UC San Diego

UC San Diego Electronic Theses and Dissertations

Title
High-Speed Hybrid Silicon-Lithium Niobate Electro-Optic Modulators & Related
Technologies

Permalink
https://escholarship.org/uc/item/3vj7v7zp

Author
Weigel, Peter Orlando

Publication Date
2018

Peer reviewed|Thesis/dissertation

eScholarship.org Powered by the California Digital Library

University of California
 UNIVERSITY OF CALIFORNIA SAN DIEGO

High-Speed Hybrid Silicon-Lithium Niobate Electro-Optic Modulators

& Related Technologies

A dissertation submitted in partial satisfaction of the

requirements for the degree
Doctor of Philosophy

in

Electrical Engineering (Photonics)

by

Peter O. Weigel

Committee in charge:

Professor Shayan Mookherjea, Chair

Professor Joseph Ford
Professor James Friend
Professor George Papen
Professor Stojan Radic
Professor Gabriel Rebeiz

2018
 Copyright
Peter O. Weigel, 2018
All rights reserved.
The dissertation of Peter O. Weigel is approved, and it is
acceptable in quality and form for publication on microfilm
and electronically:

Chair

University of California San Diego

2018

iii
 DEDICATION

To my parents, for supporting my education throughout my entire life, even when

it took me three thousand miles away.

This work is also dedicated to my friend Shane, who departed this world far too

soon. All my successes are shared with him.

iv
 EPIGRAPH

Give me a place to stand, and I shall move the world.

–Archimedes (from The Library of History of Diodorus Siculus, Fragments of Book XXVI, as
translated by F. R. Walton in Loeb Classical Library (1957) Vol. XI)

v
 TABLE OF CONTENTS

Signature Page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

Epigraph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii

Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xx

Abstract of the Dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxii

Chapter 1 Theory of Electro-Optic Modulators . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Lithium Niobate & the Pockels Effect . . . . . . . . . . . . . . . . . . 2
1.2.1 Voltage-Length Product in LN MZMs . . . . . . . . . . . . . . 5
1.2.2 Electro-Optic Bandwidth Theory for a Traveling Wave Mach-
Zehnder Modulator . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Silicon & the Plasma Dispersion Effect . . . . . . . . . . . . . . . . . 13
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

Chapter 2 Design of Hybrid Electro-Optic Modulators . . . . . . . . . . . . . . . . . . 16

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Hybrid Si-LN Optical Mode Design . . . . . . . . . . . . . . . . . . . 17
2.2.1 Optical Mode for the Electro-Optic Modulator . . . . . . . . . 19
2.2.2 Optical Mode for Si Waveguide Applications . . . . . . . . . . 20
2.2.3 Optical Mode for the Edge Coupler . . . . . . . . . . . . . . . 21
2.2.4 Bonding Markers & Electrode Markers . . . . . . . . . . . . . 23
2.3 Full Optical Design of the EOM . . . . . . . . . . . . . . . . . . . . . 23
2.4 Design of the Electrical Lines . . . . . . . . . . . . . . . . . . . . . . 24
2.4.1 EOM Region . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.2 Electrical Bends . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.3 Electrical Pads . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

vi
Chapter 3 A Process Design Kit for Hybrid Si-LiNbO3 Devices . . . . . . . . . . . . . 28
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2 Technology Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.1 Features of the Technology . . . . . . . . . . . . . . . . . . . . 29
3.3 Design Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Process Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.5 Design Rules for the Hybrid Si-LN Platform . . . . . . . . . . . . . . . 34
3.5.1 Non-Hybrid Si Components . . . . . . . . . . . . . . . . . . . 34
3.5.2 Hybrid Si-LN Components . . . . . . . . . . . . . . . . . . . . 35
3.6 Static Library Components . . . . . . . . . . . . . . . . . . . . . . . . 38
3.6.1 Edge Coupling Taper . . . . . . . . . . . . . . . . . . . . . . . 39
3.6.2 Thin-Si Waveguide . . . . . . . . . . . . . . . . . . . . . . . . 39
3.6.3 Thin-Si Spline Bends . . . . . . . . . . . . . . . . . . . . . . . 39
3.6.4 Thin-Si 3-dB Directional Coupler . . . . . . . . . . . . . . . . 41
3.6.5 Mach-Zehnder Interferometer . . . . . . . . . . . . . . . . . . 44
3.6.6 Hybrid Si-LN Taper . . . . . . . . . . . . . . . . . . . . . . . 45
3.6.7 Hybrid Si-LN Waveguide . . . . . . . . . . . . . . . . . . . . . 46
3.6.8 Hybrid Si-LN Electro-Optic Phase Shifter . . . . . . . . . . . . 47
3.6.9 Coplanar Waveguide Transmission Line . . . . . . . . . . . . . 47
3.7 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

Chapter 4 Fabrication of Hybrid Electro-Optic Devices . . . . . . . . . . . . . . . . . . 51

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2 Film-on-Insulator Wafers . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.3 SOI Patterning & Oxide Polishing . . . . . . . . . . . . . . . . . . . . 53
4.4 Bonding Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4.1 Types of Bonds . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.5 Reducing Thermal Stress in Heterogeneous Bonded Materials . . . . . 58
4.6 Handle Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.7 Metallization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.8 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

Chapter 5 All-Optical Hybrid Silicon-Lithium Niobate Devices . . . . . . . . . . . . . 75

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2 Hybrid Modes and Mode-Transition Tapers . . . . . . . . . . . . . . . 78
5.3 Waveguides, Directional Couplers and Photonic Circuits . . . . . . . . 83
5.4 Discussion and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.5 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

vii
Chapter 6 Characterization of Hybrid Electro-Optic Modulators . . . . . . . . . . . . . 92
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.2 Optical Characterization of an Electro-Optic Modulator . . . . . . . . . 92
6.2.1 Optical Transmission through an Electro-Optic Modulator . . . 94
6.2.2 Propagation Loss . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.3 Spline Bending Loss . . . . . . . . . . . . . . . . . . . . . . . 96
6.2.4 Method to Optimize Optical Polarization State . . . . . . . . . 97
6.3 Electrical Characterization of an Electro-Optic Modulator . . . . . . . . 98
6.3.1 Lossy Transmission Line Theory . . . . . . . . . . . . . . . . . 100
6.3.2 Electrical Extraction from VNA Measurements . . . . . . . . . 103
6.3.3 Measuring Vπ L . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.3.4 Low Frequency Electro-Optic Response . . . . . . . . . . . . . 105
6.3.5 High Frequency Electro-Optic Response . . . . . . . . . . . . . 105
6.4 Electro-Optic Characterization of an Electro-Optic Modulator . . . . . 106
6.4.1 Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.4.2 Electro-Optic Mach-Zehnder Modulator Design . . . . . . . . . 109
6.4.3 Electro-Optic Measurements . . . . . . . . . . . . . . . . . . . 110
6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.6 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

Chapter 7 Second Harmonic Generation in Hybrid Waveguides . . . . . . . . . . . . . 118

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.2 Optical Second-Harmonic Generation . . . . . . . . . . . . . . . . . . 119
7.2.1 Derivation of Power at the Generated Second Harmonic Frequency119
7.3 Design of Folded Hybrid Silicon Carbide-Lithium Niobate Waveguides
for Efficient Second Harmonic Generation . . . . . . . . . . . . . . . . 128
7.3.1 A Brief Background of Integrated Photonic Circuits and Nonlin-
ear Optics in LN . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.3.2 Overview of the Proposed Structure . . . . . . . . . . . . . . . 131
7.3.3 Additional Theory of the Nonlinear Interaction Region . . . . . 133
7.3.4 Silicon Carbide-Lithium Niobate Second Harmonic Generation
Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.3.5 Impact of Variations of Design Parameters . . . . . . . . . . . . 136
7.3.6 Loss and Disorder . . . . . . . . . . . . . . . . . . . . . . . . 137
7.3.7 Bent Hybrid Waveguide Region . . . . . . . . . . . . . . . . . 139
7.3.8 Effective Mode Area (Aeff ) . . . . . . . . . . . . . . . . . . . . 142
7.3.9 Lithium Niobate Confinement Factor (LNCF) . . . . . . . . . . 143
7.4 Concluding Thoughts . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.5 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

viii
Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
A.1 Alternative Vπ L Derivation . . . . . . . . . . . . . . . . . . . . . . . . 149
A.2 Deriving the Electro-Optic Response from Coupled Mode Theory . . . 151
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

ix
 LIST OF FIGURES

Figure 1.1: Top view of a standard MZM in the push-pull configuration, meaning the two
arms of the MZM have (ideally) equal and opposite electro-optic phase shifts
(∆φ = the total phase shift of the device). Gray lines represent the optical
waveguides, gold boxes represent the electrical lines (“electrodes”)... . . . . . 2
Figure 1.2: Vπ L is evaluated versus: neff (a), G (b), and Γmo (c). . . . . . . . . . . . . . . . 9
Figure 1.3: Circuit model showing where the electro-optic modulator fits between the signal
generator (with applied voltage Va and internal impedance of ZG ) and the load
impedance ZL . The voltage along the modulator is Vm (z), which varies along
the length of the EOM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Figure 1.4: The results shown in this figure are simulated 3-dBe electrical bandwidth values,
in GHz, based on Equation 1.23. a Assumes perfect index matching (∆n = 0)
and perfect impedance matching (ZG = Z0 = ZL = 50 Ω). Each curve represents
a different electrical conductor loss... . . . . . . . . . . . . . . . . . . . . . . 13

Figure 2.1: a A schematic representation of the hybrid Si-LN EOM cross-section in the
push-pull configuration. Material labels are provided below the image, and ma-
terial dimensions are as follows: hSi = Si waveguide height; wSi = Si waveguide
width; hsub = height of SOI substrate... . . . . . . . . . . . . . . . . . . . . . . 18
Figure 2.2: Simulated LNCF (a) and Aeff (b) for combinations of wSi and hLN , with
hSi = 150 nm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Figure 2.3: a Optical mode with wSi = 650 nm, hLN = 600 nm, and hSi = 150 nm, under the
LN film. b Same mode as a, but outside of the LN region. c Top view of the
transition from the EOM mode (narrow 320 nm wide Si waveguide, left side),
to a wide Si-LN waveguide, and finally to a wide all-Si mode. . . . . . . . . . 21
Figure 2.4: a Simulation of power transmission of an edge coupler between a lensed tapered
fiber with MFD = 2.5 µm and a Si waveguide of height 150 nm surrounded
by SiO2 . The maximum power coupling is 89% for a Si width of 220 nm. b
Logarithmic plot of the normalized electric field component... . . . . . . . . . 22
Figure 2.5: a GDS layout of EOM chip with arrays of alignment markers for electrode
mask alignment and LN bonding alignment. b Edge-to-edge alignment is when
the chip marker (black) and the mask marker (gray) are designed to align at the
edge. This allows for high precision alignment, since both markers are visible. 23
Figure 2.6: Top view of the full EOM layout. DC = directional coupler, PLD = path-length
difference. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Figure 2.7: This is the GDS layout of a portion of the electrical lines, shown in white on
a black background. The ground lines are expanded at the pads to ensure that
the probes can make contact, and the signal line is left unaltered for minimal
reflections. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

Figure 3.1: a Linear waveguide taper optical transmission simulation in MODE. b Di-
rectional coupler waveguide optical simulation in FDTD. c Direct current
(DC) electric-field simulation in DEVICE. d Circuit simulation of electro-optic
modulator in INTERCONNECT. . . . . . . . . . . . . . . . . . . . . . . . . . 31

x
Figure 3.2: Cross-section after bonding and thermal anneal. Each layer’s thickness is given,
along with resistivity values for the Si handles. . . . . . . . . . . . . . . . . . 34
Figure 3.3: Cross-section after the full fabrication process. Each layer’s thickness is given,
along with resistivity values for the Si handles. . . . . . . . . . . . . . . . . . 35
Figure 3.4: a Minimum edge waveguide spacing. b Maximum Si waveguide width. c
Minimum spline bend radius. . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Figure 3.5: a Linear plot of a simulated hybrid mode. wSi = 320 nm and hLN = 600 nm. The
first and second rows of plots are x, y, and z components of the E- and H-fields,
respectively. The third row of plots, labeled Sx , Sy , and Sz , are components of
the time-average instantaneous Poynting vector. . . . . . . . . . . . . . . . . . 36
Figure 3.6: Log plot of a simulated hybrid optical mode when wSi = 200 nm and hLN = 900
nm, i.e. one of the gray boxes in the upper right corner of Table. 3.2. In this
case, the Ex , Hy , and Hz components are nearly slab modes, while the Ey , Ez ,
and Hx components are non-zero. . . . . . . . . . . . . . . . . . . . . . . . . 38
Figure 3.7: a Top view of optimized input/output waveguide taper. Taper length Ltaper
is varied from 10 µm to 500 um in MODE Solutions. b Plot of edge taper
transmission versus taper length at λ = 1550 nm for the taper in a. . . . . . . . 39
Figure 3.8: Simulation of Enorm (the norm of all three E-field components) of the non-
hybrid mode. Since Ey and Ez are so small for this TE-like mode, Enorm ≈ Ex .
wSi = 650 nm. As with all Si waveguides in this document, hSi = 150 nm. . . . 40
Figure 3.9: Cubic spline curve with knots P0 and P3 , and control points P1 and P2 . . . . . . 40
Figure 3.10: a 90◦ spline bend with knots P0 and P3 , and control points P1 and P2 . b Cubic
spline waveguide 90◦ bend with width wSi and characteristic length Lc . Lc is
measured from the middle of the input waveguide to the edge of the output
waveguide (or vice versa). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Figure 3.11: a Cubic spline S bend. b Cubic spline waveguide S bend with width wSi , lateral
offset Lx , and longitudinal length Ly . . . . . . . . . . . . . . . . . . . . . . . 42
Figure 3.12: Generic directional coupler. Lcoup is the straight-section coupling length, and
Lex is the additional coupling length in the region that angles away from the
straight region. The coupling matrix comes from section 6.2.2 in [4]. . . . . . 42
Figure 3.13: a GDS layout of a symmetric directional coupler. Lcoup = 50 µm, Lex = 40
µm of 90◦ spline bends, the gap in the straight region is 280 nm, and each
waveguide has a width of 650 nm. b Coupling coefficient |κ|2 for the coupler
in a as a function of wavelength. . . . . . . . . . . . . . . . . . . . . . . . . . 43
Figure 3.14: a Schematic of an asymmetric directional coupler. Coupling occurs most
strongly in section L2 , less strongly in sections L1 and L3 , and not at all in
section L4 . b This plot shows the transmission through C and D for an input in
B. Total transmission is around 98% across the full wavelength range... . . . . 44
Figure 3.15: Top view of an MZI structure. Black lines represent waveguides and the path-
length difference on the top arm is highlighted in yellow. Each 50/50 splitter is
either a symmetric DC or an asymmetric DC (there are other types of WG-based
splitters, but they are not part of this design kit). . . . . . . . . . . . . . . . . . 45

xi
Figure 3.16: a Ex field of optical mode in hybrid Si-LN region with wide (wSi = 650 nm)
Si. b Narrow (wSi = 320 nm) Si. c Top view of hybrid Si-LN taper. The taper
is in the Si ridge layer, and LN film is bonded over the entire area. d Plot of
transmission versus taper length. . . . . . . . . . . . . . . . . . . . . . . . . . 45
Figure 3.17: a Plot of simulated effective mode area (left axis, black circles) and LNCF
(right axis, red circles). b Plot of simulated effective index (left axis, black
circles) and group index (right axis, red squares). The dashed vertical blue line
in each part shows where the mode we chose to use sits. . . . . . . . . . . . . 47
Figure 3.18: 3D depiction of the final cross-section from Fig. 3.3, but for the single Si
waveguide of a hybrid Si-LN EO phase shifter. Only the straight section of the
PS is shown here. Each layer’s thickness is given, along with resistivity values
for the Si handles. The EO PS has length “L.” . . . . . . . . . . . . . . . . . . 48
Figure 3.19: Plot of (3.4) when varying: a Characteristic impedance Z0 , with ∆n = 0.10 and
αRF = 1 dB/(cm-GHz1/2 ); b RF losses αRF , with Z0 = ZG = ZL = 50 Ω and ∆n
= 0.10; and c RF-to-optical index difference ∆n, with Z0 = ZG = ZL = 50 Ω and
αRF = 1 dB/(cm-GHz1/2 ). L = 1 cm for all simulations. . . . . . . . . . . . . . 49

Figure 4.1: The basic crystal ion slicing process is depicted in four steps: a The first wafer
(W1) is ion-implanted. b W1 is bonded to the second wafer (W2), comprised
of some handle material covered with a thin, smooth insulator material. c The
bonded stack is heated until W1 cracks off (“slicing”)... . . . . . . . . . . . . . 52
Figure 4.2: a A standard 220 nm SOI wafer from SOITEC is thinned to 150 nm with RIE.
b Photoresist is spun on the wafer and patterned with DUV lithography. c RIE
is used to etch the photoresist pattern into the thin Si layer. d Approximately
1 µm of oxide is deposited via HDP CVD onto the patterned wafer. . . . . . . 54
Figure 4.3: a SEM image of an over-polished waveguide. The measured value of only
20 nm is 130 nm below the targeted thickness, and is too thin to guide light.
b Wafer map of oxide thickness over Si features after CMP and wet etch in
optimized process flow (Fig. 4.2f). . . . . . . . . . . . . . . . . . . . . . . . . 54
Figure 4.4: a The thin-film stack used in hybrid photonic devices based on silicon-on-
insulator (SOI) and lithium niobate-on-insulator (LNOI). b A simplified bulk
LN-BCB-Si bonded structure used in the calculations shown in Fig. 4.5. . . . . 59
Figure 4.5: a Calculated maximum stresses in the upper-handle layer (LN or Si) of the
complete thin-film stack and the simplified LN-BCB-Si stack shown in Fig. 4.4.
Stress along each axis of the LN handle is nearly identical for the full stack and
for the simplified stack, suggesting that the details... . . . . . . . . . . . . . . 62
Figure 4.6: Depiction of LNOI with a Si handle, after implantation and before splitting off
the bulk LN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Figure 4.7: a through e show camera images of 1 cm × 1 cm pieces of LN or LNOI that
were bonded to bulk Si pieces of a similar size through a 1 µm BCB layer. a
is bulk LN, b is LNOI with a LN handle, and c through e are LNOI with a Si
handle, which has been removed by dry-etching... . . . . . . . . . . . . . . . . 64

xii
Figure 4.8: Energy-dispersive x-ray (EDX) spectrum analysis. The left plot is of a Si-
backed LNOI sample from the manufacturer, and the right plot is of a LN film
bonded through BCB to bulk Si. These two samples exhibit similar spectra,
indicating a similarity in material properties. . . . . . . . . . . . . . . . . . . 66
Figure 4.9: a The patterned SOI sample, bonded to a 4.2 mm × 16 mm film of LN through
a thin BCB layer. b A zoomed in image of a corner of the bonded region. c
The specific device we measure, magnified from a with sections of unchanging
straight waveguide removed for visual convenience. . . . . . . . . . . . . . . . 67
Figure 4.10: a Wafer cross-section as received from Sandia National Laboratories, as de-
picted in Fig. 4.2f. b Bond the LN-on-insulator (LNOI) die (from a NanoLN
wafer) to an SOI die, singulated from the Sandia wafer. c Remove the handle
(SiO2 and substrate) from the LN film after bonding. . . . . . . . . . . . . . . 68

Figure 5.1: Hybrid LN-Si photonic circuits. a Silicon photonic components were fabri-
cated using deep ultraviolet (DUV) lithography. Singulated dies (size: 25 mm ×
16 mm) were bonded to diced pieces (size: 21 mm × 17 mm) of an unpatterned
Lithium Niobate (LN)-on-insulator wafer. . . . . . . . . . . . . . . . . . . . . 77
Figure 5.2: Waveguide modes. a Calculated hybrid optical mode profiles for different
Si rib widths. The panels show the magnitude of the electric field in the TE
polarization, with the E-field vector oriented along the crystal axis. As the Si
rib width w decreases, the modal effective index (neff ) decreases... . . . . . . . 79
Figure 5.3: Adiabatic mode transition. a Gradual linear reduction of the Si rib width to
transition between cross-sections A and B. b Numerical calculations of the
modal effective index versus waveguide width, with yellow circles indicating
the initial and final points of the taper. . . . . . . . . . . . . . . . . . . . . . . 81
Figure 5.4: Waveguide characterization. a Transmission measurements of test struc-
tures consisting of 650 nm wide Si waveguides with SiO2 cladding (labeled
‘Si/SiO2 ’). b Transmission versus length of waveguides and bends (paper-clip
structures)... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Figure 5.5: Hybrid Si-LN optical circuit. a Optical microscope image of an interferome-
teric hybrid Si-LN circuit which uses both waveguide cross-sections (C.S.) A
and B, four adiabatic transitions (Transition) in each lightpath, two directional
couplers (D.C.), and twenty 90-degree bends in each lightpath. . . . . . . . . . 86
Figure 5.6: Si/SiO2 Mach-Zehnder interferometer (MZI). a An SiO2 -clad Si photonic
MZI was fabricated on the same chip but outside the bonded LN region. The
input-output waveguides pass under the LN bonded region and emerge on the
other facet of the chip. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

Figure 6.1: a Schematic of the EOM (not to scale, not showing electrodes), including
two 3-dB directional couplers (DC) and a waveguide segment for path-length
difference (PLD). Three optical waveguide modes are used, labeled as A, B,
and C. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

xiii
Figure 6.2: a Optical transmission through EOM device with greater than 15 dB extinction
ratio across the entire measured wavelength range of 1525 nm to 1575 nm. b
Pictorial representation of optical setup. SM = single mode, PM = polarization
maintaining, and L/T = lensed/tapered. . . . . . . . . . . . . . . . . . . . . . 94
Figure 6.3: Transmission for various waveguide lengths at λ = 1550 nm is shown for the
all-Si mode (a, Mode A) and the Si-LN mode (b, Mode C). Propagation loss
is significantly lower for the hybrid, narrow-Si mode, at only 0.64 dB/cm
compared with 1.34 dB/cm for the all-Si mode... . . . . . . . . . . . . . . . . 95
Figure 6.4: a Measured (circles) spline bending loss for various characteristic lengths (Lc )
at λ = 1550 nm. Lines represent linear fits of transmission measurements for a
number of devices with varying amounts of quarter-circle spline bends. . . . . 97
Figure 6.5: a Scattering matrix representation of a two-port electrical network. b ABCD
matrix representation of a two-port electrical network. c Circuit schematic for a
long, lossy transmission line of length d. R, L, C, and G are the per unit length
series resistance, series inductance, shunt capacitance... . . . . . . . . . . . . . 100
Figure 6.6: a Measured electrical S-parameters of the EOM’s coplanar-waveguide transmis-
sion line. b Left y-axis: extracted microwave phase index nm and microwave
loss αm (units of dB/cm). Right y-axis: characteristic impedance Zc (Ω). . . . . 103
Figure 6.7: A plot of optical power versus applied dc voltage. The dc voltage supply is
a Keithley 2420. Device length is 5 mm. ER = extinction ratio. Fitting was
performed in MATLAB. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Figure 6.8: a Experimental setup for low frequency modulation response. PD = photode-
tector. Red arrows represent the optical signal, black arrows represent the
electrical signal. b Measured response of a low frequency EOM from 100 MHz
up to 6 GHz (the limit of the VNA). . . . . . . . . . . . . . . . . . . . . . . . 106
Figure 6.9: a Thin film x-cut lithium-niobate (LN) on insulator dies were bonded at room
temperature to segmented dies of a patterned and planarized silicon-on-insulator
(SOI) wafer which contained fabricated silicon photonic waveguide circuits.
No etching or patterning of the LN film was performed. . . . . . . . . . . . . 108
Figure 6.10: a Normalized optical transmission of the Mach-Zehnder interferometric electro-
optic Modulator (MZM), versus dc voltage at optical wavelength = 1560 nm.
Fitted Vπ L = 6.7 V.cm for device length L = 0.5 cm. . . . . . . . . . . . . . . 111
Figure 6.11: Measured electro-optic response up to 106 GHz. . . . . . . . . . . . . . . . . 113

Figure 7.1: Measured n and k curves of PECVD amorphous SiC. . . . . . . . . . . . . . . 132

Figure 7.2: a Cross-section of hybrid SiC-LN waveguide. b Top view of our proposed
structure, consisting of a quasi-phase matched region (the “PPLN” section), an
adiabatic taper to convert the SiC width from a narrow rib to a wide rib, and
two 90◦ spline bends to wrap the SiC back around into the PPLN section. . . . 133
Figure 7.3: a Sweep of η0 versus SiC width for SiC heights of 100 nm, 150 nm, 200 nm,
and 250 nm. b Mode pictures for wSiC = 120 nm and hSiC = 250 nm for the SH
mode (top) and FH mode (bottom). . . . . . . . . . . . . . . . . . . . . . . . 135
Figure 7.4: How the phase matching FH wavelength shifts as a function of wSiC , hSiC , hLN ,
and temperature is shown in a, b, c, and d, respectively. . . . . . . . . . . . . . 136

xiv
Figure 7.5: Simulated PSH when loss is added to the model (a) and when duty cycle error is
added to the model (b). In both cases, the input power (PFH ) is 10 mW and the
nonlinear interaction length is assumed to be 1 cm. In b, each curve represents
a different RMS poling period error, from 0 nm (no error) to 250 nm. . . . . . 138
Figure 7.6: a Top view of the 180◦ bend, plus tapers and additional straight waveguides
(“Str.”), connecting adjacent QPM NLO sections. The phase along the structure
is tracked in b to ensure that an integer multiple of ∆k = 2π rad. occurs at the
output of the bend. c Plot of device area AD versus Lc ... . . . . . . . . . . . . 139
Figure 7.7: Simulated spline bending losses for the SH and FH waves are shown in parts a
and b, respectively. Losses are significantly lower for the SH wave because of
its reduced mode size. Simulated taper losses are shown in part c for both SH
and FH waves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Figure 7.8: Sweep of Aeff versus SiC width at λ = 775 nm (part a) and λ = 1550 nm (part b)
for SiC heights of 100 nm, 150 nm, 200 nm, and 250 nm. c and d show mode
pictures for wSiC = 250 nm and hSiC 250 nm (part c) and wSiC = 150 nm and
hSiC 250 nm (part d) for the SH mode (top) and the FH mode (bottom). . . . . 143
Figure 7.9: Sweep of LNCF versus SiC width at λ = 775 nm (part a) and λ = 1550 nm
(part b) for SiC heights of 100 nm, 150 nm, 200 nm, and 250 nm. c and d show
mode pictures for wSiC = 100 nm and hSiC 100 nm (part c) and wSiC = 200 nm
and hSiC 100 nm (part d) for the SH mode (top) and the FH mode (bottom). . . 145

xv
 LIST OF TABLES

Table 3.1: Process outline to fabricate hybrid Si-LN devices (active and passive). . . . . . 32
Table 3.2: Table of acceptable combinations of values for Si width and LN thickness. In
the upper right of the table, one or more of the field components (Ex , Ey , Ez , Hx ,
Hy , or Hz ) has coupled into the LN slab. In the lower section of the table, TE2
exists and can couple with TE0 . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

Table 4.1: Material properties of all materials used in our numerical calculations of thermally-
induced stress. All data is taken at room temperature. E is the modulus of
elasticity, and α is the coefficient of thermal expansion (CTE). x, y, and z are
used to denote the crystal axes of anisotropic LN. . . . . . . . . . . . . . . . . 60

Table 7.1: Fitted parameters of the Tauc-Lorentz model for our PECVD SiC film. See [17]
for more details on the Tauc-Lorentz model. . . . . . . . . . . . . . . . . . . . 132

xvi
 ACKNOWLEDGEMENTS

I would like to thank my adviser, Shayan Mookherjea, for overseeing this thesis work. He
knows better than anyone the struggles I have suffered and the successes I have enjoyed over the
years. I can not thank him enough for his support and tutelage during my graduate studies.
I would also like to thank all of my co-authors for their assistance throughout this experience.
The Sandia team led by Anthony Lentine has been nothing short of superb. There is no doubt
that this work would never have been anything more than an amalgamation of simulations if not
for the efforts of Anthony and others. In particular, I would like to thank Douglas Trotter for
patiently answering all of my fabrication-related questions, even when it meant getting pulled into
an unexpected Friday afternoon conference call. Sri Sriram and Vincent Stenger at Srico, Inc., thank
you for allowing me to spend a few weeks at your headquarters in Ohio. There’s really nothing quite
as impressive as seeing how a small business makes beautiful products. To my collaborators here at
UC San Diego: thank you for the (sometimes long) talks about high-speed electrical measurements
and components, and for sharing equipment, sometimes the day before a deadline. More than once.
Cooper Levy, Kelvin Fang, and Hasan Al-Rubaye – I would still be staring at noisy 6 GHz VNA
measurements if not for your help.
To the group members who guided me in my early years, I owe you a debt I can never
repay. Ryan Aguinaldo developed many of the early simulation techniques for this project before
I joined the research group, and freely offered his time and expertise to get me started. I cherish
our meandering, sometimes even circular, discussions on the meaning of Maxwell’s equations.
Marc Savanier’s knowledge of nonlinear optical theory and fabrication experience were invaluable
throughout the course of this research. It was truly an honor to work alongside him. And of course I
would be remiss not to mention Ranjeet Kumar and Jun Rong Ong for their constant willingness to
see me succeed and to help where they could (as well as the not-so-infrequent outing to Buffalo
Wild Wings whenever Ranjeet decided he wasn’t vegetarian for a day).
Of course, this acknowledgments section would be incomplete without mentioning the
efforts of Jie Zhao, another PhD student in our research group. Jie’s tireless efforts, particularly in

xvii
the clean room, have gotten us through many a tight deadline. I look forward to seeing her own
thesis work progress over the next couple of years.
Chapter 3, in part, is a reprint of the material as it appears in Conference on Lasers and
Electro-Optics (CLEO) 2017. Peter O. Weigel and Shayan Mookherjea, paper JTu5A.126. The
dissertation author, together with his adviser, led the research efforts for this work and co-authored
the paper.
Section 4.5 of Chapter 4, in full, is a reprint of the material as it appears in Optical
Materials 2017. Peter O. Weigel & Shayan Mookherjea, Optical Materials Vol. 66, 605-610 and in
Conference on Lasers and Electro-Optics (CLEO) 2017. Peter O. Weigel and Shayan Mookherjea,
paper. SM3K.1. The dissertation author, together with his adviser, led the research efforts for this
work and co-authored the papers.
Chapter 5, in part, is a reprint of the material as it appears in Scientific Reports 2016. Peter
O. Weigel, Marc Savanier, Christopher T. DeRose, Andrew T. Pomerene, Andrew L. Starbuck,
Anthony L. Lentine, Vincent Stenger & Shayan Mookherjea, Scientific Reports 6, 22301. The
dissertation author, together with his adviser, led the research efforts for this work and co-authored
the paper.
Chapter 6, in part, is currently being prepared for submission for publication of the material.
Peter O. Weigel, Jie Zhao, Kelvin Fang, Hasan Al-Rubaye, Douglas Trotter, Dana Hood, John
Mudrick, Christina Dallo, Andrew T. Pomerene, Andrew L. Starbuck, Christopher T. DeRose,
Anthony L. Lentine, Gabriel Rebeiz and Shayan Mookherjea. Portions of Chapter 6 are reprinted
from Conference on Lasers and Electro-Optics (CLEO) 2018. Peter O. Weigel, Jie Zhao, Douglas
Trotter, Dana Hood, John Mudrick, Christina Dallo, Andrew T. Pomerene, Andrew L. Starbuck,
Christopher T. DeRose, Anthony L. Lentine and Shayan Mookherjea, paper SF2I.4. The dissertation
author, together with his adviser, led the research efforts for this work and co-authored the papers.
Chapter 7, in part, is a reprint of the material as it appears in Journal of the Optical Society
B 2018. Peter O. Weigel & Shayan Mookherjea, JOSA B 35(3), 593-600 and in Conference on
Lasers and Electro-Optics (CLEO) 2017. Peter O. Weigel, Marc Savanier and Shayan Mookherjea,

xviii
paper SW4M.5. The dissertation author, together with his adviser, led the research efforts for this
work and co-authored the papers.

xix
 VITA

2013 B.S. in Electrical Engineering summa cum laude, Lehigh University

2014 M.S. in Electrical Engineering, University of California San Diego
2018 Ph.D. in Electrical Engineering (Photonics), University of California San
Diego

PUBLICATIONS

Peter O. Weigel, Jie Zhao, Kelvin Fang, Hasan Al-Rubaye, Douglas Trotter, Dana Hood, John
Mudrick, Christina Dallo, Andrew T. Pomerene, Andrew L. Starbuck, Christopher T. DeRose,
Anthony L. Lentine, Gabriel Rebeiz and Shayan Mookherjea, “Bonded Thin Film Lithium Niobate
Modulator on a Silicon Photonics Platform Exceeding 100 GHz Bandwidth,” Manuscript under
review.
Peter O. Weigel, Jie Zhao, Douglas Trotter, Dana Hood, John Mudrick, Christina Dallo, Andrew
T. Pomerene, Andrew L. Starbuck, Christopher T. DeRose, Anthony L. Lentine and Shayan
Mookherjea, “Foundry-compatible Hybrid Silicon / Lithium Niobate Electro-Optic Modulator,” in
Conference on Lasers and Electro-Optics (CLEO), paper SF2I.4 (2018).
Peter O. Weigel and Shayan Mookherjea, “Design of folded hybrid silicon carbide-lithium niobate
waveguides for efficient second-harmonic generation,” Journal of the Optical Society B 35(3),
593-600 (2018).
Peter O. Weigel, Marc Savanier and Shayan Mookherjea, “Hybrid Lithium Niobate Waveguide
for Efficient Quasi-Phase-Matched Optical Frequency Conversion,” in Conference on Lasers and
Electro-Optics (CLEO), paper SW4M.5 (2017).
Peter O. Weigel and Shayan Mookherjea, “Process Design Kit and Modulator Simulations for
Hybrid Silicon-Lithium Niobate Integrated Optics,” in Conference on Lasers and Electro-Optics
(CLEO), paper JTu5A.126 (2017).
Peter O. Weigel and Shayan Mookherjea, “Hybrid Silicon / Lithium Niobate Waveguide Micro-chips
Stable to 300◦ C,” in Conference on Lasers and Electro-Optics (CLEO), paper SM3K.1 (2017).
Peter O. Weigel and Shayan Mookherjea, “Reducing the thermal stress in a heterogeneous material
stack for large-area hybrid optical silicon-lithium niobate waveguide micro-chips,” Optical Materials
Vol. 66, 605-610 (2017).
Peter O. Weigel, Marc Savanier, Christopher T. DeRose, Andrew T. Pomerene, Andrew L. Starbuck,
Anthony L. Lentine, Vincent Stenger and Shayan Mookherjea, “Lightwave circuits in lithium
niobate through hybrid waveguides with silicon photonics,” Scientific Reports 6, 22301 (2016).
Marc Savanier, Peter O. Weigel and Shayan Mookherjea, “Proposal and analysis of a hybrid silicon
photonic-lithium niobate waveguide for difference frequency generation,” Optics Communications
356, 330-335 (2015).

xx
Ryan Aguinaldo, Peter Weigel, Hannah Grant, Christopher DeRose, Anthony Lentine, Andrew
Pomerene, Andrew Starbuck and Shayan Mookherjea, “Characterization of a silicon-photonic
multi-wavelength power monitor,” in Optical Interconnects Conference, 139-140 (2014).

Ryan Aguinaldo, Peter Weigel, Hannah Grant, Christopher DeRose, Anthony Lentine, Andrew
Pomerene, Andrew Starbuck, Andre Tkacenko and Shayan Mookherjea, “A silicon photonic
channelized spectrum monitor for UCSD’s multi-wavelength ring network,” in Conference on
Lasers and Electro-Optics (CLEO), paper STu1G.4 (2014).

xxi
 ABSTRACT OF THE DISSERTATION

High-Speed Hybrid Silicon-Lithium Niobate Electro-Optic Modulators

& Related Technologies

by

Peter O. Weigel

Doctor of Philosophy in Electrical Engineering (Photonics)

University of California San Diego, 2018

Professor Shayan Mookherjea, Chair

Electro-optic modulators (EOMs) serve as a technological pillar of the modern telecommuni-

cations industry. Without these devices, which convert electrical data into optical data through one of
several physical phenomena (depending on the specific technology), telecommunications channels
would be severely bandwidth-limited, particularly within data centers. To meet the ever-increasing
bandwidth demands of the industry, either more EOMs are necessary (resulting in higher power
requirements) or higher bandwidth EOM technology must be developed. This thesis discusses
the theory, design, fabrication, and characterization of foundry-compatible hybrid silicon-lithium
niobate (Si-LN) electro-optic modulators integrated on a wafer platform, a new technology with
potentially far-reaching applications.

xxii
 By bonding a thin film of ion-sliced LN crystal, which retains the crystal properties of
bulk LN, to silicon waveguides in the Mach-Zehnder modulator configuration, it is theoretically
possible to exceed the bandwidth limitations of all-Si modulators without abandoning the scalable,
dense silicon-on-insulator (SOI) platform. These hybrid devices make use of the favorable linear
electro-optic Pockels effect of LN while using the high-index Si waveguide to reduce the optical
mode area, so that low-voltage, high-bandwidth devices can be realized. This thesis focuses on
developing broadband EOMs with cutoff frequencies beyond 100 GHz. Developing this technology
on an SOI wafer platform fabricated via photolithography in a foundry facility presents a realistic
path towards next-generation high-speed, low-power integrated EOMs.

xxiii
Chapter 1

Theory of Electro-Optic Modulators

1.1 Introduction

In its most basic form, an electro-optic modulator (EOM) is a device that converts electrical
data into optical data by, in some way, imprinting an electrical waveform onto the phase of an optical
laser beam (i.e., “modulation”). The EOM in this simple form is known as a phase modulator,
but more often the phase modulator is a component in an interferometric structure such as the
Mach-Zehnder interferometer (MZI) shown in Fig. 1.1a, where the phase modulation in each arm of
the structure is used to cause interference, effectively modulating the optical intensity. When used
in this layout, the EOM is called an intensity modulator and the layout is termed a Mach-Zehnder
modulator (MZM), the electrically active form of the MZI.
While free space, table-top EOMs can be built with bulk components, because of size,
weight, and power limitations, integrated EOMs have been an area of research since the 1970s [1].
This chapter will discuss general EOM theory, particularly for integrated lithium niobate (LiNbO3
or LN) EOMs which use the Pockels effect. Silicon (Si) EOMs based on the plasma dispersion
effect will also be discussed.
All devices considered for the remainder of this chapter are assumed to be in the MZM
configuration.

1
 a b
GND
j /2
1/
2 E in 1/
2 Ein e V GND
LN
Ein V Eout x
-j /2 z
1/
2 E in 1/
2 Ein e
GND LN Handle

Figure 1.1: a Top view of a standard MZM in the push-pull configuration, meaning the two arms
of the MZM have (ideally) equal and opposite electro-optic phase shifts (∆φ = the total phase shift
of the device). Gray lines represent the optical waveguides, gold boxes represent the electrical
lines (“electrodes”), red arrows represent the flow of light, and black arrows represent electric field
lines. ‘GND’ = ground, ‘V’ = applied voltage, and Ein is the electric field component of the input
optical wave. b A cross-section of a single arm of the MZM, where the waveguiding region is
defined by an etched ridge of LN on a bulk LN handle.

1.2 Lithium Niobate & the Pockels Effect

Before discussing how lithium niobate is useful, it is first helpful to explain why lithium
niobate is useful. Lithium Niobate (LN) is a birefringent crystal, which means it has a different
refractive index depending upon the polarization and direction of propagation of light through the
crystal. In particular, LN displays negative uniaxial birefringence, meaning only the z axis (also
known as the crystal axis) has a different refractive index (uniaxial), but it is a lower value than the
refractive index along the x and y crystal axes (negative). The refractive index along the z axis is
called the extraordinary refractive index (ne ) and the refractive index along x and y is called the
ordinary refractive index (no ).
Additionally, the LN crystal can be cut into a wafer form in different ways. Each type of
cut of wafer is defined by the axis of the crystal that is perpendicular to the wafer surface. For
example, a z-cut LN crystal will have the z crystal axis perpendicular to the surface of the crystal,
while an x-cut LN crystal will have the z axis parallel to the surface of the crystal. Because of these
properties, LN lacks inversion symmetry to its crystal lattice.1
1 Though of less interest to this thesis, LN is also piezoelectric (generates charge when put under mechanical stress,
useful for acoustic wave generation) and pyroelectric (generates charge when put under thermal stress, useful as a
thermal transducer) [2].

2
 Crystals which lack inversion symmetry are classified generally as “non-centrosymmetric”
and display the Pockels effect, also known as the linear electro-optic effect, where a shift in the
crystal’s refractive index is observed under the application of an electric field. As a brief proof of
this effect (see [3] for more details), we shall start by considering an index ellipsoid in the (x, y, z)
coordinate system:

x2 y2 z2
+ + = 1. (1.1)
n2x n2y n2z

Here ni is the material’s refractive index along each of the principle dielectric axes, x, y, and z. As
mentioned earlier, the Pockels effect is a change in a material’s refractive index as a function of an
applied electric field:2

 
1
∆ 2 ≡ ri jk Ek (1.2)
n ij

where ri jk is the linear electro-optic tensor, and Ek is the kth component of the applied electric field.
k can represent any of the three principal axes (x, y, or z). Under the application of an electric field
as per Equation (1.2) (i.e. adding Equation (1.2) to Equation (1.1)) results in

  !  
1 1 1
+ r11k Ek x2 + 2
+ r22k Ek y + 2 + r33k Ek z2
n2x n2y nz (1.3)
+ 2xyr12k Ek + 2yzr23k Ek + 2zxr13k Ek = 1.

Equation (1.3) assumes that ri jk = r jik , which is valid when considering high purity non-active
crystals. Voigt index contraction is next used to simplify (1.3), where 1 = (11) = (xx), 2 = (22) =
(yy), 3 = (33) = (zz), 4 = (23) = (32) = (yz) = (zy), 5 = (13) = (31) = (xz) = (zy), and 6 = (12) = (21)
= (xy) = (yx). This convention is only for convenience, as it reduces the number of independent
electro-optic coefficients from 27 to 18. Using this contraction in matrix form, Equation (1.2)
becomes
2 Physically, this is caused by a slight shift in the crystal lattice.

3
    
 ∆(1/n2 )1   r11 r12 r13 
   

 ∆(1/n2 )2  
  r21 r22 r23 

 
    Ex
∆(1/n2 )3 r31 r32 r33  
    
   
 =  E  (1.4)
 y 
∆(1/n2 )4
  
   r41 r42 r43   
 Ez
   
  

 ∆(1/n2 )5  
  r51 r52 r53 

   
∆(1/n2 )6 r61 r62 r63

Equation (1.4) is valid for any crystalline material that meets the aforementioned symmetry
criteria. Non-centrosymmetric crystals, such as crystalline Si, have an electro-optic tensor that looks
like this:

 
 0 0 0 
 

 0 0 0 

 
0 0 0 
 

rSi = 


 (1.5)

 0 0 0 

 

 0 0 0 

 
0 0 0

which is what one should expect, considering that non-centrosymmetric materials contain an
inversion symmetry. Of more interest is the electro-optic tensor for LN, which falls under the 3m
crystal symmetry class [2]:

 
 0 −r22 r13 
 
 0 r22 r13 
 
 
 0 0 r33
 

rLN =

.
 (1.6)
 0 r42 0 
 
 
 r
 42 0 0 

 
−r22 0 0

4
Precise material properties vary depending on the composition of the LN crystal (e.g. Li+ percent-
age) as well as crystal purity and the wavelength of operation, but as a rough estimate, r22 = 3.4
pm/V, r13 = 8.6 pm/V, r42 = r51 = 26 pm/V, and r33 = 31 pm/V at a wavelength of 633 nm [3, 2].
Larger ri j values correspond to a stronger Pockels effect. The advantage of LN, in addition to its
ability to be formed into high purity crystalline wafers from a melt (as opposed to being grown in a
reactor, where uniformity, repeatability, and growth rate are concerns), is that applying an electric
field along the crystal’s z axis results in non-zero values to only ∆(1/n2 )1 , ∆(1/n2 )2 , and ∆(1/n2 )3
in Equation (1.4). Of particular interest is ∆(1/n2 )3 , because it is proportional to r33 , the largest
electro-optic coefficient in LN. Since this index shift is occurring along the z axis of the crystal
(see Equation (1.3)) – the same direction the electric field is being applied along – an optical wave
polarized along this crystal axis will respond to the material index shift with a (linearly) proportional
phase shift.3 This physical effect is at the root of all electro-optic devices built using LN.

1.2.1 Voltage-Length Product in LN MZMs

The voltage-length product, or Vπ L, is a popular figure of merit used to describe the inverse
relationship between the voltage required to modulate the output optical intensity of an MZM from a
maximum to a minimum by applying a π phase shift (Vπ ) and the physical length of the modulation
region (L). This definition will vary depending on the effect used for modulation. For MZMs based
on the Pockels effect, Vπ L can be derived from general electromagnetic theory.
We begin by considering an MZM with the top view and cross-section of Fig. 1.1. For such
a structure, the optical mode exists primarily in the LN ridge region, though some exponential tails
of the mode exist in the surrounding cladding region (typically air or silicon dioxide) which does not
display the Pockels effect. In this case, we must be careful when calculating Vπ L to only consider
that portion of the mode which is in the LN and polarized along the proper crystal axis. For this
derivation, we will assume the optical wave is well polarized along the z axis so that only the LN
3 It’sworth mentioning that the ∆(1/n2 )1 and ∆(1/n2 )2 terms, which are non-zero and proportional to r13 when an
electric field is applied along the crystal’s z axis, result in changes to the material index in the x and y crystal axes.

5
material index shift associated with r33 needs to be considered.4
We begin with the wave equation for an optical field [4]:

∇t Eo + [n2 k0 2 − β2eff ]Eo = 0, (1.7)

where n is the material index, k0 = 2π/λ0 is the free-space wavenumber, λ0 is the free-space
wavelength, βeff = 2πneff /λ0 is the optical mode’s propagation constant, and Eo ∝ exp j(βeff y−ωt) ẑ.
Equation (1.7) is the unperturbed wave equation, i.e. without any voltage. When a voltage is applied,
the LN material index changes by ∆n. In that case, n2 becomes (n+∆n)2 = n2 + ∆n2 + 2n∆n ≈ n2 +
2n∆n, where the approximation is valid because ∆n << n. Similarly, β2eff becomes (βeff + ∆βeff )2 ≈
β2eff + 2βeff ∆βeff .
When a voltage is applied, the optical field is perturbed and the wave equation becomes

∇t E0o + [(n2 + 2n∆n)k0 2 − (β2eff + 2βeff ∆βeff )]E0o = 0. (1.8)

Assuming that the perturbed field E0o ≈ Eo , we can substitute Equation (1.7) into Equation (1.8),
multiply both sides by Eo ∗ , and arrive at

[2n∆nk02 − 2βeff ∆βeff ]|Eo |2 = 0. (1.9)

The electro-optic effect takes place over all space in the transverse domain, which is accounted for
by integrating over the x,z plane:

Z Z
[2n(x, z)∆n(x, z)k02 − 2βeff ∆βeff ]|Eo (x, z)|2 dxdz = 0. (1.10)
∞

With some rearrangement of Equation (1.10), and keeping in mind that βeff = 2πneff /λ0 , we
arrive at:
4 In principle, each of the index shifts caused by a z-oriented applied electric field can be evaluated separately,
considering that they are orthogonal to one another. However, in most practical cases the polarization fraction of a
well-designed guided mode will be >99% and along the z axis, so effects due to r13 (which is nearly four times smaller
than r33 ) along the x and y crystal axes will be negligible.

6
 z)∆n(x, z)|Eo (x, z)|2 dxdz
RR
1 LN n(x,
∆neff = RR
2
, (1.11)
neff ∞ |Eo (x, z)| dxdz

n(x,z) and ∆n(x,z) correspond to the material refractive index and refractive index change, respec-
tively, at each point (x,z). Since ∆n(x,z), which is due to the Pockels effect, equals zero outside of
the LN film, the integral in the numerator is only taken over the LN. The refractive index of LN is
well known, as is the change in refractive index of LN under an applied voltage [3]:

1
∆ne ≈ − ne 3 r33 ERF . (1.12)
2

∆ne is the change in the extraordinary refractive index of LN when introduced to an external
electric field ERF , where ‘RF’ stands for ‘radio frequency’ and is commonly used to denote high
frequency alternating current (ac) electrical signals. Here we are only considering the change in the
extraordinary axis because the optical mode is TE-like in x-cut LN, meaning that the optical mode
is almost entirely polarized along the crystal axis of LN and sees only the extraordinary refractive
index, ne . ne is the extraordinary refractive index of LN when no electric field is applied, r33 is the
previously defined largest component of the linear electro-optic tensor ri jk , and ERF is the applied
electric field along the crystal (z) axis to utilize the r33 electro-optic coefficient.
By substituting Equation (1.12) into Equation (1.11) and denoting n(x, z) = ne , we have

2
−ne 4 r33
RR
LNR|E o (x, z)| ERF (x, z)dxdz
∆neff = R
2
. (1.13)
2neff ∞ |Eo (x, z)| dxdz

The integral in the numerator of Equation (1.13) can be normalized to a uniform field by substituting
VA G
ERF with G VA ERF in Equation (1.13):

2
−ne 4 r33VA G
RR
LN R|ERo (x, z)| ERF (x, z)dxdz
∆neff = , (1.14)
2neff G VA ∞ |Eo (x, z)|2 dxdz

or, more succinctly:

7
 −ne 4 r33VA
∆neff = Γmo , (1.15)
2neff G

where

2
RR
G LN R|ERo (x, z)| ERF (x, z)dxdz
Γmo = (1.16)
VA ∞ |Eo (x, z)|2 dxdz

is the normalized overlap between the optical and electrical fields in the x − z plane, and has a
maximum value of 1 when the applied field is entirely uniform, as in the case of an ideal parallel-
plate configuration.
2πneff z
The phase of an unperturbed optical wave is φ = βeff z = λ0 . When the optical field is
2π(neff +∆neff )z
perturbed by an applied electric field, the optical phase becomes φ + ∆φ = λ0 . If we
assume that the upper ground electrode of Fig. 1.1a is not present, then the electric field is only
applied along the lower optical waveguide and the phase difference between the left and right
2π∆neff
waveguides at the y-junction combiner of the EOM is ∆φ = λ0 z. When we substitute Equation
(1.15) into this expression for ∆φ, we get

ne 4 r33 πVA z
∆φ = − Γmo . (1.17)
neff λ0 G

The half wave voltage-length product Vπ L is the voltage required to achieve a phase shift
of ±π radians multiplied by the length of the device. If we set ∆φ = -π, z = L (the length of the
electro-optic interaction region), and call VA = Vπ , we can derive an expression for the half wave
voltage-length product for a thin film LN-based EOM:

neff λ0 G
Vπ L = . (1.18)
ne 4 r33 Γmo

Equation (1.18) reduces to the traditional expression (Equation (9a) in [5]) when neff = ne and the
integrals in Γmo are taken over the entire cross-section domain, not just the LN ridge. These are
practical assumptions for titanium indiffused LN (Ti:LiNbO3 ) modulators.

8
 Equation (1.18) is analyzed in Fig. 1.2, where each of the three variables (neff , G, and Γmo )
is swept through a reasonable range of values in the non push-pull configuration (in the push-pull
configuration – such as in Fig. 1.1a – all Vπ L values would simply be divided by 2).
a) 13 b) 20 c) 50
G = 12
m neff = 2.0 40
G = 12
m
12 mo = 0.5 15
mo = 0.5 neff = 2.0
V L(V-cm)

V L(V-cm)
V L(V-cm)
30
10
20
11
5 10

10 0 0
1.8 1.9 2 2.1 2.2 5 10 15 20 0 0.2 0.4 0.6 0.8 1
neff Gap ( m) mo

Figure 1.2: Vπ L is evaluated versus: neff (a), G (b), and Γmo (c).

Of interest to a designer is the theoretical limitation for this type of structure. In the
theoretical limit, Γmo = 1 and Vπ L = 2.41·105 ·neff G (V-cm) at λ0 = 1550 nm. In bulk LN devices
the limitation on the electrode gap is the width of the optical mode (as well as electrical losses which
affect modulation bandwidth limitations, but we will not consider those here). Lastly, if Γmo = 1 and
the optical mode is well confined in the LN ridge region, then it’s unlikely neff will be much lower
than 2.10, since the majority of the mode will reside in the LN and not the lower-index cladding.
In this “best-case” scenario (G = 1 µm and neff = 2.10), Vπ L = 0.5 (V-cm) in the non push-pull
configuration, and Vπ L = 0.25 (V-cm) in the push-pull configuration. This is a reasonable estimate
of the limitation on Vπ L when the EOM is optimized solely for Vπ L minimization (this ignores
optical loss limitations and fabrication limitations, such as the finite width of the LN ridge).

1.2.2 Electro-Optic Bandwidth Theory for a Traveling Wave Mach-Zehnder

Modulator

The 3-dB bandwidth of an electro-optic modulator is a metric used to describe the device’s
working frequency range. An MZM designed with straight electrical lines displays a low-pass
frequency response, and the frequency point where the response drops to 3-dB below its dc value is
called its 3-dB bandwidth.

9
 There are, however, two different definitions of 3-dB bandwidth commonly used when
discussing electro-optic modulators. Perhaps unsurprisingly, these are the electrical 3-dB bandwidth
and the optical 3-dB bandwidth. To understand the differences between these two bandwidths, some
discussion of the device’s response is required. First, consider the basic definition of the modulator’s
response described in the previous paragraph:

Po (ω)
m(ω) = . (1.19)
Po (0)

Here, Po is the optical output power, modulated with an electrical signal at frequency ω. Because of
its low pass characteristics, the response of the MZM is defined as the ratio of Po at ω Hz to Po at 0
Hz. In dB, this becomes the optical modulation response mo (ω):

 
Po (ω)
mo (ω) = 10 log10 (m(ω)) = 10 log10 . (1.20)
Po (0)

However, the modulated laser is detected through a photodetector, which has a responsivity
R (units of amps per watt) and outputs an electrical current proportional to optical power, so that

RPo (ω) Ie (ω)

m(ω) = = (1.21)
RPo (0) Ie (0)

where Ie is the electrical current out of the photodetector. In this case, the electrical bandwidth is
defined with respect to the electrical power Pe measured out of the photodetector:5

Ie2 (ω)
       
Pe (ω) Ie (ω) Po (ω)
me (ω) = 10 log10 = 10 log10 = 20 log10 = 20 log10 ,
Pe (0) Ie2 (0) Ie (0) Po (0)
(1.22)
where me (ω) is the electrical modulation response. From Equations (1.20) and (1.22), it is clear
that me (ω)|dBe = 2mo (ω)|dBo . For distinction, units of ‘dBe’ are used for the electrical response
and units of ‘dBo’ are used for the optical response. When defining the 3-dB bandwidth, the optical
5 Electrical power is proportional to current squared.

10
3-dB bandwidth will be the 6-dB electrical bandwidth. Both definitions are used in the literature,
sometimes ambiguously.
The complete derivation of the small signal frequency response m(ω) of a traveling-wave
MZM (that is, one in which the wavelength of the electrical wave is taken into account) is described
quite thoroughly in [4], and will not be included here. The final expression for the modulation
response of a modulator with coplanar waveguide electrodes, from [4], is:

RL + RG Zin (ZL + Z0 )Fu+ + (ZL − Z0 )Fu−

m(ω) = (1.23)
RL Zin + ZG (ZL + Z0 ) exp(γm L) + (ZL − Z0 ) exp(−γm L)

where RL is the load resistance, RG is the generator resistance, Zin is the device’s electrical input
impedance, ZG is the generator impedance, Z0 is the characteristic impedance of the EOM for
its electrical TEM mode,6 Fu± = (1 − exp(u± ))/u± , u± = ±αm L + j(ω/c)(±nm − no )L, γm =
αm + jωnm /c, L is the electro-optic interaction length, αm is the electrical propagation loss, c is the
speed of light in vacuum (c = 299,792,458 m/s), nm is the electrical mode’s effective index, and no
is the optical mode’s group index. A circuit model of a traveling-wave electro-optic modulator is
shown in Fig. 1.3, and Zin is defined as:

ZL + Z0 tanh(γm L)
Zin = Z0 (1.24)
Z0 + ZL tanh(γm L)

By analyzing Equation (1.23), it can be seen that three aspects of the device need to be
considered. First, the device must be impedance matched. Ideally, the device is designed so that
Z0 = ZG = ZL . ZG is usually 50 Ω, so Z0 tends to be designed for 50 Ω. Second, the electrical
and optical indices must be as close to one another as possible. This means the electrical wave
should propagate along the transmission line with the same (or close to the same) velocity as the
optical wave. Here there tends to be some confusion over whether or not to use the phase velocity
or group velocity of each wave. Because there is only one frequency component being considered,
6 For these structures, the RF mode is a TEM mode, so a characteristic impedance can be defined.

11
 L

Vm(z)
ZG Zin
Z0 ,
m ZL
Va
z

Modulator Region

Figure 1.3: Circuit model showing where the electro-optic modulator fits between the signal
generator (with applied voltage Va and internal impedance of ZG ) and the load impedance ZL . The
voltage along the modulator is Vm (z), which varies along the length of the EOM.

the electrical phase index and group index are identical and either can be used. Furthermore, in a
well-designed EOM the electrical mode is highly non-dispersive, so that electrical phase index and
group index are nearly identical. However, the optical mode’s group index should be used, not its
phase index. While a single wavelength of light is entering the EOM device, as soon as modulation
occurs, optical sidebands are generated around the wavelength of the input laser beam, caused by
the interaction of the laser light with the electrical signal. Third and lastly, the device’s electrical
propagation loss must be kept as low as possible.
From Equation (1.23), it is clear that there is a relationship between the length and bandwidth
of the device. In particular, each of the three parameters described in the previous paragraph will
have some effect on the device’s bandwidth, but how significant that effect is depends on the length
of the device. These relationships are explored in Fig. 1.4, where electrical loss, index mismatch,
and impedance mismatch are varied in order to show the limitations of the traveling-wave MZM
structure. Curiously, in the case of an impedance mismatched device where Z0 > ZG = ZL (as in
Fig. 1.4c), the bandwidth actually improves for a device length L, though the frequency response is
much noisier and will result in undesired reflections that could damage some external electronics
[4].
From Fig. 1.4, it is clear that even a device with no index or impedance mismatch and
minimal electrical loss (αcond = 0.2 dB/[cm-GHz1/2 ], the solid curve in Fig. 1.4a) is limited to
less than a 50 GHz electrical bandwidth for a device length of 5 cm or longer. A typical bulk LN

12
a b c
200
cond = 0.2 200 n = 0.00 200 Z0 = 30 

cond = 0.4 n = 0.01
3-dBe Bandwidth (GHz)

3-dBe Bandwidth (GHz)

3-dBe Bandwidth (GHz)

Z0 = 40 

cond = 0.6 n = 0.05
150 150 150 Z0 = 50 

cond = 0.8 n = 0.10

cond = 1.0 n = 0.20 Z0 = 60 
n = 0.50 Z0 = 70 
100 100 100

50 50 50

0 0 0
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
Device Length (cm) Device Length (cm) Device Length (cm)

Figure 1.4: The results shown in this figure are simulated 3-dBe electrical bandwidth values, in
GHz, based on Equation 1.23. a Assumes perfect index matching (∆n = 0) and perfect impedance
matching (ZG = Z0 = ZL = 50 Ω). Each curve represents a different electrical conductor loss (αcond ,
the only significant source of electrical loss at high frequencies for a well-designed device), in units
of dB/[cm-GHz1/2 ]. b Different index matching values are shown for αcond = 0.2 dB/[cm-GHz1/2 ]
and perfect impedance matching. ∆n = nm − no . c In this case, αcond = 0.2 dB/[cm-GHz1/2 ],
∆n = 0.05, and different values of Z0 are plotted, assuming ZG = ZL = 50 Ω.

EOM device, however, has a Vπ L on the order of 10 V-cm in the push-pull configuration. In that
case, to achieve even Vπ = 5 V, the device would have to be 2 cm long. For Vπ = 1 V, L would
have to equal 10 cm, in which case the electrical bandwidth will never approach 50 GHz, much
less 100 GHz or beyond. The fundamental limitation to higher bandwidth, practical EOMs, is the
typically very large Vπ L factor for bulk LN devices. With a low enough Vπ L, there is enough margin
of error to engineer the structure for high bandwidth.
A more in-depth analysis of the limitations on Pockels-based MZM devices is included in
the next chapter, where the hybrid Si-LN EOM that is the main subject of this thesis work will be
introduced.

1.3 Silicon & the Plasma Dispersion Effect

Although this thesis focuses on devices which utilize the Pockels effect, the plasma disper-
sion effect (also known as the free carrier dispersion effect) in silicon is briefly mentioned because
of its recent widespread use in Si-based integrated optical circuits.

13
 Unlike the Pockels effect, the plasma dispersion effect in silicon is a change in refractive
index (n) and absorption (α) as a function of the change in dopants (electrons or holes). This effect
was first observed experimentally by Soref and Bennett in 1987 [6], and for λ = 1.55 µm is expressed
quantitatively as [7]:

∆n = ∆ne + ∆nh = −8.8 × 10−22 ∆Ne − 8.5 × 10−18 (∆Nh )0.8 (1.25)

∆α = ∆αe + ∆αh = −8.5 × 10−18 ∆Ne + 6.0 × 10−18 ∆Nh (1.26)

for the change in Si refractive index (∆n) and Si absorption (∆α). The subscript ‘e’ is used to
represent changes (in either n or α) due to free electron carrier concentration, while ‘h’ is used for
free hole carrier concentrations. For example, ∆ne is the change in Si refractive index due to free
electrons. ∆Ne and ∆Nh are the change in electron and hole carrier concentrations, respectively.
A Si MZM can be designed with the same optical and electrical circuitry as shown for the
simple MZM of Fig. 1.1a, except that the phase shift ∆φ/2 in each arm will be due to physical
movement of carriers in or out of the optical waveguide. Whether charges are moving into or out
of the optical region when an electric field is applied defines the type of modulator. If charges are
moving into the waveguide region under the application of an electric field, it is termed a carrier
injection MZM and the device is forward biased; conversely, if charges are moving out of the
waveguide region under the application of an electric field then it is termed a carrier depletion MZM
and the device is reverse biased.
Si MZMs have recently reached electrical bandwidths as high as 50 GHz [8]. However,
this appears to be approaching the bandwidth limitation of Si MZMs, which are around 60 GHz
[9, 10]. In addition to bandwidth, loss, power consumption, and device footprint are of concern
as well. A useful figure of merit is the voltage-loss product, obtained by multiplying Vπ L by the
device propagation loss (in dB/cm) to arrive at a figure of merit with units of V-dB. Although the
reference is a few years old at this point, the comparisons made in Table 1 of [11] are useful to

14
understand the limitations of Si modulators: reducing voltage increases optical propagation loss,
and any reasonably effective Si MZM device has propagation loss values on the order of tens of
dB/cm. While Si MZMs are attractive for their ability to be easily integrated into foundry processes,
their potential for high-bandwidth, large-scale applications are severely limited by the intrinsic loss
associated with doping silicon waveguides.

Bibliography
[1] Kaminow I, Ramaswamy V, Schmidt R, Turner E. Lithium niobate ridge waveguide modulator.
Applied Physics Letters. 1974;24(12):622–624.

[2] Weis R, Gaylord T. Lithium niobate: summary of physical properties and crystal structure.
Applied Physics A. 1985;37(4):191–203.

[3] Yariv A, Yeh P. Photonics: optical electronics in modern communications. Oxford Univ.;
2006.

[4] Ghione G. Semiconductor devices for high-speed optoelectronics. Cambridge University

Press; 2009.

[5] Alferness RC. Waveguide electrooptic modulators. IEEE Transactions on Microwave Theory
Techniques. 1982;30:1121–1137.

[6] Soref R, Bennett B. Electrooptical effects in silicon. IEEE journal of quantum electronics.
1987;23(1):123–129.

[7] Reed GT. Silicon photonics: the state of the art. John Wiley & Sons; 2008.

[8] Sun J, Sakib M, Driscoll J, Kumar R, Jayatilleka H, Chetrit Y, et al. A 128 Gb/s PAM4
Silicon Microring Modulator. In: Optical Fiber Communication Conference. Optical Society
of America; 2018. p. Th4A–7.

[9] Gardes F, Reed G, Emerson N, Png C. A sub-micron depletion-type photonic modulator in

silicon on insulator. Optics Express. 2005;13(22):8845–8854.

[10] Li ZY, Xu DX, McKinnon WR, Janz S, Schmid JH, Cheben P, et al. Silicon waveguide
modulator based on carrier depletion in periodically interleaved PN junctions. Optics express.
2009;17(18):15947–15958.

[11] Gill D, Green W, Assefa S, Rosenberg J, Barwicz T, Shank S, et al. A figure of merit-
based electrooptic Mach-Zehnder modulator link penalty estimate protocol. arXiv preprint
arXiv:12112419. 2012;.

15
Chapter 2

Design of Hybrid Electro-Optic Modulators

2.1 Introduction

The main focus of this thesis work is on hybrid silicon-lithium niobate (Si-LN) electro-optic
modulators (EOMs), the cross-section of which is shown in Fig. 2.1a in the push-pull configuration.1
The concept of the design is to use the high-index Si rib waveguide (refractive index around 3.5 for
wavelength (λ) around 1550 nm) to provide the optical mode with some horizontal confinement so
that it remains a guided mode while performing the actual electro-optic modulation in the thin LN
film, which assists the Si rib with vertical modal confinement. The LN film is not etched or sawn
to avoid complicated fabrication steps that are less likely to be adopted by foundries, as well as to
maximize optical transmission. Unlike all-Si EOMs, no electro-optic modulation is performed in
the Si layer in this hybrid EOM. Instead, the Pockels effect in LN is used for modulation. Unlike
the carrier dispersion effect in Si, the Pockels effect theoretically allows for extremely broadband
devices while maintaining low optical propagation loss [1]. Indeed, state-of-the-art, research-grade,
all-LN EOMs have reached 70 GHz electrical bandwidth in the past [2], though precise LN etching
was required and the Vπ L was around 10 V-cm (about as low a Vπ L as all-LN EOMs can achieve
1 An EOM in the Mach-Zender configuration described in the previous chapter can modulate either one or both arms
of the device; when both arms are modulated, a phase of π/2 and −π/2 are imparted on each of the arms, respectively,
and voltage required a total phase shift ∆φ is cut in half. This is called the push-pull configuration.

16
while still reaching high bandwidths).
All-Si EOMs, on the other hand, have reached as high as 50 GHz electrical bandwidth in
depletion mode with Vπ L = 0.6 V-cm, but with an optical propagation loss of 188 dB/cm in the
active region and an extinction ratio of only 5.3 dB (LN EOMs, almost universally, have 20 dB or
greater extinction ratio). This bandwidth appears to be at or close to the upper limit of Si-based
EOMs [3, 4]. Of course, these limitations are well known to the Si photonics community. For years,
the challenges to creating better Si modulators have been mainly fabrication-based, i.e. optimizing
dopant concentrations and waveguide placement. Certainly, clever Si EOM designs have advanced
the field significantly over the years [5], but device and fabrication optimization appear to be close
to their limits.
By leveraging the high-accuracy and wafer-scale foundry fabrication of silicon-on-insulator
(SOI) devices and integrating thin films of LN with such technology, it is possible to reduce the
voltage limitations of traditional all-LN EOMs without sacrificing optical propagation loss or
extinction ratio. By using an SOI substrate, wafer-scale Si-LN processing becomes a possibility
with all the potential benefits that modern CMOS fabrication has to offer. Additionally, by using
a well-defined Si rib below the LN film, extra design parameters are introduced to the device that
were not available to all-LN device designers of the past. These extra parameters, discussed below,
allow for precise bandwidth tuning without losing the low voltage and optical loss afforded by the
hybrid design.

2.2 Hybrid Si-LN Optical Mode Design

The hybrid Si-LN optical mode is in some ways a two-material core waveguide, except
the second material (LN) is quasi-infinite in width. This means the waveguide must be carefully
designed to maximize light in the LN region (where the electro-optic Pockels effect takes place)
without decoupling entirely from the Si waveguide core. An example of a well-designed optical
mode is provided in Fig. 2.1b, where the LN confinement factor (LNCF – the percentage of the

17
optical mode that is horizontally polarized and in the LN region) is 81% and the effective mode
area (Aeff ) is only 1.3 µm2 . In this case, the mode is not monotonic because the Si waveguide is
narrower than the wavelength in Si (approximately 440 nm), so the mode cannot fit in the Si rib
and must expand outwards into the nearby SiO2 and LN regions. This behavior is not observed in
weakly guided waveguides, such as fibers and bulk crystals.
LNCF is defined as

+∞
R R0
|Sy |2 dxdz
−∞ −hLN
LNCF = +∞
· 100 [%], (2.1)
R +∞
|S̄|2 dxdz
R
−∞ −∞

where hLN is the LN thickness. Aeff is defined as

R +∞
2
|E|2 dA
Aeff = R−∞
+∞ [m2 ]. (2.2)
−∞ |E|4 dA

where S is the total time-averaged Poynting vector, Sy is the time-average Poynting vector parallel
to the propagation axis y and E is the vectorized electric field component of the optical mode.

a wgnd wgap wmid b

Optical Mode
helec
b
hLN
hox LN
hSi wSi hbox

hsub SiO2
Si WG 1
m

Si SiO2
LN Au

Figure 2.1: a A cartoon of what the hybrid Si-LN EOM cross-section should look like in the
push-pull configuration. Material labels are provided below the image, and material dimensions
are as follows: hSi = Si waveguide height; wSi = Si waveguide width; hsub = height of SOI substrate;
hbox = height of buried oxide layer; hox = height of oxide between Si waveguides and LN film;
hLN = height of LN film; helec = height of electrical lines; wgap = gap width between ground planes
and signal line; wmid = width of signal line; wgnd = width of ground planes. b Simulated Si-LN
optical mode in the hybrid region. LNCF = 81% and Aeff = 1.3 µm2 . The region simulated in b is
highlighted in a with a blue box.

To arrive at the mode design of Fig. 2.1b, it helps to first decide upon the Si layer thickness

18
(hSi in Fig. 2.1a) to use. Too thick of a Si layer could result in multi-mode waveguides, as well as a
poor LNCF figure of merit; and although Aeff will be quite low as a result, only a small percentage
of the optical mode will be modulated. On the other hand, if the Si layer is too thin, it will not be
useful for on-chip optical routing since tight bends will be highly lossy, and even straight waveguide
propagation losses may be excessive. Additionally, a Si thickness should be chosen so that the
optical mode location can be adjusted between the Si region and the LN region by changing as few
design parameters as possible. With modern foundry photolithography it is quite simple to fabricate
variations in the width of features, but not in their heights. For this reason, all hybrid waveguide
designs should be designed with a single Si layer thickness, a single LN film thickness, and a
variable Si width. This means hSi must be chosen so that a change in the Si width can accommodate
both a high LNCF cross-section as well as a low LNCF cross-section. For this work, a Si layer
thickness of 150 nm was chosen.
The next step in the design process is to choose a Si width that is best for operation in the
hybrid region of the device. The goal of this hybrid Si-LN platform is not only to make efficient
broadband EOMs — it is also to integrate LN-based technologies onto SOI wafers. To do so requires
that standard Si waveguides can be used in conjunction with hybrid Si-LN devices. Therefore,
several Si widths must be used, depending on the application.

2.2.1 Optical Mode for the Electro-Optic Modulator

The optical mode of Fig. 2.1b is used in the hybrid EOM. This choice is informed by
tracking LNCF and Aeff of the optical mode, and finding a Si width that maximizes LNCF (so as
much of the light as possible is affected by the Pockels effect) without increasing Aeff to the point
where the optical signal gets absorbed by the metal lines. Because of the dependency of these
parameters on the LN film thickness (hLN ), wSi and hLN are chosen concurrently. Simulations of
LNCF and Aeff versus wSi for various hLN choices are shown in Fig.2.2. In the case where hLN is
very thick (1000 nm), the mode area becomes very sensitive to Si waveguide width. Going from a
wSi of 400 nm to 300 nm results in a five times growth in Aeff . Since fabrication error in waveguide

19
dimensions can be as large as ± 15 nm, this is too sensitive to be useful. Furthermore, such a
sensitive cross-section is too difficult to control by simply varying the Si rib width.

a 100
b
hLN = 300 nm 7
hLN = 300 nm
hLN = 400 nm
80 6 hLN = 400 nm
LNCF (%) hLN = 500 nm
5 hLN = 500 nm

Aeff ( m2)
hLN = 600 nm
60 hLN = 600 nm
hLN = 700 nm 4
hLN = 800 nm

hLN = 700 nm
40 3 hLN = 800 nm
hLN = 900 nm

20
hLN = 1000 nm 2 hLN = 900 nm

hLN = 1000 nm
1
0 0
300 400 500 600 700 800 300 400 500 600 700 800
Si Width (nm) Si Width (nm)

Figure 2.2: Simulated LNCF (a) and Aeff (b) for combinations of wSi and hLN , with hSi = 150 nm.

On the other hand, too thin a LN film (300 nm in Fig. 2.2) does not allow for a high LNCF,
limiting the electro-optic Pockels effect.2 The dimensions wSi = 320 nm and hLN = 600 nm were
chosen to achieve the optical mode of Fig. 2.1b. During device fabrication (discussed in detail later
on in this thesis), a chemical-mechanical planarization (CMP) step is performed on a layer of silicon
dioxide (SiO2 ) deposited over the Si waveguides, before attaching the LN film. For this reason, an
80 nm SiO2 layer is placed between the Si waveguides and the LN film in Fig. 2.1b to account for
the CMP tolerance of ± 50 nm. The group index of this optical mode is approximately ng = 2.3.

2.2.2 Optical Mode for Si Waveguide Applications

Although one of the primary innovations of this thesis lies in the design of the hybrid Si-LN
waveguide in the electro-optic modulator, an important benefit of using this waveguiding platform
is its versatility. By adiabatically tapering the 320 nm wide Si waveguide in the EOM up to 650 nm,
the optical mode becomes heavily confined in the Si waveguide, as is common with all-Si devices,
in which case passive as well as active all-Si devices can be realized, on the same platform as the
hybrid Si-LN EOM. Transitioning from the hybrid Si-LN EOM to the all-Si waveguide is done by
2 This approach is a rough estimate to understanding the effect of physical parameters on LNCF and Aeff , which
in turn will partially determine Vπ L. Once a few physical parameters are determined (either by intelligent choice or
fabrication limitations), it becomes more straightforward to optimize the device for low Vπ L, high bandwidth, or any
other figure of merit.

20
first tapering the Si waveguide from 320 nm to 650 nm with a linear taper, and then crossing from
the LN region into the non-LN region.
a b c
LN Wide Mode Non-LN Wide Mode

LN

SiO2 SiO2
Si 500 nm Si 500 nm

Figure 2.3: a Optical mode with wSi = 650 nm, hLN = 600 nm, and hSi = 150 nm, under the LN
film. b Same mode as a, but outside of the LN region. c Top view of the transition from the EOM
mode (narrow 320 nm wide Si waveguide, left side), to a wide Si-LN waveguide, and finally to
a wide all-Si mode. The last two modes, shown in a and b, are denoted by a and b in the figure,
respectively.

By tapering to a wide waveguide first, reflections and scattering due to the LN interface are
minimized. For the design of Fig. 2.3, there is a simulated loss of 0.1 dB when transitioning from
the LN region to the non-LN region. The mode of Fig. 2.3b is used for on-chip optical circuitry that
does not require the LN film, which can potentially include active devices in doped Si. The 650 nm
Si width is chosen because it is quite wide, which helps to confine the light in the Si region, but is
not wide enough to support higher-order optical modes which may couple into the fundamental TE
mode shown here.

2.2.3 Optical Mode for the Edge Coupler

A final optical mode cross-section is required for proper design of this device, and that is the
optical mode of the edge coupler. Edge couplers are commonly used for photonic integrated circuits
to couple light from a fiber onto a waveguide by bringing the fiber right up against the edge of the
chip. However, since Si is a high-index material, modes in Si waveguides tend to be very small
when compared to fiber modes. A well-known strategy for coupling efficiently from a high-index
contrast waveguide, such as a Si core surrounded by low-index SiO2 , into an optical fiber is to
reduce the width of the waveguide to below the wavelength of light. This causes the optical mode
to become much larger than the waveguide core. With careful design, the waveguide mode can be

21
engineered to closely match the width and shape of the fiber mode [6]. However, because this hybrid
Si-LN device is fabricated on a foundry wafer using photolithography, the minimum feature size is
limited by the photolithography tool. For most modern foundries, this feature size limit tends to be
180 nm. For standard 220 nm-thick Si waveguides, this results in suboptimal waveguide modes that
are smaller than the fiber mode. In the case of the 150 nm-thick Si waveguides described in this
thesis, and assuming an input/output fiber with a mode field diameter (MFD) of 2.5 µm (typical for
a lensed tapered fiber from Oz Optics, Ltd.), the simulated overlap between the fiber mode and the
waveguide mode is shown in Fig. 2.4a for a range of Si widths, and assuming λ = 1550 nm. The
best edge coupling occurs when the silicon width is 220 nm, with a simulated transmission loss of
approximately 0.5 dB (per edge coupler, for a total of 1 dB per device).

a 90 b
Power Coupling (%)

85

80

75

70

65 wSi = 220 nm 2
m
180 190 200 210 220 230
Silicon Width (nm)

Figure 2.4: a Simulation of power transmission of an edge coupler between a lensed tapered fiber
with MFD = 2.5 µm and a Si waveguide of height 150 nm surrounded by SiO2 . The maximum
power coupling is 89% for a Si width of 220 nm. b Logarithmic plot of the normalized electric
field component of the optical mode for the optimum Si width of 220 nm. The first five orders of
magnitude are shown, with red representing the highest field strength and dark blue representing
the weakest field strength.

The simulated plot in Fig. 2.4a reaches a maximum transmission below 100% not because
of a fabrication-limited Si width, but because of an intrinsic mismatch between the optical fiber
– whose mode has a two-dimensional Gaussian shape – and the waveguide. Other researchers
have made improvements on the edge coupler in recent years by shaping the optical mode at the
waveguide’s output [7, 8], but this typically requires electron beam lithography, custom-made fibers,
multi-material waveguide stacks, or some combination of the three.

22
2.2.4 Bonding Markers & Electrode Markers

Though not directly part of the EOM design, markers are added to the silicon layer to assist
with visual alignment of the LN film during bonding and for alignment during fabrication of the
electrical lines. In this case, rectangular markers are used for electrode alignment and square markers
are used for bonding alignment. Edge-to-edge marker alignment with mask markers (Fig. 2.5b) is
preferable to marker-overmarker alignment (Fig. 2.5c); edge-to-edge marker alignment ensures a
high level of precision since the mask and chip markers are always visible during alignment, even if
one layer is slightly out of focus.

a b

Electrode LN Bonding
Markers Markers

Figure 2.5: a GDS layout of EOM chip with arrays of alignment markers for electrode mask
alignment and LN bonding alignment. b Edge-to-edge alignment is when the chip marker (black)
and the mask marker (gray) are designed to align at the edge. This allows for high precision
alignment, since both markers are visible. c Marker-over-marker alignment is when the chip marker
and mask marker are designed to align directly over one another. This is a less accurate alignment
scheme, since one alignment marker covers the other, the depth of focus of the microscope
objective distorts the true alignment.

2.3 Full Optical Design of the EOM

Using the mode designs of Section 2.2, a full EOM in the Mach-Zehnder configuration can
be designed, with Si tapers and bends used to convert between mode types and to adjust waveguide
separation where necessary. The full hybrid EOM concept is shown in Fig. 2.6. The placement

23
of the LN film is not arbitrary: the edge of the LN film must be placed over the wide Si region to
minimize reflection and scattering loss when going from the non-LN region into the LN region.
Non-LN EOM
X Y Wide Mode
Taper Mode LN Film
650 nm
Z LN Wide
Mode 320 nm

3-dB DC
Hybrid Region
PLD

Figure 2.6: Top view of the full EOM layout. DC = directional coupler, PLD = path-length
difference.

Directional couplers are used to split the input light into both arms of the EOM and then
split again at the output. The DCs should be 50/50 splitters (3-dB) to achieve the highest possible
device extinction ratio. High precision Si foundry processes enables accurate fabrication of DCs. Y-
junctions are not used because of their sub-resolution fabrication requirements. An extra path length
is included in one arm of the EOM so that the device can be tuned by changing the wavelength of
operation instead of having to apply a dc bias voltage, which would require either a high-frequency
bias tee or separate electrical lines and probes for the dc bias.

2.4 Design of the Electrical Lines

As mentioned previously, the electrical lines are designed so that the EOM is in the push-pull
configuration. This requires coplanar waveguide (CPW) electrodes, meaning that a signal line
is evenly placed between two ground planes, with the hybrid waveguides halfway between each
ground plane and the signal plane, as in Fig. 2.1a. The advantage of the hybrid Si-LN EOM design
is that the optical mode can be optimized separately from the electrical mode, since each mode
has its own set of geometrical components that are mostly unrelated. While the optical modes of
Section 2.2 were designed by varying the height and width of the Si waveguide and the height of
the LN film, the electrical mode in the EOM is designed by adjusting the height and width of the
ground planes and signal line, as well as the gap between the ground and signal. The only overlap

24
of design parameters between the optical and electrical modes is the gap between the electrical
lines; if wgap is too small then optical transmission will suffer. This is why it is important that Aeff
for the optical mode is kept as small as possible, particularly since Vπ L is strongly affected by wgap .

2.4.1 EOM Region

As discussed in the previous chapter, the electro-optic bandwidth is limited by three factors:
characteristic impedance mismatch, index mismatch, and excessive electrical losses. All three
of these factors must be taken into account when designing the electrical lines. Fortunately,
characteristic impedance (Zc ), electrical index (nm ) and electrical losses (αm ) can all be derived
from a single mode simulation, similar to the mode simulations used for optical modes but at
electrical frequencies. Characteristic impedance can be calculated from current (I) and power (P):

P
Zc = 2 , (2.3)
|I|2

both of which can be calculated from a mode simulation, while nm and αm can be obtained directly
from the eigenvalue of the simulated mode [9].
An important consideration during electrical line design is that of the metal thickness,
material, and deposition/growth method. Metal thicknesses beyond 1 or 2 micrometers require
electroplating, and as electrodes become thicker it is more likely their sidewalls will be slanted,
depending on the photoresist used. Several types of electrical lines, of different materials, are
experimented with in this work.
Additionally, the SOI substrate material should be carefully chosen when designing high
frequency devices. A low resistivity (high conductivity) Si handle will add loss to the electrical
mode. In this work, both a low resistivity and a high resistivity handle were used, but the high
frequency EOM was designed with a high resistivity Si handle in mind.

25
2.4.2 Electrical Bends

Although the EOM is a straight device, it is much easier and more practical to make electrical
contact at 90o to where optical contact is made. While characterizing the device, this is to avoid
electrical probe arms and optical fiber mounts from coming into contact, but even for a packaged
device this is typically desirable for the end user, who will want to connect electrical lines without
disturbing the relatively delicate and permanently connected optical fibers.
As a general rule of thumb, 90o electrical bends should have a radius at least three times that
of the width of the signal line. Too large of a bend will result in unnecessary electrical propagation
losses, and too small of a bend could result in high-frequency rolloff in the device’s frequency
response. To be safe, electrical designs in this thesis use a bend radius five times greater than the
signal line width.

2.4.3 Electrical Pads

Electrical contact is made at electrical pads, which should be designed to minimize re-
flections and additional loss. Reflections are minimized by maintaining a constant characteristic
impedance. This is achieved by leaving the signal line at a constant width, and increasing the ground
plane width to ensure that a wide variety of probe spacings can be accommodated (see Fig. 2.7).

EOM Region
Bends G
S
G

Pads

Figure 2.7: This is the GDS layout of a portion of the electrical lines, shown in white on a black
background. The ground lines are expanded at the pads to ensure that the probes can make contact,
and the signal line is left unaltered for minimal reflections.

26
Bibliography
[1] Chang WS. RF photonic technology in optical fiber links. Cambridge University Press; 2007.

[2] Noguchi K, Mitomi O, Miyazawa H. Millimeter-wave Ti: LiNbO3 optical modulators. Journal
of Lightwave Technology. 1998;16(4):615.

[3] Reed GT, Mashanovich GZ, Gardes FY, Nedeljkovic M, Hu Y, Thomson DJ, et al. Recent
breakthroughs in carrier depletion based silicon optical modulators. Nanophotonics. 2014;3(4-
5):229–245.

[4] Gardes F, Reed G, Emerson N, Png C. A sub-micron depletion-type photonic modulator in

silicon on insulator. Optics Express. 2005;13(22):8845–8854.

[5] Reed GT. Silicon photonics: the state of the art. John Wiley & Sons; 2008.

[6] Almeida VR, Panepucci RR, Lipson M. Nanotaper for compact mode conversion. Optics letters.
2003;28(15):1302–1304.

[7] Wood M, Sun P, Reano RM. Compact cantilever couplers for low-loss fiber coupling to silicon
photonic integrated circuits. Optics Express. 2012;20(1):164–172.

[8] Chen L, Doerr CR, Chen YK, Liow TY. Low-Loss and Broadband Cantilever Couplers Between
Standard Cleaved Fibers and High-Index-Contrast Si {3} N {4} or Si Waveguides. IEEE
Photonics Technology Letters. 2010;22(23):1744–1746.

[9] Toney JE. Multiphysics Modeling of Electro-Optic Devices. Comsol 2011. 2011;.

27
Chapter 3

A Process Design Kit for Hybrid Si-LiNbO3

Devices

3.1 Introduction

Modern CMOS circuit designers use process design kits (PDKs), generally supplied by
foundries, to design their electrical circuits. A typical PDK will include, at a minimum, a design
rules manual outlining how components should be designed to adhere to the foundry’s capabilities.
PDKs can also include a component library so that the designer does not need to run in-depth
physical simulations, and can instead work at the circuit level. Because of this, there is a lower
barrier to entry of CMOS circuit design. Additionally, designs are completed faster and at a
lower cost. The optical circuit community would be wise to follow this model, and indeed several
members of the community have begun to do so (as is evident from the list of proprietary component
libraries in PhoeniX’s OptoDesign software, as well as the PDK recently made available by Analog
Photonics, LLC through the AIM Photonics program [1]).
In the pursuit of simplifying the photonic circuit design process, parameterized cells (P-cells)
of hybrid silicon-lithium niobate (Si-LN) components in Lumerical’s INTERCONNECT software
have been created. A “P-cell” is a table of S-parameters (that can be optical, electrical, or even

28
electro-optical) for a given component with a certain geometry. A simple embedded waveguide, for
example, has a width and a height. The P-cell for that waveguide will have pre-simulated width
and height parameters built into a table (the S parameters), and when the user changes the width
and/or height of the waveguide, the P-cell will respond accordingly. In this way, a single component
designer can simulate vast combinations of devices, and package the results into “black boxes” that
can be used by optical circuit designers in e.g. Lumerical’s INTERCONNECT software. This has
two benefits: it prevents circuit designers from having to dedicate effort to time-consuming physical
simulations, and it naturally leads to a set of design rules for the circuit designer to follow, not
unlike the design rules used by CMOS circuit designers (such as waveguide spacing, minimum
bending radii, dopant levels, etc.).
While the individual P-cells developed over the course of this research are not included in
this thesis, the design rules manual for the hybrid Si-LN optical circuit platform is provided in this
chapter, including explanations and simulation data of the various components developed for this
unique hybrid silicon-lithium niobate optical platform.

3.2 Technology Summary

This document is a design manual for hybrid silicon-lithium niobate integrated optical
components. Primary applications of this technology are Pockels effect electro-optic modulators and
χ(2) nonlinear optical devices. This document contains a description of the technology, a fabrication
process overview, design rules, and a variety of library components that can be implemented
in Lumerical’s circuit simulation tool, Interconnect. Additionally, sample MATLAB scripts are
provided for GDS design.

3.2.1 Features of the Technology

The approach taken with this technology is to focus on bonding dies of SOI to dies of LNOI.
The reasons for this are three-fold: 1) To minimize the use of expensive LNOI material; 2) To

29
better match the smaller LNOI wafer (3 in.) to the larger SOI wafer (>6 in.); and 3) To utilize both
standalone Si and hybrid Si-LN optical circuitry on the same die.
To enable and accommodate a hybrid optical mode, the Si waveguide layer on the SOI is
set at 150 nm. This value is assumed throughout the remainder of this document. Optical circuit
components are designed by adjusting other parameters, such as the width of the Si waveguides and
the thickness of the LN (lithium niobate, LiNbO3 ) film.

30
3.3 Design Tools

The simulations in this document were carried out in Lumerical’s four software packages:
FDTD Solutions, Mode Solutions, DEVICE, and INTERCONNECT. FDTD solves Maxwell’s equa-
tions in 3D or 2D, MODE provides a full waveguide design environment with various solvers, and
DEVICE is a multiphysics environment for modeling charge and heat transport. INTERCONNECT
is a circuit-level simulation tool that allows for high-level circuit simulations.
FDTD Solutions, MODE Solutions, and DEVICE were used to simulate individual compo-
nents across different combinations of geometries. The results of those simulations were stored in
files accessible by models in INTERCONNECT. Those sorts of models are known as “parameterized
cells,” or “PCells” for short. Because in-depth simulations across a wide range of parameter values
have already been performed and uploaded into INTERCONNECT via the PCells, a designer can
simply change a component’s physical dimensions in INTERCONNECT to instantly and auto-
matically adjust the dependent properties (such as S-parameters or modal effective index) of that
component.

a b c d

Figure 3.1: a Linear waveguide taper optical transmission simulation in MODE. b Directional
coupler waveguide optical simulation in FDTD. c Direct current (DC) electric-field simulation in
DEVICE. d Circuit simulation of electro-optic modulator in INTERCONNECT.

31
3.4 Process Outline

The structures designed in this document are reliant upon a series of processing steps which
are outlined in Table 3.1 below.

Table 3.1: Process outline to fabricate hybrid Si-LN devices (active and passive).

Process Step Details Options

1a) Starting SOI Wafer 250 nm 1-0-0 p-type SOI with 3 µm BOX FZ high resistivity han-
dle wafer
1b) Starting LNOI 600 nm LNOI with 2 µm BOX Si or LiNbO3 handle,
Wafer film thickness from
300 nm to 1 µm.
2) SOI Thinning 80 nm etch of Si layer to 170 nm
3) SOI Oxidation 20 nm oxidation of Si layer to 150 nm final
thickness
4) Si Layer Definition Pattern Si layer
5) Oxide Deposition SiO2 PECVD deposition
6) CMP Chemical-mechanical polish of deposited ox-
ide with final target thickness of 190 nm (40
nm thicker than Si layer)
7) Clean SOI and 5 min. megasonic in: a) acetone bath; b) IPA
LNOI bath. Follow with 15 min. in RCA-1 bath.
8) PSA Plasma surface activation in PVA TePla
PS100: 30 sec./150 W, SOI and LNOI.
9) Post-PSA Water Place SOI and LNOI in DI-water bath for 5
Bath min.
10) Bond Dry, align, and contact SOI and LNOI surfaces If bonding manually,
to be bonded. Apply uniform pressure (<150 apply pressure first
kPa). in center to initiate
bond before pressuring
edges of bond region.

32
 Table 3.1: Process outline to fabricate hybrid Si-LN devices (active and passive), continued.

Process Step Details Options

11) Thermal Anneal 200◦ C/2 hours on a hot plate. Requires that Alternative heating
SOI and LNOI materials have identical or sim- methods can work as
ilar material properties (coefficient of thermal well.
expansion and modulus of elasticity).
12) Oxide deposition 2 µm SiO2 PECVD deposition
13) BB 220 Coating Protect exposed Si features and SOI handle
with two coats of BB 220. Soft bake: 80◦ C/3
min., 180◦ C/3 min., 220◦ C/5 min.
14) SiO2 Removal HF etch: 2 sec. BOE etch: 20 min.
15) XeF2 Etching Xactix: 2.0 mT/15 sec. per cycle/200 cycles
16) SiO2 Removal HF etch: 2 sec. BOE etch: 20 min.
17) Ti/Au Evaporation E-beam evaporation: 10 nm Ti, 50 nm Au Thicker Au is accept-
able but unnecessary
18) Spin-on AZ P4620 AZ P4260: 2000 rpm/40 sec.
19) Au Electrode Defi- Pattern AZ P4620 with electrode mask
nition
20) Electroplate Electroplate Au: 10 mA/20 min./8 µm Other combinations of
current and electroplat-
ing time are possible.
21) Remove Excess Remove AZ P4620 and seed layer of Ti/Au
Material that has not been electroplated.

Fig. 3.2 shows the cross-section after bonding and thermal anneal but before handle removal.
This sort of structure is useful for hybrid Si-LN optics. Fig. 3.3 shows the cross-section after handle
removal and metallization. This fully-fabricated structure is used for electro-optic devices, where it
is necessary to remove the upper handle in order to bring the metal lines close enough to allow the
electrical field to interact with the optical field.

33
 390
m
Si Handle Wafer
< 0.05 -cm
CVD Oxide 2
m

LiNbO3 0.600
m

CVD Oxide 0.040
m

CVD Oxide Si Si Si 0.150
m
Buried Oxide 3
m

650
m
Si Handle Wafer
>1000 -cm

Figure 3.2: Cross-section after bonding and thermal anneal. Each layer’s thickness is given, along
with resistivity values for the Si handles.

3.5 Design Rules for the Hybrid Si-LN Platform

This section details the design rules/suggestions for components in the hybrid Si-LN plat-
form. These rules are based upon limitations of the components rather than limitations of the
lithography process. Lithographic capabilities are dependent upon the foundry used for fabrication.
Please note that all design rules assume an optical signal within the wavelength range
[1500,1600] nm and a Si thickness of 150 nm.

3.5.1 Non-Hybrid Si Components

Standard Si photonic circuitry (that is, optical waveguides in Si surrounded by SiO2 ) is used
for things like coupling on and off the chip, splitting/combining light, and propagating light. To
avoid undesired results, such as unintentional waveguide coupling and excessive loss, the following
design rules should be observed:

1. Edge waveguide spacing of 40 µm.

2. Maximum width of 900 nm.

3. Minimum spline characteristic length (analogous to bend radius) of 10 µm.

34
 Au Au

LiNbO3 0.600
m

CVD Oxide 0.040
m

CVD Oxide Si Si Si 0.150
m
Buried Oxide 3
m

650
m
Si Handle Wafer
>1000 -cm

Figure 3.3: Cross-section after the full fabrication process. Each layer’s thickness is given, along
with resistivity values for the Si handles.

a b c
Si Si Si Si

0.90
m
10
m

40
m
10
m

Figure 3.4: a Minimum edge waveguide spacing. b Maximum Si waveguide width. c Minimum
spline bend radius.

3.5.2 Hybrid Si-LN Components

The hybrid section of this platform takes advantage of the fact that the indices of (anisotropic)
LiNbO3 (LN) are higher than the index of SiO2 . In such a scenario the optical mode resides in
the Si and LN regions of the cross-section more so than in the SiO2 in the substrate and lateral
regions. A simulation of this optical mode is shown in Fig. 3.5 for a Si width of 320 nm and a
LN height of 600 nm. Not every possible combination of Si width (wSi ) and LN height (hLN ) will
result in an acceptable hybrid mode. If the LN is too thick and the Si is too narrow, a slab mode will
propagate in the LN film. Alternatively, if the Si is too wide then higher order modes will propagate

35
and interact with the fundamental TE0 mode of interest. A listing of acceptable geometries for this
cross-section is provided in Table 3.2. Gray boxes are wSi /hLN combinations that are unacceptable,
while white boxes represent acceptable wSi /hLN combinations.

a Ex Ey Ez b Ex Ey Ez dB
1.0 0
1
m
600 nm

LN

600 nm
SiO2 0.8 10
m
-10
Si
Hx Hy Hz Hx Hy Hz
0.6 -20

0.4 -30
Sx Sy Sz Sx Sy Sz
0.2 -40

0.0 -50

Figure 3.5: a Linear plot of a simulated hybrid mode. wSi = 320 nm and hLN = 600 nm. The first
and second rows of plots are x, y, and z components of the E- and H-fields, respectively. The third
row of plots, labeled Sx , Sy , and Sz , are components of the time-average instantaneous Poynting
vector. b Log plot of Fig. 3.5a. Because this is the TE0 mode, it is very nearly TE-polarized (Ey ,
Ez , and Hx are close to zero; this is more obvious in the linear plot shown in Fig. 3.5a). Part b of
this figure is provided in dB for comparison with Fig. 3.6. Scale bars are given in the Ex plots.

Table 3.2: Table of acceptable combinations of values for Si width and LN thickness. In the upper
right of the table, one or more of the field components (Ex , Ey , Ez , Hx , Hy , or Hz ) has coupled into
the LN slab. In the lower section of the table, TE2 exists and can couple with TE0 . TE2 exists at a
narrower Si width (900 nm) in the non-hybrid region.
PP
PP
PP hLN (nm)
P PP 300 400 500 600 700 800 900
P
wSi (nm) PP
P P
P

200 Yes Yes No No No No No

250 Yes Yes Yes Yes Yes No No

300 Yes Yes Yes Yes Yes Yes No

350 Yes Yes Yes Yes Yes Yes No

36
 Table 3.2: Table of acceptable combinations of values for Si width and LN thickness. In the upper
right of the table, one or more of the field components (Ex , Ey , Ez , Hx , Hy , or Hz ) has coupled into
the LN slab. In the lower section of the table, TE2 exists and can couple with TE0 . TE2 exists at a
narrower Si width (900 nm) in the non-hybrid region, continued.
PP
PP
PP hLN (nm)
P PP 300 400 500 600 700 800 900
P
wSi (nm) PP
P P
P

400 Yes Yes Yes Yes Yes Yes Yes

450 Yes Yes Yes Yes Yes Yes Yes

.. .. .. .. .. .. .. ..
. . . . . . . .

900 Yes Yes Yes Yes Yes Yes Yes

950 No Yes Yes Yes Yes Yes Yes

1000 No No Yes Yes Yes Yes Yes

1050 No No Yes Yes Yes Yes Yes

1100 No No No Yes Yes Yes Yes

1150 No No No No No Yes Yes

1200 No No No No No Yes Yes

1250 No No No No No Yes Yes

1300 No No No No No No Yes

1350 No No No No No No Yes

1400 No No No No No No No

The reasons why certain combinations of wSi and hLN are unacceptable for waveguiding
go back to the index of refraction for each material in the cross-section. Silicon has a higher index
than either the ordinary or the extraordinary index of LN (nSi ≈ 3.47, nLN ≈ 2.21, 2.14). When
the Si waveguide below the LN slab (Fig. 3.2 and Fig. 3.3) becomes too narrow, the optical mode
is no longer laterally confined by the Si waveguide and becomes a slab mode. Particular field

37
components can behave as higher order slab modes, like Ey , Ez , and Hx in Fig. 3.6. This is why
certain combinations of wSi and hLN are gray in the upper right of Table 3.2. Conversely, the lower
left of Table 3.2 is grayed to mark the scenario where the Si waveguide is wide enough to support
the higher order TE20 mode, which can couple with TE00 and act as a source of optical loss.

Ex Ey Ez dB
900 nm 15
m 0
LN
SiO2 -10
Si
Hx Hy Hz
-20

-30
Sx Sy Sz

-40

-50

Figure 3.6: Log plot of a simulated hybrid optical mode when wSi = 200 nm and hLN = 900 nm,
i.e. one of the gray boxes in the upper right corner of Table. 3.2. In this case, the Ex , Hy , and Hz
components are nearly slab modes, while the Ey , Ez , and Hx components are non-zero. This is in
contrast to the well-behaved mode in Fig. 3.5.

3.6 Static Library Components

We have developed several static library components for use in Lumerical’s INTERCON-
NECT circuit-level simulation software. Each component is “static” in the sense that it has fixed
values for all physical parameters, which results in dependent data like effective index and coupling
coefficient changing only with wavelength. This is in contrast to the following section, which
discusses parameterized cells (PCells) with variable input physical parameters and therefore variable
parameters of interest (effective index, coupling coefficient, etc).
It is in this section that each component is described in detail, using optimized physical
values for a Si thickness of 150 nm.

38
3.6.1 Edge Coupling Taper

We use linear tapers as edge couplers to couple from an off-chip fiber into the thin-silicon
platform. These linear tapers start at a width of 205 nm and end at a width of 650 nm, the nominal
waveguide width for this platform. The top view of this structure is shown in Fig. 3.7a. After
simulating taper loss for these width values as a function of taper length in Fig. 3.7b, we decided to
set the taper length to 200 µm. This assumes 3 µm of SiO2 below the Si waveguide, and 2 µm of
SiO2 above the Si waveguide.

a b Thin Si Linear Input Taper

1.000
205 nm 650 nm

Transmission
Ltaper 0.995
In Out 0.990
(0,0)
Si
0.985

0.980

0.975
10 100 200 300 400 500
Taper Length (
m)

Figure 3.7: a Top view of optimized input/output waveguide taper. Taper length Ltaper is varied
from 10 µm to 500 um in MODE Solutions. b Plot of edge taper transmission versus taper length
at λ = 1550 nm for the taper in a.

3.6.2 Thin-Si Waveguide

Routing light from the edge of the chip to the bonded region is done with a simple buried
channel waveguide of high-index Si (nSi ≈ 3.47) surrounded by low-index SiO2 (nSiO2 ≈ 1.44). We
use a Si width of 650 nm for these thin-Si waveguides. At λ = 1550 nm, nneff = 2.29.

3.6.3 Thin-Si Spline Bends

We use cubic spline curves (also known as cubic Bézier curves) to create our waveguide
bends. A cubic spline curve is known to be the smoothest of all functions (though not necessarily the

39
 SiO2 Si

150 nm
1
m

Figure 3.8: Simulation of Enorm (the norm of all three E-field components) of the non-hybrid
mode. Since Ey and Ez are so small for this TE-like mode, Enorm ≈ Ex . wSi = 650 nm. As with all
Si waveguides in this document, hSi = 150 nm.

most space-efficient) [?], and for this reason is a popular choice in waveguide design to minimize
bending losses [3]. The functional form of the cubic spline curve that we use is as follows:

S(t) = (1 − t)3 P0 + 3(1 − t)2tP1 + 3(1 − t)t 2 P2 + t 3 P3 , 0≤t ≤1 (3.1)

P0 and P3 are knots – the starting and ending points of the curve – while P1 and P2 are
control points; that is, points that define the shape of the curve. Each point Pi has an x- and a
y-coordinate to describe its location, and in practice (3.1) is broken down into its spatial components
when generating a spline curve. An example of a random spline curve is shown in Fig. 3.9.

P3
P1

P2

P0
Figure 3.9: Cubic spline curve with knots P0 and P3 , and control points P1 and P2 .

90◦ Bends

If P1 and P2 are set equal to each other and are arranged to form a right triangle with P0 and
P3 , as shown in Fig. 3.10a, then the spline curve becomes a 90◦ spline bend. When two such 90◦
spline bends are generated next to one another, offset by some distance wSi , and connected at their

40
endpoints, a 90◦ spline waveguide bend is created. A simple 90◦ spline bend is shown in Fig. 3.10a,
and a 90◦ spline waveguide bend is shown in Fig.3.10b. Bending loss simulations were carried
out in FDTD Solutions for a constant waveguide width wSi of 650 nm. The results are shown in
Fig. 3.10c.

a b c

Transmission (dB)
P1 = P2 0
P3
Lc = 5 μm
-0.1 Lc = 10 μm
Lc
Lc = 15 μm
wSi -0.2

1500 1540 1580

P0 (0,0) Wavelength (nm)

Figure 3.10: a 90◦ spline bend with knots P0 and P3 , and control points P1 and P2 . b Cubic spline
waveguide 90◦ bend with width wSi and characteristic length Lc . Lc is measured from the middle
of the input waveguide to the edge of the output waveguide (or vice versa). c FDTD simulation
of 90◦ waveguide bend of various characteristic lengths, Lc , for the waveguide shown in Fig. 3.8.
For Lc ≥ 10 µm, simulated bending loss is insignificant across the 1500 nm - 1600 nm range of
interest.

S Bends

If the optical signal needs to be offset laterally, an S bend can be used. We construct S
bends using the spline curve by placing the control points as shown in Fig. 3.11, where the vertical
distance from P0 to P1 and P2 is one-half the vertical distance from P0 to P3 . P1 and P0 share the
same x-coordinate, as do P2 and P3 . The S bend is characterized by its lateral offset Lx and its
longitudinal length Ly .

3.6.4 Thin-Si 3-dB Directional Coupler

The components discussed so far can be put to use as pieces of larger components, such as
directional couplers. A directional coupler is created when two (or more) waveguides come into
close enough proximity of one another for their evanescent fields to interact, resulting in a weak
“coupling” of energy between the waveguides. A simple schematic of such a device is given in

41
 a b
P0 P1
(0,0)

wSi Ly

P2 P3 Lx

Figure 3.11: a Cubic spline S bend. b Cubic spline waveguide S bend with width wSi , lateral
offset Lx , and longitudinal length Ly .

Fig. 3.12. Our directional couplers were designed to split input energy evenly, such that an input in
one of the two input ports (A and B in Fig. 3.12) would split evenly into both output ports (C and D
in Fig. 3.12). This is known as a 3-dB directional coupler.
½ Lex Lcoup ½ Lex

A C

wgap
B wSi D
(0,0)

Figure 3.12: Generic directional coupler. Lcoup is the straight-section coupling length, and Lex
is the additional coupling length in the region that angles away from the straight region. The
coupling matrix comes from section 6.2.2 in [4].

Coupling from the inputs to the outputs of a directional coupler can be calculated by matrix
multiplication (assuming that A, B, C, and D are electric field components). Considering the inputs
and outputs of Fig. 3.12, we have the following equation [4]:

    
C  t iκ A
 =   (3.2)
    
D iκ t B

where t and κ are taken to be real, and |t|2 + |κ|2 = 1. A 3-dB directional coupler (that is, one which
splits half of its input light intensity out each of the two outputs, assuming a single input) is one
where the coupling coefficient |κ|2 = 0.5.

42
Symmetric Directional Coupler

Fig. 3.13a shows the GDS layout of a symmetric directional coupler. Lcoup = 50 µm and
Lex = 40 µm (Lc = 10 µm for each of the 90◦ spline bends). Referring to the dimensions shown
in Fig. 3.12, wgap = 280 nm and wSi = 650 nm. A plot of wavelength-dependent |κ|2 is shown in
Fig. 3.13b.
a ½ Lex ½ Lex
b
A C 0.8

0.7
Lcoup

| |2


0.6

0.5

0.4
B D 1500 1520 1540 1560 1580 1600
Wavelength (nm)

Figure 3.13: a GDS layout of a symmetric directional coupler. Lcoup = 50 µm, Lex = 40 µm of 90◦
spline bends, the gap in the straight region is 280 nm, and each waveguide has a width of 650 nm.
b Coupling coefficient |κ|2 for the coupler in a as a function of wavelength.

Asymmetric Directional Coupler

The symmetric directional coupler in the previous section has one considerable drawback: it
varies strongly with wavelength. From Fig. 3.13b, |κ|2 changes by more than 0.4 over the range of
λ = [1500,1600] nm. Many, if not nearly all, practical applications of a directional coupler require a
constant coupling coefficient over a wide range of wavelength values. One way to do this is to make
the coupling length (LLcoup + Lex ) very short, but then |κ|2 is much too low for most applications. A
better design path is to make an asymmetric directional coupler [5]. This sort of design accounts for
varying wavelengths by tapering the width of one of the waveguides in the straight coupling section.
When this is done correctly, broadband coupling can be achieved. Fig. 3.14a shows the design of an
asymmetric directional coupler, and Fig. 3.14b presents simulated results of the coupling coefficient
from FDTD Solutions for the structure of Fig. 3.14a.
Referencing the variables in Fig. 3.14a, w1 = 650 nm, w2 = 400 nm, w3 = 300 nm, wg = 280
nm, L1 = L3 = L4 = 30 µm, and L2 = 325 µm. All Si features are 150 nm thick.
There is some wavelength dependency in the output of this asymmetric directional coupler –

43
 a b
|B|2
|C|
2

L1 L2 L3 L4 0.52
|B|2
|D|
2

Transmission
w1 C 0.50
A w1 w2
w3 
2 m


2 m w2
wg
w1 D
0.48

B w1
0.46
(0,0)
1500 1520 1540 1560 1580 1600
Wavelength (nm)

Figure 3.14: a Schematic of an asymmetric directional coupler. Coupling occurs most strongly
in section L2 , less strongly in sections L1 and L3 , and not at all in section L4 . b This plot shows
the transmission through C and D for an input in B. Total transmission is around 98% across the
full wavelength range shown. Power splitting from |B|2 to |C|2 and |D|2 is nearly independent of
wavelength.

assuming B as the input – but it is much less significant than with the standard directional coupler
of the previous section. Across this same wavelength range, |∆κ|2 ≈ 0.4 (|∆κ|2 is the change in
|κ|2 across the [1500,1600] nm wavelength range) for the standard directional coupler. For this
asymmetric directional coupler, ∆|κ|2 ≈ 0.06, a factor of 7 improvement. The drawback to using the
asymmetric directional coupler is that it is about seven times longer than the symmetric directional
coupler. This trade-off between the two structures should be evaluated by the designer.

3.6.5 Mach-Zehnder Interferometer

A Mach-Zehnder interferometer (MZI) can be made in a waveguide structure by splitting

light and then recombining it, either with or without a passive path-length difference (PLD) between
the two arms of the structure. A top view of a path-length imbalanced MZI is shown in Fig. 3.15.
A PLD MZI uses non-hybrid Si tapers to couple on and off the chip, non-hybrid Si WGs, either
a symmetric or an asymmetric DC as the splitter, and either hybrid waveguides or non-hybrid
waveguides in the path-length difference section.
The PDK developed for this hybrid Si-LN platform includes an optical MZI component that
can be simulated in Lumerical’s INTERCONNECT software.

44
 Out
50/50 50/50
In Splitter Splitter Out

Figure 3.15: Top view of an MZI structure. Black lines represent waveguides and the path-length
difference on the top arm is highlighted in yellow. Each 50/50 splitter is either a symmetric DC or
an asymmetric DC (there are other types of WG-based splitters, but they are not part of this design
kit).

3.6.6 Hybrid Si-LN Taper

Linear tapers were discussed in Section 3.6.1 to be used as edge couplers to couple light
from an input fiber to the chip, or from the chip to an output fiber. A similar taper concept is used
in the hybrid region of the chip to convert light from the optical mode of Fig. 3.16a to that of
Fig. 3.16b.

Figure 3.16: a Ex field of optical mode in hybrid Ssi-LN region with wide (wSi = 650 nm) Si. b
Narrow (wSi = 320 nm) Si. c Top view of hybrid Si-LN taper. The taper is in the Si ridge layer,
and LN film is bonded over the entire area. d Plot of transmission versus taper length. The final
taper design uses a taper length Ltap = 200 µm.

This taper is what allows the hybrid Si-LN platform to be so versatile. By reducing the
width of the Si ridge, the fraction of the optical mode in the LN film increases. This is because
LN (the cover material) has a higher index of refraction than SiO2 (the substrate material). As the

45
Si waveguide width is reduced, less light is confined by the Si, despite its higher index, because
its width is smaller than one wavelength in silicon (≈ 447 nm). If the Si were embedded in SiO2 ,
then the evanescent fields would spread equally into the cover and substrate. Since the cover has a
higher index, as well as a non-infinite thickness, the mode’s evanescent fields spread deeper into the
LN cover and are vertically confined by the upper edge of the LN film. Horizontal confinement
is maintained by the sidewalls of the Si ridge waveguide, and thus this hybrid Si-LN platform is
capable of guiding a mode that exists in both the Si and LN regions. Proper design of this hybrid
linear taper results in a useful hybrid mode.
Simulations of this taper were performed in MODE Solutions using the EME solver to
analyze the hybrid linear taper, whose top view is shown in Fig. 3.16c where w1 = 650 nm and w2 =
320 nm (these are the input and output taper widths, respectively). In this component, the linear
taper is made in the Si ridge layer and LN thin film (hLN = 600 nm) is bonded over the entire taper
region. Taper losses become negligible beyond a taper length Ltaper of 100 µm (Fig. 3.16d). We
chose Ltaper = 200 µm for our final design’s taper length.

3.6.7 Hybrid Si-LN Waveguide

Once the hybrid waveguide is tapered down to the desired width (in this case, that is wSi =
320 nm), it can be used in hybrid optical designs. A plot of Ex for this mode is shown in Fig. 3.16b.
This waveguide is designed so that as much of the optical mode sits in the LN region as possible,
without allowing the mode to decouple from the Si rib waveguide. A plot of mode area and LN
confinement factor (LNCF) is provided in Fig. 3.17. LNCF is defined as the percentage of power
traveling along the hybrid waveguide that is produced by the TE component of the mode:

∗
R
LN (Ex Hy )dxdy
LNCF = R∞
∗ ∗
(3.3)
−∞ (Ex Hy − Ey Hx )dxdy

46
 a b
2.00 80
70
1.50

LNCF (%)
Aeff ( m2)

60

1.00 50
40
0.50
30
0.00 20
200
Si Rib Width (nm)

Figure 3.17: a Plot of simulated effective mode area (left axis, black circles) and LNCF (right
axis, red circles). b Plot of simulated effective index (left axis, black circles) and group index
(right axis, red squares). The dashed vertical blue line in each part shows where the mode chosen
to use sits. Aeff = 0.67 µm2 , LNCF = 60%, neff = 2.04, and ng = 2.73.

3.6.8 Hybrid Si-LN Electro-Optic Phase Shifter

A single hybrid Si-LN waveguide from Section 3.6.7 can be made into a high-speed electro-
optic (EO) phase shifter (PS) by launching a radio-frequency (abbreviated as RF, accounts for the
frequency range of several kHz to 300 GHz) wave down a transmission line that runs parallel to the
hybrid waveguide. The transmission line must be physically close to the hybrid waveguide – on
the order of micrometers – to maximize the overlap of the RF and optical electric fields. A typical
cross-section for this structure, using x-cut LN and a coplanar waveguide (CPW) structure for the
transmission line, is shown in Fig. 3.18.

3.6.9 Coplanar Waveguide Transmission Line

The lumped element model for circuit elements is valid only when the RF frequency fRF
c
<< nd , where c is the speed of light, n is the phase index of the RF mode in the circuit element,
and d is the length of the element. As an example, consider a fairly typical phase index of 2.2 in a
10 cm element; in such a scenario, an element with an operating frequency well below 1.4 GHz can
be treated as a lumped element and, in the case of electro-optic devices, is generally treated as an
RC-limited low-pass filter.
The hybrid EO devices discussed in this work are designed with high speed (tens to hundreds

47
 Au Au

8 m 8 m

LiNbO3 0.600
m

CVD Oxide 0.040
m

CVD Oxide Si 0.150
m
Buried Oxide 3
m L
650
m
Si Handle Wafer
>1000 -cm

Figure 3.18: 3D depiction of the final cross-section from Fig. 3.3, but for the single Si waveguide
of a hybrid Si-LN EO phase shifter. Only the straight section of the PS is shown here. Each layer’s
thickness is given, along with resistivity values for the Si handles. The EO PS has length “L.”

of GHz) in mind, in which case the lumped element model is invalid and is replaced with a traveling
wave transmission line model [6]:

∆φ(ω) RL + RG Zin (ZL + Z0 )F(u+ ) + (ZL − Z0 )F(u− )

m(ω) = = · · , (3.4)
∆φ(0) RL Zin + ZG (ZL + Z0 )eγm L + (ZL − Z0 )e−γm L

1−eu±
where m(ω) is the normalized modulation response, F(u± ) = u± , and u± = j(±βm − βo )L ± αm L
= ±αm L + j ωc (±nm − no )L. RL is the load resistance, RG is the source resistance, and Zin is the RF
input impedance of the loaded transmission line, defined as:

ZL + Z0 tanh(γm L)
Zin = Z0 . (3.5)
Z0 + ZL tanh(γm L)

ZG in (3.4) is the source impedance (typically this will be ZG = RG = 50 Ω). ZL is the load
impedance, Z0 is the characteristic impedance of the transmission line, γm = αm + jβm is the
complex RF propagation constant, αm is the RF attenuation coefficient in cm−1 , βm is the phase
constant (also in cm−1 ), nm is the RF phase index, no is the optical group index, and L is the

48
interaction length along which the RF and optical modes interact. An additional term ∆n = nm − no
is defined for convenience and future use.
From (3.4), the following are evident:

1. ∆n must be made as small as possible;

2. αm must be minimized;

3. ZL and Z0 should be made equal, if possible.

Each of these design conditions is investigated separately in Fig. 3.19a-c, and a more in-depth
analysis is given in [6]. For the purposes of design optimization, what is important to know is that
these three conditions limit electro-optic bandwidth. Note that in Fig. 3.19, RF losses are given in
terms of per (GHz)1/2 . This is because in a well-fabricated electro-optic modulator, RF losses are
dominated by conductor losses, which scale with the square root of frequency. By using units of
dB/(cm-GHz1/2 ), a single loss value can be used to describe losses across a range of frequencies.
In the hybrid Si-LN modulator, this assumption is fairly accurate as long as the Si substrate has a
(relatively high) resistivity of at least 1000 Ω-cm.

a 1
b 1
c 1
Modulation Index, dB (optical)

Modulation Index, dB (optical)

Modulation Index, dB (optical)

0 0 0

-1 -1 -1

-2 -2 -2

-3 ZG = ZL = 50
, Z0 = 40
-3 RF = 0.5 dB/(cm-GHz1/2) -3 n = 0.05

-4 ZG = ZL = 50
, Z0 = 50
-4 RF = 1.0 dB/(cm-GHz1/2) -4 n = 0.10
ZG = ZL = 50
, Z0 = 60
RF = 1.5 dB/(cm-GHz1/2) n = 0.20
-5 -5 -5
1 10 100 1 10 100 1 10 100
Frequency (GHz) Frequency (GHz) Frequency (GHz)

Figure 3.19: Plot of (3.4) when varying: a Characteristic impedance Z0 , with ∆n = 0.10 and αRF =
1 dB/(cm-GHz1/2 ); b RF losses αRF , with Z0 = ZG = ZL = 50 Ω and ∆n = 0.10; and c RF-to-optical
index difference ∆n, with Z0 = ZG = ZL = 50 Ω and αRF = 1 dB/(cm-GHz1/2 ). L = 1 cm for all
simulations.

49
3.7 Acknowledgments

Chapter 3, in part, is a reprint of the material as it appears in Conference on Lasers and

Electro-Optics (CLEO) 2017. Peter O. Weigel and Shayan Mookherjea, paper JTu5A.126. The
dissertation author, together with his adviser, led the research efforts for this work and co-authored
the paper.

Bibliography
[1] Timurdogan E, Su Z, Poulton C, Byrd MJ, Xin S, Shiue RJ, et al. AIM Process Design Kit
(AIMPDKv2. 0): Silicon Photonics Passive and Active Component Libraries on a 300mm
Wafer. In: Optical Fiber Communication Conference. Optical Society of America; 2018. p.
M3F–1.

[2] Farin G. Curves and surfaces for computer-aided geometric design: a practical guide. Academic
Press; 1992. Print.

[3] Bogaerts W, et al. Compact single-mode silicon hybrid rib/strip waveguide with adiabatic bends.
IEEE Photonics Journal. 2011;3(3):422–432.

[4] Aguinaldo RF. Silicon Photonics with Applications to Data Center Networks. University of
California San Diego; 2014.

[5] DeRose CT, Watts M, Young RW, Trotter DC, Nielson GN, Zortman W, et al. Low power and
broadband 2 x 2 silicon thermo-optic switch. In: Optical Fiber Communication Conference.
Optical Society of America; 2011. p. OThM3.

[6] Ghione G. Semiconductor devices for high-speed optoelectronics. Cambridge University Press;
2009. Print.

50
Chapter 4

Fabrication of Hybrid Electro-Optic Devices

4.1 Introduction

Fabrication of this device was performed at two locations. First, 150 mm silicon-on-insulator
(SOI) wafers from SOITEC, Inc. were fabricated at Sandia National Laboratories in Albuquerque,
New Mexico with standard foundry processing techniques (i.e. photolithography, reactive ion
etching, silicon oxidation, etc.) as well as a chemical-mechanical planarization (CMP) step for
<1 nm surface roughness. Wafer dicing/edge polishing, LiNbO3 (LN) film bonding and handle
removal, and metallization were performed at UC San Diego. Details of these fabrication efforts are
detailed in this chapter.

4.2 Film-on-Insulator Wafers

Modern silicon photonics is built not on bulk silicon wafers, but on silicon-on-insulator
(SOI) wafers. Modern SOI wafers are a sub-class of “film-on-insulator” wafers produced by the
Smart-Cut process (also known as crystal ion slicing, or CIS), first patented in 1994 by Michel
Bruel [1]. This process, depicted in Fig. 4.1, implants a bulk wafer with ions of a certain type
(though typically hydrogen or helium) at a particular energy, chosen so that the ions reach a desired
mean depth into the material (Fig. 4.1a). That bulk, implanted wafer is then bonded to another

51
wafer (the second wafer can theoretically be any material as long as it is smooth and flat enough for
bonding; bonding is discussed in detail in Section 4.4) and heated until the implanted ions expand
until they cause the first wafer to crack along the implanted layer (Fig. 4.1b-c). A final polishing
step is performed, and what results is a wafer comprised of a thin film of crystalline material – with
the same material properties as its bulk wafer form [2, 3, 4, 5] – bonded to (usually) an insulator
such as silicon dioxide (SiO2 ) on some bulk material handle (Fig. 4.1d).

a Ion Implantation b Bond c Heat and "Slice" d Polish/Planarize

W1 W1
Film (from W1)

W1 W2
W2 W2 W2

Figure 4.1: The basic crystal ion slicing process is depicted in four steps: a The first wafer (W1)
is ion-implanted. b W1 is bonded to the second wafer (W2), comprised of some handle material
covered with a thin, smooth insulator material. c The bonded stack is heated until W1 cracks
off (“slicing”), leaving a thin rough layer of crystalline material. d Polish and planarize the thin
crystal layer.

Film-on-insulator wafers have been used in guided-wave optics since they were invented
[6]. They feature thin, crystalline films that can be selectively patterned to form small waveguides
with high mode confinement and low loss. SOI wafers in particular are attractive as a guided-wave
material for several reasons, a few of which are listed here: silicon waveguides can potentially be
integrated with CMOS technology, though this requires a CMOS foundry to develop a fabrication
process which accounts for optical waveguides; silicon has a high index (approximately 3.5) in
the telecommunications windows above λ = 1.15 µm (its bandgap), which is useful for densely
packed photonic circuits; as a semiconductor, silicon can be p or n doped to create useful structures,
including p-i-n waveguides with uses as heaters, switches, and modulators [7, 8, 9].
More recently, research into LN films-on-insulator (LNOI) has emerged due to the availabil-
ity of commercial LNOI (NanoLN, Jinan Jingzheng Electronics Co. Ltd.). LN was the guided-wave
material of choice until the mid to late 90s [10], when SOI wafers started to become available and

52
a large portion of the field moved towards developing optics on SOI (now commonly called Si
photonics). However, (bulk) LN remains one of the most ubiquitous materials in the industry, and
is still used for electro-optic devices worldwide. With the emergence of readily available LNOI
wafers, there is renewed interest in LN photonics [11].
The fabrication in this chapter begins with an SOI wafer from SOITEC and an x-cut LNOI
wafer from NanoLN. Both wafers are polished to <1 nm surface roughness and <20 µm flatness at
their respective fabrication facilities.

4.3 SOI Patterning & Oxide Polishing

SOI patterning was performed at Sandia National Laboratories on standard 220 nm-thick
Si on a 3 µm SiO2 layer. The Si layer was thinned down to 150 nm via reactive ion etching
(RIE, “dry” etching) before Si patterning was performed using wafer-scale deep ultraviolet (DUV)
photolithography and RIE, followed by high-density plasma chemical-vapor deposition (HDP
CVD) of approximately 1 µm of SiO2 . A chemical-mechanical planarization/polishing (CMP)
process was developed at Sandia to bring the SiO2 thickness down to around 350-400 nm with low
surface roughness and good flatness. Further oxide thinning was performed with a highly diluted
hydrofluoric (HF) acid etch – 100 parts aqueous ammonium fluoride (NH4 F, 40%) to 1 part aqueous
HF (49%) – to reduce the oxide thickness even further. This process flow is shown in Fig. 4.2.
However, the initial fabrication efforts did not incorporate a wet etch step after CMP. This
resulted in over-polishing of the wafers, to the point where sections of the Si waveguides were
too thin to propagate light; in some cases, the Si waveguides disappeared entirely. By targeting
a thicker oxide layer with CMP and then slowly etching down to the desired thickness, bondable
devices could be fabricated. An oxide map of a properly polished and wet etched wafer is presented
in Fig. 4.3b. Negative values indicate where the oxide thickness is below the top surface of the
patterned thin Si layer. From Fig. 4.3b it is clear that wafer flatness after CMP can be improved,
though individual dies were typically flat enough for bonding. Surface roughness was consistently

53
 a Thin Si SOI Wafer b Photolithography c Si Patterning
Photoresist

150 nm Si
3
m SiO2 3
m SiO2 3
m SiO2

Si Substrate Si Substrate Si Substrate

d SiO2 Deposition e CMP f Wet Etch

200-250 nm Si WGs
~1 m <100 nm

SiO2 SiO2 SiO2

Si Substrate Si Substrate Si Substrate

Figure 4.2: a A standard 220 nm SOI wafer from SOITEC is thinned to 150 nm with RIE. b
Photoresist is spun on the wafer and patterned with DUV lithography. c RIE is used to etch the
photoresist pattern into the thin Si layer. d Approximately 1 µm of oxide is deposited via HDP
CVD onto the patterned wafer. e A CMP recipe, customized for this fabrication process, is used
to bring the oxide thickness to several hundred nanometers. f A diluted HF wet etch is used to
further reduce the oxide thickness. WG = waveguide.

low enough for bonding (<1 nm rms). Fig. 4.3c shows a properly fabricated SOI wafer, as delivered
to UCSD from Sandia.
a b c
100
SiO2 Thickness (nm)

Si WG
50

0
20 nm
-50

Figure 4.3: a SEM image of an over-polished waveguide. The measured value of only 20 nm is
130 nm below the targeted thickness, and is too thin to guide light. b Wafer map of oxide thickness
over Si features after CMP and wet etch in optimized process flow (Fig. 4.2f). As is typical with
first-attempt wafer CMP processes, wafer flatness is not yet ideal. This can be improved with
optimization efforts. c Fully fabricated 150 mm SOI wafer from Sandia, corresponding to the
wafer map of b.

54
4.4 Bonding Theory

Bonding, as its name implies, is the process of bringing two materials into contact in such a
way that they remain in contact after external forces are removed. While bonding is used colloquially
to describe everyday adhesives, such as glue and tape, in the context of this work the word “bonding”
refers to the action of bringing two planar, smooth surfaces into contact and being held together by
various intermolecular forces, namely Van der Waals forces (including hydrogen bonds, also known
as dipole-dipole bonds) and covalent bonds [12]. Under the correct conditions – to be discussed in
this section – large surfaces of dissimilar materials can be brought into contact and remain bonded,
despite changes in stress caused by sudden temperature variations.

4.4.1 Types of Bonds

The different types of bonds used in this thesis are defined as follows:
Direct Bonding: Room-temperature bonding between the two materials of interest; uses van der
Waals forces
Fusion Bonding: Covalent bonding between the two materials of interest; can be done at room
temperature under the right conditions [16], but typically requires a thermal annealing process [12]
Adhesive/Indirect Bonding: Using an adhesive layer between the two materials of interest during
bonding

These bond types are not mutually exclusive; it is possible to have a direct fusion bond, an adhesive
direct bond, or even an adhesive fusion bond.

Direct Bonds & Fusion Bonds

A direct bond is one in which two materials are physically placed in close enough proximity
(on the order of angstroms) for intermolecular forces to bind them to one another after external
forces are removed. Direct bonding was first observed by Sir Isaac Newton, though he had no
concept of “bonding” at the time; in reference to two pieces of polished glass coming into contact,

55
he described it as a “central black spot” at the point of “optical contact” [13]. Newton’s name
still bears relevance to the science of bonding, as the colored rings surrounding a bonded spot are
commonly called Newton rings.
There are different definitions of direct bonds and fusion bonds throughout the literature,
where in some cases the only attributed distinction between the two is the temperature at which
the bonding occurs [13]. This is not technically correct. Instead, the type of intermolecular bond
defines whether the materials are direct bonded or fusion bonded. A fusion bond is one in which two
materials are joined by chemical bonds, while a direct bond is simply two materials held together
by van der Waals forces.
Direct bonding became a subject of great interest for the semiconductor and Si photonics
communities (amongst others) after Bruel’s 1994 patent of the Smart-Cut process [1]. Direct
bonding via the Smart-Cut process quickly accelerated from a research curiosity to a mainstream
industry practice. In the silicon photonics community, for example, nearly all silicon-on-insulator
(SOI) wafers are purchased from SOITEC, Inc., as opposed to being fabricated separately at each
laboratory or institution.
These intermolecular forces are typically van der Waals [12], though much stronger covalent
bonding at room temperature is possible in vacuum with plasma surface activation (PSA) [16].
However, due to the versatility of bonding by hand, particularly with small dies of varying sizes, all
bonding discussed in this thesis was performed at room temperature and pressure.

Adhesive Bonds

An adhesive bond is similar in some ways to a direct bond, in that some direct contact, often
at room temperature, occurs. However, the two layers desired to be bonded are no brought into
direct contact. Instead, at least one of them is coated with a thin adhesive before bonding and then
curing. In some cases, a partial cure precedes the bonding step. Adhesive bonding is advantageous
due to the physical properties of adhesives – namely, they tend to be in liquid or gel form before
being cured, so that typically strict bonding requirements such as low surface roughness and good

56
flatness become much less severe. The downsides to adhesive bonding are the separation between
the two materials being bonded, and aging/thermal limits of the adhesive material.

Practical Considerations during Bonding

This thesis deals with all three types of bonding discussed previously. The adhesive used in
this thesis is benzocyclobutene, or BCB (many other types of adhesives can be used for bonding,
depending on the application [14]). BCB is a polymer commercially available in an aqueous form
that can be spin coated onto wafers or dies and subsequently hardened through a thermal annealing
process (i.e. curing). However, BCB has some thickness to it (though while in solution it can
be diluted so that a cured film has a thickness below 50 nm [15]), which may not be suitable for
certain applications. And like all polymers, BCB has a limited thermal budget (glass transition
temperature Tg <350◦ C) and can degrade over time. The primary advantage to using BCB and
polymers like it is the reduction in bonding fabrication challenges, particularly at the wafer scale
where the spin-coated BCB layer can be made extremely uniform in place of a wafer polishing
process [15].
If the fabrication challenges associated with unassisted direct bonding can be overcome, it is
generally the most preferred form of bonding; direct bonds can be performed at room temperature
and annealed to form fusion bonds, which can be as strong as covalent bonds in a homogeneous
crystal with the proper annealing process and surface conditions [12], and require no polymer
adhesive interlayer material. Particularly for high purity crystalline wafers, direct bonding is
attractive for both its robustness and cleanliness. That being said, direct bonding requires pristine
surface quality of the two wafers being bonded – even a single particle can ruin an entire wafer
bond if the particle is unfortunately large – as well as low surface roughness and very good flatness
across the wafer to ensure that the full wafer is bonded.

57
4.5 Reducing Thermal Stress in Heterogeneous Bonded Mate-

rials

Despite advances in the last decade or so in silicon photonics, lithium niobate continues to
find uses in advanced areas of integrated photonics due to its favorable nonlinear properties and
low propagation loss [17, 18, 19, 20]. Recently, researchers have been studying hybrid integrated
photonic devices in which silicon-on-insulator (SOI) is bonded to lithium-niobate-on-insulator
(LNOI), creating single-mode hybrid waveguides with sub-micron cross-sectional dimensions
[21, 22, 23, 24, 25, 26]. Bonded materials are sometimes subject to additional fabrication steps,
for example, to pattern waveguides [27], and such processing generates stress in the layers due
to a difference in coefficients of thermal expansion (CTEs) between the different materials and
can cause irreversible damage. Similar to the work done earlier in bonding laser material to SOI
[28, 29], it is important to identify the parameters and properties of the inter-layer bonding material,
such as benzocylobutene (BCB), to mitigate the stresses [30, 14, 15, 31, 32].
The approach to hybrid SOI-LNOI devices we study here is to bond a SOI chip (diced from
SOI wafers from Soitec, Inc., and patterned using a standard silicon photonics foundry process) to
an LNOI chip (LNOI from NanoLN Jinan Jingzheng Electronics Co. Ltd.) which has a thin layer of
LN crystal, an SiO2 buffer layer and a LN or Si handle [25]. A typical cross-section of a LNOI-SOI
bonded chip is shown in Fig. 4.4. The chips are initially brought into contact at room temperature,
and then the bond is strengthened by a thermal anneal. Compared to other approaches, such as
those based on etched or deposited thin films [33, 34, 35], the advantage of this approach is that
both the Si and LN thin film layers are crystalline and of high quality, which may benefit device
performance and enable opto-electronic functionality.
However, if this stack is directly bonded and the LNOI has a LN handle, it can only be held
together at room-temperature by weak hydrogen bonds. We have found that the stack debonds
when the temperature is raised to about 70-80◦ C. Researchers who have studied direct bonding
of bulk Si and LN under optimum conditions have found that the two materials either debond or

58
 a b c

Upper Handle: 500
m R

LN or Si
LN 500
m F1 1, E1 t1 F1
2
m h1
LN Film 750 nm F2 2, E2 t2 F2
100 nm 100 nm
150 nm y
z x h2
3
m
575
m F7 7, E7 t7 F
h7 7
Si 650
m y=0
L0

Figure 4.4: a The thin-film stack used in hybrid photonic devices based on silicon-on-insulator
(SOI) and lithium niobate-on-insulator (LNOI). b A simplified bulk LN-BCB-Si bonded structure
used in the calculations shown in Fig. 4.5. c A schematic drawing of the bonded stack, labeling
the dimensions and axial forces used in equations (4.1)-(4.4).

crack by 150◦ C [16, 36]. It may be hypothesized that debonding of Si and LN, two materials with
substantially different properties (see Table 1), can be mitigated by using a layer of amorphous,
low-optical-loss adhesive, such as benzocyclobutene (BCB, cyclotene 3022-35 resin from The
DOW Chemical Company). BCB has a high bond strength and thermal stability, but has a high
thermal resistance and a low refractive index that is substantially different from that of either Si
or LN. Calculations show that in order to maintain a hybrid optical mode similar to that of the
directly-bonded case, the BCB layer must be no more than 100 nm thicker than the SOI patterned
features; however, thin layers of BCB may not be sufficient to mitigate the built-up stresses. In
studies of bonded III-V-to-SOI materials, BCB layer thicknesses of 50 nm have been used [15].
After thorough cleaning of a SOI chip and a LNOI chip, a layer of BCB was spun onto the
Si chip; some of the chips used a BCB thickness of 300 nm (the summation of the height of the SOI
features and the calculated 100 nm separation between SOI and LN), whereas other chips used a
thickness of 1 µm. The chips were partially cured at 180◦ C for 45 minutes before being manually
aligned, bonded and hard-baked for 14 hours at 190◦ C. By placing a weight over the chips, a force
of 9.8 N was applied over a bonded area of 1 cm × 1 cm during the hard-bake process.
Contrary to intuition, the thickness of the BCB layer did not play a role in managing thermal
stress. We observed that LNOI with a LN handle cracked or debonded by the end of the hard

59
bake, regardless of BCB thickness. However, if the LNOI stack was based on a Si handle, the
bonded films did not crack or separate. This experimental observation is discussed using a simple
computational model and supported by further test experiments on test (bulk Si) samples.
We use a strain model which accounts for self-induced thermal strain (since the materials
will experience strain as temperature changes), stack-induced thermal strain (i.e., thermal strain
due to a mismatch in CTEs of materials held together in the stack), and bending-induced thermal
strain (a difference in CTEs causes a bonded stack to bend, resulting in an additional expansion or
contraction of the layers in the stack) based on the seminal theory outlined in [37].

Table 4.1: Material properties of all materials used in our numerical calculations of thermally-
induced stress. All data is taken at room temperature. E is the modulus of elasticity, and α is
the coefficient of thermal expansion (CTE). x, y, and z are used to denote the crystal axes of
anisotropic LN.

Material E (GPa) α (10−6 -K−1 )

Si 130[38] 2.6[39]
LN, z 199[40, 41] 3.4[42]
LN, y 173[40, 41] 13.4[42]
LN, x 173[40, 41] 13.4[42]
SiO2 75[43] 1.0[43]
BCB 2.9[44] 52[45]

A schematic of the structure is shown in Fig. 4.4c, and consists of seven layers, each with
a different axial force, and a common bending radius R. We require eight equations to solve for
the total strain in each of the seven layers. The width of each cross-section is not shown, but it
is assumed that they are all 1 cm, the nominal width of the test pieces used in this study. αi is
the CTE for the i-th layer, Ei is the modulus of elasticity for the i-th layer, and ti is the thickness
of the i-th layer, where i = 1, 2, . . . , 7. The CTEs and moduli of elasticity at room temperature
are provided in Table 7.1. The moduli of elasticity in anisotropic LN were recalculated from the
stiffness tensor of [40]. Calculating a modulus of elasticity from a stiffness tensor is discussed in
[41]. We assume all materials operate in the elastic regime, governed by Hooke’s law: σ = Eε, and

60
that there is no slip between bonded layers (i.e., two adjacent layers will stretch equally at their
shared boundary). Additionally, we assume that the radius of curvature R is much larger than the
sum of the thicknesses of all layers, and that the length L0 of the stack is much smaller than R, so
that the small-angle approximation [tan(θ) ≈ θ] is valid. Under these assumptions, the equation for
the strain in the i-th layer is:

Fi ti
εi = αi ∆T + + . (4.1)
Ei Ai 2R

We solve the following equations for the eight variables Fi and R,

i=7
∑ Fi = 0, (4.2)
i=1

i=7
1 i=7
∑ hiFi = R ∑ EiIi, (4.3)
i=1 i=1

Fj tj F j+1 t j+1
α j ∆T + + = α j+1 ∆T + − , (4.4)
E j A j 2R E j+1 A j+1 2R

(4.2) is the sum of thermally-induced axial forces, (4.3) is the sum of moments caused by
those forces, and (4.4) comes from the equality of strain at each boundary between layers, so that
j = 1,...,6 (because there are six boundaries between the seven layers). We assume there are no
external forces. Ii is the area moment of inertia of each layer (i = 1,...,7): Ii = wt3i /12, where w is the
width of the stack. ∆T = Tf - T0 , where Tf is the final temperature and T0 is the initial temperature
at which bonding occurs, and Ai = ti wi is the area of the cross-section of the i-th layer. hi is the
distance from y = 0 to the point of application of Fi . This is depicted in Fig. 4.4c.
We numerically solved for this set of equations for the seven-layer structure shown in
Fig. 4.4a for both a LN upper handle and a Si upper handle. Our calculations for the stress are
shown in Fig. 4.5a, where stress is calculated from Hooke’s law. Since LN is anisotropic and has
different material properties along different crystal axes, we ran the simulation for z-axis LN as
well as for y-axis LN (x-axis stress will be the same as y-axis stress in LN, see Table 7.1), assuming

61
the LN film had the same orientation as the LN handle. As we expected, z-axis stresses were much
lower than y-axis stresses for each LNOI handle material, simply because the CTE along the z-axis
of LN is much closer to the CTE of Si than is the CTE along the y-axis of LN (from Table 7.1).

a Stress in Upper Handle b Stress in SOI-LNOI Layers

1400 250
LN y-axis, Full Stack Si Handle
1200 LN z-axis, Full Stack LN Film y-axis
200
1000 Si Handle, Full Stack
Stress (MPa)

LN Film z-axis

Stress (MPa)
LN y-axis, LN-BCB-Si 150
800
600 100
LN z-axis, LN-BCB-Si
400
50
200
0 0
100 200 300 400 500 100 200 300 400 500
Temperature (oC) Temperature (oC)

Figure 4.5: a Calculated maximum stresses in the upper-handle layer (LN or Si) of the complete
thin-film stack and the simplified LN-BCB-Si stack shown in Fig. 4.4. Stress along each axis of
the LN handle is nearly identical for the full stack and for the simplified stack, suggesting that the
details of the thin films in the full stack have little bearing on the thermal stress in the upper handle
layer. b Thermal stress in the LN film of SOI-LNOI with a Si upper handle. Si upper handle stress
is also provided. z-axis LN stress is slightly higher than y-axis LN stress because the modulus of
elasticity is higher for the z-axis than the y-axis of LN. The strain in the LN film is almost identical
along both axes.

The numerical calculations suggest that the principal limitation in SOI-LNOI bonded stability
is not the thickness of the BCB layer between the two materials as we had originally hypothesized,
but, instead, whether or not the SOI handle (Si) and LNOI handle (LN or Si) are similar. To
confirm this, we also studied a simpler stack, consisting of (bulk) Si and LN pieces, shown in
Fig. 4.4b. Calculations shown in Fig. 4.5a nearly overlap the results of SOI-LNOI bonding along
the respective LN axes, suggesting that the thermal stresses are indeed similar to the SOI-LNOI
case. Experimentally, we attempted to bond a bulk piece of Si to a bulk piece of LN through a BCB
layer (Fig. 4.4b), and found that the bulk LN cracked at a relatively low temperature, regardless of
the BCB thickness.
A material stack of particular interest is that of Fig. 4.6. This stack consists of a bulk piece
of Si coated with a relatively thin (1 µm) layer of deposited SiO2 , bonded to a bulk piece of LN that

62
has been implanted to a desired depth with a certain species of ions (often either H+ or He+ ). This
material stack is significant because it is the precursor to the LNOI with a Si handle that we use in
this work. To go from Fig. 4.6 to a piece of LNOI, one must heat the material stack of Fig. 4.6 to at
least 200◦ C, at which point the bulk LN splits apart at the implanted layer, leaving a thin film of LN
bonded to the Si/SiO2 substrate [46, 34]. Yet as we established above, bonding bulk Si to bulk LN
and raising the temperature over 150◦ C results in thermally-damaged LN.

LN 500
m

~1
m

SiO2 1-3
m

575
m

Figure 4.6: Depiction of LNOI with a Si handle, after implantation and before splitting off the
bulk LN.

Here, an explanation as to why implanted bulk LN bonded to bulk Si does not crack at high
temperatures is provided. The common implant species for crystal-ion slicing are H+ and He+ ions,
which are light elements that exist as gases at room temperature. When enough of these ions are
implanted into the bulk LN, they will behave as tiny pockets of gas, allowing the LN to expand and
slip at that layer as temperature is increased. With rising temperature, the ions will become more
energetic and eventually form a single gas bubble. At this point the LN bulk is separated from the
LN film by the aggregated ions, and the LN splits [47]. Because the LN can slip and deform slightly
at the implant layer, thermal stress does not build up as it would if the LN were not implanted, and
the materials are left intact. The resulting LNOI can then be used for hybrid device fabrication.
The required fabrication processes hybrid SOI-LNOI devices include patterning, etching,
and metalization, to name a few, which often raise (or cycle) the temperature of the stack. The
benchmark studies of Takagi et al. [36] show that bulk Si-LN bonded at room temperature (without
any intermediate layer) cracks around 150◦ C. Our results with BCB-assisted bonding show that this

63
 a b c
Thermal cracks Particles between
BCB and LN

Thermal cracks
LN Film
LNOI
Bulk LN (LN Handle)
BCB BCB BCB

d e f
Particle between
BCB and LN Wet etch
undercut LN 780 nm

BCB 1045 nm
LN Film LN Film
1
m
Si
BCB Si

Figure 4.7: a through e show camera images of 1 cm × 1 cm pieces of LN or LNOI which were
bonded to bulk Si pieces of a similar size through a 1 µm BCB layer. a is bulk LN, b is LNOI with
a LN handle, and c through e are LNOI with a Si handle, which has been removed by dry-etching
(note that the SiO2 layer is still on the LN film in c and d). c shows the chip after substrate removal.
d shows the same chip after 10 thermal cycles between room temperature and 300◦ C. e shows the
chip after the protective SiO2 and surrounding BCB have been removed by respective HF and
Piranha acid baths. A cross-section SEM image of e taken at an angle of 52◦ from the vertical axis
is presented in f, showing the ∼1 µm thick BCB layer bonding a 780 nm thick film of LN to a
bulk Si substrate.

temperature budget can be substantially exceeded. We bonded bulk Si chips to bulk LN [Fig. 4.7a],
LNOI with a LN handle [Fig. 4.7b], and LNOI with a Si handle [Fig. 4.7c, handle removed]. The
first two cracked in several places below 190◦ C, but the third did not crack during bonding or in the
tested temperature range, which went up to 300◦ C.
The chip was further processed by extended XeF2 dry etching to remove the Si handle
from the LNOI side, i.e., the topmost layer in Fig. 4.4a. This chip therefore has a thin film of LN
bonded via BCB to Si, and is accessible from the top side for further processing (e.g., poling or
lithography on the LN film). Unfortunately, in our university clean room, we were not able to
achieve a completely particle-free bonding process, and, despite scrupulous attempts at cleaning, a
couple of dust particles were trapped beneath the LN layer, as indicated in Fig. 4.7c-e; however,

64
these particles did not affect the results discussed here. We cycled the chip 10 times between room
temperature and 300◦ C without cracking, as shown by a visual comparison between Fig. 4.7c and
d. Because the substrate of the chip is Si, there is no undesired pyroelectric effect during thermal
cycling, as there would be with bulk LN (particulary Z-cut bulk LN).
It is conjectured that an even higher temperature budget may be achieved. Assuming that
cracks first occur, in this configuration, along the y-axis (which is a reasonable assumption, since
the thermal stress is much greater along the y-axis than the z-axis), our model estimates a maximum
stress of approximately 340 MPa from Fig. 4.5a. Stress in the LN film of a bonded SOI-LNOI stack
with an upper handle of Si is shown in Fig. 4.5b, and does not reach the failure point of 340 MPa
until well over 500◦ C. The highest temperature our stack can withstand is thus set by the BCB layer,
which has a glass transition temperature of approximately 360◦ C when fully cured. Experimentally,
we did not exceed 300◦ C to avoid the BCB glass transition temperature.
The SiO2 layer and surrounding BCB were then removed – Fig. 4.7 – for EDX (energy-
dispersive x-ray) spectrum analysis. Results of the EDX measurement are shown in Fig. 4.8. BCB
on bulk Si peaks primarily in the ranges of carbon and oxygen, while our BCB-bonded LN film on
bulk Si (Fig. 4.7c-e) peaks in the range of oxygen, with a smaller peak at niobium. The spectrum of
Si-backed LNOI (a manufacturer-provided sample with known properties [48]) is nearly identical
to our BCB-bonded sample. This indicates that the LN crystal film has not deteriorated after the
300◦ C thermal cycling or the preceding bonding and handle removal.
Transferring LN films from one handle to another, without damaging the LN crystal, is an
important accomplishment for the advancement of hybrid silicon-lithium niobate photonics, as
commercially available 3-inch wafers of LNOI can be transferred to any SOI-based photonic chip
to create hybrid photonic circuits. An alternative approach to developing hybrid waveguides is to
ion slice small pieces of LN and place the pieces on the patterned silicon substrate for bonding
[50]. The approach we have taken avoids having to handle delicate LN pieces without a supporting
handle, and allows for larger area, regularly-sized (e.g., rectangular) LN films to be bonded.
Using the knowledge gained from our research into thermal stress of a bonded stack, we

65
 300
Si-backed LNOI LN on BCB on Si

200

Counts
Nb Nb
100

0 O O

0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5

Energy (keV) Energy (keV)

Figure 4.8: Energy-dispersive x-ray (EDX) spectrum analysis. The left plot is of a Si-backed
LNOI sample from the manufacturer, and the right plot is of a LN film bonded through BCB to
bulk Si. These two samples exhibit similar spectra, indicating a similarity in material properties.
Oxygen and niobium spectral indicators are well known and have been documented [49].

were able to transfer a LN film to a patterned SOI chip over a significant area – in this case
4.2 mm × 16 mm – to create a large hybrid region covering a couple dozen devices shown in
Fig. 4.9a. After the film transfer and bonding, we deposited 1 µm of PECVD SiO2 at 300◦ C to
improve optical coupling on- and off-chip through tapered silicon waveguides. We investigated one
of the devices in the bonded hybrid region of the chip – a 50/50 Y-splitter in silicon with waveguides
that taper down to 320 nm from 650 nm, propagate for 5 mm, then taper up to 650 nm before
recombining in a 50/50 Y-recombiner – which is shown in Fig. 4.9c. The mode transitions from the
bonded region to the unbonded region (the transition into the bonded region is shown in Fig. 4.9a
but not in (c)) and is followed by an S-bend comprised of two 25 µm quarter-circle bends to offset
the output from the input; this prevents light from coupling directly from the input fiber into the
output fiber without passing through the waveguide.
After subtracting fiber losses and fiber-to-chip coupling loss, we measured the insertion loss
through this device to be 10.2 dB over a total chip length of 26 mm, comparable to the measured
losses of [25]. This includes two transitions from a non-bonded silicon waveguide to a bonded
Si-LN waveguide, two 50/50 Y-junctions (shown with IR imaging in Fig. 4.9d-e, taken with a
Xenics XEVA-566 infrared camera through a 10x Mitutoyo microscope objective), and a 5 mm
hybrid propagation section, between the Y-junctions, where the silicon strip narrows down from a

66
 Y-Splitter Narrow Si Y-Recombiner
a c Input Mode: Panel f Mode: Panel g Mode: Panel f Transition Output

Marker Film 25
m Bends

Edge
4.2 mm
300 m 320 m 300 m
Bonded
LN Film (d) Y-Splitter (e) Y-Recombiner
16 mm

300 m 300 m

b Edge of film f g
SiO2 SiO2
LN LN
BCB BCB
SiO2 Si 1 m SiO2 Si 1 m

Figure 4.9: a The patterned SOI sample, bonded to a 4.2 mm × 16 mm film of LN through a
thin BCB layer. b A zoomed in image of a corner of the bonded region. c The specific device we
measure, magnified from a with sections of unchanging straight waveguide removed for visual
convenience. The transition out of the bonded region is shown before the output, and white arrows
indicate the direction of optical propagation through the waveguide. d and e are top-down infrared
images of 1550 nm light passing through the Y-splitter and Y-recombiner, respectively. These
pictures were taken with a Xenics XEVA-566 infrared camera through a 10x Mitutoyo microscope
objective. f and g show the simulated electric fields of the modes in the 650 nm and 320 nm wide
silicon strips.

width of 650 nm to 320 nm. Simulations of the modes in the 650 nm and 320 nm wide silicon strips
are shown in Fig. 4.9f and g, respectively. While these losses are higher than those seen in all-LN
thin ridge structures [51], we are far from full optimization of our process.
This work shows that whereas certain types of bonded Si-LN stacks cannot be heated to
elevated temperatures without damage, 1 µm thick BCB-assisted bonding of thin films of LN on
a Si handle, which matches the Si handle of the SOI chips, can withstand elevated temperatures
of at least 300◦ C over multiple heat-and-cool cycles, allowing for further fabrication processes
to be performed. We have shown that we can transfer a pre-fabricated LN film from a simple
LNOI platform to a patterned SOI chip, with little to no damage to the LN film, over a sizable
area, and deposit PECVD SiO2 at 300◦ C without disturbing the LN film. In particular, we observe
optical transmission through a hybrid Si-LN waveguide that goes through a 50/50 Y-splitter and

67
then a Y-recombiner, an optical structure that can be used to make electro-optic and thermo-optic
chip-scale devices. This is promising for the continued development of hybrid Si-LN technologies
with thin crystalline device layers.

4.6 Handle Removal

In order to fabricate a full hybrid Si-LN electro-optic modulator, further processing is

required after bonding and annealing. As alluded to in Section 4.5, the LN film’s handle must be
removed in order to fabricate electrodes close enough to the LN film for any observable electro-optic
effect to occur. The general process flow is shown in Fig. 4.10.

a Wafer as received from b Bond LNOI c Remove LN Handle

Sandia Labs LN Substrate
SiO2
Si WGs
<100 nm LN Film (< 1
m) LN Film (< 1
m)

SiO2 SiO2 SiO2

Si Substrate Si Substrate Si Substrate

Figure 4.10: a Wafer cross-section as received from Sandia National Laboratories, as depicted
in Fig. 4.2f. b Bond the LN-on-insulator (LNOI) die (from a NanoLN wafer) to an SOI die,
singulated from the Sandia wafer. c Remove the handle (SiO2 and substrate) from the LN film
after bonding.

Wafer handle removal has become more commonplace over the last decade or so, thanks in
large part to the the microelectromechanical systems (MEMS) community. Due to this recent interest
in wafer handle removal and wafer-scale thin film transfer, some manufacturers have begun to offer
temporary bonding agents specifically designed to transfer thin films from one handle to another. In
this work, an adhesive called BrewerBond 220 (BB220) was purchased from BrewerScience, Inc.
BB220 has the consistency of liquid wax, and can be spin coated onto a sample before curing and
temporary bonding. Also, like wax bonding, BB220 can be cooled to room temperature to harden
after an initial cure, and then softened again at higher temperatures to slide debond the transferred

68
film from its original substrate.
However, the most important aspect of BB220 in this work is its ability to withstand
hydrofluoric (HF) wet etching and XeF2 dry etching. To remove the LN handle, which was chosen
to be Si for its reduced thermal mismatch and easy removal (Section 4.5) a LNOI die was bonded
to an SOI die. Next, the sample was coated with BB220 everywhere except the Si backside of the
LNOI, which was removed with a deep XeF2 etch in a Xactic isotropic etcher. This was followed
by an HF etch to remove the 2 µm SiO2 layer between the LN and the Si handle. Finally, the BB220
was removed in a toluene-based solvent. At this stage in the fabrication process, the sample was
ready for electrode fabrication.

4.7 Metallization

The metallization process was performed after handle removal as the final step in the
fabrication process of the hybrid Si-LN EOM. First, a seed layer of 5 nm Cr and 60 nm Au was
deposited in an e-beam evaporator. After seed layer deposition, a thick 10 µm positive photoresist
was spin coated onto the sample and lithographically etched to expose metal where the electrical
lines were needed, as well as an open rectangle at the edge of the sample.
To form the electrical lines, the exposed sample was electroplated in a homemade non-
cyanide gold electroplating bath. The sample was clipped into place on the exposed rectangle at the
edge of the chip; potential current density concerns were mitigated by the flat Au seed layer across
the entire sample below the photoresist. Despite the homemade electroplating setup, relatively
repeatable electroplating results were possible by carefully monitoring the electroplating bath
temperature, homogeneity, current, time, and electrode thickness during electroplating. Currents
between 3 mA and 10 mA were used during device fabrication.
Afterwards, the photoresist was removed in acetone and the Cr/Au seed layer components
were removed in their respective wet etchants.

69
4.8 Acknowledgments

Section 4.5 of Chapter 4, in full, is a reprint of the material as it appears in Optical

Materials 2017. Peter O. Weigel & Shayan Mookherjea, Optical Materials Vol. 66, 605-610 and in
Conference on Lasers and Electro-Optics (CLEO) 2017. Peter O. Weigel and Shayan Mookherjea,
paper. SM3K.1. The dissertation author, together with his adviser, led the research efforts for this
work and co-authored the papers.

Bibliography
[1] Michel Bruel. Process for the production of thin semiconductor material films, December 20
1994. US Patent 5,374,564.

[2] M Levy, RM Osgood Jr, R Liu, LE Cross, GS Cargill III, A Kumar, and H Bakhru. Fabri-
cation of single-crystal lithium niobate films by crystal ion slicing. Applied Physics Letters,
73(16):2293–2295, 1998.

[3] AM Radojevic, M Levy, H Kwak, and RM Osgood Jr. Strong nonlinear optical response in
epitaxial liftoff single-crystal linbo 3 films. Applied physics letters, 75(19):2888–2890, 1999.

[4] Tarek A Ramadan, M Levy, and RM Osgood Jr. Electro-optic modulation in crystal-ion-sliced
z-cut linbo 3 thin films. Applied Physics Letters, 76(11):1407–1409, 2000.

[5] T Izuhara, I-L Gheorma, RM Osgood Jr, AN Roy, H Bakhru, Yiheli M Tesfu, and ME Reeves.
Single-crystal barium titanate thin films by ion slicing. Applied physics letters, 82(4):616–618,
2003.

[6] Lionel C Kimerling. Silicon microphotonics. Applied Surface Science, 159:8–13, 2000.

[7] Qianfan Xu, Bradley Schmidt, Sameer Pradhan, and Michal Lipson. Micrometre-scale silicon
electro-optic modulator. nature, 435(7040):325, 2005.

[8] Ryan Aguinaldo, Alex Forencich, Christopher DeRose, Anthony Lentine, Douglas C Trotter,
Yeshaiahu Fainman, George Porter, George Papen, and Shayan Mookherjea. Wideband silicon-
photonic thermo-optic switch in a wavelength-division multiplexed ring network. Optics
express, 22(7):8205–8218, 2014.

[9] JDB Bradley, PE Jessop, and AP Knights. Silicon waveguide-integrated optical power monitor
with enhanced sensitivity at 1550 nm. Applied Physics Letters, 86(24):241103, 2005.

[10] M Lawrence. Lithium niobate integrated optics. Reports on Progress in Physics, 56(3):363,
1993.

70
[11] ScienceDaily. Now entering, lithium niobate valley: Reserachers demon-
strate high-quality optical microstructures using lithium niobate. Online.
www.sciencedaily.com/releases/2017/12/171221133650.htm.

[12] Thomas Plach, K Hingerl, S Tollabimazraehno, G Hesser, V Dragoi, and M Wimplinger. Mech-
anisms for room temperature direct wafer bonding. Journal of Applied Physics, 113(9):094905,
2013.

[13] Marin Alexe and Ulrich Gösele. Wafer bonding: applications and technology. Springer
Science & Business Media, 2004.

[14] Frank Niklaus, Göran Stemme, J-Q Lu, and RJ Gutmann. Adhesive wafer bonding. Journal
of applied physics, 99(3):2, 2006.

[15] Shahram Keyvaninia, Muhammad Muneeb, Stevan Stanković, PJ Van Veldhoven, Dries
Van Thourhout, and Günther Roelkens. Ultra-thin dvs-bcb adhesive bonding of iii-v wafers,
dies and multiple dies to a patterned silicon-on-insulator substrate. Optical Materials Express,
3(1):35–46, 2013.

[16] Hideki Takagi, Ryutaro Maeda, Naoe Hosoda, and Tadatomo Suga. Room-temperature
bonding of lithium niobate and silicon wafers by argon-beam surface activation. Applied
physics letters, 74(16):2387–2389, 1999.

[17] Stephan Krapick, Benjamin Brecht, Harald Herrmann, Viktor Quiring, and Christine Silber-
horn. On-chip generation of photon-triplet states. Optics express, 24(3):2836–2849, 2016.

[18] F Thomas, S Heidmann, M De Mengin, N Courjal, G Ulliac, A Morand, P Benech, E Le Coarer,

and G Martin. First results in near and mid ir lithium niobate-based integrated optics interferom-
eter based on swifts-lippmann concept. Journal of Lightwave Technology, 32(22):3736–3742,
2014.

[19] Ian W Frank, Jeremy Moore, James K Douglas, Ryan Camacho, and Matt Eichenfield.
Entangled photon generation in lithium niobate microdisk resonators through spontaneous
parametric down conversion. In Lasers and Electro-Optics (CLEO), 2016 Conference on,
pages 1–2. IEEE, 2016.

[20] Abu Thomas, Mackenzie A VanCamp, Andrew Fraine, and Alexander V Sergienko. Low-noise
quantum frequency conversion in titanium-diffused lithium niobate waveguide. In Frontiers in
Optics, pages FTh1D–4. Optical Society of America, 2015.

[21] MMR Howlader, T Suga, and MJ Kim. Room temperature bonding of silicon and lithium
niobate. Applied physics letters, 89(3):031914, 2006.

[22] Yoo Seung Lee, Gun-Duk Kim, Woo-Ju Kim, Sang-Shin Lee, Wan-Gyu Lee, and William H
Steier. Hybrid si-linbo 3 microring electro-optically tunable resonators for active photonic
devices. Optics letters, 36(7):1119–1121, 2011.

71
[23] Idan Bakish, Ran Califa, Tali Ilovitsh, Vlada Artel, Georg Winzer, Karsten Voigt, Lars
Zimmermann, Eyal Shekel, Chaim Sukenik, and Avi Zadok. Voltage-induced phase shift in
a hybrid linbo3-on-silicon mach-zehnder interferometer. In Integrated Photonics Research,
Silicon and Nanophotonics, pages IW4A–2. Optical Society of America, 2013.

[24] Li Chen, Qiang Xu, Michael G Wood, and Ronald M Reano. Hybrid silicon and lithium
niobate electro-optical ring modulator. Optica, 1(2):112–118, 2014.

[25] Peter O Weigel, Marc Savanier, Christopher T DeRose, Andrew T Pomerene, Andrew L
Starbuck, Anthony L Lentine, Vincent Stenger, and Shayan Mookherjea. Lightwave circuits in
lithium niobate through hybrid waveguides with silicon photonics. Scientific reports, 6:22301,
2016.

[26] W Qiu, M-P Bernal, A Ndao, C Guyot, NM Hameed, N Courjal, H Maillotte, and FI Baida.
Analysis of ultra-compact waveguide modes in thin film lithium niobate. Applied Physics B,
118(2):261–267, 2015.

[27] Martin F Volk, Sergiy Suntsov, Christian E Rüter, and Detlef Kip. Low loss ridge waveguides
in lithium niobate thin films by optical grade diamond blade dicing. Optics express, 24(2):1386–
1391, 2016.

[28] Günther Roelkens, Dries Van Thourhout, and Roel Baets. Ultra-thin benzocyclobutene bonding
of iii-v dies onto soi substrate. Electronics Letters, 41(9):561–562, 2005.

[29] Alexander W Fang, Hyundai Park, Oded Cohen, Richard Jones, Mario J Paniccia, and
John E Bowers. Electrically pumped hybrid algainas-silicon evanescent laser. Optics express,
14(20):9203–9210, 2006.

[30] J Almerico, S Ross, P Werbaneth, J Yang, and P Garrou. Plasma etching of thick bcb polymer
films for flip chip bonding of hybrid compound semiconductor-silicon devices. Tegal Crop,
2001.

[31] Cheng-Ta Ko and Kuan-Neng Chen. Wafer-level bonding/stacking technology for 3d integra-
tion. Microelectronics reliability, 50(4):481–488, 2010.

[32] JJ McMahon, E Chan, SH Lee, RJ Gutmann, and J-Q Lu. Bonding interfaces in wafer-level
metal/adhesive bonded 3d integration. In Electronic Components and Technology Conference,
2008. ECTC 2008. 58th, pages 871–878. IEEE, 2008.

[33] Jong-Gul Yoon and Kun Kim. Growth of highly textured linbo3 thin film on si with mgo
buffer layer through the sol-gel process. Applied physics letters, 68(18):2523–2525, 1996.

[34] Payam Rabiei, Jichi Ma, Saeed Khan, Jeff Chiles, and Sasan Fathpour. Heterogeneous lithium
niobate photonics on silicon substrates. Optics express, 21(21):25573–25581, 2013.

[35] Shuang Li, Lutong Cai, Yiwen Wang, Yunpeng Jiang, and Hui Hu. Waveguides consisting
of single-crystal lithium niobate thin film and oxidized titanium stripe. Optics express,
23(19):24212–24219, 2015.

72
[36] Hideki Takagi, Ryutaro Maeda, and Tadatomo Suga. Room-temperature wafer bonding of si
to linbo3, litao3 and gd3ga5o12 by ar-beam surface activation. Journal of Micromechanics
and Microengineering, 11(4):348, 2001.

[37] Stephen Timoshenko. Analysis of bi-metal thermostats. JOSA, 11(3):233–255, 1925.

[38] Matthew A Hopcroft, William D Nix, and Thomas W Kenny. What is the young’s modulus of
silicon? Journal of microelectromechanical systems, 19(2):229–238, 2010.

[39] CA Swenson. Recommended values for the thermal expansivity of silicon from 0 to 1000 k.
Journal of physical and chemical reference data, 12(2):179–182, 1983.

[40] RS Weis and TK Gaylord. Lithium niobate: summary of physical properties and crystal
structure. Applied Physics A, 37(4):191–203, 1985.

[41] Allan F Bower. Applied mechanics of solids. CRC press, 2009.

[42] F Pignatiello, M De Rosa, P Ferraro, S Grilli, P De Natale, A Arie, and S De Nicola. Measure-
ment of the thermal expansion coefficients of ferroelectric crystals by a moiré interferometer.
Optics Communications, 277(1):14–18, 2007.

[43] Jie-Hua Zhao, Todd Ryan, Paul S Ho, Andrew J McKerrow, and Wei-Yan Shih. Measurement
of elastic modulus, poisson ratio, and coefficient of thermal expansion of on-wafer submicron
films. Journal of applied physics, 85(9):6421–6424, 1999.

[44] AJG Strandjord, RH Heistand, JN Bremmer, PE Garrou, and TG Tessier. A photosensitive-bcb

on laminate technology (mcm-ld). In Electronic Components and Technology Conference,
1994. Proceedings., 44th, pages 374–386. IEEE, 1994.

[45] Michael E Mills, Paul Townsend, Dan Castillo, Steve Martin, and Albert Achen. Benzo-
cyclobutene (dvs-bcb) polymer as an interlayer dielectric (ild) material. Microelectronic
Engineering, 33(1-4):327–334, 1997.

[46] Gorazd Poberaj, Hui Hu, Wolfgang Sohler, and Peter Guenter. Lithium niobate on insulator
(lnoi) for micro-photonic devices. Laser & photonics reviews, 6(4):488–503, 2012.

[47] B Aspar, H Moriceau, E Jalaguier, C Lagahe, A Soubie, B Biasse, AM Papon, A Claverie,

J Grisolia, G Benassayag, et al. The generic nature of the smart-cut R process for thin film
transfer. Journal of Electronic Materials, 30(7):834–840, 2001.

[48] Lutong Cai, Yun Kang, and Hui Hu. Electric-optical property of the proton exchanged phase
modulator in single-crystal lithium niobate thin film. Optics express, 24(5):4640–4647, 2016.

[49] Jill Chastain, Roger C King, and J Moulder. Handbook of X-ray photoelectron spectroscopy: a
reference book of standard spectra for identification and interpretation of XPS data. Physical
Electronics Division, Perkin-Elmer Corporation Eden Prairie, Minnesota, 1992.

[50] Li Chen, Jonathan Nagy, and Ronald M Reano. Patterned ion-sliced lithium niobate for hybrid
photonic integration on silicon. Optical Materials Express, 6(7):2460–2467, 2016.

73
[51] Arnaud Gerthoffer, Clément Guyot, Wentao Qiu, Abdoulaye Ndao, Maria-Pilar Bernal, and
Nadège Courjal. Strong reduction of propagation losses in linbo3 ridge waveguides. Optical
Materials, 38:37–41, 2014.

74
Chapter 5

All-Optical Hybrid Silicon-Lithium Niobate

Devices

5.1 Introduction

Lithium Niobate (LN) was once considered the most promising of materials for integrated
optics [1], but despite a rich set of properties, the technology of LN integrated optics has not evolved
as much as integrated optics in III-V semiconductors and silicon (Si) photonics [2, 3]. Although
high performance stand-alone LN devices have been shown [4, 5] and a technique of ion-sliced
thin-film LN has been developed [6, 7, 8], the technology of LN integrated optics has continued
to rely on traditional waveguide fabrication techniques based on ion-exchange [9], diffusion and
serial writing [10] or mechanical sawing [11, 12], all of which are very different from the modern
lithographic techniques and foundry processing available in Si or III-V photonics.
To address this problem, we develop an approach to developing hybrid waveguides in LN
based on a two-material core cross-section, consisting of few-hundred nanometer single-crystal
thin films of Si and LN bonded face-to-face. We used manufacturer-provided Si-on-insulator and
LN-on-insulator wafers which have a SiO2 cladding layer which is a few microns thick, and a
handle material (Si and LN, respectively) for the substrate which is several hundred microns thick.

75
We first pattern all features required for waveguiding in the Si wafer using deep ultraviolet (DUV)
lithography (see Methods section). This allows much finer features to be patterned than is possible
in LN (which has previously constrained LN waveguides to be multi-moded [13]). Since LN is not a
CMOS-compatible material and is not processed in CMOS foundries, we developed designs which
allow all the waveguiding circuitry to be defined in Si, requiring only a single back-end process step
of bonding LN to form the functional chip. DUV lithographic fabrication of Si photonic features is
more precise, scalable and manufacturable than the traditional waveguide-fabrication techniques in
LN, and the LN layer can be left as an unpatterned slab that is then bonded to the patterned Si [14].
By not patterning, etching or sawing LN, its material properties are kept pristine [15].
Fig. 5.1 shows images of a bonded chip whose dimensions are that of a typical field size of
a stepper projection DUV lithography system (few centimeters squared). Upon bonding, a hybrid
waveguide is formed, in which the geometric dimensions of the Si features and LN film dictate
how the optical mode is distributed between the Si and LN layers. Techniques have been developed
in the past decade for bonding LN of various orientations to Si wafers without incurring thermal
stress mismatches [16, 17, 18, 19, 20, 21]. Bonding achieves much higher quality and better optical
properties than growth of LN on Si using sol-gel processes [22] or chemical vapor deposition [23].
We directly bonded the LN chip to the Si chip i.e., without any intervening polymeric ‘glue’ layer as
used in some approaches [13, 24], thus permitting the maximum control over the mode distribution.
As will be discussed below, light propagates mainly in the silicon layer in certain sections
of the layout, and primarily in the LN layer where desired, and makes transitions between the two
layers at several locations while remaining in the transverse-electric (TE) polarized single-mode
regime, which is highly desirable for Si photonics. Thus, we can also design devices that are
outside the Si-LN bonded region and behave entirely like conventional Si photonics. In this way,
the technology for Si photonics can be leveraged to enhance LN integrated optics, and vice versa.
The main advantages of this approach are: (a) the requirements on LN fabrication are
reduced to a minimum, i.e., a single bonding step to the manufacturer-supplied wafer after all silicon
patterning has been completed (etching LN is possible [25, 26] but is technologically less mature

76
a b Waveguide Bonding
Bonding surface for LN
0.32 - 0.65 μm
“shoulder”

Si 25 μm 0.15 μm

c Si-LN Si d 0.15 mm
Bonding "shoulder"

After singulation & bonding:

0.35 mm
d
c
1.7 cm

Si y
2.1 cm z
LN Waveguides

Figure 5.1: Hybrid LN-Si photonic circuits. a Silicon photonic components were fabricated
using deep ultraviolet (DUV) lithography. Singulated dies (size: 25 mm × 16 mm) were bonded
to diced pieces (size: 21 mm × 17 mm) of an unpatterned Lithium Niobate (LN)-on-insulator
wafer. This particular LN chip was labeled ‘b13’ by scratch marks on the LN substrate (0.5 mm
thick, opposite to the bonded surface). b Schematic of the cross-section showing how waveguides
and bonding “shoulders,” which are at the same height as the waveguides, are conveniently
formed in one lithographic etch step on the Si wafer. c Optical microscope image showing
waveguides transitioning between the portion of the hybrid chip which is not covered by LN (i.e.,
conventional SiO2 -clad Si photonic waveguides), and that which is bonded to LN (and uses hybrid
waveguides). Arrows colored red indicate the back-and-forth direction of light propagation in
certain representative sections. d Optical components can be defined in the SiO2 -clad Si section,
or in the LN-Si bonded section.

77
than etching Si, which is available in every foundry process); and (b) both the cross-sectional mode
area and the length of components are greatly reduced, compared to traditional LN ion-exchanged
or diffused waveguides, allowing complex circuits to be realized in a small area. The tradeoffs are
that: (a) the bonding step should be performed at a low temperature since Si and LN have different
coefficients of thermal expansion, and (b) the resulting hybrid modes have a higher refractive
index dispersion compared to low-index contrast (e.g., doped-glass) integrated optics, and thus,
components will be optimized for a desired vacuum wavelength range, e.g., the telecommunications
band around 1550 nm.

5.2 Hybrid Modes and Mode-Transition Tapers

The dimensions of the Si features and LN film determine what fraction of the optical mode
resides in Si and in LN. Two distinct waveguide cross-sections, labeled ‘A’ and ‘B’, are selected,
as shown in Fig. 5.2a, and used in the microchip layout. These cross-sections vary only in the
width of the Si structures, which allows for convenient fabrication in a single step of lithography.
The numerical values of the widths and heights chosen here are suitable for wavelengths between
1530 nm and 1565 nm (i.e., the telecommunications C-band).
Cross-section A consists of 650 nm wide and 150 nm tall Si features, bonded to LN. The
lowest-order mode of the waveguide is quasi-TE-polarized and is similar to the quasi-TE0 mode
in conventional silicon photonics (see Fig. 5.2a). Wider waveguides are close to becoming multi-
moded, which is undesirable. A waveguide using cross-section A can transition from an SiO2 -clad
silicon photonic section to the bonded Si-LN region, as shown in Fig. 5.1c, with less than -0.3 dB
loss calculated by numerical simulation (Lumerical software package). We have not experimentally
confirmed the mode-transition loss at this time. As shown by Fig. 5.2b, waveguides using cross-
section A can support compact bends with radii of 3-5 µm without incurring bending losses higher
than that of a standard rib waveguide (650 nm × 220 nm surrounded by SiO2 , labeled ’Si/SiO2 ’ in
Fig. 5.2b) with a radius of 0.5-1 µm.

78
 a A B
w [nm] 760 650 540 430 320 210

LN
Si
SiO2 neff 2.462 2.401 2.306 2.157 2.034 2.015
Aeff [ m2] 0.22 0.22 0.23 0.32 1.25 3.76
LN [%] 30.1 31.9 36.2 50.9 87.0 93.8

b -1
10
Bend Loss (dB/90°)

A ss
-2
o
10
Si/SiO2 .L
op
Pr B
B/cm
3 d
-3
10

-4
10
0.5 1 5 10 20 50 100 300 700
Bend Radius (µm)
Figure 5.2: Waveguide modes. a Calculated hybrid optical mode profiles for different Si rib
widths. The panels show the magnitude of the electric field in the TE polarization, with the E-field
vector oriented along the crystal axis. As the Si rib width w decreases, the modal effective index
(neff ) decreases, the effective area (Aeff ) increases, and the fraction of light in LN (ΓLN ) also
increases. The dotted boxes indicate the two cross-sectional designs (‘A’ and ‘B’) used in the
chip. b The calculated bending loss of the ‘A’ cross-sectional mode is much lower than that of
the ‘B’ mode, and is not too different from the waveguides used in silicon photonics (labeled
as ‘Si/SiO2 ’). Therefore, ‘A’ is used for bends and compact routing, and ‘B’ is used in straight
waveguide sections when most of the light should “see” LN. The dashed line labeled ’3 dB/cm
Prop. Loss’ represents the propagation loss at all bend radii for a propagation loss of 3 dB/cm to
give an idea of practical minimum losses for bent waveguides.

79
 Cross-section B consists of 320 nm wide Si features at the same height as cross-section
A. With an air upper cladding, this waveguide is unable to support a waveguide mode by itself,
and thus requires the bonded LN layer to support a well-defined mode at C-band wavelengths.
Cross-section B cannot be used for compact low-loss bends (see Fig. 5.2b). Instead, it is used
when we want the light to interact with the crystal properties of LN; otherwise, cross-section A
is used to route light on the hybrid chip. Therefore, it is desirable to maximize the fraction of
light in LN for the mode in cross-section B. The confinement factor is defined as follows: ΓLN =

∗ 0 Re [ (E × H∗ · ŷ) dA], where ŷ is the direction of light propagation and
RR RR
Re [ LN (E × H · ŷ) dA ]
the range of integration in the integral in the numerator is restricted to the LN region. In waveguides
using cross-section A, ΓLN ≈ 32% whereas in waveguides using cross-section B, ΓLN ≈87%. The
transition between the two cross-sections is discussed below.
A modified version of cross-section A, in which SiO2 replaces LN as the upper cladding, is
used outside the LN bonded region to demonstrate all-Si components made on the same platform
and at the same device level. Another cross-section with a reduced Si width of 180 nm is used
for couplers at the edge of the Si chip, as is commonly used in Si photonics [27], and can also be
fabricated in the same step.
All these waveguide shapes are simple rectangles without relying on slots, tilted sidewalls or
other difficult-to-fabricate shapes. There is no patterning in the LN layer which eliminates many of
the challenges faced in the past [13]. Conceptually, the waveguide structure is similar to that of the
strip-loaded waveguide film studied in the 1970’s [28], but scaled to the sub-micron regime. The
modal area of cross-section A (0.22 µm2 ) is, in fact, smaller, and that of cross-section B (1.25 µm2 )
is only slightly larger, than those of the smallest-area waveguides reported in LN, fabricated by the
ion-milling technique [29].
As shown in Fig. 5.1d, large rectangular areas were defined in the Si device layer at distances
of about 30 µm from the waveguide edge. These “bonding pads” between the Si and LN chips
are far from the waveguide core and serve no optical purpose, but instead provide a large surface
area for bonding. They are precisely at the same height as the Si rib features, since they are

80
formed in the same lithographic step. They are similar to “dummy fill” features inserted to assist in
chemical-mechanical polishing, but without a fragmented pattern. The etched trenches between
the Si ribs and the bonding pads not only provide optical confinement but may assist in bonding by
defining convenient outgassing channels, and for local stress relaxation.

a Section Section c
Adiabatic 25
A taper B
20

Loss (dB/cm)
X Y
650 nm 320 nm 15
Z
10
150
m
5 TE0
B TM0
b 0
2.6 500 750 1000 1250 1500 1750
TE0 LN thickness (nm)
A
Effective index

2.5 d
2.4 TE1 2
TE0
2.3 Loss (dB/cm) 1.5
TM0
2.2 1

2.1 B TM1 0.5

B A
2 0
200 400 600 800 1000 1200 200 300 400 500 600 700
Si width (nm)
Si width (nm)
Figure 5.3: Adiabatic mode transition. a Gradual linear reduction of the Si rib width to transition
between cross-sections A and B. b Numerical calculations of the modal effective index versus
waveguide width, with yellow circles indicating the initial and final points of the taper. The
TE-polarized mode does not hybridize with, or convert into, the TM-polarized mode in the taper. c
Plot of simulated propagation losses for cross-section B with a varying LN film thickness. (Inset:
Ex component of the fundamental TE0 mode, (left) bound for cross-section B, i.e., 750 nm LN;
and (right) leaky for 1000 nm thick LN.) The losses decrease as the mode approaches a TE slab
mode. d Plot of simulated propagation losses for our hybrid geometry (150 nm tall Si, 750 nm
thick LN) with a varying Si width. Similar to c, losses increase suddenly when the x-component
of the E-field is no longer confined, and then decrease as the mode approaches a TE slab mode.

Along the direction of light propagation, adiabatic tapered transitions were defined in the
Si layer (see Fig. 5.3a) as a linear change in the Si rib width over a distance of 150 µm. The
quasi-TE-polarized mode TE0 has the highest effective refractive index in cross-section A (and is
used extensively in conventional Si photonic waveguides, which are wider than they are tall [30]).

81
In cross-section B, it is actually the quasi-TM-polarized mode TM0 which has the higher refractive
index. This is due to both the anisotropic properties of LN (LN has a lower material index along the
z-axis) and the low index lateral air cladding on either side of the Si. However, the taper-induced
coupling between these modes is zero to first order because the electric field of the former is a
symmetric mode, and that of the latter is an anti-symmetric mode [31]. Thus, we are able to draw
the waveguide down to a narrow width that pushes a very large fraction of the light into LN while
maintaining its state of polarization all the way from the feeder waveguides. For the Si rib widths
used in this design, the higher-order quasi-TE polarized mode (labeled TE1 in Fig. 5.3b) is not
guided; however, by using wider Si rib widths, tapers can couple between this anti-symmetric mode
and the TM0 mode, which may be useful for polarization rotators etc. Thus, compared to other
hybrid Si-LN structures [26, 20, 32], our design achieves the highest TE-polarized (lowest-order)
mode-fraction in LN while also providing a pathway for integration to Si photonics by first coupling
light into cross-section A (which is similar to a traditional Si photonics cross-section), and then
transitioning into cross-section B adiabatically. Attempting to couple from an external input to
the quasi-TE mode in cross-section B would render the device very susceptible to roughness and
bending losses, because it is not the fundamental mode. Widening the Si waveguide effectively
pulls the light back into Si again, and, because it remains in the TE-polarization, allows for the
benefits that have been realized by Si-only waveguides, such as tight bends and compact directional
couplers, among others.
Another important consideration is that the propagation loss of a hybrid mode can increase
significantly due to lack of transverse confinement, if the LN thickness falls within a range of values,
depending upon the Si rib dimensions, at which one of the vectorial field components becomes
slab-like (in the LN region). This effect, studied and measured in standard rib waveguides [33], has
recently been highlighted for quasi-TE modes in x-cut LN proton exchanged channel waveguides
[34]. Similarly, in our configuration, it originates from the TE ↔ TM mode coupling at the Si strip
boundaries. Quasi-TE mode leakage then occurs when the modal effective index becomes smaller
than that of the TM slab mode in LN, i.e., when the minor TM-polarized field is no longer a bound

82
mode and propagates in the LN planar waveguide. Due to the LN birefringence, this condition
cannot be met for the quasi-TM modes which is therefore immune to leakage. This is shown in
Fig. 5.3c: for LN thicknesses between 850 nm and 1750 nm (not used in our present design), the
propagation loss would be significantly higher than at other thicknesses. A similar behavior is
observed when the width of the Si region is reduced and the LN thickness is fixed, as in Fig. 5.3d.

a Si/SiO2 b Cross-Section B
0 Measured Data 0 Measured Data
Linear Fit Linear Fit
-5 -5
Transmission (dB)

Transmission (dB)
-10 -10

-15 -15

-20 -20

-25 -25

1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5
Length (cm) Length (cm)

Figure 5.4: Waveguide characterization. a Transmission measurements of test structures con-

sisting of 650 nm wide Si waveguides with SiO2 cladding (labeled ‘Si/SiO2 ’). b Transmission
versus length of waveguides and bends (paper-clip structures) at the wavelength of 1.55 µm for
mode cross-section ‘B’ in the straight sections and cross-section ‘A’ in the semi-circular bends
(labeled ‘Cross-section B’).In both a and b, the numerical fit is shown by the solid lines. The
y-axes are normalized to be 0 dB for 0 cm length waveguides.

5.3 Waveguides, Directional Couplers and Photonic Circuits

Photonic circuits can be assembled from the basic building blocks of waveguides and
directional couplers. Hybrid waveguides of lengths between 1.8 cm and 4.6 cm were fabricated
with the majority of the length consisting of waveguides using cross-section B. Cross-section A was
used at the semi-circular bends, in order to fit the longer waveguides into a compact footprint using
“paperclip” structures. The longest waveguide involves 10 transitions between cross-sections A and

83
B. Since the edge facets were roughly diced and not polished, the fiber-to-chip insertion loss was
high (about 9.5 dB without index-matching liquid) and non-uniform. As shown in Fig. 5.4b, by
measuring transmission versus length across several chips, we were able to build up an ensemble
of measurements from which a propagation loss of 4.3 ± 2.1 dB/cm was extracted at 1550 nm
wavelength with only minor changes across the wavelength range 1530 nm to 1570 nm. At this
time, we are unable to separate the bending loss from the straight waveguide propagation loss, but
the former is expected to be very small based on the large bending radius of 25 µm. Numerical
simulations reported in Fig. 5.2b suggest that the bending loss should drop to less than 0.001 dB per
90◦ bend for bending radius greater than 3 µm using cross-section A.
By way of comparison, the propagation loss of the same Si waveguides with 3 µm SiO2
top-cladding (rather than LN) was measured to be 3.1 ± 2.1 dB/cm in Fig. 5.4a, indicating that
bonding and incorporation of LN as the top cladding did not significantly worsen the propagation
characteristics. In fact, the propagation loss is similar to that measured in rib Si photonic waveguides
[30] despite the thinner Si layer in our structures.
The loss values in these hybrid Si-LN waveguids using cross-section B are significantly
lower than other reported values, e.g., 16 dB/cm in etched thin-film LN waveguides [13] and 6 –10
dB/cm in thin-film LN waveguides (660 nm thickness, not too different from the 750 nm thickness
used here) with oxidized titanium stripe [35]. Although the loss is still an order-of-magnitude higher
than that of traditional in-diffused waveguides which have a much larger mode area, our circuits are
also more than an order-of-magnitude more compact. Furthermore, we expect that with improved
Si waveguide fabrication (e.g., roughness reduction), and filling of the air pockets on the lateral
sides of the Si rib with a low-optical-loss gap-fill dielectric material, the overall propagation loss
will decrease down to around 1 dB/cm.
Directional couplers were defined by lithography in the Si layer, but act on the hybrid mode
using cross-section A, with a gap of 0.4 µm between the waveguide edges. The typical length was
only 150 µm, compared to a typical length of about 5 mm for directional couplers with diffused
or ion-exchanged waveguides. Because the mode in cross-section A resides primarily in Si, there

84
was no measurable crosstalk at milliwatt power levels, unlike the traditional waveguide LN devices
which are susceptible to photorefractive artifacts [36, 37]. However, long-term (> 1 day) tests have
not yet been performed since the chips are not fully packaged and we rely on manually-adjusted
fiber coupling to the silicon waveguides.
Optical circuits can be designed if two fundamental parameters are known: the optical
(modal, i.e., “effective”) refractive index of a waveguide, which determines the rate of accumulation
of optical phase with distance, and the directional coupling coefficient of two adjacent waveguides.
Both these parameters are functions of the optical wavelength. To measure these parameters and
compare with numerical calculations, we designed an interferometric test structure (see Fig. 5.5)
which can also perform several functions useful in integrated optics such as spectral filtering,
interleaving or multiplexing / de-multiplexing for wavelength-division multiplexing (WDM). Such
a device does not use microring or standing-wave (e.g., Fabry-Perot) resonators, in which the
intensity enhancement (due to the infinite impulse response nature of the transfer function) may
cause photorefractive artifacts in LN. The circuit consists of several building blocks where the
optical pathway is indicated by the red arrows and consists of four transitions, two directional
couplers and twenty 90-degree bends, incurring a cumulative insertion loss of about -15 dB. (The
circuit does not need to be so complicated to realize only a Mach-Zehnder interferometer, but
showcases some of the capabilities of the platform.)

5.4 Discussion and Analysis

To experimentally show the large difference in the effective modal refractive index between
the A and B cross-sections, we measured two distinct MZIs in which the path imbalance of one arm
with respect to the other was formed using waveguides of cross-section A and B, of length LA and
LB , respectively. In each case, the free-spectral range (FSR) of the MZI was approximately 10 nm
near the central wavelength of 1550 nm.
Figure 5.5b shows the wavelength variation of the effective refractive index of the A and

85
a C.S. ‘A’ Transition C.S. ‘B’

out1 bends
in1
D.C. D.C.
out2 in2

bonding shoulders

b 2.6 c 1
A e 127 μm
Effective index

2.4 0.8
s
0.6
2.2 |κ|2
s
0.4 MZI #1
2.0 B MZI #2
e
0.2 Simulation
1.8 0
1540 1580 1620 1540 1580 1620
Wavelength (nm) Wavelength (nm)

Figure 5.5: Hybrid Si-LN optical circuit. a Optical microscope image of an interferometeric
hybrid Si-LN circuit which uses both waveguide cross-sections (C.S.) A and B, four adiabatic
transitions (Transition) in each lightpath, two directional couplers (D.C.), and twenty 90-degree
bends in each lightpath. The image is shown using a simulated dark-field colour map for clarity,
and with added shading to highlight the different sections. b Extracted modal effective index versus
wavelength for MZIs with path-length difference (PLD) regions comprised of hybrid waveguides
with cross-sections A (dashed lines) and B (solid lines). Simulated values using an eigenmode
solver software (A: dotted, and B: solid lines) are also shown, assuming a nominal (design) Si
rib waveguide width, and manufacturer-specified LN film thickness. Shaded regions indicate the
standard errors provided by the fitting routine. c Extracted coupling coefficient of the directional
coupler versus wavelength for the MZIs reported here (dashed and solid lines). MZIs with PLD
cross-sections A and B use the same type of directional coupler, hence the results are similar. The
simulation results shown by the squares were calculated using the supermode equations and a
numerical simulation of the eigenmodes. Shaded regions indicate the standard errors provided by
the fitting routine.

86
B waveguides. The two branches (cross-section A and cross-section B) are sufficiently far apart
that the results confirm a clear distinction between the A (Si-like) and B (LN-like) guided modes.
The group velocity dispersion (GVD) coefficients of the two modes at λ = 1550 nm are as follows:
DA ≈ -3100 ps/nm-km for cross-section A, and DB ≈ -4300 ps/nm-km for cross-section B, which
are of the same order-of-magnitude as that of a typical silicon photonic waveguide whose modal
effective area is similar to that of cross-section A, DSi ≈ -1500 ps/nm-km [38].
Figure 5.5c shows the wavelength variation of the coupling coefficient of a directional
coupler which was formed using waveguides of cross-section A, with a gap of 0.4 µm between the
waveguide edges. As may be expected, the change in the magnitude of the coupling coefficient over
the wavelength range of 1520 nm–1620 nm is similar to the behavior seen in all-Si waveguides [39].
On the same chip, outside the LN-bonded region, waveguides and devices may be designed
as usual in Si photonics, i.e., with SiO2 cladding. Because of the high refractive index of Si,
light is tightly confined for waveguides of cross-section ‘A’ whether LN or SiO2 is used as the
upper cladding. We designed and measured a Mach-Zehnder interferometer outside of the LN-
bonded region, using waveguide with cross-section ‘A’ and with SiO2 replacing LN as the upper
cladding (and with SiO2 side- and lower-claddings in our fabrication process), as reported in Fig. 5.6.
Although our test chip shown in Fig. 5.1 was designed for LN covering nearly all of the Si surface, it
is equally possible to design “mixed” chips with smaller-sized LN pieces, which combine traditional
Si photonic components with hybrid LN-Si photonic components on a monolithic platform.
We have restricted this fabrication process for reasons of cost and complexity to exclude
dopants and electrodes as part of the Si chip, which could be used for electro-optic effects. Simula-
tions show that gold electrodes can be positioned directly on the LN thin-film layer (surface opposite
to the bonded interface to Si) at a lateral distance of only 0.4 µm from the Si edge for cross-section
A and 4 µm from the Si edge for cross-section B, for an estimated additional propagation loss of
0.1 dB/cm. Alternatively, electrodes can be fabricated on the LN layer after substrate removal,
following the traditional approach [13].
These results show that, even though LN is not a CMOS-compatible material, foundry-

87
 Figure 5.6: Si/SiO2 Mach-Zehnder interferometer (MZI). a An SiO2 -clad Si photonic MZI
was fabricated on the same chip but outside the bonded LN region. The input-output waveguides
pass under the LN bonded region and emerge on the other facet of the chip. b The measured ‘Thru’
(red) and ‘Cross’ (black) characteristics, and their sum (blue) follow the expected behavior for
an MZI, and demonstrate that conventional Si photonic devices can be fabricated alongside the
hybrid Si-LN ones (the y-axis (dB) is in arbitrary units).

fabrication technologies can play a very useful role in a new generation of LN integrated optics.
While LN has always been a desirable material for its nonlinear and electro-optic properties, it
has not been possible in the past to make compact and complex waveguide circuits as is possible
nowadays in Si photonics using precise and highly-repeatable DUV lithography. We have shown a
suite of hybrid building blocks from which optical circuits can be assembled alongside traditional
Si photonics components. While improved wafer-scale bonding techniques are being developed by
industry for commercial applications, here we have shown chip-scale direct bonding of chips that
are a few centimeters squared (the size of a typical field size of a DUV stepper lithography system),
with enough bond strength to permit dicing and simple packaging for test and measurement. A
similar approach may also be applied to design and investigate optical circuits using other thin-film
materials in place of LN, leveraging the advanced foundry fabrication capabilities of Si photonics
as a waveguiding template for the hybrid modes without needing to pattern the thin-films.

88
5.5 Acknowledgments

Chapter 5, in part, is a reprint of the material as it appears in Scientific Reports 2016. Peter
O. Weigel, Marc Savanier, Christopher T. DeRose, Andrew T. Pomerene, Andrew L. Starbuck,
Anthony L. Lentine, Vincent Stenger & Shayan Mookherjea, Scientific Reports 6, 22301. The
dissertation author, together with his adviser, led the research efforts for this work and co-authored
the paper.

Bibliography
[1] Lawrence M. Lithium niobate integrated optics. Reports on Progress in Physics.
1993;56(3):363.

[2] Smit M, Van der Tol J, Hill M. Moore’s law in photonics. Laser & Photonics Reviews.
2012;6(1):1–13.

[3] Hochberg M, Baehr-Jones T. Towards fabless silicon photonics. Nature Photonics.

2010;4(8):492.

[4] Sohler W, Hu H, Ricken R, Quiring V, Vannahme C, Herrmann H, et al. Integrated optical

devices in lithium niobate. Optics and Photonics News. 2008;19(1):24–31.

[5] Chiba A, Sakamoto T, Kawanishi T, Higuma K, Sudo M, Ichikawa J. 16-level quadrature am-
plitude modulation by monolithic quad-parallel Mach-Zehnder optical modulator. Electronics
letters. 2010;46(3):227–228.

[6] Levy M, Osgood Jr R, Liu R, Cross L, Cargill III G, Kumar A, et al. Fabrication of single-
crystal lithium niobate films by crystal ion slicing. Applied Physics Letters. 1998;73(16):2293–
2295.

[7] Rabiei P, Gunter P. Optical and electro-optical properties of submicrometer lithium niobate
slab waveguides prepared by crystal ion slicing and wafer bonding. Applied physics letters.
2004;85(20):4603–4605.

[8] Poberaj G, Koechlin M, Sulser F, Günter P. High-density integrated optics in ion-sliced lithium
niobate thin films. In: Integrated Optics: Devices, Materials, and Technologies XIV. vol. 7604.
International Society for Optics and Photonics; 2010. p. 76040U.

[9] Tamir T, Tomlinson W. Guided-wave optoelectronics. Applied Optics. 1989;28:2262.

[10] Thomas J, Heinrich M, Zeil P, Hilbert V, Rademaker K, Riedel R, et al. Laser direct writing:
Enabling monolithic and hybrid integrated solutions on the lithium niobate platform. physica
status solidi (a). 2011;208(2):276–283.

89
[11] Ulliac G, Guichardaz B, Rauch JY, Queste S, Benchabane S, Courjal N. Ultra-smooth
LiNbO3 micro and nano structures for photonic applications. Microelectronic Engineering.
2011;88(8):2417–2419.

[12] Takigawa R, Higurashi E, Kawanishi T, Asano T. Lithium niobate ridged waveguides with
smooth vertical sidewalls fabricated by an ultra-precision cutting method. Optics Express.
2014;22(22):27733–27738.

[13] Guarino A, Poberaj G, Rezzonico D, Degl’Innocenti R, Günter P. Electro–optically tunable

microring resonators in lithium niobate. Nature photonics. 2007;1(7):407.

[14] Mookherjea S, Savanier M. Heterogeneous waveguides and methods of manufacture. Google

Patents; 2018. US Patent App. 15/550,493.

[15] Han H, Cai L, Hu H. Optical and structural properties of single-crystal lithium niobate thin
film. Optical Materials. 2015;42:47–51.

[16] Takagi H, Maeda R, Hosoda N, Suga T. Room-temperature bonding of lithium niobate and
silicon wafers by argon-beam surface activation. Applied physics letters. 1999;74(16):2387–
2389.

[17] Howlader M, Suga T, Kim M. Room temperature bonding of silicon and lithium niobate.
Applied physics letters. 2006;89(3):031914.

[18] Takigawa R, Higurashi E, Suga T, Shinada S, Kawanishi T. Low-temperature bonding of a

LiNbO 3 waveguide chip to a Si substrate in ambient air for hybrid-integrated optical devices.
In: Optomechatronic Micro/Nano Devices and Components II. vol. 6376. International Society
for Optics and Photonics; 2006. p. 637603.

[19] Lee YS, Kim GD, Kim WJ, Lee SS, Lee WG, Steier WH. Hybrid Si-LiNbO 3 microring electro-
optically tunable resonators for active photonic devices. Optics letters. 2011;36(7):1119–1121.

[20] Chen L, Xu Q, Wood MG, Reano RM. Hybrid silicon and lithium niobate electro-optical ring
modulator. Optica. 2014;1(2):112–118.

[21] Chiles J, Fathpour S. Mid-infrared integrated waveguide modulators based on silicon-on-

lithium-niobate photonics. Optica. 2014;1(5):350–355.

[22] Yoon JG, Kim K. Growth of highly textured LiNbO3 thin film on Si with MgO buffer layer
through the sol-gel process. Applied physics letters. 1996;68(18):2523–2525.

[23] Sakashita Y, Segawa H. Preparation and characterization of LiNbO3 thin films produced by
chemical-vapor deposition. Journal of applied physics. 1995;77(11):5995–5999.

[24] Chen L, Reano RM. Compact electric field sensors based on indirect bonding of lithium
niobate to silicon microrings. Optics Express. 2012;20(4):4032–4038.

[25] Hui H, Ricken R, Sohler W. Etching of lithium niobate: from ridge waveguides to photonic
crystal structures. ECIO, Eindhoven. 2008;.

90
[26] Rabiei P, Steier WH. Lithium niobate ridge waveguides and modulators fabricated using smart
guide. Applied Physics Letters. 2005;86(16):161115.

[27] Almeida VR, Panepucci RR, Lipson M. Nanotaper for compact mode conversion. Optics
letters. 2003;28(15):1302–1304.

[28] Uchida N. Optical waveguide loaded with high refractive-index strip film. Applied optics.
1976;15(1):179–182.

[29] Hu H, Ricken R, Sohler W. Lithium niobate photonic wires. Optics express.

2009;17(26):24261–24268.

[30] Vlasov YA, McNab SJ. Losses in single-mode silicon-on-insulator strip waveguides and bends.
Optics express. 2004;12(8):1622–1631.

[31] Bures J. Guided optics. John Wiley & Sons; 2009.

[32] Rao A, Patil A, Chiles J, Malinowski M, Novak S, Richardson K, et al. Heterogeneous

microring and mach-zehnder lithium niobate electro-optical modulators on silicon. In: CLEO:
Science and Innovations. Optical Society of America; 2015. p. STu2F–4.

[33] Ogusu K, Tanaka I. Optical strip waveguide: an experiment. Applied optics. 1980;19(19):3322–
3325.

[34] Cai L, Han SLH, Hu H. Waveguides in single-crystal lithium niobate thin film by proton
exchange. Optics Express. 2015;23(2):1240–1248.

[35] Li S, Cai L, Wang Y, Jiang Y, Hu H. Waveguides consisting of single-crystal lithium niobate

thin film and oxidized titanium stripe. Optics express. 2015;23(19):24212–24219.

[36] Schmidt R, Cross P, Glass A. Optically induced crosstalk in LiNbO3 waveguide switches.
Journal of Applied Physics. 1980;51(1):90–93.

[37] Mueller CT, Garmire E. Photorefractive effect in LiNbO 3 directional couplers. Applied
optics. 1984;23(23):4348–4351.

[38] Turner Amy C, Manolatou C, Schmidt Bradley S, Lipson M, Foster Mark A, Sharping Jay
E, et al. Tailored anomalous group-velocity dispersion in silicon channel waveguides. Opt
Express. 2006;14:4357–4362.

[39] Aguinaldo R, Shen Y, Mookherjea S. Large dispersion of silicon directional couplers obtained
via wideband microring parametric characterization. IEEE Photonics Technology Letters.
2012;24(14):1242–1244.

91
Chapter 6

Characterization of Hybrid Electro-Optic

Modulators

6.1 Introduction

Characterizing electro-optic modulators consists of three components: optical, electrical,

and electro-optical characterization. Proper characterization requires an understanding of both
optical integrated circuits and microwave electronics, and at high frequencies equipment limitations
become a nontrivial concern. For this reason, extracting high frequency electro-optic information –
namely, the device’s electro-optic response – is an interesting experimental challenge.

6.2 Optical Characterization of an Electro-Optic Modulator

Optically, the full electro-optic modulator consists of a Mach-Zehnder interferometer with a

path-length difference in one arm, and several Si waveguide width changes as shown in Fig. 6.1a.
Broadband directional couplers are used for 50/50 splitters. Simulated effective mode indices for
different low order waveguide modes in the hybrid Si-LN region are shown in Fig. 6.1b, with modes
‘B’ and ‘C’ of Fig. 6.1a shown in Fig. 6.1c. ΓSi is the optical confinement factor in Si, ΓLN is the
optical confinement factor in LN, and ne f f is the mode’s effective (i.e. phase) index.

92
a Mode 'A' Mode 'C'
b 2.5
Taper LN Film TE0

Effective Index
2.4
650 nm Mode 'B'
2.3 TE1
Mo Out
2.2
In
X Y
2.1 TE2
3-dB DC TM0
Hybrid Region Mode 'C'
2.0
SOI Chip PLD Z
200 400 600 800 1000 1200 1400
Si Width (nm)
c Mode 'A' Mode 'B' Mode 'C'
Air Air
Si = 64% LN Si = 58% LN Si = 5.0%
SiO2 neff = 2.28 LN = 12% LN = 81%
neff = 2.33 neff = 1.97
SiO2 SiO2
Si 1
m Si 1
m 1
m
Si

Figure 6.1: a Schematic of the EOM (not to scale, not showing electrodes), including two 3-dB
directional couplers (DC) and a waveguide segment for path-length difference (PLD). Three optical
waveguide modes are used, labeled as A, B, and C. Modes A (Si under SiO2) and B (Si under LN)
have Si rib width w = 650 nm whereas mode C has w = 320 nm. b Dispersion curves (effective
index versus w) in the hybrid region; w values for modes B and C are chosen to stay within the
single-mode region of operation. An adiabatic waveguide transition (variation in w) is designed
to evolve from mode B to C and vice versa. c Calculated Poynting vector components along the
direction of propagation. Modes A and B are Si-guided and have a similar confinement fraction
in Si. Mode C, with LN confinement factor (ΓLN ) greater than 80%, is used in the phase-shifter
segments.

93
6.2.1 Optical Transmission through an Electro-Optic Modulator

A fabricated electro-optic modulator has the optical transmission as shown in Fig. 6.2a, with
an interferometer pattern caused by the path-length difference in one arm of the EOM. The spacing
between nulls, known as the free spectral range (FSR, ∆λFSR ) varies from 5.4 nm around 1525 nm
to 5.8 nm around 1575 nm. From theory [1],

λ2
∆λFSR ≈ (6.1)
ng LFSR

where λ is the wavelength, ng is the optical group index in the path-length difference region, and
LFSR is the path-length difference between the two arms, since it is this path-length difference that
is causing a wavelength-dependent phase difference. For LFSR = 96.26 µm, ng is between 4.47 and
4.44. These sorts of relatively high group indices are common for high-index Si waveguides.

a b PM L/T Fiber
0
PM
Power (dB)

-5 Laser Fiber
-10 EOM
SM Fiber Chip Detector
-15
Paddles
-20
-25 PM PM L/T Fiber
1530 1540 1550 1560 1570
Fiber
Wavelength (nm)

Figure 6.2: a Optical transmission through EOM device with greater than 15 dB extinction ratio
across the entire measured wavelength range of 1525 nm to 1575 nm. b Pictorial representation
of optical setup. SM = single mode, PM = polarization maintaining, and L/T = lensed/tapered.
Paddles are used to precisely control polarization going onto the chip.

The experimental setup for this measurement is described in Fig. 6.2b. For optical trans-
mission, the measurement is relatively straightforward; a laser source with good linear polarization
may not even require additional polarization control. For this work, commercially available optical
fiber “paddles” (hollow, adjustable ceramic cylinders designed to hold tightly wound spools of
fiber) are used for additional polarization control to ensure that laser light entering the chip is
horizontally polarized more details on attaining a high degree of horizontal polarization is described
in Section 6.2.4. Lensed/tapered polarization maintaining fibers from Oz Optics, Ltd. were used to

94
couple on and off the chip using waveguide edge couplers. Lensed tapered fibers have a mode field
diameter of 2.5 µm, small enough to efficiently match the input/output waveguide mode. All fiber
after the paddles is PM to ensure the highly linear polarization state is maintained. From the clean
spectrum of Fig. 6.2a – which has no ripples that would normally be caused by polarization effects –
it is clear that the polarization of the input light is well controlled and horizontally polarized.

6.2.2 Propagation Loss

All waveguides have some of propagation loss. “Propagation loss” refers to a per unit length
loss term describing how much of the optical signal is attenuated simply by traveling through the
waveguide. Waveguide loss is typically described in dB per centimeter. Optical fibers, for example,
have propagation losses of less than 0.5 dB/km, or 5e-6 dB/cm.1 A good Si waveguide, on the other
hand, has a propagation loss around 1 dB/cm.

a Mode A b Mode C
-3 -1
Measured Data Measured Data
-4 Linear Fit Linear Fit
Transmission (dB)

Transmission (dB)

-5
-2
-6

-7
-3
-8
= 1550 nm
= 1550 nm
-9  = 1.34 dB/cm  = 0.64 dB/cm
-10 -4
2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 2 2.5 3 3.5 4 4.5 5
Length (cm) Length (cm)

Figure 6.3: Transmission for various waveguide lengths at λ = 1550 nm is shown for the all-Si
mode (a, Mode A) and the Si-LN mode (b, Mode C). Propagation loss is significantly lower for
the hybrid, narrow-Si mode, at only 0.64 dB/cm compared with 1.34 dB/cm for the all-Si mode.
The y-axes are normalized to be 0 dB for 0 cm length waveguides.

For this Si-LN device, two sets of propagation loss measurements were performed to describe
the propagation loss in the non-hybrid region (mode ‘A’ in Fig. 6.1c) and in the hybrid, narrow Si
region (mode ‘C’ in Fig. 6.1c). The results at λ = 1550 nm are shown in Fig. 6.3. Propagation losses
in Mode C are only 0.64 dB/cm compared with 1.34 dB/cm in Mode A. These results may seem
1 The extremely low loss of optical fibers has been a significant motivator of research into optics for the last 35 years.

95
surprising at first, but with some thought they become clear. LN is a highly pure crystalline insulator
with intrinsically low propagation loss (as early as the late 1980s, LN waveguide propagation losses
were shown to be as low as 0.15 dB/cm [2]), and Mode C is designed with the intention of pushing
as much of the optical mode into the LN film as possible while still maintaining a small mode area.
By reducing the Si rib width and increasing ΓLN , less light scatters off impurities in the lossier Si
and overall propagation loss is significantly reduced.
Still, propagation losses in both cases can be reduced by improving the fabrication process.
In particular, Si sidewall roughness can be further reduced with additional oxidation steps. SiO2
is deposited via plasma-enhanced chemical vapor deposition (PECVD) in a university cleanroom
above Mode A only after bonding the LN. Performing this deposition in a industry grade dedicated
SiO2 PECVD chamber would likely reduce optical losses as well. Lastly, it is unclear whether the
bonding process somehow reduces the optical quality of the LN film (perhaps simply by handling
the sample) or if the bond interface is a source of optical loss caused by, perhaps, strain-induced
crystal defects. It is also possible that the propagation loss of Mode C can be optimized by adjusting
the Si rib height and width and the LN film thickness (though this would of course impact the
performance of the electro-optic device).

6.2.3 Spline Bending Loss

In addition to straight waveguides for propagation, spline bends are used for bent or angled
optical routing. As a low-loss waveguide bend, spline functions are mathematically superior to
quarter circle bends [3]. The most significant source of loss in a quarter-circle waveguide bend is
at the point where the straight waveguide and the quarter-circle bend meet, due to the asymmetric
mode supported by a circularly bent waveguide. Spline bends, however, have zero curvature (infinite
bending radius) at their inputs and outputs, elegantly avoiding the in/out coupling issue without
adding too much additional length to the bend. Based on the results of Fig. 6.4a, 30 µm spline bends
are used for all electro-optic modulators (in the broadband directional couplers and path-length
difference sections), though 15 µm bends or even 10 µm bends would have been acceptable as well.

96
 a b
Transmission (dBm) -5
Lc = 5 μm
-15 Lc = 15 μm
Lc = 10 μm

2 mm
Lc = 15 μm
-25
Lc = 10 μm
-35
-45 Lc = 5 μm
0 50 100 150 200 250
No. of Bends

Figure 6.4: a Measured (circles) spline bending loss for various characteristic lengths (Lc ) at
λ = 1550 nm. Lines represent linear fits of transmission measurements for a number of devices with
varying amounts of quarter-circle spline bends. For Lc = 5 µm, 10 µm, and 15 µm, αb = 0.18 dB/90◦ ,
0.0023 dB/90◦ , and 0.0015 dB/90◦ , respectively. b GDS layout of array of spline bends of various
characteristic length and number of bends.

In Fig. 6.4, each spline is defined by its characteristic length Lc . As described in Chapter 3,
Lc is analogous to the bend radius of a quarter-circle bend. The full length of a 15 µm quarter-circle
bend is 23.6 µm, whereas the full length of a 15 µm spline bend is 27.0 µm. Similarly, a 30 µm
quarter-circle bend has a length of 47.1 µm versus 54.1 µm for a 30 µm characteristic length spline
bend. So although the spline bend has a marginally longer path length than a typical quarter-circle
bend, this extra length is vastly outweighed by the optical propagation benefits. GDS layout of
the various spline bend test structures are included in Fig. 6.4b to show how that despite the large
number of devices required for these tests, they can be densely packed without crosstalk due to the
high confinement of the Si waveguide, even though the Si waveguide is only 150 nm thick.

6.2.4 Method to Optimize Optical Polarization State

Due to the importance of optical polarization control for high-quality optical transmission
measurements, it is worth taking the time to understand how to minimize undesired polarization
components.
Typical lasers are designed to output a single, linear polarization state, either horizontal
(TE) or vertical (TM). Polarization maintaining (PM) fibers, as the name implies, maintain this
polarization state along their length by breaking the symmetry of a single mode (SM) circular fiber
core to prevent coupling between TE and TM fundamental modes. Imagine an ideal laser with

97
100% of its output light polarized horizontally. In such a scenario, PM fiber can connect directly
from the laser to the next component in the experiment, without any need for polarization control
and stability. However, since realistic lasers can sometimes have a noticeable fraction of light in an
orthogonal polarization state (or even a circular or elliptical polarization state, if the laser is poorly
designed) polarization control after the laser becomes exceedingly useful.
To minimize undesired polarization components, fiber paddles are used after the laser. As
described previously, fiber paddles are simply spools of SM fiber in three adjacent, adjustable
ceramic cylinders. Each paddle can have a different number of fiber spools in it. By spooling fiber
into the paddles, some polarization rotation occurs in the stressed fiber. By then rotating the paddles,
torsional stress on the fiber adds an additional polarization rotation. In this way, the three paddles
act as tunable knobs to control the polarization state.
The simplest way of minimizing undesired polarization components is to pass the optical
signal through a set of paddles, through a polarization rotator, and into a photodetector. The
polarization rotator should be set to pass light linearly polarized orthogonal to the desired linear
polarization state; if horizontally polarized light is desired, the polarizer should be set to pass
vertically polarized light, and vice versa. Then, the desired polarization state can be obtained by
minimizing light at the undesired polarization angle. This is much more accurate than maximizing
the desired polarization state, since low power measurements are more sensitive to fine adjustments.
When done correctly, a polarization extinction ratio of 30 to 1 or greater can be achieved.

6.3 Electrical Characterization of an Electro-Optic Modulator

Electrical characterization of electro-optic modulators can be extracted from scattering

matrix measurements through a vector network analyzer (VNA). The scattering matrix of an
electrical network measure incident and reflected voltages through the various ports of the network.
In the case of a two-port network (one input and one output), the scattering matrix is defined as

98
     
−
V1 S11 S12  V1+ 
 =  ; (6.2)
    
V2− S21 S22 V2+

S11 , S12 , S21 , and S22 are called the S-parameters of the network. Physically, Sii can be thought of
as the reflection at port i and Si j as the transmission from port j through port i.
Another useful matrix for two-port analysis is the ABCD matrix:

    
V1  A B  V2 
 =  . (6.3)
    
I1 C D I2

Block diagrams of (6.2) and (6.3) are shown in Fig. 6.5a-b, respectively. Note that the direction of
current flow at port 2 has been redefined for the ABCD matrix, following standard convention [4].
Conversions between S-parameters and ABCD parameters are well known and are as follows (from
[4]):

(1 + S11 )(1 − S22 ) + S12 S21

A= (6.4)
2S21
(1 + S11 )(1 + S22 ) − S12 S21
B = Z0 (6.5)
2S21
1 (1 − S11 )(1 − S22 ) − S12 S21
C= (6.6)
Z0 2S21
(1 − S11 )(1 + S22 ) + S12 S21
D= . (6.7)
2S21

Z0 is the characteristic impedance of the measurement system, and is typically 50 Ω.

99
 aV +
V 2-
b c
I+
I R
1 I1 I2
I1 L I I2
+ + + +
Electrical Electrical + G +
V1 V2 V1 V+
V C V V2
Network _ Network _ _ _
_ _

V 1- V 2+
x
d

Figure 6.5: a Scattering matrix representation of a two-port electrical network. b ABCD matrix
representation of a two-port electrical network. c Circuit schematic for a long, lossy transmission
line of length d. R, L, C, and G are the per unit length series resistance, series inductance,
shunt capacitance, and shunt impedance of the transmission line. ∆x is a short length of the
total transmission line. V1 , V+∆V, V, and V2 are various voltages along the length of the line
(V1 = input voltage, V2 = output voltage), with corresponding currents shown above each voltage.

6.3.1 Lossy Transmission Line Theory

In the case of a long, lossy transmission line (Fig. 6.5c), the ABCD parameters can be related
to the transmission line’s characteristic impedance and propagation constant. Performing loop
analysis (KVL) on Fig. 6.5c results in

∆V = Iz∆x, (6.8)

which can be rearranged and taken to the limit as ∆x goes to zero to get

dV
= Iz. (6.9)
dx

A similar node analysis (KCL) results in an expression for the differential current in the line:

dI
= V y. (6.10)
dx

Z ≡ zd = (R+ jωL)d is the series impedance of the line, and Y ≡ yd = (G+ jωC)d is the
shunt impedance of the line. Variables z and y are the per unit length series and shunt impedances
of the line, respectively. (6.9) and (6.10) can be combined and rearranged into

100
 d 2V
−V zy = 0, (6.11)
dx2
√
whose roots are located at ± yz:

√ √
V = A1 ex yz
+ A2 e−x yz
. (6.12)

z and y can be related to the characteristic impedance and propagation constant of the transmission
line from the circuit components of Fig. 6.5c [4]. The characteristic impedance is

s r
R + jωL z
Zc = = , (6.13)
G + jωC y

and the complex propagation constant is

p √
γ= (R + jωL)(G + jωC) = zy. (6.14)

2πkm 2πnm
γ = α + jβ, where α = λm and β = λm ; km is the imaginary component of the effective index of
the electrical signal, nm is the real part of the effective index of the electrical signal, and λm is the
wavelength of the electrical signal. The variables y and z can be replaced with Equation (6.13) and
Equation (6.14) in Equation (6.12):

V = A1 eγx + A2 e−γx (6.15)

A1 γx A2 −γx
I= e − e . (6.16)
Zc Zc

V2 +Zc I2 V2 −Zc I2
If x = 0 then V = V2 , I = I2 , and A1 = 2 and A2 = 2 . If x = d, V = V1 and I = I1 .
With some algebra, the following expressions for V1 and I1 – the input voltage and current – are
obtained in terms of V2 and I2 :

101
 V2 + Zc I2 γd V2 − Zc I2 −γd
V1 = e + e (6.17)
2 2
V2 /Zc + I2 γd V2 /Zc − I2 −γd
I1 = e − e . (6.18)
2 2

Substituting the exponentials in Equation (6.17) and Equation (6.18) with hyperbolic functions and
again performing some simple algebra results in

V1 = V2 cosh(γd) + Zc I2 sinh(γd) (6.19)

sinh(γd)
I1 = V2 + I2 cosh(γd). (6.20)
Zc

Combining Equation (6.19) and Equation (6.20) with the definition of the ABCD matrix in
Equation (6.3) finally gives expressions for A, B, C, and D in terms of Zc and γ:

A = cosh(γd) (6.21)

B = Zc sinh(γd) (6.22)
1
C= sinh(γd) (6.23)
Zc
D = cosh(γd). (6.24)

The ABCD values for a transmission line can be experimentally determined by carefully
measuring the transmission line’s S-parameters and using Equations (6.4)-(6.7). Finally, from
Equation (6.21) and Equation (6.22):

102
 γ = acosh(A)/d (6.25)
p
Zc = B/ A2 − 1. (6.26)

6.3.2 Electrical Extraction from VNA Measurements

Using the theory developed in Section 6.3, it is relatively straightforward to electrically

characterize an electro-optic modulator.
Electrically, an EOM is simply a two-port device; in particular, it is a two-port lossy
transmission line that matches the modeling of Fig. 6.5c. This means that Equations (6.21)-(6.26)
can be used to determine the characteristic impedance, microwave effective index, and microwave
propagation loss of the coplanar waveguide transmission line used for the EOMs in this work.
Measurements were performed in the High Speed Circuits Lab at UC San Diego using
an Agilent E8361A vector network analyzer up to 67 GHz, high enough to extract microwave
parameters. The measured |S21 |2 and |S11 |2 curves are shown in Fig. 6.6a and the extracted
microwave parameters are shown in Fig. 6.6b.

a b
0 nm
(dB/cm) Zc (Ω)
8
-10 50
6
-20
dB

4 40
-30 |S21|2
-40 2
|S11|2 30

0.01 10 20 30 40 50 60 67 0.01 20 40 60 80 100

Frequency (GHz) Frequency (GHz)

Figure 6.6: a Measured electrical S-parameters of the EOM’s coplanar-waveguide transmission

line. b Left y-axis: extracted microwave phase index nm and microwave loss αm (units of dB/cm).
Right y-axis: characteristic impedance Zc (Ω).

The data of Fig. 6.6b are fits of the raw extracted data. There are several things to note
about the extracted data in Fig. 6.6b. First, the microwave index is nearly flat at high frequencies

103
(> 10 GHz or so). This is not by accident: since index matching is a critical bandwidth limitation –
particularly at higher frequencies – and the optical index does not change with microwave frequency,
it is imperative that nm be as flat as possible so that index matching can occur across a wide range
of frequencies. In Fig. 6.6, nm ≈ 2.08. Second, the microwave loss, which is in units of dB/cm,
has a distinct curvature to it that approximately follows an x1/2 shape. This is because microwave
losses at high frequencies tend to be dominated by conductor losses, i.e. the losses due to the
metal electrodes, which scale as x1/2 . With this in mind, a commonly used loss value known as
the conductor loss can be extracted by dividing α by the square root of frequency, in GHz, to
get αcond = 0.65 dB/(cm-GHz1/2 ). This value, along with nm and Zc , can be substituted into the
equation for the electro-optic bandwidth of a modulator (Equation 1.23) to obtain a theoretical
electro-optic response, as long as the optical group index is known as well.

6.3.3 Measuring Vπ L

The Vπ L term is the product of the halfwave voltage (the voltage required to switch the
optical output intensity from a maximum (“one”) to a minimum (“zero”), and the active device
length L. Experimentally, electrostatic (dc) data are obtained by applying a dc voltage to one end of
the CPW line and measuring the optical intensity at a particular wavelength of interest. By sweeping
the voltage and recording the optical intensity for each voltage point, a sin2 curve is created of
optical power versus applied voltage. The difference (in voltage) between the maximum optical
power and minimum optical power is Vπ , and multiplying that value by the device’s length (in cm)
gives the Vπ L figure of merit.
The Vπ L of the device of Fig. 6.7 is 6.7 V-cm. This compares well with the simulated
value of 6.1 V-cm. The difference is likely a combination of fabrication imperfections (such as
non-constant electrical line widths/heights/gaps) a non-exact LN r33 value in the LN film used
here, and slight variations to the Si waveguide dimensions. From these relatively minor errors can
result the 9.8% error between simulation and measurement. Note that for Vπ L measurements at
zero-frequency, the other end of the CPW can be left as an open circuit since only an electric field is

104
 Optical Power (dBm)
0

Measurement
-10 Fit

V
L = 6.7 V-cm

-20
 = 1561 nm
ER = 25 dB
-30
-20 -10 0 10 20
Voltage (V)

Figure 6.7: A plot of optical power versus applied dc voltage. The dc voltage supply is a Keithley
2420. Device length is 5 mm. ER = extinction ratio. Fitting was performed in MATLAB.

required to induce an optical phase shift in the device. For non-zero frequencies a load is required
to prevent reflections and for impedance matching.

6.3.4 Low Frequency Electro-Optic Response

At low frequencies,2 a VNA can be used with a photodetector to measure the frequency
response of an EOM. Port 1 of the VNA is used as a signal, which gets fed in to the electrical input
of the EOM. The electrical output is loaded with a 50 Ω impedance, and the optical output is put
directly into a photodetector whose output goes into Port 2 of the VNA. The VNA can then be
swept across its frequency range and, as long as proper calibration has been performed to account
for all sources of electrical loss (cables, adapters, probes, the photodetector, etc.), the frequency
response produced by the VNA will be that of the EOM.
An example of this measurement is shown in Fig. 6.8 for a low-frequency EOM device. The
setup is shown in Fig. 6.8a and the measurement is shown in Fig. 6.8b.

6.3.5 High Frequency Electro-Optic Response

At higher frequencies, an optical spectrum analyzer (OSA) can be used to extract the
electro-optic response. In this experiment, a signal generator is used to generate a sine wave that is
2 Here “low” means 6 GHz, but really this method works at any frequency as long as a VNA and a photodetector are
available for the desired frequency range.

105
 a b
1

EO Response (dBo)
100 MHz
VNA 0

-1
PD
-2
Laser EOM -3 Measured
Theory
-4
0 2 4 6
Frequency (GHz)

Figure 6.8: a Experimental setup for low frequency modulation response. PD = photodetector.
Red arrows represent the optical signal, black arrows represent the electrical signal. b Measured
response of a low frequency EOM from 100 MHz up to 6 GHz (the limit of the VNA). Units of
dBo are used to denote that this is the optical modulation response.

launched onto the EOM device through high frequency GSG (ground-signal-ground) probes. The
electrical output again is loaded with a 50 Ω impedance, just as in the low frequency measurement
of Section 6.3.4. However, instead of sending the modulated optical output into a photodetector
connected to the second port of the VNA generating the microwave signal, it is sent into an OSA.
Due to the nature of phase modulators, the modulated optical signal will have optical sidebands
spaced at frequencies corresponding to the electrical frequency of the input sine wave; the power
in these sidebands is proportional to the frequency response of the modulator [5]. By carefully
measuring these sidebands, high frequency electro-optic measurements can be performed with only
a signal generator and an OSA.

6.4 Electro-Optic Characterization of an Electro-Optic Modu-

lator

Electro-optic modulation, the imprinting of a radio-frequency (RF) waveform on an optical

carrier, is one of the most important photonics functions, being crucial for high-bandwidth signal
generation, optical switching, optical waveform shaping, data communications, ultrafast measure-
ments, sensing and ranging, timing and telemetry, and RF photonics. Although silicon (Si) photonic
electro-optic modulators can be fabricated using wafer-scale technology compatible with the semi-

106
conductor industry, Si depletion-mode phase electro-optic modulators (EOMs) are limited to an
electrical 3-dB bandwidth of about 50 GHz [6], whereas applications will benefit by having access to
higher RF frequencies. Bulk lithium niobate (LN) modulators [7] and etched LN modulators [8] can
scale to 70-100 GHz bandwidths, but are not integrated with the Si photonics toolkit and fabrication
process adopted widely over the last decade. Here, we show an ultra-high-bandwidth Mach-Zehnder
EOM, based on silicon photonics, made using conventional lithography and wafer-scale fabrication,
bonded to an unpatterned LN thin film using a low-temperature (200◦ C) back-end process without
etching or patterning the LN film. Our design allows for compatibility with silicon photonics, with
light input/output and optical components, including directional couplers, low-radius bends, and
path-length difference segments, realized in a foundry Si photonics process while avoiding high
temperature processing [9]. This hybrid LN-Si MZM achieves beyond 100 GHz 3-dB electrical
bandwidth.

6.4.1 Fabrication

The fabrication approach for the final devices developed for this thesis work is similar to the
general approach of Chapter 4, but with a few changes.
In our fabrication approach, depicted in Fig. 6.9, MZMs were built on a silicon photonics
platform, using photolithography on silicon-on-insulator wafers (220 nm Si thickness, 3 µm oxide
thickness) and did not require sub-resolution features unlike most plasmonic or polymeric slot
modulators [10, 11]. Silicon thinning (down to 150 nm) and feature patterning were followed by
oxide deposition and subsequent chemical mechanical polishing and oxide thinning by a timed
wet etch (diluted hydrofluoric acid) process. After die segmentation, commercially-procured x-cut
thin-film LN on insulator (NanoLN, Jinan Jingzheng Electronics Co. Ltd.) was bonded over a large
area ( 1 cm2 ) with a pressure of 45 kPa, but not processed further (e.g., no etching [12, 13] or sawing
[14] of LN was performed). Oxide bonding was done at room temperature after surface cleaning
and surface plasma activation steps. The bonded sample was thermally annealed at 200◦ C for one
hour under pressure. The bonded stack has been shown to withstand repeated temperature-cycling to

107
a Thin film LN dies b c
(bonded where needed) Aluminum 0.75 cm

SiO2
LN
Oxide die bond

2.5 cm
SiO2

2.1 cm
SOI
SiP n
io
R eg
d
Si WGs bri
Hy ion
g
Re
0.5 cm
SOI photonic circuits SiP n
io
(1000's of devices per wafer) R eg

d
SiP Region Hybrid Region SiP Region
5 mm
LN Out
3-dB DC
PLD 3-dB DC
GSG Electrodes

Figure 6.9: a Thin film x-cut lithium-niobate (LN) on insulator dies were bonded at room
temperature to segmented dies of a patterned and planarized silicon-on-insulator (SOI) wafer
which contained fabricated silicon photonic waveguide circuits. No etching or patterning of the
LN film was performed. b Exploded representation of the EOM, where an unpatterned, un-etched
LN thin film was bonded to a Mach-Zehnder interferometer fabricated in Si. Aluminum electrodes
were deposited on a 50 nm SiO2 layer over the LN film. SiP Region denotes the SiO2-clad region
outside the bonded LN film, containing Si waveguide circuits, such as feeder waveguides, bends,
directional couplers, and path-length difference segments. c Top view of a representative fabricated
hybrid Si-LN EOM test chip, which contains 60 EOM waveguide structures in parallel (in the
north-south direction); for this report, test electrodes for use in push-pull configuration were only
fabricated on one EOM device. d Composite microscope image of the EOM. DC: directional
coupler, PLD: path-length difference, GSG: ground-signal-ground, SiP: Si photonics.

108
at least 300◦ C [15], sufficient for the post-processing required here. In fact, several fabricated chips
were repeatedly processed, after bonding, through multiple cycles of electrode formation, removal
and re-formation, in search of the optimal dimensions. No debonding or noticeable degradation
to the stability or quality of the samples occurred during these additional process steps. The LN
die handle was removed, followed by coplanar waveguide electrode formation using aluminum
deposition, with total electrode thickness of 1.6 µm. A fully fabricated chip is shown in Fig. 6.9c;
a microscope image of the EOM is provided in Fig. 6.9d. The electrodes used here are more
than 15 times thinner than those used in Ref. [7], thus improving fabrication practicality. The
silicon photonic features were made on a high-resistivity Si handle wafer, potentially mitigating
piezoelectric resonances from traditional LN substrates [16]. As described below, the optical
input and output from the MZM section were through (crystalline) silicon photonic waveguiding
structures.

6.4.2 Electro-Optic Mach-Zehnder Modulator Design

For the MZM reported here, the silicon photonics region outside the bonded LN area
included four types of optical waveguide structures: fully-etched tapers for light input and output
from the chip (edge couplers), single-mode broadband directional couplers (>15 dB extinction ratio
from 1525-1575 nm, maximum of approximately 30 dB), path-length difference (PLD) segment
(including spline curve bends), and adiabatic waveguide tapers for inter-layer transitions (Si-to-LN
and vice-versa). Precise foundry processing of the Si photonic features results in accurate and
repeatable formation of the directional coupler splitting ratio. Since the LN layer is neither patterned
nor etched in our design, there was no alignment issue at the bonding step; the Si features alone
determine the optical propagation path.
The adiabatic waveguide tapers were designed to achieve a vertical inter-layer transition
(from Si to LN, and the reverse). As shown in Fig. 6.1, the design uses the TE-polarized fundamental
guided mode, which is also used in conventional silicon photonics at 1.5 µm wavelengths [17].
Since the refractive index of Si at these wavelengths (approximately 3.5) is significantly higher

109
than that of LN (approximately 2.2), the large index difference enables control of the mode size
and location (i.e., mainly in the Si rib or the LN slab) through lithography of the Si layer alone.
Thus, only the width of the Si waveguide (w) was varied in our design; when w > 600 nm, light
at 1.55 µm is mostly confined within the Si rib with confinement factor ΓSi = 64% (Mode A) and
ΓSi = 58% (Mode B). For w = 320 nm, light is guided in Mode C and sees the LN slab layer, with
confinement fraction in the LN layer calculated as ΓLN = 81% and ΓSi = 5%. Longitudinal Poynting
vector simulations of these modes are shown in Fig. 6.1c. The Mode B to Mode C transition loss
is estimated as 0.1 dB from simulations and is described in more detail in previous work [18].
A benefit of these high-bandwidth modulators is that ΓLN has less variation with small errors in
fabricated waveguide dimensions than plasmonic or polymeric slot waveguide MZMs.
Vertical, inter-layer transitions to and from the hybrid LN-Si region occur only where
needed, inside the perimeter of the bonded region. Optical losses between Modes A and B are
minimized by keeping the Si waveguide wide (w = 650 nm) when crossing into the hybrid region.
Thus, the edges of the bonded thin film, even if rough on the scale of the optical wavelength, do
not significantly affect optical propagation. This makes the back-end integration of thin-film LN
simple and feasible, without requiring precision etching or patterning of either LN or silicon after
bonding. Complex waveguiding circuits can be built up with a single bonded layer and multiple
vertical transitions, as shown elsewhere [18], but were not required here. A design library of hybrid
LN-Si components was created to aid in simulations and design within the Lumerical Interconnect
simulation environment [19], of which portions of the design manual are included in Chapter 3.

6.4.3 Electro-Optic Measurements

Light input and output was achieved using tapered single-mode, polarization-maintaining
fibers, whose positions were controlled using micro-positioning stages. From test structures, an
optical propagation loss of -0.6 dB/cm in the hybrid LN-Si region was measured. The propagation
losses in the Si-only regions were about -1.3 dB/cm and are kept short in this design. The edges
of the silicon photonic chip were lightly polished, but not fully prepared or packaged; hence, the

110
edge coupling loss was about -3 dB per edge and the total fiber-to-fiber insertion loss was -13.6 dB.
The calculated intrinsic loss of the full MZM (not including edge couplers), based on the measured
propagation loss (-0.6 dB/cm in the EOM, -1.3 dB/cm outside the EOM) and the device length (0.5
cm for the EOM, 1.67 cm outside the EOM), simulated inter-layer transition loss estimates (-0.1 dB
each), should be about -2.9 dB. However, the actual insertion loss was around -7.6 dB. The extra
4.7 dB is likely due to higher-than-expected loss in the broadband directional couplers and tapers,
as well as an estimated (from simulations) additional -0.4 dB loss due to the electrical lines, which
pass directly over the optical mode (see Fig. 6.9d). It is also possible that the propagation losses in
this particular device are higher than expected due to the multiple re-fabrication performed on this
chip in search of the optimum electrode structure. At low speeds, the MZM demonstrated a high
extinction ratio (> 20 dB) as shown in Fig. 6.10a, with Vπ L = 6.7 V.cm at dc for an L = 0.5 cm
device.

a b c 60
0
Optical Power (dB)

0 8
-10 50
-10 -20
6 nm

dB

4
m (dB/cm) 40
-30 Zc
-20 |S12|2dB
Measurement -40 2
|S22|2dB nm = 2.25 30
Fit
-30 -50
-20 -10 0 10 20 0.01 20 40 60 80 100 0.01 20 40 60 80 100
Voltage (V) Frequency (GHz) Frequency (GHz)

Figure 6.10: a Normalized optical transmission of the Mach-Zehnder interferometric electro-optic

Modulator (MZM), versus dc voltage at optical wavelength = 1560 nm. Fitted Vπ L = 6.7 V.cm for
device length L = 0.5 cm. b Measured electrical S-parameters of the MZM’s coplanar-waveguide
line used to model the transmission line characteristics. c Left y-axis: extracted microwave phase
index nm and microwave loss αm (dB/cm) over the dc-110 GHz frequency range. Right y-axis:
characteristic electrode impedance Zc (Ω).

RF measurements were performed on a bare-die chip using 50- probes rated to 110 GHz and
using laboratory equipment and RF waveguide components also rated and calibrated to about 110
GHz. The RF driving waveform was either from an RF oscillator (up to 67 GHz) or three different
frequency multipliers (sequentially covering the range of frequencies up to 106 GHz). GSG probes
were used for both launch and termination. Calibration of the signal pathway was performed using

111
a high-frequency RF power sensor. To inform a computational model of the expected behavior,
electrical S-parameters were measured using a Vector Network Analyzer up to 110 GHz, as shown
in Fig. 6.10b, and analyzed using standard algebraic transformations and lossy transmission line
circuit analysis [4]. The minor artifact near 90 GHz is attributed to repeated testing by contact
probes, and does not disrupt the smoothly-varying general trend seen in the data. The microwave
refractive index, nm , was fitted to a power-law equation and the characteristic impedance, Zc, was
fitted to a first-order polynomial of the RF frequency. As shown in Fig. 6.10c, nm = 2.25 and Zc
varied between 53.4 and 55.1 Ω from dc to 110 GHz. The microwave loss, αm , when fitted to a
power-law equation, followed approximately an f1/4 dependence, in contrast with the typical f1/2
dependence in traditional EOMs [7]. Because we have relatively thin electrodes and a Si substrate,
αm is not due solely to conductor loss, and instead, includes a combination of conductor, substrate,
and radiation losses.
Using the electro-optic bandwidth theory derived in Chapter 1 and the extracted parameters
of Fig. 6.10c (which used Equations (6.25) and (6.26)), a calculated electro-optic bandwidth curve
is shown in Fig. 6.11 (solid black line) along with the measured electro-optic response using the
OSA sideband technique of Section 6.3.5. Specifically, the method of Ref. [5] was used to detect
signals and modulation sidebands at an optical wavelength of 1560 nm. With the modulator biased
at quadrature, the difference (log scale) between the optical intensity of the first sideband and carrier
signal was used to extract the modulation index, and thus the frequency response, from 106 GHz
down to 2 GHz (providing a safe margin for the 0.18 GHz resolution of the OSA). The peak-to-peak
RF drive amplitude was about 1 V. Frequency multipliers were used for the frequency range above
67 GHz up to 106 GHz. The electro-optic response is shown in Fig. 6.11, and an (electrical) 3-dB
bandwidth was observed to lie well beyond 106 GHz, the limit of our measurement capabilities.
The measurement matches well with the calculated response (solid black line in Fig. 6.11). Electro-
optic measurements at frequencies beyond 67 GHz require RF multipliers and time-consuming
calibrations using appropriate waveguides, cables, probes and detectors for each frequency band, as
observed elsewhere [20]. Some of the scatter in the measurements at the highest frequencies arises

112
from the calibration of the frequency extenders, which have nonlinear and discontinuous dispersion
curves. The measured flat-spectrum modulation response is consistent with our simulation based
on electrical S-parameter measurements, which predicts flat frequency response to even higher
frequencies.
a
Extracted from Sideband 1
EO Response (dBe)

0
-3
-5
Measured Simulated
-10
10 20 30 40 50 60 70 80 90 100 110
Frequency (GHz)
b
Extracted from Sideband 2
EO Response (dBe)

0
-3
-5
Measured Simulated
-10
10 20 30 40 50 60 70 80 90 100 110
Frequency (GHz)

Figure 6.11: Electro-optic response of the EOM for both sidebands (a and b) from the optical
spectrum analyzer. Solid black line: calculated response from electrical S-parameters of Fig. 3c;
black circles: electro-optic response from sideband OSA measurements.

The product Vπ L = 6.7 V.cm (with a 0.5 cm length) in this device compares favorably with
other high-bandwidth LN MZIs and commercial technology, but the present devices were not
designed to minimize Vπ L. Recently, a fully-etched thin-film LN-on-SiO2 MZM was reported,
which achieved low Vπ L = 2.2 V.cm [8]. In contrast with our device, that device is not integrated
with other integrated photonics components, and relies on etching LN, which may lead to concerns
such as heat and pyroelectric charge buildup, structural defect formation and Nb depletion [21, 22].
An order-of-magnitude lower Vπ L product can be achieved using plasmonic [23] or graphene [24]
modulators, which are less widely adopted than LN modulators at this time.

113
6.5 Conclusion

In this thesis work, an un-patterned thin-film of LN was simply bonded at room temperature
to the patterned and planarized Si waveguide circuits, with an anneal step at 200◦ C. In contrast,
fabrication of bulk titanium-indiffused LN modulators and doped III-V or Si modulators require
at least 600◦ C and typically 900-1000◦ C [25, 26], limiting them to either standalone or front-end-
of-line device fabrication. Based on our approach, integration of electronic circuits alongside
a very-high-bandwidth LN EOM may be envisioned. More complex integrated optics can be
realized, which use a wider range of silicon photonic components such as filters, interferometers
and detectors alongside one or several ultra-high-bandwidth modulators, formed using single-step
bonding. Integration may also avoid the challenges of traditional packaging of LN EOMs: the
70-GHz unpackaged LN modulator of Ref. [7] was reported to achieve a 3-dB bandwidth, when
packaged into a stand-alone module, of only about 30 GHz [27].
In summary, an electro-optic Mach Zehnder modulator (MZM) based on single-mode silicon
(Si) photonic circuits bonded at low temperature to an unpatterned, un-etched thin-film of lithium
niobate (LN) and utilizing the well-known Pockels electro-optic effect has been shown. Both theory
and measurements support the performance of this device as a greater-than-100-GHz electrical
bandwidth EOM, realized using a new design and fabrication process that brings lithium niobate,
the traditional electro-optic material-of-choice in the first few decades of integrated optics, into
compatibility with silicon photonics, the more recent platform for more complex integrated optics.
The input and output are in silicon photonics and, through the use of inter-layer vertical waveguide
transitions, the device is not sensitive to the rough edges, if any, of the LN thin film. The fabrication
process, which does not require etching or sawing of LN, is based on a standard silicon photonics
foundry fabrication flow. Such a device can bring ultrawide electro-optic bandwidths to integrated
silicon photonics, and benefit applications in analog and digital communications, millimeter-wave
instrumentation, analog-to-digital conversion, sensing, antenna remoting and phased arrays.

114
6.6 Acknowledgments

Chapter 6, in part, is currently being prepared for submission for publication of the material.
Peter O. Weigel, Jie Zhao, Kelvin Fang, Hasan Al-Rubaye, Douglas Trotter, Dana Hood, John
Mudrick, Christina Dallo, Andrew T. Pomerene, Andrew L. Starbuck, Christopher T. DeRose,
Anthony L. Lentine, Gabriel Rebeiz and Shayan Mookherjea. Portions of Chapter 6 are reprinted
from Conference on Lasers and Electro-Optics (CLEO) 2018. Peter O. Weigel, Jie Zhao, Douglas
Trotter, Dana Hood, John Mudrick, Christina Dallo, Andrew T. Pomerene, Andrew L. Starbuck,
Christopher T. DeRose, Anthony L. Lentine and Shayan Mookherjea, paper SF2I.4. The dissertation
author, together with his adviser, led the research efforts for this work and co-authored the papers.

Bibliography
[1] Yariv A, Yeh P. Photonics. Oxford University Press Oxford, UK:; 2007.

[2] Suchoski P, Findakly TK, Leonberger F. Stable low-loss proton-exchanged LiNbO 3 waveguide
devices with no electro-optic degradation. Optics letters. 1988;13(11):1050–1052.

[3] Farin G. Curves and surfaces for computer-aided geometric design: a practical guide. Elsevier;
2014.

[4] Pozar DM. Microwave engineering. John Wiley & Sons; 2009.

[5] Shi Y, Yan L, Willner AE. High-speed electrooptic modulator characterization using optical
spectrum analysis. Journal of lightwave technology. 2003;21(10):2358.

[6] Sun J, Sakib M, Driscoll J, Kumar R, Jayatilleka H, Chetrit Y, et al.. A 128 Gb/s PAM4 Silicon
Microring Modulator; 2018. Paper TH4A.7.

[7] Noguchi K, Mitomi O, Miyazawa H. Millimeter-Wave Ti:LiNbO3 Optical Modulators. Journal

of Lightwave Technology. 1998;16:615–619.

[8] Zhang M, Wang C, Chen X, Bertrand M, Shams-Ansari A, Chandrasekhar S, et al. Ultra-High

Bandwidth Integrated Lithium Niobate Modulators with Record-Low Vpi. vol. TH4A.5. OFC;
2018. .

[9] DeRose CT, Gehl M, Long C, Boynton N, Martinez N, Pomerene A, et al. Radio frequency
silicon photonics at Sandia National Laboratories. 2016 IEEE Avionics and Vehicle Fiber-
Optics and Photonics Conference (AVFOP); 2016. p. 3–4.

115
[10] Zhu S, Lo GQ, Kwong DL. Phase modulation in horizontal metal-insulator-silicon-insulator-
metal plasmonic waveguides. Optics Express. 2013;21:8320–8330.

[11] Ayata M, Fedoryshyn Y, Heni W, Baeuerle B, Josten A, Zahner M, et al. High-speed plasmonic
modulator in a single metal layer. Science. 2017;358:630–632.

[12] Mercante AJ, Yao P, Shi S, Schneider G, Murakowski J, Prather DW. 110 GHz CMOS
Compoatible thin film LiNbO3 modulator on silicon. Optics Express. 2016;24:15590–15595.

[13] Ulliac G, Calero V, Ndao A, Baida F, Bernal MP. Argon plasma inductively coupled plasma
reactive ion etching study for smooth sidewall thin film lithium niobate waveguide application.
Optical Materials. 2016;53:1–5.

[14] Courjal N, Devaux F, Gerthoffer A, Guyot C, Henrot F, Ndao A, et al. Low-loss LiNbO3
tapered-ridge waveguides made by optical-grade dicing. Optics Express. 2015;23:13983–
13990.

[15] Weigel PO, Mookherjea S. Reducing the thermal stress in a heterogeneous material stack for
large-area hybrid optical silicon-lithium niobate waveguide micro-chips. Optical Materials.
2017;66:605–610.

[16] Nightingale JL, Becker RA, Willis PC, Vrhel JS. Characterization of frequency dispersion in
Ti-indiffused lithium niobate optical devices. Applied Physics Letters. 1987;51:716–718.

[17] Vlasov YA, McNab SJ. Losses in single-mode silicon-on-insulator strip waveguides and bends.
Optics Express. 2004;12:1622–1631.

[18] Weigel PO, Savanier M, DeRose CT, Pomerene AT, Starbuck LA, Lentine AL, et al. Lightwave
Circuits in Lithium Niobate through Hybrid Waveguides with Silicon Photonics. Scientific
Reports. 2016.

[19] Weigel PO, Mookherjea S. Process Design Kit and Modulator Simulation for Hybrid Silicon-
Lithium Niobate Integrated Optics. CLEO; 2017.

[20] Hoessbacher C, Josten A, Baeuerle B, Fedoryshyn Y, Hettrich H, Salamin Y, et al. Plasmonic

modulator with >170 GHz bandwidth demonstrated at 100 GBd NRZ. Optics Express.
2017;25:1762–1768.

[21] Ren Z, Heard PJ, Marshall JM, Thomas PA, Yu S. Etching characteristics of LiNbO3 in reactive
ion etching and inductively coupled plasma. Journal of Applied Physics. 2008;103:034109.

[22] Kirkby CJ. Low-energy ion-beam processing damage in lithium niobate surface-acoustic-wave
optical waveguide devices and its post-manufacture removal. Journal of Materials Science.
1992;27:3637–3641.

[23] Haffner C, Heni W, Fedoryshyn Y, Niegemann J, Melikyan A, Elder DL, et al. All-plasmonic
Mach-Zehnder modulator enabling optical high-speed communication at the microscale.
Nature Photonics. 2015;9:525–528.

116
[24] Sorianello V, Midrio M, Contestabile G, Asselberg I, Campenhout JV, Huyghebaerts CG,
et al. Graphene-silicon phase modulators with gigahertz bandwidth. Nature Photonics.
2018;12:40–44.

[25] Wooten EL, Kissa KM, Yi-Yan A, Murphy EJ, Lafaw DA, Hallemeier PF, et al. A Review
of Lithium Niobate Modulators for Fiber-Optic Communications Systems. IEEE Journal of
Selected Topics in Quantum Electronics. 2000;.

[26] Stolk PA, Gossman HJ, Eaglesham DJ, Jacobson DC, Rafferty CS, Gilmer GH, et al. Physical
mechanisms of transient enhanced dopant diffusion in ion-implanted silicon. Journal of
Applied Physics. 1997;81:6031–6050.

[27] Howerton MM, Burns WK. Broadband traveling wave modulators in LiNbO3. Cambridge
University Press; 2002.

117
Chapter 7

Second Harmonic Generation in Hybrid

Waveguides

7.1 Introduction

The electro-optic modulation discussed so far in this thesis is a special case of a set of
optical phenomena known as nonlinear optics. The “nonlinear” portion of that name refers to
the frequency mixing process whereby input signals combine in some way (through addition or
subtraction of frequencies; the “mixing” process) to generate one or more output signals at different
frequencies. The electro-optic Pockels effect in lithium niobate (LN) can be thought of as a type
of nonlinear optical process where a high frequency optical signal, typically on the order of tens
or hundreds of THz, interacts with a much lower electrical signal, well below 1 THz, to generate
a new electromagnetic wave. Since the frequency of the electrical signal is several orders of
magnitude lower than that of the optical signal, the generated electromagnetic wave produced by
the electrical-optical interaction is assumed to have the same wavelength as the input optical signal,
and the electrical signal is treated as a perturbation. This approximation allows for index matching
in electro-optic devices, as opposed to the more precise phase matching required in optical-optical
nonlinear interactions.

118
 In particular, LN-based nonlinear optical devices have strict phase matching criteria that have
hindered their development from research into useful products. The design discussed in this chapter
proposes bonding Si-based circuitry, and silicon carbide waveguides in particular (SiC), to a LN
film. With the proper design and further development efforts, it may be possible to monolithically
integrate electrical circuitry into a LN-based nonlinear optical device to stabilize the highly-sensitive
phase matching conditions.

7.2 Optical Second-Harmonic Generation

In the field of nonlinear optics, there are several types of nonlinear processes. In addition to
the aforementioned electro-optic Pockels effect, one of the most commonly used processes is second
harmonic generation (SHG). Every nonlinear process must conserve energy in some way, and so
each nonlinear process is defined by its energy conservation condition. SHG is the process whereby
a pump laser in a second-order nonlinear material, such as LN, generates an electromagnetic wave
at twice the input laser’s frequency (ω + ω = 2ω). This chapter will only consider the SHG process.

7.2.1 Derivation of Power at the Generated Second Harmonic Frequency

To begin this derivation, and as with any derivation, a few assumptions must be made. First,
it is assumed that the optical modes are linearly polarized horizontally along the crystal (ẑ) axis of
the x-cut LN. The form we use for the electric field is

Ei (x, y, z,t) = Ai (y)Ei (x, z)e j(βi y−ωit) + A∗i (y)Ei∗ (x, z)e− j(βi y−ωit) (7.1)

where i is an index used to number the optical waves (this notation becomes helpful when considering
more than one wave), Ai is the longitudinal amplitude, Ei is the transverse wave function of the
mode, βi is the (real) phase constant (βi = 2πni /λi where ni is the effective mode index and λi is the

119
free space wavelength), and ωi is the angular frequency of the wave (ωi = 2πc/λi ). Some authors
(e.g., [1]) include a factor of 1/2 in this description of Ei . Admittedly, there is a justified reason for
doing so if the wave is defined by the standard definition of a plane wave1 :

Ei = E0 cos(ωit − βi y + φ)

= ℜ[Ai (y)Ei (x, z)e j(βi y−ωit) ] (7.2)

1 1
= Ai (y)Ei (x, z)e j(βi y−ωit) + A∗i (y)Ei∗ (x, z)e− j(βi y−ωit) ,
2 2

where E0 is a real-valued vector describing the amplitude of the wave and all additional phase
information is contained in φ.
However, when working with optical waves, the exponential form of Ei is easier to handle
algebraically and is commonly used, and so the form of Ei in Equation (7.1) is used here with the
understanding that Ei = 2E0 cos(ωit − βi y + φ).
As with most electromagnetic derivations, we begin with Maxwell’s equations, which in
differential form and in SI units are

∇ · Di = ρ (7.3)

∇ · Bi = 0 (7.4)
∂Bi
∇ × Ei = − (7.5)
∂t
∂Di
∇ × Hi = Ji + . (7.6)
∂t

Ei represents the electric field in free space, while Di is used to describe how the electric field
behaves in materials (likewise, Bi describes magnetic fields in free space and Hi describes magnetic
fields in materials). In Gaussian units, Di = Ei in free space, agreeing with the implications of the
previous sentence. However, because we choose to work in SI units, we introduce a constant ε0 so
1 Note x+x∗ (a+ jb)+(a− jb)
that for x = a + jb: 2 = 2 = ℜ[x].

120
that Di = ε0 Ei and Bi = µ0 Hi in free space. ε0 is the vacuum permittivity and µ0 is the vacuum
permeability. These are constants with known values: ε0 =8.854187817...×10−12 F/m (Farads per
meter) and µ0 =1.256637061...×10−6 H/m (Henries per meter).
For passive optical waveguides in non-magnetic materials: ρ = 0 (there are no free charges),
Ji = 0 (there is no electrical current), and Bi = µ0 Hi (the material is non-magnetic). Taking the cross
product of Equation (7.5) and knowing that ∇ × (∇ × V) = ∇(∇ · V) − ∇2 V for some vector V, we
find that

∂2 D i
∇(∇ · Ei ) − ∇2 Ei = −µ0 (7.7)
∂t 2

by using Equation (7.6). In a material, Di = ε0 Ei is no longer necessarily valid. An additional term

Pi is added to account for how the material responds to Ei :

Di = ε0 Ei + Pi (7.8)

where Pi is called the dielectric polarization density and responds to the electric field Ei dependent
on the material. Pi can be expanded as a Taylor series:

(1) (2)
Pi = a1 Ei + a2 EE + ... = Pi + Pi + ... (7.9)

The expansion of Equation (7.9) goes to infinity, but since we are considering interactions in LN
we only care about first-order (linear) and second-order (nonlinear, i.e. the interaction of multiple
input waves) – this is why we only look at two teRMS on the right-hand side of Equation (7.9).
Because the first term is linear, it is simply a linear response to Ei . This is true for all optical
waves in a material, even when there are multiple optical waves concurrently propagating in the
same material. However, when multiple waves concurrently propagate in a material, they can
generate new waves under the proper conditions. The second-order nonlinear term in Equation (7.9)
represents the behavior of the material when subject to two optical waves. Likewise, a third-order
term would represent the material’s response to three optical waves, a fourth-order term would

121
represent the material’s response to four optical waves, and so on. These interactions are governed
by energy conservation, so that the newly generated wave in second-order nonlinear interactions has
a frequency equal to the sum (or difference) of the initial waves. More specifically, E in Equation
(7.9) is the sum of the two input waves E1 and E2 , where each has the form of Equation (7.1). In
that case, we find that

(2)
Pi =a2 [A21 E12 e2 j(β1 y−ω1t) + A22 E22 e2 j(β2 y−ω2t)

+ 2A1 A2 E1 E2 e j((β1 +β2 )y−(ω1 +ω2 )t)

(7.10)
+ 2A1 A∗2 E1 E2∗ e j((β1 −β2 )y−(ω1 −ω2 )t) + c.c.]

+ 2a2 [A1 A∗1 E1 E1∗ + A2 A∗2 E2 E2∗ ].

Equation (7.10) contains every possible second-order nonlinear interaction. They can be
organized by the frequencies in the exponential of each term, which have to add up to ωi by energy
conservation. What this means is that 2ω1 = ωi and 2ω2 = ωi in the first two teRMS describe second
harmonic generation (SHG), ω1 + ω2 = ωi in the third term describes sum frequency generation
(SFG), ω1 − ω2 = ωi in the fourth term describes difference frequency generation (DFG), and
ω1 − ω1 = ωi = 0 and ω2 − ω2 = ωi = 0 in the last two teRMS describe optical rectification (OR).
We don’t consider the complex conjugate teRMS because they describe the same processes as the
first four teRMS, but with negative frequencies.
Two waves at fixed frequencies will not excite all these interactions. For example, a single
wavelength of light can generate a second-order harmonic by SHG. That same single wavelength
of light can never meet the energy conservation conditions of SFG or DFG because they require
multiple wavelengths. However, the generated signal from SHG could potentially interact with its
first harmonic source via SFG or DFG to generate a third wavelength of light. We can determine
the efficiencies of these interactions by using coupled mode theory.
First, we will follow standard convention and replace a1 and a2 in Equation (7.9) and
Equation (7.10) with a1 = ε0 (εr − 1) and a2 = ε0 χ(2)33 = 2ε0 d33 . εr is the relative permittivity of

122
 (2)
a material, and χ33 is the second-order nonlinear susceptibility of a LN crystal when all three
interacting waves are polarized along the ẑ-axis of the crystal [1]. Likewise, d33 is the nonlinear
(2)
coefficient and defined (by historical convention) as d33 = 21 χ33 .
Next, we can reorganize Equation (7.8) so that

(2)
Di = ε0 εr Ei + Pi . (7.11)

(2)
Because of the addition of Pi in Equation (7.11), ∇ · Ei = 0 is only exactly true for isotropic
(2)
materials (so that εr does not change across space) where Pi is zero and there are no nonlinear
interactions. However, ∇ · Ei ≈ 0 is a good assumption in most nonlinear cases. Using this
assumption and substituting Equation (7.11) into Equation (7.7) gives the nonlinear wave equation:

(2)
∂2 Ei ∂2 P i
∇2 Ei − µ0 ε = µ 0 , (7.12)
∂t 2 ∂t 2

where ε = ε0 εr . Because we are considering SHG, we know from Equation (7.10) that

h i
(2) 2 2 2 j(βFH y−ωFH t)
PSH = 2ε0 d33 AFH EFH e + c.c. (7.13)
h i
(2)
PFH = 4ε0 d33 ASH ESH A∗FH EFH
∗ j((βSH −βFH )y−(ωSH −ωFH )t)
e + c.c. (7.14)

where “c.c” is used to denote the complex conjugate and the i subscript has been replaced with “SH”
for the interaction that would generate a second harmonic wave from the first harmonic (FH) pump
and ωSH = 2ωFH . Now Equation (7.12) can be expanded into its three constituent equations – one
for each interacting wave:

123
 (2)
∂2 ESH ∂2 PSH
∇2 ESH − µ0 ε = µ 0 (7.15)
∂t 2 ∂t 2
(2)
2 ∂2 EFH ∂2 PFH
∇ EFH − µ0 ε = µ0 (7.16)
∂t 2 ∂t 2
(2)
∂2 EFH ∂2 PFH
∇2 EFH − µ0 ε = µ 0 (7.17)
∂t 2 ∂t 2

There are two equations for EFH because SHG requires two pumps at the FH wavelength to maintain
energy conservation.
For now we will only consider Equation (7.16). ∇2 EFH can be expanded so that

(2)
∂2 EFH ∂2 EFH ∂2 PFH
∇t2 EFH + − µ 0 ε = µ0 (7.18)
∂y2 ∂t 2 ∂t 2

where ∇t is the transverse component of ∇2 . We can ignore the complex conjugate on each side of
the equation and still maintain the equality. In that case, the second-order derivative with respect to
ŷ can be expanded to get

∂2 AFH ∂AFH
∇t2 EFH + 2
EFH e j(βFH y−ωFHt) + j2βFH EFH e j(βFH y−ωFHt)
∂y ∂y
(2)
(7.19)
∂ 2E
FH ∂2 PFH
−β2FH EFH − µ0 ε 2
= µ0 .
∂t ∂t 2

Here we can make a useful observation: the transverse component of EFH is independent
of the transverse component of ESH . In other words, the mode profile of EFH is not affected by
the presence of ESH . This means that ∇t2 EFH will be the same regardless of whether or not ESH is
present. We can imagine a waveguide that produces exactly the same mode profile but without any
nonlinear properties, in which case the right-hand side of Equation (7.18) goes to zero and we are
left with the linear wave equation:

124
 ∂2 EFH ∂2 EFH
∇t2 EFH + − µ0 ε = 0. (7.20)
∂y2 ∂t 2

In a linear waveguide, AFH does not change with propagation length because it does not
couple into the SH mode. In this case, ∂2 EFH /∂y2 simply reduces to -β2 EFH and we have

∂2 EFH
∇t2 EFH − β2 EFH − µ0 ε = 0. (7.21)
∂t 2

By substituting Equation (7.21) into Equation (7.19) and using the slowly varying amplitude
approximation (∂2 AFH /∂y2  2βFH ∂AFH /∂y), we have

(2)
∂AFH j(βFH y−ωFH t) ∂2 PFH
j2βFH EFH e = µ0 . (7.22)
∂y ∂t 2

From Equation (7.14) and remembering that ωSH − ωFH = ωFH , Equation (7.22) becomes

∂AFH
j2βFH EFH e jβFH y = −4ω2FH µ0 ε0 d33 ASH ESH A∗FH EFH
∗ j(βSH −βFH )y
e (7.23)
∂y

after taking the second-order derivative of PFH with respect to time and removing like teRMS from
both sides of the equation. We are interested in learning how AFH changes along the propagation
length of the device, so by moving teRMS around and integrating both sides of Equation (7.23)
across the transverse plane, we arrive at an expression for ∂AFH /∂y:

∂AFH 4π
=j ASH A∗FH Γ∗ e− j∆βy (7.24)
∂y nFH λFH

where nFH is the effective mode index of the FH wave, ∆β = 2βFH − βSH , and

Z +∞
∗
Γ= ESH (x, z) · deff (x, z) : EFH (x, z)EFH (x, z) dA (7.25)
−∞
R +∞
where : is a double-dot product, the normalization is chosen to be −∞ |Ei (x, z)|2 dA = 1 and

125
 




0 0<x<∞


deff (x, z) = d33 −hLN ≤ x ≤ 0 (7.26)





0
 −∞ < x < −hLN .

In Equation (7.26), hLN is the thickness of the LN film. Equation (7.24) is the coupled mode
equation for AFH in SHG in an x-cut LN film-based waveguide.
A similar derivation can be performed with Equation (7.15) to find a coupled mode equation
for ASH :

∂ASH 4π
=j A2 Γe j∆βy (7.27)
∂y nSH λFH FH

In most applications, the assumption that ∂AFH /∂y ≈ 0 is used because the amount of energy
transferred into the SH mode is typically much less than the initial energy in the FH mode. This
is known as the “undepleted pump” approximation. In that case, AFH in Equation (7.27) does not
vary along the propagation length and a simple integration along the propagation direction can be
performed to get

4π 2

j∆βL

ASH (y = L) = A Γ e −1 (7.28)
nSH λFH ∆β FH

where L is the length of the nonlinear region.

While Equation (7.28) is helpful for seeing the relationship between the SH and FH waves,
what is more useful is an expression for power in the SH wave. Power in the mode can be found
by calculating the Poynting vector S from SSH = ESH × HSH , finding the time averaged Poynting
vector, and integrating the real part along ŷ over the transverse plane. Using the definition of E in
Equation (7.1), these operations can be described by

Z ∞
PSH = 2 ℜ [ESH × H∗SH ] · ŷ dA. (7.29)
−∞

126
 An expression for HSH can be found from Equation (7.5). Because we are only interested in
the power propagating along the device (i.e. along the ŷ-axis) and the electric field is nearly entirely
polarized along the ẑ-axis, the only component of HSH that we are concerned with is that which is
polarized along the x̂-axis. From Equation (7.5) we have

∂Ez
= jωSH µ0 Hx (7.30)
∂y

after taking the derivative of H with respect to time. From Equation (7.1) and ignoring the complex
conjugate we have

Ez = ASH (y)ESH (x, z)e j(βSH y−ωSHt) . (7.31)

We take the derivative to get

∂Ez ∂ASH 1
= Ez + jβEz . (7.32)
∂y ∂y ASH

By substituting Equation (7.27) and Equation (7.28) into Equation (7.32) and replacing “L”
in Equation (7.28) with “y,” we get

∆β e j∆βy βSH
Hx = j∆βy
Ez + Ez . (7.33)
ωSH µ0 e −1 ωSH µ0

The cross product of ESH and H∗SH in Equation (7.29) simplifies to Ez H∗x . The real part of that
product is

|Ez |2
 
∆β
ℜ[Ez H∗x ] = βSH + . (7.34)
ωSH µ0 2

In a device designed for nonlinear optics, ∆β is ideally zero and practically much smaller than βSH ,
so that in all reasonable cases ∆β/2  βSH . In that case we can ignore the second term in Equation
(7.34) and can use Equation (7.29) to get

127
 2βSH
Z ∞
PSH = |ASH |2 |ESH |2 dA, (7.35)
ωSH µ0 −∞

but from our normalization condition the integral equals 1 and the power in the SH wave reduces to

PSH = 2cε0 nSH |ASH |2 . (7.36)

In fact, the form of Equation (7.36) is valid for any TE waveguide mode defined by the electric field
in Equation (7.1). Thus,

PFH = 2cε0 nFH |AFH |2 . (7.37)

From Equation (7.28) and Equation (7.36), we have

128π2 cε0 ∆βL 2

 
2 4
PSH = |Γ| |AFH | sin . (7.38)
nSH λ2SH ∆β2 2

We will now make two important substitutions. First, we will replace the sine in Equation (7.38)
with a sinc: sin(x)2 = x2 sinc(x)2 ; second, we will use Equation (7.37) to replace |AFH |4 . After
making these substitutions, we finally arrive at an expression for power in the SH wave under the
undepleted pump approximation:

8π2 |Γ|2
 
2 2 2 ∆βL
PSH = P L sinc [W]. (7.39)
nSH n2FH cε0 λ2FH FH NL 2

7.3 Design of Folded Hybrid Silicon Carbide-Lithium Niobate

Waveguides for Efficient Second Harmonic Generation

In this section, a design for a highly efficient folded hybrid silicon carbide-lithium niobate
(SiC-LN) SHG device is discussed, based on [2]. Although the focus of this thesis has been on
electro-optic devices, the theoretical work shown here will hopefully provide a strong case for why

128
the thin film LN platform should be considered for nonlinear optics, as well as even an idea of how
such devices can be fabricated in the not-so-distant future.

7.3.1 A Brief Background of Integrated Photonic Circuits and Nonlinear

Optics in LN

Equation (7.39), the power in a SH wave during SHG, consists of leading teRMS and a
sinc2 (x) term, where x is a function of the phase mismatch ∆β. It turns out that for most waveguides
designed for SHG, there is significant enough dispersion between λFH and λSH that the phase
mismatch is prohibitively large, in which case the nonlinear efficiency is negligible and no second
harmonic wave is generated. Several solutions to this problem have been investigated and attempted
over the years and are documented in [3]. However, for LN photonic circuits, the d33 coefficient is
desirable due to its size and the fact that two similarly polarized waves at the FH and SH wavelengths
can be used efficiently. In this case, the most promising technique for producing high efficiency
nonlinear devices is periodic poling of the LN crystal. This is known as periodically poled lithium
niobate, or PPLN. PPLN is one of the most well-known and widely used materials, but has not been
studied in the proposed configuration: a nonlinear device that can be folded into a compact region
without etching PPLN, leveraging the technology to make complex integrated photonic circuits in
bonded thin-film lithium niobate [4]. This configuration both minimizes the longest edge of the
device on a microchip, and achieves very high conversion efficiency.
High-index-contrast silicon or III-V photonics are used to drive modern integrated optics
toward achieving more functionality in a smaller device and energy footprint, and tighter integration
with microelectronics. Scaled-up integrated photonic circuits with dozens of electrically-driven
or active (e.g., laser or detector) components have been densely integrated within a few squared
millimeters [5, 6]. However, nonlinear optical (NLO) devices are difficult to scale-up, for a few
different reasons.
Crystalline lithium niobate (LiNbO3 or LN) has been an attractive material for nonlinear
optics (NLO) due to its wide transparency window, low loss, and its large second-order nonlinear

129
coefficient, d33 [7] (five times higher than that of AlN [8], for example). However, NLO devices
based on LN (and similar materials) mostly use waveguides with titanium indiffusion or annealed
proton exchange to create a relatively weak core-cladding index difference [9, 10, 11], resulting
in a large modal effective area, large minimum bending radius and low nonlinear efficiencies.
Thus, a long and straight device is typically used to generate appreciable amounts of power by
second-harmonic generation or wavelength conversion. Unlike the square footprints achieved in
silicon or III-V photonics (which are amenable to microelectronics packaging), LN NLO devices
tend to be narrow in one dimension (tens of microns) and very long in the other (several centimeters)
[10].
This work presents and studies a way to overcome this limitation. As a representative case
study, designs and simulations of an NLO LN device for wavelength conversion from 1550 nm to
775 nm wavelengths are presented and discussed. Two aspects of this approach enable the improved
structure: (a) a high-index contrast material, silicon carbide (SiC), is used as a “ridge” to load a thin
slab of LN, thus enabling a confined mode with a small effective area Aeff at both the fundamental
and second-harmonic wavelengths with (simultaneously) low loss, low dispersion, and ease of
fabrication; and (b) bends are designed to periodically compensate for walk-off, thus enabling a very
compact device with simplified poling and packaging requirements. High conversion efficiencies
are predicted even with realistic assumptions for loss, disorder as affecting the periodic poling, and
imperfect fabrication of bends.
This design is possible thanks to advances in crystal ion slicing over the past twenty years
[12, 13] and the introduction of commercially available LN thin films on handle (abbreviated as
LNOI). In fact, a new class of optical waveguides has recently emerged, wherein a semiconductor or
glass rib material is bonded or deposited to a LN thin film to form a hybrid waveguide. These hybrid
devices have used Si ribs on LN (Si-LN) to make all-optical interferometers [4] and electro-optic
modulators [14], and SiN-LN to make nonlinear optical (NLO) waveguides [15] and electro-optic
modulators [16]. While SiN-LN hybrid waveguides are useful for nonlinear conversion into the
visible regime, the low refractive index of SiN (n≈2) compared with LN (n≈2.13 along the crystal

130
axis) prevents the SiN-LN structure from being used in more complex geometries. Conversely,
Si-LN devices are useful because of the high index of Si. By changing the width of the high-index
Si rib, the optical mode can be confined in the Si to make low-loss, low-radius bends and complex
waveguiding devices that require such bends [4]. Yet the low bandgap of Si (1.1 eV) prevents the
hybrid Si-LN waveguide from being used for nonlinear processes in the visible regime.

7.3.2 Overview of the Proposed Structure

Here, we propose a new hybrid waveguide design: amorphous silicon carbide (SiC) which
can be deposited by plasma-enhanced chemical vapor deposition (PECVD) onto a LN film, and
patterned into waveguides. Such a SiC film has a band gap of Eg,SiC = 2.18 eV and a refractive index
in the range of 2.9-3.1, as shown in Fig. 7.1. SiC deposition was performed in an Oxford Plasmalab
80+ PECVD at a temperature of 350◦ C, an RF power of 20 W, a chamber pressure of 1000 mT, a
SiH4 flow rate of 300 sccm, and a CH4 flow rate of 50 sccm. Under these conditions, we observed a
deposition rate of 7.73 nm/min. The SiC used in this work has a much higher refractive index than
that of SiN or LN. The n (real refractive index) and k (imaginary refractive index) curves of Fig. 7.1
were fit from ellipsometry over the wavelength range of 200-2000 nm to the Tauc-Lorentz model
[17] using the Levenberg-Marquardt non-linear fitting algorithm in Regress Pro, a free spectroscopic
ellipsometry fitting software. The fitting parameters and the figure of merit during the fit (χ2 , not the
same as the nonlinear optical susceptibility) are provided in Table. 7.1. The SiC film has a bandgap
of Eg = 2.18 eV, nearly twice as wide as that of Si.

Recent works in hybrid LN waveguides have used Si (Eg,Si = 1.1 eV, nSi ≈3.5 [18, 19, 4, 14])
and SiN (Eg,SiN = 5.0, nSiN ≈2 [19, 15, 16]). SiC has a wider band gap than Si, and a higher refractive
index than both SiN and LN [20]. A wider band gap material allows for nonlinear interactions
below 1.1 µm (the band gap of Si), particularly in the visible regime; a higher refractive index than
LN allows for a larger variety of hybrid waveguide designs, as well as the potential for high-density

131
 Table 7.1: Fitted parameters of the Tauc-Lorentz model for our PECVD SiC film. See [17] for
more details on the Tauc-Lorentz model.

Fit Parameter Value

AL 313.795
E0 (eV) 2.99442
C 2.79093
Eg (eV) 2.1766
ε∞ 1.30125
χ2 71.1215

4
n

k
1 n
k

0
0.2 0.6 1.0 1.4 1.8
Wavelength (
m)

Figure 7.1: Measured n and k curves of PECVD amorphous SiC.

on-chip circuitry similar to what is possible with silicon photonics. SiC is also becoming available
through foundry fabrication of devices at wafer-scale, potentially benefiting integrated optics using
this material.
The design described in this work consists of two sections with the cross-section of Fig. 7.2a:
1) a traditional straight waveguide NLO section; and 2) several iterations of a straight waveguide
NLO section plus a taper plus a 180◦ bend, as in Fig. 7.2b, to fold the device into a compact footprint
and reduce the poling length significantly. Because of the high index of SiC, tight bends can be
made in the SiC layer to keep the length of the device short without reducing the total nonlinear
interaction length.
As poling length decreases, poling domain width increases based on the number of bends
in the device. The design described here has a width of approximately 600 µm, though wider or

132
a b PPLN Tapers Bends
x SiO2
SiC
z In
300 nm x-cut LN
y
z
SiO2

Si Out

Figure 7.2: a Cross-section of hybrid SiC-LN waveguide. b Top view of our proposed structure,
consisting of a quasi-phase matched region (the “PPLN” section), an adiabatic taper to convert the
SiC width from a narrow rib to a wide rib, and two 90◦ spline bends to wrap the SiC back around
into the PPLN section.

narrower folded structures can be designed. This distance is comparable to the poling width of bulk
LN devices, which are typically z-cut and poled through the substrate.

7.3.3 Additional Theory of the Nonlinear Interaction Region

As stated earlier, the NLO interaction consider here is second harmonic generation (SHG)
from 1550 nm to 775 nm, using fundamental TE-like modes with a cross-section of Fig. 7.2a, where
the x-cut LN is a 300 nm film. As will be shown throughout this chapter, choosing such a thin LN
film and cladding it with low index SiO2 increases the efficiency of the SHG process by reducing
the optical mode area well below the mode sizes of traditional LN devices.
Because the device of interest here makes use of PPLN, Equation 7.39 is no longer precisely
correct. Physically, periodically poling the LN crystal is the act of flipping the sign of the d33
nonlinear coefficient, as in Fig. 7.2b. This can be thought of as a square wave, which can be defined
mathematically as

  
2πy
d(y) = d33 sgn cos , (7.40)
Λ

where d(z) is the value of the nonlinear coefficient d3 3 along the length of the device, sgn is the
sign, or signum, function, and Λ is the grating period (the physical period of a single section of
poled crystal). By representing d(y) in Equation 7.40 as a Fourier series, we find [3]:

133
 2
dm = d33 , m = odd. (7.41)
mπ

In Equation 7.41, m denotes the order of the grating structure. For example, the structure
shown in Fig. 7.2b is a first-order grating structure, where each positive and negative section of d33
is of equal length. In this case, d33 becomes (2/π)d33 , and Equation 7.39 becomes:

 2
8π2 |Γ|2
 
2 2 2 2 ∆βL
PSH = P L sinc [W]. (7.42)
π nSH n2FH cε0 λ2FH FH NL 2

Higher-order grating structures result in significantly less efficient devices since the 2/π term is
squared in Equation (7.42), though the requirements on poling period length become less strict.
It should be noted that even values of m can be used, since an even value of m corresponds to a
complete phase mismatch period in one LN domain region, effectively having no impact on the
phase matching conditions.
Variations of Equation (7.42) can be derived, and are well cataloged in [3]. Of interest in
this chapter is the SH power for a lossy waveguide, which is:

sin2 ( ∆βL2 NL ) + sinh2 ( (αFH −αSH /2)LNL

 2 2 2 2 " #
2 8π |Γ| LNL PFH 2 −(αFH +αSH /2)LNL 2 )
PSH = 2 2
e × (α −α /2)L
.
π nSH nFH ε0 cλFH ∆βL NL
( 2 ) +( 2 FH SH NL 2
)
2
(7.43)
αSH and αFH are the SH and FH mode losses, respectively, in units of inverse length. Equation (7.43)
is used in Fig. 7.5a. As is commonly done [21], we define the normalized nonlinear conversion
efficiency as:

 2
8π2 |Γ|2
 
2 %
η0 = · 100 · 0.012 . (7.44)
π nSH n2FH cε0 λ2FH W − cm2

In Equation (7.44), only the effective mode indices, wavelength, and Γ can be altered by the
designer. In practice, Γ is optimized to improve η0 , as the mode indices are much less sensitive to
changes in the geometry and wavelength is typically pre-determined.

134
7.3.4 Silicon Carbide-Lithium Niobate Second Harmonic Generation De-

sign

The three variable geometric parameters in the cross-section of Fig. 7.2a are the width (wSiC )
and height (hSiC ) of the SiC rib, as well as the height (hLN ) of the LN film. As stated in the previous
section, hLN = 300 nm is chosen to keep the mode area small.

a 0 b 0 = 1280 %/(W-cm2)
1400
b  = 775 nm
1200
(%/[W-cm2])

1000
800 2
m
600 hSiC = 100nm
400 hSiC = 150nm  = 1550 nm
0

hSiC = 200nm
200
hSiC = 250nm
0
50 100 150 200 250 2
m
SiC Rib Width (nm)

Figure 7.3: a Sweep of η0 versus SiC width for SiC heights of 100 nm, 150 nm, 200 nm, and
250 nm. b Mode pictures for wSiC = 120 nm and hSiC = 250 nm for the SH mode (top) and FH
mode (bottom).

Lumerical’s MODE Solutions was used to simulate the modes of the FH and SH waves for a
given waveguide cross-section. η0 was calculated from Equation (7.44) for a variety of combinations
of wSiC and hSiC . These simulated η0 values are shown in Fig. 7.3a, with modes from the highest
η0 design (wSiC = 120 nm, hSiC = 250 nm) shown in Fig. 7.3b, where η0 = 1280 %/(W-cm2 ). A
value of d33 ≈ 25 pm/V was used for these simulations [22].
The local maximum of each curve in Fig. 7.3a exists because of an inherent trade-off
between Aeff (effective mode area) and LNCF (LN confinement factor) for the two interacting
modes; if the modes are too big then η0 decreases, but the only portion of the modes that contributes
to second-order NLO is that which exists in the LN film layer. The ideal waveguide design should
result in low Aeff values and high LNCF values at both FH and SH wavelengths. However, because
the SiC rib has a higher index than that of the LN film, if the SiC is made too large then LNCF will
decreases and η0 will be reduced. For a more detailed analysis of Aeff and LNCF, please refer to

135
sections 7.3.8 and 7.3.9.
Each of the curves in Fig. 7.3a has a maximum over 1200 %/(W-cm2 ) and a grating period
Λ between 3.418 µm and 3.421 µm.

7.3.5 Impact of Variations of Design Parameters

The effects of typical fabrication and real-world conditions on the operation of the hybrid
SiC-LN SHG NLO device are nontrivial. Some amount of detailed analysis is required not only to
understand fabrication tolerances, but also to have a good sense of the amount of post-fabrication
tuning (by, for example, varying device temperature or operating wavelength) that is available.

a 156
b c d
206 304

Temperature (oC)
slope = 0.17 nm/nm slope = 1.09 nm/nm slope = -0.65 nm/nm 60 slope = 61.5 oC/nm
302
wSiC (nm)

hSiC (nm)

hLN (nm)
152 202 40
300
148 198 20
Simulation Simulation 298 Simulation Simulation
Linear Fit Linear Fit Linear Fit Linear Fit
144 194 296 0
1520 1540 1560 1580 1546 1550 1554 1546 1550 1554 1549.6 1550 1550.4

FH (nm) FH (nm) FH (nm) FH (nm)

Figure 7.4: How the phase matching FH wavelength shifts as a function of wSiC , hSiC , hLN , and
temperature is shown in a, b, c, and d, respectively.

The wavelengths of the guided modes which can participate in a quasi-phase matched (QPM;
a device that uses periodic poling or some other scheme to force ∆β = 0) interaction are determined
by the LN grating period Λ, fixed during the design and fabrication. As discussed in Section 7.2.1,
∆β = βSH -2βFH -2π/Λ. If we set ∆β = 0 (so the device is (quasi)-phase matched) and rearrange this
equation we obtain:

2π
Λ= (nSH − nFH ) , (7.45)
λFH

where nSH and nFH are the SH and FH mode effective indices and are dependent upon λFH , wSiC ,
hSiC , and hLN . Equation (7.45) sets the condition for Λ during the waveguide design at a particular
wavelength (in this case, λFH = 1550 nm). Ideally, ∆β = 0 and second harmonic power is always at

136
a maximum. In reality, fabrication errors will alter nSH and nFH , and ∆β 6= 0 for λFH = 1550 nm.
Instead, ∆β will equal zero at some other combination of wavelengths.
 
This can be thought of as a shift of the sinc2 ∆βL2 NL curve in Equation (7.42). In Fig. 7.4a-c,
plots are shown of the simulated shift of this peak with fabrication errors for a designed structure
of wSiC = 150 nm and hSiC = 200 nm. Each simulated curve was matched with a linear fit (a good
approximation for the most part, though there is some quadratic behavior in Fig. 7.4b) to compare
 
the plots. The sinc2 ∆βL2 NL peak is most sensitive to the width of the SiC rib in Fig. 7.4a, where a
1 nm deviation causes a nearly 6 nm change in FH wavelength. This is not surprising when one
considers Fig. 7.9 in Section 7.3.9: when the SiC rib changes, LNCF – and therefore the mode
 
effective index – will also change, resulting in a shift in the sinc2 ∆βL2 NL peak. Because LNCF is
less sensitive to changes in the height of the hybrid waveguide (both SiC and LN), hSiC and hLN
 
have less of an impact on the change in the sinc2 ∆βL2 NL peak.
The plots of Fig. 7.4a-c assume the device is at 25◦ C. In Fig. 7.4d, the effect of temper-
 
ature on the sinc2 ∆βL2 NL peak is simulated, where temperature-dependent material refractive
indices cause ∆k to vary with temperature. The change in refractive index of amorphous SiC as a
function of temperature around room temperature is approximately linear, and displays a value of
dn
dT = 9.2 × 10−5 ◦ C−1 [23]. Similarly, the temperature-dependent refractive index models of [20]
and [24] are used for LN and SiO2 , respectively. The simulated device is fairly stable from 0-60◦ C
(Fig. 7.4d). In this range, the FH wavelength varies only by approximately 1 nm, low enough to be
tuned with a common semiconductor diode laser to account for thermal drift.

7.3.6 Loss and Disorder

In addition to cross-section geometry fabrication tolerances, one must also consider the
impact of a realistic level of propagation loss at the FH and SH wavelengths, and quantify the
expected reduction in conversion efficiency that would result. In a real device, there will be
some amount of propagation loss that will degrade the generated SH output power. When loss is
introduced into the SHG model with Equation (7.43), the peak second harmonic output power is

137
reduced (Fig. 7.5a). In fabricated waveguides, losses for thin-film LN is 0.3–1.2 dB/cm [25, 26, 15]
and around 5 dB/cm for amorphous SiC [27]. For a fairly reasonable hybrid waveguide propagation
loss of 2 dB/cm, peak SH power is reduced by 2.98 dB, or approximately a factor of 2. In the case
of 5 dB/cm propagation loss – the upper loss limit for this waveguide, since SiC is its most lossy
material – peak SH power is reduced by approximately a factor of 5.

a 1.0 b 1.0
PFH = 10 mW PFH = 10 mW
0.8 LNL = 1 cm 1.22 dB LNL = 1 cm
0.8
2.98 dB
PSH (mW)

PSH (mW)
0.6 0.6
0.147 nm
0.4 0 dB/cm 0.4 0 nm
1 dB/cm 50 nm
0.2 2 dB/cm 0.2 100 nm
5 dB/cm 250 nm
0 0
1549.8 1550.0 1550.2 1549.8 1550.0 1550.2
Wavelength (nm) Wavelength (nm)

Figure 7.5: Simulated PSH when loss is added to the model (a) and when duty cycle error is added
to the model (b). In both cases, the input power (PFH ) is 10 mW and the nonlinear interaction
length is assumed to be 1 cm. In b, each curve represents a different RMS poling period error,
from 0 nm (no error) to 250 nm.

A similar analysis for Fig. 7.5a was performed in Fig. 7.5b, but instead of adding loss to the
waveguide it was assumed that the QPM grating period had different amounts of duty cycle error.
“Duty cycle error” refers to variations in the periodicity of domain inversion regions of the LN film.
For this design, Λ = 2.79 µm. While these grating periods can be expected to be periodic on average,
local disorder is inevitable during poling of the domain inversions. Fig. 7.5b uses direct numerical
integration of Equations (7.15)-(7.17) to calculate PSH under the undepleted pump approximation.
For realistic values of duty cycle error (0-0.25 µm), there is at most a 1.22 dB drop in maximum PSH .
In the case of a 0.1 µm duty cycle error, the drop in maximum power and the FWHM bandwidth
reduction are both negligible. For this reason we expect the hybrid SiC-LN waveguide to be quite
robust to QPM grating period duty cycle error.

138
7.3.7 Bent Hybrid Waveguide Region

A high-index contrast hybrid waveguide such as SiC-LN can be formed into low-loss,
low-radius bends to connect straight NLO waveguide sections, as shown in Fig. 7.2b. A top view
of a single 180◦ bend with tapers and phase markings is shown in Fig. 7.6a, where an adiabatic
linear taper is used to connect the nonlinear mixing section with the bent section. A straight wide
SiC waveguide is placed between the adiabatic taper and the bend to account for phase mismatch
that occurs between the FH and SH waves in the taper and bend. If the FH and SH waves are out
of phase, then the newly generated SH wave in each additional nonlinear mixing section could
destructively interfere with the previously generated SH waves. The phases φ in Fig. 7.6b are
∆βL radians along the lengths of the tapers, wide straight waveguides, and bends.
a PPLN Taper Str. b c
4000 15
F = 571 (2) rad.
Phase Mismatch (rad.)

Lc
LN
3000 Lt
=0 =A =B Spline

AD (mm2)
10
SiC D,E
Bend 2000 A,B 
=C C
5
1000
=F =E =D
0 0
Bent Region 0 200 400 600 0 100 200 300 400
Waveguide Length (
m) Length ( m)

Figure 7.6: a Top view of the 180◦ bend, plus tapers and additional straight waveguides (“Str.”),
connecting adjacent QPM NLO sections. The phase along the structure is tracked in b to ensure
that an integer multiple of ∆k = 2π rad. occurs at the output of the bend. c Plot of device area AD
versus Lc (solid black curve) when Lt = 100 µm and Lt (dashed red curve) when Lc = 40 µm. For
both curves, LNL = 1 cm and n = 10.

Two 90◦ spline bends are used in Fig. 7.6 to create the 180◦ bend. In order to minimize the
total footprint of the device, we derive an approximate equation for its area:

n
AD = wD × hD ≈ 2Lc LNL + 4Lt Lc n + 4Lc 2 n, (7.46)
n+1

where AD is the footprint of the device, wD = 2Lc n, hD = Ls +2Lt +Lc , Lc is the characteristic length
of the 90◦ spline bend, LNL is the total nonlinear interaction length (the sum of all the straight
narrow section lengths), n is the number of 180◦ turns, and Lt is the length of the adiabatic taper.
As one should expect, AD is proportional to the square of Lc , meaning that Lc , not Lt , is more

139
critical when considering a compact design. This behavior is shown in Fig. 7.6c. For this reason, an
important design consideration for this structure is minimization of the characteristic spline bend
without incurring significant optical losses. In addition, reducing Lc reduces the width of the device,
resulting in a narrower poling width.
Simulated spline bending losses are shown in Fig. 7.7a,b at 775 nm and 1550 nm, respectively.
These simulations were performed in Lumerical’s FDTD Solutions software. Losses are higher
for the 1550 nm wave because the modes are larger and less confined in the SiC rib for a given rib
width. For our design, we chose a characteristic length of 30 µm and a SiC rib width of 700 nm.
From Fig. 7.7a,b, this is a good compromise between low losses and short Lc .
Since the narrow (nonlinear section, width of 150 nm) and wide (spline bend, width of
700 nm) SiC rib widths – see Fig. 7.6a – are now determined for this design, connecting them with a
linear taper is simply a matter of determining Lt . Taper losses for various Lt at both the FH and SH
wavelengths, also simulated with Lumerical’s FDTD Solutions software, are provided in Fig. 7.7c.
We chose a value of Lt = 300 µm.
Using the device parameters chosen in this work, and for LNL = 1 cm and n = 10, AD = 0.94 mm2 .
This means a 1 cm long device can be reduced to less than a 1 mm2 area with a maximum dimension
of less than 1.6 mm. The loss penalty for such a device is 2.4 dB for the FH wave and 0.4 dB for
the SH wave, and are almost entirely due to taper losses. These losses can be reduced by choosing a
longer taper length, though this would increase the footprint of the device.
The waveguiding components in the bent region (see Fig. 7.6a) are subject to fabrication
tolerances, and these fabrication tolerances can impact the SH output power because of interference
between separately generated SH waves. In Fig. 7.7d, direct numerical integration was used to
calculate PSH versus wavelength for several waveguide width errors, similar to what was done in
Fig. 7.5b with duty cycle error. In Fig. 7.7d, PSH is calculated for RMS waveguide width errors
between 0 nm and 10 nm along the length of each linear waveguide component for a structure with
10 bent regions and 1 mm nonlinear sections, for a total nonlinear interaction length of 1 cm. The
input FH power is 10 mW.

140
 a 90o Spline, = 1550 nm b 90o Spline,  = 775 nm
5 5

Bend Loss (dB/90o)

Bend Loss (dB/90o)

LN wSiC = 500 nm LN
1 SiC Lc 1 wSiC = 600 nm SiC Lc
wSiC = 700 nm
wSiC wSiC = 800 nm
wSiC
10-1 10 -1

wSiC = 500 nm
10-2 wSiC = 600 nm 10-2
wSiC = 700 nm
wSiC = 800 nm
10-3 10-35
10 20 30 40 10 15 20
Characteristic Length (
m) Characteristic Length (m)

c 4
Linear Taper d 1.0
Taper Loss (dB/taper)

= 1550 nm PFH = 10 mW 0 nm

1 = 775 nm 0.8 LNL = 1 cm 1 nm

5 nm

PSH (mW)
0.6 10 nm
0.082 nm

10-1 LN Taper Length 0.4

SiC 0.2
150 nm 700 nm
10-22 5 10 25 50 100 300
0
1549.8 1550.0 1550.2
Taper Length (m) Wavelength (nm)

Figure 7.7: Simulated spline bending losses for the SH and FH waves are shown in parts a and
b, respectively. Losses are significantly lower for the SH wave because of its reduced mode size.
Simulated taper losses are shown in part c for both SH and FH waves. d Simulated SH output
power in a bent hybrid SiC-LN NLO, SHG device with 10 bends and 10 nonlinear sections 1 mm
in length each. Different RMS waveguide width errors between 0 nm (no error) and 10 nm in the
bent region (Fig. 7.6a) were simulated.

The full width half max (FWHM) of the curves in Fig. 7.7d – even with 0 nm RMS
waveguide width error – is 0.082 nm compared with the FWHM of 0.147 nm in Fig. 7.5. This
reduction in FWHM can be explained by the additional phase matching requirements of the bent
region: in the bent region, a particular device geometry has been carefully chosen so that the phase
difference between the SH and FH waves at the output of the bent region is precisely a multiple
of 2π (Fig. 7.6b). As wavelength changes, that phase matching condition is no longer met and
separately generated SH waves partially interfere to reduce the SH output power at wavelengths
other than 1550 nm. However, the device is tolerant to RMS waveguide width errors up to 5 nm.
At a 10 nm waveguide width RMS error the curve of Fig. 7.7d loses its shape, though certain
wavelengths still generate sizable output SH power.
Because the phase-matching curve of the device narrows with the addition of waveguide
bends, fabrication of this device will be slightly more challenging than standard nonlinear optical
device fabrication. However, unlike traditional LN devices which are made by induffision or

141
implantation, features in the SiC rib can be fabricated with photolithography or electron beam
lithography, in which case the ribs width and height can be made accurate down to fewer than 10
nm - within the range of fabrication errors considered in Fig. 7.4. A tunable pump laser will likely
be required to offset any fabrication error in a practical device. At a pump wavelength of 1550 nm,
tunable lasers are commercially available.
If these fabrication challenges can be overcome, it may be possible to use this platform in
diverse applications such as, for example, maintaining the phase matching condition as is done with
adiabatically-chirped poling periods [28], but with a long sequence of carefully designed waveguide
bends for rephasing instead of poling.

7.3.8 Effective Mode Area (Aeff )

In typical non-hybrid NLO waveguide geometries, a reduced Aeff (effective mode area) is
analogous to a more efficient device since the intensity of each mode for a given amount of power
is increased in the NLO material. The definition of Aeff used here is:

R +∞
2
|E|2 dA
Aeff = R−∞
+∞ [m2 ]. (7.47)
−∞ |E|4 dA

In this hybrid SiC-LN waveguide, optical modes are vertically confined by both the SiC rib and
the LN film, and horizontally confined by the SiC rib. By reducing the thickness of the hybrid
waveguide (the combined thickness of the SiC rib and LN film), a lower Aeff can be obtained.
Since amorphous SiC does not display a χ(2) nonlinearity, any portion of the optical mode
that is not in the LN will not contribute to SHG. This is apparent in Fig. 7.8, where although Aeff
decreases as wSiC and hSiC increase for both the SH and FH modes in Fig. 7.8a and Fig. 7.8b, respec-
tively, η0 (calculated with (7.44)) is actually higher in the geometry of Fig. 7.8d (wSiC = 150 nm,
hSiC = 250 nm) than in Fig. 7.8c (wSiC = 250 nm, hSiC = 250 nm), despite the modes of Fig. 7.8c
having lower Aeff values. This is because the SH mode is almost entirely contained in the SiC rib in
Fig. 7.8c, whereas in Fig. 7.8d it has expanded enough into the LN to offset the efficiency penalty

142
incurred for increasing the mode sizes.

a b
λ = 775 nm λ = 1550 nm
3.0 12
hSiC = 100 nm hSiC = 100 nm
2.5 hSiC = 150 nm 10 hSiC = 150 nm
2.0 hSiC = 200 nm 8 hSiC = 200 nm
Aeff ( m2)

Aeff ( m2)
hSiC = 250 nm hSiC = 250 nm
1.5 6
1.0 4
d d
0.5 2 c
c
0 0
50 100 150 200 250 50 100 150 200 250
SiC Rib Width (nm) SiC Rib Width (nm)

c
0 = 110 %/(W-cm2) d
0 = 515 %/(W-cm2)

 = 775 nm  = 775 nm

2 m 2 m

 = 1550 nm  = 1550 nm

2 m 2 m

Figure 7.8: Sweep of Aeff versus SiC width at λ = 775 nm (part a) and λ = 1550 nm (part b) for SiC
heights of 100 nm, 150 nm, 200 nm, and 250 nm. c and d show mode pictures for wSiC = 250 nm
and hSiC 250 nm (part c) and wSiC = 150 nm and hSiC 250 nm (part d) for the SH mode (top) and
the FH mode (bottom). Despite the reduced mode areas of the design of c, the design of d has a
higher normalized conversion efficiency due to the locations of the modes.

7.3.9 Lithium Niobate Confinement Factor (LNCF)

The outcome of Section 7.3.8 is that a hybrid waveguide for NLO needs to be designed so
that the interacting modes do not only overlap in the smallest area possible, but that they do so

143
specifically in the LN region.
To track the portion of each mode that is in the LN film, we introduce a parameter called the
LN confinement factor (LNCF):

+∞
R R0
|Sy |2 dxdz
−∞ −hLN
LNCF = +∞
· 100 [%], (7.48)
R +∞
|S̄|2 dxdz
R
−∞ −∞

where S is the total time-averaged Poynting vector and Sy is the time-average Poynting vector
parallel to the propagation axis y. We only consider the y-component of the Poynting vector in the
numerator because we are only interested in knowing how much TE-like light resides in the LN
film for a given mode, and only the longitudinal component of the Poynting vector is non-zero in
the case of an optical mode that is 100% TE-polarized.
Simulated sweeps of LNCF versus wSiC for various hSiC in Fig. 7.9 reveal that the LNCF is
fairly stable for the FH mode across the SiC rib geometries we consider, but is sensitive to changes
in both wSiC and hSiC for the SH mode. We know from Fig. 7.8 that too narrow of a SiC rib will
dramatically increase the mode areas and reduce conversion efficiency, but we also know from
Fig. 7.9a-b that if we do not pick a narrow enough SiC rib then too much of the mode, especially
the SH mode, will be confined in the SiC rib. Knowing this, we select two reasonable choices for
the SiC geometry and show their interacting modes in Fig. 7.9c (wSiC = 100 nm, hSiC = 100 nm)
and Fig. 7.9d (wSiC = 200 nm, hSiC = 100 nm). Indeed, in both cases η0 is higher than in the designs
chosen in Fig. 7.8. However, the more efficient of the two designs is not the one where LNCF is at a
maximum (Fig. 7.9c), but when it is well below its maximum for the SH wave (Fig. 7.9d).

7.4 Concluding Thoughts

Despite the more than 50 years of research and discovery in the field of nonlinear optics,
it still remains a significant challenge to routinely design and fabricate highly efficient, integrated
nonlinear optical devices. This is in large part due to a combination of deviations from designs

144
a λ = 775 nm
b λ = 1550 nm
100 100
c
80 80
c d
LN CF (%)

LN CF (%)
60 d 60

40 hSiC = 100 nm 40 hSiC = 100 nm

hSiC = 150 nm hSiC = 150 nm
20 hSiC = 200 nm 20 hSiC = 200 nm
hSiC = 250 nm hSiC = 250 nm
0 0
50 100 150 200 250 50 100 150 200 250
SiC Rib Width (nm) SiC Rib Width (nm)
c η0 = 850 %/(W-cm2)
d η0 = 1090 %/(W-cm2)

λ = 775 nm λ = 775 nm

2 μm 2 μm

λ = 1550 nm λ = 1550 nm

2 μm 2 μm

Figure 7.9: Sweep of LNCF versus SiC width at λ = 775 nm (part a) and λ = 1550 nm (part
b) for SiC heights of 100 nm, 150 nm, 200 nm, and 250 nm. c and d show mode pictures for
wSiC = 100 nm and hSiC 100 nm (part c) and wSiC = 200 nm and hSiC 100 nm (part d) for the
SH mode (top) and the FH mode (bottom). d has a higher η0 than c because of its reduced Aeff ,
despite having a lower LNCF for the SH wave.

145
during fabrication and the strict design tolerances of integrated nonlinear optical devices. Much of
the work in this chapter on hybrid thin film LN-based devices has been done with the hope that it
may be useful as a foundation for the study of fabricating nonlinear optical devices in a foundry-
compatible process (as with the electro-optic modulators discussed in previous chapters), where
tolerances are tighter and feedback devices can be integrated on-chip. Perhaps in the not-so-distant
future foundry-fabricated, highly efficient, compact nonlinear optical devices will be developed that
feature monolithic integration with electrical circuits to monitor and perhaps eventually stabilize the
device for continuous, repeatable nonlinear mixing.

7.5 Acknowledgments

Chapter 7, in part, is a reprint of the material as it appears in Journal of the Optical Society
B 2018. Peter O. Weigel & Shayan Mookherjea, JOSA B 35(3), 593-600 and in Conference on
Lasers and Electro-Optics (CLEO) 2017. Peter O. Weigel, Marc Savanier and Shayan Mookherjea,
paper SW4M.5. The dissertation author, together with his adviser, led the research efforts for this
work and co-authored the papers.

Bibliography
[1] Yariv A, Yeh P. Photonics: optical electronics in modern communications. Oxford Univ.;
2006.

[2] Weigel PO, Mookherjea S. Design of folded hybrid silicon carbide-lithium niobate waveguides
for efficient second-harmonic generation. JOSA B. 2018;35(3):593–600.

[3] Sutherland RL. Handbook of nonlinear optics. CRC press; 2003.

[4] Weigel PO, Savanier M, DeRose CT, Pomerene AT, Starbuck AL, Lentine AL, et al. Lightwave
circuits in lithium niobate through hybrid waveguides with silicon photonics. Scientific reports.
2016;6:22301.

[5] Lee BG, Rylyakov AV, Green WM, Assefa S, Baks CW, Rimolo-Donadio R, et al. Monolithic
silicon integration of scaled photonic switch fabrics, CMOS logic, and device driver circuits.
Journal of Lightwave Technology. 2014;32(4):743–751.

146
 [6] Sun J, Timurdogan E, Yaacobi A, Hosseini ES, Watts MR. Large-scale nanophotonic phased
array. Nature. 2013;493(7431):195.

[7] Nikogosyan DN. Nonlinear optical crystals: a complete survey. Springer Science & Business
Media; 2006.

[8] Jung H, Guo X, Zhu N, Papp SB, Diddams SA, Tang HX. Phase-dependent interference
between frequency doubled comb lines in a χ (2) phase-matched aluminum nitride microring.
Optics letters. 2016;41(16):3747–3750.

[9] Schreiber G, Suche H, Lee Y, Grundkötter W, Quiring V, Ricken R, et al. Efficient cascaded
difference frequency conversion in periodically poled Ti: LiNbO3 waveguides using pulsed
and cw pumping. Applied Physics B. 2001;73(5-6):501–504.

[10] Schreiber G, Hofmann D, Grundkoetter W, Lee YL, Suche H, Quiring V, et al. Nonlinear
integrated optical frequency converters with periodically poled Ti: LiNbO 3 waveguides. In:
Integrated Optics Devices V. vol. 4277. International Society for Optics and Photonics; 2001.
p. 144–161.

[11] Amin J, Pruneri V, Webjörn J, Russell PSJ, Hanna D, Wilkinson J. Blue light generation in
a periodically poled Ti: LiNbO3 channel waveguide. Optics communications. 1997;135(1-
3):41–44.

[12] Levy M, Osgood Jr R, Liu R, Cross L, Cargill III G, Kumar A, et al. Fabrication of single-
crystal lithium niobate films by crystal ion slicing. Applied Physics Letters. 1998;73(16):2293–
2295.

[13] Rabiei P, Gunter P. Optical and electro-optical properties of submicrometer lithium niobate
slab waveguides prepared by crystal ion slicing and wafer bonding. Applied physics letters.
2004;85(20):4603–4605.

[14] Chen L, Xu Q, Wood MG, Reano RM. Hybrid silicon and lithium niobate electro-optical ring
modulator. Optica. 2014;1(2):112–118.

[15] Chang L, Li Y, Volet N, Wang L, Peters J, Bowers JE. Thin film wavelength converters for
photonic integrated circuits. Optica. 2016;3(5):531–535.

[16] Jin S, Xu L, Zhang H, Li Y. LiNbO 3 thin-film modulators using silicon nitride surface ridge
waveguides. IEEE Photonics Technology Letters. 2016;28(7):736–739.

[17] Jellison Jr G, Modine F. Parameterization of the optical functions of amorphous materials in

the interband region. Applied Physics Letters. 1996;69(3):371–373.

[18] Edwards DF, Ochoa E. Infrared refractive index of silicon. Applied optics. 1980;19(24):4130–
4131.

[19] Semiconductor V. General properties of Si, GE, SiGe, SiO2 and Si3 N4 ; 2002. Online. Available
from: www.virginiasemi.com/pdf/generalpropertiesSi62002.pdf.

147
[20] Edwards G, Lawrence M. A temperature-dependent dispersion equation for congruently grown
lithium niobate. Optical and quantum electronics. 1984;16(4):373–375.

[21] Fejer MM, Magel GA, Lim E. Quasi-phase-matched interactions in lithium niobate. In:
Nonlinear Optical Properties of Materials. vol. 1148. International Society for Optics and
Photonics; 1990. p. 213–225.

[22] Shoji I, Kondo T, Kitamoto A, Shirane M, Ito R. Absolute scale of second-order nonlinear-
optical coefficients. JOSA B. 1997;14(9):2268–2294.

[23] Della Corte FG, Montefusco ME, Moretti L, Rendina I, Rubino A. Study of the thermo-
optic effect in hydrogenated amorphous silicon and hydrogenated amorphous silicon carbide
between 300 and 500 K at 1.55 µm. Applied Physics Letters. 2001;79(2):168–170.

[24] Leviton DB, Frey BJ. Temperature-dependent absolute refractive index measurements of syn-
thetic fused silica. In: Optomechanical Technologies for Astronomy. vol. 6273. International
Society for Optics and Photonics; 2006. p. 62732K.

[25] Volk MF, Suntsov S, Rüter CE, Kip D. Low loss ridge waveguides in lithium niobate thin
films by optical grade diamond blade dicing. Optics express. 2016;24(2):1386–1391.

[26] Cai L, Wang Y, Hu H. Low-loss waveguides in a single-crystal lithium niobate thin film.
Optics letters. 2015;40(13):3013–3016.

[27] Shoji Y, Nakanishi K, Sakakibara Y, Kintaka K, Kawashima H, Mori M, et al. Hydrogenated

amorphous silicon carbide optical waveguide for telecommunication wavelength applications.
Applied Physics Express. 2010;3(12):122201.

[28] Leshem A, Meshulam G, Porat G, Arie A. Adiabatic second-harmonic generation. Optics

letters. 2016;41(6):1229–1232.

148
Appendix A

A.1 Alternative VπL Derivation

As an alternative to the Vπ L equation derived in Chapter 1, the following definition may

be used. This definition may be more useful depending on the simulation tool and method. Many
thanks to Dr. Ryan Aguinaldo for his discussions regarding this and many other mathematical
proofs.
From well-known electro-optic theory (see, for example, [1]), we know that:

1
∆ne ≈ − n3e r33 Ez (A.1a)
2

1
∆no ≈ − n3o r13 Ez (A.1b)
2

Equation (A.1a) describes an index change along the extraordinary axis of the crystal, and Equa-
tion (A.1b) describes an index change along the ordinary axes of the crystal; in both equations it is
assumed the RF electric field is polarized entirely along the extraordinary axis. We can rewrite the
RF field in its most general form as:

ERF (r̄) = VRF · ψ(r̄) (A.2)

149
where VRF is the voltage of the RF signal and ψ(r̄) is some space-dependent function. Because the
changes in indices described by Equation (A.1) are linear with respect to the RF field, we expect
the change in the effective index to also be linear with respect to the RF field:

Z Z
∆neff = C ERF (r̄) · f (r̄)d 2 r = CVRF ψ(r̄) · f (r̄)d 2 r = CVRF γ. (A.3)

Here we use C to represent some combination of constants, f (r̄) to denote some spatial
function related to the normalized optical field, and γ to constitute a two-dimensional summation of
the space-dependency attributed to the RF field and the optical field. Note that neff is the effective
index of the propagating optical signal before an external field is applied. Now, if we rearrange
(A.3) to be in terms of the constants C and γ, we can say:

∆neff
= Cγ. (A.4)
VRF

∆neff can be related to the phase by observing the space-dependent phase. For a fixed modulator
length L, the space-dependent phase shift is written as:

2π
∆φ = ∆βL = ∆neff L (A.5)
λ

In the desired case where ∆φ = π, Equation (A.5) becomes:

λ
∆neff,π = , (A.6)
2L

where ∆neff,π is the change in the effective index required to achieve a π radian phase shift. Using
the form of Equation (A.4) along with Equation (A.6), we have:

λ
= Cγ. (A.7)
2LVπ

Noticing that the right-hand side of Equation (A.4) is identical to the right-hand side of
Equation (A.7), we equate the left-hand sides of the respective equations and rearrange to obtain a

150
final equation for Vπ L that is in terms of the applied RF voltage and the change in effective index,
∆ne f f :

λ VRF
Vπ L = . (A.8)
2 ∆neff

Equation (A.8) allows for the calculation of Vπ L from any single VRF point. With this
expression, the Vπ L value for a given device structure can be computed by, for example, computing
the optical mode with and without an applied voltage. Coupling an electrostatic model with an
optical model is fairly straightforward in COMSOL Multiphysics, in which case this expression for
Vπ L is beneficial.
Lastly, the change in optical effective index can be written in a more obvious form:

∆neff = neff (VRF ) − neff (0). (A.9)

A.2 Deriving the Electro-Optic Response from Coupled Mode

Theory

Although the electro-optic bandwidth equation in Chapter 1 is accurate, it is possible to

arrive at a similar expression with coupled mode theory. In fact, several works over the years
have provided partial explanations of how to arrive at an accurate bandwidth model for a lossy
electro-optic modulator, but a complete derivation is difficult to find in the literature ([2, 3, 4]). We
take this opportunity to provide a complete derivation from coupled-mode theory, beginning with
the wave equation in a nonlinear medium (see [5] for details of how to arrive at this step):

1 ∂2 E j 1 ∂2 P j
∇2 E j − = , (A.10)
c2 ∂t 2 ε0 c2 ∂t 2

where j = 1, 2, or 3, corresponding respectively to the input optical signal, the input electrical signal,
and the output modulated optical signal; c is the speed of light in a vacuum, ε0 is the free space

151
permittivity, E is the electric field, and P is the polarization density field in the medium. For this
derivation, we will take our guided optical mode to be propagating along the x̂-axis in a LN crystal.
Our RF electric field and optical electric field components are assumed to be polarized along the
ẑ-axis (extraordinary axis) of the crystal, and therefore we can assume the electric field to be of the
form:

E j = A j (x)ei(k j x−ω j t) ẑ. (A.11)

A j (x) is the longitudinal-dependent amplitude of the electric field, k j is the propagation

constant of the field, and ω j is the frequency of the field. By examining the particular case where j
= 3 (output field), we can replace “j” in (A.11) with “opt,” and by standard coupled-mode theory
we arrive at the lossless differential equation for Aopt :

∂Aopt 2iωopt
= d33 Aopt Aelec ei∆kx . (A.12)
∂x cnopt

d33 is the component of the nonlinear susceptibility tensor which corresponds to our crystal and
applied field orientations. An important difference between Equation (A.12) and standard coupled-
mode equations (again, see [5]) is our assumption herein that the frequency of the input optical
signal is approximately the same as the frequency of the output optical signal. Truly, ω3 = ω1 +
ω2 , but the RF frequency ω2 (in the GHz frequency regime) is on the order of 1000 times less than
the optical input frequency ω1 (in the THz frequency regime). Thus the electrical signal is nearly
constant, with respect to an optical wavelength, along the length of the LN crystal, and so we can
make the assumption that A1 ≈ A3 . For this reason, we are able to place Aopt in the right-hand side
of Equation (A.12).
Though it is tempting to then say k1 ≈ k3 , this assumption will result in a static phase
mismatch ∆k, which is defined as:

∆k = k1 + k2 − k3 . (A.13)

152
 Specifically, if k1 ≈ k3 , then ∆k would always be equal to k2 , making us unable to minimize
our phase mismatch. Instead, we will expand this phase mismatch equation and realize that the
refractive index in the waveguide core at a frequency ω1 will be, in practicality, identical to the
refractive index in the waveguide core at a frequency ω3 . Thus:

2πn1 2πn2 2πn3

∆k = k1 + k2 − k3 = + − , (A.14)
λ1 λ1 λ1
1 1
where λ3 exactly equals λ1 + λ12 under the second-order nonlinear effect, and n j is the effective
index of the corresponding signal in the waveguide. As mentioned earlier, we can now say with
great accuracy that n1 ≈ n3 to obtain:

2π 2π
∆k = (n2 − n3 ) = (nRF − nopt ). (A.15)
λ2 λRF

This immediately tells us that we can minimize our phase mismatch (i.e. the phase difference
between the RF field and the optical field) by designing our modulator such that nRF is as close to
nopt as possible. If loss were not a factor, then the theory up to this point could be used to maximize
optical bandwidth (such as in [1]). Since we are considering loss as well – both optical loss and RF
loss – we must adjust Equation (A.12). This is done simply by phenomenologically subtracting a
loss term from (A.12), so that we have ([6]):

∂Aopt 2iωopt
= d33 Aopt Aelec ei∆kx − αopt Aopt . (A.16)
∂x cnopt

We make one final assumption before solving this differential equation. Since our pump
(the RF signal) is much stronger than the optical signal, we can take advantage of the undepleted
pump approximation and say that ARF (x) is only affected by material absorption:

ARF (x) = ARF (x = 0)e−αRF x . (A.17)

Finally, we are able to solve Equation (A.16) from the beginning of the modulation region

153
(x = 0) to the end of the modulation region (x = L), thereby obtaining an expression for the output
optical signal:

Aopt (x = L) = Aopt (x = 0)e−αopt L ei∆φ , (A.18)

where:

" #
∆kL αRF L ∆kL αRF L
2ωopt d33 αRF L ∆kL cos( ) sinh( ) − i sin( ) cosh( )
∆φ = LARF (x = 0)e− 2 ei 2 2 2
αRF L
2 2
. (A.19)
nopt c 2 − i ∆kL
2

Our next step is to consider the output of the MZM in terms of optical intensity, since
any practical use of the EOM requires data conversion back to the electrical domain by use of
a photodetector. The MZM output is a linear combination of the signal passing through the
unmodulated arm and the signal passing through the modulated arm. Using Equation (A.18) as our
description of the modulated signal, we can obtain an output amplitude Aout :

Aout (x) = Aopt (x = 0)e−αopt L ei∆φ + Aopt (x = 0)e−αopt L = Aopt (x = 0)e−αopt L (1 + ei∆φ ), (A.20)

And thereby calculate the proportional value of Iout :

Iout ∝ |Aout |2 = Aout A∗out = 2Aout (x = 0)2 e−2αout L [1 + cos(∆φ)]. (A.21)

From the right-hand side of Equation (A.21), we can see that Iout is proportional to a cosine
function of the phase change of Equation (A.19). Because we desire high-speed modulation, we
want to modulate about a voltage that resides in the linear region of this cosine function; this is
done to maximize the output intensity difference between an applied voltage of VRF and an applied
voltage of zero. Since the linear region of a cosine wave has the highest magnitude slope of the
cosine function, the linear region is where we will obtain the greatest high/low output intensity

154
difference. Therefore, we can make the assumption that ∆φ is small about π/2 radians (where π/2
is the halfway point between the maximum and minimum values of the cos(∆φ) function), and
cos(∆φ) ≈ π/2 − ∆φ.
We now have the relationship:

Iout ∝ (1 + π/2 − ∆φ). (A.22)

To relate Equation (A.22) to the optical bandwidth, we first need to define the bandwidth in terms
of the output intensity. For an EOM, we would like to know how Iout changes as VRF varies between
its high and low values at different electrical frequencies. Mathematically, we can write this as:

∆Iout |RF
change in optical output intensity ∆VRF
normalized EO response = = ∆Iout |dc
. (A.23)
change in RF voltage
∆Vdc

Here we normalize our EO response to the dc case to determine how our output degrades, as a
function of frequency, from the best-case scenario of sub-Hz frequency (which is what we are
calling dc). In Equation (A.23), ∆VRF and ∆Vdc are identical, since the value of the high and low
voltages are the same regardless of the electrical frequency at which the device is being operated.
We can use Equation (A.22) to say:

∆Iout |RF = (1 + π/2 − ∆φhigh )|RF − (1 + π/2 − ∆φlow )|RF = −(∆φhigh − ∆φlow )|RF = −∆φ|RF
(A.24a)

∆Iout |dc = (1 + π/2 − ∆φhigh )|dc − (1 + π/2 − ∆φlow )|dc = −(∆φhigh − ∆φlow )|dc = −∆φ|dc .
(A.24b)
From Equation (A.19), we define ∆φ|dc as:

155
 2ωopt d33
∆φ|dc = LARF (x = 0), (A.25)
nopt c

And, taking the magnitude of the EO response, we obtain:

" #1/2
L
α L
− RF sinh2 ( αRF 2 ∆kL
2 ) + sin ( 2 )
|m( fRF )| = e 2
L 2
, (A.26)
( αRF ∆kL 2
2 ) +( 2 )

Where m( fRF ) is the electro-optic response of the device, and fRF is the RF electrical
frequency. By converting Equation (A.26) into dB, one can choose whether to use the optical
bandwidth definition (10log10 (m)) or the electrical bandwidth definition (20log10 (m)). From
Equation A.26, it is clear that two of the limiting factors of the bandwidth for a traveling-wave EOM
structure in a MZM configuration are the optical-electrical index mismatch and the electrical losses.

Bibliography
[1] Yariv A, Yeh P. Photonics: Optical Electronics in Modern Communications. Oxford University
Press; 2007. Print.

[2] D Janner MB D Tulli, Pruneri V. Waveguide Electro-Optic Modulation in Micro-Engineered

LiNbO3 . Journal of Optics A: Pure Applied Optics. 2008;10(104003).

[3] Haeyang Chung WSCC, Adler EL. Modeling and Optimization of Traveling-Wave LiNbO3
Interferometric Modulators. IEEE Journal of Quantum Electronics. 1991;27(3):608–617.

[4] Donnelly JP, Gopinath A. A Comparison of Power Requirements of Traveling-Wave LiNbO3

Optical Couplers and Interferometric Modulators. IEEE Journal of Quantum Electronics.
1987;QE-23(1):30–41.

[5] Boyd RW. Nonlinear Optics, 3rd ed. Academic Press; 2008. Print.

[6] Sutherland. Handbook of Nonlinear Optics. Marcel Dekker, Inc.; 2003. Print.

156