close

Вход

Забыли?

вход по аккаунту

?

High-Speed Hybrid Silicon Mach-Zehnder Modulator and Tunable Microwave Filter

код для вставкиСкачать
UNIVERSITY OF CALIFORNIA
Santa Barbara
High-Speed Hybrid Silicon Mach-Zehnder Modulator
and Tunable Microwave Filter
A Dissertation submitted in partial satisfaction of
the requirements for the degree
Doctor of Philosophy
in
Electrical and Computer Engineering
by
Hui-Wen Chen
Committee in charge:
Professor John E. Bowers, Chair
Professor Larry A. Coldren
Professor Nadir Dagli
Professor Evelyn Hu
March 2011
UMI Number: 3456108
All rights reserved
INFORMATION TO ALL USERS
The quality of this reproduction is dependent on the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
UMI 3456108
Copyright 2011 by ProQuest LLC.
All rights reserved. This edition of the work is protected against
unauthorized copying under Title 17, United States Code.
ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
The dissertation of Hui-Wen Chen is approved.
Professor Evelyn Hu
Professor Nadir Dagli
Professor Larry A. Coldren
Professor John E. Bowers, Chair
January 2011
High-Speed Hybrid Silicon Mach-Zehnder Modulator
and Tunable Microwave Filter
Hui-Wen Chen
c January 2011
Copyright typeset in LATEX
iii
Acknowledgments
For these many years at UCSB, I have always been thankful to have this chance
to study here with all the excellent faculties and intelligent colleagues. This experience does change my view and definitely will be my most valuable memory in
the future.
First I would like to thank my advisor, John Bowers. During these past years,
he has provided excellent research environment that not only helped me as a
researcher but also changed my vision as a person. The philosophy of working
hard, playing hard really changes my view and I believe this will be one of the
most valuable experiences in my life. I would like to acknowledge my committee
members Prof. Larry Coldren, Prof. Nadir Dagli, and Prof. Evelyn Hu for their
wealth of knowledge in photonic devices and semiconductor fabrications. Without
the guidance they provided in the past few years, this thesis will not be possible.
Next I would like to thank all the senior members in Bowers group for their
generous support and help. Ying-hao Kuo deserves my special thanks. During
that one and half years, he taught me all the knowledge about fabrication techniques, design thoughts, and testing ability. The time (or late time) we spent in
the cleanroom was definitely the most crazy hours in my life. I would also like
to thank Alex Fang for his support not only as a colleague, but also as a friend.
From him I learn how to deal with events fairly in larger aspects, and of course
to improve my English vocabulary by all the “conversation” they carry on in the
office. Di Liang always provided inspirited ideas when I have trouble in either in
process and research. Hyundai help me to have a clear mind when it goes to the
fundamental physics. I won’t forget all the good time we spent in each conference
iv
to get drunk.
I would like to give special thanks to Jon Peters for his help in fabrication. I
cannot image how miserable it would be without his efforts. Zhi Wang’s contribution to PhASER project also make this thesis possible.
Of course, life will be much more boring without all the Bowers group. Brian,
Emily, Hsu-Feng and Gan has been really supportive during my early years.
Anand is always my best buddy in the office and thank him for being the target in
the office for so long. Although is is always really difficult to find a restaurant for
Geza, Sid, and Molly, thank them for bearing with me making weird noise in the
office all the time. I would like to thank the rest of the group members during my
time here at UCSB: Daoxin, Yongbo, Martijn, Jason, Andy, Shane, JeHyeong,
Jock, Jared, Mike, Tony, Sudha, John Magnani, Ashok, Gehong, Ben, Daryl,
Chong, Paolo, and Stefano. As for Bowers group assistant Alyssa, Christine, and
Christina, thank you for taking care with all the POs and reimbursements. This
really makes my graduate student life much easier.
The support from other group members is always very grateful to do research
here. I would never be able to complete my measurement without John Mach and
John Garcia’s knowledge about high-speed testing and it was fun to do testing
with them. I would like to thank Jae and Selim for their experience and suggests
for the electrode design. I am also grateful to know Erik and Rob for their sharing
knowledge of RF testing in the PhASER project. I would like to acknowledge rest
of the colleagues at UCSB: Henrik, Kim, Erica, Demis, Byungchae, Matt Dummer,
Sasa, Yan, Ashish, and John Parker.
For those who ever work in the UCSB nanofab, it is definitely one of the best
v
environments to work with. I guess we are all spoiled by the excellent support and
knowledge from all staffs: Jack Whaley, Brian Tibeault, Ning Cao, Bill Mitchell,
Tony Bosh, Don Freeborn, Adam Abrahamsen, and Brian Lingg.
Life at UCSB will be boring without all the friends hanging out each Friday
night. Jason, Chinhan, Vincent, Tsungli, Jeff, Yvonne, Janet, Sherman, Lily,
Pinta, and the volleyball team.
My family, Dad, Mom, Hui-Jen, and Yen-Chu. They are always there no
matter how far I go. My aunt deserves my deeply thanks for taking care of my
whole family.
Finally, I would like to thank Kuan-Chung. It has been a tough five years, but
we got through it. Without his support, understanding and confidence in me, it
will never be possible for me to finish my journey.
Back to five years ago, I came to this continental alone. Five years later, I
have all the cherished memory shared all the friends and colleagues at UCSB. I
wish you all good and hope we will have chance to meet again one day.
vi
This dissertation is dedicated to my mother.
vii
Vita of Hui-Wen Chen
EDUCATION
Jan. 2011
Ph.D., Electrical and Computer Engineering
University of California, Santa Barbara, California, USA
Jun. 2005
M.S., Communication Engineering
National Chiao Tung University, Hsinchu, Taiwan
Jun. 2003
B.S., Electrical Engineering
National Tsing Hua University, Hsinchu, Taiwan
AWARDS
2009
Best Student Paper Award, Microwave Photonics Conference
2005
Taiwan Merit Scholarship (Four-year Ph.D. scholarship)
PROFESSIONAL EMPLOYMENT
Jun. - Sep. 2010
Consultant
Aurrion, LLC, Santa Barbara, CA
Jun. - Aug. 2008
Project mentor
National Nanotechnology Infrastructure Network (NNIN)
Research Experience for Undergraduates program
PUBLICATIONS
Journal Publications
1. “Device and Integration Technology for Silicon Photonic Transmitters,” H. Park, M. N. Sysak, H.-W. Chen, A. W. Fang, D. Liang, L. Liao, B.
R. Koch, J. Bovington, Y. Tang, K. Wong, M. Jacob-Mitos, R. Jones, and J. E.
Bowers, IEEE JSTQE, invited, to be published.
viii
2. “Hybrid Silicon Photonics for Optical Interconnects,” M. J. R. Heck,
H.-W. Chen, A. W. Fang, B. R. Koch, D. Liang, H. Park, M. N. Sysak, and J. E.
Bowers, IEEE JSTQE Special issue: Green Photonics, to be published
3. “40 Gb/s hybrid silicon Mach-Zehnder modulator with low chirp,”
H.-W. Chen, J. D. Peters, and J. E. Bowers, Optics Express, Vol. 19 No. 2,
2011.
4. “Integrated Microwave Photonic Filter on a Hybrid Silicon Platform,”
H.-W. Chen, A. W. Fang, J. D. Peters, J. Bovington, D. Liang, and J. E. Bowers,
IEEE TMTT/JLT Special issue: Microwave Photonics, Nov. 2010
5. “25Gb/s hybrid silicon switch using a capacitively loaded traveling
wave electrode,” H.-W. Chen, Y.-H. Kuo, and J. E. Bowers, Optics Express,
Vol. 18, No. 2, 2010
6. “Resonant normal-incidence separate-absorptioncharge-multiplication
Ge/Si avalanche photodiodes,” D. Dai, H.-W. Chen, J. E. Bowers, Y. Kang,
M. Morse, and M. J. Paniccia, Optics Express, Vol. 17, No. 19, 2009
7. “Frequency response and bandwidth enhancement in Ge/Si avalanche
photodiodes with over 840GHz gain-bandwidth-product,” W. Sfar Zaoui,
H.-W. Chen, J. E. Bowers, Y. Kang, M. Morse, M. J. Paniccia, A. Pauchard, J.
C. Campbell, Optics Express, Vol. 17, No.15, 2009
8. “Dynamic Characterization of Distortion in Hybrid Silicon Evanescent
Phase Modulators,” N. Nunoya, A. Ramaswamy, H.-W. Chen, M. N. Sysak
and J. E. Bowers, IEEE Photonics Technology Letters, Vol. 21 no. 13, pp. 83335, 2009
9. “Hybrid silicon modulators,” H.-W. Chen, Y.-H. Kuo, J. E. Bowers, Chinese
Optics Letters, Vol. 7, No. 4, Invited, 2009
10. “Hybrid silicon evanescent approach to optical interconnects,” D. Liang,
A. W. Fang, H.-W. Chen, M. N. Sysak, B. R. Koch, E. Lively, O. Raday, Y.-H.
ix
Kuo, R. Jones, J. E. Bowers, Applied Physics A: Materials Science and Processing,
invited, 2009
11. “Monolithic germanium/silicon avalanche photodiodes with 340 GHz
gain-bandwidth product,” Y. Kang, H.-D. Liu, M. Morse, M. J. Paniccia,
M.Zadka, S. Litski, G. Sarid, A. Pauchard, Y.-H. Kuo, H.-W. Chen, W. Sfar
Zaoui, J. E. Bowers, A. Beling, D. C. McIntosh, X. Zheng and J. C. Campbell,
Nature Photonics, DOI:10.1038, 2009
12. “High speed hybrid silicon evanescent Mach-Zehnder modulator and
switch,” H.-W. Chen, Y.-H. Kuo, and J. E. Bowers, Optics Express, 16(25),
20571-20576, 2008
13. “A Hybrid Silicon-AlGaInAs Phase Modulator,” H.-W. Chen, Y.-H. Kuo,
and J. E. Bowers, Photonics Technology Letter, 20(23), 1920-1922, 2008
14. “High speed hybrid silicon evanescent electroabsorption modulator,”
Y.-H. Kuo, H.-W. Chen, J.E. Bowers, Optics Express, 16(13), 9936-9941, 2008
15. “Photonic Integration on Hybrid Silicon Evanescent Device Platform,”
H. Park, A. W. Fang, D. Liang, Y.-H. Kuo, H.-H. Chang, B. R. Koch, H.-W. Chen,
M. N. Sysak, R. Jones, and J. E. Bowers, Advances in Optical Technologies,
Article ID 682978, 2008
Conference Publications
1. “25 GHz Hybrid Silicon Mach-Zehnder Modulator using High-Speed
Push-Pull Slotline Design,” H.-W. Chen, J. D. Peters, and J. E. Bowers, to
be published in OFC 11
2. “Over 40 GHz Traveling-Wave Electroabsorption Modulator Based on
Hybrid Silicon Platform,” Y. Tang, H.-W. Chen, J. D. Peters, U. Westergren,
and J. E. Bowers, to be published in OFC 11
x
3. “Hybrid integration of III-V and Si for Photonic Integrated Circuits,”
G. Kurczveil, S. Jain, D. Liang, H.-W. Chen, M. J. R. Heck, J. E. Bowers, Frontiers in Optics 2010, Rochester, NY, October 26, (2010)
4. “Low-Power, Fast Hybrid Silicon Switches for High-Capacity Optical Networks,” H.-W. Chen, J. E. Bowers, Photonics in Switching 10,Monterey,PMC3, 2010
5. “Integrated Triplexer on Hybrid Silicon Platform,” H.-H. Chang, Y.H. Kuo, H.-W. Chen, R. Jones, A. Barkai, M. J. Paniccia, and J. E. Bowers1,
OFC/NFOEC, San Diego, OThC4, 2010
6. “Hybrid Silicon Tunable Filter Based on a Mach- Zehnder Interferometer and Ring Resonantor,” H.-W. Chen, A. W. Fang, J. Bovington, J. D.
Peters, J. E. Bowers, Microwave Photonics 09, Valencia, Spain, We 2.3, 2009,
Best student paper award
7. “Monolithic Ge/Si Avalanche Photodiodes,” Y. Kang, M. Morse, M. J.
Paniccia, M. Zadka, Y. Saad, G. Sarid, A. Pauchard, W. Sfar Zaoui, H.-W. Chen,
D. Dai, J. E. Bowers, H.-D. Liu, D. C. Mcintosh, X. Zheng, J. C. Campbell,
Group IV Photonics, San Francisco, WB6, 2009
8. “Equivalent circuit model of a Ge/Si avalanche photodiode,” D. Dai,
H.-W. Chen, J. E. Bowers, Y. Kang, M. Morse, and M. J. Paniccia, Group IV
Photonics, San Francisco, WB2, 2009
9. “A High Speed Mach-Zehnder Silicon Evanescent Modulator Using
Capacitively Loaded Traveling Wave Electrode,” H.-W. Chen, Y.-H. Kuo,
J. E. Bowers, Group IV Photonics, San Francisco, FC4, 2009
10. “Equivalent circuit model of a waveguide-type Ge/Si avalanche photodetector operating at 1550nm,” D. Dai, H.-W. Chen, J. E. Bowers, Y.
Kang, M. Morse, and M. J. Paniccia, The 36th International Symposium on
Compound Semiconductors (ISCS-2009), 2009
xi
11. “High Speed Silicon Modulators,” H.-W. Chen, Y.-H. Kuo, J. E. Bowers,
OECC 2009 Hong Kong,ThG1, Invited, 2009
12. “Dynamic Distortion Characteristics of Silicon Evanescent Detectors
and Phase Modulators,”N. Nunoya, A. Ramaswamy, H.-W. Chen, H. Park,
J. E. Bowers, Optical Fiber Communication (OFC),paper no. OMR8, San Diego,
CA, 2009
13. “Origin of the Gain-Bandwidth-Product Enhancement in SeparateAbsorption-Charge-Multiplication Ge/Si Avalanche Photodiodes,” W.
Sfar Zaoui, H.-W. Chen, J. E. Bowers,Y. Kang, M. Morse, M. J. Paniccia, A.
Pauchard, J. Campbell, OFC 2009, San Diego, 2009
14. “Hybrid Silicon Modulators,” H.-W. Chen, Y.-H. Kuo, J. E. Bowers, Photonics West 2009, Proc. of SPIE Vol. 7220, Invited, 2009
15. “Hybrid Silicon Evanescent Modulators,” H.-W. Chen, Y.-H. Kuo, and J.
E. Bowers, LEOS Annual,TuDD1, Invited, Newport Beach, CA, 2008
16. “Wafer Bonded Silicon Photonics,” D. Liang, A. W. Fang, H.-W. Chen, J.
E. Bowers, IEEE Avionics, Fiber- Optics and Photonics Technology, Paper WA1
(invited), San Diego, CA, USA, 2008
17. “A 10Gb/s Mach-Zehnder Silicon Evanescent Modulator,” H.-W. Chen,
Y.-H. Kuo, and J. E. Bowers, IEEE GFP, WA3, Sorrento, Italy, 2008
18. “Hybrid silicon photonic integrated circuits for optical networking,” E.
F. Burmeister, H.-W. Chen, J. P. Mack, Y.-H.Kuo, J. E. Bowers, Photonics in
Switching 2008, Japan, 2008
19. “Hybrid silicon evanescent phase modulator based on carrier depletion
in offset multiple-quantum-well,” H.-W. Chen, Y.-H. Kuo, J. E. Bowers,
CLEO, San Jose, 2008
20. “A hybrid silicon evanescent electroabsorption modulator,” Y.-H. Kuo,
H.-W. Chen, J. E. Bowers, OFC 2008, San Diego, CA, 2008
xii
Abstract
High-Speed Hybrid Silicon Mach-Zehnder Modulator
and Tunable Microwave Filter
by
Hui-Wen Chen
The routing of data electronically within microprocessors is becoming increasingly
challenging due to the large volumes of data being transferred. Optical interconnects on
silicon are an attractive alternative to traditional electronic interconnects because they
provide higher bandwidth and have the potential to be compatible with low cost, high
volume, mature CMOS processing. Recent efforts in silicon photonics have focused on
developing a wide range of optical components, such as lasers, amplifiers, photodetectors, that can be integrated on a single platform. Among all, a lot of work has focused
on modulators as they are crucial for the generation and transmission of high-speed
signals such that increasing demands on data capacity can be satisfied.
In this work, a hybrid silicon modulator based on a Mach-Zehnder interferometer
architecture is developed to efficiently send large amount of information with large optical bandwidth, high-speed operation and good modulation efficiency. The tradeoff
between modulation efficiency and speed in pure silicon can be overcome by utilizing
the hybrid silicon platform, where the carrier depletion effect inside the III-V materials is introduced to create index shift. The demonstrated device has a voltage-length
product of 2.5 Vmm and 20 dB extinction ratio. Moreover, the modulation bandwidth
xiii
can be promoted to a higher regime by employing a capacitively loaded traveling-wave
electrode design. With the developed techniques, we successfully demonstrate a highspeed modulator with 25 GHz bandwidth and 10 dB extinction ratio at 40 Gb/s. In
addition, a 2x2 hybrid silicon switch, which is a candidate for large-scale optical network, is also reported based on the developed architecture with 0.5 dB power penalty at
40 Gb/s. The application can be further extended to an integrated level by establishing a microwave tunable filter using thermal modulators and amplifiers on the hybrid
silicon platform. The details of the design, fabrication, and characterizations of the
aforementioned devices will be presented.
xiv
Contents
Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
viii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xiii
1 Introduction
1
1.1
Silicon Modulators and the Hybrid Silicon Platform (HSP) . . . . . . .
2
1.2
Traveling-Wave Electrode . . . . . . . . . . . . . . . . . . . . . . . . . .
5
1.3
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8
1.4
Outline of This Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8
2 Optical Waveguide Design and Proof of Concept
2.1
2.2
12
Optical Waveguide Design . . . . . . . . . . . . . . . . . . . . . . . . . .
12
2.1.1
Carrier Depletion Effect . . . . . . . . . . . . . . . . . . . . . . .
12
2.1.2
Epitaxial Layer Design and Bandgap Engineering . . . . . . . . .
17
2.1.3
Passive Waveguide Design . . . . . . . . . . . . . . . . . . . . . .
25
Proof of Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
31
2.2.1
Modulation Efficiency and Proof of Concept . . . . . . . . . . . .
31
2.2.2
Optical Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . .
34
2.2.3
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
35
xv
3 Coplanar Waveguide TWE Mach-Zehnder Modulator
3.1
37
Traveling-Wave Electrode Design . . . . . . . . . . . . . . . . . . . . . .
38
3.1.1
General Transmission Line and Frequency Response . . . . . . .
38
3.1.2
Equivalent Circuit Model . . . . . . . . . . . . . . . . . . . . . .
44
3.1.3
Electrical Characteristics and Device Parameters . . . . . . . . .
48
3.2
Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55
3.3
Device Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
3.3.1
Static Characteristics . . . . . . . . . . . . . . . . . . . . . . . .
60
3.3.2
High Frequency Performance . . . . . . . . . . . . . . . . . . . .
61
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
64
3.4
4 Capacitively Loaded TWE MZM
4.1
67
Capacitively Loaded TWE Design . . . . . . . . . . . . . . . . . . . . .
67
4.1.1
Push-Pull Structure: Slotline v.s. CPW . . . . . . . . . . . . . .
67
4.1.2
Capacitively Loaded TWE Circuit Model . . . . . . . . . . . . .
70
4.2
Fabrication Improvements . . . . . . . . . . . . . . . . . . . . . . . . . .
76
4.3
Device Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . .
80
4.3.1
Static Characteristics . . . . . . . . . . . . . . . . . . . . . . . .
80
4.3.2
Electrical Properties of the CL Slotline
. . . . . . . . . . . . . .
82
4.3.3
Large Signal Modulation . . . . . . . . . . . . . . . . . . . . . . .
86
4.3.4
Chirp Properties . . . . . . . . . . . . . . . . . . . . . . . . . . .
89
4.4
Switch Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . .
90
4.5
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
5 Application: Tunable Microwave Filter
5.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xvi
99
99
5.2
Device Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.3
Experiment Setup and Results . . . . . . . . . . . . . . . . . . . . . . . 104
5.4
5.3.1
Filter Response in Optical Domain . . . . . . . . . . . . . . . . . 104
5.3.2
Filter Response in Frequency Domain . . . . . . . . . . . . . . . 111
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6 Conclusion and Future Work
117
6.1
Summary of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.2
Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.2.1
Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.2.2
Large-scale optical interconnects . . . . . . . . . . . . . . . . . . 120
6.2.3
Programmable microwave filter . . . . . . . . . . . . . . . . . . . 120
Appendices
122
A SU8 Dry-Etch Development
123
A.1 Lithographically Defined SU8 . . . . . . . . . . . . . . . . . . . . . . . . 123
A.2 SU8 Dry-Etch Development . . . . . . . . . . . . . . . . . . . . . . . . . 125
xvii
List of Figures
1.1
Circuit model for (a)lumped (b)traveling-wave modulator. . . . . . . . .
2.1
(a)A p-i-n diode with lightly doped intrinsic layer where the carriers stay.
6
(b)With external bias, the carriers are swept to the n layer. . . . . . . .
13
2.2
Energy dependence of the proportionality constant for band-filling. . . .
15
2.3
The total index change introduced by the carrier depletion effect in a
p-i-n diode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4
16
Bandgap heaven of III-V materials. The bandgap are drawn in black
while the offsets are labeled in blue with the unit in meV. . . . . . . . .
17
2.5
Illustration of the epitaxial layers of a hybrid silicon modulator. . . . . .
18
2.6
Material composition for different values of strain and bandgap in the
Inx Gay Al1−x−y As. The material parameters can be found in [6, 7]. . . .
2.7
19
Density of states of the QW listed in Table 2.1. The shadow area represents the total carrier density and the dash line is ∆Ec . . . . . . . . . .
xviii
22
2.8
(a)The built-in depletion region width for a SCH layer next to a p doped
cladding layer of 1 × 1018 cm−3 . (b)Estimated index change as a function of NSCH at different bias conditions. (c)3 dB cutoff frequency as a
function of NSCH with series resistance of 5 Ω. (d)Overall performance
at different value of NSCH . . . . . . . . . . . . . . . . . . . . . . . . . .
2.9
23
(a)Simulated band diagram using parameters listed in Table 2.2. Both
heavy and light (red curve) hole valence bands are plotted. (b)The electrical field at different biases. (c)The conduction band energy at different
biases. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
2.10 The cross section of a hybrid silicon modulator. . . . . . . . . . . . . . .
27
2.11 (a)The confinement factor in the silicon waveguide and the active region
as functions of H and W. (b)Simulated mode profile in the hybrid section
with H=0.5 µm and W=1 µm. . . . . . . . . . . . . . . . . . . . . . . .
28
2.12 (a)Schematic illustration of the III-V/Si taper. (b)The coupling loss as
function of taper length. (c)The coupling loss as a function of different
alignment shifts in x direction. . . . . . . . . . . . . . . . . . . . . . . .
29
2.13 (a)Schematic illustration of two 2x2 MMI and a phase section. (b)The
transmission without any phase in the MZI. (c)The transmission with π
phase shift in the MZI. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
30
2.14 (a)A Fabry-Perot structure used to measure the index shift. (b)Illustrated
spectra with different levels of reverse bias. . . . . . . . . . . . . . . . .
32
2.15 Experimental results of: (a)The total index change with correct bonding alignment. (b)The index change with incorrect bonding alignment.
(c)The index change introduced by the Pockels effect. (d)The index
change due to the rest of the physical effects. . . . . . . . . . . . . . . .
xix
33
2.16 The measured ER and Vπ L of a 500 µm long MZM. . . . . . . . . . . .
34
3.1
General transmission line circuit with source and load. . . . . . . . . . .
38
3.2
(a)Frequency responses for different values of ZL when ZD = 35Ω. (b)Average
voltage as a function of frequency for different values of ZL . . . . . . . .
3.3
(a)Frequency response for different values of ne when no =3.45. (b)Frequency
response for different α. . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4
42
43
(a)Cross section of a hybrid silicon CPW modulator and its dimensions.
(b)Equivalent circuit model for a TWE device. . . . . . . . . . . . . . .
45
3.5
(a)Impedance of Zu . (b)Impedance of Yu . . . . . . . . . . . . . . . . . .
48
3.6
Attenuation constant due to different device components . . . . . . . . .
49
3.7
Transmission line characteristics as a function of p contact resistance. .
52
3.8
Transmission line characteristics as a function of n contact resistance. .
52
3.9
Transmission line characteristics as a function of intrinsic region width.
53
3.10 Transmission line characteristics as a function of mesa width. . . . . . .
53
3.11 Transmission line characteristics as a function of probe metal thickness.
54
3.12 Transmission line characteristics as a function of doping concentration of
cladding layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
54
3.13 The transition between passive and hybrid sections. . . . . . . . . . . .
56
3.14 Process flow for a hybrid silicon modulator. . . . . . . . . . . . . . . . .
57
3.15 (a)Top view of a device with a CPW electrode. (b)The optical image of
the device under microscope. (c)Cross section (along A-A’) of the hybrid
waveguide. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
3.16 Modulation efficiency of a hybrid silicon modulator with different input
power at 1540nm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xx
60
3.17 Transmission line characteristics of a 500 µm hybrid silicon modulator
with CPW electrode. The black lines are the measured data and the
green lines are calculated based on 2 µm wide QW. . . . . . . . . . . . .
62
3.18 EO frequency response and simulation with different termination. . . . .
63
3.19 EO frequency response and simulation with different termination. . . . .
64
4.1
Bias condition for a push-pull structure. . . . . . . . . . . . . . . . . . .
68
4.2
Circuit diagram for (a)CPW electrode with common ground. (b)CPW
electrode without common ground. (c)Slotline electrode . . . . . . . . .
4.3
69
(a)Top view of a device with a CL slotline electrode (b)Cross section of
loaded (along A-A’) and unloaded sections (along B-B’) of the hybrid
waveguide. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4
70
(a)Cross section of a hybrid silicon CL TWE modulator and its dimensions. (b)Distributed transmission line circuit. (c)Distributed equivalent
circuit model per unit length. . . . . . . . . . . . . . . . . . . . . . . . .
72
4.5
Top view of a CL slotline. . . . . . . . . . . . . . . . . . . . . . . . . . .
75
4.6
(a)Device Impedance as a function of fill factor at different Lm . (b)Frequency
response at 20 GHz as a function of fill factor at different Lm . . . . . . .
76
4.7
Modified process flow for CL slotline modulator. . . . . . . . . . . . . .
77
4.8
(a)A cross section of the SU8 via with ICP etch. (b)The cross section of
the device with plating metal. . . . . . . . . . . . . . . . . . . . . . . . .
4.9
79
(a)Top schematic of a 500 µm MZM with slotline design. (b)Optical
image of a 500 µm MZM with fill factor of 0.8. . . . . . . . . . . . . . .
79
4.10 (a)Schematic circuit of the device with DC bias and RF driven signal.
(b)Normalized transmission as a function of reverse bias at 1550 nm. . .
xxi
80
4.11 (a)The introduced index change (b)The slope of index change (c)The 3
dB modulation bandwidth (d)The device loss of a 500 µm long MZM as
a function of reverse bias
. . . . . . . . . . . . . . . . . . . . . . . . . .
81
4.12 Experimental electrical property of a 500 µm hybrid silicon MZM with
CL slotline at different fill factor. . . . . . . . . . . . . . . . . . . . . . .
83
4.13 Simulated electrical property of a 500 µm hybrid silicon MZM with CL
slotline at different fill factor . . . . . . . . . . . . . . . . . . . . . . . .
84
4.14 Frequency response with different fill factor . . . . . . . . . . . . . . . .
85
4.15 Modulation bandwidth measured at -3 V for a MZM with F =0.9 at 1550
nm.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
86
4.16 (a)25 Gb/s and 40 Gb/s electrical drive signal with 231 -1 NRZ PRBS.
(b)Modulated signal with different fill factor. . . . . . . . . . . . . . . .
88
4.17 (a)Measured resonant frequency where u is uth resonant position. (b)Measured
chirp spectrum with 36 km SMF. Inset: Chirp parameters at different bias
condition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
90
4.18 Energy per bit in network devices. (modified from Tucker, LEOS Annual,
2008). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
91
4.19 (a)Normalized transmission of a hybrid silicon switch for all port configurations. (b)Measured extinction ratio, crosstalk, and rise time for each
port at 1550 nm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
92
4.20 (a)Rise and fall time of the switch for all porsts. (b)BER versus optical
received power for all ports configurations at 40 Gb/s with 231 -1 NRZ
PRBS.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xxii
93
4.21 Power consumption as a function of number of ports with different switch
architectures. Inset (a): Illustration of an 8x8 WRC [11]. Inset (b): a
typical 8x8 Benes network (photo courtesy Martijn Heck). . . . . . . . .
5.1
94
(a)The schematic of a hybrid silicon tunable filter. (b)An SEM picture
of the fabricated device . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.2
(a)Schematic figure of three different configurations realizable with the
hybrid silicon filter. (b)Simulated response of the filter operated as a:
ring, MZI, and unit cell. . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.3
The experimental setup to measure the filter response. An oscilloscope
is used instead of an OSA to resolve the spectrum. . . . . . . . . . . . . 105
5.4
(a)Measured ring response of a 5 mm long delay loop at different current
levels. (b)Experimental MZI response. . . . . . . . . . . . . . . . . . . . 106
5.5
Experimental cell responses with SOAr at 140 mA. The injected current
of the SOAf is adjusted from 20mA (a) to 40mA (c) to show the difference
on cell responses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.6
(a)The wavelength shift of ring and MZI responses at different current
level of MODr , where the ring FSR is 0.164 nm. (b)The wavelength shift
of MZI response at different current level of MODf . . . . . . . . . . . . 109
5.7
(a)Cell response without any thermal modulation. (b)Cell response with
Ir =0 mA and If =13 mA. (c)Cell response with Ir =0 mA and If =19.5
mA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.8
Schematic of the setup to measure the filter function in the frequency
domain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
xxiii
5.9
(a)Measured filter responses at various wavelength in the frequency domain. The SOAr is biased at 140 mA and SOAf is biased at 32 mA.
(b)Simulated filter responses corresponding to the experiment results. . 112
A.1 SEM pictures for SU8 (a)After development and 10 minutes O2 descum.
(b)Step(a) and 30 minutes 150 ◦ C bake. (c)Step(a) and 30 minutes 220
◦C
bake. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
A.2 SU8 profile at different bias power. The ICP power is 350 W, pressure is
1 Pa with O2 /CF4 =40/4 sccm. . . . . . . . . . . . . . . . . . . . . . . . 125
A.3 SU8 profile at different etch pressure. The ICP power is 350 W, bias
power is 50 W with O2 /CF4 =40/4 sccm. . . . . . . . . . . . . . . . . . . 126
A.4 SU8 profile using hard mask. . . . . . . . . . . . . . . . . . . . . . . . . 126
xxiv
List of Tables
1.1
Comparison of silicon modulators. . . . . . . . . . . . . . . . . . . . . .
4
2.1
Compositions of SCH, QWs and barriers . . . . . . . . . . . . . . . . . .
22
2.2
Epitaxial layer structure of the hybrid silicon modulator . . . . . . . . .
25
2.3
Dimensions and Indexes for each layer in Figure 2.10 . . . . . . . . . . .
27
2.4
Dimensions and Indexes in Figure 2.14 . . . . . . . . . . . . . . . . . . .
31
3.1
Parameters of Figure 3.1. . . . . . . . . . . . . . . . . . . . . . . . . . .
38
3.2
Variables of Figure 3.4(a). Material parameters and dimensions can be
found in 2.2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
3.3
Parameters used in Figure 3.7 to Figure 3.12. . . . . . . . . . . . . . . .
50
4.1
Variables for simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
5.1
Definition of terms in Equation 5.1. . . . . . . . . . . . . . . . . . . . . . 102
xxv
Chapter 1
Introduction
Communication networks have grown rapidly over the past decade due to continually
increasing capacity demands on computation, multimedia and data services. Early
on in the development of high-speed network infrastructure, key enabling components,
such as modulators and switches, were purely electronic and as such operated at higher
data rates at the cost of consuming increasing power. Furthermore, higher capacity
was achieved by miniaturizing the underlying silicon technology. Today, a state of the
art router has a capacity of 92 Tb/s with more than 0.2 million CPUs. This router,
however, dissipates around 1 MW regardless of the cooling system used. The tremendous
power consumption of electronic routers has become a major concern, and limits their
performance as capacity increases. Hence, it is absolutely vital to develop an alternative
to replace existing technologies.
Optical interconnects on silicon are an attractive alternative to traditional electronic interconnects because they provide higher bandwidth and have the potential to
be compatible with low cost, high volume, mature CMOS processing. By adopting the
1
Chapter 1. Introduction
modulation formats such as wavelength division multiplexing (WDM) or differential
quadrature phase-shift keying (DQPSK) [1], the bit rate could be several times larger
than the symbol rate. A lot of work has focused on modulators as they are crucial for
the generation and transmission of signals with high-speed and wide optical bandwidth.
In this chapter, we will start to address the importance of the development of modulators with high speed and large optical bandwidth. The applications of this technology
will also be introduced to show the ability for the demonstration of more complicated
interconnects.
1.1
Silicon Modulators and the Hybrid Silicon
Platform (HSP)
In this section, we will discuss different approaches to implement silicon modulators and
analyze the advantages and disadvantages individually.
Microring modulator (RM)
Micro-ring modulators have been widely explored due to their ultra compact footprint
and low drive voltage. The micro-ring has a natural resonant frequency based on the device dimension. By introducing an index change inside the ring, the resonant frequency
can be shifted and thus a large modulation depth occurs near the resonance. Microring
modulators utilizing carrier injection have been demonstrated by several groups [2, 3, 4].
The radius of such a ring modulator can be as small as 5 µm with only 1.8 V voltage
difference acquired to shift the resonant frequency and results in a 17 dB extinction
ratio (ER). The device is well known for its extreme compactness and sharp modu-
2
Chapter 1. Introduction
lation; applications such as comb generators, comb switches, and optical filters have
also been demonstrated based on the mirco-ring structure [3, 4]. However, due to the
nature of the ring structure, the optical bandwidth of the ring modulator is generally
smaller than 1 nm so that cascaded rings or more complicated structures are needed
for a multi-wavelength system. Moreover, the device is very sensitive to temperature,
bias condition, and fabrication variations. Temperature various within 1 ◦ C results in
shifting resonance by a full width at half maximum, hence extra components such as
heaters are required for tuning purposes.
As for high speed communication, the reported micro-ring resonator has demonstrated large signal modulation at 12.5 Gb/s with 9 dB ER. However, this can only
be achieved by utilizing a pre-emphasized electrical driven signal (Vpp = 8 V plus 3.5
V pre-emphasized pulse) since carrier injection is a relatively slow process limited by
the nanosecond carrier lifetime. It is possible to increase the modulation bandwidth if
carrier depletion is used to introduce the phase change. Nevertheless, either the size of
the ring or the overlap between optical mode and carrier swept region has to be increase
because carrier depletion is a weaker effect than carrier injection to introduce the index
shift.
Electroabsorption Modulator (EAM)
The electroabsorption modulator has been a very common device used in optical communication on the pure III-V platform over the past decade. By applying bias to the
material, the conduction and valence band edges are tilted, and this results in red shift
of the absorption spectrum. Silicon, however, is well known as an indirect bandgap material and the absorption efficiency is fairly weak compared to III-V material. Therefore,
two different approaches, germanium (Ge) grown on silicon and III-V bonded to silicon,
3
Chapter 1. Introduction
have been used to solve this problem. The SiGe EAM based on the Franz-Keldysh
and quantum confined Stark effects was reported recently [5, 6]. The waveguide SiGe
EAM demonstrated in [5] has small footprint around 50 µm in length and low energy
consumption per bit. However, additional absorption caused by the indirect bandgap
of Ge introduces higher propagation loss at zero bias, as shown in Table 1.1.
Another approach is to fabricate the EAM on the hybrid silicon evanescent platform
[7, 8], which consists of III-V material bonded with patterned silicon-on-insulator (SOI)
wafers such that the active devices can be implemented using III-V material properties
while low loss waveguides can be made on the SOI. The hybrid silicon EAM has 10
dB ER at -5 V and only 5 dB propagation loss for a 100 µm device at zero bias. Not
only does it have a large modulation bandwidth of 42 GHz, it also has large signal
modulation with 9.8 dB ER at 50 Gb/s. The large ER at high speed is necessary and
important for practical application in optical communication networks.
Mach-Zehnder Modulator (MZM)
The third option is the Mach-Zehnder interferometer (MZI). By changing the phase
on one or both arms of the MZI, either constructive or deconstructive interference is
achieved, and consequently intensity modulation can be expected. In order to introduce
π phase shift, electrooptic (EO) effects, such as carrier injection or carrier depletion, are
generally used. The device footprint is generally on the millimeter scale due to the weak
silicon EO effect [10, 12]. A high speed silicon modulator using the carrier depletion
effect with 1 dB ER at 40 Gb/s and Vπ L = 40 Vmm was reported in [10]. In contrast,
a silicon MZM based on forward bias carrier injection can have a small footprint (100
µm), yet has a carrier lifetime limited bandwidth (around 5 ns) [13, 11]. Although large
modulation of 10 Gb/s was demonstrated by using a pre-emphasized electrical driven
signal, extra electrical power is required. The third column in the MZI category is the
4
Chapter 1. Introduction
Table 1.1: Comparison of silicon modulators.
RM
Group
Platform
Ref.
Vpp (V)
Length(µm)
Vπ L(Vmm)
ER(dB)
f3dB (GHz)
ERRF (dB)∗1
Data rate(Gb/s)
Optical BW (nm)
Loss(dB/device)
Power handling(mW)
EAM
MZM
Cornell
Si
UCSB
Hybrid Si
MIT
SiGe
UCSB
Hybrid Si
Intel
Si
IBM
Si
[2]
1.8
10
–
17
–
9
12.5∗2
≤1
–
–
[8]
2
100
–
11
42
9.8
50
30
5
–
[5]
7
50
–
10
1
–
–
15
14.8∗3
–
[9]
4
500
2
20
25
10
40
100
4.5
28
[10]
6
1000
40
≥20
30
1
40
75
5.4
–
[11]
1
100
0.06
18
–
N/A
10∗2
110
12
–
∗1 The ER measured at the maximal data rate.
∗2 Pre-emphasized electrical drive signal.
∗3 The loss is calculated where the “one” level is used.
focus of this thesis and uses a hybrid silicon approach.
Modulator Comparison
In order to compare different types of modulators, Table 1.1 lists some of the important
parameters for each specific modulator. As can be seen, the ring modulator is compact
and operates at low voltages, but the optical and modulation bandwidths are both limited. It is suitable for high density and narrow bandwidth applications. In contrast, the
EAM has a slightly larger footprint, decent modulation efficiency and high modulation
bandwidth up to 42 GHz. For a reconfigurable communication system which requires
larger optical bandwidth, the EAM is a good candidate. The MZM has a larger footprint and requires higher bias condition; however, it has a wider optical bandwidth,
5
Chapter 1. Introduction
Rg
(a)
Rs
Cd
Vg
(b)
Vg
RL
Ru L u
Rg
Gu
Cu
Rs
Cd
RL
Figure 1.1: Circuit model for (a)lumped (b)traveling-wave modulator.
higher large signal modulation, and potential applications such as switches and routers
make it very competitive for future all optical communication networks.
1.2
Traveling-Wave Electrode
There are two standard approaches, lumped and traveling-wave, to design high-speed
modulators. Lumped circuits, as shown in Figure 1.1(a), are the easiest and most
conventional method to drive modulators. In general, the signal is fed into the device
from the center and experiences reflection from both ends of the device. Since there
is no termination at the end of the device (RL = ∞), the signal is totally reflected
and interacts with the input signal. The impedance mismatch between the source
6
Chapter 1. Introduction
(Rg = 50 Ω) and the device also introduces reflections, which degrades the modulation
bandwidth. This is especially pronounced at high frequencies. For a lumped circuit,
the frequency response depends on how effectively the supplied voltage is used to drive
the diode and is expressed as:
1
H(ω) = 1 + iω(Rg + Rs )Cd (1.1)
where Rg is the system impedance, and Rs and Cd are the series resistance and capacitance of the device, respectively. |H(ω)|2 is the frequency response. As can be
seen, the 3 dB cutoff frequency, which occurs when Rg + Rs equals |1/ωCd |, is inversely
proportional to the value of total resistance and capacitance. In general, Rg is larger
than Rs so it would be the dominant term in determining the cutoff frequency. For a
hybrid silicon MZM, Rs is less than 10 Ω and Cd is around 1 pF, which indicates a cutoff
frequency around 3 GHz. To further increase the bandwidth, one of the most conventional approaches is to decrease the device capacitance by reducing the total length,
but this leads to lower modulation efficiency and higher series resistance. This makes
the lumped modulator a poor approach for high-speed operation due to the tradeoff
between modulation efficiency and bandwidth.
On the other hand, the TWE design is a better approach because the cutoff frequency is limited by the distributed impedance and capacitance. As shown in Figure 1.1(b), the signal is applied to one end of the device, propagates along the waveguide
direction, and is terminated at the end of the transmission line. In this case, the signal only sees distributed components and hence the RC limitation present in a lumped
circuit design no longer exists. With carefully design of the transmission line, one can
reduce the reflection at the interface between source and the device. In addition, a
7
Chapter 1. Introduction
proper termination RL can help increase the bandwidth by eliminating the reflection
from the end. Therefore, the modulation bandwidth can be significantly improved and
over 10 GHz operation can be expected. The theoretical frequency response and circuit
model will be explained in detail in Section 3.1.
1.3
Applications
In this work, we also show the applicability of the developed modulator in realizing
a range of components for both digital and analog communications. In the digital
domain the modulator is used in a 2×2 optical switch and is shown to have 0.5 dB
power penalty at 40 Gb/s. It has the potential to meet the stringent energy and power
switching constraints of future large-scale optical networks. The design and performance
of the switch will be presented in Section 4.4.
In the analog domain, a thermal modulator is integrated with amplifiers and other
passive components on the HSP to realize a tunable microwave filter. The details of the
design, fabrication, and performance of the aforementioned devices will be presented in
Chapter 5.
1.4
Outline of This Thesis
The scope of this thesis is to develop a high-speed silicon modulator for optical communication, data transmission and computation. The application of this modulator
is not limited to existing long-distance transmission, but also can be utilized in photonic integrated circuits (PIC) for rack-to-rack, board-to-board, or even chip-to-chip
interconnects.
8
Chapter 1. Introduction
In Chapter 2, we will focus on the material design for an efficient hybrid silicon
modulator. The discussions and trade-offs between bandwidth, speed, and efficiency of
the modulator are addressed. Experimental data are shown to demonstrate the material
properties concept.
Chapter 3 focuses on the TWE design, and the circuit model to analyze the influences
of different design parameters. By utilizing the established model, we demonstrate a
hybrid silicon modulator up to 10 Gb/s. The bandwidth limitations are analyzed in
order to improve the design.
In Chapter 4, we introduce a different TWE design, a slotline-based design, to
increase the modulation efficiency and bandwidth by employing a push-pull structure.
In addition, the concept of the capacitively loaded electrode is added to the TWE
design such that the electric properties, such as propagation loss, device impedance,
and signal velocity, of the TWE can be enhanced. With this improved design, a hybrid
silicon modulator with 40 Gb/s large signal modulation is fabricated and demonstrated.
Meanwhile, we also demonstrate its application as a high-speed switch by changing the
passive waveguide design.
Chapter 5 presents one of the applications of the HSP. A tunable microwave filter
utilizing a low-loss long-delay loop and active components on this platform makes it
possible to implement strong resonance peaks in the GHz range.
Finally, Chapter 6 summarizes the contributions of this thesis and provides directions for future research.
9
Bibliography
[1] C. R. Doerr, D. M. Gill, A. H. Gnauck, L. L. Buhl, P. J. Winzer, M. A. Cappuzzo,
A. Wong-Foy, E. Y. Chen, and L. T. Gomez, “Monolithic demodulator for 40-gb/s
dqpsk using a star coupler,” J. Lightwave Technol., vol. 24, no. 1, p. 171, Jan
2006. [Online]. Available: http://jlt.osa.org/abstract.cfm?URI=JLT-24-1-171
[2] Q. Xu, B. Schmidt, S. Pradhan, and M. Lipson, “Micrometre-scale silicon
electro-optic modulator,” Nature, vol. 435, no. 7040, pp. 325–327, May 2005.
[Online]. Available: http://dx.doi.org/10.1038/nature03569
[3] N. Sherwood-Droz, H. Wang, L. Chen, B. G. Lee, A. Biberman, K. Bergman,
and M. Lipson, “Optical 4x4 hitless Silicon router for optical Networks-on-Chip
(NoC),” Opt. Express, vol. 16, no. 23, pp. 19 395–19 395, Nov 2008. [Online].
Available: http://www.opticsexpress.org/abstract.cfm?URI=oe-16-23-19395
[4] B. G. Lee, A. Biberman, P. Dong, M. Lipson, and K. Bergman, “All-Optical Comb
Switch for Multiwavelength Message Routing in Silicon Photonic Networks,” IEEE
Photonics Technology Letters, vol. 20, no. 10, pp. 767–769, 2008.
[5] J. Liu, M. Beals, A. Pomerene, S. Bernardis, R. Sun, J. Cheng, L. C. Kimerling,
and J. Michel, “Waveguide-integrated, ultralow-energy GeSi electro-absorption
modulators,” Nat Photon, vol. 2, no. 7, pp. 433–437, July 2008. [Online].
Available: http://dx.doi.org/10.1038/nphoton.2008.99
[6] J. E. Roth, O. Fidaner, R. K. Schaevitz, Y.-H. Kuo, T. I. Kamins, J. S.
Harris, and D. A. B. Miller, “Optical modulator on silicon employing germanium
quantum wells,” Opt. Express, vol. 15, no. 9, pp. 5851–5859, Apr 2007. [Online].
Available: http://www.opticsexpress.org/abstract.cfm?URI=oe-15-9-5851
[7] Y. hao Kuo, H.-W. Chen, and J. E. Bowers, “Hybrid silicon travelingwave electroabsorption modulator suitable for 50 Gb/s optical interconnects,”
Opt. Express, no. 13, pp. 9936–9941, Jun 2010. [Online]. Available: http:
//www.opticsexpress.org/abstract.cfm?URI=oe-16-13-9936
10
BIBLIOGRAPHY
[8] Y. Tang, H.-W. Chen, S. Jain, J. D. Peters, U. Westergren, and J. E. Bowers, “50
Gb/s hybrid silicon traveling-wave electroabsorption modulator,” Opt. Express, vol.
to be published, 2011.
[9] H.-W. Chen, J. D. Peters, and J. E. Bowers, “Forty Gb/s hybrid silicon
Mach-Zehnder modulator with low chirp,” Opt. Express, vol. 19, no. 2, pp.
1455–1460, Jan 2011. [Online]. Available: http://www.opticsexpress.org/abstract.
cfm?URI=oe-19-2-1455
[10] A. Liu, L. Liao, D. Rubin, H. Nguyen, B. Ciftcioglu, Y. Chetrit, N. Izhaky,
and M. Paniccia, “High-speed optical modulation based on carrier depletion in a
silicon waveguide,” Opt. Express, vol. 15, no. 2, pp. 660–668, Jan 2007. [Online].
Available: http://www.opticsexpress.org/abstract.cfm?URI=oe-15-2-660
[11] J. V. Campenhout, W. M. Green, S. Assefa, and Y. A. Vlasov, “Lowpower, 2x2 silicon electro-optic switch with 110-nm bandwidth for broadband
reconfigurable optical networks,” Opt. Express, vol. 17, no. 26, pp. 24 020–24 029,
Dec 2009. [Online]. Available: http://www.opticsexpress.org/abstract.cfm?URI=
oe-17-26-24020
[12] D. Marris-Morini, L. Vivien, J. M. Fédéli, E. Cassan, P. Lyan, and S. Laval, “Low
loss and high speed silicon optical modulator based on a lateral carrier depletion
structure,” Opt. Express, vol. 16, no. 1, pp. 334–339, Jan 2008. [Online]. Available:
http://www.opticsexpress.org/abstract.cfm?URI=oe-16-1-334
[13] W. M. Green, M. J. Rooks, L. Sekaric, and Y. A. Vlasov, “Ultracompact, low RF power, 10 Gb/s siliconMach-Zehnder modulator,” Opt.
Express, vol. 15, no. 25, pp. 17 106–17 113, Dec 2007. [Online]. Available:
http://www.opticsexpress.org/abstract.cfm?URI=oe-15-25-17106
11
Chapter 2
Optical Waveguide Design and
Proof of Concept
In this chapter, we will introduce the material system used for the hybrid silicon MachZehnder modulator (MZM). The physical effects utilized to alter refractive index are
first addressed and the advantage of the carrier depletion effect will be discussed followed by the epitaxial design of the III-V material. Finally, the dimensions of the
silicon waveguide and their effect on the optical confinement factor in different layers is
analyzed.
2.1
2.1.1
Optical Waveguide Design
Carrier Depletion Effect
As mentioned in the Chapter 1, the carrier depletion effect (CDE) [1, 2, 3]is preferred
for high-speed modulators since this physical mechanism is a fast process and is not the
12
Chapter 2. Optical Waveguide Design and Proof of Concept
(a)
(b)
P
N
-
P
N
N
N
Figure 2.1: (a)A p-i-n diode with lightly doped intrinsic layer where the carriers
stay. (b)With external bias, the carriers are swept to the n layer.
limitation in terms of the relevant frequency range. In order to implement a modulator
utilizing the CDE, a p-i-n diode with a lightly doped intrinsic layer is used as illustrated
in Figure 2.1. By reverse biasing this diode, the electric field and depletion region width
are changed, forcing the carriers to swept out from the intrinsic region. This results
in the desired index shift. The effects that are involved in causing an index shift are
explained below:
• Pockels Effect
The Pockels effect, also known as the linear electrooptic (EO) effect, is an anisotropic
effect proportional to the applied electric field (E in Equation 2.1). With a uniform electrical field, this effect is proportional to the bias voltage.
1
∆nP ockels = n3 r41 E
2
(2.1)
where n is the refractive index of the material, and r41 is the Pockels coefficient,
where r41 is 1.6 × 10−12 m/V for GaAs and 1.68 × 10−12 m/V for InP.
This effect is strongly dependant on the orientation of the applied fields and crystal
13
Chapter 2. Optical Waveguide Design and Proof of Concept
plane, i.e., the index change (∆n) is positive for TE polarization if light propagates
along h11̄0i direction and is negative if light propagates along h110i direction. In
the hybrid silicon platform (HSP) utilized in this work, special attention needs to
be paid to the orientation of the crystal plane during the bonding process.
• Quantum Confined Stark Effect (QCSE)
The QCSE effect is a second order EO effect proportional to the square of the
applied electric field inside the MQW structure and can be written as:
1
∆nQCSE = n3 RE 2
2
(2.2)
where R is the empirical quadratic EO coefficient and is 0.25×10−18 (m2 /V 2 ) [4].
• Plasma Effect
The number of carriers interacting with the optical mode changes with external
bias. The relationship between carrier density difference and the index change
can be expressed as:
∆nplasma =
−e2 λ20
∆Nd
·
8π 2 c2 0 n m∗e
(2.3)
where ∆Nd is the number of depleted carriers (cm−3 ) and m∗e is the effective
mass of the electrons in the intrinsic layer. In our case, we only consider the
index change introduced by electrons since the intrinsic layer is slightly N doped.
• Band-Filling Effect
Similar to the plasma effect, the band-filling effect [1] is also introduced by a
change in carrier concentration in the intrinsic region. In addition to the plasma
effect, carrier depletion also causes a variation of the absorption coefficient that
results in a refractive index change. This is due to the shift of the absorption edge
14
Chapter 2. Optical Waveguide Design and Proof of Concept
10
x 10
-20
material bandgap
P bf
8
6
4
2
0.65
0.7
0.75
0.8
0.85
0.9
1.0
Photon Energy (eV)
Figure 2.2: Energy dependence of the proportionality constant for band-filling.
by carriers accumulated at the bottom of the conduction band edge. The total
index change introduced by the band-filling effect can be simplified as :
∆nbf = Pbf · ∆Nd
(2.4)
where Pbf is a proportionality constant and is highly dependant on the photon
energy of the input light signal as illustrated in Figure 2.2. Pbf increases exponentially as the photon energy of the light gets closer to the material bandgap
(0.91 eV).
Therefore, the total index change can be obtained by summing all the mentioned
effects with the positive and negative Pockels effects. The confinement factor of each
15
Chapter 2. Optical Waveguide Design and Proof of Concept
1.5
Total
(positive)
1
∆n (x10 -3 )
Total
(negative)
Band
filling
0.5
QCSE
Pockels
0
Plasma
0
1
2
3
4
5
Reverse Bias (V)
Figure 2.3: The total index change introduced by the carrier depletion effect in a
p-i-n diode.
layer is included in our calculation for a more accurate estimation of the index change.
∆np = ∆nQCSE + ∆nplasma + ∆nbf + ∆nP ockels
∆nn = ∆nQCSE + ∆nplasma + ∆nbf − ∆nP ockels
(2.5)
The estimated index change as a function of reverse bias is shown in Figure 2.3
where the index is proportional to the magnitude of bias. Since the Pockels effect is
linearly dependant on the applied electric field, it is not surprising that the index change
linearly increases as the bias increases. On the other hand, the QCSE effect is smaller
than the Pockels effect at lower bias, but increases faster than the Pockels effect as the
16
Chapter 2. Optical Waveguide Design and Proof of Concept
bias becomes larger due to its quadratic dependance on the electric field. The bandfilling effect is about two times larger than the Pockels and plasma effects, but is about
the same to QCSE at -5 V. This suggests that the band-filling effect is a much more
efficient effect in general, and it can approximately double the modulation efficiency by
depleting the carriers via bandgap engineering and index shift.
Of all the effects that introduce index change using this carrier depletion effect, The
Pockels effect is the only one sensitive to wafer orientation. With different alignment
between optical waveguide and the crystal plane, the index change between positive
and negative Pockels effect results in about 1.5 times difference in total index change
as shown in Figure 2.3. Consequently, the voltage-length product becomes larger if the
Pockels effect is negative even with identical device layout. In our case, the major flat of
the III-V wafer needs to be perpendicular to the optical waveguide on the silicon wafer
during bonding in order to have the Pockels effect be positive.
2.1.2
Epitaxial Layer Design and Bandgap Engineering
In order to implement a hybrid silicon modulator using CDE, we start our design by
choosing a material system that maximizes index change. A quaternary such as InGaAsP is very common and well researched for photonic devices. However, the bandgap
offset of this system is relatively small compared to InGaAlAs as shown in Figure 2.4.
The conduction band offset in InGaAlAs can be two times larger than that in the
InGaAsP system such that there is more freedom in band structure design. In addition, the InGaAlAs is of special interest for high-speed design due to its potential to
selectively oxidize layers with high concentrations of Al. This oxidation can create an
undercut layer, e.g. the intrinsic layer for a diode, to reduce the total device capacitance
17
Chapter 2. Optical Waveguide Design and Proof of Concept
200
250
500
450
AlAs
2170
GaAs
1420
InGaAs
760
InAlAs
1460
InGaP
1900
InP
1350
200
170
550
200
Figure 2.4: Bandgap heaven of III-V materials. The bandgap are drawn in black
while the offsets are labeled in blue with the unit in meV.
p contact
P InGaAs
1e19
Cladding
P InP
1e18
+
SCH
n AlGaInAs
QW
n AlGaInAs
SCH
n contact
n+ AlGaInAs
+
n InP
n.i.d
3e18
Figure 2.5: Illustration of the epitaxial layers of a hybrid silicon modulator.
without introducing any mechanical instability. By manipulating the Al concentration,
oxidation time and temperature, a device feature can be precisely controlled on the
sub-micrometer scale[5].
To have a general understanding of the epitaxial design of the carrier depletion
18
Chapter 2. Optical Waveguide Design and Proof of Concept
modulator, a conventional epitaxial stack is illustrated in Figure 2.5, where the layers
from top to bottom are p contact, cladding, top SCH, QW, bottom SCH, and n contact.
The top p contact layer is used to create an ohmic contact alloy between contact metal
and the semiconductor by annealing. With a doping concentration of 1 × 1019 cm−3 ,
which is the limit of the growth vendor, the contact resistance is minimized. Similarly,
the n contact layer at the bottom has a doping concentration of 3 × 1018 cm−3 . The
p cladding on top of the SCH layer is used to separate the optical mode inside the
waveguide from the metal on top. If the optical mode overlaps with the contact metal,
significant optical loss is introduced, which consequently reduces the modulation depth
and degrades the device performance. The thickness of this cladding layer depends on
the thickness of other layers, including the silicon waveguide that will be discussed in
Section 2.1.3. Here, we choose 1.5 µm for this cladding layer to provide a safe margin.
Next, we need to decide the compositions of the quaternaries to match our requirements for a hybrid silicon modulator. First, the PL peak of the separate confinement
heterostructure (SCH) is fixed at 1.3 µm. The bandgap of the SCH layer needs to be
large enough such that there is limited absorption of the optical signal at 1.5 µm. Also,
it is important to keep this SCH layer lattice matched to the InP substrate to avoid any
strain imbalance. Therefore, the composition of the SCH can be decided by mapping
the requirements onto Figure 2.6 such that the compositions of Indium (In), Gallium
(Ga), and Al for SCH are 0.53, 0.305, and 0.165, respectively. With this starting point,
the compositions of the quantum wells (QWs) and barriers are constrained within a
certain range due to the growth technique. In reality, it is preferable to keep one of
the three compositions (In, Ga, and Al) constant over the entire growth because it is
easier to control the material quality and reliability. If the composition varies during
the growth processing, it is much more difficult to control the growth conditions, i.e.
19
Chapter 2. Optical Waveguide Design and Proof of Concept
1
1.46 µm
0.9
0% +0.3%
-0.4%
0.8
0.7
Ga(y)
0.6
1.3 µm
0.5
0.4
1.16 µm
0.3
0.2
0.1
0
0.2
0.3
0.4
0.5
0.6
0.7
In(x)
Figure 2.6: Material composition for different values of strain and bandgap in the
Inx Gay Al1−x−y As. The material parameters can be found in [6, 7].
gas flow, chamber pressure, and temperature such that the final composition may deviate from the original design. Since the strain shown in Figure 2.6 is very sensitive to
In composition, we choose to keep Ga constant over the growth, which provides more
freedom to satisfy the design rules for QWs and barriers that will be discussed later in
this section.
To have a proper design for QWs and barriers based on the CDE, several aspects
need to be considered.
1. Strain balance
Lattice match is always an issue for quaternaries since their lattice constants
are usually different from that of the InP or GaAs substrate. One of the most
20
Chapter 2. Optical Waveguide Design and Proof of Concept
conventional approaches to design a lattice matched material is to counteract the
strains between two different layers, e.g. QWs and barriers. The strain, defined
as S = (a − a0 )/a0 , is compressive when a > a0 and tensile when a < a0 . With
proper design, the total strain, expressed in Equation 2.6, could be zero and the
structure becomes mechanically stable.
Snet = dQW · SQW + db · Sb
(2.6)
where dQW and db are the thicknesses of QW and barrier, respectively, while SQW
and Sb represent the strains in the QW and barrier.
2. PL peak
PL peak and bandgap design are the most important parameters for photonic devices. Lasers and amplifiers have bandgaps very close to the operating light signal
to provide strong interaction between the photons and the electrons through the
absorption process. In contrast, when designing modulators, especially phase
modulators, it is very important to keep the bandgap large enough to avoid undesired absorption that introduces additional loss in the device. In general, the
separation between the light wavelength and the PL peak of the modulator has to
be more than 160 nm because the PL red shifts with external bias. However, the
index change introduced by the carrier depletion effect decreases if the separation
is too large because the band-filling effect depends on the position of bandgap
edge.
3. Band offset
For a QW structure, the fundamental state (E0 ) strongly depends on the barrier
21
Chapter 2. Optical Waveguide Design and Proof of Concept
height, i.e. the band offsets in both the conduction (∆Ec ) and valence bands
(∆Ev ). The offset should be deep enough to accumulate carriers for carrier depletion. However, if the offset is too large, such that the first order state (E1 )
exists, then the carriers will stay in this state rather than being depleted. Hence,
the best approach is to set ∆Ec equal to E1 to maximize the carriers stored inside
the QW. Here we fix the QW thickness at 8 nm and barrier thickness at 5 nm for
larger carrier confinement in the QW and better depletion efficiency. The band
offset, therefore, can be calculated to fulfill these requirements.
Table 2.1: Compositions of SCH, QWs and barriers
Layer
Material Composition
Thickness
SCH
QW
barrier
In0.53 Ga0.305 Al0.165 As,1.3µm
In0.574 Ga0.315 Al0.111 As,1.46µm, +0.3%(15x)
In0.468 Ga0.315 Al0.217 As,1.16µm, -0.41%(16x)
8 nm
5 nm
By considering all these three aspects, the compositions of the SCH, QW, and barrier
are listed in Table 2.1. Here, the PL peak is at 1360 nm, ∆Ec is 131 meV, and E0 is
43 meV. The density of states as a function of conduction band energy is shown in
Figure 2.7, where the shaded area represents the total carrier concentration of 1×1018
in this QW.
The doping concentration in SCH and QW layers is one of the most important
parameters of the epitaxial design because it is a major source of index change. In this
case, the carriers inside the SCH layers need to be totally depleted such that voltage
drop inside the QW is maximized, which indicates that the SCH thickness must be
adjusted based on the doping concentration. Figure 2.8(a) depicts the built-in depletion
width as a function of doping concentration in SCH (Nd,SCH ) when the cladding doping
22
Chapter 2. Optical Waveguide Design and Proof of Concept
2
x 10
44
1.8
1.6
E1
DOS (J -1 m-3 )
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
E0
50
100
Energy (meV)
∆E c
150
Figure 2.7: Density of states of the QW listed in Table 2.1. The shadow area
represents the total carrier density and the dash line is ∆Ec
is 1 × 1018 cm−3 . As can be seen, the depletion region width, also defined as SCH
thickness (dSCH ) in our case, is around 0.35 µm when Nd,SCH is 1 × 1016 cm−3 and
decreases to 0.1 µm when Nd,SCH is 1 × 1017 cm−3 . The combination of Nd,SCH and
dSCH , however, affects the modulation efficiency (voltage-length product), modulation
bandwidth, and optical propagation loss simultaneously. Figure 2.8(b) and (c) show
the influence of Nd,SCH on index change and modulation bandwidth. Here we assume
the optical confinement factor inside SCH (ΓSCH ) is 9.11% when dSCH is 0.1 µm and
varies linearly with dSCH . The bandwidth is calculated using an RC cutoff frequency
model with a series resistance of 5 Ω. If Nd,SCH increases (dSCH decreases), as shown
in Figure 2.8(b), then the modulation efficiency increases because more carriers can be
depleted to introduce index change and stronger electrical fields exist inside the QW.
Each line represents a different bias voltage (1-5 V) across the diode. In contrast, the
modulation bandwidth degrades (Figure 2.8(c)) since the total device capacitance is
23
Chapter 2. Optical Waveguide Design and Proof of Concept
0.4
(b)
0.2
0.1
0
16
10
10
17
10
-3
3V
fc *∆n (GHzx10-3 )
20
10
17
SCH Doping (cm
10
-3
10
17
SCH Doping (cm
(d)
10
1V
)
30
fc (GHz)
0.4
0
16
10
18
40
0
16
10
5V
0.6
0.2
SCH Doping (cm
(c)
1
0.8
0.3
∆n (x10 -3 )
Depletion width (µm)
(a)
18
)
10
-3
18
)
15
5V
10
3V
5
1V
0
16
10
10
17
SCH Doping (cm
10
-3
18
)
Figure 2.8: (a)The built-in depletion region width for a SCH layer next to a p
doped cladding layer of 1 × 1018 cm−3 . (b)Estimated index change as a function
of NSCH at different bias conditions. (c)3 dB cutoff frequency as a function of
NSCH with series resistance of 5 Ω. (d)Overall performance at different value of
NSCH .
larger due to the decrease of dSCH . The combination of bandwidth and modulation
(total efficiency) as a function of Nd,SCH is shown in Figure 2.8(d) to illustrate the
optimal doping concentration. It is clear that Nd,SCH larger than 1 × 1017 cm−3 is
not preferred since the bandwidth drops very quickly in this region. With Nd,SCH
24
Chapter 2. Optical Waveguide Design and Proof of Concept
smaller than 1 × 1017 cm−3 , the variation of total efficiency is not significant such that
it gives some room for other design concerns. Here we choose Nd,SCH as 1 × 1017 cm−3
to have an adequate dSCH . A large value of dSCH is not preferred in terms of the
optical confinement because ΓSCH might be larger than ΓQW , resulting in insufficient
modulation and large coupling loss from the silicon waveguide.
Table 2.2: Epitaxial layer structure of the hybrid silicon modulator
Layer
Material Composition
p contact
In0.53 Ga0.47 As
Cladding
InP
SCH
In0.53 Ga0.305 Al0.165 As,1.3µm
QW
In0.574 Ga0.315 Al0.111 As,+0.3%(15x)
(λP L =1.36µm) In0.468 Ga0.315 Al0.217 As,-0.41%(16x)
SCH
In0.53 Ga0.305 Al0.165 As,1.3µm
n contact
InP
Super lattice
In0.85 Ga0.15 As0.327 P0.673 ,(2x)
InP(2x)
Bonding
InP
Doping
Thickness
P-1e19
P-1e18
N-1e17
N-1e17
N-1e17
N-3e18
N-3e18
N-3e18
N-3e18
N-3e18
0.1 µm
1.5 µm
0.1 µm
8 nm
5 nm
0.05 µm
0.11 µm
7.5 nm
7.5 nm
10 nm
As a summary, the parameters of the designed epitaxial layers are shown in Table
2.2, and the band diagram neglecting doping concentration for SCH, QWs and barriers
is shown in Figure 2.9(a). The splitting of the valence band due to the strained QWs is
also plotted for heavy holes and light holes. Moreover, the band diagram with doping
concentration is calculated utilizing commercial software (Silvaco), and the simulated
results, including electric field and conduction band energy, are shown in Figure 2.9(b)
and (c). It is clear that the carriers are well confined in the QWs at 0 V, and are
depleted from the QWs with external applied bias due to the change of fundamental
state in the wells.
25
Chapter 2. Optical Waveguide Design and Proof of Concept
1.5
(a)
Bandgap (eV)
Ec
1
Cladding
SCH
QWs
SCH
E vhh
0.5
E vhh (red)
(b)
Electrical Field (kV/cm)
400
300
4V
200
3V
2V
100
1V
0V
0
2V 3V
4V
1V
(c)
Ec (eV)
1.5
0V
1
0.5
0
1.5
1.6
1.7
1.8
1.9
2
Distance from top (µm)
Figure 2.9: (a)Simulated band diagram using parameters listed in Table 2.2. Both
heavy and light (red curve) hole valence bands are plotted. (b)The electrical field
at different biases. (c)The conduction band energy at different biases.
26
Chapter 2. Optical Waveguide Design and Proof of Concept
2.1.3
Passive Waveguide Design
On of the advantages of the HSP is that it utilizes two different material systems,
silicon and III-V. The optical mode is tightly confined in the silicon for passive sections
(waveguides without III-V on top) due to the high index contrast between silicon and
the buried oxide (BOX) layer on the silicon-on-insulator (SOI) wafer. This provides the
possibility for sharp waveguide bends and also strong interaction between photons and
electrons. On the other hand, the optical mode in the hybrid section is controlled by
the dimensions (thickness and width) of the silicon/III-V waveguides since the refractive
indexes of both materials are similar. A cross section of the hybrid section is shown
in Figure 2.10 where the indices and the dimensions of each layer are listed in Table
2.3, where H, W, and R represent the height, width and rib etch height of the silicon
waveguide, respectively. G is the trench gap, which is 2 µm on the sides of the silicon
waveguide to avoid any leakage of the optical mode.
By adjusting the silicon height and width, the shape of the mode is changed accordingly and the simulated confinement factors in silicon and active (QWs and barriers)
p contact
p cladding
top SCH
bottom SCH
Si
H
R
QW
G
n contact
w
BOX
Si substrate
Figure 2.10: The cross section of a hybrid silicon modulator.
27
Chapter 2. Optical Waveguide Design and Proof of Concept
Table 2.3: Dimensions and Indexes for each layer in Figure 2.10
Layer
p contact
Cladding
top SCH
QW/barriers
bottom SCH
n contact
silicon
BOX
Thickness (µm)
Width (µm)
Index
0.1
1.5
0.1
0.2
0.05
0.15
H
1
4
4
2
2
20
20
W
N/A
3.4146
3.1563
3.4146
3.4467
3.4146
3.1563
3.4764
1.45
region (Γsi and Γactive ) are depicted in Figure 2.11(a). For a silicon waveguide with
fixed width, Γsi increases as H increases. On the other hand, Γactive is reduced since the
effective index of the silicon waveguide becomes larger and pulls the mode down. The
dependence of confinement factors on the silicon waveguide width is obvious as well.
The optical mode could be pushed up into the III-V region by shrinking the width (W)
such that efficient interaction between electrons and photons is possible. However, if
Γactive is so large that most of the mode is inside the active region, there will be large
coupling loss between passive and the hybrid sections. Therefore, a balanced design with
adequate Γsi and Γactive is necessary as a compromise between modulation efficiency and
coupling loss. Hence, we choose H=0.5 µm and W=1 µm such that the Γsi is around
50%, Γactive is about 20 % and the other portion of the mode is in SCH, n contact, and
cladding layers. The simulated optical mode profile is shown in Figure 2.11(b). As can
be seen, most of the optical mode sits inside the silicon waveguide while the remainder
of the mode is inside the III-V region.
Due to the mode mismatch between the passive and hybrid sections, a taper needs
to be utilized to minimize reflection and increase coupling efficiency. Figure 2.12(a)
28
Chapter 2. Optical Waveguide Design and Proof of Concept
100
(a)
W = 1.0 µm
W = 1.5 µm
W = 2.0 µm
80
si
Γ (%)
Γactive (%)
60
40
20
0
0.4
0.45
0.5
0.55
0.6
H (µm)
3
Height (µm)
(b)
2
1
0
-5
-4
-3
-2
-1
0
1
2
3
4
5
Width (µm)
Figure 2.11: (a)The confinement factor in the silicon waveguide and the active
region as functions of H and W. (b)Simulated mode profile in the hybrid section
with H=0.5 µm and W=1 µm.
shows a schematic of the taper section where we adjust the dimensions in both silicon
and III-V layers to achieve better coupling efficiency. The silicon waveguide is tapered
from 1.5 µm to 1 µm while the III-V taper is changed from 0.6 µm to 2 µm and vice
versa when the optical mode propagates from hybrid to passive sections. The simulated
coupling loss is displayed in Figure 2.12(b) with various taper lengths. As can be seen,
the coupling loss is around 3 dB per taper, which corresponds to Γsi =50 %, if the taper
29
Chapter 2. Optical Waveguide Design and Proof of Concept
0
(b)
Coupling loss (dB)
(a)
Silicon waveguide
Taper length
x
-2
-3
0
III-V waveguide
(c)
50
Taper length ( µm)
100
0
Coupling loss (dB)
y
-1
-1
-2
-3
0
0.1
0.2
0.3
0.4
0.5
x s (µm)
Figure 2.12: (a)Schematic illustration of the III-V/Si taper. (b)The coupling
loss as function of taper length. (c)The coupling loss as a function of different
alignment shifts in x direction.
section did not exist. The loss is reduced to 0.5 dB and is constant for any taper length
larger than 10 µm. To be conservative, we choose a taper length of 60 µm. In addition
to the taper length, alignment accuracy affects the coupling efficiency. The alignment
accuracy is highly dependent on the fabrication tools available. The accuracy is within
± 0.6 µm using an I-Line stepper and can be improved to ± 0.2 µm by using automatic
alignment. Figure 2.12(c) shows the coupling loss at different values of the alignment
shift in x direction (xs ). The loss does not have significant variation over ± 0.5 µm
30
Chapter 2. Optical Waveguide Design and Proof of Concept
(a) 1
3
φ
2
-5
(b) φ = 0
0
1-> 4
Transmission (dB)
Transmission (dB)
0
-10
-15
-20
1-> 3
-25
-30
1500
4
1550
1600
Wavelength (nm)
-5
(c) φ = π
1-> 3
-10
-15
-20
1-> 4
-25
-30
1500
1550
1600
Wavelength (nm)
Figure 2.13: (a)Schematic illustration of two 2x2 MMI and a phase section. (b)The
transmission without any phase in the MZI. (c)The transmission with π phase shift
in the MZI.
range. The mode profile, however, is altered and a higher order mode could be excited
if the alignment is not good enough.
Another important component in the passive design is the splitter to separate the
incoming optical signal into the two arms of the MZI. In order to have better fabrication
tolerance and avoid the use of an Ebeam writer, we choose the multimode interferometer
(MMI) as the splitter. A schematic drawing of such an MMI used in our device layout
is shown in Figure 2.13(a) where the MMI length is 80 µm and the width is 7.5 µm with
a silicon waveguide of 0.5 µm height, 1 µm wide, and 0.5 µm rib etch. By adjusting the
phase section inside the MZI, constructive interference occurs at different output ports
(port 3 or 4). Figure 2.13(b) and (c) are the simulated results of the transmission with
31
Chapter 2. Optical Waveguide Design and Proof of Concept
and without 180 degree phase shift in the MZI from 1500 nm to 1600 nm. The state
of transmission is defined as “Cross” when φ = 0 and “Bar” when φ = π. As can be
seen, the MMIs have an optical bandwidth around 100 nm with loss less than 3 dB,
regardless of the state. The cross talk is below -20 dB within 50 nm around 1550 nm
and slightly increases to -15 dB at 1500nm.
2.2
2.2.1
Proof of Concept
Modulation Efficiency and Proof of Concept
(a)
nh L h
nsi L si /2
n L
∆λ
(b)
0V
1V
Figure 2.14: (a)A Fabry-Perot structure used to measure the index shift.
(b)Illustrated spectra with different levels of reverse bias.
To experimentally verify the index change introduced by the CDE, a test structure
shown in Figure 2.14(a) is utilized to extract the index change. All the parameters
32
Chapter 2. Optical Waveguide Design and Proof of Concept
Table 2.4: Dimensions and Indexes in Figure 2.14
Length
Index
Total
L = Lsi + La
nh + nsi
Hybrid section Lh = 500 µm nh = 3.2982
Passive section Lsi = 2347 µm nsi = 3.4764
are listed in Table 2.4. This structure includes a 500 µm hybrid section and an approximately 2 mm long silicon waveguide with polished facets on both sides to create
a resonant cavity. By measuring the shift of the resonance peak at various biases, the
introduced index change can be extracted from Equation 2.7.
∆n =
∆λ n · L
·
λ
Lh
(2.7)
where λ is the wavelength of the optical signal and ∆λ is the shift of the resonant peak
shown in Figure 2.14(b).
The index change measured by this test structure is shown in Figure 2.15. As can be
seen, Figure 2.15(a) and (b) represent the total index change (Equation 2.5) with correct (∆np ) and incorrect (∆nn ) bonding alignments at different operating wavelengths,
respectively. It is obvious that the modulation depth will be better when the index
change introduced by the Pockels effect (∆nP ockels ) has identical sign with that from
other effects (∆nr = ∆nQCSE + ∆nplasma + ∆nbf ). ∆np is about 1.5 times larger than
∆nn at 1550 nm, which is consistent with the simulation in Figure 2.3. As a summary,
it is extremely important to have correct alignment during bonding such that maximal
index change can be obtained.
Based on the experimental data shown in Figure 2.15(a) and (b), the ∆nP ockels
and ∆nr can be extracted. The results are shown in Figure 2.15(c) and (d). It is not
33
(a)
1.5
(b)
1.5
∆n (x10 -3 )
1
∆n (x10 -3 )
Chapter 2. Optical Waveguide Design and Proof of Concept
1
0.5
0.5
1525, 1550, 1575 nm
1525, 1550, 1575 nm
0
0
1
2
3
4
0
5
0
1
2
3
4
5
(c)
1.5
(d)
1.5
1
∆n (x10 -3 )
Reverse Bias (V)
∆n (x10 -3 )
Reverse Bias (V)
1
0.5
0.5
1525, 1550, 1575 nm
0
0
1
2
3
4
5
Reverse Bias (V)
0
0
1
2
3
4
5
Reverse Bias (V)
Figure 2.15: Experimental results of: (a)The total index change with correct bonding alignment. (b)The index change with incorrect bonding alignment. (c)The
index change introduced by the Pockels effect. (d)The index change due to the
rest of the physical effects.
surprising to see that ∆nP ockels does not vary too much at different wavelengths because
it is only related to the magnitude of the electric fields across the depletion region. On
the other hand, ∆nr , shown in Figure 2.15(d), is a function of operating wavelengths
and becomes smaller when the wavelength increases. This is because the band-filling
effect is more pronounced when the operating wavelength is closer to bandgap of the
QWs. Again, all the experimental results agree with the simulation shown in Figure 2.3.
34
18
18
16
16
14
14
12
12
10
10
8
8
6
6
4
4
2
2
0
1500
1520
1540
1560
1580
VπL (Vmm)
Extinction Ratio (dB)
Chapter 2. Optical Waveguide Design and Proof of Concept
0
1600
Wavelength (nm)
Figure 2.16: The measured ER and Vπ L of a 500 µm long MZM.
2.2.2
Optical Bandwidth
One important parameter for MZMs to be used in WDM systems is the spectral bandwidth required to modulate signals at different wavelengths. Figure 2.16 shows the
relative extinction ratio (ER) and voltage-length product (Vπ L) of a 500 µm long MZM
at various wavelengths. The ERs are above 12 dB from 1500 nm to 1600 nm while
the Vπ Ls are below 3 Vmm over a 100 nm optical bandwidth. The 100 nm optical
bandwidth is about three times larger than that (about 30 nm) of electroabsorption
modulators (EAMs) [8]. In general, MZMs have larger optical bandwidths than EAMs
because electrorefraction effects are the main effects used to generate index change for
MZMs, and these effects are less wavelength sensitive compared to electroabsorption
effects.
35
Chapter 2. Optical Waveguide Design and Proof of Concept
The optical loss is around 3 dB/mm for the hybrid section and 0.7 dB/mm for
the passive section. The large propagation loss in the passive section is due to overlap
between the optical modes and the rough sidewalls. This should be able to be improved
if a careful process is developed to reduce the roughness. In addition, the taper loss is
around 1-1.5 dB per taper, which is affected by the alignment accuracy and dosage of
proton implant.
2.2.3
Summary
In this chapter, we explore and design the material system for a high-speed modulator by
considering the trade-off between device footprint, optical bandwidth, and modulation
bandwidth. The bandgap and material compositions are researched to achieve the
requirements of the aforementioned aspects while the physical effects used to introduce
the index change are discussed. The concept of carrier depletion effect on the HSP is
verified by the experimental data and ready for high-speed design.
36
Bibliography
[1] J.-F. Vinchant, J. A. Cavailles, M. Erman, P. Jarry, and M. Renaud, “InP/GaInAsP
guided-wave phase modulators based on carrier-induced effects: theory and experiment,” IEEE Journal of Lightwave Technology, vol. 10, no. 1, pp. 63–70, 1992.
[2] J. G. Mendoza-Alvarez, L. A. Coldren, A. Alping, R. H. Yan, T. Hausken, K. Lee,
and K. Pedrotti, “Analysis of depletion edge translation lightwave modulators,”
IEEE Journal of Lightwave Technology, vol. 6, no. 6, pp. 793–808, 1988.
[3] H. Ohe, H. Shimizu, and Y. Nakano, “Ingaalas multiple-quantum-well optical phase
modulators based on carrier depletion,” IEEE Photonics Technology Letters, vol. 19,
no. 22, pp. 1816–1818, 2007.
[4] S. Nishimura, H. Inoue, H. Sano, and K. Ishida, “Electrooptic effects in an InGaAs/InAlAs multiquantum well structure,” IEEE Photonics Technology Letters,
vol. 4, no. 10, pp. 1123–1126, 1992.
[5] Y. C. Chang, C. S. Wang, L. A. Johansson, and L. A. Coldren, “High-efficiency,
high-speed VCSELs with deep oxidation layers,” Electronics Letters, vol. 42, no. 22,
pp. 1281–1282, 2006.
[6] T. Ishikawa and J. E. Bowers, “Band lineup and in-plane effective mass of InGaAsP
or InGaAlAs on InP strained-layer quantum well,” IEEE Journal of Quantum Electronics, vol. 30, no. 2, pp. 562–570, 1994.
[7] S. L. Chuang, Physics of Optoelectronic Devices, J. W. Goodman, Ed.
Interscience Publication, John Wiley & Sopns, Inc., 1995.
Wiley-
[8] H. Park, M. N. Sysak, H.-W. Chen, A. W. Fang, D. Liang, Y. Tang, L. Liao, B. R.
Koch, R. Bovington, J. Jones, and J. E. Bowers, “Device and Integration Technology
for Silicon Photonic Transmitters,” IEEE Journal of Selected Topics in Quantum
Electronics, vol. Invited, to be published, 2011.
37
Chapter 3
Coplanar Waveguide TWE
Mach-Zehnder Modulator
This chapter focuses on the traveling-wave electrode (TWE) design. Details of design rules, including general transmission line concepts for semiconductor devices and
a theoretical circuit model, are introduced in Section 3.1. Theoretical estimation of
modulation bandwidth based on different parameters is also discussed in order to have
better understanding of the circuits. The fabrication of hybrid silicon modulators is
then depicted in Section 3.2. Based on the theoretical analysis and developed process,
the performance of a hybrid silicon MZM is shown in Section 3.3. The device electrical
properties are in good agreement with the simulation. In addition, EO response and
large signal modulation are shown in Section 3.3.2 to demonstrate optical transmission capabilities of the device. Finally, the chapter concludes with a brief discussion of
performance of the device and possible improvements for a faster modulator.
38
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
3.1
3.1.1
Traveling-Wave Electrode Design
General Transmission Line and Frequency Response
Zg
Vg
l
Vin
ZD
z
z=0
ZL
x
x=0
Figure 3.1: General transmission line circuit with source and load.
Parameters
Table 3.1: Parameters of Figure 3.1.
Definition
Vg
Vin
Zg
ZD
ZL
l
γ
β0
Voltage provided by the source
Voltage at the interface between source and device
Source impedance
Transmission line impedance
Load impedance
Device length
Electrical propagation constant
Optical propagation constant
To find a theoretical expression for the frequency response of a modulator with
a TWE, the derivation starts by finding the voltage at an arbitrary point along the
transmission line. By adding the interaction between optical and electrical signals, a
39
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
normalized frequency can then be extracted [1, 2]. A simple transmission line circuit
is shown in Figure 3.1 where all the parameters are listed in Table 3.1. The voltage,
current, and reflection at a specific point z can be written as:
V = V + + V − = V0+ e−γz + V0− e+γz
I = I+ + I− =
Γ=
V0+ −γz
(e
− e+γz )
ZD
V−
V0− +2γz
=
e
V+
V0+
(3.1)
(3.2)
(3.3)
The reflections at some special positions are listed here:
V0− +2γl
e
V0+
(3.4)
V0−
−2γl
+ = ΓL e
V0
(3.5)
ΓL = Γ(z = l) =
Γ0 = Γ(z = 0) =
In addition, the relationship between Vg and V0+ can be expressed as:
V0+ =
Vg 1 + Γ0
2 1 + Γ20
(3.6)
by starting the derivation from Vg = Vin + Iin Zg . For a more general expression, the
voltage and reflection at point z can be rewritten as (x = l − z):
Γ = ΓL e−2γx
(3.7)
V = V + (1 + Γ) = V0+ e−γ(l−x) (1 + ΓL e−2γx )
(3.8)
Since the modulation is the result of the interaction between electrical and optical
40
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
signals, one can imagine that their two velocities have to be kept identical to have
maximum modulation efficiency. Any deviation of propagation velocity causes phase
mismatch between those two signals and consequently results in reduction of modulation
depth. By multipling Equation 3.8 with eiβo (l−x) and combing Equation 3.6, the effective
voltage for modulation at z can be expressed as:
VEO =
Vg
eiβo l
(1 + Γ0 ) iβe l
(ei(βe −βo )x + ΓL e−i(βe +βo )x )
2
e
+ Γ0 ΓL e−iβe l
(3.9)
where βe = −iγ. Next, the average voltage across the entire device length can be found
by integrating Equation 3.9 from 0 to l.
Vavg =
=
Z
1 l
VEO
l 0
Vg
eiβo l
(1 + Γ0 ) iβe l
[V + + ΓL V − ]
2
e
+ Γ0 ΓL e−iβe l
(3.10)
where all the parameters can be found below:
ωno
c
ZL − ZD
ΓL =
ZL + ZD
−
− sinφ
V − = e−iφ
φ−
βe + βo
φ− =
l
2
ωne
− iα,
c
ZD − Zg
=
,
ZD + Zg
+
+ sinφ
= eiφ
,
φ+
βe − βo
=
l,
2
βo =
βe =
Γ0
V+
φ+
After normalizing Equation 3.10 to DC, the frequency response of a TWE modulator
can be found:
1 + Γ0 ΓL V + + ΓL V − 2
H(ω) = 1 + ΓL eiβe l + Γ0 ΓL e−iβe l 41
(3.11)
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
Equation 3.11 address several important figure of merit for the frequency response. The
difference between βe and βo describes the phase velocity mismatch whereas α represents
the electrical propagation loss. The reflections, on the other hand, are included in Γ0
and ΓL to estimate the effects introduced by the variation of device impedance and load
impedance. To have a better understanding of the TWE modulator, the influences of
those three parameters are discussed below individually.
• Impedance Mismatch
To isolate the effect of impedance mismatch, we assume that ne = no and α = 0,
which results in V + = 1. In general, the system impedance Zg is fixed at 50
Ω so device impedance ZD and load impedance ZL are the only variables to
be considered. Figure 3.2(a) shows the frequency response of a 500 µm long
modulator with ZD = 35Ω. As can be seen, the frequency response is the best
when a matched load is used (ZL = 35Ω in this case), and can change several
dB for different terminations. Any mismatched ZL causes reflection, introduces a
counter propagating wave on the transmission line, and degrades the modulation
efficiency. For a 50 Ω termination, the response drops as frequency increases,
and so does the modulation power, which is proportional to Vavg . On the other
hand, for a termination smaller than ZD , the frequency response has an overshoot
at higher frequency, which means it helps to increase the bandwidth of a TWE
modulator, but may leads to pattern effects for large signal modulation. Moreover,
the modulation power is actually smaller at DC as shown in Figure 3.2(b) for small
termination. This indicates possible reduction of signal to noise ratio (SNR) for
large signal modulation as well. It is important to choose the proper value of ZL
for a optimal operation.
42
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
(a) 4
(b) 1.5
Z L =50Ω
3
Z L =20Ω
1
1
Z L =35Ω
0
Z L =35Ω
|Vavg|
H(dB)
2
0.5
Z L =20Ω
-1
Z L =50Ω
-2
0
10
20
30
Frequency (GHz)
0
40
0
10
20
30
Frequency (GHz)
40
Figure 3.2: (a)Frequency responses for different values of ZL when ZD = 35Ω.
(b)Average voltage as a function of frequency for different values of ZL .
• Velocity Mismatch
As mentioned previously, the modulation involves both optical and electrical signals. A optimal modulation should occurs when these two signal propagate at
identical phase velocities ( ωnc e =
ωno
c ).
To explore the influence of velocity mis-
match, the frequency response is calculated by assuming a lossless (α = 0) transmission line with matched impedances (Zg =ZD =ZL ), and the results are shown
in Figure 3.3(a). It is not surprising that the response drops more when the velocity difference is larger at a fixed frequency. However, the frequency drop due
to velocity mismatch is less than -1 dB even with large value of index difference
(∆n = ne − n0 = 1.55). This means that influence of velocity mismatch is limited because the modulator is only 500 µm. For such a short device, the velocity
mismatch is not as important as other two factors and can be ignored.
• Propagation Loss
43
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
(a)
(b) 0
0
ne =3
-1
ne =5
-0.4
-0.6
-3
-4
-0.8
-1
α=100dB/GHz/m
-2
H(dB)
H(dB)
-0.2
α=200dB/GHz/m
-5
0
10
20
30
Frequency (GHz)
-6
40
0
10
20
30
Frequency (GHz)
40
Figure 3.3: (a)Frequency response for different values of ne when no =3.45.
(b)Frequency response for different α.
The electrical propagation loss of a transmission line can be decomposed into
three major parts: metal loss, dielectric loss, and radiation loss. The first term is
due to the skin effect of a conductor. As the frequency increases, the skin depth
becomes smaller, results in pushing the electromagnetic waves towards the surface
of conductor, which introduces the metal loss. Since the metal loss is inversely
√
proportional to skin depth (Equation 3.12), then it is proportional to f .
r
δ=
2
ωµ0 σ
(3.12)
where ω is the frequency, µ0 is the permeability constant, and σ is the metal
conductivity. The dielectric loss, on the other hand, is not significant at low frequency but becomes the dominant term since it is proportional to the loss constant
(tan(δ) ∝ f ). The radiation loss, on the other hand, is relatively small compared
to other two terms and can be ignored. Figure 3.3(b) shows the frequency re-
44
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
sponse under the assumption of impedance and velocity match. As can be seen,
the 3 dB cutoff frequency is 30 GHz if the loss is 200 dB/GHz/m. The bandwidth
can be increased further to 60 GHz if the loss is reduced by half. In fact, for a
TWE modulator design, loss is the most important issue to be addressed and the
bandwidth could be improved by using a low-loss TWE.
3.1.2
Equivalent Circuit Model
So far the frequency response has been discussed on the circuit level by considering the
TWE modulator to be a section of transmission line with characteristic impedance ZD
and propagation constant γ. In this section, an equivalent circuit model is built up to
convert device structure dimensions and material parameters into a transmission line
model such that ZD and γ can be related to all the design variables.
Table 3.2: Variables of Figure 3.4(a). Material parameters and dimensions can be
found in 2.2.
Layer
width thickness permittivity resistivity
Signal line of CPW
Ground line of CPW
Metal via
P contact
P Cladding
Intrinsic (SCH+QW)
N contact
ws
wgnd
wv
wpc
wc
wi
wnc
dmp
dmn
dv
dpc
dclad
di
dnc
m
m
m
pc
clad
i
nc
1/σm
1/σm
1/σm
ρpc
ρclad
∞
ρnc
The device geometric cross section is shown in Figure 3.4(a). As can be seen, the
coplanar waveguide (CPW) is utilized because it has lower propagation loss than other
transmission line structures. This minimizes the loss from the probe to the device. In
addition, it is a metallic layer to physical access the p and n contacts of the modulator.
45
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
(a)
ws
y
wv
x
wgnd
wc
wt
wi
wg
dv
d pc
d clad
d mp
d i d mn
d nc
substrate
(b)
Zm
Z ml
Rc
Gl
Zu
Zt
Zi
Yu
Zb
Figure 3.4: (a)Cross section of a hybrid silicon CPW modulator and its dimensions. (b)Equivalent circuit model for a TWE device.
In this case, the signal line is applied to the top of the mesa, and the ground line is
applied to the n contact on both sides of the mesa without any complicated design.
However, it is important to notice that the electromagnetic field is actually similar to
microstrip line in this case since most of the electromagnetic field is concentrated in
the depletion region, i.e. active region (SCH+QW) [3]. An equivalent circuit model of
multilayer semiconductor has been developed [4] and good agreement with experimental
46
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
results has been obtained [5]. Here, we start from the developed circuit model with some
modification to fit our structure. An equivalent circuit model of the TWE modulator
is shown in Figure 3.4(b). The circuit can be divided into two major parts: impedance
Zu and admittance Yu . For a transmission line, the characteristic impedance ZD and
propagation constant γ can be given by:
r
ZD =
Zu
,
Yu
γ=
p
Zu Yu
(3.13)
All the variables are discussed individually as below.
• Rc (Ω-m) : Contact resistance
The contact resistance is from the annealed alloy at the interface between metal
and semicoductor. The contact resistances of p contact and n contact have to
be calculated separately since the metal stacks and the alloy are different. Here
we only consider the contact resistance based on the contact area. The transfer
length of the annealed alloy is included in Zb
Rc = ρspec,p ·
1
1
+ ρspec,n ·
wc
wgnd
(3.14)
where ρspec is the specific resistance.
• Zt (Ω-m) : Impedance between metal via and the intrinsic region
For a multilayer structure, the impedance of each layer (in the y direction) can
be described as a parallel circuit of a resistance and a capacitance. The total
impedance of the layers on top of the active region is then given by Equation
3.15.
Zt =
ρpc
1
1
ρclad
·
+
·
wpc 1 + i ωρpc pc wc 1 + i ωρclad clad
47
(3.15)
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
• Zi (Ω-m) : Impedance of the intrinsic region
The active region is totally depleted with applied external electrical field. Therefore, it can be described as a capacitance. Assume the device capacitance is
Ci =
a wi
di ,
then Zi is given in Equation 3.16
Zi =
1
i ωCi
(3.16)
• Zb (Ω-m) : Impedance between bottom of active region and n contact
Zb is similar to Zt , but mainly in the x direction in the thin n contact layer. The
total impedance has a factor of 1/2 because of the symmetrical mesa structure.
Zb =
wi /2 + wg + wt
ρnc
1
·
·
2 1 + i ωρnc nc
dnc
(3.17)
where wg is the gap between the edge of active region and n metal and wt is
the transfer length of a contact. wt can be obtained by transmission line method
(TLM) measurement.
• Zml (Ω/m) : Impedance of CPW inductance [Lm (H/m)]
Zm l = i ωLm
(3.18)
• Zm (Ω/m) : Metal impedance
For a metal line in the air, the impedance is described in Equation 3.19 by as-
48
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
suming the wave impedance of the air is much larger than that of the metal [4].
ηm
ηm
dm
Z=
· coth (1 + i)
w
δm
r
i ωµ0
= wave impedance =
σm
r
2
δm = skin depth =
ωµ0 σm
(3.19)
The total metal impedance can be found in Equation 3.20 where Zms and Zmv
are the impedance from signal line and via, respectively. Here we ignore the
impedance from the ground metal since it is much smaller than signal line
Zm = Zmv || Zms
(3.20)
• Gl (S-m) : Conductance of semiconductor multilayers
This term represents the longitudinal resistance along the propagation direction
introduced by doped semiconductor.
Gl =
wpc dpc wc dclad wnc dnc
+
+
ρpc
ρclad
ρnc
(3.21)
After obtain each variable, the overall impedance and admittance can therefore written
as:
Zu = (Zm + Zml ) ||
1
Gl
Yu = (Rc + Zt + Zi + Zb )−1 .
49
(3.22)
(3.23)
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
(a)
10
8
(b)
10
0
1/G l
6
1/Yu ( Ω-m)
Z u(Ω/m)
10
Z ml
10
4
10
10
Zi
-2
Zc
Zt
-4
Zb
Zm
10
2
0
10
20
30
10
40
-6
0
10
Frequency (GHz)
20
30
40
Frequency (GHz)
Figure 3.5: (a)Impedance of Zu . (b)Impedance of Yu .
6
Loss(dB/mm)
5
4
3
Zc
2
Zm+Zml
1
0
Zb+Zt
0
10
20
30
40
Frequency (GHz)
Figure 3.6: Attenuation constant due to different device components
3.1.3
Electrical Characteristics and Device Parameters
This section analyzes the influences of device parameters and physical dimensions based
on the theoretical model developed in previous sections. First, the impedance of each
50
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
component is shown in Figure 3.5. As can be seen, the impedance due to Gl is much
larger than the combination of Zm and Zml , which indicates the metal line is the dominant term in Zu . The impedance along the transverse direction is also shown in Figure 3.5(b). It is clear that most of the voltage drop occurs on the capacitance at low
frequency for a given bias across the device. The capacitance impedance, however, decreases at higher frequency such that it is less efficient to drive the device because some
of the energy dissipates in other layers. Among all the impedance contributed to Yu ,
the contact resistance Zc is the most important one since it is only two times smaller
than Zi at 40 GHz. If Zc is larger than Zi , then most of the voltage drop is wasted on
the contact resistance and results in frequency roll off. In addition, the propagation loss
due to different sources is analyzed to locate the maximum loss source. As can be seen
in Figure 3.6, the loss is decomposed into three parts by setting the real parts of the
others to zero. For example, loss from Zc is calculated by assuming that the real parts
of Zm , Zml , Zb , and Zt are zero. At low frequency, the loss from contact resistance (Zc )
and metal (Zm +Zml ) are about the same. However, the loss from Zc is larger than the
other two when the frequency is over 15 GHz. At higher frequency, Zc becomes the
major loss source and introduces around 2.5 dB loss for a 500 µm modulator at 40 GHz.
This indicates the modulation bandwidth is significantly affected by the contact resistance and should be reduced with improved Zc by optimizing the annealing conditions
(temperature, time, and metal stacks).
The effects of different values of p contact resistance (ρspec,p ) and n contact resistance
(ρspec,n ) are analyzed in Figure 3.7 and Figure 3.8, respectively. The parameters used
for simulation are listed in Table 3.3. As expected, the loss increases as ρspec,p increases
while the device impedance does not change too much over 40 GHz range. A bad
anneal of the p contact can result in huge loss difference and consequently decrease the
51
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
Table 3.3: Parameters used in Figure 3.7 to Figure 3.12.
Layer
Signal line of CPW
Ground line of CPW
Metal via
P contact
P Cladding
Intrinsic (SCH+QW)
N contact
SU8
width(µm)
thickness(µm)
ws =10
wgnd =16
wv =2.5
wpc =4
wc =4
wi =2
wnc =20
–
dm =2
dm =2
dv =3
dpc =0.1
dclad =1.5
di =0.3
dnc =0.15
5
r
resistivity(Ω-m)
–
1/σm
–
1/σm
–
1/σm
13.91 ρpc = 1.47 × 10−5
12.50 ρclad = 4.17 × 10−5
13.43
–
12.50 ρnc = 3.88 × 10−5
2.80
–
1. σm = 4 × 107
2. ρspec,p = 6 × 10−5 , ρspec,n = 6 × 10−6
3. The effective index of the optical mode is 3.45.
modulation bandwidth. The anneal temperature, however, is limited to middle of the
300-400 ◦ C range due to the low-temperature bonding process such that the best ρspec,n
is about 1x10−5 on the hybrid silicon platform. On the other hand, the influence from
ρspec,n is much smaller. Figure 3.8 shows that the ZD , α, and bandwidth barely change
within two orders of magnitude difference of ρspec,n because the contact area of n (wgnd )
is much larger than that of p (wc ) regardless of the transfer length (wt ).
The physical dimensions of the device are also of great interest for modulation
bandwidth. To explore the effect of individual dimensions, some of the most important
ones are analyzed here. First, the electrical parameters with different intrinsic region
widths (wi ) from 1 µm to 4 µm are shown in Figure 3.9. By reducing the intrinsic
region width by four times, the device impedance increases from 15 Ω to 30 Ω due to
huge reduction of the device capacitance, which also results in 8 dB/mm loss difference
at 30 GHz. Consequently, the 3 dB cutoff frequency is expected to change from 4.7
GHz to 20.5 GHz when wi is varied from 4 µm to 1 µm. However, the tradeoff between
52
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
50
50
40
2
ρ =6E-6 -> 6E-4 Ω-cm
Impedance( Ω)
Impedance( Ω)
40
cp
30
20
10
0
-10
30
20
2
ρ =6E-7 -> 6E-5 Ω-cm
10
cn
0
0
10
20
30
-10
40
0
Frequency (GHz)
10
20
30
40
10
6
2
8
Loss(dB/mm)
ρcp= 6E-6 Ω-cm
8
Loss(dB/mm)
10
Frequency (GHz)
6E-5 Ω-cm 2
4
2
2
ρcn=6E-7 -> 6E-5 Ω-cm
6
4
2
6E-4 Ω-cm 2
0
0
10
20
30
0
40
Frequency (GHz)
20
30
40
0
2
ρcp=6E-6 -> 6E-4 Ω-cm
-2
-2
2
-4
S21 (dB)
S21 (dB)
10
Frequency (GHz)
0
-6
-8
-10
0
ρ =6E-7 -> 6E-5 Ω-cm
-4
cn
-6
-8
10
20
30
-10
40
Frequency (GHz)
10
20
30
40
Frequency (GHz)
Figure 3.7: Transmission line characteristics as a function of p contact resistance.
53
Figure 3.8: Transmission line characteristics as a function of n contact resistance.
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
50
50
40
wi=1µm
30
2µm
20
3µm
4µm
10
0
-10
Impedance( Ω)
Impedance( Ω)
40
10
20
30
w c= 2, 4, 6, 8µm
20
10
0
1µm -> 4µm
0
30
-10
40
w c= 2, 4, 6, 8µm
0
Frequency (GHz)
10
Loss(dB/mm)
Loss(dB/mm)
6
40
2µm
4
2
w i =1µm
0
10
20
30
6µm
4
8µm
2
0
40
4µm
6
Frequency (GHz)
0
10
20
30
40
Frequency (GHz)
0
0
-2
-2
w i =1µm
-4
S21 (dB)
S21 (dB)
30
w c =2µm
8
3µm
-6
4µm
3µm
-8
-10
20
10
w i =4µm
8
0
10
Frequency (GHz)
10
2µm
20
w c= 2, 4, 6, 8µm
-4
-6
-8
30
-10
40
Frequency (GHz)
10
20
30
40
Frequency (GHz)
Figure 3.9: Transmission line characteristics as a function of intrinsic region
width.
54
Figure 3.10: Transmission line characteristics as a function of mesa width.
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
50
30
20
10
0
-10
1E17cm-3
1E18cm-3
1E19cm-3
40
Impedance( Ω)
40
Impedance( Ω)
50
dm=0.1 µm
dm=1 µm
dm=10 µm
30
20
10
0
0
10
20
30
-10
40
0
Frequency (GHz)
10
dm=0.1 µm
dm=1 µm
dm=10 µm
6
4
2
0
10
20
30
4
2
0
40
0
10
20
30
40
Frequency (GHz)
0
0
dm=0.1 µm
dm=1 µm
dm=10 µm
-4
1E17cm-3
1E18cm-3
1E19cm-3
-2
S21 (dB)
-2
S21 (dB)
40
6
Frequency (GHz)
-6
-8
-10
30
1E17cm-3
1E18cm-3
1E19cm-3
8
Loss(dB/mm)
Loss(dB/mm)
8
20
Frequency (GHz)
10
0
10
-4
-6
-8
10
20
30
-10
40
Frequency (GHz)
10
20
30
40
Frequency (GHz)
Figure 3.11: Transmission line characteristics as a function of probe metal thickness.
55
Figure 3.12: Transmission line characteristics as a function of doping concentration of cladding layer.
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
optical loss, optical modulation efficiency, and the bandwidth makes it impractical to
have a extremely small wi . Both optical loss and voltage-length product increase when
wi becomes smaller. In addition, mechanical instability becomes an issue at small wi
and the mesa tends to collapse easily during processing if wc is much wider than wi .
Figure 3.10 shows the transmission line characteristics at different wc . It shows that wc
does not have strong influence on bandwidth and impedance, and can have large impact
on loss constant. Therefore, having a device with small wc can benefit the bandwidth.
Based on the lithographical capability at UCSB nanofab, a wc of 4 µm is chosen to keep
fabrication feasible.
In addition to wc and wi , the influence of probe pad thickness is also of interest
because it is the second major source of loss as mentioned in previous section. As can
be seen in Figure 3.11, all the electrical characteristics are similar once dm is larger
than 1 µm because dm is larger than the skin depth (less than 0.4 µm over 7 GHz).
Once the metal thickness is larger than the skin depth, the overall field penetration
into the conductor is fixed. In contrast, not all the electrical signals are transmitted
via the conductor if dm smaller than the skin depth, and this results in a large loss.
For our hybrid silicon modulator, we chose dm =2µm to avoid large loss caused by field
penetration.
Another interesting parameter is the doping concentration in the cladding layer. The
cladding layer is 1.5 µm thick to avoid overlap between metal pad and the optical mode
as mentioned previously. Therefore, the electrical signal penetrates into this layer due
to shallow skin depth. The electrical properties based on different doping concentration
are shown in Figure 3.12. It can be seen that the doping concentration does not affect
the bandwidth too much from 1E17 cm−3 to 1E19 cm−3 . This gives more freedom to
design the epitaxial layers in terms of optical design. In general, the doping is around
56
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
1E18 as a compromise between resistance and optical loss.
3.2
Fabrication
The hybrid silicon device fabrication flow consists of silicon waveguide formation, low
temperature wafer bonding and post-bonding fabrication.
In order to have good optical confinement, the silicon waveguides are fabricated on
silicon on insulation (SOI) wafers, which are prepared in the oxidation furnace at 1050◦ C
to control the top silicon layer thickness at 0.5 µm and form a 150 nm thick oxide layer
as a hard mask. This thermal oxide is then etched to create waveguide patterns using a
CHF3 inductively coupled plasma (ICP) etch. Immediately after the mask is formed, the
silicon waveguides are etched half way through the silicon thickness in the same chamber
using Cl2 /BCl3 gas combination. In order to have less cross talk between two arms of
the MZI, the silicon waveguides around the MMIs are further etched down to the buried
oxide layer. Using the same hard mask left from silicon waveguide etch can eliminate
an extra alignment step. Figure 3.13 shows the top view of fabricated deep etch. The
final step of silicon preparation is to create vertical outgassing channels (VOCs) [6],
which assists in quenching hydrogen outgassing during bonding. First, the thermal
oxide mask is removed in buffered HF to create a fresh surface. One has to be careful
during this step since the buried oxide layer is exposed and can be removed by buffered
hydrofluoric acid (HF) as well. This step is executed with care such that the oxide
underneath silicon waveguides is not totally removed and results in suspended silicon
waveguides with instable mechanical support. Next, a 400 nm thick plasma-enhanced
chemical vapor deposition (PECVD) silicon dioxide (SiO2 ) hard mask is deposited on
the sample surface. This SiO2 can protect the silicon waveguides during the VOCs etch
57
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
20μm
VOC
III-V and Si tapers
deep etch Si
Figure 3.13: The transition between passive and hybrid sections.
process. An array of of 8x8 µm VOCs with 50 µm spacing (Figure 3.13) is then etched
using identical chemistry for silicon waveguide etch. The sample is now ready for the
bonding process.
The surface condition of both silicon and III-V epitaxial wafers is very important for
a successful bond. Several processing steps are taken prior to physical contact between
those two chips. In order to remove organic residue and particles on top of the sample
surface, the samples are spun and swabbed under a spray rinse of ACE/ISO/H2 O. The
silicon wafer is then dipped in Piranha solution (H2 SO4 :H2 O2 =3:1) to further etch any
possible organic contamination. Both samples are inspected carefully under an optical
microscope before removal of oxide layers on top of the surfaces. The samples are then
treated with oxygen plasma and NH4 OH to create OH− bonds on the surfaces. With a
performance of physical contact, a spontaneous bond is formed by connecting the OH−
bonds. This bond is strong enough to hold both pieces in contact unless large external
force is applied to the interface. The sample is then placed into a special fixture to
apply necessary pressure and is baked at 300◦ C for one hour. After bonding, the thick
substrate of InP needs to be removed prior any post-bond processing. The sample is
mounted on a carrier wafer surrounded by crystal bond. Next, the substrate thickness
58
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
(1) Bonding
(2) Mesa etch
p contact
Cladding
SiO2
Silicon
(3) QW etch
(4) n metal deposition
n contact
QW/SCH
n metal
(5) SU8 passivation
(6) Probe metal deposition
Probe metal
SU8 polymer
SU8 polymer
p contact
QW/SCH
Cladding
n contact
n metal
SiO2
Silicon
Figure 3.14: Process flow for a hybrid silicon modulator.
is reduced from 360 µm to around 100 µm by an automatic lapping tool. The sample is
then dipped into a diluted hydrochloric acid (HCL) solution (HCL:H2 O=3:1) for about
10 minutes to completely remove the substrate while the p-contact layer is used as a stop
etch layer. The crystal bond is left on the sides of the wafer to avoid lateral etch of the
bonded material, and hence increase the bonding yield. The cross section of the device
59
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
after substrate removal is shown in Figure 3.14 as step (1). The mesa structure is then
fabricated using a self-aligned dry etch process. Figure 3.14 shows the schematic cross
section in each process step. First, a stack of Pd/Ti/Pd/Au p-contacts and a thin layer
of silicon nitride is used as the hard mask for reactive ion etch (RIE) to form a 4 µm
wide mesa (step(2)). A solution with H3 PO4 :H2 O2 :H2 O=1:5:15 is applied to the sample
to create the undercut while several circular patterns with different radii were laid out
on the mask to monitor and control the distance of the undercut (step(3)). The sample
is then dipped into diluted HCL (HCL:H2 O=1:10) to remove native indium oxide on the
sidewall of the QW/SCH layers to avoid current leakage prior to deposit silicon nitride
for passivation. After forming n-contacts with Ni/Au/Ge/Ni/Au metal stack (step(4)),
a 30 second 360◦ C precise anneal in rapid thermal annealer (RTA) is applied to form
an alloy between the metal stack and semiconductor. A 5 µm thick SU8 is coated and
lithographically defined to provide additional mechanical support for the thin bonding
layer and to keep the traveling wave electrode away from the underlying ground in order
to reduce the impedance mismatch and to minimize parasitic capacitances (step(5)).
Finally, a 2 µm thick layer of Ti/Au is deposited using Ebeam evaporator to form the
transmission line structures and contact pads (step(6)).
3.3
Device Characterization
The top view of a 500 µm long MZM is shown in Figure 3.15(a). It has two MMIs, each
6 µm wide and 40 µm long, at the input and output functioning as the splitter and the
combiner. Due to the mode mismatch between the passive and hybrid sections, two 60
µm long tapers are added to minimize reflection and increase coupling efficiency. An
optical image of a fabricated device is also shown in Figure 3.15(b). A thin layer of
60
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
A
(a)
(c)
Probe metal
SU8 polymer
A‘
QW/SCH Cladding
(b)
n contact
n metal
SiO
Si
Oxide 2
Figure 3.15: (a)Top view of a device with a CPW electrode. (b)The optical
image of the device under microscope. (c)Cross section (along A-A’) of the hybrid
waveguide.
silicon nitride, shown as the orange cross region in the figure, is deposited at the end
of the process to protect the device from scratches. Figure 3.15(c) illustrates the cross
section of the hybrid section, which is similar to the cross section mentioned in Section
3.2, but with two arms. The device has a 4 µm cladding width while the QW/SCH
layers are under-cut to 2 µm to reduce the device capacitance. Moreover, the two arms
of the MZM are electrically isolated by etching through the n-contact layer between
them. The silicon waveguides have a height of 0.46 µm, a slab height of 0.19 µm, and
a width of 0.94 µm.
3.3.1
Static Characteristics
The devices are first anti-reflection coated to eliminate the undesired resonance and
reduce the reflection on both facets. Two lensed fibers were then used to couple the
light in and out of the silicon waveguides. The experimental results of transmission as
a function of reverse bias with different input optical power is shown in Figure 3.16. As
61
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
1
Normalized Transmission
0.9
0.8
0.7
0.6
−7.5dBm
2.5dBm
12.5dBm
0.5
0.4
0.3
0.2
0.1
0
0
1
2
3
4
5
6
Reverse Bias (V)
Figure 3.16: Modulation efficiency of a hybrid silicon modulator with different
input power at 1540nm.
can be seen, all three curves have a highest transmission at bias other than zero due
to the additional phase change introduced by the tapers. The voltage-length product
decreases from 1.95 V-mm to 1 V-mm as the input optical power increases from -7.5 dBm
to 12.5 dBm. The reduction in voltage-length product can be attributed to the excess
carriers generated at higher optical intensity. The shift in the highest transmission point
to higher reverse bias voltages also indicates the existence of the extra carriers due to
two photon absorption (TPA), where stronger electrical field is required to completely
deplete the QW/SCH layers. The extinction ratios (ER) are 12.77 dB, 11.51 dB and
8.64 dB for input powers of 12.5 dBm, 2.5 dBm, and -7.5 dBm, respectively. The ER
is around 10 dB only because the silicon waveguides were not etched all the way to the
BOX layer around the MMIs in these particular devices. It can be further improved
either by deeply etching the silicon or by using a push-pull design to reduce the loss
imbalance between the two arms.
62
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
40
2
Cd (pF/mm)
Z D (Ω)
30
real
20
10
imaginary
1.5
1
0
-10
0
10
20
30
0.5
40
0
20
20
15
15
10
5
0
20
30
40
Frequency (GHz)
Index
Loss(dB/mm)
Frequency (GHz)
10
10
5
0
10
20
30
0
40
0
Frequency (GHz)
10
20
30
40
Frequency (GHz)
Figure 3.17: Transmission line characteristics of a 500 µm hybrid silicon modulator
with CPW electrode. The black lines are the measured data and the green lines
are calculated based on 2 µm wide QW.
3.3.2
High Frequency Performance
To investigate the high-speed performance of the devices, the electrical properties of the
transmission line is first characterized using an Agilent 8164A PNA network analyzer.
Device impedance (ZD ), electrical propagation loss (α), phase velocity (βe ) and other
parameters are extracted from Equation 3.24 by converting measured S parameters to
63
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
an ABCD matrix.
2B
p
D − A ± (A + D)2 − 4


s
2
1
A+D
A+D
−1 
γ = α + i β = ln 
±
L
2
2
ZD =
(3.24)
Comparison between the experimental data and the simulation results is shown in Figure 3.17 where simulation results are obtained by substituting the physical dimensions
and variables into theoretical model established in Section 3.1. The device impedance
is around 18 Ω and agrees with the simulation. The measured device capacitance Cd ,
however, is slightly larger than the capacitance value from a 2 µm QW region due to
0.325 pF parasitic capacitance. This also results in a larger value of propagation loss
compared to the theoretical estimation shown in Figure 3.9.The small difference of loss
constant between experimental data and simulation is possibly due to metal roughness
introduced by uneven SU8 surface. For a 500 µm modulator, the loss is around 2.5 dB at
10 GHz and 4 dB at 20 GHz. This predicts that the bandwidth of a CPW hybrid silicon
modulator is not larger than 10 GHz. Overall, the measured electrical characteristics
are within the expectation from our circuit model.
Next, the device electrical-optical responses are measured using HP 8703A Lightwave component analyzer. As shown in Figure 3.18, the device without any impedance
termination has a 3 dB cutoff frequency at 3.5 GHz (green curve) and the simulation
fits with the response well. The bandwidth, however, can be improved by putting a
termination at the end of the CPW electrode to reduce the reflection. Here we use a
customized probe with 50 Ω impedance at the tip and terminated with 50 Ω cap. The
response (black curve) together with its simulation (black dash line) are illustrated in
64
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
Normalized S21 (dBe)
2
0
Z L =25 Ω
-2
-4
Z L =open
-6
-8
-10
1
10
Frequency (GHz)
Figure 3.18: EO frequency response and simulation with different termination.
Figure 3.18 and the 3 dB bandwidth increases to 8 GHz, which should be sufficient for
10 Gb/s data transmission. The overshoot in the frequency response is caused by an
extra inductance from an imperfect termination probe and could be eliminated with
further adjustment.
After the frequency response of the devices is determined, the MZM is then tested
with large signal modulation to explore its potential for high speed communication. A
231 -1 pseudorandom bit sequence (PRBS) pattern generator connected to an electrical
amplifier is used to provide the drive signal. The device is biased at -3.8 V with 1.5
V swing while the bias at other arm is adjusted to achieve best signal quality. The 10
Gb/s modulated light is then collected by a lensed fiber and amplified with an EDFA
following by a filter to eliminate the ASE noise before the signal is sent to an Agilent
digital communication analyzer (DCA). The signal has an ER of 6.3 dB, as shown in
Figure 3.19, which is smaller than the ER (11.6 dB) measured at DC bias due to the
65
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
10Gb/s
Figure 3.19: EO frequency response and simulation with different termination.
partial voltage drop across the series resistance and cladding layer. The eye is clearly
open and sufficient for 10 Gb/s operation except for the noisy one level due to the
frequency overshoot at 1.4 GHz.
3.4
Summary
In this chapter, we first developed the circuit model for a traveling-wave electrode device, estimated the transmission line characteristics, and derived the frequency response
based on this model. Some important parameters and figures of merit for the modulation bandwidth are discussed to improve the device high-speed performance. Next,
the fabrication on the hybrid silicon platform is illustrated and introduced. Critical
processing techniques are addressed to reach the balance between device performance
and equipment capability. Finally, we discussed the device static and high-speed performance and compared with the theoretical model. A high speed Mach-Zehnder silicon
66
Chapter 3. Coplanar Waveguide TWE Mach-Zehnder Modulator
evanescent modulator is demonstrated utilizing the carrier depletion effect inside the
offset MQW. The modulator has a modulation efficiency of 2 V-mm, a modulation
bandwidth of 8 GHz, and clear open eye diagram at 10 Gb/s with 6.3 dB ER.
The high-speed performance of this CPW modulator, however, is not enough for
current demands on data transmission. A modulator with higher bandwidth and data
rate needs to be developed to fulfill the need for high capacity transmission. Therefore,
a better electrode design to reduce propagation loss, increase device impedance and
reduce reflection at the interfaces is important. The details of a more advanced TWE
design will be discussed in next chapter to enable data transmission up to 40 Gb/s.
67
Bibliography
[1] I. Kim, M. R. T. Tan, and S.-Y. Wang, “Analysis of a new microwave low-loss
and velocity-matched III-V transmission line for traveling-wave electrooptic modulators,” IEEE/OSA Journal of Lightwave Technology, vol. 8, no. 5, pp. 728–738,
1990.
[2] S. H. Lin and S.-Y. Wang, “High-throughput GaAs PIN electrooptic modulator with
a 3-dB bandwidth of 9.6 GHz at 1.3 µm,” Appl. Opt., vol. 26, no. 9, pp. 1696–1700,
May 1987. [Online]. Available: http://ao.osa.org/abstract.cfm?URI=ao-26-9-1696
[3] S. Zhang, “Traveling-wave Electroabsorption Modulators,” Ph.D. dissertation, University of California, Santa Barbara, 1999.
[4] H. Guckel, P. A. Brennan, and I. Palocz, “A Parallel-Plate Waveguide Approach
to Microminiaturized, Planar Transmission Lines for Integrated Circuits,” IEEE
Transactions on Microwave Theory and Techniques, vol. 15, no. 8, pp. 468–476,
1967.
[5] K. S. Giboney, “Traveling-wave Photodetectors,” Ph.D. dissertation, University of
California, Santa Barbara, 1995.
[6] D. Liang and J. E. Bowers, “Highly efficient vertical outgassing channels for lowtemperature InP-to-silicon direct wafer bonding on the silicon-on-insulator substrate,” Journal of Vacuum Science Technology B: Microelectronics and Nanometer
Structures, vol. 26, no. 4, pp. 1560 –1568, July 2008.
68
Chapter 4
Capacitively Loaded TWE MZM
In this chapter, we will focus on an improved TWE design for the hybrid silicon modulator using the carrier depletion effect. A novel capacitively loaded TWE and its design
details will be discussed in Section 4.1. Fabrication adjustments for this specific TWE
design are presented in Section 4.2. Finally, Section 4.3 includes the experimental results
and a comparison to a theoretical model.
4.1
Capacitively Loaded TWE Design
4.1.1
Push-Pull Structure: Slotline v.s. CPW
Operating modulators in a push-pull configuration is one of the most common methods used to improve modulator performance. It was first proposed to reduce the drive
voltage while maintaining similar modulation depth [1]. Integrated devices using this
structure to reduce the footprint by decreasing the voltage-length product [2, 3] were
also demonstrated. In addition, studies have shown chirp reduction due to the differ-
69
Chapter 4. Capacitively Loaded TWE MZM
RF signal
Arm1
RF signal
Arm2
DC bias
point
Figure 4.1: Bias condition for a push-pull structure.
ential drive signal [4] applied to the two arms of the MZI. Hence, one can improve the
modulation efficiency and chirp of a modulator with appropriate device/structure design. In order to apply the push-pull design to our hybrid silicon modulator, we need to
understand the bias condition (both DC and RF) of a push-pull structure and choose a
proper electrode design to fit the scheme. The bias condition of a conventional push-pull
structure is illustrated Figure 4.1. The basic idea of a push-pull structure is that two
arms in the MZI have identical DC bias, but opposite RF drive signal. Therefore, the
drive voltage can be reduced to half and the nonlinear phase change can be canceled
due to the differential drive.
For our hybrid silicon modulator, the CPW (ground-signal-ground) TWE cannot
satisfy the bias condition for push-pull. As can be seen in Figure 4.2(a), if the n side
of the diode from two arms of the MZI are connected, then the DC bias is identical for
two arms. However, the RF drive signal is identical as well, which indicates that there
is no modulation since two arms have the same phase change. In contrast, if one of the
diodes is flipped (Figure 4.2(b)) and the n side of one diode is connected to the p side
of the other, then the RF drive signal has the same amplitude but opposite sign. This
70
Chapter 4. Capacitively Loaded TWE MZM
(a) G
(b) G
vRF
-Vb
S
-vRF
-Vb
vRF
Vb
S
vRF
-Vb
G
G
(c) S
vRF
-Vb
2
vRF
-Vb
2
Vb
G
Figure 4.2: Circuit diagram for (a)CPW electrode with common ground. (b)CPW
electrode without common ground. (c)Slotline electrode
is the common circuit model for most electrooptic modulators. However, this scheme
requires forward bias in one arm. This will result in very slow frequency response in our
case because the speed of carrier injection effect is limited by carrier life time (around
ns).
A different TWE design shown in Figure 4.2(c), is introduced to fulfill the requirements for high-speed and a push-pull circuit. Unlike the CPW, the slotline only has
on signal line and one ground line, which provides the flexibility to separate DC and
RF bias. As can be seen, the circuit is similar to Figure 4.2(a) where two arms have
a common DC ground. However, the RF ground is only connected to one side of the
diode such that differential RF drive is applied to both arms. With this scheme, the
71
Chapter 4. Capacitively Loaded TWE MZM
(a)
DC bias pad
A
1
3
2
4
B
Ls
Lp
A’
(b)
Loaded Section(A-A’)
SU8 polymer
B’
Unloaded Section(B-B’)
Probe metal
Implanted mesa
p cladding
n contact
QW/SCH
SiO2
Si
Figure 4.3: (a)Top view of a device with a CL slotline electrode (b)Cross section of
loaded (along A-A’) and unloaded sections (along B-B’) of the hybrid waveguide.
diodes of two arms have the same DC bias condition while maintaining opposite drive
voltage at high speed. In contrast to conventional push-pull structure, this scheme is
different in two key ways. First, the total device capacitance is reduced by half because
two diodes are in series. This will directly benefit the frequency response and increase
the bandwidth. Second, the voltage-length product does not decrease for this scheme
since the RF voltage drops evenly in both arms.
4.1.2
Capacitively Loaded TWE Circuit Model
In general, a conventional TWE such as the CPW TWE in Chapter 3 is good enough
for high-speed operation up to 10 Gb/s. However, for a continuous TWE incorporating
a pin diode, the structure is similar to a microstrip line rather than the TWE itself
because the electric fields propagate inside the depletion region of the diode instead of
residing in between the gaps of signal and ground electrodes. Moreover, the electric
72
Chapter 4. Capacitively Loaded TWE MZM
fields penetrate the top p cladding and bottom n contact layers depending on the skin
depth. The skin depth of the doped epitaxial layers can be still around several µm at
40 GHz due to the small conductivity of the semiconductors so one would expect to see
large propagation loss of the driven electrical signal. In order to reduce the propagation
loss, it is important to ensure that the propagating electric fields have minimal overlap
with the doped semiconductor. One of the options that has been widely demonstrated
for high speed operation is a capacitively loaded (CL) TWE [5, 6]. As illustrated in
Figure 4.3(a), the small pads extending from the transmission line can provide the
necessary electrical signal to drive the device while the TWE is kept away from the
semiconductor. The loaded sections (small pads) are capacitive if the Bragg frequency
of each periodic section (Lp ) is much larger than the highest frequency of interest.
Furthermore, the overall device impedance and the phase velocity of the electrical signal
can be adjusted by changing the distributed capacitance of the transmission line. This
can help reduce velocity mismatch between the electrical and optical signal and also
the reflection at interfaces. As mentioned in Section 4.1.1, we adopt a slotline TWE
to implement a push-pull structure [7]. The electrical signal travels along the slotline
and the bias voltage is applied using a probe pad connected to the n-contact layer.
One important parameter of this design, the fill factor (F), is defined as Ls /Lp in
Figure 4.3(a), where Ls is the length of each loaded section and Lp is the periodical
length. The total modulation length (La = Ls × N ) is the summation of each individual
loaded section, where N is the number of loaded sections. One can also specify the total
electrode length (Lt ) as Lp × N .
To better explain the design details, the cross section of a CL TWE is shown in
Figure 4.4(a), where definitions of all the dimensions are listed in Table 3.2 except wgap ,
which is the distance between two arms. Basically, the circuit model is a combination
73
Chapter 4. Capacitively Loaded TWE MZM
(a)
d mp
y
ws
wv
dv
d pc
d clad
wc
x
wi
wgap
di
d nc
substrate
(b)
Rm
Lm
Gm
(c)
Zm
Rm
Cm
Z ml
Zdiode
Lm
Gm
Cm
Zdiode
Z implnt
Rc
Zu
Zt
Yu
Ci
Cm
Zb
Ci
Zt
Rc
Figure 4.4: (a)Cross section of a hybrid silicon CL TWE modulator and its dimensions. (b)Distributed transmission line circuit. (c)Distributed equivalent circuit
model per unit length.
74
Chapter 4. Capacitively Loaded TWE MZM
of a TWE plus a loaded diode as shown in Figure 4.4(b). Since the transmission line
is partially in contact to the mesa, the fill factor has to be added to all the parameters
related to the diodes. In contrast, parameters related to the transmission line itself can
be kept as in Chapter 3. The details of each variable in Figure 4.4(c) is explained as
follows:
• Rc (Ω-m) : Contact resistance
The contact resistance slightly different than the CPW design since there is no n
contact resistance between signal and ground line.
Rc = F · ρspec,p ·
1
wc
(4.1)
where ρspec is the specific resistance of the metal-semiconductor contact.
• Zt (Ω-m) : Impedance between metal via and the intrinsic region
ρpc
1
1
ρclad
Zt = F ·
·
+
·
wpc 1 + i ωρpc pc wclad 1 + i ωρclad clad
(4.2)
• Zi (Ω-m) : Impedance of the intrinsic region
Ci = F ·
a wi
,
di
Zi =
1
i ωCi
(4.3)
• Zb (Ω-m) : Impedance in the n contact layer between the two arms
Zb =
wgap
ρnc
·
1 + i ωρnc nc dnc
where wgap is the gap between the two arms.
75
(4.4)
Chapter 4. Capacitively Loaded TWE MZM
• Zmc (Ω-m) : Impedance due to the transmission line capacitance
This is the transmission line capacitance (Cm ) without any III-V contact.
Zmc =
1
i ωCm
(4.5)
• Zml (Ω/m) : Impedance due to CL inductance [Lm (H/m)]
Zm l = i ωLm
(4.6)
• Zm (Ω/m) : Metal impedance
ηm
dm
Z=
· coth (1 + i)
w
δm
(4.7)
Zm = Zmv || Zms
(4.8)
• Zimplant (Ω/m) : Impedance from implant region
This term represents loss introduced by the implant region, where k is an empirical
constant.
Zimplant = k · ω · F
(4.9)
After obtaining each variable, the overall impedance and admittance can be written as:
Zu = Zm + Zml + Zimplant
(4.10)
Zdiode = 2 × (Rc + Zt + Zi ) + Zb
Yu = [(Zdiode + Zb ) || Zmc ]−1
76
(4.11)
Chapter 4. Capacitively Loaded TWE MZM
ws
G
50 Ω
feed-in electrode
For G in = 60 µm
Lm (nH/m)
ws (µm)
500
240
600
135
700
78
ZD
Figure 4.5: Top view of a CL slotline.
Similarly, the frequency response of the device can be obtained by substituting
Equation 4.10 into Equation 3.11.
For all the device parameters in CL slotline design, the gap and electrode width
are the most important ones. Figure 4.5 illustrates the top view of a CL slotline with
a feed-in electrode. The 50 Ω feed-in electrode has a gap of 9 µm and a width of 162
µm. In order to reduce the reflection at the interface between feed-in electrode and the
device electrode, the width difference has to be minimized. However, change of electrode
width results in different value of transmission line inductance (Lm ) and hence affects
the frequency response. In addition, the fill factor is an important parameter because
it alters the phase velocity, device impedance and loss constant. A proper choice of
fill factor can optimize the modulation bandwidth. Here, the influence of Lm and fill
factor are shown in Figure 4.6 with a fixed length of active region La . It is clear that
for a given Lm , ZD is closer to 50 Ω as the fill factor decreases (Figure 4.6(a)). This
indicates that the capacitance from the diode is distributed over the total length of
the slotline, and average capacitance (capacitance per unit length) decreases as the fill
factor decreases. Consequently the overall device impedance is higher. However, the
77
Chapter 4. Capacitively Loaded TWE MZM
50
-4
Lm=700nH/m
Z D (Ω)
45
Loss at 20 GHz(dB)
Lm=600nH/m
Lm=700nH/m
40
35
-4.5
Lm=500nH/m
-5
Lm=600nH/m
-5.5
Lm=500nH/m
30
0.2
0.4
0.6
0.8
-6
0.2
1
Fill factor
0.4
0.6
0.8
1
Fill factor
Figure 4.6: (a)Device Impedance as a function of fill factor at different Lm .
(b)Frequency response at 20 GHz as a function of fill factor at different Lm .
frequency response does not follow this trend as shown in Figure 4.6(b). As the fill factor
decreases, the total device length increases. Therefore the total loss increases as well.
This figure shows that a device with higher fill factor is better for higher bandwidth.
The fill factor, however, cannot equal to one because the electric fields will propagates
on top of the mesa and introduce an extra loss term which is not included in our circuit
model. On the other hand, the benefit of choosing a larger value of Lm based on ZD
is also obvious as shown in Figure 4.6. This leads to higher width difference between
feed-in electrode and the CL slotline, though, which is not included in our circuit model
either. Therefore, we choose the slotline width of 135 µm as a compromise between
device impedance, frequency response, footprint and fabrication tolerance. Ls is 100
µm such that Bragg frequency of each periodic section (Lp ) is much larger than the
highest frequency of interest.
78
Chapter 4. Capacitively Loaded TWE MZM
Loaded Section
Unloaded Section
(5) Pronton Implant
n contact
n contact
n metal
n metal
(5)
(5)
(6) SU8 coating and etch
SU8 polymer
(7) Probe metal plating
Probe metal
SU8 polymer
p contact
QW/SCH
Cladding
n contact
n metal
SiO2
Silicon
Figure 4.7: Modified process flow for CL slotline modulator.
4.2
Fabrication Improvements
The fabrication of CL slotline modulator is very similar to that of CPW modulator
described in Section 3.2, but with different steps and etch technique after step (4). As
79
Chapter 4. Capacitively Loaded TWE MZM
illustrated in Figure 4.7, the sample is first implanted with protons to create the isolation
between each loaded section by decreasing the conductivity of both the p-contact and the
p-cladding layers on the unloaded sections. The order of the anneal and implant steps is
important. If the anneal is performed after implant, the high-temperature anneal leads
to drift of ions, an increase of conductivity of implanted sections, and results in weak
isolation [8]. Hence, to keep the isolation high after implant, the contact anneal has to
be done before the implant to avoid any drifting of ions. In addition, the top p contact
layer is removed in the unloaded sections for better isolation. The impedance of a 10
µm isolation region was 60 MΩ.
Utilization of polymers such as SU8 and BCB in TWE fabrication is a very common approach to eliminate parasitic capacitances and allow the freedom to design high
performance transmission lines. However, polymers are well known to have limited
lithographic resolution for small patterns. Via openings less than 5 µm are generally
difficult to define and the minimum feature size increases with polymer thickness. In
addition, this resolution is also highly dependent on topography and the material on
the chip. For the hybrid silicon MZM, the mesa width is only 4 µm, which makes it
challenging to have precise polymer via definition which is also aligned correctly. The
via shape slightly changes during the polymer cure process such that the width might be
larger than the mesa and additional parasitic capacitance can be introduced. Therefore,
it is important to have an established process that is easily repeated. Instead of lithographically defining the patterns, an inductively coupled plasma (ICP) etch is utilized
to avoid unpredictable width variation. The sample was first covered with SU8 and
cured at 240◦ C followed by a deposition of a 300 nm PECVD silicon nitride (SiN) hard
mask at 240◦ C. The pattern was then defined in Panasonic ICP by etching the hard
mask. Next, a 5 µm deep via was created with a low-power O2/CF4 ICP etch for 15
80
Chapter 4. Capacitively Loaded TWE MZM
(a)
(b)
Figure 4.8: (a)A cross section of the SU8 via with ICP etch. (b)The cross section
of the device with plating metal.
minutes. A nitride purge every five minutes during the SU8 etch is extremely important
to remove extra charges on the surface and avoid grass formation at the bottom of the
via. The fabricated via is shown in Figure 4.8(a) with an SiN hard mask on top. The
sidewall is very straight without any grass formed at the bottom. Details about the
SU8 dry etch calibration are shown in Appendix A.
After creating the via to top of the mesa, the final step is to deposit metal to make
contact between probe metal and the thin p metal. A plating technique is applied
instead of Ebeam evaporation to fill the via, which has a high aspect ratio. A thin layer
of Ti/Au (30nm/300nm) is sputtered over the entire sample such that this seed layer
covers the entire surface with different topography. The sample is then dipped into a
plating bath for 15 minutes with the pattern created by positive-tone photoresist. After
plating, the seed layer is removed by using gold etchant and HF solution.
81
Chapter 4. Capacitively Loaded TWE MZM
(a)
DC bias pad
1
3
2
4
A
B
Ls
(b)
A’
Lp
B’
Figure 4.9: (a)Top schematic of a 500 µm MZM with slotline design. (b)Optical
image of a 500 µm MZM with fill factor of 0.8.
4.3
Device Characterization
In this section, we will discuss the experimental results of the CL slotline MZM, including
the DC characteristics and high-speed performance with various fill factors. Also, a fast
switch based on this MZI stucture is demonstrated for optical interconnects.
4.3.1
Static Characteristics
The static characteristic of a 500 µm (La ) MZM with F = 0.8 was first measured by
changing the bias of one arm (V1 ) and keeping the other arm (V2 ) at 0 V. A schematic
circuit diagram is shown in Figure 4.10(a) for better understanding of the bias condition.
The normalized transmission of a 500 µm MZM is shown in Figure 4.10(b), which
indicates voltage-length product of 2.4 V-mm and 20 dB ER. The modulation efficiency
is similar to CPW MZM design since the epitaxial layers are identical. On the other
hand, the ER improves from 9 dB to 20 dB by creating better isolation of the MMIs with
82
Chapter 4. Capacitively Loaded TWE MZM
DC bias
V1
v RF
VS
Rload
V2
0
(b)
Vb
v
v
V1 = Vs − Vb + RF , V2 = −Vb − RF
2
2
v pp
v pp
vRF = −
 +
2
2
-2
Normalized Transmission
(a)
-4
-6
-8
-10
-12
-14
-16
-18
-20
0
1
2
3
4
5
6
7
8
Reverse Bias (V)
Figure 4.10: (a)Schematic circuit of the device with DC bias and RF driven signal.
(b)Normalized transmission as a function of reverse bias at 1550 nm.
deep etch silicon waveguide. The propagation loss is 3 dB/mm with 1.5 dB coupling
loss per III-V/Si taper.
For a MZM based on the CDE, device characteristics vary as a function of bias
conditions because the amount of carriers inside the intrinsic area changes as the bias
changes. Figure 4.11 shows the index change, 3 dB bandwidth, and the device loss (including taper loss) of a 500 µm long MZM. As mentioned in Section 2.1.1, Figure 4.11(a)
and (b) shows the index change increases as the bias increases and agrees with the simulations. Similarly, shown in Figure 4.11(c), the 3 dB cutoff frequency is also larger as
the bias varies from 0 V to 5 V since the device capacitance becomes smaller due to
the CDE. In addition, the optical propagation loss shown in Figure 4.11(d) depicts that
absorption increases at higher bias because QCSE is the major effect to introduce index
change. It is clear that tradeoff exists between modulation efficiency, optical loss, and
modulation bandwidth as a function of bias. This MZM can be an approach to low-loss
applications while the modulation bandwidth is not critical and vice versa.
83
Chapter 4. Capacitively Loaded TWE MZM
(c)
1.5
Bandwidth (GHz)
∆n (x10 -3 )
(a)
1
0.5
0
0
1
2
3
4
30
25
20
15
5
0
Reverse Bias (V)
(d)
1.8
1.7
1.6
1.5
1.4
1.3
0
1
2
3
2
3
4
5
4
5
Reverse Bias (V)
500 µm MZM Loss(dB)
dn/dv (x10 -4 /V)
(b)
1
4
5
Reverse Bias (V)
-4
-5
-6
-7
-8
-9
0
1
2
3
Reverse Bias (V)
Figure 4.11: (a)The introduced index change (b)The slope of index change (c)The
3 dB modulation bandwidth (d)The device loss of a 500 µm long MZM as a
function of reverse bias
4.3.2
Electrical Properties of the CL Slotline
To explore the influence of fill factor (F ) on the electrical properties of the CL slotline,
500 µm hybrid silicon MZMs with different F are fabricated by implanting the unloaded
sections. The experimental results measured at -3 V using a Agilent 8164A network
analyzer are shown in Figure 4.12. Again, all the numbers are extracted by converting
84
10
50
8
40
Impedance( Ω)
Loss(dB/mm)
Chapter 4. Capacitively Loaded TWE MZM
6
4
F = 0.4, 0.6, 0.8, 0.9
2
0
0
10
20
30
30
F = 0.4, 0.6, 0.8, 0.9
20
10
0
-10
40
0
10
1
8
0.8
6
4
2
0
10
20
30
20
30
40
0.6
0.4
0.2
F = 0.4, 0.6, 0.8, 0.9
0
10
Frequency (GHz)
C (pF/mm)
Index
Frequency (GHz)
40
0
F = 0.4, 0.6, 0.8, 0.9
0
Frequency (GHz)
10
20
30
40
Frequency (GHz)
Figure 4.12: Experimental electrical property of a 500 µm hybrid silicon MZM
with CL slotline at different fill factor.
S parameters to ABCD matrix using Equation 3.24.
As expected, the loss constant of the device increases when the fill factor decreases.
Since the overall length of the modulator is fixed, the distributed loss per unit length
becomes smaller with small F and vice versa. Similarly, as shown in Figure 4.12, the device capacitance also distributes to the entire slotline and has smaller value by reducing
the F from 0.9 to 0.4. The arrow in the figures indicate the direction of increasing F .
In addition, the overall device impedance, affected by the device capacitance, decreases
85
10
50
8
40
Impedance( Ω)
Loss(dB/mm)
Chapter 4. Capacitively Loaded TWE MZM
6
4
2
0
10
20
30
20
F = 0.4, 0.6, 0.8, 0.9
10
0
F = 0.4, 0.6, 0.8, 0.9
0
30
-10
40
0
10
1
8
0.8
6
4
2
0
10
20
30
30
40
0.6
0.4
0.2
F = 0.4, 0.6, 0.8, 0.9
0
20
Frequency (GHz)
C (pF/mm)
Index
Frequency (GHz)
10
40
Frequency (GHz)
0
F = 0.4, 0.6, 0.8, 0.9
0
10
20
30
40
Frequency (GHz)
Figure 4.13: Simulated electrical property of a 500 µm hybrid silicon MZM with
CL slotline at different fill factor
when F increases. One can expect that the reflection at the input and terminated end
of the slotline is larger in devices with larger F . Finally, the influence of F on effective
index is interesting. Since the electrical signal of an unloaded slotline propagates faster
than the optical signal due to the smaller dielectric constant of the SU8, it is expected
that this phase velocity would decrease with larger loaded sections. Figure 4.12 shows
that our estimation is correct and the effective index of the electrical signal is closer to
that of the optical signal (3.45), which reduces the velocity mismatch for modulation.
86
Chapter 4. Capacitively Loaded TWE MZM
Layer
Table 4.1: Variables for simulation.
width (µm) thickness (µm) r
resistivity (Ω-m)
slotline line
Metal via
P contact
P Cladding
Intrinsic
N contact
ws =135
wv =2.5
wpc =4
wclad =4
wi =3
N/A
dmp =2
dv =3
dpc =0.1
dclad =1.5
di =0.3
dclad =0.15
1/σm = 5 × 10−8
1/σm = 5 × 10−8
13.91 ρpc = 1.47 × 10−5
12.5 ρclad = 4.17 × 10−4
13.4 N/A
12.5 ρclad = 3.88 × 10−6
By substituting all the device dimensions (Table 4.1)and material variables (Chapter 2) to the circuit model, the simulated electrical properties is depicted in Figure 4.13
for comparison. Here we assume the gap between two arms is 30 µm and the empirical constant for implant loss in Equation 4.9 is 3 × 10−7 . Specific contact resistance
(ρspec,p )for p contact layer is found to be 4 × 10−5 by measuring the TLM patterns
while the inductance of the slotline (Lm ) is measured in an unloaded slotline on chip.
0
F=0.4
F=0.6
F=0.8
F=0.9
S21(dB)
-2
-4
-6
-8
-10
0
10
20
30
Frequency (GHz)
40
Figure 4.14: Frequency response with different fill factor
87
Chapter 4. Capacitively Loaded TWE MZM
As expected, the trend of those four parameters based on different F follows the trend
of the experimental data, and the simulation agrees with the experimental data very
well.
So far, all the experimental data indicate that the device with F of 0.4 is the best
one since it has smallest propagation loss, reflection, and velocity mismatch. In contrast,
device with larger F seems not to be preferred in terms of electrical properties. However,
the actual frequency responses of these devices tell a different story as displayed in
Figure 4.14, which indicates that the device with F of 0.9 has the largest bandwidth
compared to the others. For the device with F =0.4, the frequency response is the
worst because the total length (Lt ) of the device is the longest. Therefore, although the
electrical properties per unit length are best, as mentioned in previous section, it still
has larger frequency roll off than the others. In addition, it is preferable to have a short
device in terms of the optical loss and overall footprint.
4.3.3
Large Signal Modulation
Next, the modulation bandwidth of MZM was measured using an Agilent N4373C Lightwave Component Analyzer (LCA) at -3 V reversed bias with 4 V peak to peak drive
signal. A tunable laser, and EDFA and a polarization controller are used to generate the
light coupled into the device. The modulated signal is then coupled out from the device
using a lensed fiber and amplified by another EDFA before connecting to the LCA. The
experimental result of a device (F =0.9) with a 25 Ω termination in Figure 4.15 shows
a modulation bandwidth of 25 GHz, which suggests that this device has the potential
for 40 Gb/s NRZ operation. A simulated EO response (black dash curve) with the
electrical parameters extracted in Section 4.3.2 is also shown. Another curve with 50 Ω
88
Chapter 4. Capacitively Loaded TWE MZM
Normalized S21 (dB)
2
0
Z L =25 Ω
-2
-4
Z L =50 Ω
-6
-8
-10
0
10
20
30
Frequency (GHz)
40
Figure 4.15: Modulation bandwidth measured at -3 V for a MZM with F =0.9 at
1550 nm.
termination is plotted as a comparison to demonstrate the effect of termination. With
smaller termination, one can increase the modulation bandwidth by significant amount.
In contrast, this trade results in larger power consumption and needs to be used with
caution.
The large signal modulation was also characterized with a SHF 40G BERT system.
A 231 -1 pseudorandom bit sequence (PRBS) at 25 Gb/s and 40 Gb/s as shown in
Figure 4.16(a) is used to drive the device while an optical signal at 1550 nm is coupled
to the chip. In order to have better ER, port 2 to 4 is chosen rather than port 2 to 3
because it has an inherent 180 degree phase shift from the Mach-Zehnder interferometer
(MZI) structure. When the biases of two arms are identical, the output signal of port 2
to 3 is a “one” and it turns to “zero” while additional 180 degree phase shift difference
is applied to the device. However, due to the power imbalance introduced by QCSE
89
Chapter 4. Capacitively Loaded TWE MZM
(a) Drive Signal
40Gb/s
25Gb/s
10ps
20ps
(b) Modulated
F=0.9
16.9dB
F=0.9
13.2dB
F=0.8
15.5dB
F=0.8
11.4dB
F=0.6
11.4dB
F=0.6
8.9dB
F=0.4
9.5dB
F=0.4
7.3dB
Figure 4.16: (a)25 Gb/s and 40 Gb/s electrical drive signal with 231 -1 NRZ PRBS.
(b)Modulated signal with different fill factor.
90
Chapter 4. Capacitively Loaded TWE MZM
(Chapter 2), the “zero” is limited and significantly degrades the ER. Therefore, a better
ER can be obtained if the “zero” condition occurs when the powers of two arms are
identical, which corresponds to port 2 to 4 with 180 degree inherent phase difference.
In addition, extra phase change might be introduced by the III-V/silicon taper and
results in phase difference other than 180 degree for prot 2 to 4. This can affect the bias
condition greatly and also the ER. To solve this problem, additional phase modulators
should be placed on two arms to adjust the phase in the future.
Figure 4.16(b) shows the modulated signal from the hybrid silicon MZM driven by
4 Vpp swing while the biases (Vb and Vs ) are adjusted to achieve best signal quality.
As can be seen, the eye is clearly open without much distortion from the drive signal
for all devices at 25 Gb/s. The ERs are 16.9 dB and 15.5 dB, 11.4 dB, and 9.5 dB for
devices with F =0.9 to 0.4, respectively. On the other hand, the eye quality at 40 Gb/s
slightly degrades but is still widely open due to the limitation of the 25 GHz modulation
bandwidth. The ERs at 40 Gb/s is also labeled next the the eye diagrams, and are 13.2
dB, 11.4 dB, 8.9 dB, and 7.3 dB for F equal to 0.9 to 0.4, respectively. The ERs, to the
best of our knowledge, are the best ERs above 25 Gb/s for any silicon based modulator
and makes long-haul optical communication feasible.
4.3.4
Chirp Properties
In addition to the modulation bandwidth, chirp is one of the important metrics for data
transmission. The chirp parameter of a MZM can be expressed approximately in terms
of the phase change in each arm (∆φ1 and ∆φ2 ) as [9]:
α=
∆φ1 + ∆φ2
∆φ1 − ∆φ2
91
(4.12)
Chapter 4. Capacitively Loaded TWE MZM
With the push-pull configuration, the phase changes in the two arms have the same
amplitude but are out of phase (∆φ1 = −∆φ2 ) such that chirp-free modulation can be
achieved as shown in Equation 4.12. In reality, however, the amplitudes of the phase
change in the two arms are usually not identical because the electrorefraction effect
is bias-dependent. Therefore, an ideal zero chirp condition can only be realized if the
modulation effect is perfectly linear. The small signal chirp parameter can be obtained
by measuring the EO response by inserting a dispersive fiber between the modulator and
the LCA [10]. The interaction between fiber dispersion and modulator chirp results in
resonance dips in the spectrum as shown as in Figure 4.17(a). The relationship between
the resonance points and the chirp can be written as [10]:
fu2 L
c0
2
=
1 + 2u − arctan(α) ,
2Dλ2
π
(4.13)
where fu is the uth order of resonance, c0 is the speed of light, D is the fiber dispersion,
λ is the wavelength, and α is the chirp. An experimental curve of Equation 4.13 is
illustrated in Figure 4.17(b). By fitting the experimental resonance points, the chirp
under different biases can be calculated and the results are shown in the inset of Figure 4.17(b). It indicates that the chirp is around -0.75 over a 5 V range. This negative
chirp can counteract dispersion and consequently reduce the power penalty for long-haul
data transmission applications.
4.4
Switch Characterization
Communication networks have grown rapidly over the past decade due to continually
increasing capacity demands on computation, multimedia and data services. Early
92
Chapter 4. Capacitively Loaded TWE MZM
(a)
(b)
f *L (GHz *km)
5
4
2
-10
4
0
-0.5
2
Chirp
-20
3
-1
2
u
S21 (dB)
0
6
x 10
1
-30
0
10
20
30
40
0
50
Frequency (GHz)
-1.5
0
2
4
0
6
1
2
3
4
Reverse Bias (V)
8
5
10
2u
Figure 4.17: (a)Measured resonant frequency where u is uth resonant position.
(b)Measured chirp spectrum with 36 km SMF. Inset: Chirp parameters at different
bias condition.
1000
Energy per bit (nJ)
100
Optical Switch
10
1
0.1
This
work
AWG
Switch
PoS Ethernet Core
Tx/Rx Switch Router
PON
ONU
IPTV
Server
Electrical router/switch
0.01
Figure 4.18: Energy per bit in network devices. (modified from Tucker, LEOS
Annual, 2008).
on in the development of high-speed network infrastructure, key enabling components,
such as routers and switches were purely electronic and as such operated at higher
93
Chapter 4. Capacitively Loaded TWE MZM
data rates at the cost of consuming increasing power. Furthermore, higher capacity
was achieved by miniaturizing the underlying silicon technology. Today, a state of the
art router has a capacity of 92 Tb/s with more than 0.2 million CPUs. This router,
however, dissipates around 1 MW regardless of the cooling system used. The tremendous
power consumption of electronic routers has become a major concern, and limits their
performance as capacity increases. Hence, it is absolutely vital to develop a low power
alternative to existing switching and routing technologies. Optical switches having lower
power consumption and smaller footprints offer one such possibility. A variety of optical
switch technologies have been widely explored including wavelength routing switches
(WRS) [11, 12], broadcast-and-select switches [13], and a Mach-Zehnder interferometer
(MZI) based switch. Among these, the first technique is based on a tunable wavelength
converter (TWC) and arrayed-waveguide grating (AWG) while the second one is based
on a semiconductor optical amplifier (SOA) gate array. Both these consume at least 100
mW for each active component. In contrast, power consumption in the MZI switch can
be as low as 100 µW per 2x2 switch if the phase-shift section utilizes the electro-optic
effect with reverse bias and the switch is driven without a load by a short connection from
the amplifier. Figure 4.18 shows the energy per bit for various network devices. As can
be seen, network devices of sub-wavelength switching consume at least three orders of
magnitude larger power than the devices utilizing wavelength switching. Considering the
number of routers/switches worldwide, the difference in power consumption quickly adds
up. In addition to low-power consumption, switching speed is another key performance
factor. A fast switch can reduce the size of guard band between packages, and results
in higher-utilization networks.
The normalized transmission as a function of reverse bias of a 500 µm HSS with
fill factor of 0.8 is shown in Figure 4.19(a). Due to the phase difference generated in
94
Chapter 4. Capacitively Loaded TWE MZM
(a)
1
1-> 3
1-> 4
2 -> 3
2 -> 4
Normalized Transmission
0.9
0.8
(b)
0.7
0.6
0.5
ER
(dB)
Crosstalk
(dB)
1 -> 3
25
-19
2 -> 3
18
-29
1 -> 4
0.4
0.3
2 -> 4
0.2
0.1
0
Port
0
1
2
3
4
5
6
7
19
26
-24
-15
8
Reverse Bias (V)
Figure 4.19: (a)Normalized transmission of a hybrid silicon switch for all port
configurations. (b)Measured extinction ratio, crosstalk, and rise time for each
port at 1550 nm.
the tapers, the bias condition is adjusted to have the best ER. As can be seen, the Vpi
is about 5 V, which equals to voltage length product of 2.5 V-mm. The ER and cross
talk (XT) for all port configurations is listed in Figure 4.19(b). The ER is above 18 dB
while XT is smaller than -12 dB. As can be observed from Table I, one particular port
(2 to 4) has smaller transmission because of additional loss from imperfect fabrication,
which leads to a 3 dB increase in XT as well. In addition, the XT is generally higher for
the through ports due to the loss imbalance between the two arms of the interferometer
since the quantum confined Stark effect starts to become involved in introducing index
shift when the bias voltage is larger than 3 V.
The switch speed of the HSS is also of great interest because small latency is required for more efficient network utilization. The rise time of each port is shown in
Figure 4.20(a). The rise time (20% to 80 %) is shorter than 20 ps and the fall time is
less than 17 ps for all ports, which means that only 0.19 % of the total time interval is
95
Chapter 4. Capacitively Loaded TWE MZM
(a)
Port 1 to 3
(b)
Port 1 to 4
-5
back to back
Port 1 to 3
Port 1 to 4
Port 2 to 3
Port 2 to 4
200ps
Port 2 to 3
Port 2 to 4
Log(BER)
-6
-7
-8
-9
-10
-11
-34
-33
-32
-31
-30
-29
-28
Received Power (dBm)
Figure 4.20: (a)Rise and fall time of the switch for all porsts. (b)BER versus
optical received power for all ports configurations at 40 Gb/s with 231 -1 NRZ
PRBS.
necessary as a guard band for a standard cell of 40 Gb/s asynchronous transfer mode
(ATM) network. The utilization of the network can be around 100 times more efficient
by using the switch presented here rather than a WRS, which in general has a rise/fall
time around several nanoseconds.
A 40 Gb/s bit-error-rate (BER) measurement was also performed to explore the
signal integrity for switching applications. The power penalties for all ports, shown in
Figure 4.20(b), are below 0.5 dB at 1×10−10 BER for each port.
The power consumption of the switch in the “on” state can be calculated by measuring the photocurrent at -5 V. With an optical input power of 3 dBm, the detected
photocurrent is only 20 µA, which results in 100 µW power dissipation per switching
action. Figure 4.21 illustrated the power consumption of a WRS and the HSS when
the number of ports increases. In this estimation, we assume the power consumption
96
Chapter 4. Capacitively Loaded TWE MZM
Power consumption (mW)
10
10
10
10
4
(a)
3
WRS
2
1
(b)
This work
10
10
0
−1
0
10
20
30
40
50
60
70
Number of ports
Figure 4.21: Power consumption as a function of number of ports with different
switch architectures. Inset (a): Illustration of an 8x8 WRC [11]. Inset (b): a
typical 8x8 Benes network (photo courtesy Martijn Heck).
of each port for a WRS is 100 mW, and the power used for HSS is 100 µW per 2x2
switch with the architecture of a Benes’ network. It can be clearly seen that the power
consumption of HSS is about two orders lower than a conventional WRS at 64 ports,
which is not surprising since the TWC used in WRS dissipates more than 100 mW of
power.
4.5
Summary
In this chapter, we design a high-speed modulator by utilizing the CL slotline TWE
structure. A circuit model is first developed to estimate transmission line characteristics
and frequency response. Some important design parameters and figure of merits for the
modulation bandwidth are discussed to explore the design spcae. Next, fabrication
97
Chapter 4. Capacitively Loaded TWE MZM
improvements are introduced to establish a more stable and precise fabrication process
for high-speed devices. At the end, we demonstrate a hybrid silicon MZM with 25 GHz
bandwidth and 40 Gb/s large signal modulation with various fill factors. In addition,
a low-power, fast hybrid silicon switch based on the MZI structure is discussed. The
power consumption of 100 µW/switch and 20 ps switch time makes it very competitive
for typical demands of data transmission optical networks.
98
Bibliography
[1] F. Sterzer, “Push-Pull Optical Modulators and Demodulators,” Appl. Opt.,
vol. 2,
no. 11,
pp. 1197–1198,
Nov 1963. [Online]. Available:
http:
//ao.osa.org/abstract.cfm?URI=ao-2-11-1197
[2] J. C. Webster and F. Zernike, “Push-pull thin-film optical modulator,” Applied
Physics Letters, vol. 26, no. 8, pp. 465–467, 1975.
[3] C. M. Gee, G. D. Thurmond, and H. W. Yen, “Traveling-wave electrooptic
modulator,” Appl. Opt., vol. 22, no. 13, pp. 2034–2037, Jul 1983. [Online].
Available: http://ao.osa.org/abstract.cfm?URI=ao-22-13-2034
[4] T. C. Huang, T. C. Yang, Z. M. Chuang, Y. Chung, L. A. Coldren, and N. Dagli,
“Study of chirp-free operation for guide/antiguide modulator,” IEEE Photonics
Technology Letters, vol. 4, no. 9, pp. 1018–1019, 1992.
[5] J. Shin, C. Ozturk, S. R. Sakamoto, Y. J. Chiu, and N. Dagli, “Novel T-rail
electrodes for substrate removed low-voltage high-speed GaAs/AlGaAs electrooptic
modulators,” IEEE Transactions on Microwave Theory and Techniques, vol. 53,
no. 2, pp. 636–643, 2005.
99
BIBLIOGRAPHY
[6] J. Shin, S. Wu, and N. Dagli, “35-GHz Bandwidth, 5-V-cm Drive Voltage, Bulk
GaAs Substrate Removed Electrooptic Modulators,” IEEE Photonics Technology
Letters, vol. 19, no. 18, pp. 1362–1364, 2007.
[7] S. Akiyama, H. Itoh, S. Sekiguchi, S. Hirose, T. Takeuchi, A. Kuramata, and
T. Yamamoto, “InP-Based Mach–Zehnder Modulator With Capacitively Loaded
Traveling-Wave Electrodes,” IEEE Journal of Lightwave Technology, vol. 26, no. 5,
pp. 608–615, 2008.
[8] H. Boudinov, H. H. Tan, and C. Jagadish, “Electrical isolation of n-type and p-type
InP layers by proton bombardment,” Journal of Applied Physics, vol. 89, no. 10,
pp. 5343–5347, 2001.
[9] A. Sneh and C. Doerr, Integrated Optical Circuits and Components - Design and
Applications, E. J. . Murphy, Ed.
Marcel Dekker, Inc., 1999.
[10] F. Devaux, Y. Sorel, and J. F. Kerdiles, “Simple measurement of fiber dispersion
and of chirp parameter of intensity modulated light emitter,” IEEE Journal of
Lightwave Technology, vol. 11, no. 12, pp. 1937–1940, 1993.
[11] S. C. Nicholes, M. L. Masanovic, B. Jevremovic, E. Lively, L. A. Coldren, and
D. J. Blumenthal, “An InP Monolithic Tunable Optical Router (MOTOR) Packet
Forwarding Chip,” IEEE Journal of Lightwave Technology, vol. 28, no. 4, pp. 641–
650, 2010.
[12] T. Segawa, S. Matsuo, T. Kakitsuka, Y. Shibata, T. Sato, Y. Kawaguchi,
Y. Kondo, and R. Takahashi, “All-optical wavelength-routing switch with
monolithically integrated filter-free tunable wavelength converters and an AWG,”
100
BIBLIOGRAPHY
Opt. Express, vol. 18, no. 5, pp. 4340–4345, Mar 2010. [Online]. Available:
http://www.opticsexpress.org/abstract.cfm?URI=oe-18-5-4340
[13] P. Gambini, M. Renaud, C. Guillemot, F. Callegati, I. Andonovic, B. Bostica,
D. Chiaroni, G. Corazza, S. L. Danielsen, P. Gravey, P. B. Hansen, M. Henry,
C. Janz, A. Kloch, R. Krahenbuhl, C. Raffaelli, M. Schilling, A. Talneau, and
L. Zucchelli, “Transparent optical packet switching: network architecture and
demonstrators in the KEOPS project,” IEEE Journal on Selected Areas in Communications, vol. 16, no. 7, pp. 1245–1259, 1998.
101
Chapter 5
Application: Tunable Microwave
Filter
Integration of different components on the same platform has always been one of the
important approach to realize photonic integrated circuit (PIC). Up to date, we have
demonstrated lasers, amplifiers, photodetectors, and modulators on the hybrid silicon
platform (HSP), but a PIC is still missing in the diagram of optical communication.
It will be a leap to prove the capability of this platform if PIC can be achieved. In
this chapter, we focus on an integrated tunable microwave filter using the developed
amplifier and modulator technologies. The advantage of this platform for microwave
filter is discussed in 5.1 and design details will be presented in 5.2. Finally, the results
and interesting applications is included in 5.3.
102
Chapter 5. Application: Tunable Microwave Filter
5.1
Introduction
Microwave finite impulse response (FIR) and infinite impulse response (IIR) filters implemented using optical delay lines have been proposed and demonstrated more than
two decades ago [1]. Incoherent microwave filter responses were obtained based on the
length of the recirculating loop and the modulation frequency. However, such filters
being discrete in nature have suffered from various issues such as controllability and
complexity. Recent advances in PIC technology have made it possible to overcome
some of these issues and consequently, realize compact and environmentally insensitive
devices [2, 3]. The demonstrated FIR/IIR filters based on passive ring resonator are of
particular interest in wavelength division multiplexing (WDM) systems where a precise
filter characteristic can be realized by using a combination of the resonators [4, 5, 6].
In contrast, the application of such photonic filters in the analog and radio frequency
domain have been less explored due to coherent interference on the chip scale although
discrete photonic microwave filters have been widely investigated [7, 8, 9]. The microwave response of a filter on PIC usually consists of interference between modulation
frequencies and interference of carrier frequencies because the chip size is smaller than
the coherence length of the laser source. Structures such as Mach-Zehnder interferometers (MZI) combined with ring resonators can be utilized to engineer these filters in the
GHz response range [10]. This filter has shown tunability from 2 GHz to 15 GHz and
has a 3 dB bandwidth of 0.635 GHz. However, this particular scheme where the ring
resonators are cascaded on one arm of the MZI, only a pole or zero can be achieved at
a particular port and the filter quality is fairly sensitive to manufacturing errors, i.e.
the coupling coefficient between the ring and one arm of the MZI. In this paper, we
propose and demonstrate a microwave filter on the hybrid silicon platform (HSP) [11]
103
Chapter 5. Application: Tunable Microwave Filter
incorporating a MZI and a ring resonator where the optical paths of the MZI and the
ring resonator are partially overlapped. A complete filter response, including both poles
and zeros at one port, can be generated by utilizing low-loss silicon-on-insulator (SOI)
waveguides and semiconductor amplifiers (SOAs) [12] on the HSP. The low-loss SOI
waveguides give access to longer ring lengths unavailable in InP PIC platforms [13, 14]
and enables the realization of free spectral ranges around a GHz while the amplifiers
are used to control and achieve specific responses. The filter architecture proposed here
is ideal for applications such as band filtering for narrowly spaced signals or arbitrary
narrow band pass filtering.
5.2
Device Design
The schematic drawing of a hybrid silicon tunable filter based on the MZI and ring
resonator is shown in Figure 5.1(a) along with the SEM image of the device in Figure 5.1(b). Both SOAs and thermal modulators (MODs) are the control elements to
obtain the desired output function as well as tune the optical wavelength. The transfer
function of this device is described in Equation 5.1 where L1 , L2 , L3 , L4 , and L5 represent the lengths of different paths of the cell illustrated in Figure 5.1(a). The definition
of other parameters is listed in Table 5.1.
−j[β(L2 +L3 +L4 )+φr ]
A
e
2
T = A1 e−j[βL1 +φaf +φf ] +
−j[β(L
+L
)+φ
+φ
]
ar
r
3
5
1 − Aloop e
(5.1)
The response of such a device can be easily understood by considering this formula as
an interferometer with a ring on one arm. Without the feedback from the recirculating
loop, i.e. Aloop = 0, the equation simply expresses the transfer function of a MZI with
104
Chapter 5. Application: Tunable Microwave Filter
L1
(a)
SOA f
MOD f
SOA r
MMI
MMI
L5
MMI
L4
MMI
L2
L3
MOD r
(b)
Thermal Modulator
SOA
300 µm
Figure 5.1: (a)The schematic of a hybrid silicon tunable filter. (b)An SEM picture
of the fabricated device .
a modulator on one arm and an SOA on the other arm as shown in Figure 5.2(a)Ring whereas the second term indicates the transfer function of a ring with an SOA
and a phase modulator inside the loop. The overall transfer function of this filter
can be decomposed into three distinct responses as illustrated in Figure 5.2(a): ring,
MZI, and cell response. The first two responses can be realized by using one of the
SOAs as the optical absorbers to block the light passing through certain paths. For
example, in order to obtain ring response, the SOA on the forward path (SOAf ) is
reverse biased as an absorber and the SOA inside the loop (SOAr ) is forward biased
as an amplifier to compensate for propagation and splitter loss. In contrast, the MZI
response is obtained when SOAf is forward biased as an amplifier and SOAr is reverse
105
Chapter 5. Application: Tunable Microwave Filter
Table 5.1: Definition of terms in Equation 5.1.
Parameters Definition
A1
A2
Aloop
φf
φr
φaf
φar
β
Amplitude change along L1
Amplitude change along L2 , L3 , L4
Amplitude change along L3 , L5
Phase change from phase modulators on the forward path
Phase change from phase modulators inside the ring
Phase change from SOA on the forward path
Phase change from SOA inside the ring
Propagation constant in silicon waveguides
biased. Finally, the cell response is obtained when both SOAf and SOAr are used as
amplifiers. This cell response can be further altered by changing the injection current of
both modulators (MODf and MODr ). Identical cell response should be able to realize
at any desire wavelength as long as MODf and MODr are adjusted properly. The
simulated transfer functions of this filter based on a 5 mm delay loop are displayed in
Figure 5.2(b) where Aloop is set to 0.7. As can be seen, the difference in FSRs between
ring and MZI response can be used to engineer the desired filter shape wherein certain
nearby wavelengths need to be eliminated around the operating wavelength. In this
work, the path difference between two arms is four times smaller than the length of
the ring resonator such that multiple resonant peaks are presented in one MZI cycle.
The combination can be manipulated by changing the loop length as well as the path
difference between two arms of the MZI. Consequently, poles and zeros can be achieved
in a single cell and the resonance peaks and dips are tunable by applying the appropriate
index change using the thermal (phase) modulator. This architecture makes selective
filtering possible; and the small FSR from the large delay loop provides the ability to
filter out adjacent wavelengths on a sub-nanometer scale, which corresponds to a FSR
106
Chapter 5. Application: Tunable Microwave Filter
Ring
(b)
2
Transmission (a.u.)
(a)
MZI
Cell
Ring
MZI
Cell
1.5
1
0.5
0
1575
1575.5
1576
1576.5
1577
Wavelength (nm)
Figure 5.2: (a)Schematic figure of three different configurations realizable with
the hybrid silicon filter. (b)Simulated response of the filter operated as a: ring,
MZI, and unit cell.
of around several GHz in the frequency domain.
107
Chapter 5. Application: Tunable Microwave Filter
Laser
EDFA
5mm delay loop
PC
Oscilloscope
Trigger
PD
Figure 5.3: The experimental setup to measure the filter response. An oscilloscope
is used instead of an OSA to resolve the spectrum.
5.3
Experiment Setup and Results
5.3.1
Filter Response in Optical Domain
The experimental setup used to collect the filter responses is illustrated in Figure 5.3. A
temporally-swept tunable laser and a 1 GHz oscilloscope were used instead of an optical
spectrum analyzer (OSA) because the FSR of interest was smaller than the resolution of
a conventional OSA. An Agilent 8163A tunable laser was used as a laser source, followed
by an erbium-doped fiber amplifier (EDFA) and a polarization controller are used to
generate the signal. The polarization is maintained at TE because the quantum wells
of the SOA are compressively strained and have larger interaction with TE polarized
signal. A lensed fiber was then utilized to couple the light into the hybrid silicon filter.
The output signal was collected by a lensed fiber and coupled to a photodetector (PD)
attached to the oscilloscope. By adjusting the swept speed of the wavelength, a time
varying signal corresponding to the wavelength can be measured where the resolution
is much higher than a conventional OSA. Hence, device/ring structures with large loop
108
Chapter 5. Application: Tunable Microwave Filter
Transmission(µW)
(a)
0.8
0.7
140mA
0.6
0.5
0.4
0.3
0.2
70mA
0.1
0
0mA
1575
1575.5
1576
1576.5
1577
Wavelength (nm)
Transmission(µW)
(b)
0.8
0.7
0.6
30mA
0.5
0.4
0.3
0.2
20mA
0.1
0
1575
0mA
1575.5
1576
1576.5
1577
Wavelength (nm)
Figure 5.4: (a)Measured ring response of a 5 mm long delay loop at different
current levels. (b)Experimental MZI response.
lengths can be resolved.
The experimentally responses of the ring and MZI are shown in Figure 5.4(a) and
(b), respectively. First, the ring response was measured with SOAf reverse biased and
109
Chapter 5. Application: Tunable Microwave Filter
the bias current of SOAr varied from 0 mA to 140 mA (before lasing occurs). As can
be seen in Figure 5.4(a), the response is almost independent of wavelength at low bias
current because the loss inside the loop is very high. The ring resonance with FSR =
0.164 nm then becomes stronger as the bias current increases. The maximum extinction
ratio (ER) of 6 dB occurs at 140 mA, where the ring is just below the lasing threshold.
The variation over different wavelength of peak transmission is due to the resolution
of the instrument, not due to the inherent characteristics of the filter. In addition,
the MZI response was also measured by using the SOAr as an optical absorber. The
experimental curve with 0.654 nm FSR is shown in Figure 5.4(b). The largest ER of
8 dB occurs when the amplitudes from two arms of the interferometer are equal while
SOAf is biased at 20 mA. The amplitude becomes larger as the current increases to
30 mA, yet the ER is reduced due to additional amplified spontaneous emission (ASE)
noise. Next, the cell response was measured by fixing the bias current of SOAr at 140
mA so that the ring response is maximized. The cell response resembles more closely
the ring response, shown in Figure 5.5(a), when the SOAf does not have enough gain to
compensate the loss from SOA itself. As the amplitude of both ring and MZI response
are almost identical (Figure 5.5(b)), the cell response has both poles and zeros over the
measured wavelength range, which is exactly what we expect from simulation. As the
bias current of the SOAf increases, the amplitude of the MZI is larger than that of the
ring response and consequently dominates the cell response, which is also degraded by
higher noise floor from increasing ASE.
As can be seen in Figure 5.4 and Figure 5.5, the ERs of ring, MZI and cell responses
are limited by ASE generated from the on-chip SOAs and the C-band EDFA in the measurement system. To improve the signal quality of the filter, there are two approaches
can be applied in the future. First approach is to replace the C-band EDFA by an
110
Chapter 5. Application: Tunable Microwave Filter
2
(a)
(b)
(c)
Transmission(µW)
1
Ring
MZI
Cel
20mA
0
2
30mA
1
0
2
40mA
1
0
1575
1575.5
1576
1576.5
1577
Wavelength (nm)
Figure 5.5: Experimental cell responses with SOAr at 140 mA. The injected current of the SOAf is adjusted from 20mA (a) to 40mA (c) to show the difference
on cell responses.
L-band EDFA since the on-chip SOAs have gain characteristics in the range around
1575 nm. An L-band EDFA can significantly increase the signal to noise ratio of the
optical input, and consequently improve the filter ER. The second approach is a structure change, where the input MMI in Figure 5.1(a) is replaced by a tunable coupler.
Currently the MZI is not optimized because of power imbalance between two arms. If a
tunable coupler is used to adjust the splitting ratio of the incoming signal, the current
level of the on-chip SOAs can be reduced. The ASE, therefore, can be decreased as well
and introduce better filter response. The tunable coupler can be implemented utilizing
111
Chapter 5. Application: Tunable Microwave Filter
the technique developed on this platform, which consists of phase modulators on two
arms of a MZI.
One of the key components in this filter is the phase modulator. It is used to
introduce the necessary index shift in the filter response which in turn allows for different
filter shapes to be realized. A resistor in the form of a thin layer of NiCr was deposited
next to the waveguide. Passing current through the resistor generates the necessary
heat to thermally tune the index. Figure 5.6(a) shows the relative wavelength shift in
terms of ring FSR by adjusting the current (Ir ) of the thermal modulator inside the ring
(MODr ). The modulator is 637.5 µm long and 10 µm wide. As can be seen, a shift over
one FSR, which corresponds to 2π phase change, can be achieved when Ir equals 15
mA and 11 mA for ring and MZI response, respectively. The thermally induced index
shift is positive, which corresponds to a shift towards longer wavelength (red shift). In
contrast, the forward path thermal modulator (MODf ) introduces a blue shift (towards
shorter wavelength). This is shown in Figure 5.6(b) where it can also be seen that the
phase shift introduced by this 950 µm MODf can achieve a single FSR at 13 mA. A
blue shift is observed rather than red shift because the MODf is on the shorter arm of
the MZI such that any positive index change will reduce the phase difference between
two arms. The relation between wavelength change and injected current of both MODs
can be described in Equation 5.2, where a1 , a2 , and a3 are coefficients extracted from
fitted curve in Figure 5.6.
dλring = a1 Ir2 ,
dλM ZI = a2 Ir2 + a3 If2
(5.2)
To verify the tunability of the filter response using thermal modulation, experimental
results are presented in Figure 5.7. All three figures have the same amplifier bias
112
(a)
3.0
(b)
0
Number of FSR(Ring)
2.5
Number of FSR(Ring)
Chapter 5. Application: Tunable Microwave Filter
-0.5
2.0
1.5
MZI
1.0
Ring
0.5
0
0
5
10
15
20
MODr Current (mA)
MZI
-1.0
-1.5
-2.0
0
5
10
15
20
MODf Current (mA)
Figure 5.6: (a)The wavelength shift of ring and MZI responses at different current
level of MODr , where the ring FSR is 0.164 nm. (b)The wavelength shift of MZI
response at different current level of MODf .
condition; namely, SOAr and SOAf are biased at 100 mA and 24 mA, respectively.
Figure 5.7(a) is the cell response when If and Ir are zero. By increasing If to 13 mA,
the ring response does not change and the MZI response shifts one FSR down to a lower
wavelength (Figure 5.6(b)). The combined cell response is shown in Figure 5.7(b). It is
clear that the cell response maintains a similar shape but is shifted to the left by one
FSR. The response can be translated further by increasing If to 19.5 mA as shown in
Figure 5.7(c). The ER of the filter degrades by 2 dB due to increased thermal effects at
higher injection currents into the thermal phase modulator. However, the overall filter
response is similar to Figure 5.7(a).
5.3.2
Filter Response in Frequency Domain
To explore the microwave characteristics of the silicon hybrid filter, responses of interest
in the frequency domain were measured using the experimental setup illustrated in
113
Chapter 5. Application: Tunable Microwave Filter
1
(a)
If = 0mA
(b)
If = 13mA
(c)
If = 19.5mA
0.8
0.6
0.4
Normalized Transmission
0.2
0
1
0.8
0.6
0.4
0.2
0
1
0.8
0.6
0.4
0.2
0
1575
1575.5
1576
1576.5
1577
Wavelength (nm)
Figure 5.7: (a)Cell response without any thermal modulation. (b)Cell response
with Ir =0 mA and If =13 mA. (c)Cell response with Ir =0 mA and If =19.5 mA.
Figure 5.8. An Agilent E8703A lightwave component analyzer (LCA) with an external
tunable laser source is used to provide a modulated optical signal. This signal with
double sideband modulation is amplified, filtered to eliminate ASE noise of the EDFA,
and then coupled into the hybrid silicon filter. The LCA has a bandwidth of 20 GHz,
which is similar to the FSR of the ring resonator shown in Figure 5.5(a). Since the
coherent microwave response is a combination of the interference between modulation
frequencies and the interference between carrier frequencies, arbitrary filter shapes can
be generated by adjusting the wavelength of the external laser. A microwave filter with
114
Chapter 5. Application: Tunable Microwave Filter
EDFA
LCA
EDFA
Filter
PC
Filter
DUT
Laser


 


Figure 5.8: Schematic of the setup to measure the filter function in the frequency
domain.
either a pole or a zero can be achieved if the carrier wavelength is at the center of a
pole or a zero in the optical domain. As shown in Figure 5.9(a), a arrow indicates the
carrier frequency while the shadow region represents ± 20 GHz bandwidth of the LCA.
The microwave response is then a summary of the optical response within this 40 GHz
bandwidth. When the carrier wavelength differs from the peak of a pole or the dip of
a zero, a combined response of pole and zero will be observed. Figure 5.9(b) shows the
microwave transfer function of the cell response by changing the carrier frequency from
1575.00 nm to 1575.22 nm. The SOAf is kept at 32 mA while the SOAr is biased at 140
mA so that ring response is maximized and the amplitudes of ring and MZI response
are at similar level. As can be seen, the frequency spacing between a pole and a zero
changes when the carrier frequency is adjusted. For example, the carrier frequency of
1575.10 nm is in the middle between a pole and zero; thus the frequency difference is
about 10 GHz, which is exactly half of the FSR of the filter. It is clear that tunability
over 20 GHz of this filter can be realized by varying the carrier frequency by 0.22 nm.
115
(a)
Transmission (µW)
Chapter 5. Application: Tunable Microwave Filter
2
40 GHz
1
0
1576.5
1575

(b)
(c)
1575.00 nm
−20
−30
1575.05 nm
Transmission (dB)
Transmission (dB)
−25
−30
1575.10 nm
−20
−25
−30
1575.15 nm
−20
−25
−30
1575.22 nm
−20
−30
5
10
1575.05 nm
0
−10
−20
1575.10 nm
0
−10
−20
1575.15 nm
0
−10
−20
1575.22 nm
0
−10
−20
−25
0
1575.00 nm
0
−10
−20
−25
−20
1576
Wavelength (nm)
15
20
Frequency (GHz)
0
5
10
15
20
Frequency (GHz)
Figure 5.9: (a)Measured filter responses at various wavelength in the frequency domain. The SOAr is biased at 140 mA and SOAf is biased at 32 mA. (b)Simulated
filter responses corresponding to the experiment results.
The best ER of pole and zero are 4 dB and 8 dB, respectively. Again, the ER is limited
by the ASE noise. Similar responses can also be achieved by adjusting the thermal
modulator to change the phase of the carrier instead of shifting the carrier wavelength.
As a comparison, the simulation results of the device are depicted in Figure 5.9(c) where
the loop gain is set to 0.7. It can be seen that the measured filter responses agree well
116
Chapter 5. Application: Tunable Microwave Filter
with the theoretical calculation within the frequency range of interest. Multiple poles
and zeros within 20 GHz can also be further demonstrated if the loop delay is longer.
5.4
Summary
A hybrid silicon microwave filter based on a MZI and a ring resonator was demonstrated.
A filter with complete functions in the optical domain, including ring, MZI and cell
responses, was measured by utilizing the SOAs and modulators as control elements.
By taking advantage of the low-loss waveguide and gain element on the hybrid silicon
platform, filter responses in the GHz frequency range were also demonstrated. The
microwave filter responses have large tunability over 20 GHz range by slightly changing
the carrier frequency. The experimental results in both optical domain and frequency
domain match closely the simulated responses. In the future, SOAs as feedback elements
can be added to the cell so that accurate and stable control of the cell response can be
realized. Meanwhile, the thermal modulator can be replace by the hybrid silicon MZM
to reduce the response time such time in-time monitor and fast switch function are
possible. Cascaded cells with similar function would be another interesting approach to
increase the complexity and ER of the filter response. Additionally, more complicated
cell functions are also feasible by modifying the length of the delay loop as well as the
coupler splitting ratio.
117
Bibliography
[1] J. E. Bowers, S. A. Newton, W. V. Sorin, and H. J. Shaw, “Filter response of
single-mode fibre recirculating delay lines,” Electronics Letters, vol. 18, no. 3, pp.
110–111, 1982.
[2] B. E. Little, S. T. Chu, H. A. Haus, J. Foresi, and J.-P. Laine, “Microring resonator
channel dropping filters,” IEEE Journal of Lightwave Technology, vol. 15, no. 6,
pp. 998–1005, 1997.
[3] B. E. Little, J. S. Foresi, G. Steinmeyer, E. R. Thoen, S. T. Chu, H. A. Haus,
E. P. Ippen, L. C. Kimerling, and W. Greene, “Ultra-compact Si-SiO2 microring
resonator optical channel dropping filters,” IEEE Photonics Technology Letters,
vol. 10, no. 4, pp. 549–551, 1998.
[4] S. T. Chu, B. E. Little, W. Pan, T. Kaneko, and Y. Kokubun, “Second-order
filter response from parallel coupled glass microring resonators,” IEEE Photonics
Technology Letters, vol. 11, no. 11, pp. 1426–1428, 1999.
[5] J. V. Hryniewicz, P. P. Absil, B. E. Little, R. A. Wilson, and P.-T. Ho, “Higher
order filter response in coupled microring resonators,” IEEE Photonics Technology
Letters, vol. 12, no. 3, pp. 320–322, 2000.
[6] F. Xia, M. Rooks, L. Sekaric, and Y. Vlasov, “Ultra-compact high order
ring resonator filters using submicron silicon photonic wires for on-chip optical
interconnects,” Opt. Express, vol. 15, no. 19, pp. 11 934–11 941, Sep 2007. [Online].
Available: http://www.opticsexpress.org/abstract.cfm?URI=oe-15-19-11934
[7] S. Sales, J. Capmany, J. Marti, and D. Pastor, “Experimental demonstration of
fibre-optic delay line filters with negative coefficients,” Electronics Letters, vol. 31,
no. 13, pp. 1095–1096, 1995.
[8] N. You and R. A. Minasian, “A novel high-Q optical microwave processor using
hybrid delay-line filters,” IEEE Transactions on Microwave Theory and Techniques,
vol. 47, no. 7, pp. 1304–1308, 1999.
118
BIBLIOGRAPHY
[9] J. Capmany, D. Pastor, and B. Ortega, “New and flexible fiber-optic delay-line
filters using chirped Bragg gratings and laser arrays,” IEEE Transactions on Microwave Theory and Techniques, vol. 47, no. 7, pp. 1321–1326, 1999.
[10] M. S. Rasras, K.-Y. Tu, D. M. Gill, Y.-K. Chen, A. E. White, S. S. Patel,
A. Pomerene, D. Carothers, J. Beattie, M. Beals, J. Michel, and L. C. Kimerling,
“Demonstration of a Tunable Microwave-Photonic Notch Filter Using Low-Loss
Silicon Ring Resonators,” IEEE Journal of Lightwave Technology, vol. 27, no. 12,
pp. 2105–2110, 2009.
[11] A. W. Fang, H. Park, O. Cohen, R. Jones, M. J. Paniccia, and J. E.
Bowers, “Electrically pumped hybrid AlGaInAs-silicon evanescent laser,” Opt.
Express, vol. 14, no. 20, pp. 9203–9210, Oct 2006. [Online]. Available:
http://www.opticsexpress.org/abstract.cfm?URI=oe-14-20-9203
[12] H. Park, Y. hao Kuo, A. W. Fang, R. Jones, O. Cohen, M. J. Paniccia, and J. E.
Bowers, “A hybrid AlGaInAs-silicon evanescent preamplifier and photodetector,”
Opt. Express, vol. 15, no. 21, pp. 13 539–13 546, Oct 2007. [Online]. Available:
http://www.opticsexpress.org/abstract.cfm?URI=oe-15-21-13539
[13] E. J. Norberg, R. S. Guzzon, S. C. Nicholes, J. S. Parker, and L. A. Coldren, “Programmable Photonic Lattice Filters in InGaAsP-InP,” IEEE Photonics Technology
Letters, vol. 22, no. 2, pp. 109–111, 2010.
[14] R. S. Guzzon, E. J. Norberg, J. S. Parker, L. A. Johansson, and L. A. Coldren,
“Monolithically integrated programmable photonic microwave filter with tunable
inter-ring coupling,” in Proc. IEEE Topical Meeting Microwave Photonics (MWP),
2010, pp. 23–26.
119
Chapter 6
Conclusion and Future Work
6.1
Summary of Thesis
In this thesis, we successfully demonstrated a high-speed hybrid silicon modulator as
a building block in the optical communication as well as an interconnects in future
data transmission architectures. The accomplishments of each chapter are summarized
below.
In Chapter 2, the fundamental physical principle of a carrier depletion modulator
on the hybrid silicon platform are discussed. To introduce efficient index change, the
material composition is determined by considering the bandgap, strain, and PL peak.
In addition, the dimensions of the passive silicon waveguide are chosen to control the
optical mode profile inside the hybrid section. By combining the material properties
and waveguide dimensions, the modulation efficiency of this hybrid silicon modulator
is 2 Vmm at 1550 nm, and is below 3 Vmm from 1500 nm to 1600 nm with extinction
ratios (ER) larger than 13 dB.
120
Chapter 6. Conclusion and Future Work
In order to overcome the RC limit of the lumped electrode, a traveling wave electrode
design is introduced in Chapter 3. We establish a distributed circuit model to analyze the
electrical properties of this electrode design and estimate the modulation bandwidth of
the modulator. Meanwhile, new fabrication techniques for the HSP are also developed to
realize the device. The modulator with CPW electrode design has a device impedance of
18 Ω, a modulation bandwidth of 8 GHz with a 25 Ω termination load, and demonstrated
operation under large signal modulation of 6 dB ER at 10 Gb/s.
After the first demonstration of high-speed modulator using CPW design, an analysis is conducted to improve both the modulation efficiency and frequency response. In
Chapter 4, we replace the CPW transmission line by slotline electrode to implement
push-pull architecture for better modulation efficiency and reduction of device capacitance. In addition, the concept of a capacitively loaded electrode is introduced to adjust
the electrical properties of the device such as device impedance, electrical propagation
loss, and phase velocity. The demonstrated modulator with this electrode design has a
device impedance of 35 Ω, a modulation bandwidth of 25 GHz with a 25 Ω termination load, and demonstrated operation under large signal modulation of 13 dB ER at
40 Gb/s. The transmission experimenta also shows the chirp of -0.75 over the entire
reverse bias range, which can potentially counteract the dispersion in standard single
mode fiber. A fast switch based on a similar device layout is also demonstrated with
0.5 dB power penalty at 40Gb/s while the rise and fall times are all shorter than 20 ps.
As a possible application for this modulator, a microwave filter consisting of amplifiers and phase modulators, is developed in Chapter 5 to demonstrate the potential for
integration on the hybrid silicon platform. With the assistance of low-loss waveguides
and gain elements, tunable filter responses are demonstrated not only in the optical
domain but also in the frequency domain, where responses within a GHz range are only
121
Chapter 6. Conclusion and Future Work
possible with a long delay loop around several millimeters in length.
6.2
6.2.1
Future Work
Transmitter
One of the ultimate goals for this hybrid silicon modulator (HSM) is to integrate it
with hybrid silicon lasers (HSL) to implement a transmitter as a component in optical
interconnects for future applications. A hybrid silicon transmitter can be realized by
integrating the HSL with the HSM demonstrated in this thesis. With the help of 40
Gb/s data rate and 100 nm optical bandwidth, large amount of data can be generated
to fulfill the increasing demands of data capacity while the modulation efficiency is good
enough, more than 10 dB ER, to keep the signal quality over long-haul transmission.
For example, 100 Gb/s data rate can be implemented with four channels operating at
25 Gb/s at different wavelengths.
One of the issues to integrate HSL and HSM is that the PLs for these two devices
are different, where the laser PL is around 1540 nm and modulator PL is around 1360
nm. Conventional integration technique used to integrate different epitaxial layers, such
as regrowth at 700 ◦C, is impossible on the HSP because the bonding temperature is
only 300 ◦C. Any process over 300 ◦C degrades the bonding surface and consequently
affects the device quality.
Among all the technologies, quantum well intermixing (QWI) is one possible approach, where HSLs [1, 2, 3, 4, 5] integrated with EAMs have been demonstrated with
over than 10 dB ER [6, 7]. However, it would require extra effort to shift the laser PL
far enough to meet the PL requirement of the MZM because the well shape changes
122
Chapter 6. Conclusion and Future Work
and the confinement of carriers degrades as the shift increases. Another approach to
realize the integration is selective bonding where different epitaxial layers are bonded to
the patterned silicon wafer separately. This technique does not require further changes
in the design of the epitaxial layers, but development of a process flow is necessary to
protect the silicon waveguides between two III-V epitaxial layers from being damage.
6.2.2
Large-scale optical interconnects
As mentioned in Section 4.4, the developed hybrid silicon switch has low-power consumption, which is critical for large-scale interconnects such as a 32x32 switch. In order
to have a fully functional switch array, additional phase modulators on each arm of the
MZI have to be added to compensate the phase change introduced by the III-V/silicon
tapers. The switch array also requires careful design to protect the passive waveguides
and minimize the propagation loss. Active components can be added to this array in the
future to compensate optical loss while a more complex process flow has to be developed
as mentioned in Section 6.2.1.
6.2.3
Programmable microwave filter
In this thesis, we demonstrate a tunable microwave filter by integrating long-delay loops,
phase modulators and hybrid silicon amplifiers. Various filter functions are implemented
by controlling individual elements on the chip. In the future, a programmable filter
response is possible if the active elements are connected to electronic chips or circuits
with feedback loops. In addition, more complex filter functions can be expected if
multiple unit cells with different lengths of delay loops are cascaded.
123
Bibliography
[1] A. W. Fang, H. Park, O. Cohen, R. Jones, M. J. Paniccia, and J. E.
Bowers, “Electrically pumped hybrid algainas-silicon evanescent laser,” Opt.
Express, vol. 14, no. 20, pp. 9203–9210, Oct 2006. [Online]. Available:
http://www.opticsexpress.org/abstract.cfm?URI=oe-14-20-9203
[2] A. W. Fang, R. Jones, H. Park, O. Cohen, O. Raday, M. J. Paniccia, and
J. E. Bowers, “Integrated AlGaInAs-silicon evanescent race track laser and
photodetector,” Opt. Express, vol. 15, no. 5, pp. 2315–2322, Mar 2007. [Online].
Available: http://www.opticsexpress.org/abstract.cfm?URI=oe-15-5-2315
[3] B. R. Koch, A. W. Fang, O. Cohen, and J. E. Bowers, “Mode-locked
silicon evanescent lasers,” Opt. Express, vol. 15, no. 18, pp. 11 225–11 233,
Sep 2007. [Online]. Available: http://www.opticsexpress.org/abstract.cfm?URI=
oe-15-18-11225
[4] A. W. Fang, E. Lively, Y.-H. Kuo, D. Liang, and J. E. Bowers, “A distributed
feedback silicon evanescent laser,” Opt. Express, vol. 16, no. 7, pp. 4413–4419,
Mar 2008. [Online]. Available: http://www.opticsexpress.org/abstract.cfm?URI=
oe-16-7-4413
[5] D. Liang, M. Fiorentino, T. Okumura, H.-H. Chang, D. T. Spencer, Y.-H.
Kuo, A. W. Fang, D. Dai, R. G. Beausoleil, and J. E. Bowers, “Electricallypumped compact hybrid silicon microring lasers for optical interconnects,” Opt.
Express, vol. 17, no. 22, pp. 20 355–20 364, Oct 2009. [Online]. Available:
http://www.opticsexpress.org/abstract.cfm?URI=oe-17-22-20355
[6] S. R. Jain, M. N. Sysak, G. Kurczveil, and J. Bowers, “Integration of hybrid silicon
DFB laser and electro-absorption modulator using quantum well intermixing,” in
Proc. 36th European Conf Optical Communication (ECOC) and Exhibition, 2010,
pp. 1–3.
[7] S. R. Jain, M. N. Sysak, G. Kurczveil, and J. E. Bowers, “Integrated broadband
hybrid silicon DFB laser array using quantum well intermixing,” in Proc. 22nd
IEEE Int. Semiconductor Laser Conf. (ISLC), 2010, pp. 139–140.
124
Appendices
125
Appendix A
SU8 Dry-Etch Development
SU8 is a polymer popular in micromachined and electronic devices because small open
features can be defined in it lithographically. Some specific types of SU8 are widely
used in MEMS fabrication for high-aspect ratio pillars or vias. However, the adhesion
between SU8 and gold is generally poor compared to other materials such as silicon or
silicon nitride. Since we use this SU8 to open a via between the thin metal contact
on the device and the probe metal, one can expect the surface tension between SU8
and gold to distort the shape of the opening. In this appendix, we will start from a
lithographically defined via and move to dry etch techniques used to form an opening
with straight sidewalls. SU8 3005 around 5 µm thick is used in all processes discussed.
A.1
Lithographically Defined SU8
In order to simulate the shape of a SU8 opening on top of the III-V mesa, we first
prepare a dummy silicon sample with 100 nm gold and 20 nm silicon nitride on top. A
special photoresist (XHRiC) is spun on top of the sample before SU8 coating to eliminate
126
Chapter A. SU8 Dry-Etch Development
(a)
(b)
(c)
undercut
undercut
undercut
Figure A.1: SEM pictures for SU8 (a)After development and 10 minutes O2 descum. (b)Step(a) and 30 minutes 150 ◦ C bake. (c)Step(a) and 30 minutes 220 ◦ C
bake.
any possible reflection from the bottom of the feature during exposure. This sample
is then soft baked, exposed in the auto stepper, and developed in a special developer.
The sample undergoes a 10 minutes O2 descum to remove SU8 residue before the first
inspection. The cross section of the via after this step is shown in Figure A.1(a). With
the exposure condition of 0.8 second and focus offset of -30, the via is undercut at
the interface between the XHRiC and SU8. This is particularly undesirable for the
later plating process due to the discontinuity between the sidewall and the bottom. If
the metal seed layer at the bottom of the via is not conductive, gold will not grow
at that surface and this results in a discontinuity between probe metal and the mesa.
Figure A.1(b) and (c) are the via shapes after a 30 minute bake at 150 ◦ C and 220 ◦ C,
respectively. As can be seen, the profile does not change too much and the undercut
of SU8 is still present. We also calibrate the process with different variables, such as
exposure time, exposure focus, surface treatment before coating and descum condition,
but none of them show straight sidewalls. Therefore, it is necessary to develop another
technique to eliminate the undesired undercut.
127
Chapter A. SU8 Dry-Etch Development
Bias power: 50W
Bias power: 100W
Bias power: 150W
Bias power: 200W
Figure A.2: SU8 profile at different bias power. The ICP power is 350 W, pressure
is 1 Pa with O2 /CF4 =40/4 sccm.
A.2
SU8 Dry-Etch Development
In general, people use polymer dry etches to remove residue or to blanket etch the
sample for certain purposes. Research has been done for BCB to create small feature
sizes, but there have been no reports of features with high aspect ratio. Therefore, it
is not possible to use an already-published process flow. Also, the polymer dry etch
process is strongly dependent on the etch conditions. The etch rate and profile strongly
vary based on the tool parameters. To start the calibration, a 300 nm Ebeam deposited
128
Chapter A. SU8 Dry-Etch Development
1 Pa
0.5 Pa
Figure A.3: SU8 profile at different etch pressure. The ICP power is 350 W, bias
power is 50 W with O2 /CF4 =40/4 sccm.
layer of Titanium (Ti) is used as the hard mask and the patterns are created using a liftoff technique. We set the initial ICP power to 350 W because the sample temperature
increases very quickly if the ICP power is high. In general, oxygen is a very common
gas used to etch polymer while small amount of CF4 is added to keep the etch rate
consistent over long etch time. Without any CF4 , the fluorinated polymer generated
during the process will be resistive to oxygen and hence reduce the etch rate.
The profile of the via after being etched at different bias powers is shown in Fig-
Figure A.4: SU8 profile using hard mask.
129
Chapter A. SU8 Dry-Etch Development
ure A.2 with O2 /CF4 flow of 40/4 sccm and pressure of 1 Pa. The etch is performed
for 15 minutes while a 100 sccm nitrogen (N2 ) purge is added to the process after every
5 minutes of O2 /CF4 etching. This is used to remove accumulated charges on the edge
of the Ti hard mask (in the location indicated by the circle), which block ions from
entering the via during etch and affects the etch. It is obvious that all the bias conditions successfully create a via opening with high-aspect ratio. The etch, however, is
not feasible since there is a lot of polymer residue (“grass”) at the bottom due to the
accumulated charge at the edge of the Ti mask.
To decrease the charging effect, we first try to reduce the plasma density in the
chamber by lowering the pressure from 1 Pa to 0.5 Pa. The result of a 15 minute
etch with N2 interval purge is shown in Figure A.3. As can be seen, there is barely
any improvement with lower pressure. In fact, the via width becomes larger with low
pressure because the generated ions are less directionally confined.
To totally eliminate the charge effect that results in the grass, the Ti hard mask
has to be replaced by another nonconductive material. Among all of the dielectrics,
silicon dioxide (SiO2 ) and silicon nitride (SiN) are the most popular materials used as
hard masks. SiO2 , however, has very poor adhesion to SU8. Hence, we use a 300 nm
PECVD SiN (240 ◦ C) to replace the Ti. The etch conditions are similar to the ones
used with the Ti mask (O2 /CF4 = 40/4 sccm, 1 Pa, ICP power: 350W, bias power:
50W). Figure A.4 shows that the polymer grass disappears after using the SiN hard
mask while still maintaining a high-aspect ratio profile. The etch is easy, repeatable,
and the etch rate is around 0.4 µm/min with the selectivity between SU8 and SiN more
than 1:40.
130
Документ
Категория
Без категории
Просмотров
0
Размер файла
46 756 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа