Upload
vantu
View
238
Download
3
Embed Size (px)
Citation preview
INTERFACE ENGINEERING OF CAPACITIVE MICROMACHINED ULTRASONIC TRANSDUCERS FOR
MEDICAL APPLICATIONS
A DISSERTATION
SUBMITTED TO THE DEPARTMENT OF MECHANICAL ENGINEERING
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE
OF
DOCTOR OF PHILOSOPHY
Der-Song Lin
June 2011
http://creativecommons.org/licenses/by-nc/3.0/us/
This dissertation is online at: http://purl.stanford.edu/xk708cv9822
© 2011 by Der-Song Lin. All Rights Reserved.
Re-distributed by Stanford University under license with the author.
This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.
ii
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Butrus Khuri-Yakub, Primary Adviser
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Roger Howe
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Thomas Kenny
Approved for the Stanford University Committee on Graduate Studies.
Patricia J. Gumport, Vice Provost Graduate Education
This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.
iii
iv
ABSTRACT
Capacitive micromachined ultrasonic transducers (CMUTs), have been widely
studied in academia and industry over the last decade. CMUTs provide many benefits
over traditional piezoelectric transducers including improvement in performance through
wide bandwidth, and ease of electronics integration, with the potential to batch fabricate
very large 2D arrays with low-cost and high-yield. Though many aspects of CMUT
technology have been studied over the years, packaging the CMUT into a fully practical
system has not been thoroughly explored. Two important interfaces of packaging that
this thesis explores are device encapsulation (the interface between CMUTs and patients)
and full electronic integration of large scale 2D arrays (the interface between CMUTs and
electronics).
In the first part of the work, I investigate the requirements for the CMUT
encapsulation. For medical usage, encapsulation is needed to electrically insulate the
device, mechanically protect the device, and maintain transducer performance, especially
the access of the ultrasound energy. While hermetic sealing can protect many other
MEMS devices, CMUTs require mechanical interaction to a fluid, which makes fulfilling
the previous criterion very challenging. The proposed solution is to use a viscoelastic
material with the glass-transition-temperature lower than room temperature, such as
Polydimethylsiloxane (PDMS), to preserve the CMUT static and dynamic performance.
Experimental implementation of the encapsulated imaging CMUT arrays shows the
device performance was maintained; 95 % of efficiency, 85% of the maximum output
pressure, and 91% of the fractional bandwidth (FBW) can be preserved. A viscoelastic
finite element model was also developed and shows the performance effects of the
coating can be accurately predicted. Four designs, providing acoustic crosstalk
suppression, flexible substrate, lens focusing, and blood flow monitoring using PDMS
layer were also demonstrated.
v
The second part of the work, presents contributions towards the electronic
integration and packaging of large-area 2-D arrays. A very large 2D array is appealing for
it can enable advanced novel imaging applications, such as a reconfigurable array, and a
compression plate for breast cancer screening. With these goals in mind, I developed the
first large-scale fully populated and integrated 2D CMUTs array with 32 by 192
elements. In this study, I demonstrate a flexible and reliable integration approach by
successfully combining a simple UBM preparation technique and a CMUTs-interposer-
ASICs sandwich design. The results show high shear strength of the UBM (26.5 g),
100% yield of the interconnections, and excellent CMUT resonance uniformity (σ = 0.02
MHz). As demonstrated, this allows for a large-scale assembly of a tile-able array by
using an interposer.
Interface engineering is crucial towards the development of CMUTs into a practical
ultrasound system. With the advances in encapsulation technique with a viscoelastic
polymer and the combination of the UBM technique to the TSV fabrication for
electronics integration, a fully integrated CMUT system can be realized.
vi
ACKNOWLEDGEMENTS
The work presented in this thesis would not have been possible without the vision,
guidance, collaboration, and support of many people. They have together made my life at
Stanford an educational, enjoyable, and exceptional experience.
First, I would like to thank my Ph.D. principal advisor, Professor Butrus (Pierre)
T. Khuri-Yakub, for granting me the privilege to work under his supervision so that I
could pursuit a research topic I have dreamed for years. Through all my Ph.D. study,
Pierre offered continuous instruction and insights that he has never hesitated to share his
wealth of knowledge in any occasions. He is always accessible for advice no matter in
town or on the trip. I would like to express my great appreciation for his support,
understanding, and responsiveness during my time at Stanford.
I would like to thank the other members of my committee, Professor Thomas
Kenny, Professor Roger T. Howe, and Professor Mark Brongersma. I would like to
convey my appreciation to Professor Kenny for guiding me the research resources when I
first came to Stanford. The opportunity of working closely with his group on an MEMS
temperature regulator project has been one of the most enjoyable and educational
summers during my days at Stanford. I was honored to know him technically and
personally. Professor Howe is always passionate in sharing his knowledge. His EE312
course was particularly useful to me in learning about MEMS and my research topic in
broader points of view. I also am indebted to Professor Brongersma, who served in my
Ph.D. qualification exam committee and as the chair of my Ph.D. defense committee. I
would also like to extend my thanks to Professor Beth Pruitt for training me about
MEMS through my early days at Stanford. Her teaching was truly inspiring and
ultimately led me to the research of using MEMS for biomedical applications.
This work would not have been possible without the funding provided by the
National Institutes of Health (NIH) under Grant Number 1R01CA1152677. The majority
vii
of this work made use of the National Nanofabrication Users Network (NNUN) facilities
funded by the National Science Foundation (NSF) under award number ECS-9731294.
I have learned valuable lessons through the collaborations with Robert Wodnicki,
Dr. Charley Woychik, and Dr. Kai Thomenius from GE Global Research Center, and
Bernd Otto of Pac Tech USA Inc. They provided me tremendous resources and
introduced me to the industry practices. They have helped to enrich my research scope
and equip me for future career.
I was fortunate to have worked closely with Dr. Xuefeng (Steve) Zhuang,
Professor A. Sanli Ergun, Professor Mario Kupnik, and Dr. Serena H. Wong in different
stages of my Ph.D. Sanli is my first personal fabrication mentor at Stanford. He spent
considerable time training me in the field of CMUTs technology. He showed much
patience during my first learning of CMUT design, clean room activity and
characterization. I worked with Steve together for many projects and shared most of my
research with him. The work presented in this thesis would not be possible without his
continuous and timely support technically and mentally. I am deeply indebted to him for
his mentorship and friendship inside and out of the lab. My discussions with Mario were
always encouraging and inspiring my research to the next level. His distinct high
standards for the quality of research and publication has contributed one of the most to
the successful completion of my Ph.D. work. Our time together with his sense of humor
has kept our friendship after his return to Europe. Also special thanks to Serena, who is
very technically resourceful friend to me. She is extremely helpful in many aspects
including troubleshooting of my experiments as well as revisions to papers. I was lucky
to know her technically and personally.
I would like to express my gratitude for the great conversations and valuable
friendship from other former and current members of the Khuri-Yakub group: Dr. David
Yeh, Dr. Goksen Yaralioglu, Herb Te-Jen Ma, Dr. Amin Nikoozadeh, Dr. Omer Oralkan,
Dr. Ira O. Wygant, Prof. Ching-Hsiang Cheng, Dr. Yongli Huang, Prof. Baris Bayram,
Srikant Vaithilingam, Kazutoshi Torashima, Kwan-Kyu Park, Hyunjoo Jenny Lee, Min-
Chieh Ho, Amr Ahmed Essawi Saleh, Georgios Papadimitriou, Jessica Faruque, Yukio
Furukawa, Ron Watkins, Dr. Yukihide Tsuji, and Jung Woo Choe. I would like to thank
viii
for the assistances from Ginzton Lab staffs: Larry Randall, Tim Brand, Pauline Prather
and Heber Taina. In particular, Tom Carver has spent tremendous time contributing to the
development of PDMS encapsulation and UBM pads for CMUTs. The discussion with
him was always informative and extremely helpful.
I would like to thank SNF staff team for their assistance in training me and
keeping my works smoothly in the clean room. My special appreciation to many lab
members, especially, Eric Perozziello for many helpful processing advice and urgent
wafer rescue during the nights.
I was honored to know a number of MEMS professionals who inspired me for the
research in this field before my study in U.S. and advised me even until this day.
Professor Long-Sun Huang from National Taiwan University (NTU) was the first teacher
to initiate me into BioMEMS study. Professor Cheng-Hsien Liu from National Tsing-
Hua University provided me invaluable guidance on how to survive at Stanford.
Professor Tsung-Tsong Wu has provided me solid training of Ultrasonics in the early
years of my study at NTU, which in turn becomes the important foundations of the
successful completion of my Ph.D. study.
I would like to reserve my special thanks for many wonderful friends who have
enriched my life at Stanford in many ways. I would like to thank all the brothers and
sisters from South Bay Bible Church (SBBC) for being my physical and mental support
through all my struggles and spiritual growth. I am indebted to them for being like a true
family to me. My appreciation is also to the friends from Stanford Taiwanese Student
Association (STSA) and STSA softball team for sharing all the happy time. In particular,
I am grateful to Dr. Yu-Chi Chang, Prof. Pei-Chen Su, and Dr. Po-Ta (Joseph) Chen for
their life sharing and encouragement at all times.
Finally, I would like to express my deepest gratitude to my family. I was deeply
indebted to my parents, both teachers, for their teaching about disciplines and
responsibilities to the society. I thanks for their unconditional love, sacrifices and the
utmost support from the very beginning.
ix
I was especially indebted to my spiritual companion, my wife, Yachen (Nina) Liu,
for being exceptionally understanding, and extremely supportive and encouraging
throughout the years. She joked sometimes she is truly a Ph.D. short for “Push Husband
to Doctor”. I cannot express my deepest appreciation for her constant mental companion
by volunteering countless sleepless nights to be with me in the lab or aside the fab. By
sharing all the years together, I fully understand how well she completes me. Our little
angel daughter, Koen Elyse Lin, was born on 04/27/’10 exactly one month before my
defense exam. The new family member has made our family full of love and blessing
since then. I dedicate this thesis to my wife, parents and my daughter, to whom I owe all I
have ever accomplished.
x
TABLE OF CONTENTS
Abstract ................................................................................................................. iv
Acknowledgements .............................................................................................. vi
Table of Contents .................................................................................................. x
List of Tables ...................................................................................................... xiii
List of Figures ..................................................................................................... xiv
Chapter 1 Introduction ...................................................................................... 1 1.1 Motivations ............................................................................................. 1
1.2 Contributions ........................................................................................... 5
1.3 Thesis Structure ...................................................................................... 8
Chapter 2 CMUT Basics and Characterization ............................................ 10
2.1 Background of CMUTs ......................................................................... 10
2.2 Modeling for CMUTs ........................................................................... 13
2.2.1 Parallel-plate capacitor model ........................................................... 13
2.2.2 Equivalent circuit model ................................................................... 19
2.3 Characterization methods for CMUTs .................................................. 21
2.3.1 Characterization in air ....................................................................... 21
2.3.2 Characterization in immersion .......................................................... 22
2.4 Static Operating Point Evaluation ......................................................... 24
2.4.1 Background and Motivation ............................................................. 24
2.4.2 Materials and Methods ...................................................................... 25
2.4.3 Results and Discussion ..................................................................... 29
2.4.4 Conclusions ....................................................................................... 34
Chapter 3 Frontside Interface Engineering of CMUTs ............................... 35 3.1 Encapsulation Design ............................................................................ 35
xi
3.1.1 Static operation and dynamic performance of encapsulated CMUTs
36
3.1.2 Viscoelasticity of polymer ................................................................ 37
3.2 Fabrication Process ............................................................................... 40
3.2.1 CMUT Fabrication ............................................................................ 40
3.2.2 PDMS encapsulation ......................................................................... 42
3.3 Viscoelastic Finite Element Model ....................................................... 48
3.4 Results and Discussion ......................................................................... 51
3.4.1 Characterization for static behavior .................................................. 51
3.4.2 Model parameter studies ................................................................... 54
3.4.3 Characterization for dynamic performance ...................................... 58
3.5 Conclusion ............................................................................................ 67
Chapter 4 PDMS-Encapsulated CMUTs ....................................................... 69
4.1 Crosstalk Suppression ........................................................................... 69
4.1.1 Overview ........................................................................................... 69
4.1.2 Coating thickness design ................................................................... 71
4.1.3 Experimental method and results ...................................................... 73
4.2 Flexible CMUTs ................................................................................... 79
4.2.1 Overview ........................................................................................... 79
4.2.2 Design and fabrication ...................................................................... 80
4.2.3 Test results and discussion ................................................................ 82
4.3 PDMS Coating as Lens Material for Focusing ..................................... 85
4.3.1 Overview ........................................................................................... 85
4.3.2 Lens design ....................................................................................... 86
4.3.3 Materials and methods ...................................................................... 87
4.3.4 Results and discussion ...................................................................... 91
4.4 Ultrasound Beam Tilting for Blood Flow Imaging ............................... 93
4.4.1 Introduction ....................................................................................... 93
4.4.2 Methods and results .......................................................................... 94
Chapter 5 Backside Interface Engineering of CMUTs .............................. 101
xii
5.1 Large Scale 2D Arrays ........................................................................ 101
5.1.1 Motivations ..................................................................................... 101
5.1.2 Reconfigurable array ....................................................................... 102
5.2 2D CMUT interconnection and 3D CMUT/IC Intergrations ............. 104
5.2.1 Surface micromachining through-wafer via ................................... 105
5.2.2 Through-wafer trench-framed interconnections ............................. 106
5.3 Interposer Design and UBM Preparation for Large Scale 2D Array .. 107
5.3.1 Single large die versus array-tiling assembly ................................. 108
5.3.2 CMUT design and fabrication ........................................................ 111
5.3.3 Under-bump-metallurgy design and fabrication ............................. 116
5.4 Large Scale Tiling CMUTs ................................................................. 126
5.4.1 Tiling assembly module design ...................................................... 127
5.4.2 Tiling assembly results ................................................................... 128
5.4.3 Acoustic measurement results ........................................................ 129
5.5 Small Pitch Design .............................................................................. 130
5.6 Conclusion .......................................................................................... 133
Chapter 6 Current and Future Work .......................................................... 134
Bibliography ...................................................................................................... 136
xiii
LIST OF TABLES
Table 1-1. Comparison of transducer technologies. ........................................................... 1
Table 1-2. Requirements of the encapsulation for a medical imaging ultrasonic
transducer. ................................................................................................................... 2
Table 2-1. Performance index as functions of gap-height ................................................ 24
Table 2-2. Design parameters and values ......................................................................... 25
Table 2-3. RMS surface roughness measured by AFM .................................................... 31
Table 3-1. Properties of various polymer materials commonly used for MEMS
applications [19], [21]. .............................................................................................. 39
Table 3-2. Parameters of the CMUTs for imaging and HIFU. ......................................... 40
Table 3-3. Acoustic properties of PDMS. ......................................................................... 51
Table 3-4. CMUTs for imaging. ....................................................................................... 60
Table 3-5. The coating effect of the CMUTs for imaging. ............................................... 60
Table 3-6. CMUTs for HIFU (radius of 70 µm and 50 µm). ............................................ 61
Table 4-1. Physical parameters of the CMUT array for crosstalk study .......................... 75
Table 4-2. Device parameters of flexible CMUTs. ........................................................... 80
Table 4-3. Attenuation of Sylgard 160 compared with various doped RTV or HTV ...... 88
Table 4-4. Physical parameters of the CMUT array for lens study. ................................. 89
Table 5-1. Device parameters of CMUT for breast cancer screening. ........................... 114
Table 5-2. CMUT fabrication process flow. ................................................................... 115
Table 5-3. Bump shear test results of various UBM metal stacks. ................................. 123
xiv
LIST OF FIGURES
Figure 1-1. The roadmap of CMUT integration with electronics. ...................................... 4
Figure 2-1. Basic CMUT structure: a cross-section diagram of a CMUT cell. ................ 11
Figure 2-2. CMUT anatomy in cell, element and array level. .......................................... 12
Figure 2-3. CMUT parallel plate capacitor model ............................................................ 13
Figure 2-5. Equivalent circuit model of CMUT (F=0 in transmit and V=0 in receive). C0:
device capacitance, n: electromechanical coupling coefficient, Zp: plate mechanical
impedance, Zm: medium impedance, Rs: source impedance, Cp: parasitic
capacitance, Rp: parasitic resistance. ........................................................................ 19
Figure 2-6. Experimental setup for input impedance measurements. ............................... 22
Figure 2-7. Typical electrical input impedance of a CMUT element. .............................. 22
Figure 2-8. Experimental setup for pulse-echo measurements. ........................................ 23
Figure 2-9. Typical pulse-echo wave-form and band shape for a 185-um by 185-um
CMUT element in a 2D array. .................................................................................. 23
Figure 2-10. The comparison of the surface roughness from the bottom side of the gap
between (a) The CMUTs with and without sealing, and (b) The CMUT with doped
polysilicon and thermal oxide and the one with doped substrate ............................. 27
Figure 2-11. The methodology to quantify the roughness of each layer. ......................... 28
Figure 2-12. Behavior of center deflection versus DC bias and real part of input
impedance versus DC bias: (a) Theoretical part, and (b). Measurements part. ........ 29
Figure 2-13. (a) SEM photograph (cross-section) of one CMUT cell showing the gap-
height, and (b) SEM photograph of surface roughness from the bottom side of the
gap ............................................................................................................................. 30
Figure 2-14. Thermal Oxide is roughened due to imprinting by overlying polysilicon
layer ........................................................................................................................... 32
xv
Figure 2-15. Doping reduces surface roughness of polysilicon ........................................ 33
Figure 2-16. Roughness contributed from doped polysilicon. .......................................... 33
Figure 3-1. Encapsulation approach for various MEMS devices. .................................... 36
Figure 3-2. Viscoelastic curves of PDMS in terms of the Young’s modulus and acoustic
impedance over (a) loading frequency, or (b) temperature based on the theory of
time-temperature superposition. Tg and fg represent the glass-transition temperature
under static loading, and the glass-transition loading frequency in room temperature,
respectively. .............................................................................................................. 38
Figure 3-3. CMUT device fabrication using wafer bonding technique for (a) imaging
CMUTs and (b) HIFU CMUTs. The process flow include (1) first oxidation and
BOE oxide etch, (2) second oxidation and dry etching, (3) wafer bonding, (4) handle
wafer and BOX removal and etch back for substrate access, (5) top electrode
deposition and patterning, and (6) PDMS coating. ................................................... 41
Figure 3-4. Imaging CMUTs pictures of (a) SEM cross sectional view of a cell, (b) 6
elements of one array with wire bonding, and (c) CMUT cells with plate deflection
under DC bias of 80% of pull-in voltage measured by white light interferometer.
Similarly, HIFU CMUTs pictures are shown in (d)-(f). ........................................... 42
Figure 3-5. Chemical structure of PDMS ......................................................................... 43
Figure 3-6. SEM of PDMS surface (a). before plasma treatment, and (b). after plasma
treatment (after [40]). ................................................................................................ 44
Figure 3-7. Relationship between the surface topography in micro scale and hydrophilic
behavior in macro scale (after [40]). ......................................................................... 44
Figure 3-8. Silanol group of PDMS surface treatment (after [40]). .................................. 44
Figure 3-9. HEMA grafted PDMS: creation of the hydroxyl terminated function group
(after [40]). ................................................................................................................ 44
Figure 3-10. Aging induced hydrophobicity recovery effect on PDMS with various
surface treatments (after [40]). .................................................................................. 45
xvi
Figure 3-11. Surface energy of various solids and liquids. ............................................... 46
Figure 3-12. Four biasing schemes for CMUT operation. ................................................ 47
Figure 3-13. Electrical input impedance results and the optical microscopic images of the
CMUTs before and after the surface-conductivity-induced interconnection failure. 47
Figure 3-14. Schematic of (a) the non-viscoelastic and (b) the viscoelastic FEM. FLUID
29, PLANE 42, PLANE 182, and TRANS 126 represent the element types used in
the ANSYS model. The plots also show the roller boundary conditions (constrained
to x-direction), and the anchor boundary conditions (constrained to both x and y
directions). (c) The Prony Series expansion used for PLANE 182 element in the
viscoelastic FEM can be represented by a spring in parallel with a series set of a
spring and a dashpot. ................................................................................................. 51
Figure 3-15. Static behavior verification: capacitance vs. DC bias voltage for the imaging
CMUTs: (a) The experimental data, and (b) the modeling data. The measured
capacitance larger than the modeling is due to the parasitics from the electrical
interconnections of the experimental setup. .............................................................. 52
Figure 3-16. The electrical impedance measurement of the CMUTs without coating,
showing the (a) real parts, and (b) imaginary parts at DC bias before and after the
pull-in. ....................................................................................................................... 53
Figure 3-17. The electrical impedance measurement of the CMUTs with 150µm PDMS
coating, showing the (a) real parts and (b) imaginary parts at DC bias before and
after the pull-in. ......................................................................................................... 54
Figure 3-18. The real part of the electrical impedance versus the DC bias measured from
the CMUTs (a) without and (b) with PDMS coating. ............................................... 54
Figure 3-19. The viscoelastic FEM results showing the effect of different PDMS coating
thicknesses: (a) 200-µm, and (b) 75-µm of GE RTV 615 PDMS coating. t = 0 s
corresponds to the beginning of the 50-ns pulse. The data were retrieved from the
PDMS-water interface. ............................................................................................. 56
xvii
Figure 3-20. The viscoelastic FEM results showing the effect from different types of
PDMS: (a) 150 µm of GE RTV 615, and (b) 150 µm of Sylgard 160 on the CMUT
for imaging. The average displacement from plate-PDMS and PDMS-water
interface were both shown with a 1-nm offset apart in Y-scale for clearer
visualization. The main signals (a-1, b-1) arrive at time zero, and then the excitations
propagate upward into the PDMS-water interfaces. The transmission parts were
shown at the PDMS-water interfaces at the arrival time of 139 ns (a-2), and 158 ns
(b-2) individually. The reflection part (a-3) travels back to the plate-PDMS interface
and arrives at the time of 278 ns, while there is no reflection for the Sylgard 160
case (b-3). .................................................................................................................. 57
Figure 3-21. Comparison between the measurements (with and without the 150 µm of GE
RTV 615 coating) and the viscoelastic FEM (with the 150 µm of GE RTV 615
coating) results for imaging CMUTs: (a) and (b) the time domain, and (b) and (c) the
frequency domain. The data were retrieved from the PDMS-water interface. ......... 59
Figure 3-22. The results from the non-viscoelastic FEM with sE for imaging CMUTs
with 150 µm of GE RTV 615 coating. The data were retrieved from the plate-PDMS
interface. sE leads to the slower speed of sound, so there was no data from PDMS-
water interface before 1.2 µs. .................................................................................... 60
Figure 3-23. Comparison between the (a) measurements and (b) the results from
viscoelastic model in time domain, and (c) those in frequency domain for HIFU
device with CMUT cell radius of 70 µm with 150 µm of GE RTV 615 coating. The
data was retrieved from the plate-PDMS interface. .................................................. 62
Figure 3-24. Comparison between the (a) measurements and (b) the results from
viscoelastic model in time domain, and (c) those in frequency domain for HIFU
device with CMUT cell radius of 50 µm with 150 µm of GE RTV 615 coating. The
data was retrieved from the plate-PDMS interface. .................................................. 63
Figure 3-25. Displacement and output pressure at CMUT surface: surface area
displacement results for the devices (a) without and (b) with PDMS coating;
xviii
comparison of the (c) average displacement and (b) average pressure between the
cases with and without PDMS coating. .................................................................... 65
Figure 3-26. Total output pressure from the CMUT with and without PDMS coating. ... 67
Figure 4-1. Schematic of surface-guided-mode crosstalk with energy guided at the
transducers surface. ................................................................................................... 71
Figure 4-2. Suggested PDMS coating thickness for the crosstalk-suppression. (a) The
energy envelop, )1exp(0 −⋅= pp in oil without coating. (b) The energy envelop
with coating by using Ccoating =1080 m/s (from GE RTV 615 PDMS). .................... 72
Figure 4-3. Experimental setup of optical and electrical measurement for crosstalk
characterization. ........................................................................................................ 74
Figure 4-4. Inter- and intra-element crosstalk measurement. ........................................... 76
Figure 4-5. The crosstalk measurement comparison between the CMUTs (a) without and
(b) with a lossy PDMS coating for the frequency of interest. .................................. 77
Figure 4-6. The impulse response comparison between PDMS coatings with different
attenuation. ................................................................................................................ 78
Figure 4-7. Schematic of (a) a catheter for intravascular ultrasound imaging and (b) the
detail of a catherter-based side-looking probe. ......................................................... 79
Figure 4-8. PDMS thickness as a function of spin speed. ................................................ 81
Figure 4-9. Schematic cross-section of CMUT elements with PDMS encapsulation and
trenches-refilling substrate. ....................................................................................... 82
Figure 4-10. Device photographs: (a) front side view, (b) back side view showing the
flexibility pushed by a wire, (c) magnified view of through-wafer PDMS dispensing
holes, (d) cross-sectional view showing the trench, the membrane, and the PDMS
coating and filling. .................................................................................................... 82
Figure 4-11. Test results: (a) electrical input impedance in air, (b) resonant frequency. . 83
xix
Figure 4-12. (a) A photograph of cracks aside the mechanical linking bridge, (b)
Schematic of current linking design, (c) Schematic of proposed ribbons (springs)
design. ....................................................................................................................... 84
Figure 4-13. Schematic acoustic pressure profile of comparison between CMUT arrays
(a) without lens coating, and (b) with lens coating. .................................................. 85
Figure 4-14. Design of lens focusing. ............................................................................... 87
Figure 4-15. Lens mold manufactured by using mold-transfer technique. ....................... 89
Figure 4-16. A 1D 128 elements linear CMUT array (a) before lens coating, and (b-c)
after lens coating. ...................................................................................................... 90
Figure 4-17. Pressure field measurement for CMUTs with (a) conformal PDMS coating,
and (b) lens PDMS coating. ...................................................................................... 92
Figure 4-18. System concept for a new carotid screening tool based on a 5-plane CMUT
array. (a) CMUT array and the interface board. (b) Asymmetric acoustic lens on
each 1D array for fixed angle off-axis beam steering in the elevation direction. (c)
Diagram of cross-sectional carotid images obtained using the proposed scheme. ... 95
Figure 4-19. Active array area illustration. ....................................................................... 96
Figure 4-20. (a) A 4-inch silicon wafer after the completion of fabrication. The center two
devices are 5-plane CMUT arrays; the periphery contains test devices. (b) A 5-plane
CMUT array diced off from the wafer. ..................................................................... 96
Figure 4-21. Resonant frequency distribution across a 1D array. ..................................... 97
Figure 4-22. Pulse-echo signal, and corresponding spectrum. ......................................... 98
Figure 4-23. Normalized output pressure as a function of DC bias and AC excitation
voltages. .................................................................................................................... 98
Figure 4-24. (a) Illustration of off-axis beam steering using an off-centered cylindrical
lens. (b) Tilt angle as a function of the ratio of acoustic velocities in the lens material
and in water (n = Cwater/Clens). .................................................................................... 99
xx
Figure 4-25. Setup of using a 1D array to obtain five cross-sectional color flow Doppler
images of a phantom. .............................................................................................. 100
Figure 5-1. Schematic of the dual ultrasonic imaging plates on the current mammography
setup. ....................................................................................................................... 102
Figure 5-2. Reconfigurable 2D CMUT arrays. ............................................................... 103
Figure 5-3. Different approaches for through-wafer interconnect. (a) Through-wafer via,
(b) through-wafer trenches. ..................................................................................... 105
Figure 5-4. 3D schematic of the trench-isolated CMUT with a supporting frame. ........ 107
Figure 5-5. Large-scale CMUT arrays can be made of (a) a single large die, or (b) an
array tiling assembly. .............................................................................................. 109
Figure 5-6. 9-layer HyperBGA® substrate ..................................................................... 110
Figure 5-7. Illustration of signal routing in the interposer. ............................................. 111
Figure 5-8. Serial resistance versus the resistivity of the SOI substrate [93]. ................ 112
Figure 5-9. Temperature profile in CZ Silicon growth [94]. The freezing interface of the
Silicon crystal is concave into the melt. .................................................................. 113
Figure 5-10. Fusion bonding yields comparison between the SOI with resistivities of (a)
0.01~0.025 Ω-cm, and (b) 0.001~0.004 Ω-cm. Bonding yield of (a) is 100% for all
21 arrays, and down to (b) 66.7% (14 out of 21 arrays). ........................................ 113
Figure 5-11. Schematic comparison between the previous single-die UBM technique and
the proposed pre-trench UBM for tile able sandwich structure. ............................. 117
Figure 5-12. Schematic process of the pre-trench UBM using electroplating (a) before
and (b) after oven reflow. ........................................................................................ 118
Figure 5-13. Plating process steps .................................................................................. 119
Figure 5-14. Comparison between the pillars (a) with and (b) without footing. ............ 119
Figure 5-15. Schematic of the post-trench UBM with (a) straight sidewalls of trenches,
and (b) trenches with negative sidewall slope and footing. .................................... 120
xxi
Figure 5-16. Post-trench UBM and solder-bumping in the area with and without BCB
trench-filling. .......................................................................................................... 120
Figure 5-17. An optical microscopy showing the crack through the photoresist layer after
the e-beam evaporation of UBM metal stacks. The crack was due to the intrinsic
compressive stress of the 0.5-µm Ni film. .............................................................. 122
Figure 5-18. SEM pictures show the UBM pad with 0.03µm Au and 0.5um Ni has huge
surface roughness by observing the grain size around tens of nanometers. ........... 122
Figure 5-19. SEM pictures show the UBM pad shear interface of the brittle failure mode.
................................................................................................................................. 123
Figure 5-20. Suggested bump shear strength and the results from the pre-trench UBM
using evaporated Al/Ti/Ni/Au. ................................................................................ 124
Figure 5-21. A SEM pictures of the first trenched CMUT arrays with excellent quality
solder bumps showing (a) an excellent yield on a 16 by 16 2D array with excellent
solder-ball-height uniformity. (b) A close-up of a signal pad shows the solder ball on
top of the trench-isolated 90x90-µm-wide and 255-µm-tall pillar. ........................ 125
Figure 5-22. (a) Electrical input impedance of one element, and (b) resonance frequency
distribution across the 16x16 array. ........................................................................ 126
Figure 5-23. Schematic cross-section of the modular package design. .......................... 127
Figure 5-24. Schematic signal routing scheme for a multi-row linear imaging array. ... 128
Figure 5-25. Top view photograph of the first large-scale tile-able CMUT array assembly.
................................................................................................................................. 128
Figure 5-26. Cross-section SEM of the assembled CMUTs-interposer-ASICs sandwich
structure. .................................................................................................................. 129
Figure 5-27. Photograph showing topside view of the 1x2 functional CMUT array on a
laminate interposer solder attached to a board. ....................................................... 129
Figure 5-28. Photograph showing the cross-sectional view of the functional prototype.130
Figure 5-29. The pulse/echo results from the CMUT-interposer assembly module. ...... 130
xxii
Figure 5-30. Beam theory calculated induced maximum tensile and shear stress at the
bottom of the signal pillar under the bump-shear force acting on the top of the pillar.
................................................................................................................................. 131
Figure 5-31. Finite element analysis agrees with the beam theory calculation with only
6.7% error. .............................................................................................................. 131
Figure 5-32. Design guideline for pitch (pillar width) and substrate thickness (pillar
length) with experimental-determined fusion-bonding strength. The tensile fusion
bonding strength was based on the bump-shear test of 3.5 g results with the beam
theory calculation. The yellow region indicates the verified safe zone for various
pillar width and length designs. .............................................................................. 132
Figure 5-33. (a) Cartoon showing the bump-shear test on a signal pillar. (b) SEM on the
bottom of a post-shear-failure pillar showing the presence of a BOX layer. The shear
failure interface is through the Si-BOX interface but not the Si pillar. (c)
Microscopic optical picture shows the bottom of the trench with the fringing patterns
within the 90x90 µm square. The fringing pattern indicates the delaminated BOX
from device Si layer. ............................................................................................... 132
1
CHAPTER 1 INTRODUCTION
1.1 MOTIVATIONS
Capacitive micromachined ultrasonic transducers (CMUTs) have been widely
studied in academia and industry over the last decade. CMUTs provide many benefits
over traditional piezoelectric transducers including improvement in performance through
wide bandwidth, and ease of electronics integration, with the potential to batch fabricate
very large 2D arrays with low-cost and high-yield (Table 1-1). Though many aspects of
CMUT technology have been studied over the years, packaging the CMUT into a fully
practical system has not been thoroughly explored. Two important interfaces of
packaging that this thesis explores are device encapsulation (the interface between
CMUTs and patients) and full electronic integration of large-scale 2D arrays (the
interface between CMUTs and electronics).
Table 1-1. Comparison of transducer technologies.
Piezoelectric transducer CMUT
Fabrication method Ceramic technology MEMS technology
Array fabrication Difficult and high cost,
very difficult for 2D array, ring array
Easy and low cost, arrays with through-wafer
interconnects Frequency range Relatively narrow Broad
Bandwidth Moderate, matching layers required Wide
Array uniformity Moderate High
Thermal stability Low High
IC integration No Yes
Output pressure High Relatively low but improving
For the front-side interface, the requirements for the CMUT packaging in terms of
encapsulation were investigated. For medical usage, this packaging material should
2
electrically insulate the device from the environment, protect the device against humidity
and other corrosive substances, and also be biocompatible (Table 1-2). At the same time,
this encapsulation should maintain transducer performance, especially the transfer of
energy to and from the medium. While other microelectromechanical system (MEMS)
devices such as accelerometers and gyroscopes are typically protected from the
environment by hermetic sealing, CMUTs require mechanical interaction to a fluid,
which makes fulfilling the previous criteria very challenging. Operation of a CMUT
requires both a DC and AC voltage, such that attributes a static and dynamic behavior.
Figures of merit for CMUTs include the reception sensitivity and transmit efficiency,
which are determined as the static operation point; it also includes the dynamic
performance, i.e. total output pressure and fractional bandwidth (FBW) (Table 1-2). A
solution by using a viscoelastic polymer to address the encapsulation challenges and
fulfills the AC and DC requirements is presented. By using this approach, four designs,
providing acoustic crosstalk suppression, flexible substrate, lens focusing, and blood flow
monitoring using PDMS layer is also demonstrated.
Table 1-2. Desirable attributes of the encapsulation material of a medical imaging ultrasonic transducer.
Provide protection
Maintain performance
Provide other benefits
p Electrical insulationp Humidity & corrosive substance barrierp Ablation resistancep Biocompatibility
Figures of Merit :p Sensitivity & efficiencyp Max output pressurep Fractional bandwidth (FBW)
p Acoustic crosstalk suppressionp Lens focusingp Flexible substratep Blood flow imaging
A full packaging towards the CMUT development into a practical ultrasound
system requires also the full electronic integration (the interface between CMUTs and
3
electronics). The second part of the thesis, presents works on the electronic integration
and packaging of CMUT arrays, specifically on the large-area 2-D arrays. A very large
2D array is appealing for it can enable advanced novel imaging applications. A number
of specific applications exist in which large area ultrasound transducer arrays can be
used. These include cancer screening and continuous non-invasive blood pressure
monitoring [1]. A novel breast cancer screening can be realized by placing very large
ultrasonic transducer arrays on both upper and lower compression plates in a manner
similar to X-ray based mammography. By using the dual ultrasonic imaging plates,
compound images of a lesion using data from both the upper and lower plates can
minimize shadowing artifacts. Two-way projection formed 3D data set can provide
through-transmission measurements including the speed of sound and attenuation.
Complete 3D data sets can be acquired in renderings made with 3D compounding to
optimize achievable contrast and maximize speckle suppression. Imaging techniques such
as elastography can be applied in a more rigorous manner than before including use of
3D [7, 8].
Another motivation of developing large area ultrasound transducer arrays using
CMUT technology is to enable advanced novel interconnection technique for a mosaic
annular reconfigurable array. A major challenge in large area transducers is the large
number of interconnects existing between the signal processing electronics and the
transducer array. With the large number of transducer elements (10,000 to >1,000,000),
each with its own respective signal processing circuitry, significant power, cost, and area
penalties exist. One attractive way to reduce the number of signal processing channels for
such a large area array is through the use of a mosaic annular reconfigurable array [4][5].
The mosaic array architecture groups a number of sub-elements together along iso-phase
lines to form larger transducer elements which are then each connected to a single system
channel. In this way, an array that has tens of thousands of active acoustic sub-elements
can be reduced to a much smaller number of system processing channels (e.g. 20-100). In
order to realize such array architecture, it is necessary to integrate switching electronics
immediately behind the acoustic array. Compared to traditional PZT-based ultrasonic
transducers, CMUTs were made by adopting MEMS technology. It allows CMUTs to be
4
processed in a similar way to the ASICs, which makes them more amenable to standard
packaging flows.
CMUTs Interface with Electronics
1D array using wire-‐bonding
2D CMUT arrays using through-‐wafer-‐via
2D CMUT arrays using trench-‐frame technology
Integration w/ electronics
Interconnections
2004 2006 2008 2010
16 x 16 M x N1 x N 16 x 16
Wire-‐bonding Via-‐first TSV Via-‐last TSV
(C. Cheng ’04)
(X. Zhuang ’08)
Large scale 2D CMUT arrays using trench-‐frame technology
CMUT
interposer
C CMUT
interposer
C
Interposer
Figure 1-1. The roadmap of CMUT integration with electronics.
CMUT interconnections and integration with electronics have been improving
over the past few years (Figure 1-1). Following the 1D CMUT array, the first successful
2D interconnections for 2D CMUT array was demonstrated by Cheng et al. [6] by using a
“through-wafer-via” technology. However, this via-first through-silicon-via (TSV)
technique is not compatible with the fusion bonding for CMUT cell formation. Because
the via has to be processed first, the change of wafer surface condition make it extremely
difficult for a good wafer bonding. As the fusion-bonding technique has been
demonstrated to provide many advantages on CMUT design flexibility and fabrication
simplifications, a 2D interconnection technology compatible with fusion bonding was
required. This issue was addressed by Zhuang et al. [7] by demonstrating a via-last 2D
interconnection technology using a trench-frame structure. However, the challenge is
how to extend this approach to a larger scale, and with smaller pitch-size. An enabling
technology for a large-scale 2D array requires a simple, flexible and reliable
interconnection and IC integration technique. As an example of fabricating a large-scale
5
2D CMUT array using wafer bonding, a tile-able CMUT array assembly and a CMUT-
interposer-ASIC sandwich structure were proposed in this work.
Interface engineering is crucial towards the development of CMUTs into a
practical ultrasound system. With the advances in encapsulation technique with a
viscoelastic polymer and the combination of the UBM technique to the TSV fabrication
for electronics integration, a fully integrated CMUT system can be realized.
1.2 CONTRIBUTIONS
The primary contributions of this thesis include two parts. First, for the front side
interface engineering (the interface between CMUTs and patients):
• A viscoelastic polymer with Tg lower than room temperature, such as PDMS,
was proposed as a solution for CMUT encapsulation, which preserves the
static operation and maintains the dynamic performance.
• A viscoelastic finite element model was developed which can predict both
the DC and AC behavior of a PDMS-coated CMUT.
• Various CMUT applications by using PDMS encapsulation were
demonstrated, including acoustic crosstalk suppression, a flexible substrate,
lens focusing, and a blood flow imaging.
Second, for the backside interface engineering (the interface between CMUTs and
electronics):
• The first successful tile-able assembly module of CMUTs-interposer-ASICs
sandwich with flexible pitch sizes (185 µm for CMUTs and 150 µm for
ASICs ).
• The first large-scale fully populated 2D CMUTs array (32 by192 elements).
• A design guideline for finer element-pitch design: pillar interface strength
(fusion bonding strength and BOX strength) and the length of pillars.
• A simple pre-trench UBM compatible with trench-frame interconnect
technology, which provides great and uniform shear strength, and enables the
6
100% yield of the solder bumping and the interconnections of CMUTs-
interposer-ASICs module.
Most of the work contained in this dissertation has been previously published.
The publication related to Chapter 2 is:
1. D.-S. Lin, X. Zhuang, S. H. Wong, A. S. Ergun, M. Kupnik, and B. T.
Khuri-Yakub, "Characterization of fabrication related gap-height variations in capacitive micromachined ultrasonic transducers," in Proc. IEEE Ultrason. Symp., 2007, pp. 523-526.
The publications related to Chapter 3 include:
2. D.-S. Lin, X. Zhuang, S. Wong, M. Kupnik, and B. T. Khuri-Yakub, "Encapsulation of Capacitive Micromachined Ultrasonic Transducers using Viscoelastic Polymer," IEEE/ASME J. Microelectromech. Syst., in press.
3. D.-S. Lin, X. Zhuang, S. H. Wong, M. Kupnik, and B. T. Khuri-Yakub, “A Viscoelastic Finite Element Model for Polymer Coatings on CMUTs,” presented at the IEEE Ultrasonics Symposium, Beijing, China, Nov. 2008.
The publications related to Chapter 4 include:
4. B. Bayram, M. Kupnik, G. G. Yaralioglu, O. Oralkan, D.-S. Lin, X. Zhuang, A. S. Ergun, A. Sarioglu, S. H. Wong, and B. T. Khuri-Yakub, "Characterization of cross-coupling in capacitive micromachined ultrasonic transducers," in Proc. IEEE Ultrason. Symp., 2005, pp. 601-604.
5. B. Bayram, M. Kupnik, G. G. Yaralioglu, Ö. Oralkan, A. S. Ergun, D. Lin, S. H. Wong, and B. T. Khuri-Yakub, "Finite element modeling and experimental characterization of crosstalk in 1-D CMUT arrays," Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, vol. 54, no. 2, pp. 418- 430, Feb. 2007.
6. X. Zhuang, D.-S Lin, O. Oralkan, and B. T. Khuri-Yakub, “Fabrication of flexible transducer arrays with through-wafer electrical interconnects based on trench refilling with PDMS,” IEEE/ASME J. Microelectromech. Syst., vol. 17, pp. 446-452, Apr. 2008.
7. X. Zhuang, D.-S. Lin, O. Oralkan, and B. T. Khuri-Yakub, "Flexible transducer arrays with through-wafer electrical interconnects based on trench refilling with PDMS," in Proc. IEEE MEMS Conference, 2007, pp. 73-76.
8. X. Zhuang, D.-S. Lin, O. Oralkan, and B. T. Khuri-Yakub, “Flexible CMUT Arrays with Through-Wafer Electrical Interconnects Based on Trench Refilling with PDMS”, presented at the Transducer Conference, Los Angeles, CA, Apr. 2007.
7
9. D.-S Lin, X. Zhuang, J. Faruque, O. Oralkan, S. Napel, R. B. Jeffrey, and B. T. Khuri-Yakub, “Carotid Peak Blood Velocity Detection using a 5-Plane CMUT Array with Asymmetric Acoustic Lens: Initial Results,” to be published in Proc. IEEE Ultrason. Symp., 2009.
The publications related to Chapter 5 include:
10. D.-S. Lin, X. Zhuang, R. Wodnicki, C. G. Woychik, O. Oralkan, M. Kupnik, and B. T. Khuri-Yakub, “Packaging of Large and Low-Pitch Size 2D Ultrasonic Transducer Arrays,” to be published in Proc. IEEE MEMS Conference, Hong Kong, 2010 (in print).
11. R. Wodnicki, D.-S Lin, K. Thomenius, R. Fisher, C. Woychik, A. Byun, X. Zhuang, O. Oralkan, S. Vaithilinigam, and B. Khuri-Yakub , “Multi-row linear array using cMUTs and multiplexing electronics,” to be published in Proc. IEEE Ultrason. Symp., 2009 (in print).
12. X. Zhuang, I. O. Wygant, D.-S Lin, O. Oralkan, M. Kupnik, and B. T. Khuri-Yakub, “Wafer-bonded 2-D CMUT arrays incorporating through-wafer trench-isolated interconnects with a supporting frame,” IEEE Trans. Ultrason., Ferroelect., Freq. Contr. , vol. 56, pp. 182-192, Jan. 2009.
13. X. Zhuang, D.-S. Lin, A. S. Ergun, O. Oralkan, and B. T. Khuri-Yakub, "Trench-isolated CMUT arrays with a supporting frame," in Proc. IEEE Ultrason. Symp., 2006, pp. 1955-1958.
14. X. Zhuang, I. O. Wygant, D.-S. Lin, M. Kupnik, O. Oralkan, and B. T. Khuri-Yakub, "Trench-Isolated CMUT Arrays with a Supporting Frame: Characterization and Imaging Results," in Proc. IEEE Ultrason. Symp., 2007, pp. 507-510.
I also had the opportunity to coauthor a number of papers prior to my work in
CMUT technology:
1. A. Nikoozadeh, I. O. Wygant, D.-S Lin, Ö. Oralkan, A. S. Ergun, D. N. Stephens, K. E. Thomenius, A. M. Dentinger, D. Wildes, G. Akopyan, K. Shivkumar, A. Mahajan, D. J. Sahn, and B. T. Khuri-Yakub, "Forward-looking intracardiac ultrasound imaging using a 1-D CMUT array integrated with custom front-end electronics," Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, vol. 55, no. 12, pp. 2651-2660, Dec. 2008.
2. S. H. Wong, M. Kupnik, X. Zhuang, D.-S Lin, K Butts-Pauly, and B.T. Khuri-Yakub, “Evaluation of wafer bonded CMUTs with rectangular membranes featuring high fill factor,” IEEE Trans. Ultrason., Ferroelect., Freq. Contr. , vol. 55, pp. 2053-2064, Sep. 2008.
The contributions presented in this thesis have widespread applications in many
other transducer arrays, including gas, bio/medical, and imaging sensor array and many
8
other actuator arrays. The work presented in this dissertation will be presented from the
perspective of micromachined ultrasound imaging array for medical applications.
1.3 THESIS STRUCTURE
In Chapter 2 of this thesis, it is first presented a brief theoretical analysis and
experimental characterization of CMUTs. A parallel-plate model of CMUTs is used for
the analysis. Important device operation parameters, such as plate pull-in voltage, center
frequency, fractional bandwidth, output pressure and receive sensitivity are analyzed.
Many experimental characterization techniques are introduced, utilizing the design
feedback during and after the fabrication. The gap-height variation as one of the most
critical issues from the CMUTs fabricated by a sacrificial-release process was addressed.
Using the experimental characterization techniques, the fabrication-related gap-height
variation was identified which suggest the wafer-bonding fabrication technique is
advantageous.
In Chapter 3, the requirements for the CMUT encapsulation are investigated. A
solution is proposed using a viscoelastic material with the glass-transition-temperature
lower than the room temperature, such as Polydimethylsiloxane (PDMS), to preserve the
CMUT static and dynamic performance. Experimental implementations of the
encapsulated CMUT arrays are demonstrated (for both imaging and high-intensity-
focused-ultrasound CMUTs). Both the experimental results of the maximum output
pressure, and the fractional bandwidth (FBW) are presented. A finite element model
addressing the viscoelastic polymer coating was also developed; the performance effects
of the coating are predicted and compared to the experimental results.
In Chapter 4, demonstrations of four PDMS encapsulated CMUT designs,
providing acoustic crosstalk suppression, flexible substrate, lens focusing, and blood flow
monitoring follow. In the first part of this chapter, a lossy top coating by using PDMS
was proposed to feature the encapsulation and the crosstalk suppression at the same time.
The coating thickness and material attenuation are designed. The crosstalk reduction is
demonstrated experimentally for two kinds of PDMS. The second part is to design the
low Young’s modulus from PDMS into the substrate to provide the flexibility of CMUT
9
arrays. By using a PDMS encapsulation and refilling trenches, a flexible CMUT array
was demonstrated for side-looking intravascular ultrasound imaging. Then the PDMS
was used as a lens material featuring a mechanical focusing for a 1D CMUT array.
PDMS selection in terms of the acoustic impedance, attenuation, and viscosity is
discussed. The pressure fields and beam profiles from the CMUTs with conformal
coating and lens coating are measured for comparison. Last, a new screening method for
carotid stenosis is proposed. The key components of the system include a 5-plane CMUT
array and asymmetric acoustic lenses using PDMS are designed, fabricated and tested.
Chapter 5 presents works on the backside interface engineering of CMUTs,
specifically on the electronic integration and packaging of large-area 2-D arrays. First,
the background of the 2D CMUT interconnections and 3D CMUT/IC intergrations are
discussed. Next, the interposer design is introduced and a simple UBM preparation for
large scale 2D array is presented. Then, the first large-scale fully populated and
integrated 2D CMUTs array with 32 by 192 elements is demonstrated. The results show a
flexible and reliable integration approach by successfully combining a simple UBM
preparation technique and a CMUTs-interposer-ASICs sandwich design. The results
show high shear strength of the UBM, great yield of the interconnections, and excellent
CMUT resonance uniformity. As demonstrated, this allows for a large-scale assembly of
a tile-able array by using an interposer.
Significant contributions have been made towards the fields of the encapsulation
and large-scale packaging of CMUTs. The results obtained suggest a number of areas
where CMUT encapsulation/packaging can be further improved are summarized in
Chapter 6.
10
CHAPTER 2 CMUT BASICS AND CHARACTERIZATION
In this chapter, a basic introduction to CMUT technology will be presented, starting with the basic structure and operation of a conventional CMUT cell. Important CMUT characteristics and the figure of merit are introduced. Simple analytical models are also presented to develop intuition concerning the design and operation of the device. Then, characterization methods are discussed which are essential for fabrication in-field
feedback and post-fabrication comparison with the theory and modeling results. A
fabrication-induced gap-height variation is evaluated as an example.
2.1 BACKGROUND OF CMUTS
Ultrasonic transducers can be classified according to the physical mechanism
upon which they are based to convert electrical energy into ultrasonic energy, and vice
versa. Magnetostriction, piezoelectricity and electrostatics are some of the physical
mechanisms used to generate and detect ultrasound. With the advent and maturation of
silicon micromachining, new possibilities for making ultrasonic transducers have
emerged, both for piezoelectric based and capacitor based devices, and for both airborne
and immersion applications. Thin vibrating plates of silicon, silicon nitride, alumina,
diamond, and other materials can be easily made and vibrated with piezoelectric,
capacitive, magnetic, thermal, and other types of driving mechanisms.
Historically, piezoelectric crystals, ceramics, polymers and recently
piezocomposite materials have been predominantly used to generate and detect
ultrasound. Although the idea of electrostatic transducers is as old as the early
piezoelectric transducers, piezoelectric materials have dominated ultrasonic transducer
technology. F. V. Hunt writes in his book Electroacoustics: The Analysis of
Transduction, and its Historical Background [8]: “After a month of careful study, during
which both magnetostriction and piezoelectricity were considered and then rejected,
Langevin decided that it would be better to fall back on the “singing condenser”... (March
1915) Numerical estimates indicated that, if electric field strengths on the order of a
million volts per centimeter (108 V/m) could be maintained, electrostatic forces as large
11
as a kilogram per square centimeter would come into play...” The reason why
electrostatic or capacitive transducers have not been popular is the high electric fields that
need to be maintained to achieve acceptable efficiencies.
Recent advances in microfabrication technology have made it possible to build
capacitive ultrasound transducers competing with piezoelectric transducers. With the use
of integrated circuit manufacturing techniques it is possible to make capacitors with sub-
micron gaps where electric fields of over 108 V/m can be sustained. The merit of such a
high electric field is that it results in transducers where the electromechanical coupling
coefficient can get close to unity, and thus be very competitive with the best piezoelectric
materials.
A CMUT is simply a device with two plate-like electrodes biased with a DC
voltage and driven with an additional AC signal to harmonically move one of the plates.
The main components are the cavity, the plate, and the electrode. The plate is clamped on
side posts (Figure 2-1), with size and thickness determined by the frequency requirement
of the specific application. The cavities underneath the plates are usually sealed to
enhance device performance and reliability by preventing environmental hazards, such as
moisture and dust particles, from either mechanically hampering the plate motion, or
electrically shorting the capacitor structure. A dielectric insulation layer is often
presented in the cavity to prevent electrical shorting between the two electrodes of the
CMUT.
Bottom Electrode
Substrate
Vacuum
Plate
Bottom Electrode
Substrate
Top ElectrodePlate
Vacuum
DC
AC
Figure 2-1. Basic CMUT structure: a cross-section diagram of a CMUT cell.
The CMUT plates deflect when subjected to an electrostatic or mechanical
actuation force. Thus, they serve to couple energy between the electrical and mechanical
12
domains. In the transmit mode, an AC actuation voltage superimposed on an optional DC
bias voltage is applied on the electrodes residing on the CMUT plates. The resulting
electrostatic force causes the plates to vibrate in the ultrasonic frequency range. The plate
vibration couples acoustic energy into the surrounding medium. In the receive mode, a
DC bias voltage is first applied on the capacitor structure of the CMUT. The impinging
ultrasound waves cause the plates to vibrate, resulting in an AC current that can be
detected by an external circuit.
2D array with 16x16 elements
1D linear array with 64 elements
Four CMUT elements with 35cells per element
100 cells per element
Element level
Nine CMUT cells
Five CMUT cells
Cell levelArray level
Figure 2-2. CMUT anatomy in cell, element and array level.
CMUTs provide several advantages over piezoelectric transducers: they can be
batch-produced with micromachining processes to tight parameter specifications, which
is difficult for piezoelectrics; they are easier to fabricate than piezoelectric transducers;
batch processing also enables the fabrication of transducer arrays with different
geometries and operating frequencies on a single wafer. Added to this advantage is the
broad bandwidth resulting from using over-damped plates in immersion applications, the
control afforded by integrated circuit manufacturing techniques in making single element,
13
1-D, 1.5-D, and 2-D arrays, and the ease of integration of the transducers with electronic
circuits.
2.2 MODELING FOR CMUTS
2.2.1 PARALLEL-PLATE CAPACITOR MODEL
A simple model is introduced in this section that allows the derivation of simple
formulas for some of the important CMUT parameters, such as the pull-in voltage, center
frequency, fractional bandwidth (FBW), maximum achievable output pressure, and
receive sensitivity. In this model, the CMUT top plate is constraint to only a piston-like
displacement (i.e., no deflection), and it forms a parallel-plate capacitor with the fixed
bottom electrode (Figure 2-3). The piston (top plate) is held over an effective electrostatic
gap, 0g , by a spring, k .
top electrode
bottom electrode
x
tp
g0
V
movable plate
Spring, kDamper, b
Figure 2-3. CMUT parallel plate capacitor model
In this model, the plate is simplified as a rigid plate with an equivalent mass m
supported by a spring with an equivalent spring constant k anchored on a substrate. A
damping factor b is used to represent all mechanical and acoustical damping mechanisms
in the CMUT. V represents the voltage that is applied between the plate and substrate.
One of the major assumptions in this model is that the plates endure a piston-like motion.
In reality, the plate displacement profile is non-uniform, with the center having the
maximum displacement and the side posts having zero displacement. Further, the piston-
like motion assumption also neglects higher order vibration modes of the plate that are
inherent in actual devices. Another assumption is the linear spring, which may also
14
deviate from reality especially when the plate displacement is significant when compared
to the thickness of the plate. Despite these limitations, this simple model provides us with
an understanding of the most important characteristics of a CMUT.
2.2.1.1 Plate pull-in voltage
A pull-in voltage exists for the CMUT plate, which occurs when the electrostatic
force overcomes the mechanical restoring force, and the plate abruptly snaps down to the
substrate. Usually a CMUT is designed to operate at a bias voltage less than the pull-in
voltage.
First, the equilibrium exists between the electrostatic force, eF , and the
mechanical restoring force, mF , exerted by the spring as long as the bias voltage not
larger than the pull-in voltage.
0=+=∑ em FFF , (2-1)
where =mF )( ggk eff − , 2
20
2gAV
F dce
ε−= , 0ε is the permittivity of free space and
A denotes the piston electrode area. Then the pull-in voltage can be obtained by knowing
the pull-in occurs when the electrostatic force gradient overcomes the gradient of the
mechanical restoring force.
Akg
VkgAV
gF pi
pipi
pi
Vg pipi0
3
3
20
,
0)(
εε
=⇒=−=∂
∂ ∑ . (2-2)
From (Equation 2-1) and (Equation 2-2), one can show that the pig is 2/3 of the
effective electrostatic gap, independent of other parameters.
effpipi
effpi
pieffpi
pi
pi ggg
gkgAV
gggAV
k32
22 2
20
3
20 =⇒−=−=⇒=
εε. (2-3)
By plugging this result into (Equation 2-2), the pull-in voltage can be expressed
as:
15
0
3
278
εAkg
V effpi = . (2-4)
The pull-in voltage is an important design parameter. In many applications, a
CMUT is used for both transmission and reception of acoustic waves; the pull-in voltage
is a trade-off between transmission and reception performances. Details of this tradeoff
will be discussed in sections 2.2.1.5.
2.2.1.2 Center frequency
The center frequency of a mass-spring-damper system is determined by the
equivalent mass meq and the spring keq as eqeqo mk /=ω while operating in air or
vacuum. This is equivalent to the natural resonant frequency of the plate because the
damping effect from air or vacuum can be ignored. Once the system is operating in
immersion, the center frequency is also determined by the complex damping factor
ir bbb += . Analytical formulae exist in some special cases, e.g. thin circular or square
plate that 10<< a/t <<100, where a is the lateral dimension of the plate (e.g., width,
diameter), and t is the plate thickness [9].
Using the general assumptions of Kirchhoff’s plate bending theory, the governing
differential equation for the motion of a thin plate is given by
Pt
ttD =∂∂+∇−∇ 2
224 ωρωσω , (2-5)
where σ , t , ρ , P and ω denote the stress within the plate, plate thickness, plate
density, the external pressure applied on the plate, and the plate displacement [9], [10],
[11]; In addition, D denotes the flexural rigidity of the plate given by
)1(12)(2
3
υσ−+= tED , (2-6)
where E and υ are the Young’s modulus and Poisson’s ratio of the top plate
material, respectively.
16
By simplifying the case, 0=σ , and following the derivations in [12], it can be
shown that, for a clamped circular plate, the first harmonic frequency (i.e., lowest
resonant frequency) in air or vacuum can be expressed as:
eq
eqo m
kEat =
−=
)1(95.2
22 νρω . (2-7)
While the damping effect from the medium is not negligible, assuming <<a the
acoustic wavelength (λ ), the resonant frequency of the system becomes:
ta
Eah
mo
ρρ
νρω
67.01
)1(95.2
22
+
−= , (2-8)
where mρ is the density of the medium. The equation refers to a more general
expression of the resonant frequency, implying the resonance shifts down with the
presence of the damping.
2.2.1.3 Fractional bandwidth
A wide bandwidth is preferred for better axial resolution. For a second order
spring-mass-dashpot resonant system, one can identify the higher and lower 3-dB cutoff
frequencies [13] as:
⎟⎟⎟
⎠
⎞
⎜⎜⎜
⎝
⎛+⎟⎟⎠
⎞⎜⎜⎝
⎛+= 1
21
21
2
QQoH ωω , (2-9)
and ⎟⎟⎟
⎠
⎞
⎜⎜⎜
⎝
⎛+⎟⎟⎠
⎞⎜⎜⎝
⎛+−= 1
21
21
2
QQoL ωω , (2-10)
where r
i
b)bm(k
Q+
= is the quality factor of the system, which is proportional
to the ratio of the stored energy over dissipated energy at the operating frequency. Based
on the definition of a fractional bandwidth, one can obtain:
17
)(1
0 i
rLH
bmkb
QFBW
+==
−=
ωωω . (2-11)
The complex damping factor b is contributed by the loss within the CMUT
structure as well as the medium loss. For CMUTs operating in immersion, medium loss
dominates, which the loss can be represented by the medium load impedance [14]:
om ZaKjKaaKZ 22
22
1++= , (2-12)
where λπ /2=K . It is seen from Equation 2-11 and Equation 2-12 that ib can be
minimized as Ka becomes very large (a >>λ ), such that a preferred large FBW can be
obtained. It implies the post area between vibrating plates should be minimized to limit
the imaginary part of the medium loading.
2.2.1.4 Transmission efficiency and maximum output pressure
The transmission efficiency, TXS , is expressed as the ratio of the pressure output
delivered to the medium through the piston movement of the plate to the electrical
actuation voltage. As the AC voltage excitation )t(Vac applies on the plate introduces an
electrostatic force, the equilibrium can be written as:
( ))(2)(22
))(( 222
02
20 tVVtVV
gA
gtVVA
F acdcacdcacdc
e ++=+
=−εε
(2-13)
The 2)(tVac term can be decomposed into a second harmonic term and a DC term.
The second harmonic term can be ignored for analysis at the fundamental frequency. The
resulting DC terms with the 2dcV set the static operation point. Thus, at a given DC
operation point, the )(2 tVV acdc term contributes to the acoustic pressure output:
)()()(
)()( 20 tnVtCEVg
tVAVtFAtP acac
acdc ≡=−==⋅ε , (2-14)
where g/AC 0ε= and g/VE dc= are the capacitance of the parallel plate
capacitor and the electrical field strength in the gap between the two electrodes. The
18
equation leads to the transmission efficiency, AntVtPS acTX == )()( (in units of Pa/V)
showing that CMUTs with a larger capacitance and stronger electric field (i.e. a small gap
g and larger bias voltage Vdc) have better coupling efficiency between the electrical and
acoustic domains in transmission.
Another essential parameter is the maximum output pressure. The output pressure
of a CMUT with frequency ω can be written as:
xZxZP mm ω)Re()Re( == , (2-15)
where x is the amplitude of the acoustic wave, which is equivalent to the CMUT
plate displacement. The maximum displacement, hence the maximum output pressure, is
limited by the gap between the two electrodes. Moreover, the pull-in effect of a parallel-
plate capacitor may further limit plate displacement. As a result, a large electrode gap is
desirable for higher output pressure; however, a small gap is needed for better
transduction efficiency. Thus there is a trade-off in terms of the gap-height for CMUT
design.
2.2.1.5 Receive sensitivity
In receiving mode, the input is the incident acoustic pressure impinging on the
plate, and the output is the current through the capacitance change (in units of A/Pa). The
electrostatic force is introduced by a bias voltage Vdc, which is used to set the operation
point of the device. The bias voltage applied on the CMUT also converts the capacitance
change into electric current, which is detected by the front-end circuit of the CMUT. The
receive current can be written as:
xgAV
tCVI dc
20)( ε
=∂
∂= . (2-16)
By plugging g/AC 0ε= , g/VE dc= , and )Re( xZP m = into (Equation 2-16), we
can find the formula for receive current becomes:
19
AZPCExCEIm
⋅⋅=⋅= . (2-17)
Thus, the receive sensitivity mm
RX ZnAA
ZCE
PIS =⋅== can be obtained. It is seen
from the expressions of TXS and RXS that both are proportional to the electromechanical
transformer ratio, n. A two-way sensitivity, RXTXS / in units of A/V, can be defined by the
product of TXS and RXS :
mRXTX Z
nS2
/ = . (2-18)
Therefore, for CMUTs sensitivity design, a small gap is desired. However, a large
gap is desired for maximum achievable output pressure. Thus, the gap height is one of
most critical device parameters and good control over it is required in manufacturing
CMUTs to avoid significant deviation of the device performance from the design target.
2.2.2 EQUIVALENT CIRCUIT MODEL
Mechanical systems can be converted into electrical circuits by using the analogy
between the mechanical and the electrical domains. One way to implement this analogy is
to replace the forces in the mechanical domain by voltage sources and velocities by
electrical currents. Then, an equivalent circuit of the system is constructed. This method
becomes an even more powerful tool for the analysis of electromechanical systems where
some parts of the system are already in the electrical domain. In addition, it will be
possible to use circuit simulation tools, like SPICE, to simulate the behavior of the
CMUT, individually or in connection with supporting electronic circuits.
mp Rloss mm
Rm
1/kp-1/ksoft
C0Cp
Rp
RS
V FZp
1 : n
+
- x
Q
Zm
Figure 2-4. Equivalent circuit model of CMUT (F=0 in transmit and V=0 in receive). C0: device capacitance, n: electromechanical coupling coefficient, Zp: plate mechanical
20
impedance, Zm: medium impedance, Rs: source impedance, Cp: parasitic capacitance, Rp: parasitic resistance.
Figure 2-4 shows the equivalent circuit for a CMUT transducer. In the electrical
part, C0 is the clamped capacitance of the device at the bias voltage. Spring softening
capacitance and the mechanical plate impedance constitute the mechanical part. The two
parts are coupled together through an electromechanical transformer (1:n). The lumped
elements representing a CMUT in the mechanical port can be obtained from the mass-
spring-damper system model. The CMUT top plate is modeled with a spring, pk , and a
mass, pm . The medium impedance is simply modeled using a damper, mR , and a mass,
mm . We can derive the transfer function of the CMUT in TX mode by setting F = 0 and
in RX mode by setting V=0 in Figure 2-4. The voltage across mR represents the force on
the top plate. Assuming that sR , pR , and pC are zero, and lossR and mm can be ignored,
the magnitude of the TX and RX transfer function are given by
21
22
202 )1()(1
−
⎥⎦
⎤⎢⎣
⎡−+== ∑ωωω
mTX R
mAn
VPS , (2-19)
and 21
22
202 )1()(1
−
⎥⎥⎦
⎤
⎢⎢⎣
⎡−+== ∑ωωω
mmRX R
mRAn
PIS , (2-20)
where ∑∑=mk
0ω , psoft kkk +=∑ , mpt mmm +=∑ , and 0
2
Cnksoft =
It is seen from the expression for the transfer function in Equation 2-19 and
Equation 2-20 that both the maximum sensitivities in TX and RX occur at 0ωω = . By
plugging 0ωω = into Equation 2-19 and Equation 2-20, it is not surprising the results are
the same as those derived from the parallel-plate capacitor model:
21
AnSTX =max, , and m
RX RnAS =max, (2-21)
2.3 CHARACTERIZATION METHODS FOR CMUTS
2.3.1 CHARACTERIZATION IN AIR
In air or vacuum, the medium loading on CMUT plates is negligible, and the
plates show good resonance behavior. The input impedance measurements, in these
cases, are helpful in determining the following parameters: resonant frequency, quality
factor, device capacitance, static collapse voltage, leakage resistance and series
resistance. These parameters, together with the plate and gap dimension measurements,
provide valuable insights to parameters that are more difficult to measure on finished
devices, such as Young’s modulus and residual stress in the plates.
The input impedance can be directly measured using an impedance analyzer, or
by measuring the scattering parameter (S11) using a network analyzer and converting to
impedance. In either case, a bias T is needed to enable the superimposition of an AC
voltage on top of a DC bias on the CMUT while protecting the measurement instrument
from being exposed to the DC voltage (Figure 2-5).
The real part of the input impedance peaks at the open circuit resonant frequency
(Figure 2-6). This frequency decreases with increasing DC bias voltage. This
phenomenon is due to the presence of parasitic capacitance. The amplitude of the
resonance peak will increase with increasing bias voltage due to the higher transduction
efficiency. When the electrostatic force surpasses mechanical restoring force from the
plate spring, the plate pull-in occurs. When this happens, the resonant frequency
increases significantly because of much stiffened plate spring [15], [16]. The pull-in
voltage can therefore be determined experimentally by noticing this sudden change in the
resonant frequency.
22
High voltage power supply
Bias “T”Network Anaslyzer
CMUT
Figure 2-5. Experimental setup for input impedance measurements.
Rea
l par
t of i
mpe
danc
e (Ω
)
Frequency (MHz)Im
agin
ary
part
of im
peda
nce
(Ω)
Frequency (MHz) Figure 2-6. Typical electrical input impedance of a CMUT element.
2.3.2 CHARACTERIZATION IN IMMERSION
For many applications, such as medical imaging, the CMUT is operated in
immersion or in contact with tissue. For these applications, some of the key
characteristics for the transducer are output pressure, receive sensitivity, center frequency
and fractional bandwidth. The output pressure and the receive sensitivity determine the
penetration depth; center frequency is determined by the application; the frequency
response determines the resolution of the system.
The transmit/receive behavior of a CMUT can be characterized in the pulse-echo
measurement (Figure 2-7). Usually, the transducer is biased at a certain DC voltage, and
a broadband electrical excitation pulse is superimposed on top of the DC voltage. A plane
reflector or the fluid-air interface is located in the far field of the transducer element. The
acoustic wave that bounces back from the reflector will change the CMUT capacitance,
causing a voltage change across the CMUT. This electrical signal is amplified and read
out onto an oscilloscope. A Fourier Transform can be performed on the received
waveform to determine the band shape (Figure 2-8). This band shape can be corrected for
23
the medium attenuation and diffraction as well as the excitation pulse shape to yield the
true transducer frequency response of the transducer [17].
The absolute value of the total output pressure can be measured with a needle type
hydrophone. Instead of a plane reflector, a hydrophone is placed perpendicular to the
transducer surface in far field. The hydrophone gives a pressure reading. The pressure at
the surface of the transducer can be inferred from this pressure level, after taking into
consideration of the medium attenuation and diffraction.
Oscilloscope Preamplifier
DC bias network
Function generator
High voltage power supply
CMUT
Water/oil tank
Use fluid/air interface as a reflector
Figure 2-7. Experimental setup for pulse-echo measurements.
5 10 15 20-30
-25
-20
-15
-10
-5
0
Frequency [MHz]
Nor
mal
ized
pre
ssur
e [d
B]
Impulse Frequency Response
1.6 1.8 2 2.2 2.4 2.6x 10-6
-4
-2
0
2
4
6 x 10-4
volta
ge (V
)
time (sec)
Pulse Signal
Figure 2-8. Typical pulse-echo waveform and band shape for a 185-um by 185-um CMUT element in a 2D array.
24
Other dynamic performance methods including a laser-interferometry for CMUT
plate displacement measurement is introduced in section 4.1. More static characterization
methods, such as SEM, AFM and a white-light interferometry, are discussed in detail in
section 2.4 as ways to evaluate the gap-height variation.
2.4 STATIC OPERATING POINT EVALUATION
It was discussed in section 2.2 the gap-height directly affects the static operation
point of the CMUT which in turn affects the receive sensitivity and the total possible
output pressure. In this section, the characterization of fabrication related gap-height
variations in CMUTs using sacrificial release process is reported (Figure 2-9). The
CMUTs under investigation were found to have lower pull-in voltage, smaller maximum
plate deflection and missing pull-in behavior as compared to theory. The in-cavity
deposition was examined by SEM and the surface topography on the bottom side of the
gap by AFM. The AFM measurements were evaluated for devices with and without in-
cavity-deposition as well as devices fabricated with doped polysilicon film or doped bulk
silicon for the bottom electrode. A methodology to determine the contribution of each
layer in the fabrication process to surface roughness is presented. The results show that
surface roughness reduces the actual gap-height, and the polysilicon layer is the main
contributor.
2.4.1 BACKGROUND AND MOTIVATION
To design CMUTs arrays for medical imaging, we optimize gap height, plate
radius, and plate thickness for the optimum total acoustic output pressure, transmission
efficiency, and reception sensitivity. All of those are strong functions of gap-height
(Table 2-1).
Table 2-1. Performance index as functions of gap-height
Performance Index Function of gap Acoustic output pressure ∝ gap Transmission efficiency ∝ 1/gap2 Reception sensitivity ∝ 1/gap2
For high frequency designs, a small gap height, as small as 50nm, is needed to
maintain a reasonable pull-in voltage. Fabricating such a small gap with the sacrificial
25
release process is very challenging. Any deviation from the expected gap-height could
lead to large changes in performance.
2.4.2 MATERIALS AND METHODS
Fabricating a CMUT using sacrificial release process is widely used because the
surface micromachining approach is compatible with the IC fabrication process. In
addition, the through-wafer-via, a well-established technology for the electronics
integration, can be easily incorporated into the process to enable a 2D array. Using a
CMUT array fabricated by this technology with a 10MHz center frequency design, a gap
of 120nm was chosen to maintain a reasonable pull-in voltage. The transducer’s
specifications are given in Table 2-2.
Table 2-2. Design parameters and values
Design parameter Target value
Center frequency 10MHz Radius 15.5µm Plate thickness 650nm Gap height 120nm Insulator thickness 180nm
To investigate the pull-in behavior and compare with the models, the frequency
response of the transducer and the deflection of the plate were measured under applied
DC voltage. The frequency response was measured using a vector network analyzer (HP
8751A) under applied DC voltage (SRS PS 310 high voltage power supply, Stanford,
CA). The static deflection was also measured using a white light interferometer (Zygo
NewView 2000) under applied voltage. The static deflection measurement enables the
detection of not only the plate pull-in but also the total maximum deflection that should
be equivalent to the gap height.
26
Si substrate
Poly-silicon
Doped polysilicon
LPCVD Si3N4
Thermal oxide
LTO
Aluminum
Through wafer via etch and thermal oxidation
Conductive poly-silicon deposition and doping
Poly-silicon fill
Doped poly-silicon and patterning
Si3N4 (etch stop layer) deposition
Poly-silicon deposition and active area patterning
Membrane deposition and etch hole opening
KOH release of the membranes
Seal the etch hole
Sputter and pattern the metal
Figure 2-9. Sacrificial release process: (a) Substrate doping, etch-stop layer deposition (LPCVD silicon nitride), first sacrificial layer deposition (LPCVD poly-silicon) and patterning, (b) second sacrificial layer deposition for reduced etch channel height regions, (c) active area definition, (d) plate deposition, (e) define sacrificial release etch holes and etch silicon nitride, (f) plate release in KOH, (g) plate sealing with more silicon nitride deposition, (h) top electrode deposition and patterning
It was verified that the process produced the desired gap-height by using a
scanning electron microscope (FEI XL30 Sirion SEM) to view the cross-section of a cell.
The gap-height was confirmed by measuring the height at the edge of the cell. This is
because other points at the center of the plate are deflected downwards from stress in the
nitride, so only the edge is the correct measurement of the gap-height. A SEM
measurement was also performed to view the surface roughness of the bottom side of the
gap by peeling off the CMUT plate.
In order to identify the source of the surface roughness, a Digital Instruments
Nanoscope 3000 Atomic Force Microscope (AFM) was used to compare devices with
27
and without in-cavity-deposition as well as devices with heavily doped polysilicon film
versus heavily doped bulk silicon as bottom electrode (Figure 2-10).
Unsealed-cavity CMUTs Sealed-cavity CMUTs
Doped-substrate CMUTsDoped-polysilicon CMUTs
vs.
vs.
(b)
(a)
Figure 2-10. The comparison of the surface roughness from the bottom side of the gap between (a) The CMUTs with and without sealing, and (b) The CMUT with doped polysilicon and thermal oxide and the one with doped substrate
Si substrate
Amorphous-silicon
Doped polysilicon
Silicon-nitride
Thermal oxide
A
B
C
D
E
�
�
�
�
�
�
Figure 2-11. A comprehensive surface topography characterization for each material layer at different process step. This approach start from five wafers (A, B, C, D, and E)
28
with one deposition step behind compared to each wafer process run (starting from step 1, 2, 3, 4, or 5).
─
─
─
Si substrate
Amorphous-silicon
Doped polysilicon
Silicon-nitride
Thermal oxide Figure 2-12. The methodology to quantify the topography/roughness change due to each layer from the full run device.
The results of the AFM on the actual device motivated us to do a more systematic
study using the AFM to measure the surface topography of different material layers at
different points in the process (Figure 2-12). Starting from five wafers, every wafer starts
from each offset process step. This approach enabled us to measure the roughness of
individually new-deposited layers (wafer A1, B2, C3, D4, and E5), and also to monitor
the surface roughness change through the real process run (wafer A1~A6).
In addition, the contribution of surface roughness from each layer is available
from the comparison as shown in Figure 2-12. Since the roughness is the combination of
the material itself in association with the experienced thermal cycles, the surface
roughness of interest has to include the effects of the neighboring layers during thermal
cycling. It is difficult to etch back layer by layer without altering the surface topography,
so we compared the difference of the results between the case with and without the layer
of interest.
29
2.4.3 RESULTS AND DISCUSSION
2.4.3.1 Static deflection and impedance measurements
Static deflection measurements and the frequency response were used to verify
the response of our designed plate with the model. According to the design, a pull-in
behavior around 65 V and a deflection of 120 nm was expected. Instead, it was found
that the pull-in behaviors were missing and the deflection saturated at 80 nm instead of
120 nm of the gap-height (Figure 2-13), which motivated to study the gap-height
variations.
Center Deflection V.S. Bias (P14-RingTest)
0
20
40
60
80
100
0 5 10 15 20 25 30 35 40
Bias Voltage (V)
Cen
ter
Defl
ecti
on
(n
m)
Real Part of Impedance (Ω)
Bias Voltage (V)
Frequency (MHz)
0 5 10 15 20 25 30 35 40
10
15
20
25
30 50
100
150
200
250
300
350
Bias Voltage (V)
Frequency (MHz)
Sign of the Imaginary Part of Impedance (Ω)
0 5 10 15 20 25 30 35 40
10
15
20
25
30
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Bias Voltage [V]
30
20
10
Peak
freq
uenc
y [M
Hz]
Bias Voltage
Peak
freq
uenc
y
Bias Voltage
Cen
ter D
efle
ctio
n
Bias Voltage [V]
0 40
0 40
Missing gap
Cen
ter D
efle
ctio
n (n
m)
No frequency jump
No deflection jump
Deflection jump
Frequency jump
Figure 2-13. Behavior of center deflection versus DC bias and real part of input impedance versus DC bias: (a) Theoretical part, and (b). Measurements part.
2.4.3.2 SEM measurements
The gap height from the edge of the cell was measured to be 120±10 nm from the
SEM picture [Figure 2-14(a)], which is very close to the design. The measurement
showed that the total in-cavity deposition was well controlled by the reduced-height
sealing channel.
30
In addition, SEM was used to view the bottom side of the gap [Figure 2-14(b)]
and a roughness was observed the same order as gap-height, which may account for the
gap-height variations.
120 ± 10 µm
Figure 2-14. (a) SEM photograph (cross-section) of one CMUT cell showing the gap-height, and (b) SEM photograph of surface roughness from the bottom side of the gap
2.4.3.3 AFM comparisons of various CMUTs
There are two sources of the roughness at the bottom of the cavity: in-cavity
deposition and individual roughness of various layers.
First it was compared the surface roughness of an unsealed and sealed cavity and
found very close surface roughness of 3.8 nm and 3.9 nm RMS [Figure 2-10(a)],
respectively. This indicates the sealant does not contribute to the surface roughness. Next,
it was investigated whether roughness of the underlying layers contributed to the
roughness by comparing the cases in Figure 2-10(b). CMUTs with doped polysilicon and
underlying thermal oxide showed 9.5 nm RMS, which is much higher than 3.9 nm RMS
of CMUTs with doped bulk silicon substrate. It means the cumulative roughness from
underlying layers significantly contributes to gap-height variation.
2.4.3.4 AFM differential comparisons of a process mimicking the real fabrication run
Given that the underlying layers contributed to the surface roughness, the
roughness of each individual deposited layer as well as combinations of layers through
31
thermal cycles in the process were investigated. First the contribution of individual layers
were investigated (Table 2-3: wafer A-1a, B-2a, C-3, D-4, and E-5). It was found that the
polysilicon produces the most roughness, 108±24 nm, which is 16-120 times greater than
the roughness of any other individual layer.
Table 2-3. RMS surface roughness measured by AFM
Process step Temp. Time
A B C D E
0 Prime wafer - 0.1 nm 0.1 nm 0.1 nm 0.1 nm 0.1 nm 1a Thermal oxide 1000oC
4h.50m 0.2 nm - - - -
1b Thermal cycles - - - - - - 1c Etch-backh to the
oxide - 3.0 nm - - - -
2a LPCVD polysilicon 620oC 1h.20m
17.5 nm 23.9 nm
- - -
2b P diffusion doping 900oC 1h.30m
10.4 nm 13.9 nm
- - -
3 LPCVD low-stress nitride
785oC 1h
9.7 nm 11.8 nm
1.2 nm - -
4 LPCVD amorphous silicon
560oC 57m
8.0 nm 10.5 nm
1.4 nm 0.3 nm -
5 LPCVD low-stress nitride
785oC 5h.05m
8.4 nm 9.9 nm 1.7 nm 1.5 nm 1.5 nm
6 TMAH wet-release - 9.2 nm - - - -
Next, it was investigated the roughness of successive layers during processing
(Table 2-3: wafer A0~A6). A-1a shows that the oxide layer itself does not add roughness,
and then the surface became rougher with 3 nm RMS (Figure 2-15) in A-1c. The reason
is that the oxide has glass transition temperature is between 900 oC to 1100 oC [18]. The
oxide was softened due to the much lower viscosity at 1000 oC. Meanwhile, the grain
growth of the polysilicon layer punctures the oxide. Consequently, the oxide is imprinted
by polysilicon during high temperature and left with a higher roughness.
32
Figure 2-15. Thermal Oxide is roughened due to imprinting by overlying polysilicon
layer
In Table 2-3: A-2a, the deposition of LPCVD polysilicon and doping with
phosphorus increases surface roughness from 3 to 17.5 nm RMS. It is the largest
roughness rise through the whole process. In contrast, the largest reduction of surface
roughness occurred after the doping step, which reduced the value from 17.5 nm RMS to
10.4 nm RMS (Table 2-3: A-2b). This reduction may be a result of the 900 °C annealing
used to dope the polysilicon, which not only joined and enlarged the grain size in x-y
direction but also relaxed the high surface energy from columnar grain structure [19] to
the typical hemisphere grain shape (Figure 2-16) [20].
33
Figure 2-16. Doping reduces surface roughness of polysilicon
The relative smooth nitride layer was grown after the doping, which provide a
conformal deposition and results in a smoother surface from 10.4 nm RMS to 9.7 nm
RMS [Table 2-3: A-3]. Through steps A-4~6, the nitride shows a surface roughness
value of 9.2nm RMS, which is very close to our measurement from actual devices.
Figure 2-17. Roughness contributed from doped polysilicon.
2.4.3.5 Roughness contribution from various layers
In order to see the contribution of surface roughness from each layer, it was
compared three pairs of wafers with and without the layer of nitride, thermal oxide and
34
polysilicon. Both of the cases with and without nitride and oxide [Figure 2-14(b-I, III)],
were similar in surface roughness, measuring 1.7 nm RMS versus 1.5nm RMS and 8.4
nm RMS versus 9.9 nm RMS (Table III: C-5 vs. D-5 and A-5 vs. B-5), respectively.
Compared to the case shown in figure 4b-II, the roughness 9.9 nm RMS with polysilicon
layer compared to 1.7 nm RMS without polysilicon (Table 2-3: B5 vs. C5) proves that
the doped polysilicon layer is the main contributor to the final surface roughness.
2.4.4 CONCLUSIONS
The characterization of fabrication related gap-height variations in CMUTs was
studied in this section. The results show that the surface roughness reduces the actual
gap-height, and the polysilicon is the layer most responsible for the increased surface
roughness. Therefore, surface roughness should be accounted for in small gap designs if
the nitride sacrificial release method is used. Otherwise, an SOI wafer bonding for thin
gap devices is preferred for CMUTs with small gap height design.
35
CHAPTER 3 FRONTSIDE INTERFACE ENGINEERING
OF CMUTS
The packaging of a medical imaging or therapeutic ultrasound transducer should
provide protective insulation while maintaining high performance. For a capacitive
micromachined ultrasonic transducer (CMUT), an ideal encapsulation coating would
therefore require a limited and predictable change on the static operation point and the
dynamic performance, while insulating the high DC and AC actuation voltages from the
environment. To fulfill these requirements, viscoelastic materials, such as
Polydimethylsiloxane (PDMS), were investigated for an encapsulation material. In
addition, PDMS, with a glass-transition temperature below room temperature, provides a
low Young’s modulus that preserves the static behavior; at higher frequencies for
ultrasonic operation, this material becomes stiffer and acoustically matches to water. This
chapter presents the modeling and implementation of the viscoelastic polymer as the
encapsulation material. A finite element model (FEM) is introduced that addresses
viscoelasticity. This enables a correct calculation both the static operation point and the
dynamic behavior of the CMUT. CMUTs designed for medical imaging and therapeutic
ultrasound were fabricated and encapsulated. Static and dynamic measurements were
used to verify the FEM, and show excellent agreement. This work will help in the design
process for optimizing the static and the dynamic behavior of viscoelastic-polymer-
coated CMUTs.
3.1 ENCAPSULATION DESIGN
The packaging of a micromachined ultrasonic transducer is crucial for medical
applications. This packaging material should electrically insulate the device from the
environment, protect the device against humidity and other corrosive substances, and also
be biocompatible. At the same time, this encapsulation should maintain transducer
performance, especially the transfer of energy to and from the medium. While other
microelectromechanical system (MEMS) devices such as accelerometers and gyroscopes
36
are typically protected from the environment by hermetic sealing, CMUTs require
mechanical interaction to a fluid (Figure 3-1), which makes fulfilling the previous criteria
very challenging. In this work, it was discussed the requirements on such a coating
material for a CMUT used in immersion applications. Then it was presented a model of a
viscoelastic material (e.g. PDMS), and the model was verified using experiments.
l Accelerometerl Gyro
?Immersion
Hermetic sealing
Air window l Pressure sensorl Microphone
l Optical sensorl Micromirrorarray
Glass plate Optical fiber
l Ultrasonic transducer− Medical imaging− Therapeutic ultrasound
(a)
(b)
(c)
(d)
Figure 3-1. Encapsulation approach for various MEMS devices.
3.1.1 STATIC OPERATION AND DYNAMIC PERFORMANCE OF ENCAPSULATED CMUTS
CMUTs have demonstrated promising performance, ease and flexibility of
fabrication, and ease of electronics integration, which makes them advantageous over
conventional piezoelectronic transducers [21], [22]. The encapsulation material of
CMUTs should provide the matching acoustic impedance to the medium (1.5 MRayl for
water or 1.63 MRayl for average human soft tissue [23]) to enable maximum
transmission of energy into the medium. Based on transmission line theory (Equation
3-1), the incident wave transmit efficiency is strongly related to the acoustic impedance
matching condition [14].
37
2)(1cm
cmenergy ZZ
ZZT+−
−= (3-1)
where energyT , cZ , and mZ are the energy transmission ratio, and the acoustic
impedance of the coating material and the medium respectively. The matching acoustic
impedance can be further related to the Young’s modulus based on the elasticity theory
[14].
)1()21)(1(12
υυυ
ρ −−+⋅⋅= ZE , (3-2)
where E , υ , Z , ρ represent the Young’s modulus, Poisson’s ratio, acoustic
impedance, and density, respectively. In order to match the acoustic impedance of 1.5
MRayl, a Young’s modulus on the order of GPa (with a Poisson’s ratio of 0.40 and
density of 1000 kg/m3) is needed (Equation 3-2).
Operation of a CMUT requires both a DC and AC voltage. The DC voltage sets
the operation point of the device, and establishes the transmit and receive sensitivities
[24] and the electromechanical coupling efficiency [25]. Since this point determines the
performance of the device, it is important to predict and limit the impact of the
encapsulation material on the pull-in voltage variation. In addition, a practical thickness
of the coating should be of the order of hundreds of microns to protect the device against
the regular contact ablation force and the electrical breakdown. Considering these factors
and the Young’s modulus of the CMUT plate material, a Young’s modulus for the
coating material on the order of MPa is essential to limit the pull-in voltage variation.
Fulfilling the AC and DC requirements of this encapsulation material are
challenging unless the coating materials are used that exhibit viscoelasticity. This work
presents the first modeling and verification of the viscoelasticity of polymers for CMUT
immersion imaging applications.
3.1.2 VISCOELASTICITY OF POLYMER
Many polymers used for MEMS applications exhibit viscoelastic behavior [26],
[27], [28]. The Young’s modulus and Poisson’s ratio change as functions of loading
38
frequency. This occurs because a relaxation time is needed under a corresponding strain
rate. As the loading frequency increases, the behavior of the polymer changes from
rubbery to glassy [26], resulting in a change in the Young’s modulus and the Poisson’s
ratio from the slow limit values ( sE , sυ ) to the fast limit ones ( fE , fυ ). The Young’s
modulus can differ up to several orders of magnitude, with the modulus being stiffer at
higher frequencies [27]. Poisson’s ratio decreases from close to 0.5 down to 0.35~0.4
[29], leading the )21)(1()1(υυ
υ−+
− part of Equation (3-2) to reduce by one order of magnitude.
Due to these changes, the acoustic impedance increases from hundreds of kRayl up to
MRayl range as the loading frequency is increased.
For CMUTs that operate in the MHz range, a preferred rubber to glassy transition
frequency should be smaller than 1 MHz [Figure 3-2(a)]. At low frequencies, the
Young’s modulus is in the MPa range; as the loading frequency is increased to the
ultrasonic region, the Young’s modulus [Figure 3-2(a)] approaches GPa, which translates
to a good matching acoustic impedance. In terms of temperature, a material should be
chosen so that it is in the rubbery state at room temperature for static loading and a glass
transition temperature that is lower than room temperature [Figure 3-2(b)] [26], [27].
(a)
100 kRayl
Loading frequency
Fast
Slow
E Z
1 MRayl
MPa
GPa
1MHzStatic
(b)
Rubbery
Glassy1 MRayl
100 kRaylMPa
GPa
Temperature Room Temperature
E Z
Tgfg
Figure 3-2. Viscoelastic curves of PDMS in terms of the Young’s modulus and acoustic impedance over (a) loading frequency, or (b) temperature based on the theory of time-temperature superposition. Tg and fg represent the glass-transition temperature under static loading, and the glass-transition loading frequency in room temperature, respectively.
Among the polymers listed in Table 3-1 [28], PDMS fulfills the needed
requirements because its glass transition temperature ( gT ) is much lower than the room
39
temperature. This characteristic provides the low static Young’s modulus for preserving
the CMUT’s pull-in voltage; at the same time it offers the matching acoustic impedance
needed for efficient acoustic transmission. Finally, PDMS is a biocompatible substance
that provides protection against biofouling, thermal instability, and dielectric breakdown
[9]. However, since it is not photo-patternable and plasma-etchable, a mold-casting
technique will be utilized to fabricate these encapsulation layers.
Table 3-1. Properties of various polymer materials commonly used for MEMS applications [28], [30].
Glass transition Modulus Density Temperature
(oC) (GPa) (g/cm3)
PDMS -125 ~0.001 1.00~1.64 Polyimide 390 ~3.3 1.42~1.53 EPON SU-8 194 ~4.5 1.2 PMMA 105 ~2.5 0.9 Parylene 160 ~2.5 1.1~1.4
PDMS has been routinely used as a lens material for piezo-electric ultrasound
transducers [31]; processing techniques and material properties of PDMS are well
understood. Previously, polymer coatings including PDMS were used to electrically
insulate CMUTs in immersion applications [32][33][34][35]; to create flexible arrays
[36][37] and to improve acoustic crosstalk suppression [38][39][40][41]. However, the
study of using polymer coatings on a CMUT has been based on a “go and redesign” base
such that a fundamental and systematic study providing a design rule has been absent. In
addition, while analytical and FEM modeling of CMUTs with different shapes, gap
heights, plate topologies, plate stress have been used to optimize bandwidth, output
pressure and sensitivity [42][43][44], none of these studies have considered the effects of
the viscoelasticity of coating material layers on the CMUT’s performance. In order to
successfully perform a CMUT design that accounts for the effects of PDMS coating
layers, a FEM was introduced that considers viscoelasticity and can be used to analyze
the pull-in voltage and frequency response of CMUT for immersion.
40
3.2 FABRICATION PROCESS
3.2.1 CMUT FABRICATION
The bandwidth and transmission efficiency of CMUTs can be optimized for
diagnostic and therapeutic ultrasound, e.g. high intensity focused ultrasound (HIFU), by
changing the cell size, gap height and plate thickness. A PDMS coating on different
geometry CMUT cells has different effects. In order to demonstrate the implementation
of PDMS encapsulation on CMUTs designed for different applications, both imaging and
HIFU CMUTs were modeled, designed, and fabricated.
An imaging CMUT should be designed to provide high center frequency with
wide bandwidth. A relatively thinner plate thickness and gap height with a smaller cell
size were used. HIFU CMUTs are required to operate at single frequency but generate
high output pressure, so a thicker plate and larger gap height were used (Table 3-2) [45].
Table 3-2. Parameters of the CMUTs for imaging and HIFU.
Device Imaging HIFU Configuration 1D linear array Single element Radius of CMUT cell 21.5 µm 50/ 70 µm Thickness of Si plate 2.0 µm 6.0 µm Gap height 0.15 µm 0.40 µm Thickness of SiO2 insulator 0.30 µm 0.60 µm
Among the two developed CMUT fabrication technologies available, wafer
bonding and sacrificial-release techniques [46][47], the fusion-bonding approach was
chosen since the plate thickness was 6 microns. In the sacrificial release process, forming
such a thick plate using chemical vapor deposition (CVD) is not possible due to residual
stress and deposition time for cells of those lateral dimensions. Figure 3-3 shows the
process used for fabricating the CMUT arrays based on a wafer bonding technique. First,
the highly conductive Si wafers (0.01-0.025 ohm-cm) are selected. Then 400-nm and
150-nm-high cavities are formed by a two-step thermal oxidation-and-etch, a buffered
oxide etch (BOE) and then a dry plasma etch. The second step etching is to remove the
bird beak formed by the second oxidation step. Next, two kinds of silicon-on-insulator
(SOI) wafers are selected, with the device layer thicknesses of 6 µm and 2 µm,
respectively. The SOI wafers are then bonded on top of the etched wafers using fusion
41
bonding. The handle wafers are then removed using wafer grinding (Silicon Quest
International, Inc., Santa Clara, California) and then wet etching in a hot
tetramethylammonium hydroxide (TMAH) solution. The buried oxide layer is then
removed by BOE, leaving the 6-μm and 2-μm plate covering the cavities. For the
imaging CMUT, elements within an array are defined by separating the silicon plates
with a dry plasma etch step. Then aluminum is deposited, and hot and ground electrodes
are defined on the plate and the silicon substrate, respectively. Figure 3-4 shows various
views of the finished devices. The last step of the fabrication is the PDMS encapsulation
of CMUTs. To optimize the PDMS coating in terms of the coupling efficiency, the
adhesion issue was discussed in detail in section 3.2.2, and various coating techniques
were covered in section 4.2 and 4.3.
Si SiO2 Al PDMS
(1)
(2)
(3)
(4)
(6)
(a) (b)
(5)
Figure 3-3. CMUT device fabrication using wafer bonding technique for (a) imaging CMUTs and (b) HIFU CMUTs. The process flow include (1) first oxidation and BOE oxide etch, (2) second oxidation and dry etching, (3) wafer bonding, (4) handle wafer and BOX removal and etch back for substrate access, (5) top electrode deposition and patterning, and (6) PDMS coating.
42
(a) (b) (c)
(d) (e) (f)
Silicon plate
Vacuum cavity
SiO2 insulator
Vacuum cavity
Silicon plate
SiO2 insulator
Aluminum
Aluminum
Si substrate
Si substrate
x (µm)
y (µm)
x (µm) y (µm)
Disp
. (nm
)
Disp
. (nm
)
Ground wire
Signal wire
Ground wireSignal wire
One array element
One CMUT cell
One CMUT cell
Figure 3-4. Imaging CMUTs pictures of (a) SEM cross sectional view of a cell, (b) 6 elements of one array with wire bonding, and (c) CMUT cells with plate deflection under DC bias of 80% of pull-in voltage measured by white light interferometer. Similarly, HIFU CMUTs pictures are shown in (d)-(f).
3.2.2 PDMS ENCAPSULATION
Polydimethylsiloxane (PDMS) is known for its useful properties, such as
flexibility, low Tg, and low surface energy [48]. These characteristics are associated with
the requirements for CMUT encapsulation; a Tg lower than room temperature was
required to maintain both static and dynamic performance of CMUTs. Good adhesion,
through the modification of surface energy, was essential for the energy coupling
between the CMUT/PDMS interface. To better utilize those characteristics, it is
important to understand the chemistry of the individual elements of the polymer as well
as the behavior of the functional group.
PDMS is formed by alternating silicon and oxygen atoms as backbone with
functional group of methyl (Figure 3-5). The flexibility comes from the bendable Si-O-Si
bond, which can vary between 135o and 180o. Bending flexibility occurs when there is a
large hindrance between non-bonded atoms, where there are unfavorable torsion angles.
43
Flexibility has an important role in the low Tg; the higher the flexibility, the lower the Tg.
These polymers can be made less flexible for better abrasion resistance by adding bulky
side groups, while resulting in a higher Tg.
Figure 3-5. Chemical structure of PDMS
Low surface energy, or surface tension, is another key feature of PDMS,
influencing the surface wettability and thus the adhesion. The mechanism for the low
surface energy is that the methyl groups have virtually no interactions with each other.
Alternatively, the low surface energy can be related to the low boiling point.
The adhesion is essential for an efficient ultrasonic energy coupling through the
CMUT/PDMS interface. PDMS can be surface treated to be hydrophilic for better
adhesion. Adhesion studies for two scenarios, the cross linked PDMS and the semi liquid
form of PDMS, were discussed in the following sections.
3.2.2.1 Adhesion on cross-linked PDMS
An efficient microfluidics device, with PDMS-formed microchannel, requires
sidewall wettability. Researches on microfludics have demonstrated knowledge to
improve the hydrophilicity of PDMS surface, through the roughness manipulation or the
functional group grafting [49]. A plasma treatment, roughening the surface (Figure 3-6),
can increase the local contact angel, and enable the hydrophilicity in macro behavior
(Figure 3-7). A formation of silanol group serves the same purpose to increase the
wettability of PDMS surface (Figure 3-8). The aging induced hydrophobicity recovery
(Figure 3-10) can be prevented with a further HEMA grafting, creation of the hydroxyl
terminated function group (Figure 3-9).
44
Figure 3-6. SEM of PDMS surface (a). Before plasma treatment, and (b). After plasma treatment (after [49]).
Figure 3-7. Relationship between the surface topography in micro scale and hydrophilic behavior in macro scale (after [49]).
Figure 3-8. Silanol group of PDMS surface treatment (after [49]).
Figure 3-9. HEMA grafted PDMS: creation of the hydroxyl terminated function group
(after [49]).
45
Figure 3-10. Aging induced hydrophobicity recovery effect on PDMS with various surface treatments (after [49]).
3.2.2.2 Adhesion on semi liquid PDMS
Researches on adhesive bonding have studied the adhesion of the semi liquid
form of PDMS to the substrate. PDMS was utilized as a bonding agent for the benefits of
low bonding temperatures and insensitivity to the topology [50]. Adhesion between the
semi liquid PDMS adhesive and the substrate is from the covalent and Van Der Waals
bonds, where the atoms of two opposing surfaces must be less than 0.3–0.5 nm apart
[50]. In light of the proximity requirement, the wetting of the surfaces is critical; the solid
surface must have a greater surface energy than the liquid for wetting to occur. Typically,
a primer, silicone powder content mixed in a methanol and ethanol solvent, was required
to improve the wettability. The solvent provides low viscosity with less than 10
centipoises (cP) compared to over thousands of cP of semiliquid PDMS. Low viscosity,
along with the ultralow surface energy (Figure 3-11) versus the solid substrate, helps the
silicone content to fill with the troughs of the surface profile for closer contact distance.
46
TungstenDiamond
PlatinumSilicon
AluminumTin
Glass
Ice
Epoxies
PolyethyleneTeflon Ethanol/ Methanol
EpoxiesPolyurethanes
Water Glycerol
Surface Energy (mJ/m2)
Figure 3-11. Surface energy of various solids and liquids.
3.2.2.3 Primer induced metal interconnections failure
CMUT behaves like a parallel plate capacitor, where a DC bias was required
between two electrodes. There are four possible biasing schemes: electrically grounding
the top or bottom electrode, and biasing another electrode with positive or negative
polarity (Figure 3-12). It was reported thermal decomposition of ethanol could increase
the surface conductivity [51]. Consequently, the biasing scheme needs to be carefully
selected to prevent from the current leakage induced interconnection failure.
Electrical input impedance was performed with all four biasing schemes. DC bias
was swept from 0 to 190 V, while the CMUT cells and interconnections were inspected
under the optical microscope for integrity. It was observed only the case with biased top
electrode and positive polarity suffers the interconnection failure due to the loss of
Aluminum at DC of 190 V (Figure 3-13). Cause to failure was suspected due to an
electrochemical reaction comparable to electromigration, because the failure only
happens with positive polarity when the top electrode is biased.
47
Conductive polysilicon
NitrideAluminum
Primer
Conductive polysilicon
NitrideAluminum
Primer
Conductive polysilicon
NitrideAluminum
Primer
Conductive polysilicon
NitrideAluminum
Primer
Figure 3-12. Four biasing schemes for CMUT operation.
Real Part of Impedance (Ω)
Bias Voltage (V)
Freq
uenc
y (M
Hz)
60 80 100 120 140 160 180
2
4
6
8
10
12
14 -8
-6
-4
-2
0
2
4
6
8
10
x 105
Bias Voltage (V)
Freq
uenc
y (M
Hz)
Sign of the Imaginary Part of Impedance (Ω)
60 80 100 120 140 160 180
2
4
6
8
10
12
14-1
-0.5
0
0.5
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.2
0.4
0.6
0.8
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.2
0.4
0.6
0.8
1
Bias Voltage (V)
Freq
uenc
y (M
Hz)
Real part of impedance (Ω)
Bias =0~180V Bias >190V
Figure 3-13. Electrical input impedance results and the optical microscopic images of the CMUTs before and after the surface-conductivity-induced interconnection failure.
48
3.2.2.4 Plasma etching for surface treatment
Alternatively to the primer, a plasma etch on the substrate can activate the surface
to be hydrophilic. Treatments in oxygen plasma can remove organic residue to improve
the surface wetting while forming a thin oxide layer at the bonding interface [52], [53].
Surfaces etched in an Ar/H2 plasma showed an appreciable increase in the bonding
strength after a thermal desorption of hydrogen at temperatures between 400 °C and 600
°C [54], [55]. SF6/O2 mixtures plasma etching can result in a surface layer of silicic acid
(SiO2⋅nH2O), with the presence of moisture, which leads to the formation of Si-OH bond,
behaving hydrophilic [56].
Among the plasma etching options, Ar/H2 is not favorable because the presence of
the aluminum for the current design. Hence, the oxygen plasma and SF6/O2 will be
utilized for contamination removal and surface modification to improve the adhesion.
3.2.2.5 Encapsulation assembly
To encapsulate CMUT with PDMS, the devices were diced and then wire bonded
to a custom-designed printed circuit board (PCB) after device level fabrication. Before
application of the encapsulation material, we performed an oxygen plasma cleaning step
on the CMUTs surface and applied a primer (GE SS4120) on the PCB to increase the
interface adhesion. Then a 150-μm-thick layer of GE RTV 615 PDMS was mold casted
by pressing a Polystyrene plate on the CMUTs with a spacer. This was followed by a
vacuum chamber degassing step, a curing cycle at 50oC for 12 hours, and a mold-
releasing step. Finally, the exposed wires for interconnections were encapsulated by
epoxy for water-immersion tests. The CMUTs for imaging without PDMS coating layers
were also prepared for comparison.
3.3 VISCOELASTIC FINITE ELEMENT MODEL
In order to properly account for the viscoelastic property of the PDMS on the
CMUT plates, a FEM was developed (ANSYS 11.0, ANSYS Inc., Canonsburg, PA).
Previously, Lohfink et al. demonstrated an axial symmetric model of a single circular
CMUT plate with a wave-guide liquid column [43]. However, they did not consider the
PDMS coating. This modeling approach was extended by adding a PDMS layer on top of
49
the CMUT plate, and by accounting for the viscoelastic effects by creating two variations
of FEMs.
The first FEM (non-viscoelastic FEM) considers the CMUT with the coating
material without viscoelasticity. We used PLANE42 elements [57] for both the plate and
the PDMS layer [Figure 3-14(a)], using either sE or fE as an appropriate material
property for static or frequency response simulation, respectively. In a second FEM
(viscoelastic FEM), which considers the viscoelasticity, we used the element type
PLANE 182 [57]. PLANE 182 is a 2-D 4-node structural solid element as PLANE 42,
but has special feature to address viscoelasticity.
PLANE182 can feature a linear viscoelastic model based on the generalized
Maxwell elements, a viscous damper and a purely elastic spring connected in series.
Bergstrom, et al. [58] introduced a micro-mechanism about this model by illustrating the
relaxation of the free chains within the polymer network. This model adequately
describes the constitutive model of many amorphous polymers by decomposing the
mechanical behavior into two parts: an equilibrium network under long time stress
relaxation and a second network capturing the rate-dependent deviation from the
equilibrium state. The Prony series expansion was used in ANSYS to express this model
mathematically. The series constants were implemented in ANSYS code by using “TB,
PRONY” command to determine the viscoelastic moduli as functions of time, )(tE and
)(tυ [57], [59]:
)exp()()(0Tt
sfs EEEtE −⋅−+= ; (3-3a)
)exp()()(0Tt
sfst −⋅−+= υυυυ ; and (3-4b)
11)1()21)(1(CE
f
fff ⋅
−−+
=υ
υυ, (3-5c)
where 4.6=sE MPa [30], 48.0=sυ , 40.0=fυ [29], 211 lVC ⋅= ρ , lV is the
longitudinal wave velocity, and 0T =1e-4 second, the relaxation time constant selected
based on the glass-transition loading frequency in room temperature [27]. The equivalent
50
mechanical model of the Prony Series expansion used in our model comprises a spring in
parallel with a series set of a spring and a dashpot [Figure 3-14(c)].
Depending on the loading condition (DC or AC superimposed on DC) of the
CMUT plate, a static analysis and transient response analysis was developed. From the
static analysis the pull-in voltage was obtained, which then was compared to the
measured pull-in voltages of the fabricated devices with and without PDMS coating. The
transient response analysis was used to calculate both the time-history and frequency
response. A transient analysis rather than a harmonic analysis was chosen to analyze the
effect of the echoes from the PDMS-water interface due to the acoustic impedance
mismatch. Voltage loading was applied in two steps. First, a voltage step with a ratio of
the calculated pull-in voltage (80% for the imaging CMUTs and 60% for the HIFU
CMUTs) was used. Then, after waiting for one second, a unipolar 10-Vpp 50-ns square
pulse was superimposed. The ambient pressure of 101,325 Pa was applied before the
voltage load to consider the fact that the CMUTs feature an evacuated gap. The
displacements from the plate and the PDMS surface were retrieved. The time derivative
of the average displacement of the CMUT plate was calculated and then multiplied by the
medium acoustic impedance to calculate the average surface pressure response. A FFT
was performed on the Gaussian-windowed pressure pulse to calculate the frequency
response. The attenuation from the PDMS layer was then compensated based on the loss
[60]:
Loss in dB = rf ⋅⋅ βα , (3-6)
where α = 0.4 dB/MHz/cm for GE RTV 615, β = 1.4, f is the frequency in
MHz and r is the PDMS thickness in cm.
First, evaluating the differences between the reflected echoes from the PDMS-
water interface of a 75-μm thick coating and a 200-μm thick coating tested the transient
model. Then the effect of the acoustic impedance mismatch from different types of
PDMS (GE RTV 615 and Sylgard 160, Table 3-3) was analyzed. Next, the simulation
results were compared to measurement results obtained from fabricated devices with
different designs (Table 3-2). To demonstrate how the coating material without
51
viscoelasticity fails to preserve the CMUT operation point and performance, we
processed the same calculations with the FEM that considers the coating without
viscoelasticity.
Table 3-3. Acoustic properties of PDMS.
Type Vl (m/s) ρ (kg/m3) Z (MRayls) GE® RTV 615 1080 1020 1.1 Sylgard® 160 950 1580 1.5
(b) (c)(a)
ssE υ,),( sf EE −
FLUID 29
PLANE 42
PLANE 42
spring
dashpot
)( sf υυ −
TRANS126
Perfect absorbing boundary
Fluid-structural coupling layer
FLUID 29
PLANE 182
PLANE 42
TRANS126
Perfect absorbing boundary
Fluid-structural coupling layer
Figure 3-14. Schematic of (a) the non-viscoelastic and (b) the viscoelastic FEM. FLUID 29, PLANE 42, PLANE 182, and TRANS 126 represent the element types used in the ANSYS model. The plots also show the roller boundary conditions (constrained to x-direction), and the anchor boundary conditions (constrained to both x and y directions). (c) The Prony Series expansion used for PLANE 182 element in the viscoelastic FEM can be represented by a spring in parallel with a series set of a spring and a dashpot.
3.4 RESULTS AND DISCUSSION
3.4.1 CHARACTERIZATION FOR STATIC BEHAVIOR
The pull-in voltage can be measured by sweeping the electrical impedance
frequency for resonance shift or capacitance change [61], or by observing the plate
deflection under white light interferometer. By using the impedance analyzer (Model
4294A, Agilent Co., Palo Alto, CA), both real part and imaginary part of the electrical
impedance can be measured for pull-in voltage characterization.
52
According to the CMUT equivalent circuit model, the CMUT cell capacitance,
active part along with the parasitic, can be retrieved from the imaginary part of the
impedance at the frequency away from the resonance. The capacitance-bias voltage
measurement was selected when the thick PDMS layer damps the resonance and opaque
type of PDMS blocks the optical access. The pull-in of the plate causes a discontinuity in
the capacitance bias-voltage curve (Figure 3-15).
Compared to the capacitance calculated from the imaginary part of the
impedance, resonance from the real part can also be used to characterize the pull-in
voltage by showing the resonance shift. The pull-in voltage measurement results were
compared with the results from the FEM models with and without the coating layer.
50 100 1509
9.5
10
10.5
11
50 100 1509
9.5
10
10.5
11
50 100 1506.5
7
7.5
8
8.5
50 100 1506.5
7
7.5
8
8.5
Measurement: No coating
Measurement: 150-mm coating
Cap
acita
nce
(pF)
Cap
acita
nce
(pF)
Bias voltage (V)
Pull-in at 142 V
Pull-in at 142 V
Previous FEM : No coating
150-mm coating
Viscoelastic FEM ornon-viscoelastic FEM with Es: Pull-in at 142 V
Pull-in at 142 V
Cap
acita
nce
(pF)
Cap
acita
nce
(pF)
(a)Bias voltage (V)
(b)
non-viscoelastic FEM with Ef: Pull-in at 155 V
Figure 3-15. Static behavior verification: capacitance vs. DC bias voltage for the imaging CMUTs: (a) The experimental data, and (b) the modeling data. The measured capacitance larger than the modeling is due to the parasitic capacitance from the electrical interconnections of the experimental setup.
53
The capacitance and the resonance were measured with and without the 150-μm
PDMS coating to verify the prediction of static operating point. The pull-in voltage of the
device without PDMS was 142 V [Figure 3-15(a), Figure 3-16]. With PDMS, a group of
resonances were observed instead of a single one (Figure 3-17). First, capacitance was
calculated from the imaginary impedance value at the frequency away from the
resonance. The pull-in voltage was found at 142 V [Figure 3-15(a)]. Then, the real part of
the impedance was examined, showing the multi-resonance peaks, which come from the
reverberations within the PDMS layer. The pull-in of the plate causes a discontinuity to
the multi-resonances at 142 V [Figure 3-18(b)]. A pull-in voltage of 142 V was found in
both approaches, thus demonstrating that the PDMS coating is significantly less stiff than
the plate at DC and does not affect the pull-in voltage. For non-viscoelastic FEM, only
the model with sE can fit the measurement while fE provides overestimated stiffness
and results in higher pull-in voltage [Figure 3-15(b)]. The viscoelastic static modeling
result [Figure 3-15(b)] is consistent with the measurements by showing the same pull-in
voltage of 142 V. This confirms that the model can simulate the PDMS modulus as the
slow limit under a static loading and correctly predicts that a 150-μm PDMS coating does
not alter the pull-in voltage.
5 10 15 20 25 30 35 400
1
2
3
4
5
Impe
danc
e [kΩ
]
Frequency [MHz]5 10 15 20 25 30 35 40
-5
-4
-3
-2
-1
0
1
2
Impe
danc
e [kΩ
]
Frequency [MHz]
Pull-in at 142 V
(a) (b) Figure 3-16. The electrical impedance measurement of the CMUTs without coating, showing the (a) real parts, and (b) imaginary parts at DC bias before and after the pull-in.
54
5 10 15 20 25 30-3
-2.5
-2
-1.5
-1
-0.5
0
Impe
danc
e [kΩ
]
Frequency [MHz]5 10 15 20 25 30
0
0.1
0.2
0.3
0.4
0.5
Impe
danc
e [kΩ
]
Frequency [MHz]
Pull-in at 142 V
(a) (b) Figure 3-17. The electrical impedance measurement of the CMUTs with 150µm PDMS coating, showing the (a) real parts and (b) imaginary parts at DC bias before and after the pull-in.
20 40 60 80 100 120 140
5
10
15
20
25
30
50
100
150
200
250
300
350
400
450
500
550
Freq
uenc
y [M
Hz]
Bias voltage [V]
With Coating
20 40 60 80 100 120 140
5
10
15
20
25
30
35
0
500
1000
1500
2000
2500
3000
3500
4000
Freq
uenc
y [M
Hz]
Bias voltage [V]
Without coating
(a) (b) Figure 3-18. The real part of the electrical impedance versus the DC bias measured from the CMUTs (a) without and (b) with PDMS coating.
3.4.2 MODEL PARAMETER STUDIES
First, the design parameter study by using the viscoelastic FEM was
demonstrated. It was first verified the speed of sound and the effects from mismatched
acoustic impedance in the transient model. This model uses the design parameters from
the CMUT for imaging as an example.
3.4.2.1 Effects from different PDMS coating thicknesses
The arrival times of the first impulse response wavelets (main signals) [Figure
3-19(a) and (b)] are 185 ns and 69 ns for the 200-μm and 75-μm thick coatings,
respectively. For both cases, 1080 m/s of the speed of sound of PDMS was calculated, by
55
using the travel time and the distance from the CMUT surface (plate-PDMS interface) to
the PDMS-water interface, i.e., coating thickness. This speed matches with the input
parameter, the longitudinal wave velocity of GE RTV 615 PDMS. The speed also
confirms that the excited wave is traveling as the longitudinal wave instead of the higher
order, extensional wave.
Because of the acoustic impedance mismatch between the GE RTV 615 PDMS
(1.1 MRayl) and water (1.5 MRayl), echoes from the PDMS-water interface are present
after the main signals [Figure 3-19(a) and (b)]. The arrival time of the echo is exactly
three times of the one-way traveling time through the PDMS layer. Except for the
presence of the echoes, the main signals from both cases are identical to each other,
indicating that the thickness of the coating does not change the CMUT plate
characteristics and it does not alter the main signal.
From the comparison between these two different coating thicknesses [Figure
3-19(a) and (b)], the main effect on the impulse response is the relative location of the
secondary echo signal. For the 200-μm coating, a time window can be applied to filter
away the echo safely. For the 75-μm coating, the echo overlaps with the main signal and
filtering is impossible. Consequently, the presence of the echo alters the frequency
spectrum [Figure 3-19(b)]. For many ultrasound applications, such as imaging, the
application of the time window is not practical. This is because the echo from the PDMS-
water interface may arrive at the same time as the real signal, similar to the case of Figure
3-19(b). Therefore, to design a good encapsulation, it is preferred to use a PDMS material
that has good acoustic impedance matching to the medium. This point will be illustrated
further in the following subsection.
56
-0.4 -0.2 0 0.2 0.4 0.6 0.8-3
-2
-1
0
1
2
3
Time [µs]
Dis
plac
emen
t [nm
]
-0.4 -0.2 0 0.2 0.4 0.6 0.8-3
-2
-1
0
1
2
3
Time [µs]
Dis
plac
emen
t [nm
]
75 µm of coating200 µm of coating
2 4 6 8 10 12 14-25
-20
-15
-10
-5
0
dB re
Pre
ssur
e/V
Frequency [Hz]2 4 6 8 10 12 14
-30
-25
-20
-15
-10
-5
0
Time [µs]
Dis
plac
emen
t [nm
]5.8 MHz 12.2 MHz5.5 MHz 11.9 MHz
Frequency (MHz) Frequency (MHz)
Time (µs) Time (µs)
Dis
plac
emen
t (nm
)N
orm
aliz
ed p
ress
ure
(dB)
Dis
plac
emen
t (nm
)N
orm
aliz
ed p
ress
ure
(dB)
(a) (b)
185 ns 69 ns
Main signal Main signal
EchoEcho
-3 -3
Figure 3-19. The viscoelastic FEM results showing the effect of different PDMS coating thicknesses: (a) 200-μm, and (b) 75-μm of GE RTV 615 PDMS coating. t = 0 s corresponds to the beginning of the 50-ns pulse. The data were retrieved from the PDMS-water interface.
3.4.2.2 Effects from different types of PDMS
We also demonstrate that the model could accurately reflect the influence of
different types of PDMS (Figure 3-20). Because Sylgard 160 is better matched to water
than GE RTV 615, the return echo from the PDMS-water interface disappears [Figure
3-20(b)].
The transmission and reflection coefficient of GE RTV 615 PDMS (1.1 MRayl)
to water (1.5 MRayl) is 84.6% and 15.4%, respectively. The peak amplitude of the main
signal [Figure 3-20(a)-1] is 2.79 nm. The measured transmitted [Figure 3-20(a)-2] and
the reflected waves [Figure 3-20(a)-3] are 2.33 nm and 0.37 nm, respectively. This
matches with the calculated transmitted and reflected wave, which had amplitude of 2.36
57
nm and 0.43 nm, respectively. The minor difference comes from the fact that we compare
only the peak amplitude of the signal instead of the full energy.
Similarly, in the case using Sylgard 160 PDMS whose acoustic impedance closely
matches that of water, the reflection coefficient is zero, which means there is no reflected
wave [Figure 3-20(b)-3, and (b)-4]. Therefore, the transmitted wave [Figure 3-20(b)-2]
has the same amplitude (2.3 nm) as the main signal from CMUT surface [Figure 3-20(b)-
1].
We also compared the peak amplitude of the main signals and the transmitted
waves between both types of PDMS. Due to the heavier mass loading of Sylgard 160
(1580 3mkg ) compared to GE RTV 615 (1020 3mkg , Table 3-3), the main signal of
the Sylgard 160 coated device is smaller (2.48 nm of peak amplitude [Figure 3-20(b)-1])
than the GE RTV 615 coated device (2.79 nm [Figure 3-20(a)-1]). However, because of
matched acoustic impedances, the transmitted wave from the Sylgard 160 case has a
higher peak amplitude (2.48 nm [Figure 3-20(b)-2]), compared to the GE RTV 615 case
(2.33 nm [Figure 3-20(a)-2]). Therefore, to design a coating without compromising the
amplitude, the acoustic impedance matching is important to maximize the transmitted
energy; moreover, it has to be optimized by considering the tradeoff from the effect of the
increased PDMS density.
-5 0 5 10x 10-7
-4
-3
-2
-1
0
1 x 10-9
Time [s]
Dis
plac
emen
t [m
]
-5 0 5 10x 10-7
-4
-3
-2
-1
0
1 x 10-9
Time [s]
Dis
plac
emen
t [m
]D
ispl
acem
ent (
nm)
Dis
plac
emen
t (nm
)
Time (µs)-0.5 0 0.5 1.0
Time (µs)-0.5 0 0.5 1.0
PDMS-waterinterface
Plate-PDMSinterface
-5 0 5 10x 10-7
-4
-3
-2
-1
0
1 x 10-9
Time [s]
Disp
lace
men
t [m
]
-5 0 5 10x 10-7
-4
-3
-2
-1
0
1 x 10-9
Time [s]
Dis
plac
emen
t [m
]
(a) (b)
(1)
(2)
(3)
(4)
Plate-PDMSinterface
(1) (3)
PDMS-waterinterface
(2)
(4)
Figure 3-20. The viscoelastic FEM results showing the effect from different types of PDMS: (a) 150 µm of GE RTV 615, and (b) 150 µm of Sylgard 160 on the CMUT for imaging. The average displacement from plate-PDMS and PDMS-water interface were
58
both shown with a 1-nm offset apart in Y-scale for clearer visualization. The main signals (a-1, b-1) arrive at time zero, and then the excitations propagate upward into the PDMS-water interfaces. The transmission parts were shown at the PDMS-water interfaces at the arrival time of 139 ns (a-2), and 158 ns (b-2) individually. The reflection part (a-3) travels back to the plate-PDMS interface and arrives at the time of 278 ns, while there is no reflection for the Sylgard 160 case (b-3).
3.4.3 CHARACTERIZATION FOR DYNAMIC PERFORMANCE
The frequency response was measured in vegetable oil that is electrically
insulating for devices without coating and in water for devices with coating. The CMUTs
are biased at a ratio of the pull-in voltage (80% for imaging CMUTs and 60% for HIFU
CMUTs) and excited by a unipolar 10-Vpp 50-ns square pulse, generated by a function
generator (Model HP 8116, Hewlett Packard Corp., Palo Alto, CA). Next, the pressure
was measured using a calibrated hydrophone (Model HNP-0400, Onda Corp., Sunnyvale,
CA 94089) in the far field (2.25 mm for the imaging CMUTs and 10 mm for HIFU
CMUTs). A fast Fourier transform (FFT) was then performed on the Gaussian-windowed
pressure signal. The pressure at the CMUT surface was calculated by compensating for
attenuation and diffraction [25]. The time domain pressure waveform and the frequency
spectra were compared to the modeling results.
3.4.3.1 Encapsulation effect on FBW
To address the coating effect, the measurement with 150 µm GE-RTV-615 PDMS
coating was compared to the one without coating (Figure 3-21). Because of acoustic
impedance mismatch between the PDMS-water interface, the additional echo was
observed comparing the coating case to the no-coating one [Figure 3-21(a)]. For the same
reason, the signal amplitude was decreased due to the reflection loss. The measurements
show that the coating increases the center frequency by 5%, and which agrees with the
viscoelastic FEM predicted increase of 6% [Figure 3-21(d), and Table 3-5]. The
fractional bandwidth (FBW) was decreased by 21% after the coating based on the
measurement and was 14% based on the modeling [Figure 3-21(c) and (d), and Table
3-5].
The viscoelastic FEM results of the 150 µm coated devices were compared to the
measurement (Figure 3-21). The comparison shows that the center frequencies of
59
measured and simulated devices differ by only 1% (Table 3-4). In addition, the FBW is
only overestimated by 9% in the model. The agreement can be further improved by using
a real 3D model that includes the element-to-element crosstalk and the effect of the finite
element size of the CMUT.
5 10 15-30
-25
-20
-15
-10
-5
0
Nor
mal
ized
pre
ssur
e (d
B)
Frequency (MHz)
-0.2 0 0.2 0.4 0.6-150
-100
-50
0
50
100
150
200
Pres
sure
(kPa
)
Measurement:no coating
Time (µs)
Measurement:150-µm coating
-0.2 0 0.2 0.4 0.6-150
-100
-50
0
50
100
150
200
Pres
sure
(kPa
)Time (µs)
Modeling:150 µm coating
Measurement:150-µm coating
Viscoelastic model 150-µm coating
Measurement:150-µm coating
5 10 15-30
-25
-20
-15
-10
-5
0
Nor
mal
ized
pre
ssur
e (d
B)
Frequency (MHz)
Measurement: no coating
Measurement: 150-µm coating
(a) (b)
(c) (d)Frequency (MHz)Frequency (MHz)
Figure 3-21. Comparison between the measurements (with and without the 150 µm of GE RTV 615 coating) and the viscoelastic FEM (with the 150 µm of GE RTV 615 coating) results for imaging CMUTs: (a) and (b) the time domain, and (b) and (c) the frequency domain. The data were retrieved from the PDMS-water interface.
For further comparison, the transmit impulse response from the non-viscoelastic
FEM was also performed. The sE was selected to preserve the static operation point
(Figure 3-22). Compared with the viscoelastic results, the response is under damped
which results in a smaller FBW in frequency domain. This occurs because sE leads to
60
the smaller acoustic impedance of PDMS (2) and lower speed of sound ( ρZVl = ). The
lower speed of sound also translates into the longer echo time, so we cannot see the slow
echo in the figure.
Table 3-4. CMUTs for imaging.
Modeling Experiment Error flower-3dB 6.1 MHz 6.3 MHz 3% fupper-3dB 12.7 MHz 12.3 MHz 3% fcenter 9.4 MHz 9.3 MHz 1% FBW 70% 64% 9%
Table 3-5. The coating effect of the CMUTs for imaging.
No coating 150-µm coating Coating effect Experiment Experiment
Modeling
fcenter 8.9 MHz 9.3 MHz 5% 9.4 MHz 6%
FBW 81% 64% -21% 70% -14%
-0.5 0 0.5 1-60
-40
-20
0
20
40
60
80
5 10 15-30
-25
-20
-15
-10
-5
0
dB re
Pre
ssur
e/V
Frequency [Hz]
Nor
mal
ized
pre
ssur
e (d
B)
Pres
sure
(kPa
)
Frequency (MHz)
-3
Time (µs) Figure 3-22. The results from the non-viscoelastic FEM with sE for imaging CMUTs with 150 µm of GE RTV 615 coating. The data were retrieved from the plate-PDMS interface. sE leads to the slower speed of sound, so there was no data from PDMS-water interface before 1.2 µs.
61
For the HIFU CMUTs comparison between the measurement and the modeling
results, we evaluated the low frequency design (radius = 70 µm) (Figure 3-23) and the
high frequency design (radius = 50 µm) (Figure 3-24). For the design with 70 µm radius,
the characteristic of the impulse response in time domain was simulated well; A 50-ns
pulse makes the discontinuity of the waveform from measurement [Figure 3-23(a)] that
we also found in the modeling result [Figure 3-23(b)]. For the design with 50-µm radius,
the 50-ns pulse does not result in any discontinuity as the case in low frequency design in
both the measurement and modeling results [Figure 3-24(a) and (b)]. The characteristics
of the impulse response from measurement and modeling, including the ringing, are
comparable.
Table 3-6. CMUTs for HIFU (radius of 70 µm and 50 µm).
Radius Modeling Experiment Error
70 µm flower-3dB 2.2 MHz 2.2 MHz 0% fupper-3dB 5.3 MHz 4.7 MHz 13% fcenter 3.7 MHz 3.4 MHz 8%
50 µm flower-3dB 5.4 MHz 4.7 MHz 15% fupper-3dB 8.5 MHz 7.7 MHz 10% fcenter 6.9 MHz 6.2 MHz 11%
The simulated results agree with the measurements in frequency domains as well
as in time domain. The center frequency was measured at 3.4 MHz compared to the
modeling at 3.7 MHz, which is only 8% of difference for the low frequency design
(Table 3-6). Similarly, the measured center frequency shows 6.2 MHz versus 6.9 MHz
from the modeling, reflecting only 11% of difference for the high frequency design
[Figure 3-24(c), and Table 3-6]. In addition to the center frequency, the modeled
frequency spectrum pattern matches the measurement very well. The only minor
discrepancy is the frequency offset. Possible causes for this discrepancy include: the
imperfect boundary constrain of the HIFU CMUT plate edge due to the thicker plate
thickness; the excitation pulse in the experiment might not be exactly the same as
simulation. Nevertheless, the HIFU device was always tuned to operate at a single
62
frequency and the off-resonance spectrum is much less important than that for imaging
CMUTs.
-0.5 0 0.5 1-20
-15
-10
-5
0
5
10
15
20
-0.5 0 0.5 1-20
-15
-10
-5
0
5
10
15
20
Pres
sure
[kPa
]
Time [µs]
Pres
sure
(kPa
)
Time (µs)
Pres
sure
(kPa
)
Time (µs)(a) (b)
- Measurement - Modeling
5 10 15-30
-25
-20
-15
-10
-5
0
Nor
mal
ized
pre
ssur
e [d
B]
Nor
mal
ized
pre
ssur
e (d
B)
Frequency (MHz)(c)
— Measurement
— Modeling
-3
Figure 3-23. Comparison between the (a) measurements and (b) the results from viscoelastic model in time domain, and (c) those in frequency domain for HIFU device with CMUT cell radius of 70 μm with 150 μm of GE RTV 615 coating. The data was retrieved from the plate-PDMS interface.
63
-0.5 0 0.5 1-15
-10
-5
0
5
10
15
20
-0.5 0 0.5 1-15
-10
-5
0
5
10
15
20
Pres
sure
[kPa
]
Time [µs]
Pres
sure
(kPa
)
Time (µs)
Pres
sure
(kPa
)
Time (µs)(a) (b)
- Measurement - Modeling
5 10 15-30
-25
-20
-15
-10
-5
0
dB re
Pre
ssur
e / V
olt
Frequency [Hz]
Nor
mal
ized
pre
ssur
e [d
B]
Nor
mal
ized
pre
ssur
e (d
B)
… Measurement
— Modeling
-3
Frequency (MHz)(c)
Figure 3-24. Comparison between the (a) measurements and (b) the results from viscoelastic model in time domain, and (c) those in frequency domain for HIFU device with CMUT cell radius of 50 μm with 150 μm of GE RTV 615 coating. The data was retrieved from the plate-PDMS interface.
3.4.3.2 Encapsulation effect on output pressure
To ensure the PDMS encapsulation can maintain the transmit efficiency and total
output pressure, the plate displacement at CMUT surface and output pressure in far-field
were measured individually for the devices with and without PDMS coating.
There are two different coating effects on the output pressure: first, a top coating
with the CMUT plate behaves as a compound plate, and second, the coating layer with
64
the medium behaves as multilayer medium for wave propagation. Coating layer as
medium could introduce attenuation and impedance-mismatch-induced reverberation. To
characterize the coating effect purely from the compound plate effect, the surface output
pressure can be measured first. Then the output pressure in far-field can be characterized
as an accumulative effect due to the compound plate and the multilayer medium.
In order to measure the dynamic response at CMUT plate surface, i.e.,
CMUT/PDMS interface, a clear type of PDMS, GE RTV 615, was chosen for optical
measurement. As the clear PDMS has acoustic impedance not perfectly matched to that
of medium, a thick 1-mm PDMS coating was selected to prevent overlap of the echo
from the PDMS/water interface. The output pressure was measured in vegetable oil for
devices without coating and in water for devices with coating. Similar to the setup for
frequency response measurement, the CMUTs were biased at 80% of the pull-in voltage
(PS300, Stanford Research Systems, Sunnyvale, CA, USA) and excited by a unipolar 10-
Vpp 50-ns square pulse, generated by a function generator (Model HP 8116, Hewlett
Packard Corp., Palo Alto, CA). Next, the transient response displacement was measured
using an optical fiber interferometer OFV-511 (Polytec GmbH, Waldbronn, Germany)
connected to a vibrometer controller OFV-2700/2 (Polytec) that contained a modified
wideband displacement decoder OVD-30 (Polytec) with an extended frequency range (50
kHz-30 MHz). It was measured at the CMUT surface by focusing at the plate/PDMS
interface and plate/oil interface, for devices with coating and without coating
individually. Surface area scan of 150 by 150 microns was performed to cover around 9
cells with the step size of 2 microns. The average displacement-time data was then
calculated, and transformed into output pressure according to (Equation 3-7).
ZtuP ⋅∂∂⋅= π2 , (3-7)
where P is output pressure, u is displacement, and Z is the acoustic impedance of
medium. Both time domain displacement and pressure waveform from the device with
coating were compared to that without coating.
65
No coatingWith coating
Dis
plac
emen
t (nm
)
x (µm)y (µm)
Dis
plac
emen
t (nm
)
x (µm)y (µm)(a) (b)
2 2.5 3 3.5x 10-6
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
Time [µs]
Dis
plac
emen
t [nm
]
Out
put p
ress
ure
[Pa]
2 2.2 2.4 2.6x 10-6
-10
-5
0
5
x 104
Time [µs](c) (d)
− no coating− with coating
− no coating− with coating
Figure 3-25. Displacement and output pressure at CMUT surface: surface area displacement results for the devices (a) without and (b) with PDMS coating; comparison of the (c) average displacement and (b) average pressure between the cases with and without PDMS coating.
Figure 3-25(a) and (b) show the displacements of the area scan retrieved from an
arbitrary time point for the cases with and without the 1-mm PDMS coating. The
averaged displacement comparison shows the one with the coating responses larger
[Figure 3-25(c)]. It was expected because the medium load for CMUT with coating is
smaller because of the smaller acoustic impedance (1.1 MRayl from PDMS instead of 1.5
MRayl from oil). Then the output pressure was compared and shown comparable
amplitudes [Figure 3-25(d)]. The peak-to-peak output pressure from the case with coating
is only 5% smaller than the one without coating. The result indicates the transmit
efficiency at CMUT surface can be maintained up to 95% with a 1-mm RTV 615 PDMS
coating.
Next, to study the effect on output pressure from the coating as a multi-layer
medium, the output pressure measurement in the far field was performed. The coating
66
effects on the loss include the impedance-mismatch-induced reflection and the
attenuation within the coating.
Previously in the setup for frequency response measurement, the CMUTs were
biased and excited by a relatively small AC excitation; the small AC setup was to
measure the linear response, and characterize the static operating point. As the gap height
was shared by the DC static deflection and the plate vibration, the dynamic response in
terms of the total output pressure was determined by the optimized combination of DC
bias and AC excitation.
To study the coating effect on the total output pressure, CMUTs with and without
a 150 microns of GE RTV 615 PDMS were measured under applied DC bias (DC bias
supply: PS300, Stanford Research Systems, Sunnyvale, CA, USA) and AC excitation
(generated by a function generator HP8112A, Agilent Technologies Inc., Palo Alto, CA,
USA). The DC and AC were swept through 50% to 90% and 50% to 100% of the pull-in
voltage (93V) respectively. Next, the pressure was measured using a calibrated
hydrophone (Model HNP-0400, Onda Corp., Sunnyvale, CA 94089) in the far field at
2.25 mm. The pressure at the CMUT surface was calculated by compensating for the
medium attenuation and diffraction [25]. Finally, the peak-to-peak output pressure
corresponding to various DC and AC in a step of 5V was illustrated (Figure 3-26).
Figure 3-26 shows the output pressure increased along with the DC and AC. One
minor distortion under high AC/DC [Figure 3-26(a)] is due to the charging which reduce
the effective DC bias and the transmit efficiency. The maximum total output pressure was
measured at 84 V of DC bias and 93V of peak-to-peak AC excitation. The device with
coating shows 1.02 MPa compared to 1.21 MPa from the device without coating. The
pressure was maintained 84.3%, versus 95% maintained at the CMUT surface before the
coating (Figure 3-25).
150 microns of GE RTV 615 PDMS introduces the attenuation less than 0.1 dB
(Equation 3-6), equivalent to 1%. The reflection introduces 15.4 % of loss (Equation
2)(1cm
cmenergy ZZ
ZZT+−
−= (3-1). The total loss from these two effect combined is
67
16.4% matched with the experimental result, 15.7%. Therefore, CMUT with PDMS
coating can feature high output pressure with a careful selection of the attenuation and
acoustic impedance as two of the main design parameters.
4050
6070
80 40
60
800
200
400
600
800
1000
1200
200
300
400
500
600
700
800
900
1000
1100
1200
4050
6070
80 40
60
800
200
400
600
800
1000
1200
200
300
400
500
600
700
800
900
1000[kPa] [kPa]
Out
put p
ress
ure
[kpa
]
Out
put p
ress
ure
[kP
a]
DC Bias [V]AC excitation [V] AC excitation [V]
DC Bias [V]
Measurement:Without coating
Meaurement:With 150-µm coating
1.21 MPa 1.02 MPa
Figure 3-26. Total output pressure from the CMUT with and without PDMS coating.
3.5 CONCLUSION
PDMS exhibits viscoelasticity with a glass-transition-temperature lower than
room temperature that makes it an ideal coating material for CMUT operation. PDMS
coating preserves the static operation point due to the low Young’s modulus at DC. The
low modulus value comes from the rubbery state under static loading at room
temperature. At the same time, it fulfills the acoustic matching requirement in ultrasonic
operating frequency region. The matching helps to secure the transmission efficiency and
prevent from echo reverberations.
A viscoelastic FEM was developed which can predict both the DC and AC
behavior of a PDMS-coated CMUT. The viscoelastic FEM has been demonstrated to
have good agreement with the measurement results. The model can correctly simulate the
static operation point, frequency-dependent stiffness, mass loading, and results in the
correct acoustic impedance. Based upon the correct static and dynamic parameters, the
hydrodynamic and acoustic behavior can be simulated. The model and measurements of
static results match perfectly. The center frequencies and FBW match within 1% and 9%
for imaging CMUTs, and 8~11% for HIFU CMUTs. The discrepancy occurred because
68
of the difference of the CMUT plate boundary constrain condition and the excitation
pulse between the model and experiments.
It was also demonstrated additive effects on performance due to the presence of
this encapsulation. The measurements show that a 150-µm coating of GE RTV 615
PDMS preserves the pull-in voltage of CMUTs at 142 V, and offset the center frequency
by only 5%. The FBW was decreased by 21% and is due to the echo from the acoustic
mismatch at the PDMS-water interface. A coating material with a better matching to
water and limited attenuation can be used to reduce the effect on FBW.
Finally, the output pressure was measured on the PDMS-encapsulated CMUTs.
The result indicates the transmit efficiency at CMUT surface can be maintained up to
95% with a 1-mm RTV 615 PDMS coating; while the total loss was addressed from the
attenuation and the energy reflection due to acoustic impedance mismatch. With the
experimental demonstration and the correct modeling, the study provides the
encapsulation strategy of CMUTs, and will help in the design process for optimizing the
static and the dynamic behavior of viscoelastic-polymer-coated CMUTs.
69
CHAPTER 4 PDMS-ENCAPSULATED CMUTS
This chapter, following the front side interface engineering discussion, presents
the applications of PDMS-encapsulated CMUTs. PDMS is a viscoelastic polymer with Tg
lower than room temperature, represents one of the optimized encapsulation materials for
CMUTs to achieve large output pressure, and maintain high sensitivity. By taking
advantages of PDMS properties, such as the speed of sound, attenuation and flexibility,
applications are demonstrated, such as crosstalk reduction, forming a flexible array, lens
focusing, and blood flow measurement capability.
4.1 CROSSTALK SUPPRESSION
4.1.1 OVERVIEW
Ultrasound transducers in immersion are subject to the crosstalk between the
neighboring array elements. This crosstalk degrades the transducer performance; the
issue was previously addressed for piezoelectric transducers [62], [63] and capacitive
micromachined ultrasonic transducers [17]. In general, crosstalk increases the effective
element aperture and the ringdown time of a transducer, resulting in a poor angular
response and range resolution [64]. In imaging experiments using CMUTs, degradation
in the axial resolution and bright patterns in the near field were observed due to crosstalk
[22]. HIFU applications require the precise control of the focal point for power deposition
to the pathological regions without destroying the surrounding healthy tissues [65].
However, because of a wider beam width and higher side lobes, the crosstalk impairs the
transducer’s capability to focus the ultrasound tightly [66].
Experimental, analytical, and finite element methods are used to determine the
causes and effects of crosstalk in CMUTs. Common experimental methods are optical
displacement and electrical received signal measurements. Other methods such as pulse-
echo and radiation pattern measurements are also employed to analyze crosstalk [17].
The dispersive guided mode wave was categorized as the main contributor (-22 dBs
normalized to the maximum transmitter displacement) compared to other mode, e.g., 0S
70
Lamb wave mode (-65 dBs), and 0A Lamb wave mode (-40 dBs) [38]. One of the
characteristics of the surface guided mode is that most of the energy is restricted to the
transducer surface (Figure 4-1).
Different approaches have been studied for crosstalk suppression. Previously,
Bayram et al. introduced a pull-in mode operation, effectively increasing the center
frequency, as a way to tweak the energy spectrums for crosstalk reduction [38], [67].
Berg, et al. proposed a second periodicity of the CMUT cell spatial configuration [68].
Certain periodicity provides an energy band gap such that the unwanted crosstalk wave
can be blocked. A different way to modify the element-element boundary condition was
studied by Bayram et al. [69]. By using a physical separation approach, a lossy wall
between elements was studied as a reflective/absorbing interface for crosstalk reduction
[70]. Another numerical study shows that a transfer-function-matrix approach can be
used to derive modified transmit waveforms on adjacent elements to reduce the acoustic
crosstalk [71]. However, this technique relies on programmable waveforms, which
introduce circuitry complexity.
All of the above studies address the change of CMUT design rule or impose
additional fabrication steps. It is advantageous for a simple crosstalk suppression
approach immune from major design modifications that might cause a tradeoff. A lossy
top thin coating with optimized thickness can effectively suppress the crosstalk wave
[39]. A PDMS top coating was proposed to feature the encapsulation and to suppress the
crosstalk at the same time. Thanks to the thin coating, the attenuation of the
transmit/receive energy normal to the transducer surface can be limited. This section
introduces the design and calculation of the lossy top coating. Experimental results
showing the crosstalk suppression effect due to coating was demonstrated.
71
Figure 4-1. Schematic of surface-guided-mode crosstalk with energy guided at the transducers surface.
4.1.2 COATING THICKNESS DESIGN
The thickness of the lossy top coating is the main design parameter for crosstalk
reduction while limiting the transmit/receive energy attenuation. Previously, a thin
coating (7 µms) of PDMS (Dow-Corning 4105 RTV) was coated on CMUTs for the
study of the crosstalk suppression [38]. The result shows that the thin PDMS coating is
ineffective in reducing crosstalk. Crosstalk suppression is only -1 dB or negligible. This
result, suggesting a thicker or a more lossy type of PDMS, was the main motivation for
this study.
To obtain the required coating thickness, the energy distribution along z direction
(normal to transducer surface) of the surface acoustic wave should be calculated. The
pressure of the surface acoustic wave ( p ) in a simplest form is
)exp(0 tjxjzpp ωξα −+−⋅= (4-1)
, where α is the decay constant in z direction, 22 αξ += k is the wave number
along the surface, ck ω= is the wave number for free space sound propagation, and c is
the speed of sound in the fluid. Thus, the phase velocity of this surface wave ( phc ) is
given by
72
2
2
2
2 222 11ω
α αω
αω
ξω
ck
phc
kkc
+=
+=
+== (4-2)
As the surface wave is exponentially decayed in z direction (Equation 4-1), the
energy envelope confined by )1exp(0 −⋅= pp refers to that the energy is mostly within
the thickness, α1=z . Given the phc and c, the relationship between the suggested coating
thickness (in terms of α1 ) and the frequency,
πω2
=f can be created (Equation 4-2).
Figure 4-2 shows the suggested PDMS coating thickness as a function of frequency for
each given phase velocity.
According to Eccardt, et. al. [72], phase velocity can be approximated by using
the effective stiffness density s (units: 2mmN ) and mass density m (units: 2mkg ) at
the interface,
cccccc
mscph ≤
+=
+=
−22 )(1)(1 *2ω
ωρ (4-3)
(a)
2 4 6 8 100
100
200
300
Frequency [MHz]
Thick
ness
[ µm
]
Without coatingCoil=1430 m/s
Cphase=1300 m/s
Cphase=1000 m/s
Cphase=700 m/s
)( 1α
)( 2πω
(b)
2 4 6 8 100
100
200
300
Frequency [MHz]
Thick
ness
[ µm
]
)( 1α
)( 2πω
With PDMS coatingCcoating=1080 m/s
Cphase=1000 m/s
Cphase=700 m/s
Figure 4-2. Suggested PDMS coating thickness for the crosstalk-suppression. (a) The energy envelop, )1exp(0 −⋅= pp in oil without coating. (b) The energy envelop with coating by using Ccoating =1080 m/s (from GE RTV 615 PDMS).
73
In addition to the interface stiffness and mass dependence, the phase velocity
decreases along with the frequency; this frequency dependence comes from the
dispersion of the interface acoustic wave. For various CMUT designs (affecting s and
m , and thus phc ) and frequencies of interest, Figure 4-2(b) indicates a useful guideline
for coating thickness. For example, the curve shows a 150 µm coating of PDMS can
effectively attenuate the frequency component larger than 3 MHz of the surface acoustic
wave with a ≤phc 1000 m/s. Considering all the encapsulation requirements from
Chapter 3, a 150 µm coating thickness was selected to provide dielectric/abrasive
protection while minimizing the transmit/receive attenuation, and offer sufficient
crosstalk reduction at the same time.
To enable an optical measurement, a clear type of PDMS, GE RTV 615 PDMS,
was selected for this study. GE RTV 615, with an attenuation of 0.8 dB/MHz/cm
(Equation 3-1), is one of the less lossy PDMS among various other opaque types. Thus,
the study with RTV 615 PDMS was intended to provide a lower bound of crosstalk
suppression effect. A lossier type of PDMS, Dow-Corning Sylgard 160 PDMS, was also
studied; more crosstalk reduction was expected although an optical measurement is
impossible such that a pitch-catch pressure measurement was performed instead.
4.1.3 EXPERIMENTAL METHOD AND RESULTS
The experimental setup (Figure 4-3) enabled both optical and electrical
measurements to be performed simultaneously. A DC bias supply (PS300, Stanford
Research Systems, Sunnyvale, CA, USA) was used to bias all the elements of the array,
and a function generator (HP8112A, Agilent Technologies Inc., Palo Alto, CA, USA)
was used to excite the transmit element with a 20-ns 10Vpp unipolar pulse. Electrical
signals were measured on a digital oscilloscope (Infiniium 500 MHz, 2 GS/s, Agilent
Technologies Inc., Palo Alto, CA, USA) by probing individual array elements. The
optical measurements were made using an optical fiber interferometer OFV-511 (Polytec
GmbH, Waldbronn, Germany) connected to a vibrometer controller OFV-2700/2
(Polytec) that contained a modified wideband displacement decoder OVD-30 (Polytec)
with an extended frequency range (50 kHz-30 MHz). A common personal computer (PC)
74
was used to process the data transferred over a GPIB-IEEE 488 bus from the
oscilloscope. LabViewTM Version 7.6 software (National Instruments, Austin, TX, USA)
was used to control the x-y stage and gather the optical and electrical data.
VibrometerController &
Decoder
Microscope&
Interferometer
AC Coupler
Digital OscilloscopeHP Infiniium
500 MHz, 250MS/s
Personal Computerwith
LabViewTM
DC BiasSupply
&FunctionGenerator
XY-Stage
Received signal u(x,t)
Lens (20x)
Laser
RS-232
Figure 4-3. Experimental setup of optical and electrical measurement for crosstalk characterization.
The 1D 132-element CMUT arrays with the design parameters listed in Figure 4-3
were used for this study. The CMUTs with and without PDMS coating were immersed in
vegetable oil for comparison. Using the microscope and the x-y stage, the CMUT surface
was aligned with the laser light. As the optical measurements spatial resolution is around
5 μm, determined by the laser spot diameter, the uniform 10 μm sampled data were used
by performing a piecewise linear approximation. Considering the acousto-optical
correction, the actual plate displacements are obtained by using the refractive index of
75
soybean oil (noil = 1.47) in the optical path calculation. For the case of CMUTs with
PDMS coating, the refractive index of the PDMS, 1.46, was used instead for the
correction.
Table 4-1. Physical parameters of the CMUT array for crosstalk study
Plate width, µm 40 by 40 (square) Plate thickness, µm 1.50 Vacuum cavity height, µm 0.15 SiO2 insulation layer thickness, µm 0.3 Element pitch, µm 300 Number of plate/element 6 by 120 Number of elements in an array 132 (4 dummies) Array dimension, mm 3 by 30 PDMS coating thickness, µm 150 Center frequency, MHz (in immersion)
7.5
Pull-in voltage, V 92
First, the optical surface displacement scan was performed on the edge 300 by
300 µm of the excited element and the same area of the neighboring element (Figure
4-4). This measurement was designed to demonstrate the inter- and intra-element
crosstalk. As the stationary area have different separations between cells within the
element (40 µm) and between elements (10 µm), the different stiffness and mass density
result in an impedance-discontinuity interface for wave to transmit and reflect [72]. These
out-of-phase displacements were considered to add the tail ringing to the main pulse (in
time domain), thus deteriorate the bandwidth (in frequency domain).
Then, the center 41 elements of the CMUTs with and without GE RTV 615
PDMS coating were biased, and the center element was excited. The optical
measurements were performed over the center of the 6-cells of each element. These plate
displacements on the line scan were averaged over each cell. Retrieving the data
neighboring to the excited one, the plate displacement was plotted in frequency-spatial
(ω-k) domain (Figure 4-5). The results show the maximum crosstalk amplitude happened
at the frequency of 2.8 MHz with a phase velocity of 1350 m/s ( mMHz 120808.2 ). The
energy ranges up to 6.2 MHz (-40 dB normalized to the peak amplitude from the excited
element) with an energy band gap at around 4 MHz. Thus, the second relative maximum
76
amplitude was observed, located at the frequency of 4.6 MHz with a phase velocity of
1210 m/s ( mMHz 138006.4 ). The energy band gap comes from the arrangement of the
plates within/between the array elements that influences the preferred frequency of the
guided mode [73]. The CMUT design for this study has a center frequency of 7.5 MHz
and a 90% FBW, which represent the frequency of interest between ~4.1 MHz to 10.9
MHz. Therefore, the crosstalk energy below 4.1 MHz is less harmful. As the attenuation
of PDMS increases along with the frequency ( 4.1floss ∝ ), the PDMS coating for
crosstalk suppression was expected to be efficient within the frequency of interest.
T=0.32 µs T=0.39 µs T=0.47 µs
T=0.55 µs
inter-element crosstalk
intra-element crosstalkx (µm)
y (µ
m)
Figure 4-4. Inter- and intra-element crosstalk measurement.
Figure 4-5(b) shows the ω-k result from the CMUTs with coating. The maximum
amplitude was shifted to lower frequency (1.85 MHz) with a slower phase velocity
(1011m/s) compared to that in Figure 4-5(a). The frequency shift is because of the
increasing PDMS attenuation along with the frequency; the slower phase velocity comes
from the lower speed of sound within the coating (Equation 4-3). The crosstalk amplitude
within the frequency of interest (the second relative maximum at around 4 MHz) is 5.6
dB lower compared to the counterpart in Figure 4-5(a). The results show that a thin
PDMS coating with optimized thickness can suppress the crosstalk frequency within the
frequency of interest.
77
k, W
aven
umbe
r[1/
m]
k, W
aven
umbe
r[1/
m]
0 1 2 3 4 5 6 7-9000
-8000
-7000
-6000
-5000
-4000
-3000
-2000
-1000
0
0 1 2 3 4 5 6 7-9000
-8000
-7000
-6000
-5000
-4000
-3000
-2000
-1000
0
Frequency (MHz) Frequency (MHz)
Without coating With coating
(a) (b)
Coating
reverberations
Figure 4-5. The crosstalk measurement comparison between the CMUTs (a) without and (b) with a lossy PDMS coating for the frequency of interest.
Compared to GE RTV 615 PDMS, a lossier type of PDMS was expected to
provide a more effective crosstalk reduction. Dow Corning Sylgard 160 PDMS with
attenuation (3.7 dB/MHz/cm) was selected and coated on a consistent 1D CMUT array
for comparison. Because a lossier type of PDMS is opaque that prohibits the optical
CMUT surface measurement, instead a hydrophone pressure measurement in the far field
was performed. The results show that a lossy thin coating attenuates the tail (-4.5 dB)
more effectively than the main pulse (-1.4 dB) (Figure 4-6). As the tail, effectively
increasing the ring down time, can be attributed to the out-of-phase plate displacements
due to crosstalk, a lossier thin PDMS coating was observed to be effective in crosstalk
suppression with an acceptable main signal loss.
78
1.2 1.4 1.6 1.8 2-2
-1
0
1
2
3
Time (µs)
Volta
ge (m
V)-1.4dB
-4.5dB
α=0.8 dB/MHz/cm α=3.7 dB/MHz/cm
Volta
ge (m
V)
Time (µs) Figure 4-6. The impulse response comparison between PDMS coatings with different attenuation.
In conclusion, a lossy top coating by using PDMS was proposed to feature the
encapsulation and the crosstalk suppression at the same time. The PDMS coating
thickness and material attenuation were identified as the design parameters. The
optimized coating thickness was suggested and a guideline for various phase velocity of
the surface crosstalk wave was provided. A 150 µm coating of PDMS was selected and
demonstrated to be effective in crosstalk suppression through an optical surface
displacement measurement. The result demonstrates the first PDMS encapsulated CMUT
with the crosstalk reduction effect. The impulse response for a lossier and opaque PDMS
coated CMUTs was demonstrated, showing the tail attenuation is more effective than the
main-pulse attenuation. To study further the crosstalk reduction with a lossier opaque
PDMS coating by using an optical displacement measurement is appealing for future
research.
79
4.2 FLEXIBLE CMUTS
4.2.1 OVERVIEW
Flexible transducer arrays allow conformal coverage of an object in applications
such as medical imaging, ultrasound therapy, and nondestructive evaluation. Among the
medical imaging applications, intravascular ultrasound imaging (IVUS) requires side-
looking capability, which can be offered by flexible transducer arrays wrapping around
catheter tips (Figure 4-7). Using IVUS imaging, the degree of narrowing in blood vessels
can be determined; plaque composition can be analyzed; and stent placements can be
visually guided.
side-looking
forward-looking
Transducer array
Frontend ICs
Coaxial guide wire
Microcoaxial cable bundles
(a) (b) Figure 4-7. Schematic of (a) a catheter for intravascular ultrasound imaging and (b) the detail of a catherter-based side-looking probe.
In Chapter 3, PDMS, a viscoelastic polymer with Tg lower than room temperature,
was studied as the CMUT encapsulation material to maintain performance while
providing electrical insulation. The low Tg of PDMS (-125 oC) comes with its high
flexibility at room temperature, i.e. a Young’s modulus of only couples of MPa. The low
Young’s modulus, compared to that of Si (~168 GPa), can be designed into the substrate
to provide the flexibility of CMUT arrays. By using a PDMS encapsulation and refilling
trenches, a flexible CMUT array was demonstrated in this section for side-looking
intravascular ultrasound imaging. Most of the work contained in section 4.2 has been
previously published in [36].
How to fabricate a flexible transducer array has been studied in recent years. For
example, one method to achieve a flexible array is by aggressively thinning down the
80
silicon substrate (thickness < 100 µm) [74]. The thin substrate provides flexibility,
however also imposes reliability issue and handling difficulties [75]. Polymer-based
structures have also been used as a flexible carrier for MEMS sensors [76], [77], [78],
[79]. An easy flexible substrate preparation, which is compatible to the present CMUT
process, is based on polymer-coated trenches [80], [81]. The selected approach to achieve
flexible CMUTs is by etching through-wafer trenches to isolate array elements, and
refilling the trenches with PDMS while using PDMS as the encapsulation material.
4.2.2 DESIGN AND FABRICATION
To demonstrate the flexibility of CMUTs in the x and y axes, 2D arrays were
designed and fabricated. The physical parameters of the CMUT designed in this study are
summarized in Table 4-2.
Table 4-2. Device parameters of flexible CMUTs.
Plate length, µm 220 Plate width, µm 35 Plate thickness, µm 1.83 Vacuum cavity height, µm 0.15 SiO2 insulation layer thickness, µm 0.3 Element pitch, µm 250 Number of plate/element 8 Metal coverage on plate 100% Number of elements in an array 256 (16 by 16) Array dimension 4 mm by 4 mm PDMS coating thickness, µm 30
The PDMS coating technique was selected between spin coating, mold casting,
dip flowing, and precast-then-transferring. To incorporate with the trench-refilling
process, it is not suitable for any precast approach. While dip flowing offers a quick-and-
dirty way for PDMS coating, yet thickness is not well controlled; spin coating and mold
casting were suggested. A clear type of PDMS, i.e. one-part 1-4105 (Dow Corning
Corporation, Midland, MI) was used for further CMUT surface inspection. The spin-
coating thickness versus spinning speed was characterized (Figure 4-8). As the final
substrate thicknesses were tested among 50-200 µm, a 30 µm PDMS thickness was
selected to provide a reasonable flexibility. Thus, the spin-coating technique was
81
suggested for this thickness design, because the mold casting involved mold-releasing
step, which is not favorable for thickness less than 50 µm.
To build flexible CMUT arrays, the fabrication process was modified from the
regular CMUT wafer bonding process. First, a highly conductive Si wafer was used, and
thermal oxide was patterned to form the cavities. Before the fusion bonding for CMUT
plates, deep trenches were patterned and etched into the silicon substrate using DRIE.
These trenches are for electrical isolation and the PDMS-refilling for substrate separation.
The widths varied between 6 µm and 20 µm, and the depth was about 150 µm. The wafer
was then fusion bonded to an SOI wafer to form the CMUT plate. After sputtering an
aluminum layer metalized the plate, holes with diameters of 4 µm to 10 µm were opened
to the trenches as inlets for the PDMS refilling. Then it was performed the PDMS
dispensing, degassing, and curing, followed by the substrate etch back. The array
elements were electrically isolated from one another, but mechanically linked by the
PDMS in the trenches and on top of the plate. Finally, the metal pads were patterned on
the backside for interconnections. The finished flexible 2D CMUT array was
demonstrated in Figure 4-10.
Empirical fitMeasurement
Spin speed (rpm)
PDM
S th
ickn
ess (µm
)
Figure 4-8. PDMS thickness as a function of spin speed.
82
Si SiO2 Al PDMS Au Figure 4-9. Schematic cross-section of CMUT elements with PDMS encapsulation and trenches-refilling substrate.
1 mm
PDMS (30 mm)
Si membrane (1.83 mm)
Trench filled with PDMS
(15 mm)
1 mm
PDMS(30 µm thick)
Si membrane(1.83 µm thick)
Trench filled with PDMS
(15 µm wide)
(a) (b)
(c) (d) Figure 4-10. Device photographs: (a) front side view, (b) back side view showing the flexibility pushed by a wire, (c) magnified view of through-wafer PDMS dispensing holes, (d) cross-sectional view showing the trench, the plate, and the PDMS coating and filling.
4.2.3 TEST RESULTS AND DISCUSSION
The transducer arrays were characterized in air. The CMUT was biased at various
DC voltages and a network analyzer (Model 8751, Hewlett-Packard Company, Palo Alto,
CA) was used to measure the electrical input impedance. A micro-probe (Model ACP40-
83
W-GS-150, Cascade Microtech, Inc., Beaverton, OR) was used to make electrical contact
on the backside of the CMUT array. Real and imaginary parts of the impedance are
shown in Figure 4-11(a). It was expected the PDMS coating over damps the CMUT and
results in the low quality factor of the resonance. The resonant frequency in air was then
extracted along with various DC voltages. Figure 4-11(c) shows the dependence of the
open circuit resonant frequency on the DC bias voltage. The decreased resonant
frequency is attributed to the spring softening effect of an electrostatically actuated
system with increased DC bias voltage.
(a) (b) Figure 4-11. Test results: (a) electrical input impedance in air, (b) resonant frequency.
These arrays were flexible as demonstrated [Figure 4-10 (b)], however, the
bending force generated cracks aside the mechanical linking parts [Figure 4-12(a)]. The
defects were suggesting a better stress released mechanical linking structures rather than
the continuous blanket with holes [Figure 4-12 (b)]. Silicon ribbons (springs) have been
used as interconnects for MEMS devices to provide better compliance without
interrupting electrical continuity [82], [83], [84]. Such ribbons were suggested to
incorporate into future designs [Figure 4-12 (c)].
84
(a) (b) (c)
50 µm
Figure 4-12. (a) A photograph of cracks aside the mechanical linking bridge, (b) Schematic of current linking design, (c) Schematic of proposed ribbons (springs) design.
85
4.3 PDMS COATING AS LENS MATERIAL FOR FOCUSING
4.3.1 OVERVIEW
1D linear array is widely used in today’s diagnostic medical imaging providing a
2D B-mode sonography. While 1D array was attributed the electronic focusing in
azimuth direction, a lens coating is required to enable the mechanical focusing in
elevation (Figure 4-13). The benefit is to improve the elevation resolution, i.e. thinner
slice thickness.
A conventional 1D piezoelectric transducer array requires matching layers in
addition to the lens coating. The encapsulation material optimized for CMUT operation
can be a viscoelastic polymer with Tg lower than room temperature. PDMS features the
required specifications, and attributes different speed of sound from that of the medium,
e.g. 1530 m/s for soft tissue. In light of all the considerations for encapsulation and
mechanical focusing, PDMS was chosen as a lens material.
Azimuth
Depth
Elevation(a) (b) Figure 4-13. Schematic acoustic pressure profile of comparison between CMUT arrays (a) without lens coating, and (b) with lens coating.
86
4.3.2 LENS DESIGN
Lens design starts from the calculation for determining the radius of curvature
(ROC) of a convex acoustic lens:
)1( −⋅=lens
mediumgeo C
CFROC , (4-4)
where geoF is the geometric focal point, mediumC is the speed of sound of the
medium, e.g. human tissue, and lensC is the speed of sound of lens material ( lensC <
mediumC for a convex lens in this study). Compared to a concave lens, a convex
counterpart serves for an ideal ergonomic contact and hence a better acoustic energy
coupling through the lens/ human tissue interfaces. To determine geoF , given the center
frequency ( λmediumCf = ) and aperture size (D), several design tradeoff need to be
considered, such as resolution (slice thickness), depth of view (DOV), and focal gain,
DOV = 22)(1.7 geogeo FDF
∝⋅⋅λ , (4-5)
Resolution = geogeo FDF
∝⋅λ , and (4-6)
Focal gain = geogeo FF
apertureofArea 1∝⋅λ
. (4-7)
Among the three design parameters, focal gain is less concerned because the
azimuth electronic focusing provides the primary gain. Therefore, resolution and DOV
are tradeoff to each other in terms of geoF . As an example of a 6.5 MHz array design and
D of 3 mm, 11.3 mm of geoF was selected resulting a resolution of 1.1 mm and DOV of
around 23.2 mm. Thus, ROC was calculated as 6.55 mm given the other parameters as
illustrated in Figure 4-14.
Given the geoF and D, S parameter was calculated,
87
S = λ⋅⋅ 422.1 D
geoF = ~1.2, (4-8)
where S > 1 indicating a weak focusing, and resulting in a shorter DOV [14].
D=3.0mm
ROC=6.55mm
Fgeo=11.3mm
Clens=947 m/s
Cmedium=1497 m/s
F# = 3.8λ=230 µm
Fresnel’s limit = 9.8mm
Figure 4-14. Design of lens focusing.
4.3.3 MATERIALS AND METHODS
PDMS was carefully selected considering several parameters. First, the
impedance of the lens material should be close to that of the medium, similar to the
encapsulation discussion in chapter 3. Second, the attenuation of the lens material should
be low to maintain high sensitivity. Last, a low viscosity was required for the better
workability during the mixing, mold, and degassing processes. The attenuation loss can
be expressed as follow:
Loss in dB = 24.1 ⋅⋅⋅ TLfα , (4-9)
where α is the lens material attenuation, f is the operational frequency, and
22 )2(DROCROCTL −−= , the thickness of the lens. The total loss is proportional to
TL, which can be reduced by lowering lensC . Consequently, a low lensC (apart from
mediumC ) was preferred for an efficient focusing to limit the lens thickness and hence the
total attenuation.
88
The attributes of the base silicone rubber without dopants include clear color,
density of ~1000 kg/m3 and speed of sound at ~1000 m/s. The off-the-shelf products are
available, e.g. GE RTV 615, and Dow Corning Sylgard 182/184. The base silicon rubber
usually addresses low attenuation, < 0.1 dB/mm/MHz, which is preferable as the lens
material. As the acoustic impedance is around 1 MRayl (1000 kg/m3 x 1000 m/s)
mismatched to that of the medium, matching acoustic impedance could be achieved by
adding dopants to increase the density. It was studied that different types of inorganic,
metal particles (TiO2, Pt, and Yb2O3 etc.), in different sizes (10 nm to 2 µm) mixed into
the base RTV or HTV to achieve matching impedance and obtaining low attenuation
[85], [60]. By mixing with high-density material, particle size and dopant volume ratio
can be minimized, and hence the attenuation increase was limited.
Among several commercially available RTV formulations, Dow Corning®
Sylgard 160 was selected for its matching acoustic impedance (Z=1.5 MRayls), and
workable viscosity (8800 cps). The attenuation was measured at 0.38 dB/mm/MHz by
using a commercial calibrated piezoelectric transducer probe. This low attenuation figure
is comparable to the state of the art attenuation obtained from recently reported doped
RTV or doped HTV (Table 4-3) [85], [60].
Table 4-3. Attenuation of Sylgard 160 compared with various doped RTV or HTV
Attenuation Acoustic impedance (dB/mm/MHz) (MRayls) Dow Corning® Sylgard 160 0.38 1.5 10 nm of Pt within base RTV 0.71 1.26 16 nm of Yb2O3 within base RTV 0.93 1.35 16 nm of Yb2O3 within base HTV 0.73 1.43
For mold manufacturing, the machinability, mold-releasing feasibility, and
toughness were considered into the design. To fulfill the requirements, a mold transfer
technique was used: from aluminum to silicon rubber mold, and then from silicon rubber
to polyurethane mold (Figure 4-15). Aluminum was chosen for machining easiness.
Polyurethane is inert to silicon rubber easing the mold-release; at the same time
polyurethane is tough enough to prevent from distortion and abrasion during the mold
casting and releasing.
89
Figure 4-15. Lens mold manufactured by using mold-transfer technique.
To minimize the total thickness of the lens coating, it was essential to limit the
bonding wire profile, e.g. to be less than 150 µm. While the curvature part (TL)
contributes around 174 µm, the total thickness was measured at 345 µm above the
transducer surface. In order to study the lens focusing effect, another 1D array was coated
with a conformal PDMS layer of the same thickness for comparison.
A 1D 132 elements (128 active elements plus 2 dummies on both sides) 300 µm
pitch CMUT array was fabricated using fusion bonding technique. The key physical
parameters of the CMUT array are summarized in Table 4-4. After the fabrication, the
array was attached to a printed circuit board (PCB) for further wire bonding and lens
coating. The same array and the PCB design were also used for the blood flow study
described in section 4.4, that the board was connected to higher-level interconnections.
Table 4-4. Physical parameters of the CMUT array for lens study.
Number of elements 132 Azimuth element pitch (µm) 200 Elevational element pitch (mm) 5 Elevational element size (mm) 4 Square plate width (µm) 40 Vacuum cavity height (µm) 0.15 Insulation layer thickness (µm) 0.3 Plate thickness (µm) 1.5 Aluminum layer thickness (µm) 0.3 Silicon substrate thickness (µm) 400 Chip length (mm) 53 Chip width (mm) 37
90
(a)
(b) (c)
39.6 mm
300 µm
Figure 4-16. A 1D 128 elements linear CMUT array (a) before lens coating, and (b-c) after lens coating.
The pressure response was measured in DI water, which sat for 4 hours for
degassing. One CMUT element was biased at 73V, 80% of the pull-in voltage and
excited by a 15-cycles 10-Vpp continuous wave (CW) at 6.5 MHz, generated by a function
generator (Model HP 8116, Hewlett Packard Corp., Palo Alto, CA). Next, the pressure
was measured using a calibrated hydrophone (Model HNP-0400, Onda Corp., Sunnyvale,
CA 94089) at the depth ranging from 5 mm to 20 mm. The step of the depth
measurement is 250 µm considering the hydrophone tip size 300 µm, and the signal
wavelength of around 230 µm. At each depth, an elevation line scan from -2 cm to +2 cm
with a step of 150 µm was performed, controlled by a computerized linear stage. As the
measurement covered large area including the near field, where rapid phase changes, the
alignment is very critical for an accurate pressure field result. The alignment was
performed through a loop crosscheck between an optical alignment and an acoustic
pressure measurement feedback. The optical way was using an optical microscope to
focus on the PCB; the measurement feedback was to find the maximum pressure point
from the elevation line-scan at three different depths.
91
To be noted that this study was to characterize one CMUT element with lens
focusing. A practical scheme of 1D linear array operation requires a group of elements,
e.g. 32 elements, with or without electronic focusing along azimuth. The azimuth
electronic phase focusing was not available, such that the geometric diffraction was
expected along the azimuth direction. For the same reason, the intensity of the pressure
field decreases along with the depth.
4.3.4 RESULTS AND DISCUSSION
Figure 4-17 demonstrates the pressure fields and beam profiles from the CMUTs
with conformal coating and lens coating. Compared to the conformal coated CMUTs
[Figure 4-17(a)], it was expected that the lens coated CMUTs show more significant
focusing and the focal point is at around 9 mm. The relative peak amplitude around the
focal point was measured at 10.8 kPa, versus 9.0 kPa measured for the conformal case.
The limited focusing gain was expected for the weak focusing design.
The main design purpose for lens on a 1D array is the elevation resolution
improvement, i.e. thinner slice thickness. With conformal coating, the slice thinness, i.e.
elevation resolution, is 2.0 mm at F# of 3, and 1.4 mm at F# of 7 [Figure 4-17(b)]. With
lens coating, the results show that the slice thickness is reduced to only 8mm at F# of 3,
and 1.2 mm at F# of 7. These results demonstrate a direct PDMS lens coating can provide
encapsulation and mechanical focusing on CMUTs.
92
x [mm]
y [m
m]
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
5
10
15
20
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3x [mm]
y [m
m]
-2 -1 0 1 2
5
10
15
20
2
3
4
5
6
7
8
9
10
11
12
x [mm]
y [mm
]
-2 -1.5 -1 -0.5 0 0.5 1 1.5 25
10
15
20
2
3
4
5
6
7
8
9
10
11
12
x [mm]
y [mm
]
-2 -1.5 -1 -0.5 0 0.5 1 1.5 25
10
15
20
2
3
4
5
6
7
8
9
10
11
12
x [mm]
y [m
m]
-2 -1.5 -1 -0.5 0 0.5 1 1.5 25
10
15
20
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
x [mm]
y [m
m]
-2 -1.5 -1 -0.5 0 0.5 1 1.5 25
10
15
20
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
kPa
dB
Dep
th (m
m)
Dep
th (m
m)
Dep
th (m
m)
Dept
h (m
m)
dB
kPa
F# = 7
F# = 3
10.5 kPa
DO
VElevation (mm) Elevation (mm)
(a) (b)
Figure 4-17. Pressure field measurement for CMUTs with (a) conformal PDMS coating, and (b) lens PDMS coating.
93
4.4 ULTRASOUND BEAM TILTING FOR BLOOD FLOW IMAGING
One application of PDMS encapsulated CMUTs is an operator-independent
carotid artery screening by using asymmetric lenses coating on a 5-plane CMUT array.
The array is used for color flow Doppler detection of peak blood velocity (PBV), which
is correlated with carotid stenosis. Asymmetric acoustic lenses provide fixed angle off-
axis beam steering for color flow Doppler. In the proposed system, first, five parallel
short-axis B-mode and color flow images of the carotid artery are obtained. Then, a
custom software package automatically finds the center points of the blood flow,
computes the vessel trajectory, Doppler angle and angle-corrected peak velocities. The
fabricated CMUT arrays showed a center frequency of 8.1 MHz and a fractional
bandwidth of 103% in immersion. In resonant frequency measurement in air, the
uniformity across the array is excellent, with a standard deviation of 0.18% of the mean.
Initial experiments on asymmetric acoustic lenses verified the off-axis steering capability.
Using the in-house software package and a commercial 1D array, the calculated peak
blood velocity has an error of less than 6%, when the Doppler angle is less than 45o.
4.4.1 INTRODUCTION
Stroke is the third leading cause of death in the United States [86]. In many cases,
atherosclerotic stroke is caused by carotid stenosis (narrowing of the carotid artery due to
fatty deposits on the artery side walls). Doppler ultrasound is a widely available screening
tool for carotid stenosis. In the screening procedure, the peak blood velocity (PBV) in the
carotid artery, which is correlated to carotid stenosis, is measured. However, in the
current procedure, the technologist needs to align the ultrasound probe longitudinally
with the carotid artery, and carefully position the probe in order to locate the peak blood
velocity. Then, the technologist needs to manually position cursors on the ultrasound
machine to determine the Doppler angle, which is essential for the accurate calculation of
the PBV. Thus, the current procedure is highly dependent on the technologist. It not only
requires a highly trained professional, but also makes the procedure lengthy (15-30 min).
Moreover, the measurement results are prone to intra- and inter-operator variability.
An asymmetric lens was proposed to use on each of the five parallel 1D CMUT
arrays on a single substrate to obtain five parallel B-mode and color flow images of the
94
carotid artery. Custom-built board-level electronics interface the 5-plane array with a
commercial ultrasound imaging system. In the elevation direction asymmetric acoustic
lenses steer the acoustic beam at fixed angles for color flow Doppler measurements. A
custom software package automatically computes the vessel trajectory, Doppler angle
and angle-corrected peak velocities for the 5-plane transducer array.
4.4.2 METHODS AND RESULTS
4.4.2.1 System overview
The goal of the proposed carotid screening system is to eliminate operator
dependence, thus making it more reproducible and clinically accessible than the
conventional method. This goal can be achieved by using five 1D transducer arrays
located on a single substrate. CMUTs are well suited for this application because of the
demonstrated performance advantages for carotid imaging [87], and the ease of array
manufacturing, due to parallel micro-fabrication. Asymmetric acoustic lenses are formed
on each array in the elevation direction. A custom-designed circuit board interfaces the
CMUT array with a commercial ultrasound system (in this case a Siemens Antares
system) via a standard probe cable. The board provides the needed DC biasing network,
and the electronic circuits to switch between the five individual arrays. Minimal
modifications to the existing ultrasound systems are needed to accommodate the CMUT
arrays and the interface board. Thus, five cross-sectional B-mode and color flow images
of the carotid artery can be obtained. A software package developed in-house then
determines the center point of each cross-sectional slice of the vessel in the color flow
image. The program uses this information to automatically compute the vessel trajectory,
Doppler angle and angle-corrected peak velocities for the 5-plane transducer array. The
system concept is illustrated in Figure 4-18.
95
(a)
(b)
(c) Figure 4-18. System concept for a new carotid screening tool based on a 5-plane CMUT array. (a) CMUT array and the interface board. (b) Asymmetric acoustic lens on each 1D array for fixed angle off-axis beam steering in the elevation direction. (c) Diagram of cross-sectional carotid images obtained using the proposed scheme.
4.4.2.2 5-Plane CMUT arrays
CMUT arrays are designed to be interchangeable with a Siemens VF 13-5 SP
ultrasound probe to minimize modification to the ultrasound system. Thus, the geometry
of each 1D array (200 µm by 4 mm element size) as well as the center frequency (7.5
MHz) is determined by those of the VF 13-5 SP probes. To improve the device
uniformity, two additional dummy CMUT elements are included on either side of the 128
active elements. The active array area is roughly 25 mm by 26.4 mm (Figure 4-19). The
CMUT cavities are square shaped, with a side length of 40 µm. The plate is made of 1.5-
µm-thick single-crystal silicon. The vacuum cavity is 0.15 µm, over a 0.3 µm thick
silicon oxide insulation layer. For such a design, the expected plate resonant frequency is
12.11 MHz in air at 40 V DC bias, and the collapse voltage is 91 V. The key physical
parameters of the CMUT arrays are the same as summarized in section 4.3.
96
Figure 4-19. Active array area illustration.
The arrays are fabricated using SOI wafer-to-wafer fusion bonding [47]. Device
pictures after the completion of fabrication are shown in Figure 4-20. In the first
demonstration, a simple fan-out structure is used for electrical connections to the array
elements from the front side of the substrate. The total chip size is enlarged to 53 mm x
37 mm due to the electrical connection traces. In future designs, through-wafer trench-
isolated interconnects will be used to realize a reduced chip size. This will not only result
in a more ergonomic ultrasound probe, but also result in more chips on a silicon wafer,
thus reducing the cost.
(a) (b) Figure 4-20. (a) A 4-inch silicon wafer after the completion of fabrication. The center two devices are 5-plane CMUT arrays; the periphery contains test devices. (b) A 5-plane CMUT array diced off from the wafer.
First, the CMUTs were tested in air using an impedance analyzer (Model 8751,
Hewlett-Packard Company, Palo Alto, CA). The measured resonant frequency is 11.98
MHz at 40 Vdc, 0.13 MHz (1.1%) less than design target of 12.11 MHz. Across a 1D
97
array, the resonant frequency showed a uniform distribution, with a standard deviation of
0.18% of the mean (Figure 4-21). Therefore, in air testing, the fabricated CMUTs showed
excellent match to design targets.
Figure 4-21. Resonant frequency distribution across a 1D array.
Pulse-echo experiments were performed using a pulser-receiver (Model 5058PR,
Olympus NDT, Waltham, MA 02254) with arrays immersed in soybean oil. The oil-air
interface was used as a plane reflector. The CMUT showed a center frequency of 8.1
MHz, and a 6-dB fractional bandwidth of 103% (Figure 4-22). The notch in the
frequency spectrum at around 11 MHz is due to acoustic reverberation in the 400-µm
silicon substrate. It can be pushed out of the operating spectrum by making the substrate
thinner, or by attaching the substrate to a backing with matched impedance and with high
attenuation. The acoustic output pressure increased as the DC bias voltage and the AC
excitation voltage increased (Figure 4-23). Using an 80-V DC bias and a 100-V, 50-ns
AC excitation, the output pressure on the CMUT surface was measured to be 1.6 MPa
peak-peak.
98
Figure 4-22. Pulse-echo signal, and corresponding spectrum.
Figure 4-23. Normalized output pressure as a function of DC bias and AC excitation voltages.
4.4.2.3 Asymmetric acoustic lenses
For five parallel 1D arrays built on a single substrate, asymmetric acoustic lenses
can provide an ergonomic means of achieving off-axis steering. Such off-axis beam
steering is needed for Doppler measurements. Similar to section 4.3, PDMS was used as
the lens material. The off-centered cylindrical acoustic lens was investigated in this
section. The basic principle of off-axis steering is illustrated in [Figure 4-24(a)]. A simple
calculation yields that, to achieve as large of a steering angle as possible (small Doppler
angle), the ratio between the acoustic velocity in the lens material and that in water,
should be as large as possible [Figure 4-24(b)]. Common types of PDMS possess acoustic
99
velocities that would yield maximum steering angles of 20o-25o. In a first experiment, an
off-centered cylindrical lens was formed on a 1D CMUT element. A four-degree steering
angle was measured. A better lens material with a slower acoustic velocity is desired to
achieve a larger steering angle.
(a) (b) Figure 4-24. (a) Illustration of off-axis beam steering using an off-centered cylindrical lens. (b) Tilt angle as a function of the ratio of acoustic velocities in the lens material and in water (n = Cwater/Clens).
4.4.2.4 Peak blood velocity calculation
By cooperating with Jessica Faruque from Professor Sandy Napel’s group at
Stanford University, we developed a software package, which allows the peak blood
velocity (PBV) calculation. The PBV is automatically computed once color flow images
are obtained from the five parallel 1D arrays. The computation is carried out using a
software package developed in-house. In our preliminary experiments, the five color flow
Doppler images, overlaid on B-mode images, are obtained using a commercial 1D array.
A mechanical fixture is built so that the 1D array is tilted at a fixed angle, and can slide
along a rail. A straight tube vessel phantom was imaged with pumping blood-mimicking
fluid through the tube to simulate blood flow in an artery (Figure 4-25). The software first
removes the background B-mode images from the color flow Doppler images extracted
100
from the ultrasound imaging system. Then, the software determines the center points of
the vessel cross sections in the color flow images. Using this information, the vessel
trajectory and thereby the angle between the transducer and the vessel is calculated. The
PBV can then be found out by correcting for the Doppler angle. When the Doppler angle
is less than 45°, the PBV calculated using the proposed scheme has an error of less than
6% when compared to that determined by the conventional method.
Figure 4-25. Setup of using a 1D array to obtain five cross-sectional color flow Doppler images of a phantom.
4.4.2.5 Summary and future work
A new screening method for carotid stenosis is proposed. The key components of
the system include a 5-plane CMUT array, asymmetric acoustic lenses, an electronic
interface board, and a software package for automatic PBV calculation. The design,
fabrication and testing of the first generation 5-plane CMUT arrays were completed. The
arrays showed uniform element-to-element performance, and are suitable for the
proposed application. Preliminary experimental results validate the viability of the
asymmetric acoustic lenses. An in-house software package has been developed for PBV
calculation once the five color flow Doppler images are obtained. For future work, a 5-
plane CMUT arrays with through-wafer interconnects can be developed for compact
packaging. Different types of PDMS were suggested to find the best lens material for
improved steering angle.
101
CHAPTER 5 BACKSIDE INTERFACE ENGINEERING
OF CMUTS
The work on the backside interface engineering of CMUTs presents contributions
towards the electronic integration and packaging of large-area 2-D arrays. A very large
2D array is appealing for it can enable advanced novel imaging applications, such as a
reconfigurable array, and a compression plate for breast cancer screening. To achieve
these goals, the first large-scale fully populated and integrated 2D CMUTs array with 32
by 192 elements was developed. In this study, I demonstrate a flexible and reliable
integration approach by successfully combining a simple UBM preparation technique and
a CMUTs-interposer-ASICs sandwich design. The results show high shear strength of the
UBM (26.5 g), 100% yield of the interconnections, and excellent CMUT resonance
uniformity (σ = 0.02 MHz). As demonstrated, this allows for a large-scale assembly of a
tile-able array by using an interposer.
5.1 LARGE SCALE 2D ARRAYS
5.1.1 MOTIVATION
Ultrasonic imaging, shared with other imaging modalities, can be benefited by
large aperture size in terms of lateral resolution (5-1).
Lateral resolution ∝ 1 / Aperture size (5-1)
In addition, a number of specific applications exist in which large area ultrasound
transducer arrays can be used. These include cancer screening and continuous non-
invasive blood pressure monitoring [1]. A novel breast cancer screening can be realized
by placing very large ultrasonic transducer arrays on both upper and lower compression
plates in a manner similar to X-ray based mammography. A screening mammography
every 1-2 year was recommended by the “U.S. Preventive Services Task Force” for
women aged 40 and older. Although, mammography is the most commonly used
modality for early detection of breast cancer, there exists 15% of false-positive require
102
further ultrasound testing, and at least 10% of false-negative due to dense tissues
obscuring the cancer. One-way projection manner and poor resolution of mammography
cause the high false-positive and -negative rate, and results in unnecessary biopsy. By
using the dual ultrasonic imaging plates (Figure 5-1), compound images of a lesion using
data from both the upper and lower plates can minimize shadowing artifacts. Two-way
projection formed 3D data set can provide through-transmission measurements including
the speed of sound and attenuation. Complete 3D data sets can be acquired in renderings
made with 3D compounding to optimize achievable contrast and maximize speckle
suppression. Imaging techniques such as elastography can be applied in a more rigorous
manner than before including use of 3D [7, 8].
Bottom CMUTplate
Top CMUT plate
Camera unit
X-ray beam
Film plate
Figure 5-1. Schematic of the dual ultrasonic imaging plates on the current mammography setup.
5.1.2 RECONFIGURABLE ARRAY
A major challenge in future ultrasound imaging systems that features large area
transducers is the large number of interconnects existing between the signal processing
electronics and the transducer array. As systems move to larger and larger 2D arrays at
finer pitches, existing technologies for interconnect are increasingly inadequate.
Depending on the application, the element count for the large area transducer will range
from 10,000 to >1,000,000. Given the large number of transducer elements, each with its
own respective signal processing circuitry, significant power, cost, and area penalties
exist.
One attractive way to reduce the number of signal processing channels for such a
large area array is through the use of a mosaic annular reconfigurable array [4][5].
103
Ultrasound systems using reconfigurable array architectures can benefit from good image
quality with a significant reduction in the number of system channels. As illustrated in
Figure 5-2, the mosaic array architecture groups a number of sub elements together along
iso-phase lines to form larger transducer elements which are then each connected to a
single system channel. In this way, an array that has tens of thousands of active acoustic
sub elements can be reduced to a much smaller number of system processing channels
(e.g. 20-100). This greatly reduces the requirements on the system and makes possible
low power and low complexity electronics systems for large area arrays. For example, the
entire surface of both compression plates can be made to function as reconfigurable
transducer arrays. The reconfigurabilty provides the capability to form larger arrays in
any desired configuration such as an annular array; the annular array aperture will result
in the thinnest possible slice thickness, which, unlike existing devices, will be
dynamically focused over the entire depth.
In order to realize such array architecture, it is necessary to integrate switching
electronics immediately behind the acoustic array. These switching circuits, which are
realized using dedicated ASICs, connect directly to each respective sub element and can
be programmed to short these elements to one another in a reconfigurable manner. One of
the main challenges, which exist with such a system, is how to interconnect the large
number of transducers with a respective switching circuit on the adjacent ASICs.
Compared to traditional PZT-based ultrasonic transducers, CMUTs were made by
adopting MEMS technology. It allows CMUTs to be processed in a similar way to the
ASICs, which makes them more amenable to standard packaging flows.
Figure 5-2. Reconfigurable 2D CMUT arrays.
104
5.2 2D CMUT INTERCONNECTION AND 3D CMUT/IC INTERGRATIONS
The packaging regarding to CMUT interconnection and CMUT/ IC integration
can adopt the efforts developed in IC industry. The goal is to achieve a high density, and
high fill-factor 2D integration between CMUTs and ICs. One way of achieving this
integration is using System-on-a-Chip (SOC) to make both the electronics and CMUTs
on the same substrate (monolithic integration) [88][89][90]. SOC, featuring a complete
system built on a single chip, was expected to offer small product size, high speed, and
good reliability. Because CMUT is directly built on the substrate with the IC using a post
IC fabrication process, thermal, material, and processing method choices of the CMUT
are strictly limited [91]. Therefore, monolithically integrating IC and CMUT processing
generally compromise the CMUT performance, increases the process complexity, and
thus reduces device yield, which in turn increases cost.
The multichip module (MCM) or 2D System-in-Package (SIP) technology, on the
other hand, simply puts IC and CMUT chips together in a single package. In this case,
CMUTs and ICs are fabricated from each separated substrates, such that the performance
can be optimized. In 2D SIP, wire bonding can be used to connect 1D CMUT array
elements to ICs. However, direct wire bonding is not feasible for arrays with tight space
constraints, e.g., those for catheter-based applications. In addition, wire bonding is
difficult to apply on 2D CMUT arrays. First, the bonding wires cannot span over the
active surface, which causes interference with the acoustic performance. Second, the
routing of the electrical connection, from all elements to the bonding pads on device
edges, may cause huge area penalty, which in turn degrade CMUT performance, and
increase the parasitic capacitance. Thus, it is advantageous to interconnect the CMUT
elements by flip-chip bonding with through silicon via (TSV) using 3D SIP technology.
The main benefits of 3D SIP over other alternatives, e.g. SOC or 2D SIP, are
performance, form factor and cost [92]. Flip-chip assembly provides much shorter
interconnections through solder bumping. Thus, parasitic was greatly reduced, allowing
higher-speed signals. Because of the nature of ultrasonic transducer, CMUT surface are
required to interact with the outside world. This is in contrast to many popular MEMS
devices such as accelerometers and MEMS resonators, which can be isolated
105
hermetically. Therefore, it is necessary to provide electrical contact to CMUT elements
from the backside of the substrate. Developing proper through-wafer interconnect
methods is essential in this approach.
5.2.1 SURFACE MICROMACHINING THROUGH-WAFER VIA
Through-wafer interconnects can be implemented using two different approaches:
through-wafer vias (Figure 5-3). In the through-wafer via implementation, a conductive
material is used to fill the vias through the several-hundred-micron-thick silicon
substrate. The vias connect the front side of a silicon wafer, where the CMUT elements
reside, to the backside, where the flip-chip bond pads are located. In the IC industry, the
popular material for the conductor in the vias is copper (by electroplating), a standard
interconnect material [92]. However, copper becomes inappropriate for the high
temperature steps necessary for the CMUT fabrication processes. Polysilicon-filled
through-wafer vias are high-temperature compatible, as demonstrated by Chow et. al. in
2002 [93]. The method was first adopted for CMUTs by Cheng et. al. [6][94][95]. The
polysilicon is deposited using low pressure chemical vapor deposition (LPCVD), and is
heavily doped to reduce resistivity. The resulting resistance and parasitic capacitance are
suitable for CMUT applications. A surface micromachined CMUT array successfully
integrated with IC has been demonstrated [6][94][95]. By using this approach, a 16 by 16
array size can be realized [6][94][95].
(a) (b)
InsulatorConductor
Figure 5-3. Different approaches for through-wafer interconnect. (a) Through-wafer via, (b) through-wafer trenches.
In the meantime, wafer-bonded 1D CMUT array was developed showing superior
performance, simpler process, better plate uniformity, and allowing more flexible
geometry over the surface micromachined counterparts [47]. For example, large area
106
plates can be reliably made using wafer bonding, because the SOI provides stress free
film and the wet releasing step is no longer required. Wafer bonding offers the freedom to
choose the plate geometry in both the thickness and lateral directions. Wafer bonding also
ensures well-controlled plate and cavity quality, which improves device performance and
uniformity.
In order to realize a wafer-bonded 2D CMUT array, a through-wafer-via
technology compatible with the wafer bonding technique is required. However, the
through-wafer via method involves complex front and back side processes, including
multiple thin film depositions, doping and etching before the wafer bonding step.
Particularly, it is a “via-first” approach, such that the fusion-bonding step was proceeded
after the via processes. These processes introduce surface roughness, induce stress
imbalance on the wafer and cause wafer warping. As fusion bonding is extremely
sensitive to the substrate surface condition, so far no successful fusion bonded CMUTs
with through-wafer via was reported. To address this issue, Zhuang et al, have presented
a new TSV technology, through-wafer trench-isolated interconnections with a supporting
frame [7].
5.2.2 THROUGH-WAFER TRENCH-FRAMED INTERCONNECTIONS
In the through-wafer trench isolation approach, the silicon substrate is used as the
conductive material, while the trenches between array elements providing electrical
isolation. One such implementation was demonstrated by Lemmerhirt et al [96], however,
each interconnect is with a large footprint which becomes incompatible with CMUTs for
small pitches. Also, an extra connector structure is needed to hold each interconnect,
making the fabrication more complex.
Zhuang et al, developed a trench-isolated through-wafer interconnects on a
supporting frame successfully integrated with a wafer-bonded, fully populated 2D CMUT
arrays [7]. In this approach, the CMUT arrays are built on an SOI wafer and all electrical
connections to the array elements are brought to the backside of the wafer through the
highly conductive silicon substrate. Neighboring array elements are separated by trenches
on both the device layer and the bulk silicon (Figure 5-4). A mesh frame structure,
providing mechanical support, is embedded between silicon pillars, which electrically
107
connect to individual elements. A contact via on the device layer was opened by using
DRIE, and filled by doped polysilicon to build up the conductivity. In this way, a limited
size of via footprint, smaller than one CMUT cell, can be provided. The major benefits
over the through-wafer via technique is that the via was build after the fusion bonding
step. This “via-last” TSV technique, comparing to the “via-first” TSV, can be
successfully integrated to the wafer bonding process.
Backside trench
Frontside trench
Contact viaOne CMUT element
Figure 5-4. 3D schematic of the trench-isolated CMUT with a supporting frame.
A 16x16-element CMUT array with 250-µm-pitch was successfully fabricated
with good yield (100%), and great uniformity of pulse-echo amplitude (σ= 6.6%). This
results shows that through-wafer trench-isolation with a supporting frame is a viable 2D
interconnection solution to integrate with wafer bonding for delivering good
performance. Yet, the challenge is how to extend this approach to a larger scale, and with
smaller pitch-size.
5.3 INTERPOSER DESIGN AND UBM PREPARATION FOR LARGE SCALE 2D
ARRAY
An enabling technology for a large scale 2D array requires a simple, flexible and
reliable interconnection and IC integration technique. As an example of fabricating a
108
large scale 2D CMUT array using wafer-bonding, an interposer sandwich structure and a
pre-trench UBM preparation were proposed in this section. First, the approaches between
a single large die versus an array tiling assembly were discussed. The study introduced an
interposer design, which provides a flexible assembly scheme for various pitches
combination while enabling a large scale array with high yield. Second, the material and
fabrication of CMUTs considering the large-scale assembly was addressed. Last, a pre-
trench UBM preparation compatible to the trench-frame interconnection technique was
demonstrated. As the key parameter for enabling large-scale array tiling through flip-chip
bonding is the excellent uniformity of the solder height, the first solder-bumped trenched
2D CMUT array using the new UBM preparation was demonstrated.
5.3.1 SINGLE LARGE DIE VERSUS ARRAY-TILING ASSEMBLY
To fabricate a large-scale CMUT arrays, it can be realized as a single large die
[Figure 5-5(a)] or a tiling assembly module consisting of many smaller dies [Figure
5-5(b)]. Single large 2D array fabrication and assembly introduce several new challenges,
which are absent for small 2D array such as a 16x16 array. For example, to fabricate
100% yield on single large array is challenging. Moreover, under filling is more difficult
for larger die. The flatness requirement for an IC assembly integration is more
challenging as the die size becomes larger. Besides, it is not practical for ASIC die pitch
size being the same as of CMUT arrays.
Alternatively, array assembly by tiling provides high yield from more 100% yield
small arrays. The flexible size and pitch for both CMUT and ASIC sides are available. It
is also easy for bumping, soldering, and under filling. In addition, the flatness of CMUT
die is not as critical as of large single die.
109
ASIC ASIC
Interposer
ASIC
CMUT die #1 CMUT die #2
CMUT die
(a)
(b) Figure 5-5. Large-scale CMUT arrays can be made of (a) a single large die, or (b) an array tiling assembly.
To achieve an array tiling assembly module, several challenges were first
introduced. Upon using tiling array, sectioning of the sensor might affect the acoustic
performance depending on the die separation. The current center frequency design needs
this space to go down to 50 µm as the total non-active distance (separation plus the die
boundaries, ~22 µm) to be less than half of the wavelength (185 µm). Array assembly by
tiling requires an interposer featuring low dielectric loss, ease of integration, and flatness
fulfilling the imaging and assembly requirements.
A high-density Teflon interposer, HyperBGA® technology, produced by Endicott
Interconnect (EI) was used to design a modular package in which the sensor CMUT array
is on the topside and the supporting electronics on the backside. HyperBGA technology is
to use a PTFE-based (Teflon) interposer by laminating multiple layers for signal routing.
PTFE provides a low dielectric constant (εr = 2.7) and low dielectric loss (tan δ = 0.003),
which translates to superior electrical performance [97].
A standard HyperBGA® is composed of nine low loss copper metallization layers
(Figure 5-6). Specifically, the laminate contains two signal layers utilized for a majority
of the signal routing, Signal 1 and Signal 2. Both of these layers are completely
embedded in a true strip line environment (sandwiched between either Voltage 1 or
Voltage 2 and the center copper-invar-copper (CIC) ground plane). In addition to the
signal routing layers, are two redistribution signal layers, top surface redistribution (TSR)
and bottom surface redistribution (BSR). The redistribution layers are designed to
establish wiring channels, which allow the signal routing layers to efficiently match the
110
C4 footprint to the desired Ball Grid Array (BGA) footprint. The C4 layer on the very top
of the substrate contains only the C4 pads while the BGA layer on the very bottom of the
laminate contains only the BGA pads. Interlayer connections are accomplished with
microvias and plated through-holes (PTHs).
C4 padTSRVoltage 1Signal 1CICSignal 2Voltage 2BSRBGA pad
Figure 5-6. 9-layer HyperBGA® substrate
EI’s existing 9-layer HyperBGA® substrate was used for the assembly evaluation
and as a testing vehicle for signal routing to an imaging board through BGA. The pitch-
size difference between CMUT array and ASIC together with interconnection line width
and array size determine the total number of the routing layers. While 9 layers are
sufficient for this evaluation step, more signal routing layers can be added as a
modification enhancement to this existing 9-layered interposer substrate.
In general, with element pitch difference, one extra routing layer was added once
every “one element fall behind” (Figure 5-7). At CMUT pitch of 185 µm and ASIC pitch
of 150 µm (35µm difference for each cell), ASIC side falls behind by one element every
5.3 elements (185 µm / 35 µm = 5.3). The required routing layer count can be calculated
in two steps by first considering in one dimension (on the page), and then the second
dimension (into the page). For an array with 32 elements (mirrored 64 elements), 7
routing layers are calculated for one dimension:
32 elements / 5.3 elements = 6.1 layers; 6+1 = 7 routing layers. (5-2)
Then, for the second dimension (dimension into the page of Figure 5-7),
considering 25 µm for both the line width and line space, a 185 µm wide column can fit
around 3 routing channels:
111
columnchannelsmm
mspacelinewidthline
pitchCMUT /7.32525
185 =+
=+ µµ
µ . (5-3)
The routing layers can be reduced from 7 layers to 3 layers (7/3=2.3; 2+1=3) by
considering the second dimension. In this case, 4 routing layers will be used for
symmetry resulting totally 11 layers for future functional 150 µm-pitch ASIC/ interposer/
functional 185 µm-pitch CMUT assembly.
ASIC array with 150µm pitch
CMUT transducer array with 185 µm pitch
Interposer with signal routing layers
Mirrored
“one element fall behind”
Figure 5-7. Illustration of signal routing in the interposer.
5.3.2 CMUT DESIGN AND FABRICATION
The design goal of CMUT for breast cancer screening requires around 8 MHz
center frequency in immersion with 100% of FBW and 1 MPa minimum output pressure.
Based on the center frequency, 185 µm pitch was selected to prevent from a grating lobe.
To achieve the design criteria, an enhanced equivalent circuit model using MATLAB
script [8][10] was used, that includes imaginary part of mass loading effect [43]. Square
shaped plates, over circular shape, were chosen to achieve higher device fill factor.
Compared to rectangular plates, square plates yield a smaller fill factor, but increase the
separation between the adjacent modal frequencies, which reduced acoustic crosstalk.
The electrical contact via was located at the center of each element, and occupies one cell
per element.
Wafer-level yield was studied because the large-scale tiling assembly module
requires much more functional CMUT dies. Several SOI wafer specifications were
112
discussed as follow for yield improvement, including resistivity (doping concentration),
and the Si device layer and BOX layer thicknesses.
Previously, the interconnect resistance as a function of the silicon wafer resistivity
was calculated to be just a few ohms if the silicon wafer resistivity is below 0.05 Ω-cm
[Figure 5-8]. Low interconnect resistance benefits the output efficiency and receiving
sensitivity, however, there is a tradeoff between the resistance and the fusion-bonding
yield. As superior fusion bonding requires close bonding surfaces proximity (< ~10Å),
variation of surface topography greatly reduces the bonding yield. The thermal oxidation
growth rate differs due to “dopant enhanced oxidation” effect. The lower the resistivity,
the higher variation of the thermal oxide thickness. The commercially available SOI
wafer usually specifies resistivity variation, such as 0.01~0.025 Ω-cm. This axial
variation comes from the nature of CZ crystal growth that has freezing interface concave
into the melt [98] (Figure 5-9). While targeting the insulation layer of 3000 Å, the SOI
with resistivity of 0.01~0.025 Ω-cm (doping concentration of 3.3e18~8.1e18 cm-3) results
in 3000 Å ~ 3005 Å of oxide thickness variation over the wafer. However, as the
resistivity was lowered one order to 0.001~0.004 Ω-cm, the oxide thickness
nonuniformity increases up to 36 nm (3000 Å ~ 3354 Å). A short loop run comparing
two resistivities with all other parameters controlled was performed. The results show the
fusion-bonding yield largely was reduced by using the SOI with lower resistivity [Figure
5-10]. Taking the tradeoff into consideration, the SOI with resistivity of 0.01~0.025 Ω-
cm was selected.
Figure 5-8. Serial resistance versus the resistivity of the SOI substrate [99].
113
seedSilica crucible
Si-crystal
Si-melt
heater
Figure 5-9. Temperature profile in CZ Silicon growth [100]. The freezing interface of the Silicon crystal is concave into the melt.
(a) (b)
Good bonding yield across the wafer
Poor bonding arrays
Figure 5-10. Fusion bonding yields comparison between the SOI with resistivities of (a) 0.01~0.025 Ω-cm, and (b) 0.001~0.004 Ω-cm. Bonding yield of (a) is 100% for all 21 arrays, and down to (b) 66.7% (14 out of 21 arrays).
An increased BOX layer thickness is desired for a reduced parasitic capacitance.
The total parasitic capacitance as a function of the BOX layer thickness is calculated
indicating that a BOX layer of 2 µm results in a parasitic capacitance that is less than 0.4
pF. Compared to a typical device capacitance of ~ 2 pF for such an element, this is a
good compromise (< 20%). In addition, an increased Si device layer is desired for
114
assembly rigidity and yield. Regarding to the trench-frame-pillar structure, the frame
provides the rigidity of the CMUT array, while the device and BOX layers, where the
pillars attached, provides the rigidity to the flip-chip bonded module. The thicker the
device layer, the higher of the rigidity of the attached assembly module. This rigidity
becomes crucial for a good yield of the sandwich-assembly process. However, there is a
fabrication limitation of using very thick device and BOX layers. The thicker the Si
device/ BOX layers, the slower and longer is the BOX layer etching (Table 5-2-4). A test
run showed a multi-coated 2 µm baked Shipley 3612 photoresist provides sufficient
masking for the etching (~2 hour) of the 2 µm BOX layer at bottom of the 20 µm device
layer. For the same etching, a single coated Shipley 3612 resist is too thin and thick
Shipley SPR220 resist without baking suffered from burning during the etching. By using
this modified recipe, the device layer thickness can be increased from 10 µm to 20 µm,
which is essential for the assembly feasibility reported in section 5.4. An even thicker
device layer design will require a HF vapor etching of the BOX layer.
Table 5-1. Device parameters of CMUT for breast cancer screening.
Number of elements 256 (16x16) Element pitch, µm 185 Number of plates/element 24 Insulating layer thickness, µm 0.3 Vacuum cavity height, µm 0.15 Plate width, µm 28 Plate length, µm 28 Plate thickness, µm 1.0 Device layer thickness, µm 20
BOX layer thickness, µm 2.0 Silicon substrate thickness, µm 255 Front side trench width, µm 4 Backside trench width, µm 27.5 Frame width, µm 40
With all the design parameters (Table 5-1), the CMUT elements were first
fabricated on an SOI wafer with a 20-µm thick silicon device layer. Then, the electrical
contact vias were formed in each element. In a subsequent fabrication step, the backside
isolation trenches and the embedded silicon supporting frame were fabricated
simultaneously. The detailed fabrication steps are described below (Table 5-2).
115
Table 5-2. CMUT fabrication process flow.
Cartoon of process flow Description of the fabrication I.
CM
UT
cel
ls fo
rmat
ion
(1~3
) 1. Define cavities on an SOI wafer
Silicon Oxide
CMUT cavities (0.15µm of gap) with 0.3µm of insulation layer were defined on the device layer of a double-side-polished SOI wafer using double thermal-oxidation and BOE techniques.
2. Separate elements on device layer
The front electrical pads (element scale) were electrically separated by forming the frontside trenches (4µm wide) using a DRIE “Bosch process” step.
3. Wafer bonding and Remove handle wafer
Second SOI wafer carrying a silicon plate (1µm thick) was fusion bonded to the patterned oxide of the first SOI wafer and annealed at 1000oC for 2.5 hours. The handle layer of the second SOI wafer was removed in a heated TMAH solution.
II. F
ront
side
inte
rcon
nect
ions
(4~7
)
4. Contact via opening
Nitride
The nitride layer deposited atop of the BOX layer acts as a diffusion barrier in the upcoming doping step. Contact vias were opened in the center cell of each element by DRIE “Bosch process” with the BOX layer as etch stop. A dry plasma etcher was used to remove the BOX layer to open the access to the substrate layer.
5. Poly deposition, & doping
Doped polysilicon
A conformal deposition of a conductive material , phosphorus doped polysilicon, was used in the contact via to provide electrical continuity between the CMUT device layer and the silicon substrate.
6. Via protection and etch back to plate
It was etched back to the CMUT plates by removing doped polysilicon, nitride, and BOX layers. The via areas were protected during the etch-back.
116
7. Front side metallization and electrical isolation
Aluminum
The CMUT plates were remetallized with aluminum. The aluminum with plates, top electrodes of CMUT cells, were then isolated from the contact vias by etching back the surrounding trenches to the oxide insulation layer.
III.
Bac
ksid
e in
terc
onne
ctio
ns (8
~9)
8. Back side grinding and polishing
The silicon substrate was thinned down from the back side to 255 µm by mechanical grinding and polishing. Thinning down the silicon substrate reduced the series resistance and parasitic capacitance. In addition, this benefited a shortened back side trench etching time.
9. UBM, backside trench opening, and IC intergration
Interposer
ASIC
UBM Solder bump
A newly developed process steps including the under-bump metallurgy (UBM), backside trench and the IC integration will be discussed in detail in section 5.3.3.
5.3.3 UNDER-BUMP-METALLURGY DESIGN AND FABRICATION
In previous work, an UBM was implemented to enable the integration of a 16x16
2D CMUT array with ASIC. This was accomplished by using a post-trench UBM formed
by a 45o-tilting metal evaporation (Figure 5-11). This approach was for demonstration
purpose for a relatively small 2D array. Because the UBM pad is unconfined, the use of
solder is limited to the IC side during solder installation. For the same reason, it also
degrades the uniformity of the solder ball height, which makes large scale assembly
impossible. A large-scale tiling sandwich assembly requires a flexible, simple, and
reliable UBM preparation, which is compatible to the trench-frame process. In the
following sections, the design, fabrication, and test results of the newly developed post-
trench and pre-trench UBMs are described. The first solder-bumped trench-framed
117
functional 16x16 2D CMUT array using the new pre-trench UBM is demonstrated. The
first successful large-scale tiling CMUT assembly is shown in section 5.4.
b-2. Back side etch.
b-3. Dicing and solder bumping.
interposer
ASIC
a-1. Backside grinding and etching.
a-2. UBM patterning, and dicing (solder bump on IC).
ASIC
SiO2Si
Doped polysilicon
UBM
Al
b-1. Backside grinding and UBM patterning.
b-2. Back side etch.
b-3. Dicing and solder bumping.
interposer
ASIC
a-1. Backside grinding and etching.
a-2. UBM patterning, and dicing (solder bump on IC).
ASIC
SiO2Si
Doped polysilicon
UBM
Al
b-1. Backside grinding and UBM patterning.
Figure 5-11. Schematic comparison between the previous single-die UBM technique and the proposed pre-trench UBM for tile able sandwich structure.
There is a new IBM C4NP Cu-cored Pb-free flip chip process [101]. This process
is currently being used for fine pitch solder bumps on large die with thousands of I/Os
with great success. The C4NP Cu-cored process offers the advantages of stress reduction
and improvement in electrical performance. The center Cu sphere encased in the solder
enables a higher stand-off, especially for fine pitch interconnects, helping to facilitate the
possible further under fill process and improve the overall mechanical fatigue life of the
interconnect. The Cu sphere, which comprise a large percentage of the volume of the flip
chip interconnect, has a low resistivity which can therefore increase the current carrying
capacity for the interconnect. The Cu sphere can be selectively placed to only those
interconnections designed for optimum electrical performance. The highly stressed
interconnects can be electrically connected without any Cu spheres to maximize their
118
reliability. An adapted plating process was used at Fraunhofer IZM, Berlin, Germany to
deposit solder on the trenched 4” CMUT wafers. The process step (Figure 5-12) is first
sputtering of TiW/Cu as plating base. After applying the lamination of dry film resist and
patterning by lithography, electroplating of Cu UBM and Sn-Pb solder was performed. At
last, the dry film resist was stripped away, and plating base was removed by wet etching.
electroplated solder
electroplated UBM
Sputtered Cu as plating base
Sputtered Ti/W as adhesion layer and diffusion barrier
Al chip pad with overlapped passivation(a) (b)
Figure 5-12. Schematic process of the pre-trench UBM using electroplating (a) before and (b) after oven reflow.
A challenging part of this process is to form a continuous plating seed layer by
sputtering into the trench-frame structure. In order to study this issue, a mechanical test
vehicle (a trench-framed SOI wafer without front side active device) was used for
Fraunhofer’s plating process development. After the trenched CMUT wafers were
processed, the deposition of TiW/Cu plating seed layer [Figure 5-13(a)], the resist
lamination and lithography [Figure 5-13(b)] were performed accordingly. Then, after
electroplating, it was observed not all of the bump positions were plated and variation
presents in solder heights [Figure 5-13(c)]. Cross sectioning was performed for both a
non-bumped and a bumped pillar after plating in order to correlate any feature of the
trench with a possible cause for the inability to plate a solder bump. It was determined
that the suspected cause for the poor bump plating resulted from an interrupted plating
base in the trench. Upon cross-sectioning a trenched CMUT with no bumps, one can
clearly identify the problematic area as an undercut of the Si pillar [Figure 5-14(b)]. This
is due to the etching no uniformity within the wafer. The over etching causes charging of
119
the etch-stop layer (oxide), that redirects the etchant to attack laterally. Figure 5-15 also
shows schematically how the negative slopes of the side walls, and footing of the pillars
cause the discontinuity of the plating seed layer, and fail the electroplating.
(a) (b) (c) Figure 5-13. Plating process steps
(a) (b) Figure 5-14. Comparison between the pillars (a) with and (b) without footing.
In order to verify the cause and provide a possible solution for the seed layer
discontinuity, a process has been developed to fill the base of the trenches with
Benzocyclobutene-based polymers (BCB). The bumping results in the areas with and
without the BCB trench filling was shown in Figure 5-16. Successful bumping was
observed while the trench was filled with BCB polymer; in the contrast, the missing
bumping happened, as the trench has no filling BCB polymer. The process provides
possible solution, however, introduces additional step. Due to the nature of the
120
complexity of the post-trench UBM, pre-trench UBMs were studied and described in the
following sections.
Dry film
•UBM/Solder
Plating baseFooting Negative
trench slope Figure 5-15. Schematic of the post-trench UBM with (a) straight sidewalls of trenches, and (b) trenches with negative sidewall slope and footing.
Figure 5-16. Post-trench UBM and solder-bumping in the area with and without BCB trench-filling.
A commercially available pre-trench UBM is available at Pac-Tech, Santa Clara,
USA. By using a solder-jetting technique, versus the electroplating, UBM step is
independent of the solder bumping and can be deposited before the trench opening. Then
the solder bumping was performed as a last step after the trench deep etching. However,
this chemical plating techniques requires passivation layer for UBM pad definition.
Passivation layer, such as oxide, can be grown at as low as 450oC, e.g. low temperature
oxide (LTO), or 425oC, e.g. spin-on glass. Because the presence of the aluminum as the
top electrode of CMUT, the thermal cycle introduced by the oxide make this chemical
plating UBM technique become incompatible to the existing CMUT process.
121
In order to provide a simple and reliable UBM to incorporate the trench-frame
interconnection technology, a pre-trench UBM was developed by e-beam evaporation of
an Al/Ti/Ni/Au metal stack. Among the metal stack, Ti serves as an adhesion layer and
diffusion barrier. While Au provides wettability to the solder, Ni acts as additional
standoff and diffusion barrier, and becomes the intermetallic layer for the primary
bonding. The e-beam evaporated Ni provides better quality in terms of voids, which
minimizes the solid-solubility induced reliability issue. The Ni thickness was chosen to
be thin enough to minimize stress, yet sufficiently thick to endure double reflow of the
solder. Three different metal stacks in terms of the thicknesses of Ni (0.5µm, 0.4µm, and
0.3µm,) and Au (0.03µm and 0.1µm,) were studied. The process step was first the wafer
substrate cleanness to secure the UBM/substrate and resist/substrate adhesion. Then the
pad was defined by using lift-off bi-layer resist stack (LOR, MicroChem Corp., Newton,
MA), where LOR 5B was used according to the total UBM thickness. After the
lithography, a sequence of Al, Ti, Ni and Au were evaporated without breaking vacuum.
The UBM pads were released following the Acetone bath as a lift-off process step. At
last, the wafers were performed solder bumping and bump shear test at Pac-Tech, Santa
Clara, USA.
Among various Ni/Au thicknesses, the wafer with 0.5 µm Ni/ 0.03 µm Au was
observed showing the cracks through the photoresist layer after the metal evaporation
(Figure 5-17). This is due to an intrinsic tensile stress within the Ni film over the
adhesion strength of resist/ substrate interface. As a result, a misalignment of the metal
pad was introduced through the movement of the resist. It suggests a thinner Ni layer
should be used to minimize the intrinsic stress. After performing the bump shear test, the
strength results show only 8.4 g along with a brittle failure mode (Table 5-3). The bump
shear strength is much lower than the suggested value, 26 g, according to a 70 µm pad
size (Figure 5-20). A second oven reflow was tested to see if a larger shear strength can
be achieved, though only a very limited strength was increased (8.4g to 11.4g). Upon
examining a SEM, the pad before bumping shows the surface roughness with the grain
size around tens of nanometers (Figure 5-18). Nickel surface oxidize fairly quickly if a
noble metal, Au, is not thick enough to protect the surface from oxidation. The Au layer
122
thickness, 0.03 µm, compared to the grain size implies a thicker Au was needed to
maintain the wettability. Another SEM shows the bump shear failure interface (Figure
5-19). The nickel layer shows a full-pad intermetallic layer is not present, and implies the
non-wettability of Ni causes a brittle failure and low bump shear strength.
Pad
Photoresist
Exposed Si substrate
Crack through photoresist layer
100 µm
Figure 5-17. An optical microscopy showing the crack through the photoresist layer after the e-beam evaporation of UBM metal stacks. The crack was due to the intrinsic compressive stress of the 0.5-µm Ni film.
(a) (b)
(b)
Figure 5-18. SEM pictures show the UBM pad with 0.03µm Au and 0.5um Ni has huge surface roughness by observing the grain size around tens of nanometers.
Then a 0.4 µm Ni/ 0.1 µm Au UBM was studied by taking advantages of a thinner
Ni to reduce the intrinsic stress and thicker Au for better wettability protection. Though
the intrinsic stress induced resist cracking still presented, the shear strength demonstrated
a promising improvement. The average shear strength was increased to 24.3 g along with
a ductile shear failure mode indicating the preferable failure interface through the solder.
123
A large deviation, 5.8 g of standard deviation, and the substrate failure mode simply
suggested a more powerful substrate cleaning preparation.
(a) (b)
(b)
Solder ball
UBM pad after shear test
UBM pad underneath was not shown
Figure 5-19. SEM pictures show the UBM pad shear interface of the brittle failure mode.
Table 5-3. Bump shear test results of various UBM metal stacks.
UBM* 0.5 µm Ni/ 0.03µm Au
0.4 µm Ni/ 0.1µm Au
0.3 µm Ni/ 0.1µm Au
Shear mode Brittle mode
Ni
Partial ductile mode
Solder
Substrate mode
Si substrate
Partial ductile mode
Solder
Ductile mode
Solder
Number of test 30 30 60 60
Bump shear strength (g)
Mean 8.4 11.4** 24.3 26.5
StdDev 1.7 0.8** 5.8 1.3
* UBM includes 0.3µm Al/ 0.015µm Ti before Ni and Au. ** Two oven reflows.
124
40 60 80 100 1200
20
40
60
80
100 Minimum sample averageMinimum individual shear reading
Shea
r st
reng
th (g
)
Ball bond diameter (µm) 0.3 µm Al0.015 µm Ti
0.5 µm Ni
0.03 µm AuBrittlemodex
Ductile mode
0.3 µm Al0.015 µm Ti0.3 µm Ni
0.1 µm Au
x
Before reflowAfter reflow
Failure interface
Before reflowAfter reflow
Failure interface
Figure 5-20. Suggested bump shear strength and the results from the pre-trench UBM using evaporated Al/Ti/Ni/Au.
After all the aforementioned studies, the 0.3 µm Al/ 0.015 µm Ti/ 0.3 µm Ni/ 0.1
µm Au was finally used and provides the strongest bump shear strength (µ = 26.5 g) with
a preferred ductile shear failure mode. It also shows an excellent uniformity (σ = 1.3 g)
after using a standard RCA substrate cleaning with double dump-rinse and spin-dry. The
0.3 µm Ni limits the intrinsic stress and prevents from the resist-cracking induced pad-
misalignment. Only 80 nm of the Ni layer was consumed after the second oven-reflow,
indicating that 0.3 µm of Ni is adequate for a double-reflow flip-chip-bonding
application, which was used in our final assembly with the interposer. After solder
bumping, the first trenched CMUT arrays with excellent quality solder bumps were
demonstrated (Figure 5-21). The solder balls showed accurate height (µ = 59.7 µm versus
the target of 60 µm) and excellent uniformity (σ = 0.9 µm), which is essential for large-
scale tiling assembly. In conclusion, this simple UBM preparation, compatible with the
trench- frame technology, does not require electro plating or chemical plating techniques
with passivation layer for UBM pad definition; by optimizing the metal stack thicknesses,
a 0.3 µm Al/ 0.015 µm Ti/ 0.3 µm Ni/ 0.1 µm Au UBM can meet the flip-chip bonding
design criteria.
125
90 µm
Ground pad
(a)
(b)
Figure 5-21. A SEM pictures of the first trenched CMUT arrays with excellent quality solder bumps showing (a) an excellent yield on a 16 by 16 2D array with excellent solder-ball-height uniformity. (b) A close-up of a signal pad shows the solder ball on top of the trench-isolated 90x90-µm-wide and 255-µm-tall pillar.
The electrical input impedance was measured to determine the interconnection
yield and the performance of the ultrasonic transducers after solder bumping. An
impedance analyzer (Model 4294A, Agilent Co., Palo Alto, CA) was performed to carry
out this measurement in air (Figure 5-22). The probing was on the solders of one ground
element and sweeping across all 255 signal elements through the solder-UBM-pillar-via-
transducer interconnect. At the same time the impedance was measured by sweeping the
frequencies of a small AC excitation (50 mV) superimposed on top of a DC bias of 80 V
(80% of the plate pull-in voltage) [Figure 5-22(a)]. The results show 100% yield of the
interconnections from all 255 elements and great uniformity of the resonant frequency of
the ultrasonic transducers (µ = 15.4 MHz, σ = 0.02 MHz) [Figure 5-22(b)]. These
promising results indicate the new UBM preparation is compatible to the trench-framed
process, and suitable for developing large-scale tiling assembly.
126
5 10 15 20-40
-20
0
20
40
Frequency (MHz)
Inpu
t im
peda
nce
(kΩ)
Real partImaginary part
(a)
510
15
5
10
155
10
15
20
Reso
nant
freq
uenc
y (M
Hz)
Element numberElement number 1
µ= 15.4 MHzσ= 0.02 MHz
510
15
5
10
155
10
15
20
Reso
nant
freq
uenc
y (M
Hz)
Element numberElement number 1
µ= 15.4 MHzσ= 0.02 MHz
(b)
Figure 5-22. (a) Electrical input impedance of one element, and (b) resonance frequency distribution across the 16x16 array.
5.4 LARGE SCALE TILING CMUTS
This section discusses the results of the tiling CMUT arrays. By using the newly
developed UBM technique with the interposer technology, a flip-chip bonded CMUTs-
interposer-ASICs sandwich assembly module can be realized. First, the design of a 4x12
tiling assembly module featuring a multi-row linear imaging array was discussed. Then a
127
test assembly of a 2x12 tiling CMUT assembly was demonstrated. Finally, a full
functional 1x2 tiling assembly module was built and a pulse-echo acoustic measurement
was realized.
5.4.1 TILING ASSEMBLY MODULE DESIGN
In the design of the final functional module, there is a 4 x 12 array of die closely
spaced on an organic interposer substrate to create a 12 mm x 36 mm CMUT transducer
probe. The main purpose of this package was to test the functionality of a large CMUT
sensor array for various medical applications. As one of an example for demonstration,
the interposer was designed with a routing scheme for a multi-row linear array. For this
design, a 40 mm HyperBGA® interposer was used that accommodated a 4x12 array of
CMUT devices in the center of the module (Figure 5-23). The routing scheme, as a 1.5D
linear array requires, features three groups of elements to address the phase focusing
capability in elevation direction (Figure 5-24). Essentially the functional CMUT region
covers an area of about 6 mm x 36 mm. Two internal rows of functional CMUT devices
were required while the two outer rows are dummy CMUT die for boundary condition
continuity. On the backside of this functional CMUT assembly module, three non-
functional ASIC chips were attached to simulate the stresses generated along with the
packaging. Each ASIC die has ~4100 I/O pads with a pitch down to 150 µm. Both
eutectic Sn-Pb and Pb-rich solder bumps will be evaluated. The module backside also
features 648 1.0-mm-pitch BGA pads. The BGA pads use a eutectic Sn-Pb solder alloy.
The module was attached to an imaging motherboard using a standard BGA solder attach
process.
Interposer
Imaging board
Functional CMUTs : 2 inner rows (yellow) Mechanical CMUTs
: 2 outer rows (blue)
Figure 5-23. Schematic cross-section of the modular package design.
128
One 16x16 CMUT array
Signal routingSignal routing
Figure 5-24. Schematic signal routing scheme for a multi-row linear imaging array.
5.4.2 TILING ASSEMBLY RESULTS
To demonstrate the feasibility of the large-scale tile-able array assembly, 2x12
solder-bumped trench-framed dummy CMUT arrays were flip-chip bonded to an
interposer. 1x3 non-functional ASIC chips were attached on the backside of the
interposer by flip-chip bonding. The results show the successful assembly with CMUT
arrays placed only 100 µm in separation (Figure 5-25). The whole module was
maintained at the specified coplanarity requirement of ±10 µm. Assembled CMUTs have
totally 6144 (2x12x256) interconnections with pitch down to 185 µm, and each ASIC die
has ~4100 I/O pads with a pitch of 150 µm. This promising results demonstrate the first
viable large-scale tiling CMUT assembly and enable many advanced imaging modalities.
Ultimately, this would be the building block to create not only CMUT sensor arrays, but
also a generic packaging solution for any type of sensor array with the backside support
electronics.
interposer
CMUT
Figure 5-25. Top view photograph of the first large-scale tile-able CMUT array assembly.
129
CMUT
interposer
C CMUT
interposer
C
0.2 mm Figure 5-26. Cross-section SEM of the assembled CMUTs-interposer-ASICs sandwich structure.
5.4.3 ACOUSTIC MEASUREMENT RESULTS
In cooperation with GE Global Research, Niskayuna, NY, a functional module
having a 1x2 CMUT array was attached to an imaging motherboard for initial acoustic
measurements. The imaging board provides connection to the system for acoustic testing
and imaging. Figure 5-27 shows a topside view of the 1x2 functional trenched CMUT
arrays flip-chip bonded to an interposer. Figure 5-28 demonstrates the whole module was
bonded to the imaging board using a standard BGA solder attach. The functional CMUT-
interposer module was tested for pulse/echo response in order to generate the time
domain impulse response as shown in Figure 5-29(a). The resulting spectrum is shown in
Figure 5-29(b). This promising results demonstrate the success of the packaging and
electronics integration method for a large-scale tile-able 2D array.
Interposer
CMUTs
Imaging board
Figure 5-27. Photograph showing topside view of the 1x2 functional CMUT array on a laminate interposer solder attached to a board.
130
Interposer CMUTs
Imaging board
Figure 5-28. Photograph showing the cross-sectional view of the functional prototype.
Frequency (MHz)Time (µs)
Volta
ge (m
V)
Nor
mal
ized
pre
ssur
e (d
B)
0 5 10 15 20-60
-50
-40
-30
-20
-10
0
0 0.5 1 1.5 2-100
-50
0
50
100
150
(a) (b) Figure 5-29. The pulse/echo results from the CMUT-interposer assembly module.
5.5 SMALL PITCH DESIGN
It was demonstrated the 185-µm tall 90x90-µm wide pillar could endure the
interposer assembly process. For finer element pitch sizes and wafer thicknesses, a design
guideline was developed by using a beam theory calculation and finite element analysis
(FEA). First, the induced maximum tensile and shear stress were calculated at the bottom
of the signal pillar under the bump-shear force acting on the top of the pillar (Figure
5-30). The pillar bottom reflects the Si-BOX interface or the interface within the BOX
layer. The analysis predicted less than 4.5 g of bump shear force can result in a tensile
stress level of tens of MPa, which reflects the fusion bonding strength [102] and the
tensile strength of thermal-grown silicon dioxide. The FEA result shows the same stress
level thus agrees with the calculation (Figure 5-31).
This analysis was further verified by a shear strength test on the solder-bump-
pillar [Figure 5-33(a)]. Five sample data all show between 3.4 g and 3.6 g (µ = 3.5 g, σ =
131
0.1 g), which agree with our prediction. The post-failure patterns [Figure 5-33(b), and
(c)], showing the remaining oxide on the bottom of the trench, indicated that the failure is
through the BOX-Si interface and/or the BOX layer of the silicon on insulator (SOI)
wafer. For even finer pitch designs, the fusion bonding adhesion should be enhanced,
and/or the pillar length should be shortened.
Pillar width (µm)
150 180 210 240 270 3000
20
40
60
80
100
Pitch (µm)
Tens
il st
ress
& s
hear
stre
ss (M
Pa)
55 85 115 145 175 205
0
20
40
60
80
100
Width (µm)
1 g
3 g5 g
10 g
20 g30 g
5 g10 g
30 g
Current designElement pitch (µm)
Indu
ced
max
imum
stre
ss at
the
pilla
r bot
tom
(MPa
)
*Bump shear test result
- Tensile- Shear
Figure 5-30. Beam theory calculated induced maximum tensile and shear stress at the bottom of the signal pillar under the bump-shear force acting on the top of the pillar.
Figure 5-31. Finite element analysis agrees with the beam theory calculation with only 6.7% error.
132
1
11
3.5
3.5
3.5
3.5
5
5
5
5
10
10
10
1020
20
20
20
3.5 3.5 3.5
5 5 5 5
10 10 10
20 20 20
Pillar length (µm)
Pilla
r wid
th ( µ
m)
50 100 150 200 250 300
40
60
80
100
120
140
160 Shear test (g) for tensile failureShear test (g) for shear failure
Assuming tensile bonding strength = 80MPa& shear bonding strength = 10MPa [3]
*
x
* 185-µm pitch designx 250-µm pitch design
Figure 5-32. Design guideline for pitch (pillar width) and substrate thickness (pillar length) with experimental-determined fusion-bonding strength. The tensile fusion bonding strength was based on the bump-shear test of 3.5 g results with the beam theory calculation. The yellow region indicates the verified safe zone for various pillar width and length designs.
Shear force
A A
A A BB BOX
pilla
r
fram
e
pilla
r
B B
Delaminated BOX
Contact via
Device Si
Frame
BOX
90 µmContact via
(a) (b) (c)
Figure 5-33. (a) Cartoon showing the bump-shear test on a signal pillar. (b) SEM on the bottom of a post-shear-failure pillar showing the presence of a BOX layer. The shear failure interface is through the Si-BOX interface but not the Si pillar. (c) Microscopic optical picture shows the bottom of the trench with the fringing patterns within the 90x90 µm square. The fringing pattern indicates the delaminated BOX from device Si layer.
133
5.6 CONCLUSION
The work on a packaging and electronics integration method of large-scale tile-
able 2D array devices with small pitch-sizes has been demonstrated. A pre-trench UBM,
by e-beam evaporating Al/Ti/Ni/Au metal stacks, was developed to incorporate to the
trench-frame interconnect technology. This simple UBM was proved to provide good
bump shear strength and excellent solder ball height uniformity along with a preferred
ductile shear failure mode. Test results show 100% yield of the interconnections and
great integration uniformity as indicated by the air resonance performance over 255
transducer elements.
A tile-able module assembly was demonstrated and the CMUTs-interposer-ASICs
sandwich by a complete flip-chip attach. Future work includes the assembly of large-
scale functional CMUT chips by tiling and the acoustic tests on such assemblies.
A design guideline for various element-pitch and wafer-thickness based on
trench-frame interconnection technology was developed. In the current design, the 185-
µm tall 90x90-µm wide pillar can endure the interposer assembly process. The bump-
shear test shows that the failure is through the Si-Box interface and/or the BOX layer. For
designs with smaller pitch, the fusion bonding adhesion should be enhanced (if possible),
and/or the pillar length should be shortened.
134
CHAPTER 6 CURRENT AND FUTURE WORK
The frontside interface study in this thesis indicates the encapsulation material
optimized for CMUT requires a viscoelastic polymer with Tg lower than room
temperature. Various polymer alternatives other than PDMS exist, such as polypropylene
(PP), polybutane (PB), polyisobutylene (PIB), and polyethylene (PE) etc. Investigation of
these polymers regarding to a lower attenuation and better abrasion resistance is another
topic of research for the future.
As CMUT has been used as a platform technology, featuring many applications in
addition to medical imaging [103], [104], different encapsulation requirements exist.
Several applications require the close interaction of the CMUT surface with the
substance. For example, CMUTs can be designed as the actuator arrays for particle
transportation in immersion by utilizing the surface acoustic wave. In that case, a thick
polymer coating with attenuation property becomes unsuitable. Encapsulation material
other than polymer is also required for a CMUT as a high-temperature gas-flow sensor.
For all the above applications, coating by the atomic layer deposition (ALD) technique is
appealing because of the capability of pinhole free quality with thickness of only tens of
nanometers. A related topic is the ALD passivated CMUTs for specific sensor or actuator
applications.
Currently, integration of functional 16x16 2D CMUT array dies into the bigger
tiling assembly (256 by 128) and the CMUT-interposer-ASIC sandwich module is under
way. The next goal is to demonstrate the wire-phantom imaging compared with that from
a commercial imaging system. To achieve this goal, a CMUT-element-level defect-free
or a defect-immune assembly module is required. A CMUT sensitivity achieved by the
optimized DC bias is lost if one defect results in a short circuit. The short circuit drains
current from the DC supply and at least reduces the optimum bias operating voltage. Two
approaches are being investigated to reduce the dependence on short circuits: 1) the
capability of the batch characterization of the CMUT element yield and next the isolation
of the defective elements from the system-level interconnections; 2) a defect-immune
135
design by incorporating a fuse-like interconnections to each CMUT element to prevent
DC leakage.
To expand the current design (7.5 MHz) to a higher frequency is another topic of
research for the future. One of the challenges following the smaller-pitch design is a
tighter tiling assembly, i.e., a smaller dead area of a CMUT die. In the current design, the
conventional wafer saw dicing was used to achieve a 25 µm chipping level. It is desired
to use a chipping-free technique to result in a dead-boundary free CMUT die. One of the
solutions is to use the deep etching for die separation. Another option is to use a novel
laser dicing technique. The internal absorption laser process, versus the conventional
surface absorption one, generates no debris by creating the cracks within the substrate.
The following crack extension is expected to provide a chipping-free singulation of the
CMUT die. Incorporation with these methods into the CMUT fabrication steps and the
tiling assembly process is the research topic for the future high frequency design.
136
BIBLIOGRAPHY
[1] J.L. Orbay, W.G. De Quevedo, A. Espinosa, and M. Moore, “Non-Invasive Blood
Pressure Monitoring Device and Method,” U.S. Patent 20100106016Al, Apr. 29,
2010.
[2] C.M. Sehgal, S.P. Weinstein, P.H. Arger, and E.F. Conant, “A review of breast
ultrasound.,” Journal of mammary gland biology and neoplasia, vol. 11, Apr. 2006,
pp. 113-23.
[3] Z. Hui, O. Bing, B.-M.L. Luo, X. Feng, Y.-L. Wen, and H.-Y. Yang, “Comparison
of ultrasound elastography, mammography, and sonography in the diagnosis of
solid breast lesions,” Journal of Ultrasound in Medicine, vol. 26, 2007, p. 807–815.
[4] R. Fisher, K. Thomenius, R. Wodnicki, R. Thomas, S. Cogan, C. Hazard, W. Lee,
and D. Mills, “Reconfigurable Arrays for Portable Ultrasound,” Proc. IEEE
International Ultrasonics Symposium, 2005, pp. 495-499.
[5] K. Thomenius, R.A. Fisher, D.M. Mills, R.G. Wodnicki, C.R. Hazard, and L.S.
Smith, “Mosaic Arrays using Micromachined Ultrasound Transducers,” U.S.
Patent 6865140 B2, Mar. 8, 2005.
[6] C.-H. Cheng, A.S. Ergun, and B.T. Khuri-Yakub, “Electrical through-wafer
interconnects with 0.05 pico farads parasitic capacitance on 400 um thick silicon
substrates,” Proc. Solid-State Sensor and Actuator Work-shop, 2002, pp. 157-160.
[7] X. Zhuang, I. Wygant, D.-S. Lin, M. Kupnik, Ö. Oralkan, and B.T. Khuri-Yakub,
“Wafer-bonded 2-D CMUT arrays interconnects with a supporting frame,”
Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, vol. 56,
2009, pp. 182-192.
[8] F.V. Hunt, Electroacoustics: The Analysis of Transduction, and its Historical
Background, New York: Acoustical Society of America, 1982.
[9] E.V. and T. Krauthammer, Thin Plates and Shells: Theory, Analysis, and
Applications, New York: Marcel Dekker, Inc., 2001.
137
[10] W.P. Mason, Electromechanical Transducers and Wave Filters, 2nd ed., New
York: D. Van Nostrand Co., 1948.
[11] S.P. Timoshenko and S. Woinowsky-Krieger, Theory of Plates and Shells, 2nd ed.,
New York: McGraw-Hill Book Company Inc., 1959.
[12] W.H. Peake, “The lowest resonant frequency of a water-loaded circular plate,” J.
Acoustic Society of America, vol. 26, 1954, pp. 166-168.
[13] J.D. Irwin, Basic Engineering Circuit Analysis, 7th ed., Upper Saddle River, New
Jersey: Prentice-Hall, 1996.
[14] G.S. Kino, Acoustic Waves: Devices, Imaging, and Analog Signal Processing,
Englewood Cliffs, NJ: Prentice-Hall, 1987.
[15] O. Oralkan, B. Bayram, G.G. Yaralioglu, a S. Ergun, M. Kupnik, D.T. Yeh, I.O.
Wygant, and B.T. Khuri-Yakub, “Experimental characterization of collapse-mode
CMUT operation.,” Ultrasonics, Ferroelectrics and Frequency Control, IEEE
Transactions on, vol. 53, Aug. 2006, pp. 1513-23.
[16] Y. Huang, E. Haeggström, B. Bayram, X. Zhuang, A.S. Ergun, C.-H. Cheng, and
B.T. Khuri-Yakub, “Comparison of conventional and collapsed region operation of
capacitive micromachined ultrasonic transducers.,” Ultrasonics, Ferroelectrics
and Frequency Control, IEEE Transactions on, vol. 53, Oct. 2006, pp. 1918-33.
[17] X. Jin, O. Oralkan, and F.L. Degertekin, “Characterization of one-dimensional
capacitive micromachined ultrasonic immersion transducer arrays,” Ultrasonics,
Ferroelectrics and Frequency Control, IEEE Transactions on, vol. 48, May. 2001,
pp. 750-760.
[18] V. Senez, D. Collard, B. Baccus, M. Brault, and J. Lebailly, “Analysis and
application of a viscoelastic model for silicon oxidation,” Applied Physics, Journal
of, vol. 76, 1994, pp. 3285-3296.
[19] M. Madou, Fundamentals of Microfabrication, 2nd ed., New York: CRC, 2001.
[20] J. a Greenwood and J.B.P. Williamson, “Contact of nominally flat surfaces,”
Proceedings of the Royal Society A: Mathematical, Physical and Engineering
Sciences, 1966, pp. 300-319.
[21] A. Eisenberg, “An Ultrasound That Navigates Every Nook and Cranny,” The New
York Times, 2004.
138
[22] O. Oralkan, A.S. Ergun, J.A. Johnson, M. Karaman, U. Demirci, K. Kaviani, T.H.
Lee, and B.T. Khuri-yakub, “Transducers : Next-Generation Arrays for Acoustic
Imaging ?,” Ultrasonics, Ferroelectrics and Frequency Control, IEEE
Transactions on, vol. 49, 2002, pp. 1596-1610.
[23] “Velocity of sound in some biological materials,” [Online]. Available:
http://www.drgdiaz.com/tables.shtml, 2010.
[24] A. Caronti, R. Carotenuto, and M. Pappalardo, “Electromechanical coupling factor
of capacitive micromachined ultrasonic transducers,” J. Acoust. Soc. Amer, vol.
113, 2003, p. 279–288.
[25] G.G. Yaralioglu, A.S. Ergun, B. Bayram, E. Hzggstrom, and B.T. Khuri-Yakub,
“Calculation and Measurement of Electromechanical Coupling Coefficient of
Capacitive Micromachined Ultrasonic Transducers,” Ultrasonics, Ferroelectrics
and Frequency Control, IEEE Transactions on, vol. 50, Feb. 2003, pp. 449-456.
[26] M. Sinha and D.J. Buckley, Physical Properties of Polymers Handbook, New
York: Springer-Verlag, 2007.
[27] D. Bosc and P. Mauguen, “Acoustic properties of transparent polysiloxanes,”
Journal of Applied Polymer Science, vol. 40, Jul. 1990, pp. 135-142.
[28] C. Liu, Foundations of MEMS, Upper Saddle River, NJ: Pearson Education, Inc.,
2006.
[29] N.W. Tschoegl, W.G. Knauss, and I. Emri, “Poisson ’ s Ratio in Linear
Viscoelasticity – A Critical Review,” Mechanics of Time-Dependent Materials,
vol. 6, 2002, pp. 3-51.
[30] “Silicone Solutions for Electronic Devices and Component Assemblies,”
Momentive Perform. Mater., Albany, NY, 2007.
[31] T.L. Szabo, Diagnostic Ultrasound Imaging: Inside Out, Burlington, MA: Elsevier,
2004.
[32] D.M. Mills and L.S. Smith, “Real-time in-vivo imaging with capacitive
micromachied ultrasound transducer (CMUT) linear arrays,” Proc. IEEE
International Ultrasonics Symposium, 2003, pp. 568-571.
139
[33] D.W. Greve, J.J. Neumann, I.J. Oppenheim, S.P. Pessiki, and D. Ozevin, “Robust
capacitive MEMS ultrasonics transducers for liquid immersion,” Proc. IEEE
International Ultrasonics Symposium, Ieee, 2003, pp. 581-584.
[34] E. Jeanne, C. Meynier, F. Teston, D. Certon, N. Felix, M. Roy, and D. Alquier,
“Protection layer influence on capacitive micromachined ultrasonic transducers
performance,” Materials Research Society Symp., Warrendale, PA: 2008, pp.
1052-DD06-27.
[35] X. Zhuang, a Nikoozadeh, M. a Beasley, G.G. Yaralioglu, B.T. Khuri-Yakub, and
B.L. Pruitt, “Biocompatible coatings for CMUTs in a harsh, aqueous
environment,” Journal of Micromechanics and Microengineering, vol. 17, May.
2007, pp. 994-1001.
[36] X. Zhuang, D.S. Lin, O. Oralkan, and B.T. Khuri-Yakub, “Fabrication of Flexible
Transducer Arrays With Through-Wafer Electrical Interconnects Based on Trench
Refilling With PDMS,” Microelectromechanical System, IEEE/ASME Journal of,
vol. 17, 2008, p. 446–452.
[37] X. Cheng, J. Chen, I.-M. Shen, P.-C. Li, and M.H. Wang, “Fabrication and
Assembly of a Monolithic 3D CMUT Array for Imaging Applications,” Proc.
IEEE International Ultrasonics Symposium, Oct. 2007, pp. 515-518.
[38] B. Bayram, M. Kupnik, G.G. Yaralioglu, A.S. Ergun, D.-S. Lin, S.H. Wong, and
B.T. Khuri-yakub, “Finite element modeling and experimental characterization of
crosstalk in 1-D CMUT arrays,” Ultrasonics, Ferroelectrics and Frequency
Control, IEEE Transactions on, vol. 54, 2007, pp. 418-430.
[39] S. Berg and A. Ronnekleiv, “Reducing Fluid Coupled Crosstalk Between
Membranes in CMUT Arrays by Introducing a Lossy Top Layer,” Proc. IEEE
International Ultrasonics Symposium, Ieee, 2006, pp. 594-597.
[40] K. Niederer, P.-C. Eccardt, H. Meixner, and R. Lerch, “Micromachined transducer
design for minimized generation of surface waves,” Proc. IEEE International
Ultrasonics Symposium, 1999, pp. 1137-1139.
[41] E. Campbell, L. a J. Davis, G. Hayward, and D. Hutchins, “Cross-Coupling in
Sealed cMUT Arrays for Immersion Applications,” 2007 IEEE Ultrasonics
Symposium Proceedings, Ieee, 2007, pp. 2135-2138.
140
[42] H. Koymen, M. Senlik, A. Atalar, and S. Olcum, “Parametric linear modeling of
circular cMUT membranes in vacuum,” Ultrasonics, Ferroelectrics and
Frequency Control, IEEE Transactions on, vol. 54, 2007, pp. 1229-1239.
[43] A. Lohfink and P.-C. Eccardt, “Linear and nonlinear equivalent circuit modeling
of CMUTs.,” Ultrasonics, Ferroelectrics and Frequency Control, IEEE
Transactions on, vol. 52, Dec. 2005, pp. 2163-72.
[44] G.G. Yaralioglu, M.H. Badi, A.S. Ergun, and B.T. Khuri-Yakub, “Improved
equivalent circuit and finite element method modeling of capacitive
micromachined ultrasonic transducers,” Proc. IEEE International Ultrasonics
Symposium, Ieee, 2003, pp. 469-472.
[45] S.H. Wong, M. Kupnik, K. Butts-Pauly, and B.T. Khuri-Yakub, “Advantages of
Capacitive Micromachined Ultrasonics Transducers (CMUTs) for High Intensity
Focused Ultrasound (HIFU),” Proc. IEEE International Ultrasonics Symposium,
Ieee, 2007, pp. 1313-1316.
[46] A. Ergun, Y. Huang, X. Zhuang, O. Oralkan, G. Yarahoglu, and B.T. Khuri-Yakub,
“Capacitive micromachined ultrasonic transducers: fabrication technology,”
Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, vol. 52,
2005, p. 2242–2258.
[47] Y. Huang, A.S. Ergun, E. Hæggström, M.H. Badi, and B.T. Khuri-Yakub,
“Fabricating Capacitive Micromachined Ultrasonic Transducers with Wafer-
bonding Technology,” Microelectromechanical System, IEEE/ASME Journal of,
vol. 12, Apr. 2003, pp. 128-137.
[48] L. Sussman, “Polysiloxanes,” [Online]. Available:
http://www.eng.buffalo.edu/Courses/ce435/Polysiloxanes.
[49] D. Bodas and C. Khan-Malek, “Hydrophilization and hydrophobic recovery of
PDMS by oxygen plasma and chemical treatment—An SEM investigation,”
Sensors and Actuators B: Chemical, vol. 123, Apr. 2007, pp. 368-373.
[50] F. Niklaus, G. Stemme, J.-Q. Lu, and R.J. Gutmann, “Adhesive wafer bonding,”
Journal of Applied Physics, vol. 99, 2006, pp. 031101-031128.
[51] W. Mokwa, “Decomposition of Ethanol and Acetaldehyde on Clean ZnO Prism
and Oxygen Faces,” Surface Science, vol. 117, 1982, pp. 659-667.
141
[52] C.A. Desmond, K. Hobart, F. Kub, G. Campisi, and M. Weldon, Semiconductor
Wafer Bonding: science, Technology, and Applications IV, PV 97-36, Pennington:
The Electrochemical Society, 1997.
[53] M. Wiegand, M. Reiche, and U. Gosele, “Semiconductor Wafer Bonding: Science,
Technology, and Applications V,” C.E. Hunt, H. Baumgart, and G. U., eds.,
Pennington: The Electrochemical Society, 1999.
[54] U. Gosele, with Q. Y. Tong, Semiconductor Wafer Bonding: Science and
Technology, Electrochemical Society Press, 1999.
[55] H. Takagi, R. Maeda, N. Hosoda, T.R. Chung, and T. Suga, “Semiconductor
Wafer Bonding: Science, Technology, and Applications V,” C.E. Hunt, H.
Baumgart, and G. U., eds., Pennington: The Electrochemical Society, 1999.
[56] M. Reiche, U. Gosele, and M. Wiegand, “Modification of Si ( 100 ) -Surfaces by
SF 6 Plasma Etching – Application to Wafer Direct Bonding,” Crystal Research
and Technology, vol. 35, 2000, pp. 807-821.
[57] ANSYS 11.0, Canonsburg, PA: Ansys, Inc., 2007.
[58] J. Bergström and M. Boyce, “Constitutive Modeling of The Large Strain Time-
Dependent Behavior of Elastomers,” Journal of the Mechanics and Physics of
Solids, vol. 46, 1998, p. 931–954.
[59] S. Imaoka, “Viscoelasticity (Memo Number: STI0807A),” Ansys, Inc.,
Canonsburg, PA, 2008.
[60] Y. Hosono, Y. Yamashita, and K. Itsumi, “Effects of fine metal oxide particle
dopant on the acoustic properties of silicone rubber lens for medical array probe.,”
Ultrasonics, Ferroelectrics and Frequency Control, IEEE Transactions on, vol. 54,
Aug. 2007, pp. 1589-95.
[61] A. Ergun, G. Yaralioglu, O. Oralkan, and B. Khuri-Yakub, MEMS/NEMS
Handbook Techniques and Applications, Springer US, 2006.
[62] D. Certon, N. Felix, E. Lacaze, F. Teston, and F. Patat, “Investigation of cross-
coupling in 1-3 piezocomposite arrays,” Ultrasonics, Ferroelectrics and
Frequency Control, IEEE Transactions on, vol. 48, 2001, pp. 85-93.
142
[63] D. Certon, N. Felix, L.P.T.H. Hue, F. Patat, and M. Lethiecq, “Evaluation of laser
probe performances for measuring cross-coupling in 1-3 piezocomposite arrays,”
Proc. IEEE International Ultrasonics Symposium, 1999, pp. 1091-1094.
[64] J.F. Dias, “An experimental investigation of the cross coupling between elements
of an acoustic imaging array transducer,” Ultrason. Imag., vol. 4, 1982, p. 1137–
1139.
[65] C. Zanelli, “Design and characterization of a 10 cm annular array transducer for
high intensity focused ultrasound (HIFU) applications,” Proc. IEEE International
Ultrasonics Symposium, 1994, pp. 144-148.
[66] S. Zhou, G.L. Wojcik, and J.A. Hossack, “An approach for reducing adjacent
element crosstalk in ultrasound arrays,” Ultrasonics, Ferroelectrics and Frequency
Control, IEEE Transactions on, vol. 50, 2003, pp. 1752-1761.
[67] B. Bayram, M. Kupnik, G.G. Yaralioglu, Ö. Oralkan, D.-S. Lin, X. Zhuang, A.S.
Ergun, A.F. Sarioglu, S.H. Wong, and B.T. Khuri-yakub, “Characterization of
cross-coupling in capacitive micromachined ultrasonic transducers,” Proc. IEEE
International Ultrasonics Symposium, Ieee, 2005, pp. 601-604.
[68] S. Berg and A. Ronnekleiv, “P4M-9 Reduction of Crosstalk in CMUT Arrays by
Introducing Double Periodicities,” Proc. IEEE International Ultrasonics
Symposium, Oct. 2007, pp. 2155-2158.
[69] B. Bayram, G.G. Yaralioglu, M. Kupnik, and B.T. Khuri-Yakub, “Acoustic
crosstalk reduction method for CMUT arrays,” Proc. IEEE International
Ultrasonics Symposium, Ieee, 2006, pp. 590-593.
[70] Y. Roh and B.T. Khuri-Yakub, “Finite element analysis of underwater capacitor
micromachined ultrasonic transducers,” Ultrasonics, Ferroelectrics and Frequency
Control, IEEE Transactions on, vol. 49, Mar. 2002, pp. 293-298.
[71] S. Zhou and J.A. Hossack, “Reducing inter-element acoustic crosstalk in
capacitive micromachined ultrasound transducers,” Ultrasonics, Ferroelectrics
and Frequency Control, IEEE Transactions on, vol. 54, 2007, p. 1217–1228.
[72] P.-C. Eccardt, A. Lohfink, and H.-G. v Garssen, “Analysis of crosstalk between
fluid coupled cmut membranes,” Proc. IEEE International Ultrasonics Symposium,
Ieee, 2005, pp. 593-596.
143
[73] M. Wilm, A. Reinhardt, V. Laude, and S. Ballandras, “Three dimensional
modeling of micromachined-ultrasonic-transducer arrays operating in water,”
Ultrasonics, vol. 43, 2005, pp. 457-465.
[74] K.A. Wong, S. Panda, and I. Ladabaum, “Curved micromachined ultrasonic
transducers,” Proc. IEEE International Ultrasonics Symposium, 2003, pp. 572-576.
[75] “Ultrathin silicon, product description,” Virginia Semiconductor, Fredericksburg,
VA.
[76] N. Chen, J. Engel, S. Pandya, and C. Liu, “Flexible skin with two-axis bending
capability made using weaving-by-lithography fabrication method,” Proc. IEEE
MEMS Conference, 2006, pp. 330-333.
[77] J. Engel, J. Chen, and C. Liu, “Development of polyimide flexible tactile sensor
skin,” J. Micromechanics and Microengineering, vol. 13, 2003, pp. 359-366.
[78] K. Noda, K. Hoshino, K. Matsumoto, and I. Shimoyama, “Fabrication of the
flexible sensor using SOI wafer by removing the thick silicon layer,” Proc. IEEE
MEMS Conference, 2006, pp. 122-125.
[79] H.K. Lee, S.I. Chang, and E. Yoon, “A flexible polymer tactile sensor: fabrication
and modular expandability for large area deployment,” Microelectromechanical
System, IEEE/ASME Journal of, vol. 15, 2006, pp. 1681-1686.
[80] Y. Xu, Y.-C. Tai, A. Huang, and C.-M. Ho, “IC-integrated flexible shear-stress
sensor skin,” Microelectromechanical System, IEEE/ASME Journal of, vol. 12,
2003, pp. 740-747.
[81] A. Huang, J. Lew, Y. Xu, Y.-C. Tai, and C.-M. Ho, “Microsensors and actuators
for macrofluidic control,” IEEE Sensors J., vol. 4, 2004, pp. 494-502.
[82] J.F. Hetke, J.L. Lund, K. Najafi, K.D. Wise, and D.J. Anderson, “Silicon ribbon
cables for chronically implantable microelectrode arrays,” IEEE Trans. Biomed.
Eng., vol. 41, 1994, pp. 314-321.
[83] K. Huang and P. Peumans, “Stretchable silicon sensor networks for structural
health monitoring,” Proc. SPIE, 2006, pp. 322-331.
[84] R. Dinyari, S.-B. Rim, K. Huang, P.B. Catrysse, and P. Peumans, “Curving
monolithic silicon for nonplanar focal plane array applications,” Appl. Phys. Lett,
vol. 92, 2008, pp. 091114-3.
144
[85] K. Itsumi, Y. Hosono, N. Yamamoto, and Y.J. Yamashita, “Low acoustic
attenuation silicone rubber lens for medical ultrasonic array probe.,” Ultrasonics,
Ferroelectrics and Frequency Control, IEEE Transactions on, vol. 56, Apr. 2009,
pp. 870-4.
[86] “American Stroke Foundation,” [Online].
Available:http://www.americanstroke.org/content/view/36/56/.
[87] D.M. Mills, “Medical imaging with capacitive micromachied ultrasound
transducer (CMUT) arrays,” Proc. IEEE International Ultrasonics Symposium,
2004, pp. 384-390.
[88] R.A. Noble, R.R. Davies, M.M. Day, L. Koker, D.O. King, K.M. Brunson, A.R.D.
Jones, J.S. McIntosh, D.A. Hutchins, T.J. Robertson, and P. Saul, “Cost-effective
and manufacturable route to the fabrication of high-density 2D micromachined
ultrasonic transducer arrays and (CMOS) signal conditioning electronics on the
same silicon substrate,” Proc. IEEE International Ultrasonics Symposium, 2001,
pp. 941-944.
[89] R.A. Noble, R.R. Davies, D.O. King, M.M. Day, A.R.D. Jones, J.S. McIntosh,
D.A. Hutchins, and P. Saul, “Low-temperature micromachined cMUTs with fully-
integrated analogue front-end electronics,” Proc. IEEE International Ultrasonics
Symposium, 2002, pp. 1045-1050.
[90] C. Daft, S. Calmes, D.D. Graca, K. Patel, P. Wagner, and I. Ladabaum,
“Microfabricated ultrasonic transducers monolithically integrated with high
voltage electronics,” Proc. IEEE International Ultrasonics Symposium, 2004, pp.
493-496.
[91] B. Selikson and T.A. Longo, “A study of purple plague and its role in integrated
circuits,” Proceedings of the IEEE, 1964, pp. 1638-1641.
[92] P. Garrou, “3D IC integration: an emerging system level integration architecture,”
3D Inegration and Packaging Roadshow, San Jose: 2008.
[93] E.M. Chow, V. Chandrasekaran, A. Partridge, T. Nishida, M. Sheplak, C.F. Quate,
and T.W. Kenny, “Process compatible polysilicon-based electrical through-wafer
interconnects in silicon substrates,” Microelectromechanical System, IEEE/ASME
Journal of, vol. 11, 2002, pp. 631-640.
145
[94] C.H. Cheng, A.S. Ergun, and B.T. Khuri-Yakub, “Electrical through-wafer
interconnects with sub-picofarad parasitic capacitance,” Proc.
Microelectromechanical Systems Conference, 2001, pp. 18-21.
[95] C.H. Cheng, E.M. Chow, X. Jin, A.S. Ergun, and B.T. Khuri-Yakub, “An efficient
electrical addressing method using through-wafer vias for two-dimensional
ultrasonic arrays,” Proc. IEEE International Ultrasonics Symposium, 2000, pp.
1179-1182.
[96] D.F. Lemmerhirt and K.D. Wise, “Air-isolated through-wafer interconnects for
microsystem applications,” Proc. Transducers Conference, 2003, pp. 1067-1070.
[97] R.D. Mcbride, S.G. Rosser, and R.P. Nowak, “Modeling and Simulation of 12.5
Gb/s on a HyperBGA® Package,” Electronics Manufacturing Technology
Symposium, 2003, pp. 1-5.
[98] D.P. Lukanin, V.V. Kalaev, Y.N. Makarov, T. Wetzel, J. Virbulis, and W. von
Ammon, “Advances in the simulation of heat transfer and prediction of the melt-
crystal interface shape in silicon CZ growth,” Journal of Crystal Growth, vol. 266,
May. 2004, pp. 20-27.
[99] X. Zhuang, “Capacitive Micromachied Ultrasonic Transducers with Through-
Wafer Interconnects,” Stanford University, 2008.
[100] “Modeling of Crystal Growth from the Melt,” Soft-Impact, Ltd., [Online].
Available: http://softimpact.ru/melt.php.
[101] J.-W. Nah, S.L. Buchwalter, P.A. Gruber, D.Y. Shih, and B.K. Furman, “C4NP
Cu-cored Pb-free Flip Chip Interconnections,” 2008 Electronic Components and
Technology Conference, 2008, pp. 1351-1356.
[102] Y.T. Cheng, L. Lin, and K. Najafi, “Localized Silicon Fusion and Eutectic
Bonding for MEMS Fabrication and Packaging,” Microelectromechanical System,
IEEE/ASME Journal of, vol. 9, Mar. 2000, pp. 3-8.
[103] G.G. Yaralioglu, I.O. Wygant, T.C. Marentis, and B.T. Khuri-Yakub, “Ultrasonic
mixing in microfluidic channels using integrated transducers.,” Analytical
chemistry, vol. 76, Jul. 2004, pp. 3694-8.
[104] K.K. Park, H.J. Lee, G.G. Yaralioglu, a S. Ergun, O. Oralkan, M. Kupnik, C.F.
Quate, B.T. Khuri-Yakub, T. Braun, J.-P. Ramseyer, H.P. Lang, M. Hegner, C.
146
Gerber, and J.K. Gimzewski, “Capacitive micromachined ultrasonic transducers
for chemical detection in nitrogen,” Applied Physics Letters, vol. 91, 2007, p.
094102.