender. We
can imagine SATURN being used as a power harvester for heavy industry equipment monitoring
sensing systems
.
71
Figure 3.34: SATURN embedded in the face mask acts as a human breath-based power harvester
unique insight into some health data like the number of coughs, breathing rate, etc. Inspired
by this, Figure 3.34 shows the SATURN device embedded into N95 masks. Simple fab-
rication and a lightweight and configurable construction allow SATURN to be embedded
easily into different types of masks. A SATURN embedded in a face mask can generate
continuous power from human breath, coughing, or talking. Such acoustic vibration inputs
are similar to an acoustically loud sound environment, resulting in surface vibration in the
mask. Preliminary results for power harvesting showed that coughing results in 2.3 uJ/min
and 4.8uJ/min. This power collected over time can be used to power upcoming intermittent
computing platforms for physiological data monitoring sporadically during the day [32].
In conclusion, inspired by earlier work in materials science that demonstrated the op-
portunities for harvesting energy from the triboelectric effect, I have presented the design,
evaluation and potential use of the SATURN microphone (section 3.2 to section 3.9). The
thin and flexible two-layer SATURN microphone design has favorable audio performance
when compared to other passive and active microphones and requires no power. SAT-
URN’s simple fabrication process and ease of deployment on a variety of surfaces enable
new opportunities for audio sensing over large indoor/outdoor areas for both mobile and
stationary objects. SATURN may enable battery-less remote sensing for acoustic events,
which has potential applications in controlling noise pollution, workplace safety, environ-
72
mental monitoring, and military situation awareness. Next, I will detail other interaction
sensors.
In this section, I will detail the design and working of a flexible-thin finger swipe gesture-
based sensor that is made with easily available materials and fabrication techniques. The
swipe gesture sensor allows for sensing of direction and identity as detailed below:
73
– paper, copper foil, and kapton (plastic) tap for fabricating the electrodes Figure 3.35a.
The patterns of copper foil are cut using a silhouette paper cutter. Next, teeth sensors of
three different capacitance values are designed by the layering of copper foil of different
sizes and kapton tape of varied thicknesses. For this first, teeth of different sizes were cut
from the copper foil and pasted onto the paper. It is followed by layering kapton tape in-
dividually on each tooth individually to create different capacitances. For example, digit
1 has three layers of Kapton sheets and will have the least amount of capacitance change
when someone touches it, while digit 3 has one layer and thus has the maximum capaci-
tance change when someone touches it (Figure 3.35b). I performed a short study with 5
participants to confirm that each of the digits fabricated is differentiated enough in their
capacitance change when a human finger touches it (Figure 3.35c). Measurement of ca-
pacitance was done using an LRC meter was used at 10 Khz. I seldom observe borderline
cases, and in the future capacitative sensors can be further tuned by adjusting the surface
area of the tooth and the number of layers of Kapton or other materials on top of it. The
final prototype is shown in Figure 3.35d.
Figure 3.36: Swipe direction gesture sensor: a. Sensor prototype b. Monotonically increasing and
decreasing freq change signature
Based on the design of capacitative teeth, which allows for three differentiated capac-
itative values, I developed a swiping surface that monotonously increases or decreases its
capacitance depending on the direction of swiping Figure 3.36a. The capacitative values,
because of their frequency dependence, are often read as frequency changes. Thus, Fig-
ure 3.36b shows a monotonically increasing and decreasing frequency spectrum for the
74
right and left swipes, respectively. The direction sensor built has been integrated with the
MARS wireless communication sticker. The details of integration and signal processing
for detection are provided in subsection 4.13.4.
3.10.3 Identity
Figure 3.37: Swipe-based identity sensor: a. Sensor prototype b. Unique frequency pattern
The identity sensor is spun off of the basic capacitative swipe-based sensor, where the
differentiated capacitative tooth created are arranged in unique patterns (Figure 3.37). In
the future, ID sensors can be built beyond just vertical barcodes and extended to several
shapes and even more expansive sets of teeth based on the materials chosen. The identity
sensor built has been integrated with the MARS wireless communication sticker, and the
details of integration are provided in subsection 4.13.5.
The touch sensor has a basic button design and a network of inductive sensors. While the
design is not novel, the touch sensor itself serves as an enabler later for sustainable wireless
interactive stickers in chapter 4.
L11 and L12 are two inductors placed below the arrow key buttons, which act as a
switch in the circuit. Initially, when no game is being played, L11 is in series with L12,
so the effective L1eq = L1||(L11 + L12) in steady-state. When any one of the buttons is
pressed (e.g., the up arrow key), the inductor is parallel to it, and L1 is effectively shorted.
This shorting makes the inductance L1eq = L1||L12. Similarly, when down is pressed,
75
Figure 3.38: Inductive touch sensor: a. Schematic diagram of the prototype b. Circuit diagram
then the inductance becomes L1eq = L1||L11. Each change to L1eq results in a change in
frequency when integrated with the LC oscillator-based sensing system.
Like sticky notes, I imagine that thin-flexible sensors embedded in interaction stickers
can be made into many different shapes and sizes to attach to other objects. For example,
a rectangular interaction sticker attached under each item on a printed food menu (Fig-
ure 3.39) can allow a customer to order the item by simply sliding their finger across it.
This remote control capability can be used in other scenarios, such as lighting controls
(Figure 3.39). In addition, the MARS interaction sticker can be designed to convey its
meaning visually; for example, it can mimic the design of a game controller (Figure 3.39)
76
or reveal multiple-choice options on a desk in a classroom for interactive polling.
In this section, I will critically look at some limitations of the sensors built in this chapter
and provide possible future alternatives and areas of improvement.
As part of the self-powered operation goal, components that require external power to op-
erate, such as a pre-amplifier, have been removed. It comes at the price of reduced audio
sensitivity, and as a result, the SATURN patch’s performance is equivalent to an active mi-
crophone till 5kHz. The glue anchor points limit the maximum def f that can be achieved.
Thus, after 105 dBSPL, increasing the acoustic pressure further results in a voltage sat-
uration. A more comprehensive mathematical and electrical model of the two layers of
SATURN, paper, and PTFE and their interactions can provide a deeper understanding of
structural device parameters like attachment points for patches of different sizes, shapes,
and flexibility. This model could also help design SATURN microphone patches tuned at
different resonant frequencies for different applications.
As a result of the design and the materials selected for fabrication, the SATURN patch may
be affected by environmental conditions such as humidity, wind, and extreme vibration.
Most of the tests conducted were under controlled environments, and further analysis is
required to determine the effect of such adversarial conditions, as well as the overall dura-
bility of the design over time. SATURN can also be affected by electric field coupling, like
powerlines. This may be improved with appropriate shielding techniques.
A similar issue may be observed in capacitance-based swipe sensors. Further, the swipe
sensor based on the materials used may experience capacitance overload. It limits the
77
number of unique capacitance-based digits that can be built. One way to overcome this is
using inter-digitated structures in sensor teeth rather than single teeth that are at different
heights; e.g., cooktop demonstrated a swipe sensor with several inter-digitated teeth of
varying lengths robust to different kinds of swipe directions. [136].
In the future, other examples of Self-powered sensors can be explored. Inspired by SAT-
URN and capacitative ID sensors together, utilizing everyday materials, e.g., using paper,
copper foil, and plastic, self-powered binary-patterned Triboelectric Nanogenerator based
identity sensors can be built.
3.13.4 Simpler Manufacturing Process and Experimenting with Materials of Varied Shelf-life
78
and other low-cost scalable manufacturing techniques. Some applications of SATURN mi-
crophone patches suggest a disposable use, such as a label on a bottle. The low cost of
SATURN makes that a possibility. Other applications, such as enabling a tabletop surface
or wall to do auditory scene analysis or a building/road surface triggering noise alerts, con-
sider the long-term use of such material. The former short-term applications require very
inexpensive manufacturing processes, while the latter requires durability.
When I discussed the use of SATURN patches to perform ambient monitoring of acoustic
scenes, I introduced the opportunity to transmit the harvested audio signal of the SATURN
patch via mechanisms like analog backscatter [178]. It opens the possibility of more self-
sustained application scenarios, which I consider the most compelling direction for our
future research. To take advantage of SATURN microphone as a self-powered sensor with
high acoustic sensitivity, I should either connect it to low power processor [61, 149, 162],
which allows for both operation and recognition of sound in about a few tens of micro-
watts as shown in Figure 3.41 or send the audio to the remote base station for recognition
using analog backscatter [178] which would only consume a few micro-watts which can be
harvested from the environment.
Figure 3.41: Low-power real-time local sensing and recognition system design: schematic diagram
of self-powered analog sensor like SATURN microphone connected to analog recognition SoC like
Field Programmable Analog Arrays [61]
79
3.14 Conclusion
Given the power, cost, and form factor constraints, there are two types of sensors presented
in this chapter. First are self-powered sensors, and second is the traditional capacitative and
inductive sensors that encode their values in material design to do the sensing. Both the
sensors have pushed for using cheap thin easily available materials. I have also explored
the space of potential applications for such material as part of a self-sustainable interactive
system that is concretely built in further chapters. Broadly speaking, this Chapter lays
down the following guidelines and contributions for the selection of interaction sensors for
sustainable interactive stickers :
• Use of everyday materials reduces cost and optimizes on form factor: Sensors
and displays that are built leveraging easily available materials like plastic and paper
with simple fabrication and assembly result in an overall reduction in the cost of the
system.
• Self-powered Sensing: Phenomenon being sensed also powers the sensing and elim-
inates the need for powering explicitly in some sustainable systems or at the least
removes the need for amplifiers.
• The right fabrication technique can optimize the power consumption: Leverag-
ing novel physics and materials science advances, the device design and fabrication
can be optimized to have lower power consumption while still maintaining the re-
quired functionality and sensitivity, e.g., the introduction of glue dot points in SAT-
URN.
80
In Chapter 4, I will be including the self-powered sensors, e.g., SATURN, and other
passive sensors for direction, identity, and touch as part of a backscatter-based wireless
communication.
81
CHAPTER 4
COMMUNICATION: NO/LOW POWER WIRELESS STICKERS BASED ON
AM/FM BACKSCATTER COMMUNICATION
4.1 Introduction
Figure 4.1: Sustainable wireless sensing with low/no-power simple circuitry leveraging analog
backscatter communication: a) ZEUSSS adopts amplitude modulation of sensor information with
incoming RF wave b) MARS adopts frequency modulation/shifting that allows multiple tags to
communicate information simultaneously
Sustainable Wireless Interaction Stickers enable indoor physical objects and surfaces
to be instrumented with a thin, self-sustainable cheap material that can provide interaction
sensing and communication capabilities. I demonstrated an array of interaction sensors in
chapter 3 that follow the system design parameters of power, cost, and form factor. Next,
in this chapter, I look at ways to build low/no power wireless circuits and systems that
are consistent with the system design parameters while maintaining functionality. I em-
82
phasize low-complexity fully-analog circuits to achieve low-cost and ultra-low-power (few
microwatts) operations. In addition, this chapter leverages specific strategies like backscat-
ter communication, low-threshold voltage transistor, and transistor as a passive voltage-
controlled-resistor. The communication circuitry is built on thin substrates to maintain the
sticky note form factor. I introduce two projects below, ZEUSSS and MARS, that itera-
tively build onto each other to accomplish sustainable communication aims:
• ZEUSSS: ZEUSSS (Zero Energy Ubiquitous Sound Sensing Surface), allows phys-
ical objects and surfaces to be instrumented with a thin, self-sustainable sticker
that provides acoustic sensing and communication capabilities. I combine the self-
powered mechanical vibration sensing property of SATURN with JFET (Junction
Field Effect Transistor) as a voltage-controlled resistor to enable amplitude modulation-
based analog backscatter communication. The ZEUSSS tag has a thin, flexible form
factor and employs no-additional power harvester other than SATURN itself. While
ZEUSSS has simple circuitry and low cost, it is limited in scalability as it operates
on a single channel.
83
power requirements, cost, and part count and explore different interaction and activ-
ity sensing scenarios suitable for indoor environments.
In the next sections, I will first explain the theory of analog backscatter communication
(section 4.2). That is followed by details regarding ZEUSSS (section 4.3 to section 4.8).
Further, I explain how the learnings and limitations of ZEUSSS motivated the creation
of MARS. I go into the details of MARS’ system design and working, applications, and
performance (section 4.11 to section 4.16). I end the chapter with some general leanings
and observations that can be applied to building low-power communication systems (sec-
tion 4.17).
I will first explain the concept of backscatter communication in a more generic way.
The word backscatter in this situation means reflection. I can explain analog backscatter
by using a simple analogy of the reflection of light rays (Figure 4.2a). When light hits a
mirror, it bounces off the surface at an angle equal to the angle of the incoming light wave.
However, when light hits an irregular surface, it is diffused, resulting in modulation in both
the intensity and angle of reflection of the light waves. Analysis of the received diffused
light can help deconstruct the shape of the irregular surface from which it is reflected. Sim-
ilar to light waves, when radio frequency (RF) waves are incident on a specially designed
84
tag, they get modulated in amplitude and phase. This reflected RF from the tag can be
processed at the receiver to extract information about the phenomenon which caused the
modulation.
Next, I will go into theory for analog backscatter and demonstrate how the modulation
of the reflected wave depends on the audio/sensor input being sensed. Figure 4.2b illustrates
a typical backscatter system that consists of a high radio frequency transmitter and receiver
setup with wireless tag(s) located in its vicinity. The power density of electromagnetic
waves incident on a tag placed at a distance r from the reader is given as
Pt Gt
S= (4.1)
4πr2
Where Pt is the power transmitted by the reader and Gt is the gain of the receiver an-
tenna [38]. The incident power is collected by the aperture of the tag antenna and delivered
to the terminating impedance. Simultaneously, a fraction of the incident power is scattered
by the antenna. The analysis in this work of received and backscattered power is based on
the Thevenin equivalent circuit model shown in Figure 4.2c where VA is the open circuit
voltage on the antenna terminals, ZA = RA + jXA is the complex antenna impedance and
ZT = RT + jXT is the impedance across the antenna terminals [18]. Passive backscatter
tags modulate the impedance across the antenna terminals, which causes a change in the RF
field backscattered by the tag. These changes in the field are detected by the receiver and
are used to decode the information communicated by the tag. The power re-radiated can be
computed using the equivalent circuit model shown in Figure 4.2b and can be written as –
Pt Gt λ2 Gr 2
Pbackscatter = Sσ = 2
· |1 − Γ|2 (4.2)
4πr 4π
where, σ , the scalar radar cross section (RCS) of the antenna is defined as
λ2 Gr 2
σ= |1 − Γ|2 (4.3)
4π
85
and Γ is the reflection constant defined on impedances of the antenna and the sensor as
ZT − ZA ∗
Γ= (4.4)
ZA + ZT
For maximum power transfer ( |Γ| = 0 ), the antenna should be terminated by the complex
conjugate of its impedance i.e., ZT = ZA ∗ . Also, the backscattered field from the tag
antenna undergoes path loss in the reverse direction, and the signal power received by the
reader located at a distance r from the tag is given as
Pbackscatter Ae Pt G2t λ2 σ
Preceived = = (4.5)
4πr2 (4π)3 r4
In the setup for ZEUSSS and MARS, the impedance of the sensor (Zsensor or ZT ) varies
as a function of the sensing quantity, thereby modulating the backscatter power with sen-
sor information. Combining Equation 4.4, 4.3, 4.5 together analog backscattered power
PanalogBack as a function of the impedance of the sensor can be written as:
λ4 Pt Gt 2 Gr 2 2
4RA
PanalogBack = · = f (Zsensor ) (4.6)
(4π)3 r4 |RA + Zsensor |2
The receiver captures this information and performs signal processing on it to recover
the sensor information. Next, I will be looking at how this theory can be leveraged by
ZEUSSS.
A ZEUSSS prototype tag consists of three main components: SATURN, a junction gate
field-effect transistor (JFET), and a printed antenna (Figure 4.3) [11]. The SATURN micro-
phone is self-powered. When the sound waves move the SATURN structure, the distance
between the two plates of the resulting in a change in capacitance change and generation
of Vspeech , which is directly proportional to the volume of the speech at the diaphragm. It
86
is typically ∼ 1V for nearby speech and 100 mV from 1m for a 4x4 cm2 patch.
Figure 4.3: ZEUSSS system design: Self-sustaining sound sensing and communication system
using SATURN and analog backscatter technique
The next component in the system is the junction-gate field-effect transistor (JFET)
which is one of the simplest types of field-effect transistors. JFETs are three-terminal
(drain, source, gate) semiconductor devices that can be used as electronically controlled
switches or resistors or to build amplifiers. A JFET has two regions of operation, as shown
in Figure 4.4: the ohmic region (also called the triode/linear region) and the saturation
region. This segregation is done on the basis of Vpinchof f or Vp , which is achieved when
drain-to-source conduction stops. In ZEUSSS, we will be exploiting JFET’s property as a
passive voltage-controlled resistor in the ohmic region.
87
a JFET with VGS = 0 bias point operates in the triode/ohmic region and can be used as a
trans-impedance (i.e., voltage to impedance) amplifier or passive VCR (voltage controlled
resistor). Other transistors cannot be used for this purpose because MOSFETs and BJTs
require a bias voltage to operate in the active/saturation region; JFETs are by default active
when VGS = 0. Thus, in the ZEUSSS circuit the SATURN-JFET connection does not re-
quire an additional bias voltage at the gate of the JFET. Next, JFET is optionally connected
to the LC matching circuit, which is connected to the third main component, the antenna.
Thus, SATURN, the JFET in the triode region, and the antenna together are the building
blocks of ZEUSSS.
Next, I will look at some theories of how the backscatter of audio from SATURN hap-
pens and how the system can be tuned for performance. In the ohmic region of the JFET,
2IDSS
ID = (VGS − Vp − VDS /2)VDS (4.7)
Vp2
The voltage from SATURN, Vspeech , changes the impedance between the drain and
source terminals (RDS ) of the JFET. The JFET impedance in terms of Vspeech can be math-
ematically written with the help of Eq. 4.7 and 4.8 as Eq 4.9.
Vp2 1
RDS = (4.8)
2IDSS (VGS − Vp − VDS /2)
With VDS approaching 0 when RDS o is the impedance of the JFET for zero voltage at
the gate terminal, we get :
RDS o
RDS = Vspeech
(4.9)
1− Vp
Vp2
Here RDS o = 2IDSS
and VGS = Vspeech . Next, to maximize the signal backscattered by the
microphone, the impedance of the JFET is tuned to match the impedance of the antenna
using an L-C matching network. Thus, the impedance of the SATURN is seen by the
88
antenna. RSA (referred ZA previously in Eq 4.4) is
ω 2 L2 Vspeech
RSA = = R0 (1 − ) (4.10)
RDS Vp
R0 is the impedance at the antenna terminal for no speech. The backscatter signal is based
Ra −RSA
on the reflection coefficient Γ = Ra +RSA
, which leads to
2Ra R0
Sbackscattered ≈ (4.12)
R0 + Ra
Ideally, we want the JFET resistance/impedance to vary from maximum open to close on
the smith chart for a particular frequency between the lowest (0V) to maximum Vspeech
input applied. In conclusion, from the theory discussed above and in section 4.2, sounds
and vibrations in the environment result in the generation of voltage in SATURN, which
in turn changes the impedance in the circuit via the JFET, which acts as voltage controlled
resister in the ohmic region. This change in impedance changes the radar cross-section of
the antenna resulting in amplitude modulation of the incoming RF waves.
The experimental system consists of three main components: the transmitter (TX), ZEUSSS
tags, and the receiver (RX). I use an Ettus Research N210 USRP with a UBX-40 USRP
daughterboard as a transceiver in a monostatic backscatter configuration with a separate
antenna for transmitting and receiving as shown in Figure 4.5. For our experiments, the
USRP produces a 915 M Hz carrier wave (16 dBm) (within the FCC limit of 30 dBm).
89
The transmitter and receiver antenna (circularly polarized 6 dBi) were placed in the room,
approximately 3 feet from the MARS tags.
The selection of the right voltage controlled resistor is essential for the optimal working
of the ZEUSSS sticker. After experimenting with 20+ JFETs, I selected the MPF-102 by
Fairchild for the system that gave the clearest and most reliable audio signal. Below, I have
characterized its performance for modulating the antenna impedance. Figure 4.6 shows the
change in resistance at the antenna side, Rds with respect to the input voltage Vgs . It is
not a graph that can be found in transistor datasheets since the voltage applied to the drain
side is zero, which is unlike normal common source configuration. The antenna is tuned to
low resistance, around 50 Ohm, and can only see resistance changes at about 1000 Ohm.
Bigger changes are considered as load or too big for the antenna to detect. Thus, voltages
from -300 mV to +800mV can be detected.
Next, the minimum current requirement Igs is plotted against the input voltage Vgs .
The current requirement is 100nA until +300mV and rises suddenly after that. The orange
region is the voltage and current that SATURN can provide for speaking nearby. It can
be seen that it is not possible for SATURN to drive the resistance change after +300mV.
90
Figure 4.6: Behavior of a JFET as a voltage controlled resistor (VCR): The graph shows the change
in resistance with respect to input voltage Vgs
Combining the two results shows that the SATURN voltage needs to be restricted to +/-
Figure 4.7: Minimum current requirement Igs vs Vgs for the transistor to act like a VCR (blue).
Current and voltage produced by SATURN (orange).
300mV with enough current to result in a sufficient change in impedance such that it can
be detected.
The impedance change can be further improved by the introduction of a voltage divider.
More investigation into transistor device physics is warranted for improving the stability
of this system. In the future, the transistor can also be replaced by a varactor that does
imaginary impedance changes. The varactor would be easier to print, making the whole
91
circuit design passive and printable. In the future, a more thorough analysis of the JFET
physics and model is warranted to come closer to the theoretical model where backscatter
is maximized.
Figure 4.9 demonstrates some examples of backscattered audio received. It was generally
observed that speaking closely to a ZEUSSS tag placed on the table at a 2 m distance from
the mono-static TX/RX station gives a reasonably good signal with 15dBm power. This
distance can be increased further with a higher power. With 27 dB Tx, we found the range
of audio (PESQ score > 1.5) to be increased to 5.5 m in the monostatic configuration of
TX/RX.
Figure 4.9: Example spectrograms of backscatter audio received from ZEUSSS tag: speech, speech
at different volumes, and whistling
The thin and flexible form factor of ZEUSSS allows it to be placed on top of different
physical surfaces for self-sustainable audio sensing and communication. We imagine sce-
narios where many different ZEUSSS tags in the home can cheaply extend the range of
audio input for home assistants (e.g., Amazon Echo or Google Home), which are currently
limited to a single room. Figure 4.10 shows ZEUSSS embedded as part of the conference
brochure to act as an extended microphone in an auditorium. In addition to audio sensing,
92
Figure 4.10: ZEUSSS can potentially be used as an extended microphone in an auditorium
the ZEUSSS tag can also be used as a contact microphone to sense input touches, such that
different speeds or forces of taps could be detected. It could extend input gesture and con-
trol capability for IoT applications like switching lights on or off or changing brightness.
ZEUSSS could be used as an authentication device, where users can speak, blow, whistle,
tap, or create a unique combination of these as their password to do a remote login.
Next, I articulate the contributions of the ZEUSSS in furthering the agenda of sustainable
wireless interaction stickers.
The ZEUSSS tag has extremely simple circuitry, where the JFET is the only active com-
ponent. Simple circuitry may mean it is feasible to make a fully printable prototype in the
future. In addition, when the circular life-cycle of the interactive sticky note is considered,
fewer silicon components are preferred.
93
4.8.2 Strategies for Lowering Tag Power
Backscatter communication has been an area of active research in the past decade. Follow-
ing the trend, ZEUSSS pushes for fully analog circuitry at the tag end and delegates the
more power-intensive tasks for the infrastructure. Digital components are not a require-
ment for wireless communication. They impart advantages for signal quality but are not a
necessity for operation; analog signaling is appropriate and advantageous if one is willing
to accept some trade-offs.
Generally, a JFET is used in the saturation region with a bias voltage, but the ZEUSSS tag
employs the JFET as a passive voltage-controlled resistor. Such use of a JFET as a passive
analog switch lends low-power, low-voltage operation capability to the tag in that it can
operate on SATURN voltage alone.
94
and contextual sensing domain.
ZEUSSS is an example prototype that satisfies all three system design constraints that we
aspired for sustainable computational sticky note – power, cost, and form factor (chapter 1),
but it is still limited in scalability. It does not provide for the simultaneous communication
of tags. In addition, it also has limited distance due to the presence of a high noise floor near
the central band. Thus, the logical next step is to think of alternative ways for sustainable
wireless sensing communication that can allow for different tags to communicate on their
own individual channels over sufficiently large distances for indoor applications without
compromising on the original design parameters.
I achieved the power, cost, and form factor design parameters for MARS sticky notes by
creating a circuit for frequency modulation (FM) -based backscatter communication [152,
187]. Leveraging the Zero-Vth MOSFET [90], I build a modified version of the traditional
LC Clapp oscillator to have low-startup voltage (500mV), current (∼ 2µA), and power
95
consumption (< 1µW ) up to a frequency of 1 Mhz (subsection 4.11.2). The resonant fre-
quency of the Modified Clapp Oscillator (MCO) is controlled by the inductance (L) and
capacitance values (C), which is exploited for inductance, capacitance, and voltage-based
analog sensing. This technique allows MARS to enable a diverse set of interfaces (speech,
slider, ID, and touch) that can augment objects/surfaces in the environment by simply plac-
ing a sticker. I use our previous work on a self-powered paper microphone, SATURN [14],
for speech sensing and design novel capacitance-based direction and identity sensors. With
minimal startup power, MARS tags can communicate sound up to 9m and other interac-
tions up to 12m (in monostatic configuration). I demonstrate that MARS interfaces can be
powered by two photodiodes harvesting ambient light in an office environment or a thermo-
electric generator converting power from heat flow from the user’s fingertip (section 4.12).
MARS interactive stickers are built using 9 components (2 active and 7 passive) and less
expensive power harvesters than most previous systems (section 4.12, section 4.15).
Our MARS system consists of multiple MARS tags augmented onto different objects
and surfaces in the environment within the transmitter and receiver communication range.
The transmitter is responsible for producing a high-frequency carrier radio signal (e.g.,
Fhigh in Figure 4.12) which is incident on the tags. Leveraging a custom ultra-low power
oscillator, each of the MARS tags is tuned to a unique frequency (e.g., F1-3 in Figure 4.12)
much lower than the transmitter’s carrier frequency. The unique resonant frequency allows
each tag to communicate information in its own individual channel. For example, Tag 1
sends information at F1 shifted from Fhigh . Based on the interactions the tag oscillator
shifts a single frequency or modulates a band of frequencies (speech). Figure 4.12 shows
an example of frequency shifting for interaction with Tag 1 where the sensor changes value
as a sine wave. This frequency-modulated sensed information signal from the oscillator
is communicated in real-time to the receiver by amplitude-modulating it with the incom-
96
Figure 4.12: MARS system based on frequency-shifted backscatter: MARS system consisting of
transmitter, receiver, and multiple MARS tags in the environment. The input interactions result
in a change in the electrical property (blue) resulting in frequency modulating the oscillator signal
(magenta). This signal is backscattered to the receiver (green) using an analog impedance switch
and antenna. The receiver receives frequency-modulated information at different bands from each
tag, which is demodulated back to the original sensed signal.
ing high-frequency carrier signal. This technique, where the active radio source is in the
transmitter and the tag is only responsible to modulate and reflect the incoming signal, is
called backscatter. Depending on the bandwidth of data (i.e., the single frequency or 8kHz
audio) being communicated during the interaction, this communication protocol is called
frequency shifted or modulated analog backscatter [152, 215]. I employ this protocol to
communicate speech, touch, ID and swipe information from MARS tags.
One of our main contributions is the hardware design implementation of the nano-watt
power budget MARS wireless sensing tag. The MARS tag (Figure 4.13a) consists of three
main blocks as shown in Figure 4.13b; (1) the communication block (magenta), which is
responsible for the FS-backscatter; (2) the power harvesting block (red), which leverages
ambient power sources (e.g., body heat, photodiode) to power the communication; and
(3) the sensing block (blue), which enables sensing of different electrical phenomenon
(inductance L, capacitance C, or voltage V) to support interaction sensing.
In the following subsections, I detail each of the tag blocks and their operation. I first
97
start with a primer on Zero Vth MOSFET, a piece of known technology that is essential for
the realization of the ultra-low power and startup-voltage oscillator used in the communi-
cation sub-block. Next, I explain different communication sub-blocks – oscillator, analog
switch, and antenna. I specifically focus on a detailed description of the hardware design
process and characterization of the oscillator. This section is followed by an elucidation
of the role and selection of components for the analog switch and antenna. Next is a dis-
cussion on how the tag design enables sensing based on the inductance, capacitance, and
voltage change. Finally, I will discuss how the tag gets powered by ambient power sources
like light and body heat to enable wireless sensing.
Figure 4.13: MARS tags: a. Prototype b. The three major components of the MARS tag are
wireless communication, power harvester, and sensor.
The MOSFET is a four-terminal device (body, gate, source, and drain) that forms a basis
for modern electronics. Each MOSFET device has a threshold voltage, Vth , set during
the manufacturing process, that is the minimum amount of voltage applied to the gate
needed to create a conducting path between the source and drain terminals. A typical
MOSFET requires a supply voltage, Vdd , which is a few 100’s mV higher than Vth to
operate. This difference between Vdd and Vth is called the gate overdrive. Since, the overall
device power requirement is quadratically proportional to the Vdd , one way to lower the
circuit power consumption is to lower the Vdd . As the feature size in modern MOSFET
98
devices scales down, the required Vdd decreases (Dennard scaling [35]), but the VT H does
not scale down at an equal rate [67]. If, however, the Vdd is lowered too much, it diminishes
gate overdrive, which results in performance degradation of MOSFET and increases in
static power dissipation. A potential workaround to maintain consistent behavior with a
lower supply voltage is to reduce the threshold voltage Vth so as to maintain an adequate
gate overdrive.
Inspired by applications for low-voltage (LV) and low-power (LP), analog designers
have found alternative ways to maneuver around this limited voltage headroom by manu-
facturing specially designed MOSFETs. Zero-Vth MOSFETs were introduced as a compo-
nent in the mid 90’s [25], but it was soon realized that these transistors had an exponential
increase in sub-threshold current [117] (green in Figure 4.14a) in comparison to the stan-
dard MOSFET (red in Figure 4.14a), resulting in significantly high leakage power. Thus,
Zero-Vth MOSFETs were never received with much acceptance in the hardware commu-
nity.
Recently, however, by leveraging a floating gate MOSFET design [159] and propriety
EPAD technology [37] together, manufacturers like Advanced Linear Devices have been
able to reduce Vth to ∼ 0 mV (e.g., ALD110800A [36]) while still maintaining leakage
current ∼ 1 µA. This balance results in an extremely low supply voltage, even near 100
mV, to be practical for circuits. This novel near Zero-Vth transistors hold great potential for
nano-powered analog and RF systems. The green region marked in (Figure 4.14b) shows
an overdrive voltage range of +/- 0.15V and current < 3 µA. In this dissertation, I exploit
this region to build an ultra-low startup power and voltage oscillator as discussed in the
subsubsection 4.11.2. Similar low-power oscillators have been described previously in the
literature using floating-gate MOSFET [118, 163]
Specifically, in the context of MARS, usage of a Zero-Vth transistor reduces the os-
cillator supply voltage to ∼ 0.5V with ∼ 1uA current, allowing the system impedance to
match directly with low voltage DC power harvesters (e.g., two photodiodes or the touch of
99
a 3 cm2 fingertip on a thermoelectric generator) without the need for power management.
This convenience reduces the number of components required in the system and the size of
the harvester, thus lowering cost and opening the possibility of making the entire system
printable in the future (subsection 4.16.2).
Figure 4.14: Theory of operation: Zero-Vth Transistor a. Zero threshold voltage transistors have
higher current drive capability in comparison to low threshold and conventional transistors. [90] b.
Region of operation of zero threshold voltage transistor [36] exploited in our MARS system
The communication block in the MARS Tag consists of an oscillator, an RF junction field
effect transistor that acts as an analog switch, and an antenna. The combination of the ultra-
low power oscillator and analog switch results in the nano-watt wireless sensing capability
of the MARS tag.
Low-power and Startup Voltage Modified Clapp Oscillator (MCO): Hardware Design
and Characterization
An oscillator is at the core for generating unique frequencies for the FM-backscatter-based
MARS tags. In this section, I will explore the hardware design and characterization behind
MARS’s ultra-low power and startup voltage oscillator. The oscillator should satisfy the
following design criteria:
100
1. Produce a unique frequency with minimal power consumption.
3. The Vpp of the output sine wave signal from the oscillator should be within a range
for the transistor analog switch to perform in the lower triode region to minimize the
power consumed (explained further in subsubsection 4.2).
I will explain the oscillator hardware design in three stages. First, I start with a conven-
tional common gate LC Colpitts oscillator. Second, I replace its MOSFET with a zero-Vth
MOSFET to lower the operating voltage and modify it to the Enhanced Swing Colpitts Os-
cillator (ESCO) using an additional inductor [116, 48]. Then, I explore the usage of ESCO
as a low startup voltage oscillator for FM-backscatter. Finally, I demonstrate our variant of
ESCO, a Modified Clapp Oscillator (MCO). I explain its hardware modifications that allow
for its startup voltage to be similar to ESCO but also enable lower current consumption (<2
µW) with similar SNR of the communicated unique frequency.
A conventional Colpitts oscillator (Figure 4.15a) is a type of LC oscillator consisting of
a transistor-based amplifier and a feedback network consisting of an inductor (L1 ) and two
capacitors (C1 , C2 ). The oscillation frequency is determined by the resonant frequency of
the LC tank, formed by the inductor (L1 ) connected in parallel to the two capacitors (C1 ,
C2 ) connected in series. The frequency can be calculated as F = √1 .
2π L1 Ceq
To achieve higher amplitude and lower startup voltage, the traditional Colpitts
oscillator can be modified from its traditional architecture by changing or adding
components. The first component is a zero-Vth MOSFET (e.g., the ALD110800A [36]),
101
which can replace the traditional MOSFET to lower the startup voltage of the oscillator as
discussed in subsection 4.11.1 [116].
Figure 4.15: Steps of building a Modified Clapp Oscillator (MCO): a. Start with the conventional
Colpitts oscillator configuration b. Replace the MOSFET with a zero-Vth MOSFET and replace
the Colpitts current source with a degenerative current source (an inductor) to build the Enhanced
Swing Colpitts Oscillator (ESCO) c. Complete ESCO drain-output RF-backscatter circuit
The second change is to use the Enhanced Swing Colpitts Oscillator (ESCO) config-
uration [48] (Figure 4.15b), where an inductor, used as a degenerative current source,
is used at the source of the transistor to replace the traditional current source (a resistor)
[6]. This technique allows us to reduce power consumption since the inductor only uses
reactive power, which can be fed back into the primary tank. Additionally, using an induc-
tor allows our primary tank to swing below zero at the source, thus giving a better output
range. Albeit this secondary inductor creates a secondary LC tank with C2, I prevent this
secondary tank from affecting our primary tank’s oscillation by designing the secondary
tank’s startup voltage to be much higher than our range of operation.
I tested an ESCO-based backscatter tag tuned at different frequencies ranging from
100kHz-1000kHz (Table 4.1). The ESCO drain output is fed into Cblocking , which removes
the DC bias before input into the JFET. The JFET changes the impedance connecting the
antenna to the ground using the voltage changes at its gate and effectively communicates
102
frequency-modulated information (explained in subsubsection 4.2). Parasitic capacitances
from transistors and traces can affect the oscillation frequency. Accounting for parasitic
capacitances, I perform a detailed AC analysis of the circuit (Figure 4.15b in Appendix A),
which shows the complete LC tank. Using CJF ET to denote the effective capacitance
formed by the parasitic capacitance network caused by the JFET and the antenna, CJF ET
and Cblocking are connected in series and then connected in parallel with the inductor. Since
CJF ET ¡¡Cblocking , where CJF ET has a small value in the several to 10s of pF range, it
becomes the primary effective capacitance which gets added to the LC tank. Other parasitic
capacitances from the zero-Vth MOSFET and traces are also very small (several pF). Thus,
they do not affect the oscillation frequency to a large extent. The formula for the frequency
of the ESCO is the same as the conventional Colpitts. I use fcalc to denote calculated
frequency using Cef f and L1 in Table 4.1. Table 4.1 shows details of different fmeasured
frequency setups which I built near frequency value f. I powered the ESCO-based MARS
tag setup (Figure 4.15c) using a power supply and report startup voltage (Vin ), current (Iin ),
power (P) and SNR for back-scattering a single unique frequency. I also further calculated
the number of photodiodes that would be required to run the backscatter setup if used in
the lighting of 500 lux (the final two photodiode systems worked well in practice in our
laboratory and cafe).
Figure 4.16: Steps of building MCO (Continued): a. ESCO with gate output with the addition
of Radjust b. Modified Collpits Oscillator (MCO) with frequency dependent on L1, C1, C2, C0 c.
MCO is a type of Clapp oscillator d. MCO with all three terminals oscillating. The gate oscillates
to create a negative feedback loop.
103
Table 4.1: Power consumption and backscatter signal SNR for different ESCO with drain output
configuration (PD- Photodiode)
100 106 105.144 270 17.05 4603 15 12 4.7 10 1000 1000 100 487.5
200 203 202.062 110 3.5 385 21 1 4.7 10 220 220 100 132
300 293 310.781 100 2.95 295 17 1 4.7 10 47 100 100 55.8
400 414 414.953 170 6.8 1156 16 3 4.7 10 10 47 100 31
500 502 508.402 240 11.64 2793 15 8 2 10 47 100 100 48.3
600 617 642.297 220 11.1 2442 22 6 2 10 10 47 1 30.7
700 689 672.552 200 9.6 1920 27 5 1 1 47 94 100 54.62
800 820 838.820 250 14.12 3530 24 10 1 10 20 47 1 36
900 925 954.840 380 33.1 12578 20 30 0.3 1 100 220 100 92.61
1000 999 109.0795 650 143.2 93080 27 220 0.6 1 100 30 100 33.03
While ESCO has a low startup voltage, the current is still on the higher side > 3uA.
Thus, I experimented with the addition of a third component, Radjust , a current limiting
resistor to prevent the MOSFET from drawing excess current from the power harvester
(Figure 4.16a). This resistor also comes with the side effect of creating an RC tank with
the effective capacitance of the primary tank. However, I prevent this from affecting the
primary tank’s oscillation by making the resistor value high enough to act as an impedance.
The oscillations in the primary tank will then limit their flow through the high-impedance
resistor, and enough current will flow into the lower-impedance capacitors to continue os-
cillations. Unfortunately, it comes with the side effect of increasing the startup voltage for
104
drain oscillations significantly.
I then changed the circuit a fourth time by adding the output to the gate of the Zero-Vth
MOSFET in the ESCO setup (Figure 4.16b). I have found the gate to have a lower startup
voltage requirement than the drain. Thus, despite Radjust limiting current and increasing
the startup voltage for the drain, I can utilize the gate output with a lower supply voltage.
An optional Cshif t is added to the gate of the JFET to modify the oscillation frequency. A
subtle side effect of adding Rlimit and using the gate as the output is that the effective capac-
itance, denoted as C0 , of the capacitor network formed by the parasitic capacitance of JFET
(CJF ET ), Cblocking , and optionally Cshif t appear as connected in series with the inductor L1 ,
thus changing the circuit into a Clapp oscillator-like configuration (Figure 4.16c). The de-
tailed configuration of the exact LC tank considering parasitic reactances in the circuit is
shown in Appendix A. I call this configuration the Modified Clapp Oscillator (MCO). The
oscillation frequency is still the resonance frequency of the LC tank. Ignoring other para-
sitic reactances that have smaller effects on the oscillation, the LC tank is formed by C0 ,
C1 , and C2 connected in series and then connected in parallel with L1 .
Because of the reciprocal rule for adding capacitors connected in series, the small CJF ET
value dominates C0 , and C0 , having a smaller value than C1 and C2 , dominates Ceq , making
the oscillation frequency very sensitive to the change of C0 . As a result, modifying the value
for the optional Cshif t , which is in parallel with CJF ET , can drastically shift the oscillation
frequency.
Note that the behavior of the MCO is different from conventional common gate Colpitts
and ESCO, where the drain oscillation is an amplified output of the source. In an MCO,
all three terminals oscillate. The inductor L1 offers an opposite phase output signal into
the gate node, creating negative feedback (Figure 4.16d). Increasing Radjust leads to an
increase in gate oscillation, a decrease in drain oscillation, and a decrease in overall power
consumption. Experimentally, I found that using the gate as output results in a lower start-
up voltage for the oscillator compared to using the drain as output, making the gate output
105
a better configuration. I found that it is possible to tune Radjust for different frequency
setups, so the oscillation output at the gate is around +/- 200mV, which is the ideal operating
range for the JFET (green region Figure 4.17), where the optimal SNR is produced without
generating harmonics.
I performed a similar experiment as I did for the ESCO to measure SNR. As shown in
Table 4.2, I measured that for all target frequencies from 100kHz to 1MHz, the MCO con-
figuration can be powered from 2 photodiodes under 600 lux while generating similar SNR
compared to the ESCO configuration. For the MCO experiment, fcalc is omitted because,
in practice, the expected oscillation frequency is harder to calculate as the oscillation fre-
quency depends on CJF ET , which I could not measure. Additionally, our simulation shows
an oscillating gate also creates a constantly changing transistor capacitance for the Zero-
Vth MOSFET due to the Miller effect [120], affecting the oscillation frequency.
Table 4.2: Power consumption and signal SNR for different MCO gate output configuration
(∗ connected in series, PD- Photodiode)
100 103 450 2.1 945 15 2∗ 200 29.4 1 100 220 100 30
200 202 450 2.02 909 19 2∗ 200 24.7 4.7 20 100 100 n/a
300 303 350 1.3 455 16 2∗ 68 14.7 10 47 47 100 n/a
400 389 270 0.7 189 18 2∗ 82 10 10 20 47 100 n/a
500 502 200 2.15 430 18 1 56 5.7 1 20 100 100 n/a
600 602 250 0.5 125 15 1 84 4.7 10 10 10 1 n/a
700 686 620 1.4 868 17 2∗ 380 4.7 10 47 47 10 47
800 806 440 1.64 721.6 15 2∗ 220 4.7 10 20 30 10 47
900 933 330 1.8 594 23 2∗ 39 2 4.7 10 10 10 n/a
1000 1016 460 2.13 979.8 17 2∗ 220 2 4.7 47 47 10 220
106
In conclusion, by leveraging the parasitic reactance of the system, utilizing the gate node
as the output, employing Radjust to limit current, and using an inductor as a degenerative
current source, I am able to modify the classic Colpitts oscillator into the MCO, which
satisfies the four design requirements of a desired oscillator discussed at the beginning of
this subsection.
RF-analog Switch
In the next stage, the oscillator output is fed into an RF switch which is responsible for
changing its impedance with changing voltage. To maintain the versatility of the signal
being sensed and low-power consumption, I ideally want the switch to satisfy the following
design parameters:
1. The switch should be able to have linear impedance changes with voltage input to be
able to communicate different types of sensed information.
2. Preferably not require constant Vdd for operation to lower the power consumption.
This preference reduces requirements for power management to handle voltage mis-
matches between the power harvester and the RF switch. Additionally, it also reduces
quiescent current wastage.
3. For a given signal to create impedance changes, the corresponding current consump-
tion should be as low as possible.
In our system, I use a JFET (MPF-102) as an RF switch in the common-source configura-
tion in the MARS tag (Figure 4.16b). It does not require a constant Vdd like the commonly
used ADG-902 RF switch IC in similar works [186, 152]. Figure 4.4 shows the relationship
between IGS , VGS and RDS of the JFET measured using an Agilent E5272A. The power
consumption for the impedance modulation is ideal for a low-power system. As the gate
output of the MCO goes through Cblocking and inputs into the JFET an AC signal with a
range of ∼ +/- 200mV (the green region in Figure 4.4), the current consumed by JFET is
107
Figure 4.17: Characteristics of the RF-switch JFET: Voltage versus current and resistance graph.
The region marked in green is the ideal operation region for JFET where the power consumption is
low and impedance changes are linear.
100’s of nA, which keeps the total power consumed low. In addition, in this voltage range,
the JFET operates in the ohmic/ triode region (before Vpinchof f = -1.9V) and thus will have
a desirable linear change in resistance (orange) in response to the sine wave-like signal
from the MCO, resulting in few harmonics in the frequency shifted backscatter signal.
Antenna
The carrier frequency used by the transmitter and the receiver determines the frequency of
the antenna used (e.g., 915MHz) in our current setup. I further performed antenna selection
based on the trade-offs between antenna gain, form factor, and commercial availability. I
use a flexible linear polarization patch antenna (Taoglas FXP290.07.0100A) tuned to the
UHF band with 1.5 dBi gain, resulting in a 3.5x4.5x0.1 cm3 size in the MARS tag. While
the uniformly polarized commercial whip antenna with 4 dBi gain is more robust to di-
rectionality and has a higher signal-to-noise ratio (SNR), it is too thick and big (16x0.5
x1.2 cm3 ) to be included in MARS tags. To optimize antenna size further, in the future, a
custom-printed 915 MHz dipole or PIFA antenna can be included in the prototype, which
would be much smaller in area than the current commercial one in use. Another alterna-
tive is to use commercially available chip antennas (e.g., the ACAG1204-915-T) with a
108
reasonably high gain of 3.2 dbi and a smaller form factor 1.2 x 0.4 x 0.16 cm3 .
The MARS communication block interfaces with multiple analog sensors by modifying
different electrical properties of C1, C2, and L1 that control the oscillator’s frequency. I use
C0 to tune the oscillator frequency but not for sensing. Figure 4.18a shows the positions
of inductor and capacitor passives of the LC oscillator in blue, which can be modified
directly by placing an analog sensor in its place or indirectly by placing it in parallel or
series. Below I will discuss three different approaches to add sensing to the MARS tags,
using inductance, capacitance, or self-ed voltage-generating sensors to control the tag’s
oscillation.
Figure 4.18: Overview of analog sensing modalities: a. Analog sensors can be placed in parallel
with the highlighted components or can replace the highlighted components entirely. b. Inductive-
based sensors can be added in parallel with L1 c. Capacitive sensors can be placed in parallel with
or replace C1 or C2. d. Sensors that generate voltage can be placed in parallel to a varactor which
converts the voltage changes to capacitance changes. This combination can be used to replace
C1/C2 in the oscillator.
Inductor-controlled Oscillator
I can place an inductive analog sensor in parallel/series with the inductive passive (L1).
Thus, any change in the inductance value of the sensor would also change the effective L1
of the oscillator system, effectively changing the frequency of the oscillator (Figure 4.18b).
109
I leverage this phenomenon to develop an inductive multi-button-touch game controller
(Figure 4.25).
Capacitor-controlled Oscillator
I can use an analog variable capacitance-based sensor to modify the effective capacitance
in the LC tank of the oscillator (Figure 4.18c). The sensitivity and frequency modulation
range (channel width) of the system can be tuned by adding the sensor into the capacitor
network (C0 , C1 , and C2 ) at different nodes. The low capacitance of both C1 and C2 (in
pF) gives an extensive range of capacitance sensors. I use the change in pF of capacitance
for a swipe-based touch sensor for sensing direction and ID (Figure 4.22 and Figure 4.24).
Voltage-controlled Oscillator
Several self-powered sensors produce voltage changes from the phenomenon they sense
(e.g., a photodiode from light, a piezoelectric or triboelectric generator from mechanical
vibrations). Voltage from such self-powered sensors can be fed into a variable capaci-
tor diode to produce corresponding capacitor changes (Figure 4.18d). When tuned to the
proper range, this new combined self-powered sensor and variable capacitor diode can
replace either capacitor (C1/C2) in the oscillator circuit. I have leveraged this setup to
communicate speech using the MARS tag (Figure 4.21).
In conclusion, leveraging a Modified Clapp Oscillator (MCO) (subsubsection 4.11.2), I
have been able to achieve analog sensing of inductance, capacitance, and voltage, in <
1µW with simple circuitry (2 active and 7 passive components). Low startup voltage and
low power allow MARS tags to be powered by 2 photodiodes in indoor ambient light or
with the touch of a finger on a thermoelectric generator.
110
4.12 MARS Tag Power Harvesting
For the MARS system, both the startup supply voltage (∼ 500 mV ) and the current (∼
2 µA) are relatively low for different oscillator setups (detailed later in Table 4.2) than
conventional systems. This low supply voltage and current allow for different DC power
harvesting methods, such as ambient light and body heat, to power the MARS tags, elim-
inating the need for complex circuitry or power management. Below, I discuss the two
power harvesting techniques I have so far explored.
Figure 4.19: Power harvester characterizations: a. The voltage and current produced by two photo-
diodes in a series electrical configuration. b. The voltage and current produced by a thermoelectric
cell pressed for 1, 2, and 3 seconds in a room at 73◦ F .
I employ 2 Vishay BPW34 photodiodes connected in series to power MARS tag reliably
in different frequency configurations (Table 4.2). Vishay BPW34 is a PIN photodiode with
high speed and high radiant sensitivity in a miniature, flat, top-view, clear plastic pack-
age. It is sensitive to visible and near-infrared radiation. The two photodiodes connected
together have a radiant sensitive area of 7.5 mm2 .
Lighting in a general office environment is about 500-1000 lux. I change the light intensity
of the incandescent light bulb up to 60 W and, in series, measure light intensity with a
lux meter for these experiments. The voltage and current readings were done across 56
111
Kohm resistance, similar to Radjust . Figure 4.19a demonstrates the voltage and current
characterization of two photodiodes. The power produced by 15 mm2 of photodiode area
produces 1µW plus power in ambient room light.
The area for the photodiodes helps in keeping cost minimal. I do not employ solar cells
since they are generally available in the 10cm2 range and optimized differently in their
design.
Thermoelectric power generation varies with many factors (e.g., room temperature, body
temp, area of contact, the pressure of contact). I conservatively reported (Figure 4.19b)
voltage and current from the thermoelectric generator (Perpetua Technologies) across a
100 Kohm resistor when an adult’s thumb with a body temperature of 97.8 °F touches
for 1s, 5s, and 10s in a room at 73°F. The peaks of 1s touch curve at 0.4V and 2.5 µA
demonstrate that the power generated is sufficient to power a MARS tag (Table 4.2) with
the touch of a human finger.
In this section, I discuss interaction-specific circuit design changes in MARS tags. I start
by detailing our experimental setup, and then I explain the common part of the transceiver
pipeline, which all interactions share. Next, I provide the details of several interactions that
MARS tags can support (Table 4.3). For each interaction, I demonstrate a built prototype,
the real-time signal I record by interacting with it, the transceiver-specific pipeline for
processing it, and finally, the applications MARS tags can support.
Our system consists of three main components – the transmitter (TX), MARS tags, and the
receiver (RX). I use an Ettus Research N210 USRP with a UBX-40 USRP daughterboard
112
Table 4.3: Prototype applications: Summary of input gestures that can be detected using MARS
and corresponding details of the prototype and the demo application.
Figure 4.20: FM transceiver pipeline for audio: The three signal processing steps displayed here
create the pipeline which is used for: a. Receiving the backscattered signals b. Shifting backscatter
signals to 0 Hz where they can then be c. Demodulated into human perceptible audio data
I leverage the GNU Radio Companion (GRC) software suite to create a processing pipeline
for each of our four interactions. Each GRC flow graph is optimized for a particular inter-
action modality, but they all share a similar set of initial digital signal processing steps. I
will explain them with an example of audio data (Figure 4.20). First, the data is collected at
1e6 Hz and a low pass filter is applied (Figure 4.20b). Second, a frequency translation step
is applied to shift the backscattered signal down to 0 Hz offset, e.g., from 600 KHz to 0 Hz
113
(Figure 4.20b). Finally, for frequency-modulated (FM) signals like audio, I utilize a Wide
Band Frequency Modulation (WBFM) demodulation block to separate the backscattered
signal from the backscattered carrier wave (Figure 4.20c).
4.13.3 Speech
Figure 4.21: MARS facilitated wireless speech transmission: a. Audio signals from the microphone
are utilized to change the capacitance of a varactor which modulates the oscillation frequency of the
backscattered signal. b. Conference program insert MARS wireless microphone, demonstration of
the MARS wireless microphone in a conference environment, and close up of the powered prototype
c. MARS wireless microphone attached to a foam pad for affixing to home appliances. Demon-
stration of utilizing the MARS wireless microphone, powered by a touch of a finger, to extend the
range of a smart home device.
I created a MARS tag that supports the wireless communication of audio data using our
previous work SATURN, a self-powered flexible microphone based on the principle of
triboelectric nanogenerator (TENG) [14] as shown in circuit in Figure 4.21a,c. The mi-
crophone is placed in parallel to a varactor (SMV1702-011LF), which changes the capac-
itance across its ends when the user speaks into the microphone. I also add extra blocking
caps Cr, so the microphone’s charge input is used just for changing the capacitance of the
114
varactor and does not affect the oscillation in the circuit. This microphone-varactor com-
bination could replace either C1 or C2, but since C2 gives less frequency shift compared
to C1, I put it across C2, where it was more stable. The processing pipeline followed
for audio was mentioned before in subsection 4.13.2. Audio in the tag takes a +/- 60kHz
frequency-modulated signal. Figure 4.21 B and C show example applications of MARS
audio communication. Figure 4.21b is powered by two photodiodes, while Figure 4.21c is
powered by a thermometric generator, as mentioned in section 4.12. Since both the MARS
tag and the microphone are thin, I envision that the MARS audio patches can be embedded
in a conference brochure and used as an extended microphone for central audio control.
Another application is at home, where users can touch and talk with the MARS tag, which
acts as an extended microphone for smart home devices. I characterize the quality of audio
with increasing distance based on PESQ score in section 4.14.
MARS can enable swipe-based direction control, leveraging capacitance sensing. The cir-
cuit exploits that the human body acts similarly to a bag of water, and any touch interaction
over a conductive surface directly or indirectly changes its capacitance. The details of the
fabrication of the swipe-based direction sensor are provided in section 3.10. Figure 4.22a
shows the prototype of a light dimmer, where the user increases or decreases light intensity
by swiping a finger over capacitive pads (Figure 4.22c) while the other finger remains on
the thermoelectric to power the circuit from body heat. Figure 4.22b shows the controller’s
circuit diagram and capacitative swiping interface, highlighted in green as a variable ca-
pacitor at the terminal between C1 and C2.
After developing the capacitative touch sensor and testing its capacitative changes, I next
integrate it with the MARS tag. Figure 4.22d shows the raw spectrogram from GNUradio
of a swipe right. I noticed that a small 2pF change in our circuit produces approximately 5
115
Figure 4.22: MARS facilitated capacitive dimmer demo: a. MARS swipe direction sensor de-
vice prototype. b. The capacitive touchpad acts as a variable capacitor to modify the oscillating
frequency of the MARS tag. c. The touchpad sensor is constructed out of copper tape and an
adhesive-backed polyamide sheet (Kapton). Various-size pads create varying amounts of electrical
capacitance. d. Three peaks from the touchpad are detectable on the GRC-based transceiver’s spec-
trogram E. The swipe output of the MARS tag was interfaced with a Phillips Hue light to act as a
dimmer.
kHz of frequency change. Thus, I can see in Figure 4.22d that a swipe right changed a tag
tuned to 345 kHz, first to 326 kHz (19 kHz shift)), 334 kHz (11 KHz shift), and 340 kHz (5
kHz shift), respectively. I passed this data through our offline python processing pipeline
of Fourier transform (FFT), argmax, smoothing (Figure 4.23), and linear regression. The
slope of this linear regression corresponds to the direction of the swipe. In the future, I will
evaluate building a real-time direction-sensing pipeline using this strategy.
Figure 4.23: Capacitive swipe sensor processing: a. The capacitance of each slider pad varies
slightly between participants. b. The signal processing steps for a left swipe: the spectrogram
is generated, the spectrogram is filtered into a 1D time series, and a linear regression is found
between the maximum and minimum points of the signal. c. The same signal processing pipeline is
demonstrated for a right swipe. Notably, the regression line’s slope is opposite of the left swipe.
116
4.13.5 Swipe Based Unique ID
Figure 4.24: MARS facilitated capacitive menu demo: a. The menu contains a MARS tag and
a corresponding capacitive ID sensor for each menu item. When the user swipes an ID sensor on
the menu, the signal is transmitted to a remote application for servers to manage the orders. b.
The ID sensors act as variable capacitors, which modify the oscillating frequency of the MARS tag
when they are touched. c. Pictured is an ID sensor with its capacitive pad values labeled along
with corresponding spectrograms, which illustrate the frequency changes in the backscatter signal
produced by touching the ID sensor.
I extend the idea of a capacitative swipe sensor to transmit ID information with just a finger
swipe over six capacitive teeth in a barcode. Adding these barcode IDs to MARS expands
the interaction space for one particular channel, opening possibilities for new applications.
Figure 4.24a shows a food menu with a MARS tag, which has four IDs co-located with each
food item. As the user opens the menu, the photodiode receives light and starts operating
the MARS tag, indicating table occupancy. When the user wants to order something, they
can do so by swiping on the ID next to the food item, which changes the MARS tag’s
frequency according to the physical ID swiped. Each restaurant table can have a menu with
a MARS tag tuned to a particular frequency.
Each of the four IDs is a variable capacitor and is attached to our original MARS circuit,
as shown in Figure 4.24b. Figure 4.24c shows a close-up view of one of the IDs, and the
spectrogramsćhange when the user swipes it. Over multiple finger-swipe runs on different
IDs from the menu, and with the MARS tag tuned at 345 KHz, I obtained peaks between
318-325 KHz for digit 3, 330-335 KHz for digit 2, and 339-341 KHz for digit 1.
In the future, I will employ template matching algorithms like dynamic time warping
117
(DTW) to create an online ID detection pipeline and real-time deployment of different
battery-free food menus for the restaurant scenario. Further, humans have different bulk
capacitance responses, which can shift the main tag frequency signal up and down with
a different amount. This can be handled at the receiver side by first detecting the shifted
baseband, i.e., where the peak of the signal is and then later demodulating the pattern re-
ceived.
Backscatter communication is governed by the Friss path loss equation; the power of the
radio wave reduces with distance. Thus, the closer the tag is to the transmitter, the stronger
the reflected power. For range experiments, the transmitter and receiver antennas were
118
Figure 4.25: Pong game controller demo: a. Paper game controller attached to a flexible MARS tag
b. Touch-sensitive buttons switch the MARS oscillator’s frequency by shorting a pair of inductors
c. The spectrogram shows the two controller carrier signals (at 289kHz and 349kHz) and their cor-
responding button press frequency shifts (303 and 314kHz for controller one and 366 and 379kHz
for controller two) d. The backscatter button signals are processed in GRC and are fed as control
inputs to Pong.
placed next to each other in a monostatic configuration and in line of sight with the tag.
The MARS tag was moved away from the antennas incrementally by 2 feet, and both SNR
and audio data were recorded. A MARS tag tuned to 350 KHz offset was chosen and
powered by a regulated DC power source of < 1µW to ensure reproducible measurements
with only the minimal power necessary to start the oscillator.
To calculate SNR, the recorded data was filtered using a moving average and plotted in a
spectrogram. Figure 4.26 shows changes in the SNR with distance. The maximum SNR
was measured as 45 dB at a distance of three feet. The steepest drop in SNR occurs between
distances of three and nine feet. At a distance of nine feet, the SNR is 38 dB. At the longest
distance tested (49 feet), the SNR was measured as 15 dB.
For calculating audio quality, I used a MARS tag with a flexible microphone placed next
to a speaker playing an audio file. The selected audio file was a composition of spoken
words commonly used in telecommunication audio experiments. To mimic the average
intensity of human speech sound, I configured the speaker to output in a range of 70-80
dBL. After the data was recorded through the MARS system, the PESQ score, a commonly
used measure of the quality of audio in telephony systems, was utilized [77]. A PESQ
audio score of four is the maximum and one is just understandable to human ears. After
some basic background noise filtering, our audio samples achieved a PESQ score of 2.45 at
119
3 feet. The score drops to 2 at approximately 9 feet and further reduces to 1.75 at 19 feet,
and 1.25 at 30 feet.
Figure 4.26: System level characterization: a) SNR versus distance b) Audio quality score versus
distance
Our experiments demonstrate that a bandwidth of 20 KHz is sufficient for detecting the
state of 4 touch points, single swipe direction, or single ID. This result suggests that 30 such
tags can be placed within a sensing range of 40 ft/12m (monostatic). An audio transmission
scoring PESQ 2 or higher requires +/- 60 KHz bandwidth (120 kHz total bandwidth). Even
under the maximum bandwidth of +/-60 kHz for audio, our system can support ∼ 8 tags (1
MHz / 120 kHz) within the 30ft / 9m area of sensing (monostatic configuration). Detecting
simultaneous tags is computationally intensive; the efficiency of the receiver processing
pipeline will be addressed in future work.
4.15 Comparative Case Study: Amazon Dash button, RF-band-aid, and MARS
In this section, I will provide a case study comparing MARS to a commercial prod-
uct (Amazon Dash) and the closest project in the literature with similar capabilities (RF
Bandaid) [152].
To better ground this MARS’s contributions toward the goal of interface stickers, I first
calculate the amount of power available in an office environment. I then compare what
120
energy harvesters are needed to power MARS compared to current IoT devices and recent
devices reported in the literature.
Table 4.4: Comparison between different wireless communication systems that could support
sound, swipe, identity, and touch sensing
Parts Cost ∼$1.4 $9.5; $20 w/ solar; $24 w/ >$10; $310 w/ harvester;
thin-film battery >$11.50 w/ battery
Battery n/a $15, 15 hr life (thin film) $1.50 in bulk; 7.6g; 3.8cm3
ST Microelectronics 1.8Whr AAA
EFL1K0AF39 Energizer lithium
While illumination should be above 500 lux in an office, an interface sticker could be placed
so it is not directly facing the illumination source. To take this factor into account, I assume
that the interface sticker will have at least 200 lux available, which is the value used in data
121
tables for small solar cells used for powering calculators and other small indoor devices
[15]. The total amount of power available in an office equipped with fluorescent lights
and illuminated at 200-500 lux is then between 0.33 mW/cm2 and 0.83 W/cm2 (LED
0.22 - 0.55 mW/cm2 ; incandescent 1.3-3.3 mW/cm2 ). Energy conversion for typical
small indoor amorphous silicon solar cells is currently around 9% [59]. However, previous
efforts such as the RF Bandaid also require energy storage, power management circuitry,
and, in some cases, boost converters to achieve the required start-up voltage. As reported,
the effective power harvested on the inside of a window was 8 µW/cm2 and a much lower
0.7 µW/cm2 on a desk [68] (suggesting a harvester surface area need of at least 50 cm2 ).
Body heat from the finger or hand is another source of energy readily available for inter-
action applications. A fingertip of 3cm2 at 97.8 °F has 15 mW of power available through
heat flow (50 W/m2 for body [172]). Thermoelectric generators have 0.2-0.8% efficiency
[103, 174] for heat conversion from the body at room temperature (∼20 °C) and thus can
generate 30 µW from finger touch. This level of power is reasonable for MARS, especially
with low voltages, but not for RF Bandaid or Amazon Dash.
I compare MARS to the commercial Amazon Dash product which has the capability of
touch (one button but has the internal circuitry to sense multiple capacitive buttons), sound
(MEMS microphone), and active radio transmission. The RF Bandaid [152] provides a
state-of-the-art comparison to a system in the literature most similar to MARS. The RF
Bandaid includes a microphone and senses changes in a force-sensing resistor and chest
strap. It also maps sensor output to frequency modulation for wireless backscatter trans-
mission.
Table 4.4 demonstrates the benefit of MARS’s extremely low energy operation in compar-
ison to other devices with respect to parts cost and the size of the solar harvester needed to
power the system. While Dash’s battery should last 59 years in sleep mode (3.45 µW), it
lasts only 6 hours if continuously transmitting a touch or sound (300 mW). Unlike MARS,
every second of event transmission reduces battery life in sleep mode by 1 day. Alter-
122
natively, a common solar harvester sufficient to register a button push or audio in indoor
lighting (200-500 lux) is impractical at a surface area of 10,000 cm2 and a cost of $300
[125]. Specifically, in order to match MARS’s ability for registering continuous button
interactions or sound, this large size is needed to produce the required max current of 300
mW. If the input was more occasional, a supercapacitor could be used to store harvested
energy until it was needed, and a less expensive harvester could be used. However, even
examining commercial products with solar harvesters that do not require continuous moni-
toring, such as the Logitech solar keyboard (35 cm2 ) and Phillips remote control (30 cm2 ),
shows that these devices require harvesters much larger than MARS (0.2 cm2 )[7].
Like MARS, the RF Bandaid can continuously monitor input, but it requires 160 µW of
power with a 2.6 V startup voltage. 45 cm2 of solar harvester is needed to meet these
specifications. MARS’s main advantage over RF Bandaid is its 100x lower power, which
allows a 200X smaller surface area harvester and 10X reduction in cost. However, another
advantage is the low startup voltage, which allows a better impedance match with low-
voltage power harvesters without the need for power management. This advantage reduces
the number of components in the system (9 components versus the RF Bandaid’s ∼ 200 or
the Amazon Dash’s millions), thus lowering cost, and opens the possibility of making the
entire system printable in the future, like wallpaper or house wrap [182, 1].
Like Dash or the RF Bandaid, MARS requires infrastructure, but “ubiquity” could be
achieved by embedding MARS receivers in the power outlets or light fixtures of a build-
ing and exploiting power line communication back to a central location, such as a smart
speaker. By reducing part cost, requiring a small surface area, and having a flat form factor,
MARS enables a vision of interfaces placed like stickers on surfaces in the environment; a
vision other methods have difficulty achieving.
123
4.16 MARS: Discussion, Limitations, and Future Work
While MARS interactive sticker circuit design could lead to IoT interfaces at significantly
less cost than current systems (subsection 4.16.2, section 4.15), they also hold some limi-
tations. Replication of MARS tag frequency is affected by the parasitic reactances and the
quality factor of the passive components (subsection 4.16.1). While the MARS sticker tags
are built using minimal components, there is more work to be done to make them print-
able (subsection 4.16.2). Several more strategies with respect to range (subsection 4.16.3),
placement of tags (subsection 4.16.4), power harvesting alternatives (subsection 4.16.5)
can be adopted to increase the robustness of a real-world deployment. Finally, further work
needs to be done to address user privacy concerns (subsection 4.16.6).
There are several sources of error that should be taken into account while building the
MARS tag. First, is the parasitic reactances, such as trace, transistor, and inductor capaci-
tance, that can all create a significant effect on the output frequency. Parasitics capacitances
near C0 of the MCO are emphasized since small changes in the values of C0 in series with
C1 and C2 will result in large changes in Ceq . Many of these parasitic reactances can be
circumvented in the manufacturing stage by the usage of tunable traces and passives to
achieve the desired frequency. Second, the quality factor of the passive components can
also determine the accuracy of the output frequency. Thus, higher quality passive compo-
nents should be considered when operating at higher frequencies, e.g., ceramic capacitors
made with C0G (NP0) material. Third, humidity and temperature may also play a role in
changing passive component values. However, in our indoor testing environment, there
have been no observable changes in the passive values due to changes in ambient humidity
and temperature between different rooms.
Additionally, there are further difficulties when attempting to calculate the expected fre-
124
quency of the MCO. First, the value of the capacitor network of C0 is hard to determine
because it is dependent on Cblocking , Cshif t , and CJF ET which are difficult to measure.
Second, the constantly changing gate output of the MCO results in a constantly changing
transistor capacitance for the zero-Vth MOSFET due to the Miller effect. Changes in Vdd
and Radjust can result in changes in the offset and range of the gate output oscillation, caus-
ing different effective transistor capacitance. Experiments and simulations have shown that
they can lead to a small change in the output frequency. As the value for Radjust , decreases,
the MCO will effectively turn into an ESCO configuration and a sudden jump in oscillation
can be observed.
With inexpensive, thin, and flexible interface circuits, one can imagine having interfaces in
a peelable book, similar to sticky notes, where an interface can be stuck to a wall, book, or
surface wherever it is needed. To work toward this vision, I have tried to adopt a thin form
factor when possible, such as a paper-based game controller and a post-it note-like wireless
microphone. Our photodiode-based prototypes are relatively close to the right form factor,
and recent research suggests that our actives, passives [126, 20] and thermoelectric gener-
ator prototypes could become thinner and more flexible in the near future [154]. While I
have shown some rectangular examples of a swipe-based sensor in subsection 4.13.6, the
technology enables a large design space (e.g., jog wheels) for exploring creative touch-
based nano-power wireless sensing on objects in the future.
Different strategies can be employed to increase the operational range. An obvious one
is to custom design higher gain antennas that are conformable to the size requirements
of a post-it note. The range is a function of the frequency at which the transmitter and
receiver operate. In the future, the operational range may be considerably increased by
operating in a lower frequency FCC free band (e.g., 35MHz rather than 900Mhz). Another
125
possibility for the increasing range is to build a Modified Clapp Oscillator tag in the 10s
of MHz for active transmission. Shifting from backscatter to active transmission increases
range but requires the additional overhead of a bigger power harvester, which may still be
practical for certain applications. Backscatter is often dependent on directionality and TX-
RX placement. To address this issue, I could use a bistatic TX-RX configuration where
there is a central omni-directional transmitter and multiple receivers placed strategically
around the desired region of operation. This strategy could be useful for larger spaces such
as an auditorium. Another way to increase SNR, and thus effectively the range, is to move
to an intermediate frequency like active radios and process the signal before demodulation.
The operation of each MARS tag and the SNR is affected by the direction of the antenna,
as well as when the user accidentally covers the majority of the antenna while interacting
with the tag. Interestingly, I did notice that touching the antenna partially can cause an
improvement in the SNR. In the future, rather than using a directional antenna, flexible
antennas with a uniform field may be employed to improve SNR and PESQ scores. Another
phenomenon that affects the data quality in the case of audio is the user’s closeness to the
flexible microphone patch. If the user gets too close to the flexible microphone patch, the
capacitance of the microphone varies the frequency.
Reducing the power to the nanowatt range from the microwatt range has a direct effect
on the type of power harvesters and power management circuits which can be employed
in our MARS system. For example, RF harvesting (e.g., 5G [42]) might be feasible for
perpetually running sensors. In the future, one could also imagine utilizing thinner/more
flexible versions of the harvesters, which usually tend to be lower in energy density (e.g.,
thermoelectric generators [154]).
126
4.16.6 Tangible Privacy
While the MARS system enables wireless sensing for a variety of input modalities, it also
opens research questions about privacy and the user’s perception and expectation of privacy.
One specific concern is that the technology can be embedded into everyday objects so as
to be invisible to the user [3]. In the future, I plan to conduct a user study for MARS,
specifically exploring the privacy aspects. How can I make changes in the MARS tag
design to communicate sensing intention? Are there interaction design patterns, such as
requiring the user to place their finger on the thermo-electric generator button before the
microphone works, that inform the user when the system is active?
In this Chapter, I introduced simple no/low-power circuitry for two types of analog
backscatter communication techniques. First, Zero Energy Ubiquitous Sound Sensing
Surface (ZEUSSS) which uses minimal electronic components and extends the original
self-sustaining SATURN microphone [14] with a printed, flexible antenna to enable passive
audio communication via amplitude-modulated analog backscatter. Second, Multi-channel
Ambiently-powered Realtime Sensing (MARS) leverages frequency-shifted/modulated
analog backscatter circuitry, which pushes beyond ZEUSSS for multiple tags operating
together and richness of sensors beyond just SATURN. MARS allows sub µW sensing tags
in flat form factors that can be incorporated onto surfaces such as walls and books or on
game controllers. Using the changing capacitive, inductive, or voltage properties of sen-
sors allows direct and highly efficient control of the tag’s oscillator for the transmission of
sensor data. Example applications include battery-free button-based game controllers, slid-
ers for lighting level control, sliders for identification and selection of items from a menu,
and remote microphones for auditoriums or smart home speaker systems. With further re-
search, I expect many more interfaces can be developed, and, one day, adding controls to a
127
“smart” environment will be as simple as placing stickers wherever an interface is needed.
Building ZEUSSS and MARS lends itself towards broad leanings that can be used when
designing sustainable wireless communication systems :
Chapter 5 details how the sustainable interactive wireless stickers can be extended to in-
clude display functionality for enhanced usability. It also goes into explores the selection
128
of the display technology, device fabrication, working, and systems that can be built lever-
aging it.
129
CHAPTER 5
FEEDBACK: LOW START-UP VOLTAGE AND LOW POWER DISPLAY
5.1 Introduction
Figure 5.1: VENUS display: a. Theory of operation is based on the phenomenon of elec-
trochromism that results in a change of color due to redox reaction when voltage is applied b.
VENUS display as part of a greeting card that changes color on touching
In the previous chapters, I have made iterative additions to the functionality of the sustain-
able interactive wireless stickers while maintaining the system design parameters of power,
cost, and form factor. I first added sensing (chapter 3) with SATURN and other flexible
sensors, followed by wireless communication (chapter 4) with ZEUSSS and MARS. In
this chapter, I look deeply at how visual feedback can be provided by interactive stickers.
Relaying to the user when the interactive sticker is in active operation or communicating
is a piece of important information for the user, both from a usability and privacy stand-
point. One way to provide feedback to the user is the addition of a visual display. A display
suitable for sustainable interactive wireless stickers should maintain characteristics in line
with the power, form factor, and cost system constraints; yet, it should be functionally
sophisticated enough to provide visual feedback. I describe these characteristics in detail
130
below.
1. Functionality
(a) Vibrant color changes: Usually, with low power, the functionality of the de-
vice is expected to reduce, yet the minimal functionality that is needed for sus-
tainable stickers is that of an on/off indicator.
(b) Reasonable switching speed: The display should support a change of state in
sync with the human speed of actions. Given that the average time taken to
perform micro-interactions is 1-3 seconds [17], the display can take roughly
1 second to switch from one state to another and does not need to have an
extremely fast switching speed, which is often power consuming.
2. Power
(a) Low startup voltage (∼0.5V) and power (10 µW) of operation: Display
should be able to operate in a sustainable system where power is limited and
often only available in short bursts at low voltages due to the inherent nature of
the power harvesters.
3. Cost
(b) Simple fabrication process: The display should be easy to fabricate and be
designed with inexpensive base materials to allow for an overall cost-effective
prototyping and manufacturing process.
131
4. Form factor: The selection of display technology should support a thin, lightweight,
and preferably flexible form factor.
Based on the above guidelines and the discussion in subsection 2.4.4, where I compared
three alternatives for passive non-emissive display technologies that have been previously
employed for low-power systems, I will be exploring ECD in this chapter. ECDs have the
lowest on/off switching operation voltage among the three alternatives (LCD, EPD, ECD)
and are suitable for sustainable interactive systems.
I introduce Vibrant Electrochromic-Display with Nano-ITO for Ultrathin Self-sustainable-
system (VENUS), a new type of ECD display that is specifically optimized for providing
functionality with power, cost, and form factor constraints. It leverages the phenomenon
called electrochromism which is defined as the reversible color change induced by an elec-
trochemical redox reaction. This chapter makes the following contributions:
1. Device design and simple fabrication for VENUS display: I explain how VENUS’
device design is an improvement over the traditional ECD, allowing it to fulfill the
design requirements laid before, including vibrant color change and low power. We
further explore its simple fabrication process and unique switching on/off behavior.
In the rest of the chapter, I explain the theory of color change, device design, and working
of VENUS display in its various states – oxidized, reduced, and default (section 5.2). It is
followed by VENUS display’s fabrication process (section 5.4). Next, I explain modes and
132
types of VENUS displays which can help convey different types of information. I elucidate
powering strategies possible for the low-power display that are suitable for a display in a
sustainable system (section 5.6). Further, I design interactions to switch different modes of
VENUS display guided by the power harvester’s operating principle (section 5.7). Finally,
I explore application use cases for the display part of a self-powered interactive system
(section 5.8).
Figure 5.2: Theory of operation: An example of a conjugate polymer, ECP-Magenta, changes its
color by applying +/- V due to a redox reaction.
133
stability. It is light pink colored in the charge neutrality state and switches to a colorless
state upon electrochemical oxidation by applying positive voltage potential. On the op-
posite side, it turns brighter pink when electrochemically reduced further with a negative
voltage. Not all EC materials change state from colored to colorless; some change from
one color to another [201]. Such color changes in an electrochromic device can convey
useful information to the user through visual signals. Next, I will explore electrochromism
as part of the VENUS display device design, where the focus would be to achieve color
change with the design parameters of a display for a sustainable wireless sticker.
In this section, I will first explain the traditional basic ECD device design. Next, I will
explain how VENUS optimizes these parameters by modifying the counter electrode de-
sign with nano-ITO and working electrode with ECP-Magenta, thus opening avenues for
VENUS displays to be included in a self-sustainable system.
Figure 5.3: Selection of ECP-Magenta conjugate polymer as working electrode based on voltage
of onset and clear
134
A typical ECD operates as a rechargeable electrochemical cell, usually containing two
electrodes separated by a layer of electrolyte (Figure 5.3). The device components and
their functions are described in detail below:
• Counter electrode (CE): In a typical ECD device, the CE, called ion storage or
charge balancing layer, compensates (counters) the potential of the WE primarily by
double layer formation through a Faradaic process or a combination of both capaci-
tance and Faradaic processes. In particular, the charge balancing of the EC reaction
in a device is crucial for the stability and low-power consumption of the EC device.
This layer can be EC material that alternatively changes color with working electrode
or an optically transparent metal oxide like W O3 , T iO2 , V2 O5 .
The ECD device undergoes electrochemical oxidation and the other a reduction when a
voltage difference is applied between the electrodes; the gain/loss of electrons is counter-
balanced by the movement of cations and anions to the respective electrodes through the
electrolyte to maintain charge neutrality.
135
Design of High Surface Area Counter Electrode for Vibrant Low-power VENUS Display
For this, different strategies have been adopted. Figure 5.6a shows the use of another/same
EC material as WE (Bottom) as CE (Top). This strategy often lacks the vibrancy of color in
the device as both the +/- V application results in no color change if the same EC material
is used [164]. There is also work done where Figure 5.6b CE is deposited with EC material
to be a counter-image of WE electrode so that they both alternatively show color. The WE
is clear when the CE is colored, and the WE is colored when CE is clear [89]. It often
136
results in charge unbalancing between WE and CE due to differences in areas resulting
in poor device performance. Thus, to maintain vibrancy and charge balancing together,
ECD design with the horizontal placement of CE and WE laterally next to each other has
been explored (Figure 5.6c) [9]. Since the horizontal alternative takes a large size, vertical
ECD with transparent electrodes has been explored for high color vibrancy (Figure 5.6d)
[164]. The CE is typically made of a transparent conductive material such as ITO due to
its favorable optical properties. While this device functionally works with vibrant color
changes from colored to clear, it has low stability. ITO is a poor charge-storage material
and, thus, a non-ideal counter-electrode from an electrochemical point of view. Within a
few cycles, it results in overall device performance degradation mainly due to the fouling
of the ITO glass on CE as a result of gel electrolyte and ITO reductive decomposition.
Figure 5.5: Oxidation and reduction states of VENUS display with nano-ITO
To overcome this CE ITO degradation, the more efficient devices incorporate high-
surface-area counter electrodes with an extended number of electric double-layer sites,
as opposed to just the standard ITO-coated glass electrode [4] (Figure 5.5). It allows for
more stable electrode potentials even in the absence of a reference electrode and increases
electrochemical cycling stability. Previous work has leveraged nanostructural metal oxide
137
like, but they all have high-temperature annealing step and is sensitive to thickness changes
[188]. To push for simple fabrication while achieving the design constraints of vibrancy,
low-voltage, and low power with simple fabrication, I use mesoporous indium-tin-oxide
(nano-ITO) coated on standard ITO glass as a non-polarizable counter electrode material
for the VENUS EC display.
Nano-ITO serves as an optically and electrochemically inactive counter electrode and takes
advantage of the electrochemical double-layer capacitance afforded by the high surface
area of the nano-ITO (Figure 5.6e). The nano-ITO electrode has a sufficiently large sur-
face area that the charge capacity is high enough (i.e., the effective current density is low
enough) to counterbalance the charge passed by the EC polymer layer without undergoing
electrochemical oxidation or reduction during device operation. Nano-ITO is easy to solu-
tion process and easy to deposit at low temperatures. This results in an ECD device that
is vibrant in color change at low voltage/power but also thin and easy to fabricate by sim-
ple methods like blade coating. Next, we optimize the EC material for the CE-optimised
device design.
The selection of working electrode’s electrochromic material greatly dictates the volt-
age required for the redox reaction and, thus, the power at which the ECD device oper-
ates. Among these EC materials, conjugated polymers have shown high coloration ef-
ficiency, electron richness, fast response (∼200 mS), and solution processability[165].
Conjugate polymers are basically alternating polymers based on repeat units of 3,4-
ethylenedioxythiophene (EDOT), or 3,4-propylenedioxythiophene (ProDOT), in varying
combinations that tune steric interactions and the subsequent optical absorption for fine
color control.
I will be investigating three different types of conjugate polymers for their onset voltage
(Colored reduced state) and clear voltage (colorless oxidized state) with the transparent
138
Figure 5.6: Selection of ECP-Magenta conjugate polymer as working electrode based on Voltage
of onset and clear
nano-ITO as the counter electrode and select the EC polymer to be used for the VENUS
display. Each of these ECP has a transmittance of approximately 15% at λmax. They are
139
Given the ECP-Magenta has the lowest clear voltage and produces a maximum vibrant
color change in the lowest voltage moving forward, we would be using it for the VENUS
display. See section 5.5 for more details.
The fabrication of the VENUS display is done in three steps. First is the preparation of the
counter electrode. Second the preparation of the working electrode, and the final assembly
of the two with the help of a separator where the electrolyte is contained. I explain the steps
below:
Figure 5.7: Device design crosssection with CE, WE, electrolyte, and separator
Polyethylene terephthalate (PET) (blue in Figure 5.7) precoated with indium tin oxide
(ITO) (dark grey in Figure 5.7) purchased from YNvsibile is used as substrate. The sheets
140
are cut into appropriate rectangular squares and placed on a blade coater with a conduc-
tive side facing upward. 30 microL of solution-processed Indium tin oxide nanoparticles
(NPs) with a diameter of 50 nm (ITO-50) purchased from MilliporeSigma is poured using
a pipette onto the sheet (Figure 5.8a). The blade coater set to 300um thickness of run over
the sheet at speed 30, to obtain a thin shiny coating of nano-ITO (Figure 5.8b). The nano-
ITO coated ITO/PET sheet is then placed oven at 75-80 degrees C for 72 hours to dry the
nano-ITO to obtain the transparent counter electrode (Figure 5.8 c,d).
First, a mask is placed onto ITO/PET substrate sheet, and a 4 mg/mL toluene-based ECP-
Magenta polymer ink is air-spayed onto the sheet using an airbrush with an argon pressure
of approximately 20 psi (Figure 5.9). Next, the mask was removed, and optical absorbance
spectra of the ICP-magenta sprayed into thin films were acquired using an Agilent Cary
5000 UV-Vis-NIR spectrophotometer scanning from 300-1600 nm by placing the sheet
between two holders. The optical density of 0.8/transmittance of approximately 15% at λ
max is measured using the spectrometer.
After fabricating the CE and WE electrodes, I assemble them together. First, the working
electrode with EC material is cleaned using a q-tip with IPA as required into a specific
shape (Figure 5.10a). Next, an adhesive spacer (similar to double-sided tape) is cut in the
141
shape of a frame to fit the edges of the display, and it is placed on top of the ITO/PET sheet
coated with the ECP (Figure 5.10b). An adequate volume of Li+ electrolyte gel is applied
inside the spacer frame and spread evenly using a brush (Figure 5.10c). The UV-curable
electrolyte gel used is supplied by Ynvisible Interactive Inc. The adhesive tap on another
side of the separator is peeled off (Figure 5.10d), and the counter electrode is placed with
the nano-ITO side down onto the ECP-coated ITO/PET. Effectively the two PET sheets are
arranged with the electrolyte gel sandwiched in between (Figure 5.10e). Finally, the device
is placed in a UV chamber for 30 seconds (365 nm) to cure the electrolyte gel, the final
composition of the ECD is shown in (Figure 5.10f).
Figure 5.11: Shift of VENUS display from clear mode to onset mode with application +/- V square
pulse voltage
142
Given the transparent counter electrode, the overall color of the VENUS display is con-
trolled by the working electrode’s optical color, which in turn is governed by the voltage
being applied. Based on the theory of electrochromism discussed before in section 5.2 the
device shows the three different colors and thus 3 modes of operation with three different
voltage application: Vclear , Vof f (0V) and Vonset .
The VENUS display (8mm2 ) was switched with various pulse lengths of different volt-
ages by using a EG&G PAR273 potentiostat/galvanostat controlled by the Corrware soft-
ware. Increasing voltage potential is applied to VENUS display to find Vclear at 0.37V
with a transmittance of 20% after which there is no significant visual change in color (Fig-
ure 5.11).Voltage in the opposite direction (Vonset of -0.37V) results in the colored mode
with a transmittance of 80%. We repeated the same experiment for Vclear and Vof f at 0V,
which also shows the change in color or transmittance 60%, where devices change color
from light pink to clear (Figure 5.12). In general, Vclear is a function of the internal resis-
tance and capacitance of the device that changes with the size of the device, electrolyte,
and EC material.
Figure 5.12: Shift of VENUS display from clear mode to off mode with application +/0 Volt square
pulse voltage
Defining 3 modes versus 2 traditional modes of operation: Most research and appli-
143
cations of ECD consider only completely oxidized/clear mode and reduced/onset mode
because of the high color contrast during device operation. It is important to note that to
implement a feedback indicator it is not necessary to obtain maximum transmittance, but
just a reasonable one. In employing VENUS display, control circuits, and applications, I
will also consider the off mode with Vof f , 0 volt as it also has significantly different color
change when compared to a colorless state. This significantly optimizes the power but also
simplifies the control circuitry design constraints.
VENUS display is capable of providing different types of information to the user as de-
scribed below:
5.6.1 Color
Figure 5.13: Magenta to a clear change in VENUS display can be modified to magenta to color X
(e.g. green), by augmenting it with paper of color X.
Colors inherently have meanings/emotions attached to them. For example, red is consid-
ered a symbol for stop, and green is considered associated with go. The VENUS display
inherently goes from pink/magenta color to clear. We can take advantage of the clear state
to make the VENUS display go from magenta to any other color. The display can be further
augmented with a paper strip of that color below the working electrode. For example, Fig-
144
ure 5.13 shows color changes from initial magenta to colorless and then after the addition
of a green paper strip, it changes from dark magenta to green.
5.6.2 Shape
Shapes of different types are a powerful way to convey information. The working electrode
can be etched into different shapes or emojis to convey meaning. For example, a heart
shape conveys love and strategically is powered to create meaningful sustainable interaction
(Figure 5.14).
Figure 5.14: Different shapes can be etched out in the working electrode
5.6.3 Size
The size of the VENUS display is a critical factor that determines the power that will be
consumed by the display to switch from one mode to another. Size informs the amount of
EC materials on the working electrode that will undergo redox reaction and the ions stored
at the counter electrode. It also determines the internal resistances and capacitances of the
device. All these factors influence the ease of voltage and the mean power consumption for
clear mode. EG&G PAR273 potentiostat/galvanostat controlled by the Corrware software
to the voltage and current measurements.
We find the Vclear and Iclear of the square pulse for which VENUS display turns from
transparent to colored (delta ∼60% transmittance). See Figure 5.15. The lowest power is
achieved by 3.14 mm2 display which consumes 0.25 V and 20 µA, which is a 5 µW peak
power pulse for clarity. Next is a 20 mm2 device that consumes 30 µW while a 50 mm2
consumes 480µW peak power. Such devices can be reasonably used to produce different
145
Figure 5.15: Effect of increasing size on VENUS display
types of small info-graphs. Finally, a 100 mm2 device requires 0.6V at 1mA, which is 600
µW peak power. It can be used for more complex and big infographics.
Increasing the size increases both voltage and current, and thus the power. Specifically
looking at the trend for voltage, it rises slightly from a device that is 3mm2 to 100mm2 ,
but it more or less remains near the redox potential of the ECP-Magenta, which is ∼ 0.5 V.
The current has a more sharp rise where it rises from 20µA to 1 mA. Size governs the peak
power and voltage required, and thus it dictates which device is feasible to be operated
self-sustainably or not. I will be looking at different types of power harvesting and ways of
powering devices of different sizes.
Previously to create sustainable interactive stickers we have pushed for an overall sim-
ple circuitry. I will maintain the same design parameters for the VENUS display as well
as strategies for power harvesting techniques that fit with VENUS display with minimal
power management circuitry. I will specifically leverage the fact that VENUS display’s
Vclear is orders of magnitude lower than traditional displays, which means it can be easily
impedance matched with power harvesters as explored below:
146
5.7.1 Ambient Light
Traditionally ambient light can be harvested using a photodiode, which is a p-n junction
that produces current when it absorbs photons. As explored in MARS (chapter 4) before, in
ambient room light (300 lux), a single photodiode provides 250 mV and 1.5µA DC power.
Thus, given the VENUS display consumes ∼400 mV of Vclear startup voltage, we can
arrange two photodiodes in series to add up voltages to create more than sufficient power
source for the ECD display to turn from off mode (light pink at 0V) to clear (Figure 5.16).
This setup with a switch can reasonably operate the VENUS display of < 25 mm2 . For
bigger sizes, more photodiodes can be added in series.
Figure 5.16: Few photodiodes connected in series can produce enough DC power to change the
VENUS display from default to clear operational mode in ambient room light
A thermoelectric generator (TEG) can transform thermal energy directly into electric en-
ergy through the thermoelectric effect. Figure 5.17 shows the general structure of a TEG. It
consists of a heat exchanger, a thermoelectric module (TEM), and a heat sink. The hot side
absorbs the heat and transfers it to the TEM, which typically contains a number of pairs
of p- and n-type semiconductors connected electrically in series and thermally in parallel.
Charge carriers are electrons in doped n-type semiconductors, and holes in doped p-type
semiconductors. The cold side dissipates the additional heat from the TEM. A TEG is able
to transform the temperature difference into electrical energy through the Seebeck effect.
If one side of the TEM is hotter than the other, the electrons on the hot side have more
147
kinetic energy than those on the cold side. Therefore, the electrons travel faster from the
hot side to the cold side, such that the charge carriers diffuse away from the hot side which
leads to a buildup of charge carriers at one side as shown in Figure 5.17. The hot side of the
TEM will eventually be positively charged while the cold side is negatively charged. This
buildup of charge creates a voltage potential that can be superimposed with the increasing
number of pairs of p- and n-type semiconductors. Conversely, touching the TEG module on
the other side results in the cold and hot sides being exchanged producing opposite poten-
tial. Figure 5.17 demonstrates that touching side 1 of the TEG module results in a negative
potential across the load and touching side 2 with the hand results in positive voltage. It
is similar to the square pulse generation with which we tested the VENUS display. Since
VENUS displays have a low startup voltage they can be impedance matched with a TEG
without additional voltage regulators or doublers.
Figure 5.17: Touching TEG module with hand on side 1 produces negative potential and on side 2
produces positive potential. It is similar to the square pulse produced during to testing (Figure 5.11)
to control VENUS display to go from onset mode to clear mode.
Next, I design interactions for controlling the VENUS display that exploits our understand-
ing of how harvesters like photo-diodes or thermoelectric generators work to produce DC
148
potential.
The direction of the current generated by the Thermoelectric Generator (TEG) depends on
the heating side, which is opposite to the direction of the electron flow in the TEG. Exploit-
ing this phenomenon, a novel interaction between the TEG and the VENUS is designed.
When the user touches one side of TEG with their finger as the heating side, VENUS
(100mm2 ) will display color 1 (e.g., magenta) while flipping the TEG and touching the
other side, VENUS will display color 2 (e.g., green in the clear state) (Figure 5.18). In ad-
dition to changing the applied voltage by flipping the TEG, two TEGs of opposite polarity
placed connected in parallel can also be used to control the color change of the VENUS
display.
5.8.2 Ambient Light Powered Opening/Closing Box Interaction: Clear to Off Mode
In subsection 5.7.1 I demonstrated how the DC voltage of the photo-diode can be used to
power a display, and the switch required to control the shift between the clear to default Off
state. Here we describe how opening and closing a box can be synced as a switch for the
VENUS display. A box with a window is created in such a way that the VENUS display
149
Figure 5.19: Box with conductive frame window: When the box is open the VENUS display is
in a clear state with a green indicator. In the closed state, it turns to the default state due to short-
circuiting and discharge of charge by the window frame
is placed exactly in alignment with the window if the box was closed. The window has a
protruding frame that is made conductive with copper tape Figure 5.19.
When the box is open, the light on the photodiode placed inside the box produces voltage
enough to make the VENUS display go into clear mode, thus revealing the green paper
below. When the box is closed, there are two features are work together to make the
VENUS display go to default colored mode. First, the photo-diode stops producing voltage
to the display, but this is not enough since ECD has a memory effect due to which it acts
as a super-capacitor. To discharge this super-capacitor, we leverage the second feature,
which is the conductive window. When the box is closed the conductive window frame
touches the two electrodes of the VENUS display, causing a short and discharging of the
display. Thus, we have created a box design that controls switching on/off indicators with
its opening and closing.
Based on the device design, power harvesting strategies and interactions designed, I will
explore some potential applications for the VENUS display.
150
Figure 5.20: Hand touch powered interactive cards a. TEG connected to VENUS display embedded
in card b.VENUS display in onset and clear mode c. Interactions that change the state of VENUS
display with respect to card
The VENUS display can be used to augment interactive children’s books or greeting cards.
Figure 5.20 shows a heart-shaped VENUS display embedded with leads directly connected
to a TEG. In the default state, it is magenta in color, but based on the text of the card, the
self-powered playful interactions can be designed with a thermoelectric generator, which
changes the state of the VENUS display info-graphic from clear to default and back.
151
Figure 5.21: Employing VENUS as an on/off indicator for wireless communication. In the box
closed state, the VENUS display is in off or discharged mode, and MARS is not communicating. In
the box open state, the VENUS display is in ON in clear mode, and MARS is communicating
Figure 5.21 shows the prototype of a box where MARS circuitry is placed. The MARS
oscillator, SATURN, and antenna are placed on the top lid cover that opens up (marked
in orange in the circuit diagram), and the photodiode and VENUS display is placed at the
bottom cover lid (marked in blue in the circuit diagram). The photodiodes are arranged
in such a way that in the presence of light, they produce enough power to light up both
VENUS and MARS. Two photodiodes are in series with a current limiting resistor to be
able to operate VENUS ∼ 400 mV startup voltage, and five other photodiodes are in parallel
to power up the MARS circuitry ∼ 200 mV. The arrangement of photodiodes in this way
prevents the mismatch of operating voltage between VENUS and the rest of the circuitry,
as well as saves power management circuitry costs.
When the box is closed, the photodiodes do not produce enough power for MARS circuitry
to operate, and the VENUS display is in discharged Off state. It means that its wireless
communication is inactive, and the indicator is also in-sync with the state. In addition,
152
there is a physical disconnection that happens between the top and the bottom cover on
opening and closing of the cover as means for additional usable privacy features. When the
box is open, the physical connection between the top and bottom cover is completed, which
leads to power from the photodiodes being supplied to MARS circuitry and SATURN for
active communication. The power from the photodiodes also turns the VENUS display
clear, thus green, due to the addition of green paper at the bottom. Thus, box opening and
closing controls on/off of MARS wireless communication and VENUS display.
We have demonstrated the VENUS display to be an easily fabricated low-power and voltage
alternative vibrant on/off feedback mechanism for low/self-powered systems. Below we
discuss some limitations and future work where this work can be expanded:
The devices demonstrated some degradation when left in normal bright room light for a
number of weeks. It is probably due to the oxidation of electrolytes and lack of proper
encapsulation. Thus, there is more work and device characterization to be done for the
stability of the device over long periods in real-life practical scenarios.
153
Figure 5.22: Alternate DC power sources to explore
tive power sources opens doors for sustainable self-powered wireless sensing and feedback
applications in infrastructure monitoring, smart farming, and sensing for conservation.
Additionally, these DC power sources can be voltage multiplied using simple low-voltage
and power colpitts oscillator circuit with zero-threshold voltage transistor as built for
MARS. It leads to efficient power management with minimal silicon components.
In the future, printable flexible TEG can be fabricated to power VENUS instead of using a
commercial bulky TEG. Such flexible TEG can conform to the body yet ensures efficient
heat transfer. Figure 5.22b demonstrates how such TEG devices can be built by compactly
printing p-n type materials together [44].
The VENUS display is low-voltage and power and acts as a capacitor when uncharged
and as a short circuit when charged. Due to these unique properties, we can leverage
them to create a segmented display that switches on sequentially as the voltage supplied
increases. Figure 5.23a shows a circuit with 3 VENUS displays as capacitors in parallel
to each other with resistances that act as a voltage divider. The resistor values can be
154
chosen in such a way that we attain sequential switching on behavior. The R values will
be constrained under two parameters: (1) the time to charge/discharge an ECD and (2) the
total power consumption of the display. If the resistor values are very large, the total power
consumption of the display falls dramatically. However, due to the low current capacity of
the resistors, the ECD’s would struggle to turn on / off quickly enough to be practical. If the
resistor values are very small, the ECD’s respond quickly, but the total power consumption
of the display ends beyond the capabilities of the energy harvesting. With the creation of
the right RC circuit, we can create level indicators.
Figure 5.23: a) VENUS Arrays as level indicator b)Exploit memory effect of VENUS display for
creating self-powered timers
ECDs act as super-capacitors, which means that once they go from default off to clear
mode or default off to onset/colored mode, they do not immediately come back to the
default mode and slowly discharge. The charge stored shows up as optical memory. This
effect can be leveraged to produce visual timers. More work can be done to create systems
with appropriately characterized power sources and VENUS displays to create timers of
different duration. For example, a VENUS display of size 1x1 cm2 can keep charge saved
for a few hours.
This three-state memory behavior of VENUS can also be used to create optical memory
155
digital readouts for non-interactive computing elements. That is, in intermittent computing
systems, VENUS can help communicate the charge stored in the storage capacitor.
5.11 Conclusion
In this chapter, I first introduce the VENUS display, an electrochromic device design that
allows for low-voltage operation and vibrant colors. I further showcase how VENUS can
be fabricated using simple techniques and its different modes of working. Next, I explore
ambient light and finger heat as ways of powering VENUS and design interactions that
control power input in sync with different modes of operation. I leverage this to elucidate
potential applications of VENUS as a low-power on/off indicator for sustainable interactive
objects and stickers.
156
CHAPTER 6
DISCUSSION
In this chapter, I first summarise the learnings and observations from building sustainable
interactive wireless stickers that generalize to building sustainable objects and surfaces.
Then, I discuss emerging research themes and future work based on the projects presented
in this dissertation.
In this dissertation, I rethink strategies to build systems that take power, cost, and form
factors together as their main design criteria. I fabricate novel material devices and circuits
that self-sustainably support input, feedback, and communication functionality for indoor
surfaces in a sticky note form factor. I detail below some general design guidelines that can
be followed for each of the device components to make advances in building self-sustaining
interactive wireless stickers and materials:
1. Sensors
• Use of everyday materials that reduces cost and optimizes form factor: Sen-
sors should be built leveraging easily available materials like plastic and paper
with a simple fabrication and assembly process that results in an overall reduc-
tion in the cost of the system.
• Self-powered sensing: If the phenomenon being sensed also powers the sens-
ing task itself, it eliminates the need for explicit power in some sustainable
systems (e.g., ZEUSSS) or at the least removes the need for amplifiers.
157
the material. The sensor’s design can embed the information in the device ma-
terial itself during the fabrication process (e.g., identity-based sensors have pre-
coded IDs to reduce power requirements).
158
eral norm used by both digital and analog circuit designers today is to use tran-
sistors in the saturation region that makes the supply voltage generally in the
range of 2.1-3V. A trick to lower the supply voltage in the analog domain is
to use lower threshold voltage transistors. The maturity of such transistors in
the past few years can be exploited as a way to build circuits with an order of
magnitude lower power.
• Selection of materials that have low switching voltage and power: On/off
functionality is one of the basic operations in a display. Traditional display
technologies like e-ink and LCDs require switching energies of 10-100mJ/cm2 ,
which is too high to be controlled by a power harvester with limited power
management circuitry. Thus, for a sustainable system, it is imperative to look
for alternatives with low-switching power. Low-voltage is equally important to
keep the power management complexity low or null so that the power harvester
can be directly impedance matched with the display.
• Push for device design with simple fabrication: A device with a lower num-
ber of layers, easily available materials, and simple fabrication techniques like
blade coating results in an overall low cost of manufacturing.
159
lines that optimize power, form factor, and cost together:
160
6.2 Stepping into the Era of Sustainable Computational Materials and Objects
Picture a disposable face mask that measures health biomarkers, is powered by human
breath, and can easily disintegrate into recyclable and biodegradable parts (Figure 6.2).
Imagine an eco-friendly wireless sticky note placed on a refrigerator that is powered on-
demand by the heat of a human finger to create contextualized tasks like adding items to
a grocery list that can be dissolved in water after use. Visualize agricultural sensors that
161
self-generate power from electrically active microbial communities in the soil to wirelessly
communicate water content, pH, and forest fire activity for ecological health monitoring.
Figure 6.2: Example of a sustainable computational object of the future: A disposable face mask
that measures health biomarkers, is powered by human breath, and can easily disintegrate into
recyclable and biodegradable parts
In this section, I will first mention the limitations that interactive wireless stickers have
in terms of sustainability and future ways of improving them. Next, I will broadly talk
about some emerging research areas and thrusts for this vision of sustainable computational
materials.
Sustainability is complex and hard to measure and achieve. It is dependent on a set of com-
plex factors and requires deep thinking in the entire life cycle of the computational material.
Achieving sustainability in one part of the lifecycle does not mean that the computational
material has achieved sustainability overall. Additionally, focusing on sustainability results
in creating challenges in balancing tasks of higher computational functionality like privacy
and scalability. Below I discuss some of these issues with respect to the interactive wireless
stickers.
162
Need for Less-carbon Intensive Manufacturing and Disposal: Much of the focus of
this dissertation is on sustainability in terms of operation (Figure 6.1). Interactive stickers
are powered by the ambient environment resulting in the complete removal of batteries
from the system. It is a big step as batteries are often toxic to the environment and require
intense water and power to extract base materials and manufacture.
Designing for sustainability also requires thinking about the carbon footprint of the man-
ufacturing processes required for fabricating the devices and circuits as a whole. It also
needs consideration as to what would happen to the computational material/object at the
end of its lifecycle.
• Safe Disposal: Computational stickers open the ability to place computing devices
everywhere, however, it raises a new question: what will happen to them at the end
of their life? Even the most robust device will eventually become electronic waste
(e-waste). Global e-waste is the fastest-growing waste stream and will exceed 74
163
million metric tons (Mt) by 2030 [54]. E-waste includes hazardous materials like
heavy metals and flame retardants, with the potential to pollute groundwater if buried
and air when incinerated. Disposal also can pose environmental justice concerns for
the surrounding communities [135].
Figure 6.4: SATURN device design that considers the whole lifecycle including manufacturing and
disposal
Knowing the procedures involved in the disposal of a device is crucial once it has
served its purpose. It requires thinking about the device materials from the very
start when the device is designed and manufactured. For example, SATURN was
built using materials that are not necessarily environmentally friendly. Copper is a
heavy metal that is not good for the environment. Environmentally benign alterna-
tives for copper include metals like iron, magnesium, and aluminium. Other organic
compounds like PEDOT:PSS or carbon-based inks can also be employed. PTFE is
a fluorine-based plastic, which is very harmful to the environment. To replace the
dielectric plastic, plant or animal-based proteins like cellulose and silk fibroin should
be used (Figure 6.4).
Balancing Privacy with Power, Cost, and Usability: Interactive stickers allow for com-
municating human interactions like speech, touch, and movement on everyday surfaces.
164
Such capability opens up many applications, but it also makes understanding different
threat models of the technology built equally important. Privacy-aware designs often focus
on technological and social approaches [99]. With an additional need to balance sustain-
ability (power and cost) as a system design parameter, privacy becomes an even harder
problem. One cannot assume leveraging additional computational resources without af-
fecting sustainability. Thus, privacy becomes a sustainability issue. I discuss below some
implemented strategies in this dissertation and some ideas of an exploratory nature that
may be used to tackle this non-trivial problem. Many are inspired by “privacy by design
principles” [99].
• Consent: MARS stickers only switch on when a user intentionally puts a finger on
the thermoelectric generator to power the sticker. Such an “on-demand sensor” is
based on consent, if the user has an understanding that the operation is based on their
providing power through body heat.
• Notice with tangible privacy features: Features can be designed that support tan-
gible privacy that is noticeable by the user to build trust that the microphone is off.
Section 5.9.2 demonstrates a box that, on closing, disconnects power from the MARS
circuitry and, on opening, re-connects it for wireless communication. It also adds a
low-power VENUS display-based indicator that changes color with the box opening
and closing to provide visible notice to the user about when the communication is
active.
• Locality: Computational materials focus on local sensing, which may provide users
with a natural mental model of their range of operation. For example, SATURN
microphone patches could be constructed to have a limited sensing range for the
human voice which requires the user to be present near the interactive sticker.
• Low-power encryption circuits: One major concern with backscatter signals is that
a bystander with a strong enough transceiver can listen without the knowledge of the
165
user. One way to address the problem is to have low-power encryption circuits that
operate in hundreds of mV, similar to the MARS oscillator circuit.
• Product design that makes adversarial use difficult: Design interventions can be
created to make it difficult for the adversary to operate the device yet still allow an
authentic user to be able to operate it. Imagine having a physical key that fits like a
puzzle to activate the sticker. It also creates an interesting research question about
the utility versus intrusiveness of the privacy feature. Ease of intentional use needs
to be maintained.
Balancing the Need for Scalability With Power and Cost For sustainable interactive
wireless stickers built in this dissertation, a frequency shift keying approach is adopted to
allow multiple tags to communicate together. The tags are already hardcoded with the fre-
quency at which they will operate. Also, the interactive wireless stickers in this dissertation
are limited in the scale of operation to 30-50 stickers in a range of 10 m. Several strategies
and open questions are discussed below with respect to scalability.
• Product feature that allows easy base frequency tuning: The stickers support
FSK. It can be further tuned in the product design by introducing a physical knob
that can be controlled by the user, much like the old radios. This knob can also help
the user create a mapping as to which sticker is associated with which object.
• Need for sticker circuit design to be compatible with existing protocol: Backscat-
ter has been shown to be compatible with already existing network protocols like
WiFi and BLE [87]. Alternatively, rather than making tag circuitry complex, modu-
lation of the carrier from the infrastructure can also be done.
166
• Low-power sophisticated circuits that support varied frequency operations: To
expand the bandwidth of stickers, a more complex circuitry than MARS needs to be
built that allows for frequency shifting similar to a microcontroller yet still operate
in hundreds of mV. This is an open research question.
Scalability is essentially a power and cost argument. It governs how much computing
should be added at the tag side. An application-centric approach needs to be adopted to
decide how much computation should be shifted to the infrastructure and how much should
remain locally.
Figure 6.5: Building computational objects and surfaces with a sustainability-first approach re-
quires a highly interdisciplinary and iterative approach
167
systems, and application layers. Thus, for environmental sustainability, it is pressing
to select base materials that allow for the transience of the device, which, based on
the application requirements, can be tuned for operating for a certain period and then
recycled, reused, or self-degrade at the end of the lifecycle [58].
168
Sustainable Design and Applications
Figure 6.6: Playful paper elephant with ears as sensors as a computational object
169
electronic/non-electronic parts of a computational object. In the long run, such sus-
tainable products should inspire behavioral change toward sustainability in daily life.
User Experience (UX) should set the right expectation and manage frustration: The
reduced performance of transient energy-neutral computational objects relative to
state-of-the-art will require a shift in users’ expectations. Creative UX can be used
to manage this frustration but also yield more power from the environment.
• Democratization with creative material toolkits: There is a need for creating ma-
terial toolkits and processes that empower HCI researchers and designers to explore
sustainable computational materials as a functional medium and the application sce-
narios they may support. It requires pushing for a design-driven materials innovation
process in which designers are engaged in speculative thinking about the current
novel research-grade material devices and prototypes to make them more meaning-
ful, enjoyable, usable, and privacy-aware.
170
CHAPTER 7
CONCLUSION
Weiser’s vision of ubiquitous computing promises the proliferation of a trillion IoT devices
by the year 2035 [169]. It brings into focus the research question of how everyday objects
and surfaces can be augmented or manufactured with computational capability without los-
ing their form factor or affordance. Current smart objects (e.g., smart home control, smart
fridge) look distinctively different from artifacts of daily use, such as a sticky paper note,
a cup, or a chair. They have hard bulky bodies, use rigid structural elements with millions
of electronic components, and have a power budget that necessitates the maintenance of
a wired connection or a bulky toxic battery. Truly blurring the distinction between digital
and physical worlds requires balancing an artifact’s computational functionality with three
other factors: 1) power; 2) form factors that look and feel more like everyday objects; and
3) the cost of materials and ease of manufacturing (e.g., simple circuitry). Balancing all
four together is a hard system design challenge.
Taking these design constraints into account, in this dissertation work, I have focused on
the development of ’sustainable interactive wireless stickers’ that can perform functional
computing tasks like sensing (SATURN and other interaction sensors), communication
(ZEUSSS and MARS), and feedback (VENUS) in an ultra-low-power budget with ambi-
ently harvested energy. Moreover, sustainable stickers are envisioned to be manufactured
inexpensively and employ minimal electrical components. These self-sustainable (battery-
less) stickers can potentially be augmented onto physical objects and surfaces to support
interesting interaction, human health, and infrastructure monitoring applications.
In conclusion, this dissertation research focuses on building material devices and circuits
from the elementary parts and combining them into application-oriented systems in a fun-
damentally different way than traditional IoT system research, where the focus is on build-
171
ing things from off-the-shelf components. This work is meant to inspire a new direction
for computing - ubiquitous sustainable computational materials and objects [12, 1], where
sustainability in power, manufacturing, and cost are central to the design.
172
Appendices
APPENDIX A
MARS CIRCUIT BOARDS
174
Figure A.2: Gate output AC analysis of Modified Clapp Oscillator (MCO) based backscatter
175
REFERENCES
[3] I. Ahmad, R. Farzan, A. Kapadia, and A. J. Lee, “Tangible privacy: Towards user-
centric sensor designs for bystander privacy,” Proceedings of the ACM on Human-
Computer Interaction, vol. 4, no. CSCW2, pp. 1–28, 2020.
[6] P. Andreani, X. Wang, L. Vandi, and A. Fard, “A study of phase noise in colpitts
and lc-tank cmos oscillators,” IEEE Journal of Solid-State Circuits, vol. 40, no. 5,
pp. 1107–1118, 2005.
[8] Y. Arakawa, Y. Suzuki, and N. Kasagi, “Micro seismic power generator using elec-
tret polymer film,” Proc. PowerMEMS, vol. 187, no. 190, p. 17, 2004.
[9] A. A. Argun and J. R. Reynolds, “Line patterning for flexible and laterally con-
figured electrochromic devices,” Journal of Materials Chemistry, vol. 15, no. 18,
pp. 1793–1800, 2005.
[11] N. Arora and G. D. Abowd, “Zeusss: Zero energy ubiquitous sound sensing surface
leveraging triboelectric nanogenerator and analog backscatter communication,” in
The 31st Annual ACM Symposium on User Interface Software and Technology Ad-
junct Proceedings, 2018, pp. 81–83.
176
[12] N. Arora, T. Starner, and G. D. Abowd, “Saturn: An introduction to the internet of
materials,” Communications of the ACM, vol. 63, no. 12, pp. 92–99, 2020.
[13] N. Arora et al., “Mars: Nano-power battery-free wireless interfaces for touch, swipe
and speech input,” in The 34th Annual ACM Symposium on User Interface Software
and Technology, 2021, pp. 1305–1325.
[14] N. Arora et al., “Saturn: A thin and flexible self-powered microphone leveraging
triboelectric nanogenerator,” Proceedings of the ACM on Interactive, Mobile, Wear-
able and Ubiquitous Technologies, vol. 2, no. 2, pp. 1–28, 2018.
[18] S. R. Best and B. C. Kaanta, “A tutorial on the receiving and scattering properties
of antennas,” IEEE Antennas and Propagation Magazine, vol. 51, no. 5, pp. 26–37,
2009.
[20] Y. Bonnassieux et al., “The 2021 flexible and printed electronics roadmap,” Flexi-
ble and Printed Electronics, vol. 6, no. 2, p. 023 001, 2021.
[22] G. Brooker and J. Gomez, “Lev termen’s great seal bug analyzed,” IEEE Aerospace
and Electronic Systems Magazine, vol. 28, no. 11, pp. 4–11, 2013.
177
[25] J. B. Burr and J. Shott, “A 200 mv self-testing encoder/decoder using stanford
ultra-low-power cmos,” in Proceedings of IEEE International Solid-State Circuits
Conference-ISSCC’94, IEEE, 1994, pp. 84–85.
[26] A. Butler, S. Izadi, and S. Hodges, “Sidesight: Multi-” touch” interaction around
small devices,” in Proceedings of the 21st annual ACM symposium on User inter-
face software and technology, 2008, pp. 201–204.
[29] T. Choudhury and A. Pentland, “Sensing and modeling human networks using the
sociometer,” IEEE, 2003, p. 216.
[32] A. Curtiss et al., “Facebit: Smart face masks platform,” Proceedings of the ACM on
Interactive, Mobile, Wearable and Ubiquitous Technologies, vol. 5, no. 4, pp. 1–44,
2021.
[35] R. Dennard, F. Gaensslen, H.-N. Yu, V. Rideout, E. Bassous, and A. LeBlanc, “De-
sign of ion-implanted mosfet’s with very small physical dimensions,” IEEE Journal
of Solid-State Circuits, vol. 9, no. 5, pp. 256–268, 1974.
178
[39] K. Dong et al., “3d orthogonal woven triboelectric nanogenerator for effective
biomechanical energy harvesting and as self-powered active motion sensors,” Ad-
vanced Materials, vol. 29, no. 38, p. 1 702 648, 2017.
[40] P. Dourish, Where the action is. MIT press Cambridge, 2001.
[41] J. Eargle, The Microphone Book: From mono to stereo to surround-a guide to mi-
crophone design and application. CRC Press, 2012.
[42] A. Eid, J. G. Hester, and M. M. Tentzeris, “5g as a wireless power grid,” Scientific
Reports, vol. 11, no. 1, pp. 1–9, 2021.
[46] T. H. Falk, C. Zheng, and W.-Y. Chan, “A non-intrusive quality and intelligibility
measure of reverberant and dereverberated speech,” IEEE Transactions on Audio,
Speech, and Language Processing, vol. 18, no. 7, pp. 1766–1774, 2010.
[47] X. Fan, J. Chen, J. Yang, P. Bai, Z. Li, and Z. L. Wang, “Ultrathin, rollable, paper-
based triboelectric nanogenerator for acoustic energy harvesting and self-powered
sound recording,” ACS nano, vol. 9, no. 4, pp. 4236–4243, 2015.
[48] F. Farhabakhshian, T. Brown, K. Mayaram, and T. Fiez, “A 475 mv, 4.9 ghz en-
hanced swing differential colpitts vco in 130 nm cmos with an fom of 196.2
dbc/hz,” in IEEE Custom Integrated Circuits Conference 2010, IEEE, 2010, pp. 1–
4.
[51] R. P. Feynman, “Plenty of room at the bottom,” in APS annual meeting, 1959.
179
[53] R. R. Fletcher, “Low-cost electromagnetic tagging: Design and implementation,”
Ph.D. dissertation, Massachusetts Institute of Technology, 2002.
[54] V. Forti, C. P. Balde, R. Kuehr, and G. Bel, “The Global E-waste Monitor 2020:
Quantities, flows, and the circular economy potential,” United Nations Universi-
ty/United Nations Institute for Training and Research, Tech. Rep., 2020, p. 120.
[58] K. K. Fu, Z. Wang, J. Dai, M. Carter, and L. Hu, “Transient electronics: Materials
and devices,” Chemistry of Materials, vol. 28, no. 11, pp. 3527–3539, 2016.
[59] M. Gallucci, “Next-gen solar cells can harvest indoor lighting for iot devices,” IEEE
Spectrum, 2020.
[60] C. Gao, Y. Li, and X. Zhang, “Livetag: Sensing human-object interaction through
passive chipless wifi tags,” in 15th {USENIX} Symposium on Networked Systems
Design and Implementation ({NSDI} 18), 2018, pp. 533–546.
[61] S. George et al., “A programmable and configurable mixed-mode fpaa soc,” IEEE
Transactions on Very Large Scale Integration (VLSI) Systems, vol. 24, no. 6,
pp. 2253–2261, 2016.
[65] V. Giurgiutiu, Structural health monitoring: with piezoelectric wafer active sen-
sors. Academic Press, 2007.
[66] C. Gomes et al., “Computational sustainability: Computing for a better world and
a sustainable future,” Communications of the ACM, vol. 62, no. 9, pp. 56–65, 2019.
180
[67] R. Gonzalez, B. M. Gordon, and M. A. Horowitz, “Supply and threshold voltage
scaling for low power cmos,” IEEE Journal of Solid-State Circuits, vol. 32, no. 8,
pp. 1210–1216, 1997.
[73] C. Harrison and S. E. Hudson, “Scratch input: Creating large, inexpensive, unpow-
ered and mobile finger input surfaces,” in Proceedings of the 21st annual ACM
symposium on User interface software and technology, 2008, pp. 205–208.
[74] X. He et al., “An ultrathin paper-based self-powered system for portable electron-
ics and wireless human-machine interaction,” Nano Energy, vol. 39, pp. 328–336,
2017.
[76] D. Holman, N. Fellion, and R. Vertegaal, “Sensing touch using resistive graphs,”
in Proceedings of the 2014 conference on Designing interactive systems, 2014,
pp. 195–198.
181
[79] I. Hwang et al., “Multifunctional smart skin adhesive patches for advanced health
care,” Advanced healthcare materials, vol. 7, no. 15, p. 1 800 275, 2018.
[83] H. Ishii and B. Ullmer, “Tangible bits: Towards seamless interfaces between peo-
ple, bits and atoms,” in Proceedings of the ACM SIGCHI Conference on Human
factors in computing systems, 1997, pp. 234–241.
[84] B. Islam, Y. Luo, and S. Nirjon, “Zygarde: Time-sensitive on-device deep intelli-
gence on intermittently-powered systems,” 2019.
[85] V. Iyer, J. Chan, I. Culhane, J. Mankoff, and S. Gollakota, “Wireless analytics for
3d printed objects,” in Proceedings of the 31st Annual ACM Symposium on User
Interface Software and Technology, 2018, pp. 141–152.
[86] V. Iyer, J. Chan, and S. Gollakota, “3d printing wireless connected objects,” ACM
Transactions on Graphics (TOG), vol. 36, no. 6, pp. 1–13, 2017.
[91] J. M. Kahn, R. H. Katz, and K. S. Pister, “Next century challenges: Mobile net-
working for “smart dust”,” in Proceedings of the 5th annual ACM/IEEE interna-
tional conference on Mobile computing and networking, 1999, pp. 271–278.
182
[92] M. E. Karagozler, I. Poupyrev, G. K. Fedder, and Y. Suzuki, “Paper generators:
Harvesting energy from touching, rubbing and sliding,” in Proceedings of the 26th
annual ACM symposium on User interface software and technology, 2013, pp. 23–
30.
[93] E. Karana, B. Barati, V. Rognoli, A. Zeeuw Van Der Laan, et al., “Material driven
design (mdd): A method to design for material experiences,” 2015.
[95] M. R. Khorrami, E. Fares, and D. Casalino, “Towards full aircraft airframe noise
prediction: Lattice boltzmann simulations,” in 20th AIAA/CEAS aeroacoustics con-
ference, 2014, p. 2481.
[96] S.-J. Kim and J.-S. Lee, “Flexible organic transistor memory devices,” Nano letters,
vol. 10, no. 8, pp. 2884–2890, 2010.
183
[103] A. Lay-Ekuakille, G. Vendramin, A. Trotta, and G. Mazzotta, “Thermoelectric gen-
erator design based on power from body heat for biomedical autonomous devices,”
in 2009 IEEE international workshop on medical measurements and applications,
IEEE, 2009, pp. 1–4.
[104] J. Lewis, “Understanding microphone sensitivity,” Analog Dialogue, vol. 46, no. 2,
pp. 14–16, 2012.
[105] J. Lewis and B. Moss, “Mems microphone: The future for hearing aids,” Analog
Dialogue, vol. 47, pp. 3–5, 2013.
[107] H. Li, C. Ye, and A. P. Sample, “Idsense: A human object interaction detection sys-
tem based on passive uhf rfid,” in Proceedings of the 33rd Annual ACM Conference
on Human Factors in Computing Systems, 2015, pp. 2555–2564.
[108] H. Li et al., “Paperid: A technique for drawing functional battery-free wireless in-
terfaces on paper,” in Proceedings of the 2016 CHI Conference on Human Factors
in Computing Systems, 2016, pp. 5885–5896.
[111] L. Lin, S. Wang, S. Niu, C. Liu, Y. Xie, and Z. L. Wang, “Noncontact free-rotating
disk triboelectric nanogenerator as a sustainable energy harvester and self-powered
mechanical sensor,” ACS applied materials & interfaces, vol. 6, no. 4, pp. 3031–
3038, 2014.
[112] L. Lin et al., “Triboelectric active sensor array for self-powered static and dynamic
pressure detection and tactile imaging,” ACS nano, vol. 7, no. 9, pp. 8266–8274,
2013.
[113] D. Liu et al., “A constant current triboelectric nanogenerator arising from electro-
static breakdown,” Science advances, vol. 5, no. 4, eaav6437, 2019.
[114] R. Liu et al., “Shape memory polymers for body motion energy harvesting and
self-powered mechanosensing,” Advanced Materials, p. 1 705 195, 2018.
184
[115] V. Liu, A. Parks, V. Talla, S. Gollakota, D. Wetherall, and J. R. Smith, “Ambient
backscatter: Wireless communication out of thin air,” in ACM SIGCOMM Com-
puter Communication Review, ACM, vol. 43, 2013, pp. 39–50.
[123] G. E. Moore et al., “Cramming more components onto integrated circuits,” 1965.
[126] K. Myny, “The development of flexible integrated circuits based on thin-film tran-
sistors,” Nature electronics, vol. 1, no. 1, pp. 30–39, 2018.
185
[128] K. Nakatsuma, R. Takedomi, T. Eguchi, Y. Oshima, and I. Torigoe, “Active bioa-
coustic measurement for human-to-human skin contact area detection,” in SEN-
SORS, 2015 IEEE, IEEE, 2015, pp. 1–4.
[129] A. Nathan et al., “Flexible electronics: The next ubiquitous platform,” Proceedings
of the IEEE, vol. 100, no. Special Centennial Issue, pp. 1486–1517, 2012.
[130] J. Nishimura and T. Kuroda, “Eating habits monitoring using wireless wearable
in-ear microphone,” in Wireless Pervasive Computing, 2008. ISWPC 2008. 3rd In-
ternational Symposium on, IEEE, 2008, pp. 130–132.
[132] A. Olwal, J. Moeller, G. Priest-Dorman, T. Starner, and B. Carroll, “I/o braid: Scal-
able touch-sensitive lighted cords using spiraling, repeating sensing textiles and
fiber optics,” in Proceedings of the 31st Annual ACM Symposium on User Interface
Software and Technology, 2018, pp. 485–497.
[133] M. Ono, B. Shizuki, and J. Tanaka, “Touch & activate: Adding interactivity to exist-
ing objects using active acoustic sensing,” in Proceedings of the 26th annual ACM
symposium on User interface software and technology, 2013, pp. 31–40.
[135] O. Osibanjo and I. Nnorom, “The challenge of electronic waste (e-waste) man-
agement in developing countries,” Waste Management & Research, vol. 25, no. 6,
pp. 489–501, Dec. 2007.
[136] J. A. Paradiso and L. Borque, “A robust, continuous capacitive slider control for
cooktops,”
186
[140] J. A. Paradiso and T. Starner, “Energy scavenging for mobile and wireless electron-
ics,” IEEE Pervasive computing, vol. 4, no. 1, pp. 18–27, 2005.
[141] J. A. Paradiso, “The interactive balloon: Sensing, actuation and behavior in a com-
mon object,” IBM Systems Journal, vol. 35, no. 3.4, pp. 473–487, 1996.
[142] S. N. Patel and G. D. Abowd, “Blui: Low-cost localized blowable user interfaces,”
in Proceedings of the 20th annual ACM symposium on User interface software and
technology, ACM, 2007, pp. 217–220.
[146] E. R. Post and M. Orth, “Smart fabric, or wearable clothing,” in Digest of Papers.
First International Symposium on Wearable Computers, IEEE, 1997, pp. 167–168.
[148] S. Preradovic and N. C. Karmakar, “Chipless rfid: Bar code of the future,” IEEE
microwave magazine, vol. 11, no. 7, pp. 87–97, 2010.
[150] P. Raikwal, V. Neema, and A. Verma, “High speed 8t sram cell design with im-
proved read stability at 180nm technology,” in Electronics, Communication and
Aerospace Technology (ICECA), 2017 International conference of, IEEE, vol. 2,
2017, pp. 563–568.
[151] M. Rakús, V. Stopjaková, and D. Arbet, “Design techniques for low-voltage analog
integrated circuits,” Journal of Electrical Engineering, vol. 68, no. 4, p. 245, 2017.
[152] V. Ranganathan, S. Gupta, J. Lester, J. R. Smith, and D. Tan, “Rf bandaid: A fully-
analog and passive wireless interface for wearable sensors,” Proceedings of the
187
ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, vol. 2, no. 2,
pp. 1–21, 2018.
[158] R. R. Schaller, “Moore’s law: Past, present and future,” IEEE spectrum, vol. 34,
no. 6, pp. 52–59, 1997.
[159] I. Seo, Low-voltage low-power analog circuit techniques using floating-gate MOS
transistors. University of Florida, 2004.
[160] W.-H. Seol, Y. M. Lee, and J.-K. Park, “Enhancement of the mechanical properties
of pvdf membranes by non-solvent aided morphology control,” Journal of Power
Sources, vol. 170, no. 1, pp. 191–195, 2007.
[161] G. Sessler and J. West, “Electret transducers: A review,” The Journal of the Acous-
tical Society of America, vol. 53, no. 6, pp. 1589–1600, 1973.
[164] D. E. Shen, A. M. Österholm, and J. R. Reynolds, “Out of sight but not out of
mind: The role of counter electrodes in polymer-based solid-state electrochromic
devices,” Journal of Materials Chemistry C, vol. 3, no. 37, pp. 9715–9725, 2015.
188
[165] H. Shin, S. Seo, C. Park, J. Na, M. Han, and E. Kim, “Energy saving electrochromic
windows from bistable low-homo level conjugated polymers,” Energy & Environ-
mental Science, vol. 9, no. 1, pp. 117–122, 2016.
[169] P. Sparks, “The route to a trillion devices,” White Paper, ARM, 2017.
[172] T. Starner and J. A. Paradiso, “Human generated power for mobile electronics,”
Low-power electronics design, vol. 45, pp. 1–35, 2004.
[173] M. Stead, P. Coulton, and J. Lindley, “Spimes not things. creating a design man-
ifesto for a sustainable internet of things,” The Design Journal, vol. 22, no. sup1,
pp. 2133–2152, 2019.
189
[178] V. Talla, B. Kellogg, S. Gollakota, and J. R. Smith, “Battery-free cellphone,” Pro-
ceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technolo-
gies, vol. 1, no. 2, pp. 1–20, 2017.
[181] J.-M. Valin, F. Michaud, J. Rouat, and D. Létourneau, “Robust sound source local-
ization using a microphone array on a mobile robot,” in Intelligent Robots and Sys-
tems, 2003.(IROS 2003). Proceedings. 2003 IEEE/RSJ International Conference
on, IEEE, vol. 2, 2003, pp. 1228–1233.
[184] A. Varshney and L. Corneo, “Tunnel emitter: Tunnel diode based low-power car-
rier emitters for backscatter tags,” in Proceedings of the 26th Annual International
Conference on Mobile Computing and Networking, 2020, pp. 1–14.
[185] A. Varshney, W. Yan, and P. Dutta, “Judo: Addressing the energy asymmetry of
wireless embedded systems through tunnel diode based wireless transmitters,” in
Proceedings of the 20th Annual International Conference on Mobile Systems, Ap-
plications and Services, 2022, pp. 273–286.
[187] A. Wang, V. Iyer, V. Talla, J. R. Smith, and S. Gollakota, “{Fm} backscatter: En-
abling connected cities and smart fabrics,” in 14th {USENIX} Symposium on Net-
worked Systems Design and Implementation ({NSDI} 17), 2017, pp. 243–258.
[188] L. Wang, M. Guo, J. Zhan, X. Jiao, D. Chen, and T. Wang, “A new design of an
electrochromic energy storage device with high capacity, long cycle lifetime and
multicolor display,” Journal of Materials Chemistry A, vol. 8, no. 33, pp. 17 098–
17 105, 2020.
190
[189] Y. Wang et al., “Flextouch: Enabling large-scale interaction sensing beyond touch-
screens using flexible and conductive materials,” Proceedings of the ACM on In-
teractive, Mobile, Wearable and Ubiquitous Technologies, vol. 3, no. 3, pp. 1–20,
2019.
[190] Z. L. Wang, “On maxwell’s displacement current for energy and sensors: The origin
of nanogenerators,” Materials Today, vol. 20, no. 2, pp. 74–82, 2017.
[196] R. Want, K. P. Fishkin, A. Gujar, and B. L. Harrison, “Bridging physical and virtual
worlds with electronic tags,” in Proceedings of the SIGCHI conference on Human
factors in computing systems, 1999, pp. 370–377.
[199] M. Weiser, “The computer for the 21st century,” Scientific American, vol. 265,
no. 3, pp. 94–105, 1991.
191
[200] M. Weiser, “The computer for the 21st century,” ACM SIGMOBILE mobile com-
puting and communications review, vol. 3, no. 3, pp. 3–11, 1999.
[205] C. Xu, L. Yang, and P. Zhang, “Practical backscatter communication systems for
battery-free internet of things: A tutorial and survey of recent research,” IEEE Sig-
nal Processing Magazine, vol. 35, no. 5, pp. 16–27, 2018.
[206] M. Xu et al., “A soft and robust spring based triboelectric nanogenerator for har-
vesting arbitrary directional vibration energy and self-powered vibration sensing,”
Advanced Energy Materials, vol. 8, no. 9, p. 1 702 432, 2018.
[208] K. Yamazaki et al., “Home-assistant robot for an aging society,” Proceedings of the
IEEE, vol. 100, no. 8, pp. 2429–2441, 2012.
[209] S. Yan and E. Sanchez-Sinencio, “Low voltage analog circuit design techniques:
A tutorial,” IEICE Transactions on Fundamentals of Electronics, Communications
and Computer Sciences, vol. 83, no. 2, pp. 179–196, 2000.
[211] P.-K. Yang et al., “Paper-based origami triboelectric nanogenerators and self-
powered pressure sensors,” ACS nano, vol. 9, no. 1, pp. 901–907, 2015.
192
[213] D. Zhang et al., “Flexible computational photodetectors for self-powered activity
sensing,” npj Flexible Electronics, vol. 6, no. 1, pp. 1–8, 2022.
[214] H. Zhang et al., “Triboelectric nanogenerator for harvesting vibration energy in full
space and as self-powered acceleration sensor,” Advanced Functional Materials,
vol. 24, no. 10, pp. 1401–1407, 2014.
[215] P. Zhang, M. Rostami, P. Hu, and D. Ganesan, “Enabling practical backscatter com-
munication for on-body sensors,” in Proceedings of the 2016 ACM SIGCOMM
Conference, 2016, pp. 370–383.
[216] S. L. Zhang, Y.-C. Lai, X. He, R. Liu, Y. Zi, and Z. L. Wang, “Auxetic foam-based
contact-mode triboelectric nanogenerator with highly sensitive self-powered strain
sensing capabilities to monitor human body movement,” Advanced Functional Ma-
terials, vol. 27, no. 25, 2017.
[217] Y. Zhang and C. Harrison, “Pulp nonfiction: Low-cost touch tracking for paper,” in
Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems,
2018, pp. 1–11.
[220] Y. Zigel, D. Litvak, and I. Gannot, “A method for automatic fall detection of el-
derly people using floor vibrations and sound : Proof of concept on human mim-
icking doll falls,” IEEE Transactions on Biomedical Engineering, vol. 56, no. 12,
pp. 2858–2867, 2009.
193
VITA
194