This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 128.210.126.199
This content was downloaded on 12/07/2020 at 04:31

Please note that terms and conditions apply.

You may also be interested in:

Organic Lasers and Organic Photonics: Organic photonics

F J Duarte

V International Conference of Photonics and Information Optics

NN Evtikhiev

3rd International School and Conference “Saint Petersburg OPEN 2016” on Optoelectronics, Photonics,
Engineering and Nanostructures

Special issue on integrated quantum photonics

A N Vamivakas, A Sergienko and A Fiore

New $100m photonics institute opens

Michael Banks

Illuminating the future of silicon photonics: optical coupling of carbon nanotubes to microrings
Y K Kato

Corning injects life into BT photonics lab

Matin Durrani

Vertical Cavity Surface Emitting Lasers Photonics

Kenichi Iga

Oxygen ion-implanted optical channel waveguides in Nd:MgO:LiNbO3

Feng Chen, Yang Tan, Lei Wang et al.
Silicon Photonics
Electromagnetic theory
 Silicon Photonics
Electromagnetic theory

Wouter J Westerveld
IMEC, Belgium

H Paul Urbach
Delft University of Technology, The Netherlands

IOP Publishing, Bristol, UK

ª IOP Publishing Ltd 2017

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system
or transmitted in any form or by any means, electronic, mechanical, photocopying, recording
or otherwise, without the prior permission of the publisher, or as expressly permitted by law or
under terms agreed with the appropriate rights organization. Multiple copying is permitted in
accordance with the terms of licences issued by the Copyright Licensing Agency, the Copyright
Clearance Centre and other reproduction rights organisations.

Cover image: Silicon photonic waveguide with bends. Artist impression based on a microscope
photograph. The thin white line is the 220 nm high by 400 nm wide rectangular silicon waveguide
core. This waveguide core is surrounded by 2 µm wide trenches, depicted in cyan, where the silicon
layer has been etched away and the buried oxide layer is visible. This waveguide was used to
measure waveguide propagation loss. Adapted from figure 3.5b of the Ph.D. Thesis of Wouter J.
Westerveld, Silicon photonic micro-ring resonators to sense strain and ultrasound (2014), available
online at http://repository.tudelft.nl.

Permission to make use of IOP Publishing content other than as set out above may be sought
at [email protected].

Wouter J Westerveld and H Paul Urbach have asserted their right to be identified as the authors of
this work in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.

ISBN 978-0-7503-1386-5 (ebook)

ISBN 978-0-7503-1387-2 (print)
ISBN 978-0-7503-1388-9 (mobi)

DOI 10.1088/978-0-7503-1386-5

Version: 20171201

IOP Expanding Physics

ISSN 2053-2563 (online)
ISSN 2054-7315 (print)

British Library Cataloguing-in-Publication Data: A catalogue record for this book is available
from the British Library.

Published by IOP Publishing, wholly owned by The Institute of Physics, London

IOP Publishing, Temple Circus, Temple Way, Bristol, BS1 6HG, UK

US Office: IOP Publishing, Inc., 190 North Independence Mall West, Suite 601, Philadelphia,
PA 19106, USA
 Contents

Preface viii
Further reading x
Author biographies xii
List of symbols xiii

1 Introduction 1-1
1.1 Photonics 1-1
1.2 Integrated photonics 1-1
1.3 Silicon photonics 1-2
1.4 Silicon-on-insulator technology in this book 1-3
References 1-3

2 Waveguides 2-1
2.1 Maxwell’s equations for linear, passive materials 2-1
2.2 Dielectric slab waveguides 2-2
2.2.1 TE modes 2-4
2.2.2 TM modes 2-5
2.3 General properties of modes in dielectric waveguides 2-7
2.3.1 Guided waves described by Ex, Ey, Hx and Hy 2-7
2.3.2 Guided waves described by Ez and Hz 2-9
2.3.3 Orthogonality 2-11
2.3.4 Power flow 2-13
2.3.5 Mode normalization 2-14
2.3.6 Mode expansion conjecture 2-14
2.3.7 Waveguides with losses and bends 2-16
2.4 Rectangular Si waveguides: extension of Marcatili’s approach 2-18
2.4.1 Ansatz for the shape of the field 2-18
2.4.2 Boundary conditions 2-20
2.4.3 Approximate methods 2-22
2.4.4 Avoided crossing of modes with similar propagation constants 2-26
2.4.5 Dispersion: effective group index 2-30
2.5 Rigorous numerical mode-solvers 2-32
2.6 Typical silicon-on-insulator waveguides 2-33
References 2-35

v
 Silicon Photonics

3 Components 3-1
3.1 Directional couplers 3-1
3.1.1 Coupled mode theory 3-1
3.1.2 Eigenmode expansion (EME) 3-3
3.1.3 Rigorous FDTD simulations 3-5
3.2 Multimode interference (MMI) couplers 3-8
3.2.1 Approximate description of the guided modes 3-8
3.2.2 Multimode propagation and interference 3-11
3.2.3 Imaging and power splitting 3-13
3.2.4 A 2 × 2 MMI coupler 3-15
3.2.5 Comparison with rigorous numerical simulations 3-17
3.3 Fibre to chip coupling 3-19
3.3.1 Inverted optical taper (fibre to chip coupler) 3-22
3.3.2 Basic out-of-plane grating couplers 3-23
3.3.3 Advanced grating couplers 3-28
3.4 Ring and racetrack resonators 3-30
3.4.1 Silicon ring resonators with directional couplers 3-32
3.4.2 Ring resonator resonances 3-32
3.4.3 Ring resonators with a non-uniform waveguide 3-35
3.5 Mach–Zehnder interferometers 3-36
References 3-39

4 Computational methods 4-1

4.1 Finite difference time domain (FDTD) 4-1
4.2 The eigenmode expansion method 4-6
4.2.1 Properties of unidirectional modes 4-7
4.2.2 Number of modes 4-7
4.2.3 Mode matching 4-8
4.2.4 The T-matrix scheme 4-10
4.2.5 The S-matrix scheme 4-11
4.2.6 Examples: MMI coupler and grating coupler 4-13
References 4-15

5 Devices 5-1
5.1 Ring resonator sensors 5-2
5.1.1 Ring resonators 5-2
5.1.2 Temperature sensors 5-3

vi
 Silicon Photonics

5.1.3 Sensor systems and interrogation 5-4

5.1.4 Influence of an external effect on rectangular silicon waveguides 5-4
5.1.5 Biosensors 5-6
5.1.6 Ultrasound sensors 5-7
5.2 Silicon photonic modulators based on carrier-depletion in a MZI 5-11
5.2.1 Modulation formats: on-off keying 5-12
5.2.2 Electro-optical conversion: pn-junction carrier depletion 5-12
5.2.3 Phase to amplitude: Mach–Zehnder inferometer 5-15
5.2.4 A 40 GBit/s optical modulator 5-16
5.3 Fabrication technology and design of optical circuits 5-17
5.3.1 Example: design of racetrack for ultrasound sensor 5-20
References 5-21

Appendices

A Coupled mode theory of parallel waveguides A-1

B Lossless transmission matrix B-1

C Semiconductor physics of pn-junctions C-1

vii
 Preface

With integrated optics different optical components are integrated in a single

photonic integrated circuit. Light is guided through waveguides and components
all on a planar chip. With current photonic chips, light can be generated, modulated,
processed and detected. These functions are used for high-speed optical fibre
communication and for accurate photonic sensing. Silicon photonics has become
one of the most prominent platforms for integrated optics, primarily because silicon
chips are made at very low cost with the same CMOS fabrication technologies with
which micro-electronic integrated circuits (ICs) are made. Furthermore, electronic
and photonic circuits can be combined on the same chip, and the high contrast in
refractive index between the light-guiding silicon and the silicon-dioxide surrounding
enables circuits with a small footprint.
This book is a concise treatment of fundamental electromagnetic theories of
integrated optics with a focus on silicon photonics. Only basic knowledge of
electromagnetic theory is needed. The book aims to introduce graduate and
advanced undergraduate students to silicon photonics and bring them to a level of
knowledge that enables them to understand the scientific literature. This book covers
a relevant selection of theories and the reader is referred to additional material where
appropriate. It is hoped that certain parts are also useful for experienced researchers.
After a brief introduction to silicon photonics, general properties of modes in
dielectric waveguides, are derived in chapter 2. The guided mode problem is
formulated as an eigenvalue problem for a differential operator for the transverse
components of the electromagnetic field with the propagation constant as eigen-
value. The sense in which the modes are orthogonal is clarified. Then Marcatili’s
well-known approximate method to compute modes in rectangular waveguides is
extended to the case of waveguides with high contrast which are typical for silicon
photonics. A major theme of the book is to evaluate and adapt theories developed
previously in integrated optics for low index contrast waveguides such as Marcatili’s
method, the coupled mode theory of Hardy and Streifer and the description of multi-
mode interference by Soldano and Pennings, to high index contrast waveguides of
silicon photonics.
In chapter 3, components such as directional couplers, multimode interference
(MMI) couplers, out-of-plane grating couplers, ring resonators and Mach–Zehnder
interferometers are treated analytically with coupled mode theory and eigenmode
expansions. In chapter 4 rigorous numerical methods such as the finite difference
time domain (FDTD) method and the bi-directional eigenmode expansion (EME)
method are explained. In chapter 5, three devices are described in detail namely, a
biomedical evanescent field sensor, an optical micro-machined ultrasound sensor
and a high-speed optical modulator.
The book is based on the PhD thesis of the first author, of which the research was
funded by the IOP Photonic Devices programme of RVO and by TNO, whom we
gratefully acknowledge. We also gratefully acknowledge the support of Professor
Ronald Hanson.

viii
 Silicon Photonics

We are grateful to Dr Jose Pozo, Dr Mirvais Yousefi, and Dr Suzanne Leinders

for many stimulating discussions and their feedback, especially on the parts that
appeared in the PhD Thesis. We thank Professors Peter Bientsman, Graham T
Reed, and Andrea Melloni for their encouragement. We gratefully acknowledge
Professor Dominique Gallagher and Mr Vincent Brulis of Photon Design for their
help and for the examples in chapter 4. We thank Ms Jessica Fricchione, Mr Daniel
Heatley and Mr Chris Benson of IOP Publishing for their help in completing the
book.
Most importantly, we thank our beloved families for their encouragement,
support, and patience.
Wouter J Westerveld
H Paul Urbach

ix
 Further reading

• Experimental evidence of theories presented in this book can be found in [1,

chapter 3] and in [2, section 2.1], which both include a systematic exper-
imental study of the waveguides, directional couplers, and ring resonators.
• Good textbooks on theory of dielectric waveguides are, e.g. Theory of
Dielectrical Optical Waveguides by Marcuse [3] and Optical Waveguide
Theory by Sneyder and Love [4].
• Good textbooks on integrated optics include Photonics: optical electronics in
modern communications by Yariv and Yeh [5], Fundamentals of Photonics
(2nd edn) by Saleh and Teich, and Photonic Devices by Liu [6]. These books
also contain theories and treatments of components and devices that are not
discussed in the present book, e.g. distributed bragg reflectors (DBRs) and
cascaded ring resonators.
• The textbook Silicon Photonics: An Introduction by Reed and Knights [7] is a
good introduction to silicon photonics including the history of silicon
photonics and fabrication technology. This work contains also a discussion
of arrayed waveguide gratings (AWGs).
• Advances in the field of silicon photonics including fabrication technology,
waveguides and loss mechanisms, packaging, state-of-the-art devices, etc, are
described in the advanced textbooks [8–12]. The review papers by Bogaerts
et al [13] and by Feng et al [14] both focus on silicon photonic micro-ring
resonators but also cover many of the aforementioned topcis. Additionally,
progress in the field of silicon photonics may be found in two special issues of
IEEE’s Journal of Quantum Electronics [15, 16].
• The thesis of De Vos is recommended as introduction to silicon photonic
biosensors [2]. It includes an overview of biosensors, details of the multi-
disciplinary design of a silicon photonic biosensor, an introduction to
microfluidics and a chapter entitled (Bio)Chemistry for engineers. The state
of the art of photonic silicon-on-insulator micro-ring based biosensors
including some exciting biomedical applications can be found in the review
by Kindt and Bailey [17]. A recent development is enhanced Raman
specroscopy using silicon-nitride waveguides, see e.g. [18] for an introduction
and [19] for a state-of-the-art device.
• The reviews by Reed et al [20, 21] are excellent introductions to silicon
photonic modulators. More details and additional devices can be found in the
handbook edited by Vivien and Pavesi [12, chapter 9].

Bibliography
[1] Westerveld W J 2014 Silicon photonic micro-ring resonators to sense strain and ultrasound
PhD Dissertation Technische Universiteit Delft, Delft http://repository.tudelft.nl
[2] de Vos K 2010 Label-free silicon photonics biosensor platform with microring resonators
PhD Dissertation Universiteit Gent, Belgium.

x
 Silicon Photonics

[3] Marcuse D 1991 Theory of Dielectrical Optical Waveguides 2nd edn (San Diego, CA:
Academic)
[4] Sneyder A W and Love J D 1983 Optical Waveguide Theory (London: Chapman and Hall)
[5] Yariv A and Yeh P 2007 Photonics: Optical Electronics in Modern Communications
(Oxford Series in Electrical and Computer Engineering) (Oxford: Oxford University Press)
[6] Liu J-M 2005 Photonic Devices (Cambridge: Cambridge University Press)
[7] Reed G T and Knights A P 2004 Silicon Photonics: An Introduction (Chichester: Wiley)
[8] Reed G T (ed) 2008 Silicon Photonics: The state of the art (Chichester: Wiley)
[9] Pavesi L and Lockwood D J (ed) 2004 Silicon Photonics III (Berlin: Springer) doi:10.1007/
b11504
[10] Pavesi L and Lockwood D J (ed) 2011 Silicon Photonics III: Components and Integration
(Berlin: Springer) doi:10.1007/978-3-642-10506-7
[11] Pavesi L and Lockwood D J (ed) 2016 Silicon Photonics III: Systems and Applications
(Berlin: Springer)
[12] Vivien L and Pavesi L (ed) 2013 Handbook of Silicon Photonics (Boca Raton, FL: CRC
Press) doi:10.1201/b14668
[13] Bogaerts W, De Heyn P, Van Vaerenbergh T, De Vos K, Kumar Selvaraja S, Claes T,
Dumon P, Bienstman P, Van Thourhout D and Baets R 2012 Silicon microring resonators
Laser Photon. Rev. 6 47–73
[14] Feng S, Lei T, Chen H, Cai H, Luo X and Poon A 2012 Silicon photonics: from a
microresonator perspective Laser Photon. Rev. 6 145–77
[15] Vivien L, Cheben P, Lo Guo-Qiang P, Pavesi L and Zhou Z Introduction to the issue on
silicon photonics IEEE J. Sel. Top. Quantum Electron. 20 4
[16] GuoQiang P L, Chen Y, Poon A W O, Nakamura T and Chu T 2016 Introduction to the
issue on silicon photonics IEEE J. Sel. Top. Quantum Electron. 22 4
[17] Kindt J T and Bailey R C 2013 Biomolecular analysis with microring resonators:
applications in multiplexed diagnostics and interaction screening Curr. Opin. Chem. Biol.
17 818–26
[18] Baets R, Subramanian A Z, Dhakal A, Selvaraja S K, Komorowska K, Peyskens F,
Ryckeboer E, Yebo N, Roelkens G and Thomas N Le 2013 Spectroscopy-on-chip
applications of silicon photonics Proc. SPIE 8627 86270I
[19] Dhakal A, Wuytens P C, Peyskens F, Jans K, Thomas N L and Baets R 2016 Nanophotonic
waveguide enhanced raman spectroscopy of biological submonolayers ACS Photon. 3 2141–9
[20] Reed G T, Mashanovich G, Gardes F Y and Thomson D J 2010 Silicon optical modulators
Nat. Photon. 4 518–26
[21] Reed G T, Mashanovich G Z, Gardes F Y, Nedeljkovic M, Hu Y, Thomson D J, Li K,
Wilson P R, Chen S-W and Hsu S S 2013 Recent breakthroughs in carrier depletion based
silicon optical modulators Nanophotonics 3 229–45

xi
 Authors biographies

Dr Wouter J Westerveld
Postdoctoral researcher at IMEC. Wouter J Westerveld received the
MSc degree in Applied Physics (cum laude, 2009) and the PhD
degree (cum laude, 2014) from Delft University of Technology, The
Netherlands. His multidiciplinary PhD research was on silicon
integrated optomechanical sensors to sense strain and ultrasound.
He worked as laser scientist at Mach8 Lasers, a startup (2014). He
was ultrasound engineer at Applus RTD, a global provider of non-
destructive testing services based in Rotterdam (2014–2016). He worked as post-
doctoral researcher at QuTech/Delft University of Technology on quantum control
and optical read-out of spins in diamond (2017). He currently works as postdoctoral
researcher at IMEC (Leuven, Belgium) on integrated optical ultrasound sensors.
He is interested in applied physics in general with special interest in silicon
photonics, integrated optomechanical sensors, and ultrasonic imaging.
Dr Ir Westerveld was secretary (2011) and chairman (2012) of the Student Board
of the IEEE Photonics Society Benelux chapter.

Professor H Paul Urbach

Head of Optics Research Group, Faculty of Applied Sciences, Delft
University of Technology. Paul Urbach graduated in 1981 from the
University of Groningen in The Netherlands, where he received a
PhD in 1986 on optimization of hydrodynamic propulsion. He
joined Philips Research Laboratory in Eindhoven, The Netherlands,
in 1986 and became a Principal Scientist in 1994. He became part-
time Professor Diffraction Optics at Delft University of Technology
in 2000. Since January 2008, he has been a Full Professor and the Head of the Optics
Research Group of this university.
Professor Urbach has done research on many branches of optics, such as optical
lithography for ICs, inverse problems in x-ray fluorescence, electromagnetic modeling
of of optical scattering and wave guide problems, manipulation of spontaneous
emission and nano-optics.
Professor Urbach was elected to the Board of the European Optical Society
(EOS) in 2008 where he served as President for several years.

xii
 List of symbols

∇ Nabla operator, ∇ = xˆ∂ ∂x + yˆ∂ ∂y + zˆ∂ ∂z

∇β Nabla operator for waveguide modes, see equation (2.36)
* complex conjugate
()−1 inverse
∣i 〉 mode with label i in bra-ket notation
〈i∣j 〉 scalar product of modes i and j in bra-ket notation, see equation (2.95)
Oˆ ∣i 〉 Operation notation of eigenvalue problem, see equation (2.94) and equation (2.37)
〈Ei∣Hj 〉 scalar product between modes used section 4.2, defined by equation (4.18)
α amplitude transmittance over a waveguide section with a certain length.
αp amplitude propagation loss or absorbtion coefficient, see equation (2.66).
Δα change in intensity absorbtion coefficient used in section (5.2), see equation (5.16)
β modal propagation constant, see equation (2.30)
δ various usages, in the description of the directional coupler the difference between
the corrected propagation constants of the two waveguides of the coupler, see
equation (3.4)
ε permittivity
Δεa(x , y ) difference between the permittivity profile of the coupler and permittivity profile of
single waveguide a, see equation (A.1)
η used in the description of modes of slab waveguides, equation (2.9), and rectangular
waveguides, figure 2.6.
ηoverlap coupling efficiency due to overlap of two TE-modes, see equation (3.56)
{ imaginary unit, satisfying the equation ı2 = 1
κ cross transmission of a coupler, complex amplitude coefficient, see equation (3.3) for
directional couplers, figure 3.18b for use in a ring resonator, and figure 3.21a for
use in a Mach–Zehnder interferometer
κaa, κbb correction of propagation constants in the description of the directional couplers, see
equation (3.2)–(3.4) and equation (A.37).
κab, κba direct coupling between two waveguides in a directional coupler, see equation (3.2)–
(3.5) and equation (A.37)
κ̂ab see equation (A.26)
κ∼νμ
a see equation (A.22)
γ2, γ3 used in the description of modes of slab waveguides, equation (2.20), and
rectangular waveguides, figure 2.6
γ4, γ5 used in the description of modes of slab waveguides, equation (2.9), and rectangular
waveguides, figure 2.6
Λ grating period
λ vacuum wavelength
λc resonance wavelength prior to an externally induced resonance shift
λl laser wavelength
λm wavelength of resonance number m, see equation (3.70)
ΔλFWHM resonance full-width at half-maximum, see equation (3.74)

xiii
 Silicon Photonics

ΔλFSR free-spectral-range, see equation (3.71) for ring resonators and equation (3.91) for
Mach–Zehnder interferometers
μ permeability (sometimes also used as index)
μ0 vacuum permeability
ξ used in the description of modes of slab waveguides, equation (2.20), and
rectangular waveguides, figure 2.6.
τ straight-through transmission of a coupler, complex amplitude coefficient, see
equation (3.2) for directional couplers, figure 3.21b for use in a ring resonator,
and figure 3.21(a) for use in a Mach–Zehnder interferometer
χ (i ) linear and nonlinear susceptibilities
χ external physical parameter that influences the properties of a waveguide, used in
section 5.1.1, see equation (5.1)
Ψ two-dimentional description of the electric field in the waveguide, see equation (3.27)
Ω angular velocity
ω positive angular frequency
A0 amplitude
A1−10 amplitude coefficient used in the description of modes rectangular waveguides,
figure 2.6.
B physical magnetic field, B = μH
b waveguide dimension, waveguide label, mode label, or mode coefficient
b vector with modal amplitudes of backward propagating modes, used in section 4.2,
used in section 4.2
c velocity of light in vacuum
D electric displacement
d waveguide dimension
diag(e−iβνl ) diagonal matrices containing the propagation constants βν of each eigenmode ν of a
section l the length of this section
E electric field
Ec electric field in a directional coupler
E modal electric field, see equation (2.30)
Et Transverse components of modal electric field, Ex and Ey
ExSLAB(y ) profile of a waveguide mode along the y-direction, used in the description of MMI
couplers, see equation (3.25)
e Euler’s number e ≈ 2.7 or elementary charge e ≈ 1.6 · 10−19 coulombs
F functional to solve for propagation constant β of TE-modes in slab waveguides, see
equation (2.17)
f vector with modal amplitudes of forward propagating modes, used in section 4.2
G functional to solve for propagation constant β of TM-modes in slab waveguides, see
equation (2.29)
H magnetic field
H modal magnetic field, see equation (2.30)
Ht transverse components of modal magnetic field, Hx and Hy
I optical power
J volume current density
K K = n 2k 2 − β2 , see equation (2.45)

xiv
 Silicon Photonics

k wavenumber in vacuum, k = ω c
kx used in the description of modes of slab waveguides, equation (2.20), and
rectangular waveguides, figure 2.6
ky used in the description of modes of slab waveguides, equation (2.9), and rectangular
waveguides, figure 2.6
L length, e.g. length parallel waveguides in a directional coupler, or length of the wide
waveguide in an MMI coupler
∼ ∼
L effective length of directioal coupler L =L + ΔL with ΔL a term that takes coupling
in the bends of the coupler into account
Lπ beat length of the fundamental mode and the first mode of a waveguide,
πLπ = (β0 − β1)
ΔNe concentration of free electrons in a semiconductor material, used in section 5.2 and
appendix C
ΔNh concentration of free holes in a semiconductor material, used in section 5.2 and
appendix C
n material refractive index
ne effective refractive index of a waveguide mode, see equation (2.31)
ng effective group index of a waveguide mode, see equation (2.32)
nr effective refractive index of slab waveguide, see equation (3.18)
n̂ unit vector perpendicular to a defined plane
Δn change in refractive index used in section 5.2, see equation (5.15)
Ô operator describing Maxwell’s equations for waveguide modes as eigenvalue
problem with the propagation constant β as eigenvalue, see equation (2.37)
P time-averaged power, or, in section 5.1.6, pressure
pνμ projection of mode μ of waveguide b onto mode ν of waveguide a, see equation
(A.13)
Q charge density
R waveguide bend radius
R matrix of interface reflection coefficients, see section 4.2.3
r exticion ratio, see equation (3.73), or, in section 2.3.7, radial distance as shown in
figure 2.4
S Poynting vector, see equation (2.59)
∼
S S-matrix or scattering matrix, see equation (4.42)
Sρ local strain in the direction of a waveguide, see equation (5.12)
s coupling coefficient, see equation (3.5)
T (λ ) transmission spectrum
T matrix of interface transmission coefficients, see section 4.2.3
∼
T T-matrix, see equation (4.35)
t time
Umm measure of error in the extended Marcatili method associated with discontinuities of
the electromagnetic field at interfaces, see equation (2.91)
ΔU measure to compare two electromagnetic fields, relative energy of the difference of
the fields per unit of length in the propagation direction, see equation (2.92)
V potential
V0 built-in potential difference across a pn-punction, see appendix C
Vd applied potential difference across a pn-punction, see appendix C

xv
 Silicon Photonics

w waveguide width
wp, wn with of the carrier depletion aria in the p-type material and n-type material, of a pn-
junction respectively, see appendix C
x̂ unit vector in the x-direction
ŷ unit vector in the y-direction
ẑ unit vector in the z-direction

xvi
 IOP Publishing

Silicon Photonics
Electromagnetic theory
Wouter J Westerveld and H Paul Urbach

Chapter 1
Introduction

1.1 Photonics
In photonics, light is used as a carrier of information similar to electrons in
electronics. Photonics began with the invention of the laser in 1960. Other develop-
ments followed including: the laser diode, optical fibres for transmitting information
and the erbium-doped fibre amplifier. These inventions formed the basis for the
telecommunications revolution of the late 20th century and provided the infra-
structure for the internet [1]. Next to telecommunications, another major application
of photonics is sensing, for example in the medical, aviation, and construction
industries. For these applications, a number of optical components are required such
as laser sources, modulators to encode the light with an (electronic) signal, detectors,
multiplexers, and signal-splitters.

1.2 Integrated photonics

The concept of integrated optics, already proposed in the late 1960s, is to fabricate
the aforementioned optical components in on-chip integrated circuits [2]. A photonic
integrated circuit usually consists of waveguides to confine and guide light, similar to
what optical fibres do, but in the planar surface of a chip. The minimum size of a
single-mode waveguide depends strongly on the refractive index contrast, i.e., the
difference in refractive index between the core of the waveguide and its cladding. For
example, optical fibres have a low index contrast (∼0.05) with a core of ∼8.2 μm [3],
whereas silicon waveguides with a SiO2 cladding have a high index contrast (∼2) and
core of ∼0.5 μm. In the early days of integrated optics, only low index contrast
waveguides could be fabricated. Development of integrated photonic components
has always gone hand-in-hand with improvements of lithographic fabrication
processes. The accurate fabrication of waveguides with size of the order of the
wavelength sets high demands on the micro-fabrication processes. Over recent
decades, a number of more-or-less standardized platforms have been developed,

doi:10.1088/978-0-7503-1386-5ch1 1-1 ª IOP Publishing Ltd 2017

 Silicon Photonics

comprising a set of material combinations, fabrication processes, and typical optical

components. One of these is silicon photonics.

1.3 Silicon photonics

One of the most promising platforms for integrated photonics is the silicon-on-
insulator platform, often referred to as silicon photonics. Silicon photonic
components were first reported in the mid 1980s [4] and silicon-on-insulator
waveguides were introduced a few years later [5, 6]. Photonics has been driven by
the telecommunication industry and the vision of a ‘superchip’ combining elec-
tronics and photonics of the early 1990s has inspired a lot of research since then
[7, 8]. Large companies such as IBM, Intel, Teraxion and Luxtera have launched
products in silicon photonic technology, mostly optical interconnects for high-speed
communication [9–14]. Genalyte, a San Diego based startup, produces silicon-
photonic based biosensors for a number of medical diagnostic tests [15, 16]. Indigo,
a Belgium based startup, is developing a needle-free glucose sensor [17]. On the other
hand, academic researchers of quantum photonic applications, who need stringent
single photon control, are also exploring the silicon photonics platform [18, 19].
A major advantage of this platform is that photonic integrated circuits can be
fabricated with the existing CMOS infrastructure of the semiconductor industry,
enabling low cost and reliable mass-fabrication. Moreover, silicon is a strong
material and its high refractive index contrast makes very small devices possible.
Silicon is a passive material with no direct semiconductor band gap around the
wavelength that is used for telecommunication (1550 nm free-space), hence it cannot
be used to electronically detect or generate light. Therefore germanium and silicon–
germanium alloys, which are compatible with CMOS fabrication technology, are
added to provide this functionality [20–24]. Alternatively, III–V materials can be
integrated into silicon photonic integrated circuits [25–33].
Besides silicon photonics, other platforms for integrated photonics exist, each
with specific advantages, e.g.: indium phosphide (including both passive components
as well as lasers and detectors), silicon dioxide (very low loss passive circuits), and
silicon nitride (visible wavelengths) [34–36]. It should be mentioned that certain
discrete optical components outperform integrated circuits, for example diode lasers
of III–V materials (indium, aluminium, galium, arsenide, phosphide, etc) and
modulators of lithium niobate [37, 38].
As mentioned above, silicon photonic circuits can be fabricated with CMOS
technology of the electronics industry. Although the typical critical dimensions of
silicon waveguides (100–500 nm) are fairly large compared to what is fabricated with
high-end CMOS tools, the requirements on the accuracy of the dimensions for good
optical performance are stringent and in the nanometer range. Moreover, the optical
components may consist of diverse features ranging from isolated waveguides to
dense arrays of holes. Existing CMOS processes have been tailored to these demands
of integrated photonics, and a silicon photonics platform has been offered for over a
decade by Imec (Leuven, Belgium) and CEA-LETI (Grenoble, France) in the

1-2
 Silicon Photonics

ePIXfab consortium [39]. Wafer-scale fabrication in (semi) industrial fabrication

plants (fabs) is now offered by various parties1.

1.4 Silicon-on-insulator technology in this book

Telecommunication most often uses light with a free-space wavelength around 1.55 μm
because the losses of optical fibres are then smallest. Many low cost devices such as
sources, detectors, fibres, etc have been developed for optical communication.
Therefore, the telecommunication wavelength is also attractive for other applications
such as sensing. This book focusses on devices that operate at the wavelength of 1.55 μm.
The most common type of waveguide on this platform is a rectangular silicon
waveguide on a SiO2 buried oxide (BOX) substrate with an air or a SiO2 cladding. It
has typically a height of 220 nm and a single-mode guide has a typical width of 450
nm. This type of waveguide is studied extensively in this book, although many of the
treated theories can be applied to other types of waveguides as well. For other
technologies such as rib waveguides we refer to references [43, 44].
General properties of modes in dielectric waveguide and an approximate
analytical description of modes in rectangular silicon waveguides are derived in
chapter 2. A number of optical components, such as directional couplers, multimode
interference (MMI) couplers, out-of-plane grating couplers, ring resonators and
Mach–Zehnder interferometers are treated analytically with a coupled mode theory
and eigenmode expansions in chapter 3. Two important rigorous numerical methods
are explained in chapter 4 and three illustrative devices are described in detail in
chapter 5. The treatment in this book is restricted to passive components and sources
and detectors are not considered.

References
[1] Wikipedia 2013 Photonics http://en.wikipedia.org/wiki/photonics
[2] Alferness R C 1997 Integrated optics: technology and system applications converge Opt.
Photon. News 8 16
[3] Corning Incorporated 2002 Corning® SMF-28™ optical fibre: Product information
[4] Soref R and Lorenzo J 1985 Single-crystal silicon: a new material for 1.3 and 1.6 μm
integrated-optical components Electron. Lett. 21 953–4
[5] Kurdi B N and Hall D G 1988 Optical waveguides in oxygen-implanted buried-oxide silicon-
on-insulator structures Opt. Lett. 13 175–7
[6] Cortesi E, Namavar F and Soref R A 1989 Novel silicon-on-insulator structures for silicon
waveguides SOS/SOI Technology Conf. 1989 IEEE p 109
[7] Abstreiter G 1992 Engineering the future of electronics Phys. World 5 36–9
[8] Soref R A 1993 Silicon-based optoelectronics Proc. IEEE 81 1687–706

1
The EU-funded ePIXfab consortium offers silicon photonic device fabrication at Imec (Leuven, Belgium),
CEA-LETI (Grenoble, France) or IHP (Frankfurt an der Oder, Germany) [39, 40]. A similar service is offered
by the OpSIS project based at the University of Washington in Seattle and the University of Delaware in
Newark, a consortium including the Institute of Microelectronics (IME, Singapore), Luxtera (headquarters in
Carlsbad, California), and STMicroelectronics (headquarters in Geneva, Switzerland) [41, 42].

1-3
 Silicon Photonics

[9] Vlasov Y 2015 Silicon integrated nanophotonics: from fundamental science to manufactur-
able technology (Presented at SPIE Photonics West, San Francisco CA) http://dx.doi.org/
10.1117/2.3201503.15
[10] Christy P 2016 Intels silicon photonics products could change the world of IT www.intel.
com/content/www/us/en/architecture-and-technology/silicon-photonics/451-research-sil
icon-photonics-paper.html
[11] Liao L 2017 Intel silicon photonics: From research to product (presented at meeting of the
IEEE CPMT Chapter San Francisco Bay Area, Santa Clara CA) http://ewh.ieee.org/soc/
cpmt/presentations/cpmt1703a.pdf
[12] Painchaud Y et al 2014 Silicon-based products and solutions Proc. SPIE 8988
[13] Luxtera 2016 Luxtera ships one millionth silicon photonic transceiver product (presented at
the XLII European Conference on Opical Communication, ECOC 2016, Dusseldorf,
Germany) www.luxtera.com/press-releases
[14] Luxtera 2017 Luxtera ships Industry’s first 2x100G PSM4 silicon photonics embedded
optical modules (presented at the Optical Fiber Communication Conference, OFC 2017, Los
Angeles, USA) www.luxtera.com/press-releases
[15] Genalyte 2017 Silicon photonics biosensor technology www.genalyte.com/about-us/our-
technology/
[16] Chen C 2017 Silicon valley cash is still chasing blood despite theranos bust (Bloomberg)
www.bloomberg.com/news/articles/2017-01-06/silicon-valley-cash-is-still-chasing-blood-despite-
theranos-bust
[17] 2016 Belgian startup lands EUR 7M for optical glucose monitoring chip SPIE/Optics.org
http://optics.org/news/7/12/30
[18] Pavesi L and Lockwood D J (ed) 2016 Silicon Photonics III: Systems and Applications
(Berlin: Springer)
[19] Silverstone J W, Bonneau D, O’Brien J L and Thompson M G 2016 Silicon quantum
photonics IEEE J. Sel. Top Quantum Electron. 22 390–402
[20] Assefa S, Xia F, Green W M J, Schow C, Rylyakov A and Vlasov Y 2010 Cmos-
integrated optical receivers for on-chip interconnects IEEE J. Sel. Top. Quantum
Electron. 16 1376–85
[21] Assefa S, Xia F and Vlasov Y A 2010 Reinventing germanium avalanche photodetector for
nanophotonic on-chip optical interconnects Nature 464 80–5
[22] Michel J, Liu J and Kimerling L C 2010 High-performance Ge-on-Si photodetectors Nat.
Photon. 4 527–34
[23] Liang D and Bowers J E 2010 Recent progress in lasers on silicon Nat. Photon. 4 511–7
[24] Lockwood D and Tsybeskov L 2014 Fast light-emitting silicon-germanium nanostructures
IEEE J. Sel. Top. Quantum Electron. 20 225–31
[25] Roelkens G, Liu L, Liang D, Jones R, Fang A, Koch B and Bowers J 2010 III-V/silicon
photonics for on-chip and intra-chip optical interconnects Laser Photon. Rev. 4 751–79
[26] Heck M, Bauters J, Davenport M, Doylend J, Jain S, Kurczveil G, Srinivasan S, Tang Y and
Bowers J 2013 Hybrid silicon photonic integrated circuit technology IEEE J. Sel. Top.
Quantum Electron. 19 6100117
[27] Duan G-H et al 2014 Hybrid III–V on silicon lasers for photonic integrated circuits on silicon
IEEE J. Sel. Top. Quantum Electron. 20 158–70
[28] Roelkens G et al 2015 III-V-on-silicon photonic devices for optical communication and
sensing Photonics 2 969–1004

1-4
 Silicon Photonics

[29] Davenport M L, Skendi S, Volet N, Hulme J C, Heck M J R and Bowers J E 2016

Heterogeneous silicon/III-V semiconductor optical amplifiers IEEE J. Sel. Top. Quantum
Electron. 22 78–88
[30] Zheng X et al 2016 III-V/Si hybrid laser arrays using back end of the line (BEOL) integration
IEEE J. Sel. Top. Quantum Electron. 22 204–17
[31] Duan G H et al 2016 Hybrid III-V silicon photonic integrated circuits for optical
communication applications IEEE J. Sel. Top. Quantum Electron. 22 379–89
[32] Luo X, Cheng Y, Song J, Liow T Y, Wang Q J and Yu M 2016 Wafer-scale dies-transfer
bonding technology for hybrid III/V-on-silicon photonic integrated circuit application IEEE
J. Sel. Topics Qunatum Electron. 22 8200612
[33] Bowers J E, Komljenovic T, Davenport M, Hulme J, Liu A Y, Santis C T, Spott A,
Srinivasan S, Stanton E J and Zhang C 2016 Recent advances in silicon photonic integrated
circuits Proc. SPIE 9774 977402
[34] Muñoz P 2012 Towards fabless photonic integration, VLC Photonics, whitepaper www.
vlcphotonics.com/wp-content/uploads/2014/12/Fabless_Integration_WP_v1.0.pdf
[35] Doerr C R 2013 Integrated photonic platforms for telecommunications: Inp and Si IEICE
Transac. Electron. E96.C 950–7
[36] Rahim A et al 2017 Expanding the silicon photonics portfolio with silicon nitride photonic
integrated circuits J. Lightwave Technol. 35 639–49
[37] Coldren L A, Corzine S W and Mashanovitch M L 2012 Diode Lasers and Photonic
Integrated Circuits (Hoboken, NJ: Wiley)
[38] Toney J E 2015 Lithium Niobate Photonics (Norwood, MA: Artech House)
[39] Dumon P, Bogaerts W, Tchelnokov A, Fedeli J-M and Baets R 2008 Silicon nanophotonics
Future Fab Int. 25 29–36
[40] Pozo J et al 2012 Essential: Epixfab services specifically targeting (sme) industrial takeup of
advanced silicon photonics Int. Conf. on Transparent Optical Networks (ICTON), 2012 XIV,
pp 1–3.
[41] Bourzac K 2012 Photonic chips made easier Nature 483 388
[42] Lim A-J, Song J, Fang Q, Li C, Tu X, Duan N, Chen K K, Tern R-C and Liow T-Y 2014
Review of silicon photonics foundry efforts IEEE J. Sel. Top. Quantum Electron. 20 405–16
[43] Reed G T (ed) 2008 Silicon Photonics: The state of the art (Chichester: Wiley)
[44] Xu D-X, Schmid J, Reed G, Mashanovich G, Thomson D, Nedeljkovic M, Chen X, van
Thourhout D, Keyvaninia S and Selvaraja S 2014 Silicon photonic integration platform–
have we found the sweet spot? IEEE J. Sel. Top. Quantum Electron. 20 189–205

1-5
 IOP Publishing

Silicon Photonics
Electromagnetic theory
Wouter J Westerveld and H Paul Urbach

Chapter 2
Waveguides

2.1 Maxwell’s equations for linear, passive materials

Maxwell’s equations describe the propagation of light in terms of the electric field E
and the magnetic field H. This section presents Maxwell’s equations in a convenient
form to describe passive components in silicon-on-insulator technology.
Typical passive optical components in silicon-on-insulator technology consist of
silicon (Si), silicon-dioxide (SiO2) and air (see figure 2.1 for refractive indices). These
materials have no direct band-gap around the telecom wavelength (1550 nm),
therefore, active components consist of different materials to generate and detect
(absorb) light. The active components are typically located far away (hundreds of
wavelengths) from the passive components. The passive components have no free
carriers or currents, so we can neglect these in Maxwell’s equations. The linear and
non-linear susceptibilities χ (i ) of silicon are well described in [1]. Unless intentionally
modified, silicon and SiO2 can be approximated as linear isotropic dielectrics (i.e.,
the material polarization is proportional to the electric field E ). The magnetic
susceptibility of the dielectrics can be neglected (i.e., the permeability μ is the
vacuum permeability μ0). In the the larger part of this work, material absorption is
neglected, i.e., the permittivity of all materials is real. In this work, all optics is
presented for monochromatic light with angular frequency ω and vacuum wave-
length λ = 2πc /ω, with c the speed of light in vacuum. This is not a limitation, as all
linear systems can equally be described in terms of time t or angular frequency ω via
the Fourier transform.
We describe the physical electromagnetic fields in terms of their complex amplitudes
E and H. The physical (and thus real) electromagnetic fields are the real parts of the
complex fields, i.e., Re{E} and Re{H}, respectively. Maxwell’s complex equations for
monochromatic light in an isotropic linear dielectric medium without charges are
given by [2]
∇ × E = −ıμ0ωH , (2.1)

doi:10.1088/978-0-7503-1386-5ch2 2-1 ª IOP Publishing Ltd 2017

 Silicon Photonics

Figure 2.1. Refractive index n of (a) silicon and (b) silicon dioxide, plotted as a function of wavelength.

∇ × H = ıωϵE , (2.2)

∇ · ϵE = 0, (2.3)

∇ · H = 0, (2.4)
with vacuum permeability μ0, permittivity ϵ, positive angular frequency ω, and time
evolution e ıωt . The last two equations (2.3) and (2.4) are not independent and follow
directly from the first two equations (2.1) and (2.2), as the divergence of the curl of a
vector field is zero. The refractive index is defined by n = ϵ /ϵ0 , with ϵ0 the vacuum
permittivity. The permeability ϵ(x , y, z ) profile describes how the devices (e.g.
waveguides, couplers, etc) look, and ultimately determine how the electromagnetic
fields behave.
Electromagnetic fields in a homogeneous isotropic medium obey the Helmholtz
wave equation
(∇2 + n 2k 2 )E = 0, (2.5)

(∇2 + n 2k 2 )H = 0, (2.6)
with k = ω /c the wavenumber in vacuum [2, chapter 9]. When using the wave
equation for homogeneous media, it is necessary to apply interface conditions to
ensure that the solutions also obey Maxwell’s equations across interfaces. If care is
taken that the electric and magnetic fields are free of divergence, equations (2.3) and
(2.4), then the four interface conditions are that the tangential components of the
electric E and magnetic H fields are continuous [2, chapter 7]. Otherwise, one also
has to impose boundary conditions for the normal components.

2.2 Dielectric slab waveguides

Dielectric waveguides consist of a rod of a material that has a refractive index higher
than its surroundings. Light can be guided along the z-axis of such a structure that is
translational invariant with respect to z. In today’s most common silicon-on-
insulator technology, the silicon waveguide is rectangular and fabricated on top of
a SiO2 layer with a cladding of either SiO2 or air. The easiest waveguide to study is a

2-2
 Silicon Photonics

slab waveguide. A slab waveguide is a layer sandwiched between two layers with a
lower refractive index. This basic structure demonstrates the concept of modes, and
our description of rectangular waveguides (section 2.4) is strongly related to the
equations derived for the slab waveguide. Light propagation through a slab
waveguide may be described by ray optics, even when the waveguide height is
only half a wavelength (see [3–5]). We, however, directly use wave optics.
Figure 2.2 depicts a slab waveguide. For now, we use the coordinate frame in
figure 2.2(a), in which the z-direction coincides with the direction of propagation of
the wave and the y-direction is normal to the slab. In section 2.2.2 (figure 2.2(b)), we
will use a different coordinate frame because that will be useful in the description of
rectangular waveguides (section 2.4). We simplify the analysis by assuming that
there is no variation of the electromagnetic fields in the transverse x-direction
(∂/∂x = 0), and indeed slab waveguides are infinitely wide. Maxwell’s equations
(2.1–2.4) actually include eight equations as Faraday’s law (2.1) and Ampère’s law
(2.2) are vectorial. Substituting ∂/∂x = 0 in these equations shows that they decouple
in equations either with Ex , Hy and Hz , or with Hx , Ey and Ez . Solutions to the first set
of equations are classified as transverse electric (TE) modes because the electric field
E = Exx̂ is transverse to the direction of propagation. Solutions of the second set of
equations are classified as transverse magnetic (TM) modes because H = Hx x̂ .

Figure 2.2. Silicon-on-insulator slab waveguide with height 600 nm. (a,b) Sketches of cross sections. The slab is
depicted horizontally by the grey area. Waves propagate in the z -direction, i.e., orthogonal to the (x, y )-plane
depicted here (i.e., into the paper or screen). (a) TE-modes. Coordinate frame. Electric fields of the three
supported guided modes. (b) TM-modes. Coordinate frame. Magnetic fields of the three supported guided
modes. (c) Graphical method to solve equation (2.17) for TE modes. Tangential term (dashed line) and fraction
term (solid line) both plotted versus the effective index ne , = β /k. Then ky = k 2n12 − β 2 = k 2n12 − k 2ne2 .
The three crossings correspond to the solutions plotted in (a). In section 2.2, n2 = n 4 and n3 = n5. We chose
different labels of the cladding refractive indices for the TE- and TM-modes (n 4 , n5 and n2 , n3) because that will
be useful in section 2.4.

2-3
 Silicon Photonics

2.2.1 TE modes
In figure 2.2(a), the TE-modes in a 600 nm high silicon-on-insulator slab waveguide
are sketched. Transverse electric modes have only three non-zero field components:
Ex , Hy and Hz . The magnetic components can be computed from the electric
component, so that Ex uniquely defines the electromagnetic fields. We look for
monochromatic propagating wave solutions, i.e., solutions in the form
E = Ex(y )e ı(ωt−βz )xˆ , (2.7)
with propagation constant β. Subsituting this in equation (2.5) gives
∂ 2Ex
= (β 2 − k 2n 2 )Ex. (2.8)
∂y 2
Depending on the sign of (β 2 − k 2n 2 ), the solutions of this equation are either
standing waves or exponential functions of y. Fields that exponentially increase in
the cladding with the distance to the core are not physical. The slab acts as a
waveguide when k 2n12 > β 2 > k 2n 42 , k 2n52 . In this case, the electric field inside the
core (n1) is a standing wave, while the field exponentially decays in the cladding. The
light is thus confined to the core. For β 2 > k 2n12 , the wave is above the cutoff of all
materials and does not propagate. For β 2 < k 2n 42 , k 2n52 , the wave propagates in the
upper/lower cladding and is not confined to the core. For guided waves, equation
(2.8) has the solutions
⎧C exp[ −γ5(y − b /2)] , upper cladding y > b /2,
⎪
Ex(y ) = ⎨ A cos⎡⎣ky(y + η)⎤⎦ , core b /2 ⩾ y ⩾ −b /2, (2.9)
⎪
⎩ B exp[γ4(y + b /2)] , lower cladding y < −b /2,

with

ky = k 2n12 − β 2 , (2.10)

γj = β 2 − k 2n j2 = k 2(n12 − n j2 ) − k y2 , j = 4, 5, (2.11)

and η a real constant. Expression (2.9) obeys Maxwell’s equations in all three layers
of the slab, but it also has to obey these equations at the interfaces. Electromagnetic
interface conditions demand continuity of the tangential electromagnetic field
components. Continuity of Hz in combination with Faraday’s law (2.1) and
∂/∂x = 0 are equivalent to the continuity of ∂Ex /∂y . We first calculate the relations
that follow from these conditions and then discuss their meaning.
⎧ − γ5C exp[ −γ5(y − b /2)] , upper cladding,
∂Ex ⎪
= ⎨− kyA sin⎡⎣ky(y + η)⎤⎦ , core, (2.12)
∂y ⎪
⎩ γ4B exp[γ4(y + b /2)] , lower cladding.

2-4
 Silicon Photonics

From the interface conditions at the lower interface (y = −b/2), we find

A cos⎡⎣ky(η − b /2)⎤⎦ = B, (2.13)

tan⎡⎣ky(η − b /2)⎤⎦ = −γ4 / ky, (2.14)

and at the upper interface ( y = b/2), we find

A cos⎡⎣ky(η + b /2)⎤⎦ = C , (2.15)

tan⎡⎣ky(η + b /2)⎤⎦ = γ5/ ky. (2.16)

In the description of the modal electric field, equation (2.9), B and C follow from
equations (2.13) and (2.15), respectively. The amplitude A is a normalization factor,
describing the amount of light in the mode, which does not follow from the interface
conditions. Note that the propagation constant β follows from ky via equation
(2.10). Equations (2.14) and (2.16) are two coupled equations for ky and η. By
eliminating η1, we find the following equation for ky
ky(γ4 + γ5)
F (ky, k , n1, n 4, n5, b) ≡ tan⎡⎣kyb⎤⎦ − = 0. (2.17)
k y2 − γ4γ5

For a given slab waveguide (n1, n4, n5, b) and angular frequency (k = ω /c ), Maxwell’s
equations thus demand ky to obey equation (2.17). Every solution ky of equation
(2.17) with 0 < ky < k · min⎡⎣ n12 − n 42 , n12 − n52 ⎤⎦ corresponds to a guided TE
mode. Depending on the slab waveguide, equation (2.17) has zero or more solutions
for guided waves with β between n1k and the higher value of n 4k and n5k . The tangent
is periodic and therefore multiple solutions may exist (see figure 2.2(c)). We hereby
demonstrated that a waveguide has a finite number of modes for a given frequency ω.
Each mode has its own distinct propagation constant β and modal field E (y ).

2.2.2 TM modes
Transverse magnetic modes have the magnetic field transverse to the propagation
direction. Although annoying now, we switch to a different coordinate frame in
which the x-direction is normal to the slab (see figure 2.2(b)), because this will be
convenient when treating the rectangular waveguide. In the considered frame, the
non-zero electromagnetic field components are Hy , Ex , and Ez . The waveguide is
infinitely wide in the y-direction and ∂/∂y = 0. Analogue to the derivation of TE
modes, we look for propagating wave solutions
H = Hy(x )e ı(ωt−βz )yˆ , (2.18)

1 tan[ky(b /2 + η )] + tan[ky(b /2 − η )] γ4 /ky + γ5 /ky

tan[kyb ] = tan[ky(b /2 + η) + ky(b /2 − η)] = = , using
1 − tan[ky(b /2 + η)]tan[ky(b /2 − η)] 1 − γ4γ5 /k y2
[6, equation (2.174)] for the 2nd equality sign and equations (2.14) and (2.16) for the 3rd equality sign.

2-5
 Silicon Photonics

obeying the wave equation (2.6), so that

∂ 2Hy
= (β 2 − k 2n 2 )Hy. (2.19)
∂x 2
Guided waves have k 2n12 > β 2 > k 2n 22 , k 2n32 , and the solution of equation (2.19) is
⎧C˜ exp[ −γ (x − d /2)], upper cladding x > d /2,
⎪ 3
Hy = ⎨ A˜ cos[kx(x + ξ )], core d /2 ⩾ x ⩾ −d /2, (2.20)
⎪˜
⎩ B exp[γ2(x + d /2)], lower cladding x < −d /2,

with
k x2 = k 2n12 − β 2 , (2.21)

γj2 = β 2 − k 2n j2 = k 2(n12 − n j2 ) − k x2, j = 2, 3. (2.22)

The interface conditions demand continuity of the tangential electromagnetic field

components (i.e., Hy ). Inspecting Ampère’s law (2.2) in the z-direction and using
∂/∂y = 0 gives
1 ∂Hy 1 ∂Hy ı(ωt−βz )
ıωEz = 2
= 2 e . (2.23)
n ∂x n ∂x
Hence the continuity of the tangential component Ez is equivalent to the continuity
of ∂Hy /∂x . We have
⎧ − γ / n 2 C˜ exp[ −γ (x − d /2)] , upper cladding,
1 ∂Hy ⎪
3 3 3
= ⎨ 2 ˜
− kx / n1 A sin[kx(x + ξ )], core, (2.24)
n 2 ∂x ⎪ 2
⎩ γ2 / n 2 B˜ exp[γ2(x + d /2)] , lower cladding.
The four interface conditions, obtained from the continuity of Hy and Ez at
x = −d /2 and x = d /2, give
A˜ cos[kx(ξ − d /2)] = B˜, (2.25)

n12 γ2
tan[kx(ξ − d /2)] = − , (2.26)
n 22 kx

A˜ cos[kx(ξ + d /2)] = C˜ , (2.27)

n12 γ3
tan[kx(ξ + d /2)] = − . (2.28)
n 32 kx

2-6
 Silicon Photonics

Combining2 equations (2.26) and (2.28) gives the following equation for kx, or
equivalently for propagation constant β:
n12kx(n 22γ3 + n 32γ2 )
G (kx, k , n1, n2 , n3, d ) ≡ tan[kxd ] − = 0. (2.29)
k x2n 22n 32 − n14γ2γ3

2.3 General properties of modes in dielectric waveguides

This section is about general properties of dielectric waveguides. Some important
characteristics such as the finite number of modes in a waveguides were already
observed for the slab waveguide. In this section, we derive properties and equations
for guided modes in an aribtrary waveguide with a two-dimensional refractive index
profile n(x , y ) (or permittivity profile). A dielectric waveguide is fully described by
its permittivity profile ϵ(x , y ) which is invariant in the z-direction, the direction in
which the light propagates. We look for propagating wave solutions in the form
E(x , y , z , t ) = E (x , y )e ı(ωt−βz ), H(x , y , z , t ) = H (x , y )e ı(ωt−βz ). (2.30)
The propagation constant β is often expressed in terms of the effective index
β
ne ≡ , (2.31)
k
with free-space propagation constant k = 2π /λ and free-space wavelength λ. The
first-order dispersion in the effective index ne can be expressed in terms of the
effective group index ng
∂β ∂n
ng ≡ = ne − λ e , (2.32)
∂k ∂λ
where the last equality follows from equation (2.31). In the case that we only use a
small wavelength span around a center wavelength λc , it is often possible to
approximate the wavelength-dependence of the effective index ne(λ ) by a linear
function, so that
⎡ n e ( λ c ) − ng ( λ c ) ng ( λ c ) ⎤
β ( λ ) ≈ 2π ⎢ + ⎥. (2.33)
⎣ λc λ ⎦

2.3.1 Guided waves described by Ex, Ey, Hx and Hy

For electromagnetic fields in the form of equation (2.30), Maxwell’s equations (2.1)
and (2.2) may be written as3

2 tan[kx(d / 2 + ξ )] + tan[kx(d / 2 − ξ )]
tan[kxd ] = tan[kx(d /2 + ξ ) + kx(d /2 − ξ )] = 1 − tan[kx(d / 2 + ξ )] tan[kx(d / 2 − ξ )]
, using [6, equation (2.174)] for the
second equality sign.
⎛ ∂Ez + ıβE ⎞ ⎛ ∂Hz + ıβH ⎞
⎜ ∂y y ⎟ ⎛− ıωμ0 Hx ⎞ ⎜ ∂y y ⎟
⎛ ıωϵEx ⎞
3 ⎜ ∂Ez ⎟ ⎜ ⎟ ⎜ ∂Hz ⎟ ⎜ ⎟
Written in Cartesian components: ⎜− ıβEx − ∂x ⎟ = ⎜− ıωμ0 Hy ⎟, and ⎜− ıβHx − ∂x ⎟ = ⎜ ıωϵEy ⎟.
⎜ ∂Ey ⎜ ⎟ ⎜ ⎟
⎜ ∂Ex ⎟ ⎟ ⎝− ıωμ0 Hz ⎠
⎜ ∂Hy
⎜ ∂Hx ⎟ ⎟ ⎝ ıωϵEz ⎠
⎝ ∂x − ∂y ⎠ ⎝ ∂x − ∂y ⎠
2-7
 Silicon Photonics

∇β × E = −ıωμ0H, (2.34)

∇β × H = ıωϵE, (2.35)
with
∂ ∂
∇β = xˆ + yˆ − ıβzˆ. (2.36)
∂x ∂y
We may describe the propagation of light through waveguides as an eigenvalue
problem with the propagation constant β as eigenvalue. By eliminating the
longitudinal components Ez and Hz from Maxwell’s equations, one obtains the
following eigenvalue problem for the transverse components only
⎛ Ex(x , y ) ⎞ ⎛0⎞
⎜ ⎟
E y (x , y ) ⎟ ⎜0⎟
(Oˆ − β )⎜ = ⎜ ⎟, (2.37)
⎜ Hx(x , y )⎟ ⎜ 0 ⎟
⎜ ⎟
⎝ Hy(x , y )⎠ ⎝ 0 ⎠

with Ô a second-order partial differential operator with respect to transverse

variables x and y. To show this, we note that the z-components of Faraday’s law
(2.34) and Ampère’s law (2.35) imply
ı ⎛ ∂Ey ∂Ex ⎞
Hz = ⎜ − ⎟, (2.38)
ωμ0 ⎝ ∂x ∂y ⎠

ı ⎛ ∂Hx ∂Hy ⎞
Ez = ⎜ − ⎟. (2.39)
ωϵ ⎝ ∂y ∂x ⎠

Substituting equation (2.39) in the x- and y-components of Faraday’s law (2.34) and
substituting equation (2.38) in the x- and y-components of Ampères law (2.35) gives
a set of four equations which are linear in Ex, Ey, Hx and Hy. Rearranging these
equations to the form of equation (2.37) gives
⎛ ∂ 1 ∂ ∂ 1 ∂ ⎞
⎜ 0 0 − ⎟
⎜ ∂x ωϵ ∂y ∂x ωϵ ∂x ⎟
⎛0 0 0 ωμ 0 ⎞ ⎜ 0 0 −
∂ 1 ∂ ∂ 1 ∂ ⎟
⎜ ⎟ ⎜ ∂y ωϵ ∂y ∂y ωϵ ∂x ⎟
0 0 − ωμ 0 0 ⎟ + ⎜ ⎟.
Oˆ = ⎜ (2.40)
0 ⎟⎟ ⎜ 1 ∂ ⎟
2 −1 ∂ 2
⎜⎜ 0 − ωϵ 0 0 0
⎝ ωϵ 0 ⎜ ⎟
0 0 ⎠ ⎜ ωμ 0 ∂x∂y ωμ 0 ∂x 2
⎟
⎜ 1 ∂2 −1 ∂ 2 ⎟
⎜ 0 0 ⎟
⎝ ωμ 0 ∂y 2 ωμ 0 ∂y∂x ⎠

2-8
 Silicon Photonics

Figure 2.3. Sketch of eigenvalues β of the eigenvalue problem in equation (2.37) with (2.40) showing guided,
radiation and evanescent modes. The refractive index of the core and cladding of the guide are n1 and n2 ,
respectively and k is the wave number in free space. The discrete number of guided modes, marked by x, lie on
the real axis in the interval kn2 > ∣β∣ > kn1. The continuum of radiation modes are given by ∣β∣ ⩽ kn2 . The
continuum of evanescent modes lie on the imaginary axis. See [7–9].

The solutions of the eigenvalue problem equation (2.37 with 2.40) are called modes
and the eigenvalues β are called propagation constants. Different types of modes are
associated with different eigenvalues β (see figure 2.3). The operator Ô is not
symmetrical, let alone self-adjoint, with respect to the usual L2 scalar product. Note
that for every guided wave (Ex, Ey, Hx, Hy) with positive propagation constant β,
guided wave (Ex, Ey, −Hx, −Hx) corresponds to negative propagation constant −β.
Let n1 and n2 be the refractive index of, respectively, the core and the cladding of
the waveguide. Similar to the slab waveguide, guided modes propagate along the
guide in the z direction. The guided modes have real propagation constants β with
kn2 > ∣β∣ > kn1 and there are finitely many of them. There is a continuum of
radiation modes which have real propagation constants β with ∣β∣ ⩽ kn2 . There is
furthermore a continuum of evanescent modes with imaginary propagation con-
stants β. In section 2.3.6 we explain in more detail the generally used assumption
that an arbitrary field can, in every cross-sectional plane z = constant, be written as a
superposition of the electromagnetic fields of the guided, radiation and evanescent
modes.
It is relevant to note that the operator, Ô , and the eigenvalues, β, of guided modes
are real. Therefore, we may choose the eigenfunctions of guided modes
(Ex, Ey, Hx, Hy ) to be real4. Hence all transverse electromagnetic field components
of the guided modes are in phase.

2.3.2 Guided waves described by Ez and Hz

Alternatively, it is possible to formulate Maxwell’s equations in terms of the
longitudinal field components Ez and Hz. This formulation has only two unknown
field components, and it will be used in the description of rectangular waveguides in

4
Proof: Let a real number β be the eigenvalue of a real operator Ô with a corresponding complex eigenfunction
u = u′ + ıu″, with real u′ and u″. Then the eigenproblem (Oˆ − β )u = 0 separates into (Oˆ − β )u′ = 0 and
(Oˆ − β )u″ = 0 . As both equations are identical, they span the same eigenspace, and we may thus choose the
real eigenspace as basis.

2-9
 Silicon Photonics

section 2.4. Solving equations (2.34) and (2.35) for the transverse field components
gives5
−ı ⎛ ∂Ez ∂H ⎞
Ex = ⎜β + ωμ0 z ⎟ , (2.41)
K ⎝ ∂x
2
∂y ⎠

−ı ⎛ ∂Ez ∂H ⎞
Ey = ⎜β − ωμ0 z ⎟ , (2.42)
K ⎝ ∂y
2
∂x ⎠

−ı ⎛ ∂Hz ∂E ⎞
Hx = ⎜β − ωϵ0n j2 z ⎟ , (2.43)
K ⎝ ∂x
2
∂y ⎠

−ı ⎛ ∂Hz ∂E ⎞
Hy = ⎜β + ωϵ0n j2 z ⎟ , (2.44)
K ⎝ ∂y
2
∂x ⎠

with

K (x , y ) = n( x , y ) 2 k 2 − β 2 . (2.45)

In a region where K is constant, the longitudinal components Ez and Hz satisfy (as

all other components), the reduced scalar Helmholtz equation
∂ 2Ez ∂ 2Ez
+ + K 2Ez = 0, (2.46)
∂x 2 ∂y 2
and similarly for Hz. This follows from substituting the modal electromagnetic field,
equation (2.30), in the Helmholtz equation (2.5). Furthermore, when equations
(2.41–2.44) are satisfied, the electric and magnetic field satisfy ∇ · (ϵj E) = 0 and
∇ · H = 0, respectively. Equation (2.46) is often used to derive expressions for the
field in regions where the refractive index is constant. These expressions are then
linked together using the interface conditions.
Alternatively, by substituting equations (2.41–2.44) into the z-components of
Faraday’s law and Maxwell’s law, one can derive a partial differential operator in
terms of only Ez and Hz of which β is the generalized eigenvalue and which is
equivalent to equation (2.40).

5
Equation (2.41) follows from solving the y-component of equation (2.34) for Hy, and substituting this in the
x-component of equation (2.35). Analogously, equation (2.42) follows from solving the x-component of
equation (2.34) for Hx, and substituting this in the y-component of equation (2.35). Equation (2.43) follows
from solving the y-component of equation (2.35) for Ey, and substituting this in the x-component of equation
(2.34). Equation (2.41) follows from solving the x-component of equation (2.35) for Ex, and substituting this in
the y-component of equation (2.34).

2-10
 Silicon Photonics

2.3.3 Orthogonality
We will derive the orthogonality relation which modes in a waveguide obey.
Consider two modes in the form of equation (2.30), labelled 1 and 2. A vector
identity6 implies
∇ · (E1 × H*2) = H*2 · (∇ × E1) − E1 · (∇ × H*2), (2.47)

as before * denotes the complex conjugate. Then we use Faraday’s law (2.1) and
Ampère’s law (2.2) in the first and second terms on the right-hand-side of equation
(2.47), respectively, to obtain
∇ · (E1 × H*2) = ıω(ϵE1 · E *2 − μ0H1 · H*2). (2.48)

Now we exchange labels 1 and 2 in equation (2.48), take the complex conjugate, and
add the result to equation (2.48). This gives
∇ · (E1 × H*2 + E *2 × H1) = 0. (2.49)

Modes 1 and 2 are waveguide modes of the form of equation (2.30), and therefore
equation (2.49) can be written as
∇ · (E1 × H2* + E2* × H1)e ı(β2−β1)z = 0, (2.50)

or, using a well known formula of calculus7,

∇ · (E1 × H2* + E2* × H1) + (E1 × H2* + E2* × H1) · zˆ ı(β2 − β1) = 0, (2.51)

where the equation has been divided by the factor e ı(β2−β1)z . We now use a two-
dimensional version of the divergence theorem8 where we integrate equation (2.51)
over a circular disc S in a cross-sectional plane z = constant of the waveguide:

∮∂S (E1 × H2* + E2* × H1) · nˆ dS

(2.52)
+ ∬ (E1 × H2* + E2* × H1) · zˆ dxdy ı(β2 − β1) = 0,
S

with n̂ the unit vector in the (x , y) plane perpendicular to the circle ∂S and outward
pointing with respect to S. The field of guided modes decrease exponentially with
distance to the guide. Therefore the contour integral over the circle ∂S vanishes in
the limit of infinite radius R → ∞ and equation (2.52) becomes
+∞
1
4
∬−∞ (E1 × H2* + E2* × H1) · zˆ dxdy = 0 if β2 ≠ β1. (2.53)

6
∇ · (a × b) = b · (∇ × a) − a · (∇ × b) [6, equation 2.53]
7
∇ · (f a) = f (∇ · a) + a · (∇f ) [6, equation 2.50]
8
A special case of the divergence theorem follows by specializing it to the plane instead of a volume [2, 10].
Letting S be a region in the plane with boundary ∂S , the divergence theorem then simplifies to
∬S ( ∂∂x xˆ + ∂∂y yˆ ) · a dA = ∮∂S a · nˆ dS .

2-11
 Silicon Photonics

equation (2.53) can be considered as an orthogonality of the modes. Note that it

depends only on the transverse components of the electromagnetic field, i.e., on
the x- and y-components of the electric and magnetic field. We have derived this
orthogonality for guided modes but it actually holds for any pair of modes
whether guided, radiation or evanescent, with the exception of two evanescent
modes that have opposite β [7, 9].
As will be shown in section 2.3.4, by taking the modes 1 and 2 to be identical, the
left-hand-side of equation (2.53) is the time-averaged power carried by the mode.
Although we consider equation (2.53) as an orthogonality of the modes, it is important
to realize that the left-hand side of equation (2.53) is not a scalar product. The reason is
that for equal modes the power flow is not always positive, hence for equal modes the
expression cannot be considered as a norm. In fact, if the modes have negative
propagation constant, they propagate in the negative z-direction and hence the power
flow is in that direction and the left-hand side of equation (2.53) is negative. Moreover,
evanescent modes do not propagate energy in the z-direction, hence the left-hand side
of equation (2.53) vanishes when the modes are the same and evanescent.
We proceed deriving a shorter orthogonality relation that is only valid for
unidirectional modes. Forward- and backward-traveling modes are not necessarily
orthogonal in the shorter relation. We first show that the two terms under the
integral in equation (2.53) vanish separately.
If E2 , H2 is the field of a mode with positive propagation constant β2 > 0, then
there exists a mode with propagation constant β3 = −β2 and with electromagnetic
field:
E3 = E2xxˆ + E2yyˆ − E2zzˆ , (2.54)

H3 = −H2xxˆ − H2yyˆ + H2zzˆ. (2.55)

This is easy to see from equations (2.37) and (2.40). Alternatively, equation (2.55)
can be derived by substituting equation (2.54) in Faraday’s law (2.34) and
comparing the result with β2 and E2 also substituted in Faraday’s law. Equations
(2.54)–(2.55) also obey Ampère’s law (2.35), as can be verified by first substituting β3
and H3 in Ampère’s law, then also substituting β2 and H2 in Ampère’s law, to
compare the resulting equations. The fact that the modal fields of the backwards
traveling modes are different from the forward traveling modes can be explained by
looking at the power flow. The time-averaged power flow of a mode is given by
equation (2.60). Since for backward propagating modes the power flow is opposite
to those of forward propagating modes, their fields must be different.
Because only the transverse field components contribute to equation (2.53), it
follows from equations (2.54) and (2.55) that when equation (2.53) is applied to
modes 1 and 3 we get
1
4
∬ (−E1 × H2* + E2* × H1) · zˆ dxdy = 0 if β2 ≠ −β1. (2.56)

2-12
 Silicon Photonics

By substracting equation (2.53) from equation (2.56) we find

1
2
∬ (E1 × H2*) · zˆ dxdy = 0 if β2 ≠ β1 . (2.57)

Note that the orthogonality in equation (2.57) is not true for a forward propagating
mode and its backward propagating counterpart as their propagation constants β
have equal absolute value.
Similar to the derivation of equations (2.53) and (2.57) one can derive orthogon-
ality relations without complex conjugation. For the derivation we refer to [9, chapter
31]. For modes propagating in the same direction one finds the orthagonality:
1
2
∬ (E1 × H2) · zˆ dxdy = 0 if β2 ≠ β1 . (2.58)

2.3.4 Power flow

The instantaneous power density (energy per unit area per unit time) is given by the
Poynting vector [2]
S = Re{E} × Re{H}, (2.59)
which is defined in terms of the physical (real) electromagnetic fields. The Poynting
vector of a waveguide mode is found by inserting equation (2.30) into equation
(2.59) and using Re{E} = 12 (E + E *):
1
S=
4
( (E × H*) + (E * × H) + (E × H)e ı2(ωt−βz ) + (E * × H*)e ı2(βz−ωt )).
The last two terms on the right-hand-side of this equation vanish when averaged
over a time interval long compared to the period of the mode. The time-averaged
power P carried by the mode is found by integrating the time-averaged Poynting
vector S over a cross-sectional plane z = constant. Using (E × H * +
E * × H ) = 2 Re{E × H *}, one finds [3, 4]:
1
P=
2
Re {∬ (E × H ) · zˆ dxdy}.
* (2.60)

It follows from equations (2.37) and (2.40) that the transverse electric and magnetic
fields of guided modes are in phase hence the integral at the right of equation (2.60)
is real for guided modes. This can be seen as follows: when the transverse field Et, Ht
is eigenfield of the operator Ô with real eigenvalue β, also Et*, Ht* is eigenfield with
eigenvalue β and hence so are Re{Et}, Re{Ht }. Since the propagation constants β of
the guided and radiation modes are real we conclude that the transverse field
components Et, Ht of the guided and the radiation modes may be asssumed to be
real. Hence for these modes the integral between the brackets in equation (2.60) is
real and taking the real part is superfluous. Furthermore, the expressions (2.57) and
(2.58) are identical when the fields of the modes are chosen real. Note that for
evanescent fields, the integral in equation (2.60) is purely imaginary because for
evanescent modes the power flow in the z-direction vanishes.

2-13
 Silicon Photonics

2.3.5 Mode normalization

Throughout this book, we normalize the guided waves such that they carry unit
power in the positive or negative z-direction, i.e., P = ±1 in equation (2.60), except
for some cases where it is explicitly mentioned that another normalisation is used.
Evanescent modes cannot be normalized using equation (2.60) because they carry
no power in the z-direction. It is always possible to normalize modes based on an
unconjugated orthogornality relation. In this book, it so happened that when
considering evanescent modes, only unidirectional orthogornality was needed.
Therefore, when considering evanescent modes, we will use equation (2.58) for the
orthogonality and the normalisation of all modes: guided, radiation and evanescent
modes. As was explained in section 2.3.4, for guided and radiation modes, the
transverse field components can be chosen to be real and in that case equations
(2.57) and (2.58) are identical. Because the radiation and evanescent modes are
unbounded in the transverse direction, the orthogonality relation for these modes
becomes:
1
2
∬ (E1 × H2) · zˆ dxdy = δ(β1 − β2). (2.61)

For two guided modes we then have

1
2
∬ (E1 × H2) · zˆ dxdy = δ12. (2.62)

2.3.6 Mode expansion conjecture

The mode expansion conjecture as presented hereafter will be used in section 4.2 and
appendix A. It states that at given frequency ω, the modes of a waveguide are
complete in the sense that in every cross-sectional plane z = constant, the transverse
components of an arbitrary field, not necessarily a solution of Maxwells equations,
can always be written as a linear superposition of the transverse components of the
electromagnetic fields of all the modes of the given waveguide [3, 7, 8, 11–13]. All
the modes means, guided, radiation and evanescent modes. The fact that only the
transverse components of the electromagnetic field figure in this superposition can be
understood as follows. The problem of finding the modes of a waveguide has been
formulated as an ordinary eigenvalue problem for the operator Ô , defined by
equation (2.40), which operates on the transverse field components only. It has been
shown in section 2.3.3 that the eigenfields of this operator are orthogonal in the sense
of equation (2.53), derived from the z-component of the Poynting vector.
Alternatively, the modes are also orthogonal in the sense of equations (2.61) and
(2.62). The basic assumption is now that the totality of these eigenfields is complete
in the space of all tangential fields, i.e., that the tangential components of any field
can be written as a linear combination of the tangential components of the guided,
radiation and evanescent waveguide modes.

2-14
 Silicon Photonics

We use this completeness to expand the transverse components of a time harmonic

electromagnetic field Ex , Ey , Hx , Hy of frequency ω, which is a solution of Maxwell’s
equations for some permittivity profile ϵ(x , y ) which differs from that of the given
waveguide. The transverse components of this field can then be written as
⎛ E (i ) ⎞
⎛ Ex ⎞ ⎜ x ⎟
⎜ ⎟ ⎜ Ey(i ) ⎟
⎜ Ey ⎟ =
2N

⎜H ⎟ ∑ai (z ) ⎜ (i )⎟eı(ωt−βiz )
⎜ Hx ⎟
⎜ x⎟ i=1
⎜⎜ (i ) ⎟⎟
⎝ Hx ⎠
⎝ Hy ⎠
 
guided
⎛ Ex (β ; x , y ) ⎞
⎜ ⎟
+kn 2 ⎜ Ey (β ; x , y ) ⎟ ı(ωt − βz )
+ ∫ −kn 2
b (β ; z ) ⎜
Hx (β ; x , y )⎟
e dβ (2.63)
⎜⎜ ⎟⎟
⎝ Hy (β ; x , y )⎠

radiation
⎛ Ex (β ; x , y ) ⎞
⎜ ⎟
+ı ∞ ⎜ Ey (β ; x , y ) ⎟ ı(ωt − βz )
+ ∫−ı ∞
c (β ; z ) ⎜
Hx (β ; x , y )⎟
e dβ
⎜⎜ ⎟⎟
⎝ Hy (β ; x , y )⎠

evanescent

where Ex, Ey, Hx, and Hy are the transverse components of the modes, of frequency
ω, of the given waveguide with cladding refractive index n2. Here, ai (z ), b(β; z ) and
c(β; z ) are the complex coefficients of the guided, radiation and evanescent modes,
respectively, which are all functions of z. The number N is the number of forward
propagating guided modes. Figure 2.3 shows the complex propagation constants β
of the different types of modes. So far, the conjecture has been proven for two-
dimensional lossless optical components, including TE- and TM-modes in slab
waveguides [8, 13]. The conjecture is often assumed to hold also in general, and
although no counter-examples are known, it seems that a mathematical proof for the
general case has not been given (see [8, chapter 2] for details).
One might compare this expansion with spatial Fourier decomposition, in which
an arbitrary function f (x , y ) can be written as an infinite superposition of spatial
harmonic functions.
We introduce a short-hand notation of equation (2.63) by writing the second and
third terms as a summation. The modal amplitudes ai (z ), b(β; z ) and c(β; z ) are
lumped together in ai with infinite i. This gives
∞
Et(x , y , z , t ) = ∑ ai(z )Et(i )(x, y )e ıωt , (2.64)
i=0

2-15
 Silicon Photonics

∞
Ht (x , y , z , t ) = ∑ ai(z )Ht(i )(x, y )e ıωt. (2.65)
i=0

with the summation running over all modes including discrete and continuum
modes, both in forward and backward direction, with Et , Ht a time harmonic
electromagnetic field that satisfies Maxwell’s equations for an arbitrary permittivity
profile, with Et(i )(x , y ), Ht(i )(x , y ) the transverse components of the modes of a given
waveguide and with ai (z ) are the modal coefficients which are functions of z.
Subscript t denotes the transverse field components x and y.
In a numerical implementation one would discretize the integrals over the
radiation and evanescent waves so that a summation would then anyway be
obtained. It is also possible to achieve a discrete set of modes using the approx-
imation that the waveguide is contained in a large cylinder with a perfectly
conducting metal wall. This cylinder enforces the tangential components of the
electric fields to vanish at the wall due to which the radiation and evanescent modes
form a discrete set.

2.3.7 Waveguides with losses and bends

So far, we have discussed ideal waveguides, composed of lossless materials, without
the slightest variation of the refractive index profile in the z-direction, and with
infinite cladding. However, real waveguides are never perfect and also include bends,
for example, to form a ring resonator. This section describes these phenomena in a
less formal manner than the previous sections. Rigorously including all loss
mechanisms is outside the scope of this work and computations are often impossible
due to limited knowledge about the precise material and structural properties of the
waveguides. Roughness of the sidewalls of the waveguide and bends in waveguides
break the assumption of z-invariance of the permittivity profile ϵ(x , y ).

Waveguide propagation loss

We include a propagation loss αp in our description of modes in waveguides, and
modify equation (2.30) to
E = e−αpzE (x , y )e ı(ωt −βz ), H = e−αpzH (x , y )e ı(ωt −βz ). (2.66)
Equation (2.66) does not obey the rigorous mathematical framework that was
derived up to here. We have assumed no absorption (already in Maxwell’s
equations, section 2.1) and invariance of the refractive index in the z-direction,
which led to equation (2.30). Real waveguides are not perfectly invariant in the
z-direction as the sidewalls contain nanometer-scale roughness, the material contains
nanometer-scale defects, the waveguides have width variations over a millimeter
length-scale, etc. This leads to coupling of the guided mode to other modes.
Moreover, the materials and material interfaces of real waveguides absorb electro-
magnetic energy. These effects are lumped together in the propagation loss αp. The
term exp[−αpz ] can be interpreted as modal amplitude, and the theories in this section

2-16
 Silicon Photonics

Figure 2.4. Sketch of a bent waveguide with radiation loss, top-view.

remain a good approximation when this amplitude varies slowly compared to the
other variations of the electromagnetic fields.

Waveguide bends
For waveguide bends with radii R much larger than the width of the waveguide, it is
reasonable to approximate the modes in the bend as the modes of the locally straight
waveguide. We will show that radiation loss in waveguide bends is lower when the
bending radius is larger and when the refractive index contrast is higher. We assume
that the phase front of a mode in a bent waveguide has an angular velocity Ω and
that the velocity vz of the light at a distance r′ from the center of the waveguide reads
vz = (R + r′)Ω , (2.67)
see figure 2.4. We approximate the phase velocity in the center of the waveguide by
the phase velocity of the mode in a straight waveguide, i.e., vz = ω /β for r′ = 0. The
phase front of a mode in a bent waveguide thus has angular velocity
ω
Ω= . (2.68)
βR
Around the waveguide, the velocity of the light, vz, is similar to the value ω /β at the
center, because the width of the waveguide is much smaller than the bending radius.
However, this velocity increases with increasing distance from the center of the
guide. In the cladding at a certain distance r from the center of the waveguide, the
required velocity exceeds the velocity of unguided light in the cladding (c /n2 ), leading
to radiation. We find this distance r by solving vz = c /n2 , equation (2.67) and
equation (2.68) for r [14],
⎛ β − n2k ⎞
r=R⎜ ⎟. (2.69)
⎝ n2k ⎠

This radiation process may be intuitively understood by the following hand-waving

argumentation [14]. Photons of the optical mode located at radii greater than R + r

2-17
 Silicon Photonics

cannot travel fast enough to keep up with the mode. As a result, they split away and
radiate into the cladding.
The strength of the radiation scales with the amount of the modal electromagnetic
field that is located further than the distance r from the center of the waveguide. As
can be seen in equation (2.69), a large bending radius R implies a large r, thus lower
radiation loss. High index contrast waveguides have modes that are strongly
confined around the waveguide and smaller field values at distances larger than r.
Moreover, the difference between β and kn2 is generally large for lower order modes
in high-index-contrast waveguides. Therefore sharp bends in silicon waveguide have
in general lower radiation loss than waveguides of the same geometry with lower
index contrast.

2.4 Rectangular Si waveguides: extension of Marcatili’s approach

Marcatili’s famous approximate analytical description of light propagation through
rectangular dielectric waveguides, which was published in 1969, gives accurate
results for low-index-contrast waveguides [15]. Silicon-on-insulator waveguides have
however a high-index-contrast. Marcatili’s approach has recenly been extended to
the regime of high-index-contrast waveguides by adjusting the amplitudes of the
components of the electromagnetic fields [16, 17]. Although numerical mode-solvers
are available, we believe that an analytical model is useful in order to gain insight
into the physics of the devices, and also for fast explorative simulations of photonic-
integrated circuitry [18].
We will compare the approximate analytical approaches presented in this section
with results obtained using a rigorous numerical mode solver (FMM method, see
section 2.5 for details). Typical silicon-on-insulator waveguides with air cladding are
studied, with waveguide heights of 220 nm and 300 nm. The first height is often used
because it only supports TE-like modes, whereas the second height supports both
TE-like and TM-like modes.
This section starts with an ansatz for the form of the modes followed by the
expressions that result from the boundary conditions (section 2.4.1 and 2.4.2). Then
approximate solutions are proposed (section 2.4.3). Finally, the apparent degener-
acy of ‘TE-like’ and ‘TM-like’ modes is discussed (section 2.4.4).
We have published a Matlab [19] implementation of the methods presented in this
section as a free and open-source software package entitled RECTWG [20].

2.4.1 Ansatz for the shape of the field

Figure 2.5 shows a typical SOI waveguide, whose core has a higher refractive index
(n1) than its surroundings (n2 − n5). In this section, we make an ansatz for the shape
of the modal electromagnetic fields, inspired by the assumptions that fundamental
modes have most of their energy in the center of the waveguide (such as in figure
2.5(a)), and that modes are either ‘TE-like’ or ‘TM-like’. For ‘TE-like’ modes, the
electric field is predominantly tangential to the upper surface of the waveguide,
whereas ‘TM-like’ modes have the electric field predominantly normal to the upper
surface of the waveguide. In our analysis, we choose our coordinate frame such that

2-18
 Silicon Photonics

Figure 2.5. Cross-section of a SOI waveguide. (a) Waveguide definition. Sketch of the Ex component of the
fundamental mode in colour. (Dark blue represents large Ex , white represents Ex = 0 .) (b) Outline of the
approximate analytical method.

Ex is the dominant electric field component. Consequently, Hy is the dominant

magnetic field component of such modes. In figure 2.5, the width of the waveguide,
d, is larger than its height b, and the mode is ‘TE-like’ as Ex is parallel to the upper
surface of the waveguide. However, in our analysis, there are no restrictions on the
values of d, b, and n1-n5, so that the analysis equally describes a ‘TM-like’ mode
when n2 is said to be the substrate, b the width, and d the height of the waveguide
(thus b > d ).
As derived in section 2.3.2, Maxwell’s equations may be formulated in terms of
the longitudinal field components (Ez and Hz) only. For permittivity profiles that
are invariant in the z-direction, the transverse components follow from equations
(2.41–2.44).
Following Marcatili’s original approach, we use the following ansatz for the fields
of the modes. The dominant field components, Ex(x , y ), and Hy(x , y ), are propor-
tional to cos[kx (x + ξ )] cos[ky(y + η)], with maximum field in the center of the
waveguide core. Furthermore, outside the core, the fields decay exponentially, while
the transverse profile of the field is identical to that in the core (figure 2.5(b) and 2.6(a)).
Finally, the fields in the outer quadrants are neglected because they are small. Figure
2.6b presents the ansatz for the electric and magnetic fields, expressed in their
longitudinal components (Ez and Hz). We choose to obey Maxwell’s equations in
all regions 1–5, and express β, and γ2 - γ5 in terms of kx and ky employing the wave
equation (2.46)

β= n12k 2 − k x2 − k y2 , (2.70)

γ2 = (n 1
2
)
− n 22 k 2 − k x2 , γ3 = (n 1
2
)
− n 32 k 2 − k x2 , (2.71)

γ4 = (n 1
2
)
− n 42 k 2 − k y2 , γ5 = (n 1
2
)
− n 52 k 2 − k y2 . (2.72)

Equations (2.71) and (2.72) are identical to equations (2.22) and (2.11). The errors of
the approximation manifest themselves at the interfaces between the core and the
cladding of the waveguide. Field amplitude A1 is employed to normalize the mode to

2-19
 Silicon Photonics

Figure 2.6. (a) Shape of the dominant electric field component Ex . The dominant magnetic field component Hy
has the same shape. (b) Ansatz describing the modal electromagnetic field in terms of Ez and Hz . The gray
background colour sketches the waveguide such that the mode is a ‘TE-like’ mode.

a power flux of unity. The remaining free parameters that still have to be determined
are A2–A10, ξ, η, kx and ky (13 in total, also see figure 2.6(b)).

2.4.2 Boundary conditions

At interfaces between the core and the cladding of the waveguide, continuity of the
electromagnetic field components tangential to these interfaces is required, adding
up to 4 × 4 = 16 electromagnetic boundary conditions. With these conditions
satisfied, the normal components automatically obey Maxwell’s equations.
We derive the requirements that follow from continuity of the fields at the two
horizontal interfaces, to which the dominant electric field Ex is tangential. The field
components Ex, Hx, Ez and Hz are tangential to these interfaces. From the four
boundary conditions, we find

2-20
 Silicon Photonics

βky
A2 = A1, (2.73)
ωμ0kx

A7 = A1 cos[ky(η − b /2)], (2.74)

A8 = A2 sin[ky(η − b /2)], (2.75)

A9 = A1 cos[ky(η + b /2)], (2.76)

A10 = A2 sin[ky(η + b /2)], (2.77)

together with
tan[ky(η − b /2)] = −γ4 / ky, (2.78)

tan[ky(η + b /2)] = γ5/ ky. (2.79)

Equations (2.74)–(2.77) follow from the continuity of Ez and Hz. Continuity of Ex
and Hx is most easily verified by substituting equations (2.73)–(2.79) into the four
boundary conditions corresponding to these field components at the two interfaces.
It follows from equation (2.42) that with these field amplitudes A2, A7–A10, the
electric field component Ey is zero in regions 1, 4 and 5. The last two equations,
(2.78) and (2.79) can be recognized as the eigenvalue equations for a TE mode in a
slab waveguide, namely (2.14) and (2.16), respectively. These are two equations that
determine ky, using equation (2.17), and η. These eigenvalue equations thus do not
only hold for a slab solution where ∂/∂x = 0 and Ex, Hy and Hz are the non-zero field
components, but also for our ansatz where there is a variation in the x-direction. The
results agree with the limit that the width of the waveguide goes to infinity d → ∞, as
the waveguide then becomes a slab waveguide. Eliminating η from equations (2.78)
and (2.78) gives the functional F in equation (2.17).
The dominant magnetic field component, Hy, is tangential to the vertical
interfaces, and so are Ey, Ez and Hz. From the four boundary conditions at the
two vertical interfaces, we find
ωϵ0n12ky
A2 = A1, (2.80)
βkx

A3 = A1 sin[kx(ξ − d /2)], (2.81)

A4 = A2 cos[kx(ξ − d /2)], (2.82)

A5 = A1 sin[kx(ξ + d /2)], (2.83)

A6 = A2 cos[kx(ξ + d /2)], (2.84)

2-21
 Silicon Photonics

together with
n12 γ2
tan[kx(ξ − d /2)] = − , (2.85)
n 22 kx

n12 γ3
tan[kx(ξ + d /2)] = . (2.86)
n 32 kx
Equations (2.81)–(2.84) follow directly from the continuity of Ez and Hz. The
continuity of Ey and Hy is most easily verified by substituting equations (2.80)–(2.86)
into the remaining electromagnetic boundary conditions. With these field ampli-
tudes A2–A6, the magnetic field component Hx is zero in regions 1, 2 and 3, as
follows from equation (2.43). The last two equations, (2.85) and (2.86), may be
recognized as the eigenvalue equations of a TM mode in a slab waveguide, (2.26)
and (2.28), despite the fact that our ansatz does have variation in the y-direction.
These two equations determine kx, using equation (2.29), and ξ. With kx and ky, β
follows from equation (2.70).
It can be seen that the horizontal and the vertical interfaces require a different
ratio A2 /A1, i.e., a different Hz /Ez in the core. Thus the ansatz has no solutions that
exactly obey the boundary conditions at all interfaces simultaneously. In what
follows, the 13 free parameters are chosen such that the error in the 16 boundary
conditions is minimal.

2.4.3 Approximate methods

This section presents three approximate methods for rectangular waveguides, based
on the previous ansatz. All three methods share the same equations for the
computation of the propagation constant β, but the amplitudes A2–A10 of the
modal fields are computed differently. The methods for computation of the fields
are based, respectively, on (1) the assumption of low-index-contrast, (2) continuity
of the dominant electric and magnetic field components, and (3) minimization of the
discontinuity of the electromagnetic fields. The first method is Marcatili’s original
approach [15]. The last two methods were derived for high-index-contrast SOI
waveguides [16].

Propagation constant β and spatial frequencies kx, ky

We argue that the dominant boundary conditions for determining ky and η are at the
horizontal interfaces, while the vertical interfaces influence kx and ξ more strongly.
Therefore, we compute kx from equation (2.29), ky from equation (2.17), and β from
equation (2.70). In figure 2.7a,c we compare our approximate analytical calculation
with a rigorous numerical calculation and find that the effective index, n eff = β /k , is
accurately found within 2% for typical waveguides.

2-22
 Silicon Photonics

Figure 2.7. Approximate analytical model compared with rigorous mode solver (FMM method). Typical
rectangular silicon-on-insulator waveguides with air cladding. (a, b) 220 nm high, fundamental mode. (c–f) 300
nm high, first 3 modes: TE0, TM0, TE1. We omitted one zero from conventional notation (e.g. TE 00 ), because
our waveguides have higher-order standing waves only in the direction of the width of the waveguide. (a–c)
Effective index. (b, d, e, f) Energy in the difference field between the two approximate methods and the
rigorously computed field, normalized to the energy in the rigorously computed field, i.e., equation (2.92).

Fields (1): Marcatili's approach for low-index-contrast waveguides

Marcatili has developed a widely used analytical approach for low-index-contrast
waveguides in which all refractive indices nj have similar values [3, 15]. For

2-23
 Silicon Photonics

propagating modes in these guides, knj ≈ β because modes are not guided otherwise,
so kx, ky < <knj . Choosing Ey = 0 gives a modal field profile that is continuous on
the horizontal interfaces, while it approximately satisfies the conditions on the
vertical interfaces when neglecting terms of the order of (kx /knj )2 . However, for
high-index-contrast guides, these terms can be even larger than one.

Fields (2): continuity of dominant electromagnetic field components

This method demands continuity of dominant electromagnetic field components
(Ex, Hy) across all interfaces that they are tangential to. In this section, we consider
the mode to be ‘TE-like’, such as the mode depicted in figure 2.5. For ‘TM-like’
modes in wide waveguides, a similar method can be derived (see [16]). As typical
waveguides are larger in width than in height, the height gives the strongest
confinement, therefore, we chose to match all boundary conditions at the horizontal
interfaces. This choice is supported by the argument that the vertical sides are
irrelevant for an infinitely wide waveguide. With these requirements (including ky
and kx obtained from the slab eigenvalue equations), only one amplitude parameter
is left free, although we have not yet matched Ey, Ez and Hz on the vertical
interfaces. Of these field components we chose to match Ez because Hz vanishes in
the limit of an infinitely wide waveguide (b → ∞) and Ey is a weak field component
that is largest at the corners of the waveguide. With these requirements, the free
parameters in the ansatz are fully determined. Parameters kx, ky, ξ and η are given
by the slab eigenvalue equations (2.29) and (2.17). The amplitudes of the field
components, A2 A10, are given by equations (2.73)–(2.77), together with
A3 = A1 sin[kx(ξ − d /2)], (2.87)

⎛ k 2(n12 − n 22 ) ⎞
A4 = A2 ⎜1 + ⎟ cos[kx(ξ − d /2)], (2.88)
⎝ β2 ⎠

A5 = A1 sin[kx(ξ + d /2)], (2.89)

⎛ k 2(n12 − n 32 ) ⎞
A6 = A2 ⎜1 + ⎟ cos[kx(ξ + d /2)]. (2.90)
⎝ β2 ⎠

This method is also referred to as improved Ey ≈ 0 method, because Ey is zero in

regions 1, 4 and 5.

Fields (3): least-discontinuity optimization of the ansatz parameters

We presented an ansatz for the form of the electromagnetic field of modes in a
rectangular waveguide. This ansatz was chosen such that Maxwell’s equations are
satisfied in regions 1–5, so that all errors manifest themselves at the four interfaces
between the waveguide core and the cladding regions. The method we propose is to
minimize this error, by minimizing the discontinuity of the tangential electro-
magnetic field components at the interfaces. The measure we adopt to quantify

2-24
 Silicon Photonics

this error is the average energy density that is associated with these discontinuities,
defined by:
ϵ μ
Umm = 0 (n++n−)2 · νˆ × (E + − E −) 2 dl + 0
∮ ∮νˆ × (H + − H −) 2 dl . (2.91)
4l l
The four interfaces of the waveguide are all included in the integral. The line integral
runs along the entire circumference of the waveguide in the (x , y )-plane, and
l = 2(b + d ) is the length of this circumference. E + and E − are the electric fields
just outside and inside the waveguide core region 1, so that (E + − E −) represents the
discontinuity of this field, and ν̂ is a unit vector orthogonal to the waveguide surface.
The cross product of ν̂ with the discontinuity in the field selects the tangential
components. n+ and n− are the refractive indices just outside and inside the
waveguide. At the interface, an average refractive index (n+ + n−)/2 is assumed to
calculate the energy density of the electric field components. The error Umm can be
intuitively interpreted as an energy density and with this definition of the error we
guarantee that the electric and magnetic components get appropriate similar
weightings. However, we cannot attach a rigorous physical meaning to this error
Umm . The discontinuity of the fields only occurs at interfaces, which have no physical
volume. Therefore the energy density cannot be integrated over volume in order to
obtain a total energy.
The value of Umm is minimized as a function of the amplitudes A2–A10, while ky and
η are determined from equations (2.78) and (2.79) and kx and ξ are determined from
equations (2.85) and (2.86). We determined the minimum of Umm numerically using an
unconstrained non-linear optimization as implemented in Matlab [19]. As an initial
estimate we use the modal amplitudes as computed using the previously described
improved Ey ≈ 0 method. However, equation (2.91) is quadratic in the amplitudes
A2–A10 and the minimum can thus be found analytically. Method 3 is also referred to
as amplitude optimization method, because the amplitudes A2, . . . , A10 are optimized.

Comparison
We have presented one approximate method to compute the propagation constant,
or effective index, of the modes in rectangular waveguides. Given this propagation
constant, we presented three different methods that approximate the field of these
modes: (1) Marcatili’s original approach, (2) a method based on continuity of the
dominant electromagnetic field components, and (3) a method based on minimiza-
tion of the discontinuities of the electromagnetic fields.
In figure 2.7(b) and figure 2.7(d–f), we compare modal fields computed with the
approximate methods with the fields computed with a rigorous numerical mode solver.
The measure that is used to compare two electromagnetic fields is the relative energy of
the difference of the fields per unit of length in the propagation direction, i.e.,

∬regions 1−5 (n2ϵ0 EA − EN + μ0 HA − HN )dxdy

2 2

ΔU = (2.92)
∬regions 1−5 (n2ϵ0 EN + μ0 HN )dxdy
2 2

2-25
 Silicon Photonics

where EA and EN are the analytically and rigorous numerically calculated fields,
respectively. This integral runs over all regions that are described by the analytical
solution.
It is clear that something interesting happens for waveguides with dimensions
around 833 nm width by 300 nm height. This is at the point of the apparent crossing
of the propagation constants of the TM0 and TE1 modes (see figure 2.7(c)), and this
case is addressed in the next section. Apart from this special case, the relative errors
of the method with continuity of the dominant field components is below 3%, and
the relative error of the method in which the energy associated with the discontinues
is minimized is even lower. Both methods outperform Marcatili’s original approach
for these typical SOI waveguides with high index contrast. The amplitude optimi-
zation method, which minimizes the energy associated with the discontinuities, has
the advantage that it is works for both ‘TE-like’ and ‘TM-like’ modes.

2.4.4 Avoided crossing of modes with similar propagation constants

Figure 2.7(e,f) suggest that something interesting happens at the apparent crossing
of the effective indices of the TM0-like and the TE1-like modes when the width of
the guide is changed. A detailed inspection of the waveguides with widths around
833 nm is presented in figure 2.8. Figure 2.8(a) shows the effective indices of the
waveguide modes as computed using a rigourous numerical mode solver. It can be
seen that the effective indices of the 2nd and 3rd mode in the waveguide (counted
from high to low effective index) actually do not cross each other, but show a
behaviour that is known in quantum mechanics as avoided crossing [21]. We
investigate the modes that were found numerically in terms of the analytically
computed approximate modes [16].
We denote the actual, rigorous numerically computed, modes as EiN with i the
number of the mode. We will verify that the actual modes E2N and E3N can in good
approximation be written as a superposition of the approximate TM0-like, ETM0,

Figure 2.8. Investigation of the avoided crossing of effective indices of two modes. (a) Numerically calculated
effective indices of the 2nd and 3rd mode, zoom-in of figure 2.7(c). (b) Power in the TM0-like mode when the
fields of the modes in plot (a) are written as a superposition of a TM0-like and a TM1-like mode. The curves in
plot (a) are colour-coded accordingly. (c) Relative energy in the electromagnetic difference field between the
superposition and the rigorous numerically calculated fields, to be compared with figure 2.7 (e) and (f).

2-26
 Silicon Photonics

and TE1-like, ETE1 modes. The TM0-like and TE1-like modes were calculated using
the approximate amplitude optimization method. Thus
EiN ≈ aETM0 + bETE1, (2.93)

for some real a and b, and i = 2 or 3. The phase of mode Ei is chosen such that
coefficient b is positive. The coefficient a of the TM0-like mode can be either positive
or negative. The approximate calculated modes E TM0 and E TE1, are in good
approximation orthonormal such that normalization of the guided modes Ei in
the norm of equation (2.96) implies b = 1 − a 2 .
The coefficient a of the TM0-like mode is optimized such that the difference
measured using equation (2.92) between the left- and right-hand sides of equation
(2.93) is minimum. The result is plotted in figure 2.8(b), where it can be seen that
mode 2 looks like a TM0-mode at the left of the crossing, while it looks like a TE1-
like mode on the right-hand-side of the crossing, whereas close to the crossing the
modes are an equal mixture of ETM0 and ETE1. In figure 2.8(c), it can be seen that the
error between the superposition Ei and the rigorous numerically calculated field EiN
close to the apparent crossing is small and similar to the error that was found away
from the crossing (see figure 2.7(e), (f)). Therefore we may indeed conclude that the
field around the crossing can be written as a superposition of modes of the types that
are present away from the crossing. Figure 2.9 presents the electric fields ETM0, ETE1,
E2N and E3N for a 833 nm wide by 300 nm high waveguide, where a 2 ≈ b2 ≈ 0.5. At
avoided crossing, Marcatili’s method completely breaks down.
Using the previous observations, we will derive a qualitative description of this
avoided crossing. We consider forward propagating guided modes with positive βi .
The guided modes are normalized such that they carry unit power. In section 2.3, we
formulated Maxwell’s equations as an eigenvalue problem of operator Ô , with the
propagation constant βi as eigenvalue. The operator Ô is not symmetric. However,
forward-propagating guided modes are orthonormal with respect to the relation
given by equation (2.57). We adopt bra-ket notation where a mode with label i is
described by ∣i 〉, equation (2.37) is written as:
Oˆ ∣i〉 = βi ∣i〉, (2.94)
the scalar product 〈i∣ j〉 is defined by:
1
i j ≡
2
∬ (E1 × H2*) · zˆ dxdy, (2.95)

and the mode orthogonality (2.57) for normalized forward-propagating modes is

written as:
i j = δij. (2.96)

In this book, we informally call equation (2.95) a scalar product but in fact it does
not meet all the requirements to be a scalar product and the correct mathematical
description is a sesquilinear form.

2-27
 Silicon Photonics

Figure 2.9. Amplitudes of the electric field components of different modes in a 833 nm wide by 300 nm high
waveguide. The dashed lines separate the different regions (see figure 2.5). Modes E TM0 , E TE1, E2N and E3N are
plotted from top to bottom. Colour indicates the field strength, white regions have a low field strength. Regions
with positive field strength are indicated with a plus (+) sign and red. Regions with a negative field strength are
indicated with a minus sign ( ) and blue.

We now apply the aforementioned observation that, in good approximation, the

electromagnetic fields of the modes in the waveguide, also around the crossing, can
be written as a superposition of the approximate fields. Hence
Ei ≈ a ETM0 + bETE1, or i ≈aa +bb , (2.97)
where ∣a〉 and ∣b〉 represent the TM0-like and TE1-like modes in the waveguide,
respectively, while ∣i 〉 represents the exact solutions of equation (2.94). We only
consider the 2nd and 3rd approximate solutions here. As will become clear, only
modes with similar propagation constants have to be taken into account around the
crossing. The other modes are already accurately calculated by the approximate

2-28
 Silicon Photonics

methods presented in section 2.4.3. Substituting equation (2.97) in equation (2.94)

and taking the scalar product with 〈a∣ gives
ˆ + b a Ob
a a Oa ˆ ≈ β (a a a + b a b ). (2.98)
i

If we also take the scalar product of equation (2.94) with 〈b∣ we arrive at the
(2x2)-system:
⎛ a Oa
ˆ ˆ ⎞ a
a Ob ⎛ aa ab ⎞ a
⎜⎜
ˆ
⎝ b Oa
⎟⎟
ˆ ⎠ b
b Ob
()
≈ βi ⎜ ⎟
⎝ ba bb ⎠ b
, () (2.99)

or,

1 ⎛ b b a Oaˆ − a b b Oa
ˆ ˆ ⎞ a
ˆ − a b b Ob
b b a Ob a
D
⎜⎜
ˆ + a a b Oa
⎝− b a a Oa ˆ ˆ + a a b Ob
− b a a Ob
⎟
ˆ ⎟⎠ b
≈ βi
b () ()
, (2.100)

with
D= aa bb − ab ba .
The modes that we found in our approximate analysis are almost orthonormal, so
〈a∣a〉 and 〈b∣b〉 are approximately unity and 〈a∣b〉 and 〈b∣a〉 are approximately zero.
Away from the crossing, we found that the approximate solutions ∣a〉 and ∣b〉 obey
ˆ 〉 ≈ β , 〈b∣Ob
relation (2.94) so that 〈a∣Oa ˆ 〉 ≈ β , while 〈a∣Ob
ˆ 〉 and 〈b∣Oa
ˆ 〉 are small.
a b
This allows us to write equation (2.100) as
⎛ βa + δa δab ⎞ a a
⎜ ⎟
⎝ δba βb + δb ⎠ b
≈ βi
b ()
, () (2.101)

where δa , δb , δab, and δba are quantities that are much smaller than the β’s. This
system has the eigenvalues (labelled 2 and 3 because they correspond to the 2nd and
3rd modes of the waveguide) [21]

βa′ + βb′ (βa′ − βb′)2 + 4δabδba

β2,3 = ± , (2.102)
2 2
with βa′ ≡ βa + δa and βb′ ≡ βb + δb. We assume that modes 2 and 3 are guided hence
β2,3 is real and δabδba > 0. The eigenvectors v2,3 corresponding to the eigenvalues β2,3
are (not normalized)
⎛ 2δab ⎞
⎜⎜ ⎟. (2.103)
⎝− βa′ + βb′ ± (βa′ − βb′)2 + 4δabδba ⎟⎠

The two propagation constants are closest when βa′ = βb′ but are always separated
by a minimum distance 4 δabδba , so that they never intersect. For small
δa, δb, δab, δba ≪ ∣βa − βb∣, we find the eigenvector for βa > βb to be v2 ≈ (1, 0) and
v3 ≈ (0, 1). The upper propagation constant, β2 , has a TM0-like mode in this limit,

2-29
 Silicon Photonics

while the lower propagation constant, β3, has a TE1-like mode. For βb > βa we
find v2 ≈ (0, 1) and v3 ≈ (1, 0), so that the upper propagation constant now has a
TE1-like mode while the lower propagation constant has a TM0-like mode. An
interesting case occurs when βa′ = βb′ and δab = δba . Then the normalized eigenvec-
tors of this system are v2 = 1 · (1, 1) and v3 = 1 · (1, −1), i.e., they are an equal
2 2
superposition of the eigenvectors far from the crossing.
This simple description of the avoided crossing agrees with the observations of the
numerically computed modal profiles as presented in figure 2.8 and 2.9.

2.4.5 Dispersion: effective group index

With the analytical equation for the propagation constant β or effective index ne, it is
possible to analytically calculate the dispersion in the waveguide [16]. Linear
dispersion in waveguides is often described in terms of the effective group index,
ng. This is an important quantity which, for example, describes the free-spectral-
range (FSR) of ring resonators and influences the sensitivity of waveguide-based
sensors. Silicon-on-insulator waveguides have, in fact, a very strong modal dis-
persion due to the strong confinement of the light. The group index is according to
equation (2.32) defined by ng ≡ ∂β /∂k . From equation (2.70), we find

∂β 1⎛ ∂n ∂k ∂ky ⎞
= ⎜kn12 + k 2n1 1 − kx x − ky ⎟. (2.104)
∂k β⎝ ∂k ∂k ∂k ⎠

The 1st and 2nd term between the brackets on the right-hand-side of this equation
are determined by the refractive index of the guiding material. The 3rd term is
calculated from equation (2.29). Although kx is only given implicitly, ∂kx /∂k can be
calculated explicitly. The total derivative of the left-hand-side of equation (2.29)
with respect to k, dG/dk, equals zero for solutions of G = 0. The height d does not
depend on frequency. The refractive indices nj (k ) may depend on frequency and thus
on k = ω /c . So we get

dG ∂G ∂G ∂kx ∂G ∂n1 ∂G ∂n2 ∂G ∂n3

= + + + + , (2.105)
dk ∂k ∂kx ∂k ∂n1 ∂k ∂n2 ∂k ∂n3 ∂k

or,
∂G ∂G ∂n1 ∂G ∂n2 ∂G ∂n3
+ + +
∂kx ∂k ∂n1 ∂k ∂n2 ∂k ∂n3 ∂k
=− . (2.106)
∂k ∂G
∂kx
Similarly, the 4th term of the right-hand-side of equation (2.104) is calculated from
equation (2.17) as

2-30
 Silicon Photonics

∂F ∂F ∂n1 ∂F ∂n 4 ∂F ∂n5
+ + +
∂ky ∂k ∂n1 ∂k ∂n 4 ∂k ∂n5 ∂k
=− . (2.107)
∂k ∂F
∂ky

The partial derivatives in equations (2.106) and (2.107) are straightforward to

calculate.
For the case that only the core material is dispersive, i.e. n1(k ), and where the
other refractive indices do not depend on the frequency, thus also not on k, we define
⎛k 2 γ2 ⎞ 1
α2 ≡ ⎜⎜ x4 + 34 ⎟⎟ 2 , (2.108)
⎝ n1 n 3 ⎠ n 2 γ2

⎛k 2 γ2 ⎞ 1
α3 ≡ ⎜⎜ x4 + 24 ⎟⎟ 2 , (2.109)
⎝ n1 n 2 ⎠ n 3 γ3

k y2 + γ52
α4 ≡ , (2.110)
γ4

k y2 + γ42
α5 ≡ , (2.111)
γ5
to arrive at

∂kx ⎧ ⎛ ∂n ⎞
= ⎨kxk ⎜α2(n12 − n 22 ) + α3(n12 − n32 ) + (α2 + α3)n1k 1 ⎟
∂k ⎩ ⎝ ∂k ⎠
⎛γ γ ⎞⎡ 4 k 3 2k ⎛ k 2 γ γ ⎞ ⎤ ∂n ⎫ ⎪
+ ⎜ 22 + 32 ⎟⎢ 5x − x ⎜ x4 − 22 32 ⎟⎥ 1 ⎬ (2.112)
⎝ n2 n3 ⎠⎢⎣ n1 n1 ⎝ n1 n 2 n3 ⎠⎥⎦ ∂k ⎭
⎪

⎧
⎪⎛ γ γ3 ⎞⎛ k x2 γ2γ3 ⎞ ⎛ k x2 γ2γ3 ⎞
2 ⎫
⎪
−1

⋅⎨ ⎜ 2 + 2 ⎟⎜ 4 + 2 2 ⎟ + k x (α2 + α3) + n1 d ⎜ 4 − 2 2 ⎟ sec [kxd ]⎬

2 2 2 2 ,
⎩⎝ n 2 n3 ⎠⎝ n1 n 2 n3 ⎠ ⎝ n1 n 2 n3 ⎠
⎪ ⎪
⎭

and

∂ky
=
( ∂n
kyk α 4(n12 − n 42 ) + α5(n12 − n 52 ) + (α 4 + α5)n1k ∂k1 ) . (2.113)
∂k (γ4 + γ5)(ky2 + γ4γ5) + ky2(α 4 + α5) + b(ky2 − γ4γ5)2 sec 2⎡⎣kyb⎤⎦

Figure 2.10 depicts the effective group index of typical SOI waveguides with heights
of 220 nm and 300 nm. The approximate analytical curve is computed using the
equations above where the dispersion of silicon is taken into account using
∂n1/∂k = 3.147 · 108 m−1 with k = 2π /λ for λ = 1550 nm [22]. These curves are

2-31
 Silicon Photonics

Figure 2.10. Effective group indices. Approximate analytical model compared with rigorous mode solver.
Typical rectangular silicon-on-insulator waveguides with air cladding. (a) 220 nm high, fundamental mode.
(b–d) 300 nm high, first 3 modes.

compared with a rigorous computation using the FMM method as implemented in

FimmWave. For the typical SOI waveguides considered in figure 2.10, it can be seen
that the error remains below 4%. A comparison with experimentally measured
group indices of SOI waveguides may be found in [23, section 3.8].

2.5 Rigorous numerical mode-solvers

Exact analytical solutions, i.e., closed form solutions, for the guided modes exist only
for some waveguide shapes, such as a slab waveguides or circular waveguides. For
waveguides with a rectangular cross-section, approximate models such as Marcatili’s
method and its modifications exist (see section 2.4). However, actual waveguides
might have different shapes, for example, silicon waveguides are most often
trapezoidal instead of rectangular due to the fabrication process. Moreover, wave-
guides for evanescent field sensing are designed to have even more special shapes to

2-32
 Silicon Photonics

maximize the overlap of the modal field with the material that is to be sensed.
Rigorous numerical mode solvers can handle arbitrary shaped waveguides including
losses and bends, and are therefore often used in the design of photonic waveguides.
We have used two different numerical mode solvers: the film mode-matching
(FMM) method and the finite element method (FEM), both implemented in the
FimmWave software package by Photon Design (Oxford, UK) [24, 25].
The FMM method is very suitable to solve waveguide geometries in which the
waveguide is built up from a number of vertical slices (such as rectangular
waveguides or directional couplers). In this method, the cross-section of the ridge
waveguide is divided in vertical slices, and 1-D modes are computed analytically for
each slice. The 2-D modes are found by finding a set of coefficients of the 1-D modes
that will give a field profile obeying Maxell’s equations everywhere. For the
simulations in this book, the area of the numerical simulation extends 2 μm from
the waveguide, and 200 1-D modes are used per slice.
In the finite element method (FEM), Maxwell’s equations for the modes of a
waveguide are discretized and the modes are solved on a grid. The FEM
implementation in FimmWave uses first- and second-order finite elements. The
triangular grid is automatically chosen such that it aligns with the waveguide
structure. For the simulations in this book, ∼210 grid points are used in both the
x- and y-direction.
The FMM method and the finite element method (FEM) agree very well for the
typical SOI waveguides considered in figure 2.7. The difference (between FMM and
FEM) in effective index is below 10−3 and the relative energy in the difference field is
below 10−4 .

2.6 Typical silicon-on-insulator waveguides

In this section we discuss some typical characteristics of rectangular silicon-on-
insulator waveguides, namely: (1) the uncertainty in the propagation constant, (2)
the effects of slightly slanted sidewalls, (3) the propagation loss due to sidewall
roughness, and (4) the wavelength-dependence of the effective index.
Fabrication of sub-wavelength silicon waveguides is not straightforward and real
waveguides differ from the designed ones (section 5.3). The most standard wave-
guide in SOI technology is rectangular, because it is simple and has particular
advantages. The mode is strongly confined in this waveguide, allowing for sharp
bends. The height of the waveguide is solely defined by the thickness of the silicon
layer, and thus does not require etch processes which may cause variations or
roughness. A drawback of this waveguide type is that the strong confinement also
causes the effective index to be sensitive to fabrication-induced variations partic-
ularly in the width of the guides [26, 27]. This is a problem for some devices such as
arrayed waveguide gratings.
Although the waveguides are intended to be rectangular, the sidewall angle is
often about 10 degrees, hence the bottom of a 220 nm high waveguide can be 78 nm
wider than its top (figure 2.11(a)) [28]. This is caused by the etch process in the
CMOS fabrication of the waveguides. Rigorous numerical simulations can be used

2-33
 Silicon Photonics

Figure 2.11. Trapezoidal waveguides with 10 ◦ sidewall angle compared with rectangular waveguides.
Silicon-in-SiO 2 waveguides with a height h of 220 nm. Silicon-dioxide cladding. Film mode matching
(FMM) method is used as mode-solver for the rectangular waveguides. Finite element method (FEM) is used
as mode-solver for the trapezoidal waveguides because this method handles the trapezoidal structures more
accurately (see section 2.5). (a) Sketch of the cross-section of a trapezoidal waveguide. (b) Sketch of the
cross-section of a rectangular waveguide, width w is equal to the average width of the trapezoidal waveguide
so that the areas of the two waveguide cross-sections are identical. (c) Effective index. (d) Effective group
index.

to show that the effective index and the effective group index of trapezoidal
waveguides are very well approximated by rectangular waveguides that have a
width equal to the average width of the trapezoidal guide. In figure 2.11 trapezoidal
waveguides are compared with rectangular waveguides. For silicon waveguides
embedded in silicon-dioxide with height 220 nm and widths varying from 400 nm to
1000 nm, the effective index and effective group index agree within 0.1%. Hence the
effective index and the effective group index of typical trapezoidal SOI waveguides
can be very well approximated by those of rectangular waveguides.
The propagation losses αp of SOI waveguides have many causes: linear and non-
linear absorption9 in the material (both in the bulk as well as at the interfaces
between materials), leakage into the silicon substrate, scattering from small defects
in the material, and scattering from roughness of the silicon–silica interfaces of the
waveguide. Dry etching, which is used to fabricate these guides, creates sub-
wavelength roughness at the sidewalls of the waveguides. This is the dominant
loss mechanism in sub-wavelength silicon waveguides [26]. The high index contrast
of waveguides in SOI technology allows small bending radii (3 μm) with reasonably
small propagation losses. These losses originate not only from radiation loss due to
the curvature in the waveguide. The field of the mode of a bend waveguide is pushed
outwards towards the outer side of the waveguide. This increases losses due to
sidewall roughness because the field intensity at the sidewalls is higher compared to a
straight waveguide. This also increases substrate leakage because the mode is less

9
Linear and non-linear material absorption correspond to the imaginary parts of the linear and non-linear
terms of the susceptibility χ (i ) . See, e.g. [1]. For linear absorption, waveguide propagation loss may be
described by an attenuation coefficient αp as in equation (2.66), while this coefficient depends on the intensity of
the electromagnetic field for non-linear absorption.

2-34
 Silicon Photonics

confined. Moreover, sidewalls are typically not perfectly vertical, which introduces
TE/TM conversion in the bends; this conversion gives additional loss [28].
Moreover, there is a mismatch between the mode of the straight waveguide and
the mode of the bend waveguide, especially for smaller bending radii. This causes
losses at the interfaces between the straight and the curved waveguides. Therefore,
the loss of two 90◦ turns separated by a straight guide is not necessarily the same as
the loss of a single 180◦ turn. Finding the modes of an ideal waveguide without losses
can be done with high accuracy, but the calculation of loss mechanisms is relatively
difficult [26].
Light is often coupled from optical fibers or from a free-space beam to the silicon
photonic chip using out-of-plane grating couplers. Such couplers have a bandwidth
or wavelength span of approximately 30 nm around a center wavelength λc of 1550
nm [29]. In this regime, the wavelength-dependence of the effective index can be
approximated as linear, so that we may use equation (2.33).

References
[1] Leuthold J, Koos C and Freude W 2010 Nonlinear silicon photonics Nat. Photon. 4 535–44
[2] Griffiths D J 1999 Introduction to Electrodynamics 3rd edn (Upper Saddle River, NJ:
Prentice-Hall)
[3] Marcuse D 1991 Theory of dielectrical optical waveguides 2nd edn (San Diego, CA:
Academic)
[4] Kogelnik H 1975 Theory of dielectric waveguides Integrated Optics (Springer Series on
Topics in Applied Physics vol 7) ed T Tamir (Berlin: Springer) pp 13–81
[5] Reed G T and Knights A P 2004 Silicon Photonics: An Introduction (Chichester: Wiley)
[6] Woan G 2000 The Cambridge Handbook of Physics Formulas (Cambridge: Cambrige
University Press)
[7] Marcuse D 1982 Light Transmission Optics 2nd edn (New York: Van Nostrand Reinhold)
[8] Bienstman P 2001 Rigorous and efficient modelling of wavelength scale photonic compo-
nents PhD Dissertation, Universiteit Gent www.photonics.intec.ugent.be/publications/phd.
asp?ID=104
[9] Sneyder A W and Love J D 1983 Optical Waveguide Theory (London: Chapman and Hall)
[10] Weisstein E W 2013 Divergence theorem. MathWorld–A Wolfram Web Resource http://
mathworld.wolfram.com/DivergenceTheorem.html
[11] Hardy A and Streifer W 1985 Coupled mode theory of parallel waveguides J. Lightwave
Tech. 3 1135–46
[12] Hardy A 1998 A unified approach to coupled-mode phenomena IEEE J. Quantum Electron.
34 1109–16
[13] Hardy A and Ben-Artzi M 1994 Expansion of an arbitrary field in terms of waveguide modes
IEE Proc. – Optoelectron. 141 16–20
[14] Hunsperger R G 2002 Integrated Optics: Theory and Technology (Berlin: Springer)
[15] Marcatili E A J 1969 Dielectric rectangular waveguide and directional coupler for integrated
optics Bell Syst. Tech. J. 48 2071–121
[16] Westerveld W J, Leinders S M, van Dongen K W A, Urbach H P and Yousefi M 2012
Extension of marcatilias analytical approach for rectangular silicon optical waveguides J.
Lightwave Tech. 30 2388–401

2-35
 Silicon Photonics

[17] Westerveld W J, Leinders S M, van Dongen K W A, Yousefi M and Urbach H P 2012

Extension of marcatili’s analytical approach for 220 nm high waveguides in SOI technology,
in Ann. Symp. of the IEEE Photonics Society Benelux Chapter (Mons, Belgium) pp 25–8.
[18] Melloni A, Roncelli D, Morichetti F, Canciamilla A and Bakker A 2009 Statistical design in
integrated optics CLEO/Europe and EQEC 2009 Conf. Digest (Washington, DC: Optical
Society of America) p JSI1_4
[19] 2017 Matlab—the language of technical computing. The MathWorks, Inc. Natick, MA,
USA www.mathworks.com/matlab
[20] Westerveld W J 2013 RECTWG: Matlab implementation of the extended Marcatili
approaches for rectangular silicon optical waveguides (free open-source software).
Distributed in RECTWG package for Matlab—Version 0.1. Delft University of
Technology http://waveguide.sourceforge.net
[21] Landau L D and Lifshitz E M 1977 Quantum Mechanics 3rd edn (Oxford: Pergamon Press)
sec. 79
[22] 2011 Material database and material models, distributed with FimmWave software package
(Oxford: Photon Design)
[23] Westerveld W J 2014 Silicon photonic micro-ring resonators to sense strain and ultrasound
PhD Dissertation Technische Universiteit Delft http://repository.tudelft.nl
[24] Sudbo A S 1994 Improved formulation of the film mode matching method for mode field
calculations in delectric waveguides Pure Appl. Opt. A 3 381
[25] Photon Design 2012 Fimmwave, a powerful waveguide mode solver (Oxford: Photon
Design) www.photond.com/products/fimmwave.htm
[26] Bogaerts W, de Heyn P, van Vaerenbergh T, de Vos K, Kumar Selvaraja S, Claes T, Dumon
P, Bienstman P, van Thourhout D and Baets R 2012 Silicon microring resonators Laser
Photon. Rev. 6 47–73
[27] Xu D-X, Schmid J, Reed G, Mashanovich G, Thomson D, Nedeljkovic M, Chen X, van
Thourhout D, Keyvaninia S and Selvaraja S 2014 Silicon photonic integration platform–
have we found the sweet spot IEEE J. Sel. Top. Quantum Electron. 20 189–205
[28] Selvaraja S K, Bogaerts W and Thourhout D V 2011 Loss reduction in silicon nanophotonic
waveguide micro-bends through etch profile improvement Opt. Commun. 284 2141–4
[29] Roelkens G, Vermeulen D, Selvaraja S, Halir R, Bogaerts W and van Thourhout D 2011
Grating-based optical fiber interfaces for silicon-on-insulator photonic integrated circuits
IEEE J. Sel. Top. Quantum Electron. 17 571–80

2-36
 IOP Publishing

Silicon Photonics
Electromagnetic theory
Wouter J Westerveld and H Paul Urbach

Chapter 3
Components

3.1 Directional couplers

Directional couplers can be used to couple a fraction of the light from one
waveguide to another, for example from a straight waveguide to a ring resonator.
This section starts with an intuitive introduction to the behaviour of such couplers.
Then three methods to calculate the behaviour of directional couplers are presented
and compared: coupled mode theory (section 3.1.1 and appendix A), eigenmode
expansion (section 3.1.2) and rigorous numerical FDTD simulations (section 3.1.3).

3.1.1 Coupled mode theory

A directional coupler consists of two parallel single-mode waveguides so close to
each other that a significant amount of power couples from one waveguide to the
other (see figure 3.1). To describe this system with coupled mode theory, we assume
that the electric field in the coupler can be approximated by a superposition of the
two modes of the isolated waveguides. The amplitudes of the two modes vary while
propagating through the parallel waveguides due to the coupling (i.e. light ‘leaks’
from one mode to the other). The coupled mode theory is derived in appendix A and
we present only the most important results in this section. The electromagnetic field
is approximated by
E c(x , y , z , t ) ≈ E a(x , y )ua(z )e ıωt + E b(x , y )ub(z )e ıωt , (3.1)
with ua(z) and ub(z) the complex modal amplitudes of waveguides a and b,
respectively, E a(x , y ) and E b(x , y ) the modal electric fields of the waveguides and
E c(x , y , z, t ) the total electric field. Let us consider the result of an excitation of
mode b at z = 0 (i.e. all light is in waveguide b). The transmission of a coupler with
length L̃ is given by ub(L˜ ) = τub(0), while the coupled light is given by
ua(L˜ ) = κub(0). These complex amplitudes τ and κ can be calculated using coupled
mode theory (see appendix A)

doi:10.1088/978-0-7503-1386-5ch3 3-1 ª IOP Publishing Ltd 2017

 Silicon Photonics

Figure 3.1. Sketch of a directional coupler consisting of two parallel waveguides. We consider couplers with a
silicon-dioxide cladding. (a) Cross-section of the coupler. Two 440 nm × 220 nm rectangular silicon
waveguides are separated by 200 nm. Refractive index of the guides is n1 and the SiO2 cladding has index
n2. (b) Top-view including the bends. Upper waveguide a and lower waveguide b. (c) Optical microscope
photograph of a directional coupler in SOI. The very narrow pink lines are the waveguides.

⎛ ıδ ⎞
τ = ⎜cos sL˜ − sin sL˜ ⎟e−ı(βb +κbb−δ )L˜ , (3.2)
⎝ s ⎠

⎛ ıκ ⎞
κ = −⎜ ab sin sL˜ ⎟e−ı(βb +κbb−δ )L˜ , (3.3)
⎝ s ⎠

where βb is the propagation constant of mode b, κbb is the correction to this

propagation constant originating from the other waveguide as defined in equation
(A.37), δ is half the difference between the corrected propagation constants of the
guides defined by
1
δ≡ (β + κbb − βa − κ aa ), (3.4)
2 b
while s is the coupling coefficient defined by

s= κbaκ ab + δ 2 . (3.5)

Coupling coefficient κab represents the coupling from the mode of waveguide b to the
mode of waveguide a and κba represents the coupling from the mode of waveguide a
to the mode of waveguide b. These coefficients dominate s and are defined in
equation (A.37). The guides in the coupler that we study are designed to be identical,
but we experimentally observed non-zero δ in our couplers. Equation (3.2) is valid
for two parallel waveguides, whereas the actual coupler also includes bends to
connect the parallel waveguides to the components in the circuit. We take the
coupling which happens in the bends into account by re-defining the length L̃ in
equation (3.2) as an effective coupling length L˜ = L + ΔL , with L the length of the
parallel part of the waveguides. Figure 3.2(a) illustrates the behaviour of a direc-
tional coupler in which light ‘leaks’ from waveguide b (power ∣ub∣2 ) to waveguide a.
For two different waveguides (non-zero δ), the power never fully transfers from one
waveguide to the other (figure 3.2(b)). In appendix A, we derive equation (3.2) by
following the approach of Hardy and Streifer [1, 2]. We also show in appendix A
that this method agrees well with rigorous FDTD simulations also for high-index-
contrast SOI waveguides.

3-2
 Silicon Photonics

Figure 3.2. (a) and (b) Behaviour of a directional coupler. Power in upper waveguide a and lower waveguide b.
At z = 0, all power is in the lower guide b. (a) Coupling coefficient s = 0.1, identical waveguides, δ = 0 . (b)
Coupling coefficient s = 0.1, different waveguides, δ = 0.02 . (c) Coupling coefficient calculated using three
different methods (see legend). The five groups of lines correspond to waveguide widths: 380 nm, 400 nm,
420 nm, 440 nm and 460 nm (top to bottom). The rectangular silicon waveguides are 220 nm high and have
silicon-dioxide cladding.

When using the couplers not for a single wavelength λc but for a range of
wavelengths, it is necessary to study how the behaviour of the coupler depends on
the wavelength. Dispersion is taken into account by assuming linear dispersion of
the effective index (hence the propagation constant βb(λ ) is given by equation (2.33))
and by assuming linear dispersion of the coupling s(λ ). With the definition of L̃ , it is
not necessary to include dispersion in ΔL . We found that dispersion in ΔL , and
higher order dispersion in s, are small and below our numerical and experimental
noise (see section 3.1.3). An experimental example of the behaviour of directional
couplers may be found in [3, section 3.6].

3.1.2 Eigenmode expansion (EME)

The directional coupler may be considered as one single waveguide with z-invariant
refractive index profile ϵc(x , y ) consisting of two disconnected parts. A coupler that
consists of two identical single-mode waveguides has two modes: a symmetric mode
(labeled 0) and an anti-symmetric mode (labeled 1). These are pure ‘waveguide’
modes, in the form of equation (2.30), which propagate in the z-direction with
propagation constants β0 and β1 and without distortion. Hence
E c(x , y , z , t ) = u 0E (0)(x , y )e ı(ωt−β0z ) + u1E (1)(x , y )e ı(ωt−β1z ). (3.6)
The relative phase between modes 0 and 1 changes with z due to the different
propagation constants β0 and β1. After a certain propagation distance Lπ, the

3-3
 Silicon Photonics

relative phase of the two modes is changed by π. In equation (3.6), the relation to
the individual waveguides a and b is not specified. However, we may approximate the
mode of isolated waveguide a by adding modes 0 and 1. We may approximate the
mode of isolated waveguide b by subtracting mode 1 from mode 0 (see figure 3.3).
Thus the excitation of the ‘mode’ of waveguide b at z = 0 can be approximated by
u 0 = 1/2 and u1 = − 1/2 . After propagating a distance Lπ, the sign of one mode 1
has changed, so that all light is now approximately in the ‘mode’ of waveguide a. The
smallest length over which all light transfers from waveguide b to waveguide a is hence
given by Lπ = π /(β0 − β1).
In the coupled mode theory, equations (3.1)–(3.3), it was found that all light
transfers from mode b to mode a when ∣ub(L˜ )∣ = cos(sL˜ ) = 0, thus when sL˜ = π /2.
This length L̃ is thus the same length as Lπ. This allows us to define a coupling
coefficient sEME as found with the eigenmode expansion (EME) method as
β0 − β1
s EME = . (3.7)
2
In the implemenation of the EME method, we calculated the coupling coefficients
sEME from the propagation constants ( β1 and β0 ) that were calculated with a rigorous
numerical mode-solver (FMM method). The calculated coupling coefficients sEME
agree well with coupled mode theory and with rigorous FDTD simulations
(see figure 3.2(c)).
As discussed in section 2.6, the side-walls of waveguides in SOI technology are not
perfectly vertical but have an angle of ∼10◦, hence the waveguides are trapezoidal.

Figure 3.3. Two descriptions of the modes in a directional coupler, modal electric field, Ex(x, y ) at y = 0. The
upper two curves show the modes of isolated waveguides a and b. The fields in the directional coupler can be
approximated as a superposition of those two modes with z-dependent amplitude coefficients ua(z) and ub(z)
(coupled mode theory). The lower two curves are exact solutions of the directional coupler ‘waveguide’
(eigenmode expansion method). Symmetric mode (0) and anti-symmetric mode (1). Modal profiles obtained
using rigorous mode-solver (FMM method in FimmWave).

3-4
 Silicon Photonics

It was shown that the effective index index of typical trapezoidal SOI waveguides are
well approximated by a rectangular waveguide with a width equal to the average
width of the trapezoidal waveguide. We now study the influence of the side-wall
angle of the waveguides on directional couplers using the EME method. We
computed the coupling coefficients sEME of couplers with trapezoidal and rectan-
gular waveguides, to find that they agree within 1% (for waveguide widths varying
from 380 nm to 480 nm, gaps varying from 140 nm to 240 nm, and 10◦ side-wall
angle for the trapezoidal guides). Propagation constants β0 and β1 were, for both
cases, computed with the finite element method (FEM) mode-solver.

3.1.3 Rigorous FDTD simulations

We have used rigorous finite difference time domain (FDTD) simulations to
calculate the transmittance of light through the directional coupler. Rigorous means
that we calculate the solution to Maxwell’s equations without approximations.
Space and time are discretized, and the behaviour of the electromagnetic fields over
time is calculated using Maxwell’s equations. This means that the solutions are exact
when using an infinitely small grid-size, an infinitely small time-step, exactly known
initial conditions, and an infinitely accurate computer. In reality, however, compu-
tation times and computer memory are limited of course and therefore so is the
accuracy that can be obtained (see [4, 5] for FDTD).
We have used CrystalWave, a commercial FDTD implementation by Photon
Design (Oxford, UK) [6]. For the simulations of the directional coupler, we used a
simulation domain that extends 1 μm above and below the waveguides (see figure 3.4)
and a grid-spacing of 20 nm for vacuum wavelengths around 1550 nm. The grid was
aligned to the parallel waveguides, such that the refractive index profile in this region
is not discretized or averaged. The reflections from the borders of the simulation

Figure 3.4. Set-up of the directional coupler FDTD simulation, based on screen-shots of CyrstalWave. (a)
Top-view with the exciter and the sensors in, cross, and through, indicated. (b) Cross-section at constant z
depicting the two waveguides.

3-5
 Silicon Photonics

volume was minimized using perfectly matched layers [6]. A TE-like mode is excited in
the lower waveguide (see figure 3.4) with a time-pulse that consists of a sinusoidal
signal (free-space wavelength 1550 nm) with an envelope that has a bandwidth of
200 nm (free-space wavelength). The electromagnetic energy flux through the rectan-
gular surfaces of the in-, through- and cross-sensors are recorded (see figure 3.4). The
sensors record the electromagnetic fields versus time and a Fourier transform is used to
find the energy flux as function of frequency or free-space wavelength. The power
inserted in waveguide b was normalized to one using the recordings of the in-sensor.
Equation (3.3) is used to study the behaviour of the couplers. With unit power in
waveguide b before the coupler, we write the power in the waveguide a after the
coupler, ∣ua(L˜ )∣2 , as
ua(L˜ ) 2 = A0 sin2[s(L + ΔL )]. (3.8)
For a lossless coupler as described by equation (3.3), A0 = ∣κab /s∣2 , but radiation loss
in fact also influences A0. We performed a series of FDTD simulations with length L
varying from 0 μm to 18 μm and recorded the power in waveguide a after the
coupler, ∣ua(L˜ )∣2 for each length L. Then unknowns in equation (3.8), s, ΔL and A0,
are found by fitting this equation to the ∣ua(L˜ )∣2 versus L curve. An example of the
results of such a FDTD simulation and fitting is shown in figure 3.5(a), where

Figure 3.5. Example of FDTD simulations of a directional coupler (waveguide width 440 nm, gap 200 nm).
Lengths L are: 0, 4, 8, 12, and 16 μm. (a) Power flow through the sensors, from FDTD (markers) and fitted
(curves). (b–c) Fitted values of coupling coefficient s, transmittance A0 and correction for coupling in the bends
ΔL . Plotted versus wavelength to investigate both dispersion and noise.

3-6
 Silicon Photonics

markers show the FDTD results and the solid line is a result of the fitting. For all
simulations, A0 ≈ 1, and hence radiation loss and asymmetries between the wave-
guides quantified by δ can be neglected.
The numerical error in the FDTD computation is estimated by inspecting the
results at different frequencies or free-space wavelengths (figure 3.5(b–d)). The
curves in this figure are expected to be smooth because the wavelength variation is
small, there are no resonant interference effects in the directional coupler, and the
characteristics of the coupler are smooth with respect to length L and gap g.
Therefore the variation in figure 3.5(b–d) are interpreted as computational error. It
may be seen that the numerical error in A0, s, and ΔL are on the order of 1%, 4% and
5%, respectively.
The power going straight through waveguide b (the straight waveguide) as
computed with the FDTD method is shown by circles in figure 3.5(a). We compare
this recorded power with the parameters (s, ΔL and A0) that are obtained using the
recordings of the power coupled to waveguide a (the curved waveguide). For lossless
couplers, the straight-through power in waveguide b is: ∣ub(L˜ )∣2 = 1 − ∣ua(L˜ )∣2 ,
which is plotted as the dashed line in 3.5a. There is good agreement between the
dashed line and the circles.
In figure 3.5(b, d), it can be seen that the dispersion in s is linear while the
dispersion in ΔL is negligible. A linear fit is used to find linear dispersion ∂s/∂λ in the
regime from 1525 nm to 1575 nm free-space wavelength. Figure 3.6 presents the

Figure 3.6. Characteristics of directional couplers extracted from FDTD simulations. Labels of vertical axis
are above the plots. Characteristics as a function of gap g and for different waveguide widths, namely 380 nm
(red continuous), 400 nm (blue dashed), 420 nm (green dashed), 440 nm (magenta dash-dot), and 460 nm (cyan
continuous with dots). (a) coupling coefficient s. (b) Correction for coupling in the bends ΔL (c) Dispersion in
the coupling coefficient ∂s/∂λ .

3-7
 Silicon Photonics

simulated characteristics of directional couplers with different waveguide widths and

separation gaps.

3.2 Multimode interference (MMI) couplers

In this section we introduce the multimode interference coupler (MMI) which is used
to couple light from an input waveguide to one or more output waveguides thereby
operating as a splitter. An MMI coupler basically consists of a wide multimode
waveguide with multiple narrower waveguides at the input and output sides (see
figure 3.7). In this section, we consider a multimode interference coupler that is
fabricated in silicon-on-insulator technology with a 220 nm high silicon light-guiding
layer on top of a silicon-dioxide cladding (see figure 3.7). This light-guiding slab only
supports the fundamental transverse-electric (TE) mode.
In sections 3.2.1–3.2.4, we loosely follow the paper of Soldano and Pennings [7]
and describe the behaviour of the the MMI by an approximate analytical eigenmode
expansion. This approximate method gives intuitive insight into the ‘imaging’ of the
MMI. However, the comparison with a rigorous numerical eigenmode expansion
(EME) method in section 3.2.5 reveals that the approximate propagation constants
are not sufficiently accurate to calculate quantitative results. The approximate
method is easily improved by inserting propagation constants that were computed
using rigorous numerical methods or the extended Marcatili method.

3.2.1 Approximate description of the guided modes

In this and following sections, we derive an approximate analytical description of the
MMI coupler. In this section, we approximate the shape of the waveguide modes
based on the facts that (1) the 220 nm thin silicon layer only guides a single TE-mode
and (2) the considered waveguides are larger in width than in height due to which the
the modes are most confined in the y-direction. Given this configuration we will first
approximate the waveguide modes as pure transverse electric (TE) and thereafter

Figure 3.7. Sketch of a multimode interference (MMI) coupler. (a) 3D-sketch. (b) Top-view showing wide
MMI waveguide (width w, length L), four access waveguids P1, P2, P3, and P4 each with width wA and
centered at x = ±xc . Input field Ψ0(x ) is sketched at the interface between access waveguide P1 and the wide
MMI waveguide. Modes ψν(x ) for ν = 0, 1, 2 and 3 are also sketched in the wide MMI waveguide, at arbitrary
z-positions.

3-8
 Silicon Photonics

choose an approximate description of the waveguide modes that allows us to derive

the MMI behaviour. The resulting approximate mode shape is used to compute the
modes in the wide MMI waveguide (w ∼ 4 μm) as well as in the access waveguides
(w ∼ 1 μm).

Transverse electric (TE) guided modes

The waveguide modes are approximated as pure transverse electric (TE) which is a
good approximation for low-order modes in wide waveguides (remember that an
infinitely wide slab waveguide has pure TE and TM modes). This mode has Ex, Hy
and Hz as non-zero field components and the electromagnetic fields are fully defined
by Ex because H can be computed using Faraday’s law (2.1). For a TE mode with
propagation constant β, the electromagnetic fields take the form (see section 2.2.1)
E (ν )(x , y , z , t ) = Ex(ν )(x , y )e ı(ωt−βν z )xˆ , (3.9)

H(ν )(x , y , z , t ) = H y(ν )(x , y )e ı(ωt −βν z )yˆ + Hz(ν )(x , y )e ı(ωt −βz )zˆ. (3.10)

Employing Faraday’s law (2.1) to the electromagnetic fields of TE modes as defined

in equations (3.9–3.10) we find
β
Hy = Ex. (3.11)
μ0ω

The power P that is carried by the mode is computed by inserting equation (3.11) in
equation (2.60),
β
P=
2μ 0 ω
∬ Ex · Ex* dxdy. (3.12)

The orthogonality relation for forward propagating modes, equation (2.57), may
also be simplified for TE modes,

∬ Eν,x · E μ*,x dxdy = 0 if βμ ≠ βν . (3.13)

Approximate shape of the modes

Next, we derive a convenient approximate description of the waveguide modes.
Similar to the extended Marcatili method, the shape of the electromagnetic fields
in the core of the waveguide is approximated as sinusoidal in both the x- and the
y-directions (see figures 2.5 and 2.6). Also similar, the profiles above and below the
core of the waveguide are approximated as exponentially decaying (regions 4 and 5
in the figures). In contrast to the extended Marcatili method, we set the electric field
at the left- and right-hand-sides of the core to zero (regions 2 and 3). Using this
approximation we are able to analytically reveal qualitatively the ‘imaging’
behaviour in the MMI coupler. However, in section 3.2.5 it is shown that this
approximation is not sufficiently accurate for the design of MMI couplers. As a first
correction, an effective width of the MMI waveguide is used to correct for the

3-9
 Silicon Photonics

exponential decay outside the core of the waveguide, known as the Goos–Hänchen
shift. However, then the same correction is applied to all modes and this is the cause
of why the computed propagation constants are still not sufficiently accurate.
At the horizontal interfaces of the waveguides, continuity is demanded of the
electromagnetic components that are tangential. This results in a modal electric field
that has the same profile in the y-direction as the TE mode of the corresponding slab
waveguide with the same height but with infinite width. Hence the modal electric field
Ex(ν )(x , y ) = ExSLAB( y )ψν(x ), (3.14)

where ExSLAB( y ) is the profile of the fundamental TE mode of the slab waveguide as
defined in equation (2.10) with equations (2.13), (2.15) and (2.17). As introduced
before, we now set Ex = 0 in the regions left (2) and right (3) of the waveguide. Then
ψν(x ) = 0 at the interfaces of the core of the waveguide, x = ±w/2. With the assumed
sinusoidal shape, this gives
⎧ ⎡ ⎛ ⎞⎤
⎪sin⎢kx,ν⎜x + w ⎟⎥ , − w < x < w
ψν(x ) = ⎨ ⎣ ⎝ 2 ⎠⎦ 2 2 (3.15)
⎪
⎩ 0, otherwise
with
π (ν + 1)
kx,ν = , ν = 0, 1, 2, ... m − 1, (3.16)
w
with ν the mode number and m the number of guided modes in the waveguide (also
see figure 3.7).
Upon substitution of equations (3.15), (3.14), (3.9) and (2.10) in the wave
equation (2.5) and considering the core of the waveguide, we arrive at the dispersion
relation
βν2 = n 2k 2 − k y2 − k x2,ν = n r2k 2 − k x2,ν, (3.17)
where
nr ≡ n 2 − k y2 / k 2 , (3.18)
with βν the modal propagation constant, n the refractive index of the silicon
waveguide core, nr the effective refractive index of the slab waveguide, and ky given
by equation (2.17). The derived method has some resemblance to the so-called
effective index method1 [8, section 2.8], because the 3D problem is reduced to a 2D

1
In the effective index method, the horizontal piecewise constant ( y, z ) areas with same vertical layer stack
n(x) are represented by the corresponding slab waveguide effective index nr. When applying this method to a
rectangular waveguide, it also has the shape of the ansatz in figure 2.6 but with parameters different from the
extended Marcatili method and with less accurate propagation constant β. For the SOI waveguide described
here, first the effective index nr of TE mode in the 220 nm high slab waveguide is computed. Then, the effective
index of the rectangular waveguide is approximated by the effective index of the TM mode of a slab waveguide
with left refractive index n2, guiding layer refractive index nr with thickness d, and right refractive index n3 (c.f.
figure 2.5).

3-10
 Silicon Photonics

problem by incorporating the effect of the slab waveguide in effective refractive

index nr.
For lower order modes, the propagation constant is much larger than the lateral
wave number (βv2 ≫ k x2,v ) and equation (3.17) with equation (3.16) may be
approximated by
π 2(ν − 1)2
βν ≈ n rk − . (3.19)
2n rkw 2
Henceforth we define the beat length of the two lowest order modes, Lπ, as the length
over which these modes shift exactly out of phase, i.e., πLπ = (β0 − β1). For the
example of a 4.2 μm MMI waveguide, Lπ = 43 μm. With the propagation constants
approximated as in equation (3.19) we find
2n rkw 2
Lπ ≈ , (3.20)
3π
and
πν(ν + 2)
β0 − βν ≈ . (3.21)
3Lπ

Mode normalization
For the approximate description of the guided TE modes, i.e., in the form of
equation (3.14), the power is computed as
β
P=
2μ 0 ω
∫ ExSLAB( y ) · ExSLAB (y)* dy · ∫ ψν(x) · ψν(x)* dx. (3.22)

In this section 3.2 it is more convenient to normalize the modes by taking equation
(3.13) with μ = ν equal to 1, instead of normalizing them to unit power flow. Based
on the orthogonality relation (3.13) and assuming that all modes are in the form of
equation (3.14), we normalize

∫ ψν(x) · ψμ(x)* dx = δνμ, (3.23)

where the two modes share the same slab waveguide profile ExSLAB( y ).

3.2.2 Multimode propagation and interference

In this section, we describe the propagation of the waves through the coupler. We
consider an incident wave in the one of the input waveguides propagating in positive
z-direction (see figure 3.7). At the interface between input waveguides and the wide
MMI waveguide (width W ), multiple modes of the MMI waveguide are excited.
These modes propagate with their own propagation constants to the end of the MMI
waveguide (length L) where they interfere and couple to the output waveguides.

3-11
 Silicon Photonics

The electric field in the MMI waveguide may be written as superposition of its
modes,
E( x , y , z , t ) = ∑cνEx(ν )(x, y )e ı(ωt−β z )xˆ ,
ν
(3.24)
ν

where the summation should be understood to include all guiding as well as

radiation modes (see section 2.3.5 for a detailed discussion). If the ‘spatial spectrum’
of the input, in our case kx of the input waveguide, is small enough not to excite
unguided modes, then it is sufficient to include only the guided modes in the analysis.
Additionally, we assume that reflections may be neglected and only forward
propagating modes need to be included in the approximate analysis.
Only including forward propagating guided modes and using the approximate
modal profiles, equation (3.14), we find that equation (3.24) reduces to
m−1
E( x , y , z , t ) = ∑ cνExSLAB( y )ψν(x)e ı(ωt−β z )xˆ , ν (3.25)
ν= 0
with m the number of guided modes. Taking advantage of the effective index method
we now use a 2D description of the electric field in the waveguide, Ψ( x, z), whose
phase rotates along with propagation constant β0 of the fundamental waveguide
mode. The electric field in the waveguide is then described by
E(x , y , z , t ) = ExSLAB( y )e ı(ωt −β0z )Ψ(x , z )xˆ , (3.26)

with
m−1
Ψ( x , z ) ≡ ∑ cνψν(x)e ı(β −β )z.
0 ν (3.27)
ν= 0

The modes of the MMI waveguide are excited at the interface between the input
waveguides and the MMI waveguide. The electric field at this interface, Ψ(x , 0) is
assumed equal to the incident wave in the input waveguide. The modal coefficients
cν are found using the orthogonality relation (3.23) to be

cν = ∫ Ψ(x, 0) · ψν(x)* dx. (3.28)

With the approximate propagation constants as in equation (3.21),

m−1
⎡ ν(ν + 2)π ⎤
Ψ( x , z ) = ∑ cνψν(x)exp ⎢⎣ı z⎥.
⎦
(3.29)
ν= 0
3Lπ

The relative phase that is accumulated by mode ν after propagating a distance z is

given by
ν(ν + 2)π
z. (3.30)
3Lπ

3-12
 Silicon Photonics

Equation (3.29) provides an approximate analytical description of the light

propagation in the MMI.
The electromagnetic fields at the interface z = L between a MMI waveguide and
the output waveguides are found after propagation over the length L of the MMI
waveguide. In general, it is unlikely that the electromagnetic field at this interface
neatly couples to the modes of the two output waveguides and reflections as well as
coupling to radiation modes may be expected. However, at special lengths of the
MMI coupler, the light occurs exactly as one or more shifted or mirrored copies of
the input field Ψ(x , 0) which may couple to well positioned output waveguides with
minimal reflections.
It will be convenient to treat the even and odd modes separately. The expression
ν(ν + 2) is even for even ν and odd for odd ν. The modal field ψν(x ) is symmetric for
even ν and anti-symmetric for odd ν, i.e.,
⎧ ψ (x ), for even ν
ψν( −x ) = ⎨ ν (3.31)
⎩− ψν(x ), for odd ν.

The electric field Ψ(x , z ) may be decomposed in a symmetric and an anti-symmetric

part,
Ψ(x , z ) = Ψ(x , z )symmetric + Ψ(x , z )anti-symmetric . (3.32)
At z = 0,
Ψ(x , 0) + Ψ( −x , 0)
Ψ(x , 0)symmetric =
2
= ∑ cνψν(x), (3.33)
even ν

Ψ(x , 0) − Ψ( −x , 0)
Ψ(x , 0)anti-symmetric =
2
= ∑ cνψν(x). (3.34)
odd ν

3.2.3 Imaging and power splitting

In this section, we show that for particular MMI coupler lengths L, the field at the
Ψ(x , L ) at the output interface is exactly a superposition of one or more shifted or
mirrored copies of the field at the input interface Ψ( x, z = 0). Proper positioning of
output waveguides then gives maximal coupling into these waveguides with minimal
reflections.

Mirrored image
First we consider the mirrored images in the MMI waveguide which occur at MMI
lengths of 3Lπ . In fact, they also occur at 3Lπ plus a multiple of 6Lπ . By substituting
z = 3Lπ in equation (3.29), it is seen that the phases that are accumulated by the
modes, ν(ν + 2)π , are an even multiple of π for even ν and an odd multiple of π for
odd ν . The even modes have the same phase as at the beginning of the MMI

3-13
 Silicon Photonics

Figure 3.8. Simulation of the electromagnetic field intensity in the MMI coupler. Finite difference time domain
(FDTD) simulation using the CrystalWave simulation software (Photon Design, Oxford, UK).

waveguide and the odd modes have a phase of π , i.e, their sign is reversed. As the
odd modes are anti-symmetric in x, equation (3.31), it follows that the electric field
at the output interface is a mirrored image of the field at the input interface, i.e.,
Ψ(x , 3Lπ ) = Ψ( −x , 0). In the case of a 2 × 2 MMI coupler with access waveguides
symmetrically positioned at ±xc , all energy is coupled from the upper input
waveguide to the lower output waveguide (see figure 3.8). Such a device may be
used in an integrated circuit in the case that two waveguides need to cross each other,
although better methods exist to achieve this in practice [9].

Direct image
An MMI coupler with a waveguide length being a multiple of 6Lπ has the same
electric field at the waveguide output interface as at the input interface, e.g.
Ψ(x , 6Lπ ) = Ψ(x , 0). From the previous discussion of the field at z = 3Lπ we realise
that the electric field at z = 6Lπ is mirrored twice. When z = 6Lπ in equation (3.29),
the relative phases of all modes are an even multiple of π hence the field at the end of
the MMI waveguide is identical to the field at the beginning of the MMI waveguide.

General N-fold imaging

For particular MMI lengths, the input field of the MMI coupler will be imaged N
times superimposed with each image shifted in the x-direction. The 2 × 2 MMI
coupler described hereafter is an example of this phenomenon. In a more general
approach, the summation of sine-like functions ψv(x ) in equation (3.29) can be
interpreted as a spatial Fourier expansion. According to the definition of ψv(x ) in
equation (3.15), this function is zero outside the waveguide core. This can be
interpreted as a periodic boundary condition in the Fourier analysis, with a
fundamental period of two times the waveguide width, i.e., 2W . We extend the
input field Ψ(x , 0) periodically and define the extended input field by

3-14
 Silicon Photonics

∞
Ψ̃(x ) ≡ ∑ Ψ(x + W /2 − 2vW , 0) − Ψ( −x − W /2 + 2vW , 0). (3.35)
v =−∞

It is shown in [7, 10] that at distances

p
z= 3Lπ , (3.36)
N
where p ⩾ 0 and N ⩾ 1 are integers with no common divisor, the field has the form
N −1
e ıθ
Ψ(x , 3pLπ / N ) = ∑ Ψ̃(x − xq , 0)e ıϕ ,
q (3.37)
N q=0
with
W
xq = p(2q − N ) , (3.38)
N
qπ
ϕq = p(N − q ) , (3.39)
N
where θ is an overall phase, p indicates the imaging periodicity along z, and q refer to
each of the N images along x. The above equations show that, at distances z in
equation (3.36), N images are formed of the extended input field Ψ̃(x ), located at
positions xq, each with amplitude 1/ N and phase ϕq . For the 2 × 2 MMI coupler
with 50/50 splitting presented in the following section, p = 1 and N = 2. The shortest
devices are achieved with p = 1. As shown in [7], shorter devices may be realised
provided the input waveguides are at particular positions. For example, an
symmetric input field, i.e., with Ψ(x , 0) = Ψ( −x , 0) reduces the length of N-fold
images by a factor four.

3.2.4 A 2 × 2 MMI coupler

In this section, we consider the example of a 2 × 2 MMI coupler with both input
waveguides and both output waveguides positioned symmetrically at ±xc . We
describe the input field caused by an incident wave in the fundamental mode of
the upper (P1) or lower (P2) access waveguides as Ψ +0 (x ) and Ψ−0 (x ), respectively.
These fields can by computed using equations (3.14)–(3.17) where ψν(x ∓ xc )
describes Ψ ±0 (x ) with w the width of the access waveguides and x translated. These
fields only extend locally around ±xc and are symmetric around ±xc . From
symmetry, Ψ +0 (x ) = Ψ−0 ( −x ) (see figure 3.7(b) where the field Ψ +0 (x ) is sketched).
The complex amplitudes of the fundamental modes in the upper and lower output
waveguides, a±, are given by the overlap integral

a± = ∫ Ψ±0 (x)* · Ψ(x, L) dx. (3.40)

For certain lengths L the field at the end of the MMI waveguide, Ψ(x , L ) is a
superposition of the field at the input of the MMI waveguide, Ψ(x , 0) and the same
field but mirrored in the x-direction, Ψ( −x , 0), i.e.,

3-15
 Silicon Photonics

Ψ(x , L ) = a+Ψ(x , 0) + a−Ψ( −x , 0). (3.41)

In this case, all light is coupled to the two output waveguides. As a trivial example
we consider an MMI coupler with MMI waveguide length L = 0. The field at the left
interface, Ψ(x , 0), is identical to the modal field of the upper input waveguide Ψ +0 (x ).
As L = 0, this is also the field at the right interface, Ψ(x , L ). The output waveguides
are identical to the input waveguides, therefore a+ = 1 and a− = 0 from equation
(3.40).

A 2 × 2 MMI coupler with 50/50 splitting ratio

An MMI coupler as in figure 3.7 may be used to split the power of the input
waveguide equally over two output waveguides. This is achieved by a coupler with
MMI waveguide length L = 3/2Lπ . From equation (3.29),
m−1
Ψ(x , 3/2Lπ ) = ∑ cνψν(x)e ı ν(ν+2)π .
1
2 (3.42)
ν= 0

The even modes accumulate a phase that is a multiple of 2π . The odd modes
accumulate a phase of 3π /2 plus a multiple of 2π . Hence
Ψ(x , 3/2Lπ ) = ∑ cνψν(x) + ∑ ( −ı)cνψν(x) (3.43)
even ν odd ν

1−ı 1+ı
= Ψ(x , 0) + Ψ( −x , 0). (3.44)
2 2
The last step follows from equations (3.33) and (3.34). Thus at the output of the
MMI waveguide section, the field is a superposition of the input field and the
mirrored input field. When light is inserted in the upper input waveguide centred at
x = xc then this field is imaged at the two output waveguides which are centred at xc
and −xc . The fields have amplitude fractions of ∣a±∣ = 1/ 2 thus the power is split
50%/50% over the two output waveguides. This MMI coupler is also referred to as a
−3 dB splitter.

A 2 × 2 MMI coupler with 15/85 splitting ratio

An 15/85 splitting ratio is achieved using an MMI coupler with length L = 3/4Lπ
and by setting the access waveguides at the positions xc = ±W /4 with W the width
of the MMI waveguide. Such a restriction on the position of the access waveguides
was not necessary for the 50/50 splitting ratio. Let the input field be the fundamental
mode of the upper input waveguide which is symmetric in x around xc = W /4. It
follows from equation (3.28) that such an input does not excite MMI waveguide
modes that are anti-symmetric around x = W /4, hence
cv = 0 for ν = 3 + 4m where m = 0, 1, 2, … (3.45)
Inserting z = 3/4Lπ in equation (3.29) gives

3-16
 Silicon Photonics

m−1
Ψ(x , 3/4Lπ ) = ∑ cνψν(x)e iν(ν+2)π /4. (3.46)
ν= 0

For even ν the exponential phase factor is a multiple of 2π . For ν = 3 + 4m it was

found that cv = 0 due to the input field, see equation (3.45). For ν = 1 + 4m, the
exponential phase factor is 3π /4 plus a multiple of 2π . Hence
Ψ(x , 3/4Lπ ) = ∑ cνψν(x) + ∑ e ı3π /4cνψν(x) (3.47)
even ν odd ν

or, using equations (3.33) and (3.34),

1 + e ı3π /4 1 − e ı3π /4
Ψ(x , 3/4Lπ ) = Ψ(x , 0) + Ψ( −x , 0). (3.48)
2 2
Because the field at the right interface of the MMI waveguide is in the form of
equation (3.41), it may be written as a superposition of the fundamental modes of
the two output waveguides and all light is coupled to those waveguides. The
1
amplitude of the mode in the upper waveguide a+ = 2 (1 + e ı3π /4 ) while the complex
1
amplitude of the mode in lower waveguide a− = 2 (1 − e ı3π /4 ). The relative power
coupled in those waveguides are, respectively, ∣a+∣2 ≈ 15% and ∣a+∣2 ≈ 85% .

Results
Figure 3.9 shows the behaviour of MMI couplers for different lengths. The effect of
the tapered access waveguides is neglected and only the fundamental modes of the
access waveguides are taken into account (access waveguide width wa in figure 3.7).
Effective refractive index nr = 2.83 from equation (2.17). Beat length Lπ = 42.96 μm
from equation (3.20). The fields of the waveguides are computed using equation
(3.15). The modal coefficients cν follow from equation (3.28), 40 modes are taken
into account. The field at the right interface of the MMI waveguide is given by
equation (3.29). The coupling to the output waveguides P3 (upper, +) and P4 (lower, −)
is obtained from equation (3.40).
The special MMI lengths discussed above can be recognised in figure 3.9(b). The
lengths 3Lπ /4 = 32 μm, 3Lπ /2 = 64 μm, 3Lπ = 129 μm, and 6Lπ = 258 μm result in
coupling of all energy to the output waveguides with the predicted power fractions.
In addition, it can be seen that there are more MMI lengths for which all power is
coupled to the output waveguides, e.g. the length of 97 μm results in a power
splitting of 85%/15%.

3.2.5 Comparison with rigorous numerical simulations

We compare the results of the approximate analytical model described in sections
3.2.1 and 3.2.2 with the results of rigorous numerical simulations. We apply the
bi-directional eigenmode propagation as implemented in FimmProp by Photon
Design (Oxford, UK). This method is similar to the approximate method presented
here, but with the difference that it rigorously computes the vectorial mode profiles

3-17
 Silicon Photonics

Figure 3.9. Analytical analysis of the MMI coupler consisting of silicon-on-insulator with 220 nm thick silicon
light-guiding layer (see figure 3.7). Vacuum wavelength λ 0 = 1.55 μm. MMI waveguide width W = 4.2 μm,
access waveguide width wa = W /4 , access waveguide position xc = ±W /4 . (a) Field profile Ψ(x, 0). (b) Field
profiles ψ0(x ), ψ1(x ), and ψ2(x ) of the MMI waveguide. (c) modal amplitudes cν for ν = 0…12 . (d) Relative
propagation constants β0 − βν for ν = 0…12 . (e) Power in output waveguides P3 (upper) and P4 (lower).

3-18
 Silicon Photonics

and modal propagation constants without assumptions. At the interfaces, all modes
are taken into account as well as the reflections that occur at those interfaces.
The approximate method is adjusted by introducing an effective waveguide
width Weff = 4.2764 μm which is used in the computation instead of the actual
waveguide width W = 4.2 μm. Weff is chosen such that the particular MMI lengths
corresponding to imaging align with the rigorous computation. The same effective
width is used for all modes, which is an approximation as this so-called Goos–
Hänchen shift depends on the mode number [7]. This shift per mode can be
computed using the extended Marcatili method (see section 2.4). Note that the
width wa and positions xc of the access waveguides are not altered. This leads to a
weak excitation of modes 3, 7, and 11 which are no longer perfectly symmetric
around xc (figure 3.11(a)).
In figures 3.10–3.12, the results of the approximate method and the rigorous
method are compared. It can be seen that the modal profiles (figure 3.10) and
relative propagation constants β0 − βv (figure 3.11(b)) are quite similar. However,
these slight differences cause a significant difference in the behaviour of the device.
In figure 3.12 nevertheless, the results agree qualitatively but there are significant
differences after longer propagation lengths. For example, at the length L = 267 μm
corresponding to the direct image, only 69% of the input power is coupled to the
output waveguide P3 instead of the predicted 100%.
The difference between the rigorous numerical result and the approximate
theory is mainly due to the approximation in the propagation constants. This is
followed by testing the approximate method with the propagation constants βν
computed using the numerical mode solver, instead of the approximate equation
(3.29). The results of this approach are in very good agreement with the fully
numerical results [11].

3.3 Fiber to chip coupling

Silicon photonic integrated circuits have a large refractive index contrast allowing a
small device footprint, which is convenient for most applications. However, the
small waveguides make in-and-out coupling of light into the photonic integrated
circuit (PIC) difficult, since one has to match a ∼9 μm fibre core with a ∼0.5 μm
waveguide.
An inverted optical taper can be employed to convert the modal profile of the
silicon waveguide to the modal profile of an optical fibre (section 3.3.1). Coupling
efficiencies above 90% (<0.5 dB loss) have been reported for coupling from single
mode optical fibres to TE-like modes in 450 nm × 220 nm silicon waveguides [12]. At
the end facet of the chip, the silicon waveguide is covered with a polymer that has a
refractive index higher than the silicon-dioxide BOX layer (n ≈ 1.85). The width of
the silicon waveguide is tapered down towards the end facet of the chip. This
squeezes the modal field outside the silicon into the polymer above it, which now
functions as a new waveguide core.
Alternatively, out-of-plane grating couplers are used to couple light from an
optical fibre to the planar photonic circuit. Such couplers employ a grating that

3-19
 Silicon Photonics

Figure 3.10. Modes of the MMI waveguide. Numerically computed field profiles, Ex in the middle of the
waveguide (red, solid lines). Approximate sinusoidal field profiles ψv(x ) for effective width Weff = 4.2764 μm
(blue, dashed lines). Modes 0–12. The numerically computed effective indices are listed between brackets. The
FMM method as implemented in FimmWave is used as a numerical mode solver. Silicon-on-insulator
waveguide with 220 nm thick silicon light-guiding layer. For the vacuum wavelength λ 0 = 1.55 μm. MMI
waveguide width W = 4.2 μm, access waveguide width wa = W /4 , access waveguide position xc = ±W /4 .

diffracts light from the waveguide upwards. Radiation occurs from an area on the
top surface of the PIC, allowing for a coupler that has the same dimensions as the
fibre core (figure 3.14(a)). Dielectric grating couplers date from the 1970s, and
designs for SOI technology have been presented over the last decades by, among
others, Gent University [13–16]. Basic out-of-plane grating couplers have a coupling
efficiency of about −5 dB (30%, see section 3.3.2), although more advanced grating

3-20
 Silicon Photonics

Figure 3.11. MMI coupler as in figure 3.10. (a) Modal amplitudes cv for modes 0 to 12, from the approximate
theory. (b) Relative propagation constants β0 − βν for modes 0 to 12, computed using approximate theory
(blue curve) and also computed using the numerical mode solver (FMM in FimmWave, red curve).

Figure 3.12. MMI coupler as in figure 3.10. Sum of the power in ports P3 and P4 (upper plot). Power in port
P3 (lower plot). Computed using approximate theory and also using the bi-directional eigenmode propagation
method as implemented in FimmWave. Normalized to input at port P1.

3-21
 Silicon Photonics

Figure 3.13. Inverted taper in silicon-on-insulator technology with waveguide height 220 nm on a 2 μm thick
silicon-dioxide BOX on a silicon substrate. The silicon waveguide is tapered from a width of 450 nm down to a
width of 75 nm over a length of 150 μm. The inverted taper has an overlaying polymer with size 2 μm × 2 μm
and refractive index n ≈ 1.85. Design of IBM, S J McNab 2003, [12]. (a) Sketch of the inverted taper.
(b) Modal field at the side where the waveguide width is 450 nm. (c) Modal field at the side where the
waveguide width is 75 nm and the field is coupled to an optical fibre. Courtesy of Photon Design, reprinted
from FimmProp documentation [17].

Figure 3.14. Sketch of an out-of-plane grating coupler. The silicon layer height is 220 nm and grating etch
depth is 70 nm. The grating coupler width is 10 μm, and the grating length is ∼15 μm. (a) 3D-sketch. (b) Cross-
section of the grating coupler. An efficient simulation method (2D FDTD, effective index method and
diffraction integral) is outlined, © 2011 IEEE. Reprinted, with permission from [23].

coupler designs exist with efficiencies of ∼70% (−1.6 dB, see section 3.3.3). The
advantage of grating couplers over the inverted taper is that they can be positioned at
arbitrary locations on the chip, allowing more design flexibility and offer the possibility
to place other components, such as lasers or photo-detectors, on top of the chip.

3.3.1 Inverted optical taper (fibre to chip coupler)

The most straightforward method to couple light from an optical fibre to a silicon
photonic chip is by extending a waveguide to the edge of the chip and aligning an

3-22
 Silicon Photonics

optical fibre with this silicon waveguide. However, direct coupling is inefficient due
to the large difference in effective refractive indices, geometric size, and limited mode
overlap. At least 30 dB coupling loss can be expected from the poor overlap (0.2%)
alone, because the cross-section of the optical mode in a single mode fibre is in the
order of 65 μm2 while this is 0.1 μm2 for the silicon waveguide [12].
An inverted taper may be used to efficiently couple light from the waveguide on the
silicon photonic chip to an optical fibre, or vice-versa [12, 18–20]. In [12] the coupling
form of a single mode optical fibre to a 220 nm high silicon waveguide is described
with a coupling efficiency of 90% (coupling loss below 0.5 dB). The coupler is shown
in figure 3.13 where a silicon waveguide of width of 450 nm in the middle of the chip is
tapered down over a distance of 150 μm to a width of 75 nm at the edge of the chip.
The inverted taper has a polymer overlay with a refractive index (n ≈ 1.85) higher
than the silicon-oxide BOX layer (n ≈ 1.44). At the edge of the chip, the mode is
pushed outside the silicon waveguide into the polymer which now functions as a
waveguide core (figure 3.13(c)). There is good overlap between this mode and the
mode of the single-mode optical fibre. Also, the difference in effective refractive
indices is moderate which strongly reduces the reflection at the end facets.
More recent inverted taper designs couple both TE as well as TM-polarized light
efficiently to silicon waveguides with shorter tapers [21, 22]. In those devices,
the silicon waveguides have a thickness above 300 nm to support both TE-like and
TM-like modes. In [22] several inverted taper designs are compared.
3.3.2 Basic out-of-plane grating couplers
A grating coupler consists of a very wide (e.g. 10 μm) waveguide with a grating
etched in its top surface. We call a grating coupler basic when it is one-dimensional
and all its grooves are rectangular, and identical. Advanced couplers have, for
example, concentric cylindrical gratings, different profiles of the grooves or teeth of
the grating, or an apodized grating period (see section 3.3.3). For a basic grating
coupler, the electromagnetic field varies relatively slowly with the coordinate that is
parallel to the grooves (the x-coordinate in figure 3.14(b)). In and near the coupler,
we may approximate the shape of the electromagnetic fields in that direction by the
shape of the fundamental mode of the waveguide [23].
The behaviour of the coupler can intuitively be understood by considering all the
tall-to-short interfaces on the edge of the grating grooves as ‘line-sources’ which
have a phase difference which follows from the propagation speed of the light
through the waveguide (a few such line-sources are indicated by the symbol P in
figure 3.14(b)). The effective refractive index of the grating can be estimated as the
spatially weighted average of the effective indices of the fundamental modes in the
tall and short parts of the waveguide. The fields emitted by these point-sources
constructively interfere to a far field that is a plane wave propagating under a certain
angle θq with respect to the y-axis as shown in figure 3.14(b). This angle is given by
λ
n3 sin(θq ) = ne − q , (3.49)
Λ

3-23
 Silicon Photonics

where n3 is the refractive index of the upper medium (air in figure 3.14), ne is the
averaged effective index of the grating, q is the diffraction order and Λ is the grating
period. The average effective index ne is computed as the spatial average of the effective
indices corresponding to concatenated high and low waveguides that form the grating.
For perfect vertical coupling, θq = 0◦, equation (3.49) describes a second-order
distributed Bragg reflector (DBR), which very efficiently reflects the forward propagat-
ing light in the waveguide backwards, rather than radiating it upwards. For low index
contrast waveguides coupling coefficients between the waveguide mode and a prop-
agating plane wave with wave vector whose angle with the y-axis is θq can be derived
using coupled mode theory, with the grating as a perturbation of a waveguide [24, 25].
Among other relations, equation (3.49) also follows from that derivation. To our
knowledge, the validity of this perturbation approach has not been verified for high-
index-contrast guides, and it is therefore common to use numerical methods (e.g.
FDTD). We have compared equation (3.49) with the strongest far-field radiation angle,
and found good agreement for etch depths up to 110 nm for a free-space wavelength of
1550 nm. For efficient coupling to a fibre, the grating should be designed such that only
one coupling order (q = 1) exists. Moreover, the coupling strength is a compromise
between strong upwards coupling (large coupling strength) and low backwards
reflections (low coupling strength). The ratio between upwards and downwards
radiation is strongly influenced by the height of the buried oxide (BOX) layer because
the downward propagating waves reflect at the BOX layer and interfere.
Figure 3.15 shows a numerical analysis of an out-of-plane grating coupler. The
relevant effects such as the Bragg reflector are clearly visible in plots (a) and (c). The
angle under which the coupler radiates is in good agreement with equation (3.49)
(see plot (d)) but there is also radiation in angles close to this angle (see plot (b)).
In the remainder of this section, we describe an efficient scheme for numerical
simulation of such out-of-plane grating couplers. We exploit the fact that the width
of the waveguide, and thus the grating, is about 9 μm which is much larger than
the height of the waveguide of 220 nm. This justifies the use of a 2D simulation in the
(x , z )-plane in the vicinity of the waveguide which is later corrected with the
x-profile of the fundamental mode of the waveguide. The scheme consists of four
steps: a 2D FDTD simulation in the ( y, z )-plane which describes the propagation
from the waveguide to a plane just above the coupler and assumes no variation of
the electromagnetic fields in the x-direction. Then, the effective index method is
applied to calculate the profile of the field in the x-direction, based on the width of
the grating in this plane, resulting in the 3D field. Thereafter, Rayleigh–Sommerfeld
diffraction is used to propagate the field from this plane to the plane of the fibre facet
and finally an overlap integral is used to calculate the coupling into the fibre mode.

2D calculations and the effective index method

The width of the grating waveguide is about 40 times larger than its height and light
is excited in the fundamental mode of this waveguide. Therefore the electromagnetic
fields vary slowly in the x-direction as compared to the y- and z-directions. A 2D
( y, z ) analysis is used in the vicinity of the coupler up to approximately one
wavelength above the grating. In this simulation it is assumed that the refractive

3-24
 Silicon Photonics

Figure 3.15. Two-dimensional numerical simulations of an out-of-plane grating coupler as shown in figure
3.14(b). Finite difference time domain (FDTD) simulations in the ( y, z )-plane with the assumption that there is
no variation of the electromagnetic fields in the x-direction. Forwards-propagating light is excited in the
fundamental mode of the waveguide on its left-hand-side before the grating. It is recorded how this light
propagates through the device and is partially diffracted upwards and downwards by the grating. Grating
coupler with silicon layer height 220 nm, silicon-dioxide BOX layer height 1980 nm, etch depth 66 nm, duty
cycle of grating teeth is 50%, and with 20 grating periods. (a, b) Results of a coupler with grating period
Λ = 616 nm . (b–d) Vacuum wavelength λ = 1550 nm . (a) Power that is reflected back from the grating,
propagating in the waveguide in the negative z-direction (R), power that is transmitted through at the end of
the grating at the right-hand-side (T), power that is diffracted upwards, i.e., through the plane y = y0 in figure
3.14 (Up), power that is diffracted downwards (Down). (b) Far-field intensity as a function of angle.
(c) Relative power in different directions, see legend of (a), for different grating periods Λ. (d) Angle of
maximum far field intensity (as in plot (b)) as function of the grating period for wavelength of 1550nm. Also
includes the angle that is expected from equation (3.49). (red solid curve). For details, see [11, 23].

index profile and the electromagnetic fields do not vary in the x-direction. To obtain
the variation of the electromagnetic field in the x-direction in a plane just above the
coupler, we apply a method similar to the effective index method (EIM) by
approximating the field Ex by Ex(x , y, z ) = E 2xD( y, z ) · Exmode(x ), where E 2xD is the
electric field as calculated using 2D analysis (assuming invariance in the x-direction)
and lateral field profile Exmode(x ) is approximated from the lowest order mode in the

3-25
 Silicon Photonics

waveguide. The results of this method are in good agreement with full 3D
simulations (see figure 3.16).

Propagation into the upper-half space

To obtain the electromagnetic field radiated into the homogeneous half space y > y0
by a finite aperture or source in the plane y = y0 one can use the Rayleigh–
Sommerfeld diffraction integral [26]. In our case, the finite aperture is a sufficiently
large part of the plane just above the grating coupler such that the electromagnetic
fields outside this aperture are small and hence can be neglected. The medium above
y0 is air. The diffraction integral for monochromatic light is written as [26]:

U (x , y , z ) = ∬aperture U (x′, y0′, z′)G (x, y, z; x′, y0′, z′)dx′dz′, (3.50)

where

(1 + ıkr ) ( y − y0′) exp( −ıkr ) iωt (3.51)

G= e ,
2π r r2

Figure 3.16. Three-dimensional numerical simulations of an out-of-plane grating coupler. Electric field
component Ex in the plane y0 = 1.84 μm (see figure 3.14). Silicon layer height 220 nm, silicon-dioxide BOX
layer height 1980 nm, grating period 616 nm, etch depth 88 nm, duty cycle of grating teeth is 50%, 20 grating
periods, waveguide width 9.9 μm. (a) Computed using 2D FDTD simulations with the effective index method
applied thereafter, as discussed in section 3.3.2. (b) Computed using 3D FDTD simulations (for details, see
[23]). © 2011 IEEE. Reprinted, with permission from [23].

3-26
 Silicon Photonics

2
r= (x − x′)2 + ( y − y0′) + (z − z′)2 , (3.52)

and U is any electric or magnetic field component in phasor notation with time
dependence given by e iωt . The Greens’ function G is the sum of the fields of two
in-phase point sources that are images of each other with the plane y = y0′ as
mirror. This choice of Greens’ function allows the field in the air to be described
as an integral of the field in the aperture, without requiring knowledge of the
normal derivative of the field [26]. When the field is calculated for a horizontal
plane (y constant), G depends on x − x′ and z − z′, so equation (3.50) is a 2D
convolution, which can be very efficiently calculated using fast-Fourier-transforms
(FFTs) [27]. When the plane is rotated around the x-axis as shown in figure
3.14(b), the integral with respect to x′ is still a convolution but the integral with
respect to z′ is not. The 2D equivalent of equation (3.50), where U = U ( y, z ), is
obtained by integrating equation (3.50) along x′, giving [28]:

U ( y, z) = ∫aperture U (y0′, z′)G ( y, z; y0′, z′)dz′, (3.53)

where

−ık ( y − y0′) (2) (3.54)

G= H1 (kr ),
2 r

r= ( y − y0′)2 + (z − z′)2 , (3.55)

with H1(2)(kr ) the first Hankel function of the second kind, i.e., H1(2)(kr )=
J1(kr ) − ıY1(kr ) where J1 and Y1 are the first order Bessel functions of the first
and second kind, respectively.

Coupling into a fibre

The fibre is typically positioned tenths of microns above the out-of-plane grating
coupler. The power flux through the plane S just before the fibre can be calculated
using equation (2.59) integrated over the plane of the fibre-end. The power coupling
efficiency for TE-like modes can be estimated by [29]:
2
∬ E ix(x , ρ) · E xf (x , ρ)*dS
ηoverlap = 2 2
, (3.56)
∬ E ix(x , ρ) dS · ∬ E xf (x , ρ) dS
where E ix is the x-component of electric field incident on the fibre facet and E xf is the
x-component of the electric field of the fundamental TE-mode of the fibre.
Coordinates x and ρ are in the tilted plane, parallel to the fibre facet (see figure
3.14). The fibre mode of a standard single-mode fibre can be approximated by a
Gaussian beam [30]. Details of coupling from air into the fibre are neglected since
there is a small refractive index step and a small angle of incidence of the incoming

3-27
 Silicon Photonics

wave. When the field E ix can be separated in x and ρ dependence, i.e., E ix(x , ρ )=
E ix,x(x ) · E ix,ρ(ρ ), then the overlap ηoverlap can also be separated, i.e., ηoverlap=
ηoverlap,x · ηoverlap,ρ .

3.3.3 Advanced grating couplers

The basic grating coupler has four major drawbacks, which are to a large extent
overcome by new grating coupler designs. First, in 220 nm high waveguides, light is
directed not only upwards but also in the unwanted downward direction, which
causes losses of 35% to 45% [33]. These losses can be reduced by increasing the
height of the waveguide such that the downward reflections in the grating coupler
cancel each other, thereby directing the light predominantly upwards. Such grating
couplers can be fabricated in wafer-scale CMOS processes by depositing an
amorphous silicon layer on top of the 220 nm waveguide layer [16, 33]. In this
way, efficiencies up to −1.6 dB (69%) have been reported for the coupling of
electromagnetic power from a 220 nm high silicon waveguide to a standard single-
mode optical fibre. Another approach is to use interleaved full and shallow etched
trenches. The combination of sub-wavelength trenches with heights of 220 nm high
(no etch), 150 nm high (shallow etch) and 0 nm high (full etch) allows for a design
causing constructive interference in the upward direction, see figure 3.17(b) [32, 34].
Alternatively, mirrors can be included in the substrate, but this approach has not
been demonstrated in wafer-scale CMOS processes.
The second drawback of a basic grating coupler is that the radiation is strongest
from the first grating tooth and decays exponentially with increasing distance z from
the beginning of the grating. Therefore the radiation profile has poor mode-
matching with the Gaussian-shaped mode of an optical fibre (contributing to
∼20% of efficiency loss) [33]. An apodized grating can better match the radiation
profile to the optical fibre ([35] reported the record efficiency of −1.2 dB or 76%), but
these fine grating structures cannot be fabricated in today’s wafer-scale CMOS
technology (the first groove in [35] is 44 nm wide).
The third drawback of basic grating couplers is that the adiabatic taper from the
small waveguide to a the wider waveguide is a few hundred microns long and thus
occupies a substantial footprint on the photonic chips. This drawback can be
overcome by using a focusing grating coupler, in which the light at the end of the
straight waveguide diffracts because the waveguide rapidly tapers in width (angle of
27◦) [36]. For such a cylindrical wave-front, it is possible to design a cylindrical
grating that couples the light to a single mode fibre. See figure 3.17(a) for a similar
design but where the 27◦ taper is omitted and a free-divergence region in used
instead.
The fourth drawback of the basic grating structure is that it introduces reflections
(around 6%) from the waveguide connected to the grating coupler back into this
waveguide. This often results in an undesired Fabry-Pérot cavity, for example in a
device that has both an input coupler and an output coupler which together form
this cavity. This reflection may be interpreted as occurring at the interface between
the 220 nm high waveguide and the grating with etched trenches. The difference in

3-28
 Silicon Photonics

Figure 3.17. (a) Scanning electron microscope (SEM) image of the focused grating coupler presented in [31,
© 2013 IEEE, reprinted with permission]. Light propagates from the waveguide in the bottom-left corner of
the image through the taper and the free divergence region to the focused grating. (b) Sketch of grating coupler
presented in [32], reprinted with permission. Light propagates from the waveguide at the right-hand-side of the
sketch through the taper, through the index matching region with sub-wavelength structures, to the grating
where it is diffracted upwards.

refractive index of the 220 nm high waveguide and the average refractive index of the
grating causes this reflection. This reflection may be mediated by matching the
refractive index of the 220 nm high waveguide to the refractive index of the grating
coupler by means of sub-wavelength patterning of the waveguide before the coupler
[32, 34]. Alternatively, the focussing grating coupler can be designed such that the
light reflecting back from the cylindrical grating does not focus on the waveguide,
but next to it. The reflections hereby reduced while the efficiency remains of the
order −5 dB, similar to the efficiency of the basic grating couplers [37, 38].
In our opinion, two grating coupler designs stand out. The coupler in [31] is
wafer-scale CMOS compatible and combines the small reflectionless focused grating

3-29
 Silicon Photonics

couplers with increased upwards directivity using a grating that has a higher
waveguide, thereby providing 2.2 dB (60%) coupling efficiency with −40 dB
(0.01%) back-reflections (see figure 3.17(a)). The coupler in [32] is CMOS compat-
ible2 and achieves upwards radiation with shallow and deep etched trenches, which
are more common in CMOS technology, while minimizing reflections using sub-
wavelength index-matching structures. The reported efficiency is −1.6 dB (69%) but
the reduced back-reflection is still 4% (see figure 3.17(b)).

3.4 Ring and racetrack resonators

In figure 3.18, racetrack-shaped ring resonators are shown, which are coupled to one
or two waveguides. In this section we study the case that the ring is coupled to two
waveguides, the so-called add-drop configuration. We consider the power that is
transmitted to the output waveguide, ∣b1∣2 , and the power in the so-called drop
waveguide, ∣ad ∣ 2.
First the configuration with a single coupler is considered (figure 3.18(a, b)),
following the work of Yariv [39]. All waveguides are single-mode and the complex
amplitudes of the traveling modes in the waveguides (a1, a2, b1 and b2 as defined in
figure 3.18) are normalized such that their squared magnitude corresponds to the
power in the mode. A lossless coupler without reflections is generally described by
⎛ b1 ⎞
⎜ ⎟=
⎝ b2 ⎠ − (
τ* κ
κ* τ )(aa ),
1
2
(3.57)

with ∣τ∣2 + ∣κ∣2 = 1, see equation (B.1) in appendix B. This matrix does not depend
on the specific type of coupler. After one round-trip through the racetrack, the wave
has experienced a phase delay ϕr and a loss:
a2 = αe ıϕrb2 , (3.58)
where α 2 that is the power after one one roundtrip relative to the power just before
the roundtrip (i.e., α = 1 means no loss). The power in the output waveguide, ∣b1∣2 , is
obtained by first solving equations (3.57) and (3.58) for b1. First we substitute
equation (3.58) in the second row of equation (3.57) to get
−κ *
b2 = a1. (3.59)
1 − ταe ıϕr
Substituting this in equation (3.58) to express a2 in a1 and then substitute the result in
the first row of equation (3.57), we finally get
⎛ κκ *αe ıϕr ⎞
b1 = ⎜τ * − ⎟a1. (3.60)
⎝ 1 − ταe ıϕr ⎠
We define τ = ∣τ∣e ıϕτ and use ττ * + κκ * = 1 to rewrite equation (3.60) as

2
Although the design of the coupler in [32] is compatible with CMOS deep-UV lithography, the measured
device was fabricated using electron beam (e-beam) lithography.

3-30
 Silicon Photonics

Figure 3.18. Racetrack-shaped ring resonator with one and two coupler(s). (a) Sketch with the coupling
described by indicated coefficients of the transmission matrix. (b) Layout of a racetrack resonator with 450 nm
wide waveguides, a straight track of 40 μm, a bend radius of 5 μm, and a coupler as in figure 3.1(a). (c) Sketch
of a racetrack with two couplers (add-drop configuration), © 2014 IEEE, reprinted with permission from [40].

−α + τ e−ı(ϕr+ϕτ )
b1 = a1. (3.61)
e−ıϕr − α τ e ıϕτ
Finally we compute b1b1* and use 2cosθ = (e ıθ + e−ıθ ) to arrive at [39]

α 2 + τ 2 − 2α τ cosθ
b1 2 = a1 2 , (3.62)
1 + α 2 τ 2 − 2α τ cosθ

where θ is the net phase delay of traveling through the ring and coupler
θ = ϕr + ϕτ . (3.63)

In the case of two bus waveguides with identical couplers (figure 3.18c), equations
(3.58) and (3.62) still apply, provided we include the transmission through the
second coupler in the track round-trip by replacing α by α∣τ∣, giving

(α 2 + 1 − 2α cosθ ) τ 2
b1 2 = a1 2 , (3.64)
1 + α 2 τ 4 − 2α τ 2 cosθ

and equation (3.63) changes to

θ = ϕr + 2ϕτ . (3.65)

The amplitude of the mode in the drop output waveguide is

ad = α κ e ı(ϕr /2+ϕκ )b2 , (3.66)

in which the drop bus waveguide is exactly located symmetric to the input bus
waveguide, so that the wave travels half a round-trip from b2 to the second coupler.
Using equation (3.58) we can express the power in the drop waveguide ∣ad ∣2 in terms
of the incident power ∣a1∣2
3-31
 Silicon Photonics

2
(1 − τ 2 ) α
ad 2 = a1 2 . (3.67)
1 + α 2 τ 4 − 2α τ 2 cosθ

3.4.1 Silicon ring resonators with directional couplers

We now apply the general ring theory to the racetrack resonators with directional
couplers in add-drop configuration with two coupled waveguides (figure 3.18(c)).
The racetrack including the couplers has length l. The transmission τ through the
directional couplers with effective length L̃ is given by equation (3.2). The phase
delay due to propagation through a waveguide with length l − 2L˜ is, according to
equation (2.66), equal to ϕr = −β (l − 2L˜ ). Hence the total phase delay of the ring is

θ = ϕr + 2ϕτ = −βl + 2δL˜ −2κbbL˜ + 2 arg {

cos sL˜ −
ıδ
s }
sin sL˜ . (3.68)

For typical silicon-on-insulator racetrack resonators, the second and third term at
the right-hand-side of this equation are small (δ, κbb ≪ β ). The uncertainty in the
propagation constant β due to imperfect fabrication is larger than δ and κbb. The last
term of equation (3.68) is usually small as the real part of the complex number of
which the argument is taken is much larger than the imaginary part that is
proportional to δ /s . This term can also be close to π when the real part is negative.
However, in the particular case that nearly all light is coupled from/to the resonator,
cos sL˜ ≈ 0 and the argument is then close to π /2 rad . This was also experimentally
observed in [40] (or [3, section 3.7]). We recall that the dispersion in the effective
index ne and in the directional coupler coefficient s can considered to be linear, while
the dispersion in L̃ is negligible (see sections 2.3 and 3.1, respectively). For a coupler
with two identical waveguides (δ = 0), neglecting the κbb, and with linear dispersion
of the effective index, equation (3.68) reduces to
⎡ ne − ng ng ⎤
θ = −βl = −2π ⎢ + ⎥l , (3.69)
⎣ λc λ⎦
with ne and ng evaluated at the center wavelength λc (for an example, see figure 3.19(a)).
The loss in the coupler can be approximated by the loss in the isolated waveguide,
thus the round-trip transmittance is given by α = e−αpl where αp is the propagation
loss of the waveguide.

3.4.2 Ring resonator resonances

In this section, we compute some relevant characteristics of the transmission spectra
that are described by equation (3.62). The relations derived in this section give
insight into the behaviour of racetrack resonators and are useful to design such
resonators. For simplicity, we consider ring and racetrack resonators with only one
connected waveguide. The transmission spectrum of the connected waveguide is
given by ∣b1∣2 as a function of wavelength. It shows dips for θ = 2πm, with m an

3-32
 Silicon Photonics

integer number. For the case of two identical waveguides in the coupler (i.e., δ = 0)
and neglecting κbb , the resonance wavelengths λ m are
m · λm = n e(λm ) · l . (3.70)
The free spectral range (FSR) is the difference between the resonance wavelengths of
two adjacent resonances m and m + 1 (see figure 3.19(e)). The FSR may be
approximated by linearizing the relation between λ m and m, then differentiate λ m
with respect to m and compute ΔλFSR = ∣∂λ m /∂m∣ · Δm for Δm = 1. This gives

Figure 3.19. Behaviour of racetrack-shaped ring resonators. (a–e) One coupler, device of figure 3.18(a). (a)
Phase delay θ, equation (3.69). (b) Power in output waveguide ∣b1∣2 , equation (3.62). Resonance extinction ratio
r and full-width at half-max ΔλFWHM . (c) Phase in output waveguide arg{b1}, equation (3.60). (d) Circulating
power ∣b2∣2 , equation (3.59). (e) Same as plot (b) but for larger wavelength interval. (f) Two couplers in add-
drop configuration, device of figure 3.18(c). Power in output waveguide ∣b1∣2 , equation (3.64). Power in drop
waveguide ∣ad ∣2 , equation (3.67). (All) input amplitude a1 = 1 (input power ∣a1∣2 = 1). Track length l = 111 μm,
effective index n e = 2.29 (at λc = 1.55 μm), group index ng = 4.40, round-trip power transmittance α 2 = 0.98
(round-trip loss 0.088 dB), power coupling κ 2 = 0.05, power transmittance τ 2 = 0.95.

3-33
 Silicon Photonics

∂λm λ m2 λ m2
ΔλFSR = ·1= = , (3.71)
∂m (n e
∂n
− λm ∂λe l ) ng l

where we use m = (ne(λ m ) · l )/λ m and we compute ∂λ m /∂m by (∂m /∂λ m )−1. The last
equality sign in equation (3.71) follows from equation (2.32).
At resonance, cosθ = 0 and equation (3.62) becomes
(α − τ )2
b1 2 = a1 2 . (3.72)
(1 − α τ )2
From equation (3.72) it is observed that there is no transmission at the wavelengths
of the dips when ∣τ∣ = α , hence when the round-trip loss of the racetrack is equal to
the power coupled to the racetrack. This condition is called critical coupling. The
minimum, ∣b1∣2min , and maximum, ∣b1∣2max , transmitted power occur at resonance and
in between the resonances, respectively. The extinction ratio r ≡ ∣b1∣2min /∣b1∣2max and
the full-width at half-maximum (FWHM) of the dips in equation (3.62) are (figure
3.19(b))
(α − τ )2 (1 + α τ )2
r= , (3.73)
(α + τ )2 (1 − α τ )2
and
λ m2 ⎡ 2α τ ⎤
ΔλFWHM = cos−1 ⎢ ⎥. (3.74)
πlng ⎣ 1 + α2 τ 2 ⎦

The relation for ΔλFWHM is found by solving3 equation (3.62) for cosθ at half-
maximum, i.e., with ∣b1∣2 = (∣b1∣2max + ∣b1∣2min )/2, and then employing the linearized
relation between the phase delay and vacuum wavelength. The equations in this
section, equations (3.70)–(3.74), are also valid for the case of two couplers when α is
replaced by ατ , i.e., the second coupler acts as an additional source of loss.
The relations in equations (3.73) and (3.74) explicitly show the shape of the
resonances as a function of the waveguide and coupler properties, and are very
useful in the design of resonators. The FWHM depends on the losses in the
resonator (transmittance α∣τ∣), and it scales with the free spectral range (FSR),
while the extinction ratio r scales with (α − ∣τ∣)2 so that critical coupling is most
important. Figure 3.20 shows an experimental example of critical coupling. A set of
small racetrack resonators were fabricated with the length of the straight section in
the directional coupler varying from 0.5 μm to 3.5 μm. The corresponding coupler
straight-through transmission power varies from τ 2 = 0.96 to τ 2 = 0.83 and is closest
to the ring loss α 2 = 0.9 for a coupler length of 2 μm where the low transmission at
resonance is seen.
In figure 3.19, we plotted the behaviour of the resonators shown in figure 3.18
(with the assumptions mentioned at the beginning of this section). Plots (b) and

⎡ ⎤
3
The FWHM in terms of phase θ is Δθ FWHM = 2cos−1⎢ 2α2∣ τ ∣ 2 ⎥.
⎣1 + α ∣ τ ∣ ⎦

3-34
 Silicon Photonics

Figure 3.20. Experimental example of critical coupling. Small racetrack-shaped ring resonators with straight
section of 10 μm, very small bend radius of 1.5 μm and circumference l = 30 μm. One coupled waveguide.
(a) Optical microscope photograph. (b) Transmission spectra of the racetrack resonators with different coupler
lengths L are plotted as a function of wavelength λ with respect to the wavelength of the resonance λm in nm.
Measurements (solid, blue) and fits (dashed, red). The resonance dip closest to 1550 nm is plotted. Vertical axis
is in dB, i.e., the power is logarithmic. More details may be found in [3, section 3.9].

(e) show the FWHM, extinction ratio, and FSR of these racetrack resonators. The
transmittance at resonance is not zero because the coupling constant τ = 0.95
differs from the round-trip transmission α = 0.98 . Plot (f) shows the same
racetrack as in (b) but with two identical couplers. The transmission at resonance
is lower because the effective loss ατ = 0.931 differs less from the coupling
constant. The FWHM is wider because of this higher effective loss. The peak that
appears in the ‘drop’ waveguide is also plotted. Figure 3.19(c) shows that there is a
rapid change in the phase of the light close to the resonance. This plot is in the
frequency-domain (wavelength-domain), but actually this feature is more interesting
to consider in the time-domain where the phase-delay may be used as a time-delay
[41]. Figure 3.19(d) shows that at resonance, the power circulating in the ring
exceeds the input power by a factor 40. This enhancement of the electromagnetic
field in the ring resonator is especially interesting when non-linear effects are
exploited. Photon pair sources based on photon-pair generation via spontaneous
four-wave mixing (SFWM), such as reported in [42], benefit from the large χ (3) non-
linearity of silicon, the high confinement of light in sub-wavelength SOI waveguides,
and the field enhancement in the ring resonators.

3.4.3 Ring resonators with a non-uniform waveguide

It is sometimes advantageous to use a ring resonator in which the width of the
waveguide varies. The fundamental mode of wide waveguides has a lower group
index, which implies higher sensitivity when the waveguides are used as sensors.
Also, the effective index is less sensitive to the width of the waveguide, causing less
phase-distortions when used in arrayed-waveguide-gratings (AWGs). However,
wide waveguides support multiple lateral modes which are excited when the

3-35
 Silicon Photonics

waveguide is bent. Adiabatically tapering4 the wide waveguide to a single-mode

waveguide before the bends of the resonator strongly reduces the excitation of the
higher modes. For ring resonator with a varying waveguide width and of which the
coupler(s) consist of two identical waveguides (i.e., δ = 0), the phase delay after one
round-trip is

θ=− ∮ β(ρ, λm)dρ, (3.75)

in which the integral runs over the circumference of the track and κbb is neglected.
Comparing equation (3.75) with equation (3.69) shows that the theory for uniform
guides remains applicable provided that the track-averaged effective indices are used
1
〈ne〉 =
l
∮ ne(ρ)dρ, (3.76)

1
〈ng〉 =
l
∮ ng(ρ)dρ. (3.77)

3.5 Mach–Zehnder interferometers

Mach–Zehnder interferometers (MZIs) consist of two splitters whose output ports
are connected by two waveguides, which are called the arms of the interferometer
(see figure 3.21). These interferometers are the workhorse of modulator devices
where a phase difference is generated between the two arms of the interferometer.
Another application of these interferometers is in sensing, where the phase difference
between the arms is caused by an external effect of interest.
In this section we consider an interferometer with two lossless symmetric 2 × 2
couplers, where the coupling from port a1 to ports b1 and b2 is the same as the
coupling from port a2 to ports b2 and b1, respectively. This may be a directional
coupler with two identical waveguides (section 3.1) or an MMI coupler with
two input and output ports (section 3.2). Propagation through the lossless symmetric
2 × 2 couplers is described by
⎛ b1 ⎞ τκ
⎜ ⎟=
⎝ b2 ⎠ κτ( )(aa ), 1
2
(3.78)

with
2(arg{τ} − arg{κ}) = π , (3.79)
as derived in appendix B.
We consider an incident wave in waveguide a1 with complex amplitude a1 while
waveguide a2 is dark, i.e., a2 = 0. Then (see figure 3.21(a))

4
Adiabatically tapering means tapering so slowly that hardly any higher modes are excited.

3-36
 Silicon Photonics

Figure 3.21. Sketch of Mach–Zehnder interferometer. (a) Sketch with description of couplers and modal field
amplitudes. (b) Sketch of impementation with two directional couplers.

b1 = τa1, (3.80)

b2 = κa1. (3.81)
After propagating through the two waveguides arms with lengths l1 and l2, the waves
in the arms one and two experience a phase delay of ϕ1 = β1l1 with transmittance
α1 and a phase delay of ϕ2 = β2l2 with transmittance α2 , respectively. Thus the
complex amplitudes at the interfaces between the arms and the second 2 × 2 coupler
are given by
c1 = α1e ıϕ1b1, (3.82)

c2 = α2e ıϕ2b2 , (3.83)

with α1 and α2 real (α = 1 means zero loss). The phase after propagating through one
of the two arms may be intentionally modulated to introduce a phase difference
between the two arms.
Propagation through the second coupler is also described by equation (3.78) but
with a → c and b → d . Thus, using equations (3.82) and (3.83),
d1 = τc1 + κc2 = (τ 2α1e ıϕ1 + κ2α2e ıϕ2)a1, (3.84)

d2 = κc1 + τc2 = κτ (α1e ıϕ1 + α2e ıϕ2)a1. (3.85)

The transmitted power in waveguide d1,
d1d1* = (τ 2τ 2*α12 + κ 2κ 2*+τ 2κ 2*α1α2e ı(ϕ1−ϕ2) + κ 2τ 2*α1α2e ı(ϕ2−ϕ1))a1a1*,
(3.86)
= ( τ 4 α12 + κ 4 α22 − 2 τ 2 κ 2 α1α2 cos[ϕ1 − ϕ2 ]) a1 2 ,

where in the second line τ 2κ 2*=κ 2τ 2*=−∣τ∣2 ∣κ∣2 is used, which follows5 from equation
(3.79). The transmitted power in waveguide d2 is

d2d 2* = τ 2 κ 2 (α12 + α22 + 2α1α2 cos[ϕ1 − ϕ2 ]) a1 2 . (3.87)

For couplers with equal power splitting, ∣τ∣2 = ∣κ∣2 = 12 and lossless MZI arm
waveguides α1 = α2 = 0, equations (3.86) and (3.87) reduce to

5
Using Euler’s formula 2cos(ϕ ) = exp(ıϕ ) + exp(−ıϕ ).

3-37
 Silicon Photonics

⎛1 1 ⎞
d1 2 = ⎜ − cos[ϕ1 − ϕ2 ]⎟ ∣a1∣2 , (3.88)
⎝2 2 ⎠

⎛1 1 ⎞
∣d2∣2 = ⎜ + cos[ϕ1 − ϕ2 ]⎟ a1 2 . (3.89)
⎝2 2 ⎠

Figure 3.22 shows the result of equation (3.86) for three examples including the case
of equation (3.88).
It follows from the equations above that the Mach–Zehnder interferometer is
periodic in the phase difference Δϕ = ϕ1 − ϕ2 with a period of 2π . When the
transmittance spectrum of a Mach–Zehnder interferometer is measured over a
limited wavelength range, then Δϕ is in good approximation linear in wavelength λ
and this periodicity in wavelength is referred to as the free-spectral-range (FSR). For
a MZI with two waveguides that have identical propagation constant β, the phase
difference Δϕ is given by
Δϕ = β(l2 − l1), (3.90)
with l1 and l2 the lengths of the waveguides of the upper and lower arms,
respectively. The free-spectral-range ΔλFSR is then
∂λ λ2
ΔλFSR = 2π = , (3.91)
∂Δϕ ng l2 − l1

where we used ∂(Δϕ )/∂λ = ∂β /∂λ(l2 − l1) from equation (3.90) and also ∂β /∂λ =
∂β /∂k · ∂k /∂λ with k = 2π /λ and ∂β /∂k = ng from equation (2.32).

Figure 3.22. Example of the behaviour of a Mach–Zehnder interferometer with relative power in one output
port plotted as function of path length difference. Equation (3.86), ∣d1∣2 /∣a1∣2 , for three cases: (1) 50/50 coupler
(∣τ∣2 = ∣κ∣2 = 0.5) with no transmission loss (α1 = α2 = 1), (2) 70/30 coupler (∣τ∣2 = 0.7 , ∣κ∣2 = 0.3) with no
transmission loss, and (3) 50/50 coupler with 2 dB loss in arm 2 (α1 = 1, α2 = 0.79 ). Typical SOI waveguide
(450 nm × 220 nm) with effective refractive index ne = 2.54 at vacuum wavelength λ = 1.55 μm. Path difference
ϕ1 − ϕ2 = β dl with dl plotted from 0 nm to 1500 nm.

3-38
 Silicon Photonics

In modulators the propagation constant β1 in the upper arm of the Mach–

Zehnder interferometer is externally modified for example by changing the temper-
ature of the upper waveguide. A small change in the propagation constant β1 causes
a wavelength-shift of the measured transmission spectrum. For a given MZI, the
externally induced phase shift Δϕshift can be computed from the measured wave-
length shift Δλ shift and the measured free-spectral-range ΔλFSR . The free-spectral-
range corresponds to a phase difference of 2π hence
Δλshift
Δϕshift = 2π . (3.92)
ΔλFSR

References
[1] Hardy A and Streifer W 1985 Coupled mode theory of parallel waveguides J. Lightwave
Technol. 3 1135–46
[2] Hardy A 1998 A unified approach to coupled-mode phenomena IEEE J. Quantum Electron.
34 1109–16
[3] Westerveld W J 2014 Silicon photonic micro-ring resonators to sense strain and ultrasound
PhD Dissertation Technische Universiteit Delft http://repository.tudelft.nl
[4] Taflove A and Hagness S C 2005 Computational Electrodynamics: The Finite Difference
Time Domain Method 3rd edn (Norwood, MA: Artech House)
[5] Janssen O 2010 Rigorous simulations of emitting and non-emitting nano-optical structures
PhD Dissertation Technische Universiteit Delft http://resolver.tudelft.nl/uuid:c2a93de0-21e4-
490b-a18c-09f319c2da17
[6] Photon Design 2010 CrystalWave Manual (version 4.5) (Oxford: Photon Design)
[7] Soldano L B and Pennings E C M 1995 Optical multi-mode interference devices based on
self-imaging: principles and applications J. Lightwave Technol. 13 615–27
[8] Liu J-M 2005 Photonic Devices (Cambridge: Cambridge University Press)
[9] Ma Y, Zhang Y, Yang S, Novack A, Ding R, Lim A E-J, Lo G-Q, Baehr-Jones T and
Hochberg M 2013 Ultralow loss single layer submicron silicon waveguide crossing for SOI
optical interconnect Opt. Exp. 21 29374–82
[10] Bachmann M, Besse P A and Melchior H 1994 General self-imaging properties in N × N
multimode interference couplers including phase relations Appl. Opt. 33 3905–11
[11] Westerveld W J 2009 Design of a photonic integrated circuit (pic) in silicon on insulator
(SOI) technology for a novel chaotic integrated laser lightsource (chill) MSc Thesis
Technische Universiteit Delft
[12] McNab S J, Moll N and Vlasov Y A 2003 Ultra-low loss photonic integrated circuit with
membrane-type photonic crystal waveguides Opt. Exp. 11 2927–39
[13] Dakss M L, Kuhn L, Heidrich P F and Scott B A 1970 Grating coupler for efficient
excitation of optical guided waves in thin films Appl. Phys. Lett. 16 523–5
[14] Taillaert D, Bogaerts W, Bienstman P, Krauss T, van Daele P, Moerman I, Verstuyft S, de
Mesel K and Baets R 2002 An out-of-plane grating coupler for efficient butt-coupling
between compact planar waveguides and single-mode fibers IEEE J. Quantum Electron. 38
949–55

3-39
 Silicon Photonics

[15] Taillaert D, van Laere F, Ayre M, Bogaerts W, van Thourhout D, Bienstman P and Baets R
2006 Grating couplers for coupling between optical fibers and nanophotonic waveguides
Japan. J. Appl. Phys. 45 6071–7
[16] Roelkens G, Vermeulen D, Selvaraja S, Halir R, Bogaerts W and van Thourhout D 2011
Grating-based optical fiber interfaces for silicon-on-insulator photonic integrated circuits
IEEE J. Sel. Top. Quantum Electron. 17 571–80
[17] Photon Design 2016 Fimmprop, a bi-directional optical propagation tool (Oxford: Photon
Design) http://www.photond.com/products/fimmprop.htm
[18] Shani Y, Henry C H, Kistler R C, Orlowsky K J and Ackerman D A 1989 Efficient coupling
of a semiconductor laser to an optical fiber by means of a tapered waveguide on silicon Appl.
Phys. Lett. 55 2389–91
[19] Shoji T, Tsuchizawa T, Watanabe T, Yamada K and Morita H 2002 Low loss mode size
converter from 0.3 m square Si wire waveguides to singlemode fibres Electron. Lett. 38 1669–70
[20] Almeida V R, Panepucci R R and Lipson M 2003 Nanotaper for compact mode conversion
Opt. Lett. 28 1302–4
[21] Bakir B B, de Gyves A V, Orobtchouk R, Lyan P, Porzier C, Roman A and Fedeli J M 2010
Low-loss (<1 db) and polarization-insensitive edge fiber couplers fabricated on 200-mm
silicon-on-insulator wafers IEEE Photon. Technol. Lett. 22 739–741
[22] Dewanjee A, Caspers J N, Aitchison J S and Mojahedi M 2016 Demonstration of a compact
bilayer inverse taper coupler for Si-photonics with enhanced polarization insensitivity Opt.
Exp. 24 28194–203
[23] Westerveld W J, Urbach H P and Yousefi M 2011 Optimized 3-d simulation method for
modeling out-of-plane radiation in silicon photonic integrated circuits IEEE J. Quantum
Electron. 47 561–8
[24] Yariv A and Nakamura M 1977 Periodic structures for integrated optics IEEE J. Quantum
Electron. 13 233–51
[25] Streifer W, Scifres D R and Burnham R D 1976 Analysis of grating-coupled radiation in
gaas:gaalas lasers and waveguides IEEE J. Quantum Electron. 12 422–8
[26] Goodman J 1968 Introduction to Fourier Optics (New York: McGraw-Hill)
[27] Shen F and Wang A 2006 Fast-Fourier-transform based numerical integration method for
the Rayleigh–Sommerfeld diffraction formula Appl. Opt. 45 1102–10
[28] Gisolf D and Verschuur E 2010 The Principles of Quantitative Acoustical Imaging (Houten,
The Netherlands: Eagle Publications)
[29] Hunsperger R G 2002 Integrated Optics 5th edn (Berlin: Springer)
[30] Taillaert D, Bienstman P and Baets R 2004 Compact efficient broadband grating coupler for
silicon-on-insulator waveguides Opt. Lett. 29 2749–51
[31] Li Y, Li L, Tian B, Roelkens G and Baets R 2013 Reflectionless tilted grating couplers with
improved coupling efficiency based on a silicon overlay IEEE Photon. Technol. Lett. 25 1195–8
[32] Benedikovic D et al 2015 High-directionality fiber-chip grating coupler with interleaved
trenches and subwavelength index-matching structure Opt. Lett. 40 4190–3
[33] Vermeulen D, Selvaraja S, Verheyen P, Lepage G, Bogaerts W, Absil P, Thourhout D V and
Roelkens G 2010 High-efficiency fiber-to-chip grating couplers realized using an advanced
CMOS-compatible silicon-on-insulator platform Opt. Exp. 18 18278–283
[34] Alonso-Ramos C, Cheben P, nux A O-M, Schmid J H, Xu D-X and Molina-Fernández I
2014 Fiber-chip grating coupler based on interleaved trenches with directionality exceeding
95% Opt. Lett. 39 5351–4

3-40
 Silicon Photonics

[35] Chen X, Li C, Fung C, Lo S and Tsang H 2010 Apodized waveguide grating couplers for
efficient coupling to optical fibers IEEE Photon. Technol. Lett. 22 1156–8
[36] Van Laere F, Bogaerts W, Taillaert D, Dumon P, van Thourhout D and Baets R 2007
Compact focusing grating couplers between optical fibers and silicon-on-insulator photonic
wire waveguides Optical Fiber Communication and the National Fiber Optic Engineers Conf.,
2007 OFC/NFOEC 2007 pp 1–3
[37] Vermeulen D, Koninck Y D, Li Y, Lambert E, Bogaerts W, Baets R and Roelkens G 2012
Reflectionless grating couplers for silicon-on-insulator photonic integrated circuits Opt. Exp.
20 22278–83
[38] Li Y, Vermeulen D, Koninck Y D, Yurtsever G, Roelkens G and Baets R 2012 Compact
grating couplers on silicon-on-insulator with reduced back reflection Opt. Lett. 37 4356–8
[39] Yariv A 2000 universal relations for coupling of optical power between microresonators and
dielectric waveguides Electron. Lett. 36 321–2
[40] Westerveld W, Pozo J, Leinders S, Yousefi M and Urbach H 2014 Demonstration of large
coupling-induced phase delay in silicon directional cross-couplers IEEE J. Sel. Top.
Quantum Electron. 20 1–6
[41] Feng S, Lei T, Chen H, Cai H, Luo X and Poon A 2012 Silicon photonics: from a
microresonator perspective Laser Photon. Rev. 6 145–77
[42] Engin E et al 2013 Photon pair generation in a silicon micro-ring resonator with reverse bias
enhancement Opt. Exp. 21 27826–34

3-41
 IOP Publishing

Silicon Photonics
Electromagnetic theory
Wouter J Westerveld and H Paul Urbach

Chapter 4
Computational methods

In this chapter, we describe two numerical methods that are frequently used for the
modelling of silicon photonic components. As most often in this book, we assume
passive lossless linear materials thereby excluding non-linear effects, sources, and
detectors. The finite difference time domain (FDTD) and eigenmode expansion
(EME) methods are both rigorous in the sense that there are no physical
approximations and by refining the mesh the numerical solutions converge to the
exact solution of Maxwell’s equations. The FDTD method is quite general but also
computationally demanding, thereby it is less suitable for larger devices and
generally slower than other methods described in this paragraph (section 4.1). The
eigenmode expansion method is most suited for a stack of concatenated sections that
are translational invariant over the propagation-direction. For example, the MMI
coupler, the directional coupler, or the out-of-plane grating coupler (section 4.2).
Another popular method is the beam propagation method, however, this method is
less accurate for the high refractive index contrast of silicon-in-insulator waveguides.
Rigorous numerical mode solvers have been discussed in section 2.5.

4.1 Finite difference time domain (FDTD)

In the FDTD method, Maxwell’s equations are replaced by a set of finite difference
equations. The change of the electromagnetic fields with time is computed by
integrating Maxwell’s equations. This method is very generally applicable but
memory and computational time are significant for large devices compared to the
eigenmode expansion or beam propagation methods. Another disadvantage is
inaccuracy near oblique interfaces because the grid is rectangular. We follow the
1966 paper of Yee [1].
We formulate Maxwell’s equations in the time domain in terms of the electric
field E , the magnetic field B , the electric displacement D , and the ‘magnetic’ field
H = B/μ, the volume current density J , the permittivity ϵ and the permeability μ
[4]. The permittivity ϵ and the permeability μ can be anisotropic and

doi:10.1088/978-0-7503-1386-5ch4 4-1 ª IOP Publishing Ltd 2017

 Silicon Photonics

inhomogeneous. Lossy materials and metals, with the dispersion for example
incorporated using Drude’s model, can also be modelled with the FDTD method.
For simplicitly, we consider here only isotropic and non-dispersive materials.
Maxwell’s equations in an isotropic medium are
∂B
= −∇ × E , (4.1)
∂t

∂D
= ∇ × H + J, (4.2)
∂t

B = μH , (4.3)

D = ϵE (4.4)
where J , is assumed to be a given function of space and time while μ(x , y, z ) and
ϵ(x , y, z ) are given functions of space. For passive silicon photonic components, at
the wavelengths of interest, the magnetic permeability μ = μ0 is equal to that of
vacuum and losses and dispersion can be neglected. In a rectangular coordinate
system, equations (4.1) and (4.2) are equivalent to the following system of equations
∂Bx ∂Ey ∂E
= − z, (4.5)
∂t ∂z ∂y

∂By ∂E ∂E
= z − x, (4.6)
∂t ∂x ∂z

∂Bz ∂E ∂Ey
= x − , (4.7)
∂t ∂y ∂x

∂Dx ∂Hz ∂Hy

= − − Jx, (4.8)
∂t ∂y ∂z

∂Dy ∂Hx ∂Hz

= − − J y, (4.9)
∂t ∂z ∂x

∂Dz ∂Hy ∂Hz

= − − Jz. (4.10)
∂t ∂x ∂y
We sample space in discrete points and denote a grid point in space by
(i Δx , j Δy , k Δz ) → (i , j , k ) (4.11)
and for any function of space and time we write
F (i Δx , j Δy , k Δz , nΔt ) → F n(i , j , k ). (4.12)

4-2
 Silicon Photonics

We consider the folowing equation with respect to time,

∂F (x , y , z , t )
= G (x , y , z , t ), (4.13)
∂t
where F and G are components of vector fields occuring in Maxwell’s equations. We
approximate the partial derivative with respect to time at the left of equation (4.13)
using discrete time step δt and write:
F (x , y , z , t + Δt /2) − F (x , y , z , t − Δt /2)
≈ G (x , y , z , t ). (4.14)
Δt
The same approach is used for the differentials with respect to space. It is seen that it
is convenient to know F at times t ± Δt/2 and G at times t. In a numerical
implementation, it is convenient to store F and G staggered in time, such that G is
known at times nΔt while F is known at times (n + 1/2)Δt . This staggering may also
be used for the grid points in space. In this way we get difference equations that
approximate equations (4.5)–(4.8). The difference equation that approximates
equation (4.5), is for example
⎛ 1 1⎞ ⎛ 1 1⎞
B nx+1/2 ⎜i , j + , k + ⎟ − B nx−1/2 ⎜i , j + , k + ⎟
⎝ 2 2 ⎠ ⎝ 2 2⎠
Δt
⎛ 1 ⎞ ⎛ 1 ⎞
E ny ⎜i , j + , k + 1⎟ − E ny ⎜i , j + , k ⎟
⎝ 2 ⎠ ⎝ 2 ⎠ (4.15)
=
Δz
⎛ 1 ⎞ ⎛ 1⎞
E nz ⎜i , j + 1, k + ⎟ − E nz ⎜i , j , k + ⎟
⎝ 2⎠ ⎝ 2⎠
− .
Δy

The finite difference equations corresponding to equation (4.6) and equation (4.6)
can be similarly constructed. For equation (4.8) we find
⎛ 1 ⎞ ⎛ 1 ⎞
D nx ⎜i + , j , k ⎟ − D nx−1⎜i + , j , k ⎟
⎝ 2 ⎠ ⎝ 2 ⎠
Δt
⎛ 1 1 ⎞ ⎛ 1 1 ⎞
H nz −1/2⎜i + , j + , k ⎟ − H nz −1/2⎜i + , j − , k ⎟
⎝ 2 2 ⎠ ⎝ 2 2 ⎠
=
Δy (4.16)
⎛ 1 1⎞ ⎛ 1 1⎞
H ny−1/2⎜i + , j , k + ⎟ − H ny−1/2⎜i + , j , k − ⎟
⎝ 2 2 ⎠ ⎝ 2 2⎠
−
Δz
⎛ 1 ⎞
− J nx−1/2⎜i + , j , k ⎟.
⎝ 2 ⎠

4-3
 Silicon Photonics

The equations corresponding to equation (4.9) and equation (4.10) can be con-
structed in a similar way.
Figure 4.1 depicts the grid positions where the electromagnetic field components
are stored. As discussed before, the grid points of the E and H fields are chosen such
that the finite difference equations approximate the differential equations (4.5)–(4.8)
such that for given problem the error in the computed field decreases quadratically
with grid spacing Δx , Δy and Δz . For an acceptable error the size of the spatial grid
should be a tenth of a wavelength measure in the material, or less. For stability it is
essential that the time step is chosen to be small so that errors cannot propagate over
more than a grid space. This means that Δt < cmax Δx 2 + Δy 2 + Δz 2 , where cmax
is the maximum speed of light inside the materials of the problem, i.e., cmax is the
maximum of the real part of 1/ ϵμ , taken over all materials in the configuration
In the FDTD method the discrete equations for B and D are solved consec-
1
utively. Starting from known fields B at time index n − 2 and E at time n. The B

Figure 4.1. Grid positions of the electric and magnetic field components. The E-components are in the middle
of the edges and the H-components are in the center of the faces. This grid staggered grid is known as the Yee
cell after its inventor [1]. This image has been obtained by the author(s) from the Wikispaces website (https://
fdtd.wikispaces.com/The+Yee+Cell) where it was made available by guest (117.195.247.135) under a CC BY
3.0 licence. It is included on that basis. It is attributed to guest (117.195.247.135).

4-4
 Silicon Photonics

1 1
field at time n + 2
is computed from the B field at time n − 2
and the E field at time
1
n with equation (4.15). Then the H field at time n + is computed using equation
2
(4.3). Then the D field at time n + 1 is computed from the D field at time n, the H
field at time n + 12 and the known current J at time n. Thereafter, E at time index
n + 1 is computed using equation (4.4). This procedure is repeated if the fields are
known at all desired times. When there is no dispersion, the electromagnetic fields
are only stored at one time index and therefore the FDTD method requires only
little computer memory compared to other numerical solvers such as the finite
element method. In the case of dispersive media, more memory is required because
the fields have to be stored at more times to compute the field at the next time index.
Physical quantities of interest such as the power flow or the absorbed electric energy
are computed from the fields.
The current J(x , y , z, t ) can be used to excite an electromagnetic field. The
system can be excited with a time-pulse of given bandwidth and the propagation of
this pulse is followed through the device as a function of time. Typically the fields at
specific locations, e.g. the input and output waveguides of a device, are stored as a
function of time. By Fourier tranforming the computed electromagnetic field as
a function of time, information about a whole set of frequencies can be obtained at
once. If one is only interested in time harmonic fields of a certain frequency, the
system may be exited using a continuous-wave (single-frequency) excitation starting
at some time, say at t = 0. Then the time iteration to solve the discretized Maxwell’s
equations is repeated until field values separated in time by a period are almost
identical.
As is the case for every numerical method based on discretizing Maxwell’s
differential equations, the radiation boundary conditions state that the field
scattered by the objects and radiated by the current sources (if present) propagates
outwards. In particular, these fields vanish at large distances. But they vanish rather
slowly in proportion to the reciprocal distance. Using a computational domain that
is so large that the outwards propagating fields can be set equal to zero on the
boundary is not feasible because it would require a far too large domain and hence
require too much memory and computational power. However, a very elegant
solution to this problem exists, namely the perfectly matched layer (PML), which
was first introduced by Berenger [5]. A cube is chosen as computational domain that
is as small as possible while containing all non-trivial scatterers in its interior. For
silicon photonic integrated circuits, the computational domain is typically chosen
such that the distance from the device to the edge of the cube is of the order of one
free-space wavelength, except for the input and output waveguides. This cube is
placed inside a somewhat larger cube. The volume between the larger cube and the
computational domain is called the PML. The permittivity and permeability of this
layer are chosen such that they vary smoothly as functions of the coordinate in the
PML that is locally perpendicular to the boundary of the cube. Futhermore, the
material properties are chosen equal to those of the materials in the computational
domain and adjacent to the interface between the computational domain and the
PML. This ensures that the scattered and radiated fields are not reflected by this

4-5
 Silicon Photonics

Figure 4.2. Example of 2D modelling of an out-of-plane grating coupler that is designed for perfectly vertical
coupling [2] with the finite difference time domain (FDTD) method. (Left) sketch of the coupler and simulation
layout. Electromagnetic fields are excited at the position of the yellow line, with a Gaussian spatial profile.
Electromagnetic fields are recorded at the positions of sensors indicated with the red vertical lines. (Right)
snapshot of the electromagnetic field during the simulation. Red represents positive amplitude and blue
represents negative amplitude. The figure was computed using OmniSim, omni-directional photonic simu-
lations with an FDTD engine. Courtesy of Photon Design, reprinted from OmniSim documentation [3].

interface. Furthermore, the permittivity and permeability in the PML are chosen
such that the field is absorbed. By choosing an appropriate thickness for the PML,
the field that has penetrated the PML can be made to almost vanish on the outer
boundary of the PML. This is then adopted as a new simple boundary condition on
the enlarged domain. Hence the PML means that the boundary conditions become
very simple at the cost of a somewhat enlarged computational domain, which,
however, is much smaller than the domain would have to be were PML not be used.
Figure 4.2 depicts an example of a 2D FDTD simulation of an out-of-plane
grating coupler. In this case, the electromagnetic field is excited at the position where
the optical fibre is placed. The excitation has the same field profile as the mode of an
optical fibre.

4.2 The eigenmode expansion method

Eigenmode expansion can be used to rigorously compute the behaviour of optical
devices and is especially suited for devices that consist of a number of sections that
are invariant in the propagation direction (z). The MMI coupler is a typical example
of such a device (see figure 4.3) but also radiating devices such as out-of-plane
grating couplers can be modelled. More advanced numerical techniques are needed
for devices that gradually change with z, such as the inverted taper [7, 8]. In each
section, the electromagnetic field is expanded in the eigenmodes of the section,
including guided, radiation and evanescent modes. At the interfaces, Maxwell’s
equations are satisfied by requiring that the tangential components of the electric
and magnetic field are continuous.
This section is largely adapted from the Thesis of P Bientsman [6] with examples
from Photon Design [8]. After revising the properties of unidirectional modes
(section 4.2.1) and a note on the number of modes (section 4.2.2), we present mode
matching at an interface (section 4.2.3), two schemes for the propagation of waves
through a device (T-matrix scheme, section 4.2.4 and S-matrix scheme, section
4.2.5), and two examples (section 4.2.6). Additional details on the numerical
implementation and completeness of modes may be found in [6].

4-6
 Silicon Photonics

Figure 4.3. Eigenmode expansion method. (a) Scattering at an interface: incident field, reflected field, and
transmitted field. (b) Stack of sections with interface 1 ↔ 2, z-invariant propagation section 2 ↔ 3 of length l23,
and interface 3 ↔ 4. (c) Example of how such a stack describes an 1 × 2 MMI coupler. Adapted from [6]
courtesy of Peter Bienstman.

4.2.1 Properties of unidirectional modes

Orthogonality of modes and the mode expansion conjecture are explained in
sections 2.3.3 and 2.3.6, respectively. In the eigenmode expansion method we will
treat the forward and backward propagating modes separately. We use the
unconjugated othogonality relation (2.58):
∞
1
2
∬−∞ (Ei × Hj ) · zˆ dxdy = 0 if i ≠ j. (4.17)

where Ei and Hj are the electric and magnetic fields of modes i and j, respectively.
Compared to equation (2.57) there is no complex conjugate involved. In this
chapter, we adopt the following definition of the bra-ket notation:
∞
1
〈Ei∣Hj〉 =
2
∬−∞ (Ei × Hj ) · zˆ dxdy, (4.18)

and the modes are normalized using equations (2.61) and (2.62):
〈Ei∣Hj〉 = δij. (4.19)

4.2.2 Number of modes

In numerical implementations of the eigenmode expansion method, a finite number
of N discrete modes is used. The continuum of radiation modes and evanescent
modes can be discretized by applying additional boundary conditions at a distance
far enough from the waveguide such that the influence on the guided modes can be
neglected. For example, the waveguide may be encapsulated in a large rectangular
tube with perfectly conducting metal walls, demanding the tangential

4-7
 Silicon Photonics

electromagnetic field components to be zero at these walls. This configuration has

a discrete set of radiative and evanescent modes. The examples of a 2D model of an
out-of-plane grating coupler and a three-dimensional model of a 1 × 4 MMI coupler
required approximately 30 and 40 modes, respectively (see section 4.2.6). In some
devices, such as MMI couplers or directional couplers, the coupling to radiation
modes may be neglected and it is sufficient to only include guided modes.

4.2.3 Mode matching

Consider a flat interface between waveguides I and II, coinciding with the plane z = 0
(see figure 4.3(a)). A single mode with index p is incident from medium I and gives rise
to a backward propagating electromagnetic field in medium I and a forward
propagating field in medium II. We expand the reflected field in the eigenmodes of
medium I and the transmitted field in the eigenmodes of medium II. The mode
matching technique starts by imposing continuity of the tangential components of the
total field [6, 9],
E pI,t + ∑Rk,pE kI,t = ∑Tj,pE jII,t , (4.20)
k j

H pI,t − ∑Rk,pHkI,t = ∑Tj,pH jII,t , (4.21)

k j

where {EI , H I } and {EII , H II } are the modes of media I and II, respectively.
Subscript t denotes either of the tangential components x and y. The minus sign
for the reflected H field is due to the symmetries between forward and backward
propagating modes discussed in equations (2.54–2.55). In the eigenmode expansion
method, we sum over a finite number of relevant modes. To calculate the unknown
reflection and transmission coefficients, Rk,p and Tj,p, we take the cross product of
equation (4.20) with HiI,t from the right and the cross product of equation (4.21) with
EiI,t from the left. Here, i is an arbitrary mode index. After integrating over the
interface, we get
E pI HiI + ∑Rk,p E kI HiI = ∑Tj,p E jII HiI , (4.22)
k j

EiI H pI − ∑Rk,p EiI HkI = ∑Tj,p EiI H jII , (4.23)

k j

with the inner product defined in equation (4.18). The orthogonality relation (4.19)
implies
δip + Ri ,p = ∑Tj,p E jII HiI , (4.24)
j

4-8
 Silicon Photonics

δip − Ri ,p = ∑Tj,p EiI H jII . (4.25)

j

Adding and subtracting these equations yields

2δip = ∑( EiI H jII + E jII HiI Tj ,p, ) (4.26)

j

1
Ri ,p =
2
∑( E jII HiI − EiI H jII Tj ,p. ) (4.27)
j

The summation is truncated after N relevant modes. This shows that the transmission
coefficients Tj,p for each j and incident mode p can be found by solving the linear system
(4.26). Thereafter, the reflection coefficients Ri,p for arbitrary i follow using (4.27).
We proceed by writing equations (4.26) and (4.27) as matrix equations, thereby
introducing the so-called transmission matrix T and reflection matrix R which are
the transmission coefficients Ti,j and reflection coefficients Tk,l in matrix notation. We
define the overlap matrices O and O′ by
Oij ≡ EiI H jII , (4.28)

Oij′ ≡ E jII HiI , (4.29)

and rewrite equations (4.26) and (4.27) as1

T = 2(O + O ′)−1, (4.31)

1 ′
R= (O − O )T , (4.32)
2
where ()−1 indicates matrix inversion.
As Maxwell’s equations are linear, scattering of an arbitrary incident wavefield
from medium I can be found by expanding the incident field in modes. Let
fI = (fI ,1 , fI ,2 , fI ,3 , …)T be the modal amplitudes of the forward incident waves,
which cause backwards traveling reflected waves with amplitudes bI and forward
traveling transmitted waves in medium II with amplitudes fII . These modal
amplitudes are related via
bI = R(I →II ) fI , (4.33)

1
Let us rewrite equation (4.31) to equation (4.26). Left multiplication of equation (4.31) by A ≡ (O + O′)
gives AT = 2I , with I the identity matrix. We consider element i , p of this equation and use
( )
(AT )i,p = ∑j Ai,j Tj ,p and Ai,j = 〈EiI ∣H jII 〉 + 〈E jII ∣HiI 〉 to arrive at equation (4.26) Let us also rewrite
equation (4.32) to equation (4.27). We define B = (O ′ − O ) and consider element i , p. Using
( )
(BT )i,p = ∑j Bi,jTj ,p and Bi,j = 〈E jII ∣HiI 〉 − 〈EiI ∣H jII 〉 , we arrive at equation (4.27).

4-9
 Silicon Photonics

fII = T(I →II ) fI , (4.34)

where the subscript (I → II ) has been added to empathize that the incident waves
come from medium I. Obviously, we can repeat the entire procedure for incidence
from medium II, which gives us the matrices R(II →I ) and T(II →I ). These four matrices
completely characterize the scattering that occurs at an interface.

4.2.4 The T-matrix scheme

In order to compute the scattering of a stack of sections, as occurs for example in an
MMI coupler, it is necessary to compute the scattering at each interface and the
mode propagation from one interface to the next. The previous section describes the
relation between incident waves from one side of an interface to outwards traveling
reflected or transmitted waves. The propagation of modes in a section of the stack is
straightforward as each mode ν propagates with its modal propagation constant βν .
Generally, waves propagate in both directions, for example due to reflections.
In the T-Matrix scheme, the equations are written such that the modal amplitudes
at z-position j are computed from the modal amplitudes at z-position i (see figure
4.3(b)), i.e.,
⎡f ⎤ ⎡f⎤
⎢ j ⎥ = T˜(j→i ) ⎢ i ⎥ (4.35)
⎢⎣ bj ⎥⎦ ⎣ bi ⎦

where fi and bi are column vectors containing the modal amplitudes of the forwards
and backwards traveling modes at z-position i. The transfer matrix T˜(i→j ) character-
izes the scattering from position i to position j, and may describe interfaces, e.g.
T˜(1→2), propagation, e.g. T˜(2→3), or both, e.g. T˜(1→4). The T-matrix T̃ is different from
the transmission matrix T as it describes both forwards and backwards traveling
waves simultaneously and may also include propagation.
To determine the transfer matrix T˜(1→2) at the interface we start from the
transmission and reflection matrices as used in equations (4.33) and (4.33) but
with incidence from both sides simultaneously,
f2 = T(1→2) f1 + R(2→1)b 2 , (4.36)

b1 = R(1→2)f1 + T(2→1)b 2 , (4.37)

with amplitude vectors depicted in figure 4.3(b). We solve equation (4.37) for b2 and
find
1 ⎡ ⎤
b 2 = T(2−→1) ⎣b1 − R(1→2)f1 ⎦ , (4.38)

with superscript −1 denoting the inverse matrix such that T(2−→

1
1)T(2→1) = I with I the
identity matrix. Inserting this expression for b2 back in equations (4.36) and (4.37)
and rearranging as matrix equation, we find

4-10
 Silicon Photonics

⎡ f ⎤ ⎡T(1→2) − R(2→1)T(2−→1
1)R(1→2)
1 ⎤
R(2→1)T(2−→1) ⎡ f1 ⎤
⎢ ⎥=
2 ⎢ ⎥ ⎢ ⎥. (4.39)
⎣ b 2 ⎦ ⎢⎣ 1
− T(2−→ 1)R(1→2)
1
T(2−→ 1)
⎥⎦ ⎣ b1⎦

The transfer matrix for propagation through a z-invariant section is easy to write
down using equation (2.30). For section 2 → 3,
⎡ f ⎤ ⎡ diag(e−ıβν l23) 0 ⎤⎡ f ⎤
⎢ 3⎥ = ⎢ ⎥ ⎢ 2 ⎥, (4.40)
⎣b 3⎦ ⎣ 0 diag(e ν 23)⎦ ⎣ b 2 ⎦
ıβ l

where the submatrices diag are diagonal matrices containing the propagation
constants βν of each eigenmode ν of section 2 ↔ 3, with l23 the length of this section.
After calculating the transfer matrixes for the interfaces and propagation sections,
the transfer matrix for the whole stack in figure 4.3(b) is
T˜(1→4) = T˜(3→4) T˜(2→3) T˜(1→2), (4.41)

as follows from equation (4.35).

The big disadvantage of the T-matrix scheme in a numerical implementation is
that for evanescent modes equation (4.40) contains a mixture of increasing and
decreasing exponentials, which is detrimental for numerical stability. During
calculation, very small numbers will be added to very large numbers, with a loss
in computational precision. The next section discusses the S-matrix scheme, where
this problem does not occur.

4.2.5 The S-matrix scheme

The scattering matrix, or S-matrix, relates the outwards propagating fields of a
structure to the inwards propagating fields, i.e.,
⎡f ⎤ ⎡f⎤
⎢ j ⎥ = S˜(j→i ) ⎢ i ⎥ (4.42)
⎣ bi ⎦ ⎣ bj ⎦

with fi , bi , f j and bj the modal amplitudes of the forward and backward traveling
waves at z-positions i and j and with S̃ the scattering matrix. The S-matrix directly
follows from the reflection and transmission matrices, c.f. equations (4.33) and
(4.34), but with the incident waves from both sides superimposed as in equations
(4.36) and (4.37), e.g.
⎡ f ⎤ ⎡ T(i→j ) R(j→i )⎤ ⎡ f ⎤
⎢ j⎥ = ⎢ ⎥ ⎢ i ⎥, (4.43)
⎣ bi ⎦ ⎣ (i→j ) (j→i ) ⎦ ⎣ bj ⎦
R T

for interface i ↔ j .
In the S-matrix scheme, we directly write down the S-matrix for an interface with
the adjacent propagation section, thus directly from z-position i to z-position i + 2.
We consider the interface with section 1 → 3. The interface is described by equation

4-11
 Silicon Photonics

(4.43) with i = 1 and j = 2. The propagation from position 2 to position 3 is given by

equation (4.40),
f2 = diag(eıβν l23) f3 , (4.44)

b 2 = diag(e−ıβν l23) b 3. (4.45)

Inserting f2 and b2 in equation (4.43) with i = 1 and j = 2 gives

⎡ f ⎤ ⎡ diag(e−ıβν l23) T(1→2) diag(e−ıβν l23) R(2→1) diag(e−ıβν l23)⎤ ⎡ f ⎤
⎢ 3⎥ = ⎢ ⎥ ⎢ 1 ⎥. (4.46)
⎣ b1⎦ ⎣⎢ R(1→2) T(2→1) diag(e−ıβν l23) ⎥⎦ ⎣ b 3⎦

In contrast to the T-matrix scheme, the propagation as described in equation (4.46)

only includes exponentials with modulus smaller than or equal to one, which makes
this scheme numerically always stable.
The next task at hand is to combine two scattering matrices which describe
adjacent parts of the device. These can be interface matrices, propagation matrices,
or the scattering matrix of an entire stack of sections. The case of an interface
scattering matrix S˜(1→2) with an adjacent propagation scattering matrix S˜(2→3) was
considered in equation (4.46). We now compute combined scattering matrix S˜(1→4) of
two known scattering matrices S˜(1→3) and S˜(3→4), which describes the structure in
figure 4.3(b, c). For known scattering matrices S˜(1→3) and S˜(3→4), equation (4.43)
reads
f3 = T(1→3) f1 + R(3→1)b 3, (4.47)

b1 = R(1→3) f1 + T(3→1)b 3, (4.48)

and
f4 = T(3→4) f3 + R(4→3)b4, (4.49)

b 3 = R(3→4)f3 + T(4→3)b4, (4.50)

respectively. First we solve equation (4.47) for R(3→1)b3, then left-multiply equation
(4.50) with R(3→1), and thereafter equate these two equations to find
f3 − T(1→3) f1 = R(3→1)R(3→4) f3 + R(3→1)T(4→3)b4, (4.51)
or
f3 = [I − R(3→1)R(3→4)]−1T(1→3) f1
(4.52)
+ [I − R(3→1)R(3→4) ]−1R(3→1)T(4→3)b4,
−1
with identity matrix I and superscript denoting the matrix inverse. Inserting
equation (4.52) in equation (4.49) gives

4-12
 Silicon Photonics

f4 = T(3→4)[I − R(3→1)R(3→4)]−1T(1→3)f1
(4.53)
+ T(3→4)[I − R(3→1)R(3→4)]−1R(3→1)T(4→3)b4 + R(4→3)b4.

Now we solve equation (4.50) for R(3→4)f3, then multiply equation (4.47) with R(3→4),
and thereafter equate these two equations to find
b 3 − T(4→3)b4 = R(3→4)T(1→3)f1 + R(3→4)R(3→1)b 3, (4.54)
or

b3 = [I − R(3 → 4)R(3 → 1) ]−1R(3 → 4)T(1 → 3)f1

(4.55)
+ [I − R(3 → 4)R(3 → 1) ]−1T(4 → 3)b4.

Inserting equation (4.55) in equation (4.48) gives

b1 = T(3 → 1)[I − R(3 → 4)R(3 → 1) ]−1R(3 → 4)T(1 → 3) f1 + R(1 → 3) f1

(4.56)
+ T(3 → 1)[I − R(3 → 4)R(3 → 1) ]−1T(4 → 3)b4.

From equations (4.56) and (4.53) we deduce the scattering matrix S˜(1→4) in the form
of equation (4.43) with
T(1 → 4) = T(3 → 4)[I − R(3 → 1)R(3 → 4) ]−1T(1 → 3), (4.57)

R(4 → 1) = T(3 → 4)[I − R(3 → 1)R(3 → 4) ]−1R(3 → 1)T(4 → 3) + R(4 → 3), (4.58)

R(1 → 4) = T(3 → 1)[I − R(3 → 4)R(3 → 1) ]−1R(3 → 4)T(1 → 3) + R(1 → 3), (4.59)

T(4 → 1) = T(3 → 1)[I − R(3 → 4)R(3 → 1) ]−1T(4 → 3). (4.60)

We know how to compute scattering matrices for interfaces, equations (4.31) and
(4.32), how to compute propagation in z-invariant sections, equation (4.46) and how
to compute the overall scattering matrix of two adjacent sections, equations (4.57–
4.60). Hence the overall scattering matrix of the device can be computed by starting
at the first interface and working our way through the rest of the device.

4.2.6 Examples: MMI coupler and grating coupler

In this section we consider two cases of using the eigenmode expansion to compute
the electromagnetic field. The first is the design of a 1 × 4 MMI coupler shown in
figure 4.4. It is desired that (1) most light travels from the left input waveguide to the
right output waveguides, i.e., low reflection and radiation, (2) all four output
waveguides carry similar power, and (3) that these two requirements are not
significantly affected when the MMI is fabricated to be slightly different from the
design. The following design parameters have to be chosen correctly to arrive at
the optimal design: the length of the MMI waveguide, its width, the positions of the

4-13
 Silicon Photonics

Figure 4.4. Example of the modeling of a 1 × 4 multimode interference (MMI) coupler using eigenmode
expansion. This device has a 220 nm thick silicion light-guided layer (n = 3.5) embedded in a silicon-dioxide
cladding (n = 1.44) and the wavelength is λ = 1.55 μm. (Left) sketch. (Right) intensity profile computed with
the FimmProp eigenmode expansion software. Courtesy of Photon Design, reprinted from FimmProp
documentation [8].

Figure 4.5. Example of the modeling of an out-of-plane grating coupler that is designed for perfectly vertical
coupling [2]. (Left) sketch. (Right) snapshot of one component of the electric field as computed with eigenmode
expansion software. Courtesy of Photon Design, reprinted from FimmProp documentation [8].

output waveguides, and the widths of the input and output waveguides. It is
generally necessary to perform many simulations with different parameters to arrive
at the optimal design. Therefore, the higher speed of the eigenmode expansion
method, compared to, e.g. the FDTD method, is important. The design shown in
figure 4.4 has MMI waveguide length of 30.1 μm, width of 8.2 μm, and input/output
waveguides widths of 1.5 μm. The total power transmitted from the input waveguide
to the output waveguides is 99% and the imbalance (relative difference in power)
between the output waveguides is 3 · 10−7.
In figure 4.5 the out-of-plane grating coupler proposed in [2] is shown. This
coupler was optimized for vertical coupling in the y-direction which is, for example,
desired when a downwards-emitting VCSEL laser is positioned on the planar circuit.
In this design special grating teeth are used to improve the upwards coupling. Also, a
slit is applied in front of the grating to reduce reflections. The design is in silicon-on-
insulator and has a computed coupling efficiency of 50%. To arrive at this design,
many simulations were required.
Tapers, which gradually vary with z, may be simulated using eigenmode expansion
by slicing the taper into many small sections of constant cross-section—a staircase
approximation. This z-discretisation is inefficient because the regular steps create a
non-physical grating, which can cause spurious coupling between pairs of modes.

4-14
 Silicon Photonics

Advanced methods get away from the staircase approximation by assuming the
cross-section or some other intermediate quantity varies linearly (or to some higher
power) along the small section and then computing the resulting cross-coupling
between the modes. This method is developed by Photon Design (Oxford, UK) and
implemented in their commercial FimmProp software [7, 8]. Additionally, this can
be combined with an efficient and adaptive choice of the discretisation of the taper in
the z-direction [7, 8, 10].

References
[1] Yee K S 1966 Numerical solution of initial boundary value problems involving Maxwell’s
equations in isotropic media IEEE Trans. Antennas Propag. 14 302–7
[2] Roelkens G, Thourhout D V and Baets R 2007 SOI grating structure for perfectly vertical
fiber coupling European Conf. on Integrated Optics (ECIO) (Denmark) FC4
[3] Photon Design 2016 Omnisim, omni-directional photonic simulations (Oxford: Photon
Design) www.photond.com/products/omnisim.htm
[4] Griffiths D J 1999 Introduction to Electrodynamics 3rd edn (Upper Saddle River, NJ:
Prentice-Hall)
[5] Berenger J-P 1994 A perfectly matched layer for the absorption of electromagnetic waves
J. Comput. Phys. 114 185–200
[6] Bienstman P 2001 Rigorous and efficient modelling of wavelength scale photonic compo-
nents” PhD Dissertation Universiteit Gent www.photonics.intec.ugent.be/publications/phd.
asp?ID=104
[7] Gallagher D F G and Felici R P 2003 Eigenmode expansion methods for simulation of
optical propagation in photonics: pros and cons Integrated Optics: Devices, Materials, and
Technologies VII, 69 (San Jose, CA, June 2003) Proc. SPIE 4987 290–94
[8] Photon Design 2016 Fimmprop, a bi-directional optical propagation tool (Oxford: Photon
Design) www.photond.com/products/fimmprop.htm
[9] Zaki K A, Seng-Woon C and Chunming C 1988 Modeling discontinuities in dielectric-
loaded waveguides IEEE Trans. Microw. Theory Tech. 36 1804–10
[10] Lumerical 2017 Lumerical’s eigenmode expansion (EME) solver (Vancouver: Lumerical)
www.lumerical.com/tcad-products/mode/EME

4-15
 IOP Publishing

Silicon Photonics
Electromagnetic theory
Wouter J Westerveld and H Paul Urbach

Chapter 5
Devices

In the previous chapters, integrated optical components in silicon photonic

technology were studied. A great variety of devices can be built from these
components [1–8]. In this chapter we discuss some basic but relevant examples of
sensors and modulators. Throughout this chapter, references are provided to
broader review articles and in-depth treatments.
In optical fibre communication, the transmitter sends binary data (a bit pattern of
zeros and ones) via an optical fibre to the receiver. In the most basic and frequently
used scheme, the transmitter modulates laser light with the bit pattern while the
receiving party measures the amplitude of this light and extracts this pattern. A
device modulating a laser that is continuously switched on is called a modulator.
Silicon photonics is a commercially important technology to use for these tasks
because the transmitter and receiver modules realised in silicon photonics are made
by CMOS fabrication processes of the electronics industry. This CMOS fabrication
technology offers mass production and integration of photonics and electronic
components on a single chip low cost. Commercial transceivers (combined trans-
mitter/receivers) based on silicon photonic technology are sold by various companies
such as Intel, Luxtera, and Teraxion [9–13].
Another application of silicon photonics is in sensors. Medical diagnostics
relies on accurate measurements and silicon ring resonators can be tailored to
measure biomedical markers for example in blood using refractive index sensing
[14–17]. Another application is to measure mechanical vibrations, such as ultra-
sound [18, 19].
Modulators and sensors are somewhat similar devices in the sense that light is
encoded with a quantity of interest, being either binary data or an analog physical
quantity. Waveguide based modulators or sensors are engineered such that the
effective index of the waveguide (ne = β /k ) is sensitive to this quantity. The real and
imaginary parts of the effective index may be affected, which results in a phase shift
or amplitude drop, respectively. Most devices are based on detecting a change in the

doi:10.1088/978-0-7503-1386-5ch5 5-1 ª IOP Publishing Ltd 2017

 Silicon Photonics

real part of the effective index, which amounts to detecting a phase shift. For this a
component is needed which changes the phase shift into a change of amplitude.
Sensors are most often based on ring resonators whose resonance frequencies are
very sensitive to the effective index of the ring waveguide. However, the refractive
index of silicon is very dependent on temperature hence the temperature of the ring
resonator has to be controlled or the temperature is measured by another ring
resonator. The induced shift in the resonance wavelengths can be measured by
recording the transmission spectrum. Alternatively, narrow-bandwith laser light can
be tuned to a wavelength at the flank of the resonance dip so that a shift of the
resonance causes an amplitude change of the transmitted light. Modulators have
different requirements than sensors. International standards on optical fibre com-
muncation impose stringent rules on the wavelengths that are to be used [20].
Temperature control consumes more power than is affordable in telecommunication
applications. Therefore modulators often use a Mach–Zehnder inteferometer to
translate an induced effective index change in one of its arms to an amplitude change
[21]. Interferometers are only sensitive to the difference between its arms and
temperature equally affects both arms.

5.1 Ring resonator sensors

This section introduces temperature, refractive index, and ultrasound sensors. These
sensor are applied to, e.g. medical diagnostics, chemical analysis, or industrial
testing. They are based on silicon photonic micro-ring resonators although other
configurations exist with particular advantages. The ring waveguide is placed in an
environment such that it is affected by the physical quantity of interest. Any change
in the length or effective refractive index of the waveguide causes a shift of the
resonances of the ring resonator and this shift is measured.

5.1.1 Ring resonators

The transmission properties of ring resonators are described in section 3.4. For a
ring resonator with a symmetric directional coupler (two identical waveguides), the
resonance wavelength λ m of resonance number m satisfies
m · λm = n e(λm , χ ) · l (χ ), (5.1)
with n e = β /k the effective refractive index of the waveguide, l the circumference of
the ring or racetrack, and χ a uniform physical parameter such as the temperature.
The effective index depends on the wavelength due to modal dispersion.
We derive the linear shift of the resonance wavelength λ m due to a change of the
physical parameter. Taking the partial derivative of equation (5.1) with respect to
χ gives
∂λm ∂n ∂n e ∂λm ∂l
m· = el + l + ne . (5.2)
∂χ ∂χ ∂λm ∂χ ∂χ

5-2
 Silicon Photonics

We evaluate equation (5.2) for given m at χ = 0 at which λc = λ m . Solving this

equation for ∂λ m /∂χ and substituting m from equation (5.1) at χ = 0 gives the
linearized influence of the physical parameter χ

∂λm ⎛ n ⎞ ⎛ λ ∂n λ ∂l ⎞
= ⎜ e ⎟ ⎜ c e + c ⎟, (5.3)
∂χ ⎝ ng ⎠ ⎝ n e ∂χ l ∂χ ⎠

where ng = n e − λc ∂n e /∂λ as derived in equation (2.32). The first term at the right of
equation (5.3), n e /ng , originates from the fact that the waveguide is dispersive
(the effective index n e is wavelength dependent), the second term is the change of the
effective index due to a change of the physical parameter χ, and the third term is the
change in track length due to a change of χ. The term originating from the dispersion
n e /ng is smaller than unity and therefore reduces the sensitivity of the sensor. This
effect is stronger for smaller waveguides as seen in figure 2.7(a) and figure 2.10(a).
The third term is zero when external physical effect χ only affects the effective index
of the waveguide but not the length of the resonator. This is the case for refractive
index sensors. This is also a good approximation for the influence of temperature on
ring resonators because the influence of the thermo-optic effect is much larger than
the influence of the thermal expansion of the ring resonator. When the length is not
influenced by the external physical effect χ, it can be seen in equation (5.3) that
the shift of the resonance wavelength does not depend on the circumference of the
racetrack. In the case of mechanical deformation, both the effective index and the
track length change.

5.1.2 Temperature sensors

Silicon has a strong thermooptic effect meaning that a small change in temperature
T causes a large change in the refractive index of silicon n1 (dn1/dT = 1.83 · 10−4
K−1 [22]). The influence of thermal expansion on the resonance shift is much smaller
(dl /dT = αl with α = 2.6 · 10−6 K−1). From the approximate analytical model for
rectangular waveguides (section 2.4), the temperature-induced change in effective
refractive index and the modal dispersion can be derived (section 5.1.4 and 2.4.5,
respectively). We consider a typical silicon-on-insulator waveguide of 220 nm
× 450 nm at the free-space wavelength λ = 1.55 μm and compute the effective index
n e = 2.3, and the group index ng = ∂β /∂k = 4.4. Therefore, the first term in equation
(5.3), n e /ng = 0.52. We neglect the effect of thermal expansion and compute the
influence of a change of temperature on the effective index:
∂n e /∂T = (1/k 0 )∂β /∂n e = 2.21 · 10−4 . We also neglect thermal expansion in equation
(5.3), meaning that we neglect the last term. This results in ∂λ m /∂χ = 78 pm K−1,
which agrees well with rigorously simulated and with measured values [23]. If the
thermal expansion is taken into account we get the value 84 pm K−1, when the
circumference of the ring is elongated according to the thermal expansion coef-
ficient. However, this is not the case because the silicon is attached to the silicon-
dioxide substrate with an expansion coefficient that is one order of magnitude lower.

5-3
 Silicon Photonics

5.1.3 Sensor systems and interrogation

There are different methods to interrogate the resonance wavelength shift of a ring
resonator. One method for interrogation is to measure the transmission spectrum.
This can be done using a broadband light source as input of the bus waveguide of a
ring resonator and while recording the light at the output of the bus waveguide using
a device called an optical spectrum analyser. The spectrum can also be recorded by
using a tuneable laser as input for the ring resonator and by recording the
transmitted light at the output using a photo-diode. Sweeping the laser emission
wavelength while simultaneously recording the transmitted light reveals the trans-
mission spectrum. In both methods, the recorded spectrum shows a dip at the
resonances (or a peak when the ring has two couplers and the ‘drop’ output is used),
and the exact position of the dip or peak may be found by fitting a curve. An
alternative method is to tune the wavelength of laser light to the flank of a
transmission resonance dip and record the transmitted intensity as a function of
time. A shift in the wavelength of the resonance dip with respect to the wavelength of
the laser causes a change in the transmitted intensity, hence the amplitude of the
transmitted light is modulated with the resonance wavelength shift. This method is
much faster but it can only follow the resonance peak if the shift is smaller than the
width (FWHM) of the resonance. There also exist more accurate methods, e.g. see
[24] or [25], and the references in [18, section 5.9]. When considering the accuracy of
a sensor system, it is always necessary to take both the ring resonator as well as the
interrogation system into account. For example, a ring with a waveguide with a very
small cross-section may be more sensitive to its vicinity than a typical SOI
waveguide, which is advantageous because it gives larger wavelength shift (section
5.1.5). However, such a small waveguide also has larger loss due to which the
FWHM of the resonance of the ring is wider (lower Q-factor) which causes a less
precise interrogation of the resonance wavelength (section 2.6 and 3.4.2).

5.1.4 Influence of an external effect on rectangular silicon waveguides

In this section we extend the theory of rectangular waveguides (section 2.4).
Analogue to the derivation of the effective group index in section 2.4.5, the
linearized influence of an external effect on the propagation constant of the mode
can be determined. For example, the influence of a change of temperature on the
waveguide or the influence of a change in the refractive index of the surrounding
medium can be calculated. As before, the external physical parameter is denoted by
χ. We assume that the refractive indices nj ( j = 1…5) and the waveguide its width d
and height b are known (see figure 2.5), as well as the first-order influence of the
external physical parameter on these properties, i.e., ∂nj /∂χ , ∂d /∂χ and ∂b/∂χ . Taking
the derivative of equation (2.70) with respect to χ gives
∂β 1⎛ ∂n ∂k ∂ky ⎞
= ⎜n1k 02 1 − kx x − ky ⎟. (5.4)
∂χ β⎝ ∂χ ∂χ ∂χ ⎠
Similar to the analysis in section 2.4.5, ∂kx /∂χ is found by taking the derivative of
equation (2.29), G = 0, and solving for ∂kx /∂χ :

5-4
 Silicon Photonics

∂G ∂n1 ∂G ∂n2 ∂G ∂n3 ∂G ∂d

+ + +
∂kx ∂n ∂χ ∂n2 ∂χ ∂n3 ∂χ ∂d ∂χ
=− 1 . (5.5)
∂χ ∂G
∂kx
Similarly ∂ky /∂χ is found from equation (2.17):
∂F ∂n1 ∂F ∂n 4 ∂F ∂n5 ∂F ∂b
+ + +
∂ky ∂n1 ∂χ ∂n 4 ∂χ ∂n5 ∂χ ∂b ∂χ
=− . (5.6)
∂χ ∂F
∂ky

We define
γ2 γ
αa ≡ 2
+ 32 , (5.7)
n2 n3

k x2 γγ
αb2 ≡ 4
− 22 3 2 , (5.8)
n1 n 2 n3
and α2 -α5 as in equations (2.108)–(2.111), to arrive at
⎧ ⎛
∂kx ⎪ ⎛ 4k 2 2αb2 ⎞⎞ ∂n1
= kx⎨ ⎜⎜(α2 + α3)k 02n1 + αa⎜ 5x − ⎟⎟⎟
∂χ ⎪
⎩⎝ ⎝ n1 n1 ⎠⎠ ∂χ
⎛ γ ⎛ γ ⎞⎞ ∂n
− ⎜⎜α2k 02n2 + 2 23 ⎜αb2 + αa 32 ⎟⎟⎟ 2
⎝ n2 ⎝ n 3 ⎠⎠ ∂χ
(5.9)
⎛ γ ⎛ γ ⎞⎞ ∂n ∂d ⎫
⎪
− ⎜⎜α3k 02n3 + 2 33 ⎜αb2 + αa 22 ⎟⎟⎟ 3 − αb4n12kxsec 2[kxd ] ⎬
⎝ n3 ⎝ n 2 ⎠⎠ ∂χ ⎪
∂χ ⎭
⎧ ⎛k 2
⎪ γ2γ3 ⎞ ⎫−1 ⎪

· ⎨αa⎜ 4 + 2 2 ⎟ + k x (α2 + α3) + n1 αb d sec [kxd ]⎬ ,

x 2 2 4 2
⎩ ⎝ n1 n 2 n3 ⎠ ⎭
⎪ ⎪

and
⎛ ∂n ∂n ∂n ⎞ ∂b
kyk02⎜(α 4 + α5)n1 1 − α 4n 4 4 − α5n5 5 ⎟ − ky(ky2 − γ4γ5) 2sec 2[kyb]
∂ky ⎝ ∂χ ∂χ ∂χ ⎠ ∂χ (5.10)
= .
∂χ (γ4 + γ5)(ky2 + γ4γ5) + ky2(α 4 + α5 ) + b(ky2 − γ4γ5) 2sec 2[kyb]

5.1.4.1 Influence of the temperature on SOI waveguides

In the case of a temperature change, ∂nj /∂χ is given by the thermo-optic effect. The
change in cross-section of the waveguide, ∂d /∂χ and ∂b/∂χ follow from the linear
thermal expansion coefficient. The thermo-optic coefficients of silicon and silicon

5-5
 Silicon Photonics

dioxide, as well as the linear thermal expansion coefficients, are given, e.g. in [22]
and [26], respectively.

5.1.4.2 Evanescent field sensor

In evanescent field sensors, liquids with different refractive indices flow over the SOI
waveguides. In this case, χ is the refractive index of the liquid. Hence for TM-like
modes: n3 = n 4 = n5 = χ and ∂n3 /∂χ = ∂n 4 /∂χ = ∂n5 /∂χ = 1, while the other wave-
guide properties are constant. For TE-like modes: n2 = n3 = n5 = χ and
∂n2 /∂χ = ∂n3 /∂χ = ∂n5 /∂χ = 1 (see figure 2.5).

5.1.5 Biosensors
Biosensors are refractive index sensors that measure specific molecules in blood or
other liquids. A high concentration of a certain molecule can indicate a certain
disease. Very accurate measurements of the molecular concentration are required to
enable reliable medical diagnosis. Apart from realising biosensors that are highly
accurate, there is also interest in making cheap hand-held devices to replace
expensive laboratory equipment. Affordable hand-held devices are especially
interesting for point-of-care diagnosis where a medical practitioner is directly able
to perform a diagnosis rather than waiting for laboratory results.
Refractive index sensors, also known as evanescent field sensors, probe the refractive
index in the vicinity of a waveguide through the evanescent part of the waveguide mode
(i.e, the part outside the core of the waveguide). The evanescent field typically extends
tenths of nanometers outside the core of the waveguide. A microfluidic channel is
fabricated on top of the SOI chip in order to let liquid (e.g. blood) flow over the ring
resonator surface but not over the other part of the photonic circuitry.
In bulk refractive index sensors, the refractive index of the liquid surrounding the
waveguide is measured (n2 , n3, n5 in figure 2.5). This type of sensor is not sensitive to
specific molecules and is more useful for micro-chemical applications where the
composition of the liquid is known. A change of the effective refractive index of this
type of sensors can be computed with the approximate theory for rectangular
waveguides (described in section 5.1.4).
For biomedical sensors the surface of the waveguide is functionalized with
receptors that selectively bind to specific molecules. These receptors trap the
molecules in a thin layer around the waveguide, exactly the region where the
evanescent field is largest and most sensitive.
We describe the waveguide geometry by its refractive index profile n(x , y ) or
permittivity profile ε(x , y ) = ε0n 2(x , y ). The change in the waveguide effective index
Δn e caused by a change in the permittivity profile Δε(x , y ) can be computed using
the variational theorem of waveguides [29, 30]. For TE-like modes, this change Δn e
may be approximated by
+∞
Δn e = c ∬−∞ Δε(x , y )Ex(x , y ) · Ex(x , y )*dxdy , (5.11)

with Ex(x , y ) the dominant electric field component of the TE-like mode and c the
propagation speed of light. It is seen that the sensitivity is high when the change of

5-6
 Silicon Photonics

permittivity Δε(x , y ) is large in regions where the electric field Ex is large. The
sensitivity of the sensor depends on the exact design, for example a thin waveguide
results in a larger electric field at its interfaces which gives a higher sensitivity than a
thicker waveguide where the light is more strongly confined in and near the core.
In figure 5.1 an evanescent wave sensor is shown that is used to sense the protein
avidin. The waveguide is covered with a layer of biotin which acts as a receptor for
the avidin. The working principle of the device is sketched in plots (a–d). The
binding of the advin molecules to the receptors on the waveguide induces a change in
the modal effective refractive index of the waveguide. This causes a shift in the
resonance frequency of the ring which is measured. In figure 5.1(f) measured
wavelength shifts as function of concentration are shown.
For a well-written introduction to silicon photonic biosensors including the
surface chemistry and biochemistry, we refer to the thesis of De Vos [15]. The
review of Kindt and Bailey [14] highlights some exciting biomedical applications. A
recent development in biosensors is enhanced Raman specroscopy using silicon-
nitride waveguides, see e.g. [31] for an introduction and [32] for a state-of-the-art
result.

5.1.6 Ultrasound sensors

The most familiar use of ultrasound is for the observation of unborn children;
however, ultrasonography is actually widely applied in both the medical field and in
industry. Today’s clear ultrasonic images are made by digitally focusing using an
array of transducers that record the sound at a number of positions spaced less than
a wavelength apart. Typical sound frequencies are 1–40 MHz corresponding to
wavelengths of 0.04–1.5 mm in water. Conventional ultrasound transducers employ
piezo-electric material to convert sound pressure into an electronic signal. An array
requires individual fabrication, placement and wiring of these transducers. Wiring
many coaxial cables is problematic in tight spaces such as medical intravascular
ultrasonography (IVUS), where the aim is to diagnose atherosclerosis from an
ultrasonograph of the artery wall that is obtained by bringing a catheter inside the
blood vessel.
It has been demonstrated that silicon photonic ring resonators can act as
highly sensitive ultrasound sensors [18, 19]. Initial results at 1 MHz show a
pressure detection limit of 0.4 Pa, which matches the performance of the state of
the art piezo-electric transducers while having a 65 times smaller footprint. By
combining these sensors with small (<1 mm2) silicon photonic multiplexers on
the same chip, an array of ultrasound sensors can be interrogated via a single
optical fibre. Furthemore, all sensors elements of the array can be simulta-
neously fabricated and wired using wafer-scale silicon CMOS and micro-
fabrication technology.
This optical micromachined ultrasound sensor (OMUS) is sketched in figure
5.2(a). A racetrack-shaped silicon ring resonator is on top of a 2 μm thick silicon-
dioxide membrane and covered with a 0.5 μm protective silicon-dioxide layer. The
device is fabricated by first fabricating the photonic circuitry using CMOS

5-7
 Silicon Photonics

Figure 5.1. Example of a biosensor for detecting avidin protein concentrations (reproduced from [15, 27, 28]).
The surface of the ring resonator is functionalized by linking biotin receptor molecules to the waveguide
surface. Different concentrations of avidin in phosphate buffer solution (PBS) were flowed across the sensor
surface. (a) Sketch of the ring resonator sensor with functionalized surface, before sensing. (b) After sensing,
the avidin proteins are bonded to the biotin receptor molecules. (c, d) Sketch the resonance wavelength shift
due to the binding of the avidin which causes a change of the local refractive index. (e) Scanning-electron-
microscope (SEM) picture of the ring resonator. (f) Measured wavelength shifts for different concentrations of
avidin. Each data point shows the difference in resonance wavelength of the ring resonator immersed in PBS,
before and after being in contact with the avidin solution. Redundant avidin molecules are rinsed thoroughly
with PBS, so no bulk refractive index changes are involved. In addition, the response of the sensor to another
protein, bovine serum albumin (BSA), is measured to determine the selective response to avidin. The response
to BSA concentrations is clearly lower, but this particular type of receptor molecule layer is not selective
enough for practical applications. Improved surface functionalization has been developed [14, 15, 17]. (a)–(c)
from [27], copyright 2012 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim. (d, e) Courtesy of Katrien de
Vos. (f) Reprinted with permission from [28].

5-8
 Silicon Photonics

Figure 5.2. Optical micromachined ultrasound sensor (OMUS) based on a silicon photonic micro-ring
resonator integrated in an acoustic resonant membrane. The acoustical resonance frequency of the membrane
is 0.76 MHz with a bandwidth of 19%. (a) Sketch of the sensor. (b) Microscope photograph of the sensor. The
sensor is illuminated from above and also from below such that the thin silicon-dioxide membrane is visible.
Tiny dark wires are the waveguides. (c) Operation principle of the interrogation. The deformation-induced
wavelength shift is recorded as intensity modulation of a single-wavelength laser source at the flank of the
optical resonance. (d) Measured transmission spectrum of the ring resonator. (e) Analysis of the sensor for
different laser wavelengths. Measured signal voltage swing of the OMUS with acoustic excitation (blue line),
the RMS value of the noise in the output intensity (red line), and the transmittance (dashed black line;
normalized for visibility). (f) Time responses of the OMUS (red line, time >160 μs) for transmitted acoustical
pulses with two acoustical centre frequencies (black line, time <40 μs). Figure reproduced from [19]. Reprinted
by permission from Macmillan Publishers Ltd, copyright 2015. Additional details may be found in [18, 19].

5-9
 Silicon Photonics

technology and then using micro-machining technology to etch the membrane from
the backside of the chip (figure 5.2(b) shows a microscope photograph). Incident
ultrasound waves deform the membrane and thus the resonator, thereby shifting the
optical resonance wavelengths. This shift is interrogated by inserting laser light with
a wavelength exactly at the flank of the resonance of the unperturbed ring. The shift
of the resonance wavelength due to the deformation of the membrane and ring is
detected by a modulation of the amplitude of the transmitted light (figure 5.2(c)).
An acoustical pressure P results in a non-uniform deformation of the racetrack
resonator. Four physical effects play a role in the resulting wavelength shift [33, 34].
First, the circumference of the track l increases. Second, the cross-section of the
waveguide shrinks due to the Poisson effect. Third, the refractive indices of the silicon
and SiO2 change due to the photo-elastic effect. The latter two effects together
influence the effective index n e of the waveguide. Fourth, the shift in resonance is
affected by the dispersion in the waveguide, as shown in equation (5.3). For a non-
uniform deformation of the ring resonator we write the resonance condition as

mλm = ∮ ne(ρ, λm , P )(1 + Sρ(ρ, P ))dρ, (5.12)

where the coordinate ρ runs along the track (see section 3.4.3) and Sρ is the local
strain in the direction of the track due to which an element of length dρ at given ρ
stretches to (1 + Sρ )dρ. Analogous to the derivation in section 5.1.1 we differentiate
equation (5.12) with respect to the pressure P. Assuming a uniform waveguide prior
to deformation with effective index n e and group index ng, we obtain [34]
⎛ ⎞
⎜ ⎟
∂dλm ne ⎜ λc ∂ne ∂Sρ ⎟
= · + λc , (5.13)
∂P ng ⎜ ne ∂P ∂P ⎟
⎜        ⎟
⏟
dispersion ⎝ eff.index track-length ⎠

where 〈·〉 denotes averaging over the track length and where λc is the resonance
wavelength without deformation. We denote the length of the track prior to
deformation by l and the length of the deformed track by l + Δl (P ). Considering
the last term of equation (5.12) we note that 〈Sρ(P )〉 = ∮ Sρ(P )dρ /l = Δl (P )/l , which
is the relative elongation of the track. Hence the influence of the track-length change
can be written as
∂Sρ λc ∂(Δl )
λc = . (5.14)
∂P l ∂P
For typical single-mode silicon-on-insulator waveguides, ne /ng ≈ 1/2. The influence
of the strain-induced change in the effective index turns out to be about a third of the
effect due to elongation and the two effects oppose each other. This factor was
measured for waveguides that were stretched in the long direction of the guide [34].
This results in a sensitivity of the ultrasound sensor that is about one third of the
right-hand-side of equation (5.14).

5-10
 Silicon Photonics

A description of the mechanical response of the acoustical resonant membrane,

∂(Δl )/∂P , is beyond the scope of this book. For both static and dynamic bending, the
strain (relative elongation) in the middle plane of the membrane is minimal and
maximal strain is obtained by placing the resonator on the top or bottom of the
membrane. The membrane is designed to be resonant at the acoustical frequency of
interest such that the deformation of the membrane at this frequency is enhanced
compared to off-resonant excitation. A smaller diameter or stiffer (thicker) mem-
brane results in a higher resonance frequency but lower sensitivity [18].
Ultrasonic measurements are performed in a pulse-echo sequence. The time of
this sequence is shorter than the relevant time-scale for temperature fluctuations.
Therefore, these measurements are not hampered by the strong thermo-optic effect
of silicon provided slow temperature-induced drift is compensated by tuning the
laser wavelength to the flank of the optical resonance dip prior to measuring.
Experimental results are shown in figure 5.2 for a sensor with a resonance
frequency of 0.76 MHz. The sensor was immersed in water and excited by incident
ultrasound waves. A laser is fibre-coupled to the input of the bus waveguide and a
photo-diode to the output of the bus waveguide. The transmission T (λ ) from the
input to the output of the OMUS was characterized by stepping through successive
laser wavelengths and recording the optical intensity at the output (figure 5.2(d)). Let
T0(λ ) be the transmission of the chip for a static situation without deformation.
When the optical resonator is deformed, the transmittance of the ring resonator is
shifted. We write the deformed transmittance T (λ , t ) in first order as
T (λ , t ) ≈ T0(λ + Δλ(t )), where Δλ(t ) is the time-dependent ultrasound-induced
optical resonance wavelength shift due to the dynamic deformation. The sensor is
interrogated by measuring the output intensity at one particular laser wavelength λl
over time. The sensitivity of the sensor is highest when the laser wavelength is chosen
at the flank of the resonance dip. This may be seen in figure 5.2(e) where the
ultrasound-induced swing of the optical intensity, max {T (λl , t )} t , is plotted versus
the laser wavelength λl (blue line). The sensitivity indeed is highest at the point where
the flank of the resonance dip is steepest (dashed black line). Two time-traces with an
acoustical pulse frequency below the resonant frequency of the membrane
( f0 = 0.42 MHz , top figure) and an acoustical pulse at the resonant frequency of
the membrane ( f0 = 0.77 MHz) are shown in figure 5.2(e). The acoustical resonance
of the membrane is visible in the shape of the received pulse.

5.2 Silicon photonic modulators based on carrier-depletion in a MZI

Optical fibre communication has enabled the broadband internet as we know it by a
revolutionary increase of the bandwidth of long-haul communication between cities
or countries. It is now also revolutionizing the bandwidth of interconnects on
smaller distances, such as fibre to the home (FTTH) or metropolitan-area networks
as used by companies and institutions. Developments of optical interconnects are
ongoing, boosting bandwidth at all distances including communication within
datacenters and even chip-to-chip communication. Modulators, i.e., devices that
encode binary data in laser light, are the workhorse of these interconnects which

5-11
 Silicon Photonics

have high technological and commercial demands. These include a low price of
fabrication, small device footprint, low power consumption per bit, high bandwidth
and large modulation depth. Moreover, these devices need to work at specific
telecommuncation wavelengths and their performance should not be affected by
temperature fluctuations. The interconnects for long-haul fibre communication, such
as intercity or even across the Atlantic, use the full capacity of their fibres and
require very high modulation bandwidths [10]. This is often achieved with lithium
niobate modulators [35] and complex modulation formats by which bits are encoded
in both phase and amplitude of the light. Fibre to the home (FTTH) interconnects
require cheap transducers and often use direct modulation of the current supplied to
the laser, thereby modulating the light amplitude albeit with much lower bandwidth.
Silicon photonic modulators offer a unique combination of low fabrication costs and
performance enhancements resulting from electronic–photonic integration which
are both provided by the CMOS fabrication technology. Silicon modulators are now
attractive for intermediate and short distances for which integration and large
bandwidth are important [10].
Within silicon technology, modulators can use different modulation formats, use
different physical effects to convert the electrical to an optical signal, and use
different integrated optical components to enhance this conversion. In this section,
we describe a popular implementation of silicon photonic modulators that achieves
so-called on-off keying using carrier depletion in a Mach–Zehnder interferometer.
This section is inspired by the review of Reed et al [21] which is an excellent
introduction to silicon photonic modulators and includes more details on historical
developments, junction types, and a detailed evaluation of the performance metrics.
Details and advances of the field may furthermore be found in the handbook edited
by Vivien and Pavesi [6, chapter 9] or [36]. Section 5.2.4 describes the modulator
reported by Thomson [37]. Winzer and Essiambre review [38] advanced optical
modulation formats in the broader context of optically routed wavelength-division-
multiplexed networks. Sotiropoulos [39] explains some formats in detail and
discusses the relevant example of an advanced demodulator/receiver in silicon
photonic technology.

5.2.1 Modulation formats: on-off keying

The most basic and frequently used modulation format is to encode the amplitude of
the light using so-called on-off keying (OOK), where the presence of light for a
specific duration represents a binary one, while its absence corresponds to a binary
zero. In reality, there is some light also at the binary zero and the extinction ratio ER
= Ion /Ioff is defined as the ratio of the optical intensities of the on signal, Ion and off
signal, Ioff .

5.2.2 Electro-optical conversion: pn-junction carrier depletion

Encoding of binary data in the optical mode of the light inside a waveguide is
achieved by altering the refractive index n of the waveguide material with an external
electric field. This electric field represents the binary data with frequencies of ∼50 GHz,

5-12
 Silicon Photonics

which are low compared to the frequency of the light (∼193 THz). The electro-optic
effect can be used to alter the real part of the refractive index Δn , referred to as
electro-refractive. The electro-optic affect can also alter the imaginary part of the
refractive index, referred to as electro-absorption and described by the intensity
absorption coefficient1 Δα . Physical effects that are traditionally used in semi-
conductor materials are the Pockels effect, the Kerr effect and the Franz–Keldysh
effect [21]. However, these effects are weak in pure silicon at the telecommunication
wavelength of 1.55 μm [1, 40] and therefore alternative methods are used. The
thermo-optic coefficient of silicon is large [41] but this effect is too slow for modern
telecom applications. Recent research efforts also explore mechanisms using other
materials that are compatible with silicon technology, such as germanium, to form
quantum wells for higher modulation efficiencies [6, 21]. Today’s most common
method of silicon optical modulators is based on the plasma dispersion effect, in
which the concentration of free charges in silicon changes the real and imaginary
parts of the refractive index.
The semiconductor physics of so-called pn-junctions that is necessary to under-
stand how free charges (electrons and holes) behave in this junction is explained in
appendix C.
Soref and Bennet [42] evaluate the changes in refractive index Δn from
experimentally observed absorption curves. They quantify the change in the real
refractive index Δn as well as the change in the absorption Δα resulting from a
change of the concentration of free charges. Let ΔNe be the change in the
concentration of free electrons and ΔNh the change in the concentration of free
holes. Then the following relations hold for silicon at a free-space wavelength of
1.55 μm [42]:
Δn = −8.8 · 10−22ΔNe − 8.5 · 10−18(ΔNh )0.8 (5.15)

Δα = 8.5 · 10−18ΔNe + 6.0 · 10−18ΔNh, (5.16)

where carrier concentrations are per cubic centimetre. Using these expressions, it
may be calculated that a change in carrier density on the order of 5 · 1017 cm−3
results in a Δn of −1.66 · 10−3 but is accomplished by a detrimental absorption
Δα = 7.3 due to the absorption of free carriers [21].
One of the mechanisms to electrically manipulate the free carrier concentration in
silicon photonic modulators is carrier depletion (see figure 5.3(a)). A pn-junction is
formed across the silicon waveguide using doping of the silicon to form p-type and
n-type regions. The lightly p-doped and n-doped regions have an increased
concentration of free holes (Nh ∼ NA) and and free electrons (Ne ∼ ND ), respectively,
while the depletion area around the interface of these two regions has few free
carriers (Nh, Ne small). This area becomes wider with increasing reverse bias voltage

1
The complex refractive index n̂ describes real effective index n and the intensity absorption coefficient α via
nˆ = n + α /(2k ). Considering a plane wave, the transversal component of the electric field
E = E0 exp(ıωt − ınkz
ˆ ) = E0 exp(ıωt − ıknz )exp(−αz /2). The intensity I scales with the square of the electric
field such that I = I0 exp(−αz ).

5-13
 Silicon Photonics

Figure 5.3. Sketch of an implementation of a silicon photonic modulator based on carrier depletion in the
waveguide. (a) pn-junction waveguide modulator. Reprinted from [21] by permission from Macmillan
Publishers Ltd Copyright 2010. (b and c) Specific design of Thomson et al [37]. (b) Sketch of rib waveguide
with profile of the fundamental mode with an effective index of 2.495. Plot of Ex where dark blue indicates a
large amplitude and white indicates zero amplitude. (c) Cross-section of the pn-junction/waveguide. At the
position where the optical mode is, the p-type (indicated purple) and n-type (indicated brown) materials have a
low doping concentration to avoid strong absorption (NA ∼ 3 · 1017 cm3 and ND ∼ 1.5 · 1018 cm3, respec-
tively). Further away from the optical waveguide, the p-type and n-type materials have a high doping
concentration to provide good conductivity (ND ∼ NA ∼ 1020 cm3).

Vd (see appendix C). The carrier depletion based optical modulator works as
follows. The electronic signal containing the desired modulation is connected to the
metal connections alongside the waveguide such that the voltage Vd (t ) across the pn-
junction represents this signal. The width of the carrier depletion area follows the
applied signal. In the area between the maximal and minimal width of this region,
the carrier density varies strongly from the bulk p- or n-type carrier densities to the
low carrier density of the depletion area. This dynamic change in free-hole density
ΔNh and free-electron density ΔNe causes a dynamic change in the silicon refractive
index Δn and absorption Δα via the plasma dispersion effect, equation (5.15).
Meanwhile a lightwave travels through the waveguide with the modal field strongest
in the centre of the waveguide. The phase that is accumulated by the light is strongly
dependent on the effective index Δn e of the waveguide thus on the silicon refractive
index which is being modulated. Additionally, the amplitude of the light is being
modulated due to the absorption. Although this absorption modulation could
directly be used as the modulation mechanism, it is more efficient to use the phase
modulation as the prime modulation mechanism by embedding the waveguide in a
structure such as a Mach–Zehnder interferometer.
The modulation speed or bandwidth is the key figure of merit for optical
modulators. Modulation bandwidth is usually defined by the frequency at which
the modulation is reduced to 50% of its maximal value. The speed of a modulator is
commonly characterized by its ability to carry data at a certain rate. The electronic
behaviour of the pn-junction and driving electronics is the limiting factor for the

5-14
 Silicon Photonics

bandwidth. A pn-junction has an intrinsic capacitance which poses a limit on the

maximum achievable modulation speed.

5.2.3 Phase to amplitude: Mach–Zehnder inferometer

Efficient on-off keying (OOK) amplitude modulation can be achieved by modulat-
ing the waveguide effective index and converting the thereby induced phase
modulation of the lightwave into an amplitude modulation with a phase-sensitive
component. As discussed in the beginning of this section, there are a number of
technological metrics that are important for the design of modulators. Mach–
Zehnder interferometers often provide a good trade-off [21]. Their relatively
broadband operation implies relatively large fabrication tolerances. Furthermore,
temperature effects tend to largely cancel out in Mach–Zehnder interferometers
because of the differential output.
Mach–Zehnder interferometers are discussed in section 3.5. A phase difference
between the light in the two arms causes a difference of the amplitude on its outputs.
We consider the optical power in the lower output, Iout , for the case where the MZI
has two perfect 50/50 couplers (∣τ∣2 = ∣κ∣2 = 1/2 in figure 3.21). We assume that the
modulation-induced difference in loss between the waveguides can be neglected with
respect to the total loss of the waveguides, which is particularly high in the pn-
junction, and we use the approximation: α1 ≈ α2 , with α1 and α2 the transmittance
amplitude of the upper and lower arms of the MMI, respectively. Then from
equation (3.87),
⎛1 1 ⎞
Iout = α12⎜ + cos[ϕ1 − ϕ2 ]⎟ Iin, (5.17)
⎝2 2 ⎠

with Iin the inserted optical power, α1 the amplitude transmission loss in the MZI
arms, and ϕ1 − ϕ2 the difference between accumulated phases in the upper and lower
arms. We consider the case that the effective index of the waveguide in the upper of
the arm MZI is modulated while the lower arm is constant. Without modulation
both waveguides have effective index ñ e . With modulation, the upper MZI arm has
effective index n e = n˜ e + Δn e(t ), with Δn e(t ) the time-dependent modulation. Then
ϕ1 − ϕ2 = k Δn e(t ) l , (5.18)

α1 = e−αpl , (5.19)
with l the MZI arm length, k = 2π /λ the free-space wavenumber, and αp the
amplitude propagation loss per unit length. This waveguide propagation loss
includes scattering due to waveguide surface roughness and material absorption,
but also the additional loss caused by the pn-junction. Some care has to be taken
regarding the different definitions of loss or transmittance. We have total amplitude
loss over a waveguide segment α1 and amplitude propagation loss per unit length αp
as used in equation (5.19). In equation (5.16) the material loss is expressed as
intensity propagation loss per unit length α.

5-15
 Silicon Photonics

The modulation depth, also known as the extinction ratio, is defined as the ratio
of Imax , the intensity transmitted when the modulator is adjusted to maximum
transmission, to Imin , the intensity transmitted when the modulator is adjusted for
minimum transmission. It is quoted in decibels and expressed as 10 log(Imax /Imin ).
The modulation depth increases with longer MZI arm lengths l at the cost of
increasing device loss and higher power consumption to apply the driving signal Vd.

5.2.4 A 40 GBit/s optical modulator

In this section we describe the silicon photonic modulator designed and fabricated
by Thomson and co-workers [37]. This modulator is of the previously described
type: the light is modulated using the plasma dispersion effect induced in an
electrically driven pn-junction in reverse bias which simultaneously functions as a
waveguide. This waveguide is embedded in a Mach–Zehnder interferometer. A key
innovation in this work was the development of a self-aligned fabrication process for
the doping of the pn-junction and the etching of the waveguide.
The modulator is based on a so-called rib waveguide (see figure 5.3(b)) that
simultaneously functions as pn-junction (figure 5.3(c)). Such a device needs to be
carefully designed as the requirements for low-loss waveguiding (preferably no
doping at the position of the optical field) as well as the requirements for a good pn-
junction (with necessary doping), and also a good overlap between the pn-junction
depletion layer and the optical field need to be achieved. In this particular device,
low optical absorption was achieved by using lightly doped p-type and n-type
materials. The rib region, which carries the majority of the optical power, is p-type
because free-holes give a much larger change in refractive index Δn than free-
electrons while having similar absorption Δα . This follows from equation (3.92) for
the used doping values of ∼10−17 cm−3. The target concentration NA of the p-type
material is chosen five times lower than the concentration ND of the n-type material
which causes the depletion area to be wider in the p-type material thus in the rib
region of the waveguide. This follows from equation (C.10) where the width wp of
1
the depletion layer scales with ∼(ND /NA) 2 . Modelling of the device showed that at a
reverse bias of 0 V the depletion area width is approximately 60 nm extending
almost entirely into the rib region. At 6 V the depletion area width is approximately
200 nm extending 170 nm into the p-type rib region and 30 nm into the n-type slab
region.
A potential difference cannot be supplied to the two sides of the modulator by
simply bonding two wires because the high inductance of metal wires prohibits high-
speed (GHz) transmission. Therefore, a microwave coplanar waveguide is used
which acts similarly to a coaxial cable. The electronic signal is transmitted through a
metal strip, acting in a similar way to the core of a coaxial cable, that is separated by
a well-defined gap from a metal ground plane, acting similarly to the shield of a
coaxial cable2. The signal strip and one side of the ground plane are connected to the

2
A recommended textbook is Microwave Engineering by Pozar [43] and details of coplanar waveguide design
can be found in the book by Garg, Bahl and Bozzi et al [44], chapter 7.

5-16
 Silicon Photonics

pn-junction metal connection (figure 5.3(c)). The coplanar waveguide is designed

such that the GHz microwaves co-propagate with the light with a similar velocity.
The entire structure has a 50 Ohm impedance to match the driving electronics (figure
5.4(a)) that is connected to one end of the microwave waveguide while the other end
is terminated with a 50 Ohm load to dump remaining power.
The Mach–Zehnder interferometer is built from two low-loss 1×2 MMI couplers.
In this book the results of a device with an MZI arm length of 1 mm is described.
Reference [37] also reports on a longer device with higher extinction ratio. One of the
arms of the asymmetric MZI is 180 μm longer for a more accurate analysis of the
optical response to a DC reverse bias.
The measured transmission spectra for different DC reverse bias voltages show
that the optical spectrum is shifted to smaller wavelength when the DC reverse bias
voltage is increased (figure 5.4(b)). From this shift the phase shift of the MZI arm
can be extracted by relating the wavelength shift to the free-spectral-range of the
MZI as follows from equation (3.92) (figure 5.4(c)). The performance of the
modulator is evaluated using a so-called eye diagram. An electronic signal contain-
ing a pseudorandom bit sequence (PRBS) with on-off-keying (OOK) is connected to
the modulator’s microwave line. A laser is connected to the input of the MZI. The
laser wavelength chosen at a quadrature point, i.e., such that at zero bias the phase
difference between the ams of the MZI is ±π /2 where the power Iout of the MZI
output is sensitive to the phase difference ϕ1 − ϕ2 , see equation (5.17). The output of
the MZI is measured using a fast photo-diode which is connected to an oscilloscope.
The oscilloscope is triggered with the OOK signal trigger such that it displays the
time window of approximately two bits on the horizontal scale. The resulting graph
is a superposition of all received zeros (low voltage) and ones (high voltage) which,
most interestingly, also contain the transitions from zero to one and vice versa. An
open eye means that all zeros and ones are well separated in amplitude (voltage,
vertical scale) and time (horizontal scale) thus that all bits sent to the modulator can
be received by the receiving party. Figure 5.4(d) displays the open eye pattern of this
modulator in a laboratory. Long distance communication, possibly including optical
amplifiers, add additional loss, noise, and timing errors.

5.3 Fabrication technology and design of optical circuits

Fabrication of the small waveguides in silicon-on-insulator technology is not
straightforward, only last decade’s high-end photolithography technology is suffi-
ciently accurate. The rules that apply to macro-fabrication do generally not apply to
micro-fabrication. To illustrate this, we compare the fabrication of a photonic chip
in a fab against the fabrication of a wooden table by a craftsman. Asking a fab to
make a device from a slightly different material might require a decade of develop-
ment, whereas a wood craftsman could easily work with a slightly different type of
wood. In contrast, it would be impossible for a craftsman to fabricate a million
tables, while a fab easily fabricates a millions devices.
We briefly explain the photolithography fabrication concept. First, the optical
designer makes a construction plan based on the technology which is offered by a

5-17
 Silicon Photonics

fab. The fab starts with a clean silicon-on-insulator wafer. A photo-sensitive layer is
spincoated on the wafer. This layer is illuminated with the pattern that has to be
written in this layer. (One may compare this with traditional photography, where an
image is created by light that is incident on a photographic film.) Then the
photosensitive layer is developed, transforming this layer into a mask which protects
only the illuminated patterns. The actual fabrication is done using etching, a
chemical process in which a plasma ‘eats’ the unprotected silicon. Then the residue
of the mask is removed and the wafer is cleaned. Most chips require more than one
of these fabrication sequences, for example for the fabrication of out-of-plane
grating couplers, which basically is a grating etched in the top surface of a wide
waveguide [45].
In reality, micro-fabrication when considered on the nano-scale is a rather rough
process. Note that the width of a waveguide (500 nm) is about one hundredth of the
diameter of a human hair. We will illustrate the difficulties in the processing with
some examples. In the photolithography, the imaging has to be accurate within a few
nanometers over a surface of about 1 cm2 (the footprint of chip). Multiple layers
have to be aligned to each other. This means that after fabrication of the first layer,
the wafer has to go back into the lithography machine where the image of the second
layer has to be aligned to the pattern on the wafer with an accuracy of a few tens of
nanometers. Furthermore, etching causes rough surfaces and possible non-uniform-
ity over the wafer. The top of the patterns is protected by the mask, but the
formation of the sides is determined by the etch process. The unprotected areas of
the chip consume the etchant at a higher rate than the protected patterns, giving a
variation in the concentration of the etchant over the wafer. Etching small deep
features is especially difficult as ‘fresh’ etchant hardly reaches the bottom of the
feature.
The previously mentioned challenges cause the fabricated devices to differ from
the designs. We mention a number of notable differences. (1) The world-leading
manufacturer of SOI-wafers, Soitec (Bernin, France), specifies the variation of the
height of the silicon light-guiding layer as 20 nm, which amounts to 10% of the
typical height of 220 nm [46]. (2) Deviation of the actual light dose from the ideal
one in the photolithographic process as well as uncertainties in the etching process
may cause the size of the manufactured devises to differ from the designs. (3) The
lithographic process can be optimized for only one feature size [47]. For example,
when the process is optimized such that 450 nm wide waveguides are fabricated
according to design then waveguides with other widths are generally fabricated
smaller or larger than designed. (4) The sides of the pattern (waveguide) that ideally
would be perpendicular, can in practice have an inclination angle of about 10 degrees
[48]. (5) Furthemore, the side walls are often not smooth but have have nanometer-
scale roughness [49].
For an optical designer, it is necessary to design the devices such that they remain
functional regardless of the inaccuracies caused by the fabrication. There are two
approaches to take account of these fabrication-variations. For known fabrication-
induced deviations, it is often possible to design the pattern such that is it not
identical to the desired pattern, but after fabrication it becomes equal to the desired

5-18
 Silicon Photonics

Figure 5.4. Silicon photonic modulator presented by Thomson et al [37]. (a) Microscope image of a
fabricated MZI modulator with 250 μm long phase modulators. Only one phase modulator is used in the
measurements. The diagram is annotated to show the position of the waveguides. Positions to electrically
connect the signal (S) and two grounds (G) of each of the two coplanar microwave waveguides are shown.
(b) Spectral response of the MZI with different reverse bias voltages. (c) Phase shift achieved for different
reverse bias voltages. (d) Eye diagram derived from optical pseudorandom binary sequence (PRBS) data
output at 40 Gbit/s. MZI with 6.5V RF reverse bias signal operated at quadrature. Measured extinction ratio
is 3.5 dB. Reprinted with permission from [37].

pattern. For example, when one knows the relation between the designed and
fabricated width of a waveguide, it is easy to design the width such that after
fabrication the desired width is obtained. When the deviations due to the fabrication
are unpredictable, the design should be such that it is tolerant against the deviations,
i.e., that the device performance remains acceptable in spite of the deviations.

5-19
 Silicon Photonics

5.3.1 Example: design of racetrack for ultrasound sensor

In this section we give an example how the functionality of a design can be made
sufficiently robust to fabrication error. The idea behind this basic example also
applies to the design of more complex circuits [50, 51].
We consider the design of the racetrack-shaped ring resonator that was used in
the ultrasound sensor discussed in section 5.1.6. The waveguide is a typical silcon-
on-insulator single-mode waveguide with width 450 nm and height 220 nm. The
bend radius of the ring is 5 μm because smaller bend radii give too large a loss [18,
27, 48]. The length of the straight-track of the resonator is 40 μm for optimal
sensitivity per applied pressure (deformation), see [18], section 5.4 for details. The
transmission through the bus waveguide shows dips at the resonance wavelengths of
the resonator. The ultrasound-induced shift in a resonance wavelength was
interrogated with a laser with wavelength tuned to the flank of the resonance, so
that this shift causes a change in transmitted power (figure 5.2(c)). The sensitivity of
the interrogation scales with the steepness of the flank of the resonance. We
approximate the sensitivity by (1 − r )/ΔλFWHM where r is the extinction ratio at
resonance and ΔλFWHM is the full-width at half-maximum of the transmission dip,
see equations (3.73) and (3.74).
Here, we consider the choice of directional couplers: one or two couplers (figure
3.18(a) or 3.18(c)) and the choice of coupling power ∣κ∣2 . Directional couplers are
sensitive components in the sense that fabrication-induced deviations in the width
and height of a waveguide strongly affect the coupling constant κ. Suppose that the
fabrication process causes variations in ∣κ∣2 of ±0.03. The waveguides in the
racetrack have a typical loss of 2.5 dB/cm and 0.05 dB/180 deg turn with a radius
of 5 μm [18, 27], resulting in a round-trip power transmittance α 2 = 0.98.
In figure 5.5 we show the extinction ratio r given by equation (3.73) and the
FWHM of the transmission dip given by equation (3.74) as a function of the straight
through transmission ∣τ∣2 = 1 − ∣κ∣2 for typical round-trip loss of α 2 = 0.98 (upper
row) and for low round-trip loss α 2 = 0.999 (lower row). Furthermore, the
sensitivity (1 − r )/ΔλFWHM is shown as a function of ∣τ∣2 for the two round-trip
losses. The red curves correspond to one directional coupler, whereas the blue
dashed curves are for two couplers. The highest sensitivity is achieved using
resonators with one directional coupler. With typical loss, the maximum sensitivity
is achieved when choosing ∣κ∣2 = 0.01 (∣τ∣2 = 0.99). However, fabrication-induced
variations in coupling power may also result in a device with near-zero coupling to
the resonator and near-zero sensitivity. A more robust design choice is ∣κ∣2 = 0.04
(∣τ∣2 = 0.96), which probably yields lower sensitivity but a much larger chance of a
working device.
Note that the ultrasound sensor has two directional couplers. The rationale
behind that design was based on waveguides with lower loss. As seen in the lower
row of figure 5.5, a sensor based on a low-loss resonator is very sensitive if the
coupling is as designed, but is very insensitive if the coupling deviates from the
optimum value. This is because the extinction ratio r of resonators with low loss,
α 2 ≈ 1, depends strongly on the difference α − κ between the ring round-trip

5-20
 Silicon Photonics

Figure 5.5. Design of the directional couplers for racetrack resonators for ultrasound sensing. (a, d) Extinction
ratio r using equation (3.73), (b, e) full-width at half-maximum ΔλFWHM using equation (3.74), and (c, f)
sensitivity approximated by (1 − r )/ΔλFWHM in relative power per picometer, all plotted as a function of
coupler straight-through power ∣τ∣2 , with coupled power ∣κ∣2 = 1 − ∣τ∣2 . (a–c) Typical round-trip transmittance
α 2 = 0.98, power loss (1 − α 2 ) = 0.02 . (d–f) Low-loss round-trip transmittance α 2 = 0.999, power loss
(1 − α 2 ) = 0.001. Track length l = 111 μm, effective group index ng = 4.28, and free-space wavelength
λc = 1550 nm .

transmittance α and the coupling κ. This dependance can be reduced by adding a

second coupler because this increases the loss. Two couplers close to each other will
be fabricated with similar coupling constant κ, so that α → κα and hence it follows
from equations (3.73) that the difference, ακ − κ , has a less significant effect on the
extinction ration r. For low-loss waveguides, the design with a second coupler is
more fabrication tolerant (dashed line in figure 5.5, lower row).

References
[1] Reed G T and Knights A P 2004 Silicon Photonics: An Introduction (Chichester: Wiley)
[2] Reed G T (ed) 2008 Silicon Photonics: The state of the art (Chichester: Wiley)
[3] Pavesi L and Lockwood D J (ed) 2004 Silicon Photonics III (Berlin: Springer) doi: 10.1007/
b11504
[4] Pavesi L and Lockwood D J (ed) 2011 Silicon Photonics III: Components and Integration
(Berlin: Springer) doi: 10.1007/978-3-642-10506-7
[5] Pavesi L and Lockwood D J (ed) 2016 Silicon Photonics III: Systems and Applications
(Berlin: Springer)

5-21
 Silicon Photonics

[6] Vivien L and Pavesi L (eds) 2013 Handbook of Silicon Photonics (Boca Raton, FL: Taylor
& Francis)
[7] Vivien L, Cheben P, Lo Guo-Qiang P, Pavesi L and Zhou Z Eds. 2014 IEEE J. Sel. Top.
Quantum Electron. (Issue on Silicon Photonics) 20
[8] GuoQiang P L, Chen Y, Poon A W O, Nakamura T and Chu T Eds. 2016 IEEE J. Sel.
Top. Quantum Electron. (Issue on Silicon Photonics) 22
[9] Christy P 2016 Intel’s silicon photonics products could change the world of IT www.intel.
com/content/www/us/en/architecture-and-technology/silicon-photonics/451-research-silicon-
photonics-paper.html
[10] Liao L 2017 Intel silicon photonics: From research to product (presented at meeting of the
IEEE CPMT Chapter San Francisco Bay Area, Santa Clara CA), USA http://ewh.ieee.org/
soc/cpmt/presentations/cpmt1703a.pdf
[11] Painchaud Y 2014 Silicon-based products and solutions Proc. SPIE 8988 89880L
[12] Luxtera 2016 Luxtera ships one millionth silicon photonic transceiver product (presented at
the 42nd European Conference on Opical Communication (ECOC 2016, Dusseldorf,
Germany)) www.luxtera.com/press-releases
[13] Luxtera 2017 Luxtera ships Industry’s first 2x100G PSM4 silicon photonics embedded
optical modules, (presented at the Optical Fiber Communication Conference (OFC 2017,
Los Angeles, USA)) www.luxtera.com/press-releases
[14] Kindt J T and Bailey R C 2013 Biomolecular analysis with microring resonators:
applications in multiplexed diagnostics and interaction screening Curr. Opin. Chem. Biol.
17 818–26
[15] de Vos K 2010 Label-free silicon photonics biosensor platform with microring resonators
PhD Dissertation Universiteit Gent, Belgium
[16] Ryckeboer E, Bockstaele R, Vanslembrouck M and Baets R 2014 Glucose sensing by
waveguide-based absorption spectroscopy on a silicon chip Biomed. Opt. Exp. 5 1636
[17] Genalyte 2017 Silicon photonics biosensor technology Genalyte (San Diego, CA: Genalyte)
www.genalyte.com/about-us/our-technology/
[18] Westerveld W J 2014 Silicon photonic micro-ring resonators to sense strain and ultrasound
PhD Dissertation Technische Universiteit Delft http://repository.tudelft.nl
[19] Leinders S M, Westerveld W J, Pozo J, van Neer P L M J, Snyder B, OBrien P, Urbach H P, de
Jong N and Verweij M D 2015 A sensitive optical micro-machined ultrasound sensor (omus)
based on a silicon photonic ring resonator on an acoustical membrane Sci. Rep. 5 14328
[20] International Telecommunication Union 2012 Spectral grids for WDM applications:
DWDM frequency grid, International Telecommunication Union Std. ITU-T G.694 www.
itu.int/rec/T-REC-G.694.1
[21] Reed G T, Mashanovich G, Gardes F Y and Thomson D J 2010 Silicon optical modulators
Nat. Photon. 4 518–26
[22] Photon Design 2011 Material database and material models, distributed with FimmWave
software package (Oxford: Photon Design)
[23] Westerveld W J, Leinders S M, van Dongen K W A, Urbach H P and Yousefi M 2012
Extension of marcatilias analytical approach for rectangular silicon optical waveguides J.
Lightwave Technol. 30 2388–401
[24] Rosenthal A, Kellnberger S, Bozhko D, Chekkoury A, Omar M, Razansky D and
Ntziachristos V 2014 Sensitive interferometric detection of ultrasound for minimally invasive
clinical imaging applications Laser Photon. Rev. 8 450–7

5-22
 Silicon Photonics

[25] van Gulik R J J, de Boer B M and Harmsma P J 2017 Refractive index sensing using a three-
port interferometer and comparison with ring resonators IEEE J. Sel. Top. Quantum
Electron. 23 433–9
[26] Okada Y and Tokumaru Y 1984 Precise determination of lattice parameter and thermal
expansion coefficient of silicon between 300 and 1500 k J. Appl. Phys. 56 314–20
[27] Bogaerts W, de Heyn P, van Vaerenbergh T, de Vos K, Kumar Selvaraja S, Claes T, Dumon
P, Bienstman P, van Thourhout D and Baets R 2012 Silicon microring resonators Laser
Photon. Rev. 6 47–73
[28] De Vos K, Bartolozzi I, Schacht E, Bienstman P and Baets R 2007 Silicon-on-insulator
microring resonator for sensitive and label-free biosensing Opt. Express 15 7610–5
[29] Kogelnik H 1975 Theory of dielectric waveguides Integrated Optics (Topics in Applied
Physics vol 7) (Berlin: Springer) pp 13–81
[30] Densmore A, Xu D X, Waldron P, Janz S, Cheben P, Lapointe J, Delge A, Lamontagne B,
Schmid J H and Post E 2006 A silicon-on-insulator photonic wire based evanescent field
sensor IEEE Photon. Technol. Lett. 18 2520–2
[31] Baets R, Subramanian A Z, Dhakal A, Selvaraja S K, Komorowska K, Peyskens F,
Ryckeboer E, Yebo N, Roelkens G and le Thomas N 2013 Spectroscopy-on-chip applica-
tions of silicon photonics Proc. SPIE 8627 86270I
[32] Dhakal A, Wuytens P C, Peyskens F, Jans K, Thomas N L and Baets R 2016 Nanophotonic
waveguide enhanced raman spectroscopy of biological submonolayers ACS Photon. 3 2141–9
[33] Westerveld W J, Pozo J, Harmsma P J, Schmits R, Tabak E, van den Dool T C, Leinders
S M, van Dongen K W, Urbach H P and Yousefi M 2012 Characterization of a photonic
strain sensor in silicon-on-insulator technology Opt. Lett. 37 479–81
[34] Westerveld W, Leinders S, Muilwijk P, Pozo J, van den Dool T, Verweij M, Yousefi M and
Urbach H 2014 Characterization of integrated optical strain sensors based on silicon
waveguides IEEE J. Sel. Top. Quantum Electron. 20 101–10
[35] Toney J E 2015 Lithium Niobate Photonics (Norwood, MA: Artech House)
[36] Reed G T, Mashanovich G Z, Gardes F Y, Nedeljkovic M, Hu Y, Thomson D J, Li K,
Wilson P R, Chen S-W and Hsu S S 2013 Recent breakthroughs in carrier depletion based
silicon optical modulators Nanophotonics 3 229–45
[37] Thomson D J, Gardes F Y, Hu Y, Mashanovich G, Fournier M, Grosse P, Fedeli J-M and
Reed G T 2011 High contrast 40gbit/s optical modulation in silicon Opt. Express 19 11507–16
[38] Winzer P J and Essiambre R J 2006 Advanced optical modulation formats Proc. IEEE 94
952–85
[39] Sotiropoulos N 2013 Advanced modulation formats for optical access networks PhD
Dissertation Eindhoven, the Netherlands: Technische Universiteit Eindhoven
[40] Soref R and Bennett B 1987 Electrooptical effects in silicon IEEE J.Quantum Electron. 23
123–9
[41] Cocorullo G and Rendina I 1992 Thermo-optical modulation at 1.5 μm in silicon etalon
Electron. Lett. 28 83–5
[42] Soref R and Lorenzo J 1985 Single-crystal silicon: a new material for 1.3 and 1.6 μm
integrated-optical components Electron. Lett. 21 953–4
[43] Pozar D M 2011 Microwave Engineering (Chichester: Wiley)
[44] Garg R, Bahl I and Bozzi M 2013 Microstrip Lines and Slotlines. (London: Artech House)

5-23
 Silicon Photonics

[45] Roelkens G, Vermeulen D, Selvaraja S, Halir R, Bogaerts W and van Thourhout D 2011
Grating-based optical fiber interfaces for silicon-on-insulator photonic integrated circuits
IEEE J. Sel. Top. Quantum Electron. 17 571–80
[46] Imec SiPhotonics process handbook—release 0.1, distributed with ePIXfab IMEC08 Design
Kit IMEC-Ghent University 3 2011
[47] Dumon P, Bogaerts W, Tchelnokov A, Fedeli J-M and Baets R 2008 Silicon nanophotonics
Future Fab Int. 25 29–36
[48] Selvaraja S K, Bogaerts W and Thourhout D V 2011 Loss reduction in silicon nanophotonic
waveguide micro-bends through etch profile improvement Opt. Commun. 284 2141–4
[49] Morichetti F, Canciamilla A, Martinelli M, Samarelli A, de la Rue R M, Sorel M and
Melloni A 2010 Coherent backscattering in optical microring resonators Appl. Phys. Lett. 96
081112
[50] Melloni A, Roncelli D, Morichetti F, Canciamilla A and Bakker A 2009 Statistical design
in integrated optics CLEO/Europe and EQEC 2009 Conf. Digest (Washington DC: Optical
Society of America) p JSI1_4.
[51] Bogaerts W, Fiers M and Dumon P 2014 Design challenges in silicon photonics IEEE J. Sel.
Top. Quantum Electron. 20 1–8

5-24
 IOP Publishing

Silicon Photonics
Electromagnetic theory
Wouter J Westerveld and H Paul Urbach

Appendix A
Coupled mode theory of parallel waveguides

In this appendix the coupled mode theory as derived by Hardy and Streifer [1, 2] is
summarized with some small changes. Their formalism requires fewer approxima-
tions than for example the formalism derived by Yariv [3, 4]. Most importantly, in
Yariv’s approximation the modes as pure transverse electric (TE) or transverse
magnetic (TM) which is not accurate for single-mode waveguides with a high index
contrast. A more detailed comparison between the formalisms can be found at the
end of this appendix A. Although the coupled mode formalism is generally
applicable, we apply it only to the directional couplers.
We consider a directional coupler consisting of two parallel rectangular wave-
guides with refractive index n1, embedded in a homogeneous cladding with index n2
and separated by a gap g (see figure 3.1(a). For typical SOI waveguides, this gap
is about 200 nm. The coupler is described by the permittivity profile
ϵc(x , y ) = ϵ0nc2(x , y ). We describe the electromagnetic fields in the coupler in terms
of the modes of the two waveguides, labelled a and b. Isolated waveguide a has
permittivity profile ϵa(x , y ), and waveguide b is described by permittivity profile
ϵb(x , y ). We define Δϵa(x , y ) and Δϵb(x , y ) as the difference between the permittivity
profile of the coupler and permittivity profile of the isolated waveguide, i.e.,
ϵc(x , y ) = ϵa(x , y ) + Δϵa(x , y ) = ϵb(x , y ) + Δϵb(x , y ), (A.1)

see figure A.1. In the analysis, we assume that the mode expansion conjecture as
explained in section 2.3.6 is valid. After explaining this conjecture, we rewrite
Maxwell’s equations in a form that is very useful for the description of the
directional coupler. We then expand the field in terms of modes of the single
waveguides a and b and considering only those that are important in the directional
coupler, neglecting radiation and backwards-traveling modes. This results in
intuitive equations for the directional coupler.

doi:10.1088/978-0-7503-1386-5ch6 A-1 ª IOP Publishing Ltd 2017

 Silicon Photonics

Figure A.1. Permittivity profiles, cross-section at y = 0, plotted as a function of x. Permittivity profiles of the
directional coupler ϵc(x, y ), single waveguide a ϵa(x, y ), single waveguide b ϵb(x, y ) and difference profiles
Δϵa(x, y ) and Δϵb(x, y ). In typical SOI waveguides, ϵ1 and ϵ2 are the permittivities of silicon and silicon-
dioxide, respectively.

A.1 Mode expansion conjecture

The mode expansion conjecture states that, for a given frequency ω, the transverse
components of the fields of the modes of a waveguide, all guided, radiation and
evanescent modes form a complete set in every cross-sectional plane (z = constant).
This means that the transverse components of any field can in any cross-sectional
plane be written as a linear superposition of the transverse components of the fields
of the waveguide modes (see section 2.3.6).
In the example of a directional coupler, we write the tranverse components of the
time-harmonic field that satisfy Maxwell’s equations in the configuration of the
directional coupler (permittivity ϵc(x , y )) as a superposition of the transverse
components of all the of modes of waveguide a (permittivity ϵa (x, y)), i.e.,
∞
E tc(x , y , z , t ) = ∑aν(z )Eta,ν(x, y )e ıωt , (A.2)
ν= 1

∞
Htc (x , y , z , t ) = ∑aν(z )Hta,ν(x, y )e ıωt. (A.3)
ν= 1

where E tc and Htc are the transverse components in the coupler and with the
summation over an infinite set of waveguide modal profiles Eta,ν(x , y ) and
Hta,ν(x , y ). Subscript t denotes the transverse field components x and y. Actually,
this superposition consists of a finite sum of the guided modes and an integral over
the radiation and evanescent modes, but for simplicity we will keep writing the single
infinite sum (see section 2.3.6). The modes of waveguide a are solutions of Maxwell’s

A-2
 Silicon Photonics

equations for permittivity ϵa(x , y ), but they are not a solution of Maxwell’s
equations for the directional coupler which has permittivity ϵc(x , y ). For example,
the individual modes of waveguide a are smooth outside the core of waveguide a,
whereas certain transverse components of the electromagnetic fields in the coupler
are discontinuous at the interfaces of waveguide b.
The modes of waveguide a are orthogornal in the sense of equation (2.53):
+∞
∬−∞ ( Eta,μ × Hta,ν* + Eta,ν* × Hta,μ) · zˆ dxdy = 0 if μ ≠ ν. (A.4)

The modes of waveguide b are orthogornal in the same sense. We will assume
henceforth that the guided modes of waveguides a and b are normalized as follows
+∞
∬−∞ ( Eta,ν × Hta,ν* + Eta,ν* × Hta,ν) · zˆ dxdy = 1, (A.5)

and a similar relation for the modes of waveguide b. This normalization is consistant
with [1, 2], but differs by a factor of four from normalization to unit power flow.
The mode expansion, equations (A.2) and (A.3), concerns only the transverse field
components (x , y). The z-components are not governed by these equations and
follow from Maxwell’s equations for the appropriate permittivity profile. The modal
fields E a,ν(x , y ) and H a,ν(x , y ) are solutions of Maxwell’s equations for waveguide a
with permittivity profile ϵa(x , y ). The corresponding z-component of the electric field
follows from the transverse components of the magnetic field. Using Ampère’s law
(2.2) with permittivity profile ϵa(x , y ), we find
1 ⎛ ∂Hxa,ν ∂H ya,ν ⎞
E za,ν(x , y ) = ⎜ − ⎟. (A.6)
ıωϵa(x , y ) ⎝ ∂y ∂x ⎠

Similarly the z-component of the electric field in the coupler for given transverse
magnetic field Htc follows from Ampère’s law (2.2) with permittivity ϵc(x , y ),
1 ⎛ ∂Hcx ∂Hcy ⎞
E cz(x , y , z , t ) = ⎜ − ⎟. (A.7)
ıωϵc(x , y ) ⎝ ∂y ∂x ⎠

We wish to express E cz(x , y , z, t ) in terms of the modes of waveguide a and therefore

substitute Htc from equation (A.3) and move all terms under a single summation to
arrive at
∞
1 ⎛ ∂Hxa,ν ∂H ya,ν ⎞ ıωt
E cz(x , y , z , t ) = ∑ a ν (z ) ⎜ − ⎟e , (A.8)
ν= 1
ıωϵc(x , y ) ⎝ ∂y ∂x ⎠

or, using the z-component of the electric field of the modes of waveguide a as
expressed by equation (A.6),
∞
ϵ (x , y )
E cz(x , y , z , t ) = ∑aν(z ) ϵa(x, y ) E za,ν(x, y )e ıωt. (A.9)
ν= 1 c

A-3
 Silicon Photonics

Combining equation (A.2) and (A.9) gives a description of the electric field in the
coupler, E c , in terms of the modes of waveguide a,
∞ ⎛ ϵ (x , y ) ⎞
E c(x , y , z , t ) = ∑aν(z )⎜Eta,ν(x, y ) + zˆ ϵa(x, y ) E za,ν(x, y )⎟e ıωt. (A.10)
ν= 1
⎝ c ⎠

As preparation for the derivation in the sections hereafter we also express the
transverse field components of the fundamental mode of waveguide b, {Etb,1, Htb,1},
as a superposition of the transverse fields components of the modes of waveguide a,
{Eta,ν, Hta,ν}:
∞
Etb,1 = ∑pν1 Eta,ν, (A.11)
ν= 1

∞
Htb,1 = ∑pν1 Hta,ν. (A.12)
ν= 1

The values of pνμ are computed based on the orthogonality relation (A.4) for two
modes of the same waveguide. Two modes corresponding to different waveguides
are generally not orthogonal in this sense. Now pνμ is the projection of mode μ of
waveguide b onto mode ν of waveguide a based on the orthogonality relation
(A.4), i.e.,
+∞
pμν = ∬−∞ ( Eta,μ × Htb,ν* + Etb,ν* × Hta,μ) · zˆ dxdy. (A.13)

In this Appendix one of the two modes μ or ν will always be the fundamental guided
mode of waveguide a or b, i.e., we will in fact only use p1ν and pμ1.
The last step of this section is to express the z-component of the fundamental
mode of waveguide b in terms of the modes of waveguide a. The z-component of the
electric field of the modes of waveguide b are found using an expression similar to
equation (A.6) but with a replaced by b. Analogous to the derivation of equation
(A.9), we substitute equation (A.12) in equation (A.6) but for waveguide b, move the
right-hand-side under a single summation, and then insert equation (A.6) for
waveguide a to arrive at
∞
ϵ (x , y )
E zb,1 = ∑pν1 ϵa(x, y ) E za,ν. (A.14)
ν= 1 b

Combining equation (A.11) and equation (A.14) while dividing the z-component by
ϵc(x , y ) gives
ϵb(x , y ) b,1
∞ ⎛ ϵ (x , y ) ⎞
Etb,1 + zˆ Ez = ∑pν1 ⎜Eta,ν + zˆ ϵa(x, y ) E za,ν⎟. (A.15)
ϵc(x , y ) ν= 1
⎝ c ⎠

A-4
 Silicon Photonics

A.2 Differential equations for the modal amplitudes

We have written the electromagnetic field in the coupler as a superposition of the
modes of waveguide a, with z-dependent modal amplitudes aν(z ). We now derive
differential equations that describe the z-dependence of the amplitudes of guided
modes. These equations will form the basis for the coupled mode theory.
We start with two electromagnetic fields: E c , Hc and E a,μ, Ha,μ. Both fields are
monochromatic with frequency ω hence time evolution is given by e ıωt . The first
field, labelled c, is the field in the directional coupler. The latter is guided mode μ of
waveguide a, which is a solution of Maxwell’s equations for the waveguide with
permittivity ϵa(x , y ). Its electric field is written as
E a,μ(x , y , z , t ) = E a,μ(x , y )e ı(ωt−βa,μz ). (A.16)
Fields c and a obey Ampères law (2.2) for ϵc(x , y ) and ϵa(x , y ), respectively,
∇ × Hc = ıωϵc(x , y )E c , (A.17)

∇ × Ha,μ = ıω(ϵc(x , y ) − Δϵa(x , y ))E a,μ. (A.18)

Similar to section 2.3.3, we calculate using Faraday’s (2.1) and Ampère’s (A.17)
laws:
∇ · (E c × Ha,μ*) = ıω(ϵc(x , y )E c · E a,μ*−μ0Hc · Ha,μ*−Δϵa(x , y )E c · E a,μ*),
∇ · (E a,μ*× Hc ) = ıω(μ0Hc · Ha,μ*−ϵc(x , y )E c · E a,μ*).

Adding gives:
⎛∂ ∂ ∂ ⎞
⎜ xˆ + yˆ + zˆ⎟ · (E c × Ha,μ*+E a,μ*× Hc ) = −ıωΔϵa(x , y )E c · E a,μ*.
⎝ ∂x ∂y ∂z ⎠
Following the derivation of section 2.3.3, we integrate this equation over an
(x , y)-plane, and apply the two-dimensional divergence theorem to obtain
∂
∂z
zˆ · ∬ (E tc × Hta,μ* + E ta,μ* × Htc) dxdy = −ıω ∬ Δϵa(x, y)E c · E a,μ* dxdy, (A.19)
where we used that the integral along a closed contour vanishes for increasing
distance to the origin x = y = 0. The mode expansion conjecture states that the
transverse electromagnetic fields E tc and Htc may be written as a superposition of the
fields of the modes of waveguide a. With this expansion equation (A.19) can be
expressed in the modes of waveguide a. By substituting equations (A.10) and (A.3),
together with equation (A.16) and the corresponding equation for H a,μ in equation
(A.19), one finds that time dependent terms cancel. The summation over ν and z-
dependent factors can be brought outside of the integrals:
∞ ⎛ ∂a ν (z ) ⎞
∑ν=1 ∬ zˆ · ( Eta ,ν × Hta ,μ* + Eta ,μ* × Hta ,ν ) dx dy · ⎜ + ıβa ,μa ν (z )⎟ e ıβa,μz
⎝ ∂z ⎠
(A.20)
∞ ⎛ a ,ν ϵa (x, y ) a ,ν ⎞
= − ıω ∑ a ν (z )e ıβa,μz
ν=1 ∬ Δϵa (x, y ) ⎜Et + zˆ
⎝ ϵc (x, y )
E z ⎟ · E a ,μ* dx dy.
⎠

A-5
 Silicon Photonics

The integral in the left-hand-side of this equation is the orthorormality relation (A.4)
between modes ν and μ of waveguide a, with result δμν since the guided modes are
assumed normalized by equation (A.5). Thus the summation on the left-hand-side of
this equation reduces to only one term with ν = μ. After cancelling the factor
exp[ıβa,μ ] and solving for ∂aμ /∂z we obtain for guided mode μ
∞
∂aμ a
= −ıβa,μaμ − ı∑aνκ˜ νμ , (A.21)
∂z ν= 1
where
⎛ ⎞
a
κ˜ νμ ≡ω ∬ Δϵa(x, y )⎜⎝Eta,ν + zˆ ϵϵac((xx,, yy)) Eza,ν⎟⎠ · E a,μ* dxdy. (A.22)

The expressions above describe the electromagnetic field in the coupler in terms of
the modal fields of waveguide a, with z-dependent modal amplitudes. Equation
(A.21) describes how the amplitudes of guided modes change with z. At this point,
this may seem a complicated way to write Maxwell’s equations, but it will turn out
to be very useful when only a limited number of modes are taken into account, such
as in the case of the directional coupler.

A.3 Exact differential equations for the field in the directional

coupler
Without restricting generality, we may write the transverse electromagnetic fields in
the directional coupler {E tc, Htc}, as a linear combination of the fundamental guided
mode of waveguide a, {Eta,1, Hta,1}, the fundamental guided mode of waveguide b,
{Etb,1, Htb,1} and a residual field {Etr, Htr}, i.e.,
E tc(x , y , z , t ) = ua(z )Eta,1(x , y )e ıωt + ub(z )Etb,1(x , y )e ıωt + Etr(x , y , z )e ıωt , (A.23)
with an analogous equation for Htc . The residual field {Etr(x , y, z ), Htr(x , y, z )} can
for every z always be chosen orthogonal in the sense of relation (A.4) to both of the
fundamental modes of waveguides a and b.
In equation (A.23) we express E tc and Etb,1 in terms of the modes of waveguide a
using equations (A.2) and (A.11). Then we use that the modes are orthogonal in the
sense of equation (A.4), and that the residual field is orthogonal to the fundamental
mode of waveguide a. In the resulting equation we consider only the coefficients of
Eta,1 to arrive at
a1(z ) = ua(z ) + p11 ub(z ), (A.24)
note that p11 is independent of z. Substituting equation (A.24) into equation (A.21)
gives
∞
∂ua ∂u
+ p11 b = −ı(βa1 + κ˜11a )ua − ı(p11 βa1 + p11 κ˜11a )ub − ı ∑ aνκ˜ νa1. (A.25)
∂z ∂z ν= 2

A-6
 Silicon Photonics

We introduce κ̂ab by:

∞
κˆ ab ≡ ∑pν1 κ˜ νa1, (A.26)
ν= 1

where κ̃ νa1 are given by equation (A.22). The first term of the summation in equation
(A.26), i.e., p11 κ̃11a , is also present in the second term of the right of equation (A.25).
By adding and subtracting the other terms (ν = 2,...,∞) of this summation to the
second and third term of the right of equation (A.25), respectively, one finds the
important differential equation:
∞
∂ua ∂u
+ p11 b = −ı(βa1 + κ˜11a )ua − ı(p11 βa1 + κˆ ab)ub − ı ∑ κ˜ νa1(aν − pν1 ub). (A.27)
∂z ∂z ν= 2

Coefficient κ̂ab is related to the coupling of the fundamental mode of waveguide a to

the fundamental mode of waveguide b. We will now simplify κ̂ab and express it in
terms of the modal electric fields E a,1 and E b,1 of the fundamental modes of the
waveguides a and b. Substitution of equation (A.22) in equation (A.26), and
interchanging the order of the integral and the summation yields
⎡∞ ⎛ ϵ (x , y ) a,ν⎞⎤
κˆ ab = ω ∬ Δϵa(x , y )⎢∑pν1 ⎜Eta,ν + zˆ a E z ⎟⎥ · E a,1* dxdy , (A.28)
⎢⎣ ν= 1 ⎝ ϵc(x , y ) ⎠⎥⎦

or, using equation (A.15), we obtain the expression of κ̃ab aimed for
⎛ ⎞
κˆ ab = ω ∬ Δϵa(x, y )⎜⎝Exb,1Exa,1* + E yb,1E ya,1* + ϵϵcb((xx,, yy)) Ezb,1Eza,1*⎟⎠dxdy. (A.29)

We thus have derived the differential equation (A.27) for the z-dependence of the
amplitudes ua and ub of the modal fields of the fundamental modes of waveguides a
and b. We started by expressing equation (A.23) in terms of the modes of waveguide
a and then only considered the fundamental mode to arrive at equation (A.24).
Instead one can also express equation (A.23) in terms of the modes of waveguide b
and then consider the fundamental mode of that waveguide. This gives equation
(A.27) with equation (A.29) but with labels a and b interchanged and with and pμν
replaced by pμν
* . We write the thus obtained system of differential equations as

∂u
C = −ı(BC + Kˆ )u(z ) − ıW (z ), (A.30)
∂z
with
u
u = ua ,
b ( ) (A.31)

⎛ 1 p11⎞
C = ⎜⎜ ⎟,
⎟ (A.32)
⎝ p11* 1 ⎠

A-7
 Silicon Photonics

⎛ βa,1 0 ⎞
B = ⎜⎜ ⎟⎟ , (A.33)
⎝ 0 βb,1⎠

⎛ κˆ κˆ ⎞
Kˆ = ⎜ aa ab ⎟ , (A.34)
⎝ κˆba κˆbb ⎠

⎛ ∞ κ a (a − p u ) ⎞
⎜ ∑ν=2 ˜ ν1 ν ν1 b ⎟
W (z ) = ⎜ ∞ , (A.35)
⎜∑ κ˜ ν1(bν − p * ua )⎟⎟
b
⎝ ν=2 ν1 ⎠

with κˆaa ≡ κ˜11a , κˆbb ≡ κ˜11b , κ̂ab given by equation (A.29), and κ̃ νa1 given by equation
(A.22). Equation (A.30) can be rewritten as
∂u
= −ı(B + K )u(z ) − ıC −1W (z ), (A.36)
∂z
with
κ κ
(
K = κ aa κ ab ,
ba bb )
1
κ aa =
1 − p11 p11*
(
κˆ aa − p11 κˆba + p11 p11* (βa,1 − βb,1) , )
1
κbb =
1 − p11 p11*
( )
κˆbb − p11* κˆ ab + p11 p11* (βb,1 − βa,1) , (A.37)
1
κ ab = (κˆ ab + p11 (βa,1 − βb,1 − κˆbb)),
1 − p11 p11*
1
κba =
1 − p11 p11*
( )
κˆba + p11* (βb,1 − βa,1 − κˆ aa ) .

Note that these equations are exact (provided the mode expansion conjecture is
correct).

A.4 Approximations and solutions of the differential equations

We approximate the analysis of the directional coupler by neglecting W in equation
(A.36). This term is related to the other modes of the system, that are not the
fundamental mode of waveguide a or b. More specifically, we assume that for
higher-order modes (ν ⩾ 2), κ̃ νa1aν and κ̃ νa1pν1 are much smaller then κ̂aa , κ̂ab , and βa,1
(and a similar requirement with a and b interchanged). This assumption is justified
by the results of rigorous FDTD simulations, discussed in section 3.1.3, which show
that directional couplers in SOI technology have little loss, i.e., almost all energy is
carried by the modes of waveguides a and b.
The fundamental modes of waveguides a and b are almost orthogonal, which
allows one to neglect all terms involving p11 or p11* . When neglecting these terms, C

A-8
 Silicon Photonics

reduces to the identity matrix and Kˆ = K . The results obtained with this additional
approximation did not differ appreciably from keeping these terms in the model. We
nevertheless kept the terms with p11 or p11* in the equations.
By neglecting W , i.e., neglecting the coupling to higher-order modes (radiation,
evanescent and backwards-traveling modes), equation (A.36) becomes
∂u
= −ı(B + K )u(z ). (A.38)
∂z
We consider a coupler for which in z = 0 all energy is in waveguide b, i.e. u = (0, 1).
Solving equation (A.37) for this initial condition gives1
ıκ
ua(z ) = − ab sin(sz ) e−ı(βb,1+κbb−δ )z , (A.39)
s

⎛ ıδ ⎞
ub(z ) = ⎜cos sz − sin sz⎟ e−ı(βb,1+κbb−δ )z , (A.40)
⎝ s ⎠

1
δ= (β + κbb − βa,1 − κ aa ), (A.41)
2 b,1

s= κ abκba + δ 2 , (A.42)

These equations describe the behaviour of the parallel waveguides in a directional

coupler, when all the field is in the fundamental mode of waveguide b at z = 0.
Equations (A.39) and (A.40) are identical to equations (3.2) and (3.3) which were
explained but not derived at the beginning of section 3.1.

A.5 Comparison with the literature

We have followed the approach of Hardy & Streifer [1] from 1985, and used the
unified notation introduced by Hardy in 1998 [2]. The deviation given above differs
in a number of aspects from that in [1]. Our analysis is based on the orthogonality
relation (2.53) rather than relation (2.58). Forward- and backward-traveling modes
are orthogonal in this relation, so that we may use a single summation over all
forward and backward traveling modes. In the directional coupler, excitation of the
backwards traveling modes can be neglected, and these modes are lumped together

⎛m κ ⎞
1
We write equation (A.38) as ∂u /∂z = −ıMu(z ), with M = ⎜ κ a mab ⎟, ma = βa,1 + κaa , and mb = βb,1 + κbb . We
⎝ ba b ⎠
look for solutions in the form u(z ) = a ±exp[−ıλ±z ]. Substituting this in the differential equation gives
λ±a ± = Ma ± , which we recognize as the eigenvalue equation for M with eigenvalues λ± . This equation has
solutions when det∣M − Iλ±∣ = 0 , from which we find the eigenvalues λ± = 12 (ma + mb ) ±
1 1 1
2
(ma + mb )2 − 4(mamb − κabκba ) = 2 (ma + mb ) ± 14 (mb − ma )2 + κabκba = 2 (ma + mb ) ± s , with s and δ
defined in equations (A.42) and (A.41). From the eigenvalue equation of M , we find eigenvectors
a + = (κab, δ + s ) and a− = (κab, δ − s ). If we solve the general solution of the differential equation,
u(z ) = c +a +exp[−ıλ+z ] + c−a−exp[−ıλ−z ], for initial condition u(0) = (0, 1), we find c− = −c + and
c + = 1/2s , giving equations (A.39) and (A.40).

A-9
 Silicon Photonics

with the radiative and evanescent modes in the residual field E r(x , y, z ). As in [1], we
have assumed a lossless material with real permittivity ϵc(x , y ). In the description of
the fields in terms of the modes of waveguide a, we followed Kogelnik [5] and started
our derivation with the orthogonality relation (2.53) including the complex
a
conjugates, which led to a slightly different notation for κ̃ νμ , equation (A.22), and
κ̂ab , equation (A.29), as compared to [1, 2]. In this appendix it was not nessecary to
explicitly normalize evanescent modes, for which it would be necessary to use the
unconjugated orthogonality relation. With real transversal fields, the longitudinal (z)
component of the field is imaginary, which causes a minus sign in the definition of κ̂ab
in [1,2]. The derivation given in this appendix applies to two parallel waveguides.
However, the derivation can be generalized to the more general case by changing
a,ν b,ν
ϵc(x , y ) to ϵc(x , y, z ), leading to z-dependence of E˜z , E˜z , κ̃ νμ
a b
, κ̃ νμ , κ̂ab , κ̂ba , κ̂aa , κ̂bb ,
K̂ , and K . In [2] it is stated that neglecting the radiation and the backwards
propagating waves may not be justified in the case of waveguides with high
refractive index contrast. Our comparison with rigorous FDTD simulations shows
that by far most energy is carried by the fundamental modes of the waveguides, so
that the residual fields can indeed be neglected for parallel waveguides. In contrast,
gratings in silicon introduce strong scattering, which will not only couple the
fundamental mode to another mode, but couple to a spectrum of modes.
In comparison with the work of Yariv [3] and [4], the most important difference
with our approach is that Yariv approximates the modes as pure TE or pure TM.
Typical SOI-waveguides have relevant longitudinal components of the electro-
magnetic fields, and neglecting them underestimates the coupling between wave-
guides. Furthermore, in [4] the fundamental modes of isolated waveguides a and b
are assumed to be orthogonal, which is strictly speaking incorrect but which is a
good approximation for typical SOI waveguides.

A.6 Implementation and comparison with other methods

The coupled mode theory derived in this appendix is implemented in Matlab [6] with
the modal fields computed with a numerical mode-solver (FMM method in
FimmWave) on a discretisation grid of 5 nm. In figure 3.2(c), we compare the
calculated coupling coefficient s with those obtained with other methods. It can be
seen that there is a good agreement between the methods, which indicates that the
use of the coupled mode theory is justified for typical directional couplers in high-
index-contrast silicon waveguides.

References
[1] Hardy A and Streifer W 1985 Coupled mode theory of parallel waveguides J. Lightwave
Technol. 3 1135–46
[2] Hardy A 1998 A unified approach to coupled-mode phenomena IEEE J. Quantum Electron.
34 1109–16
[3] Yariv A 1973 Coupled-mode theory for guided-wave optics IEEE J. Quantum Electron. 9
919–33

A-10
 Silicon Photonics

[4] Yariv A and Yeh P 2007 Photonics: Optical Electronics in Modern Communications (Oxford
Series in Electrical and Computer Engineering) (Oxford: Oxford University Press)
[5] Kogelnik H 1975 Theory of dielectric waveguides Integrated Optics (Topics in Applied
Physics vol 7) (Berlin: Springer) pp 13–81 doi 10.1007/978-3-662-43208-2
[6] MathWorks Inc. 2017 Matlab—the language of technical computing (Natick, MA:
MathWorks Inc.) www.mathworks.com/matlab

A-11
 IOP Publishing

Silicon Photonics
Electromagnetic theory
Wouter J Westerveld and H Paul Urbach

Appendix B
Lossless transmission matrix

In this appendix we show that power conservation allows us to write the lossless
transmission matrix U in the form
⎛ b1 ⎞ ⎛ e ıϕD τ * κ ⎞ a1
⎜ ⎟ = ⎜ ıϕ ⎟
⎝ b2 ⎠ ⎝− e D κ * τ ⎠ a2 ( )
, (B.1)

with ∣τ∣2 + ∣κ∣2 = 1 and ϕD a phase which affects the wave in input a1. First it is
shown that the transmission matrix is unitary, i.e., that its inverse equals the
transpose of its complex conjugate, U−1 = U *T , so that U *T U = I (with I the
identity matrix). The general transmission matrix equation for a system without
reflections is
⎛ b1 ⎞ ⎛ A κ ⎞ a1
⎜ ⎟ = ⎜⎝
⎝ b2 ⎠
⎟
B τ ⎠ a2
. ( ) (B.2)

We express b1 and b2 in terms of a1 and a2, and compute the power flow out of the
coupler:
b1b1* + b2b 2* = (AA* +BB*)a1a1* + (κκ *+ττ *)a2a 2*
(B.3)
+ (Aκ *+Bτ *)a1a 2* + (A*κ + B*τ )a1*a2.
This power flow out of the coupler, ∣b1∣2 + ∣b2∣2 , is equal to the power flow into the
coupler, ∣a1∣2 + ∣a2∣2 , for arbitrary a1 and a2. Hence
(AA* +BB*) = 1, (κκ *+ττ *) = 1, (Aκ *+Bτ *) = 0, (A*κ + B*τ ) = 0.
These relations are used to compute U *T U and it is found that U *T U = I . The
lossless transmission matrix U is thus unitary. Now U−1 = U *T implies
1 τ −κ ⎛ * *⎞
D (
− B A )
= ⎜ A B ⎟,
⎝ κ* τ* ⎠
(B.4)

doi:10.1088/978-0-7503-1386-5ch7 B-1 ª IOP Publishing Ltd 2017

 Silicon Photonics

with D = Aτ − κB . Hence ∣D∣ = 1, A = Dτ *, B = −Dκ *. Writing D = e ıϕD gives

equation (B.1). Note that ϕD effectively represents an overall phase added to a1 and
for some cases it is not necessary to take this phase into account.
In the remainder of this section we consider the case when the system is symmetric
in inputs and outputs 1 and 2. In this case, an incident wave in input a1 has the same
output at ports b1 and b2 as an incident wave in input a2 has in outputs b2 and b1,
respectively. This means that the transmission matrix can be written in the form
⎛ b1 ⎞
⎜ ⎟= κ τ
⎝ b2 ⎠
( )(aa ).
τ κ 1
2
(B.5)

Comparing equation (B.5) with equation (B.1), we see that for the symmetric
couplers D = τ /τ *=−κ /κ *, from which follows that D = 2 arg{τ} = −2 arg{κ}.
Hence
2(arg{τ} − arg{κ}) = π , (B.6)
and the phase difference between τ and κ is thus π /2.
As an example we consider a directional coupler with two identical waveguides,
equations (3.2) and (3.3) with δ = 0. Coupling coefficients κab and s are real. It may
be seen that the phase difference between τ and κ is indeed π /2. For the 2 × 2 MMI
coupler with 50/50 splitting ratio, equation (3.44), τ = 1 −2 ı and κ = 1 +2 ı which indeed
also differ in phase by π /2.

B-2
 IOP Publishing

Silicon Photonics
Electromagnetic theory
Wouter J Westerveld and H Paul Urbach

Appendix C
Semiconductor physics of pn-junctions

This section discusses the semiconductor physics that is necessary to understand the
pn-junction for the application in silicon photonic modulators employing the
plasma-dispersion effect with carrier depletion.
The electronic band structures of metals, semiconductors, and insulators are
discussed in figure C.1. After briefly describing the pn-junction and its relevant
properties, we follow Hook and Hall [1] to arrive at the relation between the size of
the carrier depletion area and the reverse bias voltage applied on the junction.
A p–n junction is a boundary or interface between two types of semiconductor
material, p-type and n-type, inside a single crystal of semiconductor. The ‘p’
(positive) side contains acceptor ions that cause an excess of free holes, while the
‘n’ (negative) side contains donor ions that cause an excess of free electrons (see
figure C.1). We consider the 1D pn-junction depicted in figure C.2. At the interface
between the p-type and n-type semiconductors, electrons from the n region near the
interface diffuse into the p region leaving behind positively charged ions in the n
region. The diffused electrons recombine with holes in the p region, forming
negatively charged ions in the p region. Likewise, holes from the p-type region
near the interface diffuse into the n-type region, leaving behind negatively charged
ions in the p region. These holes recombine with electrons in the n-region, creating
positive ions. The regions close to the p–n interface lose most of their free carriers
(electrons and holes) due to diffusion and recombination and are together called the
carrier depletion region. The remaining charged ions in this region (charge density
Q, negative on the p-side and positive on the n-side) cause an electric field E that
opposes the diffusion process for both electrons and holes. Without an external
applied voltage over the junction, an equilibrium condition is reached in with a built-
in potential difference ΔV = V0 that exists across the junction. The semiconductor
physics to calculate the band structure and the potential difference V0 is for the
example described in Hook and Hall [1]. We omit this and continue with the
electrostatic analysis of the pn-junction. The edges of the carrier depletion area are

doi:10.1088/978-0-7503-1386-5ch8 C-1 ª IOP Publishing Ltd 2017

 Silicon Photonics

Figure C.1. Filling of the electronic states in metals, semiconductors, and insulators. The vertical axis is energy
and the horizontal axis (width of the shown areas) is the density of available states for a certain energy. The
shading follows the Fermi–Dirac distribution (black = all states filled, white = no state filled). At finite
temperatures, the lower energy levels are occupied and the higher energy levels are empty, with a gradual
transition in the intermediate levels. These delocalized states (extending through the material) occur grouped
and form the electronic band structure of the material with band-gaps in between. As follows from the Pauli
exclusion principle, each state may contain maximally one electron. Electrical conductivity, movement of
electrons, only occurs in bands that are partially filled (sketched). The Fermi level, EF, is the hypothetical
energy level of an electron, such that at thermodynamic equilibrium this energy level has 50% probability of
being occupied. Metals have high conductivity as the Fermi level lies inside a band. Insulators have low
conductivity as the Fermi level lies inside a large bandgap and only the lower band is filled. Semiconductors
have intermediate conductivity as the Fermi level lies inside a finite bandgap such that some of the lower
(valence) band levels are empty and some of the higher (conduction) band levels are filled. Electrons that are
‘missing’ in the conduction band are called holes which behave (mathematically) very similarly to particles with
positive charge. Semiconductors may be doped with materials that donate an excess of free holes (p-type) to the
valence band or free electrons (n-type) to the conduction band. Reproduced from Nanite at Wikimedia
Commons under the Creative Commons CC0 1.0 universal Public Domain Dedication.

quite sharp and the charge density may be approximated by the step function (see
figure C.2)
⎧ − NAe −wp < x < 0
⎪
Q(x ) = ⎨+ NDe 0 < x < wn (C.1)
⎪
⎩ 0 elsewhere
with NA the concentration of acceptor ions (the concentration of free holes in the
p-type semiconductor), wp the width of the carrier depletion area in the p-type
material, ND the concentration of donor ions (the concentration of free electrons
in the p-type semiconductor), wn the width of the depletion area in the n-type
material, and e the charge of an electron. Overall charge neutrality requires
eNAwp = eNDwn. (C.2)
The electrostatic potential V is related to the charge density Q by Poisson’s equation
d2V Q (x )
=− (C.3)
dx 2 ϵ

C-2
 Silicon Photonics

Figure C.2. A 1D pn-junction in equilibrium. The upper plot is a sketch of the pn-junction with the carrier
depletion area and the carrier concentrations (on a logarithmic scale). Below are plots of the charge density
Q(x ), the electric field Ex(x ) and the potential V (x ). Adapted from Adundovi at Wikimedia Commons under
the Creative Commons Attribution 3.0 Unported license.

with ϵ the permittivity of the material. Integrating equation (C.3) for the charge density
in equation (C.1) gives the electric field (in the x-direction) E , also see figure C.2,
⎧ eNA
dV ⎪
⎪− (x + wp) −wp < x < 0
ϵ
E(x ) = − =⎨ (C.4)
dx ⎪+ eND
⎪
⎩
(x − wn) 0 < x < wn,
ϵ

C-3
 Silicon Photonics

where the integration constants were chosen such that the electric field vanishes in the
bulk semiconductor material outside the carrier depletion area. Integrating the electric
field E gives the electrostatic potential V (see figure C.2)
⎧ eNA
⎪
⎪ (wp + x )2   −wp < x < 0
2 ϵ
V (x ) = ⎨ (C.5)
⎪V − eND (w − x )2   0 < x < w ,
⎪
⎩ 0 2ϵ
n n

where integration constants have been chosen so that the potential of the p-region
outside the depletion layer is zero (this defines the zero potential). The built-in
potential difference across the junction is given by V0 which is dictated by the
thermodynamic equilibrium of the pn-junction. The potential V (x ) must be
continuous at the interface x = 0 demanding
e
V0 =
2ϵ
(NAwp2 + Ndwn2 ), (C.6)

which may be solved together with equation (C.2) for wp and wn providing the
widths of the depletion area on the two sides of the junction

⎛ 2ϵNDV0 ⎞1 2
wp = ⎜ ⎟ (C.7)
⎝ eNA(NA + ND) ⎠

⎛ 2ϵNAV0 ⎞1 2
wn = ⎜ ⎟ . (C.8)
⎝ eND(NA + ND) ⎠

Operation of the pn-junction requires application of an additional potential differ-

ence Vd across the p–n junction. If the positive potential is at the p region, the
junction is said to be forward biased and the driving voltage Vd is taken to be
positive; if the positive potential is at the n region then the junction is reverse biased
and Vd is negative. The depletion region has a high resistivity compared to the bulk
material because of the low carrier density in the depletion region. Therefore, the
driving voltage Vd exists across this region. This gives the total potential difference
across the depletion region
ΔV = V0 − Vd . (C.9)

Forward bias reduces the total potential difference whereas reverse bias increases it.
This junction acts as a diode. Forward bias reduces the potential barrier thereby
increasing the conductivity and the current flow across the junction. Reverse bias
increases the barrier so that the conductivity and the current flow remain low. A very
large reverse bias, however, causes Zehner breakdown after which the conductivity
is suddenly high and current may flow across the diode [1]. For our application to
the silicon photonic modulator the conclusion is important that the applied voltage
changes the electrostatic potential which affects the width of the depletion layer.

C-4
 Silicon Photonics

Following the analysis of equations (C.3)–(C.8), we obtain for the new potential
difference in equation (C.9):
⎛ 2ϵND(V0 − Vd ) ⎞1 2
wp = ⎜ ⎟ (C.10)
⎝ eNA(NA + ND) ⎠

⎛ 2ϵNA(V0 − Vd ) ⎞1 2
wn = ⎜ ⎟ . (C.11)
⎝ eND(NA + ND) ⎠

It follows that the width of the depletion layer is increased by reverse bias.

Reference
[1] Hook J R and Hall H E 1995 Solid State Physics (Chichester: Wiley)

C-5