CIRCUITS, DEVICES AND SYSTEMS SERIES 10

Distributed feedback
semiconductor lasers

John Carroll,
James Whiteaway
& Dick Plumb

The Institution of Electrical Engineers

Distributed feedback
semiconductor lasers

John Carroll,
James Whiteaway
& Dick Plumb

The Institution of Electrical Engineers

SPIE Optical Engineering Press
Other volumes in the IEE Circuits, Devices and Systems series:
Volume 1 GaAs technology and its impact on circuits and systems
D. G. Haigh and J. Everard (Editors)
Volume 2 Analogue IC design: the current-mode approach
C. Toumazou, F. J. Lidgey and D. G. Haigh (Editors)
Volume 3 Analogue-digital ASICs R. S. Soin, F. Maloberti and
J. Franca (Editors)
Volume 4 Algorithmic and knowledge-based CAD for VLSI
G. E. Taylor and G. Russell (Editors)
Volume 5 Switched-currents: an analogue technique for digital technology
C. Toumazou, J. B. Hughes and N. C. Battersby (Editors)
Volume 6 High frequency circuits F. Nibler and co-authors
Volume 7 MMIC design I. D. Robertson (Editor)
Volume 8 Low-power HF microelectronics G. A. S. Machado (Editor)
Volume 9 VLSI testing: digital and mixed analogue/digital techniques
S. L. Hurst
Published by: The Institution of Electrical Engineers, London,
United Kingdom

© 1998: The Institution of Electrical Engineers

This publication is copyright under the Berne Convention and the

Universal Copyright Convention. All rights reserved. Apart from any fair
dealing for the purposes of research or private study, or criticism or
review, as permitted under the Copyright, Designs and Patents Act, 1988,
this publication may be reproduced, stored or transmitted, in any forms or
by any means, only with the prior permission in writing of the publishers,
or in the case of reprographic reproduction in accordance with the terms
of licences issued by the Copyright Licensing Agency. Inquiries
concerning reproduction outside those terms should be sent to the
publishers at the undermentioned address:

The Institution of Electrical Engineers,

Michael Faraday House,
Six Hills Way, Stevenage,
Herts. SG1 2AY, United Kingdom
www: http://www.iee.org

Copublished by:
SPIE—The International Society for Optical Engineering
PO Box 10, Bellingham, Washington 98227-0010
Phone: 360/676-3290
Fax: 360/647-1445
E-mail: [email protected]
www: http://www.spie.org

While the authors and the publishers believe that the information and
guidance given in this work is correct, all parties must rely upon their own
skill and judgment when making use of it. Neither the author nor the
publishers assume any liability to anyone for any loss or damage caused
by any error or omission in the work, whether such error or omission is
the result of negligence or any other cause. Any and all such liability is
disclaimed.

The moral right of the authors to be identified as authors of this work has
been asserted by him/her in accordance with the Copyright, Designs and
Patents Act 1988.

British Library Cataloguing in Publication Data

A CIP catalogue record for this book

is available from the British Library

IEE ISBN 0 85296 917 1

SPIE ISBN 0 8194 2660 1

Printed in England by Redwood Books, Trowbridge

 Contents

Preface xiii
Acknowledgments xv
Principal abbreviations xvii
Principal notation xix

1 The semiconductor-diode laser 1

1.1 Background 1
1.2 Early developments 1
1.2.1 The first semiconductor lasers 1
1.2.2 Fabry-Perot gain and phase requirements 2
1.2.3 Some characteristics of diode lasers 4
1.3 Improvements to reduce operating currents 6
1.3.1 Heterojunctions: carrier confinement 6
1.3.2 Heterojunctions: photon confinement 7
1.3.3 Structures for'horizontal'confinement 9
1.3.4 Degree of confinement—the confinement factor 11
1.4 -Variations on conventional Fabry-Perot laser design 12
1.4.1 High-low reflective facets 12
1.4.2 External cavities 13
1.4.3 External grating 14
1.5 System requirements for single-frequency lasers 15
1.6 Introduction to lasers based on Bragg gratings 18
1.6.1 Introduction to Bragg gratings 18
1.6.2 Fabrication of gratings inside lasers 22
1.7 Some principal forms of grating laser 25
1.7.1 The distributed Bragg reflector laser 25
1.7.2 The distributed feedback (DFB) laser 26
1.7.3 More complex grating-based lasers 28
1.8 Summary 30
1.9 Bibliography 31
1.9.1 Semiconductor lasers 31
1.9.2 Optical communication systems 32
1.10 References 32
vi Contents

2 Gain, loss and spontaneous emission 37

2.1 Introduction 37
2.2 Electronic processes in semiconductors 37
2.2.1 Energy states 37
2.2.2 Occupation probabilities 40
2.2.3 Radiative recombination and absorption 41
2.2.4 Transitions and transition rates 43
2.2.5 Auger recombination 44
2.3 Absorption, emission rates and spectra 46
2.3.1 Absorption, stimulated and spontaneous emission in a
semiconductor 46
2.3.2 Stimulated-gain spectra in semiconductors 50
2.3.3 Homogenous and inhomogeneous broadening 53
2.3.4 Spontaneous-emission spectra from semiconductors 54
2.4 Semiconductor interactions with the lasing mode 55
2.4.1 Spontaneous-coupling factor 55
2.4.2 Petermann's '^factor' 58
2.4.3 Gain saturation in semiconductors 58
2.4.4 Spectral hole burning and carrier heating 59
2.4.5 Scattering losses 60
2.4.6 Free-carrier absorption 60
2.5 Henry's a factor (or linewidth enhancement factor) 61
2.6 Temperature-induced variations in semiconductor lasers 64
2.7 Properties of quantum-well-laser active regions 66
2.7.1 Introduction to quantum wells 66
2.7.2 Gain saturation and the need for multiple quantum wells 68
2.7.3 Strained quantum wells 69
2.7.4 Carrier transport 70
2.8 Summary 72
2.9 Bibliography 72
2.10 References 73
3 Principles of modelling guided waves 76
3.1 Introduction 76
3.1.1 Vertical and horizontal guiding 77
3.1.2 Index and gain guiding 77
3.1.3 Effective area and confinement factor 79
3.2 The slab guide 81
3.2.1 TE and TM guided waves 81
3.2.2 Multilayer slab guides 82
3.3 Wave equations for the TE and TM guided waves 83
3.4 Solving multislab guides 84
3.4.1 Effective refractive index 84
3.4.2 Reflection coefficient calculation 86
3.4.3 Gain-guiding example 89
 Contents vii

3.5 Scaling 90
3.6 Horizontal guiding: effective-index method 90
3.7 Orthogonality of fields 92
3.8 Far fields 92
3.9 Waveguiding with quantum-well materials 93
3.10 Summary and conclusions 95
3.11 References 96

4 Optical energy exchange in guides 97

4.1 The classic rate equations 97
4.1.1 Introduction 97
4.1.2 Rate of change of photon density 98
4.1.3 Rate of change of electron density 100
4.2 Some basic results from rate-equation analysis 103
4.2.1 Simplifying the rate equations 103
4.2.2 Steady-state results 104
4.2.3 Dynamic analysis 106
4.2.4 Problems of particle balance 110
4.3 Field equations and rate equations 111
4.3.1 Introduction 111
4.3.2 Wave propagation 111
4.3.3 Decoupling of frequency and propagation coefficient 114
4.4 Field equations with a grating 116
4.4.1 The periodic permittivity 116
4.4.2 Phase matching 117
4.4.3 Second-order gratings 120
4.4.4 Shape of grating 124
4.5 Summary 125
4.6 References 126

5 Basic principles of lasers with distributed feedback 128

5.1 Introduction 128
5.2 Coupled-mode equations for distributed feedback 129
5.2.1 Physical derivation of the coupling process 129
5.2.2 Complex gratings 131
5.3 Coupled-mode solutions and stopbands 132
5.3.1 Eigenmodes 132
5.3.2 The dispersion relationship and stopbands 133
5.4 Matrix solution of coupled-mode equations for uniform
grating laser 135
5.4.1 The field input-output relationships 135
5.4.2 Reflections and the observed stopband 137
5.5 DFB lasers with phase shifts 139
5.5.1 Phase shifts 139
5.5.2 Insertion of phase shifts: the transfer-matrix method 142
viii Contents

5.6 Longitudinal-mode spatial-hole burning 145

5.6.1 The phenomena 145
5.6.2 The stopband diagram 147
5.6.3 Influence of KL product on spatial-hole burning 148
5.6.4 Influence of phase shifts on spatial-hole burning 148
5.6.5 Spectrum and spatial-hole burning 150
5.7 Influence of series resistance 152
5.8 Simulating the static performance of DFB lasers 155
5.8.1 Light/current characteristics 155
5.8.2 Simulation of emission spectrum 159
5.9 Summary 161
5.10 References 162

6 More advanced distributed feedback laser design 165

6.1 Introduction 165
6.2 Linewidth 166
6.2.1 General 166
6.2.2 Calculation of linewidth under static conditions 168
6.2.3 Linewidth enhancement 171
6.2.4 Effective linewidth enhancement 172
6.2.5 Effective dynamic linewidth enhancement 176
6.2.6 Linewidth rebroadening 176
6.3 Influence of reflections from facets and external sources 177
6.3.1 Reflections and stability 177
6.3.2 Facet reflectivity and spectral measurements 179
6.3.3 Influence of facet reflectivity on SMSR for DFB lasers 180
6.4 Complex grating-coupling coefficients 183
6.4.1 General 183
6.4.2 Techniques for introducing complex grating-coupling
coefficients 183
6.4.3 Influence of complex grating-coupling coefficient on
static performance 184
6.4.4 Influence of complex grating-coupling coefficient on
dynamic performance 187
6.4.5 Influence of facet reflectivity 189
6.5 High-power lasers with distributed feedback 189
6.5.1 General 189
6.5.2 Techniques for obtaining high front-to-back emission
ratios 190
6.5.3 Laser-amplifier structures with distributed feedback 191
6.6 Dynamic modelling of DFB lasers 194
6.6.1 Uniform-grating DFB laser with reflective rear facet 194
6.6.2 Large signal performance of 2 x XJ8 DFB lasers with
strong and weak carrier-transport effects 197
6.7 Summary 202
 Contents ix

6.8 References 202

7 Numerical modelling for DFB lasers 209

7.1 Introduction 209
7.2 Ordinary differential equations 211
7.2.1 A first-order equation 211
7.2.2 Accuracy 213
7.3 First-order wave equations 214
7.3.1 Introduction 214
7.3.2 Step lengths in space and time—central-difference
method 215
7.3.3 Numerical stability 216
7.3.4 Gain and phase 218
7.4 Coupled reflections 219
7.4.1 Kappa coupling but no gain or phase changes 219
7.4.2 Matrix formulation 219
7.4.3 Phase jumps replacing scattering 221
7.4.4 Fourier checks 221
7.5 A uniform Bragg laser: finite difference in time and space 222
7.5.1 Full coupled-wave equations 222
7.5.2 MATLAB code 223
7.5.3 Analytic against numeric solutions 224
7.6 Spontaneous emission and random fields 226
7.6.1 Spontaneous noise and travelling fields 226
7.6.2 Null correlation for different times, positions and
directions 228
7.6.3 Spontaneous magnitude 229
7.6.4 Tutorial programs 229
7.7 - Physical effects of discretisation in the frequency domain 230
7.7.1 Discretisation process—integrals to sums 230
7.7.2 Fast Fourier transform (FFT) 232
7.8 Finite-element strategies for a spectral filter 233
7.8.1 Lorentzian filter 233
7.8.2 Numerical implementation 235
7.9 Application of the filter theory to gain filtering 237
7.9.1 General 237
7.9.2 Filtering the gain in the travelling-wave equations 238
7.9.3 Numerical implementation 240
7.10 Basic DFB laser excited by spontaneous emission 241
7.10.1 Introduction and normalisation 241
7.10.2 Field equations 243
7.10.3 Charge-carrier rate equation 243
7.10.4 Numerical programs 246
7.11 Summary 248
7.12 References 249
x Contents

8 Future devices, modelling and systems analysis 252

8.1 Introduction 252
8.2 Systems analysis 252
8.2.1 Introduction 252
8.2.2 Component modelling 253
8.2.3 System modelling 255
8.2.4 10 Gbit/s power amplification 256
8.2.5 Direct modulation: recapitulation 258
8.2.6 Simulation of integrated DFB laser and
electroabsorption modulator 259
8.2.7 Cross-gain and four-wave-mixing wavelength
conversion in an SOA 260
8.2.8 Simulation of cross-phase wavelength conversion in a
Mach-Zehnder interferometer incorporating two SOAs 263
8.3 The push-pull laser 265
8.3.1 Introduction: push-pull electronics 265
8.3.2 Symmetrical push-pull DFB laser 266
8.3.3 Asymmetry and the push-pull DFB laser 269
8.3.4 Speed of response for a push-pull DFB laser 272
8.4 Tunable lasers with distributed feedback 274
8.4.1 Introduction 274
8.4.2 Simple multicontact tunable lasers 277
8.4.3 Wide-tuning-range lasers with nonuniform gratings 279
8.4.4 Other tunable-laser structures 282
8.4.5 Tunable-laser linewidth and modulation 284
8.4.6 Modelling tunable semiconductor lasers 284
8.4.7 Multiple DFB lasers with optical couplers for WDM 285
8.5 Surface-emitting lasers 286
8.5.1 Introduction to surface-emitting lasers 286
8.5.2 Operating parameters of VCSELs compared with edge
emitters 287
8.5.3 Construction of VCSELs 291
8.5.4 Additional features of VCSELs 294
8.6 Summary 294
8.7 References 295

Appendix 1 Maxwell, plane waves and reflections 304

Al.l The wave equation 304
A1.2 Linearly polarised plane waves (in a uniform 'infinite'
material) 304
A1.3 Snell's law and total internal reflection 305
Al .4 Reflection amplitudes at surfaces: TE fields 308
A1.5 TE reflection amplitudes: three special cases 309
A1.6 Reflection amplitudes at surfaces: TM fields 309
 Contents xi

A1.7 TM reflection amplitudes at surfaces: four special cases 310

A1.8 Reflection for waveguide modes at facets 311
A1.9 References 312

Appendix 2 Algorithms for the multilayer slab guide 313

A2.1 TE slab modes 313
A2.2 TM slab modes 317
A2.3 Far fields 319
A2.4 Slab waveguide program 322
A2.5 References 323

Appendix 3 Group refractive index of laser waveguides 324

A3.1 References 328

Appendix 4 Small-signal analysis of single-mode laser 329

A4.1 Rate equations: steady-state and small-signal 329
A4.2 Carrier-transport effects 334
A4.3 Small-signal FM response of single-mode laser 336
A4.4 Small-signal FM response and carrier transport 337
A4.5 Photonic and electronic equations for large-signal analysis 339
A4.6 Reference 340

Appendix 5 Electromagnetic energy exchange 341

A5.1 Dielectric polarisation and energy exchange 341
A5.2 Electromagnetic-energy exchange and rate equations reconciled 344
A5.3 Electromagnetic-energy exchange and guided waves: field
equations 348
A5.4 References 351

Appendix 6 Pauli equations 352

A6.1 Reference 356

Appendix 7 Kramers-Kronig relationships 357

A7.1 Causality 357
A7.2 Cauchy contours and stability 359
A7.3 A proper physical basis builds in causality 360
A7.4 Refractive index of transparent quaternary alloys 362
A7.5 References 364

Appendix 8 Relative-intensity noise (RIN) 366

A8.1 References 371

Appendix 9 Thermal, quantum and numerical noise 372

A9.1 Introduction 372
A9.2 Thermal and quantum noise 373
xii Contents

A9.3 Ideal amplification 374

A9.4 The attenuator 377
A9.5 Einstein treatment: mode counting 378
A9.6 Aperture theory 379
A9.7 Numerical modelling of spontaneous noise 380
A9.8 Higher-order noise statistics 384
A9.9 References 385

Appendix 10 Laser packaging 386

A10.1 Introduction 386
A10.2 Electrical interfaces and circuits 386
A10.3 Thermal considerations 388
A10.4 Laser monitoring 388
A10.5 Package-related backreflections and fibre coupling 389
A10.6 References 391

Appendix 11 Tables of device parameters and simulated performance

for DFB laser structures 392

Appendix 12 About MATLAB programs 396

A12.1 Instructions for access 396
A12.2 Introduction to the programs 398

Index 405
 Preface

The authors have had the pleasure of working for many years within
one of the exciting areas of research and development: sources for
optical communications, an area which has seen phenomenal growth
over three decades. The semiconductor laser diode has proved to be
an essential device and one form or another of a distributed feedback
semiconductor laser (DFB laser) is a key component for modern
optical communication systems. Technically, the term DFB implies a
particular form of laser but to the authors the term has come to mean
any laser in which there is distributed feedback or where Bragg
gratings are integrated within the device.
This is a research monograph which combines a high level of
tutorial material with a research review. It will be useful for
postgraduate courses in optoelectronics studying either systems or
devices, research workers in optical systems, lasers and optoelectronic
devices, and also be of use forfinal-year-projectstudents in BEng and
MEng degree courses. Lecturers and research workers in the field of
laser diodes will find this a useful source of ideas, references and
tutorials which reach the explanations other books do not reach.
Combined with this tutorial material are accounts of the practical
operation and design of DFB lasers as used by the communications
industry along with discussions of the operational reasons for
requiring key modifications within the laser in order to meet specific
performance targets.
The tutorial material is distributed throughout the text and the
appendices. The different orders of operation of a Bragg grating are
outlined, novel concepts of the stopband diagram are explained, and
the modelling of laser interaction is based on sound fundamentals of
energy balance so that both particle models of lasers and electro-
magnetic-field models can be tied together. There are many different
ways of estimating the spontaneous emission, and again the con-
xiv Preface

sistency between the different methods is reconciled. The relationship

between gain and phase demanded by causality is also examined, tying
together different approaches; this is vital to understand the role of
the line width enhancement factor. A consistent approach to the
numerical modelling of the wave processes within a DFB laser is
extensively discussed in Chapter 7. Practical programs are supplied
through the World Wide Web (WWW). The use of MATLAB as a user-
friendly tool is emphasised with a number of sample programs.
The emphasis in the book is on the physical and numerical
modelling of laser diodes, and DFB lasers in particular. The treatment
of semiconductor materials and fabrication technology is necessarily
limited by the length of the book and the authors' expertise and
strengths, but within a whole chapter on this important topic enough
should have been said to ensure that the reader appreciates the
synergy that is required between material science, technology and
device physics in order to make successful DFB lasers. Material
parameters which are important for getting good physical representa-
tions of DFB operation are listed.
For similar reasons of length, the discussion about optical commu-
nication systems is not extensive but should be sufficient to set the
context of the DFB laser within such systems. System modelling where
the interactions between a number of sources, amplifiers, modulators
and so on is numerically modelled is a growing area and this is touched
on in a significant section in Chapter 8.
There are always idiosyncratic choices made by authors in their
selection of material and selection of references. There will be many
friends and colleagues who will be 'mortally offended' by the omission
of their best work from the reference list or from the material that is
discussed. All that one can say is that no offence was intended and it
is inevitable that with so much material published within this area a
selection has to be made. If the reader gains some of the terrific
enthusiasm and understanding that has gone into the work on DFB
lasers by research workers from all over the world, then it will all have
been worthwhile.
J.E. Carroll
J.E.A. Whiteaway
R.G.S. Plumb
 Acknowledgments

The authors would like to begin by acknowledging the help and

support of their respective establishments:
The Department of Engineering, University of Cambridge and Nortel
(Northern Telecom), Harlow Laboratories; along with British Tele-
com, BT&D Technologies Ltd., the Department of Trade and
Industry, the Science and Engineering Research Council (now the
Engineering and Physical Sciences Research Council), the Ministry of
Defence (now the Defence Evaluation Research Agency) and various
branches of the European Commission, all of whom have supported
our work on semiconductor lasers over many years.
There are numerous colleagues, students and research associates
who have either directly or indirectly contributed to our joint
understanding of the physics and modelling of laser diodes. It is
impossible to assess all their relative contributions, but each in his or
her own way will known that they have had an influence on our writing
or ability to do research for which we are indebted. Avoiding invidious
comparisons, we mention many by name below:
Mike Adams, Laurent Bacik, Peter Batchelor, Mark Bray, Andy
Carter, Andrew Collar, Phil Couch, John Devlin, Catharine Ewart,
Martyn Fice, Barry Flanigan, Kent Gardiner, Brian Garrett, Bob
Goodfellow, David Greene, Clive Irving, Jessica James, David Jones,
Ping Chek Koh, David Kozlowski, Ian Lealman, Dominique Marce-
nac, Paul Morton, Mark Nowell, David Nugent, Jeff Offside, Greg
Pakulski, David Parker, Eddie Pratt, Penny Probert, Dave Robbins,
Mike Robertson, Charlanne Scahill, Alan Shore, Tim Simmons,
Andy Skeats, Alan Steventon, Robin Thompson, Chi Tsang, Les
Westbrook, Ian White, Yuc Lun Wong, John Young, Si Fung Yu, Li
Ming Zhang.
xvi Acknowledgments

The names of all our friends and colleagues who have participated in
Laser Workshops, COST240, and key laser conferences over relevant
years are too numerous to be mentioned individually but they all have
had an influence on our thoughts about lasers and all are thanked. We
ask forgiveness of those who feel they should have been listed.
Thanks are also given to the IEE and in particular Jonathan Simpson
for support in publishing this book.
The support from our families, while apparently taken for granted,
has never been forgotten by us.
 Principal abbreviations

AR antireflection
CCCH Auger process involving three conduction-band
electrons and one heavy hole
CNR carrier-to-noise ratio
DBR distributed Bragg reflector
DFB distributed feedback
EDFA erbium-doped fibre amplifier
FP Fabry-Perot
GaAs gallium arsenide
GaAsxP,_x gallium arsenide phosphide
Ga.Al^As gallium aluminium arsenide
C^rCt Tn Ac P gallium indium arsenide phosphide
GAVCF grating-assisted vertical-coupler filter
HOFC higher-order Fourier coefficients
InP indium phosphide
IVBA intervalence-band absorption
L-I light-current
MATLAB registered trade mark of a MathWorks computing
environment
MBE molecular-beam epitaxy
MOCVD metal-organic chemical-vapour deposition
MQW multiquantum well
SCH separate-confinement heterostructure
RIN relative intensity noise
SiO2 silicon dioxide
Si3N4 silicon nitride
SMSR side-mode-suppression ratio
SNR signal-to-noise ratio
SOA semiconductor optical amplifier
TDM time-division multiplexing
xviii Principal abbreviations

TE transverse electric (often used as subscripts on

fields) (Chapter 4)
TIR total internal reflection
TLLM transmission-line laser modelling
TM transverse magnetic (often used as subscripts on
fields) (Chapter 4)
WDM wavelength-division multiplexing
 Principal notation

a arbitrary distance, atomic spacing

a, b, c, d parameters in a transfer matrix
am field losses per unit length
A linear-recombination coefficient —+AN
(usually nonradiative)
Am optical power absorption per unit time
si effective area or aperture of guide
aH linewidth broadening (enhancement)
factor^ (dnrml/dN)/(dninag/dN)

b coefficient for intervalence-band absorption

B bipolar recombination constant —•
B(NP) ~BN2 spontaneous bipolar emission
(3=27r/Am=n,w/c propagation coefficient in material with
wavelength Am
fi0, (3e, /3h, /3y, /3Z propagation coefficient associated with the
subscript (e.g. b for Bragg)
f3sp spontaneous-coupling coefficient

c velocity of light in free space

CN3 approximation for C (N2P) + C"(NP2); Auger
recombination

d, dY etc. layer thicknesses

S Po~~fii»a measure of the frequency offset
from the Bragg frequency
xx Principal notation

Sf signal bandwidth
A/ system bandwidth, gain bandwidth
A/w/ linewidth
Av/>/ spontaneous bandwidth for laser
A^/ change in frequency with change in gain
8(zl — Zo) Dirac delta function of position
Ap elemental reflection per Bragg period
AT elemental transmission coefficient per Bragg
period
8t computational step length in time
Ax uncertainty in position
A/? uncertainty in m o m e n t u m
8z space step related to St= 8z/vg

e, exp exponential
E optical electric field also as a function of zco
time, etc.: E(z), E(<o), E(t), E(y, z)
E o, •^racb A/;/(r) subscripted versions of electric field
,Er forward, reverse electric field
energy of particle
energy levels (e.g.: c, v —^conduction, valence
band)
\ > ©act,% :lad energy gaps appropriate to different mate-
rials
P quasi-Fermi levels for electrons (holes)
permittivity of free space (Chapter 4)
eA relative permittivity, same in layers 1 and 2
effective relative permittivity
&rreffr S
ri relative a n d effective, real a n d imaginary
permittivities
e gain-saturation parameter where G(N) =
G(N-Ntt)/(l + eS)

f optical frequency of operation = o)/clir

fhmgg,fmzx specific frequencies, e.g. central Bragg, max-
imum power etc.
fM modulation frequency, fM max= maximum
modulation frequency
F forward field: different functionalities, e.g.
F(t9z),F(z),F(T,Z)
Fhv FL forward-field values at input or z=L
 Principal notation xxi

ordered forward- and reverse-field

F
components

FN(%), (FP(%)) Fermi-Dirac function for electrons (holes)

(f> angle, phase shift, radians

$0 =YS0/rp, steady-state photon output

g net field gain/unit distance = rijC

g differential gain = dg/dN
g(N), g(co) net field gain/unit length: function of N, a)
etc.
g(n) net field gain/unit length in section n
(Chapter 7)
gm> gmW> gmW material optical field gain per unit length as
function of A, iVetc.
G(N) net power gain per unit time (as a function of
N)
Gm(N) material power gain/unit time
Gm differential optical-power gain = dG/dN
Gml small-signal dynamic c h a n g e s in Gm
Gthr round-trip, threshold gain
Gaerial gain of an aerial or radiating
aperture = 4TTA2/$&
Gmund round trip gain
yy vertical imaginary (evanescent) propagation
coefficient
F confinement factor

h Planck's constant
rjexl external quantum efficiency

i2n mean-square shot noise

isp, ispj, ispr spontaneous excitation, forward, reverse
/, Io9 Iu Iih current: total, steady state, small-signal,
threshold

/ current density into laser's active region

k Boltzmann's constant
xxii Principal notation

k electron's wave vector

k() = cojc propagation coefficient in free space
Kp, Klr Petermann transverse spontaneous emission
factor, subscipts p or lr
K^ Petermann longitudinal spontaneous emis-
sion factor
Ktflt Kz K/r product=total spontaneous emission
factor
Ksp s p o n t a n e o u s emission coefficient i n two-level
model
X^m stimulated emission coefficient in two-level
model
K normalised coefficient for spectral filter
K, Kfi, Krj coupling coefficients for grating
K
index+JKgain coupling coefficient for complex guide

L laser length
A, Ao free-space wavelength
A^ free-space Bragg wavelength
kelec(rm electron wavelength
A,m, Xx, A2 wavelength in material/guide, material 1, 2
etc.
\pmk wavelength of peak gain
A, A' Bragg-period first-order grating, period of
second-order grating A' = 2A

mc, mv, mr effective mass for: conduction, valence elec-

trons, reduced mass
M integer (e.g. diffraction order)
M 12 , M22 (etc.) matrix elements: row 1 columns 2 and row 2
columns 2 (etc.) of matrix M
M, Mj, MQ transfer matrix for section 1, phase shift
section etc.
ixo permeability of free space
ieiauve peniiea-Dniiy ^ = 1 in uiis LCAL;

n=nr+jnl complex-refractive-index real + imaginary

parts
nl9 n^y neff = \r&~[, V ^ , \/srep refractive index in layer
1 and 2, effective index
ng group index
 Principal notation xxiii

inversion factor (Appendix 9, Chapter 6)

n integer indicating segment number
TV, N(z) electron density (as function of z)
No, TV, steady-state, small-signal electron density
, N*, Nlr reference, threshold, transparency values of
electron density
N integers
K( density of electron states over an energy
range d%
N,, N2 electron states at energy %l9%2
acceptor density in jfr-doped material
K, K overall density of states in conduction and
valence bands
Ne =N/Ntr, a normalised electron density

p number of photons emitted per second; also

probability; also integer defining grating
order
p electron's momentum
P density of holes
Poul((o) dco optical power in spectral range co to (o + dco
Pfin> Pfout forward power at input, output (subscript r
for reverse)
n e t
Pmt> Pntisp* Pnetout output power, net spontaneous power
PSM side-mode optical power

q electronic charge
Q cavity Q ratio of stored energy to (2TT • energy
lost per cycle)

0, 9' angles giving measure of index, gain, cou-

pling coefficients through sin 8 and sinh 6'
0ain 6ivt half-cone angle of emission outside/inside
laser
#i, 0,*, 0c incident, diffracted, critical, angle
© = (fjy d/2) optical half thickness of layer
© = exp(/ a)opj phase factor in filter—for producing offset
frequency

^sp mean-square spontaneous excitation

R t , R s o u r c e , Rlaser resistances
xxiv Principal notation

R, R(T, Z) reverse field amplitude as a function of time

and space steps
RL reverse field at position z=L
$1 =9? +/& complex r a n d o m _ (normally distributed)
number such that 91 2 = 1
p amplitude reflectivity or reflectivity measured
for electric fields
Pkju Pright amplitude of reflection at left and right
facets
pr contact-layer resistivity

sspw spontaneous power coupling to forward

(reverse) fields
s, normalised 'temporal' step= vms. St
s step length in space for computation
S, S(x, y, z) optical photon density, as function of pos-
ition
Smax(z) maximum photon density at z, over an
equivalent area 6$
Ssponn Sslim density of spontaneous/stimulated photons
S' (f) d/ photon density emitted over a range of
frequencies d /
5 0 , SY photon density in the steady state
Sy, Sr photon density associated with forward,
reverse flux

t time
t(. thickness of the contact layer
T transmission coefficient at facets
T, Tn absolute temperature, noise temperature
T integer giving time steps
T j , T2 transistor labels
r time constant of spectral filter
Tp9 Tfp effective p h o t o n lifetime i n c l u d i n g t h e dis-
tributed losses
rsp lifetime for s p o n t a n e o u s radiative emission
rr spontaneous recombination time constant
including all mechanisms
rro recombination time constant when A/= Ntr
hnimband intraband relaxation time ~10~ 12 s
 Principal notation xxv

u(x, y), un(x, y) transverse-field distribution and for nth

mode
U stored electromagnetic energy
Uei Um electric and magnetic energy

effective volume for photons

V parameter for guide of width d where

v{x, y) transverse-field pattern for radiation fields,

v{y) in one dimension
Vcon contact voltage
Vj, AV^ junction voltage, change injunction voltage
vg group velocity=dco/d/3
vp phase velocity = a)/f3

wamn active-layer effective vertical width (i.e. thick-

ness)
w
effve)t(weffhor) effective vertical (horizontal) width of optical
intensity
Wv vertical-layer thickness
~o)M, ~o)OffSel normalised modulation and offset angular
frequency
a), co0, coOffSH =2irf angular (optical) frequency, central
frequency, offset frequency
coM , cor modulation and photon-electron-resonance
angular frequencies «co0
ft, O ; solid angle for either 'forward' or reverse
spontaneous emission inside and outside
guide

y co-ordinate distance

z axial distance
Z integer giving space steps
 Chapter 1
The semiconductor-diode laser

1.1 Background
The invention of semiconductor diode lasers [1-4] along with their
ability to be directly modulated with information [5,6] and the
invention of optical-fibre communication [7-9] in the 1960s-1970s
have together had far-reaching effects on telecommunications leading
to extensive optical-fibre communication networks and systems
[10-12]. This book explores one aspect of this success story. High-
performance optical-fibre systems need sources with a number of
features:
(i) stable single-frequency output;
(ii) modulation capability of gigabit/s;
(iii) stable operating lifetimes measured in 106—•lO8 hours at room
temperature; and
(iv) manufacturability.
The distributed-feedback (DFB) laser has these qualities and is now a
standard device. The design and development of these lasers has
proved to be fascinating work for those involved, and the physical and
numerical modelling which has gone into developing these lasers is
the topic of this book. Although the text concentrates on lasers with
distributed feedback, the modelling work presented is more widely
applicable and it is hoped that the reader will be as excited as the
authors by the unfolding story.

1.2 Early developments

1.2.1 The first semiconductor lasers
The earliest semiconductor diode lasers [1-4] were straightforward
p-n homojunctions formed in crystalline blocks of gallium arsenide
 2 Distributed feedback semiconductor lasers

(GaAs), or GaP^Asj _ x [4], with typical dimensions of the order of a few

hundred microns. In such lasers, two opposing parallel facets provide
reflective optical feedback, hence forming a Fabry-Perot resonator
(see, e.g. [13]) but of sufficiently small cross-section that the radial
fringes, typically associated with this optical interferometer, are not
important. At injected current densities of the order of 100 000
A/cm2, the stimulated emission from the electron-hole recombination
in the p-n junction plane is sufficient to create optical gain at
frequencies whose photon energies correspond to the band-gap
energy of the material. The optical feedback increases the light levels
in the resonator and lasing threshold can be reached, though this is
only possible in such simple p-n junctions when cooled below room
temperature and/or with short pulses to keep the thermal dissipation
low. Single-heterojunction lasers [14-16] were next developed. These
GaAs lasers had Ga^Alj.^As (#~0.3) on one side of the junction and
heterojunctions (Section 1.3.1) have the remarkable feature of
confining photons and charge carriers to the same region of the laser,
thereby reducing threshold-current densities to around 10 000 A/cm2
at room temperature. Powers of 10 W or more for 100 ns pulses were
now possible, and made the first useful lasers, though their spectral
and spatial output characteristics were relatively unstable. The double-
heterojunction (or G a ^ ^ ^ A s - G a A s - G a ^ ^ ^ s sandwich) [17-19]
laser further reduced the threshold currents and allowed continuous
operation at room temperature with current densities down to around
1000 A/cm2. Reducing the broad metal contacts to narrow stripes [20]
a few microns wide yet further improved the lasers, so that by the
1980s, Fabry-Perot diode lasers could be used confidently for data
transmission at over 100 Mbit/s through optical fibres [21,22] in real
systems. With careful design, such lasers emitted from a single small
spot, i.e. had a single 'transverse' spatial mode, but the spectrum was
generally multimoded corresponding to many 'longitudinal* modes.
The remarkable advances in lasers achieving stable single-frequency
operation with a capability for multigigabit/s direct modulation is a
key subject of this book.

1.2.2 Fabry-Perot gain and phase requirements

The earliest diode lasers were all Fabry-Perot (FP) lasers which remain
an important class of general-purpose laser. FP lasers, of small enough
cross-section, may be initially modelled by considering a resonator
which contains plane optical waves travelling back and forth along the
length of the laser. These waves will have optical (angular) frequencies
 The semiconductor-diode laser 3

a)=2iTf with an associated propagation coefficient ^8 = 2TT/AW where Xm

is the wavelength in the material/waveguide. Such a wave starting from
the left-hand plane reflector and travelling to the right is referred to as
a forward wave and has its phase and amplitude written in complex
form:
Ef(z) =E0 e{g-'^)x e-^ z (e>0 (1.1)
The amplitude decays or grows with distance because the wave suffers
scattering and other fixed losses am per unit length, but also
experiences a material optical gain gm per unit length caused by the
holes and electrons being stimulated into recombining at a rate which
increases with the injected carrier density. Note that both am and gm
vary with wavelength, but in particular that the gain gm(A.) has a limited
width of several tens of nm. The detailed mechanisms of optical gain
are discussed in Chapter 2 but at present gain is modelled by
considering a constant value of (gm— am)>0 along the laser so the
electric field at the optical frequencies oo and position z is described by
eqn. 1.1.
The laser is of length L and, for simplicity, it is assumed that there
is zero phase change on reflection from the partially reflective facets at
either end. At the right facet (z=L), the forward optical wave has a
fraction p.^ reflected and this fraction now travels back (from right to
left), and following the notation of eqn. 1.1, these reverse fields are
described by
Er(z) ={E0 pnghl e ( " ) L e-*L] e ( - " ) ( L ~ 2 ) e'^L'z) (1.2)
The time variation ejcot occurs in all terms, and will be included
implicitly only from now on.
The reverse wave now travels back to the left facet (z=0) and a
fraction p/£f/ is reflected to form the forward wave. For stable resonance,
the amplitude and phase after this single whole round trip must be
identical to the phase and amplitude for the wave when it started, or
in symbols:

Eo=EoPleflPnghle^-^Le~^L (1.3)
This results in the 'amplitude' condition for stable oscillation (lasing)
given by

/^A^e^'-^M (1.4)
or
4 Distributed feedback semiconductor lasers

divergent
output beam
reflection
optical length 0 Pright
(N half wavelengths)
reflection n
r .,
left
Figure 1.1 Operation of a Fabry-Perot laser
Showing schematic of standing electric-field waves inside the resonator

1
(1.5)
Pleft Pright

where the logarithmic term can be considered as a distributed

reflector loss. Simultaneously, the 'phase' condition must be met:
7
e -^ =l or 2/3L=2©=2NTT- (1.6)
where N is any integer.
This concept of a net round-trip complex gain, Gnund which has a
value of unity for lasing, holds for all lasers and not just for FP lasers
where Gwund=pMl p^ exp[(gm- am-jf3n)2L\. It is equivalent to the
classic condition for oscillation in electronic feedback amplifiers [23]
where the product of gain and positive feedback is unity. Put another
way, 1/(1 - Gwund) —•infinity; giving output without input! Noise in
oscillators and spontaneous emission in lasers give a non-zero input
keeping 1/(1 - Gmund) finite.

1.2.3 Some characteristics of diode lasers

For a FP laser, lasing might be expected only at \pmk where the peak
value of gain gm(Apmk) satisfies eqn. 1.4. However, as in any electronic
oscillator where random electrical noise gives the initial excitation, the
 The semiconductor-diode laser 5

Relative power spectrum

- frequency
1

0.5

A. 7V A A
-2000 0 2000
Frequency, GHz
Figure 1.2 Power spectrum for a Fabry-Perot laser
250 ^m Fabry-Perot laser operating at a central wavelength of around
1.55 jmm

magnitude of the round-trip gain Ground is ever so slightly less than unity
and it is the random spontaneous emission (discussed in Chapter 2)
which is amplified through gain and feedback, by the factor of
1/(1 - Gmund)~l03 to 106. Because the gain and spontaneous emission
spectra cover several nanometres change of wavelength, the phase
condition /3L=NTT (N integer) is satisfied at a number of frequencies,
or modes, within that band but the gain condition ensures that the
spontaneous emission is only amplified significantly at modes where
11/(1 - Ground) I is very large. As IGrowwrfl becomes smaller away from
Apeak, the amplification of the modes by the random excitation grows
weaker. The spectrum of the FP laser in general shows multiple modes
with frequencies determined from the phase condition giving a mode
spacing AFPf=(vg/2L) where vg is the group velocity for the optical
fields in the laser's waveguide. Figure 1.2 gives a typical spectrum for
a 250 juim Fabry-Perot laser operating at a central wavelength of
around 1.55 |xm showing these features.
With an ideal design, the laser's output power is proportional to the
drive current above threshold, as in Figure 1.3. This is useful for direct
amplitude modulation of the optical output by modulating the current
drive. This ideal linear behaviour is modified by heating which reduces
the optical gain and so increases the threshold. Heating and gain
variations can also alter the spatial variations in optical and electronic
6 Distributed feedback semiconductor lasers

T ' ^

3
optical
output / / >''
power, 2
mW
/ / ''''
1
/
/ /

0
20 40 60
input current, mA
Figure 1.3 Typical optical output-power/current curve for FP laser
and curves show the effect of increasing temperature T (see
Section 2.6, eqn. 2.44)

densities within the laser, thereby also changing the ideal linearity.
Modulation at gigabit rates creates additional phenomena such as a
change of frequency (called 'chirp'; discussed in Chapter 4).
It follows that lasing action can occur only in semiconductors which
exhibit optical gain; this normally implies a direct-band-gap semi-
conductor (see Chapter 2), and as a consequence silicon, so widely
used for electronic devices, is unsuited to make lasers. Furthermore,
the need for low threshold currents and stable optical modes requires
the complex technology of different materials for heterojunction
diodes as discussed briefly next.

1.3 Improvements to reduce operating currents

1.3.1 Heterojunctions: carrier confinement
It is found that, for lasing, densities of both electrons and holes, in the
conduction and valence bands respectively, are required to be
significantly greater than 1018cm"3. A major obstacle to achieving
these densities in conventional p-n junctions formed within a single
material is that carriers rapidly diffuse away from the junction and, to
compensate for this diffusion, injected currents of tens of thousands of
amperes per square centimetre are required. The previously men-
tioned double heterojunction controls this difficulty of diffusion.
 The semiconductor-diode laser 7

active width < 0.5 |um

electron quasi-Fermi
level
electron
leakage

electron curren conduction band

I TTT-

narrow gap wide gap p-type

wide gap n-type active layer confining layer
confining layer

hote valence band

leakage -v
sJ ~J

hole quasi-Fermi level

Figure 1.4 Double-heterojunction laser (schematic)
The electron and hole leakage is minimised by having energy barriers
which are several kT in height so that the carriers have inadequate energy
to surmount them

In Figure 1.4, the electronic-band diagram for a double heterojunc-

tion laser in operation is drawn schematically with holes flowing from
the right, and electrons from the left, each with the Fermi energy
defined by the injecting contact. The small slope on the bands
indicates the presence of electric fields. These are required to
overcome the ohmic resistance to the currentflow.The Fermi levels in
the materials adjacent to the contacts are typical of any forward-biased
semiconductor diode but, in the active layer where the electrons
recombine with the holes, the boundary conditions with the contacts
require continuity of the quasi-Fermi levels and so ensure strong
nonequilibrium conditions. Electron and hole leakage currents over
the heterobarriers are usually negligible at room temperature, whereas
the equivalent leakage away from a p-n homojunction would be
several orders of magnitude greater. Current densities of the order of
1000 A/cm2 can give electron densities well in excess of 1018 cm" 3 and
so give positive optical gain in the active region (Chapter 2).

1.3.2 Heterojunctions: photon confinement

It is a remarkably convenient feature of semiconductor physics that
increasing the bandgap of a semiconductor by changing its composi-
 Distributed feedback semiconductor lasers

0 > 0 critical for total internal reflection cross sectional side view

optical
intensity

ray

Figure 1.5 Rays and intensity for confinement of optical fields to an active layer
a Ray picture (total internal reflection)
b Field or modal picture

tion also normally decreases its refractive index. In Figure 1.5, the
outer cladding or confining layers have a slightly lower refractive index
than the active or gain layer. The layers act as a slab waveguide
confining the photons closely around the active layer through the
action of total internal reflection. The simple optical-ray picture is
shown in Figure l.ba. Total internal reflection (Appendix 1) requires
that the refractive indexes of each of the outer materials, n^ and ?%, are
smaller than the refractive index nx in the active region. A guided ray
has to propagate at a sufficiently shallow angle to the active region, and
from Snell's law, the angle of incidence 6>dcrilical. One may therefore
write:
sin 0>sin 9ailiral = n^/nx = \ - An/nY here An=nx - n% and n^ru, (1.7)
The angle <f> shown in Figure 1.5 then has to be 'glancing' such that
<f>= ( T T / 2 - 0 ) <COS~1 ( l - A n / n ^ V ^ A n / n ) (1.8)
where n is the mean index. GaAs has a refractive index of 3.6,
corresponding to nY, and Ga^^Al^As is lower by An— — 0.66x; typically
x=0A, so An is about 0.26.
The conditions in eqns. 1.7 and 1.8 ensure that optical rays which
propagate down the active region are reflected from side to side. This
ray picture, however, fails to show the evanescent fields in the cladding
layers which have to accompany total internal reflection. The more
accurate wave-guiding picture, shown schematically in Figure 1.56,
indicates that, although optical power does not propagate outwards
from the active layer, an evanescent wave penetrates to the order of a
wavelength into the cladding layers. The details of waveguiding in
multislab guides are discussed in Chapter 3 and Appendices 1 and 2.
 The semiconductor-diode laser 9

-1501
emitting spot s t r j p e contact

• - 0 .5 |um
Vertical'
V ^ heterojunctions

'horizontal' ~ 250 jim

Figure 1.6 Simple oxide-stripe laser (schematic)
Stripe metal contact -5-10 |mm wide helps to localise current and hence
gain and so provide 'horizontal' confinement while the heterojunction
provides 'vertical' confinement for both charge carriers and optical
fields

133 Structures for 'horizontal' confinement

The carrier and optical confinement, mentioned so far, have operated
perpendicular to the junction plane and this will be referred to as
vertical confinement. Most applications involving coupling of the laser
to an optical fibre require the laser to emit in a single mode with a well
defined spot, typically of the order of one wavelength wide which is
much less than the chip width. The light must then also be guided and
confined across the junction plane and this will be called horizontal
confinement. A high operating efficiency demands that the light and
the injected charge carriers are confined as closely as possible to the
same volume, so that the carriers must similarly be confined
horizontally as well as vertically.
Figure 1.6 illustrates one simple way of providing this horizontal
optical and electron confinement in a laser [21,22]. Contact is made to
the top of the laser through a stripe, etched through an insulating
oxide, and although the current spreads out sideways under this
contact, peak carrier densities are achieved under the middle of the
contact, and so this is where the lasing filament has its maximum
strength. The optical guiding in the junction plane is weak but
adequate and is determined by a combination of gain guiding and weak
index guiding (Chapter 3). Both of these guiding mechanisms are
altered by temperature gradients, carrier-density changes and stress
fields. Good waveguide design is essential to avoid the problems of
current spreading and weak guiding which lead to high threshold
currents, unstable modes and nonlinear light-current characteristics.
10 Distributed feedback semiconductor lasers

p contact ^ P GaAIAs

semi-insulating GaAIAs

active undoped GaAs

outline of emitting spot

\ n GaAIAs
n contact

Figure 1.7 Simplified end view of buried-heterostructure (BH) laser

Buried guiding layer (black region) may typically be 2 juum wide and
0.2 |xm thick

A theoretically good structure to solve this set of problems is the

buried heterostructure (BH) [24] where the active filament is surrounded
on all four sides by material of lower refractive index (and larger
bandgap). Figure 1.7 shows the essential features; however, the
technology associated with manufacturing such devices is difficult. In
the BH GaAs laser, semi-insulating GaAIAs regions force all the
injected current toflowstraight down through the active region, which
typically may be 2 |mm wide and 0.2 |xm thick. In this case it can be seen
that the active region is surrounded on all four sides by a material with
a wider bandgap which also has a lower refractive index, so forming an
optical waveguide (analogous to a rectangular optical fibre). Technol-
ogy and other constraints cause structures to be more complicated in
practice, especially in the longer-wavelength InP/GalnAsP-materials
system where p-n-p-n 'thyristor' current-blocking structures usually
take the place of semi-insulating material. These layers contribute
substantial stray capacitance which reduces the ability for high speed
modulation.
Another solution to the horizontal confinement problem is the ridge
laser [25] shown in cross-section in Figure 1.8. This needs only one
epitaxial semiconductor growth process, and the processing steps are
also relatively noncritical, resulting in a robust manufacturable device.
 The semiconductor-diode laser 11

contact to ridge
& p contact metal

< - - dielectric (oxide)

^ - p confining layer
"" active layer

• - - • emitting spot
-- n confining layer
(& substrate)
Figure 1.8 End viezv of 'ridge-waveguide' laser
Ridge typically may be 5-10 JJLITI wide with active layer beneath, -0.2 juim
thick

There are no semiconductor current-blocking structures which add

junction capacitances, so the stray capacitance is inherently low, and
the device is easy to modulate at high speed. However, it can be seen
that the current confinement stops just above the active layer, so
sideways loss of drive current occurs and carriers diffuse out from
under the ridge. Hence these devices typically require higher drive
currents than comparable buried heterostructures. With the ridge
guide, the optical fields extend out towards regions where the
channels, with their lower refractive index than solid semiconductor,
are etched. The effective refractive index (see Section 3.6) is then
lower at the edges of the optical field than in the centre and provides
the classic method for index waveguiding to occur. However the
refractive-index profile which creates the guide depends on the
optical-field profile which may change with power level. Careful design
is needed to provide sufficient optical guiding over the whole range of
required power levels and so avoid optical instability.

13 A Degree of confinement—the confinement factor

Heterojunctions are effective in confining charge carriers to active
narrow-bandgap material, as in the structures outlined previously.
However, the nature of waveguiding by dielectric steps results in
evanescent optical fields extending significantly into the wide-bandgap
semiconductor surrounding the active region. This effect is most easily
pictured in, say, the vertical dimension (Figure 1.9). If the active layer
with an effective thickness wavm is narrowed, a decreasing proportion,
known as the confinement factor, of the optical power travels within this
12 Distributed feedback semiconductor lasers

optical waveguide Optical intensity profile

with effective width ^effvert
confining layer
optical intensity overlap
I active layer

r
with effective active
width w
confining layer a vert

Figure 1.9 Overlap of carriers and optical field

active region. This vertical confinement factor is given by

T = wavjweffvert in Figure 1.9 where weffvert is the effective vertical width of
the optical intensity. The concept is extended in Chapter 3 to both
horizontal and vertical dimensions.
In modelling of diode lasers, the confinement factor F is an essential
parameter which 'sweeps up' a large number of details of the structure
into a single parameter that can be estimated from calculations of the
two-dimensional guiding or sometimes from curve fitting between the
theoretical and experimental current/light characteristics. Its impor-
tance stems from the fact that, with a well designed index guide, the
value of F remains substantially the same over a wide range of optical-
power levels, despite the influence of injected carriers on the local
active-region refractive index. In addition, F is almost constant over a
useful range of frequencies around that of the main mode for which F
has been calculated.

1.4 Variations on conventional Fabry-Perot laser design

1A.1 High-low reflective facets
Lasers considered so far have been 'standard' Fabry-Perot devices,
which are probably the most straightforward to fabricate. Specific
applications of lasers may require features such as:
(a) high output power (tens of milliwatts),
(b) small emitting area (-1 jmm2) matching into optical fibres,
(c) narrow (AA<1 nm) spectrum, and
(d) high-speed (Gbit/s) modulation capability through directly mod-
ulating the drive current.
More complex structures exist which enhance these useful features,
either singly or in combination. There are usually many ways of
 The semiconductor-diode laser 13

,' to ^
1 monitor.

High reflectivity coating Low reflectivity coating

Figure 1.10 High-low reflective facets

mirror lens AR coating coupling

(partial or full)
external
^ cavity cavity 1

diode laser
(a) (b)

Figure 1.11 External-cavity and cleaved-coupled-cavity lasers

a External-cavity
b Cleaved-coupled-cavity

achieving any particular characteristic from a laser, but a few

representative variations are briefly reviewed below.
In a conventional Fabry-Perot laser, the light comes out equally
from both facets. Higher efficiency is possible by forcing most of the
lasing light to come out of one facet, with only a small amount of light,
used for monitoring, from the other (Figure 1.10). This can be
achieved through the use of a high-reflectivity coating on the rear facet
and a reduced-reflectivity coating on the front (high-low facet
coatings) [26]. The product of the two field reflectivities (plefl pright) is
approximately the same as in an uncoated laser, so that the net
feedback and gain required for oscillation (eqn. 1.5) are retained at
about the same level. Extending the round-trip gain calculation in
Section 1.2.2, leads to the expression for the ratio of facet power
e m i s s i o n s : Pkft/PHght= (pright/Pkji) ( 1 " " Piejf)/ (1 ~ Pngtu) •

1.4.2 External cavities

The external-cavity laser (Figure 1.11a) is an interesting device [27,
31] where, by varying the external cavity's dimensions and the
reflectivity of the laser facet emitting into the cavity, conditions can be
14 Distributed feedback semiconductor lasers

optimised for very narrow-line (<lMHz) continuous output. The

essential features of the oscillation condition are similar to those
discussed for the Fabry-Perot laser but now the phase condition has to
be satisfied simultaneously within the laser itself and within the
external resonator. The length of the external cavity may be adjusted
so that this phase condition is met only once within the gain spectrum
of the laser's material. Changing temperature will cause such a laser to
become two moded unless it is readjusted.
With appropriate device gain and full antireflection coating of the
laser so as to produce one long effective cavity, the many modes of the
combined system can 'lock' together in phase to create a 'mode-
locked' laser giving optical pulses with peak powers at the watt level for
durations of the order of a picosecond. Repetition rates of many tens
of gigabits per second are possible, determined by the optical-pulse
round-trip time within the extended cavity. However, these pulses have
a broad spectrum appropriate to the short pulse width [28, 29].
Monolithic mode-locked lasers can now achieve pulse-repetition rates
of hundreds of gigabits per second [30].
The cleaved-coupled-cavity [31,32] (or C3) laser is a variant (Figure
1.116) on the external cavity laser, where a FP laser is cleaved into two
shorter FP lasers of incommensurate lengths with the cleave forming a
third short (typically <1 |xm-length) air-filled cavity (gap) between the
two laser sections. Such lasers have numerous attractive properties in
principle, including single-line operation with tunability of that line.
However, their manufacturability and stability in operation have
proved to be too variable for use in real systems.
Another way to achieve single-frequency operation under modula-
tion is to operate one 'master' laser continuously, and then inject
power from this into a 'slave' laser which may be modulated [33].
When carefully adjusted correctly, such a system works very well, but it
is again expensive and relatively bulky, and a cheaper and more robust
source of single optical frequencies is still required.

1.4.3 External grating

A tunable device which has become popular for many optical
measurements is yet another variant on the external-cavity theme,
where the external mirror is replaced by a diffraction grating (Figure
1.12) [34]. Single-frequency operation is ensured because feedback
only occurs at the wavelength selected by the grating, and a bonus is
that the laser may be tuned over the whole of the material-gain
spectrum by rotating the grating mechanically. Precision mechanical
 The semiconductor-diode laser 15

AR coating

output

grating lens
diode laser
Figure 1.12 Laser with external grating reflector

design allows linewidths of less than 100 kHz and tunability over tens
of nanometres to be achieved, but the mechanical assembly is too
expensive and delicate for such lasers to be used in the robust
environment of commercial communication systems, especially as the
maximum frequency for direct modulation is limited.

1.5 System requirements for single-frequency lasers

Before going further into the design and dynamic operation of single-
frequency lasers, the commercial push for such requirements should
be understood. A schematic long-distance optical-fibre link is sketched
in Figure 1.13 for an amplitude-modulated system. To keep costs
down, the system should operate with the maximum possible distance
between the repeaters which will regenerate the signal and retransmit,
because repeaters are expensive.
At one time it was envisaged that adequately long distances might be
possible only with 'coherent' detection where, like a television
heterodyne receiver, one has a tunable local oscillator (tunable laser)
mixing in a photodiode with the incoming optical signal to produce an
intermediate-frequency output which is detected and amplified. The
advent [35,36] of the low-noise optical erbium-doped fibre amplifier
(EDFA) and its subsequent development have demonstrated that the
type of system shown in Figure 1.13 is almost as sensitive as the more
complex coherent systems; consequently coherent receivers are not
discussed here.
The optical fibre [37] is, of course, a vital component, but in the
1960s and 1970s the optical wavelength was determined by the ability
to make lasers with layers of Ga1_xAlxAs giving emission in the
0.7-0.9 fxm range. However, it was soon realised that the loss process in
silica glass made the fibre's attenuation much lower around two special
wavelength 'windows' (1.3 |xm with attenuation of -0.5 dB/km and
1.55 (xm with attenuation of ~0.2 dB/km). The challenge of new
materials for lasers at these longer wavelengths was met with both
16 Distributed feedback semiconductor lasers

directly modulated laser

IHfH FIBRE LINK
[laser pump

optical amplifier
or Q B Q O D O Q Q D O Q QQQ Of
(erbium doped fibre)
laser + modulator
M-HH photo- spectral

nun '\C
input signal dispersed and detector filter
attenuated
modulated optical amplifier
carrier demodulator
TRANSMITTER signal conditioner

HfHi mm
restored/retimed signal

NN r<# ' " directly modulated laser

or
laser + modulator
RECEIVER / REPEATER FIBRE LINKS
RECEIVER/REPEATER

UjOaQflBQQQQlLQffllD

Figure 1.13 Schematic optical system with repeaters

Some repeater systems only need the optical amplifier without detection
and restoration. The fibre link may have several lasers feeding into it via
an optical coupler for WDM systems

standard fibre and laser sources becoming centred around the 1.3 |xm
and 1.55 |xm wavelengths.
Although the attenuation in optical fibres is sufficiently low to allow
links of well over 100 km using standard 1 mW FP lasers at 140 Mbit/s,
at higher bit rates there is a significant dynamic wavelength shift (or
'chirp') in the laser's emission which causes loss of information,
regardless of the laser's output power, because of the dispersion in the
fibre. Dispersion is the phenomenon whereby, because of wavelength-
dependent phenomena within a material, different wavelengths in the
fibre have different speeds of propagation. Consequently a pulse
which contains a range of wavelengths will spread as it travels.
Eventually, given a long enough fibre link, one pulse spreads into an
adjacent pulse and causes interference. At 1.55 |xm, where the
attenuation in standard silica fibre is lowest, dispersion is significant
amounting to about I7ps/kmnm. For such standard silica fibre,
dispersion is nominally zero at 1.3 |mm, but laser wavelengths are very
rarely matched exactly to the dispersion minimum of the particular
fibre in a link (due to manufacturing and installation tolerances, as
well as temperature variations) so low chirp is desirable from lasers,
 The semiconductor-diode laser 17

even at 1.3 |Jim or in special low-dispersion (dispersion-shifted) fibre

that can be formed around 1.55 |mm.
As an example of transmission in standard fibre at 1.5 |xm with
dispersion of 17ps/kmnm, a bit rate of 500 Mbit/s over 100 km
requires a spectral width ~0.15 nm (-20 GHz) or less if interpulse
mixing is to be limited to 10% of the bit period. This rules out the use
of the typical multiple-longitudinal-mode Fabry-Perot laser where, for
a cavity length of 400 jxm, the spacing of the modes is 0.8 nm
(~ 100 GHz). Increasing the bit rate to 5 Gbit/s over the same distance
would require a source spectral width of less than 0.015 nm, which is
a difficult design problem even for the single-line distributed-feedback
(DFB) laser introduced in Section 1.6 and techniques to compensate
the dispersion in the fibre caused by chirp may need to be used.
Besides the need for narrow linewidths to reduce the effects of
dispersion and so permit longer transmission distances, there is
another commercial goal of enhancing the information capacity of
optical-communication networks by transmitting different channels of
communication at sets of well defined wavelengths. Such a process is
called wavelength-division multiplexing (WDM) [38]. As the channel
spacing is decreased, one refers to dense WDM but at the time of
writing dense WDM is still a gleam in the eye of systems' designers.
Another gleam is wavelength routing whereby a particular wavelength
is assigned to a particular route or perhaps identified with some
particular transmitter.
While wavelength and linewidth control for a laser are of primary
importance in communication systems, it must not be thought that
these are the only parameters that matter. A key requirement is to
achieve an adequate carrier-to-noise ratio (CNR) so that bit-error rates
(BER) are less than 1 error in 109 bits. While increasing the optical-
carrier power is an obvious method of improving the CNR and
reducing any BER, the output-power levels for long-distance fibre
systems are restricted to the milliwatt range to limit nonlinear mixing
and other nonlinear effects in the fibre. The output of a laser will
contain amplitude noise (measured by relative-intensity noise or RIN),
frequency noise (line width broadening), and unwanted side modes
(measured by side-mode suppression ratios or SMSR). The detailed
specification of how low these levels must be usually depends on the
application, but typically the combined power of all these unwanted
signals (the noise) over the system bandwidth needs to give a CNR in
excess of 30 dB to maintain adequate BERs. Gowar [39] gives useful
power budgets for several optical-communication systems.
18 Distributed feedback semiconductor lasers

angle of
incidence angle of
diffraction

grating period
blazed A
reflection
grating

Figure 1.14 Bragg grating in reflection

The Bragg grating buried inside a laser may not be the final answer
to the problems of precise frequency determination and high CNR,
but it provides an excellent starting point and a fascinating study. This
chapter therefore concludes with an introduction to the Bragg grating
and how it may be buried inside a practical laser.

1.6 Introduction to lasers based on Bragg gratings

1.6.1 Introduction to Bragg gratings

It is likely that most readers have observed the selection of colour
(wavelength) on looking at a CD ROM under a white light at certain
angles. This is one example of a periodic system selecting the
wavelength of light through the phenomena recognised and analysed
by Bragg [40] around the 1920s. Figure 1.14 illustrates the conven-
tional reflective diffraction of a plane wave from a periodic grating
structure with elements spaced at a period A. A plane incoming wave
is incident at an angle di9 and the 'Mth-order' diffracted plane wave
emerges at the angle 6d. Constructive interference between the
diffracted waves emanating from adjacent elements of the grating
requires the condition
A(sin 0,-sin 6d)=M\ (1.9)
The integer M=0, ±1, ±2, . . . defines the order of the diffraction or the
phase shift in wavelengths between the diffracted waves from adjacent
grating elements, and A is the wavelength of the radiation.
Because a Bragg grating can select specific frequencies, it is sensible
to consider embedding a diffraction grating into a semiconductor
waveguide. This is illustrated in Figure 1.15 with concepts similar to
those in Figure 1.14, except that there are now different refractive
 The semiconductor-diode laser 19

angle of

wave-front diffraction

4TIM

^ incidence^

wave-front
Figure 1.15 Diffraction from grating embedded in laser waveguide
For a grating embedded inside a laser, 0rrr/2 and for feedback the
diffracted wave 0d~7r/2. However, one can also have 0d~O when the
diffracted wave gives radiation out from the guide. The diffracted
wavefront is reinforced because there are 2T7"M radians per pitch where
M is an integer and is known as the order of diffraction

indices and one is also thinking of obtaining a diffracted wave which

feeds backwards. Using the usual Huygen's construction for con-
structive interference from 'identical' points but at different periods of
the grating, one requires the sum (or difference) of the total optical
path in radians to be integer multiples of 2rr. In terms of distances and
the wavelengths appropriate to the different materials as in the
diagram of Figure 1.15:

A sin ft A sin 6d
(1.10)
m2

The integer M defines the diffraction order.

On applying the result of eqn. 1.10 to a propagating optical mode in
a DFB laser waveguide, one notes that the angle of incidence d{ cannot
be less than the critical angle for total internal reflection [41] d(. so
that

sin ^ sin 0 ^ = (1.11)

where n} and n^ are the refractive indices in the central and outer
regions of the waveguide, respectively. Inserting eqn. 1.11 into eqn.
1.10 gives
20 Distributed feedback semiconductor lasers

(1.12)

where an integer p = A/(2~AJ defines the grating order. Because of the

small differences in the refractive index, Am2~AOTl~Am, the effective
wavelength in the guide. In general, for a laser with feedback along the
waveguide, one requires 0^6^ IT/2 requiring, from eqn. 1.12 using
the equality sign and sin 0d=l, that the order p of the grating equals
the order M of the diffraction. Given the physical periodicity A of the
grating, the optimum free-space wavelength Xb which is most strongly
reflected is called the Bragg wavelength and satisfies the Bragg
condition:
M(A/ % )=2A (1.13)
where nejf is a mean value of the refractive index in the guide and
relates the free-space Bragg wavelength A^, with the corresponding
guide wavelength Am=(Ab/neff).
However, it is also possible to find gratings where sin 6d~0 (e.g. when
p = 2M) or sin 8d is significantly less than 1 so that the diffracted wave
is then radiated away from the guide. The second-order grating where
p = 2 is of particular note because this can be used to couple light from
a laser out into the direction perpendicular to the junction to form a
surface-emitting laser (Section 8.5.1). Table 1.1 summarises the
influence of grating order p and diffraction order M on the feedback
and radiation loss experienced by a propagating mode.
There is often confusion between what is loosely termed a pth-order
grating, and what is correctly a pure pth-order grating. The former
expression refers to a grating with a fundamental spatial periodicity of
pAm/2, and in general is described by Fourier components at the
fundamental and higher-harmonic periodicities, while the pure
grating will have only the fundamental component. A pure second-
order grating will not be able to give feedback.
A first-order grating with a period of one-half a guide wavelength
gives optical feedback in the first diffraction order. By comparison, a
second-order grating radiates normal to the grating in the first
diffraction order but gives feedback in the second diffraction order.
Because second-order gratings have larger dimensions than first-order
ones, they are often easier to define photolithographically, and so can
be preferred by manufacturers. The radiation loss need not be serious
unless such a grating is deliberately designed to couple out light as in
a surface-emitting laser (Section 8.5.1). A third-order grating gives
 The semiconductor-diode laser 21
CO
11
LL
Is
II 0 ID
0 CO
LL CC
ings

I
05
C
"I
cO .Q
05
1
i radiation los

II 0
0 CO
LL CC CE
CO CO CO
>
forv
UO1

forv

1 O T3 I D
0 0
ID
0
1
II 0 0 0
y of feed

i-1
c
di "5 g1
1 0
Sum

der ofdii ffracti

IS!
sES O
o c V
+L O "D
CO O .h=
v— d) f—
L L C7D h-
22 Distributed feedback semiconductor lasers

radiation loss only in the first and second diffraction orders at angles
symmetrically disposed about the normal, but gives feedback in the
third diffraction order. In general, feedback is obtained from a grating
of order N in diffraction order N, but diffraction orders intermediate
between 0 and N result in radiation loss for such a grating.
In principle, just as the blaze or shape of the grating in a
spectrometer alters the power division between the different orders of
diffraction, so also one can change the reflection and radiation
[42-44] by tailoring the grating-tooth shape. One may also need to
distinguish between the TE and TM polarisation within the guides
[45]. In general, the requirement to mass manufacture gratings buried
within a laser gives severe constraints on the tooth shape which is
usually controlled by the chemical etching and crystal orientation.
Details of wave propagation and gain within such gratings with a
distributed reflectivity are covered at the end of Chapter 4 and
beginning of Chapter 5. A method of calculating the reflectivity per
unit length is discussed in Chapter 4 with the help of a MATLAB
program slabexec.
1.6.2 Fabrication of gratings inside lasers
The theoretical concepts of lasers with distributed feedback [46] were
developed well before the technology could deliver a reliable and
useful method of fabrication. The early semiconductor lasers with
distributed feedback operated at low temperatures and had short
operating lifetimes [47,48] but in the mid 1970s GaAs lasers with
distributed feedback could be made to operate at room temperature
[49], and then developments with new materials led to room-
temperature operation of lasers, with distributed feedback, around
1.3 fxrn and 1.55 |xm [50,51] in the early 1980s. The details of
fabrication of semiconductor lasers and the techniques of substrate
formation and subsequent epitaxial-layer growth using metal organic
chemical vapour-deposition (MOCVD) or molecular-beam epitaxy
(MBE) are outside the scope of this book and are far from trivial. The
research and development of materials-growth techniques for forming
the confinement-layer/active-layer heterojunctions of lasers have been
vital. All that can be given here is a mere glimpse of this marvellous
interaction of materials and technology that has led to reliable
commercial devices. Figure 1.16 illustrates some principal processing
steps needed to form a semiconductor laser with a uniform embedded
grating which creates the feedback as described in Section 1.6.1.
The first step is to grow, by the methods of epitaxial growth, good
laser structures with the substrate followed by the confining layers and
 The semiconductor-diode laser 23

460 nm waveguide
period
grating

active
layer

stripe contact

Figure 1.16 Fabrication steps for ridge-DFB-laser stru cture

a Grow MQW SCH structure by MOCVD then define and etch DFB
grating; b Overgrowth of grating; c Etch ridge; d Apply stripe contact.
Completed device

active layer as discussed in Section 1.3. The active-gain region can be

a single layer of material (referred to as a bulk-gain region) or it can be
formed from one or more 'quantum-well' layers, e.g. a multiquantum
well (MQW) region which may be =^100 nm thick, surrounded by
separate-confinement-heterostructure (SCH) layers (see Section
2.7.4). While some further discussion of quantum-well material is given
in Section 2.7, the details of growth again have to be left for further
reading. Many of the steps for constructing a typical DFB laser are
similar to those for constructing a good FP laser, and the grating can
be etched into epitaxial layers below the gain region or above the gain
region (substrate or surface gratings). Here is depicted the scheme
where the grating is etched into a confining layer above the active
layer. Hence after the confining and active layers have been grown, the
grating is lithographically defined in a resist which then acts as a mask
for the grating etch. For mass production, techniques are favoured
which expose a periodic pattern in a resist using ultraviolet-laser
 24 Distributed feedback semiconductor lasers

55° p-lnP

0.27 |im waveguide layer (X = 1.1 fyim) InGaAsP

0.15|im active layer

(A.= 1.55M-m)lnGaAsP
n-lnP

effective
permittivity
for 0.46 |im * e min
mode
Figure 1.17 Schematic cross-section of DFB laser grating etched in epitaxial-layer
structure

radiation. However, electron-beam (E-beam) lithography is an alter-

native technique which uses special electron-sensitive resists and has
the advantage of greater resolution enabling first-order gratings to be
written more easily. E-beam lithography also facilitates fabrication of
more complex structures such as gratings with a changing periodicity
or incorporating 'phase shifts' as discussed shortly. After definition,
the grating is chemically wet-etched, or dry-etched using a plasma, so
as to transfer the pattern in the resist into the semiconductor. The
crystal planes and etching processes usually determine the angles and
shapes of the grating teeth.
Following removal of the resist, the grating is then overgrown with
material which completes the guiding structure and has a highly
doped top layer to make a good ohmic contact. The substrate usually
forms the lower contact. The vertical waveguide is then fabricated and
is usually either a ridge structure (Figures 1.8 and 1.16) or a buried
heterostructure (Figure 1.7). For the ridge guide, the final steps
involve depositing an insulator on the /nside ridge structure which is
often SiO2 or Si3N4. Photolithography is then used to open a window in
the insulating layer on the top of the ridge and a metallisation-
evaporation process completes the device by applying contact metal
(Figure 1.16d).
Figure 1.17 illustrates schematically the cross-section of a practical
second-order etched grating for operation around 1.55 ixm free-space
wavelength. The guide wavelength Am is 0.46 |xm which is the same
periodicity as for the grating, so that at first sight it looks as though the
feedback would all cancel, but because the teeth have an average
length of ~ Aa/4 there is net feedback along the laser as in a first-order
grating.
 The semiconductor-diode laser 25

passive
waveguide Bragg grating optical gain Bragg grating
layer reflector section reflector

Figure 1.18 Diagram of distributed-Bragg-reflector (DBR) laser

1.7 Some principal forms of grating laser

1.7.1 The distributed Bragg reflector laser
The insertion of a 'Bragg grating' into a laser has proved to be a very
good solution to providing low-cost and reliable single-frequency
optical sources capable of high-speed modulation. A device with a
uniform grating is usually known as a distributed-feedback (DFB) laser
or, in a slightly different form, the distributed-Bragg-reflector (DBR)
laser. The DBR laser is easier to understand and this is considered
first.
Figure 1.18 shows schematically a laser with a centre section of
waveguide with gain which is essentially identical to the Fabry-Perot
lasers considered earlier. The key change is that, instead of using
lumped reflections from the end facets of the laser as in a FP laser, the
passive waveguide is extended beyond the gain region into the end
sections where the grating is etched and designed to give its multiple
tiny reflections adding in-phase at the operating wavelength. The
advantage of this structure over the simple Fabry-Perot laser is that the
multiple reflections give significant feedback only around a fre-
quency—the Bragg frequency—determined by the grating pitch.
The description of a grating as in Section 1.6 gives one straight-
forward way to consider the effects of a Bragg grating, but another
simplified argument, particularly appropriate for a grating in line with
a waveguide, is to note that identical reflectors which are spaced Am/2
apart within a guide create reflections which all 'add'. The point to
note here is that, between any two reflectors (A and B), the wave
travelling forward advances by 180° from A to B but the reflected wave
also has to advance by 180° to get from B back to A, giving a 360°
round trip determining the reference phase of the reflection. Each
reflection therefore reinforces the other reflections.
26 Distributed feedback semiconductor lasers

;XX>: A, 12 \ N
ooo 12
m

n
l

X 12 X 12 X 12
m m m
reflector section centre (gain) section reflector section

Figure 1.19 Grouping of a grating into \m/2 sections where all reflections Ap from
each section add

Figure 1.19 sketches the concept of reflective blocks formed from

alternating sections of refractive indices nx and n^, where each total
block is AOT/2 thick at the desired 'Bragg' frequency where there is to
be maximum reflection or maximum feedback. Each block will give
some net reflection Ap to a wave that is travelling forward. With N
pairs there will approximately be a net reflection NAp at the Bragg
frequency, making no allowance for the loss of forward energy as the
wave is reflected back; a more accurate calculation must await until
Chapter 4. As the optical frequency moves away from the Bragg
frequency, so this reflection rapidly reduces because the reflections no
longer reinforce each other in phase. To estimate the maximum
magnitude of this effect (see Appendix 1), take each section of high or
low refractive index to be Am/4 in length so that A p - {nx — n^)/ {n^n^)
for each reflection, giving a net reflection per unit length of K:~4
(n\ ~~ vh)/1 ( n \ + %)KJ - The reflection per unit length K is known as the
grating coupling factor and, for a grating of length L, KL is a key
dimensionless parameter determining the frequency selectivity and
performance of the grating with practical values of KL taking typically
values around 2.

1.7.2 The distributed feedback (DFB) laser

In principle, to obtain oscillation there is no need to separate the
regions of gain and reflection as shown in Figure 1.18. Figure 1.20
indicates schematically the concept of a uniform distributed feedback
laser whereby, with distributed reflections and gain within the laser
cavity, an optical wave travelling in one direction is continuously
 The semiconductor-diode laser 27

active layer giving gain

gratinc

CO OC 00 CO
distributed feedback
forward

reverse

Figure 1.20 Uniform distributed-feedback laser

The concept of continuous feedback combined with gain

scattered into the optical wave in the reverse direction. Both waves
grow as they travel towards their respective facets because of the
feedback and the gain. The abrupt changes in refractive-index, with
each section of length Am/4, are purely schematic to emphasise the
main features. This is the distributed-feedback (DFB) referred to
earlier. However, it is found that the DFB laser oscillates at two possible
frequencies slightly removed from the Bragg frequency depending on
KL (Chapters 4 and 5). The reason for this is that the phase condition
for oscillation cannot be met at the Bragg frequency, as discussed
below.
Now reconsider the DBR laser. It is found that the presence of gain
enhances the reflectivity of the Bragg gratings and this increased
reflectivity dramatically reduces the required gain for the FP-gain
section. A particular utility of the argument used in Section 1.7.1 based
on the Am/2 sections above comes in the ideal design of the central
section for a DBR laser required to operate at the Bragg frequency of
the grating where there is the maximum reflection. It is possible to
ensure that the round-trip phase matching occurs, as in a Fabry-Perot
laser, if the two reflectors are identical mirror versions of each other
with a real reflection of the same sign, and then these reflectors are
separated by a central section of length NxAm/2 (Figure 1.19). The
shortest possible Fabry-Perot section is Aw/2 long (N = 1) and this then
requires the insertion of an additional Am/4, say into the lower-
refractive-index region to form a length Am/2 between two mirrored
sections of Bragg gratings (Figure 1.21a).
28 Distributed feedback semiconductor lasers

(a) DFB laser with single ^ m /4 phase shift (b) DFB laser with two J^/8 phase shifts

Figure 1.21 Schematic ofDFB-laser variants

Many commercial lasers simply insert lithographically a physical

Aw/4 phase shift into the grating mask, and the resultant devices are
known as Xm/4-shifted or phase-adjusted DFB lasers and oscillate with a
single frequency close to the Bragg frequency of the grating. These
Am/4-shifted lasers do not give optimum dynamic wavelength stability
during modulation, and so some more complex phase-adjusting
regime such as two ATO/8 phase shifts may be preferred (Figure 1.21*),
giving greater stability in the output [52]. More phase shifts have been
proposed [53], but these add to the difficulties of manufacture and the
simpler construction is usually preferred unless clear advantages are
demonstrated. High-low reflectivity facets have also been proposed to
give single-frequency operation [54], but the precise position of the
high- reflectivity facet with respect to the periodicity in the index
grating is important and it is currently impossible to maintain this few-
nanometre tolerance in production.
Although lasers with uniform gratings can give good single-mode
operation, the selection of the mode, either just above or just below
the Bragg wavelength, depends considerably on manufacturing
nonuniformities and residual facet reflections so that high yields for
one specific frequency may be a problem. These subtleties of grating
design are the key to obtaining optimum stable performance under
modulation, especially for high-speed-communication applications,
and are discussed more fully in Chapter 6.

1.73 More complex grating-based lasers

Splitting or branching the contacts allows the laser designer an extra
degree of freedom to improve the functionality of the laser. There is an
immense variety available and only a brief review is possible. Figure
1.22 shows simplified schematic side views of lasers which can give very
low dynamic chirp (i.e. changes of frequency with changes of
modulating current). In Figures 1.22a and b the total drive current to
the laser is kept almost constant under modulation, but one end is
 The semiconductor-diode laser 29

i'' h
IJUUMJ.JUUU
h
gam sections gain section pnase section gain section

X
mmmmmm
a can be used for low chirp
b and a for tunable lasers
c integrated modulator
gain sections

Figure 1.22 Multisection Bragg lasers

driven harder than the other, and under modulation the drives to the
ends are reversed, this being known as 'push-pull' operation [55] (see
also Chapter 8). The total gain and electron density in the laser remain
constant, but optical power is switched from end to end (Section 8.3).
Linewidths of less than 0.01 nm spread in wavelength are predicted to
be possible at 10 Gbit/s with these lasers. Structures such as that shown
in Figure 1.226, with a plain gain region in the middle, can be used to
give a wide range of tuning for the laser [56,57].
'DFB' lasers with more complex grating designs have been designed
to give wide-range tunability though not allowing high-speed modula-
tion [58,59]. Here the lasers are designed with an additional
periodicity of several wavelengths superimposed on the A/2 Bragg
wavelength. This then allows a different periodic reflection-spectrum
from each end of the laser, each of which can be shifted slightly by the
small refractive-index reduction caused by carrier injection. 'Vernier'
action, discussed in Section 8.4.3, then occurs so that the laser operates
only at a wavelength where reflection peaks from both of the gratings
coincide. This has allowed quasicontinuous tuning over greater than
100 nm centred on 1.5 fxm. Vernier action is also possible where the
waveguide for the laser is branched into a Y with the favoured mode
having a common peak amplification/reflection from each arm of the
Y (see Section 8.4.4) [60].
For applications which demand narrow stable linewidths, structures
such as that shown in Figure 1.22c are sometimes used [61,62]. The
main part of the laser is run continuously, and hence has a narrow
spectrum, and the modulator section simply switches the output on
and off. In practice, the modulator interacts with the master laser, and
some excess chirp results, but designs are improving, and many more
multisection, multicontact lasers with increased functionality may be
expected in the future.
30 Distributed feedback semiconductor lasers

Bragg contact
reflectors f

top diameter - 10 \i m
height ~ 20 |i m

Figure 1.23 Verticals-cavity surface-emitting laser (VCSEL)

Single devices and arrays

Another fascinating laser which uses Bragg gratings is the vertical

cavity surface-emitting laser or VCSEL (pronounced vicsel) shown
schematically in Figure 1.23 [63,64] and discussed further in Chapter
8. The 'length' of the laser is now only a few microns, and therefore
the maximum gain available is typically two orders less than that of a
conventional laser and so the mirror reflectivities typically have to be
in excess of 99%. Such a high reflectivity requires that the top mirror
has to use a multilayer stack, which is a form of Bragg-grating reflector
with either deposited multilayer dielectrics or grown semiconductor
layers of alternating high and low refractive index. These micron-size
lasers may be fabricated side by side on a single chip and, at the time
of writing, can form arrays of about 10x10 [65], but significantly
larger arrays are expected in future provided that the problems of
packaging and connections can be overcome. Such lasers offer many
exciting possibilities as sources for optical switching, routing and signal
processing, and their technology is discussed in detail in specialist texts
such as that by T.E. Sale (see Bibliography).

1.8 Summary
This chapter has set the scene for the operation and fabrication of
semiconductor lasers, emphasising why the authors see lasers with
grating reflectors as so important. The earliest lasers were examined,
showing that a key requirement is to confine the photons and
electrons to the same physical region. A geometrical factor T, known as
the confinement factor, expresses the degree to which this has been
 The semiconductor-diode laser 31

achieved. The fact that light is harder to confine closely around a given
region than it is to confine the electrons around the same region
means that F takes values typically ranging from 0.3 to 0.01, depending
in detail on the technology that is used. The simplest requirements for
lasing were put forward so that the reader understands the impor-
tance, just as in electronic feedback oscillators, of the round-trip gain
and phase in achieving a stable oscillation. Many factors influence the
net efficiency of a laser other than the facet-reflectivities (Section
1.4.1). Nonradiative recombination, waveguide absorption and scatter-
ing losses reduce the quantum efficiency for converting electron/hole
pairs to photons from the ideal of 100%. Further reductions occur
because not all charge carriers, injected at the contacts, reach the
active region. Carriers spread away from contacts, leak across blocking
layers or over heterobarriers and suffer interfacial recombination.
Typical single-facet effieciencies -10-30% dependant on structure and
processing.
The importance of single-mode lasers for modern optical commu-
nication along silica fibres over many tens of kilometres was noted. A
variety of different types of laser were outlined, and in particular it was
observed that to gain a stable single mode the favoured technology is
to incorporate some form of Bragg grating. The fundamentals of
gratings were introduced and the elements of different laser designs
with gratings buried in their structures were presented.

1.9 Bibliography

1.9.1 Semiconductor lasers

KRESSEL, H. and BUTLER, J.K.: 'Semiconductor lasers and heterojunction
LEDs' (Academic Press, New York, 1977)
CASEY, H.C. and PANISH, M.B.: 'Heterojunction lasers. Part A—Funda-
mental properties; Part B—Materials and operating characteristics'
(Academic Press, New York, 1978)
THOMPSON, G.H.B.: 'Physics of semiconductor laser devices' (Wiley,
Chichester, 1980)
HAUS, HA.: 'Waves and fields in optoelectronics' (Prentice-Hall, Englewood
Cliffs, 1984)
ADAMS, M.J., STEVENTON, A.G., DEVLIN, WJ. and HENNING, I.D.:
'Semiconductor lasers for longwavelength optical fibre communication
systems' (Peter Peregrinus, London, 1987)
BUUS, J. 'Single frequency semiconductor lasers' (SPIE Optical Engineering
Press, 1991)
PETERMANN, K.: 'Laser diode modulation and noise' (Kluwer Academic
Press, Dordrecht, 1991)
32 Distributed feedback semiconductor lasers

YARIV, A.: 'Optical electronics' (Saunders College, Philadelphia, 1991), 4th

edn.
AGRAWAL, G.R and DUTTA, N.K.: 'Semiconductor lasers' (Van Nostrand
Rheinhold, New York, 1993), 2nd edn.
OHTSU, M.: 'Highly coherent semiconductor lasers' (Artech House, Boston,
1992)
KAWAGUCHI, H.: 'Instabilities and nonlinearities in laser diodes' (Artech
House, Boston, 1994)
ZORY, P.S. (Ed.): 'Quantum well lasers' (Academic Press, Boston, 1993)
SALE, T.E.: 'Vertical cavity surface emitting lasers' (Research Studies Press,
Taunton; Wiley, New York, 1995)
COLDREN, L.A. and CORZINE, S.W.: 'Diode lasers and photonic integrated
circuits' (Wiley, 1995)
GHAFOURI SHIRAZ, H. and LO, B.S.K.: 'Distributed feedback laser diodes'
(Wiley, 1996)
MORTHIER, G. and WANKWIKEKBERGE, P.: 'Handbook of distributed
feedback laser diodes' (Artech House, 1997)

1.9.2 Optical communication systems

GO WAR, J.: 'Optical communication systems' (Prentice Hall International,
Englewood Cliffs, 1983) 2nd ed.
POWERS, J.P.: 'An introduction to fiber optic systems' (Irwin, Homewood,
1993)
GREEN, P.E.: 'Fiber optic networks' (Prentice Hall, Englewood Cliffs, 1993)
LAUDE, J.P.: 'Wavelength division multiplexing' (Prentice Hall, New York,
1993)
AGRAWAL, G.R: 'Fibre optic communication systems' (Academic Press,
Boston, 1992)
HOSS, R.J.: 'Fibre optic communications design handbook' (Prentice Hall,
Englewood Cliffs, 1990)
SENIOR, J.M.: 'Optical fibre communications' (Prentice Hall International,
Englewood Cliffs, 1988)
SANDBANK, C.P. (Ed.): 'Optical fibre communication systems' (Wiley,
Chichester, 1980)

1.10 References
1 HALL, R.N., FENNER, G.E., KINGSLEY, J.D., SOLTYS, TJ. and CARL-
SON, R.O.: 'Coherent light emission form GaAs junctions', Phys. Rev. Lett,
1962, 9, pp.366-368
2 NATHAN, M.I., DUMKE, W.P., BURNS, G., DILL, EH. and LASHER, G.:
'Stimulated emission of radiation from GaAs p-n junctions', Appl. Phys.
Lett, 1962, 1, pp. 62-64
3 QUIST, T.M., REDIKER, R.H., KEYES, R.J., KRAG, W.E., LAX, B.,
MCWHORTER, A.L. and ZEIGLER, H.J.: 'Semiconductor laser of GaAs',
Appl. Phys. Lett., 1962, 1, pp. 91-92
 The semiconductor-diode laser 33

4 HOLONYAK, N. and BEVACQUA, S.R: 'Coherent visible light emission

from GaAsP junctions', Appl. Phys. Lett, 1962, 1, pp. 82-84
5 IKEGAMI, T. and SUEMATSU, Y: 'Direct modulation of semiconductor
junction laser', Electr. Comm. Japan, 1968, 51(3), pp. 51-58
6 PAOLI, T. and RIPPER, J.E.: 'Direct modulation of semiconductor lasers',
Proc. LEE, 1970, 58, pp 1457-1465
7 KAO, K.C. and HOCKHAM, G.A.: 'Dielectric-fibre surface waveguides for
optical frequencies', Proc. IEE, 1966, 113, pp. 1151-1158
8 ADAMS, M.J.: 'An introduction to optical waveguides', (J. Wiley, Chiches-
ter, 1981); see historical introduction, pp xiii-xiv
9 CLARRICOATS, PJ.B. (Ed.): 'Progress in optical communications', IEE
reprint series 3 (Peter Peregrinus, UK, 1980)
10 SANDBANK, C.P (Ed.): 'Optical fibre communication systems' (J. Wiley,
Chichester, 1980)
11 PAL, B.P. (Ed.): 'Fundamentals of fibre optics in telecommunication and
sensor systems' (Wiley Eastern, New Delhi, 1992)
12 GREEN, RE.: 'Fiber optic networks' (Prentice-Hall, Englewood Cliffs,
1993)
13 BORN, M. and WOLF, E.: 'Principles of optics' (Pergamon Press, Oxford,
1980), 6th ed., Section 7.6.2
14 HAYASHI, L, PANISH, M.B. and FOY, P.W.: 'A low threshold room
temperature injection laser', IEEE J. Quantum Electron, 1969, 5, pp.
211-212
15 KRESSEL, H. and NELSON, N.: 'Close confinement gallium arsenide PN
junction laser with reduced optical loss at room temperature', RCA Rev.,
1969, 30, pp. 106-113
16 HAYASHI I. and PANISH, M.B.: 'GaAs G a / d ^ ^ s heterojunction injec-
tion lasers which exhibit low threshold room temperature operation',/.
Appl. Phys., 1970, 41, pp. 150-163
17 ALFEROV, ZH.L, ANDREEV, V.M., PORTNOI, E.L. and TRETYAKOV,
D.N.: 'Injection properties of n-Al^Gaj _ ^ s p-GaAs heterojunctions', Sov.
Phys. Semicond., 1969, 2, pp. 843-845 (translated from Fiz Tekh Polupovodn,
1968, 2, pp. 1016-1018)
18 HAYASHI, I., PANISH, M.B., FOY, P.W. and SUMSKI, S.: 'Junction lasers
which operate continuously at room temperature', Appl Phys. Lett, 1970,
17, pp.109-111
19 ALFEROV, ZH. I., ANDREEV, V.M., GARBUZOV, D.Z., ZHILYAEV, Y.V.,
MOROZOV, E.R, PORTNOI, E.L. and TRIOFIM, V.G.: Investigation of
the influence of the AlAs-GaAs heterostructure parameters on the laser
threshold current and the realization of continuous emission at room
temperature', Sov. Phys. Semicond., 1971, 4, pp. 1573-1576 (translated from
Fiz Tekh Polupovodn, 1970, 4, pp. 1826-29)
20 RIPPER, J.E., DYMENT, J.C., D'ASARO, L.A. and POOLE, T.L.: 'Stripe
geometry double heterostructure junction lasers: mode structure and CW
operation above room temperature', Appl. Phys. Lett, 1971, 18, pp
155-167
21 HILL, D.R.: '140 Mbit/s optical fibre field demonstration system', in
SANDBANK, C.P. (Ed.): 'Optical fibre communication systems' (J. Wiley
Chichester, 1980), Chapter 11
34 Distributed feedback semiconductor lasers

22 PAL, B.P. (Ed.): 'Fundamentals of fibre optics in telecommunication and

sensor systems' (Wiley Eastern, New Delhi, 1992), Chapters 18-20
23 AHMED, H. and SPREADBURY, P.J.: 'Analogue and digital electronics for
engineers' (Cambridge University Press, Cambridge, 1984), 2nd ed.,
Chapter 6
24 BURNHAM, R.D. and SCIFRES, D.R.: 'Etched buried heterostructure
GaAs/Ga,.vAlyAs injection lasers', AppL Phys. Lett, 1975, 27, pp. 510-512
25 LEE, T.P., BURRUS, C.A., MILLER, B.I. and LOGAN, R.A.: 'Al^Ga^s
double heterostructure rib-waveguide injection laser', IEEE J. Quantum
Electron., 1975, 11, pp. 432-435
26 HORIKAWA, H. and ISHII, A.: 'Semiconductor pump laser technology',/
Lightwave Technol., 1993, 11, pp.167-175
27 LANG, R. and KOBAYASHI, K.: 'External optical feedback effects on
semiconductor injection laser properties', IEEEJ. Quantum Electron. 1980,
16, pp. 347-355
28 HAUS, H.A.: 'Modelocking of semiconductor laser diodes', Japan. J. AppL
Phys., 1981, 20, pp. 1007-1020
29 CHEN, Y-K. and WU, M.C.: 'Monolithic colliding pulse mode-locked
semiconductor lasers', IEEEJ. Quantum Electron., 1992, 28, pp. 2176-2185
30 MARTINS FILHO, J.F. and IRONSIDE, G.N.: 'Multiple colliding pulse
mode locked operation of a semiconductor laser', AppL Phys. Lett, 1994,
65, pp. 1894-1896
31 TSANG, W.T.: 'The cleaved coupled-cavity (C3) laser' in 'Semiconductors
and semi-metals' (Academic Press, New York 1985), Vol. 22 Pt. B, Chap. 5
32 COLDREN, L.A. and KOCH, T.L.: 'Analysis and design of coupled cavity
lasers Parts 1 and 2', IEEEJ Quantum Electron., 1984, 20, pp. 659-670,
671-682
33 LANG, R.: 'Injection locking properties of a semiconductor laser', IEEEJ.
Quantum Electron., 1982, 18, pp. 976-983
34 MATTHEWS, M.R., CAMERON, K.H., WYATT, R. and DEVLIN, W.J.:
'Packaged frequency tuneable 20 kHz linewidth 1.5 |xm InGaAsP external
cavity laser', Electron. Lett, 1985, 21, pp. 113-115
35 MEARS, R., JAUNCEY, J. and PAYNE, D.: 'Low-noise erbium doped fibre
amplifier operating at 1.54 |xm\ Electron. Lett, 1987, 23, pp. 1207-1208
36 POWERS, J.P.: 'An introduction to fibre optic systems' (Irwin, Home-
wood, 1993), chapter 7
37 KAO, C.K.: 'Optical fibre' (Institution of Electrical Engineers, London,
1988)
38 LAUDEJ.P.: 'Wavelength division multiplexing' (Prentice Hall, New York,
1993)
39 GOWAR, J.: 'Optical communication systems' (Prentice Hall Inter-
national, Englewood Cliffs, 1993), 2nd ed., see chapter 26
40 BRAGG, W.H. and BRAGG, W.L.: 'X-rays and crystal structure' (G.Bell,
London, 1915)
41 RAMO, S., WHINNERY, J.R. and VAN DUZER, T: 'Fields and waves in
communication electronics' (John Wiley 8c Sons, 1994), 3rd ed., p. 310
and pp. 637-676
 The semiconductor-diode laser 35

42 STREIFER, W., SCIFRES, D.R. and BURNHAM, R.D.: 'Coupling coeffi-

cients for distributed feedback single- and double-heterostructure diode
lasers', IEEEJ. Quantum Electron., 1975, 11, pp. 867-873
43 STREIFER, W., SCIFRES, D.R. and BURNHAM, R.D.: 'Analysis of grating-
coupled radiation in GaAs:GaAlAs lasers and waveguides', IEEE J.
Quantum Electron., 1976, 12, pp. 422-428
44 STREIFER, W., BURNHAM, R.D. and SCIFRES, D.R.: 'Analysis of grating-
coupled radiation in GaAs.GaAlAs lasers and waveguides-II: Blazing
effects', IEEEJ. Quantum Electron., 1976, 12, pp. 494-499
45 STREIFER, W., SCIFRES, D.R. and BURNHAM, R.D.: 'TM-mode coupling
coefficients in guided-wave distributed feedback lasers', IEEEJ. Quantum
Electron., 1976, 12, pp. 74-78
46 KOGELNIK, H. and SHANK, C.V.: 'Coupled-wave theory of distributed
feedback lasers',/. Appl Phys., 1972, 43, pp. 2327-2335
47 NAKAMURA, M., YEN, H.W., YARIV, A., GARMIRE, E., SOMEKH, S. and
GARVIN, H.L.: 'Laser oscillation in epitaxial GaAs waveguides with
corrugation feedback', Appl Phys. Lett, 1973, 23, pp. 224-225
48 SCIFRES, D.R., BURNHAM, R.D. and STREIFER, W.: 'Distributed
feedback single heterojunction diode laser', Appl. Phys. Lett., 1974, 25, pp.
203-204
49 CASEY, H.C., SOMEKH, S. and ILEGEMS, M.: 'Room temperature
operation of low-threshold separate confinement distributed feedback
diode lasers', Appl. Phys. Lett, 1975, 27, pp. 487-489
50 UEMATSU, Y, OKUDA, H. and KINOSHITA, J.: 'Room-temperature cw
operation of 1.3-jxm distributed-feedback GalnAsP/InP lasers', Electron.
Lett, 1982, 18, pp. 857-858
51 UTAKA, K, AKIBA, S., SAKAI, K and MATSUSHIMA, Y: 'Room-
temperature cw operation of distributed-feedback buried heterostructure
InGaAsP-InP lasers emitting at 1.57 jxm,' Electron. Lett, 1981, 17, pp.
961-963
52 WHITEAWAY, J.E.A., GARRETT, B., THOMPSON, G.H.B., COLLAR, A.J.,
ARMISTEAD, CJ. and FICE, M.J.: 'The static and dynamic characteristics
of single and multiple phase-shifted DFB laser structures', IEEEJ. Quantum
Electron., 1992, 28, pp. 1227-1293
53 GHAFOURI SHIRAZ, H. and LO, B.S.K.: 'Structural dependence of three
phase shift distributed feedback semiconductor laser diodes at threshold
using the transfer matrix method (TMM)\ Semicond. Sci TechnoL, 1994, 9,
pp. 1126-1132
54 HENRY, C.H.: 'Performance of distributed feedback lasers designed to
favour the energy gap mode', IEEE J. Quantum Electron., 1985, 21, pp.
1913-1918
55 NOWELL, M.C., CARROLL, J.E., PLUMB, R.G.S., MARCENAC, D.D.,
ROBERTSON, M.J., WICKES, H. and ZHANG, L.M.: 'Low chirp and
enhanced resonant frequency by direct push-pull modulation of DFB
lasers', IEEEJ Select Topics Quantum Electron., 1995, 1, pp. 433-441
56 TOHYAMA, M., ONOMURA, M., FUNEMIZU, M. and SUZUKI, N.:
'Wavelength tuning mechanism in three electrode DFB lasers', IEEE
Photonics TechnoL Lett, 1993, 5, pp. 616-618
36 Distributed feedback semiconductor lasers

57 ISHII, H., KANO, R, TOHMORI, Y, KONDO, Y, TAMAMURA, T. and

YOSHIKUNI, Y 'Broad range (34 nm) quasi continuous wavelength
tuning in superstructure DBR lasers', Electron. Lett., 1994, 30, pp.
1154-1155
58 JAYARAMAN, V., MATHUR, A., COLDREN, L.A. and DAPKUS, P.D.:
'Extended tuning range in sampled grating DBR lasers', IEEE Photonics
Technol Lett, 1993, 5, pp. 489-491
59 OUGIER, C , TALNEAU, A., DELORME, R, SLEMPKES, S. and
MATHOORASING, D.: 'High number of wavelength channels demon-
strated by a widely tunable sampled grating DBR laser', IEE Proc.
Optoelectronics, 1996, 143, pp. 77-80
60 HILDEBRAND, O., SCHILLING, M., BAUMS, D., IDLER, W., DUTTING,
K., LAUBE, G. and WUNSTEL, K., 'The Y-laser—a multifunctional device
for optical communication-systems and switching-networks', Lightwave
Technol, 1993, 11, pp. 2066-2075
61 FELLS, J.A.J., GIBBON, M.A., THOMPSON, G.H.B., WHITE, I.H.,
PENTY, R.V., WRIGHT, A.P., SAUNDERS, R.A., ARMISTEAD, C.J., and
KIMBER, E.M.: 'Chirp and system performance of integrated laser
modulators', IEEE Photonics Technol Lett, 1995, 7, pp. 1279-1281
62 DORGEUILLE, F. and DEVAUX, R: 'On the transmission performances
and the chirp parameter of a multiple-quantum-well electroabsorption
modulator', IEEE]. Quantum Electron., 1994, 30, pp. 2565-2572
63 JEWELL, J.L., HARBISON, J.P., SCHERER, A., LEE, YH. and FLOREZ,
L.T. 'Vertical-cavity surface-emitting lasers—design, growth, fabrication,
characterization', IEEE]. Quantum Electron., 1991, 27, pp. 1332-1346
64 GEELS, R.S., THIBEAULT, B.J., CORZINE, S.W., SCOTT, J.W. and
COLDREN, L.A.: 'Design and characterization of In 02 Ga 08 As MQW
vertical-cavity surface-emitting lasers', IEEE], Quantum Electron., 1993, 29,
pp. 2977-2987
65 ZEEB E., MOLLER B., REINER G., RIES M., HACKBARTH T. and
EBELING K.J.: 'Planar proton implanted VCSELs and fiber coupled 2-D
VCSEL arrays', IEEE ] Select. Topics Quantum Electron., 1995, 1, pp.
616-623
 Chapter 2
Gain, loss and spontaneous emission

2.1 Introduction
The operation of semiconductor lasers is strongly affected by the
materials from which they are made. Spontaneous emission which
initiates lasing, optical gain which is essential to achieve lasing, and
other processes involved in lasing all use quantum processes at the
level of single atoms and electrons within the lasing material. Indeed,
without simplifications, the physics and mathematics necessary to
describe such atomic systems fully is too complicated, certainly for the
level of this book. The simplification and approximations must,
however, be done in a way which is adapted to the requirements of
semiconductor lasers, and the limitations must be understood. The
results and implications of quantum physics are discussed here but
there are only illustrative outlines of any derivations. Greater depth
may be obtained from specialist material [1,2] (see Bibliography).
In the context of lasers, the atomic behaviour of semiconductors can
be summarised under three headings:
(i) the distribution of energy states;
(ii) the occupation probability of those states; and
(iii) transitions between states.
The first few sections of this chapter will cover these headings at the
atomic level, while the later sections will apply them at a macroscopic
level to optical gain, spontaneous emission and noise.

2.2 Electronic processes in semiconductors

2.2.1 Energy states
In a semiconductor, the electrons which are of interest in the
conduction processes are not normally bound to any single atom, but
38 Distributed feedback semiconductor lasers

energy (%) energy fe)

momentum (p)
a
Figure 2.1 Energy-momentum diagrams
(i) In free space
(ii) In a periodic lattice of period a with p=hk/27T and k=27r/\dmron
may move around in the crystal. Simple Newtonian mechanics forms a
starting point where the kinetic energy of a particle % - %c is measured
from the conduction band energy edge %n and given in terms of its
momentum p and effective mass mc from

(2.1)

Such a free particle also has wave characteristics with a wavelength

given by De Broglie's relationship Ael£(.lron=h/p. Consider this particle/
wave moving through a one-dimensional periodic array of atoms
spaced a apart giving a periodic electric potential of period a forming
an approximation to a crystalline material. As far as diffraction is
concerned, the electron waves in this periodic electric potential
behave analogously to electromagnetic waves in a periodic refractive
index as outlined in Section 1.6. Waves, with frequencies having a
corresponding wavelength Aelectron=2a/N (N = l,2,3), are strongly
reflected so that, around these critical frequencies (electron energies),
there is a range of frequencies where propagation cannot occur
(similar to the concept of a stop band that is met in DFB lasers in
Chapters 1 and 5). In crystals, it is convenient to use the wave vector
\k\^2rr/\declron and to derive an energy-momentum, or %-k diagram,
which has discontinuities or jumps in the energy whenever
\k\=2ir/AeU>ctron=N7T/a (Figure 2.1). The principal disallowed energy
range for conduction and valence electrons forms the band gap of the
semiconductor with the allowed energy ranges on either side of this
band gap known as the conduction and valencebands. A complementary
 Gain, loss and spontaneous emission 39

approach starts from the discrete energy levels of the hydrogen-like

atom (for example the Bohr model) and then considers large
numbers of atoms interacting to broaden the discrete energy levels
into bands [3,4].
An analysis is straightforward for a one-dimensional array of atoms
but becomes extremely complex [5] when modelling the %-k diagram
for a real three-dimensional crystal. In many cases, an adequate
approximation for bulk material is to assume that the crystal is
spherically symmetric, and to retain the one-dimensional parabolic
relationship used above but now in all three dimensions so that
% •~%(.= \hk\2/(2m(). Pauli's exclusion principle then requires that no
two electrons (or holes) may have exactly the same energy and this
principle, combined with the fact that momentum and energy are
quantised, determines an expression for density of states (per unit
volume per unit energy) in the conduction band [6]

(2.2)

A real crystal is not spherically symmetric: the energy bands and

effective masses can vary by ~50% as the direction of momentum
changes modifying the values in eqn. 2.1 by a similar amount, but such
variations are found not to be important for lasers using unstrained
bulk active material. With lasers employing composite-layered materi-
als—quantum-well material—then the departure from spherical
symmetry combined with any anisotropic straining of the active
quantum-well regions (Section 2.7) has significant, and often bene-
ficial, effects on the energy bands and density of states [7,8].
The arguments used above for electrons in a crystalline semi-
conductor apply similarly to holes in the same material, except that a
hole may be regarded as a wave packet characteristic of one missing
electron in an otherwise full ensemble of electrons within the valence
band and so the hole has a positive charge. This absence of an electron
equivalently gives to the hole a negative mass in eqn. 2.1 (mr—• — mt)
and holes therefore appear as downward parabolas on the %-k
diagram, inverting Figure 2.1 with energy now measured down from
the valence-band edge as (%v-%). Because the movement of holes
really means movement of many electrons within an almost filled
band, it is intuitively reasonable that the associated %-k diagrams
should be more complicated than for electrons in a sparsely filled
band and that holes should be more difficult to move, or heavier than
40 Distributed feedback semiconductor lasers

heavy electron
conduction normal electron
bands
valence _
bands _»#T jL ^ • • L ^ ^T~ h e a v Y h o l e
light hole
hole split off
Figure 2.2 %-k (energy-momentum) diagram for a direct-band-gap semiconductor
Lowest energy in conduction band is directly above highest energy in
valence band. Hole-split-off band is caused by coupling of electron spins
and orbital angular momentum

electrons. A rigorous discussion is outside the scope of this book but

the sketch (for one-dimensional motion) in Figure 2.2 shows the
principal energy bands for a semiconductor, including the light and
heavy holes and bands coupled through the effects of electron spin.
It is common to show occupation of a band as being the shaded part
of the 'normal-electron' band which is convenient, because the density
of states varies with (%v - %)1/2, or (%-%()1/2 a I k\, and the shaded area
is therefore proportional to the number of electrons present. However,
this shading can be misleading, because electrons in the band can only
exist on the outer %-k line. The equally spaced dots in the k direction are
a schematic representation of energy states in the heavy-hole band and
should be thought of as similarly existing on all the other bands.

2.2.2 Occupation probabilities

The large numbers of states indicated in Figure 2.2 for the conduction
and valence bands may or may not be occupied by electrons/holes
depending on the thermal excitation and the level of the energy. The
probability of occupation for an electron state is given by the Fermi-Dirac
function
FN(%) = l/{l+exp (%-%fN)/kT} (2.3)
where %fN is referred to as the 'quasi-Fermi' level appropriate to
determine the density N of electrons. In equilibrium with no applied
voltage, %fNis the Fermi level for all the electrons, but when a forward
 Gain, loss and spontaneous emission 41

voltage is applied to a p-n junction the probability of occupation in the

conduction band increases (i.e. %fN increases) so that, for lasing,
typically %JN>%C, the energy of the conduction-band edge.
Holes are absences of electrons in a state, with densities denoted by
P, and at a given energy FP = l—FN gives the probability of a 'hole'
occupying that state but it is conventional to denote hole energies in
the reverse direction to electrons, so for holes
FP(%) = l/{l+exp(%fP-%^/kT) (2.4)
The energy %jF, like %fN, is a quasi-Fermi level determining the
occupation probability for holes. Increasing the applied voltage
decreases this quasi-Fermi level so that, for strong lasing, %jp^%xn the
energy of the valence band edge. The really important condition for
lasing is that the net applied voltage Vj to the junction given from
qVj = %fN-%fP is greater than the band gap [9], i.e.: Vj>(%c-%t)/q.
Because the positions of the quasi-Fermi levels are then inside or near
to the band edges, it is necessary to use the full Fermi-Dirac statistics
(rather than the Boltzmann approximation as in elementary texts
[10]).

2.23 Radiative recombination and absorption

If an electron moves from one state to another of lower energy,
conservation of energy requires an emission of the energy difference:
either radiatively involving a photon, or nonradiatively in which case
phonons (which are the particles associated with lattice vibrations
having quantised energies) are emitted. In optical devices it is
desirable and usual for radiative transitions to dominate over
nonradiative ones. A key factor is that, although the optical photons
which are of interest in diode lasers have energies (-1-2 eV) which are
much larger than the energies of phonons (-1-50 meV), the photon
momentum is typically less than 1% of the average phonon momen-
tum. On the scale of the %-k diagram, the photon momentum is
negligible when compared with that of phonons.
In silicon, or other indirect materials, the conduction and valence
bands have their extrema centred on very different momenta in the
energy momentum diagram (Figure 2.3). In moving an electron from
the conduction band to the valence band, a radiative recombination
would need to involve a phonon with momentum but little energy and
a photon with energy but little momentum so as to conserve both
momentum and energy. The relatively low density of phonons
combined with the complex interaction means that such radiative
42 Distributed feedback semiconductor lasers

conduction
band

energy

momentum
Figure 2.3 Transitions in indirect band-gap material
For an electron to move from the conduction band to the valence band
requires momentum and energy to be conserved. Only a photon has
enough energy to balance the energy loss and only the phonon has
enough momentum to balance the momentum change

conduction
conduction
band band

energy

photon momentum
momentum
h/X <« h/a
m emission absorption

Figure 2.4 Transitions in direct-band-gap material

The photon wavelength in the medium and frequency are \m and v,
respectively; a is the lattice periodicity with h/a giving a rough order of a
phonon's momentum in the crystal

indirect transitions are generally improbable. On the other hand, in

direct-gap materials the direct radiative transitions tend to dominate
over other more complex transitions, and even in materials such as
Ga05Al05As, which are only 'slightly indirect' light is emitted quite
well.
 Gain, loss and spontaneous emission 43

Figure 2.4 shows schematically the transition processes in direct-gap

material of interest for lasing action and indicates that optical
absorption is the reverse of radiative recombination with energy and
momentum balance applied appropriately. The rate of emission of
photons (dS/dt) emssion through recombination of electrons and holes
and the rate (dS/dt)absorptwn of absorption of photons generating
electron hole pairs has to be closely linked as discussed next.

2.2.4 Transitions and transition rates

To calculate the transition probabilities of electrons, there are several
factors to be considered. In a specialist study, the start might be the
basic quantum theory of transitions between N energy levels. This
requires the quantum formalism of an N x N complex density matrix
M, where the probability of a spontaneous transition from state i to
state jis proportional to IM^I2 with the diagonal elements IMnwl2 giving
the probability of occupation of state n. In practice, the rates of
emission and absorption are primarily discovered from measurements
on materials, but a reader might start by referring to a model due to
Kane [11] for the quantum theory which indicates that an approx-
imate transition probability between the conduction and valence
bands is very roughly proportional to the bandgap energy %g and the
electronic rest mass m0:

Here mc is the effective mass in the conduction band and Ss is the

splitting energy caused by the electrons' spin. For GaAs, the constant
of proportionality £is about 1.3, and other optical semiconductors are
of the same order. An important conclusion of Kane's theory for laser-
modelling purposes is that transition probabilities (and therefore
rates) increase with the bandgap as a first-order effect, as indicated in
eqn. 2.5.
The problem of the transitions which is to be solved can be
expressed graphically as in Figure 2.5. Here, because of the negligible
photon momentum on the scale of this diagram, radiative transitions
between bands are indicated where the electrons and holes have equal
ft, which is referred to as the strict ft-selection rule and is a useful
approximation here. Total spontaneous emission etc. is determined by
integrating for occupied states over all energies, discussed later in
44 Distributed feedback semiconductor lasers

Figure 2.5 Transitions between energy levels in the conduction and valence bands
As A^ changes so \k\ varies. Over a band of energy d%, there is a
corresponding band or range d& in I k\

Section 2.3.2. As an introduction, a simpler two-level system is

considered in Section 2.3.1
Integrating over all k as in Figure 2.5 takes into account the bands of
energy states but does not take into account impurities or defects
where the electron states are localised spatially, typically over subnano-
metre distances Ax From Heisenberg's uncertainty principle,
uncertainty7 in momentum Ap and uncertainty in position Ax are
linked by LpAx—h/2-77, so that the momentum (wave vector) becomes
ill-defined for localised states, and electrons in such states can interact
with particles over a wide range of k: the ft-selection rule becomes
relaxed, often giving better agreement between the theory of
transitions and observed results [12-14].
Localised impurities can also contribute to so-called band-tail states
just inside the band edges, and move lasing to longer wavelengths,
particularly at low temperatures. Impurity atoms and defects with
states within the bandgap also encourage recombination in multiple
steps; a high proportion of these will involve phonons and no optical
radiation, or optical radiation at lower frequencies than those of
interest. Consequently, almost all recombination via bandgap states
represents loss of charge carriers without useful radiation. Gold and
copper result in such bandgap states and are carefully excluded from
laser active regions during processing.

2.2.5 Auger recombination

One special set of transitions known as Auger recombination, unlike
recombination via bandgap impurity states, is always of concern and
 Gain, loss and spontaneous emission 45

CCCH CHHS CHHL CCCH Ph

Figure 2.6 Different Auger-recombination processes

C = conduction band, H = heavy-hole band, L = light-hole band, S = hole-
spin-split-off band, Ph = phonon
Three charge carriers are involved in each process so that3 Auger-
recombination rates are proportional to (charge-carrier density)

this concern increases as the bandgap decreases [15,16]. Auger

recombination is not a single process, but a group of different ones:
depending on definitions, over 80 such processes may be identified. A
key feature is that at least three particles (two electrons and one hole
or two holes and one electron) and four energy states will be involved,
but transitions may be band-to-band, may involve phonon assistance,
or may involve impurities. A full analysis of all Auger processes in a
semiconductor is difficult and only illustrative examples are given
below.
Individual Auger processes are identified in Figure 2.6 by listing the
bands in which states take part, using the codes C for conduction
band, H for heavy-hole band, L for light-hole band and S for hole-spin-
split-off band. The first diagram in Figure 2.6 shows an electron in state
a recombining with a heavy hole at b, and both the energy and momentum
are taken up by an electron which moves from state c to state d. It is
readily seen that even now the process is not a single one, but might
involve numerous combinations of energy and momentum. This first
CCCH process is a likely one for n-doped semiconductor, because it
needs two electrons; the next two examples require two holes, and are
therefore more likely in j^-doped material. The final diagram shows an
extension to the first three processes whereby a phonon is involved as
well as carriers and a photon.
In theory the rates of all the separate Auger processes could be
calculated, but the following summary of measured results provides a
good basis for modelling:
46 Distributed feedback semiconductor lasers

(i) Since Auger processes involve three particles, they increase with
the cube of carrier density.
(ii) All Auger rates increase with decreasing bandgap because, in
general, the range of likely transitions increases with decreasing
bandgap.
(iii) Processes such as CCCH, where an increase in temperature
increases the numbers of electrons available to participate (as in
practical lasers), will have rates which increase with temperature.
(iv) Since numbers of dopant atoms or defects are almost constant
with temperature, Auger processes involving these increase less
rapidly with temperature than the processes of (iii) above.
As a direct consequence of (ii) above, it can be seen that 1.55 jmm,
1.3 juim and 0.9 jxm lasers are steadily less affected by Auger recombi-
nation as the wavelengths decrease, as indicated by the sensitivity of
the threshold currents in lasers to temperature with Auger processes
typically becoming significant at around 50°C for 1.5 |mm devices, 70°C
at 1.3 |xm, and 120°C + for devices operating at 0.9 |j,m or less.

2.3 Absorption, emission rates and spectra

23A Absorption, stimulated and spontaneous emission in a
semiconductor
Spontaneous emission, absorption and gain in independent atoms (as,
for example, in gas lasers [17]) are more straightforward to discuss
than in semiconductors where the high occupation density of
electrons and holes in semiconductor lasers means that there are
difficulties of using Fermi—Dirac statistics and the simplifications which
arise in gas lasers do not apply. However, Fermi-Dirac statistics may be
applied in a straightforward manner using a two-level model which
illustrates the basic principles [18] and aids understanding of laser
models.
Figure 2.7 represents an idealised unit volume of material with N}
states at energy %x of which iVare occupied by electrons, and N2 states
at energy %2 of which N2 are occupied. Since it is a unit volume, the
number of electrons at the upper level is also the electron density N
and similarly the hole density P = N2 — N2. Also within the volume there
are S photons, which can be absorbed or stimulate further photons.
The first process to consider is spontaneous emission where
electrons at a higher energy spontaneously fall to a lower energy giving
out photons at the energy difference. The spontaneous power-output/
volume is proportional to the number of high-energy electrons iVand
 Gain, loss and spontaneous emission 47

N electrons vacant sites

N 1 states - Q . ^ ^ . ^

dS /dt
SPont

N 2 states MOOjiOMIQIQMOO energy

n g electrons p holes

Figure 2.7 Idealised two-energy level photonic/electronic interactions

the number of vacant states (holes, density P) available for sponta-

neous recombination; hence:
^ * = KspNP= Nref/Tsp (2.6)
where Ksp is a constant of proportionality and the recombination time
Tsp is a parameter often measured experimentally. In a typical laser with
1 8 3 9
0
The second process is stimulated emission where again electrons at
a higher energy fall into vacant states at a lower energy but in this case
are stimulated to make the transition by the photons already present.
The rate of stimulated power within the unit volume is therefore again
proportional to the same numbers of holes and electrons as above, but
also now proportional to the photon number S, and with a different
constant of proportionality K.iim and hence one obtains
(dSs(m/dt) mimwn=KstmSNP (2.7)
However, this stimulated photon rate is offset by absorption, which is
proportional to the number of holes at the upper energy (Nx — N) and
the number of electrons at the lower energy N2=(N2 — P) and also
proportional to S, the photon density, so that
p 1 - A/) (N2 - P) (2.8)
The net rate of emission is the sum of these three processes above:
(2.9)
To obtain the net rate of generation of photons being coupled into a
lasing mode, all of the stimulated photons S at the lasing frequency
above contribute, but spontaneous photons are emitted almost
48 Distributed feedback semiconductor lasers

isotropically and also over a much wider band of frequencies than the
lasing linewidth. Consequently only a very small proportion f3sp
(typically 10~6 to 10~2) couple to the lasing mode and eqn. 2.9 must be
modified to read
dS/dt=KstimS(NN^PNl - NXN2) +(3spKspNP (2.10)
Under lasing conditions when injected charge dominates over doping,
N^P, and P may be eliminated, allowing a photon-interaction
equation often used in modelling:
*= G'n(N- Nh) S+p^BN2 (2.11)
One may link the parameters in eqn. 2.11 with those in eqn. 2.10 as
follows:
p (2.12)
where Gm is the differential power gain/unit time, and Nlr is the
transparency value of electron density at which the stimulated optical
gain is cancelled by the stimulated optical absorption—the material is
transparent for the light at that frequency.
A nonzero acceptor density Na in the gain region causing P=N+Na
from electrical neutrality could, from the theory, reduce the transpar-
ency electron density. However, a high density of acceptors increases the
nonradiative recombination and in practice this technique is not
normally used deliberately, though ^dopants such as Zn may diffuse
from the />-cladding layer during growth. Here doping in the gain
region is ignored.
From eqns. 2.7 and 2.8, if gain is to exceed absorption, then one
requires
NP~ (Nj - N) (N2 - P) =NP{1 - (1 - Nx/N) (1 - N2/P)}>0 (2.13)
But iVand Pare related to Nx and N2 via the Fermi relationships of
eqns. 2.3 and 2.4:
(2.14)
N2/P={l+exp(%jp-%2)/kT} = 1/FP(%2) (2.15)
and so, for net gain (substituting the above into eqn. 2.13)
k i
t { ( « « ) ( ^ V l / ^ ( ^ ^ ^
Eqn. 2.16 confirms the earlier result that the voltage Vj applied to the
laser junction, given from the separation in quasi-Fermi levels, must be
greater than the equivalent bandgap voltage to obtain lasing action [9].
A further implication is that, if the contact resistances in the laser
 Gain, loss and spontaneous emission 49

structure (see Section 5.7) could be made very low, the carrier density
in the active region would be determined solely by the applied voltage
and the optical gain could then be regarded as a direct function of
applied voltage, rather than the electron density.
Now in thermal equilibrium, the net photon-emission rate dS/dt is
zero with no voltage applied across the junction (i.e. %JN-%J^). Also,
select the situation where spontaneous and stimulated emission cover
the same frequency ranges and angles so that the coupling factor
/3sp=l. These conditions force the system to be in equilibrium at a
temperature T with black-body thermal radiation described by
Planck's formula [19] for the photon density over a range of
frequencies, say kspf, around an optical frequency/:
S=87rfn\ Aspf/c3{cxp(hf/kT) ~ 1} (2.17)
Here n is the refractive index and ngis the group index, i.e. c/ng=vgis
the group velocity or the velocity at which the energy propagates while
c/n=vp is the phase velocity. In equilibrium, dS/dt=0, and eqn. 2.10
becomes
0=KsltmSNP- K^SNP^/N- 1) (N 2 /P- 1) + KspNP (2.18)
Also in equilibrium %jN=%jTso that eqns. 2.14 and 2.15 yield
S=(Ksp/Kslim)/{exp(%i -%2)/kT- 1) (2.19)
In the simple two-level system, hf=%x —%2 so comparing eqns. 2.17 and
2.19:
(Ksp/Kslm) =8irfn\ bspf/c5 (2.20)
A relationship having been established between spontaneous and
stimulated emission coefficients, these may now be related to the
differential of the (gain/unit time) with electron density given from
Gn. From eqn. 2.12, Gm=Kstim{H^H2)) and writing (N^N*,) =NreJ then
using this as a reference density where N=P=Nref in eqn. 2.6 so that
Ksp= l/Nr(,jTsp, o n e obtains

G ^ f f f n g r ^ J ) (2.21)
Using the group velocity vg to link the spatial distance z travelled by the
light with the time Ogives z=tvg=ct/ngand the material's optical-power-
gain/absorption per unit distance may be written as:
(1/5) dS/dz=c2(iY-A^r)/(87r/2w2T,/;A,/;j0 (2.22)
showing the important link between spontaneous recombination and
stimulated gain and absorption. This is related to the quantum theory
50 Distributed feedback semiconductor lasers

of optical amplifiers as given in Appendix 9. In this book the power

gain per unit time is written as Gm(N) and the related field gain per
unit distance is written as gm(N), and because the photon density is
proportional to the square of the optical-field amplitudes,
(1/5) dS/dz=2gm(N) = Gm(N)/vg (2.23)

23.2 Stimulated-gain spectra in semiconductors

The analysis above refers to just two levels of energy when in reality
there is a band of energies in both the conduction and valence band
as in Figure 2.5 where one is dealing with pairs of electrons and holes
just inside the conduction band and valence band, respectively. Using
the notation of Figure 2.5, and assuming parabolic %/k relationships
with ^-selection of the electron/hole interactions with negligible
photon momentum:

(2.24)
where \/mr=\/mc+\/mv gives a combined effective mass ('reduced
mass') from the effective masses in the conduction and valence bands.
Now for a spread of photon energies d(h/) there is a spread dk in
the magnitude of the electron momenta and a consequential
spread of their energies in the conduction band given from
d% = (f\2k/mf) dk=(mr/mr) d(h/). There then has to be a similar spread
of energies d(S&v) of holes because they have the same k in the valence
band, and so d(Mt) = (fi2k/m,) dk={mr/mv)d%.
Now consider a density of photons $ d(h/) spread over a range of
frequencies d/ around a frequency / interacting with a density of
available electrons in the energy states equivalent to Nj at an energy %x
in Section 2.3.1 but now %1=%=%(.+ 8%r and takes a spread d%. This
density of available electrons is found by first finding the density of
electron states (given by eqn. 2.2) with the range of energies d%
associated with the spectral spread d/:

Hx becomes 4TTI ^ {%~%)ind% (2.25)

Replacing %x in eqn. 2.14 with %, the occupation probability is FN(%)

giving the effective density of electrons available to aid the stimulated
 Gain, loss and spontaneous emission 51

emission as

(
-5) (%-~%c)l/2FN(%) d% (2.26)
h-/
Using a similar argument, the density of vacant sites or holes available
to receive the electrons for stimulated recombination is the equivalent
of N2FP(%2) in Section 2.3.1 where the energy %2=(%-hf) will be
below the valence-band edge. Bearing in mind that the spread of
energies (mc/mt) d% in the valence band corresponds to the spread d^
in the conduction band for the same ^-values, then
N2FP(%2) becomes
3/2
/9 \
47r( - J (%-%+hf)l/2FPCZ-hf)(mc/mt) &% (2.27)

Now h/along with the spread d(h/) will be held constant but the whole
range of % can be permitted, consistent with %^%(. and % —hf^%v,
i.e.
«p (2.28)
Then on summing or integrating over the whole energy range of
permitted %, the equivalent stimulated emission rate given from eqn. 2.7
now becomes
(dS'/d 0 slim misswn d (h/)
=AslmS'{j(% -%)x/2{%v~~%^f)l/2FP{%--hf)FN{%) &%} d(h/) (2.29)

(wr/ww) with A;ft-OT as in eqn. 2.7 and

the range of the integration is limited by eqn. 2.28.

Still following the methodology of Section 2.3.1, it may be seen that
the equivalent expression for stimulated absorption is the same expres-
sion as for the stimulated emission except that the occupation
probabilities FN and FP of electrons and holes, respectively, are replaced
by

(2.30)
giving the result that
52 Distributed feedback semiconductor lasers

10 _

relative

Figure 2.8 Optical gain against frequency

Simplified calcula ' >n of the approximate shape of the gain spectrum,
with ^-selection and no band tails

(l-FN)(l-FP)=FNFPexP{(hf-qVl)/kT} (2.31)
Hence one finds for the absorption
(dS'/dt)stimahsmpiton d(h/) =AstmS' exp{(h/- qVf)/kT}x

(2.32)
with the same range of % as in eqn. 2.28.
Now the gain spectrum is given from the net stimulated (emis-
sion — absorption) so that approximating the Fermi-Dirac distribu-
tions to unity over the limited range of energies, one can determine
the gain spectrum as approximately
(dS'/dz)/S'=2gtn*(ng/c)Astim x
{l-expj(h/-^)/kr}}[J(«-«,) l / 2 {« r --« + (h/-%)} 1/2 d < g] (2.33)
The integral may be evaluated by putting y2=(%-%c) to find a rough
but explicit expression for the gain spectrum with ^-selection as a
function of the junction voltage:
gm(hf, F/) = (7rn/ ( ,y8c)(h/-^) 2 [l-exp{(h/- 9 V,)/kT}] (2.34)
Figure 2.8 sketches this dependence of gain on the optical frequency
and the junction voltage. As the junction voltage increases, so the peak
optical gain increases in frequency. One can also see that as kT
 Gain, loss and spontaneous emission 53

gain, cm
200
1.8 x 10 1 8 cm' 3

150 -

100 -

1/1300 1/1280 1/1260

frequency (in reciprocal wavelength),nm

Figure 2.9 Estimates of gain against frequency (parameter is carrier density)

For a quaternary In0.75Ga0,25As0>55P0.45 undoped material: (parameter=car-
rier density). This is the material gain for the optical field and the net
effect in a laser would be reduced by the confinement factor. The bar at
the top indicates the spectral range corresponding to a change of photon
energy of kT at room temperature of 300 K. Data taken from Osinski and
Adams [21]. The general shape is common to most gain spectra with the
gain covering a range of frequencies A/ typically, where hA/~k7' The
theory agrees reasonably with experiment [22].

decreases so the gain bandwidth decreases. This simplified analysis

then demonstrates many key features of optical gain.
The results may be converted to gain as a function of electron
density by estimating the electron and hole density for a given junction
voltage. Gowar [20] gives an explicit example of how this can be done.
However, obtaining detailed calculations of the gain, as opposed to
having an understanding of the gain processes, really requires a more
detailed study of the ^-selection processes, band tailing and other
physics of optical transitions and is left to specialist references, as for
example the work shown in Figure 2.9. For many modelling purposes,
expressions for gain etc. rely on measurements on relevant material
rather than theory.

2.5.5 Homogeneous and inhomogeneous broadening

The actual linewidth A/M,/of a single laser mode is orders of magnitude
smaller than Av/,/ the width of the spontaneous spectrum. While the
54 Distributed feedback semiconductor lasers

main discussion concerning linewidth is left until Section 6.2, a brief

discussion is essential here. Ideally if there were only transitions
between a single pair of energy levels, as in an ideal gas or solid-state
laser where the active atoms are sparsely distributed and essentially
independent of each other then this lasing linewidth A /w /would be
very small. However, even when all the photon emissions are created
from identical energy transitions within the independent atoms or
molecules, the lasing linewidth is homogeneously broadened because the
photons have random phases and are emitted at random times. The
effect on the linewidth A m /of the spontaneous lifetime rsp in such a
laser is found through straightforward Fourier analysis to give

A, w /-l/(27rr 5/) ) (2.35)

The associated line shape is known as Lorentzian (Section 6.2).

This type of line broadening still takes place in semiconductor lasers
but the energy levels involved in the photon emissions are not all
identical so that strong inhomogeneous broadening should lead,
through a superposition of different random processes, to a Gaussian
line shape. However, the experimentally observed results give mixed
conclusions [23] but for many purposes the Lorentzian line shape is
still adequate with other linewidth-broadening processes dominating
as discussed in Sections 2.5, 4.2, 6.2 and 8.2.5

23 A Spontaneous-emission spectra from semiconductors

The close connection between spontaneous emission and stimulated
emission was clearly seen in the two-level model and this link still holds
in the more realistic theoretical models [24]. Spontaneous-emission
spectra may be measured from both FP lasers or LEDs, but the
measurement is not made at the point where the photons leave the
atom; the measured light has to pass through further material where it
may be absorbed and/or re-emitted, depending on the material gain
and the length traversed. This is very dependent on the device type
and structure, but measurements from surface-emitting LEDs give a
reasonable direct estimate of true material-spontaneous spectra while
those from FP lasers will need more careful interpretation because of
the feedback.
The link demonstrated between spontaneous emission and stimu-
lated emission means that the spontaneous-emission data can be
processed to give gain-frequency information at different electron
densities, provided that enough details are known of the laser
 Gain, loss and spontaneous emission 55
spontaneous power, dBm

-50

-60

-70

1/1650 1/1600 1/1550

frequency (in reciprocal wavelength), nm "1

Figure 2.10 Typical measured spontaneous spectra using a Fabry-Perot laser

Data taken from Morton et al [25] of the emission from a FP laser with
one AR-coated facet using 7 quantum-well laser material: top spectra at
-85% of threshold, bottom spectra at -25% of threshold. The spectral-
line structure is created by the many Fabry-Perot modes but the
envelope gives the spontaneous profile

structure. Figure 2.11 indicates processed results from the data of

Figure 2.10 based on Reference 25.

2.4 Semiconductor interactions with the lasing mode

2.4.1 Spontaneous-coupling factor
In Chapter 1, the lasing condition required a round-trip complex gain
of unity, thereby neglecting the existence of spontaneous emission.
This is unrealistic, particularly for describing a laser around threshold
where the spontaneous emission plays a key role. Spontaneous
emission is essentially random and, on the scale of the modelling, the
emission at a position x is uncorrelated with the emission at x' in space.
To a good first approximation, fe-vectors of electrons in a bulk
semiconductor are isotropically distributed and the band structure
does not vary sufficiently with direction to invalidate this discussion.
Spontaneous emission is therefore uniformly and randomly distrib-
uted in both direction and polarisation, and in any one direction is
uncorrelated with any other direction.
56 Distributed feedback semiconductor lasers

-1
gain, cm Af ~k7)h|,
50 r~ 1.6 x 10 1 8 cm' 3

-50

-100

1/1650 1/1600 1/1550

frequency (in reciprocal wavelength), nm'1
Figure 2.-11 Estimates of gain-frequency against carrier density
Material with 7 QWs of 80A width separated by 80A barriers (parameter
= carrier density). Such quantum-well material gain includes the net
effect of the confinement factor which would be the order of 0.15
depending in detail on the surrounding guide structure. (Data
compiled from P.A. Morton et al. [25].) The bar at the top indicates the
spectral range corresponding to a change of photon energy of k7' at
room temperature. Note how the gain peak changes with electron
density

A guided lasing mode may be regarded as a coupled group of rays

propagating with the same velocity component in the mode direction,
but at a range of different angles. The greater this range of angles, the
greater is the range of spontaneous emissions that will couple to the
mode. The coupling for the forward and backward directions will be
identical, giving a factor of 2 in spontaneous power when both couple
to a mode. When discussing forward and reverse waves separately, note
that the forward and reverse spontaneous components are uncorre-
lated. The spontaneous-emission components associated with any two
polarisation directions are similarly uncorrelated, but a single guided
mode will have only one polarisation, and only one resolved
component will be coupled to the mode, thereby reducing the
coupling of the total spontaneous emission by another factor of two.
Mathematically, one has to estimate the overlap integral between a
guided mode and spontaneous emission, in both the spatial
and spectral domains. Spectrally, the spontaneous emission covers a
broad range (A9/,/~10 000 GHz) but the lasing-mode linewidth
 Gain, loss and spontaneous emission 57

(a\ spontaneous
dipole solid angle for evert
radiation into free space

solid angle for spontaneous coupling Q

2 effective area A
= Q. '/«eff
FWHM emission angles Q ,0
vert ho

(b) laser power

spontaneous A / laser linewidth

emission power
spontaneous linewidth

frequency

Figure 2.12 Schematic of the principal elements forming the spontaneous emission-
coupling coefficient

A7 w /~ 0.1-100 MHz. To a first order, the spectral coupling can be

taken as the ratio of the modal linewidth to the spontaneous spectral
width. Figure 2.12 indicates these features schematically.
It is helpful to discuss orders of magnitude. The principal method of
guiding considered for lasers here is index guiding, where the internal
angles are a few degrees (see Chapter 3). A typical far field for a
buried-heterostructure laser might have 25° full-width half-maximum
(FWHM) horizontal (i.e. parallel to the junction) and 40° FWHM
vertical (i.e. perpendicular to the junction) as indicated in Table
A l l . l . The half-cone angles in the air, 20/wrand i.Qvem are reduced inside
the waveguide through Snell's law (see Figure 2.12&) where approx-
imating sin(20) ~20 the internal angles are simply reduced by the ratio
n
(iir/nejj9 where n^~3.3 (an effective or average value for the lasing
material waveguide as in Table A l l . l ) . Then the solid-angle emission
angle ft'~0.21 outside the laser is reduced inside the laser to a value
Cl—ft'/n^-0.02 steradians. Noting that the full solid angle is 4TT, the
angle-coupling factor is therefore ~1.6xlO~ 3 . Taking the mode
spectral linewidth, for example, as 0.15 nm, with the total spontaneous
bandwidth as 100 nm the spectral-coupling factor ~1.5xlO~~3 and
58 Distributed feedback semiconductor lasers

applying a polarisation factor of 0.5 gives the spontaneous-coupling

factor:
Ay;~0.5x 1.6x 1.5x l(T 3 x l(T 3 ~L2x 10~6 (2.36)
This is an appropriate estimate for travelling-wave rate equations, but
is doubled in conventional rate equations which consider both
forward and reverse optical waves together in a single rate equation.
The value of spontaneous-emission coupling can be evaluated in a
number of ways; the method given above is straightforward, but other
methods are examined and reconciled in Appendix 9 which also
examines the numerical modelling of spontaneous emission.

2.4.2 Petermann's 'K factor9

In some early laser devices, the guiding of the optical waves was
accomplished through the active gain. This is not studied in this book,
where index guiding is assumed to dominate. However with gain
guiding the propagating wave fronts have a significant curvature
allowing for much greater (up to 50 times ) acceptance angles for
spontaneous emission. The increase in the acceptance angle is known
as the Petermann transverse K factor, variously referred to as Kp or Klr
The reader is referred to Petermann's work [26, 27] for further study.
In DFB lasers, where optical fields may be strongly localised longitudi-
nally, an equivalent longitudinal K factor [28] also referred to as Kz
exists giving a total change in the spontaneous emission of Ktot=KzKlr.
However, with the numerical modelling here, the longitudinal (or z)
interactions and nonuniformities are included and so Kz is automat-
ically taken into account and should not be inserted as an addition.

2 A3 Gain saturation in semiconductors

Gain saturation may be regarded as arising from three different
mechanisms. If no current is driven into the diode, the semiconductor
laser is said to be unpumped, few conduction band states are occupied,
and most valence band states are full, so the optical absorption per
unit length is essentially limited by the strength of the interaction
(Imatrix elementl2, related to spontaneous lifetime) and the volu-
metric density of states. Even with very strong pumping or full
inversion, the maximum gain per unit length can only approach the
maximum absorption per unit length. This 'gain saturation' means
that the gain is nonlinear with electron density in the laser and this
applies to all materials, but the effect is especially noticeable in single-
 Gain, loss and spontaneous emission 59

quantum-well material where the small density of states can saturate

rapidly.
A similar but distinct cause of saturation is optically induced. In this
latter case, for a given drive current at high optical powers, the high
rate of stimulated transitions reduces the carrier density from the value
at low optical powers. This saturation is naturally taken into account
through the gain decreasing with electron density.
Finally, even if the average carrier density were maintained by extra
drive current at high optical powers, the high optical power at a
particular photon energy causes the distribution of electrons and holes
to depart from their equilibrium spread of energies within the
conduction and valence bands, and the gain is reduced at the lasing
frequency as discussed further in Section 2.4.4 below. This type of
saturation is dealt with by introducing a first-order 'gain-saturation
parameter' 6, where if the gain as a function of electron density in the
semiconductor is Gm(N)m at zero optical power, and Gm(N){S) is the
gain at an optical photon density S then
(2.37)
17 3
where £~ 1 to 3 x 10" cm is a useful guide to the accepted strength of
gain saturation with Sthen measured in numbers per cubic centimetre
to ensure that eS is dimensionless.

2.4.4 Spectral hole burning and carrier heating

The precise reasons for gain saturation at high optical powers are still
a cause for discussion [29], but two candidates for contributing to the
parameter e are mentioned briefly. The first main effect is the inability
of the charge carriers to redistribute themselves sufficiently quickly
within the conduction or valence bands so that, at high optical powers,
the electrons and holes in those particular energy levels (i.e. the
shaded regions shown in Figure 2.5) which are contributing to the
photon energy of highest intensity become depleted from their
conventional equilibrium values, i.e. there is 'spectral hole burning' in
the gain spectrum. Under normal conditions, charge carriers within a
band (i.e. at energies just above and below the shaded regions in
Figure 2.5) can redistribute themselves to help fill the spectral hole
towards a thermal-equilibrium state on the time scale of the intraband
relaxation time TinirabanA~\0~l* s. Because radiative-carrier lifetimes (or
mfer-band relaxation time) are around 10~9 s the effects of rinlmkmd are
usually negligible on this time scale. However, at sufficiently high
optical-power densities, the radiative lifetime is reduced by an order of
60 Distributed feedback semiconductor lasers

magnitude so that a 'hole' or notch in the gain occurs at the lasing

wavelength and the carrier distribution in the bands is no longer in
thermal equilibrium. The inclusion of the parameter e in eqn. 2.37 is
one method of accounting for this effect. Spectral-hole burning can
have noticeable effects on the linewidth of some DFB lasers [30].
There is a related effect which occurs when charge carriers are
injected from the contacts into the material confining the light. These
electrons are necessarily injected at higher than average energy, and so
need a little time to redistribute into the energies characteristic of the
band. This effect is particularly important in quantum-well lasers
where, as these carriers thermalise into the quantum wells, they arrive
at a higher than average temperature and such 'hot' carriers can then
transport out of a well and not contribute so effectively to the gain. All
this significantly affects the details of transport from contacts into
quantum wells [30] (see Section 6.6.2).

2 A.5 Scattering losses

Any slight inhomogeneities in the laser's material or waveguide will
cause optical scattering [31]. Perpendicular to a p-n junction, even
good epitaxial layers will have roughnesses of the order of one atom
step, and parallel to the junction the photolithographically defined
and etched sidewalls of a buried heterostr ucture mesa are likely to
have RMS roughnesses of the order of 100 nm. The refractive-index
steps, though small, can cause some light to be 'scattered' out of the
guide and in effect give attenuation as if there were absorption but,
unlike absorption, scattering does not generate heat (unless the light
is subsequently reabsorbed).
2.4.6 Free-carrier absorption
An electron within the conduction band may interact with a photon
and so gain the photon's energy, but because the range of energies in
the conduction band is wide some such electrons can stay within the
same energy band. This type of free-carrier photon absorption is
almost independent of the frequency because the higher energy levels
are empty and there is no equivalent bandgap to control the process.
As is normal for any interaction, the higher the density of electrons,
the greater the absorption, and in heavily doped (~1019 cm"3) w-type
substrates an absorption coefficient ~10 cm"1 can be measured.
The same occurs in principle with valence-band electrons, but
because such bands are nearly full of electrons the process statistically
involves more 'shuffling' of full and empty states to take place, and so
 Gain, loss and spontaneous emission 61

is less significant. When a low-optical loss substrate is required, it can

sometimes be advantageous to use /Mype material.

2.5 Henry's a factor (or linewidth enhancement factor)

Until now, optical emission and absorption have been considered in
terms of particles or energies. Advanced laser modelling requires that
one considers the wave nature of light. The net stimulated gain for a
propagating electric field of amplitude E^z) has to be given from:
Ej(z) =E0 exp( -jfiz) exp(gz) (2.38)
There has to be both propagation with a phase change as well as gain.
For a plane wave in a uniform (nonmagnetic) material with gain, there
are real and imaginary parts of a complex relative permittivity linked
to the real and imaginary parts of a complex refractive index:
l/2
(2.39)
If JE0 is regarded as the input and Ej(z) the output of a linear system, it
can be shown (Appendix 7) that the 'gain response' is uniquely
defined by the 'phase response' so that /3 and g are linked and
consequently the real part of the complex refractive index nr is linked
to the imaginary part of the refractive index n{. The processes of
calculating the real response from the imaginary, and vice versa, are
known as Hilbert transforms or Kramers-Kronig transforms. These
relationships apply strictly to any linear causal system where there is an
input which causes an output, but they may be extended to any
piecewise-linear or continuous system. In any case, departures from
linearity in semiconductor properties are not sufficiently great to
invalidate the application of the Kramers-Kronig relationships for the
work discussed here. The term refractive index, when used by itself,
will always mean the real part of the complex refractive index but,
when there could be some doubt, the notation nr will be used to
indicate the real part with n{ the imaginary part corresponding to
gain.
Implicit in the above is that any change in gain with frequency must
be associated with a refractive-index change in frequency and vice
versa. However, a change in gain at a particular frequency does not
necessarily result in a refractive-index change at the same frequency; the
Kramers-Kronig transforms determine how the two changes are
linked. Figure 2.13 illustrates schematically this link between gain and
refractive index changes with frequency, basing the work on eqn.
62 Distributed feedback semiconductor lasers

100
gain 1.33JV
50
1.33AT
(1/cm)
o
-0.01 0.01 -0.01 0 0.01 -0.01 0 0.01
log(atfQ) log(cnflB)
Figure 2.13 Schematic links between gain and refractive index changes with frequency
caused by Kramers-Kronig relationships
The frequency H is the frequency of the gain peak. The real refractive
index is given by the phase, measured in inverse distance to have the
same dimensions as the gain (imaginary refractive index). The real
index level can be set at an arbitrary level without altering the frequency
dependence; only the relative changes of phase are of concern. As the
electron density increases, the gain (imaginary refractive index)
increases and the real refractive index decreases. The parameter aH can
then be estimated

A7.16 and limiting the spectral range to values around the gain peak.
Even at the peak of the optical-gain mechanism, there is a separate
physical effect caused by a plasma of nearly free electrons which adds
a component into the relative permittivity, reducing the real refractive
index with increasing electron density (Appendix 7). As the frequency
moves away from the peak gain, this change in the (real) refractive
index is modified by the gain curvature through the Kramers-Kronig
relationship. At any particular wavelength, one can define the ratio
known as 'Henry's a factor' or aH:
aH= - {(dnr/dN)/(dnt/dN)} (2.40)
where the negative sign is inserted to make aH positive in this work.
When amplitude modulating a laser, by turning it on and off, there
have to be changes in the gain. From eqn. 2.39, the field gain/unit
distance g=coni/c = 2Trni/Afs, so that from eqn. 2.40 changes Agin the
gain give changes in the refractive index from
An = -aHAfsAg/27r (2.41)
The laser frequency is usually controlled by an optical length such as
the optical length of the Bragg period or the Fabry-Perot cavity, and
is therefore directly proportional to the average refractive index.
Consequently the magnitude of aH directly affects the change of
frequency A^/with a change in gain Agin a laser by approximately

~ - An/n ~ r) (aH (2.42)

 Gain, loss and spontaneous emission 63

1.6 x 10 18 c m " 3
0CL

1/1650 1/1600 1/1550

frequency (in reciprocal wavelength), nrri-1
Figure 2.14 Estimates of linewidth enhancement factor aH for quantum-well
material
QW material with seven 80A wells and 80A barriers
Parameter=carrier density
Data compiled from P.A. Morton et al [25]

where the bar denotes some appropriate average over the laser. These
dynamic wavelength shifts are known as (dynamic) chirp with dynamic
changes in frequency following dynamic changes in the gain. Different
laser structures using the same material will have different chirp
because of structure-related effects, and especially if the operating
frequency can be controlled in relation to the gain peak. Figure 2.13
shows a rough estimate of the changes in aH with operating frequency,
indicating that aH rises markedly at the lower frequencies and reaches
a minimum above the gain peak.
This effect can be understood by splitting the refractive-index
dependence into the component caused by free electrons (also
referred to as the plasma effect) where increasing the electron density
reduces the real refractive index over a broad range of frequencies
(Appendix 7, Figure A7.3). The localisation of the gain gives rise to an
associated ^-shaped change in the real refractive index, centred about
the gain peak. Above the frequency of this gain peak, the net real
refractive-index change with electron-density change reaches a mini-
mum. Figure 2.14 shows measured data indicating the reduction in aH
at high frequencies (short wavelength), as also shown schematically in
Figure 2.13.
64 Distributed feedback semiconductor lasers

In Fabry-Perot lasers, the laser must operate close to the gain peak
so that major variations in aH are not possible. In DFB lasers, the
choice of operating frequency is determined more by the grating, so
that the choice of the wavelength for peak gain gives an extra degree
of freedom and allows one to operate at frequencies where there is a
lower value of aH than for a Fabry-Perot laser, and so in principle one
can reduce the chirp. At 1550 nm it is common to offset the lasing
wavelength by around 20 nm down in wavelength from the gain peak
to take advantage of this lower value of chirp.
Quantum-well material allows one to engineer the gain profile to a
much greater extent than merely changing the composition in bulk
materials [8]. If the gain peak could be made narrower about its centre
frequency, then the ^-shaped change in the real refractive index
becomes stronger and gives a bigger offset for the plasma effect [33]
(see Figure A7.3). At the correct wavelength, it may be possible to
design quantum-well material to approach a 'chirpless' laser where
aH—*0. A detailed study of Henry's alpha factor has been undertaken
by several authors, with a notable discussion by Osinksi and Buus
[34].

2.6 Temperature-induced variations in semiconductor

lasers
There is considerable interest in understanding the changes in the
performance of lasers as temperature changes [35-37] where the
temperature of importance is that of the active area. Because of the
heating from injected carriers, this active region temperature may be
different from that of the heat sink by 2-10°C, but this difference will
be strongly affected by the technology of packaging. Even with
excellent heat sinking using Peltier coolers, local changes in tem-
perature can still alter important material parameters significantly. A
full account is beyond the scope of this book which only outlines the
effects of temperature on refractive index and wavelength and then
briefly considers the effects of temperature increases on gain and
threshold in bulk materials. Lasers using quantum-well material are no
less sensitive to temperature. Indeed temperature induced effects can
be used to measure material parameters [38].
From eqn. 1.13 for a first-order grating, the physical periodicity of
the Bragg grating A is A = 2^b/nowhere w^is the effective value of the
refractive index for the guided wave and Xb gives the Bragg wavelength
in free space for the maximum reflection. An estimate of temperature
 Gain, loss and spontaneous emission 65

coefficient of wavelength may then be obtained from

(dAb/dT)/A^(dA/dT)/A+(dneff/dT)/neff (2.43)
The material coefficient of expansion, a= (3A/97)/A, f° r InP between
0°C and 100°C is a^4.5x 10~6 KT1 ( a * 6 x 10~6 K"1 for GaAs), and is
often ignored because it is an order of magnitude smaller than the
coefficient of refractive-index change with temperature (dn/dT)/n.
The guide structure may have an effect in changing nejf with
temperature but here only the change of the underlying value n is
considered. Measurements have been discussed for this coefficient
[39], which is difficult to measure and compute close to the bandgap
where there are strong lattice resonance effects. Estimates of
(dn/dT)/n for InP (GaAs) are the order of 6(7) xlO" 5 KT1 around
1.53 jxm (1.1 fjim) leading to the conclusion that the operating
wavelength increases with temperature because the refractive index
becomes larger. One notices that increasing the injection of charge
carriers into a laser therefore has two effects: a decrease of wavelength
as the carrier density rises and reduces the refractive index (Section
2.5), but on a longer time scale the wavelength increases because of
the rise in temperature induced by the additional current.
Some features of the variation of the gain spectrum with tem-
perature can be appreciated from Figure 2.8 which plots a normalised
gain against (hf-%g)/kT An important effect, not included in that
previous discussion, is that the bandgap %g decreases in energy with
rises in temperature, thereby increasing the wavelength for the peak
gain by approximately 0.35 nm/K for both GaAs and InP. The width of
the gain spectrum (and also the width of spontaneous emission)
increases with kT but to keep the peak gain constant the applied
voltage has to increase linearly with kT thereby requiring the electron
density to increase with temperature if the gain is to be kept constant.
Menzel et al. [40] have modelled these effects in some detail with
empirical curve fitting.
From the practical view of stabilising the laser's optical output, the
effect of temperature on the threshold is most important (Figure 1.3).
Although reduction in gain with temperature increases the threshold,
the more significant increase with temperature arises because of:
(i) carrier leakage over the heterojunction barriers;
(ii) increased recombination at material interfaces, for example in
BH lasers;
(iii) increased non-radiative recombination; but especially
(iv) increased Auger recombination which rises more rapidly with
66 Distributed feedback semiconductor lasers

temperature than any of the other effects [39] and is notably

significant for 1.3 and 1.55 jxm lasers.
The effects of (i) should be small in all well designed lasers [41]. One
measure often used with threshold current is the characteristic
temperature To. This empirically relates the threshold currents at
temperatures Tx and T2 by
Ithnsh(T2)=Ilhmh(Tl) e x p { ( 7 2 - T,)/T o ) (2.44)
As a rough guide,
for 0.9 |xm lasers To~ 120-200 K
for 1.3 jxm lasers T0~60-100 K
for 1.55 firn lasers ro~4O-7O K
increasing with temperature as Auger recombination starts to be
significant.

2.7 Properties of quantum-well-laser active regions

2.7.1 Introduction to quantum wells
The physics of quantum-confined states and quantum-well lasers [7] is
a major subject but, in this book on DFB lasers, quantum-well material
is regarded as essentially a material with special parameters, partic-
ularly high differential gain, caused by the quantum-well layers. This
section limits the discussion to outlining how these special parameters
arise.
Electron-wave functions in bulk semiconductors, found through
solving Schrodinger's equation, spread over the dimensions of many
atoms. This permits average relative permittivities to be used. Equally,
conventional device dimensions are much larger than the electron
wavelengths which then allows the discrete energy levels to be
approximated by a continuum of electron energies and wavelengths.
With the dimensions of conventional heterostructure band diagrams
(Figure 2.15&), holes and electrons may exist at energies between the
active-region bandgap energy and the confining layer bandgap
energies.
Once the active region width falls below about 20 nm, and more
markedly at smaller dimensions, then quantum effects emerge and the
permitted energies are quantised between %acl and %dad (Figure 2.15&).
Transitions are only possible at energies %Q\, %02 etc. with only the
lowest states occupied except at high carrier-injection levels. The
 Gain, loss and spontaneous emission 67

f conduction band

% act
n^ ^ tf Q2
3 act 'Q1

valence band

active region well width

a b
Figure 2.15 Energy-band/distance diagrams
a Bulk material with active region widths over 20 nm so that a
continuum of energy bands is the appropriate approximation rather
than discrete levels
b Quantum-well material with active region widths in 1-10 nm range for
which quantised levels have to be considered

f
density of states

bulk

Figure 2.16 Schematic comparison of density of states in bulk material and quantum-
tuell material

density-of-states function is proportional to V(cg-Cgr) for bulk-material

conduction electrons, but for quantum wells it becomes a 'staircase'
touching the original parabola (as in Figure 2.16).
Several advantages of quantum-well material, compared with bulk
material, can be deduced from this diagram; in particular, fewer states
have to be filled to reach inversion, so threshold currents should be
reduced. Further consideration shows that if only one energy level is
filled, the spectral width of spontaneous emission and gain should be
much smaller than for bulk material, though electron-electron
collisions still broaden the energy bands. This advantage may be lost at
higher drive currents, as additional energy levels are filled.
68 Distributed feedback semiconductor lasers

Transitions in unstrained bulk material are essentially isotropic, i.e.

on average do not depend on the electron's direction of travel. In such
lasers, the excitation of a TE or TM optical mode is partly determined
by the interaction, or mode overlap as it is called, of the particular
mode with the gain region. Mode overlap is still important in QW
lasers, but now the transition probabilities, which give gain, depend on
the direction as well as the magnitude of the momentum k, and so the
polarisation of the E field and its orientation with respect to the
quantum wells also help to determine whether TE or TM modes are
favoured. In fact, any process which breaks the lattice symmetry, such
as mechanical strain, can also differentiate TE and TM gain. Higher
strain levels are possible in quantum wells without dislocations being
generated, and hence strain effects are usually larger than in bulk
materials.
Finally, it must be said that some of the improvements claimed for
lasers using quantum wells may be caused by the advances in
fabrication technology which are steadily being made over the years, so
that improvements would also be seen on bulk lasers using such
advanced fabrication.

2.7.2 Gain saturation and the need for multiple quantum wells
Many of the benefits of quantum-well materials result from the lower
density of states and the lower carrier densities needed to reach
transparency; however, these features can also be detrimental. A
separate (optical) confinement structure is essential in quantum-well
lasers to ensure that the optical filament is concentrated around the
quantum well(s); the quantum well does give some positive waveguid-
ing, but it is far too weak to confine the mode. However, because of the
nanometre scale of the quantum regions, confinement factors F<0.01
are typical, in contrast to F~0.5 for bulk lasers. Hence the required
gain for quantum wells may need to be nearly two orders of magnitude
larger than the gain in similar lasers using bulk material. This higher
gain leads to a more rapid filling of the available states at a particular
energy, making any increase in gain at that desired energy (i.e. at a
desired wavelength \0) impossible. As the drive current increases
steadily, yet higher energy levels in the wells may be filled, increasing
the gain at shorter and shorter wavelengths, but this will not increase
the gain at Ao. The gain therefore saturates at a given Ao with increasing
electron density, as was pointed out in Section 2.4.3. These serious
effects of gain saturation, especially with 1.5 |xm material, are
mitigated by stacking, say, three to seven quantum wells in an active
 Gain, loss and spontaneous emission 69

energy tensile compressive

strain unstrained strain
X
' \ I \
TE favoured \ / TM favoured

ih

momentum k
Figure 2.17 Schematic effects of strain

layer and so engineering a material which has a much higher

differential gain than bulk material but has saturation and other
properties which are intermediate between those of a single quantum-
well and bulk material.

2.73 Strained quantum wells

Strain in the active regions of lasers has been regarded as deleterious,
leading to the generation and migration of defects, and causing
premature failure. However, because quantum-well layers are so thin,
remarkably large strains are possible; in fact there is a 'critical
thickness' for any level of strain, compressive or tensile. A useful guide
is to keep the material thickness for each quantum well below a critical
value which is around 20 nm per well for 1% strain [42]. If this critical
thickness is exceeded, subsequent epitaxial growth will usually have a
poor morphology, creating large numbers of defects which can move
during subsequent processing of material and devices. Up to these
critical levels, the laser designer is then able deliberately to build strain
into quantum-well structures, thereby achieving several advantages.
One of the most direct advantages is an increase in the range of
wavelengths of materials grown on a given substrate because lattice
matching is relaxed. A well studied example [43] of this is InGaAs
grown on GaAs with about 1% compressive strain in the active region,
which allows an operating wavelength of 980 nm, which is needed for
pumping erbium amplifiers and is difficult to achieve using other
materials systems. Figure 2.17 indicates that, for a given material, the
bandgap increases under compressive strain, but more importantly the
heavy-hole band moves away faster than the light-hole band, so the
70 Distributed feedback semiconductor lasers

degeneracy of these bands is lost. Furthermore, it is found that

transitions to the light-hole band couple better to the TM than TE
mode, providing better polarisation mode selectivity.
The degeneracy of light- and heavy-hole bands causes problems via
intervalence-band absorption (IVBA), particularly in long-wavelength
lasers [44]. The effect here is that photons can (and do) cause hole
transitions from one valence band to the other, and the resulting
energy change is dissipated in the form of multiple phonons as the
excited hole relaxes back to the centre of the band. Such a process is
a source of excess optical loss. This is also one of the major causes of
poor temperature sensitivity of long-wavelength lasers, since IVBA
becomes much more prominent at the high injected carrier densities
which are needed for lasing at high temperatures. This problem is
relieved by strain, which shifts the bands apart, and reduces the
probability of transitions between them.
In longer-wavelength lasers, based on InP, both compressive and
tensile strains have been employed to reduce thresholds and IVBA,
with some success. In DFB lasers the removal of the competition
between transverse-electric-field (TE) and transverse-magnetic-field
(TM) mode is also important, but in semiconductor optical amplifiers
(SOA), by contrast, the gains for TE and TM should be closely
matched to ensure independence of the output power of the
polarisation at the input.

2.7A Carrier transport

When quantum-well lasers were first introduced, the expected increase
in frequency response for optical output with direct-current modula-
tion was not realised, and research showed that one contribution to
the problem [30] was transporting the holes (and electrons) across
waveguide layers into the active region of the quantum wells. The
quantum wells, by themselves, often provide an inadequate change in
the refractive index to guide the light effectively, and also any holes/
electrons captured by the wells tend to escape by thermal emission
over the well barriers. To combat both of these difficulties, separate
(optical) confinement-heterostructure (SCH) layers have to be intro-
duced (Figures 2.18 and A l l . l ) which help to guide the light and also
keep any escaping carriers to the physical region where they can be
recaptured by the quantum wells.
The time taken by the carriers to transport from the contacts into
the wells is the sum of a characteristic diffusion time across the
separate confinement regions and a characteristic capture time (rmp)
 Gain, loss and spontaneous emission 71

n-contact
i-contact
valence band conduction band

separate
confinement
region

quantum well
layers

separate
confinement
region

p-contact p-contact
valence band conduction band
Figure 2.18 Transport into quantum wells
Energy/depth band diagram showing the lengths dSCH of the separate
confinement zones, designed to confine the photons close to the
quantum wells. These must be kept small to allow electrons and holes to
move rapidly into the quantum wells. Graded impurity composition of
the confining regions can be used to assist in the carrier transport

into the wells. The electric field is particularly low in these separate-
confinement regions, so that carrier transport is often dominated by
diffusion with a limiting time constant rdij-^q d\CH//JL^T which is of the
order of the time taken for holes (with their lower mobility than
electrons) to diffuse across their separate confinement region. The
problem is reduced if the waveguide layers are thin (dSCHis small), and
reduced still further if a graded-composition separate confinement
heterostructure (GRIN-SCH) is used. Here either a gradient or a series
of steps in the composition of the SCH layers can change the
conduction-band edge with depth, thereby assisting the carriers to
transport from the contacts to the wells in much the same way that
varying composition in the base of a transistor can help the transport
of charge carriers through the base. The small-signal dynamics of
carrier transport are discussed in Appendix 4, along with implications
for changes in emission frequency caused by changes of the free
carrier density in the confinement regions. Once this effect is
understood and the confinement lengths are reduced appropriately,
then transport effects can often be neglected, at least for modulation
frequencies up to tens of gigahertz [45].
72 Distributed feedback semiconductor lasers

2.8 Summary
Starting from a background of degree-level knowledge about semi-
conductors and band diagrams, this chapter provides key tutorial
background for understanding photon-electron interactions in semi-
conductors so that one can appreciate how laser materials provide
optical gain and loss, spontaneous and stimulated emission, and how
these quantities change with electron density and with wavelength.
Important discussions include a two-level model for a semiconductor
which gives the key rate equations for photon and electron inter-
actions and includes the full Fermi-Dirac statistics; this progresses to
an illustrative account of determining the gain as a function of
frequency and excitation.
The various material factors which are important in laser models are
discussed. Among these effects are the coupling factor for the
spontaneous emission to the lasing mode and the changes that occur
with gain and nonuniformity (the Petermann factors); and Henry's
spectral linewidth-broadening factor aH, which describes the changes
of refractive index and gain with electron density. Important effects
such as Auger recombination, temperature dependence, strain in
material, quantum-well material and carrier transport are all touched
on to provide the required background for later chapters.

2.9 Bibliography
Some helpful books on materials:
ADACHI, S.: 'Physical properties of III-V semiconductor compounds' (Wiley
1992)
RAZEGHI, M: 'The MOCVD challenge' (Adam Hilger-IOP Publishing,
1989), vol. 1
MURARKA, S.P, and PERKERAR, M.C.: 'Electronic materials: science and
technology' (Academic Press, Boston 1989) (although this is not about
opto-electronic materials, it contains helpful accounts of a wide range of
materials technology)
PEARSALL, T.R: 'GalnAsP alloy semiconductors' (Wiley, Chichester, 1982)

IEE EMIS Series (Electronic Materials Information Service)

BROZEL, M.R., and STILLMAN, G.E. (Eds.): 'Properties of gallium arsenide'
(EMIS series, IEE, London, 1996), 3rd ed.
BHATTACHARYA, P. (Ed.): 'Properties of III-V quantum wells and super-
lattices' (EMIS series, IEE, London, 1996)
BHATTACHARYA, P. (Ed.): 'Properties of lattice matched and strained
indium gallium arsenide' (EMIS series, IEE, London, 1993)
 Gain, loss and spontaneous emission 73

ADACHI, S. (Ed): 'Properties of aluminium gallium arsenide' (EMIS series,

IEE, London, 1993)
'Properties of indium phosphide' (EMIS series, IEE, London, 1991)

2.10 References
1 SALE, T.E.: 'Gain calculations for strained InGaAs/GaAs quantum wells'
in 'Vertical cavity surface emitting lasers' (Research Studies Press,
Taunton; Wiley, New York, 1995), chap. 3
2 CORZINE, S.W., YAN, R.-H., and COLDREN, L.A.: 'Optical gain in III-V
bulk and quantum well semiconductors' in ZORY, P.S. (Ed.): 'Quantum
well lasers' (Academic Press, San Diego, 1993)
3 COLES, B.R., and CAPLIN, A.D.: 'The electronic structure of solids'
(Arnold, London, 1976), p. 88
4 SHOCKLEY, W.: 'Electrons and holes in semiconductors' (Van Nostrand,
New York, 1950)
5 KITTEL, C: 'Introduction to solid state physics' (J. Wiley, New York,
1966), lsted., pp. 316-317
6 CARROLL, J.E.: 'Physical models of semiconductor devices' (Arnold,
London, 1974), Section 6.5
7 ZORY, P.S. (Ed.): 'Quantum well lasers' (Academic Press, San Diego,
1993)
8 O'REILLY, E.P.: 'Valence band engineering in strained layer structures',
Semicond. Sd. Technol 1989, 4, pp. 121-137
9 BERNARD, M.G.A., and DURAFFOURG, G.: 'Lasers conditions in
semiconductors', Phys. Status Solid., 1961, 1, pp. 699-703
10 CARROLL, J.E.: 'Physical models of semiconductor devices' (Arnold,
London, 1974), Section 3.5
11 KANE, E.O.: 'Thomas-Fermi approach to impure semiconductor band
structure', Phys. Rev., 1963, 131, pp. 79-88
12 STERN, F: 'Calculated spectral dependence in gain in excited GaAs',/.
Appl Phys., 1976, 47, pp. 5382-5386
13 HALPERIN, B.I., and LAX, M.: 'Impurity band tails in the high density
limit. 1: Minimum counting methods', Phys. Rev., 1966, 148, pp. 722-740
14 BRINKMAN, W.F., and LEE, P.A.: 'Coulomb effects on the gain spectrum
of semiconductors', Phys. Rev. Lett, 1973, 33, pp. 237-240
15 BEATTIE, A.R., and LANDSBERG, P.T.: 'Auger effect in semiconductors',
Proc. Roy. Soc. London, 1959, 249, pp. 16-29
16 DUTTA, N.K., and NELSON, R.J.: 'The case for Auger recombination in
In^Ga, As/,.,,/. Appl. Phys., 1982, 53, pp. 74-92
17 BECK, A.H., and AHMED, H.: 'An introduction to physical electronics'
(Arnold, London, 1968), chap. 12
18 CARROLL, J.E.: 'Rate equations in semiconductor electronics' (Cam-
bridge University Press, Cambridge, 1985).
19 HECHT, E.: 'Optics' (Addison-Wesley, 1987), 2nd ed., Section 13.2
20 GOWAR, J.: 'Optical communication systems' (Prentice-Hall, Hemel
Hempstead, 1993), Section 18.2
74 Distributed feedback semiconductor lasers

21 OSINSKI, M., and ADAMS, M.J.: 'Gain spectra of quaternary semi-

conductors', IEEProc. I, 1982, 129, pp. 229-236
22 WESTBROOK, L.D.: 'Measurement of dg/dN and dn/dN and their
dependence on photon energy in A = 1.5 |xm InGaAsP laser diodes', IEE
Proc.J, 1986, 133, pp. 135-142
23 COLDREN, L.A., and CORZINE, S.W.: 'Diode lasers and photonic
integrated circuits' (Wiley, New York, 1995), Section 4.3.2
24 PEES, P., and BLOOD, P.: 'Derivation of gain spectra of laser-diodes from
spontaneous emission measurements', IEEEJ. Quantum Electron., 1995, 31,
pp. 1047-1050
25 MORTON, P.A., ACKERMAN, D.A., SHTENGEL, G.E., KAZARINOV, R.E,
HYBERTSEN, M.S., TANBUN-EK, T., LOGAN, R.A., and SERGENT, A.M.:
'Gain characteristics of 1.55-|xm high-speed multiple quantum well lasers',
IEEE Photonic Technol Lett, 1995, 7, pp. 833-835
26 PETERMANN, K: 'Calculated spontaneous emission factor for double
heterostructure injection lasers with gain induced waveguiding', IEEE J.
Quantum Electron., 1979, 15, pp. 566-570
27 ELSABER, W., and GOBEL, E.O.: 'Einstein relations for gain guided
semiconductor lasers', Electron. Lett, 1983, 19, pp. 335-336
28 WANG, J., SCHUNK, N., and PETERMANN, K.: 'Linewidth enhanced for
DFB lasers due to longitudinal field dependence in the laser cavity',
Electron. Lett, 1987, 23, pp. 715-716
29 AGRAWAL, G.P.: 'Effect of gain and index nonlinearities on single-mode
dynamics in semiconductor-lasers', IEEE J. Quantum Electron., 1990, 26,
pp. 1901-1909
30 WILLIAMS, K.A., GRIFFIN, P.S., WHITE, I.H., GARRETT, B., WHITE-
AWAY, J.E.A., and THOMPSON, G.H.B.: 'Carrier transport effects in
long-wavelength multi-quantum-well lasers under large-signal modula-
tion', IEEEJ. Quantum Electron., 1994, 30, pp. 1355-1357
31 THOMPSON, G.H.B., KIRKBY, P.A., and WHITEAWAY, J.E.A.: T h e
analysis of scattering in double-heterostructure GaAlAs/GaAs injection
lasers', IEEEJ. Quantum Electron., 1975, 11, pp. 481-488
32 PAN, X., OLESEN, H., and TROMBORG, B.: 'Influence of nonlinear gain
on DFB laser linewidth', Electron. Lett, 1990, 26, pp. 1074-1076
33 YAMANAKA, T, YOSHIKUNI, Y, YOKOYAMA, K., and LUI, W.: 'Theoret-
ical study on enhanced differential gain and extremely reduced linewidth
enhancement factor in quantum well lasers', IEEE J. Quantum Electron.,
1993, 29, pp.1609-1616
34 OSINKSI, M., and BUUS, J.: 'Linewidth broadening factor in semi-
conductor lasers—an overview', IEEE J. Quantum Electron., 1987, 23,
pp. 9-28
35 OE, K., and ASAI. H.: 'Proposal on a temperature insensitive wavelength
semiconductor laser', IEICE Trans. Electron., 1996, E-79C, pp. 1751-1754
36 PASCHOS, V., and SPHICOPOULOS, T: 'Influence of thermal effects on
the tunability of three-electrode DFB lasers', IEEE J. Quantum Electron.,
1994, 30, pp. 660-667
37 LI, X., and HUANG, W.P.: 'Simulation of DFB semiconductor lasers
incorporating thermal effects', IEEE J. Quantum Electron., 1995, 31,
pp. 1848-1855
 Gain, loss and spontaneous emission 75

38 TALGHADERJ., and SMITH, J.S.: Thermal dependence of the refractive

index of GaAs and AlAs measured using semiconductor multilayer optical
cavities', AppL Phys. Lett, 1995, 66, pp. 335-337
39 MCCAULLEY, JA., DONELLEY, V.M., VERNON, M., and TAHA L:
Temperature dependence of the near infrared refractive index of silicon,
gallium arsenide and indium phosphide', Phys. Rev. B, 1994, 49, pp. 7408-
7417
40 MENZEL, U., BARWOLFF, A., ENDERS, P., ACKERMANN, D.,
PUCHERT, R., and VOSS, M.: 'Modelling the temperature dependence of
threshold current, external differential efficiency and lasing wavelength
in QW laser diodes', Semiconductor Sci. TechnoL, 10, pp. 1382-1392
41 THOMPSON, G.H.B.: 'Physics of semiconductor laser devices' (Wiley,
Chichester, 1980), section 3.2.4
42 MATTHEWS, J.W., and BLAKESLEE, A.E.: 'Defects in epitaxial multi-
layers. I: Misfit dislocations',/ Cryst. Growth, 1974, 27, pp. 118-125
43 FALLAHI, M., DION, M., CHATENOUD, R, TEMPLETON, I.M.,
MCGREER, KA., CHAMPION, G., and BARBER, R.: 'Low divergence
electrically pumped circular grating surface emitting DBR laser on an
InGaAs/GaAs structure', Electronics Lett, 1993, 29, pp. 1412-1414
44 FUCHS, G., HORER, J., HANGLEITER, A, HARLE, V., SCHOLZ, E,
GLEW, R.W. and GOLDSTEIN, L. : 'Intervalence band absorption in
strained and unstrained InGaAs multiple quantum well structures', AppL
Phy. Lett, 1992, 60, pp. 231-233
45 WRIGHT, A.P., GARRETT, B., THOMPSON, G.H.B., and WHITEAWAY
J.E.A.: 'Influence of carrier transport on wavelength chirp of InGaAs/
InGaAsP MQW lasers', Electron. Lett, 1992, 28, pp. 191-212
 Chapter 3
Principles of modelling guided waves

3.1 Introduction

Without well controlled waveguiding to confine the light to the region

where the electrons and holes recombine most strongly, diode lasers,
and Bragg lasers in particular, would not have achieved their pre-
eminence as signal sources for communication systems. The waveguide
provides a stable platform for the electronic interaction even with
changes in optical power. Understanding waveguiding, then, is an
essential part of the study of DFB lasers. There are many fine texts on
guiding of electromagnetic waves [1-3] and many methods for solving
the problems of waveguiding ranging from finite-element and finite-
difference to integral equations, series expansions, separation of
variables, effective refractive index, beam-propagation methods and
slab-guide methods. Chiang [4] has provided a review of many
methods, giving over 200 references. This chapter therefore limits
itself to the basic concepts of optical waveguiding appropriate for
semiconductor lasers using slab guides which provides, in the authors'
view, one of the most helpful and readily accessible techniques. This
chapter, together with Appendixes 1 and 2, therefore provides only the
basic theory, but unlike many previous texts it provides the reader,
through the World Wide Web, with a versatile numerical package
(using MATLAB) which can illustrate the principles of guiding in
complex slab waveguides with both gain and loss and many layers, and
so gives the reader a powerful tool to explore some of the effects of
waveguiding. The reader who wishes to advance to discussions
specifically on DFB devices can move rapidly through this chapter,
referring back to it later as necessary.
 Principles of modelling guided waves 77

longitudinal variation of intensity

horizontal
'vertical variation
variation of intensity
of intensity

Figure 3.1 Vertical and horizontal variations

The terms transverse and lateral variations are also used in the literature
but with insufficient consensus
3.1.1 Vertical and horizontal guiding
It is helpful to make a formal distinction between vertical and
horizontal guiding, as indicated in Figure 3.1. The light is normally
propagating along the 'longitudinal' Oz direction and the terms
'vertical' and 'horizontal' refer here to the two directions which are
perpendicular to the Oz direction. Taking the normal orientation of a
laser on a test bench, these terms avoid the confusion which can arise
over the terms 'lateral' and 'transverse' which are also used.
Horizontal variations will refer to changes in the Ox direction, taken as
parallel to the junction or layer interfaces, while vertical variations will
refer to changes in the Oy direction, taken as perpendicular to the
junction or layer interfaces; Ox, Oy, Oz as usual form a right-handed
set of axes. The term transverse is occasionally used here, but will in
general mean both x and y directions, as for example in the transverse-
field pattern u(x, y). Longitudinal variations are for the present taken
to be simple forward and reverse travelling waves of constant
amplitude. These forward and reverse waves beat together to form
standing waves with a period of \J 2, but their average intensity is
taken to be uniform until later on in this work, when 'spatial hole
burning' is considered, and the average longitudinal variations can be
as shown in Figure 3.1.

3.1.2 Index and gain guiding

The reader is reminded of the principles of guiding light in a layer of
slightly higher refractive index than its surrounds. Figure 1.5 sum-
78 Distributed feedback semiconductor lasers

marised the concepts where one views the light, propagating along the
layers of dielectric, as plane electromagnetic waves (see Appendix 1)
with a cone of directions determining a pencil of optical rays. Each ray
experiences total internal reflection (TIR) at shallow ray angles of
incidence when reaching the outer boundaries of the guide. It is this
TIR which confines the light to one main region which has an
incrementally larger refractive index than the other layers. To keep the
principles clear, only nonmagnetic material, with permeability |xr
sufficiently close to unity, is considered. The complex refractive index
n=nr+jni and the complex relative permittivity er=srr-\-jeH of the
material are found to be related by n2=et.
For a plane wave in a uniform slab of such a material, the wave
propagates as exp{j(cot— (3z)\ exp gzwhere

(/3+jg)2=co2{fjL0sr((o)s0} (3.1)

so that fi-nrco/c and g= n^co/c.

There is a unique value of (3 and g for each co given 8r{(*>). For
'index-guided' lasers there are several layers of different materials with
a range of real optical refractive indices n r so that the result of eqn. 3.1
has to be modified. The whole concept of a mode within a guide is that
there is a single axial propagation coefficient fi over the whole cross-
section of the guide and one has to solve the guiding problem
matching the field boundary conditions between each layer. Never-
theless, eqn. 3.1 is still most helpful provided that it is written in the
form

(fi+jg)2 = a^{fj^ere:g{a))e0} (3.2)

with an effective value of complex relative permittivity representing
the average effects of the guide through a complex effective refractive
index neIf={nre(f+jniefj) = {sre[f)l/2. For guides where there is no gain or
loss, ?%is purely real and g=0.
In general, throughout this book, the gain is considered as a
perturbation in the refractive index so that

%e/f > nie[f with nreJf— (streir)1/2 a n d nifff—eriejr/2nrfff (3.3)

Unless stated otherwise this is the usual assumption. For a 'good' laser,
the guided transverse profile of the laser's optical fields should not
significantly alter as the gain increases. The interaction between fields
and charge carriers should then simply be proportional to gain
without serious nonlinearities over a useful range of power levels, say
up to around 5-10 mW in a laser for communication.
 Principles of modelling guided waves 79

In early lasers, the horizontal guiding was often provided primarily

by the gain [5]: 'gain-guided' lasers. From the point of view of the
mathematical analysis, layers of complex permittivity with different
gains provide reflections of the waves in a similar way to the reflections
provided by layers of different real permittivity, but the physics of gain
and index guiding rely on fundamentally different principles. When-
ever a beam of light is confined to a finite area, the beam diffracts as
it propagates and so expands after a few wavelengths unless there is
some compensating mechanism to maintain the confinement of the
optical energy to a constant area. In index guiding, the diffracted light
is reflected back into the main stream in the correct phase through
TIR, but in gain guiding this diffraction loss is compensated by the
generation of light as it propagates through an active medium with
gain. The narrower the guide, the greater the diffraction loss and the
greater has to be the gain in the material. It follows that layers which
only have changes in the optical loss cannot guide stable propagating
light because there is no gain to compensate for the diffractive loss.
In gain guiding, the gain changes as the electron density changes
and the complex interactions between light, current drive and
electron density mean that lasers, which rely on weak gain guides, tend
to be less stable under changes of operating conditions. Indeed, a laser
can be particularly sensitive to varying combinations of index and gain
guiding and in early lasers this led to instabilities or 'filamentation' in
the transverse structure [6]. This chapter presupposes that good
modern lasers are designed with strong dominant index guiding.

3.13 Effective area and confinement factor

The importance of this chapter comes in designing appropriate real
refractive-index changes in layers of material so that the optical mode
remains close to the region where the optical gain (i.e. where electrons
and holes recombine most) is strongest. Significant parameters for
laser design are the effective area si occupied by the optical fields and
the confinement factor F. The discussion of Section 1.3.4 can be
placed on a firmer footing as soon as one has discovered how the
electric field propagates, found the effective refractive index erejJ and
found the transverse modal pattern u(x,y). For example, if there are
forward and reverse propagating fields with negligible gain,
E=Fu(x, y) exp{j(co0t-/30z)}+Ru(x, y) exp{j(a)0H/30z)} (3.4)
then fi0 is determined for a given mode from eqn. 3.2. The phase
difference between the forward and reverse waves is, say, arg (i?/F) = </>.
80 Distributed feedback semiconductor lasers

w eff hor u(x,y)\'

W
eff vert

Figure 3.2 Effective-area schematic

Relationship between effective area si and field profile u(x, y) using an
equivalent 'rectangular' distribution (see also Figure 1.9).
^^wefjiwrXwef/vny Appendix 11 gives estimates of these values. (An
elliptical effective area might be used instead of a rectangular equivalent
area.)

Then the local field intensity exhibits standing waves with a periodicity
AOT/2 of the form
E*E=u(x, y)*u(x, y) {F*F+R*R+2 \FR\cos(2/30z+2(f>)} (3.5)
At first sight the periodic changes of intensity might be thought to
react with the charge carriers and set up periodic changes to the
optical gain in the active region. However, in general, these standing
waves have little effect because diffusion of the charge carriers along
the length of the guide smooths out and greatly reduces any impressed
periodicity in the electron density. There are exceptions when the
standing waves do have some effect, and these are discussed briefly at
appropriate times. Over one whole optical wavelength the product
term \FR\ cos(2/3oz+2 0) normally averages to have zero effect (but
see Section 6.4.3).
The modal pattern u (x, y) may be normalised to have a peak value
I u(x, y) \pmk=l permitting definition of an effective area for the mode
in the guide, see Figure 3.2, to be given from

\u(x,y)\2dxdy (3.6)
• / /

The equivalent average optical density S of the photons in this area

 Prin ciples of modelling gu ided ivaves 81

may then be given from the stored electromagnetic energy:

(F*F+R*R) \ erreff(a>o)so\u(x,y)\2dxdy=fia)oMS (3.7)

The optical gain of the material may now be introduced from the
imaginary part n{ (x, y) of the refractive index. The gain is confined to
the active region which occupies an area F si where the electrons and
holes recombine. The density of photons S interacts with the material
gain per unit distance gm only within the active region and F is known
as the confinement factor, given from

(F*F+R*R) I I {om,.(x, y)/c}s0 I u(x, y) 12 dx dy=hcoorgmsiS

>c area

(3.8)
A specific value for the confinement factor F is determined by taking
gm to be the peak value of {co wf(x, y)/c] and taking &iSas determined
from eqn. 3.7.
A slab waveguide, which is used to confine plane electromagnetic
waves in one dimension, provides an instructive starting point for
optical waveguiding to determine u(y) and erejj in one plane for a
particular mode. MATLAB programs are used to discover the main
effects of guiding and then an effective refractive-index model is used
for estimating u(x, y) in two dimensions.

3.2 The slab guide

3.2.1 TE and TM guided waves
In its simplest version, a slab guide can be formed from three slabs
where material of lower refractive index surrounds material with the
highest refractive index (Figure 3.3). Consider then a plane wave
propagating in one direction where the E and H fields, together with
the direction of propagation, form a right-handed set with
\E/H\ = (/uLr JUL0/'er e0)1/2. The E field is assumed to be oriented along the
plane of the junctions as in Figure 3.3a so that the E field has a
component only in the horizontal direction, i.e. transverse to the
junction; hence the nomenclature TE mode. It is found that the guide
will only act as a guide when this plane wave is launched into it at some
82 Distributed feedback semiconductor lasers

total internal reflection

total internal reflection

H

plane wave
TM O

Figure 3.3 TE and TM modes

To guide plane electromagnetic waves, the slab guides are theoretically
infinite in the x dimension

angle to the Oz axis so that Oz', the direction of launch of the plane
wave, is not exactly parallel to the Oz axis. This means that the Hy field
is not exactly parallel to the Oy axis and there is therefore a small
component of longitudinal or Hz field (see also Figure Al.l). Provided
that the angle of launching into the slab guide is shallow enough, then
the plane wave is reflected through TIR at the boundaries. Here any
one optical ray can be thought of as performing a slalom down the
guide with components of Ex, H, and Hz defined by the right-handed
set of axes shown in Figure 3.3. In general, there is a distribution, or
pencil, of rays at a number of different shallow angles at any one
position z. Appendix Al discusses plane waves further.
The TM mode is very similar to the TE mode except that it is now
the Hx field which is purely transverse to the junction (i.e. in the
horizontal direction as in Figure 3.36). The angle of launching of the
plane wave means that there has to be an E2 field as well as an E^ field
so that there is a longitudinal component of the E field.

3.2.2 Multilayer slab guides

Although a ray model can, in principle, give information about
propagation velocities through finding the angle of the ray to the axis,
 Principles of modelling guided waves 83

Figure 3.4 Multilayer slab guide (notation for TE excitation)

In the ideal theory the slab is infinite in the x direction with no x
variations; in practice the x variations are usually over a significantly
longer spatial distance than the y variations

it cannot give field or intensity distributions, which require solutions of

Maxwell's equations. In the initial cases the 'horizontal' variations of
the fields in the Ox direction are assumed to be sufficiently slow that
one can approximate with variations only in the Vertical' (or y)
direction. The three-slab guide is a well studied problem and the
reader can refer with confidence to alternative texts to obtain analytic
solutions for symmetric and asymmetric three-layer guides with TE or
TM modes [2,7]. With modern lasers, one has to have general
methods which aid designing slab guides with many more layers.
Figure 3.4 makes a start with five layers, labelling the top layer T, and
this notation is used in Appendix 2 to discuss the formalisms for a
computer program.

3.3 Wave equations for the TE and TM guided waves

The TE and TM classification shown in Figure 3.3 is discussed further
in Appendix 1 but the outcome for either mode is that, within each
uniform layer, the Efieldwhich is the main driving field for interaction
with the electrons is given from the wave equation

with HTE = (j/ co/UL0jULr) VxErE and JEVM = ( -j/ (os0er) V x HTM
With the axes as shown in the figures, Ex is the key field component for
the TE mode while Hx is the key field component for the TM mode.
The boundary conditions are determined by seeking the field
components which are tangential to the dielectric interfaces, as these
84 Distributed feedback semiconductor lasers

components are always continuous across the interface:

1 dE
Ex and Hz= are continuous for TE fields
j dy

1 r)//
Hx and Ez= - - are continuous for TM fields
dy
One notices that, for the TE fields with nonmagnetic material, where
/zr=l everywhere, the Ex field and its derivative are continuous across
the boundaries. For the TM modes there will be slight discontinuities
in Ey [= -/3HJ(cos0£t)] at boundaries between layers of different
permittivity whereas dEy/dy [ =j/3EJ is continuous.
An important caveat is required. Although interchanging E and H
can be helpful to keep a similar mathematical formalism for the
calculations of field profiles for TE and TMfields,care is needed when
considering energy or power interactions. Optical gain, in either TE or
TM waves, relies on the electricfieldinteracting with those oscillating
electronic dipoles formed by the recombining holes and electrons
which can radiate into the waveguide in the semiconductor material.
When it comes to the exchange of power and energy between
electrons and photons, it is therefore essential to consider the
interaction of the appropriate transverse E fields taken here as Ex for
the TE mode and Ey=(/3Hx/(oe) for the TM mode in the important
gain region.

3.4 Solving multislab guides

3.4.1 Effective refractive index
Appendix 2 discusses how to construct a numerical program to solve
the propagation for multilayers, takingfivelayers as an example which
can be extended to many more (e.g. 30) layers. The first numerical
illustrations in this chapter are simply for a three-slab guide, as in
Figure 3.5. The layers are lossless and only one wavelength (1.55 |mm)
is considered. The main guiding layer will be chosen to be sufficiently
thick so that three modes can appear with different vertical patterns
sandwiched between the slabs.
The method of solution is best appreciated by noting that the axial-
propagation coefficient /3 and the resulting growth rate g*for the wave
have to be the same in all the different layers, and it is possible to find
 Principles of modelling guided waves 85

Figure 3.5 Three-layer guide

Although mathematically the slab has the dimension whor—* °°, it is useful
for the purposes of calculating a finite power to have an effective and
finite horizontal width wh

an effective permittivity seff as in eqns. 3.2 and 3.3 with the fields
propagating over the whole vertical and horizontal extent of the guide
as in eqn. 3.4. The concept of an effective permittivity also can be
approached through a weighted mean relative permittivity. By the
definition of 'guiding', the optical fields have to vanish far enough
away from the main guiding layer. Now consider a solution Ex(y, z) to
the wave equation
{d%(y, z)/dz2} + {d%(y, z)/df) + kler(y)Ex(y, z)=0 (3.10)
2
with kl = co jULoso.
>z) = (3.11)
Then one may define

£/,(*)= iEx(y,z))dy (3.12)

• / :

Using

[d%(y,z)/dy2)dy=0 (3.13)

\d2EIx(z) /dz2} + %ereJ]E,x(z) =0 (3.14)

The weighted real effective relative permittivity is then given as
86 Distributed feedback semiconductor lasers

Bn(y)Es(y9 z) Ay I I Ex(y9 z) dy (3.15)

If the mode is asymmetric such that J^ Ex(y9 z) dy=O, then one has to
consider different limits between, say, yx and y2, values of y for which
dEx(y, z)/dy vanishes so that with the new limits eqn. 3.13 still holds.
The value of eqn. 3.15 lies in showing that the effective refractive index
is a weighted mean, rather than any utility in computation.
For guides with no loss or gain, the real effective refractive index
n(Jjj~verejj always lies between the maximum refractive index of all the
layers and the maximum refractive index of the two outermost layers.
This gives a clear limited area of search for appropriate numerical
solutions. It is instructive to consider the physical reasons for these
limits. Vertical variations exp( — yyy) or exp(yyy) occur with
jy-koV {srej]~ € !) when erejj>sr and such fields are said to be evanescent
To confine the optical power, the fields in both outer layers have to
decay as the magnitude of 3; increases (i.e. be evanescent). However,
when srejj<er the vertical variations are of the form cos (/3yy+(f)) where
/3v = &o v (s.,. — ereff) and can be said to give vertical propagation. If the
fields were to propagate vertically across the whole set of layers, the
power would dissipate away from the guiding region. Equally, the
modes cannot evanesce everywhere. There has to be at least one layer
where vertical 'propagation' occurs with both outer layers giving an
evanescent vertical variation with negligible energy at the far edges of
the guide. These requirements then force the effective real relative
permittivity to lie between the extreme actual values of the real relative
permittivity.

3A,2 Reflection coefficient calculation

The detailed algebra and mathematics for multislab guides is left to
Appendix 2. The method of solution starts with a trial value of an
effective refractive index. Then one considers a single frequency and
the appropriate component of an appropriately evanescing electric
field, varying vertically as Etof) exp( - yyy), which is 'fed' into the top of
the guide (i.e. decaying as ^~*oo away from the guide region). All the
fields, with the correct vertical propagation coefficients in each layer,
are matched for the required field-continuity conditions at the
boundaries. Finally, at the bottom of the guide one assumes a form of
the field which has to evanesce as exp( + yyy) (i.e. decaying as y—+ — °°)
 Principles of modelling guided waves 87

confining layer (top)

evanescent V, X \ evanescent
reflection P^top| \( ±Ttop excitation
'grows'away y\( decays away
from active area V from active area

evanescent
transmission
decays away
from active area
confining layer (bottom)

Figure 3.6 Evanescent excitation, reflection, transmission

The transversely (vertical) propagating wave in the guide area is excited
by the evanescent wave Elop leading to an evanescent wave T Ey, at the
bottom, but also a reflected wave pElof) with the incorrect form of
evanescence growing as one moves away from the guide, i.e. decaying
towards the guide. The correct solution has p = 0
with a magnitude TElop where T is a transmission factor. In general,
with an arbitrary choice of the trial effective refractive index, all
these boundary conditions will only match when there is a field
pEfop exp( + yyy) 'scattered' back at the top of the surface where p is a
reflection coefficient. Given that p is not zero, the solution is not
correct and the fields are not confined (Figure 3.6).
By scanning (both real and imaginary parts if required) the complex
relative permittivity ^^numerically, one finds distinct values where \p\
is a minimum and, for precisely the correct complex srejp one can, in
principle, find I/? 1 = 0. At this value there is the correct form of
evanescence of the field at the top and similarly the appropriate form
of evanescence for the 'transmitted' field TEtop at the bottom. The field
amplitudes decay to negligible values away from the guide, confirming
the process of guiding. Looking for a minimum in the modulus I pi of
the reflection coefficient is found to be a more stable mathematical
process (especially for lossy/gain guides with complex refractive
indices) than looking for a maximum transmission or an impedance
match at the bottom boundary [8]. Figure 3.7 shows typical results
with a simple three-layer guide found using a MATLAB program
slabexec in the directory slab. Note that the lowest-order mode has the
highest value of effective refractive index (largest value of ereff in the
scanning process) and the effective refractive index decreases as the
number of nodes increases.
88 Distributed feedback semiconductor lasers

Figure 3.7 Three-layer TE/TM-mode intensities

Index changes here support three modes for both polarisations at
1.55 jjim: note how the modal index is steadily decreasing from just below
3.4 to just above 3.25 as mode number increases. Only one TM mode
shown. Note the field intensities are the transverse IE I2 fields for both TE
and TM modes because it is the Efieldswhich interact with the electrons
and holes to give the optical gain

The example shown in Figure 3.7 makes the layer sufficiently wide
so that, at 1.55 |xm wavelength, it supports three modes. The zero-
order mode has no nulls in the Efield,while the first and second order
modes have one and two null intensities, respectively. The E-field
intensities in the TM mode are very similar in this instance to those for
the TE mode but, because of the different boundary conditions, there
are small discontinuities in the field intensities at the dielectric
interfaces. Only the second-order TM mode has been illustrated
because the other orders show hardly any significant difference from
the TE mode in their field profiles for this geometry. The modal
indices are very similar for both TE and TM modes. One notices,
particularly for the higher-order modes, how the fields evanesce most
strongly in the layer of lowest refractive index so that the optical
intensities do not penetrate as far into the bottom layer (index 3.2) as
 Principles of modelling guided waves 89

' / Y ^Ss-^
3.1994 eft. index
200/cm 0
3.277

. • 3.1989
// \ \ V"
. 300/cm / TE field \
TE field
intensity

1intensities \

gain/cm 9air\
refractive
index n=3.4
profile
3.1950 p - - - - jprofil^^
^ 1000/cm 1 \g \
I J_ gain '0' 3.2

Figure 3.8 More three-layer examples: gain/index modes

Left-hand graph shows pure gain guiding with no index change (n=3.2).
Field-intensity profile depends on gain. Figure in italics shows field gain
per centimetre while the upper figure shows the effective refractive
index. Note poor optical confinement compared with index guiding. On
the right, a narrow index guide supports only one mode

they penetrate into the top layer (index 3.25). With loss/gain in the
guide, the fields do not have strict nulls although the strong minima/
maxima remain.

3.4.3 Gain-guiding example

With a 2 |mm-thick layer (Figure 3.8) it is possible to obtain gain
guiding, but one may notice how the optical fields are poorly confined,
compared with the confinement that can be achieved with index
guiding. Practical values of gains over large volumes of material are
typically measured in the hundreds of inverse centimetres. Only with
unusually strong gains can the confinement start to match that of the
index guide. The change of field shape with increasing gain is a
problem and can lead to unstable emission patterns with pure gain
guiding. This instability of the intensity profile is made worse by
photon/electron interactions with regions of strong optical intensity
stimulating recombination of electrons and holes and reducing the
gain locally, but this has not been illustrated here.
In general, a good guide should support only one mode over the
range of frequencies of interest. To achieve this with the index guide
of Figure 3.7 one needs to make the high-index region considerably
narrower, as on the right-hand side of Figure 3.8. A comparable gain
90 Distributed feedback semiconductor lasers

guide with a 0.3 [xm-thick gain region would require a much larger
value of gain than for the 2 fxm guide because of the high diffraction
losses. Trying out numerical examples on a multipurpose slab-layer
program can rapidly give a sense of what can and cannot be achieved
in terms of guiding.

3.5 Scaling
Classic early work [9] on three-slab guides showed how to formulate
the guide in terms of normalised parameters which can be useful once
experience has been gained in the required orders of magnitude for
these normalised parameters. Here the required scaling laws are
discussed briefly to show how to change a successful design of a guide
to another wavelength.
Taking some appropriate vertical layer thickness d, a normalised
parameter can be defined:
V(y)=k,dV{sr(y)-srFff} (3.16)
This enables eqn. 3.10 to be written in the form
{d%(y, z)/d(y/d)*HV(y)%(y, z)=0 (3.17)
For a symmetrical three-slab guide where the central thickness is wx = d
with relative permittivity srX, then a V number can be defined from
koav(srl — er0) and this permits scaling of the guide with straightfor-
ward rules. Typical V numbers lie within the range 1-3 to ensure a
clear single 'zero'-order mode. With multilayers, one can see that,
provided that V(y) is maintained with y scaled with d, then the shape
of the field profile has the same shape, at least on the scale of d.
Narrowing the guide by a factor of two requires a doubling of
[\/{sr(y) — £r(,jjM which is determined by the differences of the relative
permittivity.

3.6 Horizontal guiding: effective-index method

In taking into account the vertical guiding, the effects of any
horizontal guiding have so far been ignored. Here the principles of the
'effective refractive index' method [10] are outlined to show one
useful way to estimate for horizontal guiding without having to use a
full three-dimensional numerical simulation. A convenient example
for horizontal guiding combined with vertical guiding is given from
the ridge waveguide shown schematically in Figure 3.9.
 Principles of modelling guided waves 91

Figure 3.9 Vertical and horizontal guiding under a ridge guide

An approximate solution assumes that there is a slow variation in the

O* direction so that approximate fields Ex(x, y, z) can be found at a
succession of rvalues taking into account the different set of vertical
layers at each x. Then return to the wave equation
{d2Ex(x,y, z)/dx2}+{d%(x,y, z) / df}Hd2Ex{x,y, z)/dz2}
+ k2sr(x,y)Ex(x,y,z)-0 (3.18)
By integrating over the ^-direction one reduces the problem to an
equivalent slab guide in the ^-direction. Define then:

Eeffx(x,z)=\ Ex(x,y,z)dy (3.19)

x)Eeffx(x, z) = sr(x, y)Ex(x, y, z) dy (3.20)

'-reJJ

These three equations then lead to

{d%ffx(x, z)/dx2} + {d%ffx(x, z)/dz2} + k2sre[f(x)Eeffx(x, z)=0
One then is back again at a one-dimensional problem with horizontal
changes of refractive index now dominating the calculation. The
program for solving slab guides can be invoked again, using sufficient
layers to approximate to the variations in ert^{x). The process could be
iterated. In slab guides, effective-refractive-index methods for rectan-
gular-shaped guides give varying accuracy [11,12]. In some
sophisticated guides where the vertical and horizontal geometries are
complex and similar, more sophisticated techniques may be required
such as finite elements [13,14] or beam propagation [15,16].
92 Distributed feedback semiconductor lasers

3.7 Orthogonality of fields

An important point about the optical fields is that the modes are
orthogonal. Here the distinction has to be made between guides which
are lossless (or nearly lossless) and guides with gain. The lossless guide
has modal field patterns un(x, y) where (again normalising the pattern
so that the peak values are unity)

1 for m= n I
un{x,y)*um(x,y)&x&y=8nnsin where Smn= f (3.21)
0 for m^n I

The value of sin gives the modal equivalent area. The conjugation is
significant because it leads to what Marcuse refers to as being 'energy
orthogonal' [17]. The energy storage can be split into separate modes
with all the consequences of equipartition of modal energy for
quantum statistics which is significant in mode counting for sponta-
neous emission (Appendix 9). There is an orthogonality (without the
conjugation) for guides with strong loss or gain [1] but these are not
true modes when it comes to counting modes for spontaneous
emission, as is highlighted in [18].

3.8 Far fields

The spatial distribution within the semiconductor guide is only half of
the problem of the design of a waveguide. It is often necessary to be
concerned with the far-field pattern. The conventional far-field optical
intensity is found to be proportional to the square of the modulus of
the Fourier transform of the near field just in front of the laser. In
general, the larger the radiating aperture, the narrower is the angle of
the beam. The principles of finding the far field are well recorded in
most textbooks on electromagnetic theory which have sections on
aerials [18]. However, less well recorded in texts are the effects of the
semiconductor surface which give rise to a special obliquity factor
modifying this far-field pattern. The origins of this obliquity factor are
given by the Huygen's radiation pattern, whereby any aperture, no
matter how small, has a radiation pattern which is dependent on the
obliquity and the simple far-field transform has to be multiplied by the
obliquity factor, as summarised in Appendix 2. However, for a laser
aperture this factor depends on whether the fields are TE or TM
modes. These different factors come about because of the slight
 Principles of modelling guided waves 93

1
polar obliquity
intensity factor

n = 3.5
eff TM

-1 1

Figure 3.10 Obliquity factors

Correction factors to far field dependent on direction of emission

differences of reflection at the surface of a dielectric, as shown in

Figure 3.10, and also in Figures A2.4 and A2.5.
Figure 3.10 shows that there is a slight broadening of the far-field
pattern for the TM mode in relation to the TE mode. The composite
slab structure of the semiconductor guide is taken into account in
estimating these obliquity factors through the effective refractive
index, which is taken as the appropriate mean value of n to use in such
calculations.
This section is concluded with two examples of the normalised far-
field (TE) patterns for a 'narrow' (0.5 jxm-wide) and a 'wide'
(2 |xm-wide) guide so as to indicate the orders of magnitude of the
effects (Figure 3.11). One can see how the 0.5 fxm-wide guide gives a
far field with a half angle around 30° while the 2 |mm guide gives a half
angle around 15°. The confinement factor was briefly outlined in
Chapter 1 and one may observe that, in the wider guide, the
confinement factor is close to 1 and the effective refractive index is
almost as if the wave was entirely within the central layer of highest-
index material.

3.9 Waveguiding with quantum-well materials

Before leaving this chapter on waveguiding, a short word is needed
about waveguides using material with quantum wells. In much of this
book, such material is treated as forming an equivalent bulk material
which has an enhanced gain but a reduced confinement factor
and increased optical-gain saturation. Although totally ignoring the
marvellous technology required to make these materials, this is
94 Distributed feedback semiconductor lasers

TE near field intensity

eff.ref.ind.
3.508

0 1 2 -60 -40 -20 0 20 40 60

microns angle (degrees)
1 , TE near field intensity
eff.ref.ind.
3.537
confinement
factor 0.97

0 1 2 3 -60 -40 -20 0 20 40 60

microns angle (degrees)

Figure 3.11 Near and farfieldsfor different guided layers

adequate for understanding much of the performance of quantum-

well lasers. In terms of designing waveguides to guide the light, this
assumption is inadequate. In bulk-material lasers, the active region
usually forms the main guiding layer within which the light rays are
totally internally reflected. However, with quantum-well material a
single well by itself would be inadequate to play the same role and even
with multiquantum wells one is too concerned with getting their
electrical composition right to have yet another parameter of
confining the light to consider in the optimisation problem. The
answer is to have separate-confinement heterolayers (SCH layers)
which guide the light and also have the role of keeping any escaping
electrons or holes close to their respective quantum wells. The
equivalent slab-waveguide then has many layers (Figure 3.12). Because
the quantum-well layers and barriers are so thin compared with a
transverse wavelength, one can estimate appropriate averages of the
refractive indices, but an automated slab-waveguide program can solve
the problem rigorously. The index in the SCH layers may be graded
because the composition of the separate confinement layers may be
graded, as discussed in Section 2.7.4, and this could be modelled using
the slab-waveguide techniques by a discrete step change but in many
thin layers.
 Principles of modelling guided waves 95
contact 1
outer layer

separate confinement layer *~ ^ -s ^

quantum wells
barriers
separate confinement layer

outer layer

contact

Figure 3.12 Schematic of multilayer guide

Outer layer, separate confinement layer, quantum wells and barriers, all
with different refractive indices and gains/losses

3.10 Summary and conclusions

The chapter and its associated appendixes has discussed the mecha-
nisms and mathematics required for optical guiding in
semiconductor-slab guides so that one appreciates the importance of
confining the light appropriately to the active region and can in
Chapter 4 use the transverse-field patterns to estimate the effect of the
grating on reflecting the optical fields. The MATLAB programs in the
directory slab available through the IEE Web site are started by
running slabexec . These give calculations for field profiles in slab
layers along with the far-field patterns. The default settings in the
program are for afive-layerslab guide. Trying out a few examples will
provide some insight and sense for the guiding processes. One may
observe how, with a thick guide, many modes are permitted and note
how the effective refractive index changes with increasing mode
number from a value close to the largest value of refractive index down
to a value close to the smallest value of refractive index of the different
layers. Occasionally, a result will come up where the field profiles show
clearly that the guide is failing to guide. These results happen if the
guide parameters are close to their cutoff values (i.e. the condition
where the guide is only just guiding, or not guiding, as the case may
be).
For the remainder of the book, it is assumed that the design values
of the transverse-field patterns u(x, y) are known and are sufficiently
stable with respect to changes in gain and also, because the modal
patterns are strongly gain guided, satisfy the orthogonality relationship
96 Distributed feedback semiconductor lasers

(eqn. 3.21) with respect to any other modal pattern. The MATLAB
programs provided permit the user to explore the validity of this for
any particular slab-guide design by simply changing the material gain
within the guide.

3.11 References
1 COLON, R.E.: 'Field theory of guided waves' (McGrawHill, 1960)
2 ADAMS, M.J.: 'An introduction to optical waveguides' (Wiley, 1981)
3 SNYDER, A.W., and LOVE, J.D.: 'Optical waveguide theory' (Chapman
and Hall, 1983)
4 CHIANG, K.S.: 'Review of numerical and approximate methods for the
modal-analysis of general optical dielectric wave-guides', Opt. Quantum
Electron., 1994, 26, pp. sll3-sl34
5 THOMPSON, G.H.B.: 'Physics of semiconductor laser devices' (Wiley,
Chichester, 1980), Section 6.4
6 HESS, O., KOCH, S.W., and MOLONEY, J.V.: 'Filamentation and beam-
propagation in broad-area semiconductor-lasers', IEEE J. Quantum
Electron., 1995, 31, pp. 35-43
7 HAUSS, H.A.: 'Waves and fields in optoelectronics' (Prentice Hall,
Englewood Cliffs, 1984), p. 175
8 ROZZI, T.E., and IN'TVELD, G.H.: 'Field and network analysis of
interacting step discontinuities in planar dielectric waveguides', IEEE
Trans., 1979, MTT-27, pp. 303-309
9 KOGELNIK, H., and RAMASWAMY, V.: 'Scaling rules for thin fibre optical
waveguides', AppL Opt, 1974, 8, p. 1857
10 MUNOWITZ, M., and VEZZETTI, D.J.: 'Beam-propagation computations
in one and 2 transverse dimensions', Opt. Comm., 1993, 100, pp. 43-47
11 LEE, J.S., and SHIN, S.Y: 'On the validity of the effective-index method
for rectangular dielectric wave-guides',/. Lightwave Technol, 1993, 11, pp.
1320-1324
12 CHIANG, K. S.: 'Performance of the effective-index method for the
analysis of dielectric wave-guides', Opt Lett, 1991, 16, pp. 714-716
13 SILVESTER, P.P., and FERRARI, R.L.: 'Finite elements for electrical
engineers' (Cambridge University Press, 1996), 3rd ed.
14 KOSHIBA, M., and INOUE, K: 'Simple and efficient finite-element
analysis of microwave and optical wave-guides', IEEE Trans., 1992, MTT-40,
pp. 371-377
15 YOUNG, T.P.: 'CAD tools for optoelectronic subsystems', GECJ. Res., 1994,
11, pp.110-121
16 HOEKSTRA, H.J.W.M.: 'On beam propagation methods for modelling in
integrated optics', Opt. Quantum Electron., 1997, 29, pp. 157-171 (over 70
references)
17 MARCUSE, D.: ' Quantum mechanical explanation of spontaneous
emission K-factor', Electron. Lett, 1982, 18, pp. 820-822
18 RAMO, S., WHINNERY, J.R., and VAN DUZER, X: Fields and waves in
communication electronics' (Wiley, New York, 1994), 3rd. ed., Section
12.4
 Chapter 4
Optical energy exchange in guides

4.1 The classic rate equations

4.LI Introduction
There are two basic classical methods of modelling mathematically the
operation of semiconductor lasers. The first method applies the
concepts of photon/electron particle exchange, outlined in Chapter
2, where one discusses the rate of absorption and emission of photons
along with the rate of recombination of holes and electrons, ensuring
at each stage that there is a detailed balance between photon
generation and electron/hole recombination leading to particle
conservation and energy conservation. This is the standard rate-
equation approach which is robust and well researched but can be
difficult to apply when there are strong nonuniformities, and even
more difficult when the phase of the electromagnetic field is
important. For distributed-feedback lasers, both the phase of the field
and nonuniformities are important and so one has to abandon the
photon-rate equation in favour of an approach based on interactions
between electromagnetic fields and the electric dipoles in an active
optical medium. This chapter presents both methods and shows
(mainly in Appendix 5) that they are, indeed, just different aspects of
the same physics of energy conservation and are wholly compatible
with one another.
The electromagnetic-field analysis is essential when the refractive
index/permittivity changes periodically inside the laser. The chapter
then concludes with an analysis of the coupled-mode equations which
determine how a fraction of the forward-travelling field is coupled into
the reverse-travelling field with a medium which has a periodic
permittivity, i.e. the waveguide contains a Bragg grating.
98 Distributed feedback semiconductor lasers

'vertical' /
/ <
; 'horizontal'

A
distance J / y distance
/ ' ' electron .. p^"} photon
N
^ 7 ^ density ^- -,'--
^ t J ^ density S
w^z——^
axi
arear A ^ ^st a n c e ^
•—;
_. /_ y '
\
1A
______L^
* /
/ .

/ / electronic "(confinement layer - ^ | / /

Si optical guiding layer - ' -area

i /

Figure 4.1 'Ideal' laser densities

Schematic uniform photon density 5 over an area _4 along whole length
L of laser giving a volume V=Ld. Similarly, electrons of uniform density
JV but occupying area T s£ giving a volume I T

A natural starting point uses the time-honoured rate-equation

approach. Understanding the technique is simplest with an ideal
uniform laser where there is a waveguide which determines a fixed
transverse pattern of the optical mode. In this model there is an
approximately uniform density S of photons being guided along an
area si (typically the order of a square micrometre or so) and length
L (typically a few hundred micrometres), giving a total active volume
T = JLS#. These photons S interact with a uniform density of electrons iV
in an active volume F Y = LF^ as in Figure 4.1. As indicated in previous
chapters, the photons are not as easy to confine to a sharply defined
region in the way that the electrons and holes are confined to the
active region through double heterojunctions. The concept of
confinement factor along with effective area si was discussed in
Section 3.1.3 and F(<1) is then the ratio of the effective areas of
equivalent uniform distributions of gain and photon density.

4.1.2 Rate of change ofphoton density

The rate of change of the photon number within such a laser cavity is
given from an equation of the form

(4.1)
at

Here the concepts outlined in Section 2.3 are followed and that
 Optical energy exchange in guides 99

section should be read first. The first term on the right-hand side in
eqn. 4.1 gives the rate of stimulated emission of photons by the
electrons within the laser's active region, as determined by the
material's optical power-gain coefficient Gm(N) per unit time. The
photon density is taken to be uniform so that any standing-wave effects
have been neglected, as in eqn. 3.7. Only those photons within the
region of electronic gain, of volume FT, experience gain so the
confinement factor F appears here. The second term is the optical-
power absorption per unit time as determined by a material/guide
parameter Am. This term accounts for absorption of photons outside
the gain region. [There can be an additional loss term ( — Tb N SY)
caused by IVBA, as discussed in Section 2.7.3, but it is omitted in this
first analysis.]
The photon lifetime rp gives an estimate for the rate of escape of
photons out of the cavity and can be determined in a very rough-and-
ready manner by assuming that a proportion T of the photons get
transmitted at each facet on each pass down the lasing cavity so that a
photon has an approximate effective lifetime rp~L/(Tvg) within the
cavity. More accurate estimates require the use of field calculations
(see eqn. 4.10). In the final term, N/rsp gives the spontaneous emission
caused by the electrons spontaneously recombining with holes, giving
out a random emission of light over a range of frequencies and all
directions. Only those frequencies and directions which can couple to
the lasing mode can participate in the stimulated interaction (see
Section 2.4.1) so that typically there is a coupling factor /3^~10~5<^l.
When the device is lasing strongly, one can often ignore this
spontaneous contribution but it is essential around threshold to excite
lasing action.
Obviously in eqn. 4.1 the volume T cancels but its inclusion at this
stage makes it more clear as to where and why the confinement factor
F occurs. The electron-hole recombination provides the stimulated
emission over the smaller volume F T than the volume Y occupied by
the photons as a whole. The confinement factor is then a convenient
'single-parameter' method of modelling complicated transverse dis-
tributions of photons and electrons on the assumption that these
transverse patterns are fixed by the guiding mechanisms and do not
change significantly with optical intensity or carrier density. That
assumption is not always true and then serious modifications have to
be made. It is usually a good approximation when the photons are
guided through well designed changes of the refractive index in the
material structure rather than being guided by the electronically
100 Distributed feedback semiconductor lasers

induced optical gain. Spontaneous emission could have a slightly

different confinement factor from that for the stimulated gain because
of the different distributions, but this complication is ignored here.
Later the net power gain G(N) per unit time and net field gain
g(N) per unit distance along with their differentials with electron
density need to be distinguished and related. Table 4.1 lists these
forms as used in this book. The field gain g(N) is typically measured
in units of inverse distance (with modern laser material
g~ 100-1000 cm"1—the lower values for bulk material and higher
values for quantum well material) and will be zero when the electron
density AT is at a value known as the transparency density Ntr. Around the
transparency value, it is often adequate to consider a linear change of
gain with electron density and the differential of the gain with electron
density gf (N) becomes the important parameter.

4 A3 Rate of change of electron density

The equivalent equation for the rate of change of electrons within the
interaction region T T is next considered, and this is given from

dt r q

where the first term on the right-hand side is the counterpart of the
similarly placed term in eqn. 4.1 and gives the rate of removal of
electrons from the interaction region F T caused by stimulated
recombination associated with the material optical power gain. In a
unit volume, N/rr gives the net spontaneous rate of recombination,
representing this for simplicity by a single (carrier-dependent) rate
constant l/r r . The difference between the time constants rr and rsp is
that the latter relates only to radiative recombination giving out
photons that can interact with the lasing field while the former covers
all recombinations including nonradiative recombinations.
This recombination in detail is made up from the following main
factors:
N/ rr=AN+B(NP) + C (N2F) + C"(NP2) (4.3)
The first term on the right-hand side is recombination proportional to
iVand is usually nonradiative; the second term is the bipolar radiative
recombination of electrons, density N, and holes, density P. The last
pair of terms in eqn. 4.3 represent Auger recombination where there
is no radiation given out, as discussed in Chapter 2. The fact that,
 Optical energy exchange in guides 101

Table 4.1 Relationships between different forms of gain

Subscript m for material power gain/loss per unit time

Gm(N) Material optical power gain per unit time as a function of
electron density
Gm(N) = G'm(N-Nlr)
G'm is differential of photon Gm with electron density
Am Optical-material power absorption per unit time

Upper-case G for net power gains per unit time

G=TGm(N) -Am Net power gain per unit time including confinement factor and
loss
G(N) = G'(N-Nh)
G' is differential of Gwith electron density

Lower case g for field gain/loss per unit distance

gm(N) Material optical-field gain per unit distance for fields
(dE/dz=gmE)
gJN)=g>m(N-Ntr)
g'm is differential of p h o t o n gm with e l e c t r o n d e n s i t y
am Material optical-field-absorption coefficient per unit distance

Lower case g for net field gains per unit distance

g=Tgm- am Net field gain per unit distance including confinement factor
and loss
) =gr (N~ Ntr) g' is differential of gwith electron density

Link between power gain/loss per unit time and field gain/loss per unit distance
G=2gug Gm=2gmvg Am=2amvm
where vg is group (energy) velocity in material; the factor of 2 arises because
t = GS=vgdS/dz=2vgE dE/dz= (2vgg)E2

under operational conditions, N~P within the laser means that one
can approximately simplify these terms into three major terms:
N/rr~AN+BN2+CN3 (4.4)
The analysis will show shortly that once the laser is lasing, iV remains
roughly constant or clamped and a linear approximation (N/rr) with
r r in the nanosecond range is a useful first approximation.
The final term in eqn. 4.2 is determined by the driving current /
which pushes electrons at a rate I/q into the conduction band within
the interaction region. This drive current is the source of electrons
and, under equilibrium conditions, must balance the net rate of loss of
102 Distributed feedback semicondudor lasers

electrons caused by all forms of recombination. In interpreting this

'ideal' rate equation, the current / which is referred to here is the
current which may usefully increase the electron density in the active
region and may be some 40% less than the external current being
driven into the local contact. The reason for this loss is the sideways
diffusion and spreading of the charge carriers as, for example,
indicated in Figure 1.6. For simplicity this loss has been ignored in this
analysis or alternatively lumped into an 'equivalent' nonradiative
recombination, a term which also leads to loss of carriers without
producing any useful photons. The convention here then is that / is
the useful driving current into the active region. It can also be the case
that, if the processing technology is exceptionally good, the net
impedance of the diode's contacts is less than the intrinsic junction
impedance so that the injected charge carriers are determined mainly
by the external voltage along with the appropriate diode equation
(Section 5.7), with the result that the electron density is more
uniform.
In more sophisticated analyses, as in Appendix 4, it is also
recognised that the electrons and holes, once they have been injected
into the laser contacts, take time to reach the interaction region,
especially with quantum-well material. This carrier-transport problem
is of real significance when one is dealing with such materials in order
to gain a high gigabit-per-second modulation (see Section 6.6). If the
delay or loss of charge carriers is significant (Section 2.7.4), then there
has to be another rate equation linking the actual external current Iexl
with the driving current /.
Yet another complication arises because all the experimental
evidence points to the optical gain reducing as the local optical
intensity increases at a fixed carrier density. As discussed in Chapter 2,
a useful practical approximation is to take the optical gain G(N) to be
represented by G(N- Nlr)/(1 + €S), so that the gain saturates at high
photon densities but is approximately linear in the electron density
around the transparency value Nlr. It has been argued that this
saturation of the gain arises because the electrons in the conduction
band and the holes in the valence band take time (on the
subpicosecond scale) to rethermalise as their densities within these
bands are depleted locally around those energy levels which are
participating most in the lasing process. This leads to spectral-hole
burning of the gain when the laser is lasing strongly. To a first order,
the redistribution of gain may be assumed to be immediate with the
effect approximated simply by the gain-saturation parameter.
 Optical energy exchange in guides 103

It has already been mentioned that, for a laser, N~P, an assumption

which arises through the requirement of electrical neutrality. If the
charge carriers did not neutralise one another in the gain region, then
their high densities would give rise to such strong fields that other
charge carriers would be forced into the region to ensure neutral-
isation. The time scale on which this effect occurs is also on the
subpicosecond scale and is ignored in this work with no extra rate
equation being included.
Eqns. 4.1 and 4.2 are the classic rate equations which, in various
forms, have been discussed in great detail by many authors ([1-8] are
a small selection). The variations that are possible include nonlinear
recombination, multimodes with different thresholds and coupled
lasers, along with the other factors already mentioned such as internal
impedance and charge transport from contact to quantum well.

4.2 Some basic results from rate-equation analysis

4.2.1 Simplifying the rate equations
It can be helpful, whenever possible, to group key parameters together.
For example, by definition of the 'gain region', the photons absorbed
within the gain region can create electrons within the conduction
band and these electrons are immediately then available to participate
again in stimulated photon emission . The electrons can be 'recycled'.
Photons absorbed outside the gain region do not create electrons that
are available to participate in stimulated emission by the guided
photons. One may then recognise that photons which escape from the
facets and photons absorbed outside the gain region can essentially be
grouped to give a net rate of loss of photons from the interaction
process and this requires only one parameter:

(4.5)
rp

TG(N)S/3T /sp (4.6)

rp rsp

dt
104 Distributed feedback semiconductor lasers

There are two important ways of studying the results of these

equations:

(i) small-signal analytic studies, and

(ii) large-signal dynamic computations.

Appendix 4 covers both features in more detail with MATLAB

programs available to demonstrate some main features from norm-
alised equations. A few of the main results are reported in this section
along with a small-signal analytic examination of the equations in
order to highlight some features that are either common to Fabry-
Perot lasers and distributed-feedback lasers or can be quite different.

4.2.2 Steady-state results

In the steady lasing state with negligible spontaneous emission
(/3sp—+0), it is found from eqn. 4.6 that the electron density N0—Nth
(>Ntr) where

~) /\^-rGm(N0)\ (4.8)
W/ K J
with

(4.9)
T
P

The light output is proportional to the internal photon density

SQ, SO that eqn. 4.8 shows that the light output is amplified
spontaneous emission with the amplification factor determined by
(loss - gain)" l = {( 1 /fy -TGm(No)}~x which is equivalent to the round-
trip gain referred to in Section 1.2. With small enough spontaneous
emission, it becomes clear that the steady-state electron density
N0—N(h, a constant value known as the threshold electron density
where the round-trip gain becomes infinite and eqn. 4.9 holds.
As an aside, eqn. 4.9 is equivalent to eqn. 1.5 with F = l and
Gm=2vggm> and provides a better estimate of the effective photon
lifetime for the FP laser (than in Section 4.1.2) from

l/Tfp=2vgam+(vg/L) ln(l/PlefiPrigh/) (4.10)

 Optical energy exchange in guides 105

With N0=Nth9 the net steady-state photon output rate O0=T S0/rp is also
determined from

(4U)

If the gain saturation is included through a nonzero parameter e, then

its main effect on the static characteristics is to cause the electron
density to rise slightly with increasing drive rather than be clamped at
a fixed level once the lasing starts. The concept of a fixed photon
lifetime in eqn. 4.10, while satisfactory for many FP lasers, is not so
useful for distributed-feedback (DFB) lasers where the photon lifetime
rp arises from distributed reflectivities so that a different effective
photon lifetime has to be estimated. The electron density changes with
distance and power level—so-called spatial-hole burning—and this in
turn can alter the effective photon lifetime.
From eqn. 4.11 there is a threshold current qTYNth/rr required to
create a threshold gain to start lasing. The concept of a threshold gain
is common to all forms of oscillator, and lasers are no different in this
respect. The analysis here assumes that there is no change in
confinement factor with power level so this is equivalent to assuming
stable optical-mode patterns. Above the threshold, in this ideal
uniform laser, the electron density changes little provided that one
does not wish to change the light output rapidly. The steady light
output increases linearly with an increase in the steady current (Figure
4.2). If the Auger recombination, photon absorption and optical-
scattering loss are all zero (rp=Tp) and there is no leakage or horizontal
diffusion of the electrons, then all the electrons driven into the laser by
the current above the threshold value are turned into photons. The
internal differential efficiency of the laser in such circumstances can
approach 100% and this forms one useful check on any numerical
analysis.
If the spontaneous coupling factor (3sp is no longer negligible, then
there is significant spontaneous light output below threshold where
the device acts like a light-emitting diode and the threshold current
for lasing is no longer sharply defined. Some research work envisages
the concept of a sufficiently high value of (3sp that one might talk about
the 'thresholdless' laser [9,10].
As noted in Section 1.2.3, a knowledge about only the magnitude of
the gain cannot by itself determine the frequency of lasing; conse-
106 Distributed feedback semiconductor lasers

drive-photon emission rate: D- O

drive-carrier density: D- Nr
1.5 \
normalised:
photon emission rate O
carrier density Nn
1

0.5 1 1.5 2 2.5

normalised drive D
Figure 4.2 Idealised normalised light/current characteristics
Predicted from single-mode rate equations with uniform fields. The
spontaneous coupling factor has a marked effect on removing a sharp
threshold when it reaches values of a few percent. The normalised
electron density Nn is almost constant once the stimulated emission
dominates, though with gain saturation Nn will rise slightly with
increasing drive

quently steady-state rate equations give no phase information and so

cannot determine the precise optical frequency. However, dynamic
rate equations will shortly be shown to give an indication of the changes
of frequency when the laser is modulated.

4.2.3 Dynamic analysis

With sufficiently small dynamic changes S} in the photon density from
a steady-state value So and equivalently small changes Nlf Ily Gx also
from steady-state values, the preceding results may be rearranged,
ignoring products of any two small quantities such as S} Nu to give

dS,
(4.12)
 Optical energy exchange in guides 107

where Gmi = G^A^ with GJn giving the differential material (power)
gain/unit time (Table 4.1). This pair of coupled equations becomes a
classic second-order equation

where

-Wso+i and

Note that the rates of change with time refer to modulation rates
(typically microwave rates up to several gigahertz) and not to rates of
change at optical frequencies (terahertz).
It is now helpful to list the basic results augmented by calculations
from a large-signal analysis outlined at the end of Appendix 4.
(i) Eqn. 4.14 shows that there is a damped resonant response of the
light output to the changes in the electron drive. Figure 4.3
demonstrates that gain saturation at the higher optical powers
[putting Gm—*Gm/ (1 + 65)] has a strong dynamic effect on the damping
of the resonance, and indeed without it one obtains an unrealistically
low damping of the transient. Typically, the damping rate is measured
on the nanosecond scale. The resonance frequency, referred to as the
photon—electron resonant frequency, increases as the square root of the
steady-state light output and takes values measured in the gigahertz
range. This resonance, which provides an upper limit to the useful
bandwidth of direct modulation, can be pushed up into the tens of
gigahertz by using material with high optical gain.
Horizontal carrier diffusion in the plane of the layers, referred to in
Section 1.3.3 but not considered in detail in this work, adds an
additional rate of loss of electrons and changes the details of the
damping of the electron-photon resonance. In general, these carrier
losses increase the damping and reduce the peak magnification in the
AM response (shown in Figure 4.5) with the maximum effect when the
width of the optical mode is approximately the same as the diffusion
length [11,12].
108 Distributed feedback semiconductor lasers

D = 2 . 5 ; a = 0 . 3 ; Q = 1000
normalised light

1
normalised carrier
w density
0
0 500 1000 1500
time / photon lifetime
Figure 4.3 Normalised dynamic light/current characteristics
The electron-photon resonance is shown from a large-signal dynamic
model with and without gain saturation: Q=Tr/TfJ; D is normalised drive;
a=Nlr/Nlh (Appendix 4). The drive current is switched on at 7^=0 . The
parameters are normalised as in Appendix 4. The normalising power
might be -10 mW with the photon lifetime measured in picoseconds.
Because the light has to grow from statistically varying spontaneous
emission, this emission will alter the turn-on delay of the stimulated
output

(ii) The light output (or rate of photon emission) can increase rapidly
only if the gain rises significantly above the threshold value. Similarly,
the light output can decrease only if the gain falls below the threshold
value so that dynamic changes of the light must be accompanied by
changes in the gain. For small signals, these changes of gain and light
are in quadrature (eqns. 4.12 and 4.13 and Figure 4.4) where the peak
gain (peak electron density) gives the greatest rate of increase of light
while the minimum gain gives the greatest rate of decrease of light.
Figure 4.4 also holds the key to the explanation of significant
changes in frequency and spectral-line broadening when one turns
lasers on and off by changing the gain. Recall eqn. 2.42, repeated here
for convenience, relating the changes in the frequency Agf with
changes in that gain:
 Optical energy exchange in guides 109

Figure 4.4 Light/electron-density/time

The figure shows the shaded regions where the electron density (and
hence the gain) are above the threshold value and so the light output can
increase with time. Because of a direct link between increases of electron
density and associated decreases of refractive index, the electron-density
curve also represents a dynamic shift of the central lasing frequency,
typically measured in tens of gigahertz for gigahertz modulation rates

- - Anr/nr~(A/v/27r) (aHAg/nr) (4.15)

The average changes in gain with time, as predicted by the dynamics of
a rate-equation analysis and eqn. 4.15, can give an estimate, at least for
a Fabry-Perot laser, of the change in the frequency [13] with time.
This dynamic change of frequency is called frequency 'chirp' or
linewidth broadening [14] so that a^has the obvious name of linewidth
broadening factor or linewidth-enhancement factor. In communication
lasers, this chirp can readily give more than 0.1 nm change in
wavelength, thereby broadening the spectrum of the laser many times
over the fundamental broadening required from Fourier analysis of a
modulated optical carrier.
(iii) The significance of the resonance frequency is demonstrated in
Figure 4.5 by a typical Bode plot of the power in the photodetector's
load resistor (<* l$0l2) for a given modulation power (<* \IY I2) with
increasing modulation frequency OJM. Above the photon-electron
resonance frequency, this Bode plot shows a fall in output power at 40
dB/decade even with an idealised package with negligible parasitic
capacitances, resistances or inductances and no effects from the
internal charge-carrier transport within the laser.
110 Distributed feedback semiconductor lasers

20 r

relative
response (dB)

-20

-1 0 1
log (modulation frequency /FT)

Figure 4.5 Normalised Bode response

Showing photon-electron resonance for small-signal AM response with
Fr~3 gigahertz and 0.25 ns effective recombination time. Fall off of
response above the resonance is at 40 dB per decade, indicated by broken
line

4.2.4 Problems of particle balance

The rate equations developed here do not require either detailed
quantum theory or electromagnetic theory. They rely on the concepts
of detailed balance in the numbers of particles being generated,
absorbed or lost from a given region. They cannot give explicit
information about the frequency of oscillation, even if one uses
specific information about gain as a function of frequency, as in Figure
2.8. In the form presented here, the rate equations neither make
allowance for nonuniformities in the field nor can they allow for the
optical field's phase which creates important new effects in DFB lasers.
Distributed-feedback lasers are sensitive to the nonuniformities of the
electrical fields as well as relying on careful phasing of the fields with
respect to the gratings, and so the simple rate equations cannot
succeed. Further equations are also needed if there are significant
delays or loss of charge carriers between their injection into the laser
structure at the contacts and those carriers reaching the actual
recombination regions. Despite their deficiencies, the rate equations
(eqns. 4.1 and 4.2) are key foundations on which to build the
relationships between the particle approach of the rate equations and
electromagnetic theory.
 Optical energy exchange in guides 111

4.3 Field equations and rate equations

43A Introduction
Whenever the phase of the optical field becomes an important effect
in the physics, straightforward rate equations have to be augmented.
For example, the modes of a Fabry-Perot laser have to be determined
from arguments about the field and the round-trip phase. In a DFB
laser the phase of the field, which is reflected back from the grating in
relation to the phase of the forward-travelling field, is a crucial factor.
A field analysis is therefore essential deriving, from Maxwell's
equations, the propagation of the forward field combined with some
distributed reflection, caused by the grating, of the reverse field into a
forward component. The propagation of the reverse field and the
coupling of the forward field will follow. It is also important to
understand that energy conservation as exhibited by the photon rate
equation of Section 4.1 is also demonstrable from Maxwell's classical
equations. Appendix 5 does this necessary detailed work developing:
(i) the role of electronic polarisation in the optical medium as the
key method of energy exchange between electrons and photons;
(ii) the connection between 'particle' exchange and electromagnetic
energy conservation; and
(iii) the detailed calculations for forward and reverse waves in a
guided medium with gain and periodic refractive indices.
This second part of the chapter presents a more tutorial account
referring, for detail, to Appendix 5.

4.3.2 Wave propagation

From Chapter 3, one can assume that within a lossless waveguide of a
laser there are forward- and reverse-travelling electric-field compo-
nents at an (angular) frequency co0:
E=Fu(x, y) exp{j((o0t-/30z)}+Ru(x, y) exp{j(a>Qt+(30z)} (4.16)

where the propagation coefficient /30 is identical over the whole (x, y)
cross-section of the guide. Within the permitted propagation ranges
for a single mode, there is a unique /30 for each a)0. The transverse
variation which gives the guided field pattern is given from a
distribution u(x, y) (as in Figure 4.6) normalised so that the modulus
of the peak value is unity to define an effective area si as in eqn. 3.6.
112 Distributed feedback semiconductor lasers

contact

horizontal
'vertical variation
variation of intensity
of intensity

z
Figure 4.6 TE mode
Schematic field patterns guided by a laser structure with the E field
mainly in the Ox direction. In many semiconductor diode lasers, the field
pattern in the x-direction is broader than that in the ^direction.

The frequency and propagation coefficient are related by solving

the guided-wave problem (see Chapter 3 ) to give a unique (3 for every
co from the equation:
Aco)e0} (4.17)
In a uniform medium with a plane wave throughout, the effective
permittivity enejj-erY> the real relative permittivity of the medium. In a
waveguide, the effective permittivity ene^ also gives the value of the
mean stored optical energy (electric and magnetic) per unit length
(associated with the forward wave) from eqn. 3.7.
Now consider only a forward wave and assume that this varies in
amplitude as
Ef=F(t, z)u(x,y) exp{j(a>ot--(30z)} (4.18)
Provided that the modulation frequencies within this amplitude
change are small compared with the optical frequency, the transverse-
field variation does not change markedly but there will be a 'wave
packet', as in Figure 4.7, and the envelope of the fields associated with
this wave packet within the laser will travel not at the phase velocity
vp=co0//30 but at the group velocity vg=dco0/dj30.
A straightforward mathematical description of the way in which
these forward waves move in space and time is then given from the
'advection' equation:

= (Tgm-aJF (4.19)
dz
 Optical energy exchange in guides 113

f l II II ii i i

steady field propagates at velocity: GO / (3 = v

o o p

slowly varying envelope of fields propagates

at group velocity: V g = 3 co / 3 p
Figure 4.7 Waves and wave-packets
A steady wave pattern varies as exp(ja)ot-jf3oz) but a wave packet may be
defined with the wave pattern inside the envelope varying as
exp(ja)ot—jf3oz) and moving at the velocity (O0/f30 while the envelope
F(t, z) (thick line) moves at the group velocity dco0/d/30 not in general
equal to <»o/po

where the first two terms ensure that, with no gain or loss, the fields
propagate with a fixed wave-packet envelope at the group velocity
which is the velocity at which energy propagates and is slightly lower
than the phase velocity in semiconductor materials. The terms on the
right-hand side of eqn. 4.19 give the net optical field gain using the
notation of Table 4.1. Here the field gain gm is a function of the
electron density N and may also saturate as the photon density S
increases. This field gain is reduced by the confinement factor F and
also by any absorption of the fields outside the gain region denoted by
am. The term (Tgm- am) appears regularly and, as shown in Table 4.1,
will be replaced with the net field gain per unit distance g.
Next one has to include the excitation of the fields by spontaneous
emission so that the advection equation for the forward wave becomes,
for example:

dF 1 dF ^ /A v
—+ =gF+isi)f (4.20)
u
dz vgdt
The detailed evaluation of the spontaneous excitation, written here as
ispfT is discussed in Chapters 2 and 5 and Appendix 9. This emission is
emitted over a broad range of frequencies and angles so that only a
small fraction /3sp couples to the lasing mode in the narrow waveguide,
114 Distributed feedback semiconductor lasers

with /3sp taking values ~10~ 6 . The spontaneous emission is strongly

linked to the electron density; consequently there is a confinement
factor Tsp defining the effective area T^si for the spontaneous
emission over the guide with Tsp similar in magnitude and concept to
the confinement factor F for the gain. Both fisp and Tsp can be included
in L)f so as to avoid additional cumbersome terms. Semiconductor
lasers operate typically with electron densities ~10 18 cm~ 3 and
recombination rates ~10 9 /s so that spontaneous recombinations
emitting photons which couple to the lasing field occur at ~/3sp 1027
events per cubic centimetre second. As a rough example, in a 1.55 |xm-
wavelength laser with a volume of the order of 400 jxm x 5 |jim x 0.5 fxm
there are ~10 12 photons/s emitted spontaneously which will then give
-100 nW of spontaneous emission.

433 Decoupling of frequency and propagation coefficient

When it comes to Bragg gratings with a periodicity determined from
cos(2/3/,z), it will be found to be analytically helpful to consider waves
varying as E(t, z) exp(j(3hz). However, at an arbitrary desired central
frequency co0, the wave will not in general have a propagation
coefficient of fih to couple with the Bragg grating. This problem is
overcome by 'removing' the link between the central frequency a)0 and
its associated propagation coefficient (30 by writing /30=8+(3b or
defining 8=/30 — (3h. Now write eqn. 4.18
Ef=F(t, z) exp(-j8z)u(x,y) exp{j(ajot- fibz)} (4.21)
Assume that 8 is sufficiently small so that all the previous assumptions
leading to the advection equation are still valid. Now redefine F(U z) so
that the new F is equal to the old value F(t, z) exp( —jSz). Similarly,
redefine a new value of ispf to be equal to the old value ispJ exp( —jSz).
The phase factor exp( — j8z) cancels, to leave eqn. 4.20 as:

(4.22)
vg
with Ef=F(t, z) u(x, y) exp{j(a)ot- /3hz)}. This has now allowed the fields
to have a central frequency co0 and at the same time to appear to have
a central propagation coefficient (3h as if the wave operated at the
Bragg frequency.
Because the spontaneous emission is random, the phase factor
exp(— j8z) multiplying this emission is found to be of no con-
sequence. The key random properties of spontaneous emission are
 Optical energy exchange in guides 115

described through complex autocorrelations (see Section 5.2) where

terms like the time average oi\<ispf(tl, zx)* ^ ( ^ z%)>\ determine the
magnitude and bandwidth of the spontaneous emission. It can be seen
that phase factors do not affect the value of such complex correla-
tions.
The photon density per unit guide length associated with the
forward power flow is referred to here as Sf and can, by normalising
the field values appropriately, be related to F*F=Sj so that eqn. 4.22 is
first multiplied by F* and then has the complex conjugate of the result
added to arrive at a travelling-wave rate equation for the photon
density per unit length associated with the forward field:

(4.23)

The spontaneous term is given from: sspf=ispj-F* + ispj-*F.

The reverse amplitude jRmay also be redefined (as Fwas redefined)
to obtain a reverse-propagating wave varying as
Er=R(t, z) u(x, y) exp{j(a)0t+/3hz)} where

(4-24)

paired with eqn. 4.22. Similarly, one can associate a photon density per
unit guide length Sr with this reverse-travelling field and obtain the
equation comparable to eqn. 4.23:

with sspr=isprR* + isp*R. The forward and reverse spontaneous emis-

sions are uncorrelated and so, if there is no coupling of the forward
and reverse waves through reflections or scattering, these two waves
are independent and uncorrelated. Nonlinear mixing inside the laser
can, of course, alter this result.
In some papers one may find that the advection equations are
written in terms of the power flow Pf{r) rather than in terms of Sf{r), the
photon density per unit length. Provided that the group velocity is
sufficiently constant over the packet of frequencies, power and photon
density per unit length are linearly related by the group velocity so
that
116 Distributed feedback semiconductor lasers

Pnr)=ha>vgSf{r)d (4.26)
It follows that, apart from the dimensions and calculation of the
spontaneous excitation, the advection equations for Pj{r) are the same
asforS /(r) .

4.4 Field equations with a grating

4.4.1 The periodic permittivity
The periodic change of refractive index or Bragg grating within a laser
has been met in Chapter 1 and it is convenient at this point, while
discussing field equations, to develop the coupled-mode field equa-
tions for a laser containing a Bragg grating. The whole point of finding
the travelling-wave fields with a central frequency co0 and an independ-
ent propagation coefficient f3b will now become apparent.
Suppose that inside the laser material there is a periodic permittiv-
ity
s={en(x, y) +8r/2(x, y, z)cos(2iM} (4.27)
The subscript rfz on the permittivity component indicates that it will
couple the reverse and forward waves in the z direction through its
periodicity denoted by the subscript cos(2fibz) denoting that its lowest-
order spatial period is A=w//3b, where A is a half wavelength in the
medium at the Bragg frequency. It is convenient to refer to A as the
Bragg period. In principle, the periodic term srji may be real or
complex, with the imaginary part depending on whether there is any
periodic component of the gain.
Historically, Bragg gratings grown within diode lasers had the
periodic part of their permittivity as essentially real because the grating
is most easily created by etching the cladding layers above or below the
gain region followed by regrowth of materials with different permittiv-
ity, as outlined in Chapter 1. In practice, the axial variation is not
sinusoidal as possibly suggested from eqn. 4.27 but is determined by
the etch pit that is defined to make the grating, as indicated
schematically in Figure 4.8. The magnitude of the step change of
refractive index depends on the regrowth material. With sufficiently
deep etching down into the gain region (Figure 4.8c), it is possible to
generate a periodic gain as well as a periodic permittivity, so that erfz
has an imaginary component. The reader may recall that in Section
3.1 it was argued that diffusion of the charge carriers smoothed out
variations in the gain which could be caused by optical standing waves
 Optical energy exchange in guides 117

etch and regrowth

Etching and material regrowth form grating.

Angle of etch depends on etchant + crystal
orientation. Etch depth controlled through
selective etching with one layer acting as a
stop layer as in (b). Regrowth can round
off sharp comers as dashed in (a). Deep
etching gives periodic gain as in (c).
1111111! 1

Figure 4.8 First-order grating inside laser (side-view cross-section)

with a similar periodicity. However, here the etching of the material

and the refill process limit the carrier diffusion so that the rapid
variations of gain can be embedded into the laser. Such technology has
only recently become available [15].
At each y value, Fourier analysis over one period of length A finds
the Fourier amplitude of the permittivity component varying as
cos(2/3/;z), where the component is given from

erfl(x,y) = erfz(x, y, z ) C O i ( W dz \ (4.28)

one period A
J
so that the relative permittivity variations of interest can be written as
e=en(x, y) +er/1(x, y) cos(2/3hz) (4.29)

4,4.2 Phase matching

As indicated in Section 4.3, wave packets of fields travelling in the +Oz
direction vary as F(t, z) u(x, y) exp{j(co0t— /3hz)} and when travelling in
the - O z direction vary as R(t, z) u(x, y) exp{j(a)0t+/3hz)}. The net
electric field E is then proportional to
E=F(t, z)u(x, y) exp{j(a)ot~ $bz)}+R(t, z)u(x, y) exp{j(co0t+f3hz)} (4.30)
To understand the key features of the wave coupling, take the full
wave equation in one dimension, without gain or loss or spontaneous
118 Distributed feedback semiconductor lasers

emission, and for simplicity ignore any dispersion, i.e. take the
permittivity to be independent of frequency so that phase and group
velocity are the same and take the parameter 5=0 so that (31=0% S^SQ/ULQ,
to give:

d2E d2E
{ { ) (2/3)}=0 (4.31)

Insert the fields of eqn. 4.30 into eqn. 4.31, assuming slowly varying
changes relative to the propagation coefficient or frequency so that
only first-order derivatives in time and space are retained (Appendix 5
shows how this can be done in more detail). The 'slow' variations in
the forward wave F are placed on the left-hand side to give the first-
order (i.e. ignoring terms such as srfl dF/df) rates of change of field:

dF dF
2j(3bu(x, y) -—+2j(o0srrs0fi0u(x, y) —
oz ot
=« - (o20srfl(x, y) cos(2/3hz)s0iLL0Eexp( -ja)ot+jpbz) (4.32)
The terms on the right-hand side couple strongly to the terms on the
left-hand side only if both sides vary with the same spatial frequency.
This is called the phase-matching condition. Any higher-order Fourier
components (HOFCs) such as £OJULOE exp(-ja)ot+fef3bz) change so
rapidly in space that any terms they excite over the first half of a
wavelength will be excited in the opposite phase over the next half
wavelength and the excitation effectively cancels. It is found that
elimination of HOFCs varying as~exp(/l\|/3/,z), N = ±l; ±2 etc., is avery
good approximation, when integrating eqn. 4.31 over a few wave-
lengths. Taking E as in eqn. 4.29 and writing
cos(2(3bz) =j{exp(j2(3bz) +exp( ~~j2/3bz)} (4.33)
the elimination process discussed above selects the slowly varying
envelope of the reverse fields R(z, t) and allows the variations
exp(ja)0t+j/3bz) to be cancelled to give
a>o8rf\(x, y) cos(2/3bz)s0/ui0Eexp( -j(o0t+jfibz)
- 4 o)lsrJl(x, y)eo/uLoR(t, z) u(x, y) (4.34)
Eqn. 4.32 then simplifies into

dF 1 dF !
()~—=--jj(oj0/vg){srjl(x,y)/en}R(t9z)u(x,y) (4.35)
 Optical energy exchange in guides 119

The transverse variation u{x, y) has been left in place deliberately to

emphasise that the only components which couple to the normal
mode F(t, z) u(x, y) also have to have the same normal mode field
pattern u(x, y). As stated earlier, the transverse-field pattern is unique
to a guided mode and orthogonal to any other modal pattern un(x> y)
(eqn. 3.21). To find the magnitude of the correct modal interaction,
start by writing

j((o/vg) {srfl(x, y)/srr}u(x, y) =K,fu(x, y) +JL, Knun(x, y) (4.36)

n

and then using modal orthogonality as in eqn. 3.21 and approximating

(a>/vg)~pb:

III f3k{e,n(x, y)/sj I u{x, y ) \ * d x A y (4.37)

where

I u(x, y) 12 dxdy
• / /

The reader is reminded that the periodic component er^(x, y)

cos(2f3bz) is found from Fourier analysis of the actual patterns of
etching and regrowth and eqn. 4.37 is then the second part of a
calculation to find a coupling coefficient Krf describing how forward
and reverse waves couple. This is the calculation which is performed
approximately with the program slabexec in directory slab on
selecting a Bragg grating while calculating the waveguide's field
patterns.
Now use modal orthogonality as above to remove the transverse
variation u(x,y), and make the same assumptions as in eqn. 4.22 to add
in spontaneous emission and to include the confinement factor and
attenuation in the gain. Include also the change between the Bragg
frequency and the operating frequency determined by the detuning
parameter 8=f30- /3h; then the periodic permittivity changes eqn. 4.22
to read

(4 38)
'
120 Distribu ted feedback semiconductor lasers

' etch and regrowth

n
gain and guidance region
XIII'llllllllll
Substrate////
1111111 ilia i

Figure 4.9 Second-order grating (side-view cross-section)

It is sometimes technologically easier to place the etch pits at twice the
first-order distance shown in Figure 4.8

The equivalent equation for the reverse wave is found by reversing the
sign of z:

dR 1 dR
(4.39)

With the phase of the grating chosen relative to the spatial origin as in
eqns. 4.29 and 4.31, Kjr-Krj=K. Chapter 5 revisits these equations, gives
an alternative perspective on their physics and highlights also the
meaning of the factory in front of the coupling coefficient K.

4.4.3 Second-order gratings

Lithographic constraints on defining the etch pits means that it is
sometimes technologically easier to space these pits at twice the first-
order distance apart (Figure 4.9), giving a grating period A' = 2A
known as a second-order grating (Section 1.6). Then the permittivity is
expandable in the form
e(x,y,z) =en(x9 y) +{erfl(x9 y) cos((3bz) +er/2(x, y) cos(2pbz) + • • •} (4.40)
where (3f = 7r/'A as before. However, stfl(x, y) cos( f}bz) does not have the
correct period to cause direct coupling between the forward and
reverse waves. The direct coupling requires the second-order term
with a periodic permittivity varying as cos(2/36z) extracted through
spatial Fourier analysis of the full variation of permittivity:

erfz(x, y, z) cos(2/3hz) dz\ (4.41)

one period A'
 Optical energy exchange in guides 121

radiation normal to guide

effective
t titt I
A =A, m

dielectric
constant
distributed
feedback
into
laser

no radiation normal to guide nn 11

radiation normal to guide
(a) (b)

Figure 4.10 Feedback and scattering properties of first- and second-order gratings
a First-order grating
b Second-order grating
Symmetrical radiation is assumed for second-order gratings here but
reflective layers would modify this assumption

Given appropriate designs of the grating, this component in the spatial

Fourier analysis of the grating can still have a significant amplitude.
Higher-order gratings where the etch pits are spaced at pA (p=3, etc.;
see Table 1.1) are only occasionally used.
The mathematics of the coupling for second-order gratings becomes
more complicated and the reader may be advised on a first reading to
skip lightly over the ideas rather than burrow into the details. The key
results are sketched in Figure 4.10, indicating sideways radiation
occurring as well as the forward and reverse wave coupling. The key
equations will be eqns. 4.48 and 4.49 which will contain the additional
loss and coupling terms caused by the radiation.
Assuming only vertical (y) and axial (z) variations, the required wave
equation is
d2E d2E
b*)

+sfr2(x,y) cos(2/3Az)}]e0/ufl (4.42)

As for the first-order grating, the terms in each of exp(j(o0t±jf3hz) are

isolated and equated to zero, but because of the term sfrl (x, y) cos( /3hz)
122 Distributed feedback semiconductor lasers

interacting with the E fields, there also has to be an additional

radiation-field term Emd(t, z) v{x, y) cxp(jco0t) having an axial propa-
gation coefficient (3z~0. To fit Maxwell's equations, this field has a
'vertical' propagation coefficient /3y~/3b away from the interaction
region and thereby radiates some optical power at approximately 90°
to the structure [16,17]. The required £ fields are now written as
E = F(t, z)u(x9y) exp(jco0t-j(3hz)+R(t, z)u(x,y)
x exp(j(o0t+j/3hz) +Emd(t, z) v(x, y) exp(ja)ot) (4.43)
where u(x, y) and v(x, y) are approximately real. The magnitude of
Emd(t, z) is estimated directly by putting 32/3z2—•() with d2/dt2-+ — 0)%
For compatibility with Maxwell's wave equation (eqn. 4.42) and the
phase-matching condition around /3~0, one has approximately:

t, z) +R{t, z)}u(x9 y)srJl(x, y) (4.44)

The small variations in the permittivity have relatively little effect on
the transverse propagation so that the approximation is made that
o)lsn£0juLQ-(3l. The physical clue to an approximate solution for eqn.
4.44 comes from considering an infinitesimally thin-layer excitation
8(y — y') so that the fields radiate symmetrically and vertically
(3; direction) on either side as plane waves with fields
<xexp{—j/3b\(y — yf) 1} as they propagate away from the layer. The x
variation is ignored and, on integrating over the whole range of layers
at each yf,
v{y)Emd =

e x p ( j f t ( y-y ) } d y W z)+R{ty z)} (4.45)

J/
The formal mathematics uses Green's functions [16,18] but the
physical concepts of a solution like a radiating plane wave, checked
through differentiation, show that eqn. 4.45 is a solution. The
 Optical energy exchange in guides 123

relationship between Erad and F+R can be written more compactly as

xexp( -jj3b\y-y' I) d / > [F(t> z) +R(t> z)} (4.46)

The modified eqn. 4.32 now has an extra term including Emd:
dF dF
2j/3bu(y) •r-+2j(o0erre0iuL0u(y) —
OZ 01

=* - a%erf2(y) cos(2fibz)s0{i0Ru(y) exp{j(a)ot+f3hz)} exp{ -j(co0t- (3bz)}

- o)20srfl (y) cos(/3bz)s0/n0 Eradv(y) exp(ja)ot) exp{ -j(a)ot- (5bz)} (4.47)
Note that the radiating fields have a variation ^(3;) with y that is not the
same as u(y). Ignoring rapidly changing terms in z, as explained
previously, but equating the slowly varying terms on both sides which
have the same spatial periodicity and selecting the correct modal
variation u(x, y) through the orthogonality relation, then further, on
adding in the detuning parameter 5, one finally obtains
dF 1 dF
+_ . gF+jSF=jKff2R+ ispf~ Krfl (F+R) (4.48)
0Z u„ 0 I

where

\ I fcrpiy) /eju(x, y)2dx dy

J J

and

u(x, y) dxdy
JJ
• / /

with

ll{8rf[(y)/EJu(x,y)v(y)Emddxdy (4.49)
124 Distribu ted feedback semiconductor lasers

Substituting for Emd leads to

I I ±O2erf\(y)u(x>y) I00 Srfi(y')u(x,yr)

JJ ' *n J~ *n
For symmetrical guides,

t x,y)
— exp( -j/3hy) dy dx/si (4.50)

Eqn. 4.48 has important additional terms — KrfX(F+R) when compared

with the first-order grating. It can be seen, by taking the term KrjX Fover
to the left-hand side, that this term is simply loss which reduces the net
gain. The term — K^R gives complex coupling which again adds to
loss and it is found that if this term is large enough then one can help
to select the modes which lase within a Bragg laser [11], as will be
touched on later.
The correct evaluation of the radiation-loss coefficient Krfl is more
complicated than indicated above because it depends on the relative
phasing of the fields Fand R (e.g. F+R=0 gives no radiation) which in
turn depends on conditions at the laser's facets. The radiation also
need not be symmetrical above and below the laser as assumed here
but would depend on reflective (e.g. metallic) or attenuating layers
around the waveguide. Experimentally, one will find widely varying
results depending on the detailed fabrication methods. The values of
Krj{ for second-order gratings which are approximately evaluated by
the program slabexec therefore have to be used as indicative rather
than definitive parameters, and more detailed modelling/measure-
ment is required.
The corresponding equation to eqn. 4.48 for the reverse wave is

- ^+1 Yt~ gR+j8R=JKrJ2F+ispr- Krfl (F+R) (4.51)

4.4 A Shape of grating

Considerable work has been done in demonstrating how the shape of
the grating alters the coupling coefficient [19,20]. A square-tooth
 Optical energy exchange in guides 125

shape Am/4 in length is very satisfactory but the etching processes

often prefer triangular teeth, as in Figures 4.8 and 4.9, when the top of
the etch pit should be greater than Am/4 (but less than Am/2) by an
amount that depends on the etch depth and angle. To obtain the
largest values of K, which is the general thrust of this design work,
requires maximising the appropriate spatial Fourier component of the
effective permittivity by the choice of the width and depth of the
etching within the technological constraints of the etching processes.
The algebra is 'serious' and in general requires computation to reach
a solution. The MATLAB program slabexec in directory slab permits
relatively simple shapes to be explored in a way which is appropriate to
the limited shaping capabilities for grating teeth caused by typical
etching processes.

4.5 Summary
Starting with the classic rate equations which approximate the physics
of a uniform Fabry-Perot laser, several key features about lasing were
discovered:
(i) The round-trip complex gain including the feedback has to be
unity so that both the phase and magnitude of the field gain play
a significant role in the mode selection.
(ii) There is a resonance between the electron and photon energy
exchange which limits the capabilities for modulating a laser
directly.
(iii) There are increases/decreases in the electron density which
decrease/increase the refractive index and increase/decrease
the lasing frequency as the laser is turned on/off. The
magnitude of this significant effect is determined by aH, the
linewidth-broadening factor.
(iv) In a Fabry-Perot laser there are very few mechanisms to keep
the laser lasing in just one single mode, which is the most
desirable condition for modern communication systems.
(v) The necessity to consider both the phase and the gain suggests
that one should consider travelling-wave field equations for a
laser, and the necessary mathematics was set up to permit wave
packets to be considered with independent frequency and
propagation coefficients centred about relatively arbitrary
choices so as to prepare the ground for the final main section.
(vi) The introduction of a periodic grating into the waveguide which
forms the laser introduces a frequency-selective element that is
126 Distributed feedback semiconductor lasers

the main feature of this book—the Bragg grating inside the laser
diode.
(vii) The wave packets are centred about a frequency close to, but not
necessarily identical to, the peak reflection frequency of the
grating, but the central propagation for the wave packet is taken
to be governed by the periodicity of the Bragg grating.
(viii) The field equations coupling the forward and reverse waves are
discussed and methods for finding, from Fourier analysis, the
correct coupling coefficients given the waveguide structure are
briefly outlined and a MATLAB program has been made
available for the reader to explore this aspect further.
Chapters 5 and 6 continue with the physics and design of DFB lasers.
The detailed numerical modelling is left to Chapter 7.

4.6 References
1 STATZ, H., and DE MARS, G.: 'Transients and oscillation pulses in
masers', in TOWNES, C.H. (Ed.): 'Quantum electronics' (Columbia
University Press, New York), 1960, pp. 530-537
2 IKEGAMI, T , and SUEMATSU, Y: 'Direct modulation of semiconductor
junction laser', Electron. Commun. Japan, 1968, 51(3), pp. 51-58
3 PAOLI, T , and RIPPER, J.E.: 'Direct modulation of semiconductor lasers',
Proc. IEE, 1970, 58, pp. 1457-1465
4 LANG, R., and KOBAYASHI, K.: 'Suppression of the relaxation oscillation
in the modulated light output of semiconductor lasers', IEEEJ. Quantum
Electron., 1976, 12, pp. 194-199
5 CARROLL, J.E.: 'Rate equations in semiconductor electronics' (Cam-
bridge University Press, 1990), chap. 7
6 OSINSKI, M., and ADAMS, M.J.: 'Transient time-averaged spectra of
rapidly modulated semiconductor lasers', lEEProc.J, 1985, 128, pp. 34-7
7 KAWAGUCHI, H.: 'Instabilities and non-linearities in laser diodes'
(Artech House, Boston) Section 2.1.7 and Chapter 5
8 ADAMS, M.J., and OSINSKI, M.: 'Longitudinal mode competition in
semiconductor lasers—rate-equations revisited', IEEProc.-I, 1982, 129, pp.
271-274
9 YABLONOVITCH, E.: 'Photonic band-gap crystals', / Phys. Condens.
Matter, 1993, 5, pp. 2443-2460
10 DEMARTINI, F., MARROCCO, M., MATALONI, P., MURRA, D., and
LOUDON, R.: 'Spontaneous and stimulated-emission in the thresholdless
microlaser',/. Opt. Soc. Am. B-Opt. Phys., 1993, 10, pp. 360-381
11 FURUYA, K., SUEMATSU, Y, and HONG, T: 'Reduction of resonance
like peak in direct modulation due to carrier diffusion in injection laser',
Applied Optics, 1978, 17, pp. 1949-1952
12 THOMPSON, G.H.B.: 'Physics of semiconductor laser devices' (Wiley,
1980) Section 7.3.4, 'Diffusion of damping laser oscillations', pp. 424—
433
 Optical energy exchange in guides 127

13 CARROLL, J.E., WHITE, I.H., and GALLAGHER, D.F.G.: 'Dependence of

chirp in injection-lasers on temporal optical pulse shape', IEEProc.J, 1986,
133, pp. 279-282
14 HENRY, C.H.: Theory of line width of semiconductor lasers', IEEE J.
Quantum Electron., 1982, 18, pp. 259-264
15 GLINSKI, J., and MAKINO, T.: 'Mode selectivity in DFB lasers with a
second-order grating', Electron. Lett, 1986, 22, pp. 679-680
16 KAZARINOV, R.R., and HENRY, C.H.: 'Second order distributed feedback
lasers with mode selection provided by first order radiation loss', IEEEJ.
Quantum Electron., 1985, 21, pp. 144-150
17 HARDY, A., WELCH, D.E, and STREIFER, W.: 'Analysis of second order
gratings' IEEEJ. Quantum Electron., 1989, 25, pp. 2096-2105
18 JEFFREYS, H., and JEFFREYS, B.S.: 'Mathematical physics 3rd edn., 1956'
(Cambridge University Press), Green's functions, p. 493
19 STREIFER, W., SCIFRES, D.R., and BURNHAM, R.D.: 'Coupling coeffi-
cient for distributed feedback single- and double-heterostructure diode
laser', IEEEJ. Quantum Electron., 1975, 11, pp. 867-873
20 GHAFOURI-SHIRAZ, H., and LO, B.: 'Distributed feedback laser diodes'
(Wiley, Chichester, 1996), section 2.5
 Chapter 5
Basic principles of lasers with
distributed feedback

5.1 Introduction

The rationale for inserting the frequency-selective 'Bragg' grating into

a semiconductor laser has been met in Sections 1.5 and 1.6 and the
effect of a periodic permittivity on the energy exchange has been
considered in Section 4.4. The next two chapters provide a relatively
self-contained account of the physics and some numerical modelling
for uniform and phase-shifted 'Bragg' lasers. There is a slightly
arbitrary labelling of 'basic' and 'advanced' features before the
detailed numerical time-domain modelling is discussed in Chapter 7.
This present chapter starts with a more physical derivation of the
coupled-mode equations than the mathematics of Chapter 4 and
Appendix 5 and moves on to new features such as the eigenmodes for
the analytic solutions, the influence of grating parameters on the
dispersion diagram and the 'stopband' of nonpropagating fre-
quencies.
Readers familiar with energy-band-distance diagrams in semi-
conductor devices will not be surprised to find that stopband-distance
diagrams are similarly helpful when discussing Bragg lasers. The
deficiencies of the uniform Bragg laser over single-mode selection
have already been mentioned and are pursued further here to show
the benefits of the insertion of phase shifts at judicious places into an
otherwise uniform Bragg grating. The chapter concludes with an
outline of the frequency-domain-modelling technique using transfer
matrices. This is one of a number of numerical methods for simulating
the performance of DFB lasers. The reader will be able to contrast
these with time-domain techniques discussed in Chapter 7.
 Basic principles of lasers with distributed feedback 129

Fourier component reference point

distance Z
Figure 5.1 Choice of phase position for equal Krj and Kfr

5.2 Coupled-mode equations for distributed feedback

5.2.1 Physical derivation of the coupling process
The first key paper to propose distributed feedback (DFB) in
semiconductor-laser structures was farsighted by investigating theoret-
ically not only the use of real-refractive-index ('index') gratings but
also imaginary-refractive-index ('gain') gratings [1]. The derivation
here can assume an arbitrary mixture of gain and index coupling
though initially it is assumed that the grating is a lossless first-order
grating formed from a periodic index change. As seen in Chapters 1
and 4, the reflectivity is distributed to give a mean reflectivity per unit
length of Krj of the reverse wave into a forward wave and, similarly, a
reflectivity of Kj-r of the forward wave into a reverse wave. When the
reference point is taken at an idealised step change in the dielectric as
shown in Figure 5.1, then one knows that the field reflections of a
forward wave into a reverse wave, and vice versa, are both real such that
Ap,y= — &pfr (see Appendix 1 discussing the reflection at the interface
between two dielectrics).
When the reflection reference position is moved an arbitrary
distance a forward, the reflection of the forward wave into a reverse
wave changes so that Apr/—+Aprfexp(2jf3ha) and a reverse wave into a
forward wave changes by Ap/r—>Ap/r exp( — 2j(3,Ja). The factor of 2
allows for both the extra distance a travelled in the opposite direction
by any reflection as well as the extra distance a travelled by the main
wave. However, the total physical distance 2a corresponds here to an
optical distance AOT/4, changing Apr/-—*/Apr/ and Ap^—* - j*Ap/r. It
therefore follows from this phase argument that the reflection at the
reference centre of symmetry, as shown in Figure 5.1, is purely
imaginary and, as also expected from symmetry, this reflection
argument leads to the conclusion that JKrJ=JKfr. The product of the real
130 Distributed feedback semiconductor lasers

terms (#cry, Kfr) at the centre of symmetry for this type of grating is
observed always to be positive and remains the same if the reference
point moves.
Now consider a forward field F propagating in a waveguide at an
angular frequency co0 with an axial complex propagation coefficient /30.
The complex amplitude of F at %=0 is transformed to amplitude
Fexp( —j/30A+gA) at z=A and similarly a backward wave of amplitude
R incident at z=A is transformed to amplitude R exp( -j/30A+gA) at
z=0. The gain gis the net field gain per unit length making allowance
for losses and confinement factor. The value of the propagation
coefficient may be written as
Po=P>+8 (5.1)
where f3b gives the Bragg wavelength for maximum reflectivity and
therefore, as in Section 4.3.3, 5 is a measure of the detuning of the
optical wave from the central Bragg condition. Allowing for gain,
phase shifts and coupling, one may write, for one Bragg period,
(5.2)
(5.3)
F and R are defined at the reference planes one grating period apart,
while gA and <5A are assumed to be sufficiently small compared with
unity so that the approximations in eqns. 5.2 and 5.3 hold, which for
practical gratings and gains is usually correct. The 'finite-difference'
scheme of eqns. 5.2 and 5.3 can then be rearranged with only a slight
change of notation into the differential form in the steady state:
dF
+—=(g-j8)F+JKR (5.4)

~=(g-j8)R+j*F (5.5)

Adding in time dependence, by recognising that the fields travel with

die group velocity, and adding in terms for the excitation by
spontaneous emission (as in eqns. 4.22 and 4.24) one obtains
1 dF dF
i
 Basic principles of lasers with distributed feedback 131

The spontaneous excitation uses random spontaneous 'currents' ispj

and ispr (which include the spontaneous coupling (3sp along with any
appropriate confinement factor Tsp and are discussed in more detail in
Chapter 7 and Appendix 9). The forward random spontaneous
excitation is completely independent with respect to the reverse
excitation and at any one point and time the excitations are statistically
independent relative to the excitation at other points in space or time
(to a first approximation). The excitation is similar to white noise
except that, to be more accurate, spontaneous excitation will have
some correlation with previous values on the same propagating light
cone; the weaker the correlation then the broader is the spontaneous-
emission bandwidth. These properties can be expressed by the
complex (auto) correlations as follows:
{isps/X^tl)i^f/X^k)^Csp(il-'t2) for {zx -z2)=±(tl - t2)vg (5.8)

(ipf/Azu t)ispf/r(z2, t)*)=r2sp8(zl - z2) (5.9)

where Csp(tx — t^) describes the autocorrelation properties of the
spontaneous emission to allow for the limited bandwidth, and
S(z\~h) *s a Dn*ac delta function showing a complete lack of
correlation at neighbouring points, while r2sp determines the strength
of the spontaneous emission. As noted in Section 3.3, phase factors of
unit magnitude multiplying ispj(zu tx) etc. do not alter the magnitude
of the complex autocorrelation or correlation functions. Finding the
magnitude of the spontaneous excitation is considered later.

5.2.2 Complex gratings

When calculating K from a Fourier analysis of the variations of the
refractive index as outlined in Section 4.4, the cross-section of the
guided field does not adjust its optical width at each tooth [2], so the
average modal distribution in the guide was used in that calculation to
find the appropriate harmonic components. Such an analysis can give
a complex K=Krj=Kjr=Kindex+JKgnin through etching the grating down
into the gain region, for example. This complex reflection is still
referenced from the centre of symmetry as in Figure 5.1 where Krfax\d
Kfr are equal but their complex values indicate there is a gain grating
or a loss grating in addition to the index grating. This means that the
variation in the permittivity of the grating material also has to be
complex, say proportional to etr+jeri (with both elements positive).
Positive eH represents a material with gain and so with the reference
position as in Figure 5.1, K=Kindex+JKgnin (with both elements positive)
132 Distributed feedback semiconductor lasers

represents a grating where the gain maxima (or loss minima) coincide
with the maxima in real index (and vice versa if Kgain were negative).
With Kindex>0 and Kgain>0 one may refer to a gain grating 'in phase with'
the index grating (or a loss grating 'in antiphase with' the index
grating). With Kindex>0 and Kgain<0 one refers to a gain grating 'in
antiphase with' the index grating (or a loss grating 'in phase with' the
index grating).
As before, the phase of the reference point can change the phase of
the imaginary component so that it is easy to be confused between loss
and gain in a grating, and it is safest to stick to writing the equations
with a consistent reference point. A pure index grating had a positive
Krf Kfr product and, using similar arguments, a pure gain grating has a
negative K^ Kfr product regardless of the point of phase reference.

5.3 Coupled-mode solutions and stopbands

5.3.1 Eigenmodes
An elegant method for solving the coupled-mode equations is to find
the two eigenmodes, each consisting of a forward- and backward-wave
pair that propagate indefinitely along the grating with a fixed
amplitude ratio (s+ and s_, respectively) [3]. The angular frequency co
determines the offset parameter S which describes the detuning from
the Bragg condition:

n
8= ^-^ (5.10)
c
The two eigenmodes then become
A{exp( ~j/3hz) +5+ exp(j/3bz)} exp( -jftz) (5.11)
B{exp(jM +5_ exp( -jfa)} exp(jftz) (5.12)
where /3, is a propagation coefficient offset from the Bragg value, in the
presence of coupling', /3e is to be found for each 5 and g The solution to
any problem with a uniform grating can then be obtained by matching
the eigenmodes to the boundary conditions. The solutions for s+/_ are
found by substituting eqns. 5.11 and 5.12, and into eqns. 5.4 and 5.5 to
obtain

s _[(8+jg)±{(8+jg)KfrKrfn
«
«rf
 Basic principles of lasers with distributed feedback 133

s jv+mwtfK**^ (514)
Kfr

where the ± sign is chosen so that I s+1 ^ 1 and I s_ I ^ 1. The dispersion

relationship between the frequency parameter S and the offset
propagation coefficient f}e is given from
(3e=T{(d+jgf-KfrKrfr (5.15)
Note that, in the absence of any coupling, the (complex) propagation
is just 8+jg. This dispersion relationship, eqn. 5.15, is discussed
shortly.
While the eigenmodes form an elegant mathematical technique,
analysis in terms of the 'forward'- and ' reverse '-field components/ 7 and
/?as in eqns. 5.4 and 5.5 is usually easier. The eigenmodes of eqns. 5.11
and 5.12 retain the variations exp(±jfihz) while the F and R
components have these spatial variations implicit with a uniform /3h
over the whole structure and any spatial variations in the Bragg period
accounted for by changing the S parameter. As an example, whenever
K goes to zero, such as in a short section of uniform material or at an
antireflection-coated facet which makes air appear like a uniform
semiconductor to the outgoing waves, then the continuity of the field
components F and R ensure that the amplitude reflectivity p that is
most straightforward to use is given from p=R/F For this reason the
eigenmodes, although elegant, are not pursued here.

53.2 The dispersion relationship and stopbands

The dispersion relationship given by eqn. 5.15 shows that a pure index
grating with zero gain gon\y has a propagating solution with /3e real for
82>KfrKrf>0, as shown in Figure 5.2. For 82<Kfr Kr/, the coupled wave
equations yield an evanescent solution with /3e imaginary, and the
grating has a strong reflectivity. This region is known as the grating
'stopband'. Within this band, any incident wave is reflected efficiently
and the reflections prevent propagation and instead cause a decay in
amplitude with distance (so-called evanescence with ^purely imaginary
when there is zero gain/loss). Conversely, outside the stopband, when
82>KfrKrf, the phase matching of the grating is too weak to give
constructive/destructive reflection and propagation occurs with an
oscillatory net amplitude for the (forward + backward) waves and /3e
has a real component for all values of gain.
A completely different situation arises for a pure-gain grating for
which Kfr Krf<0. If the mean gain gis zero then there is a propagating
134 Distributed feedback semiconductor lasers

0.5 . g/K
g/K 0 0 8/K 8/K
Figure 5.2 Normalised dependence of real and imaginary parts of fie on 8 and the
field gain gfor a pure index guide
Results are symmetric in 5, and gain has been taken as positive

real (P e / K)
2.5 v

imag (Pe

g/K

Figure 53 Normalised dependence of real and imaginary parts of (3e on 8 and the
field gain gfor a pure-gain grating (real K)
Results are symmetric in 8, and gain has been taken as positive

solution for all 5 (all frequencies), and no stopband exists in the

propagation coefficient )3e as shown in Figure 5.3. It is also interesting
to see that, when the magnitude of the gain exceeds the magnitude of
/c, then a stopband again occurs around the Bragg frequency. Notice
how the real and imaginary parts of (3e are effectively interchanged in
their roles between Figures 5.2 and 5.3 as are g*and 8.
 Basic principles of lasers with distributed feedback 135

8/K
phase (radians)

Figure 5.4 Dependence of the real part of fie on detuning and phase (arg Kgmn/Kindex)
of K with Kr>0
/3 can take positive or negative values, and there is symmetry in 8

In Figure 5.4 the real part of the normalised complex propagation

coefficient fij I K\ is shown to depend on the normalised detuning
5/1 K I and the phase d of the complex grating-coupling coefficient K=
I ACI exp(jd). For zero gain and for pure index-coupling coefficients, a
stopband exists in the range for detuning given by I 8/ K\ < 1 and that
for gratings with a complex coupling coefficient there is a range of
propagation coefficients which are not allowed.
A program dispbrag in directory grating can help the reader to
investigate the real and imaginary parts of the propagation coefficient
normalised to the grating coupling coefficient /?,/1 K I as a function of
the detuning from the Bragg condition normalised to the grating
coupling coefficient 8/\K\. The phase of the coupling coefficient is
taken so that Krf = Kfr but allows for complex K=\ K\ exp(jd). It will be
found that the stopband for a pure-index grating disappears in the
presence of any gain or loss coupling. The region of strongest
evanescence lies in the stop band for pure-index coupling and this
evanescence weakens as the ratio of gain / index increases.

5.4 Matrix solution of coupled-mode equations for

uniform grating laser
5.4.1 The field input-output relationships
The optical fields within a finite length of grating, having a uniform
coupling coefficient and with perfectly antireflection coated facets at
136 Distributed feedback semiconductor lasers

its ends, are now discussed. In Appendix 6 it is shown how the coupled
equations may be solved analytically and directly with a matrix
formulation where the forward and reverse fields are arranged as a
vector with

M ss]
The fields at any position z can be put in terms of the fields at z=0
(5.16)

through M(z):

M(z) = cos (

1 0 0 ~j
(g-jS) — K (5.17)
0 J 0
1

where /32L2={(8+jg)2 — K?}L2 and 5 gives the 'frequency offset'. For an

AR-coated device where F(0) =0 and R(L) =0 then the result simplifies
to require that
F(z) = (JK/pg) sin (Pgz)R(0);
R(z) = [cos(Pez)-{(g-j8)/PH sin(P,z)]R(0) (5.18)
ton(P,L)={pe/(g-j8)} (5.19)
More generally, the fields can be also rearranged to give the output
fields at each end of a laser of length L in terms of the input fields.
This rearrangement requires a little care which is again discussed in
Appendix 6. It is shown that, on writing
a=cos(p,z); d={(g-j8)/j3e} $in((3ez); b=(ic/pf) sin(&z) (5.20)
one obtains
,,2-^-^=1 (5.21)
This permits the matrix inversion to be performed analytically, and
one can write

F(t,L) J_\ 1 Jftir^/.O) 1 (5.22)

a-d\_jb \\\_R(t,L)\

Note that the idealised mathematical 'oscillation' condition is given

 Basic principles of lasers with distributed feedback 137

when there can be a significant output in the left-hand terms even

though there is a negligible (zero) input (right-hand terms), i.e. where
a— d=0 or
tan(P,L)/(P,L) = l/[(g-j8)L} (5.23)
In this form, the choice of the sign of the square root to determine (3e
is manifestly of no consequence and of course eqn. 5.23 is the same as
eqn. 5.19. The reader may check that with 5=0 and real K, f}e is
imaginary with I f}pL I >gL and it is not possible to solve eqn. 5.23 for any
gain g (of course g is real). However, two solutions are possible for
I 8\>K having Pe=Per+jPd a n d /S^/S^—jP^ giving two modes, one at the
upper- and the other at the lower-wavelength edge of the stopband.

5.4.2 Reflections and the observed stopband

With input fields Fin on the left and JR,-n on the right, then with facet
coatings giving negligible reflections one matches Fin=F(0) and
i^n=i?(L) for a laser of length L. A powerful matrix technique for
undertaking this task is given in Appendix 6. The program refl in
directory grating allows the reader to explore p=R/Fat the input for
real K. Figure 5.5 shows this reflectivity as a function of 8/\ K\ for a real
K so that one can see how the reflectivity falls off as 5/1 /c I diverges from
zero.
The stopband is, strictly speaking, the region where the fields
evanesce. With this definition the width is determined solely by the
grating-coupling coefficient, as described in Section 5.4.1 and, for real-
index gratings, is given by
151^ I/cl (5.24)
In practice the term 'stopband' is used for the spectral separation
between the two peaks in reflection which become much larger than
unity just before lasing. For a uniform grating, these peaks are situated
symmetrically just outside the strict stopband by an amount which
depends on the coupling coefficient and length of the grating [4,5].
Figure 5.5 with normalised gain of 0.5 and KL=2 shows one of these
peaks just beginning to emerge around <5/IKI~2 outside the strict
mathematical 'stopband'. The other peak would be with the negative
valueof5/kl~-2.
However, there is a further complication in this analysis in that it will
later be found desirable, when designing for a single lasing mode, to
introduce short additional sections of uniform laser which change the
phase of the forward and reverse fields with respect to one another.
The classic case is a phase shift of Am/4 situated in the middle of the
138 Distributed feedback semiconductor lasers

reflectivity
1.5-.

0.5-

Figure 5.5 Reflectivity of a uniform grating of length KL=2as a function ofgain and
frequency deviation
The reflection is symmetrical in 8 so only one-half of the diagram is drawn
to avoid too much detail

laser. Other devices introduce, for example, two Aw/8 phase shifts
appropriately displaced about the centre of the laser. For such devices,
the lasing line lies within a band of amplified spontaneous emission
which has significantly lower amplitude where the reflection sculp-
tures the emission spectrum. It is this band of low spontaneous
emission which is commonly referred to as the 'stopband' and can be
found by numerical calculations as discussed later.
Figure 5.6 shows the dependence of this extended definition of a
normalised * stopband width' (measured as a change in free-space
wavelength 2AAv/0/;) on the KL product of the grating for:
(i) a uniform grating laser,
(ii) a symmetric Am/4-phase-shifted structure, and
(iii) a DFB laser with 2 x Am/8 phase shifts positioned symmetrically in
the cavity and spaced by 70% of the cavity length (see Section
5.6.5).
The curves are plotted as dimensionless groups so as to generalise
them:

27T 1
^normalised (5.25)

The functional form of the normalised stopband width AmmMft((,rf as a

 Basic principles of lasers with distributed feedback 139

10

8--
A
normalised
6
stop band -

• 2x
•••
width /8

normalised
•«•
A uniform

KL
Figure 5.6 Normalised 'stopband width' against KL
The stopband width is defined here as the separation between the two
main peaks of spontaneous emission at high gain and so varies for lasers
with different structures as discussed in Section 5.5

function of KL means that, for low values of KL, AAV/0/, « 1/L, and for
high KL products, AAv/0/; <* /c, i.e. for very short grating lengths the
'stopband width'-—•infinity because there is no frequency selectivity,
while for high KL products the 'stopband width' <* K as one might
expect [6].

5.5 DFB lasers with phase shifts

5.5.1 Phase shifts

This concept of introducing phase shifts into a uniform laser, which
has been mentioned briefly before, is now examined further by
considering steady-state models of such lasers. The discussion begins
by considering the physics of the oscillation condition where, for lasing
action to occur, the net complete round-trip complex gain must be
unity. The phase change must therefore be equivalent to an integral
number of wavelengths. Recalling eqn. 1.4 for a Fabry-Perot laser with
reflectivities p^ and prighl:

ex
PI (5.26)
140 Distributed feedback semiconductor lasers

The confinement factor V has now been inserted and modifies the
material gain; T had not been introduced at the start of Chapter 1.
This general principle that the round-trip complex gain is unity
applies to a DFB laser as to any laser. The facet reflectivities pkfl and plighl
are replaced by equivalent grating reflectivities with appropriate
phases of reflection, and the actual cavity length has to be changed to
an equivalent cavity length. This is the essential physics which is hidden
in the initial lasing equation (eqn. 5.23).
One might think that the strong reflection at the Bragg frequency
fbmgg would ensure oscillation a t / ^ ^ but eqn. 5.23 showed no solution
at this central frequency for real K but instead indicated modes towards
the edges of the 'stopband'. This paradox arises because, although the
reflection and gain are strong enough at fbmgg, the feedback is in
antiphase which prevents lasing action. This phasing can be under-
stood by considering a microcavity formed by one grating 'tooth'
(taken to be AOT/4 long with a low index value) exactly at the centre of
the device, for example with the index stepping to the higher value at
each end of this tooth. All grating reflections at the left-hand end of
this tooth are in phase with each other and can be lumped into one
real reflection (p,oolh ^ <* K) at the step to the adjacent higher-index
tooth. Likewise, all the reflections on the right-hand end can be
collected into a reflection (p(oothrigh) at the step to the adjacent right-
hand higher-index tooth. From symmetry p(oofhrighl=p\oolh ¥=pml gives the
net reflection at the ends of this central tooth of the pure index
grating. Each single pass is Am/4 long (a phase shift of TT/2) giving the
net round-trip feedback with one pass there and one pass back as

{ionrtexp(-;i7r)}2<0 (5.27)

This is negative and so prevents lasing even if pml >1. An additional

180° phase change happens if there is some wave propagation within
the grating (/3e real) which can occur at frequencies just outside the
stopband. For oscillation, this propagation and gain must be such that
pnel forms an imaginary value of jox —j. In the ideal uniform laser, two
main lasing frequencies occur where I (3eL I ~ TT/2, symmetrically placed
about the central Bragg frequency.
An alternative method of obtaining the additional 180° at the Bragg
frequency with a real index grating can be found by inserting a Am/4
phase shift at the centre of the cavity, thus adding 180° to the round-
trip phase change within the microcavity but now keeping pnel real.
Now at the centre of the stopband pwl exp{ - j($7r+\7r)f>0, so there is
 Basic principles of lasers with distributed feedback 141

longitudinal mode approximate relative approximate effective

intensity profile carrier density profile relative permittivity

•t
relative position
L
of 2.5 • 10.735
jump

(a.u.)

--P^- 10.725 • = - ! -
intensity
distance along laser distance along laser distance along laser
400 u m W
spontaneous emission spectrum

schematic structure

Side modes

V4
phase shift
- 400 u m -
laser length
1 565 nm P574

Figure 5.7 Simulated performance ofhigh-KL KJ4 phase-shifted DFB laser operating
at low output power
Output power at each facet 0.37 mW, /cL=3, zero facet reflectivity,
2000 A/cm2 drive current, effective recombination time 4 ns, transpar-
ency density 10 cm"3
18

positive feedback giving lasing. At this central frequency the waves

were seen to be evanescent (fie imaginary) and for moderate to high
values of grating feedback (i.e. KL product > 1.25) the summed
intensities of these waves then peaks at the Aw/4 phase shift and decays
'exponentially' towards each facet, giving a very nonuniform longitu-
dinal-mode-intensity profile, as has been calculated (from numerical
analysis) in Figure 5.7.
With the Aw/4-phase-shifted laser for KL>2, the strong axial
variation in the mode's field is found to give several unsatisfactory
features in spite of providing a clean single lasing line at low output
powers (see Section 5.6). The concept of introducing phase shifts into
the structure at points other than the centre has been found to be
crucial in providing a good engineering solution to some of these
difficulties. It is therefore necessary to examine the analysis of
inserting phase shifts into structures and obtaining the oscillation
conditions.
142 Distributed feedback semiconductor lasers

5.5.2 Insertion ofphase shifts: the transfer-matrix method

Section 5.4.1 shows that by writing

F(z) =

with F(0) known then it is possible to write a 'transfer matrix' Msuch

that
F(z)=M(z)F(0) (5.28)
Now the fields F and R are the forward and reverse waves in a very
short planar-laser section. It follows that, for a phase change of </>, one
can replace this section with a phase jump where F—*Fexp(—j(f));
R—>R exp( +j<f>). The transfer matrix M^ for this phase shift gives
t) (5.29)

where M,= - ^ (5>30)

A phase shift of 4> in the middle of the laser is therefore incorporated

by concatenating transfer matrices as follows:

I ^ UMHA^HMJ I M (5.31)

where Mx is as given in eqn. 5.17 but with a length of L / 2 . If both facets

have zero reflectivity, the following equation for oscillation may be
derived after some algebra:

If 0=0, then the oscillation condition (eqn. 5.23) for the uniform DFB
laser may be recovered. A Am/4-phase-shifted DFB laser is modelled
with 0= (TT/2) or exp(j<f>) -j which is found to yield
 Basic principles of lasers with distributed feedback 143

?£j - (g-jS) (5.33)

The technique may be extended in an obvious manner for a laser with

two phase shifts placed symmetrically in the cavity where now one
would write

^ =[M,] [MJ [M3] [MJ [M,] ^ (5.34)

where Mx and M3 are given by eqn. 5.17 for the appropriate section
lengths. An analytic solution for eqn. 5.34 is not straightforward and it
is preferable to compute the results. However, one can see the utility of
transfer matrices for studying complex structures.
Figure 5.8 shows the simulated field and electron-density profiles
along the cavity of a 2xA w /8 laser analysed as a concatenation of
uniform grating sections. The significant point about the 2xA w /8-
phase-shift DFB laser is that it provides a more uniform field and
electron density along its length over a wider range of current inputs
than the single Am/4-phase-shift structure for appropriate KL products
- 2 , provided that the phase shifts are positioned correctly at
appropriate values of AL from the facets as shown. With a flat field
profile, the gain is more effectively utilised, there are narrower line
widths because of small variations in the refractive index along the
structure and there is less variation in the output wavelength with
power level than for many other types of Bragg laser.
While considering transfer matrices, it is worthwhile making a short
digression to find the transfer matrix for N pairs of teeth of a Bragg
grating, at for example its central Bragg frequency. A transfer-matrix
formulation gives
M= (Mw/4M+(i)ApMw/2M_ ( j )Ap M, /4 ) N (5.35)
where a reflection + 2A/0 from a step change in refractive index at one
position is joined to a reflection of - |Ap by a transfer matrix M^ as in
eqn. 5.30 but with a phase shift of 4>= IT/2 corresponding to the optical
length of Am/4. This unit is then completed with transfer matrices of
phase shift TT/4 on either side to complete the half-wavelength-long
unit which concatenates with N other such Bragg elements. The
elemental transfer matrix for a step change of refractive index,
assuming plane-wave fronts, is M+>A/9 given from
144 Distributed feedback semiconductor lasers

longitudinal mode approximate relative approximate effective

intensity profile carrier density profile relative permittivity
2.5
1
10.683
relative
N
(a.u.) transparent

intensity
10.676
distance along laser distance along laser distance along laser
Spontaneous emission spectrum t~ L = 300 u. m
schematic structure
AL AL
10
relative " -»'« * •
emission
Intensity
10 C X /8 X /8
phase shift phase shift
nm
1560 1570 250 u r n
separation

Figure 5.8 Simulated performance of a moderate KL 2X \J8-phase-shifted DFB laser

emitting 3 mWper facet with a moderate KL (=2.25)
Device parameters are such that there is 3.2 mW output at an input
current18 density
3
of 5000 A/cm2. The transparency density is
1.5 x 10 cm" ; KL=2.25 with a 10 300 |xm overall length, bimolecular
recombination constant of lxl0~ cmVs, Auger recombination con-
stant of 3x 10~29 cm6/s; waveguide attenuation16
of
2
40 cm"1, Henry's
linewidth-enhancement factor of 3, a=3 xl0 cm . Operation of the
device depends on having appropriate values of AL/L

|
(5.36)
-Ap/2 1 J
where T=V{1 — (1A/9)2} and §Ap= (n, - n^/{nx + n^). Then looking
ahead to a future section on Bragg reflectors for VCSELs (Section 8.5),
an estimate for the transfer matrix of 20 pairs of teeth with ^=3.5
and ng=3 (very approximately the values for GaAs and AlAs) is given
from
\20

20
-1.0119 -J0.1548
+70.1548 -1.0119
 Basic principles of lasers with distributed feedback 145

io 9349
R
i
-
-jlO.8891
> i o - 8 8 9 i ir^i
10.9349 J | ^ J
(5>37)
so that, with zero input on the right-hand side giving RL=0,
\ Ro/Fo\2=\ M2l/ M22\2=0.9916 or 99% power reflectivity. This is an
optimistic calculation, neglecting loss and assuming that the optical
energy is confined fully to the regions containing the dielectric steps,
giving the maximum reflection possible. This type of estimate will be
used later in Section 8.5.

5.6 Longitudinal-mode spatial-hole burning

5.6.1 The phenomena

'Spatial-hole burning' is a phrase which is used for the depletion of the
injected-charge carriers where the depletion is caused by strong
stimulated recombination in regions of high photon density. Spatial-
hole burning was first identified in the lateral (horizontal) direction
for broad stripe lasers which relied on gain to guide the optical mode
and also in structures where the lateral-index guiding was weak.
Because of the link between refractive index and charge-carrier
density, the depletion of the charge carriers in regions of strong
photon density increases the real refractive index and reduces the gain
or imaginary refractive index compared with the regions of lower
photon density. This can lead to self focusing in laterally guided
structures with a built-in real-index waveguide, instability in gain-
guided structures, and filamentation (constriction) of the optical
beam in broad-area devices. The strength of the mechanism is, in
general, significantly reduced by lateral diffusion of the charge carriers
because diffusion smoothes out changes in the carrier concentration.
Lateral carrier-hole burning can be detected by changes in the near-
and far-field patterns as the drive current is increased. Strong lateral
spatial-hole burning in general invalidates the approximation that the
confinement factor T for the active region is constant. However, with
sufficiently well designed index guiding, the lateral and transverse
spatial-hole burning effect is not too important and is safely
neglected.
Stimulated recombination will cause longitudinal carrier-hole burn-
ing in both Fabry-Perot and DFB lasers and create nonuniformities in
146 Distributed feedback semiconductor lasers

the carrier density N(z) and photon density S(z) along the z direction.
The changes are relatively slow, so that diffusion does not significantly
moderate the effect and diffusion is ignored. The coupling of electron
density and photon density is found from eqn. 4.2 divided by the
volume T. The drive-current density replaces the drive current
because one now has to allow for nonuniformities in the electron
density N(z). The gain Gm(N) is replaced with the differential gain
expanding in N about the transparency density. The linear-to-cubic
recombination terms of eqn. 4.4 are all considered to give

= l AN(z) - BN(z)2- CN(z)3~ GJN(z) - NJS(z) (5.38)

dt qdd

In Fabry-Perot lasers the redistribution of carriers along the cavity

with increasing optical output power has only 'second'-order effects
and, indeed, if the refractive index and gain vary linearly with carrier
density, the emission wavelength is found not to alter. With conven-
tional facet reflectivities there is only a modest nonuniformity in the
longitudinal photon-density profile and spatial-hole burning is not
usually a problem because the gain and index changes are effectively
averaged out by the requirement that the round-trip complex gain
remain at unity.
The effects of spatial-hole burning in DFB lasers are in marked
contrast to those in FP lasers because local variations in the carrier
density, and hence variations in the real refractive index and gain, give
rise to changes in the magnitude and phase of the feedback from each
section of the grating. All this changes the longitudinal-mode intensity
distribution and also alters the gain suppression of side modes relative
to the lasing mode. The lasing mode then exhibits a nonlinear light/
current characteristic which is accompanied by a frequency shift or
'chirp', as such a shift is commonly called. The output power also takes
time to stabilise following a transient in the current as a consequence
of the time constant associated with the carrier hole burning. This in
turn leads to amplitude-patterning effects under digital modulation.
Another way of viewing the problem of spatial-hole burning is that one
cannot simply move up and down the static light/current character-
istic, with rapid changes in current. Spatial-hole burning means that
the dynamic characteristics of the laser require time to relax back to
the static characteristic on a change of current. The laser designer has
to minimise this time by minimising the hole burning.
 Basic principles of lasers with distributed feedback 147

frequency frequency

upper frequency . upper frequency

Bragg frequency.

stop band (shaded) stop band (shaded)

lower frequency lower frequency

distance along laser distance along laser

Figure 5.9 Schematic of stopband/distance diagram

a Uniform DFB
b With central spacial hole burning
Changes in the carrier density change the refractive index along the laser,
thereby changing the edges of the stopband, for example as in (b). Waves
inside the stopband experience stronger reflections and less gain than
those outside the stopband. Phase adjustments to keep the round-trip
phase can more readily occur at the edge of outside the stopband.
Oscillation therefore occurs at both upper and lower edges of stopband
in (a) but the upper frequency is preferred in (b)

5.6.2 The stopband diagram

The description and physics of longitudinal spatial-hole burning is
considerably enhanced through the concept of a longitudinal stop-
band profile, as one example will show shortly. If the electron density
were uniform then there would be a uniform stopband within the
uniform grating, as indicated in Figure 5.9a. The stopband/wave-
length profile then has a pair of lines symmetrically placed with respect
to the Bragg frequency delineating the 'stopband edges' defined as the
frequencies of the two ideal lasing modes for the perfect uniform-
grating laser with zero facet reflectivities. Moving away from the
stopband, the rate of change of phase per unit length increases while
the reflectivity reduces. In contrast, moving into the stopband
increases the reflectivity and the propagation changes to evanescence
so that the number of phase changes per unit length becomes
negligible unless a phase shift is introduced. To use the stopband
diagram, one must remember that oscillation requires a round-trip
gain of unity and multiples of 2rr round-trip phase.
For a uniform laser with a high KL when the laser is first turned on,
the electron density is uniform and the two modes on either side of the
stopband can start to oscillate (Figure 5.9a). As the optical power
builds up, the optical intensity is highest in the middle of the laser and
there is significant spatial-hole burning with the carrier density
148 Distributed feedback semiconductor lasers

reducing and therefore increasing the refractive index in the centre of

the device. The stopband diagram then shows a marked curvature
towards shorter wavelengths (higher frequencies) at the edges of the
laser (Figure 5.9b). The short-wavelength (high-frequency) mode then
experiences a strong reflectivity near the facets of the laser and
maintains propagation with some phase shift and gain in the middle of
the laser, so that oscillation can be maintained. The long-wavelength
mode, in contrast, now experiences reduced reflectivity at the facets
and reduced gain and phase shift in the middle of the laser so that the
oscillation conditions cannot be maintained.
Stopband diagrams can, with a little experience, be powerful
indicators of the underlying physics in a grating structure, showing
rapidly why a DFB laser changes mode and which mode to expect. As
well as explaining the performance of uniform lasers with high and
low KL [7], stopband diagrams have proved to be most helpful in
understanding high-frequency modal oscillations within uniform
gratings [8] and also in aiding the understanding of push-pull lasers
[9] (see Section 8.3).

5.6.3 Influence of KL product on spatial-hole burning

Uniform-grating DFB laser structures, with perfectly antireflection-
coated facets, have longitudinal-mode-intensity profiles which peak
strongly at the cavity centre for KL, say around 2 or more, and peak at
the facets for values around unity or less. The longitudinal-mode-
intensity distribution is plotted in Figure 5.10 for KL products of 2.4,
1.7 and 1 with the middle value giving a nearly uniform distribution.
However, one finds in practice that the mode profile is highly
dependent on minor defects in the laser or on the facet reflections,
and that in modelling, with a perfectly uniform device with no
reflections, the oscillation mode can vary from run to run depending
on the random build up of spontaneous emission. Indeed longitudinal
spatial hole burning can make the perfectly uniform laser very
unstable with respect to which of its two modes will oscillate and there
have been a number of ideas to force uniform DFB lasers into lasing
on one stable mode such as high-low facets, longitudinal-current
variations, complex gratings and phase shifts.

5.6.4 Influence ofphase shifts on spatial-hole burning

DFB lasers with a centrally placed Am/4 phase shift in general exhibit
strong nonuniformity in their longitudinal photon density. This
 Basic principles of lasers with distributed feedback 149

mode intensity no( intensity

mode mode intensity
arbitrary units arbitrary
irbil units arbitrary units

distance along laser distance along laser

- -

carrier density carrier density—--

Figure 5.10 Dependence of longitudinal-mode-intensity profile of uniform-grating

DFB lasers on the KL product
The photons stimulate carrier recombination so that the change in
carrier density very approximately mirrors the change in photon
density, as shown schematically here, and this in turn mirrors the
changes in refractive index which controls the Bragg frequency and
Bragg stopband/distance diagrams

nonuniformity gains in strength as KL increases. When /<L~1.25, the

longitudinal mode profile is at its flattest. For KL<1.25 the highest
photon density occurs at the facets. Unfortunately, such a grating
strength can pose fabrication problems because imperfections can give
reflections comparable with the grating. Although longitudinal hole
burning is minimised around /cL~1.25, the evanescent nature of the
travelling-wave solutions at the Bragg frequency still gives rise to
significant longitudinal nonuniformity in the mode profile.
Multiple-phase-shift structures, and in particular the 2xAw/8 DFB
structure, have been particularly successful in minimising spatial-hole
burning and maintaining a near uniform field profile. This 2xAm/8
DFB structure normally has the two phase shifts positioned symmet-
rically about the centre of the cavity, as indicated schematically in
Figure 5.8. The fields can be evanescent or propagating depending on
whether the phase shifts are near the centre or ends of the laser
structure. Detailed trade-offs in the operational parameters have been
carried out as the position of the phase shifts and the strength of the
grating (KL) are changed [10, 25]. Figure 5.11 shows contours of the
nonuniformity, or variance, of the longitudinal-mode intensity dis-
150 Distributed feedback semiconductor lasers

0.5 S Uniform
DFB
0.06 0.02 0.006 0.010.02 0.04 0.08

0.1 ••

0.02 0.020.04 0.08 0.140.18 0.25 0.35 0.45^X74.

_J 1 L 1 L—L L f / rDFB

1 2 K L 3
Figure 5.11 Longitudinal-intensity-variance contours for 2 x A J8 DFB lasers
Parameters give variance as measured by k\E\2/\E\2mnx and ordinate
AL/L gives position of phase shifts: note that the values of AL/L give
two distinct modal points; see Figure 5.13

tribution as a function of KL and the separation of the 2 x Am/8 phase

shifts determined from AL/L (see Figures 5.8 and 5.13). The plot in
Figure 5.11 includes all devices ranging from a Am/4-phase-shifted DFB
laser through to a uniform-grating DFB device, in other words all
symmetric 2 x Aw/8 designs.

5.6.5 Spectrum and spatial-hole burning

As indicated earlier, a uniform intensity avoids spatial-hole burning
and leads to a more uniform electron density and uniform complex
refractive index, giving more effective utilisation of the gain and a
narrower linewidth. However, although it is important to reduce the
nonuniformity of the intensity distribution, it is also important
simultaneously to examine the side-mode-suppression ratio. Figure
5.12 shows the side-mode-suppression ratio mapped for the same
variables as in Figure 5.11. The optimum design will then have to
consider how a uniform mode distribution can coincide with large
side-mode suppression.
Figure 5.13 illustrates, for all devices between a Am/4 phase shifted
and a uniform DFB, the trade-off between flatness of the longitudinal
mode intensity distribution and spectrum for varying position of the
2 x Aw/8 phase shifts for a KL product of 1.7.
 Basic principles of lasers with distributed feedback 151

Uniform
DFB

58 62 66 70

Figure 5.12 Side-mode-suppression ratio (dB) at 1 mWfor 2 x \m/8 DFB lasers

Parameters give side-mode-suppression ratio in decibels, ordinate AL/L
giving position of the phase shift

V/8 phase shifts, phase shift

p o s i t i o n ^ AL/L
\ AI AI / 1 1

W AAAAAAAAA/ W i
i
| I J^^<^T ,A 1/1=0.5: n A uniform DFB
\J
* I = 400 u m >jS?
/ I =0:^/4 DFB O g off Bragg'

U
^ 1 • I
^ , ^ ^ ^ • *

= 0.5: uniform DFB ^>**"\^

-1 1 -10v 3 , i U

1 1 / ' relative I - ^ j U mode jump

intensity I
J''
near Bragg*
-IO°-~AAA—
U
^ A I / I =0: " X / 4 DFB
1 m
distance '
1.570 1.576
1 along laser 1

Figure 5.13 Optimisation of longitudinal-mode-intensity distribution and emission

spectrum of2x A J'8 DFB lasers
The laser is 400 (xm long with a drive current of 3000 A/cm2 and
#cL=1.7
152 Distributed feedback semiconductor lasers

Detailed mapping of these parameters as a function of phase-shift

position and KL product shows that a good optimisation is achieved for
AZ/L=0.35 and KL= 1.6. In practice, the advantage of such a structure,
over a Am/4-phase-shifted design with ACL=1.25, is a more uniform
longitudinal-mode-intensity distribution. This is obtained for a higher-
KL product which is, in fact, better matched to the fabrication
technology so that imperfections and errors in fabrication remain
small compared with the reflections from the grating and the wider
stopband also means that the periodicity does not have to be quite so
precise.

5.7 Influence of series resistance

The varying photon density along the cavity of a DFB laser gives rise to
a longitudinal variation in the stimulated emission-current density
which then changes the carrier density through the electronic/
photonic interaction (eqn. 5.38). However, until now, each section of
the laser has been driven with a fixed current. In reality, the current
will depend on the series resistance of the electronic-drive supply and
the series resistance of the semiconductor material between the metal
contact and the actual resistance of the p-n junction. The internal
series resistances can influence the longitudinal-mode spatial-hole-
burning mechanism [11-13]. Even if there were the normal 50 Cl
source impedance for the power drive to the laser, but there was a zero
contact resistance to the n- or p-side metallic contacts with negligible
voltage variation along the contacts, then there would be a flat
electron or hole quasi-Fermi level, respectively, and the electron or
hole densities would also be constant. There could then be no spatial-
hole burning, although in regions of high photon density the local
current density drawn from the contacts would be high in order to
maintain the uniform carrier density. A high resistance to both
contacts leads to a constant injected current density independent of
junction voltage and the carrier-density variation along the cavity is
then maximised.
In practice, the resistance lies between these extremes, and with
present technology it is too large to reduce hole burning effects
significantly, and the current-drive model is appropriate. The effect
can be assessed straightforwardly. If the voltage at the junction reduces
by A Vj~kT/q~25 mV along the cavity, the carrier density reduces by a
 Basic principles of lasers with distributed feedback 153

factor -2. It can be seen that to avoid spatial-hole burning it is

necessary to keep &Vj<kT/q. Now the relationship between changes
in injected-current density A/and junction voltage A V^is given by

A/=—7 (5.39)
Pch

where p( is the resistivity and tc is the thickness of the contact layer. For
jfr-InP-doped material, the resistivity pr( = l/qPl-ihoU) ~8x 10~4 Urn for a
hole density P~lx 1018 cm" 3 and mobility /^o/,~75 cm 2 /Vs [14].
Hence for a contact layer of such material with t(.= 2 |xm and
A/~1.5kA/cm 2 we find that AV^~25meV. The variations of the
injected carrier density along the laser are determined by the
variations in the stimulated photon rate. To avoid significant spatial-
hole burning in such a laser, then, it is necessary for these variations to
be much less than 1.5 kA/cm2. Typical threshold-current densities are
of this order of magnitude, so that spatial-hole burning is likely when
one is operating at two or more times threshold.
If the device was fabricated on a p-type substrate, it would be
appropriate to consider the voltage drop arising from injection
through n-type material. A typical value of the resistivity of n-lnP
doped at lxl0 1 8 cm~ 3 is about 3xl0~ 5 dm [15] which would imply
that longitudinal hole burning could be significantly reduced as a
result of the low series resistance for such a device.
A more detailed analysis involves relating the junction voltage to the
injected carrier density along the laser, taking into account the
resistivity and thickness of the intervening layers, and the voltage
difference between the contact and junction. The relationship
between junction voltage and carrier density could be obtained using
the conventional Fermi-Dirac functions for the occupation probabil-
ity, but this approach requires an integration to obtain the total carrier
density. An alternative is to use the approximate but usefully accurate
analytic relationship between Fermi-level energy and carrier density
for electrons and holes, respectively [16], defining contact parameters
Acm and Bcon from

^-^(1+0.154J -^~
V l\

and
154 Distributed feedback semiconductor lasers

V - V.
con
R
J
\7
(one segment of laser) (one segment of laser)

Figure 5.14 Voltage drive against current drive

The current /= (Vrm - Vj)/Rcon where Vron is the contact voltage and Rmn is
the contact resistance with Vj the junction voltage. If changes in /lead
to changes in Vj given by 8Vj and these are such that SV/t (Vmn - Vj), then
a constant-current input is an adequate and appropriate approximation.
Note that variation of contact resistances from segment to segment of
the laser will also cause variations in the current

exp{ (%v - %fP) / kT}] (5.41)

where (%fN—%j) is the quasi-Fermi energy difference for the conduc-
tion band and (%v — %jP) is the quasi-Fermi energy difference for the
valence band. The total junction voltage V^is given from

where %g is the bandgap energy. The carrier-rate (eqn. 5.38), now with
dA//d£=0 in the steady state and an external applied contact voltage of
Vam, is then modified to become

_/ = (V(:on VJ)J)
= I-N(r)S(z) (5.43)
qd pt(qd

As technology improves to reduce the contact resistances, it will

become important to incorporate this type of voltage drive into laser
models. One can rapidly assess the relative importance by taking any
section of the laser and looking at the Thevenin/Norton transforma-
tion from a voltage drive to a current drive, as in Figure 5.14. If, as the
current changes, the changes in the active junction voltage
SVj<(Vron— Vj), then the current-drive model remains a valid model
because variations in Vj will have little effect on determining the
current. This is equivalent to saying that the dynamic contact
resistance Rc> RjUncllon ( = 8V/SI).
 Basic principles of lasers with distributed feedback 155

5.8 Simulating the static performance of DFB lasers

5.8.1 Light/current characteristics
The dynamic time-domain modelling to be given in Chapter 7 can be
used to simulate the static performance by switching the current from
zero to its drive value and letting the laser settle down. The time taken
to reach equilibrium is limited by the time taken for the electron
densities to equilibrate and can be artificially speeded up multiplying
dN/dt by a factor of around 3-5. Too large a factor for this speed-up
will reduce the natural damping too much and can also induce
numerical instabilities so that caution is required with this approach.
Spontaneous emission can be switched off numerically once the
photon density has been established and this gives 'ideal' profiles for
the field, electron density and spectrum. Small steps in the drive help
the laser to resettle more rapidly and so steady-state-light current
characteristics can be obtained.
However frequency-domain modelling also has an important func-
tion, especially when discussing optical systems over many kilometre
lengths where time-domain techniques require many millions of time
steps to be stored. In frequency-domain analysis, the essence is to
compute the performance over a large number of individual optical
frequencies. It is usually only straightforward for conditions where
there is negligible mixing of frequency components, allowing ortho-
gonality of the spectral components to be assumed. This section
outlines this approach to numerical modelling of lasers which has
been especially useful in examining the detailed side-mode spectrum
at different output powers from the laser facets at specified drive
currents.
The first stage in any numerical analysis is to carefully define the
laser's structure, i.e. dimensions, defining the sections of planar guide
and grating, along with phase shifts between grating sections, and the
coupling coefficient and period of grating sections etc. The two facet
reflectivities and their phases relative to any adjacent grating section
are also important [17]. In the numerical modelling, all devices are
subdivided into subsections which will be assumed to have a uniform
carrier density within them, but with adjacent subsections allowed to
have differing carrier densities. The lengths of the subsections are
chosen to be appropriate to the spatial resolution required. Unlike
time-domain modelling, this spatial step size does not control the
numerical stability, and large spatial steps can be used with adequate
simulation of the device physics with the controlling factor being the
156 Distributed feedback semiconductor lasers

F F F
o M
M+l M+2 P-1 p
R p.
fields fields
section numbers
Figure 5.15 Subdivided laser for transfer-matrix methods
Electron density iVis uniform in each subsection

nonuniformity of the gain along the laser's length. For example, a

400 |xm long DFB laser with KL<2 in time-domain modelling requires
several tens of subsections but often less than ten subsections will
model the same laser satisfactorily in the frequency domain. As KL
increases, the increasing nonuniformity of the gain may require more
sections.
The key feature of frequency-domain modelling is the transfer
matrix, described earlier. The laser is divided into a number of sections
with a uniform electron density, K, g and 5 as already seen in Section
5.4, but allowing these to vary if required from section to section. The
fields along the structure may be related through the transfer matrices
at a given frequency. If the fields are labelled as in Figure 5.15, the
fields in the middle may be related at a number of frequencies through
the concatenated matrices:

(5.44)

At the facets, the appropriate field reflectivities pright and pkfl give
conditions RL=prighlFL a n d F0=pleftR().
The method is iterative and, if there are large numbers of significant
modes, then the iteration takes longer. It is therefore suited to DFB
lasers where there are perhaps only three or four significant spectral
modes which need to be tracked. The iteration starts by ignoring
stimulated emission and calculating the carrier density in each
subsection, allowing for the appropriate carrier-recombination terms
(as in eqn. 4.2 but for static conditions). The effective refractive index
of each subsection is then calculated allowing for the plasma effect and
Henry's alpha factor (Chapter 2) [18] which relates the change in the
real and imaginary refractive index with carrier density:
 Basic principles of lasers with distributed feedback 157

aH =—
(5.45)

Locally this expression is a function of the properties of the waveguide,

in terms of both the material properties and of the waveguiding
characteristics [19]. The change in the active-region refractive index
has to be diluted by the confinement factor of the mode to the active
region in order to obtain the change in the effective guide refractive
index:

(5.46)
2TT

Once the carrier density and refractive index have been found in each
subsection, the round-trip gain and phase, taking some reference
point near the centre of the laser, are computed as a function of
frequency using eqn. 5.44. From these transfer-matrix operations the
forward and reverse fields at each frequency can be computed along
the laser and from these fields the mean photon density at any
frequency in each section may be found.
The peaks in the gain spectrum are the potential lasing modes. All
the potential modes then have to be examined carefully to see which
of these modes are at or above their own threshold gain for lasing. For
each mode that is above its threshold gain, it will subsequently increase
in power and therefore its associated forward and backward field
amplitudes, at the modal frequency, in each subsection are stored
along with the total photon density over all (say, ML) such modes. The
frequencies of these modes around threshold are also stored along
with the photon densities in each section. Using this model, as
discussed later in Section 6.2, the linewidth is 'Lorentzian' and this fact
permits one to fit the solutions from a few frequencies around a
potential mode so as to find the net optical power, thereby reducing
the calculation time. Finding this photon density is an iterative process
in itself, adjusting the carrier density in each subsection to allow for
the mean stimulated emission rate for a self-consistent solution. Note
that the refractive index is not altered at this stage because this would
change the frequency of the mode significantly. The change in gain
introduces only a small frequency shift.
158 Distributed feedback semiconductor lasers
0.8 i
F>ower output, mW waveicmgth shift, nm
measured
measured
— O -
4
" O

0.6
/ / computed - 400 urn
2 c/ ' computed
A/4 phase shifted DFE

0 / ' \ 0.4 I
C) drive current, mA 10
° (5 power output, mW 5
a h
Figure 5.16 Static characteristics
For high KL ( = 3) A/4 phase-shifted DFB showing curvature of the
light/current characteristics, along with the proportional shift of the
emission wavelength across the stopband

The next step is to find the electron density in each subsection from
the steady-state carrier-rate equation, eqn. 4.2 (see also Appendix 4)
modified to take into account the different recombination mecha-
nisms and the total sum of all the photon densities SM in all the ML
modes with different differential gains GmM and transparency densities
NlrM. The electron rate equation in the steady state now reads as

L G; M {MZ)-^M}5M(Z) (5.47)
M=l

The process is iterated until the modal amplitudes and electron

densities have stabilised. Now the refractive indices in each section are
adjusted to allow for the changed carrier densities, the frequencies
then recalculated and stored and the whole cycle reiterated until the
longitudinal refractive-index profile has stabilised, giving the solution
for the selected drive current.
This approach to solving an arbitrary DFB laser structure works very
well where the device gives stable lasing in a single mode (see Figure
5.16). However, when a side mode increases in intensity, for example
because of longitudinal-mode spatial-hole burning, the model can
sometimes be unable to find a stable distribution of power between the
main lasing mode and the side mode. Even with high side-mode-
suppression ratios, intrinsic longitudinal instabilities in some grating
structures occur [20, 21] and then it is necessary to use, say, a time-
domain analysis to understand the dynamics of the instability.
 Basic principles of lasers with distributed feedback 159

1 2 M N
Figure 5.17 Spontaneous fields in Mth section (reflection and transmission)

5.8.2 Simulation of emission spectrum

When one requires detail for the spectra with kilohertz resolutions
around the central line, time-domain methods at present give
unacceptably long computation times, at least for desktop computing
by impatient researchers! An alternative method examines the
frequency domain and concentrates the computational resources
around the required limited range of frequencies. Simulations are
possible of both subthreshold spectra [22] and spectra of devices that
are lasing [23]. The laser is first broken down into short subsections
where the electron densities can be considered to be sufficiently
'uniform' and the self-consistent steady-state field solution of a DFB
laser at a specified current can be found, as in Section 5.8.1.
The approach then is to calculate, over some requisite number of
frequencies, the complex-field reflectivities of the whole laser struc-
ture, for both the forward- and backward-propagating waves (pMf, pMr),
as seen from the centre of the Mth subsection into which the laser has
been divided (Figure 5.17). In addition, the complex transmission
coefficients (TM/, TMr) from the centre of the subsection to the
emission from each facet must be calculated. These reflection factors
and transmission factors through to the output should include any
facet reflectivities (prighl, p^) at the right- and left-hand facets,
respectively, which contribute to the overall reflectivities and trans-
missivities (in some of the referenced work facet reflectivities are put in
separately).
The spontaneous emission is considered to be excited by random
uncorrelated spontaneous fields (EMf, EMr) generated at the centre of
each subsection. These excitations are linked to the random uncorre-
lated spontaneous excitation 'currents' iv/;/and isjn in eqns. 5.6 and 5.7
by (EMp EMr)=* SM(ispp ispt) where SM is the length of the Mth section.
The reverse wave EMr is immediately reflected into a forward wave
pMr EMr which adds to the forward wave EMf to give a net forward
component (pMrEMr+EMf) which is then transmitted to the right output
as TMf(pMrEMr+EMf) and also reflected into a reverse-wave component
160 Distributed feedback semiconductor lasers

pMj(pMrEMr+EMj). This latter field component is again reflected as

PMJ (pMr^Mr^^Mf) a n c l a l s o transmitted to the output as
/MTPM/ (PMr^Mr^EMf) etc. This process of reflections bouncing back
and forth is sometimes referred to as a 'bounce diagram'. A geometric
series of all these reflections and transmissions builds up to give an
output, at the right-hand facet, of

[TMJ(pMrEMr+EMf)/{l - (pMrpMf)}] (5.48)

M

The convergence of the series is assured because physically lpMr/

for practical lasers with a finite spontaneous emission although
\{l~~(pMr pM/)}l2^>l. The correlations of eqns. 5.8 and 5.9 are
simplified to give time-averaged products of the random spontaneous
fields from
I EM,,*EM,=Em*EMj=EMffEM=0

where M'*M" and l£^l = l£Ll = l£|Zl (5.49)

2
The value of the mean square spontaneous emission \E Msp I depends on
the electron densities in the Mth section and evaluation of such
spontaneous emission is discussed in Appendix 9 and Chapter 7. As
each section produces uncorrelated fields excited by the spontaneous
emission, the total output power from the right-hand facet is the sum
of all the spontaneous emission powers from each section (including
forward and reverse fields):

(5.50)

For the output from the left-hand facet, the subscripts / and r are
interchanged.
It can be seen that, if there is zero spontaneous emission, then a
finite output requires 11 — pMrpM/l2=0. The fact that (PMTPM/) is a
regular complex function of frequency means that this is zero only at
discrete real frequencies which define the permissible lasing fre-
quencies. With finite spontaneous emission, one discovers, around the
lasing frequency a)0, that II -pMrpMf\ = l - GMr-j 2(co- (o0) YMr where
GMr and YMr are parameters associated with the round-trip gain (see
 Basic principles of lasers with distributed feedback 161

relative intensity of relative intensity of

1000 spontaneous emisson 1000 spontaneous emisson
. 300 \im
0%
100 100 32% uniform DFB

1.54 mW output 5.24 mW output - * * " 30 °facet phase

left hand facet right hand facet
10 10

IAAA
1/10 1/10 I I I j I
1575 158O 1575 1580
wavelength, nm wavelength, nm

Figure 5.18 Typical emission spectra

Uniform DFB with high-low power reflectivities of 32% and 0%,
KL=2.45, 7/Ar= 23.8 mA (2263 A/cm 2 ), spectra shown for current drive
7=42 mA

also Section 6.2) as seen from the centre of the Mth section and (o0 is
the same 'lasing' frequency for all values of M. This form of the
denominator, as a function of frequency, ensures that the line shape of
any semiconductor laser is effectively a Lorentzian line shape within
the limitations of the model; and fitting a few frequencies to this shape
permits the prediction of the output around the lasing frequency,
without the need to compute the precise maximum output frequency.
Moreover, it is found that, if 11 - pMrpMj I2=O for the Mth cell, then it is
also zero for the M'th cell and the calculation of the lasing frequency
is independent of which cell is chosen as the 'central' cell but is
determined by the total effect of all the sections. However, the
closeness of the denominator to zero in eqn. 5.50 sometimes leads to
errors unless the local gain is iterated appropriately. Figure 5.18 shows
typical emission spectra calculated by this method, which has also been
formalised into a technique referred to as the power-matrix method
[24]. Figure 5.166 shows the spectral relative shifts within the stop
band for a AM/4 phase shifted DFB.

5.9 Summary
The chapter started with a discussion of the physics of the grating-
coupling process, a section which should be read in conjunction with
work in Sections 1.6 and 4.4. This section gave alternative insights into
162 Distributed feedback semiconductor lasers

this 'coupling' physics and extended the concepts to gratings with

periodic gain variations as well as periodic index variations. The
eigenmodes of the coupling equations provide an elegant solution but,
when concatenating different sections with phase shifts and facets, it is
usually more straightforward to use the 'forward' and 'reverse' field
components of the coupled-wave equations. Inside a 'stopband', these
field components evanesce rather than propagate. The term 'stop-
band' was extended from this mathematical concept of evanescing
fields to a term indicating the range of frequencies between the two
spectral modes at the edges of the stopband in a DFB. These two
modes can usually be observed in a careful spectral measurement close
to threshold looking at the external spontaneous emission which is
sculpted by the frequency selective reflections within the laser cavity.
Stopband/distance diagrams were introduced giving analogies with
the semiconductor counterpart of energy-band/distance diagrams.
Matrix methods of solving the fields at arbitrary frequencies were
developed and used to explore the stopbands along with the addition
of phase shifts within the laser structure. The concept of spatial-hole
burning was then discussed along with the use of the phase shifts to
reduce the effects of spatial-hole burning and increase the field
uniformity and also improve efficiency and linearity.
Practical laser technology is always improving, and in modelling
some modern well constructed lasers it is important to note that the
very low contact resistance can lead to voltage-driven lasers rather than
current-driven lasers and methods that can be used to extend the work
in this book (where laser models are primarily current-driven) were
discussed.
Finally, a discussion was given on the use of matrix methods of
reflection and transmission to construct the light/current character-
istics and emission spectrum starting from the random spontaneous
emission fields that excite the laser.

5.10 References
1 KOGELNIK, H., and SHANK, C.V.: 'Coupled-wave theory of distributed
feedback lasers',/ Appl Phys., 1972, 43, pp. 2327-2335
2 THOMPSON, G.H.B.: 'Physics of semiconductor laser devices' (John
Wiley 8c Sons, 1980), pp. 480-481
3 WANG, S.: 'Principles of distributed feedback and distributed Bragg-
reflector lasers', IEEEJ. Quantum Electron., 1974, 10, pp. 413-427
4 THOMPSON, G.H.B.: 'Physics of semiconductor laser devices' (John
Wiley & Sons, 1980), pp. 501-502
 Basic principles of lasers with distributed feedback 163

5 WHITEAWAY, J.E.A., THOMPSON, G.H.B., COLLAR, A.J., and ARMIS-

TEAD, C.J.: 'The design and assessment of A/4 phase-shifted DFB laser
structures', IEEEJ. Quantum Electron., 1989, 25, pp. 1261-1279
6 IGA, K.: 'On the use of effective refractive index in DFB laser mode
separation',//m./ Appl Phys., 1983, 22, p. 1630
7 ZHANG, L.M., and CARROLL, J.E.: 'Dynamics and hole burning in
uniform DFB semiconductor lasers', Int. J. Optoelectron., 1993, 8, pp.
279-291
8 MARCENAC, D.D., and CARROLL, J.E.: 'Comparison of self-pulsation
mechanisms in DFB lasers'. Presented at (Baltimore) IEEE Lasers and
Electro-Optics Society 7th annual meeting, 1994, Paper SL8.5
9 NOWELL, M.C., CARROLL, J.E., PLUMB, R.G.S., MARCENAC, D.D.,
ROBERTSON M.J., WICKES, H., and ZHANG L.M.: 'Low chirp and
enhanced resonant frequency by direct push pull modulation of DFB
lasers', IEEEf. Sel Top. Quantum Electron, 1995, 1, pp. 433-441
10 WHITEAWAY, J.E.A., GARRETT, B., THOMPSON, G.H.B., COLLAR, A.J.,
ARMISTEAD, C.J., and FICE, M.J.: 'The static and dynamic characteristics
of single and multiple phase-shifted DFB laser structures', IEEEJ. Quantum
Electron., 1992, 28, pp. 1227-1293
11 LASSEN, H.E., WENZEL, H., and TROMBORG, B.: 'Influence of series
resistance on modulation responses of dfb lasers', Electron. Lett., 1993, 29,
pp. 1124-1126
12 CHAMPAGNE, Y, and MCCARTHY, N.: 'Influence of the axially varying
quasi-Fermi-level separation of the active region on spatial hole burning
in distributed-feedback semiconductor-lasers',/. Appl. Phys., 1992, 72, pp.
2110-2118
13 BANDELOW, U., WENZEL, H., and WUNSCHE, H-J.: 'Influence of
inhomogeneous injection on sidemode suppression in strongly coupled
DFB semiconductor-lasers', Electron. Lett, 1992, 28, pp. 1324-1326
14 HAYES, J.R., ADAMS, A.R., and GREENE, P.D.: 'Low-field carrier
mobility', inPEARSALL, T.P. (Ed.): 'GalnAsP alloy semiconductors' (John
Wiley 8c Sons, 1982), chap. 8, p. 204.
15 HAYES, J.R., ADAMS, A.R., and GREENE, P.D.: 'Low-field carrier mobility'
in PEARSALL, T.P. (Ed.): 'GalnAsP alloy semiconductors' (John Wiley &
Sons, 1982), chap. 8, p. 202
16 UNGER, K. 'Spontaneous and stimulated emission injunction lasers', Z.
Phys., 1967, 207, pp. 322-331
17 SODA, H., and IMAI, H.: 'Analysis of spectrum behaviour below the
threshold in DFB lasers', IEEEJ. Quantum Electron., 1986, 28, pp. 637-641
18 THOMPSON, G.H.B.: 'Physics of semiconductor laser devices' (John
Wiley & Sons, 1980), pp. 535-537
19 OSINKSI, M., and BUUS, J.: 'Linewidth broadening factor in semi-
conductor lasers - an overview', IEEEJ. Quantum Electron., 1987, 23, pp.
9-28
20 SCHATZ, R.: 'Longitudinal spatial instability in symmetric semiconductor
lasers due to spatial hole burning', IEEEJ. Quantum Electron., 1992, 28, pp.
1443-1449
164 Distributed feedback semiconductor lasers

21 GOOBAR, E., RIGOLE, P.J., and SCHATZ, R.: 'Correlation measurements

of intensity noise from the two facets of DFB lasers during linewidth
rebroadening', Electron. Lett, 1992, 28, pp. 1542-1543
22 SODA, H., and IMAI, H.: 'Analysis of spectrum behaviour below the
threshold in DFB lasers', IEEEJ. Quantum Electron., 1986, 22, pp. 637-641
23 WHITEAWAY, J.E.A., THOMPSON, G.H.B., COLLAR, A.J., and ARMIS-
TEAD, C.J.: 'The design and assessment of A/4 phase-shifted DFB laser
structures', IEEEJ. Quantum Electron., 1989, 25, pp. 1261-1279
24 ZHANG, L.M., and CARROLL, J.E.: 'Large signal dynamic model of the
DFB laser', IEEEJ. Quantum Electron., 1992, 28, pp. 604-611
25 KINOSHITA, J., and MATSUMOTO, K.: 'Yield analysis of SLM DFB lasers
with an axially-flattened internal field', IEEEJ. Quantum Electron., 1989, 25,
pp.1324-1332
 Chapter 6
More advanced
distributed feedback laser design

6.1 Introduction

Chapter 5 considered some basic features of DFB lasers, concentrating

on the static performance. However, several key features for high-
performance lasers were not discussed such as linewidth, the influence
of reflections and especially the phase of weak-facet reflections, the
role of complex gratings and what happens on designing for power
levels in the hundreds of milliwatt range rather than the milliwatt
range of conventional communication lasers. These more advanced
problems of the static design of lasers are outlined here and the
chapter ends with a discussion on some results of modelling the
dynamic performance of DFB lasers, considering problems associated
with carrier transport into quantum wells. The dynamic performance
of DFB lasers highlights yet further the problems that have already
been met with a uniform grating in a uniform DFB laser. Change of
mode with time, dynamic instabilities, yield of devices with the right
mode etc. prove to be major problems unless the laser has additional
features.
As seen in Chapter 5, the use of one or more phase shifts offers
considerable improvement in the performance of a laser, and this
chapter concentrates on the dynamic performance of a DFB laser with
two phase shifts. It is generally agreed that one phase shift [1],
although stabilising the mode at low power levels, does not offer the
solution to all the dynamic problems. Three phase shifts have been
advocated [2], and continuous changes in the grating period/
coupling constant along the laser can be of interest [3, 4] but, from a
practical manufacturing view, two phase shifts offers a good com-
166 Distributed feedback semicondu ctor lasers

promise between keeping the design as straightforward as possible

while maintaining excellent dynamic performance over a range of
power levels [5]. While the theoretical advantages for gratings with
periodic changes in gain as well as periodic changes in refractive index
have long been recognised [6], it is only relatively recently that the
technology has indicated that such complex gratings are another
practical way of achieving the required improvements in dynamic
performance of the uniform DFB [7]. These then are some of the
topics investigated in this chapter.

6.2 Linewidth
6.2.1 General
The spectral channels in wireless-communication systems are deter-
mined by precise radio or microwave carrier frequencies with the
spacing between these channels determined by the linewidth of the
modulated carrier. At radio frequencies, the modulation bandwidth
dominates this linewidth so that one can say (for the amplitude
modulation format at a maximum frequency of fM and allowing for the
upper and lower sidebands) that the modulated carrier linewidth is
close to 2/Jvf. The principle remains with other modulation tech-
niques—the modulated linewidth is determined by the modulation
and its format. Channel spacing follows with appropriate spectral
'guard bands' to ensure that negligible interference from neighbour-
ing channels can statistically occur. The DFB laser, with its ability to
determine the optical frequency, can define channels in an optical
communication system so that its linewidth and the accuracy of
specifying its frequency become important system parameters in
determining minimum optical channel spacing for a WDM system.
Unfortunately, even when the modulation frequencies fM~0 the
linewidth of a DFB laser is significant and cannot be ignored. But
worse, as the modulation frequency increases then the carrier's
linewidth broadens much more than the value of 2fM that would be
considered for the classical radio system mentioned above. The laser
exhibits frequency shifts called dynamic chirp because of changes in
refractive index caused by changes of gain as the laser is modulated, as
outlined in Sections 2.5 and 4.2 (Figure 4.4). On switching the laser on
and off there are changes in gain, modulated by the photon-electron
resonance; hence the laser is frequency-modulated at the photon-
electron resonance frequencies by an amount which depends on the
 More advanced distributed feedback laser design 167

carrier microwave carrier diode laser

oscillator
amplitude \\ sidebands amplitude
sidebands

sidemode

a m p.i,ude

Figure 6.1 Schematic spectrum of modulated microwave oscillator and diode laser

linewidth-broadening factor aH and the change in gain (eqn. 4.15).

The different sidebands to the modulated laser's spectrum are then
very different from those of a classical oscillator and play a more
dominant role. Figure 6.1 indicates schematically these differences of
spectrum of a modulated classical (microwave) oscillator and a
modulated semiconductor laser taking the modulation frequency fM to
be a few hundred megahertz in both examples.
Consider initially the unmodulated laser. The complex optical
amplitude from a laser can, to a first approximation, be thought of as
performing a random walk (see Figure 7.5). Fourier analysis of the
laser-output field E(ml(t)=E0(t)cos[2irfl+(f>(t)} typically gives a Lor-
entzian power spectrum, at least to a useful first approximation, as shown
in Figure 6.2 (showing both linear and logarithmic representations)
along with a schematic side mode for discussion shortly.
In standard electrical systems, the full width at the half-maximum-
power points or FWHM measurement of the spectrum is often used. In
optical systems, it is sometimes the — 20 dB spectral width which is of
concern. In well constructed DFB lasers, the steady-state — 20 dB
linewidth can be as low as 1-10 MHz or, in terms of the incremental
wavelength, AA-1/1000 A (10~ 13 m). As already noted, on modulat-
ing the laser drive current, the linewidth-enhancement factor aH
causes the modulated linewidth to increase many times over the ideal
value of 2fM discussed above but also in a laser, with a high aH it can
increase the steady-state linewidth by an order of magnitude or more.
This section aims to explain some of these effects using fundamental
concepts.
Another important feature for any spectrum is its sidemode-
suppression ratio (SMSR). Sidemodes always appear in any optical
spectrum but, provided that the net power associated with these is
more than 30 dB below the power in the main optical mode, they are
not usually significant in the performance of most systems. Figure 6.2
168 Distributed feedback semiconductor lasers

frequency CO/2TC • frequency CO/2TC

a b
Figure 6.2 Linewidth and SMSR
A Lorentzian line gives a mean power-frequency relationship according
to
P(a))=P((omnx)/{l + (a)- (omix)2/(TrbLWf)2} (o>=277/)
so that the separation of the half-power points gives the FWHM of A7Mf.
(a) Linear ordinate (b) Logarithmic ordinate—frequency scale the
same.
The sidemode-suppression ratio (SMSR) is the ratio of the power in the
largest sidemode to the power in the fundamental. SMSR= - H decibels
in(b)

indicates an unsatisfactory SMSR of around - 15 dB. Sidemodes are

also considered in this chapter.

6.2.2 Calculation of linewidth under static conditions

A full dynamic time-domain model as that outlined in Chapter 7
(which allows for spatial distribution of noise sources and fields and all
significant frequencies) can automatically determine the linewidth
under both static and dynamic conditions, and demonstrate effects of
the laser's structure on the linewidth along with phenomena such as
linewidth broadening and rebroadening. The TLLM method [8] has
provided an appropriate dynamic model demonstrating the key effects
but, for the present, the aim is to give the reader a good understanding
of the physics as a well as an ability to compute laser behaviour.
The detailed explanation of the linewidth for a semiconductor laser
requires considerable discussion [9] pioneered by Henry [10-12]. The
analysis here gives a simplified physical explanation which is suffi-
ciently general to apply to most semiconductor lasers. By way of
introduction it is easier to consider first an idealised model for the
linewidth of a uniform laser (represented approximately by a single-
 More advanced distributed feedback laser design 169

moded Fabry-Perot laser). The calculation is done in stages. The first

stage is to regard the laser output as amplified spontaneous emission,
with feedback centred around a single spectral line at a)0. Consider
then the right hand optical field E{cS) just inside the laser with its
net round-trip complex-field gain given from the expression
Ground exp{ -2j/3(a))L}^{ Ground-j(a)~ co0) (2 Gmund L/vg)} for small
changes in frequency (co- (o0) from the lasing frequency co0. Here, to
obtain the right phasing of the feedback, exp{- 2j(3(co0)L} = l, i.e. a
whole number of wavelengths for the round trip in the medium. In the
absence of any significant spontaneous emission, the ideal lasing
condition is that the total round-trip complex gain is exactly unity.
However, in the presence of spontaneous emission—represented here
as additional fields Esp{oo, z) which also experience a round-trip gain
Ground but with a relative phasing which depends on the time and
position of emission—the oscillation requirement to match the net
field inside the laser on one round-trip pass is given from an
expression of the form:
E(a>) {Ground~j2((o-(o0) Yr}+ZzGrmind exp{ - 2jp((o) (L±z)}Esp(co, z) Sz
= GlhMco) (6.1)
where, for the steady-state oscillation, one may write the threshold
round-trip gain, Gthr=l and Yr=(Gwund Lng/c). The term (Lng/c) is the
effective value of the group transit time through the laser and, as this
increases, the stored electromagnetic energy U increases in the laser.
As noted in Sections 3.3 and 5.2, phase factors, multiplying random
and spontaneous emissions which are uncorrelated over the length of
the laser, do not alter the spatial autocorrelation and correlation
functions so that, to a first order, where GWMW^— Glht=l, one can average
the spontaneous emission over the whole laser and write the internal
field close to a facet in the general form
E(a))=Esptol((o)/{Glhr- Ground+j2(co-(D0)Yt} (6.2)
This result can also be interpreted as a circuit theory for a laser [13]
where (Gthr- Grmnd) is the net conductance and 2{co~ co0) Yris a circuit
susceptance. Work by Slater [14] shows that many electromagnetic
oscillators can be interpreted using such an equation, where (o0Yr=
Q Gthr with Q giving the quality factor of the circuit in the absence of
material gain or loss. Hence Q gives a measure of the stored energy
which oscillates back and forth between electric and magnetic forms at
twice the oscillation (lasing) frequency. This circuit theory then
confirms the link already made between stored energy Uand the term
Yr through the simpler argument using the group transit time.
170 Distributed feedback semiconductor lasers

The round-trip complex-gain terms Ground and Yr=GthrQ/a)0 are

properties of the oscillation/lasing mode centred at the optical
frequency co0. The normalised output field may then be written as
E0Jco) = TEspiol(aj)/{l - Ground+j2(co-u0)Q/a>0} (6.3)
where Tis an appropriate transmission term linking the interior fields
with the fields emitted from the facet. At this stage it is helpful to
normalise \E0Ut I2 8co to be proportional to power in the range Sco. With
spontaneous emission noise approximately uniform and covering a
large enough band of frequencies and T varying slowly with frequency,
eqn. 6.3 reveals that all such oscillators are expected to have a
'Lorentzian' line shape which is a term given to any laser where the net
power output per unit frequency as a function of frequency is given by
the general form
Pml((o) = I TEspil)l I 2 / [{1 - Gmm/+{2Q((o- a>0)/a*f] (6.4)
The — 3 dB linewidth A,wfm3.y be found directly:
2TTALWf=co0(l-Gmuml)/Q_ (6.5)
The net power output PnHml is found, using a standard integral
2
t
D =
f"
I
P
out(<») d<° = Pnetsp / ( 1 ~ Gf0Und) (6.6)
net out I

where Pnetsp = 7T(o01 TEsptot I 2 /2Q gives the net spontaneous emission over
a spontaneous emission-noise bandwidth -TTTCOQ/2Q in this model.
Hence the linewidth may also be written as

^Lwf/fo =Ket sp / ( QPne, out ) (6.7)

The result of eqn. 6.7 is essentially the well known Schawlow-Townes
formula [15] for the linewidth of a laser. This circuit model shows that
the linewidth decreases with increasing output power (provided that
the laser remains in a single mode) and decreases as the ability of the
laser to store energy increases (e.g. increased length of laser so that the
laser's equivalent circuit (Mactor increases). Unfortunately, this
straightforward theory that is useful for gas and solid state lasers as well
as microwave oscillators usually gives over an order-of-magnitude too
low an estimate of the static linewidth and a totally inadequate
approximation to the dynamic linewidth, and this must be explained.
 More advanced distributed feedback laser design 171

6.2.3 Linewidth enhancement

In Section 2.5 and Appendix 7, the relationship between gain and
phase was discussed and, in particular, it was shown that, because of the
Kramers-Kronig relationships, gain changes have to be accompanied
by phase changes. In the first instance, the relative changes in gain and
phase are associated purely with the properties of the material. An
increase in electron density AN results in an 'increase' Aw in the
complex refractive index, Aw=Awr + An i , where for small enough
changes,

(6.8)

with aH(>0) known as Henry's linewidth-enhancement factor. The

negative sign shows that an increase in electron density decreases the
magnitude of the refractive index. Values of aH vary around 4—8 for
bulk material and around 1-3 for quantum-well material and also vary
with wavelength. In eqns. 2.42 and 4.15 it was recalled how the value of
aH caused changes in gain to induce frequency chirp, directly
broadening the modulated linewidth over and above the conventional
Fourier limit. Less obviously, aH also enhances the linewidth in the
steady state and the previous analysis can be modified to help explain
this effect.
The first essential feature to understand is that clamping the round-
trip gain to unity makes the feedback process in the laser stabilise the
intensity of the optical fields. The spontaneous emission events
[Esp (z, (o) Sz] increase the optical intensity and to stabilise the intensity
the sum of these emission events creates a small decrement 8Ground sp{co)
in the round-trip field gain. Eqn. 6.3 therefore is rewritten as
Enul(co) {1 - Gmnnd+j2(co-u0) Q/OJ0}- 8GwundspEml(oj) =0 (6.9)

with 8Gwundsp Eoul O ) replacing TEspifll(oj) in eqn. 6.3 and recognising

the slight change in the round-trip gain caused by the spontaneous
emission. However, once one recognises such a change in the real part
of the round-trip gain one sees from eqn. 6.8 that there has to be an
associated phase change modifying eqn. 6.9 further to become

EOUI(OJ) {1 - Gmund+j2((o- ^Q/oo}-8G^#(l+jaH)E0M((o)=0 (6.10)

This phase change in the spontaneous term is now no longer random
172 Distributed feedback semiconductor lasers

as in eqn. 6.1, but is caused by the effect of the combined spontaneous

emissions. Replacing SGroundsp Eoul (to) in eqn. 6.10 with TEspt0( (co) once
again then modifies eqn. 6.4 to give

rfH) \7£sf,lof/l[l-GmiJ2+[2Q(a>-co0)/a>0f] (6.11)

and with similar assumptions as before leads to a linewidth

oul) (6.12)

the linewidth enhancement by a factor of (1 + a^) being caused by

additional phase noise through the term j a ^ in eqn. 6.10. In spite of
this enhancement factor, practical linewidths for good DFB lasers can
be reduced to the 100 kHz range with frequency selective feedback.
Recapitulating, a random increase in the spontaneous emission
reduces the carrier density and hence reduces the gain which reduces
the mode intensity allowing the carrier density to recover to re-
establish the round-trip gain close to unity. Fluctuations in the net gain
are limited, and it is the phase fluctuations which are important and
add to the linewidth through frequency modulation. Although the
round-trip gain Ground is clamped close to unity, the net overall gain for
the spontaneous emission is determined from (1 — Gwund)~l and this
value peaks at values ~103 — 104 at frequencies close to the centre of
any modal frequency and this large value of net gain will explain why
reflections even as low as 0.1% can still have considerable effect on a
laser's performance. The reader wishing for more rigour is advised to
read Henry's original work [10-12].

6.2.4 Effective linewidth enhancement

To explain the values of aH, as measured by an observation of
linewidth broadening in DFB-laser structures, the structural effects
caused by the grating have to be recognised as also changing the
effective value of aH from the intrinsic value for the basic material. The
problem is that the gain and phase terms, in the denominator of eqn.
6.11, have to be calculated for a complete pass around the laser cavity.
However, in a DFB laser, changes in the carrier density alter the
strength and phase of the feedback from the grating. The effective
value of aH in eqn. 6.12 is then a composite value which takes into
account the round-trip average over the whole structure and includes
the effects from the material, from the waveguide and from spatial-
hole burning.
 More advanced distributed feedback laser design 173

The generality of eqns. 6.10 and 6.11, which go back to the roots of
electromagnetic theory, forms a basis to estimate or measure the
effective value of aH for arbitrary laser structures. Consider a change
in output intensity which requires a small change 8Ground in the round-
trip gain of the laser; then this has to be accompanied by a phase
change to give a total round-trip change of complex gain of
8Ground(l+jaH). The oscillation/lasing condition (neglecting changes
in the spontaneous emission) is therefore effectively
E()U((co) [(1 - Ground - 5G r a w w J+72Q{(w- OJ0) / co0}-j8GroundaH]
-~(l+jaH)TEspiol=0 (6.13)
From eqn. 6.13, the change in intensity causes a change in the central
frequency of Sfo/fo = — aH(8Grmnd/2 Q). Further, because the intrinsic
3 dB linewidth is (A/VK/) = &>0(l — Ground)/27rQJ this also has to change by
8(AIwf) = — a)QSGmind/27TQ. The effective value of aH therefore may be
determined from 8f0 and <5(A/W/) to give

This value usually will not be the same as the material's intrinsic
linewidth-enhancement factor given in eqn. 6.8 but is modified by the
field distributions and waveguiding within the laser.
To use eqn. 6.14 to find numerically the effective linewidth-
enhancement factor, first calculate the steady-state solution (for
example as in Figure 6.3a for a AOT/4 phase-shifted laser with KL>2).
The mode intensity everywhere is, say, enhanced by ±1% and the new
charge carrier densities N+ (z) and N_ (z) are recalculated from, say,
eqn. 5.38. These then change the local refractive index n+ (z) and
n_ (z) which in turn leads to new central frequencies fo+ or fQ_ where
there is unity complex round-trip gain. The two linewidths for these
two conditions are calculated so that the net change in the central
frequency, given from 8fo=fo+ ~fo- , along with the net change in the
full bandwidth S(ALWf) =A, w / + — &Lwf- then give the effective value of
&Heijfromeqn. 6.14.
This approach has been used successfully to predict the effective
linewidth-enhancement factor for 2xA w /8 DFB laser structures with
varying KL product and varying phase-shift position AL/L. Figure 6.4
plots the effective linewidth-enhancement factor for a uniform
(AL/L=4) and Aw/4-phase-shifted (AL/L=0) DFB laser as a function
of output power and the intrinsic linewidth-enhancement factor of the
material. The DFB structure significantly increases the effective aH at
174 Distributed feedback semiconductor lasers

amplified con
spontaneous
intensity co_

Aco
frequency
b

super-linear increase
in power at facets

-U2 distance +LJ2

c
Figure 6.3 Schematic diagrams illustrating general method for calculating effective
linewidth-enhancement factor allowing for structural effects

low power levels where a change in the power level creates more
change in spatial-hole burning than at high power levels. With the
phase-shifted DFB the differences between the effective and intrinsic
values of aH are small at both high and low powers, with larger
differences mainly at intermediate powers.
Another striking example of structural effects on the effective value
of aH is given by a DFB laser when there is a complex coupling
coefficient Kind(,x+JKgain. For such a grating with the notation used in
this book, the peaks in the gain and the refractive-index periodicity
coincide when Kindex and Kgain are both positive ('in phase'). With aH>\
an increase in the carrier density increases the strength of the gain
grating, but weakens the index grating by a greater amount and
reduces the round-trip feedback or round-trip gain. The net differ-
ential round-trip gain with respect to carrier density is reduced thereby
increasing the effective linewidth-enhancement factor. Alternatively,
with Kindrx>0 but K^^ <0 then both the gain and index gratings increase
in strength giving a lower cavity loss. The net differential mode gain is
enhanced thereby reducing the linewidth-enhancement factor and
hence the linewidth.
A more sophisticated analysis of the result given in eqn. 6.14 has
been given for a Fabry-Perot laser [17]. Provided that one uses the
effective aH, eqn. 6.14 still holds regardless of the laser being a DFB or
FP. However, care is needed if one attempts to estimate the linewidth
 More advanced distributed feedback laser design 175

effective value of a
H
----- uniform DFB
14 A/4 phase shifted DFB

12
-\ ^ material value of, a TT
10 - '* **^ ^ n
S
** //^j; _- 10
8 "~ - p
iiij r Tr
6 "" ^ -^7r.
!l"^!U:-^z_=--=— e
4
4

2 2 .-

0 "i i i I I i i l I 11 I l l l I I I I I I I I i i i i i 11 I i j I
0 1 2 3 4
power output/facet, mW
Figure 6.4 Estimated effective linewidth-enhancement values for uniform and k/4
phase-shifted DFB lasers for different values of material linewidth-
enhancement values
Laser is 400 |xm long /with K=75 cm"1. Results at 'zero' power agree with
Amann [16] who assumed no spatial-hole burning

for a DFB laser on a similar basis to the calculation for a FP laser: the
DFB can have half an order of magnitude larger linewidths. As
outlined in Section 2.4.2, when using standard rate equations, it was
necessary to modify the spontaneous-emission terms by the Petermann
factors, Klr to account for enhanced spontaneous emission trans-
versely, and Kz to account for enhanced effects of spontaneous
emission because of longitudinal distributions of optical intensity. The
net effect is a spontaneous-emission factor Ktof = KlrxKz>l. The
reasons for a significant Kz in certain types of DFB lie in these having
end sections acting more as amplifiers rather than as feedback
elements. The factor Klnl is then very approximately a multiplier of the
spontaneous emission that would be obtained in an ideal uniform FP
laser of similar material and length. A similar factor, Kir, which applies
to gain guiding that can occur horizontally under the injecting contact
of a laser structure [18] where also there is an increased spontaneous-
emission rate into the lasing mode leading to a larger linewidth-power
product.
176 Distributed feedback semiconductor lasers

Now it must be emphasised that, if one uses a dynamic time-domain

model, similar to that in Chapter 7 where longitudinal spatial
distributions are taken into account and also good index guiding is
assumed, then one inserts the material's intrinsic value of aH into the
calculation and the main effects referred to above automatically
produce the right overall effective value of aH with an appropriate Kl0(
hidden in the numerical analysis. There is then no need to refer to an
equivalent FP though the comparison may help in understanding the
physics.

6.2.5 Effective dynamic linewidth enhancement

The dynamic linewidth is increased well over the value of the static
linewidth of eqn. 6.12, and the discussion here merely refers the
reader back to Section 4.2. To gain an estimate without detailed
numerical modelling, the appropriate value of aH will again be the
effective value averaged over the laser. From eqn. 4.15, the linewidth
will then be increased by <5(AA+LW) ~(A.JS/2TT) (aHAg/nr). To gain an
estimate of this increase, one observes that, as the modulation
frequency increases, so the change in gain has to increase to change
the waveform sufficiently rapidly within the modulation period. One
might require 10 dB power gain within a quarter of a modulation
period, so that with a period Tone finds exp (2Ag vg 7/4) > 10 giving
roughly Ag*~5 (/M/1010) cm"1 for a modulation frequency^. With
this estimate, and even with aM = l, 5(AA+/M/) -0.5 (fM/1010) nm or
0.5 nm for a 1.55 |mm laser modulated at 10 GHz.

6.2.6 Linewidth rebroadening

Although the modifications to the Schawlow—Townes formula increase
the linewidth by (l + a#), the linewidth is still predicted to decrease
monotonically with increasing output power. In practice, however, the
linewidth rebroadens at some critical power (for typical communica-
tion lasers this critical power level ~ 10-30 mW) and this requires a
more detailed consideration of the physics which is only overviewed
here.
Three main mechanisms have been identified which can give rise to
this rebroadening of the laser linewidth. To understand the first
mechanism, recall the effects of clamping of the round-trip gain to
near unity on stabilising the amplitude fluctuations caused by the
spontaneous emission (recall the derivation of eqn. 6.11). However,
the subthreshold modes, with round-trip gains significantly less than
 More advanced distributed feedback laser design 177

unity, do not clamp the gain and, in such modes, the amplitude
fluctuations induced by the spontaneous emission give them a
relatively broad linewidth. At high enough powers, the negatively
cross-correlated noise in the main and side modes, combined with a
difference in their nonlinear gain coefficients, give rise to rebroaden-
ing of the main lasing mode. A detailed analysis [19] then predicts an
extra linewidth contribution ALW+f which varies with the cube of the
sidemode intensity PSM, and as the square of the difference Ae in the
nonlinear gain coefficients e for the main and side modes:

where ninv is the inversion factor, A^ is the Petermann spontaneous-

emission factor referred to in Section 6.2.4, and T]ex( is the external
quantum efficiency.
A second mechanism requires that the sidemode has a different
intensity distribution along the laser (e.g. the '-2 mode' has an
asymmetric field distribution and the '-1 mode' has a symmetric field
distribution) so that the average clamping of the electron density
which occurs for the laser as a whole does not apply locally. Then if the
laser frequency changes with the local current density along the cavity
[20, 21] any fluctuations in the distribution of the longitudinal carriers
create dither in the central frequency and hence the linewidth
broadens. This process does not apply to Fabry-Perot lasers where
each mode has virtually the same longitudinal mode intensity distribu-
tion and the emission wavelength is not affected by redistributions of
carrier density along the cavity.
Finally, in DFB lasers it has been observed, with certain grating
designs at high enough power levels, that perturbations in the carrier/
photon-density feedback reinforce the original changes leading to
intrinsic longitudinal instabilities of the mode which broaden the
linewidth [22-25].

6,3 Influence of reflections from facets and external

sources
63.1 Reflections and stability
The laser manufacturer, having perfected the design of the grating
structure for the desired static- and dynamic-performance character-
istics, does not wish to find random extraneous reflections interfering
178 Distributed feedback semiconductor lasers

with the grating's feedback. This is an important problem because

facets can easily reflect 1-5% of the incident power, even with an
'antireflection' coating, but as noted at the end of Section 6.2.3, the
effective gain as a result of feedback is {1/(1 — Ground)} which can
exceed 1000, hence reflected power from the facets can be greatly
enhanced. One usually has to improve the AR coating from a single
'blooming' layer to two or more layer coatings to achieve less than
0.1% power reflectivity, at which level facet reflections can have little
significant effect in appropriate laser structures.
The key problem is not so much the magnitude of the reflection
from the facet but that the phase of the reflection, with respect to the
grating period, is difficult to control with present technology and is
often simply a random variable. A high-KL laser has a greater internal
reflectivity than a \OW-KL device, and hence a lower field near the
facets for a given stored energy, thereby reducing the sensitivity to
facet reflections. Unfortunately, the efficiency is also then reduced and
a trade-off has to be made, suggesting that designs which maximise the
ratio of (stored energy/threshold gain) will have an optimum
sensitivity to reflections.
The importance of a low sensitivity to reflections extends beyond
consideration of those from the facets. Reflections external to the laser,
particularly coherent reflections, can induce a wide range of effects
andfiveregimes have been identified [26] which broadly apply to both
DFB and Fabry-Perot lasers:
(i) For power reflectivities of — 50 dB or less adjacent to the laser
(falling to — 90 dB or less at 300 cm from the laser) the device
remains single mode but the linewidth increases or decreases
depending on the phase of the feedback [27, 28].
(ii) Reflections — 45 dB at any distance and close to the upper
power limit of (i) cause the oscillation line to split for out of
phase feedback, giving two low-linewidth modes whose spectral
separation depends on the strength of the reflection. The laser
will mode hop between the two modes at a rate which decreases
with increasing reflection amplitude and stops altogether above
- 45 dB, leading then to (iii).
(iii) A third regime is now entered for a small range of feedback levels
with a stable single mode with a narrow line for all feedback
phases.
(iv) Above — 39 dB reflection, the stability in (iii) breaks down again
and relaxation oscillation occurs. As the reflection is increased,
 More advanced distributed feedback laser design 179

the relaxation oscillation increases until coherence collapse

occurs.
(v) Finally, for power reflectivities above — 8 dB, the device becomes
single-mode again with the external reflection level being typical
of that obtained in external-cavity lasers. The reader is referred to
a useful analytic discussion of these feedback phenomena [29]
which gives a good physical understanding of the detailed
mechanisms involved.

63.2 Facet reflectivity and spectral measurements

A laser's spectrum is one key property of interest in optical-
communication systems, and parameters which control this spectrum
must also be measured and controlled by the manufacturer. One such
parameter is the value of KL and the measurement of this parameter is
in principle established from the 'stopband' width as determined from
the modal separation of the two DFB modes (say, just around
threshold) just on either side of the mathematical stopband as
described in Section 5.4. However, errors in this measurement process
arise because facet reflections and their phase with respect to the
grating significantly alter the laser's spectrum. Reflectivities as low as
0.05% can be required [30] in order to estimate a value of KL~ 1 to an
accuracy of about ±0.25, although this improves at higher values of KL.
Again this can be understood in general terms by remembering the
high internal optical gain with feedback so that minuscule changes in
the round-trip gain Gmund greatly affect the output through the term
1/(1 — Gwund). To understand the process in detail, the static character-
istics of DFB lasers found in Chapter 5 need to be extended to account
for facet reflectivity.
Consider, for example, the transfer-matrix equation (eqn. 5.16) with
no input and no facet reflections. With nonzero facet reflectivities, the
boundary conditions at z=L on the right and at z=0 on the left become
RL=Pngh, FL with F0=plf!p RQ. The phase of reflection relative to the
maximum index in the Bragg grating can be accounted for by forming
complex values of plejt and prighr Remember that the reflection phase is
twice the phase shift of the facet relative to the grating, since the
propagating wave passes once through the phase-shifting region
adjacent to the facet and then back again on reflection. The
determinantal equation for the oscillation frequencies for a two-phase-
shift device is now
180 Distributed feedback semiconductor lasers

P
^ = [M,] [MJ [M3] [M+] [M,] I ^ I (6.16)

Algebraically tedious, but numerically straightforward, the transfer-

matrix product can be calculated as a single matrix

M= [M,] [M,] [M3] [M^,] [M,] =

so that eqn. 6.16 then reduces to solving numerically for those

frequencies where

+M22
P (6.17)

The difficulties with a perfectly uniform DFB will be discussed shortly

and the conclusion is that adding in '2xA m /8' phase shifts at
appropriate positions, as seen already for example in Figure 5.8,
provides a good route to a useful device. The sensitivity of a DFB to
reflections is well exhibited by calculations which have been made on
such devices, with /cL=2.1, over a range of facet reflectivities of 0.1-6%
corresponding to practical reflectivities for TiO2/SiO2 to single A12O3
coatings, respectively. The detailed parameters for the device structure
and operation are given in Table A11.2 of Appendix 11.
Figure 6.5a shows the marked change in 'stopband width' (i.e. the
separation of the two band-edge modes) around critical phases for a
facet reflectivity of 3.0%, while Figure 6.5b indicates the relative
frequencies of obtaining different stopband widths as the facet phase
varies randomly. This illustrates well the difficulties of precise control
of laser parameters because of small reflections, and the near
impossibility of determining KL from the stopband when there are
residual reflections.

633 Influence of facet reflectivity on SMSRforDFB lasers

Another parameter of practical manufacturing importance is the yield
of devices with a sufficiently high sidemode-suppression ratio (SMSR)
for use with communication systems (say < — 50 dB), but again facet
reflections seriously degrade the initially expected values. The
simulations had an uncoated reflecting rear facet; an antireflection-
 More advanced distributed feedback laser design 181

relative frequency

facet 1: p h a s e 8 0
(degrees) 2.9 3
stop band width (nm)

Figure 6.5 Influence of facet phase and reflectivity on stopband width

Figure 6.56 shows a calculated histogram of stopband width over all
possible facet phases against facet reflectivity, snowing the expected
increase in the spread of measured stopband widths as the facet
reflectivity increases. The devices are DFBs with 2 X A,n/8 phase shifts

coated front facet with a random phase with respect to the grating
(corresponding to most manufacturing conditions); allowed for
longitudinal-mode spatial-hole burning and examined the main mode
and principal side mode. Values of KL-0.75 optimised the single-mode
yield for high power [31] and the rear-facet phase had a strong
influence on the maximum output power of this single mode. Later
work [32] examined the effects of a small reflectivity for the front
facet, showing that in general this reduced the single-mode yield for
K L ~ 0 . 7 5 , but that this yield was nearly independent of KL provided
that the front-facet reflectivity <5%. Threshold current and linewidth-
power product are also influenced by facet reflections [33].
Using the previous range of facet reflectivities, with phases stepped
at 45° intervals, the performance of a 2 x Am/8 DFB laser was examined
when modulated with a 40 mA peak-to-peak current drive at 2.5 Gbit/s
and with a bias level selected to give an extinction ratio of about 10:1,
values appropriate to real systems. The simulation considered the
main mode and principal side mode together with spacial hole
burning. Figure 6.6a shows the distribution of threshold currents
obtained and Figure 6.66 shows the relative frequencies, for given
values of sidemode-suppression ratio, as the facet phase is stepped.
The simulations above took into account submount parasitic
impedances to produce realistic current drives into the laser at this
frequency. This device was modelled with parameters given in Table
A11.2 and had significant speed limitations because of the time taken
182 Distributed feedback semiconductor lasers
relative
frequency
10.

10

threshold current (mA) side mode suppression ratio (dB) 60

Figure 6.6 Influence of facet-reflectivity phase

a On threshold
b On side-mode suppression ratio (laser modulated at 2.5 Gbit/s)
The devices are DFBs with 2 x A,,/8 phase shifts

to transport the charge carriers from the contacts to the quantum

wells. Excessively high facet reflectivities changed the yield of devices
achieving less than 50 dB sidemode-suppression ratio under modula-
tion of the laser current. This is one of the more sensitive tests for the
degradation of performance induced by facet reflections. Note that
there was a change from — 53 dB down to — 44 dB and even down to
- 4 d B if the reflectivity increased from 0.1% to 3% to 6%,
respectively, for the 'worst' facet phase. In addition to the features
discussed above, adverse facet reflectivities combined with a statistically
varying facet phase change the yield of good devices in production
because of changes in the following properties:
(i) the optical extinction ratio—this needs to be greater than 10:1 at
the required on/off current drives;
(ii) the curvature of the light-current characteristic, defined here as
the ratio of the incremental efficiencies at high and low power—
the smaller this curvature the better;
(iii) the spectral linewidth—broadening is caused by excessive facet
reflections combined with certain facet phases. A 6% reflectivity
can give a statistical spread that can be up to 2.5 times larger than
from lasers with a lower reflectivity of 0.1% (this causes serious
dispersion penalties for devices operating in standard fibre
systems);
(iv) amplitude patterning in the transmitted digital optical wave-
form—the patterning induced by facet reflections with certain
phases is related to the difference between the curvatures in the
 More advanced distributed feedback laser design 183

static and dynamic L/I characteristics. A measure of this is the

transmitted eye-diagram shape, and a hexagonal eye-mask test
pattern is used to determine whether a given eye diagram is
sufficiently 'open' and hence is acceptable. (This is important in
determining whether a system can distinguish between ones and
zeros for digital modulation. Typically the opening of the 'eye'
has to be larger than an appropriate hexagonal test pattern but
this test is not found to be a sensitive indicator of unsatisfactory
facet reflectivity.)
The overall conclusion for the modelled 2xA w /8 DFB laser is that
0.1% reflectivity has little influence on device performance, which is
good, no matter what phase of reflection, but 3% reflectivity gives
serious cause for concern while 6% at certain facet phases can wreck
the performance of the laser.

6.4 Complex grating-coupling coefficients

6.4,1 General
The concept that deep periodic etching into the active-gain region can
give a gain grating has already been met. In terms of Fresnel plane-
wave reflections at each step, as outlined in Section 1.7.1, with a
complex refractive index change between adjacent teeth of a grating,
the reflectivity tip per Bragg period (two reflections per period) in
Section 1.7.1 becomes:
)
)
However, this gives a significant overestimate per step because the
deep etch does not give an abrupt junction penetrating fully into the
optical fields but rather a modulation of the thickness of an active
region or an absorbing layer, as discussed next. The numerical
techniques such as those outlined in the program slabexec in
directory slab have to be used, in general.

6.4.2 Techniques for introducing complex grating-coupling

coefficients
Gain gratings have been fabricated by two methods. Early develop-
ments exploited the different rates of growth of material by
184 Distributed feedback semiconductor lasers

liquid-phase epitaxy over a periodic grating etched into a substrate

[34]. Specifically, the local growth rate for the epitaxial material
depended on the local curvature of the substrate and on the
composition of the layer being grown. Growth of the active-gain region
by this method resulted in a periodic variation in its thickness resulting
in both a periodic gain and a periodic index variation. Interestingly,
the index variation could be approximately cancelled by a simultane-
ous thickness variation of a passive-waveguiding layer leaving
substantially a periodic gain only. The second approach is to etch
partially or completely through the active region prior to regrowing on
top of the grating, as seen in Figure 4.8. This may result in low
reliability for the device unless appropriate technologies are used.
Such deep etching into the active region may alter the distribution of
carriers in a multiple-quantum-well stack [35]. Significantly, the
imaginary or gain-coupling coefficient is now dependent on the
carrier density iVin the active region which, under dynamic operation,
means that the gain coupling varies temporally with N calculated from
the usual rate equations (Section 4.1). Assuming a linear relationship
for material field-gain/unit-distance as a function of electron density,
the coupling coefficient takes the form:

g
A wrA r

where AF gives the modulation of the effective fraction of the optical

field power within the gain grating which has been modelled by step
changes in the complex refractive index.
The introduction of a loss grating [36,37] has been realised, for
example, by growing quantum wells within the laser waveguide which
are wider than those in the active region in such a way that these
additional wells provide photon absorption while those in the active
region can give gain. However, the loss mechanism can saturate as a
result of the incomplete removal of the photogenerated electron-hole
pairs and the complex dynamics for the charge carriers in such
quantum wells needs careful modelling [38].

6A3 Influence of complex grating-coupling coefficient on static

performance
A perfectly antireflection-coated DFB laser with a pure-gain grating
will lase exactly at the Bragg frequency, with frequency offset 5=0, and
will not suffer from the mode degeneracy of pure index-coupled
 More advanced distributed feedback laser design 185

devices at low power. This result can be explained by extending the

argument concerning the phase change for a single pass round the
DFB-laser cavity, as discussed in Section 5.5.1, showing the lack of
oscillation at the central Bragg frequency for an index grating. With
the gain grating, the elemental reflection for each period Ap is now
imaginary and so ploolhrighl=ploolhMl^jpml signifying a purely imaginary
single-pass reflection as seen at the edges of the 'central' tooth of the
grating. Modifying eqn. 5.27, the round-trip feedback within the
central tooth is then given by {(jpnel) exp(j2 7r)}2 but this is now positive
and consequently lasing (oscillation) can occur given sufficient gain so
that pm = l. Such pure-gain gratings are difficult to realise but the
design in [34] approaches this ideal. The more usual situation is a
mixture of index and gain coupling. Section 5.3 showed that the
presence of gain or loss coupling removes the stopband in a strict
mathematical sense in the real-eigenmode propagation coefficient.
The observable 'stopband' in the emission spectrum reduces with
increasing gain until, for a pure gain-coupled device, it is absent.
The static performance of DFB lasers with complex grating-coupling
coefficients can be simulated numerically in the same manner as
described in Chapters 5, 6 and 7 for pure index-coupled devices.
Provided that there is no saturation in loss, the loss-coupled device
merely needs to replace the real K for an index-coupled device with an
imaginary constant JK. However gain-coupled devices have a coupling
coefficient which depends on the carrier density in the active region
and this complicates the analysis with both gain and feedback
dependent on electron density which also generally varies with space.
Besides the strength of the gain component relative to the index
component, one also has the relative phase. Typically, the two
situations discussed most in the literature are Kindex+JKgain where
KindfX>0 and Kgnin>0—the 'in-phase' case, or KiwUlx>0 and Kgain<0—the
'antiphase' case. In the literature the term 'gain grating' is often taken
to include loss gratings, but the terms 'in-phase' and 'antiphase'
usually refer to gain gratings, so it is important to be clear on the
terminology. An in-phase gain grating and an antiphase loss grating
with the same IACI are 'equivalent' though in detail there may be
differences caused by the change of gain or loss with electron density.
Similarly an antiphase gain grating and an in-phase loss grating are
'equivalent'. The effect of in-phase and antiphase gain gratings on the
linewidth has already been noted and indeed the sign also helps to
select one of the two main modes in a perfectly uniform DFB [7,17,39]
with the in-phase case helping to select the longer-wavelength mode.
186 Distributed feedback semiconductor lasers

Note that for a pure gain grating it is possible to obtain a perfectly

'flat' longitudinal-mode intensity distribution. Using eqns. 5.18 and
5.19 W\t\\JK=Kgain and 5=0 so that the frequency is the Bragg frequency
^4/
F(z) = (Kgain//3,) sin((3ez)R(0); R(z) = [sinlP,(L-z)}/sin((3eL)]R(0)
(6.20)
It can be seen that, if (/3(JL) = TT/2 and {Kgain //3e) = 1, then
F(z)*F(z) +R(z)*R(z) ={sin2((3ez) +cos2(&z)}#(0)*#(0) = IR(0) I2 (6.21)
giving a uniform field intensity along the length of the laser, which will
therefore not suffer from longitudinal-mode spatial-hole burning
[40]. Alarmingly, the net field gain g is zero but the power has to be
given from the gain in ihe gain coupling giving (KgainL) = 7r/2 and this
is critical and means a critical adjustment of the transparency density
at the same time as getting the right gain coupling!
In any one longitudinal section, conventional analysis of DFB-laser
structures relates the gain, coupling and stimulated emission-current
density to the photon density averaged over a wavelength and, in
general, the use of these mean values is well justified. When, for
example, referring to a 'flat' field profile in eqn. 6.21, the standing-
wave pattern of the type discussed in eqn. 3.5 is neglected. However,
the phases of the forward wave and reverse wave could change with
respect to each other, giving a standing-optical wave pattern of the
form referred to in eqn. 3.5, but neglecting any transverse-field
pattern for simplicity:
E*E={F*F+R*R+2 \FR\ cos (2/3,%+20)} (6.22)
Now with a periodic-gain grating the gain in the device might be
written as

gi*) = gavemge +"^periodic COS (2/3bZ) (6.23)

The stimulated recombination current, averaged over one Bragg
period, is then proportional to
average[g(z)£^£]-[^^i?*i?]g a m ^+ \FR\ cos(2<f>)Kgainpmodu. (6.24)
It follows that, depending on the phase </> of the standing-wave pattern
in relation to the grating's period, the stimulated recombination
current can change by an amount up to ±Kgninperiodic/gavernge. This has
been called the standing-wave effect [41, 42]. In this book, this effect
has been ignored for simplicity in the modelling and also because for
many practical devices the gain-coupling coefficient is weak. The
 More advanced distributed feedback laser design 187

15

K L
gain
,0.06

1
index

Figure 6. 7 Contour map of nonuniformity or variance of longitudinal-mode intensity

distribution as a function of real and imaginary KL product

standing-wave effect helps to explain in a clear physical way one of the

reasons for the mode selectivity of gain gratings.
In general, the higher the proportion of gain to index coupling the
flatter the optical-field profile (once again averaging over the standing-
wave patterns) along the cavity can be made by selecting the
appropriate KL product. The result is a device which becomes more
resistant to longitudinal-mode spatial-hole burning with an increasing
proportion of gain coupling, and hence exhibits better stability with
respect to the sidemode intensities with increasing output power. In
Figure 6.7 the effect is shown on the longitudinal-mode intensity
distribution (as measured by the variance of the intensity) with
variations in the real and imaginary KL products. This plot includes all
uniform grating devices from pure index-coupled to pure gain-
coupled structures.
Figure 6.8 shows the change in the longitudinal-mode intensity
distribution as a function of the imaginary coupling coefficient for a
small real coupling coefficient Kr of 5 cm" 1 . In addition, the spectrum
is plotted for the two extreme structures showing the reduction in the
significance of the stopband with increasing imaginary coupling
coefficient.

6.4.4 Influence of complex grating-coupling coefficient on dynamic

performance
If the imaginary coupling coefficient has an amplitude comparable
with that of the real component then, as already discussed, the
188 Distributed feedback semiconductor lasers

A relative intensity
10 3 T K L * 0.2+/2
10

1568 15701572
wavelength, nm
A relative intensity relative intensity |

10

distance along laser

1566 15681570
wavelength, nm
Figure 6.8 Dependence of longitudinal-mode intensity distribution and emission
spectrum on magnitude of imaginary coupling coefficient KgainL
For a real coupling coefficient of 5 cm"1 giving a real part KL=0.2. Note
the flatness of the intensity profile around K~ninL~ TT/2 in support of the
analytic theory. Stopband shown dotted in (o) is not present in (c)

effective linewidth-enhancement factor is significantly decreased or

increased depending on whether the gain grating is, respectively, in
antiphase or in-phase with the real grating, assuming that aH>\
[43-45]. The small-signal AM bandwidth is also changed since the
effective differential gain, and hence the slope of the resonance
frequency squared against output power, is increased or decreased for
antiphase or in-phase gain gratings, respectively. This results in a
significant change in the dynamic chirp of the laser under transient
conditions. For an antiphase grating, the resulting increase in the
effective differential gain can, under certain conditions, lead to
instability and self pulsations.
The use of a weak-gvzm grating, which has a strength that is only a few
percent of that of the index grating, does not significantly affect the
dynamic behaviour via any change in the effective differential gain. A
more subtle and also useful effect is to increase the yield of devices
which lase on either the long- or short-wavelength Bragg mode
depending on whether the gain grating is in-phase or in antiphase with
the index grating, respectively [36,46,47]. In the absence of gain
 More advanced distributed feedback laser design 189

coupling, the devices as manufactured would be approximately evenly

split between the two modes at low power but this symmetry is broken
by gain coupling. This might seem insignificant, but in Section 6.6.1 it
will be shown that devices with moderate to high KL products exhibit
better stability with respect to longitudinal hole burning, and
therefore have a reduced wavelength chirp, if they are lasing on the
long-wavelength mode. This is of great practical significance for the
digital modulation of lasers for long-haul systems.
It is interesting that a second-order DFB laser which suffers from
radiation loss (see Section 4.4.3) behaves in principle in similar ways to
a gain-coupled DFB laser. The radiation loss leads to a different gain
for the two Bragg modes and results in the long-wavelength mode
being favoured [48,49]. However, the radiation loss is probably less
easy to control in manufacture than making a gain grating.

6.4.5 Influence of facet reflectivity

Sections 6.3.2 and 6.3.3 discussed the spread in several parameters
determining the static and dynamic performance for index-coupled
DFB lasers, the spread being caused by small randomly phased facet
reflectivities caused by fabrication processes. For complex coupled
devices, the effect of facet reflections is reduced as the proportion of
the total coupling coefficient which is due to gain coupling is
increased [50], Physically, this can be considered to result from the
peaks in the mode standing-wave pattern coinciding with the maxima
in gain, and thereby stabilising the mode with respect to the
perturbing effects of the reflections. It is not at present practical to
control the facet phase, but a technique for measuring it in gain-
coupled DFB lasers has been reported and correlated with the
front-to-back facet-emission ratios [51].

6.5 High-power lasers with distributed feedback

6.5.1 General
Initially, the commercial applications for DFB lasers were almost
exclusively for long-haul digital transmission at 1.5 fim wavelength
where the optical fibre has a minimum in attenuation. The significant
dispersion in the fibre at this wavelength demanded sources with a
narrow spectral width. There is now an increasing interest in high-
power (> 100 mW) DFB-laser structures for applications such as optical
pumping of erbium-doped fibre amplifiers. High-power devices for
190 Distributed feedback semiconductor lasers

optical carriers over the access network with cable television is another
clear application for lasers which give out more power than the
conventional 'communication' source. In WDM applications, power
combiners for the different wavelengths usually lose significant
amounts of power, and again higher power is required.
There is, then, a strong commercial interest in high-speed low-
wavelength chirp sources which are composed of narrow linewidth CW
DFB-laser sources coupled monolithically, or in a hybrid manner, to
external modulators. The modelling of devices with external mod-
ulators is left until Chapter 8 and for the present it is noted that this is
a tried and trusted system designer's choice for controlled low-chirp
modulation, though it adds complexity and cost. The attenuation in
such external modulators can be significant, and in hybrid arrange-
ments there is the additional loss associated with the coupling between
the laser and modulator, and consequently the source needs perhaps
10 dB more power than previously.
One way of increasing the power output from a laser is to change
from using a symmetric device, which usually wastes power by radiating
equally from both facets. A high front-to-back emission ratio can
concentrate the power where required but this leads to surprising and
significant drawbacks.

6.5.2 Techniques for obtaining high front-to-back emission ratios

Four techniques have been utilised for obtaining high front-to-back-
emission-ratio DFB laser structures with one mode clearly selected.
The first approach is to apply an antireflection coating to the front
facet, and either leave the rear facet as-cleaved (with a power
reflectivity of about 32%) or apply a high-reflectivity coating. This
straightforward approach has been employed widely by commercial
DFB-laser manufacturers. However, as repeatedly stated, the phase of
the rear facet with respect to the grating is difficult to control in a
manufacturing environment and the wrong phase has been shown to
have serious consequences on sidemode suppression, selection of
frequency etc. Present devices have random distributions of the rear-
facet phase and need careful screening to meet a required
specification. Second-order gratings [52] can help in providing some
discrimination, but not sufficiently so as to avoid the necessity of
screening.
Secondly, the introduction of gain or loss coupling into the grating
structure to give a complex K significantly helps in the modal
discrimination but not, of course, on back-to-front facet-emission
 More advanced distributed feedback laser design 191

ratios. The presence of gain modulation in phase with the index

grating leads to a preference for oscillation on the long-wavelength
side of the stopband. Likewise, antiphase gain coupling favours
oscillation on the short-wavelength side of the stopband. (After
reading Chapter 7, try out the program dfbgain in directory dfb and
change the sign of the imaginary part of K.) It is necessary to add
significant imaginary coupling to the index grating before the device
always oscillates on one side of the stopband only. For example, if
1
K=100+J5 cm" for a 300 |xm long device with a rear-facet power
reflectivity of 32% (i.e. as cleaved), then the distribution of the
selected mode over the range of all phase shifts changes from 50%:
50% between the two modes to approximately 17%: 83% in favour of
the long-wavelength mode.
A third approach is to introduce an asymmetry into the DFB grating
itself. This can be achieved, for example, by driving one half harder so
as to 'tilt' the longitudinal-mode intensity distribution so that more
radiation is emitted from one facet than the other, even though both
facets are antireflection coated to ensure that the reflection phase is
not relevant. However, such asymmetry in the optical-field hole-burns
spatial asymmetry into the electron density and gain. Low side mode
suppression and even multiple-mode emission or instability can occur.
The fourth, and most satisfactory, approach is to integrate a DFB
laser monolithically with an amplifier. The rear facet of the laser and
the front facet of the amplifier are antireflection coated and the
interface between DFB and amplifier is virtually matched because the
amplifier is the same material but with the grating omitted. A single
contact can be applied to both DFB and amplifier, allowing the device
to be operated p-side down without introducing the complexity of
patterned contacts.

6.5.3 Laser-amplifier structures with distributed feedback

The first integrated distributed-feedback laser-amplifier used a DBR-
laser structure integrated with an amplifier section and emitted at
1.3 jjim [53]. The rear facet had a high-reflectivity coating and a
passive grating section provided the feedback from the front of the
laser, the combination giving a high front-to-back emission ratio.
Regrettably, the DBR laser like a Fabry-Perot laser can operate with
different numbers of wavelengths between reflectors if the tem-
perature or current changes; consequently the lasing mode can 'hop'
to a different wavelength under changes of drive etc. The structure of
Figure 6.9 with a phase-shift section is therefore preferred.
192 Distributed feedback semiconductor lasers

amplifier drive V \ //laser drive

current \ V J L current

IF coating U. phase shift

1 waveguitte layer
= MQW layer — —
1 waveguide layer
AR v
coating 7

Figure 6.9 Schematic diagram of integrated DFB laser and amplifier

Any residual reflection from the amplifier's facet experiences

further gain before returning to the laser, so must be made extremely
low. Angling the facet at the output has been investigated for reducing
this reflection [54], but modern AR coatings are usually now adequate.
The utility of these integrated devices for high power, suitable for
pumping erbium-doped fibre amplifiers, has been well illustrated by
modules delivering ~300 mW at 1.48 jxm [55] and also a DBR laser-
amplifier emitting a CW single longitudinal mode with 110 mW at
0.98 |mm [56] , the two different wavelengths appropriate for pumping
an EDFA.
Several variant structures have been made successfully using \m/
4-phase-shifted grating DFB lasers with an amplifier [57] to
2xAm/8-grating DFB lasers with antireflection coatings. These latter
have given 45 mW at a minimum linewidth of 2.3 MHz at 35 mW [58].
Although with AR coating, the laser section radiates as much power
out of the rear facet as the front, the overall efficiency penalty with the
amplifier in place is small on account of the asymmetry introduced by
the amplifier.
All monolithically integrated laser-amplifier structures experience
some broadening of the linewidth because of spontaneous emission
from the amplifier which is fed back into the laser. Applying eqn. 6.12
indicates immediately that a higher net spontaneous emission rate into
a laser mode broadens the linewidth. An optical isolator placed
between the laser and the amplifier could reduce such broadening,
but with present isolator constructions [59,60] that is difficult to
integrate.
The magnitude of the linewidth broadening can be significant. By
way of example [57], consider an 800 |xm-long bulk-active-region Aw/4
 More advanced distributed feedback laser design 193

laser drive
booster amplifier drive

phase shifted grating

^anti-reflection
grating quantum wells
coated facet
barriers

Figure 6.10 Schematic monolithically integrated DFB laser and flared amplifier
Slightly modified from the drawing by MJ. Fice, Nortel [57] with
permission

DFB laser with a KL product of 1.2, integrated with a 600 |xm-long

amplifier. If the laser is injected with 10 kA/cm2 and the amplifier with
2 kA/cm2, then modelling predicts a linewidth of about 12 MHz for
reasonable values of the materials parameters. However, increasing the
amplifier current density to 10 kA/cm2 results in an increased source
linewidth of about 20 MHz while the amplifier gain increases from
approximately 1 to 3.5.
A new development for high-power lasers is the use of a flared
amplifier integrated with a DBR laser which has resulted in significant
increases in the available output power. The advantage of this design
is that the mode propagating through the flared amplifier is allowed to
spread under an expanding contact which, with a roughly uniform
current density, provides an increasing injected current in the regions
where the increasing photon flux requires a higher stimulated
emission current. Both longitudinal and transverse hole-burning
effects are reduced, leading, for example, to 2 W of CW power at
0.97 |xm wavelength [62] and 1.3 W CW at 0.86 fjim [63]. The use of a
phase-shifted DFB laser, using a 2xA w /8 design, integrated with a
flared amplifier, to give stable single-frequency operation combined
with high output power, has also been investigated [61] (Figure 6.10).
The ability to model such concatenated optical components along with
194 Distributed feedback semiconductor lasers

their interaction is the beginning of subsystem and system modelling

which is pursued in Section 8.2 and will form an important role for
modelling in the future.

6.6 Dynamic modelling of DFB lasers

6.6.1 Uniform-grating DFB laser with reflective rear facet
The importance of the uniform DFB laser is that it is technologically
the most straightforward Bragg laser to manufacture so that a
considerable range of studies has been carried out to elucidate the
role of the device and its two modes. These studies include:
longitudinal intensity distribution with KL [64], dependence of
stopband width on KL [65], subthreshold spectra and facet effects
[66], instabilities caused by longitudinal spatial-hole burning [67,68],
FM and AM responses [69,70], high-speed performance in a package
[71] and intermodulation distortion [72,73]. The dynamic perform-
ance of DFB lasers with a high-reflectivity rear facet and low-reflectivity
front facet, has shown experimentally a lower penalty in the received
eye diagram after propagation through dispersive fibre when lasing in
the long-wavelength mode as compared with those lasing strongly in
the short-wavelength mode.* In spite of these studies, the uniform-
index-coupled DFB has serious drawbacks which are brought out in
this section by reviewing some modelling work for such devices,
mounted in realistic packages, and driven at frequencies of 2.5 Gbit/s
with current levels adjusted to give a 10:1 extinction ratio. Device
parameters are listed in Table All.l of Appendix 11. The main
difficulty lies in obtaining an adequate yield of single-mode devices
operating in the long-wavelength mode where a satisfactory perform-
ance for operation in a real communication system can be obtained. It
follows that other methods of mode selection have to be found which
augment the uniform Bragg grating.
The modelling study here started with an investigation of the static
performance of as-cleaved/AR-coated (power reflectivities of 32% and
0%, respectively) uniform-grating lasers with a range of facet phases
between 0° and 180° similar to that described earlier. The key features
which emerged were:
(i) Emission wavelength near threshold could not be controlled and
around 50% would lase at the lower wavelength. With facet
phases around 50-100° the short wavelength seemed preferred
* LEONG, K.W., Nortel Technology, Ottawa, Canada. Private Communication.
 More advanced distributed feedback laser design 195

with threshold currents decreasing with increasing facet phase. At

facet phases around 0-40° the long wavelength was preferred
with lower threshold currents at the lower values of facet phase.
Facet phases around 120-150° gave multimoded performance or
uncertain behaviour. In general, the side modes increased with
increasing current and only at one critical facet phase did the
SMSR improve with increasing current.
(ii) Threshold current varied with facet phase by slightly less than
20%.
(iii) Incremental front-facet efficiency was reasonably independent of
facet phase, varying by about 5% of its value.
(iv) The maximum single-mode output power predicted by the model
before instability occurred was much higher for the devices lasing
on the long-wavelength mode but this occurred only over a
relatively narrow range of facet phases around 20-40°. At the
time of this work, the connection between numerical instabilities
of static simulations and dynamic instabilities of DFBs [74] was
not well understood but recent evidence suggests a strong
connection between the twoj.
The L/I curvature and wavelength chirp are both linked to the
spatial-hole burning of the carriers [75], the former being affected by
the variation in the efficiency when there is spatial-hole burning, the
latter through change of frequency with carrier-density distribution
along the cavity. The correlation shown by modelling indicates that it
is a high curvature of the total L/I output (from both facets) which
correlates best with the chirp in the laser.
This work was extended to take into account the dynamic
performance [76] and again the short-wavelength devices clearly
showed inferior and unacceptable dynamic performance when used
over 80 km of normal dispersive fibre linkf where, for a 'one', the
trailing edge of a pulse gets moved forward in time and pulse
compression occurs. Additionally, there is evidence of increased
damping of the transient response for devices lasing on the short-
wavelength mode which should improve the performance over a
dispersive link, but this seems to be less significant than the adiabatic
chirp described above. Figure 6.11 compares the received eye
diagrams for the different wavelength transmissions indicating a
relative closure for the short-wavelength device [76] which although
acceptable to some users, makes a clear difference in quality systems.
f LEONG, K.W., Nortel Technology, Ottawa, Canada. Private Communication.
X FICE, M.J., Nortel Technology, Harlow, UK. Private Communication, 12 May 1996.
196 Distributed feedback semiconductor lasers

amplitude (arbitrary units) amplitude (arbitrary units)

>

0.2 _

0.1 -

Figure 6.11 Simulatedfilteredreceived eye diagrams from uniform-grating DFB lasers

emitting on (a) long-wavelength and (b) short-wavelength Bragg modes

The main features discovered in these dynamic studies are briefly:

(a) The time-averaged sidemode-suppression ratio (SMSR) is found to
be an inadequate guide to selecting devices with good or poor
received eye diagrams. Devices with 50 dB or better SMSR could
still have an unacceptable opening of the eye for acceptance in a
mask test, indicated in Figure 6.116.
(b) The penalty at the centre of the received eye diagram after
transmission through 80 km of dispersive fibre (17.5 ps/nm km)
varied from 0.4 to 1.1 dB dependent again on the facet phase and
whether the long or short mode was lasing. This penalty is defined
as

10.log 10 [ {(Lowest 'one' level)-(Highest 'zero' level)}

{(Highest 'one' level)-(Lowest 'zero' level)}J

The work indicated that this measure is not necessarily adequate

and a better measure of acceptance is an appropriately designed
hexagonal mask which then shows up the inferior eye opening of
short-wavelength devices more clearly, and such a test is more
reliable as an indication of acceptability in a real system.
(c) Certain values of rear facet phase give totally unacceptable
performance. Rear facet phases of 30 and 80° resulted in stable
lasing in the long- and short-wavelength modes, respectively, with
the long-wavelength mode being acceptable but the short-
wavelength mode being unacceptable in its overall performance.
(d) Longitudinal-mode spatial-hole burning is stronger for the lasers
emitting on the short than the long-wavelength mode. For the
 More advanced distributed feedback laser design 197

short-wavelength lasers, this hole burning results in a larger chirp

to yet shorter wavelengths following an increase in current and
gives a larger-dispersion penalty.
(e) If the sign of the linewidth-enhancement factor is reversed (from
negative to positive), then the long-wavelength mode becomes less
stable and the short-wavelength mode offers better transmission
performance. This arises from the inversion in the sign of the
longitudinal hole burning in the refractive-index profile which
swaps the behaviour of the short- and long-wavelength modes.
The overall conclusions from this extensive modelling is that only the
long-wavelength mode of the uniform-DFB lasers offers sufficiently
stable operation with good sidemode-suppression ratio and clear eye
opening and acceptable systems performance. However, in such
uniform lasers, there will be an unacceptably low manufacturing yield
of devices operating in this desired mode unless the facet phase can be
controlled to within ±10° or unless additional features are incorpo-
rated to ensure the appropriate mode selection. The use of 2xA m /8
phase shifts, or of partial gain coupling in phase with the index
coupling, are examples with a proven track record in this respect.

6.6.2 Large signal performance of2x \m /8 DFB lasers with strong

and weak carrier-transport effects
This section gives a case history of large-signal modelling, using a
variety of techniques, for one particular type of DFB laser where there
are two phase shifts adjusted so as to give a reasonably uniform field
and electron density over a wide range of current-drive levels and
optical-power outputs.
The example given examines the degradation of the dynamic
response that occurs when the charge carriers take too long to be
transported into the quantum-well active regions. This work is based
on laser structures which have been assessed in some detail [76]. The
simulated large-signal dynamic performance of the 2x Am/8 DFB lasers
will be examined with and without significant carrier transport effects at
2.5 Gbit/s. Table A11.2 in Appendix 11 lists the material and device
parameters used for this exercise. The simulation method was
essentially that presented in Chapter 7, but with greater use made of
frequency-domain techniques to assist in a more detailed separation of
the main and sidemode characteristics.
Figure 6.12 shows the simulated performance for the case where
carrier-transport delays are around 70 ps and are found to be
198 Distributed feedback semiconductor lasers

output power, mW

5 "

wavelength, nm

1551

1551

wavelength, nm

1552J

1552.

1.E0-J
mode power, mW
1.E-2-

1.E-4- side mode

time, ns

output power

5 _
mW

time, ns
energy, pj
10 -3dB: 0.0997nms .
1 -10dB: 0.296nm
0.1 -20dB: 0.382nm
0.01 50dB SMSR
-30dB: 0.439nm
0.001
0.0001 long wavelength
sidemode^
r 1 1 i i 1
1551.2 1551.6 1552.0 1552.4 1552.8
emission wavelength, nm
(simulated spectrometer resolution = 0.1 n m )
Figure 6.12 Simulated performance of 2x\m/8 DFB laser with significant delay
(-70 ps) in injecting carriers into active regions [79]
(a) Total intensity against time (b) Wavelength of each mode against
time (c) Power in each mode against time (d) Unfiltered transmitted eye
diagram (e) Time-averaged spectrum including simulated spectrometer
resolution
 More advanced distributed feedback laser design 199

significant. In Figure 6.12a the damped turn-on transient is shown

while Figure 6.126 shows the shorter lower-emission wavelength in the
'ones' than the 'zeros', both effects arising from the presence of the
carriers in the waveguide region transporting into the active material.
The excellent sidemode-suppression ratio is seen in Figure 6.12<: where
values of -45 dB occur even during the turn-on transient when the
round-trip gain exceeds unity. The open-eye diagram in Figure 6.12d
shows the weakness of any longitudinal hole burning, but residual
effects can be seen in the shortening of the emission wavelength with
time in the 'ones' in Figure 6.126.
This wavelength shift is created by an increasing carrier density and
decreasing refractive index near the ends of the cavity, and to a lesser
extent a decreasing carrier density at the cavity centre. The simulated
eye diagram in Figure 6.12<i agrees well with experimentally observed
eye diagrams. Figure 6.12^ shows a time-averaged spectrum obtained
by summing the instantaneous energy over all modes at different
wavelengths at each time interval, including Gaussian filtering to
simulate a real spectrometer. This technique of taking the instanta-
neous energy with time allows the intrinsic laser chirp to be evaluated
but it does not include Fourier broadening caused by modulation. The
important result here is the SMSR of 50 dB.
The longitudinal-mode intensity profile peaks at the centre of the
cavity since the KL product is higher than optimum for minimum
longitudinal hole burning. The result is dynamic longitudinal hole
burning in the longitudinal active-region carrier-density profile as
shown in Figure 6.13, indicating the time scale of half a nanosecond to
stabilise even though the current may be switched instantaneously.
Some hole burning is not necessarily detrimental to the perform-
ance of a device because the hole burning dampens the turn-on
transient and hence reduces the transient chirp. It does, of course,
extend the duration of the chirp associated with the hole burning, and
the two effects must be balanced for any application [80]. The carrier-
density profile within the waveguide region also hole burns, but is
delayed relative to that in the active region as the latter acts as the
driving mechanism for the former. The delay between these two
contributions to longitudinal hole burning leads to a complex total
effect that can only be accurately analysed with a detailed model. An
important point is that, although the carrier density in the waveguide
region may be about a factor of ten lower than that in the active
region, the thickness of the waveguide region can be sufficiently large
so as to approximately compensate. The contribution to chirp from
200 Distributed feedback semiconductor lasers

/ = 1.1 x iu +40mA 6.8 time, ns

6.9
100 200 300
distance, \im
Figure 6.13 Longitudinal hole burning along laser's length changing with time
As illustrated by plot of active-region carrier-density N profile against
time. Device is switched from a 'zero' at 1.1 of threshold current to a
'one' level with 40 mA more current

the waveguide region can therefore be very significant depending on

the delay associated with the carrier transport in the laser structure.
An increase in the AM bandwidth can be obtained by reducing the
delay associated with the transport of carriers into the active region by,
for example, thinning the waveguide region or increasing the
waveguide-doping level [81]. To clarify the influence of carrier
transport, Figure 6.14 shows the modelled behaviour of the same
device as in Figure 6.12, but now with the delays caused by carrier
transport set to zero.
It can be clearly seen that the intensity transient in Figure 6.14a is
less damped than in Figure 6.12a, but also that there is a much smaller
difference in the wavelengths of the ones and zeros of the digital
modulation as shown in Figures 6.126 and 6.146. The reduced-
wavelength chirp is also very evident in Figure 6.14# which shows a
time-averaged - 20 dB spectral width of 0.243 nm compared with the
0.382 nm in Figure 6.12^ where there is carrier transport delay. Finally,
there is less longitudinal hole burning for the structure with no carrier
transport. This can be seen from the stable average emission
wavelength of the lasing mode during the turn-on transients in Figure
6.146, whereas in Figure 6.126 this is not the case as evidenced by the
 More advanced distributed feedback laser design 201
output power, mW

5 -

1 time, ns
wavelength, nm

1552.6 J
lasing mode
1552.4-
time, ns b -
wavelength, nm
Jong wavelength
1553.4 -
side mode
1553.2-

1.E0-) lasing mode

mode power, mW
46 dB
1.E-2H
long wavelength
1.E-4-L Side mode

1 2 time, ns
output power, mW

J\ J\
1
time, ns

-3dB 0.0921 nm
-10dB 0.169 nm
-20dB 0..243 nm
-30dB 0..299 nm 51dBSMSR

long wavelength

1552.2 1552.6 1553.0 1553.4

emission wavelength, nm
(simulated spectrometer resolution = 0.1 nm)
Figure 6.14 Simulated performance of 2x \m/8 DFB laser xvith negligible delay in
injecting carriers into active regions
a Total intensity against time b Wavelength of each mode against time
c Power in each mode against time d Unfiltered transmitted eye diagram
e Time-averaged spectrum including simulated spectrometer resolution
202 Distributed feedback semiconductor lasers

shortening wavelength with time. The average intensity during the

turn-on transients in Figure 6.14a and d also shows clear evidence that
longitudinal hole burning is negligible as there is no evidence of an
underlying delayed increase in intensity over about 200 ps as is seen in
Figure 6.12a and d.
Following the lessons learnt from the structure, discussed above, a
high-speed low-chirp 2x AOT/8 DFB-laser structure with a 22 GHz small-
signal AM bandwidth has been reported [79]. This was achieved by
using six 1% compressively strained 7.5 nm-thick InGaAsP quantum
wells, with 0.28% tensile-strained 15 nm-thick barriers to reduce the
overall strain. The barriers were ^doped to 5x 1017 cm" 3 . The upper
and lower waveguide layers were /nioped at 5 x 1017 cm~3 and w-doped
at 2x 1017 c m 3 and with thicknesses of 0.1 |xm and 0.18 |xm, respec-
tively. The upper />-side waveguide layer was kept as thin as possible to
minimise hole-transport effects while maintaining a sufficient thick-
ness to support a grating. The differential gain was estimated as being
25xlO~I6cm2 from measurements of resonance frequency against
output power on Fabry-Perot devices. Additionally, the laser shunt
capacitance (C[am.) was reduced from about 8 pF to 0.75 pF by
patterning the p-side metallisation, and the series resistance reduced
to 3 ft, thereby increasing the RC roll-off frequency from 4 to 70 GHz.
The submount used was also substantially improved so that the circuit
parasitics were no longer significant up to at least 20 GHz.

6.7 Summary
This chapter began by looking at how to estimate the linewidth starting
from a theory which applied to most electromagnetic oscillators but
was then found to be inadequate for semiconductor lasers because of
the plasma and Kramers-Kronig effects where changes in the gain
(changes in electron density) give changes in the refractive index
through aH, Henry's linewidth-broadening factor. This increases both
the static linewidth and the dynamic linewidth and in the latter case
has detrimental consequences in increasing the dispersion penalty for
high-bit-rate communication over long-haul optical fibres. The effec-
tive value of the linewidth enhancement as measured by the output
spectra from a device was noted often to be larger than the intrinsic
linewidth-enhancement factor of the material and this was explained
 More advanced distributed feedback laser design 203

by considering the averaging effects over the laser. The dynamic effects
of linewidth enhancement have been noted in Chapter 4.
Perfectly uniform DFB lasers need modifications to be acceptable to
ensure that the DFB operates only in the longer-wavelength mode of
its two main modes. It is only this mode which is found to have an
acceptable performance within an optical-communication system,
taking into account SMSR, eye-opening etc. One of these mode-
selection methods was seen in Chapter 5 to be the introduction of
2x Am/8 phase shifts into the grating, but even so the studies discussed
in this chapter show that operation is critically affected by the strength
and phase of the facet reflections, and reflections should be reduced
to below 0.1% for a reliable commercial product; and by the same
arguments one has to protect the laser carefully from reflections
coming from the system if the best performance is to be obtained.
An alternative change to the uniform DFB laser discussed in this
chapter is to introduce some gain coupling into the grating. This latter
can be accomplished, for example, by etching the grating through to
the gain region thereby giving some periodic gain as well as periodic
changes in refractive index. Provided that the gain coupling was 'in
phase' with the index coupling, then the desired longer-wavelength
mode could be selected. There is evidence that such complex grating
lasers may be more tolerant to reflections if the imaginary coupling
coefficient is large enough.
There is a need for more optical output power, especially for
pumping EDFAs and for use in WDM systems. An increase of at least
10 dB can be obtained by using integrated amplifiers and, when
combined with an integrated modulator, this combination is often a
'preferred' source for designers requiring high-bit-rate modulation of
DFB lasers with the capability of being used in a WDM system.
If perfectly uniform lasers were to operate consistently in the desired
long-wavelength mode, then studies discussed in this chapter show that
the phase of the facet relative to the periodic structure of the grating
has to be controlled to a fraction of a radian and, with normal
manufacturing tolerances, this would give unacceptable yields. The
2x Am/8-phase-shifted DFB laser is much better in selecting the
preferred longer-wavelength mode when appropriately designed. Part
of this design when the lasers are to be directly modulated at gigabit-
per-second rates comes in making thin enough separate confinement
heterojunction layers. The final section on modelling showed the
importance of reducing the time taken to transport the carriers across
these layers to well below the period of the modulation.
204 Distributed feedback semiconductor lasers

6.8 References
1 WHITEAWAY, J.E.A., THOMPSON, G.H.B., COLLAR, A.J., and ARMIS-
TEAD, C.J.: T h e design and assessment of A/4 phase shifted DFB laser
structures', IEEEJ. Quantum Electron., 1989, 25, pp. 1261-1279
2 GHAFOURI SHIRAZ, H., and LO, B.S.K.: 'Structural dependence of
three phase shift distributed feedback semiconductor laser diodes at
threshold using the transfer matrix method (TMM)', Semicond. Sci.
TechnoL, 1994, 9, pp. 1126-1132
3 ZHOU, P., and LEE, G.S.: 'Mode selection and spatial hole burning
suppression of a chirped grating distributed feedback laser', Appl. Phys.
Lett, 1990, 56, pp. 1400-1402
4 OKAI, M., TSUCHIYA, T., UOMI, K., CHINONE, N., and HARADA, T.:
'Corrugation-pitch-modulated MQW DFB laser with narrow linewidth
(170 kHz), IEEE Photonics TechnoL Lett, 1990, 2, pp. 529-530
5 WHITEAWAY, J.E.A., GARRETT, B., THOMPSON, G.H.B., COLLAR, A.J.,
ARMISTEAD, C.J., and FICE, M.J.: T h e static and dynamic characteristics
of single and multiple phase-shifted DFB laser structures', IEEEJ. Quantum
Electron., 1992, 28, pp. 1277-1293
6 KOGELNIK, H., and SHANK, C.V.: 'Coupled-wave theory of distributed
feedback lasers',/. Appl. Phys., 1972, 43, pp. 2327-2335
7 FLANIGAN, B.J., and CARROLL, J.E.: 'Mode selection in complex-
coupled DFB lasers', Electron. Lett, 1995, 31, pp. 977-979
8 CHAN, Y.C., PREMARATNE, M., and LOWERY, A.J.: 'Semiconductor laser
linewidth from the transmission-line laser model', ZEE Proc.—Optoelec-
tronics, 1997, 144,(4), pp. 246-252
9 OHTSU, M.: 'Highly coherent semiconductor lasers' (Artech House,
Boston, 1992)
10 HENRY, C.H.: 'Theory of the linewidth of semiconductor lasers', IEEEJ.
Quantum Electron., 1982, 18, pp. 259-264
11 HENRY, C.H.: 'Theory of phase noise and power spectrum in a single
mode injection laser', IEEE J Quantum Electron., 1983, 19, pp. 1391-1397
12 HENRY, C.H.: 'Phase noise in semiconductor lasers',/. Lightwave TechnoL,
1986,4, pp. 289-311
13 PROBERT, P.J., and CARROLL, J.E.: 'Lumped circuit model prediction of
linewidth of Fabry-Perot and DFB lasers, including external cavity
devices', Proc. IEE.-J., 1989, 134, pp. 22-32
14 SLATER, J.C.: 'Microwave electronics' (Van Nostrand, 1950; Dover, New
York, 1969)
15 SCHAWLOW, A.L., and TOWNES, C.H.: 'Infra-red and optical masers',
Phys. Rev., 1959, 112, pp. 940-949
16 AMANN, M.C.: 'Linewidth enhancement in distributed-feedback semi-
conductor lasers', Electron. Lett, 1990, 26, pp. 569-571
17 LU, H., BLAAUW, C , BENYON, B., LI, G.P., and MAKINO, T: 'High
power and high speed performance of 1.3 |xm strained MQW gain
coupled DFB lasers', IEEEJ. Quantum Electron., 1995, 31, pp. 375-381
18 PETERMANN, K.: 'Calculated spontaneous emission factor for double-
heterostructure injection lasers with gain-induced waveguiding', ZEEE/.
Quantum Electron., 1979, 15, pp. 566-570
 More advanced distributed feedback laser design 205

19 KRUGER, U., and PETERMANN, K.: T h e semiconductor laser linewidth

due to the presence of side modes', IEEEJ. Quantum Electron., 1988, 24, pp.
2355-2358
20 PAN, X., TROMBORG, B., and OLESEN, H.: 'Linewidth rebroadening in
DFB lasers due to weak side modes', IEEE Photon. Technol Lett, 1991, 3, pp.
112-114
21 TROMBORG, B., OLESEN, H., and PAN, X.: Theory of linewidth for
multi-electrode laser diode with spatially distributed noise sources', IEEE
J. Quantum Electron., 1991, 27, pp. 178-192
22 OLESEN, H., TROMBORG, B., LASSEN, H.E., and PAN, X.: 'Mode
instability and iinewidth rebroadening in DFB lasers', Electron. Lett, 1992,
28, pp. 444-445
23 SCHATZ, R.: 'Longitudinal spatial instability in symmetric semiconductor
lasers due to spatial hole burning', IEEEJ. Quantum Electron., 1992, 28, pp.
1443-1449
24 LOWERY, A.J.: 'Dynamics of SHB-induced mode instabilities in uniform
DFB semiconductor lasers', Electron. Lett., 1993, 29, pp. 1852-1854
25 MARCENAC, D.D., and CARROLL, J.E.: 'Distinction between multi-
moded and singiemoded self-pulsations in DFB lasers', Electron. Lett, 1994,
30, pp.1137-1138
26 TKACH, R.W., and CHRAPLYVY, A.R.: 'Regimes of feedback effects in 1.5
|xm distributed feedback lasers',/ Lightwave Technol, 1986, 4, pp. 1655-61
27 BUUS, J., and NILSSON, O.: 'A fundamental limit for the feedback
sensitivity of semiconductor lasers'. 12th IEEE International Semiconductor
Laser Conference, Davos, Switzerland, 1990, Paper L-22, pp. 216-217
28 NILSSON, O., and BUUS, J.: 'Linewidth and feedback sensitivity of
semiconductor lasers', IEEEJ. Quantum Electron., 1990, 26, pp. 2039-2042
29 PETERMANN, K.: in 'Semiconductor lasers with optical feedback' 'Laser
diode modulation and noise', (Kluwer Academic Publishers, 1991),
chap. 9
30 KINOSHITA, J.: 'Validity of KL evaluation by stopband method for A/4
DFB lasers with low reflecting facets', Electron. Lett, 1987, 23, pp. 499-501
31 SODA, H., ISHIKAWA, H., and IMAI, H.: 'Design of DFB lasers for high-
power single-mode operation', Electron. Lett, 1986, 22, pp. 1047-1049
32 SODA, H., ISHIKAWA, H., and IMAI, H.: 'Longitudinal mode selectivity
above threshold for asymmetric mirror structure DFB lasers with low front
facet reflectivity', Electron. Lett, 1988, 24, pp. 431-432
33 WHITEAWAY, J.E.A., THOMPSON, G.H.B., COLLAR, A.J., and ARMIS-
TEAD, C.J.: T h e design and assessment of A/4 phase-shifted DFB laser
structures', IEEEJ Quantum Electron., 1989, 25, pp. 1261-1279
34 LUO, Y, NAKANO, Y, TADA, K., INOUE, T, HOSOMATSU, H., and
IWAOKA, H.: 'Purely gain coupled distributed feedback semiconductor
lasers', Appl. Phys. Lett, 1990, 56, pp. 1620-1622
35 CHAMPAGNE, A., MACIEJKO, R., and MAKINO, T: 'Enhanced carrier
injection efficiency from lateral current injection in multiple-quantum-
well DFB lasers', IEEE Photonics Technol. Lett, 1986, 8 pp. 749-751
36 LUO, Y, CAO, H.L., DOBASHI, M., HOSOMATSU, H., NAKANO, Y, and
TADA, K.: 'Gain coupled distributed feedback semiconductor lasers with
an absorptive conduction-type inverted grating', IEEE Photonics Technol.
Lett, 1992, 4, pp. 692-695
206 Distributed feedback semiconductor lasers

37 TSANG, W.T., CHOA, F.S., WU, M.C., CHEN, Y.K., LOGAN, R.A.,
TANBUN-EK, T., CHU, S.N.G., SERGENT, A.M., MAGILL, P., REICH-
MANN, K., and BURRUS, C.A.: 'Gain-coupled long wavelength
InGaAsP/InP distributed feedback lasers with quantum well gratings
grown by chemical beam epitaxy'. Proceedings of 13th IEEE International
Semiconductor Laser Conference, 1992, paper B-l, pp. 12-13
38 ZOZ, J., and BORCHERT, B.: 'Dynamic behaviour of complex-coupled
DFB lasers with in-phase absorptive grating', Electron. Lett., 1994, 30, pp.
39-40
39 LU, H., MAKINO T., and LI, G.P.: 'Dynamic properties of partly gain-
coupled 1.55 |xm strained DFB lasers', IEEEf. Quantum Electron., 1995, 31,
pp. 1443-1450
40 MORTHIER, G., VANKWIKELBERGE, P., DAVID, K., and BAETS, R.:
'Improved performance of AR-coated DFB lasers by the introduction of
gain coupling', IEEE Photonics Technol Lett, 1990, 2, pp. 170-172
41 DAVID, K., BUUS, J., and BAETS, R.G.: 'Basic analysis of AR-coated, partly
gain-coupled DFB lasers: the standing wave effect', IEEE f. Quantum
Electron., 1992, 28, pp. 427-433
42 MORTHIER, G., and VANKWIKELBERGE, P.: 'Handbook of distributed
feedback laser diodes' (Artech House Boston, 1997), chap. 3
43 LOWERY, A.J.: 'Large-signal effective a factor of complex-coupled DFB
semiconductor lasers', Electron. Lett, 1992, 28, pp. 2295-2297
44 LOWERY, A.J., and NOVAK, D.: 'Enhanced maximum intrinsic modula-
tion bandwidth of complex-coupled DFB semiconductor lasers', Electron.
Lett, 1993, 29, pp. 461-463
45 PAN, X., TROMBORG, B., OLESEN, H., and LASSEN, H.E.: 'Effective
linewidth enhancement factor and spontaneous emission rate of DFB
lasers with gain coupling', IEEE Photonics Technol. Lett, 1992, 4, pp.
1213-1215
46 LI, G.P., MAKINO, T , MOORE, R., and PUETZ, N.: '1.55 |jim index/gain
coupled DFB lasers with strained layer multiquanturn-well active grating',
Electron. Lett, 1992, 28, pp. 1726-1727
47 LI, G.P., MAKINO, T, MOORE, R., PUETZ, N., LEONG, K-W., and LU,
H.: 'Partly gain-coupled 1.55 |xm strained-layer multiquantum-well DFB
lasers', IEEEf. Quantum Electron., 1993, 29, pp. 1736-1742
48 MAKINO, T , and GLINSKI, J.: 'Effects of radiation loss on the
performance of second-order DFB semiconductor lasers', IEEE]. Quantum
Electron., 1988, 24, pp. 73-82
49 BAETS, R.G., DAVID, K, and MORTHIER, G.: 'On the distinctive features
of gain coupled DFB lasers and DFB lasers with second order grating',
IEEEJ. Quantum Electron., 1993, 29, pp. 1792-1798
50 DAVID, K, MORTHIER, G., VANKWIKELBERGE, P., and BAETS, R.:
'Yield analysis of non-AR-coated DFB lasers with combined index and gain
coupling', Electron. Lett, 1990, 26, pp. 238-239
51 ADAMS, D.M., CASSIDY, D.T., and BRUCE, D.M.: 'Scanning photo-
luminescence technique to determine the phase of the grating at the
facets of gain-coupled DFB's', IEEE f. Quantum Electron., 1996, 32, pp.
1237-1242
 More advanced distributed feedback laser design 207

52 KAZARINOV, R.F., and HENRY, C.H.: 'Second-order distributed feedback

lasers with mode selection provided by first-order radiation losses', IEEEJ.
Quantum Electron., 1985, 21, pp. 144-150
53 KOREN, U., MILLER, B.I., RAYBON, G., ORON, M., YOUNG, M.G.,
KOCH, T.L, DEMIGUEL, J.L., CHIEN, M., TELL, B., BROWN-GOEB-
ELER, K., and BURRUS, C.A.: 'Integration of 1.3 fxm wavelength lasers
and optical amplifiers', Appl. Phys. Lett, 1990, 57, pp. 1375-1377
54 NAKANO, Y, HAYASHI, Y, CHEN, N., SAKAGUCHI, Y, and TADA, K.:
'Fabrication and characteristics of an integrated DFB laser/amplifier
having reactive-ion-etched tilted end facets', Jpn. J. Appl. Phys., 1990, 29,
pp. L2430-2433
55 KOREN, U., JOPSON, R.M., MILLER, B.I., CHIEN, M., YOUNG, M.G.,
BURRUS, C.A., GILES, C.R., PRESBY, H.M., RAYBON, G., EVANKOW,
J.D., TELL, T, and BROWN-GOEBELER, K: 'High power laser-amplifier
photonic integrated circuit for 1.48 jxm wavelength operation', Appl. Phys.
Lett., 1991, 59, pp. 2351-2353
56 O'BRIEN, S., PARKE, R., WELCH, D.F., MEHUYS, D., and SCIFRES, D.:
'High power singlemode edge-emitting master oscillator power amplifier',
Electron. Lett, 1992, 28, pp. 1429-1431
57 WHITEAWAY, J.E.A., THOMPSON, G.H.B., GOODWIN, A.R., and FICE,
M.J.: 'Design of monolithically integrated single frequency laser and
booster amplifier', Electron. Lett, 1991, 27, pp. 2250-2251
58 FICE, M.J., GOODWIN, A.R., THOMPSON, G.H.B., and WHITEAWAY,
J.E.A.: 'Realisation of monolithically integrated single frequency MQW
laser and booster amplifier at 1.5 |xm wavelength', Electron. Lett, 1991, 27,
pp. 2305-2306
59 GREEN, P.E.: 'Fibre optic networks' (Prentice Hall, Englewood Cliffs,
1993), p. 87
60 GO WAR, J.: 'Optical communication systems' (Prentice Hall, 1993), 2nd
ed., p. 438
61 FICE, M.J.: 'Development of an optimised high power coherent 1.3 |xm-
1.7 jxm laser diode source'. ESA contract 8189/89/NL/PB(SC)
62 O'BRIEN, S., WELCH, D.F, PARKE, R.A., MEHUYS, D., DZURKO, K.,
LANG, R.J., WAARTS, R., and SCIFRES, D.: 'Operating characteristics of
a high-power monolithically integrated flared amplifier master oscillator
power amplifier', IEEEJ. Quantum Electron., 1993, 29, pp. 2052-2057
63 O'BRIEN, S., MEHUYS, D., MAJOR, J., LANG, R., PARKE, R., WELCH,
D.F, and SCIFRES, D.: '1.3 W cw, diffraction-limited monolithically
integrated master oscillator flared amplifier at 863 nm', Electron. Lett,
1993, 29, pp. 2109-2110
64 KOGELNIK, H.K., and SHANK, C.V.: 'Coupled-wave theory of distributed
feedback lasers',/ Appl. Phys., 1972, 43, pp. 2327-2335
65 IGA, K: 'On the use of effective refractive index in DFB laser mode
separation',//m./. Appl. Phys., 1983, 22, p. 1630
66 SODA, H., and IMAI, H.: 'Analysis of the spectrum behaviour below the
threshold in DFB lasers', IEEEJ. Quantum Electron., 1986, 22, pp. 637-641
67 LOWTERY, A.J.: 'Dynamics of SHB-induced mode instabilities in uniform
DFB semiconductor lasers', Electron. Lett, 1993, 29, pp. 1892-1894
208 Distributed feedback semiconductor lasers

68 ZHANG, L.M., and CARROLL, J.E.: 'Dynamics and hole burning in

uniform DFB semiconductor lasers', Int. J. Optoelectron., 1993, 8, pp.
279-291
69 VANKWIKELBERGE, P., BUYTAERT, E, FRANSHOIS, A., BAETS, R.,
KUINDERSMA, P.I., and FREDRIKSZ, C.W.: 'Analysis of the carrier
induced FM response of DFB lasers, theoretical and case studies', IEEEJ.
Quantum Electron., 1989, 25, pp. 2239-2254
70 CHRISTENSEN, B., OLESEN, H., JONSSON, B., LAGE, H., HANBERG,
J., ALBREKTSEN, O., and MOLLER-LARSEN, A.: 'Detailed mapping of
the local IM and FM responses of DFB lasers', Electron. Lett, 1995, 31, pp.
799-800
71 MORTON, P.A., TANBUK-EK, T., LOGAN, R.A. CHAND, N., WECHT,
K.W., SERGENT, A.M., and SCIORTINO P.F.: 'Packaged 1.55 jxm DFB
laser with 25 GHz modulation bandwidth', Electron. Lett, 1994, 30, pp.
2044-2046
72 KITO, M , ISHINO, M., OTSUKA, N., HOSHINO, N., FUJIHARA K.,
FUJITO E, and MATSUI, Y: 'Low distortion up to 2 GHz in 1.55 |jim
multiquantum well distributed feedback laser', Electron. Lett, 1992, 28, pp.
891-893
73 YAMADA, H., OKUDA, T, SHIBUTANI, M., TOMIDA, S., TORIAI, T.,
and UJI, T.: 'High modulation frequency low distortion 1.3 jxm MQW-
DFB-LDs for subcarrier multiplexed fibre-optic feeder systems', Electron.
Lett, 1993, 29, pp. 1994-1995
74 SCHATZ, R.: 'Longitudinal spatial instability in symmetric semiconductor
lasers due to spatial hole burning', IEEEJ. Quantum Electron., 1992, 28, pp.
1443_1449
75 WHITEAWAY, J.E.A., GARRETT, B., THOMPSON, G.H.B., COLLAR, A.J.,
ARMISTEAD, C.J., and FICE, M.J.: 'The static and dynamic characteristics
of single and multiple phase-shifted DFB laser structures', IEEEJ. Quantum
Electron., 1992, 28, pp. 277-1293
76 WHITEAWAY, J.E.A., WRIGHT, A.P., GARRETT, B., THOMPSON, G.H.B.,
CARROLL, J.E., ZHANG, L.M., TSANG, C.F., WHITE, I.H., and WIL-
LIAMS, K.A.: 'Detailed large-signal dynamic modelling of DFB laser
structures and comparison with experiment', Opt. Quantum Electron., 1994,
26, pp.S817-S842
77 WRIGHT, A.P., GARRETT, B., THOMPSON, G.H.B., and WHITEAWAY,
J.E.A.: 'Influence of carrier transport on wavelength chirp of InGaAs/
InGaAsP MQW lasers', Electron. Lett, 1992, 28, pp. 1911-1912
78 YAMAGUCHI, M., HENMI, N., YAMAZAKI, H., and MITO, I.: 'Analysis of
wavelength chirping for A/4 shifted DFB LD considering spatial hole-
burning along cavity'. 12th IEEE International Semiconductor Laser
Conference, Davos, Switzerland, 1990, Paper E4, pp. 66-67
79 WRIGHT, A.P., BRIGGS, A.T.R., SMITH, A.D., BAULCOMB, R.S., and
WARBRICK, K.J.: '22 GHz bandwidth 1.5 |xm compressively strained
InGaAsP MQW ridge waveguide DFB lasers', Electron. Lett, 1993, 29, pp.
1848-1849
 Chapter 7
Numerical modelling for DFB lasers

7.1 Introduction
A commonly used starting point for modelling lasers is a small-signal
analysis of the rate equations of the form discussed in Section 4.1
where perturbations from the steady state are considered [1-3]. These
solutions are often explicitly analytic and/or they can be rapidly
computed. Appendix 4 gives an outline of such an analysis including
carrier transport [4], from contacts to the radiative recombination
region, which is not included in this chapter. Small-signal methods
help to elucidate the physics of modulation and noise [5-8], especially
around steady-state values which can be computed more readily than
large-signal dynamic states. For Fabry-Perot lasers, coupling of
electron equations and photon equations has been done in a variety of
ways well reviewed by Buus [9] but the power of computers has moved
far in the last decade. DFB lasers have more complex structures and
have led to new methods specifically to aid in this understanding.
Transfer-matrix techniques, mentioned in Chapter 5, are also known as
transmission matrices [10,11] and are used for the analysis and design
of multisection and nonuniform lasers by tracking their performance
around specific frequencies. However, random spontaneous inputs to
a laser give randomly varying outputs which then need averaging. The
power-matrix method [12] is a transfer-matrix method which was
specially developed to compute the ramn-square values of optical
power even though the laser is excited by a stochastic 'spontaneous'
input. As has already been seen from Chapter 5, transfer-matrix
techniques operate well for 'single-mode' lasers with only one or two
clear weak sidemodes which can be all tracked as the laser changes
frequency, but particular care is needed with many modes or where
spectral mixing occurs. The techniques can be specially helpful for
long or complicated structures where time-domain analysis requires
excessive numerical storage.
210 Distributed feedback semiconductor lasers

Transmission-line laser modelling' (TLLM) [13-15] is a time-

domain technique which copes with many-moded operation and with
spectral mixing, all at arbitrary power levels. The concept is that
electromagnetic fields can be represented by pulses propagating in
time along 'transmission lines' [16]. Gain, loss, scattering, dispersion
and so on are introduced by using filter theory with transmission lines.
The spectral behaviour is recovered from this time-domain modelling
by using fast Fourier transforms (FFT). Spontaneous emission is
modelled by random excitation which then gives randomly changing
outputs which need appropriate averaging. Lowery was among the first
to recognise the value of computing the travelling-wave properties of
lasers [17,18] and a strong debt is owed to TLLM for many aspects of
time-domain modelling, but there can be clear philosophical differ-
ences in approaches from other time-domain work. In this book,
low-order finite-difference strategies, with attention paid to spectral
filtering, are used to compute Maxwell's electromagnetic-field equa-
tions, using complex fields, along with the rate equations which
balance up the energy flow of electrons and photons. In TLLM, the
physics [19,20] has to be translated into the terms of 'transmission
lines' but the practical differences may be only a matter of taste in
numerical modelling.
This chapter presents the basics of large-signal time-domain
modelling using the travelling-wave time/distance nonlinear partial
differential equations of the laser. The field patterns and electron
densities in the laser are computed permitting electron-photon
interactions to be visualised. The lasers are excited by random
'spontaneous noise' which leads to outputs which are never precisely
the same from run to run. However, the random output is not
normally a major drawback but the time-domain modelling of low-
frequency noise can require excessively long times of computation.
The FFT readily permits changing from a time-domain to a frequency-
domain analysis. With N^ points for the FFT, the lowest spectral
resolution is determined by a frequency (^,,/N^s) where vg is the
group velocity in the material and s is the space step. With time steps
typically in the subpicosecond range, effects around 1 kHz may require
a problematic 109-1010 steps which are not considered here. The
highest frequency {vg/2s) which can be modelled is limited by the
time step s/v^. which then must be small enough to enable all the
important physical frequencies to fall well below this highest fre-
quency.
 Numerical modelling for DFB lasers 211

This chapter aims to provide tutorial material for the reader with
limited experience in numerical modelling. The philosophy is to use
straightforward low-order techniques where the synergy between the
physics and numerical methods is reasonably good. Speed of computa-
tion is gained from modern high clock rates and cheap random-access
memory rather than the sophistication of high-order computational
techniques with large step sizes. The reader is referred to texts and
handbooks [21-24] rather than the research literature. Sections on
ordinary differential equations and hyperbolic partial differential
equations can then be selected. To help build up the reader's
confidence, the work starts with first-order ordinary differential
equations before moving in graded steps to the coupled travelling-
wave equations of the DFB. A series of tutorial MATLAB programs,
able to run on the student version of MATLAB [25], is provided via the
'net'. The reader with a full copy of MATLAB 4.0 or better can
enhance these programs, interfacing with C + + to gain computational
speed if required. In laser diodes, short enough pulses of optical
energy can disperse and change their shape through nonlinear
physical interactions, and it is important that the programs do not
confuse numerically induced gain or distortion with similar physical
effects. The good numerical analyst would design algorithms with step
lengths determined by error control [26] built into the algorithm, but
the ready availability of high-speed computation has seduced the
authors into more limited programs, checking that there are no
singularities at small step lengths, and then comparing results with
different step lengths to determine that a satisfactory value was used.

7.2 Ordinary differential equations

7.2.1 A first-order equation

By way of illustration, start with the straightforward ordinary differ-
ential equation
dF/dz = gF (7.1)
where F might represent a field with the gain per unit distance given
by g*, initially taken as constant. The continuous function F(z) is
approximated by F(Z s), where Z is an integer, and s is the step length.
The 'forward Enter method of solving this equation numerically gives
z or F(Zs+s) =F(Zs) (1+sg) (7.2)
212 Distributed feedback semiconductor lasers

12)

Z Z+l
Figure 7.1 Forward, central, and reverse differences (3, 2, 2, respectively)

This is shown schematically in Figure 7.1, and has the merit of

simplicity. Another simple technique, also indicated in Figure 7.1, is
the ' backward EuW method, which is written as
^(Zs + s)-s(dF/dz) z+1 «F(Zs) or F(Zs + s)=F(Zs)/(l -sg) (7.3)
Both methods work satisfactorily given small enough step lengths, and
the discussion of their relative merits is limited to considering
a purely imaginary g=jgi* Then with the 'forward Euler'
I.F(Zs+s) I2={l + (sg)2} I F(Zs) I2 always leads to some induced
numerical growth while with the 'backward Euler'
2 2 2
l/^Zs+s) I = li^Zs) l /{l + (sg;) } always leads to some induced numer-
ical decay no matter how small the step length, whereas the genuine
physical solution for a purely imaginary growth rate gives no growth or
decay with IF(Zs+s) l2=lf(Zs) I2.
With the previous methods, the amount of growth or decay can be
limited by sufficiently small steps but a substantial improvement in
accuracy and stability is to use the 'central-difference' method, which can
be regarded as using both of the above methods but for a half step
each and leads to a midway prediction between the forward- and
backward-difference methods as indicated in Figure 7.1 with
| | (7.4)
It can be seen immediately that, when g*=7g, then
li<\Zs+s) \2=\F(Zs) I2 as required. Variants on this central-difference
method when g; varies with Z do not all precisely preserve the 'energy'
as in eqn. 7.4 but are substantially better than the forward- or reverse-
difference methods. This central-difference technique is most useful
in solution of laser-modelling equations, and is used many times from
here onwards. It is also known as the trapezoidal rule, and is extended to
 Numerical modelling for DFB lasers 213

partial differential equations by use of Lax averaging [27] so that when

gbecomes variable with Z, i.e. g(Zs):

dF/dz—{^(Zs+ss) -F(Zs-k)}/s

(7.5)

The half step may suggest that one is introducing an additional layer
of points but this is simply an initial convenient notation to ensure
symmetry between the +ss displacement and the — £S displacement to
give a net step of s. The symmetry with Lax averaging proves to be
important in the (spatial or temporal) frequency domain because a
finite-difference solution 'samples' the electromagnetic fields leading
to the so-called 'Nyquist' limitations on frequency [28] which occur in
any discrete signal. This will be discussed later. The half-space intervals
may be 'removed' through shifting the origin by writing Z+2—>Z' + 1
and Z - 5—>Z' and then dropping the prime to give
l
(7.6)

7.2.2 Accuracy
The accuracy of these finite-difference schemes in the end depends on
the mathematical result [29] that for all values of the complex 'gain'
y=(g+j8)L
(7.7)

while, through halving % an even faster convergence is obtained

from

Lim^00[(l+ir/N)N/(l-lr/N)N]=expr (7.8)

Programs stepr and stepj in directory diff are available to help the
reader gain a 'feel' for the problem of convergence as the number of
steps N is changed. The programs integrate dF/dz=zF (Figure 7.2) and
dF/dz=jzF, respectively, from z=0-3 starting with F=l. The test
examples of either exponential growth or pure imaginary rate of
change in F(z) have been chosen because these (or mixtures) are
common in laser physics. It will be found in stepj that eqn. 7.5 is
slightly modified so as to ensure d(F*F)/dz=0 demonstrating the
considerable benefits of appropriate central-difference schemes which
can avoid numerically induced growth when there is no physical
growth.
214 Distributed feedback semiconductor lasers

df/dz=fz
with step length
f=l at z = 0.1
50
central
40
forward
30
20 analytic
f=exp( 1 / 2 z 2 )
10
0
0 1 2
Figure 7.2 Comparison of difference schemes

7.3 First-order wave equations

7.3.1 Introduction
The two equations

1 dF dF 1 dR dR
+—= 0 —=0 (7.9)
vg dt dz Vg dt dz

are the first-order wave equations for waves propagating in either the
forward or reverse directions with group velocities ±vg. In the
literature of partial differential equations, these are often referred to
as the advection equations. Their physical solutions are straightfor-
wardly given from
F=f(t-z/vg) and R=r(t+z/v) (7.10)
where f(t) and r(t) are arbitrary functions of time. It is possible to
consider immediately the equations with gain and phase changes
varying as

1 dF dF 1 dR dR
_ +— (7.11)
Vg dt dz v dt dz

because, on putting
 Numerical modelling for DFB lasers 215

F=exp{(g-j8)vgt}F R=exp{(g-j8)v^}R (7.12)

one recovers the straightforward advection equations as in eqn. 7.9 but
now as

+—= 0 - —= 0 (7.13)
vg dt dz vg dt dz

73.2 Step lengths in space and time—central-difference method

In computing eqns. 7.13 where the group velocity is constant, one can
take the step lengths St in time and dz in space to be related by
St=Sz/vg. While this simplification is used in all the applications
presented in this book, it is useful, for future work, to understand both
how more general solutions can be derived with varying group
velocities and that potential instabilities occur when attempting to
compute information before it can physically have time to arrive. To
this end, consider a spatial step 8z=$ with a temporal step S ^ s , / ^
with 8/78 = 3.
The forward function .Fmay be written now as F(T, Z), where T and
Z are integers. Taking a cue from Section 7.2, which shows that central
differences with Lax averaging offer distinct advantages for complex
equations, one particular central-difference system is examined for the
advection equation:

(7.14)
Low values of a imply that the pulse travels more slowly while larger
values of a make the pulse travel faster. Notice the use of 'Lax
averaging' in space where
F(T4,Z)^{F(T+lz4)+F(T+l Z-\)\ (7.15)
with similar Lax averaging in time. Different arrangements of Lax
averaging can give related finite-difference schemes that can then be
explored for stability and accuracy. Eqn. 7.14 can be rearranged with
future time values of F given in terms of previous values:

|,Z-|) (7.16)
The half-space values may be 'removed' by a shift of the time and space
216 Distributed feedback semiconductor lasers

origin as in the discussion between eqns. 7.5 and 7.6. If a = l one

immediately finds
F(T+1,Z+1)=F(T,Z) (7.17)
Note that in both this finite-difference solution and the correct
physical solution the wave amplitude is preserved as it propagates
along at the group velocity.

7.3.3 Numerical stability

There are many ways of arranging the finite difference approximations
to the differential equations so that one needs to have another guide
to check whether a proposed finite-difference scheme is likely to be
satisfactory. Provided that one is cautious about interpreting the
results (for reasons discussed by Iserles [30]), it can be helpful to
consider a solution F(T, Z)+femr exp(jaJT'-j0Z) where F(T, Z) is the
ideal solution but there is a distribution of small errors where 0 is some
(arbitrary) normalised propagation coefficient and aj is some resulting
frequency. The question then is whether there is some spatial
distribution of the errors that leads to unphysical growth with time (i.e.
a false complex a>) because of the numerical algorithm. In the physical
equation (eqn. 7.9), direct substitution shows that
0s = d)St (7.18)
The errors do not grow physically or distort the signal and one can
compare the physical result with the numerical result. If a 5*1, then,
from eqn. 7.14,
exp(j<o)={cos(0/2)+ajsin(0/2)}/{cos(0/2)-~ajsin(0/2)} (7.19)
The modulus of each side of eqn. 7.19 remains unity for all real 0 so
that, no matter what Fourier distribution one has of the errors, they do
not grow in time and to this extent the computation can be said to be
stable and therefore should be a satisfactory algorithm.
However, determining the phase shift per step on both sides of eqn.
7.19 gives
<b=2 arctan[a tan(/?/2)] (7.20)
It is now only at small enough 0 (slowly varying changes with step
length) that a)--+a/3 as in eqn. 7.18. Numerically induced dispersion/
distortion of the pulse shape always occurs to some extent (Figure 7.3).
Reducing the step size helps to limit the numerical dispersion.
Changing algorithms can also change this numerical distortion.
 Numerical modelling for DFB lasers 217

pulse amplitude - distance pulse amplitude - distance pulse amplitude • distance

(normalised group velocity = 2) (normalised group velocity = 0.4)

0 100 0 100 200 0 100 200

distance (number of steps) distance (number of steps) distance (number of steps)

b
Figure 7.3 Plots of pulse propagation
Figure 7.3 shows how pulses change their shape in space as they
propagate. The time of arrival at each point in space depends on the
group velocity, and, as the group velocity changes, the numerical
methods introduce some distortion. With differences of group velocity of
a few per cent less than the normalised value of 1, the distortion is usually
acceptable. Here by having a 50% variation in the normalised group
velocity one can see significant distortion induced at the pulse edges; this
could be reduced by having shorter steps giving more steps but less net
distortion over the same total distance

The acceptability of numerically induced distortion is most readily

determined in relation to specific applications by testing. Figure 7.3a
shows the distortion at the trailing edge which occurs for 'short' pulses
with the normalised group velocity giving a = 0.2. If the pulse moves
with a normalised group velocity of 0.4, the numerical distortion is
induced at the leading edge (Figure 73c). If a = l then there is no
distortion (Figure 7.3b).
The program advec in directory diff provides a basic test program
where the forward field is stored at all space points but for only two
time steps. Unless a = l, the computing scheme requires two future
points in space to be addressed by the past values of the same two space
points. This works for closed systems. There are other methods of
solving propagation along open lines [31] with one future point
determined by two past points. The normalised group velocity is then
required to be less than (or equal to) unity because, in such an open
system, any attempt at the prediction of information before it arrives
leads to instability (numerically induced growth). For the remainder
of the chapter the essential algorithm will use equal normalised space
and time steps (a = l, i.e. s=s,) providing the basis of stable algorithms
in either open or closed systems.
The reverse advection equation with the equivalent algorithm is
found by changing the sign of a to give
218 Distributed feedback semiconductor lasers

(7.21)
where again when a = 1 and the half spaces are 'removed', as explained
after eqn. 7.16, to give
i?(T+l,Z)=/?(T,Z+l) (7.22)
The requirement of a constant group velocity is not too serious a
limitation. It can be relaxed easily if the group velocity does not
change with time or frequency but changes only with space. Uniform
time steps 8t are then maintained but the spatial steps may be adjusted
to correspond to vg{z) 8t at each local space point Only if vg(t, z)
varies with time as well as space is it necessary to consider a more
complicated system, and this is not done in this book.

7.3.4 Gain and phase

Reverting to eqns. 7.11 and 7.12 with the velocity parameter a = 1 allows
the insertion of gain and phase changes, and central differences with
Lax averaging may be used to give
F(T+1, Z) -F(T, Z - l)4(g-j8)s{F(T+l, Z)+F(T, Z - 1)} (7.23)
or
l l l
4 - l ) (7.24)
The simplicity of the algorithm is enhanced by keeping g and 8
evaluated at Z — 1 on the right-hand side of eqn. 7.24, even though this
requires smaller steps. The result in eqn. 7.24 is in effect saying that
(l + ys)/(l — ys) is a good approximation to exp(2ys) for complex
7=2(g~j8) and indeed, to the level of approximation, one may use
either {(l + ys)/(l — ys)} or exp(2ys). However the form of eqn. 7.24
is more useful for later on when shaping the gain with frequency using
a numerical filter in the time domain.
The reverse advection equation with gain and phase changes is
similarly given from
R(T+l,Z)={(l4sg-J2-s8)/(l-2-sg+jl2sS)}R(T,Z+l) (7.25)
Central differences with Lax averaging improve the accuracy and
stability for partial differential equations just as they did for complex
ordinary differential equations.
 Numerical modelling for DFB lasers 219

7.4 Coupled reflections

7.4.1 Kappa coupling but no gain or phase changes
With operation at the Bragg frequency (8=0) and no loss or gain in
the waveguide, then the coupled equations of interest are

1 dF dF 1 dR dR
J J
vg dt dz vg dt dz

With a constant K it is straightforward to eliminate R to obtain

1 d2F d2F (7 27)

M?-J£ -"*F
=
-
xrg dt dz
One may write an analytic solution as F=exp{cd(vgt/ s) — j0(z/s)} where
a spatial-step length s is used to normalise the spatial propagation
coefficient and temporal frequencies, 0 and d>, respectively, related by
02 = CO2~(KS)2 (7.28)
The physical frequency corresponding to the phase change per time
step a) is a modulation frequency coM—a measure of the frequency
deviation from the central frequency coo which has been removed in
the formulation of eqns. 7.26. When the phase shift per step is such
that \a)\<KS there are 'evanescent waves' giving an imaginary spatial
phase shift per step 0 (corresponding to the DFB's stopband).
For a straightforward first-order finite-difference scheme, one treats
the forward and reverse waves on a symmetrical footing. There are a
number of ways of using Lax averaging on the right-hand side of the
coupled equations and a useful way forward as judged from later work
on energy conservation is given from

(7.29)

f,Z-|)}
(7.30)

7.4.2 Matrix formulation

One now wishes to re-arrange eqns. 7.29 and 7.30 into a form which
demonstrates that the 'energy' that is scattered by a lossless index
grating is conserved. Start with the equations written in matrix terms:
220 Distributed feedback semiconductor lasers

1 -J\KS\ \F\(T+\),
i 1 JL |

1 JZKS \\ r\(J

I 72ACS 1 \ \ R ( T

Write

cos jd
(7.32)

with tan \6=\ /cs and eqn. 7.32 then gives a matrix U{6) such that
iflid) U(0) = l wheref implies conjugation and transposing, f/is said to
be a unitary matrix. Changing the origin of time and space as before to
'remove' the half steps in eqn. 7.31 one can get back almost to the
conventional advection relationships linking ^{(T+l), (Z+l)} and
J?{(T+1), Z} with F{T, Z} and R{T, (Z+l)}:

V{(T+l), (Z+l)}] [ cos^ jsin^ l F F(T,Z\ 1

i?{(T+l),Z} J Li sine cos d j [«IT,(Z+1)}J
(7.33)

where sin 0=KS/(1+\K2 S 2 ). Because cos2^+sin2^=l, the scattered

fields are linked by a unitary matrix in eqn. 7.33 which ensures that the
energy, determined by F*FZ +/?*i?z+1, is conserved to F*Fz+l +R*RZ as
time steps on one step and so ensures a stable numerical process which
reflects the physics of the continuous system. (Eqn. 7.33 is not a
transfer matrix because the field F on, say, the right-hand side of the
matrix is at the opposite ends of a spatial step to the field R on the
same side of the matrix.)
There is an interesting corollary which occurs with gain-coupled DFB
lasers where, in the extreme limit, K=JKgain is purely imaginary so that
eqn. 7.33 changes to become
 Numerical modelling for DFB lasers 221

fcoshfl' sinhfl'lf F(T,Z) 1

L i?{(T+l),Z} J [sinh & cosh ffJ [f?{T, (Z+1)|J '

where sinh 0r = Kgain s/(l — 4 K2ain s2) the gain-coupling coefficient. Now
one finds thatF*Fz+1 — i?*i?z is correctly conserved, again supporting
the strength of the central-difference scheme combined with Lax
averaging in obtaining a useful approximation to the physics.

7.4.3 Phase jumps replacing scattering

In the design of DFB lasers, it has become common (Chapters 5 and
6) to introduce one or two sections at appropriate points into the
Bragg grating adding an extra phase shift exp( —j(f>) but with no extra
scattering of the forward and reverse waves within that section. The
phase jump introduces an additional phase delay on top of any normal
propagation delay so that a first approximation replaces the scattering
matrix (eqn. 7.34) with the phase-jump matrix:

r \ y \ +1), (£.+ i)\ , , ~.~r v j ^ , « , ^ v . , —, ' / 7 ^ F ; \

/J{(T+1),Z} R{T, (Z+l)}

In the simplified 'universal' DFB program which is demonstrated with

this tutorial text, phase jumps are included by such substitution of a
phase-jump matrix for a scattering matrix at appropriate points within
the structure.

7.4.4 Fourier checks

The Fourier techniques used previously to check stability can again be
used on eqn. 7.33 by allowing an arbitrary variation exp(j/3 Z) in space
(Z integer) and looking at the normalised frequency cb which is
required for the variation exp(jd)J) in time step number T. After some
algebra, the dispersion relationship is obtained:
cos &>=cos 6 cos 0 (7.36)
For all real 0, the corresponding value of to is real because cos 6< 1. For
small enough values of s (where d>—»() and /3—»0 with cos 0=*1 — 2*:2s2)
then (A)2tt(K$)2+/32 so that the dispersion relationship is correct for
sufficiently small step lengths. Because co is real for all distributions of
error (all values of /?), the method is normally computationally stable
222 Distributed feedback semiconductor lasers

having no numerically induced growth with time. These Fourier

techniques will need revisiting when there is physical growth of the
laser fields. Sufficient tools are now assembled to consider the
complete finite-difference scheme for a uniform Bragg laser.

7.5 A uniform Bragg laser: finite difference in time and

space

7.5.1 Full coupled-wave equations

The full 'DFB equations' can be recalled from Chapter 4:

(7 37)
'

(7 38)
-

As explained in Chapter 4, these equations have removed the rapid

optical frequencies at some chosen central frequency of the wave
packet close to the Bragg frequency. The parameter S is a measure of
this detuning. The amplitudes F and R then vary at 'microwave'
frequencies with the wave packet propagating with the group velocity
vg which is taken here to be approximately constant at its value around
the central or Bragg frequency. The net gain, with a confinement
factor and loss, has been simplified here to just a single term g
independent of deviations from the central frequency of the laser. The
value K represents the reflection per unit length, while g—j8
represents the field gain and phase change per unit length so that S
can allow for phase-velocity variations caused by changes of refractive
index or also changes in the chosen central frequency.
Using the substitution of eqn. 7.12 defines modified fields Fand R
apparently without gain or loss as

1 dF dF 1 dR dR
_„+___ = y#c/j — =jKp (739)
J J ;
vg dt dz vg dt dz

The finite-difference scheme then should give

 Numerical modelling for DFB lasers 223

(7.40)
where sin 0=KS/(1+4~K S ). Frequently in a DFB, one finds that there
2 2

are long sections with a uniform K SO that a series of space steps can all
have the same values of ft
The stability criterion should be revisited because now, at each space
step, the forward field changes by exp(j$Z) and with /3~0 an increase
at each time step is given from exp{(g—jS)s}. This growth at each step
is the correct physics and does not indicate numerical instability.
Numerical instability would arise if the rate of increase in time became
unbounded or unphysical as the value of 0 either increased or took
critical values. Here, following the work leading to eqn. 7.36, the
dispersion relationship is
cos(d)~sS-jsg)=cos 0cos 0 (7.41)
The gain remains bounded for the high spatial frequencies and one
may expect convergence and adequately low distortion for sufficiently
small steps of length s.

13.2 MATLABcode
A follow-up to eqn. 7.40 must show how to concatenate the spatial steps
of length s. Initially, the field numbers are considered from 1 to N f 1
where there are N sections making the laser's length L=N>. For the
moment, the laser is considered to be uniform. The 'present' fields
(forward and reverse), labelled as ff(n) and fr(n) where n refers to the
space-step numbering, are stored as vectors ff and fr. The 'new' fields,
created at the next time step, are stored as ffn and frn .The 'complex
magnification' per step is
mL=exp{(gL-j*dL)/N}-(2*N + gL-j # dL)/(2*N-gL+rdL) (7.42)
Here gL — j*dL is the code for (g~~ jS)L with L the laser length. The last
term is the central-difference approximation to the exponential. This
magnification is stored as an array mL(n) defining gain and phase
changes at each spatial step and is updated at each time step as
required. From eqn. 7.33 , the Bragg grating introduces a reflection
term sin 8 { = KS/(1+4K2S2) ~KS} and a transmission term
cos 0~ v (1 — K2S2). Numerical 'energy' conservation requires
2 2
cos 6 4-sin 6=1 so, although rounding sin 6 to ACS only gives a small
224 Distributed feedback semiconductor lasers

error (typically <1%), rounding cos 0 to unity is not recommended.

An array of reflections/ transmissions are now formed at each step:
sint(n) = ks(n) and cost(n)=V[1 -{ks(n)} A 2] (7.43)
where, in MATLAB terminology, "p indicates raising to the power of p
while for future reference * indicates multiplication. Here ks is the
value of kappa xs product for each section and is permitted to vary
from section to section. The 'guts' of the difference equations for the
fields in the middle of the DFB then may be written as
forn = 1:N; (7.44)
ffn(n4-1) = mL(n)^cost(n)^ff(n)+j^mL(n)*sint(n)^fr(n + 1); (7.45)
nr=N + 1-n; (7.46)
frn(nr) = mL(nr)*cost(nr)*fr(nr+1)+j*mL(nr)*sint(nr)*ff(nr); (7.47)
end (7.48)
The value of dl_ in eqn. 7.42 is determined by the offset frequency of
this input field from the Bragg frequency. A program called dfbamp in
the directory dfb 'tops and tails' eqns. 7.44-7.48 with input routines
giving zero input on the right and a unit amplitude field entering on
the left so that the device acts as a coherent forward-wave 'amplifier'.
Comparison with the analytic solutions, discussed in Section 7.5.3,
shows that the program works robustly with as few as 16-30 or so
sections in a device where KL=2, gL=0.5 with SL=4 as an example
taking around 10 s on a 166 MHz 486 processor to compute sufficient
steps to allow the device to settle down near to its steady state.
MATLAB programs often operate more efficiently when 'vectorised'
but this can be done later.

7.5.3 Analytic against numeric solutions

Pauli matrices (Appendix 6) provide compact analytic steady-state
solutions for eqns. 7.37 and 7.38 to compare with a finite-difference
program. From eqn. A6.16 one may write

(7.49)

where a=cos(/3,z); d=(/jL//3e) sin((3ez); b= {K/fie) sin(/3ez); jui=(g~jS);

(32L2= — (JUL2+K2)L2 and S gives the 'frequency'-offset parameter. With
only a unit input at z=0; F(0) = l, /?(L)=0 leading to
 Numerical modelling jm DFB lasers 225

forward propagating power

reverse propagating power

R(32) = 0

10 20 30
distance
Figure 7.4 Numerical solution for a uniform-DFB-structure amplifier
KL=2; gL=0.5; 8L=4 with a 31-section laser (32 field points) calculation.
On the scale of this drawing one cannot distinguish the difference
between the analytic and the numerical solutions from dfbamp

R(0) =jb(L)/{a(L) — d(L)\ so that, at a general value of z, one obtains

an analytic result:

a(z)+d(z) 1
(7.50)
ib(L)/[a(L)-b(L)}

This can be calculated and compared with the finite-difference

solution from dfbamp as in Figure 7.4.
The amplifier's fields take longer to settle to a steady state as one
gets closer to the lasing condition (i.e. the effective photon lifetime
increases) so that, to avoid apparent errors in steady-state values, a
longer time of computation is then required to achieve a steady state.
The reader can gain familiarity with this response time by changing
the program's parameters. It is suggested that the results of Figure 7.4
be contrasted with results obtained by putting kL=2, gL=0.7; dL=O
where the fields evanesce inside the laser and take longer to settle
down.
Different coherent inputs create different interference patterns
inside the laser and completely different profiles are obtained for the
steady state fields when a DFB is excited coherently in different ways.
The reader is encouraged to save dfbamp in new files, say dfbampi
(etc.), and to modify these to allow for inputs at both ends of the laser
226 Distributed feedback semiconductor lasers

to check this phenomenon, remembering that a student edition of

MATLAB may limit the number of field points that can be used in a
program.

7.6 Spontaneous emission and random fields

7.6.1 Spontaneous noise and travelling fields

As discussed in Chapter 4, a laser is excited through random
spontaneous emission. The quasiclassical way of modelling the
resulting fields is to start with a single-moded optical field
a exp j4> exp(jco0t) = (ax+jay) exp(ja)ot) with the fixed single-mode
optical frequency (o0, with amplitude a and phase (f) giving a complex
amplitude (ax+]ay). Added to this 'coherent' complex-field amplitude
are random Gaussian distributions of (real and imaginary) fields. A
program spont in directory spontan gives the reader an indication of
this modelling process. The sequential snapshots in time indicate
movement of the complex phase relative to the steady phasor
exp(jco0t) which is treated as a constant (see Figure 7.5a). The
frequency-domain representation is found using the fast Fourier
transform and shows white noise (incoherent fields) superimposed on
the narrow-line coherent signal (Figure 7.5b). This net random signal
is then filtered to give narrow spectral-band noise (Figure 7.5d). On
filtering, the (coherent+noise) output phasor performs a random
walk as sketched in Figure 7.5c.
As the 'coherent' amplitude gets large relative to the noise, any
uncertainty or randomness in either the net amplitude or the phase of
the coherent fields becomes negligible. To demonstrate further
features of the model, consider a laser amplifier without feedback
where there are only forward fields advancing with gain, according to
the advection equation. The complex field varies with some constant
central frequency which for the purposes of calculation may be
'removed' as in Figure 7.5. The coherent complex-field amplitude is
amplified but also has spontaneous emission added as the fields
travel:

dF 1 dF
-r+-—=gF+ispf(t,z) (7.51)
Hl
dz vg dt
Excitation of the fields is through the random spontaneous emission,
represented by the term isp(t, z) whose magnitude is studied in
 Nu men cal modelling for DFB lasers 227

power spectrum
coherent + coherent +
white noise field
filtered noise field
imaginary
(a) random walk

(c)
-100 frequency 100
filtered power spectrum
20 •

dB
'instantaneous (d)
instantaneous field
-20
-40
-60 real
-100 frequency 100
Figure 7.5 Complex coherentfieldwith added noise
A program spontd provides numerical demonstrations of adding sponta-
neous emission (e.g. white noise) to a coherent signal (i.e. a constant
complex field) at the central frequency. The demonstration shows
(a) A display of the sampled instantaneous fields with time
(b) The white-frequency-domain spectrum apart from the coherent
spike
(c) The effect in the time domain of filtering in the spectral domain
(discussed in Section 7.9) to give
(d) A coherent signal plus narrowband noise where the coherent vector
performs a random walk in the phase plane as in (c)

Appendix 9. Initially omitting the spontaneous emission in eqn. 7.51,

a difference scheme with space and time steps s and s/vg9 respectively,
may be written as
F{T+l)s/vg9
(7.52)
which rearranges to
F{(T+l)s/vg9 (Z+l)s} = {(l+igs)/(l -jgs)}F(Ts/vg9 Zs) (7.53)
An essential feature is that, at each (T, Z) step, spontaneous fields are
added through the spontaneous excitation:
2H, (7.54)
where 2ft, = 9f?+/2/ has 9? and ^ each normally distributed random real
numbers. The randomness of both 9? and ^ ensures there is no
correlation between the spontaneous emission at any two points of
space and time. The strength of the spontaneous emission (Appendix
9) is related to the local electron density iV so that with this scheme it
makes sense to link the spontaneous excitation with the mean electron
228 Distributed feedback semiconductor lasers

F1 F2 F3 F4

t
o

N1 ! N2 ! A/3

1 2 3 4
• distance
Figure 7.6 Field and spontaneous excitations
At different space points, for one time

density on either side of the field point at which the spontaneous

emission is inserted (Figure 7.6) leading to
F{(T+l)s/vg9(Z+l)s) = {(l+Tgs)/(l-Tgs)){F(Ts/vg9Zs)
+sispf(Ts/vg,Zs)} (7.55)
where, writing Ne = N/Nlr for a normalised electron density to avoid
confusion with any integer N,
isftf(Ts/vg9Zs)*T[Ne(Ts/vg9Zs)+NelTs/vg9 (Z-l)s}] (7.56)

7.6.2 Null correlation for different times, positions and directions

Once a diode laser is lasing, it is found that the electron density
typically changes by only a few per cent so that the precise variation of
the spontaneous emission with the changes of electron density proves
to be of less concern than might initially be thought. Of more concern
is the possibility of spurious numerically induced correlations of the
randomly excited signals within the laser. It is again emphasised that,
to avoid such correlation, both d£ and % above are determined
separately and randomly for both forward- and reverse-travelling
spontaneous emission so as to ensure all these excitations are
uncorrelated with each other.
Uncorrelated inputs at each time step ensure a white-spontaneous-
noise power spectrum over the finite bandwidth of vg/s which is the
Nyquist limit caused by the sampling action. Appendix 9 shows how to
tailor this spectrum and, as indicated earlier, such tailoring may be
required if vg/s extends for an excessive range of physical frequencies.
For the basic models outlined in this chapter, a 'white' spontaneous
spectrum is mostly used because the laser's gain is filtered and the
 Numerical modelling for DFB lasers 229

shape of the spontaneous spectrum at outlying frequencies becomes

unimportant. The mean-square value of the spontaneous random
generator 9i is chosen to be 2ft 2 = 1 (ensemble or temporal average) so
that the spontaneous noise is then determined by rsp in eqn. 7.54.

7.6.3 Spontaneous magnitude

The precise formulation of rsp depends on the normalisation of the
fields E Here the normalisation is such that F*F=Sfs&vg gives the
numbers of photons per second with Sf, the density of photons
associated with the forward flow along the guide of area «s4. Equation
A9.47 shows:
^ (7.57)
2
where (Klr\ vg/9>7renejjLspf) is an effective volume with an order of
magnitude of 10 (|xm)3 for 1.55 |xm-wavelength lasers and Klr is the
transverse Petermann factor also discussed in Appendix 9 and in
Chapter 2 but taken as 1 for good-index guiding lasers. The term BPN
is the spontaneous (bimolecular) recombination rate per unit volume
given charge-carrier densities P and N with B a parameter that has
specific measured values for different materials. The parameter Tsp is
the confinement factor appropriate for spontaneous emission equiva-
lent to (but because of different distributions of photons and charge
carriers not necessarily identical to) the confinement factor F for gain.
Typically, in laser modelling P~N, reflecting the charge neutrality.
Provided that the laser remains single-moded, Appendix 9 shows a
slightly surprising result that the increase in spontaneous emission,
caused by an increased guide volume per unit length where the area $&
is increased, is effectively cancelled because the increase in guide
aperture also leads to increased 'aerial gain' for the spontaneous
dipoles within the guide, and hence a reduced solid angle for the
effective coupling of the spontaneous emission into the guide.

7.6.4 Tutorial programs

In directory spontan the programs amp1, amp2 and amp3 provide
three demonstrations based on an 'ideal' amplifier of eqn. 7.55. The
stimulated gain is taken as proportional to PN (the charge-carrier
product) as well as being proportional to the photon density. The
spontaneous emission is similarly proportional to PiVand this random
emission is inserted between the field points as in Figure 7.6. The first
demonstration (amp1) shows how the (white) spontaneous-noise
230 Distributed feedback semiconductor lasers

power increases linearly with the amplifier length when there is no

gain (ampif filters the noise to give a coloured spectrum but
essentially is the same demonstration as in ampi). The second
demonstration (amp2) shows how, with a fixed gain per unit length
and increasing length of amplifier, the ratio of noise power to signal
power reaches a limit in this ideal 'quantum' amplifier. The third
demonstration (amp3) shows that the link between gain and spontan-
eous emission means that there is no benefit in increasing the gain per
unit length indefinitely because the (signal/gain)/(noise-power/gain)
reaches a limit—the quantum limit. The fields are 'normalised' merely
to demonstrate these links between gain, length of amplifier and
signal-to-noise power and do not at present relate to physically
significant magnitudes.

7.7 Physical effects of discretisation in the frequency

domain

7.7.1 Discretisation process—integrals to sums

At first sight, the process of improving the approximation when using
finite differences is simply a question of making the step length
sufficiently short. However, no matter how short the step length, one
is effectively replacing a continuous function, say of time f(t), by its
discretised or sampled version f[T(s,/vg)} where (s(/vg) is the
temporal step length and T is an integer. Because of the wide use of
and teaching about digital signal processing, it is believed that the
physics of discrete signals will be relatively well understood by most
readers so that only a brief resume is needed of some points to watch.
Figure 7.7a indicates a continuous signal which might represent the
modulated amplitude of, say, an optical pulse along with its baseband
spectrum (Figure 7.76), and then indicates the effect of discretisation
or sampling on that signal (Figure 7.7c) and spectrum (Figure 7.7d).
The computation can only take a finite number of samples, say N (this
integer N has no necessary relationship with the number of segments
of the laser) leading to a total time of observation (s,/vg)N. The
corresponding smallest spectral interval in angular frequency is then
27r(i//r/s,N) and the spectrum ranges from - ir(s(/vg) to + w(st/vg),
again in angular frequency. The limits of this spectral range are
referred to as the Nyquist range or the Nyquist limits.
It may not be apparent immediately that discretising the signal over
a limited time and taking the discrete Fourier transform (DFT) forces
 Numerical modelling for DFB lasers 231

4
temporal spectral
amplitude amplitude

time angular frequency

Nyquist limits

—nvo +Jcv
g
spectral
temporal amplitude d
amplitude
l
ill,
time angular frequency
step st/Vg step 2n Vg /N

Figure 7.7 Discretisation: schematic relation between time step and spectral limits

the signal to be effectively periodic but this is the effect, as shown

schematically in Figure 7.8a [32] (see also [28], for example). There is
no physical significance in the signals outside the main interval of
either time or frequency. The only reliable information lies within the
N time steps of the temporal observation and the N spectral steps
(placed symmetrically about the origin). For an adequate approxima-
tion to the physics, both the length of observation and the spectral
width have to be adequate to cover the required range of physical
phenomena. If one attempts to model physical processes with a
spectrum wider than the Nyquist limits or tries to obtain information
over time scales longer than the observation interval, then one obtains
errors. Figure 7.8c shows the effects of modelling high-speed phenom-
ena with an inadequate computational bandwidth so that the physical
spectrum extends beyond the Nyquist limits, forcing the spectrum
from one spectral period to overlap with the spectrum in the next
period (so called aliasing errors). The frequency lines in Figure 7.8c
have been slightly displaced to make it clear that there is an overlap of
the spectral amplitudes associated with the neighbouring periods of
the spectrum. The results for the discretisation of space follow those
found with time but with spatial (angular) frequencies or propagation
232 Distributed feedback semiconductor lasers

A main period
temporal (a)
amplitude

l1,
time-

main period (b)

spectral
amplitude
I ;
!. i 1 _L 11
angular frequen<

main period
4 1
: ' (c)
spectral
amplitude
, 1l Jjili I Ilk 11,
angular frequen
aliasing aliasing
Figure 7.8 Discretisation: periodicity and aliasing

coefficients /3 replacing the temporal (angular) frequencies a> and

similar Nyquist limits on propagation.

7.7.2 Fast Fourier transform (FFT)

A major advance in DFT analysis came with the fast Fourier transform
(FFT) [33] (see also [28,32], for example) where, by choosing the
number of steps N = 2M with M integer, one is enabled to calculate the
DFT in exceptionally efficient ways with the calculation time increas-
ing for large N as fMN rather than as N2. FFT algorithms are standard
tools and MATLAB has its own version. Programs specti through to
spect7 in the directory fftexamp may be helpful in refreshing the
reader's mind about some key points (listed below) to bear in mind
when using a fast Fourier transform:
(i) There is an equivalent of Parseval's theorem showing that
'energy' in the time domain equals the 'energy' in the frequency
domain.
(ii) With a maximum observation time T, the minimum frequency
that can be accurately measured is 1/T and this is also the
minimum change of frequency that can be measured. If one
 Numerical modelling for DFB lasers 233

wishes to discriminate two frequencies separated by Af, then

T > I / A / is required.
(iii) With the student MATLAB 4 which has been tested by the
authors, one cannot find the FFT for more than 8192 = 21S points
but there is nothing in the algorithms which prevents a larger
number (other than time of computation).
(iv) The maximum frequency that can usefully be measured is
N/2T=vg/(2 s,) where st/vg is the temporal step in the finite-
difference scheme. Attempts to measure higher frequencies may
appear, erroneously, to give a result, but at a lower frequency
(aliasing). In the modelling work here, the potential bad effects
of aliasing are reduced by choosing finite-difference algorithms
so that the signal is effectively filtered to give a finite spectral
bandwidth with optical gain falling to zero at the Nyquist-band
limits. However, nothing can invent information which is not
present so that the step length still has to be short enough to
ensure that the numerical bandwidth encompasses the required
physical bandwidth.

7.8 Finite-element strategies for a spectral filter

7.8.1 Lorentzian filter
It is worth opening the discussion with the consideration of represent-
ing g(t) by a sampled approximation $[g[(T+l) (st/vg)} + g{T(st/vg)}]
where T is an integer. This representation has been referred to as a Lax
average. By considering a phase factor of Gxp{ja)M(st/vg)} per
temporal step, one can see that, in the spectral domain, this process
gives a filter with the characteristic amplitude response
cos{coM(s(/2vg)}. At the Nyquist-band edges, its magnitude falls to zero,
so that Lax averaging has a positive beneficial role in limiting the
magnitude of any quantity to the central spectral region determined
by the sampling time. One could naively then simply change the step
length to obtain the best approximate filter shape for the gain.
However, a more sophisticated strategy is worth considering where the
central frequency and the central gain curvature can be altered even
with a fixed step length. This leads to the Lorentzian filter.
There are at least two particular instances in laser-diode modelling
where a Lorentzian filter can be helpful:
(i) filtering the spectral shape of the gain spectrum acting on the
optical fields, and
234 Distributed feedback semiconductor lasers

(ii) shaping the spectrum of the spontaneous noise.

The gain may be g(co0) at some 'central' frequency co0 but this changes
as the frequency is changed and reduces to negligible amounts at the
edges of the laser-frequency band. One useful filter replaces the single
value g(co0) by g(a)0)/{l+ r2 (co~ &>0)2}1/2, the so-called 'Lorentzian
filter'. Of course the measured gain/frequency relationships, as for
example seen in Figures 2.8, 2.9 and 2.11, are not as simple as this, but
fortunately only the peak gain value, peak gain frequency and gain
curvature are really significant in determining the operation of the
main modes in the laser [34] and these features can be matched by an
appropriate Lorentzian filter.
In taking this filter shape, the reader is reminded that the central
lasing frequency has been removed from the laser fields by writing
E(i) exp(jco0t) and removing the phase term exp(jco0t). The field
amplitudes E(t) vary at 'base-band' or modulation frequencies
OJM = o)~ a)0 and have to be filtered by 1/(1 + r2 OJ^) 1/2 taking, for initial
simplicity, g(<o0) = l.
The Lorentzian spectral filter is based on a temporal first-order
relaxation equation, linking an output b()ul with an input aIn through
(l + rd/dt)bOul = aIn (7.58)
In the frequency domain, the spectral amplitudes at a frequency ± cou
from the central frequency a)0 are related by
\Bf)ul\2=\Aln\2/{l + (coMr)2} (7.59)
[In the real frequency domain in eqn. 7.59, it may appear not to matter
if, in place of eqn. 7.58, one used (1 - rd/dt)bOu( = aIn, but this leads to
any small disturbance growing exponentially in the time domain and
therefore is not a physical filter which satisfies the Kxamers-Kronig
relationships.] To implement eqn. 7.58 digitally, remember yet again
that sampling the signal with a normalised time step s( = s,) automat-
ically limits the useful physical range to the Nyquist limit, and
computational errors can occur in attempts to use frequencies outside
this limit.
Using central differences with Lax averaging and initially taking half
steps:

= fl/n(+i)+a/n(-i) (7.60)
Writing K=(2r vg/s), and changing to unit time shifts, this filter
rearranges to give
 Numerical modelling for DFB lasers 235

(7.61)
This program is given in filti under the directory filter. The theory can
be tested out analytically by noting that, in the normalised frequency
domain, there is a phase shift of exp(jd)M) per whole time step (i.e.
(bM-(x)Ms/vg) and then use of eqn. 7.60 with half steps gives the
spectral form
BOu(=A[n cos(j(bM)/{co${jd)M)+jKsin(Td)M)} (7.62)
As already noted, Lax averaging replaces aIn with [aIn (+2) + % ( - 2M/2
as in eqn. 7.60 and filters the signal as cos(2 coM) with the Nyquist
limitations — 7T<coM< TT. The parameter fallows the curvature of the
gain with frequency in the central part of the filter to be changed so
that, if JK=1, there is a spectral power filter varying as I cos(2 (x>M) '2- If
0 < K< 1 the spectrum is flattened with respect to the situation with
K=\ while K> 1 narrows the gain bandwidth, increasing the gain
curvature. The filter is stable provided that K> 0.
It is frequently more useful to have the spectral filter offset from the
central laser frequency by an amount co0gsel so that (oM—*coM — (o0jj-sel:
I B(hu 12= \Aln I V [1 + {(coM - (o<gM) r ) 2 ] (7.63)
The implementation of this offset frequency in the conventional
frequency domain replaces exp(ja)Mt) with exp[j(a)M — (*)Ojfsel)i\. In the
normalised frequency domain with one positive space step, there is a
phase shift of exp( — j(oojpe(). This phase shift can be incorporated into
the digital filter by writing @ = Gxp(jd)0^sel) and then putting

7.8.2 Numerical implementation

The checking of this filter theory is done in the program filti within
the directory filter using either or both analytic theory and a white-
noise stochastic excitation, which is another interesting variant for
modelling and helps to build confidence that the filter really works on
a stochastic input. Figure 7.9a shows a typical result where K=1.5 and
no phase shift is included while Figure 7.9b shows the same result with
a phase shift. In these two figures, the ' — 3 dB'-gain points are not
conventional half-power points but are the normalised frequencies
where the real (output/input) falls to 1/V2 of its peak value. The
reasons for this are discussed in Section 7.9 on gain filtering.
The same filter can be used if required to filter spontaneous noise.
For spontaneous noise, the 3 dB points refer to the frequencies where
236 Distributed feedback semiconductor lasers

Frequency -Gain Spectrum Frequency -Gain Spectrum

net real.... imag (pos -.-. n e g — ) net real.... imag (pos-.-. n e g — )
dB dB
K parameter and offset radians 1.5 , 0 K parameter and offset radians 1 , 0.8
normalised centre ( normalised centre
[gmax/g(0)](dB) 0 5 frequency 0.2539 [gmax/g(0)](dB) 1.428
5 frequency
-0.2578 3dB points 0.2578 -0.1094 3dB points 0.6172
0 0

(a) -5 00
-5

-10 -10

-15 -15

Normalised Frequency Normalised Frequency © M

Frequency - Power Spectrum

dB
K parameter and offset radians 1 , 0.8
0011 2539
ESSS? " ° gainnuix(dB) 0
5
-0.2441 3dB points 0.7539
0
(C)

-5

-10

-15

-7t Normalised Frequency ©M n

Figure 7.9 Spectralfilterforfinite-differencescheme
(a) No offset frequency
(b) Offset frequency
(c) White-noise input

the noise power falls to half of its central value. Figure 7.9c shows the
output from the same filter as given in Figure 1.9b but now
implemented with a white-noise input (using the program filtin).
On examining the result of Figure 1.9 c and considering this applied
to gain, the reader may be disturbed at seeing the 'output' rise again
at the Nyquist-frequency edges. This problem, if it is a problem, is
removed by an alternative filtering demonstrated in the programs filt2.
In this alternative program, the gain always falls to zero at the Nyquist-
band edges but the gain profile is not symmetrical about the gain peak.
 Numerical modelling for DFB lasers 237

A combination of both techniques in filti and filt2, in principle,

provides a versatile armoury for tailoring the gain with frequency or
shaping the spontaneous spectrum with frequency. Further ideas for
filtering the gain, using the principles of TLLM combined with digital
filtering, are provided by Nguyen et al [35]. It turns out that, in DFB
modelling, it is probably more important to use a filter to tailor the
gain with frequency than filter the spontaneous emission. The
spontaneous spectrum can be prefiltered in a lookup table without
greatly slowing down the computation, but this risks introducing
unwanted periodicities if the table is too short.

7.9 Application of the filter theory to gain filtering

7.9.1 General
Care is needed in implementing the gain filter of Section 7.8 because
the field gain over a distance z is given from Gjidd =exp(j gdz) while the
power gain over the same distance is given from Gpoxm.=exp(2 \ gdz).
The integrals are not straightforward when the gain varies with time.
Now the physics requires the gain g per unit length to be some
function of frequency so that the temporal filter must affect g, as in
Section 7.8, rather than act on GfieU or Gpower, where there would be a
more complicated spectral action. Note that, even if g((o) = g(co0), a
real value at co=(o0f then on moving away from this central frequency,
the Kramers-Kronig relationships (Appendix 7) force the gain to take
complex values g((o)—+grmi(co)+jgimag(a))9 where the magnitude of
gimng((o) depends on the local curvature of grml(a>). However, it is only
gmd that changes the power gain per step; the imaginary component
merely changes the phase change per step. It was for this reason that,
in Figure 7.9a and b, the bandwidth was related only to the real part of
the gain.
A further complication comes from the fact that, as the gain
increases with an increase in the electron density, so, at the same time,
there is an associated reduction in the permittivity of the material
leading to a complex differential gain g(l+jaH) which changes with
frequency and electron density. Here aH is known as Henry's
linewidth-broadening factor [36, 37], taking values typically between 1
and 7 depending on the material and structure of the lasing waveguide
[38]. However, it has to be recognised that the net complex refractive
index is made up of several contributions with each contribution
varying in distinctive ways. Appendix 7 addresses this issue with three
238 Distributed feedback semiconductor lasers

components: gain, loss and the plasma effect. At the central gain peak
g(o)0) at angular frequency OJ0 it is changes in the plasma effect (or
oscillation of a plasma of nearly free electrons) to which is attributed
changes in the phase term jg((o0)aH, with the value at the trans-
parency density taken as the reference level. Over the frequencies
where the gain is strong, this phase contribution is assumed to remain
approximately constant. The Lorentzian filter is then applied only to
the net-gain/loss term and not to the term in aH:
-ii}0)} (7.65)
with
grml(a))=g(co0)/{l + i*(w-<*0)2} (7.66)
and
^(ai--a) o ) 2 } (7.67)
For each spatial step of length s, the complex magnification mL of eqn.
7.42 is given by exp{g(&>) (l+jaH)s} and is now modelled as
mL=((1+Tgs)/(1-ygs))*exp(j*ahs) (7.68)
where the MATLAB vector ahs is the array of values of {g(<o0)aHs} that
are allowed to vary at each step dependent on the central gain
associated with the electron density at that step, but are taken to be
independent of the frequency of lasing. The real part of this
magnification at the central frequency is given by exp{g(a)) s} per step
represented in eqn. 7.68 by its central difference approximation of
( l + | g s ) / ( l - | g s ) as in eqn. 7.42 (with the substitution that L/N = s)
but now as the frequency changes, g(co) s, or equivalently the term gs,
will be modelled in accordance with the filter principles just outlined.
At the time of writing, the definitive paper in this area of complex gain
modelling with frequency has perhaps not yet been written, and more
needs to be done numerically, experimentally and physically, but [35]
provides important ideas for transmission-line laser modelling.

7.9,2 Filtering the gain in the travelling-wave equations

The application of the filter theory to the gain needs care. Consider
eqns. 7.44-7.48 along with the definitions of mL, sint and cost in eqns.
7.42 and 7.43. Initially, consider just a real gain dependent on the
electron density which will change much more slowly with time than
the temporal changes in the optical fields. Consequently, from eqn.
7.68 one can rewrite the central-difference approximation hidden in
 Numerical modelling for DFB lasers 239

eqns. 7.44-7.48 for the effect of gain in the nth section as

(7.69)
where
C(n) = exp{j ahs(n)}{cost(n)ff(n)+j sint(n)fr(n + 1)}.
The programs for the distributed lasers store sets of the current values
of the discretised spatial forward/reverse fields {ff(n) fr(n)} along the
laser. Then there is also the set of spatial fields {ffo(n) fro(n)} for the old
(or previous) time step, and from the current fields and the old fields
are generated the updated or new set of fields all along the laser at the
next time step (ffn(n) frn(n)}. The assumption is made that the changes
in electron density occur sufficiently slowly over two time steps so that,
between these three temporal values of 'old', 'present' and 'new', all
quantities involving electron densities can be taken to be values at the
present time step. The phase changes caused by Henry's a factor are
assumed here to vary at a similar rate to that of the electron density.
Consequently, in this approximation, all the gains, phase factors from
aH and transmissions/reflections from the grating (i.e. g, ahs, cost and
sint) are evaluated at their values for the current time step in eqn. 7.68.
For time steps of 0.1 ps, this is usually an entirely reasonable
approximation with electron densities varying on the 10 ps scale for
100 GHz modulation frequencies.
Now rearrange eqn. 7.69 so that the gain g(n) is entirely on one side
and write
B(n + 1)«ig(n)sA(n) (7.70)
where
B(n + 1) = {ffn(n + 1)-C(n)}with A(n)={ffn(n + 1) + C(n)}.
The indexing of A(n) and relating it to B(n + 1) is a matter of arbitrary
choice but, having made the choice, eqn. 7.70 determines the
definitions. Remember that n denotes the index of the spatial
position, and then consider that each 'element' B of the spatial field is
the 'output' from an 'input' A with a gain multiplier fg(n)s rather than
unity as previously considered in Section 7.8. Use the notation Bo and
Ao to indicate the stored versions of the old (or previous) time step.
Applying Section 7.8 to update the output B as in eqn. 7.64, using the
same definition of K( = K) and writing @ = exp(y dF) = Ep with dF
determining the normalised frequency offset as in filti. All this leads
to
1)} (7.71)
240 Distributed feedback semiconductor lasers

where
Bo(n + 1) = {ff(n + 1)-Co(n)} with Ao(n) = {ff(n + 1) + Co(n)}
with
Co(n) = exp{j ahs(n)}{cost(n)ffo(n)+j sint(n)fro(n + 1)}
Eqn. 7.71 is then unravelled, retaining the assumption that the gain
and phase changes caused by the electron density vary sufficiently
slowly, to give the result
ffn(n + 1) = ma(n)C(n>-mb(n)Co(n) + mc(n)ff(n + 1) (7.72)
where
ma = [{K+1+Tg(n)s}/{K+1-yg(n)s}] (7.73)
mb=Ep[{K-1-yg(n)s}/{K+1-Tg(n)s}] (7.74)
mc=Ep[{K-1+yg(n)s}/{K+1-ig(n)s}] (7.75)
The updated reverse field frn(n) is found in a similar fashion and
gives
frn(nr) = ma(nr)D(nr)-mb(nr)Do(nr) + mc(nr)fr(nr) (7.76)
where
D(nr) = exp{j ahs(nr)}{cost(nr)fr(nr + 1)+j sint(nr)ff(nr)}
and
Do(nr) = exp{j ahs(nr)}{cost(nr)fro(nr+1)+j sint(nr)ffo(nr)}
Equations 7.72-7.76 achieve the objectives of giving the new fields
{ffn(n) frn(n)} at each time point in terms of the present fields {ff(n)
fr(n)} and old fields {ffo(n) fro(n)}. The gain filter now is a function of
modulation frequency and has been designed to give zero gain at the
Nyquist-band edges, adjusted for any frequency offset. This sig-
nificantly reduces the risks of aliasing errors or unphysical
oscillations.

7.93 Numerical implementation

A program gain in the directory filter implements the above theory
and shows how the noise can be amplified over a controllable narrow
band (Figure 7.10). Note that if loss is included then the gain is
negative (attenuation) at the Nyquist limits, even though the peak gain
has been offset by a considerable fraction of the Nyquist-frequency
range. This reinforces the point made towards the end of Section 7.8
that it is the real part of g which determines the net gain. This real net
 Numerical modelling for DFB lasers 241

gain-frequency (white noise input)

K= 2

20 frequ. offset 1

-10

—71 normalised frequency n

Figure 7.10 Gain/frequency with gain filter
Amplification of spontaneous emission

gain gives strong attenuation at the Nyquist limits unless there are
exceptional frequency shifts. Further shaping of the digital filter could
be accomplished by making K complex as, for example, when using
the program filt2 in the directory filter.

7,10 Basic DFB laser excited by spontaneous emission

7A 0.1 Introduction and normalisation
This section brings together the previous work and provides numerical
models for a range of DFB lasers. Several programs in directory dfb are
common to each master program determining the type of DFB (e.g.
dfbuni for the uniform laser). Not all MATLAB terms (which are in
sans-serif type face here) correspond 1:1 with the mathematical
symbols so that the reader must make an appropriate translation with
the help of the interface provided by this section and 1readme.txt files
available on the web.
Normalisation can aid in evaluating the relative magnitudes of
parameters. For example, normalising the electron density N to the
value Nlr required for gain transparency suggests that with a high-gain
242 Distributed feedback semiconductor lasers

and low-loss material, N/Ntr (—>Ne in programs) will perhaps vary

around 1-2 and, once lasing starts, will vary little and slowly compared
with the optical fields. At Ne~l, it is also possible to set a
recombination time rr for the electrons and holes so that the rate of
feed of electrons into a given volume Ts&L will be around NlrTs&L
electrons per unit time which can also give a normalising value for the
photons emitted per unit time. When a laser lases strongly with
negligible losses and only radiative recombination (T,~TV/,), any
increase [A(N(tTsiL/rr)] in the input electrons per unit time must
give the same change of photons out per time—this is a useful
numerical check.
Parameters measured in inverse length are normalised by multi-
plying by the length of the laser, e.g. (KL) giving the coupling, (g.mL)
the material field gain; and (aL) the material optical-field-loss
coefficient. In MATLAB code in the programs provided these are
written respectively as kL, gL and aL.
This computation is now for a laser and not an amplifier, so that
there is no longer a well prescribed central operating frequency, at
least not until the computation has been made. The Bragg offset-
frequency parameter 8 giving a phase change jS per unit length in
eqns. 7.37 and 7.38 becomes now a measure of how the central Bragg-
grating frequency departs from the central gain peak. A prime cause of
this change is given by the fact that, as the electron density increases
the gain increases by /\gm and there is a reduction in the refractive
index which changes the optical length of the grating and changes the
Bragg frequency. All this can be accounted for by a phase change
jaH Agm per unit length associated with the gain where aH is Henry's
linewidth-broadening factor. A reference level is required, and here
this is taken by choosing 8L as zero whenever the electron density is at
the transparency value N(r. This is not a necessary assumption, but
provides a simplification in this book which usefully limits the
computed frequency ranges. All this leads then to a complex material
gain given by gmL-j8L (written as gL+j*dL in the MATLAB code) with
8=aH/\gm where Ag*w measures deviation of gain from the transparency
value. The fields F and R are coded as arrays ff and fr with N values
giving the fields at each of the N spatial steps at the current time.
Typically, as orders of magnitude, (KL) and (aL) ~1 with T(gmL) -10
where F is the confinement factor -0.3 for bulk material lasers and
aH -4, again dependent on the material. For quantum-well materials,
the confinement factor is smaller because the gain is confined to the
narrow quantum wells, but the gain is higher, leading to similar or
 Numerical modelling for DFB lasers 243

larger values of T(gmL). The electromagnetic fields F and R are

normalised so that F*F and i?*i?give the photons per second flowing
from left to right and from right to left, respectively, at a local position
z so that the photon density per unit length is (F*F+R*R)/vg taken
over the effective guide area <s4.

7A 0.2 Field equations

The field-advection equations at the position z may then be written as

L =j{KL) R+ [n {gmL)
\ V Yt+fz) ~j

-j(8L)\-(aL)]R+(Lisl,r) (7.78)

These equations are computed numerically, including the gain filter,

using the codes that have been developed. The gain peak can shift with
electron density and temperature but this is not implemented in any of
the sample programs. The value of KL could change at each space step
to model graded gratings but is taken here as constant. The terms Lisp
represent the spontaneous excitation, evaluated from the bipolar
radiative recombination, taken as locally proportional to N2 and
include the spontaneous confinement factor which is similar to the
gain confinement factor P. The latter gives an effective gain below the
material gain and reduces the effective phase shift associated with the
complex gain.

7.103 Charge-carrier rate equation

The assumption of electrical neutrality means only a single rate
equation is required for the carrier density iV (electrons) ~P (holes) in
order to determine the local recombination and stimulated emission.
This density N changes with distance along the laser and so is a
function of the section number. The electron number per unit length
of the interaction is then NT$& where s& is the effective optical cross-
section and r<s$ is the effective electronic cross-section, with T the
confinement factor assumed to be independent of optical power level.
The electronic rate equation per unit length is then given from
244 Distributed feedback semiconductor lasers

(F*F+R*R\ I
)7 (7.79)
vg ) qL

where

N ^

The optical density (photons per unit length) is given from

(F*F+R*R)/vg as explained above. In eqn. 7.80 the recombination
really has three terms:
(i) a linear term AN (partly attributed to nonradiative recombina-
tion) but A=0 is assumed here for simplicity;
(ii) a quadratic or 'bimolecular' term BNP-N Bn(N/Ntr) (attributed
to radiative recombination); and
(iii) a 'cubic term' (CN2P+C"NP2) ~NCn(N/Nlr)2 (attributed to
nonradiative Auger recombination which does not contribute to
the lasing field but removes carriers from the photonic inter-
action, so reducing the quantum efficiency).
The approximations refer to the assumption of electrical neutrality
N~P. The effective electron recombination time T T O ~ 1 0 ~ 9 S , at the
normalising carrier density A^-10 2 4 m~3, should be contrasted with
the photon lifetime ~10~ 12 s.
Normalisation is now made in eqn. 7.79, writing N/Ntr—+Ne with this,
in the MATLAB code, going to the array Ne varying from step to step.
With uniform material then Ntr is uniform throughout the device but
could be allowed to vary. The effective cross-sectional area for
electronic/photonic interaction is T$l so that vgNtr Ys& gives a
normalising electron flow in numbers per second:

+ gM
r vr/
vgrr

(7.8!)
vgrm [q(NfrTML)}

where rr is now a function of Ne with rro the value when Ne = 1.

 Numerical modelling for DFB lasers 245

Knowing that the operational electron density will typically be a few

times Ntr so that Ne~2-4 provides an order of magnitude for the
threshold drive current I-NeqN(rTdL/rm. Such rough order-of-
magnitude checks can be reassuring in initial computations. Taking
L-100 jjim, vg ~1014 |xm/s and rr -10 ~9 s, this normalisation indicates
that (L/vg Tr) -10" 3 which is more than three orders of magnitude
smaller than (L/s) (-30) where s is the spatial-step length for the
computation. Hence the electron density in general changes on a
much slower time scale than changes in the photonic fields. As
indicated earlier, this relatively slow variation in eqn. 7.80 allows
several key approximations to be made in the modelling:
(a) The electron density is approximately constant over two or three
time steps even though the optical field may change significantly
in these time steps.
(b) A forward finite-difference scheme has adequate accuracy for the
electron equation (eqn. 7.80) which is composed of entirely real
quantities. For the more rapidly changing complex optical fields
in eqns. 7.77 and 7.78, a second-order central-difference scheme is
used giving greater accuracy.
(c) Note that in the instantaneous stimulated emission current given
from the real term proportional to g(F*F+R*R), the instanta-
neous gain gis fundamentally real. Provided the laser is operating
around a reasonably flat peak gain, then gain variations caused by
the output being at different frequencies need not be added to the
time-domain equations and indeed such routines have not been
added in this book. However, if the laser is deliberately designed to
operate away from the gain peak this gain has to be reduced
appropriately.
While iteration would work, one recognises that the real
stimulated recombination is more correctly given from the terms
{(gF)F*+(gF)*F+(gR)R* + (gR)*R} (7.82)
The method of the complex Lorentzian filter can then be used on
the complex terms (gF), (gf)*> (gR)> a n d (gR)*« Alternatively, the
methods of eqn A.5.30 could also be extended by writing
g—[g(co0)-j(dg/dco)Qd/dt-j(d2g/dco2)0d2/dt2] (7.83)
Now, with o)Q giving the peak gain taking (dg/dco)0 = 09 consider
eqns 7.83 and 7.82 and integrate by partial integration over one
cycle of time. The partially integrated term and its conjugate are
approximately zero, giving a real stimulated current (which is
subsequently restricted to be non-negative) from
246 Distributed feedback semiconductor lasers

g{F*F+R*R) -+{g(a>0) (F*F+R*R)

+T(d2g/do)2)0[dF/dt*dF/dt+dR/dt*dR/dt]} (7.84)

With negative gain curvature at the peak gain, any modulation of

the fields will reduce the effective gain and reduce the stimulated
emission current as the rate of modulation (i.e. deviation of
frequency from the gain peak) increases.

7.10.4 Numerical programs

Program dfbuni in directory dfb brings these results together for a
uniform DFB laser with real coupling constant K. A vectorised system
of equations helps to shorten the code and is well suited to MATLAB
with the 'guts' of the program in the subroutine dfbrun. This program
demonstrates that, at first switching on the drive to the laser, the
electron density first has to increase, and to shorten the duration of
the turn-on process, the electron density is started close to its
transparency value. The first couple of runs simply show spontaneous
noise appearing until a threshold electron density is reached when
oscillations appear in the output waveform. Because the perfectly
uniform DFB cannot discriminate the thresholds of the upper or lower
mode close to the stop bands, both modes initially appear and which
one finally wins depends upon the value of KL and the spontaneous
emission. The reader may explore this directly but also may prefer to
start with a program such as that for the phase-shifted DFB that gives
the 'same' results on each run.
The program dfb14 is a very similar DFB but now with a Aw/4 phase
shift in its centre. Figure 7.11 gives examples of the typical output that
one expects to find from this program. Here there is strong variation
with distance of both the electron density (spatial-hole burning) and
the optical-intensity profile within the laser. This laser now switches on
in one clear mode which is in the middle of the conventional stopband
for the uniform DFB.
The program dfb218 is again a similar DFB, as previously, but now
with two phase shifts of Am/8, one on either side and offset from the
laser's centre. This program shows that the lower frequency is the
favoured mode of lasing with considerably less spatial-hole burning
than for the single central phase shift. The program dfb238 simply
changes these phase shifts from Aw/8 to 3Aw/8, and now the higher
frequency is preferred. As discussed in Chapter 5, the detailed position
of the phase shifts affects the performance and 'dfb238' lasers are
generally less stable with a higher drive level than 'dfb218' lasers.
 Numerical modelling for DFB lasers 247

tog output v. time, ps tog output v. time, ps

•601
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 10 15 20 25 30
field section number
a b c
normalised power spectrum v. frequency. GHz normalised electron density v. dtetance photon/s (solid sum, dotted forward, dashed reverse)

1.5

•600 -400 -200 0 200 10 15 20 25 30 5 10 15 20 25 30

frequency, GHz section number field (section) number
d e

Figure 7.11 Typical output for dfbH (a DFB with one central KJ4 phase shift)
a initial run; b continuation; c field profile with spontaneous emission;
d spectrum; e electron density-distance; / field profile without
spontaneous emission

Recent changes in technology have made devices with gain coupling

possible. Here the value of K is complex and the sign of the imaginary
component determines whether the laser prefers the upper or the
lower frequency around the stopband of the DFB. Strictly, the value of
the imaginary component should change with electron density.
However, here a constant complex K=Kn>al+JKgftin is inserted in a
program called dfbgain. By changing the sign of Kgain, the modal-
selection properties of gain gratings can be demonstrated. The
program would be very similar for a second-order grating but in a
second-order grating there would be additional attenuation in the
unscattered fields.
The program has the ability to run with bandwidth-limited (i.e.
filtered) spontaneous emission. This slows down the program and has
little practical effect because the gain filter is included as standard and,
as stated earlier, once the gain becomes attenuation at the Nyquist-
band edges the detailed shape of the spontaneous emission has little
effect. The laser can also be run without the spontaneous emission
248 Distributed feedback semiconductor lasers

provided that lasing is well established. This can provide 'ideal'

theoretical field profiles. To operate this facility, wait until the laser has
settled and stopped between runs, go into the keyboard and type
spontO and return. Appendix 12 summarises the programs that are
available, while 1readme.txt files, provided within each directory, give
more operational detail.

7.11 Summary
The chapter started with discussions on the numerical solution of an
ordinary differential equation with complex rates of growth, high-
lighting the importance of the straightforward central-difference
method. The advection equations or travelling-wave equations were
considered, focusing on a central-difference technique with appro-
priate steps in space and time. Gain and phase changes were
introduced along with coupled reflections. At each stage, numerical
stability was considered and model test programs given. A uniform
Bragg-grating amplifier was considered and the numerical model of a
forward travelling wave inside a Bragg grating was implemented in
MATLAB and checked against an analytic model.
The processes of spontaneous emission are modelled by uncorre-
lated Gaussian excitations injected into each section in the laser, and
tutorial models of the spontaneous-emission processes were made
available in MATLAB code.
The process of discretisation leads to limitations, and these were
briefly discussed, particularly in relation to the fast Fourier transform.
Strategies for Lorentzian spectral filters were outlined and then
implemented to ensure that the gain inside the laser varied with
frequency in an appropriate manner. The use of stochastic excitation
to investigate broadband behaviour of a filter was demonstrated.
Finally, a basic program for a DFB laser operating around the peak
gain with spontaneous emission was discussed and implemented in
MATLAB code.
It is a pleasure to acknowledge that many of the ideas in this chapter
owe a considerable debt to research students Chi Tsang [39] who
participated in the European COST-240 exercise [40] , Mark Nowell
[41] and Barry Flanigan [42], Si Fung Yu [43] and David Jones [44]
and especially to research associates Li Ming Zhang [43,45] (see also
[12]) and Dominique Marcenac [45-47]. Their work has been pulled
together with a more consistent use of central-difference methods for
laser-diode DFB analysis than appears to have been previously
 Numerical modelling for DFB lasers 249

established. It is hoped that the cut-down programs for use with

student MATLAB will also encourage the reader to gain a better
understanding of the numerical processes and some of their limita-
tions, looking on fast numerical processing as an enjoyable
experimental process.

7.12 References
1 AGRAWAL, G.R, and DUTTA, N.K.: 'Semiconductor lasers' (Van Nos-
trand Reinhold, New York, 1993), 2nd ed., Section 6.6.1
2 THOMPSON, G.H.B.: 'Physics of semiconductor laser devices' (J. Wiley,
Chichester, 1980), Section 7.3.2
3 CARROLL, J.E.: 'Rate equations in semiconductor electronics' (Cam-
bridge University Press, Cambridge, 1985), Section 6.4.2
4 NAGARAJAN, R., ISHIKAWA, M., FUKUSHIMA, T., GEELS, R.S., BOW-
ERS, J.E., and COLD REN, LA.: 'High speed quantum well lasers and
carrier transport effects', IEEE J. Quantum Electron., 1992, 28, pp.
1990-2007
5 OHTSU, M.: 'Highly coherent semiconductor lasers' (Artech House,
Boston, 1991), chap. 2
6 PETERMANN, K.: 'Laser diode modulation and noise' (Kluwer Academic
Publishers, Dordrecht, 1991)
7 VANKWILEBERGE, P., MORTHIER, G., and BAETS, R. 'CLADISS—a
longitudinal multimode model for the analysis of static and stochastic
behaviour of diode lasers with distributed feedback', IEEE J. Quantum
Electron., 1988, 24, pp. 2160-2169
8 MORTHIER, G., and VANKWIKELBERGE, P.: 'Handbook of distributed
feedback lasers' (Artech, Boston, 1997)
9 BUUS, J.: 'Principles of semiconductor laser modelling', IEE Proc.—-J.,
1985, 132, pp. 42-51
10 COLDREN, LA., and CORZINE, S.W.: 'Diode lasers and photonic
integrated circuits (j. Wiley, New York, 1995), Section 3.5.2
11 BJORK, G., and NILSSON, O.: 'A new and efficient numerical matrix
theory of complicated laser structures. Properties of asymmetric phase
shifted DFB-laser'/ Lightwave Technol, 1987, 5, pp. 140-146
12 ZHANG, L.M., and CARROLL, J.E.: 'Large signal dynamic model of the
DFB laser', IEEEJ. Quantum Electron., 1992, 28, pp. 604-611
13 LOWERY, A.J.: Transmission line modelling of semiconductor lasers: the
transmission-line laser model', Int.J. Numer. Modell, 1989, 2, pp. 249-265
14 LOWERY, A.J.: 'A qualitative comparison between two semiconductor
laser amplifier equivalent circuit models', IEEE.J. Quantum Electron., 1990,
26, pp.1369-1375
15 LOWERY, A.J., GURNEY, P.C.R., WANG, X-H., NGUYEN, L.V.T., CHAN,
Y.C., and PREMARATNE, M.: 'Time-domain simulation of photonic
devices, circuits and systems'. Proceedings of Physics and Simulation of
Optoelectronic Devices, SPIE Photonics West, 1996, 2693, pp. 624-635
250 Distributed feedback semiconductor lasers

16 HOEFER, W.J.R.: 'The transmission-line matrix method—theory and

applications', IEEE Trans., 1987, MTT-35, pp. 370-377
17 LOWERY, A.J.: 'A new dynamic semiconductor laser model based on the
transmission line laser model', IEEProc.—J, 1988, 135, pp. 126-132
18 WONG, YL, and CARROLL, J.E.: 'A travelling wave rate equation analysis
for semiconductor lasers', Solid-StateElectron., 1987, 30, pp. 13-19
19 LOWERY, A.J.: 'A new dynamic semiconductor laser model based on the
transmission line modelling method', IEE Proc.—/., 1983, 134, pp.
281-289
20 LOWERY, A.J.: Transmission-line modelling of semiconductor lasers: the
transmission-line laser model', Int.J. Numer. Modell, 1990, 2, pp. 249-265
21 COHEN, A.M.: 'Numerical analysis' (McGraw Hill, London, 1973)
22 ISERLES, A.: 'Numerical analysis of differential equations' (Cambridge
University Press, 1996)
23 LAPIDUS, L., and PINDER, G.F.: 'Numerical solution of partial differ-
ential equations in science and engineering' (Wiley , New York, 1982)
24 PRESS, W.H., TEUKOLSKY, S.A., VETTERLING, W.T., and FLANNERY,
B.P.: 'Numerical recipes in C (Cambridge University Press, 1992), 2nd
ed.
25 T h e student edition of MATLAB' (Prentice Hall, Englewood Cliffs,
1995)
26 ISERLES, A.: 'Numerical analysis of differential equations' (Cambridge
University Press, 1996), chap. 5
27 LAPIDUS, L., and PINDER, G.E: 'Numerical solution of partial differ-
ential equations in science and engineering' (J. Wiley, New York, 1982), p.
566
28 IFEACHOR, E.C., and JERVIS, B.W.: 'Digital signal processing' (Addison-
Wesley, Wokingham and Reading, 1993)
29 ABRAMOWITZ, M., and STEGUN, I.A.: 'Handbook of mathematical
functions' (Dover, New York, 1965), p. 70
30 ISERLES, A.: 'Numerical analysis of differential equations' (Cambridge
University Press, 1996), p. 312
31 CLOUDE, S.: 'Electromagnetic book' (UCL Press, London, 1995)
32 CUNNINGHAM, E.P.: 'Digital filtering' (Houghton Miflin, Boston, 1992)
33 COOLEY, J.W., and TUKEYJ.W.: 'An algorithm for machine computation
of complex fourier series', Math. Comput., 1965 19, pp. 297-301
34 WESTBROOK, L.D.: 'Measurement of dg/dN and dn/dN and their
dependence on photon energy in A =1.5 fxm InGaAsP laser diodes', IEE
Proc.—J. 1986, 133, pp. 135-142
35 NGUYEN, L.V.T., LOWERY, A.J., GURNEY, P.C.R., NOVAK, D., and
MURTONEN, C.N.: 'Efficient material gain models for the transmission-
line laser model', Int.J. Numer. Modell. Electron. Netw., Devices Fields, 1995, 8,
pp. 315-330
36 HENRY, C.H.: Theory of linewidth of semiconductor lasers', IEEE J.
Quantum Electron., 1982, 18, pp. 259-264
37 HENRY, C.H.: 'Phase noise in semiconductor lasers',/. Lightwave Technol,
1986, 4, pp. 298-311
 Numerical modelling for DFB lasers 251

38 OSINSKI, M., and BUUS, J.: 'Linewidth broadening factor in semi-

conductor lasers—an overview', IEEE Quantum Electron., 1987, 23, pp.
9-28
39 TSANG, C.F.: 'Dynamics of distributed feedback lasers'. PhD dissertation,
Cambridge, 1993
40 MORTHIER, G., BAETS, R., TSIGOPOULOS, A., SPHICOPOULOS, T.,
TSANG, C.F., CARROLL, J.E., WENZEL, H., MECOZZI, A., SAPIA, A.,
CORREC, P., HANSMANN, S., BURKHARD, H., PARADISI, A., MON-
TROSSET, I., OLESEN, H., LASSEN, H.E., and SCHATZ, R.: 'Comparison
of different dfb laser models within the European cost-240 collaboration'.
IEE Proc.—Optoelectron., 1994, 141(2), pp. 82-88
41 NOWELL, M.C.: Tush-pull directly modulated laser diodes'. PhD
dissertation, University of Cambridge, 1994
42 FLANIGAN, B.J.: 'Advances in push-pull modulation of lasers'. PhD
dissertation, University of Cambridge, 1996
43 ZHANG, L.M., YU, S.F., NOWELL, M.C., MARCENAC, D.D., CARROLL,
J.E., and PLUMB, R.G.S.: 'Dynamic analysis of radiation and side mode
suppression in second order DFB laser using time-domain large signal
travelling wave model', IEEE J. Quantum Electron., 1994, 30, pp.
1389-1395
44 JONES, DJ,. ZHANG, L.M., CARROLL, J.E., and MARCENAC, D.D.:
'Dynamics of monolithic passively mode-locked semiconductor lasers',
IEEEJ. Quantum Electron. , 1995, 31, pp. 1051-1058
45 TSANG, C.E MARCENAC, D.D., CARROLL, J.E., and ZHANG, L.M.:
'Comparison between 'power matrix model' and 'time domain model' in
modelling large signal response of DFB lasers', IEE Proc. Optoelectron.,
1994, 141, pp. 89-96
46 MARCENAC, D.D.: 'Fundamentals of laser modelling'. PhD dissertation,
University of Cambridge, 1993
47 MARCENAC, D.D., and CARROLL, J.E.: 'Quantum mechanical model for
realistic Fabry-Perot lasers', IEE Proc—J., 1993, 140, pp. 151-171
 Chapter 8
Future devices, modelling and
systems analysis

8.1 Introduction
Computer-aided design (CAD) has been a vital initial ingredient in the
realisation of very large-scale integrated (VLSI) electronic circuits.
Now in optoelectronics projects, mathematical modelling, although at
present lacking the sophistication of CAD in VLSI, is being recognised
as a key component in the design of devices and systems [1]. One
thrust of the book has been to lay out the physical- and mathematical-
modelling techniques for the electromagnetic and electronic
interactions within distributed feedback lasers to give better explana-
tions and to facilitate new designs. This chapter outlines further areas
in optoelectronics where such modelling will be at the forefront in the
design of more effective prototypes. Optical systems where arrays of
devices may be interconnected are one such important area and are
discussed in Section 8.2. Novel concepts such as that of the push-pull
laser (Section 8.3), tunable lasers with Bragg gratings (Section 8.4),
and surface-emitting lasers (Section 8.5) are all candidate areas for
applications of the modelling principles given in this text. The future
for mathematical modelling, coupled with a sound physical under-
standing, is bright, extensive and assured.

8.2 Systems analysis

8.2.1 Introduction
The time-domain techniques (Chapter 7) for solving the electro-
magnetic interactions in lasers have led to powerful and efficient
routines which are capable of solving a wide range of (single- and)
 Future devices, modelling and systems analysis 253

multimodal problems covering Fabry-Perot lasers, uniform and phase-

shifted DFB lasers, all of which may be driven with digital or analogue
modulation. Similar routines have also been built using transmission-
line laser modelling [2-4], an alternative time-domain technique with
many of the virtues of the techniques discussed in Chapter 7.
Considerable effort has been put into developing a general optical-
system simulator based on embedded-device models that are
interconnected with passive components so as to mimic system
performance [5]. The laser simulator can, for example, model high-
speed digital-modulation performance, fast pulse generation, passive
mode locking, and dynamic instabilities. Extending such time-domain
analyses to optical subsystems and systems containing many compo-
nents will provide excellent insights into the interactions between
these components over a wide range of frequencies. This section gives
examples of some initial applications of such an optical-systems
simulator.

8.2.2 Component modelling

The first ingredient for a 'systems simulator' is to have numerical
models for all the components within the system. The basic numerical
techniques described in Chapter 7 can be applied to Fabry-Perot and
DFB lasers, and also to semiconductor optical amplifiers (SOA) which
are essentially Fabry-Perot lasers with weak (or zero) reflection at their
facets. Some geometrically complicated components may require
techniques such as beam-propagation [6] or 'finite elements' [7] to
solve the electromagnetic fields but these lie outside the scope of this
work. The numerical techniques can be applied to both the electronic
drives and the optical outputs/inputs. With an armoury of modelling
methods, one can analyse combinations of lasers, electro-absorption
modulators, ideal optical transmitters , y-junctions, lengths of dis-
persive fibre and so on, enabling complex systems to be tested
numerically with a high degree of confidence in subsequent experi-
mental trials.
Key features incorporated into such opto-electronic device/system
modelling are:
(i) spatial-hole burning in the density of the charge carriers along
the length of devices as the optical intensity varies;
(ii) time delays in injecting carriers from contact to active lasing
regions (carrier-transport effects via three-level models, for
example);
254 Distributed feedback semiconductor lasers

(in) nonlinear gain (e) as the photon density changes (introduces

damping into the response of a laser to modulation drive);
(iv) wide varieties of carrier recombination (linear, bimolecular and
Auger recombination and general terms of the form DNP);
(v) multiple-mode operation (time-domain techniques automat-
ically take many modes into account while frequency-domain
techniques using, for example, transfer matrices often require
one to track each mode separately);
(vi) losses can be fixed or dependent on charge-carrier densities
(e.g. waveguide absorption and scattering, and intervalence
band-absorption, respectively);
(vii) realistic gain variations with carrier density and wavelength
(linear/log, peak-gain against carrier-density relationships, and
changes of wavelength for the peak gain with carrier density);
(viii) realistic spontaneous-emission spectra with wavelength (changes
of wavelength for the peak spontaneous emission as carrier
density changes);
(ix) submounts with parasitic capacitances and inductances appro-
priate for the design of modulated laser/ampli-
fier/electroabsorption modulators;
(x) arbitrary drive-voltage waveforms into multicontact optoelec-
tronic devices;
(xi) reflections and radiation loss between sections of laser/ampli-
fier/passive devices (note that full time-domain modelling of
lasers attached to many kilometres of nonlinear fibre is difficult
because lasers typically need subpicosecond time steps requiring
billions of steps to model a millisecond delay fully); and
(xii) relative-intensity noise (RIN) due to inclusion of spontaneous
emission [evaluating RIN in the kilohertz region also poses
challenges for time-domain techniques for the same reasons as
in (xi)].
Various devices (e.g. the electroabsorption modulator) will need data
about absorption and refractive index variations and such variations
can be modelled using high-order (e.g. fifth-order) polynomial fits to
experimental or theoretical data at discrete sets of bias voltage,
frequency etc. and then stored in look-up tables as part of the model,
with interpolation providing data at intermediate points.
One useful component is the 'ideal optical transmitter' which
generates optical waveforms with no amplitude patterning or intrinsic
wavelength chirp though, of course, it contains the fundamental
linewidth broadening caused by the information modulation (Fourier
 Future devices, modelling and systems analysis 255

broadening). Such an ideal transmitter aids investigations of distor-

tion, introduced for example by an SOA, without the complicating
effects of an optical input having a nonideal waveform. Investigating
the propagation of 'information-bandwidth-limited signals' along
dispersive fibre links is another example where the ideal transmitter is
of benefit: a realistic chirped source initially just confuses the issues. To
model assemblages of subsystems and systems, it is essential to add y-
junctions, attenuators, filters, partial reflectors, fibre links and
four-port couplers. All these can be modelled at a variety of levels of
complexity but many of these devices have linear input-output
transfer functions with well defined spectral characteristics over the
range of required power levels. It is then sometimes better to convert
the time-domain input, using the FFT, to the frequency domain where
transfer-matrix operations are performed to give the output which is
then transformed via the inverse FFT back to the time domain. For
linear elements this is a good straightforward technique with wide
validity, however, considerable care is needed if the element is within
a loop. The inclusion of fibre gratings, modelled like a Bragg grating
in a laser, will be another component which leads to new and
interesting applications in systems.

8.2.3 System modelling

A systems model will have a library of default and bespoke models of
all the required devices which can then be selected and intercon-
nected in arbitrary configurations to form the optical subsystems and
systems. Such libraries will contain routines for interconnecting
devices and subsystems and the number of these interconnects will be
limited by the data storage available, remembering that each device/
subsystem will need to store such entities as drive voltages/currents,
input/output fields etc. for evaluation and for use in a later
computation. Typically, one will examine at various points in the
system:

(a) optical spectra and waveforms with selected time windows and, if
required, effects of dispersion would be included;
(b) baseband spectra and filtered waveforms using RC, Butterworth or
Bessel filters, typically with 1-7 poles;
(c) eye diagrams giving a qualitative indication of the system
performance;
(d) instantaneous frequency against time (from rate of change of
phase of complex field);
256 Distributed feedback semiconductor lasers

(e) instantaneous and mean powers over selected time windows; and
(/) data-transfer techniques for concatenating simulations and trans-
fer to other computer systems and output hardware.
To find wide acceptance within any research/development organisa-
tion, such model libraries need to ensure that the software is very 'user
friendly' and does not need significant prior knowledge of program-
ming languages or complex physics. The difficult problem of the user
interface is not addressed here but programs like MATLAB 5 are
coming to the rescue with more user-friendly methods for data input
and output.
In this section, results are presented from:
(i) the modelling of SOA post-amplification of output from high-
speed digitally modulated optical sources;
(ii) DFB laser integrated with electroabsorption modulator allowing
for residual reflections from modulator-output facet;
(iii) cross-gain and four-wave-mixing wavelength conversion in a
semiconductor optical amplifier; and
(iv) cross-phase wavelength conversion in a Mach-Zehnder inter-
ferometer structure incorporating two semiconductor optical
amplifiers.

8.2,4 10 Gbit/s power amplification

In the first example, an ideal optical transmitter injects a 10 Gbit/s
waveform with no patterning or wavelength chirp into the SOA. The
simulator generates the output from the SOA which includes the
amplitude patterning and wavelength chirp induced by the amplifier-
saturation process. The signal then is dispersed by a standard fibre and
the received eye diagram of this dispersed waveform is examined as a
function of the input optical power to the SOA by post-processing the
stored waveforms.
Figure 8.1 a shows the unfiltered transmitted eye diagram at
10 Gbit/s over one period with a pseudorandom sequence taken over
an interval of 4.1 ns. The output is plotted every 4.1 ps although the
time step in the computation is 0.05 ps. The optical output is then
dispersed by 70 km of standard fibre detected and filtered (fifth-order
Bessel, 7 GHz bandwidth) with Figure 8.1c showing the received eye
diagram and Figure 8.1b showing the instantaneous frequency of the
SOA output against time, for a 1 mW optical input to the SOA.
There is little amplitude patterning on the output from the SOA for
this design, and the received eye diagram is symmetric with an
 Future devices, modelling and systems analysis 257

Unfjltered transmitted 'eye' r Filtered received 'eye'

power, mW
15- 0.001 -i

10-
0.0005-
5 -

0
-0.05 0.05 0.10 0.15 -0.05 0 0.05 O10 O15
relative time, ns relative time, ns
a

Instantaneous frequency, GHz configuration

80- semiconductor
ideal optical 1 mW
optical
transmitter amplifier

-80- filter detector

u
relative time, ns 4

b d
Figure 8.1 Simulated amplifier performance for 1 mW peak optical input power

adequate opening. The plot (b) of 'instantaneous frequency' or

instantaneous rate of change of phase shows fluctuations which are
found to be linked to the modulation. There is a high amplitude of
frequency noise in the 'zeros' and a lower amplitude of frequency
noise in the 'ones'. This effect is caused by the increased significance
of the spontaneous emission noise and hence an enhanced phase
noise when the input power to the SOA is low. Further, the gain of the
SOA is larger in the 'zeros' and therefore also enhances the phase
noise.
The same plots are shown in Figure 8.2 but now with a 5 mW optical
input to the SOA. The SOA was not designed to operate at these power
levels, and more amplitude patterning is clearly seen with saturation
occurring over the duration of the pulse. There is also an increased
instantaneous frequency excursion at turn-on and turn-off with a
negative-frequency chirp at the leading edge of the 'ones'. This chirp
is associated with the gain saturation process in the SOA. The filtered
received eye diagram now shows an asymmetry because the isolated
'ones' are delayed relative to sequences of 'ones'. This is in agreement
with experimental work* and the model was used to investigate its
cause. The time constant associated with the gain-saturation process in
* COLLAR, AJ,: Private communication, 1966
258 Distributed feedback semiconductor lasers

unfiltered transmitted 'eye' filtered received 'eye'

power, mW power, mW
60 J 4-I

4a

20.

-0 05 6 0.05 0 0 0.15 -0.05 0 0.05 0.10 0.15

relative time, ns relative time, ns

instantaneous frequency, GHz configuration

40
ideal optical
5mW semiconductor
optical
0 - transmitter amplifier

-40.
filter detector
relative time, ns

Figure 8.2 Simulated amplifier performance for 5 mW peak optical input power

SOAs causes the isolated 'ones' to propagate through a reducing

average carrier density and increasing refractive index, resulting in a
dynamic chirp to longer wavelengths, unlike 'ones' in a sequence of
'ones'. In a dispersive standard fibre system, a temporal delay relative
to the rest of the waveform is found. In addition, the information
bandwidth broadens the spectrum which, in the presence of disper-
sion, changes the pulse shape.

8.2.5 Direct modulation: recapitulation

The next system to be discussed is the laser with an external modulator
instead of using direct modulation. From the earliest days, direct
modulation of the current driving a semiconductor laser has been
recognised as the low cost way of modulating the optical output from a
laser [8] but there are limitations which are well recognised. The first
of these is dynamic 'chirp'. To increase (reduce) the amplitude of the
optical output from a laser, the gain has to be increased (reduced). As
noted in Section 2.5, the linewidth-broadening factor aH relates such
changes of gain with necessary changes in the refractive index and
resulting changes in the frequency. As seen in Section 6.2.5, this
dynamic 'chirp' can increase the spread of frequencies to ~100 GHz
for 10 Gbit/s modulation—well beyond the spectral broadening
 Future devices, modelling and systems analysis 259

required from Fourier analysis of any conventionally modulated signal.

This chirp, if not limited, leads (as discussed in Section 1.5) to
spreading of the 'ones' into the 'zeros' giving rise to a loss of signal—
a dispersion penalty that can be unacceptable even over a few
kilometres.
Besides the dynamic chirp, there is also a fundamental limit to the
bit rate for such direct modulation of a laser because of the photon-
electron resonance discussed in Chapter 3. This limiting maximum
modulation frequency fMmax, combined with package limitations, is
usually significantly below 50 GHz depending on the differential
optical gain for the material. For modulation frequencies fM>fMmax> the
optical output-modulation efficiency typically falls off at over 12 dB per
octave increase in fM.
The limitation imposed by both dispersion and photon-electron
resonance may be mitigated by using a continuously operated DFB
laser but integrated with a modulator after the laser [9-12] which
turns the light on and off in a quite separate process. Such a
combination, often with an amplifier between the laser and modulator,
can also provide 'prechirp' which can so arrange the amplitude and
phase of the frequency components forming the modulated optical
pulse that the changes in group velocity with frequency within the fibre
cause the pulse shape to compress along standard fibre before the
pulse disperses again. This dispersion compensation combined with
the removal of the photon—electron resonance offers excellent
prospects for communication systems over standard fibre up to and
beyond 10 Gbit/s with distances in excess of 100 km or more [13-16].
This option is currently preferred for high-quality systems, in spite of
the extra cost and complexity.

8.2.6 Simulation of integrated DFB laser and electroabsorption

modulator
As mentioned above, external modulators have the potential benefit of
amplitude modulating a laser signal either without the large 'chirp'
that is caused by a^ in directly modulated lasers or with a modified
chirp formed to compensate for dispersion, but this latter is not
considered here. Integration of the laser and (electroabsorption)
modulator should give potential benefits of lower cost, higher output
power and ease of packaging. However, the problem with external
modulators that is exposed in this section is that, even with < 1 %
reflections from an antireflection-coated output facet, the modulator
can still cause complications by passing such reflections back through
260 Distributed feedback semiconductor lasers

0% modulator output facet reflectivity

power, mW power, mW
1 4

T 5 4
relative time, ns relative time, ns

0.2% modulator output facet reflectivity

power, mW power, mW
15. 4

10.

5.

1 2 3 4 5 1 \ I Y 5
relative time, ns relative time, ns
Figure 83 Simulation of effect of reflections from output facet of an integrated DFB
laser and electroabsorption modulator on dynamic performance of source
Time step for the computation is 0.0956 ps while the curves are plotted
every 4.11 ps

the modulator and interacting with the laser resulting in transient

oscillations which lead to amplitude and phase variations being
superimposed on the laser's nominally CW output field. Figure 8.3
illustrates the effects of changing the reflections from no reflection at
the output facet of the modulator to 0.2% facet reflection where
significant effects in terms of noisy power fluctuations and ringing on
the 10 Gbit/s-modulated waveforms can be observed. However, the
noise in the modulator output is less intrusive in the presence of
modulation but the modelling clearly explains why reflections <0.1%
are often specified.

8.2.7 Cross-gain and four-wave-mixing wavelength conversion in

anSOA
A good example of the use of the time-domain-modelling method
using multiple devices with multiple frequencies is given by consider-
ing wavelength conversion using an SOA subsystem. Figure 8.4 shows
an ideal optical transmitter launching a 10 mW CW optical
 Future devices, modelling and systems analysis 261
power, mW
10 passive Y-junction
ideal optical
transmitter
at 1558 nm bit stream output
semiconductor
1 time, ns
optical amplifier
10J Ideal optical 10mWCW at 1556 nm
power, at 1557 nm
(mW) transmitter
at 1557 nm at 1558 nm
nJ
at 1559 nm
1 time (ns) 5 time step for simulation 0.1 ps
time step for plotting 4.1 ps
Figure 8.4 Four-wave mixing for frequency shifting

power at 1557 nm into a 350 |mm-long SOA and a

second ideal transmitter at 1558 nm launching a 10 Gbit/s
'10110111011110111110100100010000100000' digitally modulated
signal with 10 mW peak and 1 mW minimum optical power. The
output spectrum shows signals at 1557 and 1558 nm, as expected, and
also the four-wave-mixing products at 1556 and 1559 nm, as indicated
in the spectrum of Figure 8.5.
The output spectrum is then optically filtered and Figure 8.6 shows
the output power against time centred on the wavelengths of 1556,
1557, 1558 and 1559 nm using passbands each with a total width of
1 nm. The modulated signal at 1558 nm emerges amplified but with
superimposed amplitude patterning associated with the SOA gain-
saturation process. The gain recovery in the zeros can also be seen.
The cross-gain saturation causes the CW signal at 1557 nm now to have
superimposed on it an inverted version of the data but with a poor
extinction ratio. The gain saturation leads to a completely different
shape to the waveform for the inverted signal as the modulation is now
driven entirely by the gain-saturation process, and so even an
instantaneous change in the data leads to a delayed response in the
inverted signal.
The four-wave-mixing products at 1556 and 1559 nm are shown in
Figure 8.6 and are typically over a factor of 100 weaker than the main
signals at 1557 and 1558 nm. Both mixing products (1556 and
1559 nm) show the data in noninverted form, with the 1556 nm signal
being the stronger of the two and exhibiting less distortion but having
a less satisfactory extinction ratio. It must be emphasised that these are
not optimised results but simply indicative of the abilities of time-
domain modelling to handle multiple frequencies.
The modelled example displays four-wave mixing as a result of the
fluctuations induced in the carrier density and hence variations in gain
262 Distributed feedback semiconductor lasers

fft time step 0.5 ps frequency spacing 244 MHz:-

u
10 - 8192 points for FFT optical spectrum
over interval 0.9 to 5 ns
10000-
energy density, J/Hz
100

1 -

1552 1556 1560 1564

Figure 8.5 Simulated output spectrum from amplifier

power, mW power, mW
10(h
0.2 1559 nm

50-
0.1-
1557 nm

1 2 3 4 3 4
time, ns time, ns

power, mW
power, mW
1556 nm
1558 nm
0.15- I
il
i—. i—

J 1
50- 0.1-
0.05-
^—I
1 2 3 4 1 2 3 4 5
time, ns time, ns

Figure 8.6 Simulated cross-gain and four-wave-mixing wavelength-converted signals

at amplifier output
 Future devices, modelling and systems analysis 263

caused by the beating of the two optical-input signals. From the

conventional analysis, the modulation of the carrier density decreases
rapidly with frequencies above the photon-electron resonance.
However, spectral-hole burning and carrier heating (see Section 2.4.4)
may contribute to modulation of the gain up to much larger
wavelength separations, and it is the gain modulation, for example,
which could give the reported conversion efficiencies of up to 0 dB for
16 nm of wavelength separation (corresponding to 2THz) [17, 18].
Two-photon absorption and the Kerr effect may also contribute to this
remarkable result, but the numerical models presented here would
need augmenting appropriately.

8.2.8 Simulation of cross-phase wavelength conversion in a

Mach-Zehnder interferometer incorporating two SOAs
Signals can be sent to different locations via different routes (space-
division multiplexing) using different time slots (time-division
multiplexing or TDM) and via different wavelengths (wavelength-
division multiplexing or WDM). One can then see that a wavelength
converter will be a key component for future optical systems.
Semiconductor-amplifier devices, combined with optical filters, offer
significant potential in making such a device. Again, because there are
wide ranges of frequencies and strong nonlinearities to be considered,
time-domain modelling makes good sense. Preliminary results are
shown here of simulating the performance of a cross-phase wavelength
converter operating at 5 Gbit/s. The Mach-Zehnder interferometer
contains an SOA in each arm and is injected with a CW optical input
at 1557 nm and data at 1558 nm using ideal optical transmitters, as
shown in Figure 8.7. There are four passive y-junctions as indicated
which are here assumed to be lossless. In order to make the conversion
process non-inverting a phase shift of IT radians is added in the upper
arm of the interferometer. This ensures that when injecting a T on
the data, which is designed to induce a IT phase shift in the lower SOA,
the interferometer becomes balanced which in turns results in
maximum output power at 1557 nm.
The optical data input at 1558 nm consists of a 4011011101001000'
sequence with 4.6 mW in the 'ones' and 0 mW in the 'zeros'. Figure 8.8
presents the simulated output waveform at 1557 nm which has the
inverted data superimposed by cross-gain saturation. The chirp
introduced by the lower SOA gives negative-wavelength chirp on the
leading edge of output pulses, which helps propagation in dispersive
fibre, with Figure 8.8d showing the predicted filtered received eye
264 Distributed feedback semiconductor lasers

ideal load

semiconductor
optical
amplifier output
ideal optical
transmitter waveform
semiconductor
CW optical waveform optical
at 1557nm amplifier
ideal optical
transmitter
Mach Zehnder with SOA
in each arm ! Modulated input optical
waveform at 1558 nm
Figure 8.7 Schematic diagram of cross-phase ivavelength-conversion system

power, mW
radians
amplified probe waveform wavelength offset 1557 nm

5 -

phase of amplified probe waveform

10- 0—
b

relative time, ns relative time, ns

power, mW converted output power, mW
0.04-,
received filtered eye diagram
0.03.

0.02-

0.01-

0.00
-0.1
relative time, ns relative time, ns

Figure 8.8 Mach-Zehnder wavelength converter

a Simulated output waveform from lower amplifier at 1557 nm modu-
lated by counter-propagating 1558 nm signal via cross-gain saturation
b Phase shift imposed on this signal by the amplifier saturation process
c The output from the wavelength converter
d Eye diagram after 70 km of standard 2fibre giving dispersion of 17 ps/
nm km and dispersion slope of 0 ps/nm km with a loss of 0.2 dB/km
 Future devices, modelling and systems analysis 265

diagram after propagation through 70 km of fibre with 17 ps/nm km

of dispersion.
These examples illustrate the power of a general optoelectronic
simulator which has inclusive software to enable it to be applied to
these very different applications. Such tools will mature over the next
decade to the state now achieved by CAD for VLSI.

8.3 The push-pull laser

8.3.1 Introduction: push-pull electronics

Push-pull operation is well known in electronics where, for example
(Figure 8.9), a pair of transistors T\ and T2 are coupled at their
emitters (marked E). Because of a sufficiently large resistance Rt ('the
long tail') there is a constant current /flowing equally between the two
transistors in the steady state. On driving the input (the base of Tx)
positive, the current in T2 is enhanced by 8/ but, because of the
constant-current source at the coupled emitters, the current into T2 is
reduced by 8/. The base at transistor T2 is shown to be fixed at 'ground'
potential but could equally well have been driven by a second input
provided that the polarity was opposite to that at the base input of T2.
The 'push-pull' outputs are then capable of driving another similar
circuit. This circuit is a remarkably high-speed circuit because both
transistors are operated with either unity voltage gain or unity current
gain which then gives the potential for working at the limit of their
gain-bandwidth product signified by the transition frequency ft. The
emitter E is the output for the first transistor (Tx) for driving transistor
T2. Consequently, the transistor Tj is effectively operated as an emitter
follower with unity voltage gain. The transistor T2 is effectively
operated in a common- (grounded-) base mode which then has unity
current gain. Although each transistor is operating at unity voltage or
current gain, the net power gain can still be in excess of 10 dB because
of the change from a low input impedance to a higher output
impedance. This emitter-coupled pair (or long-tail pair) provides the
classic circuit for achieving the highest modulation speeds in
transistors. It is typically used, with appropriate transistors, as an
output stage for electronic drivers for lasers even up to 20 or 40 Gbit/s.
With such drivers having two outputs Ox and O2 in antiphase, there is
a natural speculation as to whether two contacts on a single laser
operated in a push-pull mode could create benefits for high-bit-rate
laser sources.
266 Distributed feedback semiconductor lasers

-V
Figure 8.9 Emitter-coupled amplifier giving push-pull output
The output from T, which drives T2 is the emitter E. Consequently, T, is
driven as an emitter follower, while T2 is driven as a grounded-base
transistor. The (near) constant current I ensures that increment 81 in the
current flowing in T, is a decrement 81 in the current flowing in T2.
Hence outputs 0, and O2 are in antiphase or push-pull. The configura-
tion is also known as a 'long-tail pair'

8.3.2 Symmetrical push-pull DFB laser

The concept of push-pull laser operation is that modulation of the
optical output is achieved by shifting light to and fro from facet to facet
rather than increasing and reducing stored energy. Moving energy to
achieve modulation is well known in the mode-locked Fabry-Perot
laser where picosecond pulses of energy propagate up and down the
laser, appearing at the facets with repetition rates in excess of 100 GHz
[19,20] and with optical spectra close to the Fourier limits (so-called
bandwidth-limited spectra). Bistable DFB lasers also can similarly give
modulation at many tens of gigahertz by moving the energy back and
forth from one facet to the other facet [21]. Unfortunately, neither
mode locking nor bistability can give high-bit-rate pulses on demand
and this is the contribution of push-pull modulation of split-contact
lasers. It has been shown that the concept cannot work satisfactorily
with conventional Fabry-Perot lasers because localised gain does not
lead to localised optical intensity in these devices.* However, in a DFB
laser the distributed feedback tends to localise the optical intensity to
regions of optical gain and there is considerable potential [22,23], as
discussed briefly here.

* YU, S.F.: Private communication, 1990.

 Future devices, modelling and systems analysis 267

current current. current current

left mMMMMMMMMAMM right left \puuuu!jJU\J\A^ right

facet current drives into split contacts facet facet I C u r r e n t d r i v e s j n t 0 sp | it contacti facet

optical energy high X

t Optical energy low
output
output
gain + propagation gain + propagation
evanescence
frequencyj stop t frequency
band
'Bragg1 _^ i _ lasing_H _'Bragg'
frequency frequency
frequency
distance

Figure 8.10 Stopband and energy/distance diagrams for push-pull DFB operation
a Right output high; b Right output low
Consider a symmetrical split-contact DFB laser as in Figure 8.10. The
modulation of this laser is achieved by increasing IR and simultaneously
decreasing IL so that the stored energy moves from the left to the right.
The digital output changes from low (zero) to high (one). To change
back to the low state, one decreases IR and simultaneously increases IL.
From the symmetry, one can see that an output train of information Q
at the right-hand facet is matched by a complementary output Q at the
left-hand facet. Again, from the symmetry it follows that the steady
frequency in the 'zero' state must be identical to that in the 'one' state.
The dynamic chirp or change of frequency is limited to the transient
region where the energy actually moves, and it will be seen shortly that
there are ways of optimising this transient change.
To understand how the push-pull laser works without delving into
deep mathematics, use can be made of the stopband/distance
diagrams introduced in Section 5.6.2 and one can consider how these
diagrams change as the modulation changes. Figure 8.10 shows the
laser structure and the optical energy/distance and stopband/
distance diagrams schematically. When the greatest current is driven
into the right-hand contact as in Figure 8.10a, the gain is greatest in
the right-hand section and light generated in the middle of this section
propagates to the left and right but, because of the Bragg grating, the
light evanesces on the left leaving very little light emerging from the
left-hand facet. More light emerges from the nearer right-hand facet.
The laser grating is designed so that the laser naturally prefers to lase
in the — 1 mode (the lower-frequency mode in the stopband diagram).
268 Distributed feedback semiconductor lasers

This selection of the mode can, in principle, be accomplished by

having either an appropriate second-order grating [24] or an
appropriate gain grating [25]. The preference for the —1 mode is
further enhanced by the reflection from the left-hand section caused
by the Bragg grating within that section being operated within its
stopband (see the stopband diagram). The right-hand Bragg section
provides adequate gain and reflection combined with phase shifting
caused by the propagation to give a unity round-trip (complex) gain.
The spatial-hole burning in the carrier density created by the optical
energy within this half section will also aid the modal selection of the
- 1 lasing mode given, say, that K ( L / 2 ) ~2.
Now switch the current in an appropriate mirror image as in Figure
8.10 and the steady-state condition is the same mirror image (Figure
8.10b), of the situation in (a). The right-hand Bragg section how ceases
to provide enough gain for sustained lasing within that section. The
stopband diagram readjusts to the changes of refractive index as the
electron density is reduced. There is then a reflection from this right-
hand Bragg section as there was previously from the left-hand section
and this change of reflection actually helps the optical stored energy to
move from the right-hand over to the left-hand section. The fact that
the two states are mirror images must mean that they operate at the
same lasing frequency. The chirp between the on and off states which
is so evident in a conventional laser should therefore be significantly
reduced except when there are switching transients.
Both modelling and experimental work indicate that it helps to have
three sections rather than two, with a steady-current drive in the
central section providing a source of stored photonic energy [26]
which helps to reduce the chirp further. Typically, with such push—pull
laser designs, one will find that the total length is over twice that of a
conventional laser so that the end-modulated sections (of length Ls
and operated in push-pull) each have a value of KLS that is similar to
the value of KL for the conventionally operated laser of length L.
Figure 8.11 (after Flanigan [27]) shows the qualitative characteristic
differences that are measured between a conventionally operated DFB
laser and an unoptimised symmetrical push-pull-operated DFB laser
highlighted by the amplitude-modulated waveforms and the localised
frequency as a function of time. This latter is measured by an
electronically tunable Fabry-Perot etalon which can be tuned to show
the frequency for maximum energy through this etalon at any
particular time during the output pulse [28]. The results for the
conventionally modulated laser demonstrate the close link between
 Future devices, modelling and systems analysis 269

k ,d A/
AJUUUUUUU1 output lUUUULAJUUU^^ output
current drive into facet facet
current drives into split contacts
single contact

2.6 |-
relative
relative wavelength shift,
wavelength shift, Angstroms
Angstroms

0-5 time (ns) 1 1.5 0 0.5 1 1.5

conventional modulation push-pull modulation

Figure 8.11 Output intensity and time-resolved chirp from commercial single-contact
MQWDFB laser contrasted with unoptimised 3-contact push-pull bulk
DFB laser
Pattern is 010 at 2.5 Gbit/s. The total bias level for single contact is
60 mA compared with 90 mA for push-pull and the modulation depth
is 2 VH, (data after Flanigan [27])

the amplitude modulation (including the photon-electron reso-

nance) and the dynamic chirp. The push-pull laser demonstrates that
the photon-electron resonance has markedly changed the chirp with
it now most significant only during the transients between the ones
and zeros where it has the same direction at both turn-on and turn-off,
in contrast to the chirp arising from conventional modulation.

833 Asymmetry and the push-pull DFB laser

The switching transients in the frequency output of the laser can be
changed by changing the details of the drive to the push-pull laser.
There are at least four ways of modifying the drive to a push-pull laser
so as to influence the dynamic chirp and improve/change the laser's
transmission performance over long distances. Figure 8.12 shows these
schematically:
(i) modulation-depth asymmetry,
(ii) single-ended modulation which is a limiting case of modulation-
depth asymmetry where only one contact (typically at the
emitting end) is modulated,
2 70 Distributed feedback semiconductor lasers

IL Id.c. IR

luUUlAAJUUlAiULaAAAJUUUUUUUI

iT Ax

>*

Figure 8.12 Schematic diagram illustrating various asymmetrical modulation

schemes (afterFlanigan [27])
a A/, , negative-modulation-depth asymmetry b Single-ended modula-
tion 7 A/,,,,,,, positive-DC-bias-level asymmetry d AT positive temporal
asymmetry

(iii) DC bias level asymmetry, and

(iv) temporal asymmetry.
One could also consider building spatial asymmetry into the laser
but this has not been investigated extensively.
The results on the time-resolved chirp for two such asymmetries are
shown in Figure 8.13 which shows the effects of asymmetry in the
modulation-current amplitudes. Note that a flat time-resolved chirp
implies that the fundamental spectral spread in frequencies caused by
Fourier's theorem is the only spread in frequencies that remains. The
time-resolved chirp measures the excess spread in the frequency.
Experimental work has confirmed that the push-pull configuration
for operating DFB lasers can achieve useful results. Pakulski has
achieved a 0.8 dB penalty for symmetrical push-pull operating at
2.5 Gbit/s over a 370 km link using quantum-well devices [29]. When
single-ended modulation was applied, the penalty could be reduced to
0.3 dB, indicating that asymmetry may help in reducing dispersion
penalty. However, temporal asymmetry offers new possibilities and has
yet to be tried at the time of writing. This is discussed briefly next.
Figure 8.14 considers temporal offset in an ideal symmetrical push-
pull laser. The frequency changes at the leading and trailing edges of
 Future devices, modelling and systems analysis 271

10.0'
+20mA
8.0 0mA
(symmetrical)
-10mA
6.0
-20mA
-30mA
4.0 -40mA
(single ended)
in-phase
2.0
waveleijgth shift,
Angst oms
0
time, ns 1.0

Figure 8.13 Effect of modulation-depth asymmetry on the experimentally meas-

ured time-resolved chirp for a 010 1.5 Gbit/s pattern, using a push-pull
modulated three-contact bulk-material device
The fact that a slight negative asymmetry drive appears to give the 'best'
results may indicate constructional asymmetries within the laser. The
asymmetry is as defined in Figure 8.12 (after Flanigan [27])

the pulse have great significance when it comes to propagation along

dispersive fibre. Notice how, in Figure 8.14, the pulses with +60 ps and
— 60 ps have the opposite changes in the wavelength shift at the
leading and trailing pulse edges. The significance of these shifts is
shown well in modelled 'eye' diagrams ( Figure 8.15) for propagation
of pseudorandom pulse trains over standard fibre. With positive
temporal asymmetry, the frequency shifts in the leading- and trailing-
edge frequencies partly compensate for the fibre's dispersion by
tending to control the speed of the pulse edges of the ones, such that
these edges do not move into the zeros. This keeps the ones well
defined and the eye clearly open up to 400 km, and even reasonably
open at 800 km with this 2.5 Gbit/s modulation, indicating that this
form of push-pull modulation might compete with similar compensa-
tion schemes using external modulators [14]. With a negative
temporal asymmetry, these frequency shifts at the trailing and leading
edges of the pulse, combined with the fibre's dispersion, tend to move
these edges more rapidly into the zeros and so close the eye more
rapidly. Then at 400 km the eye is virtually closed. Zero temporal
asymmetry is not shown, but is somewhere in between these two
conditions.
272 Distributed feedback semiconductor lasers

4.8

+60 ps
4.0
+40 ps
32
wavelength +20 ps
shift,
Angstroms 2.4 Ops
(symmetric)
-20 ps
1.6
-40 ps
0.8
-60 ps
0.
time, ns 1
Figure 8:14 Modelled effect of temporal asymmetry on the time-resolved chirp for a
three-contact MQW device push-pull modulated at 1.5 Gbit/s with a
010 pattern.
The output was taken from the right facet, and a positive temporal
asymmetry means that the modulation switch on the left contact leads
the switch on the right contact. The DC bias levels on the end contacts
were 22.5 mA, the centre contact bias was 50 mA. The modulation
depth on the end contacts was 40 mA.The asymmetry is as defined in
Figure 8.12 (after Flanigan [27]).

83 A Speed of response for a push-pull DFB laser

Multigigabyte-per-second operation is required of directly modulated
lasers for modern communication systems. While poor packaging can
always limit the speed of response (Appendix 10), the internal physics
of the laser creates the fundamental limits, for example the photon-
electron resonance in a conventional laser even with negligible
package parasitics. With the push-pull laser it is found that this
resonance is not quite so important as the spectral separation between
the — 1 mode with its symmetrical fields and the — 2 mode with its
asymmetrical fields. Consider the split-contact laser just at the point
where switching occurs with the drive approximately uniform. The field
patterns within the laser have to be formed from a superposition of the
modal patterns of fields created by modes within the laser (Figure
8.16). The — 1 mode has a symmetrical envelope with distance along
the laser and the frequency of the fields is f_v The — 2 mode is
asymmetrical and, relative to the — 1 mode, beats at a frequency
/_!—/_ 2 , and so the net field pattern of these two modes will have
 Future devices, modelling and systems analysis 273

400km

0 0.2 0.4 0.6 0.8 1 1.2

a b
Figure 8.15 Effect of temporal asymmetry on simulated eye diagrams for a three-
contact push-pull-modulated MQW laser
a + 60 ps modulation offset
b - 60 ps modulation offset (as defined in Figure 8.12)
The bit rate is 2.5 Gbit/s. The steady bias is 22.5 mA on the end contacts
and 50 mA on the centre contact. The modulation-depth current is
±40 mA on each end contact (after Flanigan [27]).

motion

distance Asym. Asym. distance

Figure 8.16 Beating of two modal patterns

Symmetrical field envelope S contains fields at a frequency /_, and
asymmetrical field envelope A containsfieldsat a frequency/_ 2. The net
field pattern JVthen beats back and forth at a frequency/_, —/_2

energy concentrated either nearer the right-hand facet or nearer to

the left-hand facet. The speed at which this energy can move from one
facet to the other is therefore limited by l / ( / _ i - / - 2 ) > the reciprocal
of this beat frequency. This does not mean that one excites the - 2
mode into lasing, because that mode is always well below cutoff, but of
course it means that, if one attempts to modulate the laser at
274 Distributed feedback semiconductor lasers

fM> (/_! —f-9), the spread of frequencies extends to the - 2 mode.

Detailed numerical modelling confirms this result [30]. Although the
modelling indicates the potential for modulation of these types of laser
at 100 Gbit/s given optical material with sufficient gain, it is worth
pointing out that, with standard fibre, dispersion rapidly closes the eye
after a few kilometres even at 50 Gbit/s.
In conclusion, the push-pull laser is a device which requires
extensive modelling to design and optimise. At the time of writing it is
not clear that it will compete with the external modulator although it
addresses the same issues of giving high-bit-rate modulation while
helping to shape the spectra to compensate for the fibre's dispersion.
A significant drawback comes when there is a long string of ones (or
zeros) which gives a consequential change of the temperature between
the two ends of the laser and so causes a small but troublesome
temperature-induced change of frequency.

8.4 Tunable lasers with distributed feedback

8.4.1 Introduction
The emphasis in this book has been on single mode laser emitters with
narrow and stable linewidths. However, future high-capacity optical
networks need to employ tunable sources and detectors to increase
capacity and flexibility [31], and for this purpose trial systems have
been built with 80 or more channels in a 30 nm window centred
around 1515 nm [32] using erbium-doped fibre amplifiers. Tunable
lasers that are capable of wide tuning ranges are complex, expensive
and have variable tuning ranges because of sensitivity to manufactur-
ing tolerances. Sensing and measuring instruments can already accept
the expense of currently available devices, so that high-volume
communication applications should promote the improvement and
cost reduction of tunable lasers.
In Chapter 2 it was pointed out that the gain spectra typically
covered a frequency range A/~kT/h, giving wavelength ranges to half-
maximum gain of AA~A2kT/hc, or in other words, to a first order,
A A-50 nm centred on 1500 nm, reducing to AA~15 nm centred at
850 nm. These ranges can be extended to ~80 nm around 1550 nm
central wavelength by pumping harder to increase the width of the
gain spectrum and making use of low-loss material with adequate
feedback. Figure 8.17 indicates schematically how quantum-well
material, at high drive, can significantly increase the bandwidth for
 Future devices, modelling and systems analysis 275

effective
gain (cm" ) | Af~k7?h
100

bulk

-100 in
1/1650 1/1600 1/1550
frequency (in reciprocal nm)

Figure 8.17 Schematic spectral widths of quantum-well and bulk materials

Greater spectral broadening with increasing drive is possible with
quantum-well material than with bulk material because the reduced
density of states for the lowest energy transitions forces the higher
energy levels to be used—see Figure 8.18

conduction band

lowest A 2nd lowest 2nd

transition transition transition transition
energy energy energy
energy^
valence band '
narrow quantum well wider quantum well
Figure 8.18 Gain-bandwidth broadening
The narrower the well, the greater the spectral separation between the
first transition and the second transition. Narrow quantum wells favour
a wide gain spectrum at sufficiently high drive levels. Careful design is
required, though, in relation to saturation of the levels leading to gain
saturation

gain compared with that of conventional bulk material. At low gain

levels when only the lowest level is filled, the range of carrier energies
is lower than in bulk material; but if the material is pumped strongly,
the next quantised energy level begins to be filled and the energy
range can be greater than in bulk material. Tuning ranges of over
110 nm have been reported using such quantum-well material [33].
276 Distributed feedback semiconductor lasers

Two points should be made about designing quantum-well struc-

tures for a large gain bandwidth. First, as shown schematically in Figure
8.18, deep narrow wells are more desirable than wide shallow ones
because of the wider energy (frequency) separation between the first
and subsequent transition energies. Secondly, gain saturation occurs at
the wavelength corresponding to the first transition energy because
filling of these quantum-well states is an essential mechanism of gain
broadening. The design of lasers using this effect is therefore critical,
with relatively restricted ranges of operation with respect to both
temperature and output power.
Obtaining a material with a wide gain spectrum centred on the
desired lasing frequency is only part of the problem of obtaining a
tunable laser. The round-trip complex gain has to be unity at the lasing
frequency; i.e. amplitude and phase from the feedback filter have to be
matched independently at the desired lasing frequency, but not at any
other frequency so as to avoid spurious or multiwavelength operation.
An external grating (Section 1.4.2) has been one extremely successful
method of achieving wideband tunability [34] and here the frequency
of peak amplitude response is determined by rotating the grating, but
to achieve a continuous tuning, a phase adjustment is also provided by
means of slight changes to the grating-laser separation. However, the
emphasis in this section is not on mechanically tunable or switched
lasers but on monolithic electronically tuned DFB lasers where the
principal methods of tuning are from changing the refractive index
by:
(i) temperature (using, for example, electrical heating);
(ii) carrier density ( via the plasma effect, see Appendix 7); and
(iii) electric field (via the Franz-Keldysh [35] or quantum-confined
Stark effects [36,37]).
Temperature tuning [38,39] can, of course, be applied to any laser.
The temperature coefficient of wavelength for InP-based single
frequency lasers at 1300 nm and 1500 nm is about 0.09 nm/degC, and,
given a temperature range of operation AT of at least 40 degC (e.g.
+ 10°C to +50°C), a tuning range of 3 nm or more is easily available.
Appropriately designed DFB lasers are capable of operating at from
- 40 to +80°C, which leads to a tuning range of around 10 nm for a
nominal 1500 nm laser. This approach is simple, predictable and
reliable, but requires a Peltier heater/cooler to control the tem-
perature, and the stabilisation to a new temperature then takes tens of
milliseconds or more. At high temperatures the optical gain for a given
carrier density is decreased, so the threshold current increases and, in
 Future devices, modelling and systems analysis 277

-i
/ WWWWV\AAA\A/WVW
s vwwwvw
i VWSAMAA/V
1 2 3

2 contact DFB 2 contacts: gain +DFB reflector 3 contacts: gain + phase +DFB

Figure 8.19 Some developments of multicontact DFB lasers

The narrow guide indicates material operating at the lasing wavelength
while the wider guide indicates a guide made from material with a wider
bandgap so as to avoid absorption losses

general, the power output and modulation efficiency decrease. At low

temperatures the gain spectrum narrows (kT is decreased) and it
becomes difficult to keep the correct mode, determined by the Bragg
grating, aligned in frequency with the gain peak. A mode jump or
multimode operation may well occur at some temperatures .
If the mean carrier density in a laser is increased, the mean
refractive index decreases through the plasma effect and shortens the
wavelength of the lasing modes; however, in most lasers, it is observed
that the carrier densities are approximately clamped around some
threshold value, and spatial variations generated by multiple contacts
have to be employed to achieve significant tuning by this means: about
1% variation is possible. Smaller ranges than often expected are found
in practice because the current, injected to increase the carrier
density, will heat the laser and hence lengthen the wavelength, as noted
in Section 2.6. Because one effect is electronic and the other thermal,
they have very different associated time constants; this is observed in
the FM response of DFB lasers [40] which changes sign at some
modulation frequency in the megahertz range because at higher
frequencies the thermal effects are too slow to follow the modulation.

8.4.2 Simple multicontact tunable lasers

When DFB lasers were introduced in the 1980s, it was immediately
noted that a measure of electrically controlled tuning was desirable,
and several laboratories subsequently demonstrated some tuning by
simply taking a DFB laser and splitting its top contact into two or three
sections. Since then a host of variations on this theme has been
reported (e.g. [41-43]) and a brief summary is given below.
Figure 8.19 shows a series of increasingly complex tunable lasers
incorporating a Bragg grating. The simplest design, laser 1, with only
two contacts and a uniform grating, is not the easiest to understand.
278 Distributed feedback semiconductor lasers

When contacts A and B are driven identically, the device behaves like
a conventional DFB laser. One way of making a tunable laser drives one
contact (say A) alone until the laser reaches threshold and then, on
increasing the current through contact B, tuning occurs. The
important feature is that the asymmetry along the laser's length in the
drive leads to an associated asymmetry in the effective linewidth-
broadening factor aH and also asymmetry in the gain saturation
enabling the gain and phase tuning to be partly decoupled to different
sections of the laser. In this device it is helpful to use quantum-well
active regions so that, on the side which is close to gain saturation, the
aH factor is increased, thereby increasing the change of frequency for
a given change in the carrier density. At 1500 nm, reductions in
wavelength of 2-3 nm are possible by increasing the current in
contact B before heating causes the refractive index, and thus the
wavelength, to increase again. The device is easy to make but the
tuning range is limited if relatively predictable.
Laser 2 may be regarded as a Fabry-Perot cavity (contact C) with
feedback determined by one single-facet mirror, and one reflecting
stack or Bragg grating (contact D) which is primary in determining the
FP mode that is selected. The net round-trip gain from the reflector
and the FP section must be appropriate but the important thing is that,
at approximately the lasing wavelength, increasing current into D directly
reduces the refractive index of the waveguide via the plasma effect,
and hence reduces the wavelength of the peak reflection from the
Bragg grating, thereby tuning the laser. To avoid the difficulties of
interaction between gain and frequency in the Bragg section, this
section can be fabricated in a wide-bandgap waveguide which neither
absorbs nor gives gain to the lasing wavelength. Any spontaneous
emission from this guide lies outside the frequencies of operational
significance. In general with only two contacts, sufficient change in
wavelength will lead to a jump to an adjacent Fabry-Perot mode
('mode hopping') so as to maintain the required round-trip phase
matching.
This drawback of mode hopping is avoided in laser 3, where the
Fabry-Perot section of laser 2 is now divided into two distinct regions.
The region with contact F is a section of plain waveguide made from
material with a bandgap which is too large to absorb or give gain at the
lasing wavelength but can have its refractive index changed through
the plasma effect and so provide the correct round-trip phase as the
lasing frequency changes without creating a mode hop. The region
with contact E provides the required round-trip gain (with some effect
 Future devices, modelling and systems analysis 279

on the phase) as the Bragg reflector is tuned via the plasma effect as
in laser 2.
Variations can be created, for example, by having the region, with
contact F, formed from an intrinsic layer sandwiched within a P-N
junction (forming a P-I-A/junction). Then with a variable reverse bias
with strong enough electric fields (>10V/|mm) the conduction and
valence-band edges within the I-region shift so as to increase the
bandgap energy. This is known as the Franz-Keldysh effect [35] in
bulk materials, or the quantum-confined Stark effect [36,37] in
quantum-well materials, and is primarily employed in electroabsorp-
tion modulators for high-speed low-chirp applications. The change in
absorption also implies (from the Kramers-Kronig relationships) that
there are changes in the refractive index which may be employed for
tuning a laser. However, with those structures and voltages which are
compatible with laser operation, the total refractive-index change
available is typically less than 0.1% with a fractional tuning range of the
same order. Four-section devices exist, similar to laser 3 but with
tunable Bragg reflectors at both ends and phase- and gain-compensat-
ing regions.

8.43 Wide-tuning-range lasers with nonuniform gratings

Wider tuning ranges than those of Section 8.4.2 may be obtained in a
relatively complex laser by use of interrupted [44,45] or super-
structure [46-48] gratings. It is possible to show that, if one inserts
periodic plain sections into an otherwise uniform grating then one
can generate a comb of narrow reflections, with each reflection 'line
width' A\ r comparable with those from a single grating &kslof)9 but with
the comb covering a much wider range of frequencies A\/Mm, (Figure
8.20). Spatial Fourier analysis of short grating 'patches' of length Lg
gives the principal comb-envelope width AKlune— Ao/ N^= \l / 2Lg where
Ng is the number of grating lines in each patch. The comb spacing is
determined from AAn/?/,— A]/2Lr where Lr is the periodicity of the
interrupted grating structure. These approximate relationships may
be derived from similar arguments of wave interference which were
used for determining the fundamental reflection from a Bragg grating
(Section 1.6). In order to obtain a wide comb with many lines,
Lg/Lr<\> but this fraction is also the factor by which the peak
reflection per unit length (or effective-coupling coefficient) of the
grating is reduced from its value when the grating is uniform.
Consequently, long interrupted gratings are required to get ade-
quately high reflectivity for laser operation but, as a recompense for
280 Distributed feedback semiconductor lasers

comb spacing AA, cmb

Figure 8.20 Reflections from modified gratings

the long length and low effective-coupling coefficient, narrow line-

widths are achieved which can be well under 1 MHz [49].
Instead of the periodic interruption of the grating it is also possible
to generate reflection combs from continuous superstructure gratings,
with a periodically varying pitch. The interrupted gratings may be
regarded as having a spatial periodic amplitude modulation of the
reflections per unit length while a varying pitch is equivalent to a
spatial periodic phase modulation of the reflections per unit length.
Superstructures therefore have the advantage that the coupling
coefficient remains higher than for interrupted gratings and conse-
quently shorter reflectors are possible for the required feedback,
thereby leading to less scattering loss and less free-carrier absorption
in the waveguide when carriers are injected for tuning. The dis-
advantage is that superstructures are more difficult to fabricate than
interrupted gratings.
These structures are used with vernier tuning, as illustrated in Figures
8.21 and 8.22. Here the device is a DBR type of laser with interrupted
gratings with different pitches at either end. The central FP section of
this device provides the gain. Adequate feedback will occur only when
a line in both reflection combs aligns, as shown at Ao (Figure 8.21).
Shifting the refractive index of grating 1 slightly will shift its whole re-
flection comb, and eventually lines at A+1 will line up (or A^) in the
other direction. A small shift for the whole comb, created by the plas-
ma effect and carrier injection, results in a larger shift for the
lasing wavelength, with 'shift-multiplication ratio' « AAl/(AAl — AA2).
This multiplication ratio cannot be increased indefinitely in order to
get larger tuning ranges, because of the finite comb linewidths AAr.
 Future devices, modelling and systems analysis 281

schematic of a DBR with a pair of different interrupted gratings

spatial domain corresponding wavelength domain

L.

mm*
-1 AQ ^+1

Figure 8.21 Pairs of periodic gratings and their reflection combs

tuning TV

anti-reflection coatings

Figure 8.22 Four-section vernier-tuned laser

(i) Interrupted grating mirror A
(ii) Gain section
(iii) Phase-adjusting section
(iv) Interrupted grating mirror B

If AA{ and AA2 are too similar, there is insufficient discrimination

between the adjacent comb lines and the laser becomes multimoded
or has an inadequate sidemode-suppression ratio, especially near the
extremes of the tuning range.
Figure 8.22 shows schematically a laser structure incorporating a
pair of interrupted gratings. For the same reasons as in laser 3 (Figure
8.19) a phase-adjusting section has been added to the gain section to
ensure phase matching at the desired wavelength and to reduce mode
hopping. As with other tunable lasers, there are several currents that
must be precisely controlled over the lifetime of the laser.
282 Distributed feedback semiconductor lasers

. Contact A y Contact B

IIP v///)>/////////,

contact C y
modes coincide

T M11 in 11 I I I combA + B
^ I
mode I I I I l l lI I I I I I combA + C
amplitude
frequency

Figure 8.23 An asymmetric Y laser (schematic)

The two sets of Fabry-Perot modes formed by the two branches of arm
A have slightly different frequency spacings for the modes. Only if both
modes align in frequency will lasing occur, and then a small change in
the refractive index in one arm permits a significant shift in frequency
through the vernier effect [50,51]

8.4.4 Other tunable-laser structures

Another laser employing the vernier tuning action of Section 8.4.3 is
the 'Y laser [50,51]. These lasers contain a Y-branched waveguide so
that two coupled sets of Fabry-Perot modes exist, and strong lasing
occurs where the modes coincide, as indicated in Figure 8.23. They
were introduced originally as single-frequency lasers for similar
applications to DFB lasers; their construction is relatively easy using
well tested technology of Fabry-Perot lasers. When properly adjusted,
their performance is good, but multimode operation can occur at
drive currents which are not exactly the correct design values, which in
any event vary with temperature. However, in principle they have given
the order of 50 nm tuning range by vernier action obtained by
adjusting the relative currents into contacts A, B and C in Figure 8.23.
The simplest concept of operation occurs when the waveguides under
contacts B and C are transparent, so that their refractive indices may
be changed by carrier injection without much gain change; however,
operation is still possible even with gain in all the waveguides. Using
the same arguments as in Section. 8.4.3, the combs shown in Figure
8.23 may have too little separation of frequency to allow sufficient
frequency discrimination, and multimode operation would probably
occur.
 Future devices, modelling and systems analysis 283

contact A
contact B

active I;

passive tuning
guide layer

lasing
filament
Contact C

Figure 8.24 End-view schematic of twin-guide laser

With only two branches, modal discrimination can be a problem but

excellent tuning has been demonstrated by extending the principle of
operation by adding one (or more ) further Y^unctions [52], so that at
least three sets of modes have to coincide for lasing to occur, but this
requires temperature control and four precisely controlled drive
currents!
A related concept is the grating-assisted vertical-coupler filter
(GAVCF) laser which gives a wide tuning range in principle by
adjusting only one current [53,54]. Here a pair of waveguides with
slightly different effective indices is 'phase matched' at one wavelength
by a coarse grating (e.g. pitch 20 |mm), so that power couples from one
guide to the other. Small changes in the index of one guide cause
much larger changes in the phase-matching wavelength. Unfortu-
nately, such GAVCF devices have an insufficiently narrow passband to
ensure single-mode operation so that a further auxiliary and adjust-
able filter must be added, again adding complexity to both
manufacture and control.
A structure which gives a smaller tuning range, but is relatively easy
to control, is the tunable twin-guide structure shown in Figure 8.24.
Tuning ranges of over 10 nm in the 1500 nm band are possible [55],
which is adequate for some applications. Referring to Figure 8.24,
lasing gain is provided in the upper waveguide shown by injecting
current from contact A to B, whereas separately the lower waveguide has
its refractive index changed by carrier injection using a material
composition so as to ensure transparency. Upper and lower guides are
closely coupled so that the pair act as a single composite guide with
adjustable index; problems with injection current and heating still
exist, but the structure is well suited to optoelectronic integration
[56].
284 Distributed feedback semiconductor lasers

In summary, while there are many candidates for the role of the
tunable semiconductor laser, the really broadband devices are all
difficult to characterise and require complex arrangements for the
control of several drive currents. These features combined with the
thermal limitations are a recurring problem of such tunable lasers and
become particularly acute if the laser degrades with time. The control
problem will need to be solved before such lasers become suitable for
mass applications.

8.4.5 Tunable-laser linewidth and modulation

Tunable DFB lasers tend to use long cavities and low effective-coupling
coefficients, which can result in very satisfactory linewidths of the
order of 1 MHz for nominal 1500 nm lasers [57,58]. In cases where
tuning and phase controls are not optimally adjusted or the vernier
methods give inadequate gain and phase margins, the sidemode-
suppression ratios are inadequate and multimode or noisy operation
may occur. For critical systems using tunable lasers, it is essential to
monitor the laser output for wavelength, spectral purity and sidemode
suppression.
Direct frequency modulation, with moderate deviations, at constant
amplitude is relatively straightforward for most monolithic tunable
lasers by modulating the current (~ 1 GHz/mA is achievable) but
direct amplitude modulation is more difficult. This is because systems
require an amplitude-modulation depth of 10:1 or greater but, to
achieve this modulation depth, large and rapid changes in the gain are
required which also change the phase characteristics of any gain
section. Changes of frequency even as low as 1 part in 104 give
unacceptable dispersion penalties in the transmission of gigabyte-per-
second information over 100 km fibre systems. For exacting
applications, the answer is to use an external modulator along with an
optical isolator to prevent changing feedback into the laser. For less
critical applications, it may be that some hybrid of a two-section
tunable DFB laser and a two-section push-pull laser could achieve an
adequate result.

8.4.6 Modelling tunable semiconductor lasers

In principle, the time-domain model described in detail in Chapter 7
will work as well for tunable semiconductor lasers as for the fixed-
wavelength grating-based lasers on which it was developed. However,
there are several extra problems:
 Future devices, modelling and systems analysis 285

(i) Lasers with interrupted gratings need many sections to character-

ise them fully, so that well over 100 sections are likely to be
required (at least one section for each 'plain' section, and one
section for each grating section throughout the reflectors).
Probably the easiest way to deal with this is to take the section
lengths in the reflectors to the nearest integral number of
distance/time steps, and then add a phase-shifting function to
every section. This will result in a relatively complex program
which will be outside the array-handling capabilities of the
Student edition of MATLAB used in this book.
(ii) Thermal effects are frequently important in tunable lasers. The
time-domain model is not at present suited to simulation of
thermal transients lasting milliseconds or more, since with the
number of sections and time steps the computer running time
could become excessive. In some cases, this can be circumvented
by precalculating the thermally dependent parameters with other
modelling techniques, and then inserting these parameters into
the normal time-domain model at accelerated time intervals, but
care is needed when using this technique.

If only static tunable-laser performance is required, the character-

istics of complex-interrupted-grating reflectors can be calculated by
any of the usual methods employed for multilayer dielectric filters.
Laser performance can then be calculated using a transfer-matrix or
power-matrix model.

8.4.7 Multiple DFB lasers with optical couplers for WDM

The strength of the tunable laser lies in being able to tune to many
wavelengths, but this also leads to weaknesses for some applications:
the careful balance of control currents required to set the appropriate
wavelengths can be upset by aging or temperature changes; the low
threshold discrimination that is often necessary to obtain the widest
tuning ranges means that dynamic increases in gain can excite
unwanted modes; and feedback from external sources with spurious
wavelengths may cause drastic changes of output by exciting near-
threshold modes. An alternative for WDM systems is to have an array
of fixed- but different-wavelength lasers which are switched in or out as
required.
Electron-beam lithography, for example, may be used to write a set
of gratings side-by-side but each with slightly different pitches, forming
an array (-eight devices) of edge-emitting lasers on a single chip with
286 Distributed feedback semiconductor lasers

predetermined wavelength spacing (~ 1-2 nm) [59-61]. The outputs

from the different spatial positions are then combined externally, or
integrated on the chip. For applications where eight or so wavelengths
are required, such a device is attractive, because wavelength control is
straightforward, spurious outputs are unlikely, and it is also possible to
drive and modulate two or more lasers at different wavelengths
simultaneously and independently for WDM applications.
The straightforward and popular approach is to combine the
wavelengths in a 1:N star coupler or some integrated equivalent but
such a device without frequency selectivity necessarily has a 1 / N power
loss which has to be made up by amplification either before or after
the coupler. Amplification before the coupler may be preferred
because, unless the amplification is strictly linear after the coupler, the
nonlinearity or saturation can cause too much crosstalk between the
channels. Frequency-selective couplers need not have the same loss as
star couplers but are challenging to make typically using integrated
two-dimensional optics to permit precise dimensioning of guides,
couplers and gratings and with present technology such integrated
eight-way couplers can still have around 7 dB of loss [62]. There is
room for novel work in this area [63, 64], but it will be difficult to beat
the straightforward methods.
Optical interference and crosstalk between such lasers need not be
a problem, even for less than 2 nm wavelength spacing [65], while
thermal crosstalk [66] between adjacent lasers will probably be
measurable, but not serious in most applications. For applications
where several tens or even hundreds of wavelengths are required, the
fully tunable laser remains the more attractive option [67] and the
compact nature of such semiconductor lasers contrasts well with other
types of tunable laser [68].

8.5 Surface-emitting lasers

8.5.1 Introduction to surface-emitting lasers
The DFB structures so far discussed emit from their ends (edge
emitters) rather than from their top or bottom surfaces, but there is a
steadily increasing motivation to design lasers which emit from the
chip's surface [69-71] for the following reasons:
(a) Distributing the optical output over a larger area of surface can
reduce the peak power density at the emission facets and so
facilitate yet higher-power lasers without the present difficulties of
 Future devices, modelling and systems analysis 287

facet damage. With appropriate designs, one can also envisage

improved heat sinking for high-power lasers leading to yet higher
power outputs.
(b) The divergence of a laser beam is inversely proportional to the
emission area (solid angle of emission ~ 11 = 477 Al/s&) so that an
increased emission area ,s4 can give narrow beams which are useful
for applications such as free-space optical communication. Appro-
priate circular far-field patterns from surface emitters, as
compared with asymmetric patterns from edge emitters, should
also increase the coupling efficiency to circular optical fibres.
(c) Surface emitters can readily be arranged into two-dimensional
(NXMj arrays (Figure 1.23) which could be useful for optical data
switching and processing, whereas edge emitters can easily form
only linear (1 X N) arrays.

Figure 8.25 shows two main different approaches to forming surface

emitters. The first approach (Figure 8.25a and b) uses one form or
another of edge emitter to feed either a second-order grating
(Sections 1.6.1 and 4.4.3) or a 45° mirror so as to give surface emission
[72-76]. High-power lasers have successfully been made by these
techniques [77] and the angle of emission can also be steered over a
limited range [78]. The second and very different approach is the
vertical-cavity surface-emitting laser (VCSEL) [79,80,81], where the
lasing filament is perpendicular to both the chip surface and the
epitaxially grown layers of the structure.
VCSELs have become an exceedingly important new development,
well deserving a research monograph [82] which contains the
appropriate historical perspective and far more technical detail than
can be accomplished here. The work here is limited to introductory
material highlighting some of the significant challenges for the future
to make VCSELs fulfil their potential.

8.5,2 Operating parameters of VCSELs compared with edge

emitters
It is instructive to begin by comparing a conventional edge-emitting
Fabry-Perot laser (having a cavity length L^~240|uim and facet
power reflectivity i?^~30%) with a VCSEL (cavity length Lvrsel~\ |xm
and facet power reflectivities i^sr/, to be determined) as shown in
Figure 8.26.
Assume to start with that the net material field gains per unit length
(after scattering losses are subtracted) are gedge and gvcsd for the
288 Distributed feedback semiconductor lasers

o
optical output optical output

\ \ gain section
2nd order grating reflector & output coupler

a
optical output
optical output
optional anti-reflector / \
gain

\ 45 deg. mirror bragg reflectors \

gain section (optional grating) (dielectric stack)

Figure 8.25 Grating-coupled DBR, mirror-coupled DFB and VCSEL

conventional laser and VCSEL, respectively, with confinement factors

of r ^ ~ 0 . 5 and TvrsH~ 1. The lasing power-balance equations are
dge exp (4 r ^ gedge L, % ) = 1 exp (4 Twsd gvcsd LV(.J = 1 (8.1)
or, on rearranging to give the power gain explicitly,
1
2 g^= ( 1 / T ^ Ledge) In (l/Redge) - 100 cm "
2 gtmd= 10 000 In ( 1 / i U cm"1 (8.2)
If VCSELs are to operate at high temperatures or need the long
operating lives of conventional lasers, the gain for lasing and the
resulting carrier density in the active region must not be significantly
higher than in conventional edge-emitting lasers. For comparable
power gains around 100 cm"1 (i.e. field gains of 50 cm" 1 ), it can be
seen that the vertical-cavity surface emitter requires both top and
bottom power reflectivities to be greater than 99%. However, these
values cannot readily be achieved by metal reflectors or single
dielectric interfaces but require Bragg reflectors formed from multiple
A/4 layers of high- and low-index semiconductor or vacuum-deposited
dielectric stacks whose design and fabrication needs discussion.
Given a value of 99% for reflectivity of a Bragg grating, Figure 8.27
gives a rough estimate of the minimum number of quarter-wavelength
 Future devices, modelling and systems analysis 289

Length L e d g e Length LVCsei

Reflectivity
lasing filament I ^vcsel

Reflectivity lasing filament

/?edge
Edge emitter VCSEL
Figure 8.26 Schematic edge emitter and VCSEL

0.2

0.15
99% reflectivity
8n/Sn
0.1

0.05

0 10 20 30 40 50 60 70 80
approximate number of layer pairs

Figure 8.27 Approximate number of layer pairs

Required to achievel 99% power reflectivity as a function of
8n/Xn = (n] —n2)/{nl- r7h2). The analysis here does not account for
significantly different refractive index of any terminating layer, nor for
optical loss

layer pairs that will be required to achieve this reflectivity, assuming

that there is a change in refractive index between each layer in any pair
of Sn=(nl — n2) with (n1 + n2)=y£n. The calculation (see eqns.
5.35-5.37) neglects any differences in the refractive indices of the
incident and final layers and assumes no optical loss. Optical loss limits
the reflectivity, no matter how many layer pairs are used.
Figure 8.27 shows that when the adjacent layers have a normalised
step in the refractive index (Sn/Hn) falling below, say, 0.05, then
30-40 layer pairs are required to obtain adequate reflectivity, and this
is a relevant problem at present for materials such as InGaAsP/InP
290 Distributed feedback semiconductor lasers

forming appropriate combinations for emission at 1.3 or 1.5 juim

communication wavelengths. The large number of layers of semi-
conductor material gives optical losses, further significantly limiting
the available reflectivity. By contrast, Bragg stacks formed from high-
index-step dielectrics such as amorphous silicon and silicon dioxide
where na.S7~3.4 and ftm~1.9 will give adequate reflectivity with
around five or so pairs and also have sufficiently low optical absorption
so that the reflectivity is not affected at this level. Bragg coupling
coefficients range from K~ 2000 cm"1 for the dielectric stacks to
1
K ~ 2 0 0 cm" for the semiconductor stacks, and from such values one
estimates the associated stopband, or spectral width of high reflection,
to ranges from tens of nanometres for the dielectric stacks to a few
nanometres for the semiconductor stacks, so that the material gain
peak and the reflection peak have to be aligned more carefully than in
edge emitters because of this need for high gain.
It might be thought that VCSELs would have a much faster optical-
response time to direct-current modulation than edge emitters, simply
because of their short length and low volume. However the high
reflectivity of the facets means that the photon lifetimes are, in fact,
comparable with those of edge emitters and more careful discussion is
required. The small geometry of VCSELs makes them almost free from
longitudinal spatial-hole burning and consequently (provided that the
radial dimensions are small enough not to permit strong radial-hole
burning) the dynamic behaviour can often be adequately dealt with by
the standard rate equations of Section 4.1 provided that appropriate
changes are made. The first such change notes that the dielectric
mirrors store significant optical energy without any optical gain so that
now a longitudinal confinement factor (of order 0.2) has to be
introduced in addition to the transverse confinement factor, but this
latter is normally high in VCSELs. The net confinement factor is then
comparable with that for an edge emitter. A second notable change
has to be made in the spontaneous-coupling coefficient /3^ which can
be significantly larger (> 10~2) than for edge emitters [83-86] because
the short, low volume cavity increases the lasing linewidth and the high
reflectivity mirrors reduce the spontaneous linewidth compared with
an edge emitter. Novel microcavity VCSEL structures are candidates
for increasing f$sp and reducing the threshold current towards the so-
called 'thresholdless laser' mentioned in Section 4.2.2. A third
important change is that Auger recombination needs greater emphasis
in the modelling. Auger recombination is already a significant factor in
determining the threshold currents in edge-emitting long-wavelength
 Future devices, modelling and systems analysis 291

lasers, and may well be even more serious with VCSELs unless these
lasers can operate with lower threshold-current densities. The strong
temperature dependence of the nonradiative Auger recombination
will also be of concern.
Making allowance for such changes, analysis and experiment does
confirm that there is an overall increase in the speed of response [87]
for the following reasons:
(i) VCSELS tend to use high-gain material, at high carrier densities
(~5 X 1018 cm" 3 ), which then reduces the stimulated and sponta-
neous carrier lifetime, though it restricts the maximum
temperature of operation;
(ii) the small dimensions of VCSELs give small electrical capacitances
which are easier to charge and discharge; and
(iii) the resonance frequency increases as V(/— I(h) (Appendix 4) and
the low threshold currents in VCSELs means that one can drive
the laser at, say, 10 times threshold whereas edge emitters are
rarely driven at more than three times threshold.
If one uses time-domain modelling for VCSELs, some changes from
the techniques used for edge-emitting DFB/DBR lasers are advised.
The significant refractive-index steps in the Bragg reflectors, com-
bined with their length, suggests that the appropriate space step for
the computation is that of a single layer or perhaps layer pair within
the stack compared with the tens of grating periods used previously.
Using such small step lengths gives excellent results for computing the
reflection spectra of the stack mirrors, but also requires correspond-
ingly smaller time steps ~10~15s, rather than ~10~13s for
conventional DFBs.
8.5.3 Construction of VCSELs
VCSELs are fabricated using virtually identical technologies to those
developed for edge emitters but one critical process (facet cleaving) is
replaced by a different critical process: reflector stack fabrication.
Figure 8.28 indicates some varieties. Testing is potentially much easier
than with edge emitters since it may be done without dicing chips from
their wafer; this is a very considerable advantage.
Formation of dielectric-mirror stacks by vacuum evaporation of
materials such as SiO2, Mg2F, TiO2 etc. is a well established and reliable
technology. In-situ monitoring of reflectivity during deposition is often
employed, so that drifts in deposition rates, optical densities etc. can
292 Distributed feedback semiconductor lasers

active
region

oxidised proton
annulu bombarded

Figure 8.28 Varieties of VCSEL

All devices have circular emitting facets the order of 5-10 jim diameter
(a) Deposited top mirror, bottom grown, buried-heterostructure cur-
rent confinement
(b) Grown mirrors, mesa current confinement
(c) Grown mirrors, large mesa with oxidised current confining annulus
(d) Deposited top mirror, grown bottom, proton-bombarded current
confinement

be compensated, and high accuracy for the spectral characteristics of

the mirror is assured. Good lasers may be fabricated [88], but injecting
the current into the VCSEL junction with insulating mirrors remains a
problem, with solutions such as annular contacts being employed
(Figure 8.28c).
Conductive stacks grown from alternating layers of high- and low-
index semiconductor [89] (all doped n-type or all p-type) make for a
relatively simple laser structure (as in Figure 8.286). However, there
are associated difficulties caused by the stack of heterojunctions where,
without careful design [90,91], the heterobarriers at the many
interfaces inhibit conduction, with lasers needing 10 V of drive or
more to achieve threshold. This problem can be avoided by doping the
whole mirror stack very heavily, but high optical losses can then result.
An elegant solution is either to grade the interfaces, or to employ
'delta doping', i.e. heavy doping in very thin layers at the interfaces
[92,93].
Although metal organic vapour-phase epitaxial (MOVPE) growth is
 Future devices, modelling and systems analysis 293

a preferred growth technique for large-scale production, the stringent

material tolerances required for VCSEL devices lead to the preference
for growing VCSEL mirrors by molecular-beam-epitaxy (MBE) where
in-situ monitoring of the growth process and optical reflectivity is
easier [94,95], but unfortunately the difference in volatility of In and
P makes MBE less easy to use for InP-based devices. The detailed
choice of method(s) depends on the material and device design
[96-100], but the difficulties of achieving the right compromises have
made it easier to fabricate devices, with the GaAlAs/GaAs and GalnAs/
GaAs systems giving outputs around 0.8 |mm and 1.0 fxm, respectively,
where the step-index variations are sufficient to permit about 25-30
layer pairs to be used. Significantly more layers are usually required for
InGaAsP/InP devices centred around the 1300 nm and 1550 nm
bands. Although a higher step in the refractive index is possible at
1550 nm than at 1300 nm, the maximum gain tends to be lower and
low-loss mirrors are a major difficulty at 1550 nm. Sophisticated
techniques may provide a solution such as diffusion bonding a thin
'film' of VCSEL-active regions onto a substrate of an entirely different
material in which low-loss mirrors can be formed. Even if a layer of
dislocations is formed at such a bond, it need not have over-serious
consequences on the operating lifetime of the laser provided that the
dislocation layer and active region are sufficiently far apart. Two-
dimensional arrays of VCSELs fabricated on the top of standard silicon
integrated circuits would be especially attractive for optical signal
processing, combining the best of Si VLSI and VCSEL technology.
Progress towards these goals is being made using interfacial matching
layers or diffusion bonding [101-104].
Because of the short length of the VCSEL, it is essential to have high
optical gain which is best obtained by using quantum-well material.
However, the position of any quantum wells is now more critical than
in an edge-emitting laser. The precision required is illustrated by
Figure 8.29, where it can be seen that the active layer (quantum well)
now lies parallel to the wave fronts in the laser, unlike the edge-
emitting laser. This gain region then should ideally be situated at a
where the optical standing wave, at the desired longitudinal modal
frequency, provides the strongest electric field. If the layer is at the
node b, then only a weak interaction can occur, giving an excessive
threshold current. Multiple wells should ideally be close to an
appropriate antinode for maximum gain. The deposition of the layers
then has to be significantly more accurate than Aw/4—the quarter-
wavelength inside the material.
294 Distributed feedback semiconductor lasers

lasing mode standing waves

active layer multilayer mirror

position "b
spacer layers
active layer
position "a"
multilayer mirror

Figure 8.29 Active-layer position in relation to the standing-electrical-field wave

8.5.4 Additional features of VCSELs

The slab-waveguide model outlined in Chapter 3 is inadequate for the
circular VCSEL devices other than to give an indication that the lateral
dimensions are likely to permit more than one lateral mode, even
though the cavity dimensions and selective mirrors ensure only one
longitudinal mode. If a number of lateral modes is present, the far-
field patterns are complicated [105]. The diameter of the active core
ranges from 5 to 50 (xm with the result that waveguiding for the
principal mode is weak and guiding is susceptible to perturbations: the
drive current can heat regions leading to thermal waveguiding; radial
spatial-hole burning in the junction plane can depress the carrier
density along the axis of the device. All this can cause significant
changes in the mode guiding, especially if the laser is driven well above
its threshold level [106-108] and provides the future challenges.
The polarisation is not usually well determined in VCSELs, which is
a serious drawback for long-haul optical-communication systems
[109]. Slight radial asymmetries can provide positive discrimination
for polarisation; for example anisotropic strain generated by growing
on misoriented substrates has had some success [110,111].

8.6 Summary
Four areas have been covered in this final chapter. The first has been
systems modelling where advanced numerical models can predict the
performance of assemblages of devices all of which are modelled
separately but can be brought together. It is believed that, as device
modelling improves in accuracy, this area will grow extensively and it
will reduce the costs of trying out new systems and speed up their
 Future devices, modelling and systems analysis 295

design. The second area discussed is a specialist area where numerical

modelling has led to a new device: the push-pull laser. Even if this
device does not become a production device, it highlights many
mechanisms of importance within DFB lasers and illustrates how
numerical modelling can aid invention.
The third and fourth areas are of growing importance. Tunable
lasers or switchable laser systems will give WDM systems the versatility
and flexibility they require. Which particular device or technique will
win and become the industry standard for a WDM source is, at the
time of writing, an open question. The final area of discussion is that
of surface-emitting lasers and particularly VCSELs where the potential
for growth in data processing and communications is vast. However,
one recognises that there are serious challenges for designing and
fabricating 1.55 |xm wavelength VCSELs where the desirable materials
at present give too small a refractive-index step and the nonradiative
Auger recombination can inhibit lasing unless low enough current
densities are achieved. It is possible that full and detailed three-
dimensional dynamic models of VCSELs may be helpful in their
optimisation, which may well be more critical than in edge emitters.
This shows up future modelling requirements, where techniques of
finite elements, for example, combined with the increasing power in
personal computers, will make such computation and design as easy as
the current time-domain models for edge-emitting DFBs.

8.7 References
1 LOWERY, A.J.: 'Computer-aided photonics design', IEEE Spectrum, 1997,
34, pp. 26-31
2 LOWERY, A. J.: 'Semiconductor device and lightwave systems perform-
ance modelling'. Technical digest of Optical Fiber Communication, 97
(OFC'97), OSA Technical Digest Series, 1997, 6, paper ThG6, pp. 267-268
3 LOWERY, A.J.: 'Transmission line modelling of semiconductor lasers: the
transmission line laser model', Int. J. Numer. Model, 1989, 2, pp. 249-265
4 LOWERY, A.J.: 'Transmission-line laser modelling of semiconductor laser
amplified optical communications systems', TEE Proc. J, 1992, 139, pp.
180-188
5 LOWERY, A.J., GURNEY, P.C.R., WANG, X-H., NGUYEN, L.V.T., CHAN,
Y.C., and PREMARATNE, M.: 'Time-domain simulation of photonic
devices, circuits and systems', Proceedings of Physics and Simulation of
Optoelectronic Devices, at SPIE Photonics West, 1996, 2693, pp. 624-635
6 YOUNG, T.P.: 'CAD tools for optoelectronic subsystems', GEC J. Res.,
1994, 11, pp. 110-121
296 Distributed feedback semiconductor lasers

7 SILVESTER, P.P., and FERRARI, R.L.: 'Finite elements for electrical

engineers' (Cambridge University Press, 1966), 3rd edn.
8 PAOLI, T., AND RIPPER, J.E.: 'Direct modulation of semiconductor
lasers', Proc. LEE, 1970, 58, pp. 1457-1465
9 OJALA, P., PETTERSSON, C, STOLTZ, B., MORNER, A.C., JANSON,
M., and SAHLEN, O.: 'DFB laser monolithically integrated with an
absorption modulator with low residual reflectance and small chirp',
Electron. Lett, 1993, 29, pp. 859-860
10 HASHIMOTO, J.I., NAKANO, Y, and TADA, K.: 'Influence of facet
reflection on the performance of a DFB laser integrated with an optical
amplifier/modulator', IEEEJ. Quantum Electron., 1992, 28, pp. 594-603
11 ZHANG, L.M., and CARROLL, J.E.: 'Semiconductor 1.55 micron laser
source with gigabit/second integrated electroabsorptive modulator',
IEEEJ. Quantum Electron., 1994, 30, pp. 2573-2577
12 MORITO, K, SAHARA, R., SATO, K., KOTAKI, Y, and SODA, H.: 'High
power modulator integrated DFB laser incorporating strain compressed
MQW and graded SCH modulator for 10 Gb/s transmission', Electron.
Lett, 1995, 31, pp. 975-976
13 BINSMA, J.J.M., KUINDERSMA, P.I., VANGESTEL, P., VANDERHOF-
STAD, G.L.A., CUIJPERS, G.P.J.M., PEETERS, R., VANDONGEN, T, and
THIJS, P.J.A.: 'DFB lasers with integrated electroabsorption modulator',
Opt. Quantum Electron., 1996, 28, pp. 455-462
14 GNAUCK, A.H., KOROTKY, S.K., VESELKA, J.J., NAGEL, J., KREM-
MERER, C.T., MINFORD, K.W., and MOSER, D.T.: 'Dispersion penalty
reduction using an optical modulator with adjustable chirp', IEEE
Photonics Technol Lett, 1991, 3, pp. 916-918
15 REICHMANN, K.C., MAGILL, P., KOREN, U., MILLER, B.I., YOUNG,
M., NEWKIRK, M., and CHEN, M.D.: '2.5 Gb/s transmission over 674 km
at multiple wavelengths using a tunable DBR laser with an integrated
electroabsorption modulator,' IEEE Photonics Technol. Lett, 1993, 5, pp.
1098-1100
16 PARK, Y.K., NGUYEN, TV., MORTON, P.A., JOHNSON, J.E., MIZU-
HURA, O., JEONG, J., TZENG, L.D. et at. 'Dispersion penalty free
transmission over 130 km standard fiber using a 1.55 mm 10 Gb/s
integrated EA/DFB laser with low extinction ratio and negative chirp,'
IEEE Photonics Technol Lett, 1995, 8, pp. 1255-1257
17 D'OTTAVI, A., MARTELLI, E, SPANO, P., MECOZZI, A., SCOTTI, S.,
DALL'ARA, R., ECKNER, J., and GUEKOS, G.: 'Very high efficiency four-
wave mixing in a single semiconductor traveling wave amplifier', Appl
Phys. Lett 1996, 68, pp. 2186-2188
18 D'OTTAVI, A., IANNONE, E., MECOZZI, A., SCOTTI, S., SPANO, P.,
DALL'ARA, R., GUEKOS, G., and ECKNER, J.: 'Ultrafast gain and
refractive index dynamics of semiconductor amplifiers measured by four-
wave mixing', Conference Digest, 14th IEEE International Semiconductor
Laser Conference, 1994, paper T2.3, pp. 63-64
19 CHENG, Y-K, and WU, M.: 'Monolithic colliding-pulse mode-locked
quantum well lasers', IEEEJ. Quantum Electron., 1992, 28, pp. 2176-2185
 Future devices, modelling and systems analysis 297

20 MARTINS-FILHO, J.F., and IRONSIDE, C.N.: 'Multiple colliding-pulse

mode-locked operation of a semiconductor laser', Appl Phys. Lett, 1994,
65, pp.1894-1896
21 MARCENAC, D.D.: 'Comparison of self-pulsation mechanisms in DFB
lasers', Presented at IEEE Lasers and Electro-Optics Society 7th Annual
meeting, Baltimore 1994, Paper SL8.5
22 NOWELL, M.C., ZHANG, L.M., CARROLL, J.E., and FICE, M.J.: 'Chirp
reduction using push-pull modulation of three-contact lasers', IEEE
Photonics Technol Lett, 1993, 5, pp. 1368-1371
23 NOWELL, M.C., CARROLL, J.E., PLUMB, R.G.S., MARCENAC, D.D.,
ROBERTSON, M.J., WICKES, H., and ZHANG, L.M.: 'Low chirp and
enhanced resonant frequency by direct push-pull modulation of DFB
lasers', IEEEJ. Sel. Topics Quantum Electron., 1995, 1, pp. 433-441
24 KAZARINOV, R.F., and HENRY, C.H.: 'Second order distributed
feedback lasers with mode selection provided by first order radiation
losses', IEEEJ. Quantum Electron., 1985, 21, pp. 144-150
25 MAKINO, T.: 'Mechanisms for high single mode stability of gain-coupled
DFB lasers with periodically etched quantum wells', Electron. Lett., 1995,
31, pp.1579-1580
26 GRIFFIN, P.S., WILLIAMS, K.A., WHITE, I.H., and FICE, M.J.: 'Sub-
stantial chirp reduction of dynamic wavelength chirp of 2 Gbit/s
pseudorandom optical signal using 3-contact InGaAsP/InP multi-quan-
tum well distributed feedback laser.' Electron. Lett., 1992, 28, pp.
2045-2046
27 FLANIGAN, B.J.: 'Advances in push-pull modulation of lasers', PhD
dissertation, University of Cambridge, 1996
28 OLSEN, CM., and OLESEN, H.: 'Time-resolving wavelength chirp with
Fabry-Perot etalons and gratings: a theoretical approach,' / . Lightwave
Technol., 1991, 9, pp. 436-441
29 PAKULSKI, G., BLAAUW, C , CLEROUX, M., and KIM, H.B.: 'Adjustable
chirp in three-electrode directly modulated distributed feedback lasers'.
Proceedings of IEEE Lasers and Electro-Optics Society 8th annual meeting, San
Francisco, 1995, 1, paper SCL 5.2
30 MARCENAC, D.D., NOWELL, M.C., and CARROLL, J.E.: 'Theory of
enhanced amplitude modulation bandwidth in push-pull modulated
DFB lasers,' IEEE Photonics Technol. Lett., 1994, 30, pp. 2064-2072
31 HAWKER, I., TANDON, V., COTTER, D., and HILL, A.: 'New network
infrastructures for the 21st century. Part 1: Ultra high capacity
wavelength routed networks', Br. Telecomm. Eng, 1995, 13, pp. 103-111
32 YOUNG, M.G.: TnP-based components for wavelength division multi-
plexing', Proc. SPIE, 1994, 2153, pp. 251-258
33 ISHII, H., TOHMORI, Y, YAMAMOTO, M., TAMAMURA, T, and
YOSHIKUNI, Y: 'Modified multiple phase shift superstructure grating
DBR lasers for broad wavelength tuning', Electron. Lett, 1994, 30, pp.
1146-1147
34 TABUCHI, H., and ISHIKAWA, H.: 'External grating tunable MQW laser
with wide tuning range of 240 nm', Electron. Lett., 1990, 26, pp. 742-743
298 Distributed feedback semiconductor lasers

35 PANKOVE,J.L: 'Optical processes in semiconductors' (Dover, New York,

1971)
36 MILLER, D.A.B., CHEMLA, D.S., DAMEN, T.C., GOSSARD, A.C.,
WIEGMANN, W., WOOD, T.H., and BURRUS, C.A.: 'Electric field
dependence of optical absorption near the band gap of quantum well
structures', Phys. Rev. B, 1985, 32, pp. 1043-1060
37 CAI, B., SEEDS, A.J., and ROBERTS, J.S.: 'MQW tuned semiconductor
lasers with uniform frequency response', IEEE Photonics Technol. Lett.,
1994, 6, pp. 496-498
38 WOODWARD, S.L., KOREN, U., MILLER, B.L, YOUNG, M.G., NEW-
KIRK, M.A., and BURRUS, C.A.: 'A DBR laser tunable by resistive
heating', IEEE Photonics Technol. Lett., 1992, 4, pp. 1330-1332
39 DUTTA, N.K., HOBSON, W.S., LOPATA, J., and ZYDZIK, G.: 'Tunable
InGaAs/GaAs/InGaP laser', Appl. Phys. Lett, 1997, 70, pp. 1219-1220
40 PANDIAN, G.S., and DILWALI, S.: 'On the thermal FM response of a
semiconductor-laser diode', IEEE Photonics Technol. Lett, 1992, 4, pp.
130-133
41 GRIFFEL, G., and CHEN, C.H.: 'Static and dynamic analysis of tunable
2-section high-speed distributed-feedback laser utilizing the gain lever
effect', IEEEJ. Quantum Electron., 1996, 32, pp. 61-68
42 TOHYAMA, M., ONOMURA, M., FUNEMIZU, M., and SUZUKI, N.:
'Wavelength tuning mechanism in three electrode DFB lasers', IEEE
Photonics Technol. Lett, 1993, 5, pp. 616-618
43 HILLMER, H., ZHU, H.L., and BURKHARD, H.: 'Enhanced tunability of
asymmetric three section InGaAsP/InP distributed feedback lasers', / .
Appl Phys., 1993, 73, pp. 1035-1038
44 JAYARAMAN, V., HEINBUCH, M.E., COLDREN, L.A., and DENBAARS,
S.P.: 'Widely tunable continuous wave InGaAsP/InP sampled grating
lasers', Electron. Lett, 1994, 30, pp. 1492-1494
45 OUGIER, C., TALNEAU, A., DELORME, R, RAFFLE, Y, LANDREAU, J.,
and MATHOORASING, D.: 'Sampled-grating DBR lasers with 80
addressable wavelengths over 33 nm for 2.5 Gbit/s WDM applications',
Electron. Lett, 1996, 32, (17), pp. 1592-1593
46 ISHII, H., TOHMORI, Y, YOSHIKUNI, Y, TAMAMURA, T., and
KONDO, Y: 'Multiple phase shift superstructure grating DBR lasers for
broad wavelength tuning', IEEE Photonics Technol. Lett, 1993, 5, pp.
613-615
47 ISHII, H., TOHMORI, Y, YAMAMOTO, M., TAMAMURA, T., and
YOSHIKUNI, Y: 'Modified multiple phase shift superstructure grating
DBR lasers for broad wavelength tuning', Electron. Lett, 1994, 30, pp.
1146-1147
48 KANO, R, ISHII, H., TOHMORI, Y, and YOSHIKUNI, Y: 'Character-
istics of superstructure grating (SSG) DBR lasers under broad range
wavelength tuning', IEEE Photonics Technol. Lett, 1993, 5, pp. 611-613
49 TOHMORI, Y, ANO, F.K., ISHII, H., YOSHIKUNI, Y, and KONDO, Y:
'Wide tuning with narrow linewidth in DFB laser with superstructure
grating (SSG)', Electron. Lett, 1993, 29, pp. 1350-1352
 Future devices, modelling and systems analysis 299

50 DUTTING, K., HILDEBRAND, O., BAUMS, D., IDLER, W., SCHILLING,

M., AND WUNSTEL, K.: 'Analysis and simple tuning scheme of
asymmetric Y lasers', IEEEJ. Quantum Electron., 1994, 30, pp. 654-659
51 KUZNETSOU, M., VERLANGIERI, P., DENTAI, A.G., JOYNER, C.H.,
BURRUS, C.A.: 'Asymmetric Y-branch tunable semiconductor laser with
1.0 THz tuning range', IEEE Photonics Technol Lett., 1992, 4, pp.
1093-1095
52 KUZNETSOV, M., VERLANGIERI, P., DENTAI, D.G., JOYNER, C.H.,
and BURRUS, C.A.: 'Widely tunable (45 nm, 5.6 thz) multi quantum well
three branch Y3 lasers for WDM networks', IEEE Photonics Technol. Lett.,
1993, 5, pp. 879-882
53 OBERG, M., NILSSON, S., STREUBEL, K., WALLIN, J., BACKBON, L.,
and KLINGA, T.: '74 nm wavelength tuning range of an InGaAsP/InP
vertical grating assisted codirectional coupler laser with rear sampled
grating reflector', IEEE Photonics Technol Lett., 1993, 5, pp. 735-737
54 KIM, I., ALFERNESS, R.C., KOREN, U, BUHL, L.L., MILLER, B.I.,
YOUNG, M.G., CHIEN, M.D., KOCH, T.L., PRESBY, H.M., RAYBON, G.,
and BURRUS, C.A.: 'Broadly tunable vertical coupler filtered tensile
strained InGaAs/InGaAsP multiple quantum well laser', Appl. Phys. Lett.,
1994, 64, pp. 2767-2769
55 WOLF, T., ILLEK, S., RIEGER, J., BORCHERT, B., and AMMAN, M.C.:
'Tunable twin guide (TTG) distributed feedback (DFB) laser with over
10 nm continuous tuning range', Electron. Lett, 1993, 29, pp. 2124-2125
56 THULKE, W., ACHATZ, V., BORCHERT, B., ILLEK, S., and WOLF, T:
Tntegratable tunable twin guide laser on semi-insulating substrate', Jpn.
J. Appl. Phys. 2, Lett, 1993, 32, pp. L914-L916
57 DIECKMANN, A.: 'Linewidth broadening in wavelength tunable laser
diode due to absorption and parasitic currents', Jpn. J. Appl. Phys. 2, Lett,
1994, 33, pp. L1222-L1223
58 AMANN, M.C., HAKIMI, R., BORCHERT, B., and ILLEK, S.: 'Linewidth
broadening by 1/f noise in wavelength tunable laser diodes', Appl. Phys.
Lett, 1997, 70, pp. 1512-1514
59 ZAH, C.E., PATHAK, B., FAVIRE, F.J., ANDREAKIS, N.C., BHAT, R.,
CANEAU, C , CURTIS, L., MAHONEY, D.D., YOUNG., W.C., and LEE,
T.P.: 'Monolithic integrated multiwavelength laser arrays for WDM
lightwave systems', Optoelectron. Devices Technol, 1994, 9, pp. 143-166
60 YOUNG, M.G., KOREN, U., MILLER, B.I., CHIEN, M., KOCH, T.L.,
TENNANT, D.M., FEDER, K., DREYER, K., and RAYBON, G.: '6-Wave-
length laser array with integrated amplifier and modulator', Electron.
Lett, 1995, 31, pp. 1835-1836
61 KATOH, Y, KUNII, T, MATSUI, Y, and KAMIJOH, T: '4 wavelength
DBR laser array with waveguide couplers fabricated using selective
MOVPE growth', Opt. Quantum Electron. 1996, 28, pp. 533-540
62 OKAMOTO, K, TAKIGUCHI, R., and OHMORI, Y: '8-channel flat
spectral response arrayed-waveguide multiplexer with asymmetrical
Mach-Zehnder filters', IEEE Photonics Technol. Lett, 1996, 8, pp. 373-374
300 Distributed feedback semiconductor lasers

63 PARK, C.Y, KIM, D.B., YOON, T.H., KIM, J.S., OH, K.R., LEE, S.W., LEE,
S.M., AHN, J.H., KIM, H.M., and PYUN, K.E.: 'Fabrication of wavelength-
tunable InGaAsP/InP grating-assisted codirectional coupler filter with
very narrow bandwidth', Electron. Lett, 1997, 33, pp. 773-774
64 ASGHARI, M., ZHU, B., WHITE, I.H., SELTZER, C.P., NICE, C ,
HENNING, I.D., BURNESS, A.L., and THOMPSON, G.H.B.: 'Demon-
stration of an integrated multichannel grating cavity laser for WDM
applications', Electron. Lett., 1994, 30, pp. 1674-1675
65 MILLER, B.L, DREYER, K, BEHRINGER, R.E., KOREN, M.G., CHIEN,
M., RAYBON, G., and CAPIK, R.J.: 'Low chirp wavelength selectable 1 x
6 arrays suitable for WDM applications', 15th IEEE International Semi-
conductor Laser Conference, 1996, paper Wl.3, pp. 129-130
66 SATO, K, and MURAKAMI, M. 'Experimental investigation of thermal
crosstalk in a distributed feedback laser array', IEEE Photonics Technol
Lett, 1991, 3, pp. 501-503
67 AMMAN, M.C.: 'Recent progress on wavelength tunable laser diodes',
Inst. Phys. Conf Sen, 1994, 136, pp. 257-264
68 MOLLENAUER, L.F., and WHITE,J.C. (Eds.): 'Tunable lasers' (Springer
Verlag, 1987)
69 OKUDA, H., SODA, H., MORIKI, K., MOTEGI, Y, and IGA, K.:
'GalnAsP-InP surface emitting injection laser with buried hetero-
structures',Jm./. Appl. Phys., 1981, 20, pp. L563-L566
70 MOTEGI, Y, SODA, H., and IGA, K: 'Surface emitting GalnAsP-InP
injection laser with short cavity length', Electron. Lett, 1982, 18, pp.
461-463
71 IGA, K, ISHIKAWA, S., OHKOUCHI, S., and NISHIMURA, T: 'Room
temperature pulsed operation of GaAlAs-GaAs surface emitting injec-
tion laser', Appl. Phys. Lett, 1984, 45, pp. 348-350
72 OU, S.S., YANG, J.J., andJANSEN, M.: '635 nm GalnP/GaAlInP surface
emitting laser diodes', Appl Phys. Lett, 1993, 63, pp. 3262-3264
73 TAKAMORI, T, COLDREN, L.A., and MERZ, J.L.: 'Folded cavity
transverse junction stripe surface emitting laser', Appl. Phys. Lett, 1989,
55, pp. 1053-1055
74 SAITO, H., and KONDO, Y: ' 4 x 4 surface emitting 1.55>m InGaAsP/
InP laser arrays with microcoated reflectors fabricated by reactive ion
etching', fpn.J. Appl. Phys. Lett, 1991, 30, pp. L599-L601
75 GOODHUE, W.D., DONNELLY, J.P., WANG, C.A., LINCOLN, G.A.,
RAUSCHENBACH, K., BAILEY, R.J., and JOHNSON, G.D.: 'Monolithic
2-dimensional InGaAs/AlGaAs and AHnGaAs/AlGaAs diode laser arrays
with over 50% differential quantum efficiencies', Appl. Phys. Lett, 1991,
59, pp. 632-634
76 STEGMULLER, B., WESTERMEIER, H., THULKE, W., FRANZ, G., and
SACHER, D.: 'Surface emitting InGaAsP/InP distributed feedback laser
diode with monolithic integrated microlens', IEEE Photonics Technol. Lett,
1991,3, pp. 776-778
77 EVANS, G.A., CARLSON, N.W., BOUR, D.P., LURIE, M., DEFREEZ,
R.K., and BOSSERT, D.J.: '14-W peak power grating surface-emitting
laser array', Electron. Lett, 1990, 26, pp. 1380-1381
 Future devices, modelling and systems analysis 301

78 YU, S.F., PLUMB, R.G.S., and CARROLL, J.E.: 'Spatial active optical
switching by using grating coupled surface emitting DFB lasers', Electron.
Lett., 1993, 29, pp. 1147-1148
79 JEWELL, J.L., HUANG, K.E, TAI, K., LEE, Y.H., FISCHER, R.J.,
MCCALL, S.L., and CHO, A.Y.: 'Vertical cavity single quantum well laser',
Appl. Phys. Lett., 1989, 55, pp. 424-426
80 JEWELL, J.L., HARBISON, J.P., SCHERER, A., LEE, Y.H., and FLOREZ,
L.T.: 'Vertical-cavity surface-emitting lasers—design, growth, fabrication,
characterization', IEEEJ. Quantum Electron., 1991, 27, pp. 1332-1346
81 IGA, K., ISHIKAWA, S., OHKOUCHI, S., and NISHIMURA, T.: 'Room
temperature pulsed operation of GaALAs-GaAs surface emitting injec-
tion laser', Appl. Phys. Lett., 1984, 45, pp. 348-350
82 SALES, T.E., 'Vertical cavity surface emitting lasers', (Wiley, New York,
1995)
83 SHTENGEL, G., TEMKIN, H., BRUSENBACH, P., UCHIDA, T., KIM, M.,
PARSONS, C , QUINN, W.E., and SWIRHUN, S.E.: 'High-speed vertical-
cavity surface-emitting laser', IEEE Photonics Technol. Lett., 1993, 5, pp.
1359-1362
84 BABA, T., HAMANO, T., KOYAMA, E, and IGA, K: 'Spontaneous
emission factor of a microcavity DBR surface emitting laser', IEEE J.
Quantum Electron., 1991, 27, pp. 1347-1358
85 BABA, T., HAMANO, T, KOYAMA, E, and IGA, K.: 'Spontaneous
emission factor of a microcavity DBR surface emitting laser, 2, effects of
electron quantum confinements', IEEEJ. Quantum Electron., 1992, 28, pp.
1310-1319
86 HAMANO, T., IGA, K., and BABA, T.: 'Spontaneous emission behavior
and its injection level dependence in 3-dimensional microcavity surface-
emitting lasers', Electron. Commun. Jpn. 2, Electron., 1996, 79, pp. 46-54
87 SANDUSKY, J.V., and BRUECK, S.R.J.: 'Observation of spontaneous
emission microcavity effects in an external-cavity surface-emitting laser
structure', Appl Phys. Lett, 1996, 69, pp. 3993-3995
88 WATANABE, I., KOYAMA, E, and IGA, K.: 'GalnAsP-InP CBH surface
emitting laser with a dielectric multilayer reflector', Jpn. J. Appl Phys. 1,
Regul Pap. Short Notes, 1987, 26, pp. 1598-1599
89 CHAILERTVANITKUL, A., IGA, K., and MORIKI, K.: 'GalnAsP-InP
surface emitting laser (A=1.4[xm, 77 K) with heteromultilayer Bragg
reflector', Electron. Lett., 1985, 21, pp. 303-304
90 HOUNG, Y.M., TAN, M.R.T., LIANG, B.W., WANG, S.Y, YANG, L., and
MARS, D.E.: 'InGaAs(0.98 jxm)/GaAs vertical cavity surface emitting
laser grown by gas-source molecular beam epitaxy', / Crystal Growth,
1994, 136, pp. 216-220
91 SUGIMOTO, M., KOSAKA, H., KURIHARA, K., OGURA, I., NUMAI, T.,
and KASAHARA, K.: 'Very low threshold current density in vertical cavity
surface emitting laser diodes with periodically doped distributed Bragg
reflectors', Electron. Lett., 1992, 28, pp. 385-387
92 ZEEB, E., MOLLER, B., REINER, G., RIES, M., HACKBARTH, T., and
EBELING, K. J.: 'Planar proton implanted VCSELs and fiber-coupled 2-D
VCSEL arrays', IEEE J. Select. Topics Quantum Electron., 1995, 1 , pp.
302 Distributed feedback semiconductor lasers

93 KIM, J.Y., HAYASHI, Y, MUKAIHARA, T., OHNOKI, N., KOYAMA, E,

and IGA, K.: 'Improvement of current density-voltage characteristics of
GaAs/AlAs distributed Bragg reflectors in InGaAs/AlGaAs surface-
emitting lasers',/. Korean Phys. Soc, 1995, 28, pp. 495-498
94 IGA, K., NISHIMURA, T., YAGI, K., YAMAGUCHI, X, and NIINA, X:
'Room temperature pulsed oscillation of GaAs/GaAs surface emitting
junction laser grown by MBE\Jpn.J. Appl. Phys. 1 ReguL Pap. Short Notes,
1986, 25, pp. 924-925
95 GOURLEY, P.L., and DRUMMOND, X.J.: 'Visible, room temperature,
surface emitting laser using an epitaxial Fabry-Perot resonator with
AlGaAs/AlAs quarter wave high reflectors and AlGaAs/GaAs multiple
quantum wells', Appl. Phys. Lett., 1987, 50, pp. 1225-1227
96 KOYAMA, E, XOMOMAXSU, K., and IGA, K.: 'GaAs surface emitting
lasers with circular buried heterostructure grown by metalorganic
chemical vapor deposition and two-dimensional laser array', Appl. Phys.
Lett, 1988, 52, pp. 528-529
97 YANG, L., WU, M.C., XAI, K., XANBUNEK, X, and LOGAN, R.A.:
'InGaAsP (1.3 jxm)/InP vertical cavity surface emitting laser grown by
metalorganic vapor-phase epitaxy', Appl. Phys. Lett, 1990, 56, pp.
889-891
98 HIBBSBRENNER, M.K., SCHNEIDER, R.P., MORGAN, R.A., WALXER-
SON, R.A., LEHMAN, J.A., KALWEIX, E.L., LOXX, J.A., LEAR, K.L.,
CHOQUEXXE, K.D., and JUERGENSEN, H.: 'Metalorganic vapor phase
epitaxial growth of red and infrared vertical cavity surface emitting laser
diodes', Microelectron.J., 1994, 25, pp. 747-755
99 LEI, C., ROGERS, X.J., DEPPE, D.G., and SXREEXMAN, B.G.: 'InGaAs-
GaAs quantum well vertical cavity surface emitting laser using molecular
beam epitaxial regrowth', Appl. Phys. Lett., 1991, 58, pp. 1122-1124
100 LEI, C , ROGERS, X.J., DEPPE, D.G., and SXREEXMAN, B.G.: 'ZnSe/
CaF2 quarter wave Bragg reflector for the vertical cavity surface emitting
laser',/ Appl. Phys., 1991, 69, pp. 7430-7434
101 OHISO, Y, AMANO, C , IXOH, Y, XAXENO, K., XADOKORO, X,
XAKENOUCHI, H., and KUROKAWA, X, '1.55 fxm vertical-cavity sur-
face-emitting lasers with wafer-fused InGaAsP/InP-GaAs/AlAs DBRs',
Electron. Lett, 1996, 32, pp. 1483-1484
102 CHUA, C.L., LIN, C.H., ZHU, Z.H., LO, Y.H., HONG, M., MANNAERXS,
J.P., and BHAX, R.: 'Dielectrically-bonded long-wavelength vertical-cavity
laser on GaAs substrates using strain-compensated multiple-quantum
wells', IEEE Photonics TechnoL Lett, 1994, 6, (12), pp. 1400-1402
103 EGAWA, X, HASEGAWA, Y, JIMBO, X, and UMENO, M.: 'Room
temperature pulsed operation of AlGaAs/GaAs vertical-cavity surface-
emitting laser-diode on Si substrate', IEEE Photonics TechnoL Lett, 1994, 6,
(6), pp. 681-683
104 SCHRAUD, G., GROXHE, H., and SCHRODER, S.: 'Substrateless
singlemode vertical-cavity surface emitting GaAs/GaAlAs laser diode',
Electron. Lett, 1994, 30, (3), pp. 238-239
105 DELLUNDE, J., VALLE, A., and SHORE, K.A.: Transverse mode
selection in external cavity surface emitting laser diodes',/. Opt Soc. Am.
B, Opt Phys., 1996, 13, (11), pp. 2477-2483
 Future devices, modelling and systems analysis 303

106 SCOTT, J.W., GEELS, R.S., CORZINE, S.W., and COLDREN, L.A.:
'Modelling temperature effects and spatial hole burning to optimise
vertical cavity surface emitting laser performance', IEEE J. Quantum
Electron., 1993, 29, pp. 1295-1308
107 WILSON, G.C., KUCHTA, D.M., WALKER, J.D., and SMITH, J.S.: 'Spatial
hole-burning and self focusing in vertical cavity surface emitting laser
diodes', Appl Phys. Lett, 1994, 64, pp. 542-544
108 WIPIEJEWSKI, T, PETERS, M.G., THIBEAULT, B.J., YOUNG, D.B., and
COLDREN, L.A.: 'Size dependent output power saturation of vertical
cavity surface emitting laser diodes', IEEE Photonics Technol. Lett., 1996, 8,
pp. 10-12
109 FIEDLER, U., REINER, G., SCHNITZER, P., and EBELING, K.J.: 'Top
surface-emitting vertical cavity laser diodes for 10 Gbit/s data transmis-
sion', IEEE Photonics Technol. Lett, 1996, 8, pp. 746-748
110 NUMAI, T, KURIHARA, K., KUHN, K., KOSAKA, H., OGURA, I.,
KAJITA, M., SAITO, H., and KASAHARA, K.: 'Control of light output
polarization for surface emitting laser type device by strained active layer
grown on misoriented substrate', IEEE]. Quantum Electron., 1995, 31, pp.
636-642
111 TAKAHASHI, M., VACCARO, P., FUJITA, K., WATANABE, T, MUKAI-
HARA, T., KOYAMA, E, and IGA, K.: 'InGaAs-GaAs vertical cavity surface
emitting laser grown on GaAs (311) A substrate having low threshold and
stable polarization', IEEE Photonics Technol. Lett., 1996, 8, pp. 737-739
 Appendix 1
Maxwell, plane waves and reflections

This appendix provides a summary of plane-wave interactions at

dielectric interfaces and a summary of special cases which are of
importance in laser-diode design, in particular, indicating one reason
why TE modes are slightly more strongly reflected from a cleaved facet
than TM modes.

Al.l The wave equation

Start with Maxwell's two famous equations [1-3] for the electro-
magnetic fields:
curl E= - iar jmodH/ dt (juir= 1 in this book) (Al. 1)
curl H=ereodE/dt (A1.2)
In this work, the optical waveguide is formed from uniform layered
slabs of different relative permittivity er=n2=(nr+j n{)2, giving the
material's complex refractive index n with nr determining the real
refractive index and n? determining the optical gain (Appendix 5).
Taking the notation of Figure 3.1, the normal to the optical slab guides
is in the Oy direction and the TE (TM) waves may be classified as those
modes with the E (H) field in the plane of the slabs i.e. in the Ox
direction. With a uniform permittivity within each slab, elimination of
Hfrom eqns. Al.l and A1.2 leads to the wave equation
V2E-mr/jL0sre0d2E/dt2=0

A1.2 Linearly polarised plane waves (in a uniform 'infinite*

material)
Taking an infinite uniform isotropic slab of material where there is no
variation over the xy plane (d/dx=d/dy=0) but only electric fields Ex
 Appendix 1 305

Figure A 1.1 Right-handed sets for plane waves

When a plane wave is launched at an angle to the Oz axis with E parallel
to Ox then note there is a component Hz and a component field Hv.
This is typical in slab guides for TE modes where a pencil of different
directions for p is possible with ft not quite parallel to the Oz axis. See
Appendix 2

with optical frequencies co, these fields propagate in the Oz direction

as:
Ex=Fexp\j(a>t-J3z)}
where, from eqn. A1.3,

with c= (l//i o eo)1/2 = 3x 108 m/s, the velocity of light in vacuum. The
phase velocity of light within material is cw=3x 108/ nr m/s which varies
slightly with the optical frequency. In general, with optical gain or loss
where n{ is nonzero, (3 is complex.
The //fields are given from eqn. Al.l by putting d/dx=d/dy=0 with
d/dt-^jo) so that only a component Hy is found. For forward-
propagating fields:
(curl H)x= - dHy/dz=jPHy=ere0dEx/dt=ja>ereoEx (A1.6)
Hy=(8rs0/firfjL0)]/2Ex-^Ex/Hy=(377/n) O (A1.7)
In isotropic material, the E field could lie along any direction but the
orientation offieldswith respect to the vector direction of propagation
/?= ((3X, (3y, f3z) must be maintained. For plane linearly polarised waves,
the forward E field lies along one direction with Ep i^and /^forming
a right-handed set for the forward wave as shown in Figure Al.l and
similarly for the reverse wave Er9 Hr and /3r form a right-handed set.

A1.3 SnelPs law and total internal reflection

When plane waves are incident on an interface, they give rise to
transmitted and reflected plane waves, as indicated in a ray diagram
306 Distributed feedback semiconductor lasers

I R

Figure A 1.2 Notation for Snell s law

Total internal reflection occurs when sin 6t>\, i.e. 0t is imaginary

Wave variation along

interface cannot vary
across the boundary ^
Figure A 1.3 The boundary conditions
The continuity of any fields cannot be satisfied unless they have the
same spatial periodicity actually over the interface

showing the choice of axes in Figure A1.2. Because all tangential fields
have to be continuous, the periodicity of these three waves, resolved
along the interface must be identical (Figure A1.3) to one another.
This periodicity is determined from /3zi, the axial component of the
propagation vector(0, fiyi, /3zi) of the incident wave, and similarly
replacing the subscript i with r and t for the reflected and transmitted
waves, respectively. Note that (3xi/r/l=0 shows that the light rays lie in the
yz plane. Hence, from the equal periodicity condition,
/3zl=(3zr=f3zt
The ray angles of the incident transmitted and reflected directions are
 Appendix 1 307

Figure A 1.4 Total internal reflections

given from:
tan 6l=pu/pyi tan 0=Pzr/Pyr tan 6t=Pzt/Py( (A1.9)
From eqn. A1.5, one has for the i and rwaves in region 1 and t wave in
region 2, respectively,

Hence from eqn. A1.8 one recovers Snell's law for reflection and
refraction in lossless material with real refractive indices
sin di=sin ^ ( r ^ / n j s i n 0t (Al.ll)
The 'critical ray' has ^=90° when sin fli=(n2/n1)<l. When sin dt>\
it is not possible to find any real solution to Snell's law of eqn. A l . l l
but explicit expressions can still be found for the vertical 'propagation'
in the region 2:

Pyt=±j{(p0nl)2 sin2 0- (pon2)2}l/2=±jyyl (A1.13)

The 'y propagation coefficient1 is now imaginary, giving 'evanescent
variations exp(±yyly) in the y direction. With the next boundary 'far
away', only the decaying evanescent solution is physically possible in
region 2 ( Figure Al .4) with the evanescent fields storing energy only
significantly within a wavelength from the boundary, but note that this
energy does not propagate away from the boundary. Power flow in the
y direction is solely in region 1. This is the physics of total internal
reflection (TIR).
308 Distributed feedback semiconductor lasers

i i \
T 2
Figure A1.5 Reflection/transmission amplitudes at surfaces
F=£forTE fields
F=J/forTM fields

A1.4 Reflection amplitudes at surfaces: TE fields

If the E field lies parallel to the interfacial plane as sketched in Figure

A1.5, the field excites dipoles in the dielectric which in turn excite
E-fields in the same direction in both region 1 and 2. The TE fields are
therefore maintained throughout the slab. The H2 fields are provided
by (taking fJLr=l) and using eqn. Al.l:
(curl E)= - dEx/dy=j(3y£x= - fJLrniodHz/dt= - jo)^Hz (A1.14)

giving
S77(co/c)H=-/3yEx (A1.15)
The continuity of the tangential-field components Ex and Hz requires
Exi+Exr=Exl ^yiE^liyrEx=^JExt (A1.16)

Noting that fiyi/r=± (wn 1 /c)cos 0,-;fiyt=- (co n<>/c)cos 0t and 0,-= 0r
nx cos O^— nYcos 0rExr=n2 cos 0tExt (Al.l7)
Eliminating appropriately from eqns. A1.15 and Al.l7 leads, after a
little work, to reflection and transmission coefficients:
cos
p=Exr/Exi={(nl cos 0,-^2 &t)/ (ni c o s 0,-+w2 cos 0,)} (Al.l8)
T=Ext/Exi={2nx cos 0J(nx cos 0,-+n2 cos 0,)} (A1.19)
 Appendix 1 309

A1.5 TE reflection amplitudes: three special cases

(i) Normal incidence: 02= 6t=0 gives the so-called 'normal' incidence
where the amplitude reflection coefficient p=Exr/Exi is given from eqn.
A1.18by
A)=(wi-n 2 )/(w 1 + w2) (A1.20)
This is used in considering Bragg gratings.
(ii) Total internal reflection: cos 6t is purely imaginary, and p=Exr/Exi
is then of the form p= (C- jK) / (C+jK) where Cand J^are appropriate
real numbers; whence lpl = l confirming that there can be no power
lost. Total internal reflection is fundamental to the understanding of
guided waves along slab structures.
(iii) Near-normal incidence: For small angles close to the normal
incidence cos 86t — 1 — %8df then, making use of Snell's law,
cos 9t—l-2(nl/n2)2 89f. The reflection coefficient for the TE modes
may be written, again using eqn. A1.18, as

The near-normal incidence is typical of the wavefronts close to the

facets of a laser with a range of 80{.

A1.6 Reflection amplitudes at surfaces: TM fields

Significantly different results occur when the H field is parallel to the
interface, taking F=H in Figure A1.5. Again the arguments about
dipole excitations of E fields ensure that TM field will be preserved
across the boundary. The tangential fields which are continuous are
now the Hx fields and the Ez fields which are related through eqn.
A1.2:
(curl H)= - dHx/dy=j(3yJlx=srsodEz/dt= -jcos.s^ (A1.22)
(l/377)n2(co/c)E=pyHx (A1.23)
Similarly, the transverse electric field may also be found from eqn.
A1.2:
(l/377)n2(co/c)Ey={3zHx (A1.24)
Because the E field reflection coefficient for the TM mode will be
required in order to compare with the E field reflection coefficient of
the TE mode, the continuity equations for the tangential H fields and
E fields are written as:
310 Distributed feedback semiconductor lasers

(A1.25)

Noting again that fiyi/r=±{a)nl/c) cos 6-, /3yi/r= - (con^/c) cos 0t and still

{n.EJ cos e{) - {n.EJcos 6t) = (n2Ezt/cos d{) (A1.26)

Comparing with the TE case, one can make the appropriate changes
to find
6t)}/{(nx/cos ^-) + (^/cos 0t)}] (A1.27)
E^E^in,/cos OJ/ifa/cos fl.O + ^ / c o s dt)}] (A1.28)
The E field reflection coefficient corresponding to p=Exr/Exi in eqn.
A1.18 should take the full magnitude of the E field, so that now
p={EJcos 6r)/{EJcos 6t)
={(nx cos 9t— T?2 cos 9i)/(nx cos O^n^ cos 0{)} (A1.29)

A1.7 TM reflection amplitudes at surfaces: four special

cases
(i) Normal incidence and (ii) Total internal reflection: These remain in
most essentials unaltered from the TE case. The reflection coefficient
at normal incidence has to be identical because, with normal
incidence, the E field is also parallel to the surface and there physically
cannot be any difference. For the total internal reflection, the total
amplitude-reflection coefficient remains of modulus unity, but there
are different phase changes compared with the TE case. The similar
results here indicate that waves can be guided and Bragg gratings will
still operate in essentially the same manner for either TE or TM
modes. There will be significant differences because of the boundary
conditions and slight differences at near normal reflections as seen
below.
(ii) Near normal incidence: The third special case of nearly normal
incidence is important for laser structures because it helps to show why
TE rather than TM modes are selected in structures where there are
facets. When cos 80^1 -\80?\ cos 0t^l -\{n x /TI^) 86? and now the
reflection coefficient for the TM modes can be written as pTM using
Po= {(^i
 Appendix 1 311

(A1.30)
giving

(iv) Brewster's angle: A new special case arises in that it is now possible
to have pTM-Q at a critical angle: the 'Brewster' angle. This happens
when
0,)}=O (A1.32)
Using SnelFs law and writing n^ (1 — sin2 #,) = ?%2 (1 — sin2 6t)
gives sin ^ =/22/V(n2+72|) or tan d^n^/rix (A1.33)
This is the Brewster angle of incidence where there is no reflection of
the TM wave and only a TE wave is reflected. This is the principle on
which Polaroid glasses help to cut out some of the glare from
reflection. This angle is of special significance in gas lasers, and has
also been used in semiconductor lasers to cut out facet reflections (at
one polarisation) at the semiconductor/air interface. However, the
steepness of the Brewster angle (#i~70°) for a typical semiconductor/
air interface makes this unattractive compared with the present
technology for low-reflectivity surfaces where multiple dielectric thin
films are deposited with precision over the surface and can keep
reflections below 0.1% for near normal incidence.

A1.8 Reflection for waveguide modes at facets

In many situations with an index-guided mode, the reflection at the
facet may be calculated with sufficient accuracy from the effective
relative refractive index and unity (air). However, it is important to be
able to understand the significant difference between the observed
reflectivities of TE and TM modes.
All waveguide modes can be regarded as haying a pencil of rays
covering a mean-square range of incident angles 86? with this average
taking into account the different weighting of each ray angle with the
ray intensity. Hence from eqns. A1.21 and A1.30jt follows that the ratio
of TM modes to TE-mode reflectance- (1 - 2 86? neff) where n^is the
effective refractive index of the guide. The slightly higher reflectivity
for the TE mode gives a lower threshold to a laser and this is one
reason for the selectivity of certain laser structures in oscillating in TE
rather than TM modes . However, the detailed calculation of reflection
312 Distributed feedback semiconductor lasers

of the wave front within a waveguide is not entirely straightforward [4]:

the wavefront will curve to some extent, especially with strong gain,
and radiation modes can affect the outcome. When it comes to the TE
gain margins for DFB lasers, there are further detailed
problems concerned with the effects of the grating reflections
in TE/TM mode reflections which this appendix has not addressed
but are left for further reading [5].
The difficulty of precise calculation of reflection is compounded, of
course, when multilayer antireflection or high-reflectivity layers are
evaporated onto the facets or grown into the laser, as with VCSELs,
where optical diffraction can reduce the reflectivity calculated from a
pure-plane-wave analysis [6]. To gain low facet reflectivities for lasers,
careful experimental monitoring is usually performed rather than
using precise calculation.

A1.9 References
1 STRATTON, J.A.: 'Electromagnetic theory' (McGraw Hill, New York,
1941)
2 RAMO, S., WHINNERY, J.R. and VAN DUZER, T.: 'Fields and waves in
communication electronics', (J.Wiley, New York, 1993), 3rd Ed.
3 LONGAIR, M.S.: 'Theoretical concepts in physics' (Cambridge University
Press, 1984) (case study 2 gives an excellent historical perspective)
4 HERZINGER, CM., LU, C.C., DETEMPLE, T.A., and CHEW, W.C.: T h e
semiconductor waveguide facet problem', IEEEJ. Quantum Electron., 1993,
29, pp. 2273-81
5 YU, B.M., and LIU, J.M.: 'Gain margin analysis of distributed feedback
lasers for both transverse electric and magnetic-modes', IEEEJ. Quantum
Electron., 1992, 28, pp. 822-832
6 SALE, T.E.: 'Vertical cavity surface emitting lasers' (Research Studies Press,
Taunton; Wiley, New York, 1995)
 Appendix 2
Algorithms for the multilayer
slab guide

A2.1 TE slab modes

This appendix gives a systematic approach to providing a program for
solving propagation, confinement factor and far-field emission when
electromagnetic waves are guided by slab waveguides. Almost arbitrary
numbers of layers can be computed with complex refractive indices.
Chapter 3 covers slabs with only three layers, where there are many
fine texts that can provide back-up material [1,2]. Here, by examining
a five-layer system, the way forward to a semi-automated method for an
arbitrary number of layers with complex refractive indices is demon-
strated. Multilayer guides are particularly relevant in discussing DFB
lasers where the longitudinally periodic profile of the permittivity has
lateral variations across the waveguide which can be taken into account
by having a series of layers with different patterns of permittivity, giving
an average permittivity and an average periodic component which can
be calculated using the programs.
The classification of TE was given in Appendix 1, where TE plane
waves had an Hz component because the associated set of orthogonal
E, H and fi vectors has /? at a slight angle to the actual Oz as in Figure
Al. 1. This plane wave then 'slaloms' between the slabs with (3Z common
to all layers but with varying vertical 'propagation' f}y from layer to
layer. Five layers of arbitrary material will demonstrate the procedures.
Here the top layer is labelled with the symbol T and other layers
labelled from 1 to 4 (etc.) sequentially.
Now in each layer, by definition of a mode, one has to have the same
value for the axial propagation /3Z and for uniformity of notation one
now has a vertical propagation as exp( — yry) in the rth layer for the
wave 'travelling' in the ' + / direction and varying as exp( + yry)
'travelling' in the '—/ direction. Temporal variations are taken as
314 Distribu ted feedback semiconductor lasers

exp(+ja>0 and the magnetic fields relate to the electric fields from
jmodHz/dt=dEx/dy that with the wave equation give

It is then observed that this problem is like a transmission-line

problem [3] where there is, effectively, a terminating impedance Zrfor
the top layer being fed from layers of different impedances and optical
thicknesses. The impedance Zr is the wave impedance for the layer r
with ZQ the characteristic wave impedance of free space. YQ~ 1/ ZQ is the
wave admittance of free space. Starting from the top layer labelled T,
the wave in this layer T is evanescing as exp( —7r;y). The wave
admittance for this layer is jY0(yT/ko) and this is the 'terminating'
impedance for the layer 1. The sign of the real part of yr is important
because the wave must evanesce away from the main guide and may
not grow or propagate if the guide is to sustain a guided mode. The
layer 1 is of thickness dx and one has a 'forward' wave exp( — yxy) with
a wave admittance of jYo(yl/k0) and a 'reverse' wave varying as
exp(y1))) with a wave admittance of -jY0(yl/ki)). The notation here
uses a subscripted r+ on the E fields for the 'forward' wave and a
subscripted r— on the 'reverse' wave. The interface in question is
denoted by subscripts ( )pq so that ( ) 71 denotes the interface between
the Tand 1 layers etc. Then at the interface we have matching E fields
and H fields:
(Exl + +Exl _ ) n = (ExT+ y, (Exl + - Exl _ ) n = yT{ExT+ - ExT_ ) n
(A2.2)

Figure A 2.1 MuItilayer slab guide

In the ideal theory, the slab is infinite in the ^-direction with no x
variations; in practice the x variations simply have to be slow enough
 Appendix 2 315

Eqn. A2.2 may be rewritten in a matrix form as

7\
iclA 7,-77.
(A2.3)

Moving a distance — dx (i.e. back from the interface) to reach the

interface 2/1, the propagation inside the layer is accounted for
through

(A2.4)

Then at the 2/1 interface one has the fields at the start of region 2
given from

U2+ 4 %+% % r i | r « + , (A25)

£
| £ J L^~7 7+7 J l

Then at the 3/2 interface one has the fields at the end of region 2
given from

This process is repeated until the bottom layer is reached. For

example, with the five layers as illustrated, define

r ., r .t> o BJy.*r, y,-y,\ (A27)

exp( - yA) J L ^ ~ Tr r<+ % J
1 Z) 1T (A2.8)

71727374 v
I tx^ J 4 3
316 Distributed feedback semiconductor lasers

It is again important to get the sign of y4 right because, just as for the
top layer T, the bottom layer (here layer 4) has to have an evanescent
wave with only Ex4_ evanescing away from the guiding layer. One also
notices that £M+ is zero and consequently we have the situation that

7i727s74

leading to the determinantal equation for guiding:

M n =0 (A2.ll)
Solving eqn. A2.ll numerically is not found to be the best technique
because, especially with complex refractive indices, the convergence of
search routines is not assured. The better method (see Figure 3.6) is to
note that, with only the correct form of evanescent wave 'transmitted'
at the bottom of the slab but an arbitrary propagation coefficient, in
general there will be a 'reflected' evanescent wave at the top with the
incorrect sign (i.e. growing as it moves away from the guide: y—>°o) as
well as the 'incident' evanescent component with the correct sign (i.e.
decaying as it moves away from the guide). A 'reflection coefficient' is
found by assuming that ExT__ at the top of the guide is not quite zero
but \p\ = \ExT_ /ExT+\ = \Mn/Ml2\ a n d it is found that log Ipl displays
clearer minima when one is close to a solution than any attempt to find
where Mn goes to zero. The value of lpl=0 ensures that there is only
one evanescent wave at the top and, by construction, only one at the
bottom of the slab and both have the correct sign to ensure
confinement of the optical wave with the amplitudes decaying away
from the guiding layers. Although at first sight it is an apparently
minor mathematical alteration to consider \Mn/Mn\ rather than
IMnl, this method turns out to be a major step towards a more stable
computational process with complex guides.
The numerical problem starts by making a series of estimates of the
effective refractive index and calculates log I pi, noting those indices
which give the lowest value. By alternately scanning the imaginary
parts and the real parts (rather like balancing an old-fashioned
impedance bridge in an electrical laboratory) one can converge
rapidly on values for the effective refractive index which make
\p\< 0.001, which ensures a reasonable accuracy to the effective
refractive index and an extremely good accuracy for the intensity
profile.
 Appendix 2 317

A2.2 TM slab modes

TM modes follow closely the scheme of solution for the TE mode. The
same value of (3Z holds for each layer, with vertical 'propagation' as
exp( — yry) in the rth layer for the wave 'travelling' in the ' + /
direction and varying as exp( + yr y) 'travelling' in the ' — / direction.
Temporal variations exp(+ja>t) with srsodEz/dt= — dHJdy along with
the wave equation give
2 ni 2 i 2

Z=EJHX= -jyr/ere0co= -j
2 o =( / i 0 /e 0 ) 1 / 2 (A2.12)
As before, one starts with the top layer Twith a single evanescent wave
varying as exp ( — yry) with a wave impedance of — j7^{yje,.#b)- The
layer 1, of thickness du has a 'forward' wave exp( — yxy) with a wave
impedance of —jl^^yjefc) and a 'reverse' wave varying as exp(yxy)
with a wave impedance of jZ^{yr/erko). As in the TE case, r+ and r—
indicate, respectively, the 'forward' and 'reverse' types of evanescent
wave in the layer r. Then at the interface we have matching H fields
and E fields, respectively:
(Hxl++HxX _ ) n = (HxT++Hxl^_) n

-H^),, (A2.13)

1= (ST7I) ~

l - Jn
Jn \{sryi)~ (exyT) {sTyx) + (exyT)
(A2.14)
Moving a distance — dx (i.e. back from the interface) to reach the
interface 2 / 1 , one accounts for the propagation inside the layer using

Then at the 2/1 interface one has the fields at the start of region 2
given from

L^-Jis \_{eiy2) - {s2y\) (^y2) + (^yi) J L

(A2.16)
318 Distributed feedback semiconductor lasers

Then at the 3/2 interface one has the fields at the end of region 2
given from

(A2.17)
12

As for the TE mode, this process is repeated until the bottom layer is
reached. For example, with the five layers as illustrated, define

(A2.18)
M= D^sD^H^H^ r (A2.19)

Jx4+ l
=M\
nx4x4-_ 143

(A2.20)
leading to the requirement for a solution, similar to that for the TE
mode, that
l/9l = IM n /M 12 l=0 (A2.21)
The solution technique is followed as for the TE mode.
The word of warning given elsewhere is repeated: it is the E fields
which interact with the electronic dipoles in the material caused by
electrons and holes recombining to give gain. Consequently it is the E
fields which are important in determining the strength of the
interaction and in determining the confinement factor in a laser.
When discussing the TM modes, any computing programs constructed
may well find it easiest initially to compute the H field but the final
interaction should compute the total E fields to indicate the strength
of the interaction with the electronic gain along the guide and to
estimate the confinement factor using the E fields.
 Appendix 2 319

E(y)

Figure A2.2 Farfieldcontributions

A2.3 Far fields

Not only does one need to find the field intensities within the guide,
but also one requires the far-field pattern. The first effect on the far
field is given from the phase changes of all the contributions to the far
field from the near field. Consider Figure A2.2 where a distribution of
near fields in the y direction leads to a distribution of the far field as
a function of the angle ft
Each element of the near field contributes to the far field but with
a phase shift caused by the additional distance y sin 8 with the
associated phase shift (2TT y/ A) sin ft As seen in Figure A2.2, the far
field is contributed to in the same proportion by each element of the
near field apart from this phase shift which adds a further distance of
travel so that fields can cancel or reinforce one another. On summing
all the elemental contributions from the near field, one finds what is,
in effect, a Fourier transform of the near field:

Efar{ 6) oc Ex( y) exp ( -jk^y sin 6) dy (A2.22)

•/.

This transform gives a first result for the far-field intensity as

proportional to \Efnr(0)\2.
However, there is another significant effect. This is caused by the
Huygens pattern of radiation from any aperture [4], shown in Figure
A2.3. It can be shown that any elemental aperture, no matter how
small, re-radiates into free space with a field polar diagram determined
by 2 (1 + cos 6). One notices that there is some reradiation of the field
in the reverse directions where 2'ir<\d\<'n'. However, this standard
 320 Distributed feedback semiconductor lasers

forward radiation
+cos9]

Vz [1 - ICOSGI] reverse radiation

Figure A2.3 Huygens' re-radiation pattern

Vz [1 + COS6 ]

i
/2 p^ [1 - cos e ]

refractive index n

Figure A2.4 Radiation with E field normal to plane of emission

The radiating aperture (white) with its known fields are placed just in
front of the laser facet (shaded) to show more clearly that the reverse re-
radiated light is reflected from the facet

pattern is valid only if there is no reflection of this reverse radiation.

With the semiconductor present, the material with refractive index
n> 1 causes this reverse radiation to be reflected, as indicated in Figure
A2.4 where one uses the reflection coefficients for TE and TMfieldsas
given in Appendix Al. Hence, for a TE mode, where
PTF(0) ={cos 6- V(n 2 -sin 2 0)}/{cos 6+ V(w 2 -sin 2 9)}
the net far-field intensity is formed from the initial effect combining
with a modified Huygens factor to give
Ifar=\Efar(9)\2OTE(9) (A2.23)
 Appendix 2 321

-cose]

refractive Index n
Figure A 2.5 Radiation with H field normal to plane of emission
The radiating aperture (white) with its known fields are placed just in
front of the laser facet (shaded) to show more clearly that the reverse
reradiated light is reflected from the facet

where OTE(0) is a power obliquity factor given from the expression

l{l(l+cos 0)+/o 7E (0)|(l-cos 0)}l2 and leads to
O7,:(0)=cos2 0l{l + V ( n 2 - s i n 2 0)}/{cos 0+V(n 2 »sin 2 0)}\2
for -7r/2<d<7r/2 (A2.24)
Now when the plane of the far field lies along the E field, one at first
sight simply replaces prE{0) by pTM{0) given by the expression
{V(n 2 -sin 2 0) - n2 cos 6}/{n2 cos 0+V(rc 2 -sin 2 0)}. However, a little
more care is required, as shown in Figure A2.5. The E-field
components, resolved tangential to the exciting E field, must lie in the
same sense in both the reverse and forward directions, as seen in
Figures A2.4 and A2.5. This is seen to force the direct and reflected E
field components to be in antiphase for their far-field contributions.
When the far-field components combine, one has to have a field-
obliquity factor obtained by subtracting the reflected fields from the
direct fields to obtain |{(l + cos 0) ~pTM(0) (1 ~cos 0)}.
The resulting far-field pattern is now given from
lfar=\Efar(0)\2Om(O) (A2.25)
where Om{0) is the obliquity factor given from

for -7r/2<0<7r/2 (A2.26)

Experimentally observed narrowing of the far fields is caused by these
322 Distributed feedback semiconductor lasers

Figure A2.6 Replacing etched/refilled thick layer with many thin layers for calculation
of a slab wavguide which includes a Bragg grating
Permittivity is then averaged along the length of the guide to find the
field patterns

obliquity factors, as was noted by Hockham [5]. The treatment here is

simpler than Hockham's and owes much to Lewin [6] who noted the
significance of the reflection from the dielectric material. If both the
result here and Hockham's results are normalised, the obliquity
factors give practically identical results for the TE mode with either
calculation, but it is hoped that the derivation here is more physically
intuitive as well as being based on sound physics. The reader may also
be interested in angled mirrors, and is referred to the literature for
this more advanced material [7].

A2.4 Slab waveguide program

Included with the series of MATLAB 4.0 programs is a program
slabexec in a directory slab. The 1 readme file will guide the reader for
obtaining results with multislab guides with complex permittivities and
selecting either the TE or TM mode. For quantum-well materials, one
can have a succession of thin slabs modelling individually the
quantum-well layers, the barriers and the separate confinement layers
required to confine the light around the quantum wells. There is, in
principle, no difficulty in having the large number of layers as
required, other than the limitations imposed by a student edition of
MATLAB 4.0. Field intensities, confinement factors and far-field
patterns are calculated (see Figures 3.7 and 3.8). Bragg-grating
calculations can also be taken into account by using a stack of thin
layers with varying periodic patterns of permittivity, as shown
schematically in Figure A2.6. The program takes the average permittiv-
ity over the length of each of the thin layers, calculates the
transverse-field pattern and then estimates the coupling coefficient
 Appendix 2 323

using the form of eqn. 4.37. However, the reader is advised to become
familiar with running the slab-waveguide program without gratings
before requesting the calculation for a Bragg grating.

A2.5 References
1 SNYDER, A.W., and LOVE, J.D.: 'Optical waveguide theory' (Chapman and
Hall, London, 1983)
2 ADAMS, M.J.: 'Introduction to optical guided waves' (Wiley, Chichester,
1981)
3 ROZZI, T.E., and IN'TVELD, G.H.: 'Fields and network analysis of
interacting step discontinuities in planar dielectric waveguides', IEEE
Trans., 1979, MTT-27, pp. 303-309
4 RAMO, S., WHINNERY, J.R., and VAN DUZER, X: 'Fields and waves in
communication electronics' (Wiley, 1994), 3rd ed., pp. 618-627
5 HOCKHAM, G.A.: 'Radiation from a solid state laser', Electron. Lett, 1973,
9, pp. 389-391
6 LEWIN, L.: 'Obliquity factor for radiation from solid state laser', Electron.
Lett., 1974, 10, pp. 134-135
7 LAU, T., and BALLANTYNE, J.M.: 'Two-dimensional analysis of a dielectric
waveguide mirror',/ Lightwave Technol, 1997, 15, pp. 551-558
 Appendix 3
Group refractive index of laser
waveguides

The group velocity is the velocity of a wave packet (i.e. the velocity of
energy) that is centred on a central carrier frequency f = (0/ 2TT. The
group velocity is different from the phase velocity for two reasons: (i)
the waveguide changes the propagation coefficient as a function of
frequency; and (ii) the material permittivity changes with frequency so
that the propagation coefficient in the material changes with
frequency. This appendix illustrates these two roles using a symmetric
three-layer waveguide so that one can appreciate the physics through
putting numbers into an analytic solution.
The analytic solution for the propagation coefficient as a function of
frequency is well documented for the TE mode in a symmetrical three-
slab guide [1-4] . The guide is illustrated in Figure 3.5, where the
effect of weak transverse guiding is neglected and only the strong
lateral guiding is considered. The thickness of the central layer is taken
here to be d and it has a relative permittivity erl and is surrounded by
layers with relative permittivity er2. The axial-propagation coefficient /3Z
determines, through the wave equation with an angular frequency co,
that the central region has a lateral propagation coefficient f3':

where /^=V (co2 e0 /x0) = (co/c) =2TT/ A with A the free-space wavelength.
In the outer regions there is an evanescent wave decaying spatially in
y with a rate
y)=Pl-kler2 (A3.2)
There are several different notations used in the literature giving
equivalent results, but the use of symmetry of the guide gives a
 Appendix 3 325

straightforward result [3,4] that the field profile with continuity of the
E fields must be of the form:
EX=EO cos((3yy) [0<y<d/2] (A3.3)

e x p { - yy(y- d/2)} [d/2<y] (ASA)

Continuity of the H field is equal to continuity of dEJdy at the

boundary where y=d/2, leading to the requirement that, for
propagation along the guide, the lateral propagation and evanescent
constants have to satisfy a transcendental equation:

(A3.5)

This equation may also be rewritten as

JeA.JjL.*^ (A3.6)

where Ser= (srl — s^). A normalised thickness for such a slab guide may
be given in terms of the 'Vparameter' where V= (co d/c) V(Ser) and
then, defining the optical phase change across half the width of the
guide ® = (fiy d/2), the equation which determines the guide's main
mode of operation is:

&
cos 0 = 2 - (A3.7)

As V—•O, so the guide width tends to zero but as @ increases (2 0/1/)

is limited to magnitude 1 and so V has to increase.
The phase velocity vp along the guide for this zero-order mode is
given from (co//3Z) and this is found from

(odV(Ser) 0
\^IL 2 ~ (A3.8)
c =V= cos 0
326 Distributed feedback semicondu ctor lasers

c/n

0.29
c/n V
0 1 2
normalised frequency, V
Figure A3.1 Phase velocity against normalised guide frequency
The outera nlayers of the slab guide have the lower refractive index
n.2 = V £,2 d with thin enough guides the energy resides in the outer
layers giving the higher phase velocity. As the optical thickness
increases, V increases and the energy is stored in the central part of the
guide. At high enough V, the phase velocity is c/n} where nx= Ve r] but
at such high Vnumbers the guide is overmoded

&<*= -402 (A3.9)

cod
=—=c/V{erl - &r(cos (A3.10)
J1
(3d
From the equations, observe that when cos 0 - 0 then vp~c/v(srl).
Physically, this arises because the optical width of the guide is so large
that all the stored optical energy is, effectively, in the central region.
However, when cos @ ~ 1 where the optical width is small then all the
energy is effectively in the cladding and so vp~c/v(er2)- The
fundamental mode only extends usefully up to lvalues of around IT
when the first higher-order mode starts to propagate so that @ is only
considered up to values around TT/4 rather than the full range. Figure
A3.1 plots the phase velocity as a function of the normalised frequency
V for materials given in Table A3.1; these are typical materials for
heterostructure DFB lasers. Typical V numbers for d=0.1 to 0.2 jxm-
thick guiding layers at wavelengths around 1.55 fxm are in the range
0.6-1.3.
 Appendix 3 327

To find the group velocity v^ differentiate /3Z with respect to co to

obtain

c ^ = Si = f_+YP & r O)@'COS 0 sin 6+0.5 ^ ; f^, -

da> vg Vp c c
(A3.ll)
f
where 0 ' = d0/dw; s rl=dsrl/d(o; 8s'r=d(8sr)/d(o and, from eqn. A3.8,
calculate da>/d0 to obtain

(A3.12)
+ 0 tan©

The equation for the reciprocal of the group velocity may be

summarised as

+fc,,1M(.), m.,.^ (A3.i3)

vg vp c 1 + 0 tan 0 c
where

A(a,) =0.5 °p; B(co) =0.5 ^ - ^ (cos

With no dependence of the material permittivities on frequency,

A(co) = B(oj) =0. Under these conditions, one notices from the algebra
that the group velocity equals the phase velocity at the end points
where cos 0 = 1 or 0, again consistent with the concept that all the
energy effectively resides in the cladding layer or the central layer,
respectively. Then using 0 as a parameter one can plot the phase and
group indices as a function of the normalised frequency parameter V,
with and without material effects, by the inclusion or omission of A(co)
and B(co).
The results are best considered for specific materials and cases.
Table A3.1 gives refractive indices n and their rates of change with free-
space wavelength A at wavelengths around 1.55 (xm for two typical laser
materials. The connection between dn/dA and efr=dsr/dw is given
from n = V# r with a)K = 2 TTC yielding a)g'r= — 2n A dn/dA. The reader
will find it instructive to use a MATLAB program called group in
directory slab to explore these effects.
328 Distributed feedback semiconductor lasers

approxfrhate
4.4 useful range - " " 'group index:
guide + material
4
index
value
3.6 • group index: guide only
- V-V-----
3.2 phase index

0 1 2 3 4
normalised frequency, V
Figure A3.2 Effective group and phase indices against normalised frequency
Group and phase indices as appropriate for a guide formed from a
cladding layer InP with a refractive index of 3.168 and a core layer of
In^Ga^ASyP^, with a refractive index 3.531. For a core-layer thickness of
0.2 |xm, one finds V~ 1.25 for A ~ 1.55 |xm

Table A3.1 Refractive index/permittivity and differentials

Material Refractive Relative dn/dA

index n permittivity er at A =1.55 atA=1.55|xm

InP 3.168 10.0362 - 9.8064 x10 4 0.963067

ln1.xGaxAsyP1_y;y=0.9 3.531 12.4680 - 7.4322 x10 5 8.13536

A3.1 References
1 HAUSS, H.A.: 'Waves and fields in optoelectronics' (Prentice Hall,
Englewood Cliffs, 1984), p. 175
2 KOGELNIK, H., and RAMASWAMY, V.: 'Scaling rules for thin-fiber optical
waveguides', Appl. Opt, 1974, 8, p. 1857
3 COLDREN, L.A., and CORZEINE, S.W.: 'Diode lasers and photonic
integrated circuits' (Wiley, 1995)
4 ANDERSON, W.W.: 'Mode confinement and gain injunction lasers', IEEE
J. Quantum Electron., 1965, 6, pp. 228-236
 Appendix 4
Small-signal analysis
of single-mode laser

This appendix provides a more detailed account of the classic rate-

equation analyses of appropriately uniform lasers and shows the
first-order effects of carrier transport into the active quantum-well
region, and the effects of spontaneous emission and gain saturation on
damping of the photon-electron resonance. The appendix ends with
large-signal rate equations showing the influence of four key parame-
ters which shape the main features of the large-signal response.

A4.1 Rate equations: steady-state and small-signal

This appendix evaluates the small-signal responses of a laser diode to
modulation of the drive current more comprehensively than in
Chapter 4. The photonic equation (eqn. 4.1) is now written as

The effective photon lifetime r'p has been used as in Chapter 4. The
intervalence-band-absorption term bNS is added here for complete-
ness and the spontaneous emission is recognised as being formed from
bipolar recombination so that, when the holes and electrons are of
equal density, the radiative spontaneous emission is proportional to
N2. The gain is linearised about the transparency density Ntr but gain
saturation with optical density is included. The spontaneous-emission
coupling /3sp is now assumed to have absorbed the confinement factor.
The electronic eqn. 4.2 is also written more fully as
330 Distributed feedback semiconductor lasers

=^- AN- BN* - CN*- G'm(N- Ntr) ~ (A4.2)

at qa 1 + 6,3
where the effective recombination rate N/rr in eqn. 4.2 has been
expanded to become
(A4.3)
The first (linear) term is often identified with nonradiative recombina-
tions and the final (cubic) term identified with Auger (nonradiative)
recombinations. The current density into the active area is / and the
active region thickness is d (J/qd= I/qW).
As is standard for a small-signal approach, the terms split into steady-
state and dynamic terms varying at the modulation angular frequency
coM:
S = S0+Sl exp(jcoMt)
N=N0^Nl exp(ja)Mt) (A4.4)
Mo+Ji exp(j(oMt)
Substituting eqns. A4.4 into eqn. A4.1 and selecting the steady-state
term,

(A4.5)

If p'spoM BNl—*Q but nevertheless So remains significant, then the

electron density iV0 is pinned to what is known as the threshold
electron density Nth:

(Gm- b- beS0)>0 (A4.6)

For small enough gain-saturation terms eSQ, the electron density is

pinned and this is the value required to start the lasing. To see how the
electron density approaches its threshold value, one looks at the
steady-state conditions from eqn. A4.2:

AN0+BN*+ CN30+ G'm(NQ - Nlr) - ^ - (A4.7)

qa l +€o
 Appendix 4 331

With So zero, the electron density increases with the current input
until N=N(h, when

^AN^BNl+CNl (A4.8)
qd
As / continues to increase, the optical output then increases almost
linearly with drive current above threshold, provided that the gain
saturation is negligible:

qdGm(Nth-Ntr)

The dynamic first-order equations are found by retaining only the first-
order terms in eqns. A4.1 and A4.2 using eqns. A4.4. In particular, the
small-signal version of eqn. A4.1 may be rewritten after some
manipulation as

EjiV, (A4.10)
V W
where

—= (#' BNl/ So) + rG'm(N0 - Nlr) -

With negligible spontaneous emission [(/3spN0/So)—»0] and also with

negligible optical gain saturation, the changes in the electron density
and changes in photon density are in quadrature (i.e. the greatest rates
of increase/decrease of photons occur at the peaks and troughs of the
electron-density changes with time). However, when the output is low
enough just around threshold, so that (/3spN0/So) is no longer a
negligible number, the spontaneous emission damps the response of
the light to changes in electron density with a damping time constant
T
di~$o/ (PspBN02). Note that too much intervalence-band absorption
(b too large) kills the possibility of lasing by making Hi negative.
Because, as an order of magnitude /?^~10^ 5 , the spontaneous
332 Distributed feedback semiconductor lasers

emission is neglected in the remaining work which applies to above-

threshold conditions but gain saturation adds the damping rate
2

The modulation changes caused by the changes in the drive current

give

i —m~-i T , _ n —m\~-v ~*ir/ / n . _o \ 2 V/VT:.11J

where

1 9
-=A+2BN0+3CN20 (A4.12)

This can be rearranged to give

fH2Si==~ (A4.13)
T
r/2/

where

and B2=Gm(N0-Nlr)
r

The steady state gives

1
ram(N0~Ntr) rl-=r6iv0+- 7 (A4.14)
7
0 Tp
Rearranging to obtain the change in photon intensity Sx as driven by
the changes in the current density J{,

(-co2M+ja>My+a>l)S^^ (A4.15)

where

T
rf2T./l \T^2 T
d\/

Neglecting the IVBA term b gives the damping and resonance from
 Appendix 4 333

Eqn. A4.15 shows a classic damped resonant response with the overall
damping rate determined principally by the effective recombination
time but with increased damping at high output-power levels and with
gain saturation. The resonance frequency increases as JSQ [or
alternatively as ^(Io — Ilh) when the steady current increases above the
threshold value Ith, assuming that there is negligible gain saturation].
Notice that G'm in eqn. A4.16 is the differential power gain per unit
time so that knowledge of the resonance frequency as a function of
intensity 50 can yield an estimate of this differential gain. The most
important consequence of gain saturation is that it can both lead to a
marked reduction in the resonant frequency cor at high power levels
and also give additional damping of the optical response to changes in
current. Therefore at high enough optical-power levels, gain satura-
tion slows down the dynamic response of a laser to direct modulation
of the current drive.
At this point a little care is needed to note that S} already represents
an optical power so that the — 3 dB bandwidth of the optical-power
modulation for the laser occurs when
\~a>M2+jcoMy+cD2r\=2aj2r (A4.17)
2
Usually a>My<(o ., allowing the approximate — 3 dB bandwidth to be
defined as
*>M-3<*ir(V3)<», (A4.18)
Measurements of the small-signal AM response are often normalised to
the response at low frequency. The magnification at resonance is given
by

Magnification \ramam = - r = ^ Tp
' <- (A4.19)

y
)
which has a maximum when the photon density is given by

1
(A4.20)
334 Distributed feedback semiconductor lasers

Finally, it is worth noting that confusion can arise in interpreting

experimental small-signal AM-response measurements. It is usual to
plot the response in decibels normalised to the steady-state value. This
is the ratio of the power dissipated in a resistive load on the detector
to the power supplied to the laser. However, taking the load-resistor
currents at the modulation frequency coM to be ^(o^), which is
proportional to the optical power ^(a^), one has the optical
frequency response given from 10 log10{S1(cyAf)/S1(0)}dBoc
10 log10{/1(a)M)//1(0)}. However, the electrical frequency response is
given from 20 log10{/1(a>M)//1(0)}dB. Consequently, when measuring
frequency responses, dBelectrical = 2 dBoptical.

A4.2 Carrier-transport effects

The high differential gain in multiple-quantum-well (MQW) lasers led
to an initial expectation that they would offer improved high-speed
performance. In practice, the initial devices were a disappointment
until the mechanism responsible for the additional time constant was
identified as the slow transport of carriers, particularly holes, through
the separate optical-confinement layers into the active MQW region.
This is outlined in Section 2.7.4. A simple but useful model can be
developed which provides a good explanation of the basic effects.
The carriers driven into the laser waveguide are now referred to as
iVw?, rather than N as previously. These carriers diffuse through the
waveguide layer as before, possibly with some field assistance. However,
there is an important difference from bulk material in that, before the
electrons can reach the quantum wells, which are the real active
regions, they have also to diffuse through the separate confinement-
he terojunction layers of thickness dSCH, as in Figure All.l. This
characteristic time to be captured by the quantum wells is taken to be
a constant rmp and is a composite time constant taking into account
both the diffusion time across the distance dSCH and actual capture
time, which may be only a picosecond. Diffusion is often the dominant
factor. The thickness dSCH provides then a limit to how low rcap can be
made. The carriers within the quantum wells have a density Nqw but
these can then escape through their thermal energy overcoming the
barriers and diffuse back into the waveguide layer with another
characteristic time constant r m . The separate confinement region has
the role of helping these carriers to be recaptured by the quantum
wells, and this is another reason for keeping dSCH low. The carrier rate
equations describing this effect for the waveguide and active regions
 Appendix 4 335

are

(A4.21)
(Xt mAs.ni * e*r * cap

The ratio r-dqjdscH<l gives a measure of the concentration of the

charge carriers as they fall into the confinement of the quantum wells
which have an effective overall thickness dqw. The equivalent small-
signal equations can then be written as

jcoM+— jV = ^ L _ + _ V r (A4 .22)

Eliminating JVll%rand retaining only first-order co terms on the left-hand

side yields an equation which is almost identical to the previous
equation for Nx but with the principal addition that there is a further
delay

UN0<po-N,)S\-- '
qdqw(l+jo)MTmp)

(A4.24)

The photon density is now driven by Nqw, whereas previously it had

been driven by iV:

^-^^ANqu-BNlv~- l {
CNsqw- Gm(NHqw-Ntl)S (A4.25)
dt r 'hap
r r, *

The results, then, of the previous small-signal analysis are substantially

the same except for the new feature of the extra delay which is created
by the time constant rcap which summarises the effects of both diffusion
and capture along with some numerical differences created by the
factor K The escape time does not figure so prominently as the
capture time because it is has been assumed that electrons that escape
from the wells back into the guides can be recaptured by the wells.
The conclusion of this simplified analysis is that, provided that the
capture time for electrons to enter the quantum wells can be made
short enough so that cor rmp<\, it is the photon-electron resonance o>r
336 Distributed feedback semiconductor lasers

which limits the amplitude response rather than the capture rate of
charge carriers into the quantum wells. On the other hand, if it takes
too long for the wells to capture the charge carriers, there is a
significant first-order time constant which can limit the response well
before the photon-electron resonance plays a role and this pheno-
menon exhibits itself principally by an additional roll-off in the small-
signal AM response.

A4.3 Small-signal FM response of single-mode laser

In principle, rate equations for the balance between photons and
charge carriers contain no information about the frequency of
operation of the laser because they contain no phase information.
However, the photon lifetime is usually in the picosecond range while
the time scale for the change of drive current is measured in
nanoseconds or hundreds of picoseconds, and it is therefore not
surprising to find that the laser's frequency settles sufficiently rapidly
with time so that, for uniform lasers where the frequency is
determined by the optical length of, say, the Bragg pitch or Fabry-
Perot length, the changes in frequency follow the average refractive
index as given in eqn. 4.15 with Ao being the value of the free-space
wavelength:
/~ -An,./n~(A 0 /2TT) (aH&g/nt) (A4.26)
However, the differential optical-power-gain/unit-time Gm used in the
rate-equation analysis above, is linked to the differential field gain per
unit distance from:
Gm=2vgdg/dN (A4.27)
Putting nrvg~c gives the small-signal change in frequency with electron
density from
Gm/4w (A4.28)
The key to finding how the frequency/wavelength changes with drive
is therefore linked to finding Nx as a function of the drive current Jv
In the absence of any effects of charge transport, one has to rearrange
eqns. A4.10 and A4.13:

(A4.29)
 Appendix 4 337

The FM response can therefore be seen to follow closely that of the AM

response with the same resonance and damping, but the FM rolls off
as l/coM as coM increases above o>r, whereas the AM response rolls off as
1/(0% as coM increases above cor.

A4.4 Small-signal FM response and carrier transport

The carrier-transport mechanism clearly affects the current injected

into the active region, but in addition the presence of stored charge in
the waveguide layer alters the local refractive index. This in turn
changes the effective refractive index of the guided mode and leads to
wavelength/frequency chirp for the lasing modes. These changes of
refractive index are approximated as follows:

dn dn
r (A4.30)
diV wg
'""dN

The detailed results are straightforward if 'messy' and shed little light
on the physics other than to once again indicate that the key effect is
to add an additional delay so that the form of the FM response has the
general form

N(/
Wl (A4.31)
(J\ / qdqj (1 +J(OMTMP) ( - (o2Ma+ja)My+ o)2r)

where a, y and cor2 are similar to the parameters that previously would
have been found for the AM response and £ is again proportional to
j(oM as a)M increases. The important point is that transport adds an
additional roll-off term for the FM response just as it does for the AM
response.
Figure A4.1 summarises, in its upper sketch, the schematic features
of AM responses as found from the classic electron-photon rate
equations using logarithmic scales. The damping factor has been
chosen arbitrarily to give a peak magnification of 3 dB. It may be seen
that the transport time constant needs to be such that a>r rcnp< 1 for
transport effects to have negligible effects on the frequency response.
The lower sketch in Figure A4.1 shows the FM response. If there were
no nonlinear gain, the FM response would tend to zero as the
frequency fell. Nonlinear gain flattens the response. Note that the
338 Distributed feedback semiconductor lasers

10 r
relative output dB
0
3 dB
-10

-20
AM Response
-30

-40
-2 -1 coxyr/co 0
M r
1
relative output dB
10

-30
-1 co^/co

Figure A4.1 Schematic amplitude response 10 log10 (P()f)lI/Jdrive t) of laser-output

intensity to current modulation at frequency a)M and frequency response
10 logw (kf/fdnvel) withfdriwl appropriately normalised
Values are normalised to those at low frequency. Parameters chosen to
illustrate the schematic features of resonance and fall off of responses
above resonance.

resonance frequency o)r in both the AM and FM responses is the same

and is proportional to V ( / o - ~ 4 ) . This is the same resonance as is
found in the relative-intensity noise (Appendix 8). [1] gives a detailed
discussion of the effects of carriers on the FM response.
In practical devices, the FM response below a few megahertz is
significantly affected by the change in refractive index with tem-
perature. If the current increases slowly, there are increases in the
temperature which reduce the lasing frequency in opposition to the
linewidth broadening or plasma effect, which increases the frequency
with carrier density (and hence with current).
 Appendix 4 339

A4.5 Photonic and electronic equations for large-signal

analysis
This section looks at grouping and normalising variables for large-
signal analyses. The equations are as before except that now the
spontaneous emission is assumed to have a single time constant. Gain
saturation is still included. The pair of rate equations is

(A4.32)

]
dt T\dt Tp/ rr qW

The nominal threshold electron density is then given from

^=A^+F7^7 (A4.34)

and this, along with the effective photon lifetime, is used to normalise
parameters:

a^ «3 xy
r.4 eS^ (A 4.35)
'p th 'p >C

Nn=^- Ln=i^r D=-~i^ (A4.36)

Ln is the normalised light output, iVn is the normalised electron

density and D is the normalised drive with Tn the normalised time and
es the normalised gain-compression factor so that:

d^=(A^)^__
A J aT H
dTL \ —a -

dLn
7+Ln )-NH(l- frp) +D (A4.38)
^dTn \dT,

In the steady state where d/dTn—*0,

340 Distributed feedback semiconductor lasers

Nn={Ln+es(l ~ a)L 2 J/{L,,+ ( l - a) (l + eJLJpJ Ln=D-Nn(l - /3sfi)

(A4.39)
Change the variables to M=Nn+ {LJ Q) and Ln:

—a '

1
—=-[-Ln-(M-(Ln/0}(l-^)+Z)] (A4.41)
U 1n £y

Note how there are then four normalised parameters ($sp, a, Q and es
which determine the normalised light/time characteristics for a given
normalised drive D. Note also that, in general, the light changes at a
much faster rate than the electrons because of the factor l/Qin the
normalised rate equation for the electrons. The electrons vary on a
time scale governed mainly by the electron-recombination time, which
is longer by a factor of Q (typically ~ 1000) than the photon lifetime
which governs the rate of change of the photons. This is an important
point in modelling because it is the photon lifetime which determines
the temporal step that is required.
There are two MATLAB programs, fpstat and fpdyna, in directory
fabpero. These allow exploration of the qualitative effects of sponta-
neous-emission parameters and gain saturation.
In summary, the rate-equation approach has been extremely
successful in dealing with several major characteristics of Fabry—Perot
lasers where the fields are sufficiently uniform and phase effects of
fields are not important. However, considerable care is required in
applying the results from such equations to distributed-feedback lasers
where phasing, nonuniform fields, and the distribution of the
feedback change the laser's behaviour in important ways.

A4.6 Reference
1 VANKWIKELBERGE, P., BUYTAERT, E, FRANCHOIS, A., BAETS R.,
KUINDERSMA, PL, and FREDRIKSZ, C.W.: 'Analysis of the carrier
induced FM response of DFB lasers', IEEE J. Quantum Electron., 1989, 25,
pp. 2239-2254
 Appendix 5
Electromagnetic energy exchange

There are three aims of this appendix:

(i) to give a reconciliation of the classical electromagnetic-field
exchange of energy and the quantum-particle exchange of
energy;
(ii) to show how the rate equations derived from the particle balance
are consistent with Maxwell's equations; and
(iii)to show more formally why the group velocity appears in the
travelling-field equations used in Chapter 3.

A5.1 Dielectric polarisation and energy exchange

A physical model to have in mind for 'classical' laser interactions
between electrons and optical fields is that displaced charges can be
treated as dipoles. Figure A5.1 shows the concept of the equilibrium
charge and the same charge which has been displaced by an amount
xd because of the interaction with an electric field forcing the charge
to move. Each displaced charge is then equivalent to a dipole being

p =

s~\ equilibrium
\Q } charge

Figure A5.1 Displaced charge

Equivalent to a dipole and equilibrium charge
342 Distributed feedback semiconductor lasers

superimposed on the equilibrium charge. With a density N of charge

carriers, there is then a polarisation P=qNxd, where there are two
components to the displacement. The first component is dependent
on the field strength £ in a complex way which depends on the optical
frequency and the details of the optical interaction, while the second
component represents a random dipole fluctuation:

The parameter x is called the susceptibility with l+;f=e r giving the

relative permittivity. In electro-optic material, it is more correctly
considered as a tensor so that the polarisation is not necessarily in
exactly the same direction as the applied field, but in this work it is
adequate to treat ^ as a scalar. The random polarisation Pspont gives rise
to spontaneous polarisation currents through a net polarisation
current given from
J=dP/dt (A5.2)
Now the helpfulness of this model comes in forming a heuristic link
between the classical and quantum formalisms. In quantum model-
ling, one considers the electronic dipole moments p as being set up
when an electron changes its wave function from an energy state %x to
%2 where the associated pair of wave functions do not have the same
centre of symmetry. This quantum dipole oscillates at an angular
frequency <o0=(ol2=(%1 — (S2)/fi. The quantum interaction is that of a
photon exchanging energy fi^^lj-^ with the two states. The
corresponding classical picture is that of an electric field with angular
frequency &>0=a>12 interacting with the electronic dipole current at the
same angular frequency o>0. In both classical and quantum formalisms,
the strongest interaction occurs only when the frequency associated
with the oscillating dipole and the frequency associated with the field
are equal. The susceptibility x describes stimulated interactions as a
function of frequency, while any increase in energy caused by the
random spontaneous-emission energy is proportional to the square of
1 (random polarisation currents)! and so is always positive. As usual in
work here, all material is nonmagnetic: /mr=l.
The local instantaneous rate of energy-density exchange from the
electric field into the polarisation currents takes place (see eqn. A5.13)
through the local scalar product (EJ):
dU/dt=-(E.J) (A5.3)
Using ideas from classical electrical-circuit analysis, if one uses a
complex phasor field, the complex rate of energy density exchange
 Appendix 5 343

gives
(A5.4)
The Poynting vector 2?= (Ex H) gives the instantaneous power per unit
area and, to give the power from the laser, must be integrated over the
xy plane of the appropriate laser facet of effective area si:

3>zs4=l I (ExH)zdxdy (A5.5)

To see how these results can be reconciled with the particle model of
Section 4.1, one must recognise that the electrical polarisation has
three distinct components:
(i) A real susceptibility ^ r with P=xrs0E
In this first component, the dipole current {dP(t)/dt} is in quadrature
with the electric field and from eqn. A5.4 there is no mean real power
exchange. There is reactive power where the power flows back and
forth between the electric and magnetic fields, and so the effects of Xr
are reflected in the stored electric energy within the material which at
any instant is given from
U=UeoenE.E)={k^XrE.E)+UeoE.E)} (A5.6)
using the fact that srr= l+xr-
There is an important theorem in electromagnetism which states
that, at resonance or in the normal mode of a guide, the mean electric-
energy density is identical to the mean magnetic-energy density [1, 2].
The laser is both resonant and uses a guide for the optical fields so
that, at the lasing frequencies, there is an equivalence of the mean
values of magnetic and electric stored energies where, using the
phasor notation:
U=}(sosn.E.E*) = Um=j([jL0H.H*) (A5.7)
Here the fields refer to the peak values and the additional half has
appeared because this is the mean of the instantaneous energy which
peaks every half cycle.
(ii) An imaginary susceptibility Xt w ith P=jXi ®oE
This second component gives [dP(t) / dt\ directly in phase or in
antiphase with the fields and results in a direct exchange of power.
Here sign conventions are important. If one assumes, as here, a time
344 Distributed feedback semiconductor lasers

variation exp(ja)ot) at an optical (angular) frequency <o09 then the

imaginary part of x (=jXi) gi y e s a polarisation currenty=~-a>0X; e0 E>
giving a mean rate of energy input from the electrons into the field
determined from

Now E. £* is proportional to the total mean electromagnetic energy in

the guide which, in turn, is proportional to the photon density in the
material. Consequently, x* determines the stimulated emission (x*>0) or
absorption (Xi<0). In a laser, \i l$ strongly dependent on the electron
density. At the transparency density Nlrof the electrons, there is no net
absorption of emission so that XiWr) = 0- Note that, for these imaginary
components,
Xi=en (A5.8)
The connection between optical gain and Xi will be made shortly.
(iii) [dP{t)/dt]sponl=Jspml
The third component of energy exchange between the electrons and
the electromagnetic field is the spontaneous emission. This quantum
phenomenon has to be represented in a quasiclassical analysis by a
random dipole current [dP{t) / di\spon(-Jspont which is calculated in such
a way as to give the correct positive random emission of photons. The
details of estimating this value are given in Appendix 9.
These three components of the polarisation current are put into
eqns. Al.l and A1.2, but now treating the fields and currents as
functions of time with frequencies close to co0 rather than phasors,
curl E= - fiodH/dt (A5.9)
curl H= ~~ a)0sri80E+ens0dE/dt+Jsponl (A5.10)
The ground work has now been laid for finding rate equations from
Maxwell's equations as follows next.

A5.2 Electromagnetic-energy exchange and rate equations

reconciled
This section will develop rate equations corresponding to the particle
interactions given in Chapter 4, but arriving at these through the
classical electromagnetic-energy exchange. For the moment, it must be
taken on trust that it is possible to design the material structures, such
as the different slab layers, shown schematically in Figure A5.2, that are
 Appendix 5 345

Figure A5.2 Opticalfieldsand real and imaginary permittivities

Schematic profiles in a vertical direction

required in order to guide the optical fields to interact with the

recombining carriers in the active region (see Chapter 4 and
Appendix 2). Given such a guiding structure, even without knowing its
details, it is possible to consider the energy exchange between the
optical fields and the polarisation currents. The stored electro-
magnetic-energy density at any instant of time is given from
U= 2(sosrrE.E+ /UL0H.H) with the local rate of change of energy density
given from

(A5.ll)

Using the vector identity for the divergence of the Poynting vector,
which gives the electromagnetic-power flux,

div (ExH)=H.curl E- £.curl H (A5.12)

On substituting for 3H/dt and dE/dt from eqns. A5.9 and A5.10, the
instantaneous rate of energy-density exchange is given from

dU/dt=E. (curl H+(o0sris0E-Jsponl) - //.curl E

= - div(£xH) + oj0sris0 E.E- EJspont (A5.13)

Next integrate this rate of change of energy over the cross-section of

the laser so that, in the divergence, the terms 3/ dx and d/ dy integrate
over the cross-section of the laser to leave (ExH)x and (ExH)y which
are zero at the sides of the laser guide, leading to the rate of
longitudinal energy exchange:
346 Distributed feedback semiconductor lasers

(ExH)zdxdy

(D0sri80E.Edxdy\ +\ EJspmldxdy\=O (A5.14)

This gives the instantaneous energy exchange at any time.

It is possible to associate eqn. A5.14 with either the forward fields or
the reverse fields. It is also more useful to consider the mean value of
eqn. A5.14 over approximately one period of oscillation, when one
may use the equivalence of mean electric and magnetic energy over
the guide cross-section to write, for the forward field,

f f ff
I I Ufdxdy= I I sn.SQEj.Ejdxdy=h.(x)QSjSi (A5.15)

where si is the effective area containing a density Sf of photons with

energy ho>0 per photon travelling in the forward direction. The z
component of the Poynting vector 2fy gives the temporal mean
forward-power flow &/-M from

(EfxHf)zdxdy=fi(o0Sfs&vg (A5.16)
• / /

where vg is the group velocity. Here the important new point has been
made which differs from the rate equations of Section 4.1, namely that
power flow and energy stored can both be associated with the forward
(and also the reverse) directions along the guide. Eqn. A5.14 may be
recast to give the mean forward-power density and associated energy-
density exchange as in eqn. 4.23 from

1 8Sr dSr
~~+—- - 2g(iV) Sf > = (fico0si) sspf (A5.17)

The factor of 2 for 2 g(N) is again here because g(N) is used as the
field gain per unit distance as function of electron density //and 2g(N)
refers to the power gain per unit distance. Here this net optical power
 Appendix 5 347

gain is given from the temporal mean value of

co0sris0Ef.Efdxdy (A5.18)
/ /

The spontaneous emission combined with its coupling factor /3sp into
the forward-propagating field is determined from the mean temporal
value of

(tico0dvg)sspJ=-[ Ef.Jspmldxdy\ (A5.19)

It may be thought, because the phase of E^ndJspont must be random,

that this spontaneous emission could take any sign. However, it turns
out that any single spontaneous emission excitation Jsponl dx dy is
incoherent with all fields except those fields which are excited by this
particular element itself. On averaging over time and space, the net
energy gains can only be terms proportional to \Jsponl\2 and the sign of
this is such that spontaneous-emission power is always given from the
material into the fields.
For energy propagating in the — Oz direction, the above arguments
are mirrored and there is, similarly, the equation (eqn. 4.25):

=V
Equations A5.17 and A5.20 combine to give S= (Sj+Sr). The averaging
over one cycle, as in eqn. A5.15, is kept in future equations but not
shown explicitly. Integrating over the laser's length with Y = s&L gives:

U dx dy dz=

SIS e77£0E.Edxdydz=ha)0SV (A5.21)

The power flowing out at the 'right' and 'left' facet ends gives the rate
of escaping photons from
348 Distributed feedback semiconductor lasers

(ExH)zdxdy\

f/J (ExH)zdxdyY (A5.22)

The mean rate of net stimulated emission over the whole volume is
given from

o0sris0E.Edxdydz (A5.23)

Note that, because the calculation is concerned with the fields, it has
by default included the confinement factor and any loss which was put
in separately in Section 4.1. The net output of the radiative
spontaneous emission (N/TSP) to the lasing mode taking into account
both directions of propagation of energy is given from

(A5.24)

Combining both forward- and reverse-power-flow equations permits

one to see the principles of assigning quantities from the electro-
magnetic-energy balance to give the equivalent 'photon'-particle rate
equation as in eqn. 3.1 but taking the net gain and value of j3sp to
include the appropriate confinement factors which do not now appear
explicitly:

^ j > ) (A5.25)

A5.3 Electromagnetic-energy exchange and guided waves:

field equations
Here the discussion concentrates on a TE index-guided field to
demonstrate key points about energy exchange in waveguides and to
develop the travelling-wave equations for forward and reverse waves in
a guide to demonstrate why the group velocity appears.
 Appendix 5 349

Consider a TE field with Ex and

dz
as the predominant components. The magnetic-field components may
be eliminated from Maxwell's equations to obtain the wave equation
for Ex, retaining a frequency-dependent real permittivity for the
material but ignoring the frequency dependence of the much smaller
imaginary permittivity:

djsponlx
y) }8
* °
(A5.26)
A single-mode solution to this equation is solved for steady-state
amplitudes with negligible loss and gain at one particular frequency o>
and a corresponding value of /3. These values of a) and yS are the same
over the whole cross-section of the guide, so that
Ex=Ef(z, t) u(x,y) where Ef{z, t)=Eoexp (](a)t-/3z)}
For this steady-state solution Eo cancels with

P22{uL0£tr(x, yy a))so}u=O (A5.27)

After solving this equation for the real Vertical' variation u(x, y) with
an index-guided wave (as, for example, in Chapter 3) relate co and /?
giving an effective permittivity srreg-:
(A5.28)
Now suppose that the mode no longer has a steady-state amplitude but
that
Ef (z, t) =F(z, t) exp{j(cobt- ftz)} (A5.29)
where F(z, t) varies slowly in space and time with respect to
exp{j(cob t— f3hz)}. In using phasor notation, one can 'equate' co with
(—jd/dt) but now one needs to separate the low frequencies, varying
with rates noted by (d/dt), from the optical frequency ca/r This is done
by writing <0—>a)h—j(d/dt) and a)2—>cob2 — 2 cobj {d/dt) to a first order
in the slow changes denoted by {d/dt). It follows from Taylor's
theorem that the frequency-dependent permittivity becomes a tempo-
rally changing permittivity
350 Distributed feedback semiconductor lasers

er(co) ->er(a>b) + e ; K ) x ( -jd/dt) (A5.30)

where e'r(a)b)=der((o) /diol^.
Similarly, /3—+/3h+j(d/dz) where now (d/dz) refers to the slow
variations with space with fi2—+$b+2Pbj{d/dz) to a first order in
(d/ dz). Then eqn. A5.28 is generalised to a first order in (d/dt) to
become

r(x, y9 (ob)u[ Q>lF-j2a>b— I - ja)lfiososfr(x, y, o)h)u-—=j(ohfi0jsponl

y ot j ot'
(A5.31)
where sr=err+j eri with e.ri (^e^) giving the gain which determines the
growth. The growth is usually sufficiently small that one may, to a first
order, neglect e n \ Note that one has multiplied the spontaneous
emission by a factor exp{j((&bt—$bz)} but this does not alter its
magnitude nor alter its random properties, and apart from a phase
change there is no significant effect so it is not necessary to make this
phase factor explicit.
The steady-state terms must cancel giving:
pl=a>l{^sneff{a>()eQ} (A5.32)
The group velocity is found from eqn. A5.32 using the expression
vg=da>/dP:

T W {f*0Sn,ff(<»b) fiol+^ IT ^neff(^b)^o) (A5.33)

JJ
vg do) pb pb
Removing the steady-state terms in eqn. A5.31 and with a little
rearranging, such as cancelling the j operator, gives

uF= - I — I fioj (A5.34)

\*PbJ
 Appendix 5 351

The result is more useful if one uses the effective modal area:

•J J *"•• ,y)2dxdy (A5.35)

where u is real and one also writes

y)
u(x,yydxdy=gsl/pb (A5.36)
J J 2e^

Sr )vjspont(x, y)u(x, y) dxdy=ispfd (A5.37)

J
b)

where ispj includes the spontaneous-coupling coefficient, i.e. only

those spontaneous interactions feeding into the guide at the right
frequency. Then, with eqn. A5.33, the equation for the forward field
with gain becomes ( eqn. 4.22 with 5=0)

Similarly, for the reverse field (eqn 4.24),

dR 1 (dR\
(A5.39)
vg

A5.4 References
1 COLON, R.E.: 'Field theory of guided waves' (McGraw Hill, 1960)
2 SLATER, J.C.: 'Microwave electronics' (Van Nostrand, 1950)
 Appendix 6
Pauli equations

This appendix gives the detailed calculations for finding the steady-
state fields in a uniform DFB laser with uniform gain, and thereby
finding the threshold conditions. The results are essential if one wishes
to make comparisons with the numerical algorithms to estimate the
accuracy of these algorithms. The appendix also provides a tutorial on
the use of Pauli matrices for coupled differential equations which
enables one to extend the concept of exp(yz) to exp(Mz) where Mis
a matrix: a useful concept for linear differential equations.
Take the coupled-wave equations

1 r) /? r) /?

*ft(>'"t ( A 6
-2)

Pauli matrices are widely used in the quantum theory of spin and
polarisation of light and the way in which the fields are coupled, but
they can also be used to provide neat solutions for other physical
problems (see, for example, [1]) and in particular solutions for
coupled linear equations. Define then the vector F and the Pauli
matrices

cr9 =

to write eqns. A6.1 and A6.2 as

 Appendix 6 353

+a
^Jt "Tz~JK(riF~ ^S-jS)F=O (A6.4)

If, for example, there were no gain, phase or space variations, then

- — ~JKaxF=0 (A6.5)
vgdt

The reader who has not met the concept of matrix exponentials will
rapidly appreciate that one may, using a Taylor series, formally write
exp(jatf)=l+/irtf- (1/2!) ( ^ ) 2 - 7 ' ( l / 3 ! ) (criy)*+> • • (A6.6)
Using crf=l (more strictly this is the identity matrix), one recovers the
identity
expiry) =cos(y) +jax sin(y) (A6.7)
A solution then for eqn. A6.5 using Pauli matrices yields
F( t) = cos (Kvgt) +jcrx sin (Kvgt) (A6.8)
The object now is to integrate in space with a constant-frequency input
with the offset frequency determined by 5 so that we rearrange eqn.
A6.4 by multiplying through by cr3 to obtain

dF 1 dF
—+or 3 -a3(g~j8)F+Kcr2F=0 (A6.9)
dz Vg dt

Neglecting the time variation for the moment, the formal solution of
this equation is
F(z) =exp{ -j(o-3D- a2JK)z}F(0) (A6.10)
where, for a shorthand in this appendix only, D=8+j g.
However, because cr3 a2=~~ a2 a^ it may be shown by using Taylor
series as in eqn. A6.6 that exp{ — j(a3 D— cr2 JK)Z} cannot be the same
as expK-703 D)z] exp{(- CT2K)Z). The problem, stated briefly, is that
the matrices <r3 and a2 do not commute and so the matrices formed
from exp{( —703 D) z] and exp{( — <T2K) Z) cannot commute. One has to
consider
(o"$D~ CF2JK) (cr3D— CT2JK) = (D2 — K2) — (asa2+a2as)jDK= (D2 — K2)
(A6.ll)
showing that
- a2JK)2n=(D2-K2) n
(A6.12)
354 Distributed feedback semiconductor lasers

(o-sD- a2JK)2n+l = (D2- K2)n(cr3D- CT2JK) (A6.13)

and hence, using the power-series expansions for cos and sin,
exp{ -j(crsD-~ cr2JK)z}=cos(fiez) —j{(cr3D— (T2JK) //3J sin(/3,z) (A6.14)
where

Apply eqn. A6.14 to find the solutions for eqn. A6.9 for a length z:

S 3
F(z) =cos(&z)F(0) + ^ ^ {(g-j8)crs-K(T2}F(0) (A6.15)

where /32L2={(S+jg)2 — K2}L2 and S gives the 'frequency offset'.

Expanding out in full,

F(z) a+ d jb F(0)
(A6.16)
R(z) —jb a—d R(0)

where a=cos(/3l,z), d={(g—j8)//3} sin(/S,z) and b=(K/fif) sin(^z).

Observe that a2-d2-b2=l so that

Ja+d * ] =1

This then permits eqn. A6.16 to be arranged more usefully so that the
two outputs are in terms of the two inputs:

F(L) | J _ I J b ] \ F(0) (A6.17)

jb 1 J LR(L)
R(0) J a-d\_jb
It is helpful to consider a unit input, ^(0) = 1 with R(L) =0, and then
the reflection is given from R(0) =jb/ (a— d) while the transmission is
F(L) = l/(a— d). With no gain or loss, dis always imaginary while band
a are always real so that U?(0) 12+\F{L) 12= LF(O) 12.
Figure A6.1 then plots solutions using this approach. It is important
to check the basic algebra of the calculation. One way is to look at the
 Appendix 6 355

KL=2
KL=2

-5 +5 -5 5L +5

Figure A6.1 Power reflection and transmission for uniform DFB

Note that, when the structure is lossless and has no gain, I TI 2 + IJRI2 = 1.
Note also that the DFB acts as a stopband filter on transmission but the
overall gain has peaks at the edges of this stopband. Eventually, as the
gain increases, so the peak reflection and peak transmission tend to
almost the same value. The classical oscillation condition is given when
the reflection and transmission power gain become infinite so that zero
power can excite a finite output

power reflection and the power transmission through a uniform

section of DFB where (KL) =2. This is shown for three values of gain as
the frequency deviation from the central structure, as measured by SL,
varies. Typically the range of the stop-band has SL in excess of KL. The
oscillation condition would give infinite reflection for a finite input.
With the uniform distributed-feedback laser of length L, there is no
temporal growth rate for the inputs or outputs in the steady state. In
the situation where the spontaneous emission is negligible, one has to
have an output for zero input which means that (a— d)=0:

cos(PeL)={(g-j8)/P,) si (A6.18)
One can explore this threshold equation, for example with program
dfbthr in directory grating from:

error = l (1/ftL) tan(0,L) -l/[(g-j8)L)} I (A6.19)

At threshold with zero power output, the total intensity along the laser
can be shown as being proportional to \a(z) — d(z) 12+ \b(z) 12;
however, this is not of much interest at operating power levels because
spatial-hole burning, with variations in gain and photon density, play
key roles. The field profiles given in Chapter 6 or from computation
using the techniques of Chapter 7 are then more relevant.
356 Distributed feedback semiconductor lasers

0.2
error KL = 2 1
/ / \

0
[/gL =1.8 N
\r \

-2 -1 mode number +1 +2
+5
Figure A 6,2 Error
Error is zero where oscillation modes appear. This is shown for KL=2
and one can see here four modes. Two modes with gain gL~ 1 with ±1
modes while, if the gain is increased to gL~ 1.8, these modes closest to
the stopband grow rapidly with time and the ±2 modes further away
from the stopband just reach threshold

A6.1 Reference
1 GOLDSTEIN, H.: 'Classical mechanics' (Addison-Wesley, Reading, 1980),
2nd ed., p. 156 and exercises on pp. 185-186
 Appendix 7
Kramers-Krdnig relationships

A7.1 Causality
The Kramers-Kronig relationships provide fundamental rules for the
relationships between the real and imaginary parts of the spectrum of
any real physically realisable quantity when expressed as a complex
function of frequency. One cannot, for example, design the optical
gain spectrum to be any desired function of frequency without
discovering that the phase spectrum is then closely prescribed.
Frequently, such connections appear to be abstract and mathemat-
ically based. This appendix looks at three different ways of discovering
these relationships which should help the reader to understand the
fundamental nature and physics of the Kramers-Kronig relationships.
The appendix includes at the end a collection of approximations for
the real refractive indices of relevant laser materials.
The first way relies on a real 'causal' system where no real output
o{ t) can occur from any system until after a. real input i(t) has occurred.
An elegant way to proceed splits this real output o(t) from such a system
into even and odd parts, as in Figure A7.1, noting that

oft) oo(t)

f\. y\J \/v, /v

Figure A7.1 Splitting output into even and odd parts
An impulse response over positive time = even -t- odd functions over all
time
358 Distributed feedback semiconductor lasers

Now assume that the Fourier transform of o(t) is Or((o) •\-jOi(<a) in the
spectral domain so that the Fourier relationships give:

«.(')= I \Or(
(A7.2)

,(«)= I {o.(t) cos(tot)}dtdt OO i{co)=\ {~oo(t) sin(cot)}dt

i(u))=\

(A7.3)
Using eqn. A7.1 links the real and imaginary parts of the spectrum [1]
from

Or(a>) =
lit 0,-(<w') sin(6>' I/I)}
do/
2TT
cos(wif)
I
df (A7.4)

d(of
0,-(«) = cos(a)ft)} sin(a>UI) d^ (A7.5)
2TT J

The fact that cosine and sine are Hilbert-transform pairs permits one
to show that ( ^ - ^ ( O , ) and Oz=3€ (Or) where it corresponds to
taking the Hilbert transform in frequency [2].
To apply this to optics, consider a short length h of optical medium
with a real gain/loss g((o)h and imaginary phase shift S(f>(a))h as
functions of frequency, and an input I(a)) (plane waves with frequencies
co) with a corresponding output 0(a)) related to I((o) by
0(co) ==/(a)) exp h{g(a)) +j8(f>(co)} or
H I +g((o) h+jS<f>((o) h}I(w) (A7.6)
Splitting into real and imaginary parts:
Or(co)={l+g(w)h}I(a)) and Of.(a>) = &|>(<»)A/(a>) (A7.7)
If one allows a delta function or real impulse of field as an input in the
time domain with /(a») = l, there cannot be any real output o(t) for
 Appendix 7 359

t<0. This therefore means that the results of eqns. A7.4 and A7.5 have
to hold again and force an intimate connection between the gain and
phase as a function of frequency. These relationships hold for any
physically realisable outputs, of which an optical material's complex
gain/phase is but one such quantity.

A7.2 Cauchy contours and stability

A second and equivalent form of these relationships is derived from

the requirement that any realisable physical quantity has to be stable
with time so that even optical gain which leads to lasing has to be stable
with time. To explain this further, oscillations and instability occur only
when the physical system with gain is placed into an environment with
adequate feedback . The oscillation or lasing does not just happen on
finding a semiconductor material with gain. Consider then the
complex polarisation of a medium given as x(^) = {Xr(C0) +7X^)1 where
changes in xA60) a n d X?(^) provide the changes in refractive index
and gain, respectively. Now x(^) ls a complex function of frequency
and can be described through its poles and zeros in the complex
frequency plane. It can only have 'stable' poles, i.e. poles in the
complex frequency plane, where co—^co'+ja with a>0. Such poles, in
the frequency domain, lead to a transient response in the time domain
of the form exp(jo)' t) exp( — at) with a > 0 for t> 0. Contrast this with
a material where x(^) has poles with a<0 which are associated with
transients of the form exp(jcof t) exp( + laU). Then almost any
disturbance, no matter how small, will grow without bound and without
any physical feedback] Such material cannot therefore be found.
To discover the required restrictions, apply Cauchy's integral
theorem to the complex susceptibility x(^)> ta ^ing an appropriate
contour, as shown in Figure A7.2, where there are no poles inside the
contour. The contour of radius 5,—* Ogives the contribution — jrrx{(*>)
while the contribution from the large contour goes to zero as R—*
infinity. Hence, in the limit on the real-frequency axis,

dco' = -j7rx{(o) (A7.8)

I-. (o'-a)

The next step is to take the real and imaginary parts so, with the
contribution around the semicircles tending to zero, one has
360 Distributed feedback semiconductor lasers

complex
CO co'+./a
-R, plane

Figure A7.2 Contour for Cauchy

R t -*oo a n d <5C—0

x(a>r)
. da)' = O

XX<o)=-- if
TT - . ( « ' - (o)
d(o' (A7.9)

(A7.10)

or Xr-"^ (Xi) a n ( l ^ = ^ (/Vr) where 3€ , as defined from the

relationships above, corresponds to taking the Hilbert transform in
frequency. Although this appears to be a different form from eqns.
A7.4 and A7.5, it is the same result using different expressions for ways
of finding Hilbert transforms.

A7.3 A proper physical basis builds in causality

Provided that one uses a physical model for a system, then one finds
that the above difficult mathematics can be circumvented and one
always satisfies the Kramers-Kronig relationships. For example, if each
electron were 'bound' to a 'centre' with some damping and restoring
force, then a physical equation of damped harmonic motion of an
electronic dipole (see Appendix 5) is developed:
 Appendix 7 361

d x d^c
7-2 +lyl —+n 2 x= ? £(<)/m* (A7.ll)
Of (Xt

With £(0 varying as Ew exp(jcot), one finds a susceptibility (see

Appendix 5 again) proportional to x(co) /E^ which, for a density of
electrons N9 is of the form
Xb*s(<») =ftjL*«/ (f! 2 +>y- co2) where H ^ = (Nf/eom*)
=ft^.ma{(a2 - OJ?) ->y}/{(O 2 - a/)2+ (yo,)2) (A7.12)
The even and odd relationships between the real and imaginary parts
of the susceptibility appear naturally, and can be shown [3] to satisfy
the Hilbert relationships that are given in eqns. A7.9 and A7.10.
There can be several different components for the susceptibility
with different resonances, losses etc. For example, ifft—»0and y—>0,
appropriate for a free electron plasma with no damping forces, one
finds a negative component of real susceptibility at frequencies co > Ct
giving a negative refractive index. This is the component responsible
for the plasma effect referred to in relation to the linewidth-
broadening factor:
Xfree electron^) = ~ ^plasma/ *? (A7.13)

This is a useful guide to the contribution to the susceptibility given by

the free electrons in the conduction band, where one notes that
increasing the electron or hole density reduces the permittivity or real
refractive index of the material. Some damping or absorption might
be added to simulate the free electron absorption which occurs in
practice, but this has been neglected here.
Suppose the material changes from loss to gain; then the (Kramers-
Kronig) stability of the material requires that one still has (y/fl) > 0 so
that the material gain on inversion would, in this dipole model, have
to come from a 'negative' effective mass (the equivalent of inversion)
giving
2 2 2 2
^ ^ - ^ ) + ( y a i ) i (A7.i4)
Hence, in general, there are at least three main forms of the variation
with frequency around any local frequency: plasma, loss and gain.
These components have to be added appropriately in terms of their
susceptibilities using the Clausius-Masotti (also known as the Lorenz-
Lorentz) relation [4,10], but the broad overall effect locally around
any one frequency is to be able to split the susceptibility into general
forms such as:
362 Distributed feedback semiconductor lasers

-0.02 [((O/Q) - 1] ->0.02

2.5 |

real susc. .0.02 o.O2

x loss (©j
-2.5
real susc.
loss: imag. susc
0.02

Figure A 7.3 Shape of contributions for real susceptibility and imaginary

susceptibility
The same normalising frequency 12 is taken for all components for the
centre of gain or centre of loss and the plasma effect. Parameters
ygnin/fl=0.05 for broader gain and ygain/Q = 0.025 for narrower gain
line; y/ftW/O = 0.45. Note that x ls not shown to any physical scale; the
sketch is simply to show the shape of the contributions with changing
frequency

X=Xgcun(to) +Xjree electronic) (A7.15)

+CliplmJ(Cli - w2) -jcoy}/m2 - a?)** ijcoY) ~ tljlmma/

(A7.16)
It is important not to expect eqn. A7.16 to give valid results over a wide
range of frequencies, because it is not possible by these simple
methods to produce the asymmetric gain curves that necessarily arise
from the existence of the bandgap (e.g. no gain below the bandgap
frequency), nor does eqn. A7.16 give the correct asymptotic behaviour
[5]. The results give a few key dependences that are physically
permitted and sketched in Figure A7.3. (If one knows the spectrum
over the whole frequency range, one can apply the full Kramers-
Kronig relationships.)
Figure A7.3 uses eqn. A7.15 to produce a schematic change of
complex susceptibilities for a narrow range of frequencies around the
central gain peak. Note how, as the gain narrows, so the associated
excursion in the real permittivity increases [6]. This is the essential
basis to reductions in aH by use of high-gain, narrow-bandwidth
quantum-well materials.

A7.4 Refractive index of transparent quaternary alloys

More elaborate models for the permittivity of materials are found in
the literature, basing the translation from the imaginary part of the
permittivity into the real part through the use of Kramers-Kronig
 Appendix 7 363

relationships. One example is given here of the single-effective-

oscillator model which can be used to provide an approximate
expression for the refractive index of the quaternary alloys used to
fabricate the waveguide structure for semiconductor lasers. One notes
that, if one takes the permittivity as in eqn. A7.12 then, as the term y
becomes smaller and smaller but the resonance remains, so
simaginarf-•2£foM(0) fi7TS(a) — fl). The more sophisticated models then
suggest taking the single-oscillator model and superimposing a range
of such single oscillators over a finite range of frequencies to give more
realistic matches to the physically observed refractive index. One such
idea is discussed in [7] where putting % = hco gives the photon energy,
the loss part of the imaginary component of permittivity is taken to be
£ « „ = i%4 (^g<^<^>f) with smaginary=0 elsewhere and the parameter
7] is a 'constant' of proportionality. The Kramers-Kronig relationships
are then applied to this term using standard integrals to give a general
equation describing the dependence of the refractive index n (using
n2 = srml) on the bandgap energy % and the energy of the incident
radiation % from:

where
•n=ifZi/V&0(%\-%'§) (A7.18)
and
« = 1.2398/ A (A7.19)
with % in electron volts and A in micrometres. %0 and %d are parameters
dependent on the composition of the alloys. This expression is valid
provided that the energy of the incident radiation is less than the
bandgap energy; in other words, the material is transparent. Hence, it
cannot be used for the active region material in a laser which, in any
case, will also be influenced by the injected-carrier concentration. We
can now detail the specific parameter values applicable to the
following alloy material systems:
(i) In^Ga^ASjP^ lattice-matched to InP [7,8]
^=1.35-0.72)H-0.12/ (A7.20)
*=0.1894y (0.4184 - 0.013);) (A7.21)
where
(A7.22)
364 Distributed feedback semiconductor lasers

«rf= (12.36*- 12.71))j+7.54*+ 28.91 (A7.23)

and x can only have values between 0 (InP) and 0.4672.
(ii) In1_x_2Ga^Vl2As lattice-matched to InP (x=0.4672) [9]
«,=0.75 + 1.06z+0.87z* (A7.24)
where
«0=2.48+1.427z+2.366a? (A7.25)
2
^=25.26+11.631*+7.844z (A7.26)
Note that z can only take on values between 0 (InGaAs) and 0.48 for
alloys lattice-matched to InP.
(iii) In^Ga^As^JYy lattice-matched to GaAs [7]
%g= 1.918 - 0.593)/+0.17l/ - 0.072/ (A7.27)
x=0.51+0.49;y (A7.28)
where:
«O=3.813 - 0.663^+0.643/ - 0.143/ (A7.29)
«rf=32.755 - 2.715)1+6.056/ (A7.30)
and xcan take on values between 0.51 (GalnP) and 1.
Please note that in [7], (p. L138) the equivalent equation to eqn.
A7.27 has a negative sign for the 0.758a? term, and not positive as it
should be.* The above expressions are useful for determining the
refractive index of all the layers in a laser structure for waveguiding
calculations, with the exception of the active region.

A7.5 References
1 LEE, Y.W.: 'Statistical theory of communication' (John Wiley, 1960)
2 DEUTSCH, R.: 'System analysis techniques' (Prentice Hall, Englewood
Cliffs, 1969), p. 108
3 ERDELYI, A. (Ed.): 'Tables of integral transforms, Vol. 2' (McGraw-Hill,
1954) Use partial fractions and entries 6 and 7, Chap. XV, p. 244
4 ASHCROFT, N.W., and MERMIN, N.D.: 'Solid state physics' (Saunders
College, Philadelphia, 1976), pp. 539-544
5 GREENHOW, M. 'High-frequency and low-frequency asymptotic conse-
quences of the Kramers-Kronig relations' /. Eng. Math., 1986, 20, pp.
293-306
* GREENE, P.D.: Private communication.
 Appendix 7 365

6 YAMANAKA, T., YOSHIKUNI, Y, YOKOYAMA, K., and LUI, W.: Theoret-

ical study on enhanced differential gain and extremely reduced linewidth
enhancement factor in quantum well lasers', IEEE J. Quantum Electron.,
1993, 29, pp. 1609-1616
7 UTAKA, K., SUEMATSU, Y, KOBAYASHI, K., and KAWANISHI, H.:
'GalnAsP/InP integrated twin-guide lasers with first order distributed
Bragg reflectors at micron wavelength' ,Jpn.J. Appl. Phys., 1980, 19, L137-
140
8 BROBERG, B., and LINDGREN, S.: 'Refractive index of In(l-x)Ga(*)
As();)P(l-);) layers and InP in the transparent wavelength r e g i o n ' , / Appl.
Phys., 1984, 55, pp. 3376-3381
9 GREENE, P.D., WHITEAWAY, J.E.A., HENSHALL, G.D., GLEW, R.W.,
LOWNEY, CM., BHUMBRA, B., and MOULE, D.J.: 'Optimisation and
comparison of InP-based quantum well lasers incorporating InGaAlAs and
InGaAsP alloys'. International symposium on GaAs and related compounds,
Jersey, 1990, Inst. Phys. Conf. Ser., 112, chap. 8, p. 555
10 GUENTHER, R. D.: 'Modern optics' (Wiley, 1990) pp. 272-279
 Appendix 8
Relative-intensity noise (RIN)

The relative-intensity noise (RIN) is an important quantity in

determining whether lasers are acceptable for use in optical-commu-
nication systems. Its analytic study can require extensive algebra [1],
but in this appendix the emphasis is on the physical significance of
RIN and simulations using time-domain modelling to estimate its value
for DFB lasers.
The discovery of the optical-fibre amplifier, with its relatively low-
noise optical amplification, has maintained the interest in amplitude
modulation for optical communication. Although both analogue and
digital modulation are of interest, the discussion here focuses on
digital intensity modulation. The net photon stream is turned from a
digital-signal pattern (Figure A8.1a) imposed on a random photon
stream (Figure A8.16), the latter being created by (i) the random
spontaneous emission and (ii) shot noise in the electron current
driving the laser. So far this shot-noise term has not been considered,
but it is easily added into the numerical model, as discussed shortly.
The signal and noise combine (Figure A8.1c) and, taking the idealised
case, with 100% efficiency for the photon detector, the photon stream
is translated directly into electronic current in the load of the
photodetector:

where Pamier and Pnohe give the photon power for the optical carrier
(which is being digitally modulated) and optical noise, respectively.
The current is matched into a resistor so that the ratio of equivalent
electrical carrier power to noise power is
^^elechicalequhmlmr (^ramVr) V (Pnoise)^ (A8.2)
2
where (P) indicates a temporal average. The electrical relative-
intensity noise is the inverse, so that one writes
 Appendix 8 367

probability
P

false decision
probability
p(1, given 0)

time
Figure A8.1 Importance of relative-intensity noise
(a) Shows an ideal digital signal
(b) Photon detection within a decision window varies with a Gaussian
distribution
(c) Shows carrier power plus noise power. Decisions are made as to
whether one or zero has been transmitted according to a detection of
photons above or below the decision level within the decision window of
time
(d) Noise peaks give a probability of a false decision being made so that
a 1 instead of 0 is detected. Scale of both noise and probability are
enhanced manyfold to illustrate the problem

(A8.3)

The importance of the RIN can be illustrated in Figure A8.1c where a

decision has to be made as to whether a one or a zero has been
detected. The probability distribution of the signal, within the time
window required to make a decision, is given by p(Pcarrier) (Figure A8.1 b
sketched as approximately Gaussian). The decision level is taken here
to be at the 'midpoint' between the 'one' and 'zero' levels and the
decision is made within the time window, controlled by the bit-rate
clock (Figure A8.1c). Occasionally, on account of the contamination of
the signal or carrier by noise, a signal which is correctly a 'zero' will be
368 Distributed feedback semiconductor lasers

detected as a 'one' because the signal is above the threshold level for
long enough within the decision window to return a 'one' state. One
can increase the decision level until there is negligible probability of
this happening; however, one then has to consider the probability of
the detector returning a 'zero' when a 'one' has been transmitted. The
decision level has to be chosen to minimise both probabilities and
need not be exactly at the 'midpoint' as indicated here.
The probability of a false decision giving a 'one' is indicated by the
small shaded area under the probability curve in Figure A8.1d
indicating the integral over the relevant section of the probability
distribution of the signal. A useful model is to take this probability
distribution to be Gaussian (Figure A8.1&) with a variance which is
determined by the mean-square deviation of the photon count within
the decision window given by Seount:
~KJ~2=cr2 (ASA)
The Gaussian probability function p(Smml) for finding a photon level
Scoum given a mean level Scount0 is then
p(SC(mn() dS^d/Vv) exp{- (S^-SM
(A8.5)
Taking for simplicity Smunl0 = 0 for a zero, and take the decision level to
be SdedsionJ then the probability of the signal rising above the decision
level when a zero has been transmitted is, say, pom^wn7jno

exp( - Siml/2<r2) d{Scmnl/&V2] (A8.6)

e given zero (1 / VV) exp( - x2) dx where Q= Sdaision/ (&V2)

Hi
(A8.7)
A similar result holds for probability p ^ ^ . ^ of the signal falling
below the decision level when a one has been transmitted. Assuming
transmission probabilities pme and pm.o for ones and zeros, then the net
probability of a false signal at any detection decision is
zero given one ~*~ Pxjtro > Pone given zero \ **•* * * /

In general, the decision level is adjusted to give the minimum value for
pnf>[. The analysis is simplified here by assuming that the noise at both
 Appendix 8 369

the 'one' and 'zero' levels is the same and the decision level is adjusted
to the midlevel so that Qj=Sone/ (a 2v2) where Sone is the maximum
output-photon count. For most digital systems, a maximum value of
this false probability is usually taken as 10"9 or even smaller, so that the
probability of a false decision is then

^ r ^ ^ ^ < 1 0 " 9 - Q > 6 (as a useful guide) (A8.9)

Hence at the peak power of the laser one requires a signal-to-noise

ratio which is given from

20 logl0[(Scmntmax)/crV2]>20 logl0(12)~21.5 dB (A8.10)

Analogue-television transmission using amplitude modulation is par-

ticularly demanding on the carrier-to-noise ratio for good-quality
pictures, but these demands can be alleviated by using frequency
modulation. Bearing in mind that photodetectors are not 100%
efficient, one can show that, even with frequency modulation, one is
likely to require the laser to have a net RIN of around — 40 dB or lower
[2]. Analogue applications may require RIN requirements below
— 50 dB for the net RIN integrated over the signal bandwidth. Because
the signal bandwidth varies with different systems, it is considered to
be more useful to plot the RIN of the laser against frequency in
dB/(Hz) terms. In a laser, it is found that the RIN peaks at the
photon-electron resonance fr discussed in Section 3.2.3, which is
typically in the gigahertz range with/ oc (Idrive- IthmhM)0'5.
To model RIN correctly, the laser model given in Chapter 7 should
be modified to provide shot noise into the electron current. Shot noise
should be modelled as a Poisson stream with a probability of one
electron arriving in rarrival seconds where rarrival then determines the
mean current. The mean arrival rate in any time interval T is then
^mean- T/7'arrival with a variance that is also N^rtn. However, with practical
currents of a few milliamperes or more, the rate of arrival of electrons
is so high that the Poisson stream is well approximated by having a
Gaussian distribution with a mean and variance equal to the equivalent
Poisson stream, even for subpicosecond time intervals. Hence if the
mean number of electrons within any interval is N^ then the probability
of Nwmw+N, electrons is approximately \lJ(27iNnmn)} exp-{|(N//NOTmw)}.
The time-step interval is s so that, if the mean current is 70 and there
are M laser sections, then there is a mean electron number per section
per step given as Nwmn=s/0/(^M). The variance in this number is also
Niman so that the current at each interval in each section is therefore
3 70 Distribu ted feedback semicondu ctor lasers

normalised drive 4 normalised drivej 2.56'

-160 -155 ikJ
it ii
ML
RIN, dB/Hz RIN, dB/Hz
-165 -160

iaP!
mIF ill
-165

H InWISH
-170
i -170
-175

WILJ
-20 -10 " f r 0
frequency, GHz
f
r 10 20
-175

1 ! ! i Iff
-20 -10 "*r0 f
r 10 20
frequency,GHz

Figure A8.2 RIN simulations

I=(q/s) (N^+VN^randn) (A8.ll)

where randn is a normally distributed random number with variance of
unity. A caveat has to be written, because it is possible for / to be
negative if randn is less than - VlMwmn. On such occasions, it would be
more appropriate to return 7=0 than a negative current. The
frequency with which this error occurs gives a rough measure of the
appropriate use of the Gaussian approximation to the Poisson
distribution for large numbers. In certain physical cases, one may be
able to provide some smoothing for the drive current [3-6] so that it
has a variance less than a Poisson stream.
The estimation of RIN using time-domain modelling demonstrates
the link between measuring low-frequency effects and the time of
calculation. The model DFB programs which are provided only resolve
frequencies down to about 3 GHz—this limitation can be removed by
running a modified program for longer. The model RIN program
(Figure A8.2) resolves down to frequencies around 25 MHz and takes
about 90 min per run on a 166 MHz Pentium PC, calculating nearly a
quarter of a million time steps and averaging appropriately. The time-
domain model, by its very nature, gives a stochastic output so that one
has to take the smoothed values to estimate the RIN but one can see
how noise fluctuates as well as causing the signal to fluctuate with time.
No two runs will be identical. The model described below uses the
program dfbrin within directory dfb and looks at RIN for a A/4-phase-
shifted DFB laser using the standard parameters within the model.
The standard numerical laser model for a A/4-phase-shifted DFB
laser (that is provided with this book) is adapted to take two runs
changing the drive levels. The differences in the resonance fre-
 Appendix 8 371

quencies/ r are found to be proportional to (Idriw~ 4r«/ww)°'5 with

normalised threshold of 2.1 and the normalised drives shown here of
4 and 2.56, giving approximately a twofold change in the resonance
frequency where the RIN peaks. The mean intensity output with a
normalised drive level of 4 is approximately 6 dB up on the drive level
at 2.56. Intensity noise is always symmetrical around the zero frequency
at baseband.
The resonance in the RIN occurs because of the photon-electron
resonance, and is the same resonance as that discussed in Chapter 4
and Appendix 4.

A8.1 References
1 COLDREN, LA., and CORZINE, S.W.: 'Diode lasers and photonic
integrated circuits' (Wiley, 1995), pp. 230-241
2 GREEN, P.E.: 'Fibre optic networks' (Prentice Hall, Englewood Cliffs,
1993), chap. 9 gives an overview about the relative importance of different
forms of noise on different systems
3 GOLUBEV, Y.M., and SOLOKEV, I.V.: 'Photon anti-bunching in a coherent
light source and suppression of the photorecording noise', Sov. Phys.-JETP,
1984, 60, pp. 234-238
4 MACHIDA, S., YAMAMOTO, Y, and ITAYA, Y: 'Observation of amplitude
squeezing in a constant current driven semiconductor laser', Phys. Rev.
Lett., 1987, 58, pp. 1000-1003
5 RICHARDSON, W.H., MACHIDA, S., and YAMAMOTO, Y: 'Squeezed
photon numbtr noise and sub-Poissonian electrical partition noise in a
semiconductor laser', Phys. Rev. Lett., 1991, 66, pp. 2867-2870
6 LI, Y-Q., EDWARDS, P.J., LYNAM, P., and CHEUNG, W.N.: 'Quantum-
correlated light from transverse junction stripe laser diodes', Int. J.
Optoelectron., 1996, 10, pp. 417-421
 Appendix 9
Thermal, quantum and numerical
noise

A9.1 Introduction

There are at least three 'different' principles that can be used to

calculate the spontaneous emission in a laser:
(i) the method given in Section 2.4.1;
(ii) a method which considers the spontaneous emission which has to
occur in an ideal waveguide amplifier to obtain the correct 'noise'
output [1,2]. This amplifier can also be turned into an attenuator
to see that the results are still consistent; and
(iii)the spontaneous emission which can also be found by the Einstein
treatment of counting photon states [3],
All these methods are tied together to show consistency with one
another. The appendix is concluded with essential material for
interpreting the spontaneous emission in a numerical formalism.
The appendix first considers thermal and quantum noise and the
ideal signal-to-noise power ratio that can be measured using a
100%-efficient photodetector. An ideal optical amplifier followed by
an ideal detector is then considered. At the output of this ideal
amplifier, the signal-to-noise-power ratio depends on the spontaneous
emission and it is argued that this ratio has to be the same as the signal-
to-noise-power ratio for the ideal direct detection of optical fields at
the input. The ideal amplification cannot improve (or worsen) the
signal-to-noise ratio. This argument gives one measure of the
spontaneous emission. The same device considered as an attenuator in
thermal equilibrium with its surrounds gives yet another measure of
 Appendix 9 373

spontaneous emission, and finally spontaneous emission can be

estimated through counting photon states.

A9.2 Thermal and quantum noise

Thermal noise at microwave frequencies provides an available power
of kTA/flowing between two matched power absorbers, connected by
a single-mode transmission line, all at a temperature T (where k is
Boltzmann's constant, T is the absolute temperature and A / is the
effective frequency bandwidth of the measurement or the system). In
thermal equilibrium, the noise power into a resistor balances the noise
power flowing out. If the two absorbers are at different temperatures,
net power flow is from the hotter to the cooler. The 'Nyquist'
relationship kT A/ forms the basis of calculations of signal-to-noise
ratios at microwave and lower frequencies [4]. However, at room
temperature, kTA/~40 pW of thermal power for every gigahertz of
bandwidth so that, with the unlimited bandwidth of free space,
astronauts would be incinerated were it not for the Bose-Einstein
correction [5] giving the available thermal noise power as
Pamalablt=hfLf/{exv{\if/kT) -1} (A9.1)
Below, say, 1000 GHz at room temperature, the Nyquist formula is
recovered because (exp(h//kT) — 1}~ ( h / / k T ) , but now eqn. A9.1
suggests that the noise at optical frequencies is negligible! Then
quantum noise [6] has to be introduced and an additional term is
conventionally included of the form
P wo ,,=|h/A/+h/A//{exp(h//kr) - 1 } (A9.2)
This again suggests a catastrophe as the bandwidth A/increases, but
the additional term of l h / A / unlike the Nyquist term, is not available
power but a measure of uncertainty in any measurement. Uncertainty
may not be recovered as usable power and is not amplified by an ideal
amplifier, but is simply present in any measurement.
The magnitude of this uncertain power depends in detail on how
one measures the field [6]. With direct detection of the optical-field
intensity by an ideal photodiode, Puncerlainty=hfAf, so one may write
Pun^mnty/P^al^f^f/P^al (A9.3)
The factor of 1/2 in eqn. A9.2 requires coherent rather than direct
detection. One way of understanding the result in eqn. A9.3 is to
consider a source of light where there are, on average, p photons per
374 Distributed feedback semiconductor lasers

second emitted over a signal bandwidth <5/(the connection between 8f

and A/emerges shortly). Then the detector receives an optical power

where/is the mean optical frequency. The direct detection process is

random, and the electrons (one per photon if the detector is 100%
efficient) emerge from the detector in a Poisson stream. The mean
'signal' current is Io= q p where q is the electronic charge and the
random arrival of the electrons gives 'shot noise' [4], with a mean-
square noise current:
il=2qlo8f (A9.5)
The electrical power into an ideal load of R ohms for the photo-
detector gives
=
\* noise electrical/ *signal electrical) output \ * ? / o ^ / ) / ( 9 P ) ^ <</ & / / * signal input ( A9.0)

Now recall that detection takes the signal down to baseband where the
noise bandwidth runs from — 8f to + 5 / and so the net bandwidth
A/=2 8f so that eqn. A9.6 confirms the result of eqn. A9.3.
With coherent optical signals, it is possible to recover a factor of 2 by
using homodyne detection which requires that the phase of the output
is known. This additional information about the phase reduces the
minimum uncertainty to i A/A/rather than the value of A/A/obtained
by power (photon) detection. The minimum uncertainty in measure-
ment of the optical-field power then directly relates to the best possible
signal-to-noise-power ratio, given that one uses ideal measurement
methods.

A9.3 Ideal amplification

Now turn to assessing the spontaneous emission which arises in the
ideal amplifier of a forward optical field where the travelling-wave
equation is given from (see eqn. 4.20)

1 dE dE
_ + —
i (t,z) (A9.7)
Vff dt dz
Here ispf is proportional to the random dipole 'current' as a function
of time that excites the spontaneous emission for the forwardfieldand
g is the field gain, at present taken to be a constant. To simplify the
analysis, redefine the field variables:
 Appendix 9 375

E=Fexp{j(cot- fiz)} and ispI{t, z)=ispJ(t, z) exp{ -j{<ot- f3z)} (A9.8)

so that ispf{t, z) varies relatively slowly with time and space like Faround
some central frequency co with /3=co/vg, so that eqn. A9.7 becomes
adequately approximated by

(A9.9)

Because isPf(t, z) and isp^tf z) differ only by a phase factor, the

autocorrelations of each of the two excitations are equal (denoting
ensemble/time averages by ):

The random properties of the spontaneous emission are not changed

fundamentally by the change of phase factor. The forward field and
then the * power flow5 are next found from integration and using the
fact that the spontaneous emission is random so that one is, in effect,
just adding random spontaneously generated components to the field
as the energy propagates along the guide:

[<
F=Fin exp(gz) + exp(gz) I ispf(t, zr) exp( — gzf) dzf (A9.11)
Jo
Squaring up on either side and ignoring terms like F*isfwhich average
to zero

F*F=Fin*Finexp(2gz) +

, * f'^f- , , ,- ,
p y n ( y fry \ I fl 7 I 1 (t 7 i ^ PVTII — (T7 I 1 ( t 7 \ PYT1 ( — GT7 i fl7
Jo Jo
(A9.12)
Spontaneous excitations are assumed to be spatially uncorrelated,
such that h,,f(t, f)% (t, z')=r (t)28(z' - z") (A9.13)
f spJ

The fields can be normalised so that P F gives the forward power

flow:
F*F=Pf= hfSjMvg (A9.14)
along the guide of effective area si where Sf is the photon density
associated with the forward power flow. Then fsp^(t)2 averaged over
376 Distributed feedback semiconductor lasers

time gives the mean-square noise power per unit length of amplifier.
Integration over the length L gives the output at the end of the
amplifier region from

f fspJ(t)2 exp( -2gz') dz' (A9.15)

For future comparison, note that the rate of change of the mean power
flow given by eqn. A9.15 can also be written at a position z as a sum of
stimulated growth plus addition of spontaneous emission:
dPf/dz = 2gPf+fspf(t)2 (A9.16)
Writing GL=exp(2gL) as the power gain along the guide gives

(A9..7,

For large enough gains with no losses of photons from input to

output,

(A9.18)

Now consider the interesting situation. At the input there was a power
Pin but with a quantum uncertainty of the power (using direct
detection) given from h/5/over a bandwidth of Sf. However, this
uncertainty is not power that can get amplified but is merely an
uncertainty in the measurement of the fields. At the end of an ideal
quantum-amplification process with a sufficiently large gain GL> the
quantum uncertainty can be negligible compared with output power
levels, but now it appears that there is an added available noise power
of fsp/(t)2/2gproduced by amplifying the available thermal noise. The
measurements at the input and the output are ideal measurement
processes and, if the amplification is ideal quantum amplification with
no absorptive loss of photons but only quantum spontaneous emission
along with ideal stimulated gain, it is argued that the signal-to-noise
ratio at the output must equal the signal-to-noise ratio Pfin/ (hfSf) at
the input. Quantifying this result gives
2gPfm/fspf(tf=P/in/hf8f or fspf(t)2=2gh/Sf (A9.19)
Here then is a statistical measure of the mean-square spontaneous
 Appendix 9 377

excitation at each point in the guide linking the spontaneous emission

with the gain, as discussed in Section 2.3. Note that, in this work, 2g*is
the net power gain per unit length making allowances for any
confinement factor F; some authors use gas the power gain. If there is
a spectral filter, there may be only a narrow band 5/of laser power that
is relevant in determining the spontaneous noise but, taking the
spontaneous emission over its whole range kspf, then g has to be
averaged appropriately.

A9.4 The attenuator

The calculation is now repeated with the field gain per unit length g
changed into a loss per unit length (— a) where one then finds that
the output power is now given from:

PfMU=APfin+ (1 - A) - ^ - (A9.20)

where A=exp( — 2aL).

Consider what happens if the input power is just the thermal power
given from
hf&spf/{exp(hf/kTJ -1) (A9.21)
where Tn is the input noise temperature and Aspf is the whole
spontaneous-noise bandwidth. If the attenuator is in thermal equilib-
rium with its surrounds, Tn must be the same 'noise temperature' for
the whole amplifier input ensuring that

- ^ - = h / A , / / { e x p ( h / / k r j -1} (A9.22)

The transition from attenuation to gain can be made via the concept
of a negative reciprocal temperature where ( 1 / T J is large near
absolute zero but becomes small by pumping sufficient electrons and
holes into the interaction region. This pumping initially effectively
increases their equivalent temperature with the attenuation a tending
to zero until transparency is reached when a- 0. Then with still harder
pumping, a changes sign to become gain g( = — a) with
(1/ Tw)—• — 1(1/ 7JI and with yet harder pumping still to give large
enough 'population inversion' (1/TJ—• —oo with exp(h//kjTJ—»0
and one recovers eqn. A9.17. The conclusion that one arrives at is that
378 Distributed feedback semiconductor lasers

eqn. A9.17 should be modified to allow for an arbitrary degree of

pumping of the laser by writing
fspf{t)2=(2g)nimthfAspf (A9.23)
where ninv=l/{l - e x p ( - h / / k I TJ)} gives the measure of the pump-
ing of the laser or the 'population inversion' of the charge carriers in
the laser. The strongest inversion makes ninv unity, but more typically
ninv~2 or so. This discussion owes much to Henry [1] and Marcenac
[2]. Ghafouri-Shiraz [7] has a more detailed treatment of the laser
amplifier including losses. In the present work, the important question
that is addressed is the magnitude of the ideal spontaneous-emission
sources.
In eqn. 5.9 no allowance was made for any difference in the
spontaneous noise in the forward and the reverse directions because,
from symmetry, the mean values for both the forward and reverse
emission are equal in magnitude: fspj(t)2=fspr(t)2 = rsp2. However, the
equality of mean magnitude must not be mistaken for correlation.

A9.5 Einstein treatment: mode counting

Next consider the Einstein treatment of stimulated and spontaneous
interactions with charge carriers in a uniform waveguide of effective
cross-sectional area si using the two-energy interaction model of
Section 2.3.1 for the interaction between photons and the charge
carriers in the medium. Recalling eqn. 2.10,
dS/dt = K^SiNM^PN, - N,N2) + 0 ^ / W (A9.24)
Make the translation that d(vg t)—>dz and use eqn. A9.14 where
Pf= hfSfs& vg and follow Section 2.3.1 using the Fermi-Dirac statistics
but with the inversion factor ninv then eqn. A9.24 becomes
dP/ dz = {Kstim(PN/ nrnv) / vg}Pf+ hffr^fNM (A9.25)
The spontaneous term /3spKspPN in eqn. A9.24 gives only that
spontaneous emission per unit length which is actually coupling to the
waveguide and being subsequently amplified. Comparing eqn. A9.25
with eqn. A9.16 shows the links
2g= (KsllJ vgnlm) PN r%rhff3spKsp PNM (A9.26)
From eqns. A9.19 and A9.26,
r%f= (2g)n^hf8fr(K^/vg)PNhf8f=hf/3^PNsl (A9.27)
From eqn. 2.12, Ksp=B where B is the bipolar recombination constant
 Appendix 9 379

for the material. The connection between jR^, and KsHm was found from
eqn. 2.19 considering the equilibrium conditions and black-body-
radiation density to give eqn. 2.20 written as
lvg) (A9.28)
On rearranging eqn. A9.27 with the result of eqn. A9.28, explicit
values for the spontaneous excitation and spontaneous-coupling
coefficients are obtained as
(A9.29)

(3sp=(l/2) (AI/4TTM) (8f/Aspf) (A9.30)

A9.6 Aperture theory

Now from basic aerial theory [8], any aerial with an aperture si
embedded wholly in a material where the central wavelength is Am has
an aerial 'gain1 Garrinl given from
GarHn^{dA7r/Xl) (A9.31)
Here it is assumed that the whole aerial is immersed in the medium so
that km is the effective wavelength appropriate to that material. The
'gain of an aerial' can also be found by finding the average solid angle
of emission and comparing this with the total 4?rsteradians so that the
effective solid angle of emission ft from the aerial is given from
Gnmfll=47r/n and hence (A2J47rs&) =il/4m
At first sight, the effective aperture of all the radiating dipoles calls
for a full analysis of the radiation of a dipole within the complex
waveguide. However, one already knows that the net field pattern of all
this spontaneous emission that actually couples to the waveguide
mode, by definition of the mode, has to have an effective cross-section
of si (the effective cross-section of the waveguide), so si is also the
effective aperture of each dipole so that Cl=A2m/M=A2/sisrrf>ij- where
Srrrjy is the real effective relative permittivity of the guide. This is exactly
the same answer as finding the solid angle of emission outside the
waveguide into air and then recognising that Snell's law reduces this
solid angle by (l/n r r ^) 2 as derived in Section 2.4. Hence, for a single
forward propagating mode,
4=1(5//A,/) (ft/477) (BMPNhf) (A9.32)
As pointed out in Section 2.4, the factor of 2 recognises that only one
polarisation couples to the waveguide; the factor (<5//A5//) recognises
380 Distributed feedback semiconductor lasers

effective
erreff aperture
optical guide

spontaneous I solid angle 8fl

dipole emission y ,
solid angle 8 n
g - 5n7e
/ rr eff radiation into free space
spc
2
.2 A

Figure A9.1 Schematic of spontaneous dipole emission

Solid angle O determined by area of guide because this becomes the
effective aperture for the whole system of spontaneous dipoles.
Comparison is made with the solid angle into free space to show that
internal solid angle is correctly reduced by seffec(ive

that only a fraction of the spontaneous emission over its whole

spectrum couples into the lasing-mode frequencies; the factor (fl/47r)
recognises that spontaneous emission covers the whole 4TT steradians
but only couples into the appropriate solid angle for the waveguide
(here il is taken as appropriate for the solid angle of, say, the forward
emission alone); while the term (Bs&PNhf) gives the spontaneous
emission over all possible polarisations. With a spontaneous-confine-
ment factor Tsp> this emission rate is modified to (rspBMPNhf) with
r v/ ,=r in this work.

A9.7 Numerical modelling of spontaneous noise

It is now essential to discuss the modifications required for the finite-
difference time-domain analysis and how to identify and model this
spontaneous-emission noise numerically, given a space step of s. A
straightforward method modifies eqn. A9.12; one finds with g-=0:

sv,(Ts/t/ g , Z's) -,"Zs)

(A9.33)
where spontaneous excitation ispf(Ts/vg, Zs) is inserted at each of the
space points [T, Z] which are separated by s so that Z and T are integers
defining the step numbers. The lack of correlation between each space
 Appendix 9 381
and time point requires for distinct Z and Z' and distinct T and T':
sispf(Ts/vg, Z's)sispf(Ts/vg9 Zs) =0 (A9.34)
sispf(Ts/vr Zs)sispf(Ts/vg, Zs)=0 (A9.35)
If one wishes to restrict the noise bandwidth below the Nyquist limit,
one should introduce temporal correlation, as will be discussed
shortly.
Now similar results can be formed for the reverse wave: there is
nothing special about the forward direction other than the sponta-
neous excitation for the forward wave is uncorrelated with the
excitation for the reverse wave. Hence one has
sispr(Ts/vg, Ts)sispr(Ts/vg, Zs)=0 (A9.36)
sispr(Ts/vgf Zs)stapr(Js/vg9 Zs)=0 (A9.37)
with
sispr(Ts/vg, Z's)sispf(Ts/vg9 Zs)=0 (A9.38)
but now for any T, T, Z and Z\ even though these may be equal. This
ensures that there is no coupling between the forward and reverse
emission when there is no scattering from the forward field into the
reverse field.
When space and time are equal in both excitation terms
sisp/T/vg, Zs)sispJ(Ts/vg, Zs) =srspf(t)2 (A9.39)
2
sp (T/vg, Zs)sispr(Ts/vg, Z s ) = s V ( 0
siispr (A9.40)
with
(A9.41)
As the length of the laser waveguide L=N s increases (retaining g=0
for this calculation), the mean spontaneous forward power increases
as
l (A9.42)
These requirements are met for the forward field by inserting, at each
space step and at every time step, a random excitation given by
sispf(Ts/vg, Zs)={l(a/47r) (Sf/AJ) (rsps&sBPNhf)\l/2gi(A9A3)
where S?l=9?+7% with % and °}J normally distributed random real
numbers such that the mean-square value 2ft 2 = 1. The randomness of
the complex 91 gives the correct zero-mean correlation in amplitude
and in phase from point to point and time to time, as required for
382 Distributed feedback semiconductor lasers

spontaneous power

(a) I
frequency
spnum
spontaneous power

(b)

frequency
spontaneous power

(c)

frequency
[
physical ] numerical! 1111II

Figure A 9.2 Spontaneous spectrum

a Approximated by rectangular distribution with uncorrelated
impulses at each time point
b Slightly curved distribution achieved by partial correlation of
impulses at adjacent time points
c Offset central frequency by adding a phase factor

spontaneous emission. A similar but independent random value of 2/1 is

used to excite the reverse field. This independence of random
excitation of the forward and reversefieldsis essential. The value given
for the bandwidth <5/in the computation needs consideration because,
in modelling, the bandwidth is determined by the step length s.
With a step length for the finite-difference scheme of s, a single
'impulse' at one point in an array of points has a 'white' spectrum
covering frequencies ±\ (vg/s) about the central frequency. The
numerical bandwidth for the spontaneous emission is then
Aspnunf= (vg/s) determined from the spatial step length, and this is
also the computing-system bandwidth. Retaining the approximation
that the spontaneous spectrum is white over this range (vg/s), one
should replace Sf/Aspf with (vg/s Av//):
s ^ ( T s / ^ , Zs) ={i(ft/47r) (vg/sAspJ) (rsp$&sBPNhf)}l/2<3l (A9.44)
Figure A9.2a indicates schematically the approximation that such a
'white' spontaneous spectrum gives.
 Appendix 9 383

spontaneous
L
power

frequency

physical j ~] numerical 1111111

Figure A9.3 Numerical-noise spectrum
With short enough space step, the numerical-noise spectrum may
extend beyond the physical spectrum but can be tailored to approx-
imate to the physical spectrum in the central region

Band limiting of the spontaneous spectrum can be achieved by

providing correlations between the spontaneous emissions and time.
This can be done with a numerical filter as in Chapter 7 or, more
simply, if approximately, by correlating the noise at just two adjacent
time steps by having, for example:
sispj{ls/vg vgy Zs)cos <f>+ispf{(T- \)s/vg, Zsjsin cf>
(A9.45)
With sin <^~0.1, for example, then the spontaneous spectrum alters as
in Figure A9.2& An offset central frequency is achieved by introducing
a phase shift; for example, writing sin cf>—*exp(j S) sin <f> where in
Figure A9.2csin 4> = 0A and the phase factor is 5=0.25. The factors 8
and <f> then constitute parameters which can be fitted to the
requirements. The filtering in time and space can be done so as to fit
with eqns. A9.36 and A9.37. Readers are encouraged, as an exercise, to
set up a MATLAB program to test for themselves this band limiting
and spectral shift of the spontaneous emission.
If one normalises F*F-^>(SfM vg) so as to give the photon numbers
per second rather than the optical power level, one omits A/from eqn.
A9.44 to give
sispf(Ts/vg, Zs)={l(O/47r) (vg/s^pJ) (T^sBPN)}l/2$l
(A9.46)
(A9.47)
2
where Vsp=(KtrA vg/8 IT srreff A5/J) is the effective volume for the
spontaneous emission per step length. The transverse Petermann
factor Klr (see Section 2.4.2) has been included for generality to cover
gain guides not discussed above, but for index guides as discussed here
it is set at 1. The shorter the step length, the wider is the bandwidth of
384 Distributed feedback semiconductor lasers

spontaneous noise, and this increase in noise is taken into account

automatically by having more steps per given length. This theory
assumes only a single transverse mode, so that the larger the cross-
sectional area or effective aperture of the guide, the higher is the
aerial gain for a small dipole radiating into that guide and the smaller
is the coupling angle of spontaneous emission into that mode. This is
why the effective area si cancels in its effect on the noise. With gain
guides, the Petermann factor can well be a function of si.
The values for the reverse spontaneous excitation are identical in
formulation to those for the forward excitation given above, but
generated with completely independent random generator 2/1 from
the generator in eqn. A9.46.
If the step length s is made so short that vg/s>Aspf, one has to
reconsider eqn. A9.45 and is forced to introduce a spectral filter, for
example by correlating the spontaneous noise at two adjacent step
points. If, for example, one had

(A9.48)
and s is sufficiently short, the numerical-noise spectrum, may extend
beyond the physical-noise spectrum, as in Figure A9.3, but could
approximate the significant central region adequately.

A9.8 Higher-order noise statistics

It has so far been assumed that the only terms which matter in the
modelling of the noise are the mean and second moment, so that
there is zero mean correlation between adjacent points of spontaneous
emission and also the correct mean-square spontaneous emission
power. The current evidence is that this is often sufficient for much
laser modelling, but one can see where the numerical model departs
from the physics. For example, if the time step is so long that, during
each time step, 100 electrons and holes on average recombine, the
amplitude fluctuations will be different in detail from the situation
where the time step is so short that on average only 1/100 electrons
and holes recombine (modelling admits of fractional electrons!). This
problem of balancing the time step and spontaneous emission to
obtain the correct granularity of emission has not been explored in
this modelling primer, and has to be left for further research.
 Appendix 9 385

A9.9 References
1 HENRY, C.H.: 'Theory of spontaneous emission noise in open resonators
and its application to lasers and optical amplifiers',/. Lightwave TechnoL,
1986, 4, pp. 288-297
2 MARCENAC, D.D., and CARROLL, J.E.: 'Quantum mechanical model for
realistic Fabry-Perot lasers', IEEProc.-J., 1993, 140, pp. 157-171
3 THOMPSON, G.H.B.: 'Physics of semiconductor laser devices' (Wiley,
1980)
4 FREEMAN, J.J.: 'Principles of noise' (Wiley, 1958)
5 KITTEL, C, and KROEMER, H.: Thermal physics' (W.H. Freeman and
Co., 1980), 2nd ed.
6 MARCUSE, D.: 'Principles of quantum electronics' (Academic Press,
1980)
7 GHAFOURI-SHIRAZ, H.: 'Fundamentals of laser diode amplifiers' (Wiley,
Chichester, 1995)
8 RAMO, S., WHINNERY, J.R., and VAN DUZER, T: 'Fields and waves in
communication electronics' (Wiley, 1994), 3rd ed.
 Appendix 10
Laser packaging

A10.1 Introduction
The extra requirements of packaging optoelectronic rather than just
electronic devices are challenging [1], and real packages involve
numerous compromises between different aspects of performance, as
well as manufacturing costs. To make the best of a well designed
device, DFB-laser chips must be packaged so as
(i) to protect them from the environment;
(ii)to enable monitoring of the optical output;
(iii)
to provide thermal heat sinking;
(iv)to provide a rigid and stable optical coupling platform to a fibre
(or other output system); and then
(v) to allow electrical interfaces between the appropriate 'DC and
modulated RF current supplies and the laser.

A10.2 Electrical interfaces and circuits

The need to obtain stability and precision of the mechanical
alignment in laser packages often results in a metal package with glass-
to-metal seals for electrical connecting pins, and glass windows or
carefully designed fibre interfaces. Rapid progress is being made with
ceramic packages incorporating stripline interfaces, and some laser
packages are now partly injection moulded. The variability of package
and laser parameters is large, and the parameters relevant to any
particular laser and its package need to be measured appropriately
and then used in any modelling. The values quoted here are for a
typical metal package with glass-to-metal seals.
 Appendix 10 387

V V,n
driver"1' ' • ' • i . • - • • • •# . -*•— '•

internal laser < .

driver external circuit package+bond laser chip

Figure A 10.1 Schematic of laser package
Showing electrical parasitic capacitances and inductances and inter-
faces

Figure A10.1 indicates such typical parasitic circuit components

which need to be considered in evaluating the overall frequency
response of a laser to direct modulation. The glass-to-metal seal of a
package may present a lead-through capacitance C/>~0.5 pF, and the
fine gold or aluminium bond wire to the laser chip usually contributes
an inductance LB~ 1 nH, giving a characteristic filter impedance ~50 ft
to match the driver's source impedance but with a cutoff frequency as
low as 7 GHz. The laser capacitance CL may or may not help
compensate to a cutoff around 10 GHz, but very careful microwave
design of special packages is needed for still higher frequencies.
'SPICE' models are likely to be available for components within the
electronic driver, and sophisticated 3D modelling and measurement
[2] may be necessary to establish the characteristic of the laser
package.
The RF equivalent circuit of lasers [3] suggests that the internal laser
dynamics adds an inductance to a series dynamic resistance ry which,
with Cj, helps to give a first-order account of the photon-electron
resonance within the laser along with the electrical-to-optical power
conversion determined by Tj. The shunt resistance Rj along with Cj
determines the damping of the modulation and relaxation rate to
equilibrium. While this gives an approximation to the laser's dynamic
characteristics, the steady current through the laser determines a
junction voltage VJ0 which, in turn, determines a steady recombination
current through Rj. The internal recombination rate should be
different dynamically from the static value, requiring more elements
than are shown here. More complete modelling of laser and package
strays becomes complex [4], and distributed equivalent circuits may
have to be used instead of lumped circuits at high enough microwave
frequencies where the laser's length becomes comparable with the
388 Distributed feedback semiconductor lasers

modulation wavelength [5]. The package elements are indicated in

Figure A10.1 as a 'lumped' equivalent-circuit model and are often
simplified further for modelling (as in Appendix 11). Laser series
resistances are typically 3-5 fl for buried heterostructure devices, and
1-2 O for ridge-type devices. The parallel laser capacitance CL varies
strongly with the laser's structure: a typical buried-heterostructure
laser may have Q~200pF, whereas comparable 'ridge'-type lasers
could have 20 pF, and specially designed lasers for high-speed
operation may have CL<b pF.

A10.3 Thermal considerations

Lasers are always sketched in this book as having their junctions close
to the upper surface. In practice, they are usually soldered junction-
side down directly onto a good heatsink (such as type-IIA diamond)
which can then give thermal impedances of around 30 degC/W,
reducing to, perhaps, 10 degC/W for 'long' lasers such as those used
as pumps for erbium-fibre amplifiers. Chips mounted with the
junction side uppermost will typically have thermal impedances 2-4
times greater, although 10 times greater is possible in some modules.
Excessive heating makes almost all laser parameters worse; this is a
particular problem with multilaser modules, where thermal crosstalk
also exists. At some frequencies, thermally induced chirp is a serious
effect, and can cause changes in wavelength >0.1 nm on timescales of
0.1-10 (xs [6]. The dynamic thermal properties of the laser are
controlled by several time constants, characteristic of parts of the laser
and its heatsinking arrangement; however, to a first order, GaAs-based
lasers may often be characterised by a single time constant of about
1 [xs, while long-wavelength InP-based lasers suffer more from direct
heating of their active regions, and often show significant heating
effects in about 0.1 |xs. These times appear long compared with
1 Gbit/s data, but many standards for data transfer allow long strings
of T s or '0's, and so thermally induced chirp and other effects may
cause major problems.

A10.4 Laser monitoring

Many laser packages contain a photodiode designed to monitor the
laser output, often merely detecting the slow variations in the average
laser power. First, the monitor diode usually detects light from the
back facet (which commonly has a coating to increase its reflection to
 Appendix 10 389

ensure that more power emerges from the front facet). Power for the
fibre is extracted from the front facet [often antireflection (AR)
coated]. In Fabry-Perot lasers, the front and back powers track each
other very well, but in DFB lasers spatial-hole burning and longitudi-
nal shifts of the lasing power can result in considerable divergences
(see Figure 5.18). In addition, efficient coupling of optical power from
laser to fibre requires submicron tolerances, and is likely to vary with
temperature, whereas power coupled to the monitor photodiode is
stable due to the relatively large sensitive area. Hence monitoring of
power in the system fibre is likely only to be accurate to ± 1-2 dB, and
long-term 'creep' of the package may increase this figure. Caution is
therefore needed in analogue or other high-linearity applications in
using the monitor diode to control the laser's precise performance.

A10.5 Package-related backreflections and fibre coupling

As seen in Chapter 1, the proximity of the laser round-trip gain to unity
makes the laser very sensitive to backreflections within its coherence
length, which certainly applies to most packages. For example,
Petermann [7] calculates that, at 5 mW output power, feedback levels
below 10~4 are required to achieve linewidths of 70 MHz starting from
an intrinsic linewidth of 12.5 MHz, assuming aH=6. The critical effects
noted for facet reflections in Chapter 6 can also occur with reflections
from elements within the package. Monitor photodiodes are therefore
usually angled with respect to the laser axis, and are often AR coated
in order to have a negligible effect on the laser. 'Selfoc' lenses are
occasionally employed to couple light from the laser-output spot to the
cleaved end of a single-mode fibre and in these cases, if all the glass/air
interfaces are flat, the typical 4% power reflection at each interface
can create serious problems of interference, and interaction with the
laser. Use of a fibre end which is rounded to form an integral lens
(Figure A10.2) not only reduces the number of glass-to-air interfaces
to one but also defocuses any reflected power, greatly reducing any
coupling back into the laser of reflected light. Typical packages
employing either discrete lenses or lens-ended fibres achieve coupling
efficiencies of 20-40% for laser power into a single-mode fibre,
although values above 70% are possible with more expensive assem-
blies and mode transformers [8]. The relatively large circular emission
areas of VCSELs combined with appropriate lenses should achieve yet
higher coupling efficiencies. These figures should be compared with
only 5-10% coupling directly from a conventional edge-emitting laser
390 Distributed feedback semiconductor lasers

High reflection coating AR coating

monitor diode T
lensed fibre seal
laser
lead I
rigid mount
throughs heat sink
for fibre
I

Figure A10.2 Schematic laser package

Lensed fibre must be rigidly mounted, but the monitor diode is not
quite so critical provided that it does not reflect light back into the
laser. The laser is mounted with its junction closest to the heat sink;
the package is normally hermetically sealed. Laser connections are not
shown, but need to have low inductances with low lead-through
capacitances. Rigid mount for fibre can be formed from a V-groove
etched in silicon [10-12]

into a flat, cleaved single-mode fibre end, which also results in

significant back reflections. Lasers without fibre pigtails but incorpo-
rating demountable optical connectors into their packages have an
acute problem for high-performance use because the laser output
should be coupled optimally to the fibre end face and therefore, by
reciprocity, any reflection from this flat end will couple optimally back
into the laser. While angling of the laser axis with respect to the axis of
the output fibre alleviates this, the better solution for critical
applications is to use a laser package incorporating a permanent fibre
pigtail, and then splice this pigtail to the system fibre.
To reduce back reflections, 'optical isolators' can be employed in
systems. These make use of the Faraday effect [9,10] of polarisation
rotation under magnetic fields in suitable materials, and can attenuate
any reflections coming back to the laser by - 3 0 dB for a single-stage
unit, or 50-60 dB for a double-stage one with transmission losses of
0.5-1 dB; but some of them generate significant reflections from their
own internal interfaces. Critical analogue applications such as multi-
channel cable television will certainly require isolators, and the rest of
the package will also need very careful design, probably involving
antireflective coating to all optical interfaces.
In WDM systems where lasers with several different wavelengths are
required, the coupling of many devices to many fibres in one package
is even more challenging than coupling a single fibre to a single
device. An ingenious solution has been found using anisotropic wet
etching of silicon along crystal planes to produce well defined V-
 Appendix 10 391

grooves. The lasers can be flip-chipped onto the silicon motherboard

containing such V-grooves which can align the fibres to the laser with
great precision [11-13], and as such are also useful in volume
production for single lasers.

A10.6 References
1 MATTHEWS, M.R., MACDONALD, R.M., and PRESTON, K.R.: 'Optical
components—the new challenge in packaging', IEEE Trans. Compon.,
Hybrids Manuf. TechnoL, 1990, 13, pp. 798-806
2 XENG, J-X.: 'A 3D electromagnetic simulator for high frequency
applications', IEEE Trans. Compon., Packag. Manuf. TechnoL B, Adv. Package
1995, 18, pp.578-595
3 TUCKER, R.S., and POPE, D.J.: 'Microwave circuit models of semi-
conductor injection lasers', IEEE Trans., 1983, MTT—31, pp. 289-294
4 DELPIANO, R, PAOLETTI, R., AUDAGNOTTO, P., and PULEO, M.:
'High-frequency modeling and characterization of high-performance
DFB laser modules', IEEE Trans. Compon. Packag. Manuf. TechnoL B, Adv.
Packag., 1994, 17, pp. 412-417
5 TIWARJ, S.: 'Transmission line delay limitations of laser bandwidths', IEE
Proc. Optoelectron., 1994, 141, pp. 163-166
6 NEIFELD, M.A., and CHOU, W-C: 'Electrical Packaging Impact on
Source Components in Optical Interconnects', IEEE Trans. Compon.,
Packag. Manuf. TechnoL, B, Adv. Packag., 1995, 18, pp. 578-595
7 HELMS, J., KURTZKE, C , and PETERMANN, K.: 'External feedback
requirements for coherent optical communication systems', J. Lightwave
TechnoL, 1992, 10, pp. 1137-1141
8 ROBERTSON, M.J., LEALMAN, I.F., and COLLINS, J.V.: 'The expanded
mode laser - a route to low cost optoelectronics', IEICE Trans. Electron.,
1997, E80C, pp. 17-23
9 GREEN, P.E.: 'Fibre optic networks', (Prentice Hall, Englewood Cliffs,
1993), p. 87
10 GOWAR, J.: 'Optical communication systems' (Prentice Hall, 1993), 2nd.
edn., p. 438
11 HALL, S.A., LANE, R., WANG, H.C., and GARERI, A.: 'Assembly of laser-
fibre arrays',/. Lightwave TechnoL, 1994, 12, pp. 1820-1826
12 PEALL, R.G., SHAW, B.J., AYLIFFE, P.J., PRIDDLE, H.F.M., BRICHENO,
T, and GURTON, P.: '1 x8, 8 Gbit/s transmitter module for optical space
switch applications', Electron. Lett, 1997, 33, pp. 1250-1252
13 CANN, R., HARRISON, P., and SPEAR, D.: 'Use of silicon vee groove
technology in the design and volume manufacture of optical devices',
Proc. SPIE, 1997, 3004, pp. 170-173
 Appendix 11
Tables of device parameters
and simulated performance for
DFB laser structures

Table A 11.1 Summary of material and device parameters used for large-signal
dynamic modelling of uniform-grating DFB laser

Vertical and horizontal waveguide parameters

Weff, vert 0.564 (0.548)* |xm effective width (vertical) for optical mode
dqw 0.05 jjun total width of quantum wells
&vert 37.6° (40.4°)* full-width half-maximum, FWHM for
emission from facet
^eff, hor 1.519 (1.587)* jxm effective width (horizontal) for optical
mode
w 2.0 \xm ridge width
@hor 24.8° FWHM for emission from facet
FMQW 0.085 (0.088)* overall confinement factor
d = 1.519x0.564=0.857 (0.869)* jxm2 effective optical area
O 2w (37.6°/180)x(24.8°/180) = 0.181 (0.191)* emission solid
angle (steradians)

* The static modelling used similar parameters but where they differ from those used
for the dynamic modelling they are shown in brackets

Gain, loss and linewidth parameters

6g/6N 3.0x10~ 18 cm 2 differential field-gain
Ntr 1.5x10 18 crrr 3 transparency density
3.0 linewidth-enhancement factor
€ 1.0x10" 17 cm 3 nonlinear-gain parameter
awg 30 cm" 1 waveguide power loss
 Appendix 11 393

Carrier recombination and IVBA parameters

A = 0 linear recombination (also written as 1 IT)
B = 1 x 10~10 crrfVs 'bimolecular' recombination—assumed
radiative
C = 1.3x 1O~28 cm6/s assumed nonradiative (Auger
recombination)
b = 2.0x10~ 1 7 cm 2 IVBA coefficient

Effective refractive indices

n9ff = 3.28 giving phase velocity vp=c/neff
neffg = 3.70 giving group velocity vg=c/neffg

Uniform grating parameters

K=100cirr 1 , KL=3, and L=300 fim
Direct-current drive with parasitic resistances/capacitors in general ignored

Table All.2 Material and device parameters used for modelling unstrained
(inGa)As-(InGa) (AsP) 6QWy 2x \J8DFB laser structure

Vertical and horizontal waveguide parameters

Weff, vert 0.56 fim effective width (vertical) for optical mode
dqw 0.039 jxr 6 QWs of 0.0065 |xm thickness
&vert ~ 47.3° full-width half-maximum, FWHM for emission
from facet
Weff,hor 1.83 fxm effective width (horizontal) for optical mode
w 2.5 |xm ridge width
#hor 22.7° FWHM for emission from facet
1 un\A/ = 0.0687 overall confinement factor
1.83 x 0.56= 1.02 Jim2 effective optical area
2TT (47.3°/180) x (22.7°/180) = 0.21 solid angle of emission in
steradians outside laser

Gain, loss and linewidth parameters

dg/d/V 3.0x10~ 1 6 cm 2 differential field gain
Ntr 1.5xi018crrr3 transparency density
aH 3.0 linewidth enhancement factor
€ = 1.0x10~ 17 cm 3 nonlinear-gain parameter
30cm~ 1 waveguide power loss
394 Distributed feedback semiconductor lasers

^guided optical mode

vert

rectangle with same area = (weffhorx weffvert)

as guided mode

Figure A 11.1 Waveguide parameters

Separate confinement parameters on left are relevant only to
multiquantum-well devices
Total thickness of active regions dmqw-Nqwdqw (N7W=number of quantum
wells)
Total confinement factor of mode to quantum wells Tmq
(Tqr0=confinement factor per well)

s)
source matching bond

R.

\
laser
bond-wire C X i
laser T '
laser

Figure A 11.2 Laser and submount parasitics: simplified model

I is drive current into the laser. The dynamic-laser-model circuit
parameters indicating the photon-electron resonance, which are
shown schematically in Figure A10.1, are not shown here because
these are included through the dynamic-modelling process
 Appendix 11 395

Carrier recombination and IVBA parameters

A = 0 linear recombination (also written as 1 /T>)
B = 1 x 1CT10 cm 3 /s 'bimolecular' recombination—assumed
radiative
C = 1.3 x 10" 2 8 cm 6 /s assumed nonradiative (Auger
recombination)
b = 2.0x1(T 1 7 cm 2 IVBA coefficient

Effective refractive indices

neff = 3.28 giving phase velocity vp-c/neff
neffg = 3.70 giving group velocity vg=c/neffg

Grating parameters
2 xA m /8 grating, #r=60cnrr 1 , KL=2A, L=350 jxm
Am/8 phase shifts positioned symmetrically in cavity and spaced 245 |jim apart,
i.e. AL=122.5 fxm

Carrier-transport parameters (see Appendix 4)

70 ps diffusion and capture time constant
140 ps escape and diffusion time constant
0.21 fi separate confinement heterostructure
width with two sections of equal length to
give
0.42 |mm total
0.56 confinement factor for SCH region
6n/6N\SCH = -1.5x10-20cm3 rate of change of refractive index with
electron density in SCH region

Laser and submount parasitics

Rlaser = 511 Claser 8.25 pF
*-bondwire =
1 .5 n H ^bondwire 0.2 il
'matching 43 ft R 50 O
 Appendix 12
About MATLAB programs

A12.1 Instructions for access

The programs accompanying this book were written in MATLAB 4,
but they have been designed to be compatible with the student edition
[1] so that a recommended 'minimum' specification for a system is:
MATLAB 4 Student Edition, 100 MHz PC/Mac with 32 MB RAM.
Users should note that these programs are supplied 'as is', on a free
basis and may be copied and modified by individual users but may not
be redistributed or used commercially without permission of the
copyright holder. While care has been exercised by the authors in
producing and testing these programs, the publishers and the authors
make no warranty for them, nor do they offer any maintenance,
support or advice for using these programs. The programs are for
tutorial purposes and not for commercial designs.
Before obtaining the programs from the World Wide Web, users are
recommended to create, within their main MATLAB directory, a
subdirectory (or folder), say laser, with further sub-directories dfb, diff,
fabpero, fftexamp, filter, grating, slab, spontan. Within MATLAB, the
effective initiation file is the M-file called matlabrc.m and this should
be backed up for future reference into a second file, say matlabrb.m.
On opening matlabrc.m, one looks for the lines of code with the
entries:
'C:\MATLAB\toolbox\matlab\demos;1,...
1);

and inserts, as in Table A12.1, the entries for interrogating the new
subdirectories. The punctuation marks at the start and ends of the
lines need to be followed in exact detail. MATLAB will then
automatically interrogate these directories/subdirectories on start up
 Appendix 12 397

Table A 12.1 Modifying matlabre. m file

'C:\MATLAB\toolbox\matlab\demos;',...
'C:\MATLAB\laser\dfb;1,...
'C:\MATLAB\laser\diff;1,...
'C:\MATLAB\laser\fabpero;1,...
etc.
"C:\MATLAB\laser\spontan;1,...

and the M-files entered into these subdirectories should be usable

directly. The instructions for automatically interrogating directories
may change with MATLAB versions and readers in general are advised
to consult the MATLAB user guide or help menu before starting. Users
who already make extensive use of MATLAB may have to ensure that
the names of directories/M-files downloaded from the WWW do not
clash with directories/M-files which are already installed.
To access the programs, users may click, in their Web browser, onto
the main IEE WWW books information page by entering appro-
priately:
http://www.iee.org.uk/publish/books/circsyst.html
Further links will lead users to the appropriate information about this
book and the directories containing the programs and detailed
information on how to load them. Different Web-browsers offer
different facilities and no information is given about these.
The information contained on the Web site combined with Table
A12.2 should aid the selection of the programs. Although the authors
have accessed these MATLAB programs from the Web site with both
Macintosh and PC hardware using MATLAB 4 and MATLAB 5, the
instructions cannot cover the complete variety of hardware and
software that readers may use. When accessing the MATLAB code from
the web, operators could make use of Microsoft Word, Notepad or
other text editor as the initial text editor and this can influence the
optimum method of access from the Web—e.g. cut-and-paste or direct-
save.
Within directories dfb or slab, it is unwise to attempt to run M-files
until all the interlinked M-files within each of these directories have
been collected. Within other directories most M-files are stand alone
and can be run without the aid of related M-files. M-files colortog and
delfig appear in a number of directories for convenience as discussed
below.
398 Distributed feedback semiconductor lasers

Table A 12.2 Directories (shown in bold) and M-files (*.m)

dfb diff fabpero ffexamp filter grating slab spontan

colortog.m advec.m fpdyna.m delfig.m colortog.m delfig.m group.m ampl.m
delfig.m delfig.m fpstat.m specti.m delfig.m dfbthr.m pinew.m ampif.m
dfb14.m stepj.m delfig.m spect2.m filti.m dispbrag.m p2thick.m amp2.m
dfb218.m stepr.m spect3.m filtin.m refl.m p3refind.m amp3.m
dfb238.m spect4.m filt2.m p4gain.m spont.m
dfbamp.m spect5.m gain.m p5layer.m delfig.m
dfbgain.m spect6.m p6bragg.m
dfbrin.m spect7.m p7layer.m
dfbuni.m p8calc.m
p.1las1.m p9search.m
p1las2.m piOrefl.m
p2comp.m pUserch.m
p3nein.m p12layer.m
p4run.m p13field.m
p4run2.m p14far.m
p5spont.m p15kappa.m
p6p!ot.m slabexec.m
p7plot.m colortog.m
p8plot.m
p9plot.m
sponto.m

A12.2 Introduction to the programs

To offer any program is to leave oneself open to criticism about
programming style, choice of language etc. The authors do not claim
to have any special flair in writing programs, and these are not
polished commercial code but endeavour to give the readers tutorial
text to help to develop their own work. There is an attempt to keep the
programs straightforward and compartmentalised and to have sub-
stantial annotation within each program. There is a small number of
programs within a suite of completely independent directories. In
each directory there is a readme file (stored for ease of identification
as 1readme.txt within each directory) which will provide information
about the programs within the directory.
The programs are within the scope of the inexpensive student
version of MATLAB 4 and usually use only the basic MATLAB features
even though the student version may allow more sophistication. The
authors have worked mainly with IBM-compatible PCs with clock rates
 Appendix 12 399

above 100 MHz, and it is worth remembering that enthusiasm can die
when results take too long to appear.
Random-access memory (RAM) is important. The MATLAB figures
produced by the code appear on our machines to use very roughly 1-2
Mbyte of memory, so that it is easy to find 16 Mbyte of memory being
used simply in storing a relatively small number of figures, and it seems
that 32 Mbyte of RAM is highly desirable for easy computation. For any
serious future user of this code, 64 Mbyte and 166 MHz clock rate is
recommended, remembering that even at the time of writing this is
already out of date! The M-file delfig, which is in several sub-
directories, permits the rapid deletion of figures to avoid clogging
memory.
Each machine, in our experience, can have slight differences in its
preferences about the directories in which the MATLAB codes are
stored. It appears to be safest to store the working programs actually
within a subdirectory within the MATLAB main directory. However,
above all, store the core programs in a back-up version on your own
machine to avoid continual reliance on the Net. Modify only copied
versions of this back-up and then back-up the modified versions!
Remember that programs are stored with the extension .m (e.g.
slabexec.m) but to run the program slabexec.m, for example, one
only types slabexec at the command-window prompt. The MATLAB
command window must, of course, be open and address the right
directory or it will simply return:
? Undefined function or variable slabexec
Ensuring that one has the right window is best accomplished by
clicking on
File , Run M-file, Browse
and then switching into the required directory and program as per
menu.
This appendix lists, under the different directories, the programs
which have been made available along with a very brief description of
them, and where sample outputs can be found within the book.
Opening the directory will give access to a README file about
programs in the directory. Opening the program as if to modify the M-
file will give access to a short account of the program, along with
'running' comments with the line header given by the symbol % .
Within the programs, there are default settings so that one usually
presses <return> or types return when asked to make an entry from
the keyboard. The default setting is then automatically returned and
400 Distributed feedback semiconductor lasers

the program should run. In this way it is hoped that the reader will at
least get started, but the real exercise for the reader is to both run and
modify the programs.
The contents of the various directories are briefly recounted below,
in approximately the order the programs appear in the book.

slab

The main program in the directory SLAB is SLABEXEC which permits

one to analyse slab waveguides with TE- (or TM-) field distributions for
an arbitrary number (~30) of layers of different complex permittiv-
ities including gain guides. Figure A12.1 shows a sequence of outputs.
The default mode of operation of slabexec plots out:
(a) five layers with slightly different refractive indices; then in the
next graph
( b) the log of the 'reflection coefficient' as defined in Figure 3.6, where
the troughs in the reflection indicate the region of the effective
refractive index for a more detailed search for the solution. If this
plot does not show clear sharp minima, it is likely that the guide
will not guide and a solution will not be found. Given clear
minima, the program looks for the detailed solution and plots
( c) the TE (TM)fieldintensity; and then finally plots
(d) the far-field pattern along with the values of the confinement
factors for the different modes.
Figures 3.7, 3.8 and 3.11 are typical of the output available from this
program.
Within slabexec there is the option to ask for a Bragg grating. This
follows the routines of Chapter 4 and Appendix 2, using a format of
input suggested by Figure A2.6. It will simply print appropriate values
of K on the waveguide patterns as estimated from the routines
discussed in Chapter 4.
Within directory slab there are two other minor programs:
(i) a program colortog which returns
Paper type has been set to A4
Figure set to black and white
This sets figures to black on white rather than white on black, and can
be helpful when producing figures for papers. It is also in other
directories.
 Appendix 12 401

slab layers - real ref. index - imaginary ref index

o
-0.5 3.4 •0.0002467
-1

-1.5 3.5 -0.0002467

-2

•2.5 3.6 0.002467

dopth
-3
3.4 •0.0002467
-3.5

•4
3.4 •0.0002467
-4.5

-5 0.1

2 3 4 -40 -20 0 20 40 60
T E E fields (mod squared) TE far fields - angular distribution

Figure A12.1 Sequence of plots obtained from the default mode for running slabexec

(ii) a program called group which provides data as in Appendix 3,

Figures. A3.1 andA3.2.
The other (linked) M-files within directory slab are used within
slabexec and are annotated either in the 1 readme file or within their
own code.

grating

There are three programs within a directory grating:

(i) Program dispbrag can help the reader to investigate the real and
imaginary parts of the propagation coefficient normalised to the
grating-coupling coefficient fij I #cl as a function of the detuning from
the Bragg condition normalised to the grating-coupling coefficient
S/\K\. This will print out figures similar to those in Figures 5.2 to 5.4.
(ii) Program ref I allows the reader to explore p= R/Fat the input for
402 Distributed feedback semiconductor lasers

Slab Layers - Real Ret. Index - Imaginary Ret Index

-0.5 3.4 -0.0002467

Bragg fill real / imag index

/ \^3'4 / I
-1

-1.5 \ 3 , /0.000246A / /
^0.0002467 '
-2 V
-2.5 3.6 0.002467
depth
-3

-3.5 . 3.4 -0.0002467 power gain (1/crn^)

-4 178.6

-4.5 3.4 -0.0002467

real imag.
-5
1 2 3 4
0.1 0.2 0.3 0.4 0.5
distance TE E Fields (mod squared)

Figure A12.2 Modified output from slabexec when inserting a Bragg grating
etched into the waveguide with an infill of different material
Note that guide is too wide and the overlap with the grating is
inadequate, leading to a low K

real K similar to Figure 5.5 which shows this reflectivity as a function of

8/1 K I for a real K, SO that one can see how the reflectivity falls off as
S/\K\ diverges from zero.
(iii) Program dfbthr returns two numbers: the approximate normalised
gain gL and offset-frequency parameter SL for a given KL>0.5
required to get to threshold in a uniform DFB laser. It can be helpful
in ensuring that, when running dfbamp in directory dfb, one is
operating with a gain that is perhaps half that required for lasing.

diff
(i) Programs stepr and stepj in directory diff integrate AF/ dz= zFand
dF/ dz = j zF, respectively, showing how step length alters the accuracy.
Figure 7.2 gives a sample of the output expected.
(ii) Program advec provides a basic test program for the advection
equations where the forward field is stored at all space points but for
only two time steps. Figure 7.3 provides a sample of the output
expected.

spontan
(i) Program spont in directory spontan gives the reader an indication
of the modelling process for spontaneous emission. It provides a
dynamic output of the static display, shown in Figure 7.5.
 Appendix 12 403

(ii) Programs amp1, amp2 and amp3 provide three short demonstra-
tions of generation and amplification based on an 'ideal' uniform laser
amplifier, demonstrating that spontaneous emission and gain are
linked. Program ampif is similar to amp 1 but has a filtered noise input
rather than relying on white spontaneous noise.

fftexamp

Programs specti through to spect7 in directory fftexamp are unre-

markable, but may help to refresh the reader about using the fast
Fourier transform. The associated 1 readme file gives more details.

filter

(i) Program filti models a numerical-frequency filter, and the use of a

white-noise input to demonstrate filtering is shown in the program
filtin. The display output is similar to that in Figure 7.9. An alternative
filtering is demonstrated in program filt2.
(ii) Program gain implements the application of the filter theory to a
laser amplifier with gain over a controllable narrow band. The display
output is similar to that in Figure 7.10.

dfb

This suite of programs gives a dynamic output from different types of

DFB lasers showing the output building up with time as the laser is
switched on; and then one can take the spectrum and/or ask for the
electron density and optical-field patterns with or without spontaneous
noise.
(i) Program dfbamp in the directory dfb models a DFB amplifier with
uniform gain with zero input on the right and a unit-amplitude field
entering on the left, so that the device acts as a coherent forward-wave
'amplifier'. The output is a dynamic version of the static display shown
in Figure 7.4 with a step field propagating into the laser from the left-
hand end and then becoming reflected as it travels along through the
laser so that one can see the process of the distributed feedback
dynamically while the laser is settling down to its steady state,
(ii) Program dfbuni brings the elements of the DFB-laser modelling
together for a uniform device with real K. Although this is the main
program about distributed-feedback lasers it is probably easier if one
simply wants to run a program to use dfb 14, as discussed below. As
404 Distributed feedback semiconductor lasers

outlined in the text, uniform-DFB lasers have a number of problems

and these physical problems of being unsure what mode will lase, as
discussed in Chapter 6, are reflected in the numerical programs. This
uncertainty in whether the higher- or lower-frequency mode of the
uniform-DFB laser will appear can be disconcerting to someone
starting to run a program and finding that the output differs from run
to run. The output form is similar to that shown in Figure 7.11.
The program drives a number of subprograms which are common
to the other forms of DFB laser, and these subprograms are best dealt
with in the 1 readme file. For example, there are material-parameter
subprograms, and a lookup table for spontaneous noise is used in
program spont2 in the directory dfb and called up as required.
(Hi) Program dfb14 uses much of dfbuni but now has a A/4 phase shift
in its centre. Figure 7.11 shows a typical output against time, a
spectrum and a field pattern (with the spontaneous noise on and
spontaneous noise off).
(iv) Program dfb218 in directory dfb again uses much of dfbuni but
now has two phase shifts of A/8, one on either side and offset from the
laser's centre.
(v) Program dfb238 also uses much of dfbgni but now has two phase
shifts, as for dfb218 but now each of 3A/8.
(vi) Program dfbgain yet again uses much of Dfbuni but now has a
complex grating with a weak gain coupling (that is carrier independ-
ent for simplicity) but enables the reader to explore the mode change
as the sign and magnitude of the gain component of the grating
change.
(vii) Program dfbrin looks at RIN for a A/ 4-phase-shifted DFB laser
using the standard parameters within the model and gives an output
similar to that in Figure A8.2.

fabpero
There are two MATLAB programs fpstat and fpdyna in directory
fabpero which are able to give the static and dynamic characteristics of
a Fabry-Perot-type laser using normalised rate equations discussed at
the end of Appendix 4. The outputs are similar to those in Figures 4.2
and 4.3.

A12.3 Reference
1 'MATLAB 1995, Student Edition' (Prentice Hall and The MathWorks Inc,
1995)
 Index

absorption 46, 51-2 fabrication 22-4

free carrier 60-61 Fourier components 20-22, 118
intervalence band 70 introduction 18-22
acceptors 48 order 20
advection equations 112-13, 214-17 period 130
amplifiers periodic gain 116-17
erbium doped 15-16 periodic permittivity 116, 121
ideal 374-7 phase matching 117-120
integration, w integration phase shifts 137-45
noise theory, see noise radiation loss 20-22, 121
aperture theory, see spontaneous reflection per period 22, 25-6, 129
emission reflectivity 137-9
AR (antireflection) coating, see facet second order 20-21, 24
reflectivity Fourier components 120
arrays of laser diodes 30, 287 radiation loss 121-4
astigmatic pattern of emission from laser shape of teeth 124-5
392-4 stopbands 133-5
attenuators 377-8 Bragg wavelength 20
Auger, see recombination Brewster angle 311
autocorrelation, see spontaneous buried heterostructures 10
emission,
correlation in space-time carrier induced index changes, see
refractive index
band-gap, see semiconductors carrier transport, see quantum wells
band tails 52-3 causality, see Kramers-Kronig
bands, see energy bands relationships
beam formation 76 charge carriers
BER (bit-error-rate) 17 clamping of density 105-6
Bernard and Duraffourg lasing variation with Fermi-level 48, 50-51,
condition 41, 48 153-4
black body radiation 49, see thermal diffusion damping 107
radiation in crystals
blaze of grating 18, 22 electrons 37-8
Bragg gratings 18-24, 114, 116-25 holes 38
complex-index variations 131-2, 183-9 lifetime of, see recombination time
fabrication 183-4 rate equation 100, 243-4
coupling of waves, see coupled-wave thermalisation of distribution 59-60
equations transport 102
diffraction order 18-22 chirp 6, 61-3, 71,108-9
dispersion relationship 133-5 clamping, see charge carriers
406 Distributed feedback semiconductor lasers

coherent detection 15 gain grating

complex complex index 116-17, 183-7,
dielectric constant 61, 78 190-91
refractive index 61, 78 dynamic effects 187-9
conduction band, see energy bands facet reflectivity effects 189
confinement static effects 184-7
carrier 6-7 high power 189-94
factor 11-12, 79-81, 98-100 high-low reflectivity 190
horizontal 9-11, 77, 90-91 integrated amplifier 191-4
photon 7-8 effects of carrier transport 197-202
vertical 77 light-current characteristic 155-8
connectorised lasers 390 phase shifts 139-45, 221
contact resistance, see resistance single X/4 137-9, 140-41, 142-3
contour integral, see Kramers-Kronig two X/8 138-9, 143-4, 149-52,
relationships 197-202
coupled cavity lasers, see laser diode simulated emission spectrum 159-61
(types) simulated static performance 155-61
coupled-mode equations, see coupled- uniform structure 140, 142, 147-8,
wave equations 194-7
coupled-wave equations 119-20, 129-32, matrix analysis 135-7
222-8, 243 oscillation modes 355-6
analytic/numeric solutions 224-5 Pauli matrix formulation 352-3
eigen-modes 132-3 reflectivity 355
finite difference scheme 219, 222-3 threshold conditions for lasing
gain coupling 220-21 355-6
MATLAB code 223-4 transmission 355
matrix formulation 135-7, 219-21, dielectric waveguide, see waveguides, slab
352-6 diffraction 18-22, 79, 90
coupling fibres to laser diodes 389-91 diffusion, effect on direct modulation
V-groove technology 390-91 107
critical angle 8 digital modulation, effects of facet
current spreading 9,31 reflections 181-3
cut-off in waveguide 95 direct bandgap, see semiconductors
direct modulation 1, 70, 176, 258-9
DBR, see laser diodes (types) discretisation 230-32
delay in turn on of light 108 dispersion in fibres 16-17, 195-6, 258-9
density of states, see energy bands dispersion relationship for grating 133-5
detuning, see frequency deviation distributions
device parameters for laser diodes 392-5 black body 49
carrier transport parameters 395 Boltzmann 41
confinement factors 392-4 Fermi-Dirac 40-41
effective mode widths 392-4
FWHM for emission 392-4 edge/surface emission comparison
gain saturation 392-3 287-8
gain/loss parameters 392-3 effective area 79-81, 98, 351
linewidth enhancement factor 392-3 effective index method for confinement
recombination parameters 393, 395 90-91
separate confinement layer thickness effective mass 38-9, 50-51
394-5 for electrons 38-9
DFB (distributed feedback) laser 17, for holes 39
26-8 effective permittivity 78
see also Bragg gratings effective refractive index 78, 84-6
asymmetric grating 191 efficiency of a laser 13, 31, 244
dynamic modelling 194-202 eigen-modes 132-3
 Index 407
Einstein relations, see spontaneous filter theory
emission application to gain filtering 237-41
electrical parasitics, see package for laser electron rate equation 245-6
diode Lorentzian 233-7
electroabsorption modulator 259-60 numerical implementation 235-7,
electromagnetic energy exchange 240-41
341-51 travelling waves 238-40
electron beam lithography 24, 285 FM (frequency modulation) response
electron concentration 48, 50-51, 153-4 336-8
emission, spontaneous, see spontaneous forward waves 2-3, 79, 111-15, 117-19,
emission 130, 136, 141-2, 214-16, 219-23,
emission, stimulated, see stimulated 305
emission free carrier absorption 60-61
energy, see also energy exchange frequency chirp, see chirp
electromagnetic 81, 341-51 frequency deviation 114, 130
exchange between energy levels 47 frequency shift with carrier-density, see
optical 97-126 chirp
rate of change 345
energy bands gain 2-5, 37-55
conduction 37-8 field 3-4, 101
density of states 39 guiding, see waveguiding mechanisms
heavy-hole 45 per unit distance 100-101
valence 38 per unit time 99-101
energy exchange 341-51 power 101
from rate equation 344-8 round-trip 4-5
reconciliation with field equation wave propagation with 3
348-51 gain-grating, see DFB
energy states 37-40 gain saturation 58-60, 102, 107-8
energy-momentum diagrams 38, 40, 42, parameter values 59
44, 45 quantum wells 68-9
erbium doped fibre amplifier (EDFA) 15 Green's function 122
error control (numerical) 211, 216-17, group refractive index 324-8
221-2 group velocity 327, 350
error program for threshold 355-6 Henry's alpha factor, see linewidth
evanescence 8, 86-7 enhancement factor
extinction ratio 261 heterojunction, see laser diode (types)
Hilbert transforms 61, 357-62
fabrication of DFB lasers 22-4 hole concentration expression 48, 51,
Fabry-Perot resonator 2-4, 209 153-4
facet reflectivity 2, 3-4, 311-12 holes, light and heavy 40, 45
antireflection coatings 15, 28 homojunctions, see laser diode (types)
high-low reflection 12-13, 28
spectral effects 179-82 ideal optical transmitter 254-5
far fields 92-3, 319-22 imaginary dielectric constant, see
Faraday effect 390 complex dielectric constant
feedback 4-5, 13, 14, 20-22, 25, 140, indirect energy gap, see semiconductors
276, 278, 280, 284, 285, index, see refractive index
see also DFB InGaAs, strained (980nm) 69
Fermi-Dirac statistics 46, 48 instability 145, 158, 191, 195, see also
FFT (fast Fourier transform) 210, 232-3 pulsations
Parseval's theorem 232 integration
fibre attenuation 15 laser and amplifier 191-3
fibre dispersion, see dispersion in fibres laser and modulator 203, 259-60
filamentation 9-10 inversion factor 378
408 Distributed feedback semiconductor lasers

^-selection 43-4, 50-53 linewidth

Kane theory, see transition rates enhancement 171-2
Kramers-Kronig relationships 61, dynamics 176
357-62 homogeneous broadening 53-4
Cauchy contour theory 359-60 influence of reflections 179
causality theory 357-9 inhomogeneous broadening 53-4
physical basis 360-62 Lorentzian 54, 167-8
laser diodes modulation increases 166-7
chirp 6, 28, 29, 61-4, 71, 108-9 power output effect 170
desirable features 1 rebroadening 176-7
early developments 1-6 Schawlow-Townes formula 170
electron-density-time characteristics static calculations 168-70
108-9 stored energy effect 169-70
impedance/resistance 102 systems effects 166-8
light-current characteristics 5-6, tunable lasers 284
105-6, 108 linewidth enhancement factor 61-4, 108,
longitudinal modes 2, 5 171
SMSR (side mode suppression ratio) changes with frequency 63-4
17, 167-8 definition 62
spectra 5 diagrams of variations 62, 63
temperature effects 64-6 effect of complex grating 174
threshold-temperature coefficient 66 effective value 172-6
transverse modes 2 estimates 175
tuning mechanisms 276 measurements 173-4
wavelength-temperature-coefficient numerical effects 242
64-5 quantum well effects 64
laser diode (types) loss, optical 3
cleaved-coupled-cavity 13-14
DBR (distributed Bragg reflector) 25
distributed feedback, see DFB Mach-Zehnder interferometer 263-5
double heterojunction 2, 6-7 MATLAB 396-404
external cavity 13-14 accessing programs 396-8
external grating 14-15 computing system requirements 396
Fabry-Perot 2-6 initiating file 397
Fabry-Perot variations 12-17 program listings 398
heterqjunctions version 4, student limitations 233
double 2, 6-8 MATLAB directory titles
single 2 dfb 191, 224, 241, 370, 403-4
homojunctions 1-2 diff 213, 217, 402
injection locked 14 fabpero 340, 404
surface emitting, see surface emitting fftexamp 232, 403
lasers filter 235-6, 403
tunable 29, 274-85 grating 135, 137,401-2
VCSEL, see surface emitting slab 87, 95, 119, 125, 183, 322, 400-1
lasers spontan 226, 229, 402-3
lattice matching 69 MATLAB directories (subject matter)
Lax averaging 213, 215, 218 advection equations, see diff
leakage current 7, 31 Bragg gratings, see grating
lifetime DFB numerical code, see dfb
carrier, see recombination time differential equations, see diff
intraband 59 FFT tutorials, see fftexamp
operational 1, 22 filter theory, see filter
photon 99, 104-5 gain guiding, see slab
spontaneous 54, 58 gain-frequency filtering, see filter
 Index 409

light-current characteristics for FP drive level 370

laser, wfabpero photon-electron resonance effects
rate equations for FP laser, s#?fabpero 370-71
slab waveguide, see slab shot noise 366
spontaneous emission, wspontan uncertainty as noise 373
MATLAB programs 396-404 normalised rate equations 106, 108, 244,
advec217 339-40
amp1-3 229 numerical distortion 216-7
ampif 230 numerical stability 216-7, 223
dfbamp 224 effects of discretisation 230-32
dfbgain 191 Fourier checks 221-2
dfbrin 370 Nyquist formula for noise, see noise
dispbrag 135 Nyquist spectral-band limits 230-32
filti 235 for FFT 232-3
filt2 236
fpdyna 340 obliquity factor 92-3, 320-22
f pstat 340 optical absorption 99
listings 398 optical component modelling methods
refl 137 253-5
slabexec 22, 87, 95,119, 124,125, beam propagation 253
183, 322 finite difference 209-48
spect1-7 232 finite elements 253
stepj 213 Fourier techniques 255
stepr 213 power matrix methods 161
matrix element for band-to-band TLLM (transmission line laser
transitions 43 modelling) 210, 253
Maxwell's equations 83, 111, 304 transfer matrix methods 141-5
MBE (molecular beam epitaxy) 22 optical gain, see gain
MOCVD (metal organic chemical vapour optical isolators 390
deposition) 22 optical systems 15-17, 252
mode locking 14, 253, 266 BER (bit-error-rate) 17
mode spacing 5, 279 modelling 255-6
modelling methods, see optical power budgets 17
components sources for WDM systems 285-6
modulation of lasers, see direct WDM (wavelength division
modulation multiplexing) 17,285-6
modulators, external with DFB lasers 189-90
electroabsorption 259-60 ordinary differential equations 211-14
prechirp effects 259 accuracy of solution 213-4
momentum central difference methods 212-13
electron 38 Euler methods 211-212
phonon 41-2 oscillation condition 2-4
photon 41 package for laser diode 386-9
equivalent circuit 387, 394
near-field pattern 94 values for 395
noise 372-84 monitor diode 388-9
attenuation 377-8 thermal considerations 388
CNR (carrier to noise ratio) 17 parabolic bands 39
ideal amplifier 374-7 particle balance 110
Nyquist formula for thermal noise 373 Pauli matrices 352-3
ultra-violet catastrophe 373 Petermann's factors 58, 175, 383
RIN (relative intensity noise) 17, phase fluctuations 172
366-71 phase jumps, see DFB laser, phase shifts
decision level 367-9 phase matching 4, 27
410 Distributed feedback semiconductor lasers

phase velocity 325-6 numerical formulation 214 et seq.

photon density (black body) 49 photon density, see rate equations
photon lifetime 99, 103,104-5, 329-30 recombination
photon-electron resonance 107-8 Auger 44-6, 100-101
plane waves 304-5 process diagram 45
plasma effect 63 temperature changes 46
polarisation 341-2 bandgap states 44
power amplification 193, 256-8 bandtails 44
propagation constant 3, 78, 91, 111-15, nonradiative 48, 100-101
324 radiative 41-3, 100-101
pulsations 148, 177, 188, 266 total 100-101
push-pull electronics 265-6 recombination time (Tr) 100-101, 107,
push-pull laser 265-74 330, 332
asymmetry 269-72 reflections, external
speed of response 272-4 coherence collapse 178-9
stopband-distance diagrams 267 laser stability 177-9
symmetry 266-9 packaging 389-90
time resolved chirp 269-72 reflections, internal, see facet reflectivity
transient modes 272-4 refractive index, see also effective
refractive index
quantum wells
carrier capture 70-71 carrier induced changes 171-2
carrier transport effects 334-8 complex value 61
AM and FM response 337-8 model for quaternary materials 362-4
carrier transport, parameter values plasma effect 361
395 single oscillator model 362-3
density of states 66-7 table of values 328
effect on waveguiding 93-5 temperature coefficient 65
effects of strain 69-70 resistance
enhancement of tuning-range 274-5 contact 48-9, 154
gain saturation 68-9 series 152-4
intervalence band absorption 70 reverse waves 2-3, 7, 115, 120,130, 136,
introduction 66-8 141-2, 214-15, 219-23, 305
multiple 68-9
SCH (separate-confinement saturation of gain, see gain saturation
heterostructure) 70-71 scattering losses 60
quasi-Fermi level 40-41, 48 SCH (separate confinement
heterostructure) 70-71, 94-5,
rate equations 98-110 394-5
carrier transport delay 334-8 selection rule for k, see ^-selection
damping 332-3 semiconductor optical amplifier (SOA)
dynamic analysis 106-10 253, 256-8, 260-64
electronic 100-103, 330 cross-phase wavelength conversion
large signal analysis 339-40 263-4
peak magnification of AM response four wave mixing 261-2
333 ideal optical transmitter 254-5
photon-electron resonance 332-3 instantaneous frequency 255, 257-8
photonic 98-100, 329 semiconductors
small signal analysis 106-7, 329-38 band-gap
FM response 336-7 direct 41-3
steady state solutions 104-6, 330-31 indirect 41-2
rates of change GaAlAs 2, 42
electric fields 111-16, 130 GaAsl,43
electron density, see rate equations GaPAs 2
energy, see energy shot noise, see noise
 Index 411

side mode suppression ratio 150-52, grating coupled DBR 287-8

167-8, 180-83, 197-201 VCSEL (vertical cavity surface-emitting
effect of phase shifts 150-52 laser) 30, 286-94
slab waveguide, see waveguides construction 291-4
Snell's law 8, 57, 305-7 polarisation 294
SOA, see semiconductor optical amplifier speed of response 291
spatial hole burning 181 susceptibility 342-4
changes with time 197-202 imaginary 343-4
emission frequency effects 146, 158 real 343
KL effects 148 switch on delay 108
lateral 145
light-current characteristic effects 146, TE (transverse electric) fields 308
158 special cases of reflection 309
longitudinal 145-52, 158 temperature, effects 64-6
spectral effects 150-52 thermal noise 373
phase-shift effects 148-50 thermal radiation 49, 373 et seq
spectra threshold (for lasing) 55, 66, 70
facet reflectivity effects 179-82 conditions 104
gain 50-54, 56 current 46, 67
optical gain 52, 53, 56 spectra 55
spontaneous emission 53-5 TLLM (transmission line laser
spectral hole burning 59-60 modelling) 210, 253
spectral-temporal interchange 245, TM (transverse magnetic) fields 309-10
349-50 special cases of reflection 310-11
spontaneous emission 4, 5, 43, 47, 54-5, total internal reflection (TIR) 8, 78,
99, 113-14, 130-31, 372-85 305-7
aperture theory 379-80 transfer matrix methods 141-5, 179-80,
autocorrelation, see correlation in 209
space-time transition rates 43-4
correlation in space-time 227-8, 375, Kane theory 43
381 transparency electron-density 48, 68
counting photon states 378-9 transport, diffusion effects 70-71
Einstein treatment 46-50, 372, 378-9 tunable laser diodes
magnitude 229 carrier density tuning 276-7
numerical modelling 226-9, 380-84 Franz-Keldysh effect 279
in DFB 241-2 linewidth 284
space-time correlation 381 modelling 284-5
numerical/physical bandwidth 382 nonuniform gratings 279-81
statistics 384 quantum confined Stark effect 279
travelling-waves 226-8 temperature tuning 276-7
tutorial programs 229-30 Vernier tuning 29, 280-82
spontaneous emission coupling tunable laser diodes (types) 274-85
explanatory diagram 57 GAVCF (grating assisted vertical-
factor 55-8,105-6, 108 coupler filter) 283
spontaneous excitation, see spontaneous interrupted gratings 279-80
emission multicontact 277-9
stimulated emission 47-9, 51-2 quantum well material 274-6
stopband-distance diagram 147-8 superstructure gratings 280
stored electromagnetic energy 343 tunable twin guide 283
strained material 69-70 Y-laser 282-3
stripe geometry lasers 9
surface emitting lasers uncertainty
advantages 286-7 amplification and uncertainty 374-7
comparison with edge-emitters 287-9 Heisenberg 44
412 Distributed feedback semiconductor lasers

noise 373-4 ridge 10-11

uniform DFB, see DFB, uniform structure slab 8, 81-90
multilayer 82-3
V-parameter 90, 325 reflection coefficient 86-7
valence band, see energy bands TE slab modes 313-16
VCSEL, see surface emitting lasers TM slab modes 317-18
velocity slab-waveguide program
group 112-13, 324-8,350
phase 112-13,324-8 322-3
vernier tuning, see tunable laser diodes waveguiding mechanisms
gain guiding 9, 78-9, 89-90
wave equation 83, 304 horizontal guiding 90-91
advection equation 214-18 index guiding 9, 77-9
central difference methods 215-18 quantum well effects 93-5
wave propagation 111-16 waveguiding modes
waveguide absorption values 392-3 scaling 90
waveguide modes TE (tranverse electric) 81-4
different orders 87-9 TM (transverse magnetic) 81-4
orthogonality 92 WDM (wavelength division
waveguides multiplexing), see optical systems
modelling 76-96
multilayer slab guides 313-18 Y-laser 29, 282-3
reflection coefficient at surface 316 zero dispersion 16
Distributed feedback semiconductor lasers
Distributed feedback (DFB) semiconductor lasers emit light in a single mode
which is essential to providing the carrier in long haul high bit-rate optical
communication systems. This comprehensive research monograph provides:
l thorough analysis of the operation and design of DFB lasers
l a high level of tutorial discussion with many valuable appendices
l the first full account of time-domain numerical modelling techniques
applicable to future optical systems as well as present devices
l Web access to a suite of MATLAB programs (student version MATLAB
4 or higher).
It is essential reading for those studying optical communications at graduate
and advanced under-graduate level, and a key book for industrial designers of
opto-electronic devices.

John Carroll is a Professor in the Department of Engineering and Head of the

Electrical Division at Cambridge University. He joined the university as a
Lecturer and Fellow of Queen’s College in 1967, forming a research group
studying Gunn and Trapatt microwave devices. He has authored three books
on semiconductor devices but for the last ten years the emphasis of his work
and publications has been the design and modelling of laser diodes. He is
currently Chairman of the Council of the School of Technology and a member
of the university’s General Board.
James Whiteaway graduated from Queen’s College, Cambridge, in 1973 and
joined Standard Telecommunications Laboratories (now Nortel) at Harlow,
UK, to work on laser diodes. He gained his PhD from Cambridge University in
1983 for his published work on semiconductor lasers. He is now the External
Research Co-ordinator for the Nortel Optical Communications Programme
Unit and leads a team researching optical device modelling. He has accomplished
pioneering work on phase-shifted DFBs which have set the industry standards
for performance of distributed feedback laser devices.
Dick Plumb graduated from Christ’s College, Cambridge, in 1973 after holding
a commission in the Royal Navy. Having gained his PhD on high-speed photo-
diodes, he joined the Standard Telecommunication Laboratories in 1977. Later
he worked for BT & D Technologies Ltd. on laser diode technology materials,
and took a leading role in a number of European programmes introducing
DFB lasers into optical communi­cation systems. He currently leads research
into high-power and tunable laser diodes at the Department of Engineering,
Cambridge University, which he joined as a lecturer and Fellow of Peterhouse
in 1991.

The Institution of Electrical Engineers

Michael Faraday House, Six Hills Way ISBN 978-0-852-96917-1
Stevenage, Herts., SGI 2AY, United Kingdom
ISBN 0 85296 917 1
Printed in the United Kingdom