InP-based polarisation independent wavelength demultiplexers
InP-based polarisation independent wavelength demultiplexers
InP-based polarisation independent wavelength demultiplexers
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>InP</strong>-<strong>based</strong> <strong>polarisation</strong> <strong>independent</strong><br />
<strong>wavelength</strong> <strong>demultiplexers</strong>
The cover shows a microscope image of a<br />
phased-array demultiplexer manufactured<br />
in an erbium-implanted Al 2 O 3 waveguide<br />
structure, see Chapter 6, section 6.4.<br />
(photo: Eduard de Kam / Hollandse Hoogte)
<strong>InP</strong>-<strong>based</strong> <strong>polarisation</strong> <strong>independent</strong><br />
<strong>wavelength</strong> <strong>demultiplexers</strong><br />
PROEFSCHRIFT<br />
TER VERKRIJGING VAN DE GRAAD VAN DOCTOR<br />
AAN DE TECHNISCHE UNIVERSITEIT DELFT,<br />
OP GEZAG VAN DE RECTOR MAGNIFICUS PROF. DR. IR. J. BLAAUWENDRAAD,<br />
IN HET OPENBAAR TE VERDEDIGEN TEN OVERSTAAN VAN EEN COMMISSIE,<br />
DOOR HET COLLEGE VAN DEKANEN AANGEGEWEZEN,<br />
OP DINSDAG 2 DECEMBER TE 13:30 UUR<br />
DOOR<br />
Cornelis van DAM<br />
elektrotechnisch ingenieur,<br />
geboren te ’s-Gravenhage.
Dit proefschrift is goedgekeurd door de promotor:<br />
Prof. dr. ir. H. Blok<br />
Samenstelling promotiecommissie:<br />
Rector Magnificus, voorzitter<br />
Prof. dr. ir. H. Blok, Technische Universiteit Delft, promotor<br />
Dr. ir M.K. Smit, Technische Universiteit Delft, toegevoegd promotor<br />
Prof. dr. G.A. Acket, Technische Universiteit Eindhoven<br />
Prof. dr. ir. R.G.F. Baets, Universiteit Gent<br />
Prof. dr. ir. W. van Etten, Universiteit Twente<br />
Prof. dr. ir. H.J. Frankena, Technische Universiteit Delft<br />
Prof. dr. B.H. Verbeek, Philips Optoelectronics<br />
Part of this work was supported by the Dutch Ministry of Economic Affairs (IOP Electro-<br />
Optics). Also parts of this work have been carried out in connection with the RACE 2070<br />
MUNDI project and the ACTS AC 065 BLISS project.<br />
The work described in Chapter 4 has been performed in close cooperation with Philips<br />
Optoelectronics Centre, Eindhoven, The Netherlands.<br />
The work described in Chapter 6, section 6.4, has been performed in close cooperation with the<br />
FOM-Institute for Atomic and Molecular Physics, Amsterdam, The Netherlands.<br />
van Dam, Cornelis<br />
<strong>InP</strong>-<strong>based</strong> <strong>polarisation</strong> <strong>independent</strong> <strong>wavelength</strong> <strong>demultiplexers</strong> /<br />
Cornelis van Dam. - [S.l. : s.n.] - Ill. -<br />
Ph.D. Thesis Delft University of Technology. - With ref. - With summary in Dutch.<br />
ISBN 90-9010798-3<br />
NUGI 832<br />
Keywords: Integrated optics / optoelectronics<br />
Copyright © 1997 Cor van Dam<br />
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system,<br />
or transmitted in any form or by any means - optical, electronic, magnetic, mechanical, or any<br />
other recording system - without the prior written permission from the publisher.<br />
Typeset by FrameMaker 5; printed by Ponsen & Looijen, Wageningen, The Netherlands
Aan Renée
Contents<br />
1 Introduction: WDM in optical communication networks 1<br />
1.1 Optical communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1<br />
1.2 Recent developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2<br />
1.3 Phased-array <strong>demultiplexers</strong> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5<br />
1.4 About this thesis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7<br />
2 PHASAR <strong>demultiplexers</strong>: a review 9<br />
2.1 Basic operation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9<br />
2.1.1 Focusing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10<br />
2.1.2 Dispersion and Free Spectral Range. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12<br />
2.1.3 Insertion loss and non-uniformity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13<br />
2.1.4 Bandwidth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14<br />
2.1.5 Channel cross talk. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15<br />
2.1.6 Polarisation dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17<br />
2.2 Phased-array design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18<br />
2.2.1 Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18<br />
2.2.2 Demultiplexer design procedure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18<br />
2.2.3 Array geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19<br />
2.2.4 Design for <strong>polarisation</strong> independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21<br />
2.2.5 Design for flattened response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23<br />
2.2.6 Design for low loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24<br />
2.2.7 Device size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25<br />
2.2.8 Correction for bending effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28<br />
2.2.9 Waveguide junctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29<br />
2.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29<br />
2.3.1 Wavelength routers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30<br />
2.3.2 Multi-<strong>wavelength</strong> receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30<br />
2.3.3 Multi-<strong>wavelength</strong> lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30<br />
2.3.4 Wavelength-selective switches and add-drop multiplexers . . . . . . . . . . . . . . 31<br />
2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
viii Contents<br />
3 Polarisation conversion in waveguide bends 35<br />
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35<br />
3.2 Computational methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37<br />
3.2.1 Effective index method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37<br />
3.2.2 Method of lines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38<br />
3.2.3 Finite element method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38<br />
3.2.4 Conformal transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39<br />
3.3 Attenuation in bends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40<br />
3.4 Tilting of the modal <strong>polarisation</strong> plane in bends . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44<br />
3.5 Polarisation dispersion in bends. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47<br />
3.6 Polarisation conversion at junctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49<br />
3.7 A compact <strong>polarisation</strong> converter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51<br />
3.8 Tolerance analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55<br />
3.8.1 Layer thickness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55<br />
3.8.2 Waveguide width . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55<br />
3.9 Avoiding <strong>polarisation</strong> conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61<br />
3.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62<br />
4 Polarisation <strong>independent</strong> PHASAR <strong>demultiplexers</strong><br />
<strong>based</strong> on nonbirefringent waveguides 63<br />
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63<br />
4.2 Nonbirefringent waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65<br />
4.2.1 The embedded square waveguide. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66<br />
4.2.2 The raised-strip waveguide. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67<br />
4.2.3 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69<br />
4.2.4 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71<br />
4.3 Loss reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71<br />
4.4 Ghost images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73<br />
4.4.1 Ghost image mechanism. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74<br />
4.4.2 Verification. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75<br />
4.4.3 Junction optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76<br />
4.4.4 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77<br />
4.4.5 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78<br />
4.5 Demultiplexer integrated with detectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78<br />
4.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81<br />
5 Polarisation <strong>independent</strong> PHASAR <strong>demultiplexers</strong><br />
<strong>based</strong> on birefringent waveguides 83<br />
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83<br />
5.2 Polarisation dispersion compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84<br />
5.2.1 Production tolerance analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86<br />
5.2.2 Coupling between array waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88<br />
5.2.3 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89<br />
5.2.4 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91<br />
5.3 Polarisation conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92<br />
5.3.1 Cross talk tolerance analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Contents ix<br />
5.4 Polarisation splitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93<br />
5.4.1 Cross talk tolerance analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94<br />
5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95<br />
6 MMI-<strong>based</strong> components 97<br />
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97<br />
6.2 MMI-MZI demultiplexer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98<br />
6.2.1 Operation principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99<br />
6.2.2 Tolerance analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105<br />
6.2.3 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111<br />
6.2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117<br />
6.3 Spatial mode filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117<br />
6.3.1 Operation principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118<br />
6.3.2 Tolerance analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119<br />
6.3.3 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121<br />
6.3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121<br />
6.4 Optical imaging of MMI patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122<br />
6.4.1 Upconversion mechanism. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122<br />
6.4.2 Measurement principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123<br />
6.4.3 MMI-coupler simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124<br />
6.4.4 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125<br />
6.4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125<br />
6.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125<br />
7 Discussion and conclusions 127<br />
A Overview of phased-array <strong>demultiplexers</strong> produced 129<br />
B Dispersion of the phased array 133<br />
C Waveguide mode effective width 135<br />
D Cross talk penalty of the tunable laser 137<br />
D.1 Measurement with a tunable laser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137<br />
D.2 Measurement with an EDFA as source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138<br />
D.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139<br />
E Excitation coefficient of the first-order mode 141<br />
F Phase correction in MMI-MZI <strong>demultiplexers</strong> 143<br />
References 145<br />
List of symbols and acronyms 157
x Contents<br />
Summary 161<br />
Samenvatting 163<br />
Dankwoord 165<br />
Biography 167
Chapter 1<br />
Introduction: WDM in optical<br />
communication networks<br />
This chapter gives a brief overview of the development of optical communications and places<br />
the subject of this thesis in the right context. Finally, an outline of the contents of this thesis is<br />
presented.<br />
1.1 Optical communication<br />
Optical fibre communication provides a solution for the growing demand for bandwidth in<br />
telecommunication systems. Two major developments - the laser and the fibre - triggered<br />
optical communication. In 1960 the first laser (Light Amplification by Stimulated Emission of<br />
Radiation) was developed by Theodore Maiman [79,85], using a pink ruby medium. In 1962<br />
Robert Hall and others followed with the invention of the semiconductor laser [85], a device<br />
now used in optical fibre communication systems, compact disk players and laser printers. The<br />
usage of optical fibres as a suitable transmission medium for light was suggested by Kao and<br />
Hockham in 1966 [72].<br />
The first fibres, however, measured transmission losses in the order of 1000 dB/km. The breakthrough<br />
took place in 1970, when Corning Glass scientists Robert Maurer, Donald Keck and<br />
Peter Schultz developed an optical fibre [71,85], capable of carrying 65,000 times more information<br />
than conventional copper wire, with optical losses of 20 dB/km - low enough for practical<br />
use in telecommunication systems. Within nine years of intensive research, this loss was<br />
narrowed to below 0.2 dB/km. By the mid-eighties fibres with attenuation minima of less than<br />
0.4 dB/km at 1.3 μm <strong>wavelength</strong> and less than 0.25 dB/km at 1.55 μm were commercially<br />
available. At that time reliable AlGaAs semiconductor lasers and Si detectors were already<br />
available, and full-scale implementation of fibres in the telephone network started. The first<br />
generation used graded-index multimode fibres with a core diameter of 50 μm in the<br />
transmission window of around 850 nm, determined by the <strong>wavelength</strong> range accessible by<br />
AlGaAs. The second generation uses the longer <strong>wavelength</strong>s (1.3 and 1.55 μm) in combination<br />
with monomode fibre, in this way enabling higher transmission capacities. For these<br />
<strong>wavelength</strong>s other semiconductor material was needed: In 1-x Ga x As 1-y P y . By varying the
2 1. Introduction: WDM in optical communication networks<br />
material composition x and y, materials with <strong>wavelength</strong>s in the entire range from 0.92 μm to<br />
1.65 μm and lattice-matched to <strong>InP</strong> can be obtained.<br />
Nowadays, optical fibre is the foundation for global multimedia telecommunications networks.<br />
More than 90 percent of the U.S. long-distance traffic is already carried via optical fibre, and<br />
more than 25 million kilometres have been installed. In the Netherlands more than 18,000<br />
kilometres fibre have been installed. Globally, over 40 million kilometres have been installed,<br />
which, if strung together, could reach 28 times from the earth to the moon and back.<br />
1.2 Recent developments<br />
The reason which makes the fibre enormously attractive is the huge bandwidth it offers in<br />
combination with high transmission speeds. At first, direct detection was used for point-topoint<br />
links, a basic system consisting of a laser, a detector and optical fibre. A more advanced<br />
principle is the coherent transmission scheme [49], in which the incoming (modulated) signal<br />
is mixed with the signal of a local oscillator laser. The advantage of this transmission scheme is<br />
the improved sensitivity and more efficient usage of bandwidth. However, the fabrication<br />
complexity of coherent receivers and the tight requirements on the local oscillator laser,<br />
restricted the attraction of this transmission scheme. Moreover, erbium-doped fibre amplifiers<br />
(EDFA) [41,82] have been developed, which also provide these advantages.<br />
The alternative is found in the concept of Wavelength Division Multiplexed (WDM) direct<br />
detection, as shown schematically in its simplest form in figure 1.1. At the transmitter side a<br />
<strong>wavelength</strong> multiplexer is used to combine the signals from lasers operating at different<br />
<strong>wavelength</strong>s. The combined signals are launched into the fibre and at the receiver side a<br />
<strong>wavelength</strong> demultiplexer is used to separate them again and to couple them to photodiodes.<br />
The diagram depicted in figure 1.1 is a typical example of a simple point-to-point link. But the<br />
most important advantage of WDM - apart from the huge transmission capacity (recently the<br />
1 Tb/s barrier has been broken [97]) - is that it allows for novel network architectures which<br />
offer much more flexibility than the current networks [23,24].<br />
λ mux λ demux<br />
fibre<br />
transmitters receivers<br />
Figure 1.1 Schematic diagram of a WDM link.<br />
In present networks all switching functions are performed in the electrical domain, for which<br />
opto-electronic conversion is required at each switch. This conversion can be avoided by<br />
switching in the optical domain, <strong>based</strong> on <strong>wavelength</strong>, using <strong>wavelength</strong> routers. In this way a
1.2 Recent developments 3<br />
passive transparent network can be created, of which an example is shown in figure 1.2. One<br />
user can send data to the desired user(s) by selecting the <strong>wavelength</strong> to transmit on.<br />
λ 1 ,λ 2<br />
R<br />
Wavelength routers, first reported by Dragone [44,45], provide an important additional<br />
function as compared to multiplexers and <strong>demultiplexers</strong>. Figure 1.3 illustrates their function.<br />
Wavelength routers have N input and N output ports. Each of the N input ports can carry N<br />
different <strong>wavelength</strong>s. The N <strong>wavelength</strong>s carried by input channel 1 are distributed among<br />
output channels 1 to N, in such a way that output channel 1 carries <strong>wavelength</strong> 1 and channel N<br />
<strong>wavelength</strong> N. The N <strong>wavelength</strong>s carried by input 2 are distributed in the same way, but<br />
cyclically rotated by one channel in such a way that <strong>wavelength</strong>s 2-4 are coupled to ports 1-3<br />
and <strong>wavelength</strong> 1 to port 4. In this way each output channel receives N different <strong>wavelength</strong>s,<br />
one from each input channel. To realise such an interconnectivity scheme in a strictly nonblocking<br />
way using a single <strong>wavelength</strong>, a huge number of switches would be required. Using<br />
a <strong>wavelength</strong> router this function can be achieved using only one single component.<br />
For a passive network, one single <strong>wavelength</strong> is entirely dedicated over the whole network to a<br />
point-to-point link between two users. More effective usage of <strong>wavelength</strong>s is offered by <strong>wavelength</strong>-selective<br />
switches, or add-drop multiplexers (ADMs). Although a number<br />
configurations are possible, they basically consist of a <strong>wavelength</strong> demultiplexer (not<br />
necessarily a <strong>wavelength</strong> router) and switches, of which the number equals the number of<br />
λ 1<br />
λ 2 ,λ 3<br />
Figure 1.2 Schematic diagram of a passive transparent network (R = router).<br />
a 1 , a 2 , a 3 , a 4<br />
b 1 , b 2 , b 3 , b 4<br />
c 1 , c 2 , c 3 , c 4<br />
d 1 , d 2 , d 3 , d 4<br />
Wavelength<br />
Router<br />
a 1 , b 2 , c 3 , d 4<br />
a 2 , b 3 , c 4 , d 1<br />
a 3 , b 4 , c 1 , d 2<br />
a 4 , b 1 , c 2 , d 3<br />
transmission<br />
(a) (b)<br />
λ 3<br />
a b c d a b c d<br />
<strong>wavelength</strong><br />
Figure 1.3 Schematic diagram illustrating the operation of a <strong>wavelength</strong> router:<br />
interconnectivity scheme (a), and <strong>wavelength</strong> response (b). It is noted that a i<br />
denotes the signal at input port a with <strong>wavelength</strong> i.
4 1. Introduction: WDM in optical communication networks<br />
<strong>wavelength</strong> channels. In figure 1.4 the configuration as produced by Tachikawa et al.<br />
[144,146,149] is shown. The demultiplexer has one main input port and one main output port,<br />
and the other outputs are routed back to the inputs of the demultiplexer using switches. By<br />
feeding these switches, signals can either be tapped from the line (and simultaneously new<br />
ones can be put on the line), or be passed through.<br />
main input<br />
mux/<br />
demux<br />
switch<br />
switch<br />
switch<br />
switch<br />
main output<br />
λ4 . . λ1 λ1 . . λ4 drop ports add ports<br />
Figure 1.4 Schematic diagram of an ADM.<br />
Whereas the ADM acts as an access node to the network, there is also the need for connecting<br />
local networks to other, possibly remote, networks. This can be achieved using an optical cross<br />
connect (OCC). In a multi-<strong>wavelength</strong> network using N <strong>wavelength</strong>s, a local network can<br />
transparently be connected to N-1 remote networks through such a cross connect. A schematic<br />
diagram is shown in figure 1.5, in which the thick and thin lines denote the multi<strong>wavelength</strong><br />
and single-<strong>wavelength</strong> channels respectively. The OCC consists of N <strong>demultiplexers</strong>, N<br />
multiplexers and N switching matrices of dimension N × N . In this example the OCC is also<br />
equipped with integrated optical amplifiers to firstly compensate for the insertion losses of the<br />
devices on chip, and to secondly balance the powers in the different <strong>wavelength</strong> channels.<br />
demux<br />
demux<br />
demux<br />
demux<br />
optical<br />
amplifiers<br />
switching<br />
matrix<br />
Figure 1.5 Schematic diagram of an optical cross connect.<br />
mux<br />
mux<br />
mux<br />
mux
1.3 Phased-array <strong>demultiplexers</strong> 5<br />
The future WDM communication network will consist of OCCs, ADMs and <strong>wavelength</strong><br />
routers, as depicted schematically in figure 1.6. By deploying more ADMs, the network can<br />
handle an increasing number of users, which is an important advantage of such a network.<br />
Another advantage is the enhanced reliability: a defect node can be “bypassed” by changing<br />
the connectivity scheme. However, this strategy fails if the node serves a large number of users,<br />
in which case other strategies are required such as e.g. dual homing on two nodes.<br />
A<br />
A<br />
1.3 Phased-array <strong>demultiplexers</strong><br />
O<br />
O<br />
Figure 1.6 WDM communication network (A = ADM, R = Router, O = OCC).<br />
It is obvious that <strong>wavelength</strong> (de)multiplexers are key components in WDM systems. Many<br />
principles have been introduced and reported for the manufacture of multiplexers and<br />
<strong>demultiplexers</strong>. Commercially available components are <strong>based</strong> on fibre-optic or micro-optic<br />
techniques [106,107] and, recently, also fully integrated devices. Since the early nineties,<br />
research on integrated-optic (de)multiplexers has increasingly been focused on grating-<strong>based</strong><br />
and phased-array (PHASAR) <strong>based</strong> devices (also called Arrayed-Waveguide Gratings) [165].<br />
Both are imaging devices, i.e. they image the field of an input waveguide onto an array of<br />
output waveguides in a dispersive way. In grating-<strong>based</strong> devices, a vertically etched reflection<br />
grating provides the focusing and dispersive properties required for demultiplexing. In phasedarray<br />
<strong>based</strong> devices these properties are provided by an array of waveguides, the length of<br />
which has been carefully chosen to obtain the required imaging and dispersive properties. As<br />
phased-array <strong>based</strong> devices are manufactured according to conventional waveguide technology<br />
and do not require the vertical etching step needed in grating-<strong>based</strong> devices, they appear to be<br />
more robust and production tolerant.<br />
Phased-array <strong>demultiplexers</strong> were introduced in 1988 by Smit [116]. The first devices<br />
operating at short <strong>wavelength</strong>s were reported by Vellekoop and Smit [161-163,117]. Takahashi<br />
et al. reported the first devices operating in the long <strong>wavelength</strong> window [151,152]. Dragone<br />
extended the phased-array concept from 1 × N to N ×<br />
N devices - the so-called <strong>wavelength</strong><br />
routers [44,45] which play an important role in multi<strong>wavelength</strong> network applications. The<br />
broad acceptance of the phased-array concept can be seen from the bargraph in figure 1.7. In<br />
this graph the development of the phased-array demultiplexer is depicted by the number of<br />
publications on single devices (including PHASAR-<strong>based</strong> system experiments, which started<br />
A<br />
O<br />
A<br />
A
6 1. Introduction: WDM in optical communication networks<br />
in 1993 [144], leading to a more exponential growth of the number of publications). A<br />
complete overview of realised phased-array <strong>demultiplexers</strong> can be found in Appendix A.<br />
1<br />
’89<br />
1<br />
’90<br />
2<br />
3<br />
’91 ’92 ’93<br />
’94 ’95<br />
Figure 1.7 Development of phased-array <strong>demultiplexers</strong>, measured by the<br />
number of publications on single devices.<br />
Devices reported so far can be divided into two main classes: silica-<strong>based</strong> devices and <strong>InP</strong><strong>based</strong><br />
devices. Most of the silica-<strong>based</strong> devices use fibre-matched (low-contrast) waveguide<br />
structures, which combine low propagation loss with a high fibre-coupling efficiency. More<br />
recently silicon-<strong>based</strong> polymer devices [56,131] and lithiumniobate devices [95,96] were<br />
reported which also used fibre-matched waveguide structures.<br />
Silica-<strong>based</strong> devices have relatively large dimensions due to the low index contrast and the<br />
corresponding large bending radii of the fibre-matched waveguides. This makes them less<br />
suitable for integration of large numbers of components on a single chip. Furthermore, silica<br />
has a limited potential for integration of active functions due to its passive character. The<br />
feasibility of integration with switches, however, was demonstrated by Okamoto who reported<br />
successful integration of thermo-optical switches with phased arrays in an integrated optical<br />
add-drop multiplexer [89,90].<br />
The first <strong>InP</strong>-<strong>based</strong> PHASAR-demultiplexer was reported in 1992 by Zirngibl et al. [176]. <strong>InP</strong><strong>based</strong><br />
devices have a better potential for integration of active functions. They exhibit<br />
considerably higher propagation and fibre-coupling losses, the latter however due to the small<br />
size of the waveguide cross section. Despite the higher propagation losses, the total on-chip<br />
device loss can be kept within acceptable limits due to the small component size which is<br />
possible because of the high index contrast and which also allows for integrating larger<br />
numbers of components onto a chip. <strong>InP</strong>-<strong>based</strong> <strong>demultiplexers</strong> cannot compete with silica-,<br />
lithiumniobate-, or polymer-<strong>based</strong> devices with respect to fibre coupling loss, which makes<br />
them less suitable for production of low complexity circuits. Their main advantage is their<br />
potential for monolithic integration of active components such as switches [167], detectors<br />
[4,5,9,138-140,182], optical amplifiers and modulators [68-70,178-181,184], and their<br />
potential to integrate large numbers of components onto a single chip.<br />
In figure 1.8 the possible future development of opto-electronic chips for WDM applications is<br />
depicted. At the moment micro-optic, fibre-optic or hybrid solutions are available, which in the<br />
near future however will be complemented by integrated solutions. The drive for integration is<br />
found in the reduction of packaging costs, mainly determined by the number of<br />
interconnections, and in an increased reliability. Integrated devices are still in research stage<br />
but promise to provide the higher feasability which will be required in future<br />
telecommunication networks. The first silicon-<strong>based</strong> devices were recently introduced to the<br />
market. These fibre-matched waveguide structures (silica, lithiumniobate and polymer) have<br />
low coupling losses, but on the other hand also low index contrast, which limits the integration<br />
4<br />
7<br />
17<br />
17<br />
’96
1.4 About this thesis 7<br />
Integration scale<br />
100<br />
MSI<br />
10<br />
SSI<br />
scale. These materials therefore will only reach the small-scale integration stage (SSI), or the<br />
lower part of the medium-scale integration stage (MSI). Due to the large index contrast, <strong>InP</strong><strong>based</strong><br />
devices can be very compact, and therefore will be able to reach the MSI stage. The high<br />
index contrast too is a cause for the high fibre-chip coupling losses, and therefore attention is<br />
now focused on improving the coupling efficiency.<br />
1.4 About this thesis<br />
1<br />
micro-optic<br />
fibre-optic<br />
hybrid<br />
1995 2000 2005 2010<br />
Year<br />
As phased-array <strong>demultiplexers</strong> are key components in future WDM networks, a detailed<br />
description of the operation and design of these components is described in Chapter 2, which<br />
has been published in the Journal of Selected Topics in Quantum Electronics [118]. The design<br />
strategy as described by Smit [117] has been followed, but as nowadays parameters such as<br />
channel spacing are given in terms of frequency rather than <strong>wavelength</strong>, the equations<br />
describing the PHASAR properties have been derived in the frequency domain. Also, elaborate<br />
attention has been paid to a number of different design requirements, such as <strong>polarisation</strong> independence,<br />
flattened <strong>wavelength</strong> response, low-loss operation, etc. Furthermore, an overview of<br />
the most important applications is given.<br />
The major part of this thesis is focused on methods for making PHASARs <strong>polarisation</strong><br />
<strong>independent</strong>. The state of <strong>polarisation</strong> of the transmitted light after propagation through a fibre<br />
over a long distance is unknown and varies in time. It is therefore important that the transfer of<br />
the demultiplexer to which the fibre is connected is insensitive to such variations. One way of<br />
making a PHASAR <strong>polarisation</strong> <strong>independent</strong> is by inserting <strong>polarisation</strong> converters in the<br />
middle of the array. It has been found that very short waveguide bends have <strong>polarisation</strong>converting<br />
properties, which can be applied in compact, low-loss <strong>polarisation</strong> converters. An<br />
extensive discussion of the <strong>polarisation</strong> conversion phenomenon in waveguide bends is<br />
presented in Chapter 3, using different computational methods. Also experimental results are<br />
discussed, part of which have been presented elsewhere [34,35], and which have led to a patent<br />
application [32].<br />
There are several methods of making <strong>demultiplexers</strong> <strong>polarisation</strong> <strong>independent</strong> and the most<br />
obvious one is to make use of <strong>polarisation</strong> <strong>independent</strong> waveguides. This method is discussed<br />
in Chapter 4. The use of non-birefringent waveguides has been initiated by Amersfoort [7], and<br />
has been extended with an elaborate analysis of the tolerances, and a discussion of<br />
experimental results, part of which have been published [164]. These experiments clearly<br />
demonstrate that <strong>polarisation</strong> <strong>independent</strong> behaviour can be achieved over a large <strong>wavelength</strong><br />
<strong>InP</strong><br />
silica lithiumniobate polymer<br />
Figure 1.8 Future development of opto-electronic ICs for WDM.
8 1. Introduction: WDM in optical communication networks<br />
range. This is an important advantage of these waveguides, together with low fibre-chip<br />
coupling losses and compact device dimensions. Furthermore, a method for loss reduction has<br />
been proposed and demonstrated [33], and the occurence of so-called “ghost” images has been<br />
investigated [37]. Finally, a packaged device integrated with photodetectors is described.<br />
Instead of making the waveguides <strong>polarisation</strong> <strong>independent</strong>, PHASAR <strong>demultiplexers</strong> can also<br />
be made <strong>polarisation</strong> <strong>independent</strong> by adaptions of the device itself. An example of this is the<br />
compensation of the <strong>polarisation</strong> dispersion of the array waveguides, as proposed by Takahashi<br />
et al. [154]. This method is discussed extensively in Chapter 5, including a tolerance analysis,<br />
and has been applied to our waveguide structure, of which experimental results are detailed.<br />
Although this method is sensitive to width and layer thickness variations, the manufacture is<br />
relatively simple, as a selective wet-chemical etchant can be used for the <strong>polarisation</strong><br />
dispersion compensating section in the PHASAR. Additionally, two methods which use<br />
<strong>polarisation</strong> conversion and splitting respectively are discussed with emphasis on cross talk tolerance.<br />
The order-matching method is discussed briefly, as a more comprehensive account of<br />
this method including experiments is given by Spiekman [134].<br />
Novel components <strong>based</strong> on Multimode Interference (MMI) are presented in Chapter 6. A<br />
novel type of phased-array demultiplexer is one in which the free propagation regions are<br />
replaced with multimode interference (MMI) couplers. Firstly in this chapter, the design of<br />
such MMI-<strong>based</strong> <strong>demultiplexers</strong> is discussed, together with a tolerance analysis. Experimental<br />
results are detailed, part of which have been published elsewhere [31]. Secondly a novel type<br />
of spatial mode filter <strong>based</strong> on a single MMI coupler is discussed. The discussion includes a<br />
tolerance analysis which reveals the advantages of this device - compact size, large optical<br />
bandwidth, and tolerant to width variations. The experimental results, which have already been<br />
presented [36], demonstrate the low-loss operation combined with a high first-order mode<br />
rejection. Finally, the interference patterns in a MMI coupler with a high erbium concentration<br />
are made visible by imaging the green light generated by upconversion of erbium atoms at high<br />
pumping power levels. In this way, the field intensity of the infrared light can be imaged with a<br />
resolution below the theoretical diffraction limit. This is demonstrated with a MMI-coupler<br />
manufactured in an aluminum oxide ridge waveguide structure implanted with erbium. This<br />
work has been carried out as a common project with the FOM-Institute for Atomic and<br />
Molecular Physics, and has been presented earlier [30,59].<br />
In Chapter 7 the results presented in this thesis are shortly highlighted and discussed.
Chapter 2<br />
PHASAR <strong>demultiplexers</strong>: a review<br />
Wavelength multiplexers, <strong>demultiplexers</strong> and routers <strong>based</strong> on optical phased arrays play a key<br />
role in multi<strong>wavelength</strong> telecommunication links and networks. In this chapter a detailed<br />
description of phased-array operation and design is presented and an overview is given of the<br />
most important applications. Part of this work has also been published in the Journal of<br />
Selected Topics in Quantum Electronics [118].<br />
2.1 Basic operation<br />
Figure 2.1a shows the schematic layout of a PHASAR-demultiplexer. The operation is<br />
understood as follows. When the beam propagating through the transmitter waveguide enters<br />
the Free Propagation Region (FPR), it is no longer laterally confined and becomes divergent.<br />
On arriving at the input aperture, the beam is coupled into the waveguide array and propagates<br />
through the individual array waveguides towards the output aperture. The length of the array<br />
waveguides is chosen in such a way that the optical path length difference between adjacent<br />
waveguides equals an integer multiple of the central <strong>wavelength</strong> of the demultiplexer. For this<br />
<strong>wavelength</strong> the fields in the individual waveguides will arrive at the output aperture with equal<br />
phase (apart from an integer multiple of 2π), and the field distribution at the input aperture will<br />
be reproduced at the output aperture. The divergent beam at the input aperture is thus<br />
transformed into a convergent one with equal amplitude and phase distribution, and an image<br />
of the input field at the object plane will be formed at the centre of the image plane. The<br />
dispersion of the PHASAR is due to the linearly increasing length of the array waveguides,<br />
which will cause the phase change induced by a change in the <strong>wavelength</strong> to vary linearly<br />
along the output aperture. As a consequence, the outgoing beam will be tilted and the focal<br />
point will shift along the image plane. By placing receiver waveguides at proper positions<br />
along the image plane, spatial separation of the different <strong>wavelength</strong> channels is obtained.<br />
In the following subsections the most important properties of a PHASAR will be analysed.
10 2. PHASAR <strong>demultiplexers</strong>: a review<br />
2.1.1 Focusing<br />
array<br />
waveguides<br />
Δα<br />
input<br />
aperture<br />
object<br />
plane<br />
transmitter<br />
waveguide<br />
FPR<br />
R a<br />
d a<br />
output<br />
aperture<br />
(a)<br />
output aperture<br />
Focusing is obtained by choosing the length difference ΔL between adjacent array waveguides<br />
equal to an integer number of <strong>wavelength</strong>s, measured inside the array waveguides:<br />
whereby m is the order of the phased array, λ c (f c ) is the central <strong>wavelength</strong> (frequency) in<br />
vacuo and N eff is the effective index of the waveguide mode. With this choice the array acts as<br />
d r<br />
image plane<br />
(b)<br />
image<br />
plane<br />
R a /2<br />
Figure 2.1 a) Layout of the PHASAR demultiplexer<br />
b) Geometry of the receiver side.<br />
θ<br />
FPR<br />
focal line<br />
s (measured along<br />
the focal line)<br />
receiver<br />
waveguides<br />
ΔL m λc mc<br />
= ⋅ -------- = --------------<br />
(2.1)<br />
N eff N eff f c
2.1 Basic operation 11<br />
a lens with image and object planes at a distance R a of the array apertures.<br />
The input and output apertures of the phased array are typical examples of Rowland-type<br />
mountings [81]. The focal line of such a mounting, which defines the image plane, follows a<br />
circle with radius R a /2 as shown in figure 1b. Transmitter and receiver waveguides should be<br />
positioned on this line.<br />
α/αi starting angle/starting angle of array waveguide i<br />
αr reference angle<br />
β/βFPR propagation constant in waveguide/FPR<br />
c speed of light<br />
D dispersion<br />
da pitch of the array waveguides in the array aperture<br />
dr pitch of the receiver waveguides in the image plane<br />
Δα divergence angle of the array waveguides in the array aperture<br />
Δfch channel spacing<br />
ΔfFSR free spectral range<br />
Δfpol TE-TM shift<br />
ΔfL L-dB bandwidth<br />
ΔΦ phase shift between adjacent array waveguides<br />
ΔL path length difference between adjacent array waveguides<br />
Hi height of array waveguide i<br />
k0 wave number in vacuo<br />
L distance between the focal points<br />
Lo central channel insertion loss<br />
Lp propagation loss in the array<br />
Lu non-uniformity<br />
l(α) path length of array waveguide with respect to α<br />
li path length of array waveguide i<br />
λc /fc central (design) <strong>wavelength</strong>/frequency<br />
m/m' order of the array/beam<br />
N/Na number of channels/array waveguides<br />
Neff /NFPR effective index in waveguide/FPR<br />
Ng group index<br />
NTE /NTM effective index for TE/TM <strong>polarisation</strong><br />
N1 /N2 transverse effective index in the ridge/in the region next to the ridge<br />
Ra length of FPR<br />
Ri radius of curvature of array waveguide i<br />
Rr radius of curvature of reference array waveguide<br />
Si Sr s<br />
straigth section length (including Ra ) of array waveguide i<br />
straigth section length (including Rr ) of reference array waveguide<br />
position along image plane<br />
θ dispersion angle<br />
θa array aperture half angle<br />
θο angular width of the far field<br />
V lateral V-parameter (normalised film parameter)<br />
w/we waveguide width/effective width<br />
Table 2.1 List of symbols used in this chapter.
12 2. PHASAR <strong>demultiplexers</strong>: a review<br />
2.1.2 Dispersion and Free Spectral Range<br />
In figure 2.1b it can be seen that the dispersion angle θ resulting from a phase difference ΔΦ<br />
between adjacent waveguides follows as:<br />
whereby ΔΦ = βΔL , β and βFPR are the propagation constants in the array waveguide and<br />
the Free Propagation Region (FPR) respectively, and da is the lateral spacing (on centre lines)<br />
of the waveguides in the array aperture.<br />
The dispersion D of the array is described as the lateral displacement ds of the focal spot along<br />
the image plane per unit frequency change. From figure 2.1b it follows (after some<br />
manipulation, see Appendix B) that:<br />
whereby f c = c ⁄ λc is the central frequency, NFPR is the (slab) mode index in the Free<br />
Propagation Region, ΔL is the length increment of the array waveguides as described before,<br />
Δα = da ⁄ Ra is the divergence angle between the array waveguides in the fan-in and fan-out<br />
sections and Ng is the group index of the waveguide mode:<br />
It can be seen that R a does not occur in the right hand expression in equation 2.3 so that fillingin<br />
of the space between the array waveguides near the apertures due to a finite lithographical<br />
resolution does not affect the dispersive properties of the demultiplexer.<br />
From equation 2.2 it can be seen that the response of the phased array is periodical. After each<br />
change of 2π in ΔΦ, the field will be imaged at the same position. The period in the frequency<br />
domain as shown in figure 2.2b is called the Free Spectral Range (FSR). It is found as the<br />
frequency shift for which the phase shift ΔΦ equals 2π:<br />
from which we derive:<br />
ΔΦ m2π<br />
θ arcsin – ( ) β ⎛<br />
⁄ FPR<br />
--------------------------------------------- ⎞ ΔΦ – m2π<br />
=<br />
≈ -------------------------<br />
⎝ d ⎠<br />
a<br />
βFPRd a<br />
D<br />
with m' = ( N g ⁄ N eff)<br />
⋅ m.<br />
The rightmost identity, which is well known from grating theory, follows by substituting<br />
N gΔL =<br />
m'c ⁄ f c (see equation 2.1). It is noted that for phased arrays, different from gratings,<br />
the Free Spectral Range is not related to the order m of the array, but to a modified order<br />
number m', which can be interpreted as the order of the beam.<br />
As the exact relation between θ and ΔΦ is non-linear (see equation 2.2), equation 2.6 is only<br />
approximate and the FSR will be slightly dependent on the input and output ports. An accurate<br />
analysis is given by Takahashi et al. [155].<br />
(2.2)<br />
ds dθ 1<br />
----- R<br />
d f a ⋅ ----- ---d<br />
f<br />
N g<br />
----------- ΔL<br />
= = = ⋅ ⋅ -------<br />
(2.3)<br />
Δα<br />
f c<br />
N g N eff f dN eff<br />
= + -----------d<br />
f<br />
N FPR<br />
2πΔ f FSR<br />
---------------------N<br />
c gΔL = 2π<br />
Δf FSR<br />
f c<br />
(2.4)<br />
(2.5)<br />
c<br />
= ------------- = ----<br />
(2.6)<br />
N gΔL m'
2.1 Basic operation 13<br />
2.1.3 Insertion loss and non-uniformity<br />
Figure 2.2a shows the field in the image plane for four different <strong>wavelength</strong>s. It is the sum field<br />
of the far fields of all individual array waveguides. As the far-field intensity of the individual<br />
waveguides reduces away from the centre of the image plane, as indicated in the figure, the<br />
focal sum field will do the same. If the <strong>wavelength</strong> is changed, it will move through the image<br />
plane and follow the envelope described by the far-field of the individual array waveguides. If<br />
we approximate the modal field of the array waveguides as a Gaussian beam and neglect the<br />
effects of coupling on the beam shape, we can derive some simple analytical equations for<br />
estimating insertion loss, channel non-uniformity and bandwidth.<br />
Using the Gaussian-beam approximation, the intensity of the far-field is found in:<br />
I( θ)<br />
Ioe 2<br />
=<br />
– θ2 2<br />
⁄ θo whereby θ o is the width of the equivalent Gaussian far field<br />
θ o<br />
λ 1<br />
= ----------- ⋅ ---------------we<br />
2π<br />
N FPR<br />
with we being the effective width of the modal field in the transmitter waveguide (as described<br />
in Appendix C). The non-uniformity Lu is defined as the intensity ratio (in dB) between the<br />
outer and the central channel. Using equation 2.7, the insertion loss of the receiver relative to<br />
the central channel is easily found by substituting the angle θmax ( θmax ≈ smax ⁄ Ra )<br />
corresponding to the outer receiver waveguide:<br />
L u<br />
– θ 2<br />
max<br />
If the FSR is chosen as being equal to N times the channel spacing Δf, as in <strong>wavelength</strong> routers<br />
(see section 2.3.1), the excess loss Lu of the outer channels will be close to 3 dB for reasons of<br />
power conservation - as for large numbers of channels, receiver waveguide 1 and the virtual<br />
receiver N+1 will experience approximately the same loss with each of them having at least<br />
3 dB excess loss relative to the central channel. For small values of N the situation may be<br />
slightly better. Thus minimising Lu will increase the FSR.<br />
The insertion loss Lo of the central channel is mainly determined by diffraction of light into<br />
undesired orders. The adjacent orders of the main focal spot will carry a fraction<br />
2 2<br />
exp( – 2Δθ FSR ⁄ θo) , with:<br />
(2.7)<br />
(2.8)<br />
whereby D is the dispersion (equation 2.3). If we neglect the power coupled into other orders<br />
the total loss L o can be estimated from:<br />
L o<br />
whereby it has been assumed that exp( – 2Δθ<br />
FSR ⁄ θo) «<br />
1 . The factor 4 is due to the fact that<br />
2<br />
θo – 10 e 2 ⎛ ⁄ ⎞ 2<br />
= ⋅ log<br />
≈ 8.7 ⋅ θ<br />
⎝ ⎠ max ⁄<br />
Δθ FSR<br />
ΔsFSR -------------<br />
Ra 2<br />
θo (2.9)<br />
D<br />
= = (2.10)<br />
2<br />
2<br />
θo -----Δf<br />
R FSR<br />
a<br />
2 2<br />
da<br />
– 2Δθ<br />
FSR ⁄<br />
– 10 ⋅ log⎛1<br />
– 4 ⋅ e ⎞ + L<br />
⎝ ⎠ p 17 e 4π – we ⁄<br />
≈ ≈ ⋅ + L (2.11)<br />
p<br />
2<br />
2
14 2. PHASAR <strong>demultiplexers</strong>: a review<br />
power is lost in two orders and that equal losses occur (because of reciprocity) at both the input<br />
and the output side of the array. The term L p denotes the total propagation loss in the array and<br />
both FPRs due to absorption and scattering. From this equation it can be seen that for low-loss<br />
devices the waveguide spacing d a in the array apertures should be minimal. For semiconductor<strong>based</strong><br />
devices, the best total loss reported is in the order of 2 dB [141]. It should be noted that<br />
equation 2.11 is a worst-case guess - coupling between the array waveguides will reduce the<br />
loss as discussed in section 2.2.6.<br />
intensity<br />
100%<br />
order m-1<br />
λ4 λ1 λ4 λ1 image plane<br />
λ4 λ1 channel 4 1 4 1<br />
frequency<br />
4 1<br />
(a) (b)<br />
Figure 2.2 Central insertion loss, non-uniformity and free spectral range (FSR).<br />
The 100% line denotes the peak intensity of the input field.<br />
2.1.4 Bandwidth<br />
central insertion loss L o<br />
order m order m-1 order m order m+1<br />
non-uniformity L u<br />
far field<br />
envelope<br />
order m+1<br />
If the <strong>wavelength</strong> is changed, the focal field of the PHASAR moves along the receiver<br />
waveguides. The frequency response of the different channels follows from the overlap of this<br />
field with the modal fields of the receiver waveguides. If we assume that the focal field is a<br />
good replica of the modal field at the input, and that the input and output waveguides are<br />
identical, the (logarithmic) transmission T(Δf) around the channel maximum T(fc ) follows as<br />
the overlap of the modal field with itself, displaced over a distance Δs( Δf ) = DΔf :<br />
whereby U(s) is the normalised modal field, D is the dispersion as defined in equation 2.3 and<br />
T(f c ) is the transmission in dB at the channel maximum. For small values of Δs (smaller than<br />
the effective mode width w e ) the overlap integral can be evaluated analytically by<br />
+∞<br />
FSR<br />
T ( Δf ) = T ( f c)<br />
+ 20log U( s)U<br />
( s– DΔf<br />
)ds<br />
∫<br />
-∞<br />
(2.12)
2.1 Basic operation 15<br />
approximating the modal fields as Gaussian fields:<br />
⎛ ⎛ ⎞⎞<br />
T ( Δf ) – T ( f c)<br />
20 ⋅ log⎜<br />
exp⎜–<br />
------------ ⎟⎟<br />
– 6.8 ⎛DΔf ---------- ⎞<br />
2<br />
⎝ ⎝ ⎠⎠<br />
⎝ w ⎠<br />
e<br />
2<br />
=<br />
≈ ⋅<br />
The L-dB bandwidth Δf L is twice the value Δf for which T ( Δf ) – T ( f c)<br />
= L dB:<br />
Δf L<br />
The latter identity follows by substitution of D = dr ⁄ Δ f ch . If we substitute we ⁄ dr ≈ 0.4 as a<br />
representative value (cross talk due to receiver waveguide spacing
16 2. PHASAR <strong>demultiplexers</strong>: a review<br />
Truncation<br />
Another source of cross talk results from truncation of the field due to the finite width of the<br />
array aperture. This causes power to be lost at the input aperture, whilst at the output aperture<br />
the side lobe level of the focal field will increase. For a proper PHASAR design, the array<br />
aperture angle should be chosen in such a way that the corresponding cross talk is sufficiently<br />
low. Figure 2.3b shows the transmitted power (solid line) and the cross talk versus the array<br />
aperture half angle θ a , defined as:<br />
θ a<br />
( N a – 1)d<br />
a<br />
= -------------------------<br />
normalised to the Gaussian width θ o as defined in equation 2.8, for different values of the<br />
relative receiver waveguide spacing d r /w. As the estimate of figure 2.3a is rather pessimistic, it<br />
is best to use the envelope depicted by the boldly dashed line. The values shown are calculated<br />
for input and output waveguides with V = 3. The dependence on the V-parameter is small. The<br />
envelope of the cross talk curves (boldly dashed line) can be used for estimating the maximum<br />
cross talk level. It is seen that for θ a >2θ o the truncation cross talk is less than -35 dB.<br />
Cross Crosstalk talk [dB]<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
-50<br />
-60<br />
V = 2<br />
V = 3<br />
V = 5<br />
Crosstalk Cross talk [dB]<br />
Mode conversion<br />
If the array waveguides are not strictly single mode, a first order mode excited at the junctions<br />
between straight and curved waveguides can propagate coherently through the array and cause<br />
“ghost” images. Because of the difference in the propagation constant between the<br />
fundamental and the first order mode, these images will occur at different locations and the<br />
“ghost image” may couple to an undesired receiver waveguide thereby degrading the cross talk<br />
performance. Mode conversion can be kept small by optimising the offset at the junctions for<br />
minimal first-order mode excitation.<br />
2R a<br />
(a) (b)<br />
(2.16)<br />
-60<br />
-10<br />
1 2<br />
d r r/w /w<br />
3 4<br />
0 1 2<br />
θ a a /θ /θ0o 3 4<br />
Figure 2.3 a) Cross talk resulting from the coupling between two adjacent<br />
receiver waveguides for different values of the lateral V-parameter of the receiver<br />
waveguides.<br />
b) Transmitted power (solid line) and cross talk as a function of the<br />
relative array aperture θ a /θ o , for different values of the relative receiver<br />
waveguide spacing d r /w (d r /w = 2.5, 3.0, 3.5). The values shown are calculated for<br />
input and output waveguides with V = 3. The boldly dashed line indicates the<br />
envelope of the cross talk curves (maximum cross talk level).<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
-50<br />
0<br />
-2<br />
-4<br />
-6<br />
-8<br />
Transmitted power [dB]
2.1 Basic operation 17<br />
Coupling within the array<br />
Cross talk can also be incurred by phase distortion due to coupling in the input and output<br />
sections in the arrays. It might be expected that this type of coupling will not greatly affect the<br />
focusing and dispersive properties of the array on similar grounds as to those mentioned in<br />
equations 2.3 and 2.4. The filling in of the gaps near the array apertures can be considered as<br />
introducing an extremely strong coupling into the input and output region, which obviously<br />
does not degrade the PHASAR performance [117]. Day et al. [38], however, observe a<br />
degradation of the cross talk performance using BPM-simulation.<br />
Phase transfer errors<br />
A fifth source of cross talk results from phase transfer errors in the phased array due to<br />
imperfections during the production process. The optical path length of the array guides is in<br />
the order of several thousands of <strong>wavelength</strong>s. Deviations in the propagation constant may lead<br />
to considerable errors in the phase transfer, and, consequently, to an increase of the cross talk<br />
level. Takada et al. [150] and Yamada et al. [170,171] have shown that improved cross talk is<br />
feasible by correcting the phase errors (e.g. by using thin-film heaters on top of the array<br />
waveguides). Phase errors may be caused by small deviations in the effective index due to local<br />
variations in composition, film thickness or waveguide width, or by inhomogeneous filling in<br />
of the gap near the apertures of the phased array. Also more systematic errors (e.g. due to discretisation<br />
in the mask pattern generation) may contribute to the cross talk [38].<br />
Background radiation<br />
As a last possible source of cross talk we mention background radiation due to light scattered<br />
out of the waveguides at junctions or rough waveguide edges. This is especially important in<br />
waveguide structures where the light is also guided besides the waveguides e.g. in shallow<br />
etched ridge guides or in waveguide structures on a heavily doped substrate where the undoped<br />
buffer layer may also act as a waveguide.<br />
Cross talk in practical devices is not limited by design but by imperfections during the<br />
production process. Typical cross talk values reported for PHASAR-<strong>demultiplexers</strong> are in the<br />
order of -25 dB for <strong>InP</strong>-<strong>based</strong> devices to even less than -30 dB for silica-<strong>based</strong> devices. Recent<br />
experiments in our laboratory, however, show cross talk levels less than -30 dB for good<br />
semiconductor devices as well, which is outlined in Appendix D. Improvement on these<br />
figures is mainly a matter of improving manufacturing technology.<br />
2.1.6 Polarisation dependence<br />
Phased arrays are <strong>polarisation</strong> <strong>independent</strong> if the array waveguides are <strong>polarisation</strong><br />
<strong>independent</strong>, i.e. the propagation constants for the fundamental TE- and TM-mode are equal.<br />
Waveguide birefringence, i.e. a difference in propagation constants, will result in a shift Δf pol<br />
of the spectral responses in respect of each other, which is called the <strong>polarisation</strong> dispersion. It<br />
can be calculated when we consider the effective <strong>wavelength</strong>s in the waveguide. Light with<br />
different <strong>wavelength</strong>s in vacuo will be coupled into the same receiver waveguide, when the<br />
effective <strong>wavelength</strong>s λ TE and λ TM of the fundamental modes in the waveguide are equal:<br />
λTM( f )<br />
c<br />
f ⋅ N TM f<br />
c<br />
--------------------------- λ<br />
( ) TE( f – Δ f pol)<br />
( f – Δ f pol)<br />
⋅ N<br />
TE f Δ f = = = --------------------------------------------------------------------- (2.17)<br />
( – pol)<br />
whereby N TE and N TM are the effective indices for both <strong>polarisation</strong>s.
18 2. PHASAR <strong>demultiplexers</strong>: a review<br />
By solving Δf pol from equation 2.17 we find:<br />
whereby N g,TE is the group index. For InGaAsP/<strong>InP</strong> DH waveguide structures, Δf pol is<br />
typically in the order of 4-5 nm. For silica-<strong>based</strong> and, more generally, for low-contrast<br />
waveguides, it will be much smaller. Also, in waveguides structures which are designed for<br />
<strong>polarisation</strong> independence, <strong>polarisation</strong> dependence may occur due to strain induced during the<br />
manufacturing process. A number of methods to reduce <strong>polarisation</strong> dependence will be<br />
mentioned briefly in section 2.2.4, and will be discussed more comprehensively in Chapter 4.<br />
2.2 Phased-array design<br />
2.2.1 Specification<br />
( N TE – N TM)<br />
Δ f pol ≈ f ⋅ -------------------------------<br />
N g,TE<br />
A PHASAR is specified by the following characteristics:<br />
• Number of channels N.<br />
• Central frequency f c and channel spacing Δf ch .<br />
• L-dB Channel bandwidth Δf L .<br />
• Free Spectral Range Δf FSR .<br />
• Maximal insertion loss L o of the central channel.<br />
• Maximal non-uniformity L u .<br />
• Maximal cross talk level.<br />
• Maximal <strong>polarisation</strong> dependence.<br />
It is noted that the non-uniformity and the Free Spectral Range can not be chosen<br />
<strong>independent</strong>ly from each other (see section 2.3.1)<br />
2.2.2 Demultiplexer design procedure<br />
(2.18)<br />
PHASARs have many degrees of design freedom and many design approaches are possible.<br />
The approach followed at TU Delft for designing multiplexers and <strong>demultiplexers</strong> is explained<br />
below. It starts with a given waveguide structure (i.e. waveguide width w and lateral Vparameter<br />
fixed). The design parameters of the PHASAR are derived subsequently from the<br />
design specifications. The design procedure of a <strong>wavelength</strong> router is slightly different (see<br />
section 2.3.1). For the reader’s convenience figure 2.1b is shown here again.<br />
• Receiver waveguide spacing dr . We start with the cross talk specification, which puts a<br />
lower limit on the receiver waveguide spacing dr . As with today’s technology cross talk<br />
levels lower than -30 to -35 dB are difficult to realise, it does not make sense to design the<br />
array for much lower cross talk. To be on the safe side, we take a margin of 5-10 dB and<br />
read from figure 2.3a the ratio dr /w required for -40 dB cross talk level. It is noted that the<br />
cross talk for TE- and TM-<strong>polarisation</strong> may be different as the lateral index contrast and,<br />
consequently, the lateral V-parameter can differ substantially for the two <strong>polarisation</strong>s.<br />
• FPR length Ra . From the maximum acceptable excess loss for the outer channel (the nonuniformity<br />
Lu ), we determine the maximum acceptable dispersion angle θmax using<br />
equations 2.8 and 2.9. The minimal length Ra,min of the Free Propagation Region then<br />
follows as Ra,min =<br />
smax ⁄ θmax whereby smax is the s-coordinate of the outer receiver<br />
waveguide (see figure 2.1b).
2.2 Phased-array design 19<br />
array<br />
waveguides<br />
Δα<br />
• Length increment ΔL. First we compute the required dispersion of the array from<br />
D = ds ⁄ d f = dr ⁄ Δ f ch (see equation 2.3). The waveguide spacing da in the array<br />
aperture should be chosen as small as possible (a large spacing will lead to high coupling<br />
losses from the FPR to the array and vice versa). With da and Ra fixed, the divergence angle<br />
Δα between the array waveguides is fixed as Δα = da ⁄ Ra (see figure 2.1b) and the length<br />
increment ΔL of the array follows from equation 2.3.<br />
• Aperture width θa . The angular half width θa of the array aperture should be determined<br />
using a graph like figure 2.3b (adapted for the specific waveguide structure used).<br />
• Number of array waveguides Na . The choice of θa fixes the number of array waveguides:<br />
N a = 2θaR a ⁄ da + 1 (see equation 2.16).<br />
This completes the determination of the most important geometrical parameters of the<br />
PHASAR.<br />
2.2.3 Array geometry<br />
R a<br />
d a<br />
output aperture<br />
For the array waveguides a number of different shapes can be applied to realise the length<br />
increment ΔL. Takahashi et al. [152] used the geometry as depicted in figure 2.5a, which is<br />
very simple from a design point of view, with a constant R for all array arms. Smit [117] and<br />
Dragone [44] applied the geometry of figure 2.5b, which contains a minimum number of<br />
waveguide junctions. This is especially important in semiconductor waveguides where<br />
junction losses and mode conversion at junctions can degrade the PHASAR performance.<br />
The freedom in the choice of array shape is bounded by the requirement that the array<br />
waveguides should not come too close to each other. For extremely low dispersion values (e.g.<br />
for duplexing 1.3 and 1.55 μm), the path length difference ΔL between adjacent waveguides<br />
becomes very small and the shapes depicted in figure 2.5 are no longer suitable. Adar et al. [1]<br />
applied S-bend-like arrays in which the dispersion of one curved section is reduced by a<br />
second section with opposite curvature and, consequently, opposite ΔL. Mestric et al. [83,84]<br />
recently also reported a S-bend shaped phased-array 1.31-1.55 μm duplexer with very low<br />
cross talk (
20 2. PHASAR <strong>demultiplexers</strong>: a review<br />
(a) (b)<br />
Figure 2.5 Phased-array waveguide geometries.<br />
The geometry as shown in figure 2.5b is also used for the design of the phased-array<br />
<strong>demultiplexers</strong> discussed in this thesis and will be described below. According to this<br />
geometry, an array guide consists of two straight waveguides connected to each other by a<br />
curved waveguide, together forming a non-concentric set.<br />
S i<br />
R a<br />
α i<br />
input<br />
aperture<br />
α i<br />
L<br />
Each array guide is completely defined by the starting angle α i , the radius of curvature R i and<br />
the straight section length S i , which includes the focal length R a for ease of calculation,<br />
according to:<br />
R i+1<br />
R i<br />
output<br />
aperture<br />
Figure 2.6 Geometry of the i-th guide of the phased-array.<br />
S i<br />
R i<br />
L ⁄ 2 – Sicosαi = ---------------------------------sinαi<br />
Lα i<br />
1<br />
-- ⎛l 2 i – ----------- ⎞ 1<br />
⎝ sinα ⎠<br />
i<br />
αicosαi =<br />
⁄ ⎛ – ------------------ ⎞<br />
⎝ sinα<br />
⎠<br />
i<br />
H i<br />
Δα<br />
(2.19)<br />
(2.20)
2.2 Phased-array design 21<br />
and:<br />
αi = α1 + ( i – 1)Δα<br />
whereby l i is the path length of the i-th array guide measured from transmitter to receiver:<br />
li = l1 + ( i – 1)ΔL<br />
In the above equations Δα is the divergence angle of the array waveguides in the array aperture<br />
and ΔL is the path length difference between adjacent array waveguides.<br />
2.2.4 Design for <strong>polarisation</strong> independence<br />
(2.21)<br />
(2.22)<br />
Several methods can be used for eliminating the <strong>polarisation</strong> dependence of the response due<br />
to waveguide birefringence. Five different methods will be briefly discussed here and more<br />
comprehensively in Chapter 4.<br />
Nonbirefringent waveguides<br />
The most obvious way to make a PHASAR <strong>polarisation</strong> <strong>independent</strong> is by eliminating the<br />
birefringence of the waveguide. This can be done by making the waveguide cross section<br />
square if the index contrast is the same in both the vertical and lateral direction as, for example,<br />
in buried waveguide structures. Bellcore [11,123] recently reported a <strong>polarisation</strong> <strong>independent</strong><br />
device <strong>based</strong> on a buried InGaAsP/<strong>InP</strong> waveguide structure with a low-contrast waveguide<br />
core (small InGaAs-fraction). Philips and TU Delft, Bellcore, and Alcatel [9,19,20,21,122,<br />
164] reported several devices <strong>based</strong> on a raised strip guide using similar material for the<br />
waveguide core. An advantage of the raised strip guide is that, due to the high lateral index<br />
contrast, it allows for very short bending radii and, consequently, for compact design. This<br />
method will be discussed more comprehensively in Chapter 4.<br />
Attempts have been made to compensate the birefringence of conventional “flat” waveguide<br />
structures by applying strained MQW-waveguides. Bi-axial compressive strain, obtained by<br />
decreasing the Ga-fraction, increases the birefringence, whereas bi-axial tensile strain reduces<br />
it. First results of this method show that <strong>polarisation</strong> dispersion changes in the order of<br />
7-12 nm are possible [166]. A complication of this approach is that the intrinsic birefringence<br />
of MQW-structures is considerably higher than that of quaternary bulk material and requires<br />
very high strains for it to be compensated. This makes the approach sensitive to well-width and<br />
composition control.<br />
The birefringence problem occurs also in silica-<strong>based</strong> waveguides, where it is due to strain<br />
induced by the different thermal expansion coefficients of silica and silicon. It can be reduced<br />
by using silica substrates instead of silicon substrates [142].<br />
Order matching<br />
The first attempt to make PHASARs <strong>polarisation</strong> <strong>independent</strong> was <strong>based</strong> on matching the FSR<br />
to the <strong>polarisation</strong> dispersion as shown in figure 2.7 [127,133,162,163,177]. If the FSR is<br />
chosen to be equal to the <strong>polarisation</strong> dispersion the m-th order beam for TE will overlap with<br />
the TM-polarised beam of order m-1, which makes the response virtually <strong>polarisation</strong><br />
<strong>independent</strong>. From equation 2.6 it can be seen that this is obtained by choosing:<br />
c<br />
ΔL = -------------------<br />
N gΔ f<br />
(2.23)<br />
pol
22 2. PHASAR <strong>demultiplexers</strong>: a review<br />
TE m-1<br />
TM m<br />
TE m<br />
TM m-1<br />
TM m<br />
TM m-1<br />
<strong>wavelength</strong><br />
For this design, the procedure described in section 2.2.2 should be slightly changed. By fixing<br />
the incremental length according to equation 2.23, the divergence angle Δα is fixed through<br />
equation 2.3 and Ra through Ra =<br />
da ⁄ Δα (see figure 2.1b). Ra being fixed in this way, the<br />
non-uniformity Lu can no longer be freely chosen. One disadvantage of this method is that the<br />
total <strong>wavelength</strong> span available for the WDM channels is limited by the <strong>polarisation</strong><br />
dispersion, which is in the order of 4-5 nm for conventional InGaAsP/<strong>InP</strong> DH structures.<br />
Another disadvantage is that the exact value of the <strong>polarisation</strong> dispersion is very sensitive to<br />
variations in layer composition and thickness, which makes it difficult to obtain a good match.<br />
Half-wave plate<br />
A very elegant method is the insertion of a λ/2-plate in the middle of the phased array. Light<br />
entering the array in TE-polarised state will be converted by the λ/2-plate and will travel<br />
through the second half of the array in TM-polarised state, and TM-polarised light will<br />
similarly traverse half the array in TE-state. As a consequence TE- and TM-polarised input<br />
signals will experience the same phase transfer regardless of the birefringence properties of the<br />
waveguides applied. This method was introduced by Takahashi et al. [153] and using<br />
polyimide half-wave plates, it has been successfully applied to silica-<strong>based</strong> [60,91] and<br />
LiNbO 3 -<strong>based</strong> devices [95,96]. As the polyimide half-wave plates have a thickness of more<br />
than 10 μm, they are only applicable to waveguide structures with a small NA which can<br />
bridge this distance with small diffraction losses. With semiconductor waveguides, the method<br />
is not practical due to the large NA of these waveguides. It could be applied successfully if a<br />
compact and fabrication tolerant integrated <strong>polarisation</strong> converter could be developed.<br />
Dispersion compensation<br />
For semiconductor-<strong>based</strong> PHASARs, a broadband solution for the <strong>polarisation</strong> dependence<br />
problem is found in compensation of the <strong>polarisation</strong> dispersion by inserting a waveguide<br />
section with a different birefringence in the phased array. This method was suggested for<br />
FSR<br />
TE m<br />
<strong>polarisation</strong> dispersion<br />
TE m-1<br />
Figure 2.7 Schematic diagram of the different diffraction orders at the receiver<br />
side for both states of <strong>polarisation</strong>: <strong>polarisation</strong> dispersion. The diagram applies<br />
to the demultiplexer’s state without order matching, because the orders TE m and<br />
TM m-1 do not overlap.
2.2 Phased-array design 23<br />
silica-<strong>based</strong> waveguides by Takahashi et al. [154] and successfully applied to <strong>InP</strong>-<strong>based</strong><br />
devices by Zirngibl et al. [183]. A more comprehensive account of this method can be found in<br />
Chapter 4.<br />
Polarisation splitter<br />
Another method for obtaining <strong>polarisation</strong> independence is by applying a <strong>polarisation</strong> splitter<br />
at the input, as shown in figure 2.8. Due to the <strong>polarisation</strong> dispersion, the position of the focal<br />
spot in the image plane for TE <strong>polarisation</strong> is shifted relative to the TM-polarised one. If the<br />
distance between the TE and the TM-polarised receiver waveguide in the object plane is<br />
chosen equal to the <strong>polarisation</strong> dispersion in the image plane, the TE and TM-polarised<br />
signals will focus on the same position and the response will become <strong>polarisation</strong> <strong>independent</strong><br />
over a broad <strong>wavelength</strong> range. This method does not apply to N × N devices.<br />
TE-TM<br />
splitter<br />
TM<br />
Figure 2.8 Application of a <strong>polarisation</strong> splitter at the input.<br />
TE<br />
2.2.5 Design for flattened response<br />
TE+TM<br />
In many applications a flattened passband is important in order to relax the requirements on<br />
<strong>wavelength</strong> control. Three methods to achieve this goal will be discussed.<br />
Multimode receiver waveguides<br />
The simplest method is the use of broad (multimode) waveguides at the receiver side<br />
[6,127,129,138,139]. If the focal spot moves along these broad receiver waveguides, almost<br />
100% of the light will be coupled into the receiver waveguide over a considerable part of the<br />
receiver waveguide aperture, thereby causing a flat region in the frequency response as shown<br />
in figure 2.9a. In this way the 1-dB bandwidth can easily be increased from 31% of the channel<br />
spacing, as shown in section 2.1.4 for a non-flattened PHASAR, to over 65%.<br />
Due to the multimode character of the receiver waveguides, this method can only be applied at<br />
the receiver side of a WDM-link, where the multimode waveguides can be coupled to a<br />
detector without additional signal loss.
24 2. PHASAR <strong>demultiplexers</strong>: a review<br />
Transmitted power [dB]<br />
0<br />
-2<br />
-4<br />
-6<br />
-8<br />
-10<br />
-0.6 -0.4 -0.2 0.0<br />
Δf/Δfch 0.2 0.4 0.6<br />
MMI-flattening<br />
A flattened response with single-mode outputs can be obtained by applying a short Multimode<br />
Interference (MMI) power splitter at the end of the transmitter waveguide [10,124,125]. This<br />
device converts the single waveguide mode at the input end of the coupler into a double image.<br />
The resulting output field pattern has a “camel-like” shape and the depth of the central<br />
depression can be controlled with the MMI width. If the image of this “camel-shaped” field<br />
moves along the single mode receiver waveguides, the response will have a flat region as<br />
shown in figure 2.9b. This method of flattening introduces insertion loss due to the mismatch<br />
between the “camel-shaped” focal field and the receiver waveguide mode.<br />
A similar effect can be obtained by applying a Y-junction and bringing the two output branches<br />
close together in the transmitter aperture. This method, however, is less compact and less<br />
robust.<br />
Shaping the phase transfer<br />
As the field in the image plane is the Fourier transform of the field at the output aperture, a field<br />
which is more or less rectangular can be realised if the field at the output aperture has a sin(x)/<br />
x distribution (x measured along the aperture). Such a sinc distribution can be approximated in<br />
a discrete manner by multiplying the field at the array aperture with a function with alternating<br />
sign in such a way that the Gaussian-like field is converted into a sin(x)/x-like field with<br />
positive and negative side lobes. The multiplication can be realised by inserting an additional<br />
half <strong>wavelength</strong> into the array waveguides terminating in the negative side lobe regions, or by<br />
increasing the optical length using thermo-optic or photo-elastic effects [87].<br />
2.2.6 Design for low loss<br />
(a) (b)<br />
-10<br />
-0.6 -0.4 -0.2 0.0<br />
Δf/Δfch 0.2 0.4 0.6<br />
Figure 2.9 Flattening of the <strong>wavelength</strong> response by using multimode receiver<br />
waveguides (a), and by applying an MMI-powersplitter at the transmitter side (b).<br />
The dashed lines indicate the response without flattening.<br />
For properly designed PHASARs realised with low-loss waveguides, the total loss is<br />
dominated by the loss occurring at the junctions between the array and the Free Propagation<br />
Region (FPR). Low losses can be obtained if the transition from the array to the FPR is<br />
adiabatical, i.e. if the gap between the waveguides reduces linearly to zero. Due to the finite<br />
Transmitted power [dB]<br />
0<br />
-2<br />
-4<br />
-6<br />
-8
2.2 Phased-array design 25<br />
resolution of the lithographical process, the gap between the waveguides, however, will stop<br />
abruptly when the waveguides come too close together. At this discontinuity, the field coming<br />
out of the array will show a modulation which is dependent on the width of the gap between<br />
the array waveguides and on the confinement of the field in the guides. Figure 2.10 shows the<br />
field for a large and a smaller gap. Due to the ripple in the field pattern, a considerable fraction<br />
of the power will diffract into adjacent orders and be lost. On reciprocity grounds, an equal loss<br />
will occur at the input aperture.<br />
To reduce this loss, the ripple of the output field should be reduced. This can be obtained by<br />
reducing the gap width (which requires better lithography) or by reducing the confinement of<br />
the waveguides. A disadvantage of the latter approach is that lowering the confinement<br />
increases the minimal bending radius and, consequently, increases the device size. Low<br />
confinement can be combined with small bending radii by applying a local contrast reduction<br />
near the array apertures using a double-etch process [33].<br />
intensity<br />
array aperture<br />
2.2.7 Device size<br />
intensity<br />
array aperture<br />
(a) (b)<br />
Figure 2.10 Fields at the input (dashed) and the output aperture (solid) of the<br />
phased array for a) a waveguide structure with strong confinement, and b) a<br />
structure with moderate confinement. For efficient coupling to the receiver<br />
waveguides, the output field should follow the dashed lines.<br />
The relative dispersion D of the array is defined as the displacement ds of the focal spot in the<br />
image plane with respect to the relative frequency variation<br />
have the following value:<br />
( ), and should<br />
˜<br />
δf δf = Δ f ch ⁄ f c<br />
D ˜<br />
d r<br />
f<br />
----- 1 -------δf<br />
dN ⎛ eff<br />
+ ⋅ ------------ ⎞<br />
⎝ d f ⎠<br />
1 –<br />
= ⋅<br />
(2.24)<br />
whereby we have assumed that N eff ⁄ N FPR ≈ 1,<br />
which is valid for most waveguide structures.<br />
The relative dispersion therefore equals the derivative dl ⁄ dα of the array guide length with<br />
respect to the starting angle α (see equation 2.3 and 2.22, and figure 2.6), resulting in the array<br />
guide length to be written as:<br />
The value l o is fixed by choosing a straight section length S r and a radius of curvature R r at an<br />
arbitrary reference angle α r :<br />
N eff<br />
l( α)<br />
lo αD˜ = +<br />
(2.25)<br />
lo 2Sr αrRr αrD˜ = + –<br />
(2.26)
26 2. PHASAR <strong>demultiplexers</strong>: a review<br />
This choice also determines the distance L between the focal points of the phased-array:<br />
resulting in three degrees of freedom to be used for design optimisation.<br />
Preferably, a phased-array design is made on the basis of graphs of R(α), S(α) and the guide<br />
height H(α) over a range of 0 to 180 degrees for different combinations of S r and R r . R(α) and<br />
S(α) follow directly from equations 2.19 and 2.20 by replacing α i with α, and H(α) is defined<br />
as:<br />
In figure 2.11 a design example is shown using representative values. It should be noted that<br />
the phased array can only be realised if the following requirements are satisfied:<br />
•<br />
S( α)<br />
> Ra,min • R( α)<br />
> Rmin dH<br />
• -------Δα > w ⋅ F with 2 < F < 3<br />
dα<br />
L = 2( Srcosαr + Rrsinαr) H( α)<br />
= S( α)sinα<br />
+ R( α)<br />
( 1 – cosα)<br />
(2.27)<br />
(2.28)<br />
The last requirement is imposed in order to avoid the gap between the array guides becoming<br />
too small. The interval over which these requirements are satisfied must at least equal the array<br />
aperture angle θa . In the presented example, a value of approximately 30 degrees is found for<br />
the aperture angle θa (θa /θo = 4). In this case the required array aperture angle is larger than the<br />
available interval, and a phased-array demultiplexer can be realised. If this is not the case,<br />
either different combinations of Sr , Rr and αr have to be tried, or Ra has to be reduced at the<br />
expense of a higher insertion loss for the outermost receiver waveguides ( Ra = smax ⁄ θmax combined with equation 2.9).<br />
Size [μm]<br />
2000<br />
1500<br />
1000<br />
500<br />
S<br />
R<br />
H<br />
R min<br />
f min<br />
0<br />
0 30 60 90<br />
α [deg]<br />
120 150 180<br />
Figure 2.11 Design example for S r = 500 μm and R r = 500 μm at a reference<br />
radius α r of 75 degrees.
2.2 Phased-array design 27<br />
The variation of the bending radius as a function of α can be minimised if the curve R(α) is<br />
flattened, which leads to a reduction of the dependence of the phased-array transfer on the<br />
radius of curvature. Flattening of the curve is done by requiring dR ⁄ dα and d to be<br />
zero at the (freely chosen) reference angle αr . The first requirement gives a local minimum in<br />
the R(α) curve, whereas the latter results in a bending point in the curve, giving a larger region<br />
of α over which the curve is flat. Both requirements lead to respectively:<br />
2 R dα 2<br />
⁄<br />
and:<br />
R r<br />
S r<br />
D˜ = ---------------<br />
(2.29)<br />
2tanα r<br />
D˜ 1<br />
--- 1<br />
2 sin 2 ⎛ ⎞<br />
= ⎜ + -------------- ⎟<br />
(2.30)<br />
⎝ α ⎠<br />
r<br />
Unfortunately, application of this concept leads to considerable device dimensions. In the best<br />
case, Rr D is the smallest reference radius obtainable ( mm is a representative<br />
value). If a concession to the flatness of the R(α) curve is allowed, the following empirical<br />
requirement can be used:<br />
˜ = D˜ = 2<br />
R r<br />
D˜ 2sin 2 = -----------------<br />
(2.31)<br />
αr With this condition the smallest possible reference radius is reduced by a factor two and,<br />
consequently, the device size. As small device dimensions are preferred, an adaption of the<br />
design strategy is still needed. The requirement for a flat R(α) curve is dropped, but in addition<br />
to the requirement of dR/dα to be zero at the reference angle α r , a second requirement is<br />
imposed:<br />
R r<br />
= Rmin (2.32)<br />
with Rmin being the minimum allowed bending radius. In our case, this concept leads to the<br />
smallest device dimensions, as the minimum bending radius is in the order of only a few<br />
hundreds of microns.<br />
The R(α) curve for each of the three proposed concepts is calculated using representative<br />
values for D , αr and Rmin of 2 mm, 75 degrees and 500 μm respectively. The graph is depicted<br />
˜<br />
in figure 2.12. It can be clearly seen that concept (a) ensures flatness of the R(α) curve over a<br />
long range of α (in this example even more than 70 degrees!), but additionally it results in the<br />
highest values for the curvature radius. Concept (b) and (c) allow for use of smaller bending<br />
radii at the cost of curve flatness. Corrections therefore have to be made in order to account for<br />
the propagation constant in the bend which depends on the curvature radius.
28 2. PHASAR <strong>demultiplexers</strong>: a review<br />
2.2.8 Correction for bending effects<br />
In the present design concept, the curved waveguides are treated as straight waveguides with a<br />
phase transfer of Φ = N effk02αR and the bending radius R defined in the centre of the<br />
waveguide. With curved waveguides, however, the modal field distribution tends to shift<br />
towards the outer edge of the bend, which effectively corresponds to an increase of the bending<br />
radius. If bending radii are applied where the shift of the modal field cannot be neglected, a<br />
correction to the bending radius is needed in order to avoid degradation of the array<br />
performance. As the phase transfer of the bend with radius R seems to be caused by a bend<br />
with radius R' = R + ΔR (see figure 2.13), the corrected bending radius R' is calculated as:<br />
leading to:<br />
R [μm]<br />
3000<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
a: eqs. 2.29 & 2.30<br />
b: eq. 2.31<br />
c: eq. 2.32<br />
0<br />
0 30 60 90<br />
α [deg]<br />
120 150 180<br />
Figure 2.12 Bending radii of the array guides as a function of array guide angle α<br />
for the three proposed design concepts.<br />
2αβ φ( R)<br />
= Φ = 2αR'β s<br />
(2.33)<br />
ΔR<br />
βφ( R)<br />
= -------------- – R<br />
(2.34)<br />
with βs being the propagation constant of a straight waveguide. The apparent increase of the<br />
bending radius can therefore be compensated for by a reduction of the bending radius by the<br />
same amount, resulting in R'' =<br />
R – ΔR for the actual bending radius.<br />
As low-loss operation is favourable, the junction losses between the straight and curved<br />
waveguides should be minimised by applying a lateral offset. The correction of the bending<br />
radius corresponds to a lateral offset in the correct direction, but in most cases it amounts to<br />
only half of the offset needed for minimum coupling loss. As the correction is in the order of a<br />
tenth of micron, the loss penalty will be negligible. It may result, however, in unwanted<br />
excitation of higher order modes, which causes “ghost” images (see section 2.1.5 and 4.4). The<br />
optimisation of the offset, both for optimum phase transfer and minimum coupling loss (or<br />
higher order mode excitation), will be explained in the next paragraph.<br />
β s
2.3 Applications 29<br />
2.2.9 Waveguide junctions<br />
The correction for bending effects results in an offset at the junction of the straight and curved<br />
array waveguides, which leads to undesired effects such as coupling loss, and, in case of<br />
bimodal waveguides, also mode conversion causing “ghost” images (see section 2.1.5 and 4.4).<br />
Usually, the bending radii are chosen in such a way that the fundamental orders for both TE<br />
and TM <strong>polarisation</strong> are guided with negligible bending loss, and higher order modes with<br />
high bending loss. However, if the higher order modes are still guided with low bending loss<br />
(which is, for instance, the case for the <strong>polarisation</strong> <strong>independent</strong> raised-strip waveguides),<br />
mode conversion is undesirable. Because of different propagation constants for the<br />
fundamental and the first order mode, the resulting image will occur at a different position<br />
along the image plane. This shift can be calculated in the same way as the TE-TM shift<br />
according to equation 2.18. Mode conversion can be kept small by optimising the offset at the<br />
junctions on minimal first-order mode excitation.<br />
Each offset between the straight and curved waveguides can be optimised separately as<br />
desired, either to maximum coupling efficiency of the fundamental mode or to minimum mode<br />
conversion. This is done by rotating each i-th straight waveguide over a small angle δα i ,<br />
additional to its original angle α i (see equation 2.21), according to:<br />
whereby Δx is the desired lateral offset. It should be noted, that the lateral shift due to the<br />
bending correction ΔR is included in this offset and should be substracted first to find the<br />
additional rotation angle δα i . With the offset Δx in the order of a few tenths of a micron, and<br />
the straight section length S i in the order of a few hundreds of microns, the additional angle δα i<br />
will be very small. Additionally, in this way neither the corrected bending radius nor the length<br />
of the straight section are changed, and therefore the phase transfer of the array is not<br />
influenced.<br />
2.3 Applications<br />
ΔR<br />
Figure 2.13 Schematic diagram of the bending correction.<br />
δα i<br />
= asin( Δx ⁄ Si) ≈ Δx ⁄ Si (2.35)<br />
In addition to the basic functions of <strong>wavelength</strong> multiplexing and demultiplexing, PHASARs<br />
are applied in <strong>wavelength</strong> routers and, in combination with other components such as
30 2. PHASAR <strong>demultiplexers</strong>: a review<br />
amplifiers and switches, in more complex devices for use in multi-<strong>wavelength</strong> networks. In this<br />
section a number of applications will be discussed further.<br />
2.3.1 Wavelength routers<br />
A <strong>wavelength</strong> router is obtained by designing the input and the output side of a PHASAR<br />
symmetrically, i.e. with N input and N output ports. For the cyclical rotation of the input<br />
frequencies along the output ports (as described in section 1.2), it is essential that the frequency<br />
response is periodical (as shown in figure 1.3b), which implies that the FSR should<br />
equal N times the channel spacing. From equation 2.6 it can be seen that this is obtained by<br />
choosing:<br />
ΔL<br />
whereby Ng is the group index of the waveguide mode, N is the number of frequency channels<br />
and Δfch is the channel spacing.<br />
For this design, the procedure described in section 2.2.2 should be changed similarly to section<br />
2.2.4 paragraph (ii). By fixing the incremental length according to equation 2.36 the<br />
divergence angle Δα is fixed through equation 2.3 and Ra through Ra =<br />
da ⁄ Δα (see figure<br />
2.1b). With this choice of FSR, the non-uniformity L u is fixed and will be in the order of 3 dB,<br />
which can be explained as follows. Channels at a frequency Δf FSR /2 away from the central<br />
frequency will experience an excess loss L u of at least 3 dB, because the focal spot<br />
corresponding to this frequency will be equally divided among two orders, which focus<br />
symmetrically around the centre of the image plane. As in a periodical design the frequency<br />
spacing between the outer channels comes close to the FSR, the outer channels will experience<br />
an excess loss L u in the order of 3 dB.<br />
Wavelength routers have been applied in various configurations in add-drop multiplexers and<br />
<strong>wavelength</strong> selective switches [47,48,64,65,90,130,143-147] and in multi-<strong>wavelength</strong><br />
networks [54]. In combination with a DFB laser used as a <strong>wavelength</strong> converter, a <strong>wavelength</strong><br />
router has also been applied as a <strong>wavelength</strong> switch [130, 135].<br />
2.3.2 Multi-<strong>wavelength</strong> receivers<br />
A multi-<strong>wavelength</strong> receiver is obtained by integration of a demultiplexer with a photodiode<br />
array. The first PHASAR receiver (reported in 1993 by Amersfoort et al. [4,5]), applied a twinguide<br />
waveguide structure where the passive region was obtained by locally removing the<br />
absorbing top layer. Integrated receivers have also been used in buried waveguide structures<br />
[182] and in <strong>polarisation</strong> <strong>independent</strong> raised-strip waveguides [9,122]. A <strong>wavelength</strong>-flattened<br />
receiver module, hybridly integrated with a silicon bipolar front-end array, has been discussed<br />
by Steenbergen et al. [138,139,140]. Recently, a low-loss (3 dB on-chip loss) 8-channel WDM<br />
receiver with 10 GHz bandwidth per channel has been reported [141].<br />
2.3.3 Multi-<strong>wavelength</strong> lasers<br />
c<br />
= ----------------------<br />
N gNΔ f<br />
(2.36)<br />
ch<br />
Today’s WDM systems use <strong>wavelength</strong>-selected or tunable lasers as sources. Multiplexing of a<br />
number of <strong>wavelength</strong>s into one fibre is done by using a power combiner or a <strong>wavelength</strong><br />
multiplexer. Integrated multi-<strong>wavelength</strong> lasers have been produced by combining a DFB-laser<br />
array (with a linear frequency spacing) with a power combiner on a single chip [12,173,174].
2.3 Applications 31<br />
Using a power combiner for multiplexing the different <strong>wavelength</strong>s into a single fibre is a very<br />
tolerant method, but it introduces a loss of 10 ⋅ logN dB, with N being the number of<br />
<strong>wavelength</strong> channels. The combination loss can be reduced by applying a <strong>wavelength</strong><br />
multiplexer, at the cost, however, of more stringent requirements on control of the laser<br />
<strong>wavelength</strong>s.<br />
An elegant solution to this problem is combining a broadband optical amplifier array with a<br />
multiplexer into a Fabry-Perot cavity as depicted in figure 2.14. This principle was first<br />
demonstrated in the MAGIC-laser [169] in a hybridly integrated form. If one of the<br />
Semiconductor Optical Amplifiers (SOAs) is excited, the device will start lasing at the<br />
passband maximum of the multiplexer channel to which the SOA is connected. In principle all<br />
SOAs can be operated and (intensity) modulated simultaneously. An important advantage of<br />
this component is that the <strong>wavelength</strong> channels are automatically tuned to the passbands of the<br />
multiplexer and coupled to the single output port with low loss.<br />
Mirror<br />
SOA<br />
SOA<br />
SOA<br />
SOA<br />
MUX<br />
Figure 2.14 Integrated multi-<strong>wavelength</strong> laser.<br />
Mirror<br />
Zirngibl and Joyner reported the first multi-<strong>wavelength</strong> lasers <strong>based</strong> on integration of a SOAarray<br />
with a PHASAR [68,178,179] and demonstrated it with a 9 × 200 Mb/s transmission<br />
experiment [180]. Despite their long cavity length, these lasers show single mode operation in<br />
a wide range of operating conditions [181]. Direct modulation speeds in excess of 1 Gb/s were<br />
reported recently [184]. Power coupled into a fibre is still low. Highest power reported so far is<br />
0.15 mW [132,136].<br />
Joyner et al. [70] reported integration of a MW-laser with an electro-absorption modulator.<br />
They used the power, radiated into an adjacent order of the phased array, to couple light out of<br />
the cavity into the modulator.<br />
A problem in MW-lasers with a small FSR is that the laser may start lasing in a wrong order<br />
and, consequently, at a wrong frequency. Doerr et al. [43] proposed and demonstrated a<br />
method to suppress the transmission for undesired orders by chirping the incremental length<br />
ΔL in the array.<br />
Tachikawa et al. [147] reported a 32-channel discretely tunable laser <strong>based</strong> on a<br />
PHASAR with SOAs, with one reflecting mirror connected to both the 4 input and the 8 output<br />
ports. The 32 <strong>wavelength</strong>s are generated by powering the proper SOA pairs.<br />
2.3.4 Wavelength-selective switches and add-drop multiplexers<br />
Add-drop multiplexers (ADMs) form a special class of <strong>wavelength</strong> selective switches. They<br />
are used for coupling one or more <strong>wavelength</strong> signals from a main input port into one or more<br />
drop ports by operating the corresponding switches. The other signals are simultaneously<br />
routed into the main output port, together with the signals applied at the proper add ports.<br />
Figure 2.15a shows the configuration as produced by Tachikawa et al. [144,146,149]. The<br />
fibre<br />
4 ×<br />
8
32 2. PHASAR <strong>demultiplexers</strong>: a review<br />
device (hybridly integrated, whereby the switching was done by changing fibre connectors),<br />
showed a fibre-to-fibre insertion loss of 3-4 dB for the add-drop signals and 6-8 dB for the<br />
transmitted signals. Through a suitable arrangement of the loop-back optical paths, the<br />
insertion loss difference between the transmitted signals can be minimised [62].<br />
demux/mux<br />
(a) RX TX<br />
(b)<br />
(c)<br />
demux<br />
TX RX<br />
demux/mux<br />
A disadvantage of this loop-back configuration is that the cross talk of the PHASAR is coupled<br />
directly into the main output port. This problem can be reduced by applying the PHASAR in a<br />
fold-back configuration as shown in figure 2.15b [47,64]. A third approach uses a separate<br />
demultiplexer and multiplexer (figure 2.15c) as reported by Okamoto et al. [89,90,93]. The two<br />
PHASAR used in this approach were placed close together in order to ensure that their channel<br />
frequencies match.<br />
In <strong>wavelength</strong> routed networks, spatial switching of arbitrary <strong>wavelength</strong> signals between<br />
multiple channels allows for efficient use of the transmission capacity by using a fixed number<br />
of <strong>wavelength</strong>s and by re-using them. For this approach a number of configurations using<br />
silica-<strong>based</strong> waveguide structures have been reported [63,64,143].<br />
The first <strong>InP</strong>-<strong>based</strong> ADM has been reported only recently [167]. The 400 GHz spaced<br />
4-channel device (in loop-back configuration) has less than 10.7 dB on-chip loss for the pass<br />
function (including switch losses, twice the demultiplexer loss, 7 waveguide crossings, and a<br />
total of 2 cm waveguide loss). The on-chip loss for the add and drop function is less than<br />
6.7 dB, and the cross talk is less than -20 dB for all paths. The device is extremely compact<br />
( mm2 3 ×<br />
6 ), which demonstrates the potential for <strong>InP</strong> integration of complex circuits on a<br />
small chip area. This potential can be further developed by including optical amplifiers to<br />
compensate the on-chip losses.<br />
mux<br />
Figure 2.15 Three different ADM configurations: a) loop-back, b) fold-back, and<br />
c) cascaded demux/mux (TX = transmitter, RX = receiver, X = switch).<br />
RX<br />
TX
2.4 Conclusions 33<br />
A configuration which does not require switches makes use of <strong>wavelength</strong> conversion and is<br />
shown in figure 2.16 [130,135]. The modulated signal is fed into a continuously operating<br />
DBR laser, of which the <strong>wavelength</strong> can be tuned by the current injection. The insertion of the<br />
modulated signal will decrease the carrier concentration, leading to a decrease of the DBR<br />
laser signal. The output signal of the DBR laser is therefore not only converted to the DBR<br />
laser <strong>wavelength</strong>, but also inverted with respect to the input. Spatial switching can be achieved<br />
by tuning the DBR injection current and by connecting the output to a phased-array<br />
demultiplexer.<br />
2.4 Conclusions<br />
DBR<br />
router<br />
Figure 2.16 Schematic diagram of a <strong>wavelength</strong> switch employing a currentinjection<br />
tunable DBR laser.<br />
The range of applications of PHASAR-<strong>based</strong> devices is growing rapidly. PHASARs have<br />
proven to be flexible components which support the possibility of a broad range of functions<br />
for use in multi-<strong>wavelength</strong> networks. Silica-<strong>based</strong> devices offer the best performance and are<br />
presently being applied most widely. They might get some competition from polymer-<strong>based</strong><br />
devices in the future. <strong>InP</strong>-<strong>based</strong> devices are most promising for manufacture of active MWdevices<br />
such as MW-lasers and receivers and, in the longer term, for more complicated circuits<br />
containing large numbers of components, such as add-drops and optical cross connects.
34 2. PHASAR <strong>demultiplexers</strong>: a review
Chapter 3<br />
Polarisation conversion in waveguide bends<br />
Very short waveguide bends appear to have <strong>polarisation</strong>-converting properties. For most<br />
applications the bends should be designed in such a way to keep <strong>polarisation</strong> conversion<br />
effects small. The <strong>polarisation</strong> converting properties can be applied especially in compact lowloss<br />
<strong>polarisation</strong> converters. The <strong>polarisation</strong> conversion phenomenon is discussed in this<br />
chapter and experimental results are presented. Part of these experiments have been presented<br />
at the IPR’96 [34] and have been published in IEEE Photonics Technology Letters [35]. They<br />
have also led to a patent application [32].<br />
3.1 Introduction<br />
As the state of <strong>polarisation</strong> is unknown at the receiver side, <strong>polarisation</strong> handling is of great<br />
importance in optical communication systems. The most obvious way to deal with this<br />
problem is the usage of <strong>polarisation</strong> <strong>independent</strong> devices, and, in particular, a <strong>polarisation</strong><br />
<strong>independent</strong> demultiplexer. A number of methods to obtain <strong>polarisation</strong> <strong>independent</strong><br />
<strong>demultiplexers</strong> have already been mentioned in Chapter 2 and are discussed more<br />
comprehensively in Chapters 4 and 5. One of these methods uses a half-wave plate inserted in<br />
the middle of a phased array and has successfully been applied to waveguide structures with a<br />
small NA, such as silica and LiNbO 3 . As this method is not practical for <strong>InP</strong>-<strong>based</strong> devices,<br />
because of the large NA of <strong>InP</strong>-<strong>based</strong> waveguides and the corresponding large diffraction of<br />
the unguided beam, a solution may be found in a compact integrated <strong>polarisation</strong> converter.<br />
Until recently research has been focused on a type of converter which uses a periodically<br />
alternating, asymmetrically loaded waveguide [53,115,157]. Schematic cross sections of<br />
produced devices are shown in figure 3.1a-c. These types of converter consist of a periodic<br />
structure, as shown in figure 3.1d (topview), of which the period is chosen in such a way to<br />
obtain constructive interference of the light, converted at a junction with light which has been<br />
converted at previous junctions. The waveguide is made asymmetrical in order to achieve a<br />
high conversion efficiency at the junctions, which results in a short device. This can be done<br />
either by a partial top load on the waveguide [115], or by tilted side walls [53,157] as shown in<br />
figure 3.1a-c. It is obvious that these types of converter require multiple masking steps,<br />
combined reactive ion etching and selective chemical etching steps. Their performance is<br />
summarised in table 3.1.
36 3. Polarisation conversion in waveguide bends<br />
Table 3.1 Performance of the three types of converter using a periodically<br />
alternating, asymmetrically loaded waveguide.<br />
Author Reference Conversion<br />
[%]<br />
Excess loss<br />
[dB]<br />
Length<br />
[μm]<br />
Shani et al. 115 90-100 2-3 3700<br />
Heidrich et al. 53 50 3 825<br />
van der Tol et al. 157 93 9-11 990<br />
(a) (b)<br />
(c) (d)<br />
Figure 3.1 Cross sections of the periodically alternating, asymmetrically loaded<br />
waveguide as proposed by Shani et al. [115] (a), Heidrich et al. [53] (b), van der<br />
Tol et al. [157] (c), and a schematic top view of Shani’s device (d).<br />
In this chapter a novel type of <strong>polarisation</strong> converter is presented <strong>based</strong> on deeply etched<br />
waveguides with a small bending radius as shown in figure 3.2. It has a potential for low loss<br />
and compact device size, and it can be integrated with other devices. The operation principle<br />
can be explained as follows. It has been found that the non-dominant field component (E x for<br />
TE and E R for TM) of both the TE and TM mode strongly increases if the bending radius is<br />
chosen sufficiently small, which means that the <strong>polarisation</strong> direction of these modes will be
3.2 Computational methods 37<br />
1.4 μm<br />
<strong>InP</strong><br />
InGaAsP<br />
Q(1.3)<br />
<strong>InP</strong> substrate<br />
rotated. If a straight waveguide is connected to such a bend, both modes will be excited. As<br />
these modes have different propagation constants, a phase difference is obtained after<br />
propagation through a bend section. By placing a next bend section with opposite curvature,<br />
<strong>polarisation</strong> conversion can be achieved. The conversion ratio can be optimised by a proper<br />
choice of the section length. Apart from the fact that tilting of the modal <strong>polarisation</strong> plane is<br />
achieved by a curved waveguide instead of an asymmetrically loaded waveguide, the operation<br />
of this type of converter is similar to those depicted in figure 3.1.<br />
The operation of such a <strong>polarisation</strong> converter has been analysed using three different<br />
computational methods, which are discussed in section 3.2. The radiation mechanism in deeply<br />
etched waveguides is discussed in section 3.3, and following that the <strong>polarisation</strong> rotation of<br />
such bends in section 3.4. The <strong>polarisation</strong> dispersion and conversion are discussed in sections<br />
3.5 and 3.6 respectively, whereafter the layout of a <strong>polarisation</strong> converter is described. Section<br />
3.8 deals with the tolerance analysis of the converter and in section 3.9 it is explained how<br />
conversion can be avoided.<br />
For the calculations and experiments detailed in this chapter, the deeply etched waveguide<br />
structure as shown in figure 3.2 is used unless otherwise stated. The <strong>wavelength</strong> was chosen at<br />
1508 nm, determined by the operating <strong>wavelength</strong> of the Fabry-Pérot laser used for the<br />
experiments. The refractive indices of <strong>InP</strong> and InGaAsP at this <strong>wavelength</strong> are 3.1745 and<br />
3.4005 respectively [46].<br />
3.2 Computational methods<br />
A number of methods for calculating the propagation constants and modal fields in curved<br />
waveguides are available in our laboratory, which will be discussed in the following sections.<br />
An overview is given in table 3.2.<br />
3.2.1 Effective index method<br />
0.3 μm<br />
0.6 μm<br />
Figure 3.2 The deeply etched waveguide structure as used for the <strong>polarisation</strong><br />
converter.<br />
The Effective Index Method (EIM) [73] is probably the most widely used method due to its<br />
ease of use. The principle of this method is briefly explained as follows, using the x, y and z for<br />
the vertical, horizontal and longitudional directions respectively. Firstly, the local effective<br />
index N(y,z) is calculated for each point (y,z) as if the structure were y-z-invariant.<br />
ϕ<br />
X<br />
R
38 3. Polarisation conversion in waveguide bends<br />
Table 3.2 Overview of the computational methods for calculation of the effective<br />
index N eff , the radiation loss α, and the modal field Ψ. (EIM = Effective Index<br />
Method, MOL = Method of Lines, FEM = Finite Element Method, and Conf.<br />
trans. = Conformal transformation)<br />
Method N eff α Ψ Comments<br />
EIM + + + 2 x 1D scalar + Conf. trans.<br />
FEM ++ − ++ 2D vectorial + Conf. trans.<br />
MOL ++ ++ ++ 2D vectorial<br />
Then the modal effective index is calculated by computing the two-dimensional mode<br />
solutions for the index profile N(y,z). This method actually transforms the 2D problem into two<br />
1D problems. With the EIM, both the mode index and the two 1D field profiles can be<br />
calculated.<br />
3.2.2 Method of lines<br />
A method suitable for analysing waveguide bends is the Method of Lines (MOL) [99]. We have<br />
a full 2D vectorial method available which is capable of calculating the effective index, the<br />
radiation loss as well as the X-, R- and ϕ-components of the waveguide mode. The program has<br />
been developed by the University of Hagen. With this method the waveguide cross section is<br />
divided into regions by horizontal lines. On these lines the vector potentials for the<br />
electromagnetic fields are calculated. For a straight waveguide they satisfy the Helmholtz<br />
equations and can be expressed analytically. For a curved waveguide, however, Bessel equations<br />
are obtained, of which the solutions are cylinder functions of large and complex order. In the<br />
transverse direction, the potentials are discretised and continuity conditions are applied on the<br />
interfaces of the single layers. The result is a system of coupled ordinary differential equations,<br />
which can be solved using standard methods.<br />
3.2.3 Finite element method<br />
The third mode solver is a vectorial Finite Element Method (FEM) [108] made available to us<br />
by the ETH-Zürich. This method is particularly suitable for electromagnetic field problems, as<br />
the general geometry of the waveguide can be quite complicated with an arbitrary refractive<br />
index profile n(x,y) in the transverse directions. For this method, the entire waveguide cross<br />
section is divided into a patchwork (or mesh) of sub-regions or elements, usually triangles or<br />
quadrilaterals. Using many elements, any cross section can be accurately approximated. The<br />
field in each element is represented by a polynomial and the field continuity conditions are<br />
imposed on all interfaces between different elements. By employing a variational expression<br />
for Maxwell’s equations, an eigenvalue matrix is obtained, which can be solved using standard<br />
methods.<br />
Figure 3.3 shows a mesh definition for a straight waveguide. It can be seen that the mesh size<br />
should be small near interfaces and equal on both sides of the interface. Additionally, as the<br />
field intensity decreases away from the waveguide core, the mesh size can be increased. The<br />
solution depends on the mesh size and therefore an increasing refinement has to be tried until<br />
the mesh dependence vanishes. It should be noted, however, that on the boundary of the
3.2 Computational methods 39<br />
Figure 3.3 Mesh definition for a straight waveguide.<br />
computation window the field is assumed to be zero, which implies that a sufficiently large<br />
window has to be chosen. In this way it is avoided that solutions are obtained which actually do<br />
not represent physical modes of the waveguide.<br />
3.2.4 Conformal transformation<br />
Using the EIM and the FEM, the complex angular propagation constant in the bend can be<br />
calculated using the conformal transformation technique [52]. This technique transforms a<br />
curved waveguide into an equivalent straight waveguide with a transformed index profile. The<br />
transformation, which is shown schematically in figure 3.4, is written as<br />
r = Rt ⋅ exp( u ⁄ Rt) , whereby Rt is an arbitrary reference radius. As the waveguide width is<br />
much smaller than the reference radius, the transformation can be linearised around r = Rt as<br />
r ≈ Rt + u.<br />
This leads to the following transformed index profile:<br />
nt( r)<br />
= n( r)<br />
⋅ r ⁄ Rt (3.1)<br />
The transformation leads to a solution in the form of a propagation constant β t in the<br />
transformed domain, from which the angular propagation constant β ϕ can be calculated as:<br />
βϕ = βt ⋅ Rt = N eff,t ⋅ k0 ⋅ R (3.2)<br />
t<br />
This angular propagation constant has the dimension rad -1 , because the propagation through a<br />
bend is described as exp( – jβϕϕ<br />
) . The radiation loss can be calculated from the imaginary part<br />
of the angular propagation constant according to:<br />
β' ϕ<br />
αϕ = 20 ⋅ log[ exp( 2 β' ϕ ⁄ π)<br />
] [ dB/90° ] (3.3)<br />
For the EIM it should be noted that, being a 2x1D scalar method, it cannot give a good<br />
prediction for the radiation loss in deeply etched waveguide structures, as it does not take<br />
radiation into the substrate into account. On the other hand, however, it is found that the modal<br />
field solutions correspond very well to the ones calculated using the MOL or the FEM.
40 3. Polarisation conversion in waveguide bends<br />
n<br />
For the 2D-FEM-analysis of a curved waveguide, a step-wise approximation of the<br />
transformed index distribution nt( x, y', z')<br />
= n( x, r, ϕ)<br />
⋅ r ⁄ Rt has to be implemented in the<br />
mesh definition. As the bend goes to the left, the flux in the left part is reduced and therefore a<br />
larger mesh can be taken than for a straight waveguide, which is shown in figure 3.5a.<br />
Consequently, the mesh size at the right side should be taken smaller. It should be noted,<br />
however, that the size of the window should be taken in such a way that the transformed index<br />
profile does not exceed the effective index as shown schematically in figure 3.5b. Otherwise<br />
the solutions will become complex which cannot be handled by our FEM mode solver.<br />
Consequently, a complex solution for the propagation constant cannot be obtained and<br />
therefore no estimation of the radiation loss can be made.<br />
3.3 Attenuation in bends<br />
ϕ<br />
r<br />
r<br />
From the equation of the conformal transformation (equation 3.1) it can be seen that the<br />
transformed index increases with increasing r. This gives three effects, which will be discussed<br />
ϕ<br />
n t<br />
u = 0<br />
r = Rt Figure 3.4 Schematic diagram of the conformal transformation.<br />
N eff<br />
(a) (b)<br />
n t<br />
u<br />
u<br />
r<br />
upper limit<br />
Figure 3.5 Mesh definition for a curved waveguide (a), and the step-wise<br />
approximated transformed index profile showing the calculation window limit (b).
3.3 Attenuation in bends 41<br />
briefly. Firstly, as the mode is guided strongest in the region with the highest index, the mode<br />
profile will “move” to the high-index region. It is shifted to the outer edge of the waveguide<br />
and will also be compressed, leading to a field mismatch at the transitions to straight<br />
waveguides. Secondly, the transformed index will always exceed the effective index at a<br />
certain value of r. At that point the mode will start radiating. The result is radiation loss, which<br />
is inherent to waveguide bends. However, if this point lies far away from the waveguide, the<br />
radiation loss is negligibly low. If the bending radius is decreased, this point comes closer to<br />
the waveguide, giving higher radiation loss. These two effects are depicted in figure 3.6. And<br />
finally, as the field intensity at the outer edge of the waveguide increases, the scattering losses<br />
increase as well.<br />
index profile<br />
Figure 3.6 Mode profiles in a straight waveguide (dashed line) and in a leftoriented<br />
curved waveguide (solid line), clearly showing an oscillatory behaviour,<br />
which indicates that the mode is experiencing radiation loss.<br />
For low-contrast waveguide structures, light radiates away from the waveguide bend in the<br />
guiding film as shown in figure 3.7a. When a high-contrast bend is used, such as the deeply<br />
etched waveguide structure of figure 3.2, light radiates into the substrate as shown in figure<br />
3.7b. This is explained by the fact that when the bending radius is decreased, the mode index<br />
decreases and phase-matching with substrate modes will occur, which in turn leads to coupling<br />
to radiation modes and a corresponding increase of the radiation loss.<br />
Transverse position [ μm]<br />
3<br />
2<br />
1<br />
0<br />
-2 -1 0 1 2 3<br />
Lateral position [μm]<br />
(a)<br />
0<br />
-2 -1 0 1 2 3<br />
Lateral position [μm]<br />
Figure 3.7 Radiation loss in a low-contrast (a) and in a high-contrast (b) bend.<br />
Transverse position [μm]<br />
3<br />
2<br />
1<br />
(b)
42 3. Polarisation conversion in waveguide bends<br />
In most cases it is important for curved waveguides to find a value for the minimum usable<br />
bending radius. Preferably this should be done in a simple and fast manner. Pennings [104]<br />
found empirically that the phase-matching condition at the outer edge of the waveguide bend is<br />
a practical criterion for radiation cut-off. Using the EIM, he derived a simple method for<br />
estimating the minimal bending radii usable for deeply etched waveguide bends, <strong>based</strong> on the<br />
width of the waveguide and on the effective index in the ridge, as well as on the mode index of<br />
a straight waveguide with the same dimensions.<br />
For the estimated cutoff radius R co he found:<br />
R co<br />
wN ridge<br />
= -----------------------------------------<br />
(3.4)<br />
2( N straight – nsub) whereby Nridge and Nstraight are the effective index in the ridge region and the mode index in a<br />
straight waveguide respectively and nsub is the substrate index. Although no magnitude of the<br />
substrate leakage is given by this prediction, the excess bend loss at the cut-off radius was<br />
found to be in the order of 1 dB/90° [104]. These experiments consisted of waveguide<br />
structures with a low transverse index contrast. For the deeply etched waveguide bend a cut-off<br />
radius of 23 μm is estimated.<br />
Experiments on waveguide bends with very small bending radii (down to 30 μm) have been<br />
reported earlier [126,128]. The results are depicted as squares in figure 3.8 and show that when<br />
a deeply etched waveguide structure (as shown in figure 3.2) is used, low radiation loss can be<br />
combined with very small bending radii - less than 0.2 dB/90° at a 30 μm radius. The<br />
estimated measurement accuracy is ± 0.2 dB. It should be noted, however, that this graph<br />
gives a slightly pessimistic view on the losses, as maximum loss values were taken from a<br />
series of measurement data. However, measured bending losses still do not exceed 0.2 dB/90°.<br />
The radiation losses as predicted by the MOL (figure 3.8, solid line) are below 0.2 dB/90° for<br />
radii as small as 20 μm, which is confirmed by the measurements (figure 3.8 - squares). No<br />
estimation for the waveguide losses could be obtained with the EIM due to the large lateral<br />
index contrast ( Δn = 3.17 – 1 ).<br />
Radiation loss [dB/90]<br />
0.50<br />
0.40<br />
0.30<br />
0.20<br />
0.10<br />
0.00<br />
Measured<br />
MOL<br />
20 40 60 80 100<br />
Radius [μm]<br />
Figure 3.8 Radiation loss: measured [126,128] (squares), and predicted with the<br />
MOL (solid line).
3.3 Attenuation in bends 43<br />
It can be seen from figure 3.8 that although the radiation loss is still less than 1 dB/90° at the<br />
estimated cut-off radius of 23 μm, the estimation gives a reasonable indication of the radius<br />
where radiation loss becomes important.<br />
As equation 3.4 was found empirically using low-contrast (both lateral and transverse)<br />
waveguide structures, we decrease the transverse index contrast of the deeply etched<br />
waveguide at a bending radius of 20 μm. For this purpose, the band-edge <strong>wavelength</strong> of the<br />
quaternary guiding layer was decreased from 1.3 μm to 1.0 μm in steps of 0.1 μm. The mode<br />
indices for TE <strong>polarisation</strong> are calculated using the MOL and it was found that the transverse<br />
index contrast decreases from 0.23 to 0.05 in steps of approximately 0.06.<br />
Transverse position [μm]<br />
Transverse position [μm]<br />
3<br />
2<br />
1<br />
0<br />
3<br />
2<br />
1<br />
0<br />
straight curved<br />
0 1<br />
0<br />
Lateral position [μm]<br />
1 2<br />
(a) (b)<br />
0 1<br />
0<br />
Lateral position [μm]<br />
1 2<br />
(c) (d)<br />
0 1<br />
0<br />
Lateral position [μm]<br />
1 2<br />
In figure 3.9 contour plots of the dominant field component of the TE-polarised mode are<br />
shown for the four different guiding layer compositions. As a reference, the dominant field<br />
components for the corresponding straight waveguide are also shown in the same graphs. It has<br />
been noted that when the transverse index contrast decreases, the mode confinement also<br />
Transverse position [μm]<br />
Transverse position [μm]<br />
3<br />
2<br />
1<br />
0<br />
3<br />
2<br />
1<br />
0<br />
straight curved<br />
Q(1.3) Q(1.2)<br />
straight curved straight curved<br />
Q(1.1)<br />
Q(1.0)<br />
0 1<br />
0<br />
Lateral position [μm]<br />
1 2<br />
Figure 3.9 Dominant lateral field component of the TE-polarised fundamental<br />
mode in a straight (left part) and curved waveguide (right part) with a radius of<br />
20 μm, for decreasing band-edge <strong>wavelength</strong>s of the guiding film: 1.3 μm (a),<br />
1.2 μm (b), 1.1 μm (c), and 1.0 μm (d). The corresponding radiation losses are<br />
0.15, 0.93, 4.4, and 15.9 dB/90° respectively and the corresponding cut-off radii<br />
are 23, 43, 336, and 905 μm.
44 3. Polarisation conversion in waveguide bends<br />
decreases and the mode leaks increasingly into the substrate. The corresponding cut-off radius<br />
increases from 23 μm to 905 μm. Consequently, the (calculated) radiation loss increases by a<br />
factor of 10 from 0.15 dB/90° (Q(1.3) guiding layer) to 15.9 dB/90° (Q(1.0) guiding layer).<br />
3.4 Tilting of the modal <strong>polarisation</strong> plane in bends<br />
As the non-dominant field component of both the TE and TM mode strongly increases if the<br />
bending radius is chosen sufficiently small, the modal <strong>polarisation</strong> plane is tilted. If a straight<br />
waveguide is connected to such a bend, both modes will be excited. These modes propagate<br />
with different propagation constants, resulting in a phase difference after a bend section. If a<br />
next bend section with opposite curvature is connected, <strong>polarisation</strong> conversion can be<br />
achieved, the ratio of which can be optimised by a proper choice of the bend section length.<br />
For investigation of the increase of the non-dominant field component, we consider figure 3.10<br />
and 3.11. These figures show the surface and contour plots respectively of the intensity for the<br />
E x (non-dominant) and the E r (dominant) field component of the TE-polarised fundamental<br />
mode in a curved waveguide for decreasing bending radii as calculated with the MOL. Use of<br />
the FEM mode solver yields similar results. These figures show that for TE <strong>polarisation</strong> the E x<br />
field component increases strongly if the radius is chosen sufficiently small. (Simultaneously,<br />
the E r field component for TM <strong>polarisation</strong> increases strongly.)<br />
For a certain value of the radius, almost 100% of the total field power is present in the E x<br />
component. This means that the <strong>polarisation</strong> angle, which is defined as:<br />
ϕ pol<br />
arctan Er,max = ⎛-------------- ⎞<br />
(3.5)<br />
⎝ ⎠<br />
E x,max<br />
is rotated by 90 degrees. This equation is valid if both field components have an (almost)<br />
identical shape and if the peaks are located in the same position. From the graphs of figure 3.11<br />
it may be concluded that this is indeed the case.
3.4 Tilting of the modal <strong>polarisation</strong> plane in bends 45<br />
Figure 3.10 Surface plots of the intensity for the E x (left column) and the E r<br />
(right column) field component of the TE-polarised fundamental mode in a<br />
curved waveguide for decreasing bending radii: 100 μm (a), 50 μm (b), 40 μm<br />
(c), and 20 μm (d).
46 3. Polarisation conversion in waveguide bends<br />
Figure 3.11 Contour plots of the intensity for the E x (left column) and the E r<br />
(right column) field component of the TE-polarised fundamental mode in a<br />
curved waveguide for decreasing bending radii: 100 μm (a), 50 μm (b), 40 μm<br />
(c), and 20 μm (d).
3.5 Polarisation dispersion in bends 47<br />
Polarisation angle [deg]<br />
80<br />
60<br />
40<br />
20<br />
0<br />
10<br />
50 100 150<br />
Radius [μm]<br />
Figure 3.12 Polarisation angle of the TE (solid line) and TM (dashed line) mode<br />
versus the bending radius, calculated with the MOL. A <strong>polarisation</strong> angle of zero<br />
(ninety) degrees indicates that the mode is horizontally (vertically) polarised.<br />
Using equation 3.5, the <strong>polarisation</strong> angles for both TE and TM <strong>polarisation</strong> have been<br />
calculated and are shown in figure 3.12. It has been noted that a <strong>polarisation</strong> angle of zero<br />
degrees indicates that the mode is polarised horizontally. It can be seen that if the radius is<br />
reduced, the TE- and TM-polarised modes evolve into hybridly polarised modes. If a straight<br />
waveguide (or a bend with opposite curvature) is connected to such a bend, both modes will be<br />
excited. The coupling efficiency is high as the intensity profiles of the hybrid modes match<br />
very well, leading to low coupling losses.<br />
3.5 Polarisation dispersion in bends<br />
The mode index for TE and TM <strong>polarisation</strong> has been calculated with three different methods<br />
and the results are shown in figure 3.13a-b. It is found that the indices obtained with the FEM<br />
and the MOL, match very well for TE <strong>polarisation</strong>, but less for TM <strong>polarisation</strong>. The index<br />
curve obtained with the EIM shows a different behaviour for small bending radii, which is due<br />
to the fact that the EIM is known to give less accurate solutions when the mode is close to cutoff.<br />
The best methods to use for these small radii are the MOL and the FEM, of which the<br />
MOL is more favourable if radiation losses are to be calculated as well.<br />
If we take a look at the index difference between the TE and TM <strong>polarisation</strong> (also denoted as<br />
the <strong>polarisation</strong> dispersion ( ΔN pol =<br />
N TE – N TM))<br />
we can see that it is almost constant for the<br />
large-radius region (80 to 100 μm) as depicted in figure 3.14. The index difference decreases<br />
towards a minimum at a bending radius between 30 and 50 m, but the predicted position<br />
depends however on the calculation method.<br />
TE<br />
TM
48 3. Polarisation conversion in waveguide bends<br />
Mode index<br />
3.40<br />
3.35<br />
3.30<br />
3.25<br />
FEM<br />
MOL<br />
EIM<br />
20 40 60 80 100<br />
Radius [um]<br />
Mode index<br />
20 40 60 80 100<br />
Radius [um]<br />
At this minimum, the power of both TE and TM mode is equally divided over the dominant<br />
and non-dominant components (crossing of the curves in figure 3.12). When the radius is<br />
decreased further, the index difference increases and the power is transferred completely to the<br />
non-dominant component.<br />
The two hybridly polarised modes propagate with different phase velocities and at a length<br />
equal to the beat length L π , defined as:<br />
whereby Δβ and ΔN are the propagation constant and the mode index difference between the<br />
TE and TM mode, the combined field will show a <strong>polarisation</strong> rotation of 90 degrees, if both<br />
modes are excited equally. The beat length is calculated with the three methods and is depicted<br />
in figure 3.15.<br />
3.40<br />
3.35<br />
3.30<br />
3.25<br />
(a) (b)<br />
Figure 3.13 Calculated mode index in the bend for different methods - for TE<br />
<strong>polarisation</strong> (a) and for TM <strong>polarisation</strong> (b).<br />
Index difference<br />
0.008<br />
0.006<br />
0.004<br />
0.002<br />
0.000<br />
FEM<br />
MOL<br />
EIM<br />
20 40 60 80 100<br />
Radius [μm]<br />
Figure 3.14 Mode index difference calculated with the three methods.<br />
FEM<br />
MOL<br />
EIM<br />
π λ<br />
Lπ = ------ = ----------<br />
(3.6)<br />
Δβ 2ΔN
3.6 Polarisation conversion at junctions 49<br />
Beat Beatlength length [μm] [μm]<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
FEM<br />
MOL<br />
EIM<br />
20 40 60 80 100<br />
Radius [μm]<br />
Figure 3.15 Calculated beat length versus bending radius for the FEM (solid<br />
line), the MOL (dashed line) and the EIM (dotted line).<br />
Apart from the fact that different values of the beat length are predicted, another problem arises<br />
for realising a <strong>polarisation</strong> converter <strong>based</strong> on these deeply etched bends. As the converter<br />
consists of a number of segments (see figure 3.1d), the length of which is chosen equal to the<br />
beat length, rotation angles in excess of 10π radians are required. It is better therefore to<br />
choose the radius in such a way that the beat length is small, and consequently the conversion<br />
per junction lower. In this way, more segments are needed, but they are considerably shorter.<br />
The best choice, therefore, may be found in using a large radius (>70 μm).<br />
3.6 Polarisation conversion at junctions<br />
Polarisation conversion takes place at junctions of waveguides with different curvatures, which<br />
can be computed from the overlap of the modal fields in both waveguides. From the graphs of<br />
figure 3.10 it can be seen that the dominant and non-dominant field components match very<br />
well with the modal fields of a straight waveguide, and therefore a high coupling efficiency is<br />
expected. Verification contour plots are drawn of both field components in a curved waveguide<br />
with a bending radius of 50 μm (calculated with the MOL). These are shown in figure 3.16,<br />
together with contour plots of the modal fields in a straight waveguide (calculated with the<br />
FEM).
50 3. Polarisation conversion in waveguide bends<br />
Transverse position [μm]<br />
Transverse position [μm]<br />
Transverse position [μm]<br />
1<br />
0<br />
-1<br />
-1 0 1 2<br />
Lateral position [μm]<br />
1<br />
0<br />
-1<br />
-1 0 1 2<br />
Lateral position [μm]<br />
1<br />
0<br />
-1<br />
-1 0 1 2<br />
Lateral position [μm]<br />
(a)<br />
(b)<br />
(c)<br />
-1<br />
-1 0 1 2<br />
Lateral position [μm]<br />
The coupling efficiency η between two fields U 1 and U 2 is calculated using the overlap<br />
integral:<br />
Transverse position [μm]<br />
Transverse position [μm]<br />
Transverse position [μm]<br />
1<br />
0<br />
1<br />
0<br />
-1<br />
-1 0 1 2<br />
Lateral position [μm]<br />
1<br />
0<br />
-1<br />
-1 0 1 2<br />
Lateral position [μm]<br />
Figure 3.16 The E x (a) and E r (b) field components of the curved waveguide<br />
mode for a radius of 50 μm, and the dominant field component of the straight<br />
waveguide mode (c). The left column shows the TE <strong>polarisation</strong> and the right<br />
column the TM <strong>polarisation</strong>.<br />
∫<br />
( U1U 2)<br />
( U1, xU<br />
2, x + U1, rU<br />
2, r)<br />
η = ------------------------------------------------------------ = ----------------------------------------------------------------------------- (3.7)<br />
2<br />
2<br />
2<br />
2<br />
U1 ⋅ ∫ dxdr ∫ U1 dxdr ⋅ ∫ U2 dxdr<br />
dxdr 2<br />
∫ dxdr U 2<br />
∫<br />
dxdr 2
3.7 A compact <strong>polarisation</strong> converter 51<br />
Using this equation, the coupling efficiency from a quasi 1 -TE mode to a quasi-TM mode has<br />
been calculated (<strong>based</strong> on fields obtained with the MOL) and is shown in figure 3.17. At the<br />
junction between two oppositely curved waveguides the coupling efficiency (depicted in figure<br />
3.17a) is very low and such a junction will therefore not contribute to <strong>polarisation</strong> rotation. The<br />
coupling efficiency at a junction between a straight and a curved waveguide on the other hand<br />
is very high, as shown in figure 3.17b. This is a result of the high fraction of power that is<br />
present in the non-dominant component, combined with the good match between the fields.<br />
And because of this match, the curve of figure 3.17b may be considered as a measure for the<br />
fraction of power carried by the non-dominant component.<br />
Coupling efficiency [%]<br />
10<br />
8<br />
6<br />
4<br />
2<br />
0<br />
20 40 60 80 100<br />
Radius [μm]<br />
(a) (b)<br />
3.7 A compact <strong>polarisation</strong> converter<br />
20 40 60 80 100<br />
Radius [μm]<br />
The <strong>polarisation</strong> conversion was analysed using the layout as shown in figure 3.18. Two Ubends,<br />
consisting of four arc segments with a segment angle α and a waveguide width of<br />
1.4 μm, are placed in series. This segment angle is varied between 10 and 80 degrees, allowing<br />
for different beat lengths between two subsequent junctions. As the 1.4 μm wide waveguides<br />
have propagation losses in the order of 8-10 dB/cm [126,128], waveguides of 3.0 μm width are<br />
used to connect the U-bends to each other. They are also used for in- and output waveguides,<br />
because they are known to have low propagation losses in the order of 1-2 dB/cm [126,128].<br />
This is due to the fact that the field intensity at the (rough) etched side wall decreases when the<br />
waveguide width increases, and therefore less scattering loss will occur for broad waveguides.<br />
For the transition between these wide waveguides and the narrow bends, an adiabatic taper is<br />
used with a length of 50 μm. The (maximum) total device size measures μm2 975 ×<br />
83 .<br />
1. Strictly taken, we cannot speak of TE- and TM-polarised modes because both types have a<br />
strong orthogonal field component. For a deeply etched curved waveguide with a small<br />
bending radius, the assignment to TE or TM <strong>polarisation</strong> of a modal solution found with<br />
the 2D-solver is done on the basis of continuous evolution of the bending radius: the mode<br />
type assignment is done according to the <strong>polarisation</strong> type of the mode when the bending<br />
radius is increased to infinity.<br />
Coupling efficiency [%]<br />
Figure 3.17 Coupling efficiency from a quasi-TE to a quasi-TM mode between<br />
two oppositely curved waveguides (a), and between a straight and a curved<br />
waveguide (b).<br />
100<br />
80<br />
60<br />
40<br />
20<br />
0
52 3. Polarisation conversion in waveguide bends<br />
The converter can be described as a number of curved and straight waveguides, connected to<br />
each other by junctions, and is depicted schematically in figure 3.19.<br />
This approach allows for a transmission matrix description, of which the complete transfer<br />
from input to output can be described by a transmission matrix T:<br />
with:<br />
whereby u TE and u TM are the complex signal amplitudes of the quasi-TE and quasi-TM modes<br />
respectively.<br />
The transmission matrix T can be decomposed as follows (see figure 3.19):<br />
with:<br />
α<br />
975 μm<br />
83 μm<br />
Figure 3.18 Layout of the <strong>polarisation</strong> converter, where the segment angle is<br />
depicted as α.<br />
in<br />
+<br />
Tsc T cc<br />
T c<br />
T c T c<br />
T cc<br />
T c<br />
-<br />
Tsc whereby T c and T s describe the transmission in a curved section with angle α and in a straight<br />
+<br />
Tsc T cc<br />
T cc<br />
α L α out<br />
T s<br />
T c<br />
T c T c<br />
Figure 3.19 Schematic diagram of the <strong>polarisation</strong> converter.<br />
T u<br />
=<br />
U out<br />
U<br />
=<br />
=<br />
T U in<br />
u TE<br />
u TM<br />
T = T uT sT u<br />
- +<br />
T scT<br />
cT ccT cT cT ccT cT sc<br />
T c<br />
-<br />
Tsc (3.8)<br />
(3.9)<br />
(3.10)<br />
(3.11)
3.7 A compact <strong>polarisation</strong> converter 53<br />
section with length L α respectively:<br />
and:<br />
T c<br />
T s<br />
=<br />
=<br />
exp( – jβϕ, TEα)<br />
0<br />
0 exp – jβϕ, TM<br />
( α)<br />
exp( – jβTELα ) 0<br />
0 exp( – jβTMLα) (3.12)<br />
(3.13)<br />
whereby the exponential terms describe the phase transfer along the waveguide. The matrix T sc<br />
describes the junction between a straight and a curved waveguide, and the matrix T cc describes<br />
the junction between two oppositely curved waveguides. The matrix elements are calculated<br />
using 2D vectorial overlap integrals according to:<br />
+ ηTE,TE – ηTE,TM - ηTE,TE ηTE,TM = , T sc =<br />
ηTM,TE ηTM,TM – ηTM,TE ηTM,TM T sc<br />
T cc<br />
=<br />
η TE,TE η TE,TM<br />
η TM,TE η TM,TM<br />
(3.14)<br />
whereby η B,A is the coupling coefficient between the field of <strong>polarisation</strong> A in the straight or<br />
curved input waveguide, and the field of <strong>polarisation</strong> B in the curved output waveguide (A and<br />
B equal TE or TM), calculated according to equation 3.7.<br />
Results of the calculation of the transmission matrix T with the MOL are shown in figure<br />
3.20a. Although the predicted conversion is very small, a beat can be clearly observed,<br />
especially for the 150 μm bending radius. In any case, low losses in the order of 0.2-0.4 dB are<br />
expected, as the radiation and the junction losses are both low. Based on these results, it is<br />
anticipated that 100% <strong>polarisation</strong> conversion is obtained after ten U-bends with 75 μm<br />
bending radius and with 55 degrees segment angle. The excess loss will then be approximately<br />
1.5 dB.<br />
Experimental results are also available: waveguides have been fabricated in a MOCVD-grown<br />
<strong>InP</strong>/InGaAsP(λ g = 1.3 μm)/<strong>InP</strong> ridge waveguide structure. A 100 nm thick PE-CVD deposited<br />
SiNx film was used as a masking layer and waveguides were etched completely through the<br />
guiding layer into the substrate employing a CH 4 /H 2 reactive-ion etching (RIE) etch/descum<br />
process to reduce the scattering losses [86]. Waveguide losses were measured at 1.0 dB/cm and<br />
2.2 dB/cm for widths of 3.0 μm and 1.4 μm respectively both for TE and TM <strong>polarisation</strong>. The<br />
side wall angle was measured to be approximately 10 degrees off verticality.<br />
A Fabry-Perot laser, operating at a <strong>wavelength</strong> of 1508 nm, was used to measure the<br />
performance of the <strong>polarisation</strong> converters. TM-polarised light was launched into the input<br />
waveguide, and at the output a <strong>polarisation</strong> filter was used for separate measurement of the TE<br />
and TM response. The measured performance of the <strong>polarisation</strong> converter is shown in figure<br />
3.20b. The curves show the behaviour as expected, which indicates that the calculation of the<br />
propagation constants is relatively accurate. The converters with a bending radius of 50 μm<br />
show the most interesting measurement results: a <strong>polarisation</strong> conversion of 85% was<br />
measured at a segment angle of 70 degrees.
54 3. Polarisation conversion in waveguide bends<br />
Conversion [%]<br />
20<br />
15<br />
10<br />
5<br />
R = 50 μm<br />
R = 75 μm<br />
R = 100 μm<br />
R = 150 μm<br />
0<br />
0 20 40 60 80<br />
Segment angle [deg]<br />
Conversion [%]<br />
(a) (b)<br />
10 20 30 40 50 60 70 80<br />
Segment angle [deg]<br />
Low conversion values (
3.8 Tolerance analysis 55<br />
3.8 Tolerance analysis<br />
In this section, the dependence of <strong>polarisation</strong> conversion on variations in the manufacturing<br />
process is described. These variations comprise the layer thickness, the waveguide width, and<br />
also the angle of the etched side wall. Variations of the etch depth do not influence the<br />
<strong>polarisation</strong> conversion as long as the guiding layer is etched through completely.<br />
3.8.1 Layer thickness<br />
The beat length between the TE- and TM-polarised modes is an important parameter of the<br />
converter of figure 3.18. It is not very tolerant to variations of the layer thickness, as shown in<br />
figure 3.22. As can be seen in this figure, the layer thickness has to be controlled accurately.<br />
The guiding layer thickness variation has to be limited to a value of ± 1 % around the designed<br />
thickness in order to keep the beat length variation below ± 10 %. The same beat length<br />
variation limits the cladding layer variation to ± 2.5 %. These are very stringent requirements<br />
on the layer thickness, which make the <strong>polarisation</strong> converter difficult to realise.<br />
Beat Beatlength length deviation variation [%]<br />
100<br />
50<br />
-50<br />
3.8.2 Waveguide width<br />
0<br />
Guiding layer<br />
Cladding layer<br />
-100<br />
-15 -10 -5 0 5 10 15<br />
Thickness deviation [%]<br />
Figure 3.22 Beat length deviation versus layer thickness deviation.<br />
Coupling efficiency<br />
Figure 3.23 shows the radius dependence of the coupling efficiency for different waveguide<br />
widths. The efficiency for TE-polarised input and TM-polarised output fields is calculated<br />
using a two-dimensional vectorial overlap integral. The efficiency for TM-polarised input<br />
fields to TE-polarised output fields is quite similar and is not shown here. Equally, from the<br />
point of power conservation, the efficiency for the coupling of TE-polarised input fields to TEpolarised<br />
output fields is complementary to the curves shown in figure 3.23. In graph (a) the<br />
efficiency is shown for a junction between two oppositely curved waveguides. It shows that the<br />
coupling efficiency remains very low and is hardly influenced by the waveguide width.
56 3. Polarisation conversion in waveguide bends<br />
Coupling efficiency [%]<br />
10<br />
8<br />
6<br />
4<br />
2<br />
0<br />
W = 1.2 μm<br />
W = 1.4 μm<br />
W = 1.6 μm<br />
20 40 60 80 100<br />
Radius [μm]<br />
(a) (b)<br />
20 40 60 80 100<br />
Radius [μm]<br />
The junction efficiency from a straight to a curved waveguide depicted in figure 3.23b on the<br />
other hand depends strongly on the waveguide width. But if the radius is decreased to<br />
15-25 μm, the width dependence is greatly reduced. This small-radius region has been zoomed<br />
in and is depicted in figure 3.24. The graph shows that high efficiencies can be obtained,<br />
especially for small waveguide widths, which can be explained by the fact that for small radii<br />
the modal field in the bend is guided by the outer edge alone, and becomes <strong>independent</strong> of the<br />
inner edge. Such a mode is called a “whispering-gallery” (WG) mode after Lord Rayleigh<br />
[109], who applied this term to acoustic waves guided by a single curved surface which he had<br />
found in St. Paul’s cathedral. The field of such a WG mode is highly compressed and becomes<br />
asymmetrical, leading to high coupling losses to the symmetrical field of the straight<br />
waveguide. Pennings [103] presented an approximal formula for estimating the width and<br />
Coupling efficiency [%]<br />
100<br />
80<br />
60<br />
40<br />
20<br />
0<br />
W = 1.2 μm<br />
W = 1.4 μm<br />
W = 1.6 μm<br />
Figure 3.23 TE to TM coupling efficiency versus bending radius for different<br />
waveguide widths: curved to curved waveguide (a), and straight to curved<br />
waveguide (b).<br />
Coupling efficiency [%]<br />
100<br />
95<br />
90<br />
85<br />
R = 15 μm<br />
R = 20 μm<br />
R = 25 μm<br />
80<br />
1.10 1.20 1.30 1.40 1.50 1.60 1.70 1.80<br />
Waveguide width [μm]<br />
Figure 3.24 TE to TM coupling efficiency versus bending radius for different<br />
waveguide widths.
3.8 Tolerance analysis 57<br />
radius at which this occurs. For the very small radii (15 to 25 μm) the width is found to be<br />
between 1.5 and 1.6 μm.<br />
Another interesting feature of figure 3.23b is the curve for the 1.2 μm waveguide width. This<br />
curve shows a different behaviour than the ones for a width of 1.4 and 1.6 μm, which is due to<br />
the fact that the assignment to TE or TM <strong>polarisation</strong> of a modal solution found with the 2Dsolver<br />
is done on the basis of continuous evolution of the mode index (see note on page 51).<br />
The problem arising from the 1.2 μm waveguides is that for a straight waveguide the mode<br />
index for TE <strong>polarisation</strong> is lower than that of the TM polarisated mode. This is shown in<br />
figure 3.25, in which the mode indices (calculated with the vectorial FEM) are depicted as a<br />
function of the waveguide width.<br />
Mode index<br />
3.285<br />
3.280<br />
3.275<br />
3.270<br />
3.265<br />
3.260<br />
TE<br />
TM<br />
3.255<br />
1.20 1.30 1.40 1.50 1.60<br />
Waveguide width [μm]<br />
Figure 3.25 Effective index of the fundamental TE- and TM-polarised mode in a<br />
straight waveguide as a function of the width.<br />
Beat length<br />
Figure 3.26 shows the beat length between the TE- and TM-polarised mode versus the bending<br />
radius at different waveguide widths. From this figure it can be seen that the waveguide width<br />
does not only have a large impact on the coupling efficiencies at the junctions but also on the<br />
beat length. Two areas are of major interest for the application of <strong>polarisation</strong> conversion.<br />
Firstly the region with radii in the order of 15-25 μm is of interest as the beat length is very<br />
short (100-200 μm) and the dependence on the waveguide width vanishes. The disadvantage of<br />
this region, however, is that to obtain a curved waveguide with a propagation length equal to<br />
the beat length, the rotation angle must be higher than 360 degrees. Additionally, the radiation<br />
losses, which are listed in table 3.3, increase with decreasing radius and width, leading to<br />
(calculated) losses in the order of 1.2-1.8 dB for a bend with a 360-degree rotation angle.<br />
The second interesting region is the one with bending radii in the range from 70 to 100 μm. In<br />
this range the beat length is almost <strong>independent</strong> of the bending radius if the waveguide width<br />
equals 1.2 or 1.6 μm. In this case the beat length will be around 265 μm, leading to a rotation<br />
angle of approximately 150 degrees and therefore the design of a <strong>polarisation</strong> converter is<br />
feasible. In addition, as the radiation loss is negligible, low-loss operation is expected.
58 3. Polarisation conversion in waveguide bends<br />
Beat Beatlength length [μm]<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
W = 1.2 μm<br />
W = 1.4 μm<br />
W = 1.6 μm<br />
20 40 60 80 100<br />
Radius [μm]<br />
Figure 3.26 Beat length versus bending radius for different waveguide widths.<br />
Table 3.3 Calculated radiation losses for TE/TM <strong>polarisation</strong> at different<br />
waveguide widths.<br />
W [μm] R = 15 μm R = 20 μm R = 25 μm<br />
1.2 0.34 / 0.46 0.18 / 0.24 0.10 / 0.12<br />
1.4 0.31 / 0.38 0.15 / 0.17 0.08 / 0.08<br />
1.6 0.30 / 0.36 0.14 / 0.15 0.07 / 0.06<br />
The performance of the <strong>polarisation</strong> converter (see figure 3.19) has been calculated for the<br />
nominal width of 1.4 μm and a width deviation ΔW of ±<br />
0.2 μm. For the bending radius a<br />
value of 75 μm has been chosen. The results are depicted in figure 3.27, which shows that the<br />
<strong>polarisation</strong> converter is very sensitive to width deviations.<br />
Side wall angle<br />
During the etching of the waveguide using a CH 4 /H 2 -plasma, polymer deposition occurs on<br />
top of the masking material, which results in broadening of the mask. The polymer can be<br />
removed by descumming with oxygen after a limited time of etching. The etching process<br />
therefore consists of a number of alternating descumming and etching steps. The ratio of the<br />
etching and descumming time influences the side wall angle, which has been investigated by<br />
our laboratory and is presented elsewhere [86].<br />
If a short descumming time is used, the etched side wall will be close to vertical, as well as<br />
being very rough. In order to obtain a smooth side wall and, consequently, low propagation<br />
losses, a longer descumming time is required. This, however, results in a less steep side wall<br />
and therefore a compromise has to be made.
3.8 Tolerance analysis 59<br />
Conversion [%]<br />
100<br />
80<br />
60<br />
40<br />
20<br />
ΔW = -0.2 μm<br />
ΔW = 0.0 μm<br />
ΔW = +0.2 μm<br />
0<br />
0 20 40 60 80<br />
Segment angle [deg]<br />
Figure 3.27 Calculation of the <strong>polarisation</strong> converter performance with a 75 μm<br />
bending radius for different waveguide width deviations ΔW.<br />
Figure 3.28 shows a scanning electron microscope (SEM) picture of the cross section of a<br />
deeply etched waveguide (1.2 μm deep). The etched side wall of the waveguide shows a<br />
deviation from the vertical of 8 to 10 degrees. As a result, the field in the waveguide will follow<br />
this interface, an effect which is used by other types of <strong>polarisation</strong> converter (see, e.g., [157]).<br />
The effect of a non-vertical side wall has been analysed by making a staircase approximation<br />
of the angled side wall using a small number of steps (3). In order to obtain the effective width<br />
deviation, the overlap to a field of a waveguide with vertical side walls has been calculated. It<br />
has been found that a side wall angle of 8 to 10 degrees has the same effect as a width deviation<br />
of 0.1 to 0.2 μm for bending radii in the range of 50-100 μm. This is shown schematically in<br />
figure 3.29.<br />
Figure 3.28 SEM picture of the waveguide.
60 3. Polarisation conversion in waveguide bends<br />
W W+ΔW<br />
(a) (b)<br />
Figure 3.29 Schematic diagram of width increase due to a non-vertical side wall:<br />
the width of the mode in (a) is equal to the mode width of (b)<br />
The performance of the <strong>polarisation</strong> converter has been calculated 1 with a side wall angle of<br />
10 degrees taken into account, by using a width deviation of +0.2 μm. The results are shown as<br />
lines in figure 3.30 for different radii of curvature. The measurement data are also depicted in<br />
the figure as symbols (see also figure 3.20). The measurements and calculations correspond,<br />
indicating that a non-vertical side wall angle has a large impact on the <strong>polarisation</strong> conversion.<br />
Conversion [%]<br />
100<br />
80<br />
60<br />
40<br />
20<br />
R = 50 μm<br />
R = 75 μm<br />
R = 100 μm<br />
R = 150 μm<br />
0<br />
0 20 40 60 80<br />
Segment angle [deg]<br />
Figure 3.30 Calculated <strong>polarisation</strong> converter performance for an increased<br />
waveguide width of 0.2 μm (lines), together with measurement results (symbols).<br />
The <strong>polarisation</strong> converters presented in this chapter have also been analysed by Lui et al. [78].<br />
He solved the wave equations with a transformed index profile with a 2D mode solver. By<br />
using the coupled-mode theory in combination with the initial field conditions and their first<br />
derivatives in the direction of propagation, a solution can be obtained for the propagation<br />
constant and the corresponding field. By taking the side wall angle into account, Lui obtained<br />
1. By using the fact that a width deviation has the same effect as a non-vertical side wall<br />
angle, the calculations can also be done with the MOL, as with the version available in our<br />
laboratory only three lateral regions can be taken into account and therefore no staircase<br />
approximation of the side wall angle can be used.
3.9 Avoiding <strong>polarisation</strong> conversion 61<br />
beat length values of 100, 120 and 130 μm for bending radii of 50, 75 and 100 μm respectively<br />
(measured values: 120,130, and 140 μm). In addition, his method also predicts a high<br />
<strong>polarisation</strong> conversion. The values increase almost linearly from 65% to 85% when the<br />
segment angle increases from 70 to 80 degrees. The bending radii at which these values are<br />
obtained are slightly less than measured (40-60 μm). The predictions obtained with this<br />
method are in the line with experimental data.<br />
In view of the above results, it has been established that a variation of the side wall angle has a<br />
large impact on the performance of the <strong>polarisation</strong> converter. The angle therefore should be<br />
controlled accurately in order to be able to manufacture repetitively devices with a high degree<br />
of <strong>polarisation</strong> conversion.<br />
3.9 Avoiding <strong>polarisation</strong> conversion<br />
To date design has been aimed at optimum <strong>polarisation</strong> conversion. In many cases, however,<br />
<strong>polarisation</strong> conversion is highly undesirable. As the graphs of figure 3.23 suggest, large radii<br />
provide a solution to this problem, but then the advantage of a deeply etched waveguide<br />
structure is lost completely, so a solution for small radii must be found. The graph in figure<br />
3.24 shows that the coupling efficiency from a TE- to a TM-polarised field at the junction<br />
between a straight and a curved waveguide decreases with increasing waveguide width.<br />
Further investigation reveals that the earlier mentioned WG modes show a negligible coupling<br />
efficiency, even at very small radii. This is depicted in figure 3.31 for different waveguide<br />
widths of the curved waveguide. The graph in figure 3.31a shows that even radii down to<br />
40 μm can be applied with coupling efficiencies below -20 dB. This low TE-TM coupling is<br />
obtained at the expense of rather high coupling losses for TE-TE coupling, which is the result<br />
of the strong asymmetry of the WG mode (figure 3.31b). They can be minimised, however, to<br />
values well below 0.5 dB, while still using very small bending radii. For instance, at a radius of<br />
50 μm only 0.3 dB coupling loss is obtained with a curved waveguide width of 2.5 μm.<br />
The fact that WG bends show negligible <strong>polarisation</strong> conversion is an important and newly<br />
discovered feature of the deeply etched waveguide bends. This paves the way for<br />
miniaturisation of components without having to cope with problems arising from <strong>polarisation</strong><br />
conversion. Furthermore, the <strong>polarisation</strong> conversion shows a negligible dependence on the<br />
waveguide width (see figure 3.31a), which makes it very tolerant to variations in the<br />
Coupling loss [dB]<br />
0<br />
-10<br />
-20<br />
-30<br />
W = 2.5 μm<br />
W = 3.0 μm<br />
W = 3.5 μm<br />
W = 4.0 μm<br />
-40<br />
20 40 60<br />
Radius [μm]<br />
80 100<br />
-3<br />
20 40 60<br />
Radius [μm]<br />
80 100<br />
(a) (b)<br />
Coupling loss [dB]<br />
0<br />
-1<br />
-2<br />
W = 2.5 μm<br />
W = 3.0 μm<br />
W = 3.5 μm<br />
W = 4.0 μm<br />
Figure 3.31 Coupling efficiency between a straight and a curved waveguide for<br />
different widths of the curved waveguide: TE to TM (a), and TE to TE (b).
62 3. Polarisation conversion in waveguide bends<br />
production process (e.g. side wall angle), and the coupling losses can be made sufficiently low<br />
(see figure 3.31b) even at very small bending radii.<br />
3.10 Conclusion<br />
In this chapter the <strong>polarisation</strong> conversion occuring in deeply etched waveguide bends has<br />
been discussed. Aiming at high conversion, a novel type of <strong>polarisation</strong> converter using deeply<br />
etched narrow <strong>InP</strong>/InGaAsP ridge waveguide bends with small bending radii has been<br />
analysed. The device has been produced and combines high <strong>polarisation</strong> conversion (>85%)<br />
with low loss (2.7 dB) and compact device size ( μm2 975 ×<br />
83 ). However, this type of<br />
converter is difficult to produce repetitively because it is very sensitive to variations in the<br />
waveguide structure.<br />
On the other hand, in many cases <strong>polarisation</strong> conversion is not desirable. Calculations have<br />
shown that the usage of whispering gallery bends can solve this problem. It has been shown<br />
that when this type of bends is used even at very small bending radii (down to 40 μm), the<br />
<strong>polarisation</strong> conversion is negligibly low. These bends are also tolerant to variations in the<br />
width and, consequently, the side wall angle. Additionally, the coupling losses at junctions to<br />
straight waveguides remain sufficiently low (a few tenths of dB’s). Therefore, they provide a<br />
good solution to the miniaturisation of integrated optical components, without having to cope<br />
with undesired <strong>polarisation</strong> conversion.
Chapter 4<br />
Polarisation <strong>independent</strong> PHASAR<br />
<strong>demultiplexers</strong> <strong>based</strong> on<br />
nonbirefringent waveguides<br />
A low-loss PHASAR demultiplexer <strong>based</strong> on raised-strip waveguides is described which has<br />
been made <strong>polarisation</strong> <strong>independent</strong> by properly shaping the array waveguides in such a way<br />
as to eliminate their birefringence. Furthermore, a method for further reducing the PHASAR<br />
loss is presented and the influence of mode conversion on the cross talk of the device is<br />
analysed. Finally, a packaged device integrated with photodetectors is described.<br />
4.1 Introduction<br />
In general, a waveguide produced in a semiconductor material has different propagation<br />
constants for TE and TM <strong>polarisation</strong>. The result is a shift of the spectral responses with<br />
respect to each other which is called <strong>polarisation</strong> dispersion. This is shown schematically in<br />
figure 4.1 (see also section 2.2.4).<br />
TMm+1 TMm+1 TMm TEm TMm-1 TEm-1 <strong>polarisation</strong> dispersion<br />
<strong>wavelength</strong><br />
Figure 4.1 Schematic diagram of the periodic <strong>wavelength</strong> response of a phasedarray<br />
demultiplexer with the <strong>polarisation</strong> dispersion indicated.
64 4. A <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>based</strong> on nonbirefringent waveguides<br />
In order to eliminate the <strong>polarisation</strong> dispersion, it is obvious that the attention should be<br />
focused on the waveguide itself. If the array waveguides are <strong>polarisation</strong> <strong>independent</strong>, the total<br />
response of the array will be <strong>polarisation</strong> <strong>independent</strong> too. Waveguides can be made<br />
<strong>polarisation</strong> <strong>independent</strong> by using a burried waveguide structure [11,123], or a raised-strip<br />
waveguide [9,19,20,21,122, 164]. A comprehensive discussion of both structures with respect<br />
to the tolerances is presented, together with experimental results of devices using the raisedstrip<br />
waveguide (part of which have already been presented [164]). This can be found in<br />
section 4.2. The experiments clearly demonstrate that <strong>polarisation</strong> <strong>independent</strong> behaviour can<br />
be achieved over a large <strong>wavelength</strong> range. This is an important advantage of these<br />
waveguides, together with low fibre-chip coupling losses and compact device dimensions.<br />
The insertion loss of a phased-array demultiplexer is mainly determined by the fibre-chip<br />
coupling losses and the losses at the transition between the array waveguides and the Free<br />
Propagation Region (FPR) as described in section 2.2.6. An optical microscope image of the<br />
transition is shown in figure 4.2. The fibre-chip coupling loss can be minimised by applying a<br />
fibre-matched waveguide structure. The losses at the transition, however, are not reduced<br />
easily.<br />
Free<br />
Propagation<br />
Region<br />
Array<br />
Waveguides<br />
Figure 4.2 Optical microscope image of the transition between the array<br />
waveguides and the Free Propagation Region.<br />
In theory, the best solution is to make the transition adiabatical, which means that the gap<br />
between the array waveguides reduces gradually to zero. Due to the finite resolution of the<br />
lithographic process, the gaps will not, however, be etched open completely leaving us with an<br />
abrupt transition, which will introduce transition loss. Additionally, the gap obtained in this<br />
way is not uniformly distributed along the array aperture (which can be seen clearly in figure<br />
4.2), resulting in an increased cross talk level due to phase errors. The transition loss can also<br />
be reduced by decreasing the confinement of the waveguides. At the transition, the fields of<br />
adjacent array waveguides will overlap and the resulting sum field will have a smaller ripple,<br />
leading to an increased overlap with the field in the FPR. However, as a decreased confinement<br />
of the waveguide requires an increase of the minimum bending radius (and thus an increase of<br />
the device size), this should be done locally at the transition using a two-step etching process.<br />
We applied this method and demonstrated its effectiveness [33]. The method is discussed in<br />
section 4.3.
4.2 Nonbirefringent waveguides 65<br />
Usually, waveguides used in phased-array <strong>demultiplexers</strong> are monomode. If the first-order<br />
mode is not cut-off, the cross talk performance of the demultiplexer can be deteriorated. At the<br />
junction between a straight and a curved waveguide, a fraction of the fundamental mode will<br />
couple to the first-order mode. Constructive interference of first-order modes in the array<br />
waveguides can lead to the occurence of so-called “ghost” images at a different position in the<br />
image plane. This shift in position corresponds to the difference in propagation constants<br />
between the fundamental and first-order mode, the mechanism being equal to <strong>polarisation</strong><br />
dispersion. We will therefore denote this shift as the modal dispersion shift. We identified that<br />
“ghost” images may contribute significantly to the cross talk of experimental devices. The<br />
occurence of these “ghost” images is discussed in section 4.4. Also a method is suggested in<br />
order to avoid them and experimental results are included.<br />
Additionally, a waveguide detector structure for the raised-strip waveguide has been designed,<br />
<strong>based</strong> on evanescent field coupling. Detectors of this type have been integrated with a<br />
PHASAR demultiplexer and the resulting optical receiver has been packaged in an industrystandard<br />
butterfly package. It is the first compact packaged <strong>InP</strong> PHASAR-<strong>based</strong> demultiplexer<br />
with integrated detectors. This work has been presented at the ECOC’97 [137] and is described<br />
in section 4.5.<br />
It has been noted that the use of the raised-strip waveguide structures was proposed by<br />
Amersfoort [7]. His work has been extended here with an elaborate analysis of the tolerances<br />
and the performing of experiments, which include insertion loss [33] and cross talk decrease<br />
(due to “ghost” images) [37], and the design and integration of detectors [137].<br />
The work presented in this chapter has been carried out in cooperation with Philips<br />
Optoelectronics Centre (POC) within the framework of the RACE 2070 MUNDI (Multiplexed<br />
Network for Distributive and Interactive Services) project, and in the framework of the ACTS<br />
AC028 TOBASCO (Towards Broadband Access Systems for CATV Optical networks) project.<br />
4.2 Nonbirefringent waveguides<br />
The most straightforward way to obtain <strong>polarisation</strong> independence is to make the array<br />
waveguides <strong>polarisation</strong> <strong>independent</strong> by eliminating birefringence. Clearly, a square<br />
waveguide embedded in a homogeneous medium with a constant refractive index, as shown in<br />
figure 4.3a, is <strong>polarisation</strong> <strong>independent</strong>, as the index contrast is the same in both the transverse<br />
and the lateral direction. But rectangular raised-strip waveguides, depicted in figure 4.3b, can<br />
also be made <strong>polarisation</strong> <strong>independent</strong> by a proper choice of the aspect ratio (height/width) of<br />
the waveguide core. The embedded square waveguide will be discussed in section 4.2.1, and<br />
the raised-strip waveguide in sections 4.2.2 to 4.2.3.<br />
InGaAsP<br />
<strong>InP</strong><br />
W<br />
(a)<br />
Figure 4.3 The buried waveguide (a), and the raised-strip waveguide (b)<br />
W<br />
W<br />
(b)<br />
D
66 4. A <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>based</strong> on nonbirefringent waveguides<br />
4.2.1 The embedded square waveguide<br />
As monomode waveguides are preferred, we first look at the waveguide dimensions and the<br />
composition of the quaternary layer in more detail. In figure 4.4 the V-parameter<br />
2<br />
2 1 2<br />
( V k0W ( nInGaAsP – n<strong>InP</strong>) ) is depicted as a function of the waveguide width (and<br />
thickness) and the composition of the quaternary layer. As can be seen in the picture, small<br />
waveguide dimensions are needed for monomode operation (i.e. ): for quaternary<br />
material with a band-edge <strong>wavelength</strong> higher than 1.12 μm the dimensions move into the<br />
submicron range.<br />
This puts rather stringent demands on the manufacturing feasibility of the photolithographic<br />
process. In order to relax these demands, the index contrast between the quaternary layer and<br />
the <strong>InP</strong> substrate can be minimised, which, on the other hand, will increase the radiation losses<br />
in the bends.<br />
⁄<br />
=<br />
V ≤ π<br />
Band-edge Bandgap <strong>wavelength</strong> [μm]<br />
1.30<br />
1.25<br />
1.20<br />
1.15<br />
1.10<br />
1.05<br />
1.00<br />
0.95<br />
π<br />
2π<br />
3π<br />
1 2 3 4 5<br />
Waveguide width [μm]<br />
Figure 4.4 Contour plot denoting the V-parameter for a square waveguide, as a<br />
function of the width (thickness) and the composition of the quaternary layer.<br />
The radiation losses are depicted in figure 4.5 for different compositions of the quaternary<br />
layer, the waveguide dimensions for each of which have been adjusted according to figure 4.4<br />
to obtain monomode operation. The graph shows that small bending radii are possible if a high<br />
index contrast is applied, but, consequently, with small waveguide dimensions. On the other<br />
hand, a low index contrast leads to an increase of the waveguide dimensions, and, additionally,<br />
the bending radius. Although this waveguide is tolerant to dimensional variations as<br />
demonstrated by Soole et al. [123], it requires accurate control of the composition of the<br />
quaternary layer and an increased complexity of the fabrication process due to an additional<br />
epitaxial growth step, during which small gaps between relatively thick and wide waveguides<br />
have to be filled completely (for instance at the junction between the FPR and the array<br />
waveguides).<br />
4π<br />
5π<br />
2π<br />
6π<br />
4π<br />
7π<br />
5π<br />
3π
4.2 Nonbirefringent waveguides 67<br />
Radiation loss [dB/90]<br />
100.000<br />
10.000<br />
1.000<br />
0.100<br />
0.010<br />
0.001<br />
4.2.2 The raised-strip waveguide<br />
Q(0.95)<br />
Q(1.00)<br />
Q(1.05)<br />
Q(1.10)<br />
200 400 600 800 1000<br />
Bending radius [μm]<br />
Figure 4.5 Radiation loss versus bending radius for monomode square<br />
waveguide for different compositions of the quaternary layer. (Q(0.95) denotes<br />
quaternary material with a band-edge <strong>wavelength</strong> of 0.95 μm.)<br />
The rectangular raised-strip waveguide, which is shown in figure 4.3b, does not require an<br />
epitaxial regrowth and has the advantage that it allows for small bending radii and,<br />
consequently, compact design, due to the high lateral index contrast, even with a small GaAsfraction<br />
in the quaternary layer (low contrast). Additionally, it can be made relatively wide and<br />
thick (ideal for small fibre-chip coupling loss) and still maintain its monomode nature.<br />
The basic principle is that the birefringence induced by the asymmetry in lateral and transverse<br />
index contrast is compensated by a small correction of the aspect ratio (height/width) of the<br />
waveguide core. As the penetration depth of the field is larger for TE <strong>polarisation</strong> than for TM<br />
<strong>polarisation</strong>, the waveguide should be made slightly wider than high in order to obtain identical<br />
field distributions and propagation constants. This was demonstrated by Chiang [26], who presented<br />
a simple equation for the correct width of the waveguide core needed to obtain <strong>polarisation</strong><br />
independence, which is <strong>based</strong> on the fact that the modal field distributions in a waveguide<br />
with high index contrast can be approximated by a sine function:<br />
2<br />
nQ 2 2<br />
2πn Q N TE<br />
– 1<br />
W = λ ---------------------------------------------<br />
2<br />
( – )<br />
whereby n Q is the refractive index of the quaternary layer, and N TE and N TM are the transverse<br />
effective indices in the centre region for TE and TM <strong>polarisation</strong> respectively. Due to the high<br />
lateral index contrast, a mode close to cut-off will radiate into the substrate, which can be<br />
examined by using Marcatili’s approximation for the propagation constant [80]:<br />
β 2<br />
2 2<br />
konQ 2<br />
kx with N x being the effective index in the centre region, and k o , k x and k y the wave numbers in<br />
vacuum in the transverse and the lateral direction respectively. Due to the high lateral index<br />
2<br />
N TM<br />
2 2<br />
N x<br />
1 ⁄ 3<br />
(4.1)<br />
= – – ky = ko – k (4.2)<br />
y<br />
2
68 4. A <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>based</strong> on nonbirefringent waveguides<br />
contrast, the lateral field distribution may be approximated by a sine function, with which we<br />
find for the lateral wavenumber:<br />
k y<br />
( m + 1)π<br />
= ---------------------<br />
whereby m is the lateral mode number. Using the above equations, combined with the fact that<br />
the cut-off condition occurs when the mode index becomes less than the substrate index n <strong>InP</strong> ,<br />
we find for the cut-off width W m of the lateral mode m:<br />
W m<br />
According to this equation, for each mode that propagates in the waveguide, a relation between<br />
waveguide width and thickness can be calculated for which the mode index equals the<br />
substrate index. This relation is called the cut-off condition, and the cut-off conditions of the<br />
three lowest-order modes are shown in figure 4.6 for a Q(0.97) and a Q(1.02) waveguide layer.<br />
The cut-off conditions are almost identical for TE and TM <strong>polarisation</strong>. Monomode operation<br />
is possible for a wide range of waveguide widths, as depicted by the hatched area in the figure.<br />
Also shown in the graph is the zero birefringence condition denoted by Δλ pol = 0. A proper<br />
choice for the waveguide width and thickness is one close to the maximum allowed value for<br />
monomode operation [39]. The field at the side walls will then be minimal, which will reduce<br />
the scattering losses.<br />
Width Q(0.97) [μm]<br />
5<br />
4<br />
3<br />
2<br />
TE TE/TM 00 /TM 00<br />
TE TE/TM 10 /TM 10<br />
TE TE/TM 01 /TM 01<br />
1<br />
1 2 3<br />
Thickness Q(0.97) [μm]<br />
4 5<br />
The dependence of the zero-birefringence condition on the composition of the quaternary<br />
material is shown in figure 4.7. It has been calculated for the correct aspect ratios for a Q(0.97)<br />
and a Q(1.02) guiding layer (because both types have been used for experiments), which are of<br />
3.4 μm and 3.0 μm width respectively with a 2.2 μm waveguide thickness. In this figure it can<br />
be seen that the raised-strip waveguide is rather insensitive to variations in the composition of<br />
the quaternary layer.<br />
W m<br />
( m + 1)λ<br />
= -----------------------------<br />
2 2<br />
2 N x – n<strong>InP</strong><br />
Width Q(1.02) [μm]<br />
(a) (b)<br />
5<br />
Δλ Δf pol = 0 Δλ Δf pol = 00<br />
4<br />
3<br />
2<br />
(4.3)<br />
(4.4)<br />
TE TE/TM 00 /TM 00<br />
TE TE/TM 10 /TM 10<br />
TE TE/TM 01 /TM 01<br />
1<br />
1 2 3<br />
Thickness Q(1.02) [μm]<br />
4 5<br />
Figure 4.6 Raised-strip waveguide cut-off conditions for a Q(0.97) layer (a) and<br />
a Q(1.02) layer (b). The hatched area denotes the monomode region. The line<br />
with TE ij /TM ij denotes as a function of the thickness the width, for smaller values<br />
of which the i-th transverse and j-th lateral TE and TM mode are cut-off.
4.2 Nonbirefringent waveguides 69<br />
A good strategy for designing a tolerant waveguide structure (from a production feasibility<br />
point of view) is to choose the InGaAs-fraction of the quaternary layer as low as possible (i.e.<br />
low band-edge <strong>wavelength</strong>). This follows directly from the graphs (a) and (b) of figure 4.8,<br />
where it can be seen that the sensitivity of the birefringence to width and thickness variations<br />
increases rapidly with increasing InGaAs-fraction. For each composition of the quaternary<br />
layer, the calculations have been performed on waveguides with dimensions close to the<br />
maximum allowed values for monomode operation.<br />
Δλ pol [nm]<br />
Δλ pol [nm]<br />
0.15<br />
0.10<br />
0.05<br />
0.00<br />
-0.05<br />
-0.10<br />
-0.15<br />
-0.20<br />
4.2.3 Experimental results<br />
W/D = 3.4/2.2<br />
W/D = 3.0/2.2<br />
Q(0.97)<br />
Q(1.02)<br />
-0.25<br />
0.95 1.00 1.05 1.10 1.15 1.20 1.25 1.30<br />
Band-edge Bandgap <strong>wavelength</strong> <strong>wavelength</strong> [μm]<br />
Figure 4.7 Composition dependence of the zero-birefringence condition.<br />
1.0<br />
0.5<br />
0.0<br />
-0.5<br />
-1.0<br />
Q(0.95)<br />
Q(1.00)<br />
Q(1.05)<br />
Q(1.10)<br />
-0.4 -0.2 0.0 0.2 0.4<br />
ΔW [μm]<br />
(a) (b)<br />
-0.4 -0.2 0.0 0.2 0.4<br />
ΔD [μm]<br />
Figure 4.8 Waveguide width (a) and thickness (b) dependence of the zerobirefringence<br />
condition.<br />
A 8-channel raised-strip-waveguide-<strong>based</strong> PHASAR has been designed and produced in<br />
cooperation with Philips Optoelectronics Centre (POC) [164]. The raised-strip waveguide was<br />
designed for <strong>polarisation</strong> independence for a Q(0.97) guiding layer, and therefore the width<br />
and thickness were chosen at 3.4 μm and 2.2 μm respectively. Additionally, due to its large<br />
Δλ pol [nm]<br />
1.0<br />
0.5<br />
0.0<br />
-0.5<br />
-1.0<br />
Q(0.95)<br />
Q(1.00)<br />
Q(1.05)<br />
Q(1.10)
70 4. A <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>based</strong> on nonbirefringent waveguides<br />
core dimensions, this strip waveguide allows for a high coupling efficiency (loss dB) to a<br />
lensed fibre. The channel spacing was designed at 2.0 nm (250 GHz) at a centre <strong>wavelength</strong> of<br />
1536 nm, determined by the specifications of the RACE 2070 MUNDI project. The gap<br />
between the array waveguides was chosen at 0.8 μm (da = 4.2 μm), the smallest feasible gap<br />
size in order to minimise device dimensions. The gap between the receiver waveguides was<br />
chosen at 1.5 μm (dr = 4.9 μm), which is sufficient to achieve a theoretical channel isolation<br />
better than -40 dB. The resulting array size is mm2 , measured between object and<br />
image plane, and the overall device size is mm2 ± 1<br />
1.4 × 1.7<br />
2.6 × 2.3 . A microscope photograph of the<br />
device is shown in figure 4.9.<br />
Figure 4.9 Microscope photograph of the realised PHASAR with raised-strip<br />
waveguides.<br />
Low-pressure OMVPE (Organo-Metallic Vapour Phase Epitaxy) has been used to grow the<br />
undoped guiding layer and RIE (Reactive Ion Etching) using Cl 2 has been applied to etch the<br />
waveguides with smooth side walls for low propagation loss. This loss is 2-3 dB/cm for both<br />
TE and TM <strong>polarisation</strong> and no radiation loss could be measured on curved waveguides using<br />
a Fabry-Pérot measurement technique [168] on uncoated waveguides.<br />
The measured response is depicted in figure 4.10a for both TE (solid lines) and TM<br />
<strong>polarisation</strong> (dashed lines). The cross talk is better than -25 dB and the on-chip loss is 4-5 dB.<br />
The figure clearly shows a <strong>polarisation</strong> <strong>independent</strong> behaviour and a <strong>polarisation</strong> <strong>independent</strong><br />
on-chip loss. To demonstrate the large bandwidth of <strong>polarisation</strong> independence, the response of<br />
a single channel was measured in different wavelenght windows, corresponding with different<br />
orders (43, 44, and 45). The measured FSR of 32 nm corresponds with the design value of<br />
33 nm. Polarisation independence was measured over a <strong>wavelength</strong> span of 64 nm, determined<br />
by the maximum wavelenght span of the tunable laser. From these measurements, as depicted<br />
in figure 4.10b, another important feature of this PHASAR becomes evident, which is the<br />
<strong>wavelength</strong> <strong>independent</strong> insertion loss.
4.3 Loss reduction 71<br />
On-chip loss [dB]<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
1510 1515 1520 1525 1530<br />
Wavelength [nm]<br />
4.2.4 Conclusions<br />
In this section the method for obtaining <strong>polarisation</strong> independence by using nonbirefringent<br />
waveguides has been presented. Two waveguide structures were analysed: the embedded<br />
square waveguide and the raised-strip waveguide. The first has the advantage that it is<br />
inherently <strong>polarisation</strong> <strong>independent</strong>, due to the identical lateral and transverse index contrast.<br />
For low-contrast waveguides, however, the bending radius becomes unpractically large,<br />
whereas the size of high-contrast waveguides gets into the submicron range, which puts<br />
stringent demands on the lithography. Additionally, an expitaxial regrowth is needed. The<br />
raised-strip waveguides at the other hand only need one single etching step for the production.<br />
Furthermore, they can be made relatively wide and thick, while maintaining their monomode<br />
nature. Also small bending radii can be used (even with a small GaAs-fraction in the guiding<br />
layer) due to the high lateral index contrast. Both structures have low fibre-chip coupling losses<br />
due to the circular mode profile, which gives a good match to the fibre mode. The raised-strip<br />
waveguide, however, has a higher potential for application because compact devices can be<br />
produced with a relatively simple manufacturing process.<br />
4.3 Loss reduction<br />
(a) (b)<br />
A significant part of the phased-array loss occurs at the junction between the array waveguides<br />
and the FPR. At this junction the field in the waveguide section shows a very deep modulation<br />
(figure 2.10a, solid line) caused by the trenches between the array waveguides. Due to the<br />
ripple in the field pattern, a considerable fraction of the power is diffracted into adjacent<br />
orders. As the coupling efficiency is found from the overlap of the sum field of the array<br />
waveguides and the far field (dashed line), filling the zeros between the individual waveguide<br />
modes (figure 2.10b, solid line) allows for an increase of the coupling efficiency. This can be<br />
obtained by the insertion of a shallowly etched transition region (TR) between the deeply<br />
etched array waveguides and the FPR [33]. This method has successfully been applied earlier<br />
to waveguide bends in order to reduce the bending and scattering losses [102].<br />
On-chip loss [dB]<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
1480 1482 1511 1513<br />
Wavelength [nm]<br />
1544 1546<br />
Figure 4.10 Measured <strong>wavelength</strong> response of all channels (a), and of a single<br />
channel for different orders of operation (b), both for TE (solid lines) and TM<br />
<strong>polarisation</strong> (dashed lines).
72 4. A <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>based</strong> on nonbirefringent waveguides<br />
AW<br />
TR<br />
FPR<br />
3.0 μm 1.0 μm<br />
The coupling from the FPR to the array waveguides and vice versa now takes place in two<br />
steps: (1) at the junction between the TR and the array waveguides (AW), and (2) at the<br />
junction between the TR and the FPR. Clearly, the first coupling becomes less efficient when<br />
the TR thickness t is increased, whereas at the same time the second one becomes more<br />
efficient, and therefore an optimum in the combined coupling efficiency can be found as shown<br />
in figure 4.12a (solid line).<br />
In this graph the calculated coupling losses (for TE <strong>polarisation</strong>) are depicted for a 2.2 μm<br />
thick, 3.0 μm wide <strong>polarisation</strong> <strong>independent</strong> raised strip waveguide [164] with a band-edge<br />
<strong>wavelength</strong> of the quaternary (InGaAsP) layer of 1.02 μm. The dashed line denotes the<br />
coupling loss between the FPR and the TR (FPR-TR), and decreases to zero with increasing t.<br />
The coupling loss between the TR and the array waveguides (TR-AW) on the other hand,<br />
increases with increasing t as depicted by the dotted line. The combined coupling loss is<br />
denoted by the solid line. The total coupling loss without TR (i.e. t = 0 μm) is 3.5 dB.<br />
Coupling loss [dB]<br />
2.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
2.2 μm<br />
Figure 4.11 Schematic topview (left) and cross section (right) including fields of<br />
the transition region (TR) between the FPR and the array guides (AW).<br />
FPR-TR<br />
TR-AW<br />
Total<br />
0.0<br />
0.5 1.0 1.5 2.0<br />
TR thickness t [μm]<br />
Coupling loss reduction [dB]<br />
2.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
t<br />
TE<br />
TM<br />
0.0<br />
0.5 1.0 1.5 2.0<br />
TR thickness t [μm]<br />
(a) (b)<br />
Figure 4.12 Coupling losses versus TR film thickness t for TE-<strong>polarisation</strong> (a),<br />
and the total predicted loss reduction (b) for TE (solid line) and TM <strong>polarisation</strong><br />
(dashed line). The experimentally measured reduction is depicted by a diamond<br />
(TE) and a triangle (TM).
4.4 Ghost images 73<br />
In figure 4.12b the total loss reduction is shown for both TE and TM <strong>polarisation</strong>. The<br />
optimum predicted loss reduction is 2.2 dB for TE <strong>polarisation</strong>, and 1.9 dB for TM <strong>polarisation</strong><br />
with t = 1.1 μm. Two PHASAR <strong>demultiplexers</strong> have been produced, one of which was used for<br />
the double etch experiment [33]. The length of the TR was chosen at 10 μm, which is - according<br />
to BPM simulations - sufficiently short to get rid of the radiation modes excited at the first<br />
junction. The waveguide has a propagation loss of 2 dB/cm, both for TE and TM <strong>polarisation</strong>.<br />
Due to the high lateral index contrast, small bending radii (200 to 500 μm) can be used, giving<br />
an array size of mm2 1.0 × 1.3 , measured between object and image plane. The overall device<br />
size is mm2 1.2 × 3.2 as the input and output waveguides are positioned with a 250 μm pitch<br />
for tapered fibre ribbon coupling. The measured improvement of the insertion loss is shown in<br />
the graph: 1.7 dB for TE <strong>polarisation</strong> and 1.3 dB for TM <strong>polarisation</strong>, which is only 0.6 dB less<br />
than predicted.<br />
4.4 Ghost images<br />
The measured response of the second PHASAR demultiplexer produced as depicted in figure<br />
4.13, shows considerably high peaks at unexpected positions. These peaks are denoted by<br />
circles in the figure and are further referred to a “ghost” images. As these “ghost” images<br />
contribute significantly to the cross talk of the device, it is of crucial importance to find and<br />
eliminate the cause. After rejecting several possibilities (such as multiple reflections within the<br />
device), we came to the conclusion that the ghost images are caused by mode conversion<br />
within the phased array. During the first experiments with the epitaxial growth of the wafer at<br />
POC, it was found that the band-edge <strong>wavelength</strong> of the quaternary material was 1.02 μm<br />
instead of the requested 0.97 μm. Together with the used (design) thickness of 2.2 μm, the<br />
width had to be fixed to 3.0 μm in order to make the waveguides <strong>polarisation</strong> <strong>independent</strong>. The<br />
resulting waveguide structure turned out to carry three modes (see figure 4.6b). In this section<br />
we will first discuss how mode conversion may cause ghost images. Then the experiments will<br />
be described which we carried out to verify this hypothesis.<br />
Transmission [dB]<br />
-30<br />
-40<br />
-50<br />
-60<br />
-70<br />
1520 1525 1530 1535<br />
Wavelength [nm]<br />
Figure 4.13 Ghost images (circles) in the response for TE (solid) and TM<br />
<strong>polarisation</strong> (dashed).
74 4. A <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>based</strong> on nonbirefringent waveguides<br />
4.4.1 Ghost image mechanism<br />
Due to the high lateral contrast, the first-order mode experiences low radiation loss in the<br />
curved waveguides. This first-order mode can be excited at discontinuities along the path of<br />
propagation through the phased array. As the group index (see equation 2.4) is different for the<br />
first-order mode, the dispersion will be different as well. The result is a “ghost” image, shifted<br />
in position along the image plane with respect to the original image. These ghost images,<br />
marked by a circle in figure 4.13, may couple into an undesired receiver waveguide, resulting<br />
in an increased cross talk level. This shift in position along the image plane corresponds to a<br />
shift in the <strong>wavelength</strong> response and is due to the difference in propagation constants between<br />
the fundamental and first order mode, analogous to what is found for <strong>polarisation</strong> dispersion.<br />
This <strong>wavelength</strong> shift will therefore be denoted as modal dispersion shift and can be calculated<br />
in exactly the same manner as for the <strong>polarisation</strong> dispersion shift according to equation 2.18.<br />
As the index of the first-order mode is smaller than the index of the fundamental mode, a<br />
positive modal dispersion shift is obtained, which means that the ghost image occurs at a<br />
shorter <strong>wavelength</strong> with respect to the original image. Furthermore, the index difference<br />
between the first-order mode and the fundamental mode is smaller for TM <strong>polarisation</strong> than for<br />
TE <strong>polarisation</strong>. The result is that the modal dispersion shift for TM <strong>polarisation</strong> is smaller<br />
than for TE <strong>polarisation</strong>.<br />
transmitter<br />
waveguide<br />
input<br />
aperture<br />
FPR<br />
4<br />
3<br />
object<br />
image<br />
5 plane<br />
plane 1<br />
For a more comprehensive analysis of the occurrence of these ghost images, a path through the<br />
phased array is considered as shown in figure 4.14. In this schematic diagram the positions at<br />
which excitation of the first-order mode (through mode conversion) may occur are marked<br />
with a circle. The positions will be discussed consecutively, the numbering of which follows<br />
those of figure 4.14.<br />
1. At this position the cross talk performance is not degraded. The first order mode may distort<br />
further signal operations.<br />
2. In this case the signal traverses the array almost completely as a fundamental mode, except<br />
for the small straight section, where a fraction is converted to a first-order mode. Therefore,<br />
only over a small part the phase transfer is disturbed and the image will remain almost in its<br />
original place along the image plane.<br />
3. In the same way as above, the phased array has now only traversed over a small straight<br />
section length as a fundamental mode and almost completely as a first-order mode. The<br />
2<br />
output<br />
aperture<br />
FPR<br />
receiver<br />
waveguide<br />
Figure 4.14 Schematic diagram of a path through the phased array, with the<br />
positions (marked by a circle) where first order mode excitation is expected.
4.4 Ghost images 75<br />
phase transfer, therefore, is determined by the first-order mode and the image will be shifted<br />
in position along the image plane with respect to the original one.<br />
4. At this position it is very unlikely that mode conversion will occur, as the far-field of the<br />
transmitter waveguide can be treated locally as a uniform field. However, an estimation of<br />
the excitation coefficient of the first-order mode can be made using a Gaussian<br />
approximation. This is discussed in Appendix E. The result is that the first-order excitation<br />
at this position is well below -50 dB and may therefore be neglected.<br />
5. Excitation of the first-order mode at this point (or at the fibre-chip junction, the effect being<br />
the same) will result in a mixed field (consisting of the fundamental and first-order mode) at<br />
the object plane. As the PHASAR is an imaging device, this field will be reproduced at the<br />
image plane at the position of the original image without affecting the dispersion properties<br />
of the array.<br />
4.4.2 Verification<br />
The assumption that the occurrence of ghost images is caused by excitation of first-order<br />
modes in the phased array has been verified by two experiments. Firstly, as the response of the<br />
demultiplexer is periodical, so should the occurrence of the ghost images. This can be seen<br />
clearly in figure 4.15. In this graph the response of a PHASAR demultiplexer (the same device<br />
of which the response is shown in figure 4.13) is shown over the full <strong>wavelength</strong> span available<br />
by the tunable laser source. The modal dispersion shift is denoted as Δλ 01,TE and Δλ 01,TM for<br />
TE and TM <strong>polarisation</strong> respectively. (The non-constant cross talk level is due to the increased<br />
spontaneous emission of the tunable laser, occurring when operating near the limits of its tuning<br />
range.) The predicted modal dispersion shift is 13.5 nm for TE <strong>polarisation</strong> and 12.2 nm<br />
for TM <strong>polarisation</strong>, which compares well to the measured modal dispersion shift of<br />
Δλ 01,TE = 16.0 and Δλ 01,TM = 10.8 nm.<br />
Transmission [dB]<br />
-25<br />
-30<br />
-35<br />
-40<br />
-45<br />
-50<br />
-55<br />
-60<br />
-65<br />
Δλ 01,TM<br />
Δλ 01,TE<br />
1460 1480 1500 1520 1540 1560 1580<br />
Wavelength [nm]<br />
Figure 4.15 Response over the full <strong>wavelength</strong> span of the tunable laser for TE<br />
(solid line) and TM (dashed line) <strong>polarisation</strong>.<br />
The second verification is described as follows. The field obtained at the image plane is the<br />
sum field of the far fields of all individual array waveguides (see section 2.1.3). Therefore, if<br />
the <strong>wavelength</strong> is changed, the image will move through the image plane and follow the
76 4. A <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>based</strong> on nonbirefringent waveguides<br />
envelope described by the far field of the individual array waveguides. In the case of a ghost<br />
image, this envelope has the shape of the far field of the first-order mode. Figure 4.16a<br />
schematically shows the far field of the first-order mode, which has a similar shape as the<br />
modal field: two maxima with a minimum in the centre.<br />
array<br />
waveguides<br />
minimum<br />
maximum<br />
receiver<br />
output<br />
waveguides<br />
aperture free<br />
propagation<br />
region<br />
image<br />
plane<br />
field<br />
intensity<br />
The assumption that ghost images originate from the presence of a first-order mode can<br />
therefore be verified by measuring the peak power in the ghost image in relation with the<br />
power in the original peak for all receiver waveguides. For the receiver waveguides placed in<br />
the centre of the image plane, the ghost peak level should have a minimum, whereas the level<br />
should increase quadratically towards the outer receiver waveguides. Figure 4.16b shows the<br />
measured ghost peak level for TE <strong>polarisation</strong> versus receiver waveguide number. The<br />
quadratic behaviour of the ghost peak level can be clearly observed.<br />
4.4.3 Junction optimisation<br />
(a) (b)<br />
1 2 3 4 5 6 7 8<br />
Receiver waveguide number<br />
Figure 4.16 The far-field intensity distribution of a first-order mode (a), and the<br />
ghost peak level with respect to the passband peak level for the eight output<br />
waveguides (b).<br />
From the analysis presented in the previous section it is apparent, that mode conversion must<br />
be minimised. This can be done by optimising the offset at the junctions not on maximal<br />
coupling for the fundamental mode, but on minimal first-order mode excitation according to<br />
the strategy presented in section 2.2.9 (equation 2.35). The bending radius nor the length of the<br />
straight section have changed, and therefore the phase transfer of the array is not influenced.<br />
As an example the strategy is applied to the raised-strip waveguide, at a thickness of 2.2 μm<br />
and a width of 3.0 μm especially designed for <strong>polarisation</strong> independence of a Q(1.02) guiding<br />
layer. The bending radius was chosen at 500 μm. In figure 4.17 the efficiency of the coupling<br />
between the fundamental mode of the straight waveguide to the fundamental and first-order<br />
mode in the curved waveguide is shown.<br />
From this graph it can be seen that a first-order coupling efficiency to a TM-polarised mode of<br />
-25 dB is obtained if the offset is optimised to optimum coupling for the fundamental TEpolarised<br />
mode. An optimum offset cannot be found for both <strong>polarisation</strong>s simultaneously, so<br />
that a compromise has to be made. This compromise can be found between the optimum<br />
offsets for each <strong>polarisation</strong> (120-150 nm), i.e. 135 nm. In any case the maximum excitation of<br />
the first-order mode is below -30 dB and the coupling loss for the fundamental order remains<br />
Ghost peak level [dB]<br />
-14<br />
-16<br />
-18<br />
-20<br />
-22<br />
-24<br />
-26
4.4 Ghost images 77<br />
negligibly low. Figure 4.18 shows the sensitivity of the coupling efficiency to the waveguide<br />
width. If width variations of ± 0.2 μm are taken into account, which is a realistic value for<br />
high-quality lithographic techniques, it shows that the first-order coupling efficiency increases<br />
rapidly, but it remains below an acceptable level of -25 dB.<br />
Offset [nm]<br />
250<br />
200<br />
150<br />
100<br />
Zero-order coupling efficiency [dB]<br />
0.00<br />
-0.05<br />
-0.10<br />
-0.15<br />
4.4.4 Experimental results<br />
-0.20<br />
-50<br />
0 50 100<br />
Offset [nm]<br />
150 200<br />
Figure 4.17 Zero and first-order coupling efficiencies for TE (solid lines) and<br />
TM <strong>polarisation</strong> (dashed lines).<br />
-20<br />
-30<br />
-40<br />
-50<br />
-50<br />
-30<br />
50<br />
-0.2 -0.1 0.0<br />
ΔW [μm]<br />
0.1 0.2<br />
-30<br />
-40<br />
-50<br />
-40<br />
-30<br />
-40-50<br />
(a) (b)<br />
A 8-channel raised-strip-waveguide-<strong>based</strong> PHASAR-demultiplexer with optimised junctions<br />
has been designed and produced in cooperation with POC [37]. The device is identical to the<br />
one used in a previous experiment (see section 4.3). The response was measured using a<br />
tunable laser for both TE (solid lines) and TM (dashed lines) <strong>polarisation</strong>, and is depicted in<br />
figure 4.19. The cross talk is better than -23 dB and the insertion loss is 8-10 dB (including<br />
fibre-chip coupling loss of ±<br />
1 dB), 1.5 dB of which due to waveguide losses. As can be seen in<br />
the graph, no ghost images are observed.<br />
Offset [nm]<br />
250<br />
200<br />
150<br />
100<br />
-20<br />
-30<br />
-40<br />
-50<br />
-50<br />
-30<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
First-order coupling efficiency [dB]<br />
50<br />
-0.2 -0.1 0.0<br />
ΔW [μm]<br />
0.1 0.2<br />
-30<br />
-40<br />
-50<br />
-40<br />
-30<br />
-40-50<br />
Figure 4.18 Sensitivity of the first-order coupling efficiency to the waveguide<br />
width for TE (a) and TM <strong>polarisation</strong> (b).<br />
-20
78 4. A <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>based</strong> on nonbirefringent waveguides<br />
Insertion loss [dB]<br />
4.4.5 Conclusions<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
-50<br />
1510 1515 1520<br />
Wavelength [nm]<br />
1525 1530<br />
Figure 4.19 Measured response of a phased-array demultiplexer with optimised<br />
junctions for TE (solid) and TM (dashed) <strong>polarisation</strong> (channel 8 defect). The<br />
graph shows an increasing TE-TM shift (from zero to 0.4 nm) to lower<br />
<strong>wavelength</strong>s. This is due to an increasing temperature during measurements, as<br />
first the TE response was measured from left to right and consecutively the TM<br />
response from right to left.<br />
The occurence of ghost images in the response of phased-array <strong>demultiplexers</strong> has been<br />
analysed. Measurement results confirm that they originate from first-order modes excited at<br />
junctions between straight and curved waveguides in the array. It has been experimentally<br />
demonstrated that ghost images can be reduced by optimising the waveguide junctions.<br />
4.5 Demultiplexer integrated with detectors<br />
This section describes the integration of a raised-strip-guide-<strong>based</strong> PHASAR demultiplexer<br />
with detectors. A detector <strong>based</strong> on evanescent field coupling is used, <strong>based</strong> on our experience<br />
with this type of detector [7]. It is relative easy to produce as no expitaxial regrowth is required<br />
and only an additional wet-chemical etching step is required to define the detector mesas. A<br />
schematic diagram of the detector is shown in figure 4.20. In this detector, the InGaAs<br />
absorption layer is sandwiched between n- and p-type layers, thereby forming a p-i-n<br />
configuration. The light from the waveguide is coupled into the absorption layer, where<br />
electron-hole pairs are created. The coupling efficiency between the guiding and the absorbing<br />
layer is improved by inserting an InGaAsP(1.3) layer in between them, which acts as an antireflection<br />
layer. For the optimisation of the layer thickness, modal propagation analysis has<br />
been used. Figure 4.21 shows the field propagation in the xz-plane for TE <strong>polarisation</strong>. From<br />
this figure it can be seen that at a detector length of 200 μm, more than 90% of the optical<br />
power is absorbed.
4.5 Demultiplexer integrated with detectors 79<br />
X<br />
Z<br />
p-InGaAsP(1.3)<br />
p-<strong>InP</strong><br />
InGaAs(1.67)<br />
InGaAsP(1.3)<br />
InGaAsP(1.0)<br />
n-<strong>InP</strong> substrate<br />
0.4 μm<br />
0.1 μm<br />
0.3 μm<br />
A 8-channel raised-strip-waveguide-<strong>based</strong> PHASAR with integrated detectors has been<br />
designed and produced in cooperation with POC [137]. The width and thickness were chosen<br />
at 3.0 μm and 2.2 μm respectively in order to obtain <strong>polarisation</strong> indepent waveguides using a<br />
Q(1.02) guiding layer. The channel spacing was designed at 200 GHz at a centre <strong>wavelength</strong> of<br />
1535 nm, determined by the specifications of the ACTS AC028 TOBASCO project. The array<br />
size is mm 2 1.4 ×<br />
1.3 , measured between object and image plane. The detectors are 8 μm wide<br />
and 200 μm long and use a common n-type bottom contact. On top of the detectors a 6 μm<br />
wide p-type contact is placed prior to which a Si 3 N 4 isolation layer is deposited. A microscope<br />
photograph of the device produced is shown in figure 4.22.<br />
Au<br />
Pt<br />
0.5 μm<br />
0.4 μm<br />
0.3 μm<br />
2.2 μm<br />
Figure 4.20 Schematic representation of the layer structure which has been used<br />
for the integrated waveguide detector.<br />
x-direction [μm]<br />
4<br />
2<br />
0<br />
0.60<br />
0.50<br />
0.40<br />
0.30<br />
0.20<br />
0.10<br />
0.50<br />
0.50<br />
0.300.40<br />
0.10<br />
0.10<br />
0.20 0.20<br />
0.60 0.70<br />
0.40<br />
0.40<br />
0.30<br />
0.20<br />
0.30<br />
0.10<br />
0.20<br />
0.20<br />
0.10<br />
0.10 0.10<br />
0.10<br />
-2<br />
0 50 100<br />
z-direction [μm]<br />
150 200<br />
Air<br />
Au<br />
<strong>InP</strong><br />
Pt<br />
InGaAsP(1.3)<br />
<strong>InP</strong><br />
InGaAs<br />
InGaAsP(1.3)<br />
InGaAsP(1.0)<br />
Figure 4.21 Field distribution in the detector structure for TE <strong>polarisation</strong>.
80 4. A <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>based</strong> on nonbirefringent waveguides<br />
Figure 4.22 Microscope photograph of PHASAR demultiplexer integrated with<br />
detectors.<br />
The response at 1 mW unpolarised input power was measured and is shown in figure 4.23a.<br />
Problems with the Si 3 N 4 isolation layer are the cause of the low external efficiency of the<br />
detectors (-13 dB, leading to an overall insertion loss of -21 dB including the 8 dB insertion<br />
loss of the PHASAR) and the large background of one of the channels.<br />
Photo current [μA]<br />
10.00<br />
1.00<br />
0.10<br />
0.01<br />
1520 1525 1530<br />
Wavelength [nm]<br />
1535 1540<br />
(a) (b)<br />
1525 1530 1535 1540<br />
Wavelength [nm]<br />
Figure 4.23 Measured photo current versus <strong>wavelength</strong> before (a) and after (b)<br />
packaging.<br />
Subsequently, the wafer was cleaved into dies of mm 2 , an AR coating was applied to<br />
the front facet and the device was mounted on a Si submount and a mm 2 3, 5 × 3<br />
6 ×<br />
6 carrier. The<br />
fibre was aligned and fixed by soldering it on a base plate mounted stud in front of the carrier.<br />
Finally, the assembly has been mounted on a Peltier cooler inside a 14-pin industry-standard<br />
butterfly package. The device is shown in figure 4.24. It is the first compact packaged <strong>InP</strong><br />
PHASAR-<strong>based</strong> demultiplexer with integrated detectors. The response measured after<br />
Photo current [μA]<br />
10.00<br />
1.00<br />
0.10<br />
0.01
4.6 Discussion 81<br />
packaging is depicted in figure 4.23b and is only 1 dB below that of the unpackaged device,<br />
which is due to a slighlty larger coupling loss. Using the Peltier cooler, the temperature has<br />
been set at 55° C in order to shift the <strong>wavelength</strong> response to the required TOBASCO<br />
specifications.<br />
Figure 4.24 First 8-channel butterfly-type PHASAR module.<br />
4.6 Discussion<br />
In this chapter the design and experiments of PHASAR <strong>demultiplexers</strong> <strong>based</strong> on raised-strip<br />
waveguides is described. The device has been made <strong>polarisation</strong> <strong>independent</strong> by properly<br />
shaping the array waveguides in such a way as to eliminate their birefringence. It is<br />
demonstrated that with this waveguide structure <strong>polarisation</strong> independence over a wide<br />
<strong>wavelength</strong> range can be obtained. Further, a method for further reducing the PHASAR loss is<br />
presented. With this method, for which only an additional etching step is required, a loss<br />
reduction in the order of 1.5 dB is possible. The influence of mode conversion on the cross talk<br />
of the device is analysed. It is demonstrated that mode conversion is the cause of the occurence<br />
of so-called “ghost” images in the response, which leads to increased cross talk. However, the<br />
mode conversion can easily be minimised by properly designing the junctions between the<br />
straight and curved array waveguides. Finally, the integration of a PHASAR demultiplexer<br />
with detectors in a raised-strip waveguide structure has been demonstrated. In addition, the<br />
device has been packaged, resulting in the first compact packaged <strong>InP</strong> PHASAR-<strong>based</strong><br />
demultiplexer with integrated detectors in a 14-pin industry-standard butterfly package.
82 4. A <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>based</strong> on nonbirefringent waveguides
Chapter 5<br />
Polarisation <strong>independent</strong> PHASAR<br />
<strong>demultiplexers</strong> <strong>based</strong> on<br />
birefringent waveguides<br />
A PHASAR demultiplexer produced in a conventional double-hetero structure, which has been<br />
made <strong>polarisation</strong> <strong>independent</strong> by inserting a <strong>polarisation</strong> dispersion compensating section, is<br />
described in this Chapter. In addition, two other methods for making PHASAR <strong>demultiplexers</strong><br />
<strong>polarisation</strong> <strong>independent</strong> are introduced and analysed.<br />
5.1 Introduction<br />
A number of methods can be applied in order to eliminate <strong>polarisation</strong> dispersion and will be<br />
discussed in this Chapter. The first attempt to do this was <strong>based</strong> on matching the Free Spectral<br />
Range (FSR) to the <strong>polarisation</strong> dispersion (see figure 5.1). It was suggested by Vellekoop and<br />
Smit [127,133], and also applied by Spiekman et al. [162,163] and Zirngibl et al. [177]. This<br />
method, also known as the so-called “order trick”, is simple from a design point of view. If the<br />
FSR is chosen equal to the <strong>polarisation</strong> dispersion as shown in figure 5.1, the m-th order beam<br />
for TE <strong>polarisation</strong> overlaps with the TM-polarised beam of order m-1, and the device will<br />
therefore be virtually <strong>polarisation</strong> <strong>independent</strong>. The disadvantage of this method lies in the fact<br />
that the total <strong>wavelength</strong> range available for the WDM channels is restricted by the <strong>polarisation</strong><br />
dispersion, which is for conventional InGaAsP/<strong>InP</strong> DH waveguide structures in the order of 4-<br />
5 nm. Additionally, Spiekman [134] showed that this method, although very tolerant to etch<br />
depth variations, is very sensitive to waveguide width and layer thickness variations: a layer<br />
thickness variation of 3% or a waveguide width variation of ±<br />
0.2 μm will cause a shift of<br />
0.2 μm between the TE and TM responses. These tolerances, however, can be relaxed if the<br />
<strong>wavelength</strong> response is flattened.<br />
In section 5.2 and 5.4 two alternative approaches for obtaining <strong>polarisation</strong> independence<br />
<strong>based</strong> on <strong>polarisation</strong> conversion and <strong>polarisation</strong> splitting respectively are briefly discussed.
84 5. Polarisation <strong>independent</strong> PHASARs <strong>based</strong> on birefringent waveguides<br />
TMm+1 TMm+1 TMm TEm TMm-1 TEm-1 <strong>polarisation</strong> dispersion<br />
FSR<br />
<strong>wavelength</strong><br />
Figure 5.1 Schematic diagram of the periodic <strong>wavelength</strong> response of a phasedarray<br />
demultiplexer with the FSR and <strong>polarisation</strong> dispersion depicted.<br />
5.2 Polarisation dispersion compensation<br />
In the previous chapter it has been demonstrated that <strong>polarisation</strong> <strong>independent</strong> behaviour can<br />
be achieved over a large <strong>wavelength</strong> range for compact PHASAR <strong>demultiplexers</strong> emplyoing a<br />
raised-strip waveguide structure (see section 4.2.3). Another broad-band solution to obtain<br />
<strong>polarisation</strong> independence is found in compensation of the <strong>polarisation</strong> dispersion. This can be<br />
done by inserting a waveguide section with a different birefringence in the phased array as<br />
shown in figure 5.2. If the waveguides in the shaded area have a <strong>polarisation</strong> dispersion<br />
different from the dispersion in the unshaded region, the total <strong>polarisation</strong> dispersion of the<br />
device can be cancelled, provided that the shape of the shaded area has been properly chosen.<br />
Figure 5.2 Schematic diagram of a PHASAR demultiplexer with <strong>polarisation</strong><br />
dispersion compensation.
5.2 Polarisation dispersion compensation 85<br />
The operation can be explained by considering the phase transfer difference ΔΦ between two<br />
adjacent waveguides, where at one of which a section with length δL and a different<br />
birefringence has been inserted (see figure 5.3):<br />
whereby N eff and n eff are the effective mode indices of the original waveguide and the<br />
compensation section respectively. It can be easily verified that ΔΦ becomes <strong>polarisation</strong><br />
<strong>independent</strong> if δL is chosen according to:<br />
with:<br />
i+1<br />
i<br />
δL<br />
The whole array can be made <strong>polarisation</strong> <strong>independent</strong> by inserting a section with length δL in<br />
the first waveguide, 2δL in the second one, 3δL in the third one, and so on. The compensation<br />
section will thus obtain a triangular shape and its total length will amount to NaδL whereby Na is the number of array waveguides. The method applies for both positive and negative values of<br />
Δn/ΔN. For values close to 1, the compensation section will become excessively long. A<br />
disadvantage is that ΔN and Δn are very sensitive to film thickness and waveguide width, so the<br />
compensation requires tight control of these parameters. This will be discussed in the next section.<br />
Absolute compensation of the <strong>polarisation</strong> dispersion (as demonstrated by Steenbergen et al.<br />
[141]) can be achieved if the following design and production stategy is used. First, a<br />
PHASAR demultiplexer is produced without the dispersion correction and the TE-TM shift is<br />
measured. From the measured TE-TM shift, the compensation section length is calculated and<br />
the correction is applied. In this way, effects of layer thickness variations from one wafer to<br />
another can be eliminated.<br />
Additionally, the insertion of a waveguide section gives rise to a shift of the centre <strong>wavelength</strong>.<br />
This can be seen by taking ΔΦ = 2πm in equation 5.1. The resulting centre <strong>wavelength</strong> λc' is<br />
then found as:<br />
whereby the first term denotes the original centre <strong>wavelength</strong>, and the second term the shift. It<br />
is noted that this shift depends on the <strong>polarisation</strong>, just as the original centre <strong>wavelength</strong>.<br />
L i<br />
L i +ΔL<br />
Figure 5.3 Schematic diagram of the <strong>polarisation</strong> dispersion compensation<br />
principle.<br />
ΔΦ = ko[ N effΔL + δL( N eff – neff) ]<br />
Δn<br />
δL = ΔL ⁄ ⎛------- – 1⎞<br />
⎝ΔN ⎠<br />
ΔN = N eff,TE – N eff,TM<br />
Δn = neff,TE – neff,TM ΔL<br />
λc' N eff ⋅ ------ ( N<br />
m eff – neff) δL<br />
= +<br />
⋅ ----m<br />
(5.1)<br />
(5.2)<br />
(5.3)<br />
(5.4)
86 5. Polarisation <strong>independent</strong> PHASARs <strong>based</strong> on birefringent waveguides<br />
5.2.1 Production tolerance analysis<br />
For the tolerance analysis of the <strong>polarisation</strong> dispersion compensation method, we use the<br />
ridge waveguide structure as depicted in figure 5.4a. It consists of a 600 nm thick InGaAsP<br />
guiding layer on an <strong>InP</strong> substrate with a 300 nm thick <strong>InP</strong> cladding layer. The ridge width is<br />
2 μm and the etch depth is chosen at 400 nm, i.e. 100 nm into the guiding layer. This allows for<br />
the usage of sufficiently short bending radii (500 μm) and, therefore, compact device<br />
dimensions.<br />
100 nm<br />
2 μm<br />
<strong>InP</strong><br />
InGaAsP(λ g = 1.3 μm)<br />
<strong>InP</strong> substrate<br />
300 nm<br />
600 nm<br />
2 μm<br />
InGaAsP(λ g = 1.3 μm)<br />
<strong>InP</strong> substrate<br />
(a) (b)<br />
Figure 5.4 Waveguide structures as used for the <strong>polarisation</strong> dispersion<br />
compensation: the original waveguide structure (a) and the waveguide structure in<br />
the compensation section.<br />
As the <strong>polarisation</strong> dispersion can be enlarged by removing the <strong>InP</strong> cladding layer, we used the<br />
structure as depicted in figure 5.4b for the dispersion compensation section. The advantage of<br />
this method is that a tolerant selective wet-chemical etchant can be applied to remove the<br />
cladding layer. The disadvantage, however, is a reduction of the lateral index contrast, by<br />
which coupling between array waveguides may become a problem. Furthermore, due to the<br />
reduced index contrast, this method cannot be applied to curved waveguides as the radiation<br />
losses will then become excessively high. The result is that an additional straight section length<br />
is needed in the centre of the phased array, which leads to a longer device.<br />
As an example, we use a PHASAR of which the design parameters are given in table 5.1. As<br />
can be seen in this table, the channel spacing was chosen at 3.2 nm (400 GHz). For lower<br />
channel spacings, the device will become less tolerant wih respect to production deviations.<br />
Figure 5.5a shows a contour plot of the TE-TM shift after <strong>polarisation</strong> dispersion<br />
compensation for different thicknesses of the Q(1.3) guiding layer and the <strong>InP</strong> cladding layer.<br />
The values are calculated using the design parameters listed in table 5.1. From this figure it<br />
becomes clear that the demands on the layer thickness are very stringent. A guiding layer<br />
thickness variation of ± 15 nm (2.5%) leads to a TE-TM shift of approximately 0.2 nm. A<br />
cladding layer thickness variation of ± 15 nm (5%) even leads to a TE-TM shift of 0.5 nm.<br />
Therefore, it is important that the layer thicknesses are accurately known before the design is<br />
made.<br />
In figure 5.5b the TE-TM shift dependence with respect to etch depth and waveguide width are<br />
shown. The standard lithographic process has to be pushed to its limits to obtain a waveguide<br />
width deviation of ± 0.1 μm. If the etch depth variation can be controlled within ±<br />
10 nm<br />
(10%), the TE-TM shift can be limited to 0.2 nm.<br />
It can, therefore, be concluded that due to the rather tight - but not unrealistic - demands on the
5.2 Polarisation dispersion compensation 87<br />
processing, this method appeals to the controllability of the production process and to the skills<br />
of the persons who actually manufacture the devices.<br />
<strong>InP</strong> layer thickness [nm]<br />
400<br />
350<br />
300<br />
250<br />
Table 5.1 Design parameters of the demultiplexer.<br />
-4.0<br />
-3.6<br />
-3.2<br />
-2.8<br />
-2.4<br />
-2.0<br />
-1.6<br />
-0.4<br />
0.0<br />
1.2<br />
2.0<br />
2.4<br />
2.8<br />
3.6<br />
4.0<br />
4.4<br />
0.8<br />
1.6<br />
3.2<br />
4.8<br />
-1.2<br />
-0.8<br />
0.4<br />
Parameter Value<br />
centre <strong>wavelength</strong> λ c<br />
1534 nm<br />
number of channels N 8<br />
channel spacing Δλ ch (Δf ch ) 3.2 nm (400 GHz)<br />
array order m 37<br />
free spectral range Δλ FSR<br />
number of array waveguides N a<br />
41.5 nm<br />
62<br />
array length 1.6 mm<br />
array height 1.1 mm<br />
TE-TM shift Δλ pol<br />
4.0 nm<br />
correction length δL 12.48 μm<br />
centre <strong>wavelength</strong> shift Δλ c<br />
index contrast ΔN<br />
index contrast Δn<br />
200<br />
500 550 600 650 700<br />
Q(1.3) layer thickness [nm]<br />
-2.0<br />
-1.6<br />
-0.4<br />
0.0<br />
1.2<br />
2.0<br />
2.4<br />
2.8<br />
3.6<br />
4.0<br />
4.4<br />
-1.2<br />
-0.8<br />
0.4<br />
0.8<br />
1.6<br />
3.2<br />
4.8<br />
Etch depth into Q(1.3) layer [nm]<br />
200<br />
180<br />
160<br />
140<br />
120<br />
100<br />
7.4 nm<br />
– 9.8 10 3 –<br />
⋅<br />
– 2.3 10 2 –<br />
⋅<br />
(a) (b)<br />
80<br />
60<br />
40<br />
20<br />
1.2<br />
1.6<br />
2.0<br />
2.4<br />
3.0<br />
3.2<br />
2.8<br />
2.2<br />
1.8<br />
1.4<br />
1.6 1.8 2.0 2.2 2.4<br />
Waveguide width [μm]<br />
Figure 5.5 TE-TM shift (nm) as a function of the guiding layer and the cladding<br />
layer thickness (a) and as a function of waveguide width and etch depth into the<br />
guiding layer (b).<br />
0.4<br />
0.8<br />
2.6<br />
1.0<br />
-0.2<br />
0.6<br />
-0.4<br />
0.2<br />
-0.6<br />
0.0<br />
0.4<br />
0.8<br />
-0.8<br />
1.0<br />
-0.2<br />
-0.4<br />
0.6<br />
-1.0<br />
-0.6<br />
0.0<br />
0.2
88 5. Polarisation <strong>independent</strong> PHASARs <strong>based</strong> on birefringent waveguides<br />
5.2.2 Coupling between array waveguides<br />
By removing the <strong>InP</strong> cladding of the array waveguides, the lateral index contrast Δn decreases<br />
by approximately 50% from 0.051 (0.071) to 0.027 (0.033) for TE (TM) <strong>polarisation</strong>, which<br />
results in a decrease of the confinement of the waveguide field. The field will therefore extend<br />
further into the film next to the waveguide and coupling between waveguides will increase.<br />
This is shown in figure 5.6 for TE <strong>polarisation</strong>.<br />
Transverse position [μm]<br />
1.0<br />
0.5<br />
0.0<br />
-0.5<br />
-1.0<br />
-2 0 2 4<br />
Lateral position [μm]<br />
-1.0<br />
-2 0 2 4<br />
Lateral position [μm]<br />
(a) (b)<br />
Figure 5.6 Field profile in the original waveguide structure (a) and in the<br />
waveguide structure where the <strong>InP</strong> cladding has been removed (b).<br />
Assuming that each waveguide only couples to its neighbouring waveguides, we use the<br />
system-mode concept for an estimate of the coupling. We consider the fundamental (even) and<br />
first-order (odd) mode of two coupled waveguides (the “system”). Light propagating in a<br />
single waveguide can be decomposed into the even and odd system mode, as depicted<br />
schematically in figure 5.7.<br />
single<br />
mode<br />
=<br />
even<br />
mode<br />
+<br />
After propagation over a certain length, both system mode have opposite phases (i.e. phase<br />
difference equals π radians) and light will be transferred completely to the other waveguide.<br />
The length L at which this occurs, is defined as:<br />
whereby β 0 and β 1 are the propagation constants of the fundamental and first-order system<br />
Transverse position [μm]<br />
1.0<br />
0.5<br />
0.0<br />
-0.5<br />
odd<br />
mode<br />
waveguide a<br />
waveguide b<br />
Figure 5.7 Decomposition of a single mode into the even and odd system mode.<br />
L( β0 – β1) = π<br />
(5.5)
5.2 Polarisation dispersion compensation 89<br />
modes of the coupled waveguides. The value of L for which this occurs equals the coupling<br />
length Lc . The coupling length is defined as Lc = π ⁄ 2C (using coupled-mode theory [160])<br />
with C being the coupling coefficient. The coupling coefficient C is thus found as:<br />
The fraction of power coupled to the other waveguide after propagation over a length L can be<br />
estimated by:<br />
We consider the worst-case coupling by using the longest side of the compensation triangle<br />
( L = N a ⋅ δL ) as the length over which coupling occurs. Figure 5.8 shows the power coupled<br />
from one waveguide to its neighbour. It shows that if the gap is chosen sufficiently large<br />
(> 5 μm), coupling effects are negligible. For the PHASAR, the parameters of which are listed<br />
in table 5.1, the gap equals 5 μm at the shortest side of the triangle, ( L<br />
than 10 μm at the longest side.<br />
= δL ) and is even more<br />
Coupled power [%]<br />
100.00<br />
10.00<br />
1.00<br />
0.10<br />
5.2.3 Experimental results<br />
C = ( β0 – β1) ⁄ 2<br />
The approach described above has been applied to a phased-array demultiplexer with a<br />
waveguide structure as designed and depicted in figure 5.4. The design parameters are listed in<br />
table 5.1 and the device has been produced in cooperation with Kees Steenbergen [141]. The<br />
etching of the waveguides was done by employing a RIE etch/descum process for low-loss<br />
waveguides [86], resulting in an optical loss of 1.4 dB/cm for both TE and TM <strong>polarisation</strong>.<br />
The removal of the <strong>InP</strong> cladding was performed by selective chemical etching with HCl/<br />
H3PO4 . The calculated index differences ΔN and Δn of the waveguide structure used are<br />
– 9.8 10 and respectively. Together with the measured TE-TM shift of 4.2 nm<br />
without correction, an incremental length δL of 12.48 μm was calculated. The number of array<br />
3 –<br />
⋅ – 2.3 10 2 –<br />
⋅<br />
(5.6)<br />
2 CL<br />
P sin ⎛------ ⎞ sin<br />
⎝ 2 ⎠<br />
2 β0 β ( – 1)L<br />
= = ⎛------------------------- ⎞<br />
(5.7)<br />
⎝ 2 ⎠<br />
0.01<br />
1 2 3 4 5 6<br />
Gap [μm]<br />
Figure 5.8 Coupled power over the longest side of the compensation triangle<br />
versus gap size.
90 5. Polarisation <strong>independent</strong> PHASARs <strong>based</strong> on birefringent waveguides<br />
2.4 mm<br />
0.8 mm<br />
Figure 5.9 Microscope photograph of the realised demultiplexer.<br />
1.1 mm<br />
guides Na equals 62, leading to an increase of the device length of slightly less than 800 μm,<br />
giving an overall array size of mm2 2.4 × 1.1 . A microscope photograph of the device is<br />
shown in figure 5.9, in which the triangularly shaped <strong>polarisation</strong> dispersion compensation<br />
section can be clearly distinguished.<br />
The device, which has been integrated with photodetectors [141], has 8 <strong>wavelength</strong> channels<br />
spaced at 400 GHz frequency (3.2 nm <strong>wavelength</strong>) with a centre <strong>wavelength</strong> of 1534 nm.<br />
Figure 5.10 shows the measured response of the demultiplexer for both TE and TM<br />
<strong>polarisation</strong>. It is seen that the response curves for both <strong>polarisation</strong>s match very well. As can<br />
be seen from the figure, the on-chip losses for TE <strong>polarisation</strong> are 3 dB and 5 dB for TM<br />
<strong>polarisation</strong>. The cross talk is better than -25 dB for the lower <strong>wavelength</strong> range, increasing to<br />
better than -20 dB to the higher <strong>wavelength</strong> range, which is due to the increased spontaneous<br />
emission of the laser occurring when operating near its <strong>wavelength</strong> limit.<br />
For TM <strong>polarisation</strong>, an additional loss was measured of about 2 dB, as can be seen in figure<br />
5.10. This is the result of the excess propagation loss of the waveguide with the etched <strong>InP</strong><br />
cladding layer. Test structures were included and consisted of straight waveguides of which the<br />
cladding layer was etched away over different section lengths. The excess propagation loss was<br />
measured for TE and TM <strong>polarisation</strong> and is shown in the graph of figure 5.11. For TE<br />
<strong>polarisation</strong>, the field does not suffer from the absence of the cladding layer, as the excess loss<br />
is <strong>independent</strong> of the section length. Only a coupling loss of 0.5 dB over two junctions was<br />
measured. TM polarised fields, on the other hand, do suffer from the absence of the cladding<br />
layer and an excess loss of 2.1 dB/mm was observed. This loss can be minimised by only<br />
partly etching the <strong>InP</strong> cladding layer, which leads to a longer section length δL and,<br />
consequently, to a longer device. In turn this may be compensated by a decreased waveguide<br />
width in the correction section [110].
5.2 Polarisation dispersion compensation 91<br />
On-chip loss [dB]<br />
5.2.4 Conclusions<br />
0<br />
-10<br />
-20<br />
-30<br />
-40<br />
1525 1530 1535 1540 1545 1550 1555 1560<br />
Wavelength [nm]<br />
Figure 5.10 Response of the phased-array demultiplexer: solid and dashed lines<br />
denote TE and TM <strong>polarisation</strong> respectively.<br />
Excess loss [dB]<br />
10<br />
8<br />
6<br />
4<br />
2<br />
L<br />
0<br />
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5<br />
Length [mm]<br />
Figure 5.11 Loss versus section length for TE (squares) and TM (triangles)<br />
<strong>polarisation</strong>.<br />
By locally removing the <strong>InP</strong> cladding layer of a conventional semiconductor waveguide<br />
structure as shown in figure 5.4, a tolerant selective wet-chemical etchant can be used to define<br />
the <strong>polarisation</strong> dispersion compensation section. Even though the production is relatively<br />
simple, this method is very sensitive to layer thickness and width variations. Therefore, a twostep<br />
production strategy is needed, in which the compensation section is <strong>based</strong> on an<br />
intermediate characterisation of the component. Although full <strong>polarisation</strong> dispersion<br />
compensation is obtained using this strategy, it makes this method less attractive for large-scale<br />
application.
92 5. Polarisation <strong>independent</strong> PHASARs <strong>based</strong> on birefringent waveguides<br />
5.3 Polarisation conversion<br />
As mentioned earlier in paragraph 2.2.4, a good method to obtain <strong>polarisation</strong> independence<br />
for silica-<strong>based</strong> and LiNbO 3 -<strong>based</strong> phased-array <strong>demultiplexers</strong> is the insertion of a λ/2-plate<br />
in the middle of the phased array. Using this method, light will always traverse half of the<br />
phased array in one state of <strong>polarisation</strong>, and the other half in the orthogonal state, regardless<br />
of the birefringence properties of the waveguides. However, as stated in section 2.2.4, it is not<br />
practical to apply this method to semiconductor waveguide structures due to their large<br />
numerical aperture, which leads to high diffraction losses when traversing the half-wave plate.<br />
Figure 5.12 Schematic diagram of a PHASAR demultiplexer with integrated<br />
<strong>polarisation</strong> converters inserted in the centre.<br />
An integrated <strong>polarisation</strong> converter, such as the one presented in section 3.7, might be an<br />
alternative for semiconductor waveguide structures. This is shown in figure 5.12. In this case<br />
the additional length of the demultiplexer will be in the order of less than 1 mm. Either the<br />
complete demultiplexer can be produced in the deeply etched waveguide structure needed for<br />
the ultra-compact bends, or only the converter part of the device can be etched deeply in a<br />
second etching step. The first option has the additonal advantage of applying small radii for the<br />
array guide bends, which leads to sub-millimeter device dimensions.<br />
5.3.1 Cross talk tolerance analysis<br />
The tolerance of this solution to deviations in the production process can be analysed as<br />
follows. The amount of light which is not converted to its orthogonal <strong>polarisation</strong> state will be<br />
imaged at the image plane, shifted in position (with respect to the original image) as<br />
determined by the <strong>polarisation</strong> dispersion. In the worst case, this light will directly couple into<br />
one of the other receiver waveguides and will cause cross talk. The cross talk CT is found as:<br />
CT 10 log PTE = ⋅ ⎛--------- ⎞<br />
⎝ ⎠<br />
(5.8)<br />
whereby P TE and P TM denote the power in the TE- and TM-polarised field respectively. The<br />
converter presented in Chapter 3 measures a conversion ratio of more than 85%, which, in this<br />
case, results in a cross talk value of -7.5 dB. Using equation 5.8, it is easily verified that for<br />
P TM<br />
output
5.4 Polarisation splitter 93<br />
cross talk levels better than -30 dB, the conversion ratio must be even at least 99.9%, which is<br />
a rather stringent demand.<br />
5.4 Polarisation splitter<br />
In the following section we introduce a novel method for obtaining <strong>polarisation</strong> independence<br />
using a <strong>polarisation</strong> splitter at the input. Due to the <strong>polarisation</strong> dispersion, the focal spot<br />
position along the image plane is differerent for both TE and TM <strong>polarisation</strong>. This is shown<br />
schematically in figure 5.13a. This shift in position Δspol can be compensated by shifting the<br />
original input field of the corresponding <strong>polarisation</strong> in the opposite direction along the object<br />
plane as shown in figure 5.13b. A <strong>polarisation</strong> splitter is then needed to spatially separate both<br />
<strong>polarisation</strong>s. This method is best applicable to 1 × N <strong>demultiplexers</strong> and a sample layout is<br />
depicted in figure 5.14. However, it can also be applied to N × N <strong>demultiplexers</strong> with channel<br />
spacings larger than the TE-TM shift (determined by the receiver spacing d r ). In this way,<br />
crossings are avoided or transmitter waveguides coming too close to each other, which is<br />
shown schematically in figure 5.15. This is the case, for instance, for the PHASAR of section<br />
5.2, the design parameters of which are listed in table 5.1. This device has a TE-TM shift of<br />
4.0 nm and a channel spacing of 3.2 nm. In combination with a receiver spacing d r of 5.5 μm<br />
and a waveguide width of 2 μm, the transmitter waveguides will overlap over 0.6 μm.<br />
ΤΕ+ΤΜ<br />
Δs pol<br />
ΤΜ<br />
ΤΕ<br />
object<br />
plane<br />
object<br />
plane<br />
phased<br />
array<br />
(a)<br />
phased<br />
array<br />
(b)<br />
image<br />
plane<br />
ΤΜ<br />
ΤΕ<br />
image<br />
plane<br />
Δs pol<br />
ΤΕ+ΤΜ<br />
Figure 5.13 Schematic diagram showing the <strong>polarisation</strong> dispersion: without<br />
compensation gives shifted images (a) and with compensation by proper<br />
positioning of the transmitter waveguides gives a single image.
94 5. Polarisation <strong>independent</strong> PHASARs <strong>based</strong> on birefringent waveguides<br />
TE+TM<br />
TE-TM<br />
splitter<br />
TM<br />
Figure 5.14 Application of a <strong>polarisation</strong> splitter at the input.<br />
TE+TM<br />
TE-TM<br />
splitter<br />
TE+TM TE-TM<br />
splitter<br />
crossing<br />
TE<br />
Δs pol<br />
object<br />
plane<br />
5.4.1 Cross talk tolerance analysis<br />
d r<br />
image<br />
plane<br />
The cross talk arising from the application of this method depends on the positioning of the TE<br />
and TM-polarised input fields with respect to each other. If proper positioning is not achieved,<br />
the TE and TM pass-bands will be shifted relative to each other and in the worst case, the light<br />
will couple into an adjacent receiver waveguide giving opportunity for cross talk. The effect of<br />
proper positioning of the two input fields can be disturbed either by a deviation of the<br />
<strong>polarisation</strong> dispersion Δs pol , or by insufficient discrimination of the <strong>polarisation</strong> splitter.<br />
The first cause can be circumvented in the same way as for the “order trick” - by flattening of<br />
the <strong>wavelength</strong> response. For the latter cause, however, a fraction of the unwanted <strong>polarisation</strong><br />
TM<br />
TM<br />
TE<br />
TE<br />
TE+TM<br />
d r<br />
TE+TM<br />
Figure 5.15 Schematic diagram of the problems occurring when the <strong>polarisation</strong><br />
dispersion shift Δs pol is larger than the receiver spacing d r .
5.5 Discussion 95<br />
may be coupled into the wrong output of the splitter. The power level of this fraction is called<br />
the splitting ratio (in dB). In a worst case situation, this power is coupled into a different<br />
receiver waveguide, resulting in an increased cross talk level. Polarisation splitters are known<br />
from literature. They have been produced in LiNbO 3 [172], silicon-<strong>based</strong> [114] and Al 2 O 3 -on-<br />
SiO 2 waveguide structures [161] (PHASAR-<strong>based</strong>). If we restrict ourselves to the <strong>InP</strong>-<strong>based</strong><br />
ones [2,101,120,156,158], we note that the splitting ratio is in the order of -20 dB for both<br />
<strong>polarisation</strong>s. As this is not sufficient to obtain cross talk levels better than -30 dB, they should<br />
be cascaded as suggested by Shani et. al. [114]. In this way, splitting ratios in the order of -35<br />
to -45 dB can be achieved.<br />
5.5 Discussion<br />
In this Chapter four methods to obtain <strong>polarisation</strong> independence have been discussed. They all<br />
have their own specific advantages and disadvantages. The compensation of <strong>polarisation</strong> dispersion<br />
provides a good solution. In this way, however, due to the high sensitivity to layer<br />
thickness and waveguide width, an intermediate characterisation of the device is required in<br />
order to obtain a fully <strong>polarisation</strong> <strong>independent</strong> device.<br />
From the design point of view, the insertion of a half-wave plate is the most simple method for<br />
obtaining <strong>polarisation</strong> independence and, in principle, it can be applied to any waveguide structure.<br />
For semiconductor structures, the half-wave plate takes the form of an integrated <strong>polarisation</strong><br />
converter. The bandwidth of this method only depends on the bandwidth of the converter.<br />
The conversion ratio, however, must be at least 99.9% in order to maintain a cross talk level<br />
better than -30 dB.<br />
Finally, a novel method has been introduced for obtaining <strong>polarisation</strong> independence by using<br />
<strong>polarisation</strong> splitters at the input of the PHASAR. The method is best suited for 1 × N devices,<br />
but can also can be applied to N ×<br />
N devices if the <strong>polarisation</strong> dispersion is smaller than the<br />
channel spacing. In order to maintain a high cross talk level, <strong>polarisation</strong> splitters should be<br />
cascaded. Additionally, the response of the PHASAR should be flattened in order to relax the<br />
<strong>polarisation</strong> dispersion tolerances with respect to the production process.
96 5. Polarisation <strong>independent</strong> PHASARs <strong>based</strong> on birefringent waveguides
Chapter 6<br />
MMI-<strong>based</strong> components<br />
A special type of phased-array demultiplexer is one in which the free propagation regions are<br />
replaced with multimode interference (MMI) couplers. In this Chapter the design of such MMI<strong>based</strong><br />
<strong>demultiplexers</strong> is discussed. A tolerance analysis is given together with experimental<br />
results, part of which have been presented at the ECIO’95 [31]. Also, a MMI-<strong>based</strong> mode filter<br />
is discussed and experimental results are given, part of which have been presented at the<br />
IEEE-LEOS Annual Meeting [36]. Finally, the interference patterns in a MMI coupler have<br />
been made visible. This is done using a MMI-coupler, produced in an aluminium oxide ridge<br />
waveguide structure implanted with erbium, and by imaging the green light generated by<br />
upconversion at high pumping power levels. This work has been carried out in cooperation<br />
with the FOM-Institute for Atomic and Molecular Physics and has been presented earlier<br />
[30,59].<br />
6.1 Introduction<br />
Multimode Interference (MMI) couplers are broad waveguides which support a high number<br />
of modes, determined by the waveguide width and the lateral index contrast. These waveguides<br />
have the so-called “self-imaging” property, which means that a replica of an arbitrary field<br />
applied at the input is reproduced at periodic positions along the direction of propagation. At<br />
fixed positions in between these intervals, multiple images of the input field are obtained with<br />
the input power equally distributed over all images.<br />
A significant advantage of MMI couplers over directional couplers is their tolerance with<br />
respect to <strong>polarisation</strong>, <strong>wavelength</strong> and etch depth variations [121]. A drawback of these<br />
couplers is their sensitivity to the waveguide width. This restriction has not, however, been an<br />
obstacle for wide usage of MMI couplers for splitting and combining functions [51], in<br />
interferometric switches [67], phase-diversity optical receivers [40,105] and in ring lasers<br />
[111].<br />
The derivation of general analytical expressions for the phases of the images in MMI couplers<br />
has led to the design and production of MMI-<strong>based</strong> <strong>wavelength</strong> <strong>demultiplexers</strong> [8,31,76]. The<br />
design and experiments of such a MMI-<strong>based</strong> demultiplexer are discussed in section 6.2 and<br />
special attention is paid to tolerance analysis.<br />
Another application of MMI couplers is a spatial mode filter. This device uses a single MMI
98 6. MMI-<strong>based</strong> components<br />
coupler to spatially separate the fundamental mode from the first-order mode. The design and<br />
experimental results of a MMI-<strong>based</strong> spatial mode filter are presented in section 6.3.<br />
Finally, the interference patterns in a MMI coupler operating at 1485 nm <strong>wavelength</strong> are made<br />
visible in green light. This has been done for an erbium-doped aluminium oxide ridge<br />
waveguide structure by imaging the green light generated by upconversion at high pumping<br />
power levels. This work, carried out in cooperation with the FOM-Institute for Atomic and<br />
Molecular Physics, is presented in section 6.4.<br />
6.2 MMI-MZI demultiplexer<br />
MMI-<strong>based</strong> <strong>demultiplexers</strong> reported so far were produced for duplexing 980 nm and 1550 nm<br />
light, and were produced by using a single MMI coupler in silica-<strong>based</strong> [50,66,98] and LiNbO3 waveguide structures [112]. This is possible due to the high ratio of the beat lengths at both<br />
<strong>wavelength</strong>s ( Lπ, λ1 ⁄ Lπ, λ2 ≈ λ2 ⁄ λ1 ≈ 1.5 ). For smaller <strong>wavelength</strong> spacings, the MMI<br />
coupler will become excessively long, which makes it less tolerant. In this case, a Mach-<br />
Zehnder Interferometer (MZI) configuration should be used, consisting of two 2 × 2 MMI-<br />
couplers connected to each other by two waveguides, the optical path lengths of which are<br />
different. This concept has been used for switches [67,159], <strong>polarisation</strong> splitters [113,120]<br />
and duplexers [58].<br />
If MMI couplers are used to replace the FPR’s of the PHASAR, the demultiplexer acts as a<br />
MMI demultiplexer in a MZI configuration, a so-called MMI-MZI demultiplexer. A sample<br />
layout of such a demultiplexer is shown in figure 6.1. In this figure, it can be seen that the<br />
number of array guides is equal to the number of input waveguides, or, in the sense of<br />
demultiplexing, to the number of waveguide channels. This is a feature by which the MMI-<br />
MZI demultiplexer can be distinguished from a conventional PHASAR demultiplexer.<br />
MMI<br />
section<br />
input<br />
waveguides<br />
Array<br />
waveguides<br />
MMI<br />
section<br />
output<br />
waveguides<br />
Figure 6.1 Novel type phased-array demultiplexer: MMI-MZI demultiplexer.
6.2 MMI-MZI demultiplexer 99<br />
A significant advantage of the MMI-MZI demultiplexer with respect to the “classical” phasedarray<br />
demultiplexer becomes apparent when deeply etched waveguides are applied, either for<br />
miniaturisation or for <strong>polarisation</strong> independence purposes. In this case, a problem arises with<br />
phased-array <strong>demultiplexers</strong>: increased junction loss at the interface between the array<br />
waveguides and the free propagation region (FPR). The lithographic resolution of the<br />
production process limits the minimum spacing of the array waveguides and therefore also the<br />
insertion loss. These problems can be avoided by using a MMI-MZI demultiplexer.<br />
6.2.1 Operation principle<br />
A N × N MMI-MZI demultiplexer consists of two identical N × N MMI couplers connected to<br />
each other by N array waveguides, the lengths of which are properly chosen. A schematic<br />
layout is depicted in figure 6.2, in which the notation of Besse et al. has been followed [13,14].<br />
i=<br />
N<br />
...<br />
2<br />
a<br />
1<br />
2W/N<br />
W/N<br />
W<br />
L MMI<br />
Figure 6.2 Schematic layout of a generalised MMI-MZI demultiplexer.<br />
j=<br />
1<br />
2<br />
...<br />
N<br />
Both MMI couplers act as N × N power splitter-combiners if the length LMMI is properly<br />
chosen: LMMI = 3Lπ ⁄ N , with Lπ = π ⁄ ( β0 – β1) , whereby β0 and β1 are the propagation<br />
constants of the fundamental and the first order mode in the MMI coupler respectively. This<br />
means, that a N-fold image of a field applied at any input i is obtained at the outputs j, when the<br />
inputs and outputs are positioned as depicted in figure 6.2 (the parameter a can be chosen<br />
freely [13,14]). The phase relations ϕj,i from input i to output j of the N × N imaging MMI<br />
coupler have been calculated by Besse et al. [13,14]:<br />
ϕ j, i<br />
=<br />
⎧<br />
⎪<br />
⎪<br />
⎨<br />
⎪<br />
⎪<br />
⎩<br />
ϕ 0<br />
whereby ϕ0 is a constant phase. Due to reciprocity, light will constructively interfere into<br />
output i' of the second MMI-coupler if ϕ j', i' =<br />
– ϕ j, i.<br />
The phase relations calculated using<br />
equation 6.1 are listed in table 6.1 with N = 4. These phases can be produced by adjusting the<br />
path lengths of the array waveguides. It has been shown that a specific set of lengths meets<br />
these requirements.<br />
j'=<br />
π<br />
+ ------- ( 2N + 1 – j – i)<br />
( j + i – 1)<br />
4N<br />
π<br />
ϕ0 π<br />
4N<br />
------- + 2N – j i + ( ) j i – ( )<br />
+<br />
i+j odd<br />
i+j even<br />
i'=<br />
N<br />
...<br />
2<br />
1<br />
(6.1)
100 6. MMI-<strong>based</strong> components<br />
Table 6.1 The phase relations for a single 4 × 4 MMI-coupler.<br />
Because of the constant phase factor ϕ0 , not the absolute phases are important for the<br />
demultiplexer, but the phase differences between adjacent inputs of the second MMI-coupler<br />
= – ( – ) in order the constituent waves at a certain output to be in phase:<br />
Δϕ j', i' ϕ j' + 1, i'<br />
Δϕ j', i'<br />
=<br />
⎧<br />
⎪<br />
⎪<br />
⎨<br />
⎪<br />
⎪<br />
⎩<br />
Thiscan be simplified to:<br />
ϕ j,i i = 1 i = 2 i = 3 i = 4<br />
ϕ 1,i π 3π/4 −π/4 π<br />
ϕ 2,i 3π/4 π π −π/4<br />
ϕ 3,i −π/4 π π 3π/4<br />
ϕ 4,i π −π/4 3π/4 π<br />
ϕ j', i'<br />
π<br />
– π + ------- [ ( 2N + 1 – j' – i')<br />
( j' + i' – 1)<br />
– ( 2N – 1 – j' + i')<br />
( j' + 1 – i')<br />
]<br />
4N<br />
π<br />
π<br />
4N<br />
------- + 2N – j' i' + ( ) j' i' – ( ) 2N – j' i' – ( ) j' i' + ( )<br />
–<br />
i'+j' odd<br />
(6.2)<br />
[ ]<br />
i'+j' even<br />
Δϕ j', i'<br />
=<br />
⎧<br />
⎪<br />
⎪<br />
⎨<br />
⎪<br />
⎪<br />
⎩<br />
j'<br />
– π + π( i' – 1)<br />
⎛1 – --- ⎞<br />
⎝ N⎠<br />
π – πi'⎛<br />
j'<br />
1 – --- ⎞<br />
⎝ N⎠<br />
Δϕ j',i' i' = 1 i' = 2 i' = 3 i' = 4 ΔΦ j'<br />
Δϕ 1',i' π/4 −π/4 3π/4 −3π/4 −π/2<br />
Δϕ 2',i' π 0 0 π −π<br />
Δϕ 3',i' 3π/4 −3π/4 π/4 −π/4 −3π/2<br />
λ 1 λ 2 λ 4 λ 3<br />
i'+j' odd<br />
i'+j' even<br />
Table 6.2 The required phase differences at the second MMI-coupler, calculated<br />
between adjacent array guides for N = 4.<br />
(6.3)<br />
These phases are spaced at a distance of ΔΦ j' =<br />
– j'2π<br />
⁄ N between adjacent points. The<br />
complete set of required phase differences is listed in table 6.2. The columns denoted by i' = 1
6.2 MMI-MZI demultiplexer 101<br />
to i' = 4 from table 6.2 show the differences Δϕj',i' which are required in order to couple all the<br />
power to output i' of the 4 × 4 demultiplexer. Investigation of the table shows that going from<br />
output i' = 1 to output i' = 2, from output i' = 2 to output i' = 4, from output i' = 4 to output<br />
i' = 3, and from output i' = 3 back to output i' = 1, the required phase difference Δϕj',i' increases<br />
with a constant amount ΔΦj' , which is listed in the far right column. This can be made clear<br />
when they are drawn in a phase diagram as a function of i', as shown in figure 6.3 with N = 4 as<br />
an example.<br />
i' = 3<br />
i' = 4<br />
j' = 1<br />
i' = 1<br />
i' = 2<br />
j' = 2<br />
i' = 1 i' = 2<br />
i' = 4<br />
i' = 3<br />
i' = 1<br />
i' = 2<br />
j' = 3<br />
Figure 6.3 Required phase differences ΔΦ j' at the input of the second MMI<br />
coupler with N = 4.<br />
i' = 3<br />
i' = 4<br />
From the values ΔΦ j' it becomes clear that, if we connect the output ports j of the first MMI<br />
coupler to the input ports j' of the second MMI coupler by waveguides with lengths L j which<br />
satisfy:<br />
ΔL j L j + 1 – L j ---------- j<br />
Δβ<br />
2π<br />
= = = -----------<br />
(6.4)<br />
NΔβ<br />
with Δβ being the channel spacing Δβ = β (λi'+1 ) - β (λi' ), then the light will shift from output i'<br />
to i'+1 if the <strong>wavelength</strong> changes from λi' to λi'+1 , according to the sequence listed in the<br />
bottom row of table 6.2.<br />
With the array guide lengths chosen according to equation 6.4, the correct absolute phase<br />
distribution is not matched at the input of the second MMI-coupler for the design <strong>wavelength</strong>.<br />
The correct phase distribution can be obtained by a small correction on the lengths, which does<br />
not affect the dispersive properties of the array. This correction is discussed in Appendix F.<br />
ΔΦ j'<br />
The spectral response of a 4 ×<br />
4 MMI-MZI demultiplexer has been calculated using a modal<br />
propagation analysis implemented in a microwave CAD tool [29,75] - Hewlett-Packard’s<br />
Microwave Desing System (MDS) - and is shown in figure 6.4. The demultiplexer has high<br />
uniformity of the output intensity of the different channels, which is inherently due to the<br />
uniform splitting ratio of the MMI-couplers. Low insertion loss in the order of 0.5 dB, and<br />
cross talk values as low as -30 dB can be obtained. The practically usable bandwidth is not<br />
determined by the 1-dB pass-band of the desired output channel, which is approximately<br />
1.1 nm, but by the cross talk level of the undesired output channels. This low cross talk passband<br />
is in the order of 0.19 nm for a -25 dB cross talk level (and only 0.11 nm for -30 dB),<br />
which leads to high demands on the <strong>wavelength</strong> accuracy of the laser sources used.<br />
Furthermore, the bandwidth of the MMI couplers, which determines the bandwidth of the<br />
demultiplexer, is inversely proportional to the number of input and output channels N [15].<br />
This is a major restriction for this type of demultiplexer and is discussed in the next section.
102 6. MMI-<strong>based</strong> components<br />
Transmission [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
1 3 4 2<br />
-30<br />
1510 1515 1520<br />
Wavelength [nm]<br />
1525 1530<br />
Figure 6.4 Calculated spectral response of a 4x4 MMI-MZI demultiplexer.<br />
A layout of how the required length differences ΔLj could be made is shown in figure 6.1. It<br />
can be seen, that the length difference monotonically increases with array guide number j,<br />
which agrees with the values of ΔΦj' listed in the last column of table 6.2. However, these<br />
values denote the phase difference between adjacent array guides. If we consider the phase<br />
difference with respect to a reference guide, we find a quadratical increase in stead of a linear<br />
increase ( – ( – ) = π ⁄ 2,<br />
3π ⁄ 2,<br />
6π ⁄ 2 for j' = 2, 3, 4 respectively). As for a proper<br />
ϕ j', i'<br />
ϕ1', i'<br />
phased-array design the phase difference between adjacent array guides should linearly<br />
increase with respect to a reference array guide, it means that the phase transfer will be<br />
disturbed. This causes considerably high sideobes, as can be seen in the graph of figure 6.4.<br />
This can be avoided if we consider the phase difference Δϕj',i' with respect to a - arbitrarily<br />
chosen - reference guide. As an example we take guide j' = 1 to be the reference guide, leading<br />
to the phase differences Δϕ j', i' = – ( ϕ j', i' – ϕ1', i')<br />
, which are listed in table 6.3, again with<br />
N = 4 as an example.<br />
Table 6.3 The required phase differences at the second MMI-coupler, calculated<br />
with N = 4 and with respect to reference array guide j' = 1.<br />
Δϕ j',i' i' = 1 i' = 2 i' = 3 i' = 4 ΔΦ j'<br />
Δϕ 1',i' 0 0 0 0 0⋅π/2<br />
Δϕ 2',i' π/4 −π/4 3π/4 −3π/4 −1⋅π/2<br />
Δϕ 3',i' −3π/4 −π/4 3π/4 π/4 +1⋅π/2<br />
Δϕ 4',i' 0 π π 0 −2⋅π/2<br />
With this approach the simple layout as shown in figure 6.1 can no longer be used, because the<br />
array guides would intersect at small angles as depicted in figure 6.5a. These crossings can be<br />
circumvented if the path length differences are made by U-bends. This geometry, presented for
6.2 MMI-MZI demultiplexer 103<br />
the first time at the ECIO’95 conference [31], has a decreased width at the expense of an<br />
increased length as depicted in figure 6.5b.<br />
Later another geometry was proposed by Lierstuen and Sudbø [76] and by Herben et al. [55]<br />
(who proposed to use it in a multi-<strong>wavelength</strong> laser). In this geometry, depicted in figure 6.5c,<br />
the crossings are circumvented by arranging the phase differences ΔΦ j' in such a way that they<br />
increase monotonically in both directions from a reference array guide towards the outer array<br />
guides. The advantage of this geometry is that all array guides have identically shaped arms,<br />
except for the pathlength difference which is obtained by inserting extra lengths of straight<br />
waveguides. On the other hand, the device has an increased height, which limits the integration<br />
scale.<br />
The number of U-bends needed in array guide j (configuration of figure 6.5b) can directly be<br />
read from the far right column of table 6.3, when the path length difference between a U-bend<br />
and a straight waveguide is calculated according to:<br />
ΔL U-bend<br />
( π ⁄ 2)<br />
= --------------<br />
Δβ<br />
(6.5)<br />
whereby R is the bending radius of the curved waveguide section, and L the horizontal length<br />
of the U-bend.<br />
(a)<br />
(b)<br />
(c)<br />
crossings<br />
Figure 6.5 Schematic layout of 4x4 MMI-MZI demultiplexer: with waveguide<br />
crossings (a), with U-bends (b), and without waveguide crossings but with<br />
increased height (c).
104 6. MMI-<strong>based</strong> components<br />
Because the U-bend consists of four curved waveguide sections with an angle α, the path<br />
length along the U-bend equals 4αR. If the length L of the U-bend is substituted with 4Rsin(α),<br />
we end up with:<br />
α – sin( α)<br />
π<br />
-------------<br />
8RΔβ<br />
which can be solved for any value of R.<br />
According to the above approach, the spectral response of a 4 × 4 MMI-MZI demultiplexer<br />
with U-bends (configuration of figure 6.5b) has been calculated and is shown in figure 6.6. If<br />
we compare the graph in this figure with the one of figure 6.4, various items catch the eye:<br />
• 100% bandwidth increase. The -1 dB band is increased from 1.1 nm to 2.2 nm and the<br />
-25 dB cross talk bandwidth is increased from 0.19 nm to 0.42 nm (the -30 dB bandwidth<br />
is increased from 0.11 nm to 0.23 nm).<br />
• Side lobe level reduction. The level of the side lobes, which lie between the channel<br />
maxima, reduces from -4 dB to -12 dB.<br />
• Side lobe reduction. A number of side lobes seem to have disappeared and the side<br />
lobes now overlap better.<br />
Still, the bandwidth is small compared to that of a conventional phased-array demultiplexer. In<br />
contrast with the PHASAR demultiplexer, it is not possible for MMI-MZI <strong>demultiplexers</strong> to<br />
flatten the <strong>wavelength</strong> response without changing the channel spacing. This is due to the fact<br />
that there is no scanning bundle in the MMI couplers as in the FPR of a PHASAR<br />
demultiplexer and can be explained as follows. In order to flatten the response, a specific phase<br />
distribution at the input of the second MMI coupler, which results in constructive interference<br />
into one of the outputs, has to be maintained over a longer <strong>wavelength</strong> range. The result will be<br />
that any phase distribution will be maintained over a longer <strong>wavelength</strong> range (including the<br />
ones that do not result in constructive interference into one of the outputs), leading to a larger<br />
channel spacing.<br />
Transmission [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
In contrast with the first structure proposed (figure 6.1), in which the path length differences<br />
were obtained by length differences between straight waveguides (of which the propagation<br />
=<br />
1 3 4 2<br />
-30<br />
1510 1515 1520<br />
Wavelength [nm]<br />
1525 1530<br />
Figure 6.6 Calculated spectral response of a 4x4 MMI-MZI demultiplexer with<br />
U-bends.<br />
(6.6)
6.2 MMI-MZI demultiplexer 105<br />
loss can be kept negligibly low), now each array guide has a different number of U-bends. As<br />
the propagation loss of a curved waveguide is higher than that of a straight waveguide, an<br />
unequal power distribution is expected at the input of the second MMI coupler. This will result<br />
in a higher cross talk and will be discussed in the next section.<br />
6.2.2 Tolerance analysis<br />
For the tolerance analysis we will consider the definition of the length L of a N × N MMI<br />
coupler more closely. If we assume a strongly guided structure, it may be approximated by<br />
[15]:<br />
L<br />
=<br />
2<br />
4 nW<br />
--- ⋅ ----------<br />
N λ<br />
whereby n is the (transverse) effective index in the MMI coupler region. Small changes of the<br />
<strong>wavelength</strong>, the width and effective index (through layer thickness variations) cause a change<br />
in the beat length L π and, consequently, performance degradation. Hence using equation 6.7,<br />
the variations of the <strong>wavelength</strong>, width, index and length of the MMI coupler are related by:<br />
It now becomes apparent that all relative changes in these parameters have the same effect,<br />
which is a change in the length at which the image is obtained. The imaging is not performed<br />
properly, leading, in case of a very small change, to defocusing of the light into the output<br />
waveguides and the result is an increased insertion loss. In more severe cases, increased cross<br />
talk is also the result.<br />
Using the relative changes of equation 6.8, particular tolerance windows for a single MMI<br />
(6.7)<br />
∂Lπ<br />
∂L<br />
∂n<br />
----------- -------- -------- 2<br />
Lπ L n<br />
W ∂ ∂λ<br />
= = = ---------- = --------<br />
(6.8)<br />
W λ<br />
coupler can be calculated (e.g. the 1-dB bandwidth). These windows should be divided by<br />
in order to obtain the tolerance windows for a MMI-MZI demultiplexer, which consists of two<br />
MMI couplers.<br />
For the analysis, the tolerance windows will be calculated and be compared with results of<br />
simulations performed with MDS. We consider the following structure: the MMI-MZI uses<br />
two MMI couplers which are 14 μm wide and 423 μm long, designed for application in a<br />
deeply etched waveguide structure, consisting of a 600 nm thick InGaAsP guiding layer and a<br />
300 nm thick <strong>InP</strong> cladding layer. The advantage of such a waveguide structure is not only that<br />
small bending radii can be applied, but also that the tolerance to etch depth variations is greatly<br />
relaxed. The input and output waveguides are 2 μm wide, which allows for a relaxed length<br />
tolerance [127,134] and on the other hand, also limits the width of the MMI coupler. The<br />
channel spacing is 3.2 nm (400 GHz) with a central <strong>wavelength</strong> of 1520 nm.<br />
Bandwidth<br />
Besse et al. [15] have shown that in order to limit the excess loss of a single MMI coupler to α<br />
(in dB), the <strong>wavelength</strong> should be limited to a range<br />
<strong>wavelength</strong> λc :<br />
2 δλ centred around the centre<br />
2 δλ Z( α)<br />
Nπd 2<br />
0<br />
≅ (6.9)<br />
-----------------λ<br />
2 c<br />
4W MMI<br />
2
106 6. MMI-<strong>based</strong> components<br />
whereby d 0 is the waist of the Gaussian field describing the input field. The exact form of Z(α)<br />
can be found elsewhere [15], only a few values are given here: Z(α) ≈ 0.8, 0.5 for α = 1 and<br />
0.5 dB respectively. According to equation 6.9, a 1-dB bandwidth of 28 nm is expected for a<br />
single MMI coupler and 20 nm for the MMI-MZI demultiplexer. A striking detail is that if the<br />
width is increased, even with the same relative amount as the number of channels (i.e. ratio N/<br />
W MMI constant), the bandwidth decreases. Figure 6.7 shows the response of the MMI-MZI<br />
demultiplexer over a longer <strong>wavelength</strong> range, calculated with MDS. From this figure the 1-dB<br />
bandwidth appears to be approximately 23 nm.<br />
Transmission [dB]<br />
0<br />
-1<br />
-2<br />
-3<br />
2<br />
1<br />
3<br />
4 2 1<br />
3<br />
-4<br />
1500 1510 1520<br />
Wavelength [nm]<br />
1530 1540<br />
Figure 6.7 Wavelength response of the 4 × 4 MMI-MZI demultiplexer.<br />
Width tolerance<br />
With respect to equation 6.9, an equivalent expression is given for the width tolerance of a<br />
single MMI coupler:<br />
from which it follows that the tolerance remains constant as long as the ratio N/WMMI is kept<br />
constant. The 1-dB width tolerance calculated using this equation equals 0.13 μm and for the<br />
MMI-MZI demultiplexer 0.09 μm. Figure 6.8 shows the transmission of the MMI-MZI<br />
demultiplexer (calculated with MDS) versus the deviation ΔW of the width with respect to the<br />
designed value. The 1-dB width tolerance measures approximately 0.10 μm. The figure also<br />
demonstrates that the lateral dimensional control of ±<br />
0.2 μm, which is a realistic value for a<br />
standard lithographic process, is not sufficient for low-loss operation.<br />
The effect of a width deviation might, however, not be as dramatic as figure 6.8 suggests. If we<br />
again consider equation 6.8, we see that a deviation in the width can be compensated by a<br />
deviation in the <strong>wavelength</strong>. This means that if the width used deviates from the designed<br />
width, the <strong>wavelength</strong> response will shift with respect to the designed central <strong>wavelength</strong>. This<br />
is depicted in figure 6.9. In this graph the contour lines denote an equal insertion loss for the<br />
central channel. It shows that a negative width deviation ΔW results in a <strong>wavelength</strong> shift<br />
towards a lower <strong>wavelength</strong> region. It also shows that the bandwidth increases linearly with the<br />
<strong>wavelength</strong> (width of the equal-loss region increasing linearly), which follows directly from<br />
equation 6.9.<br />
4<br />
δW Z( α)<br />
Nπd 2<br />
0<br />
≤ --------------------<br />
(6.10)<br />
16W MMI<br />
2<br />
1
6.2 MMI-MZI demultiplexer 107<br />
Transmission [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
-30<br />
-0.4 -0.2 0.0 0.2 0.4<br />
ΔW [μm]<br />
Figure 6.8 Transmission characteristic of the 4 × 4 MMI-MZI demultiplexer<br />
versus width deviation.<br />
Wavelength [nm]<br />
1600<br />
1550<br />
1500<br />
1450<br />
1400<br />
-11<br />
-9<br />
-8<br />
-8<br />
-7<br />
-6<br />
-4<br />
-2<br />
-3<br />
-6<br />
-5<br />
-7<br />
-9<br />
-11<br />
-11<br />
-2<br />
-4<br />
-12<br />
-10<br />
-8<br />
-10<br />
-6<br />
-6 -6<br />
-0.4 -0.2 0.0 0.2 0.4<br />
ΔW [μm]<br />
Figure 6.9 Wavelength change versus width variation.<br />
-5<br />
-7<br />
-3<br />
-7<br />
-1<br />
-7<br />
-1<br />
-6<br />
-6<br />
-2<br />
The <strong>wavelength</strong> shift due to a width deviation can be estimated using equation 6.8. If we use<br />
ΔW = -0.2 μm as an example, the <strong>wavelength</strong> shift of the envelope will be approximately<br />
43 nm. The response has been calculated for this width deviation and the result is depicted in<br />
figure 6.10. In figure 6.10a the response is shown for the design <strong>wavelength</strong> region. The cross<br />
talk and the insertion loss have increased. Figure 6.10b shows the response in the shifted<br />
<strong>wavelength</strong> region, from which it follows that the <strong>wavelength</strong> shift is approximately 45 nm. If<br />
this figure is compared with figure 6.6, it becomes apparent that the bandwidth has decreased<br />
(from 20 to 19 nm, which is verified with equation 6.9).<br />
-5<br />
-4<br />
-3<br />
-9<br />
-6<br />
-5<br />
-6<br />
-7<br />
-8 -9<br />
-2<br />
-4<br />
-5<br />
-8<br />
-6<br />
-5<br />
-7<br />
-3<br />
-9<br />
-5<br />
-1<br />
-7<br />
-8<br />
-1<br />
1<br />
2<br />
3<br />
4<br />
-3 -2<br />
-5 -4<br />
-5<br />
-6<br />
-6<br />
-4<br />
-8
108 6. MMI-<strong>based</strong> components<br />
Transmission [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
4 2 1 3 4 2 1 3 4 2 1<br />
-30<br />
1510 1515 1520<br />
Wavelength [nm]<br />
1525 1530<br />
-30<br />
1470 1475 1480<br />
Wavelength [nm]<br />
1485 1490<br />
(a) (b)<br />
Layer thickness tolerance<br />
A change in the transverse effective index also affects the length of the MMI coupler, as<br />
follows from equation 6.8. The effective index is dependent on the <strong>polarisation</strong> and the<br />
temperature, both of which can be controlled accurately. The index, however, also depends on<br />
the layer thickness, especially the guiding layer thickness, as most of the power is present in<br />
that layer. Figure 6.11 shows the response versus layer thickness variations.<br />
Transmission [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
Figure 6.10 Response for ΔW = -0.2 μm for the <strong>wavelength</strong> region according to<br />
the design (a) and for the shifted <strong>wavelength</strong> region (b).<br />
2<br />
-30<br />
500 550 600 650 700<br />
Q(1.3) layer thickness [nm]<br />
4<br />
3<br />
The graph of figure 6.11a shows interesting results. As the transverse effective index is mainly<br />
determined by the guiding layer thickness, a variation of this layer has a large impact on the<br />
response: a change in the layer thickness shifts the <strong>wavelength</strong> response. From this graph it can<br />
also be seen that an increasing layer thickness results in a relaxed tolerance with respect to loss<br />
and cross talk. This is explained by the fact that the relative change ∂n<br />
⁄ n is small.<br />
Transmission [dB]<br />
A variation in the cladding layer, on the other hand, which only guides a small part of the<br />
power, leads to a small relative index change and therefore the response is hardly influenced.<br />
As can be seen from figure 6.11b, the cladding layer thickness has to be controlled within<br />
Transmission [dB]<br />
0.0<br />
-0.5<br />
-1.0<br />
-1.5<br />
-2.0<br />
-2.5<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
1,2,3<br />
-10<br />
-15<br />
-20<br />
-25<br />
-3.0<br />
-30<br />
200 250 300 350 400<br />
<strong>InP</strong> layer thickness [nm]<br />
(a) (b)<br />
Figure 6.11 Response for guiding layer variation (a) and for cladding layer<br />
variation (b).<br />
4<br />
0<br />
-5<br />
Transmission [dB]
6.2 MMI-MZI demultiplexer 109<br />
± 15 nm in order to maintain a cross talk level of less than -30 dB, whereas the guiding layer<br />
thickness has to be controlled within 5 nm for the same cross talk level.<br />
Waveguide loss tolerance<br />
In addition to these production deviations, waveguide losses may also have an impact on the<br />
performance due to the different lengths of the array arms. The losses will vary from one array<br />
guide to another, leading to a non-uniform intensity distribution at the input of the second MMI<br />
coupler. This will especially be the case if the required path length differences are obtained by<br />
different numbers of curved waveguide sections in the array guides, as in the structure of figure<br />
6.5b. The imaging in the second MMI coupler is not performed properly and therefore power is<br />
not only coupled to the desired output, but also to undesired outputs, leading to unwanted high<br />
cross talk values.<br />
To analyse this effect, the phase transfer in the second MMI-coupler will be considered, from<br />
which it is found that for one particular output all contributions add up coherently, leading to a<br />
proper image. For all other outputs each phase transfer term appears to have one counterphase<br />
term, leading to extinction of the fields. If these terms have different amplitudes due to<br />
attenuation in the demultiplexer, the extinction factor is reduced, i.e. the cross talk increases.<br />
From this analysis the sensitivity to loss variations in the different array guides, which leads to<br />
power imbalance at the input of the second MMI-coupler, can be calculated as follows and<br />
considering the configuration of figure 6.2. It should be noted that the splitting ratio of the<br />
MMI couplers is assumed to be uniform; the MMI couplers do not necessarily need to be<br />
without loss. The field at output i' of the second MMI-coupler is defined as:<br />
A i'<br />
If we use input i = 1 as an example (calculations for other values of i are analogous) and<br />
evaluate this formula for odd i', we arrive at:<br />
Ai', odd<br />
=<br />
+<br />
j' = odd<br />
This can be simplified to:<br />
Ai', odd<br />
N<br />
∑<br />
N<br />
∑<br />
j' = even<br />
+<br />
=<br />
N<br />
∑<br />
N<br />
∑<br />
N<br />
∑<br />
– i', j'<br />
A j'e ϕi' j'<br />
A j'e jϕ – = =<br />
+<br />
i', j' (6.11)<br />
j' = 1<br />
1<br />
------- A1e N<br />
1<br />
------- A1e N<br />
N<br />
∑<br />
j' = odd<br />
j' = even<br />
, A j' e jϕ<br />
j' = odd<br />
π<br />
j ϕ0 + π + ------- ( j' – 1)<br />
( 2N – j' + 1)<br />
4N<br />
π<br />
j ϕ0 + ------- j'( 2N – j')<br />
4N<br />
1<br />
------- A1e N<br />
1<br />
------- A1e N<br />
e<br />
e<br />
N<br />
∑<br />
j' = even<br />
π<br />
– j ϕ0 + π + ------- ( i' – j')<br />
( 2N – i' + j')<br />
4N<br />
π<br />
– j ϕ0 + ------- ( i' + j' – 1)<br />
( 2N – i' – j' + 1)<br />
4N<br />
j π<br />
– ------- [ – ( j' – 1)<br />
( 2N – j' + 1)<br />
+ ( i' – j')<br />
( 2N – i' + j')<br />
]<br />
4N<br />
j π<br />
– ------- [ – j'( 2N – j')<br />
+ ( i' + j' – 1)<br />
( 2N – i' – j' + 1)<br />
]<br />
4N<br />
(6.12)<br />
(6.13)
110 6. MMI-<strong>based</strong> components<br />
In a similar manner we can find for even i':<br />
Ai', even<br />
=<br />
N<br />
∑<br />
j' = odd<br />
+<br />
j' = even<br />
1<br />
------- A1e N<br />
1<br />
------- A1e N<br />
In order to evaluate the interference of the fields from input j' to an arbitrary output i', we can<br />
prove that for any odd (even) number j' there can always an even (odd) number j'' be found in<br />
such a way that the contribution from input j' cancels out the one from input j''. In other words,<br />
the phase transfer from input j' to output i' equals the phase transfer from input j'' to output i',<br />
modulo an odd multiple k of π radians:<br />
which holds for k = ± 1,<br />
± 3,<br />
…, N ⁄ 2.<br />
Generally stated, this means that every phase transfer term has one which is in counterphase.<br />
We look at equation 6.13, and only take into account the phases:<br />
After some manipulations we find for j'':<br />
N<br />
∑<br />
j π<br />
– ------- [ – ( j' – 1)<br />
( 2N – j' + 1)<br />
+ ( i' + j' – 1)<br />
( 2N – i' – j' + 1)<br />
– 4N ]<br />
4N<br />
j π<br />
– ------- [ – j'( 2N – j')<br />
+ ( i' – j')<br />
( 2N – i' + j')<br />
+ 4N ]<br />
4N<br />
ϕ j' = ϕ j'' + k ⋅ 4N<br />
– ( j' – 1)<br />
( 2N – j' + 1)<br />
+ ( i' – j')<br />
( 2N – i' + j')<br />
= – j''(<br />
2N – j'')<br />
+ ( i' + j'' – 1)<br />
( 2N – i' – j'' + 1)<br />
+ k ⋅ 4N<br />
j''<br />
(6.14)<br />
(6.15)<br />
(6.16)<br />
⎧<br />
⎪ 2N ( k + j' – 1)<br />
⎪<br />
--------------------------------- + ( 1 – j')<br />
i',j' odd<br />
=<br />
i' – 1<br />
⎨<br />
(6.17)<br />
⎪ 2N ( k + j' – 2)<br />
⎪<br />
--------------------------------- + ( 1 – j')<br />
i',j' even<br />
i'<br />
⎩<br />
Similar equations can be found if other inputs of the first MMI coupler are used. With<br />
equations 6.13 and 6.14 the phase contributions at the second MMI coupler have been<br />
calculated from input j' to output i' and are drawn in phase diagrams, as shown in figure 6.12.<br />
j' = 4<br />
j' = 3<br />
i' = 2<br />
j' = 2<br />
j' = 1<br />
j' = 1<br />
j' = 3<br />
i' = 3<br />
j' = 2<br />
j' = 4<br />
j' = 3<br />
j' = 2<br />
i' = 4<br />
Figure 6.12 Phase contributions from input j' to output i' at the second MMI<br />
coupler with N = 4.<br />
j' = 4<br />
j' = 1
6.2 MMI-MZI demultiplexer 111<br />
In this figure it can be seen that every contribution has one in counterphase. This means that at<br />
these outputs the sum field equals zero. For output i' = 1 all contributions add up coherently (all<br />
phase contributions equal zero). This figure makes it also clear, that a disturbance on the<br />
uniformity of the input powers will have a drastic impact on the cancellation of the fields and<br />
the result will be an increased cross talk level. The cross talk level can be estimated by taking<br />
into account the U-bend losses. If we use ε for a single U-bend loss, then the amplitude of the<br />
field at input j of the second MMI coupler can be denoted as:<br />
whereby nj is the number of U-bend sections for guide j (nj = [2,3,1,0] for j = [1,2,3,4]), as can<br />
be seen in figure 6.5. With the above formula we can calculate the field at output i' using<br />
equation 6.13 and 6.14. By approximating equation 6.18 with 1 – n jε , which holds for small<br />
values of ε, the result can be simplified to (for simplicity we have omitted the term A1 /2):<br />
With the above equations the cross talk values have been calculated with N = 4 and are shown<br />
in figure 6.13. As can be seen in this figure, the U-bend loss should be kept below 0.4 dB in<br />
order to keep the cross talk below -30 dB. It is stressed again that the MMI couplers are not<br />
assumed to be without loss, which means that the insertion loss of the MMI couplers should be<br />
added to the insertion loss values found in 6.13, in order to obtain the insertion loss of the<br />
demultiplexer.<br />
Crosstalk Cross talk level level [dB]<br />
6.2.3 Experimental results<br />
0<br />
-10<br />
-20<br />
a j ( 1 – ε)<br />
n j<br />
= (6.18)<br />
A1' = 4 – 6ε<br />
A2' = – 4εcos( π ⁄ 4)<br />
A3' = – 4εj sin(<br />
π ⁄ 4)<br />
A 4' = 2ε<br />
-30<br />
output 1<br />
output 2 & 3<br />
output 4<br />
-3<br />
-40<br />
-4<br />
0.0 0.5 1.0<br />
U-bend loss [dB]<br />
1.5 2.0<br />
Figure 6.13 Cross talk level versus U-bend loss with N = 4.<br />
(6.19)<br />
A number of experiments have been performed using different <strong>demultiplexers</strong> geometries,<br />
which will be described below.<br />
0<br />
-1<br />
-2<br />
Insertion loss [dB]
112 6. MMI-<strong>based</strong> components<br />
First experiment: conventional MMI-MZI demultiplexer<br />
For the first experiment the conventional MMI-MZI demultiplexer geometry, as depicted in<br />
figure 6.1, has been used. The device has been produced in an undeeply etched waveguide<br />
structure consisting of a 0.66 μm thick InGaAsP guiding layer on an <strong>InP</strong> substrate covered<br />
with a 0.32 μm thick <strong>InP</strong> layer. The designed etch depth was 80 nm into the guiding layer,<br />
which results in a calculated TE-TM shift of 4.0 nm. The 4-channel device was designed to<br />
operate at a central <strong>wavelength</strong> of 1536 nm with 1 nm channel spacing, thus aiming at<br />
<strong>polarisation</strong> <strong>independent</strong> operation by means of the so-called “order trick”.<br />
Both 1 × 4 and 4 × 4 MMI-couplers are 22 μm wide, the first being 284.5 μm long, the latter<br />
1133.5 μm. The width was chosen in such a way that large gaps were obtained in order to<br />
minimise coupling effects between adjacent array waveguides, which are 2.0 μm wide (both at<br />
input and output). The device uses bends with a radius of 650 μm, having negligible radiation<br />
loss. The overall size is mm2 , mm2 3.0 × 2.7 1.8 × 1.3 of which is occupied by the device<br />
itself. The rest is occupied by curved waveguides placed in such a way that the output<br />
waveguides lie in line with the input waveguides in order to simplify the measurement process.<br />
The measured response for TE <strong>polarisation</strong> is shown in figure 6.14a, which can be compared<br />
with the calculated response shown in 6.4. The measured response for TM <strong>polarisation</strong> is<br />
shown in figure 6.14b. No anti-reflection (AR) coating has been applied to the cleaved end<br />
facets. The propagation losses of a straight waveguide were measured to be 1.1 dB/cm for TE<br />
and 2.8 dB/cm for TM <strong>polarisation</strong>. The excess loss of the device is measured relative to a<br />
waveguide of the same length as the device. For TE <strong>polarisation</strong> it was found to be 1.2 dB for<br />
the best channel, which is only 0.7 dB more than calculated. The response for TM <strong>polarisation</strong><br />
shows a higher excess loss in the order of 4 dB, approximately 2 dB of which is due to the<br />
propagation loss difference. The cross talk is better than -15 dB and -10 dB for TE and TM<br />
<strong>polarisation</strong> respectively. Additionally, a residual TE-TM shift of 0.5 nm was measured, which<br />
can be accounted for by the deviation of the used etch depth (110 nm instead of 80 nm).<br />
Excess loss [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
-30<br />
1<br />
2<br />
3<br />
4<br />
1530 1532 1534 1536 1538 1540<br />
Wavelength [nm]<br />
(a) (b)<br />
1530 1532 1534 1536 1538 1540<br />
Wavelength [nm]<br />
Figure 6.14 Measured response of the first MMI-MZI demultiplexer for TE (a)<br />
and TM <strong>polarisation</strong> (b).<br />
Second experiment: MMI-MZI demultiplexer with U-bends<br />
For the second experiment a MMI-MZI demultiplexer with U-bends, as depicted in figure 6.5b,<br />
has been produced in a deeply etched waveguide structure in order to facilitate the usage of<br />
extremely small bending radii, which are necessary for small device dimensions. The device<br />
Excess loss [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
-30<br />
1<br />
2<br />
3<br />
4
6.2 MMI-MZI demultiplexer 113<br />
was designed to operate at a central <strong>wavelength</strong> of 1520 nm with 4 nm channel spacing. The<br />
MMI-couplers are 21 μm wide and 664 μm long, with 3 μm wide input waveguides in<br />
order to relax production tolerances [134]. The U-bends have a radius of 80 μm and a width of<br />
1.4 μm, ensuring monomode operation. The total device size is μm2 (< 0.3 mm2 4 × 4<br />
2800 × 106<br />
!),<br />
which is the smallest demultiplexer reported so far.<br />
The measured response for TE <strong>polarisation</strong> is shown in figure 6.15a. Due to an increased width<br />
of ± 0.2 μm, the <strong>wavelength</strong> response is shifted towards a higher centre <strong>wavelength</strong>. The losses<br />
of the device are considerable (9-11 dB), partly due to problems in the mask, which resulted in<br />
rough waveguide edges. The losses of straight waveguides are 2.3-2.6 dB/cm for 3.0 μm wide<br />
waveguides, up to 4.1-4.9 dB/cm for the 1.4 μm narrow waveguides.<br />
Measurements of the U-bends yielded 0.6 dB loss per single U-bend, which, according to<br />
figure 6.13, cannot be the only reason for the poor cross talk performance (8-10 dB).<br />
Therefore, the amount of TM-polarised light at the output has also been measured, while<br />
exciting TE-polarised light at the input. The measurement results are shown in figure 6.15b. It<br />
can be clearly seen that the power of the TM-polarised output signals is in the same order of<br />
the TE-polarised output, which indicates that <strong>polarisation</strong> conversion up to 50% occurs in the<br />
U-bends.<br />
Excess loss [dB]<br />
Transmission [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
-30<br />
1530 1535 1540 1545 1550 1555 1560<br />
Wavelength [nm]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
3<br />
4<br />
2 1 3 4<br />
2<br />
1<br />
-30<br />
1530 1535 1540 1545 1550 1555 1560<br />
Wavelength [nm]<br />
(a) (b)<br />
1 3 4 2<br />
-30<br />
1510 1515 1520<br />
Wavelength [nm]<br />
1525 1530<br />
(c)<br />
Excess loss [dB]<br />
Excess loss [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
3<br />
-30<br />
1535 1540 1545<br />
Wavelength [nm]<br />
1550 1555<br />
(d)<br />
1<br />
2<br />
3<br />
4<br />
4 2 1 3<br />
Figure 6.15 Measured response of the second MMI-MZI demultiplexer for TE<br />
(a) and TM <strong>polarisation</strong> (b), when exciting TE <strong>polarisation</strong> at the input and<br />
simulated response for TE <strong>polarisation</strong>: without (c) and with (d) <strong>polarisation</strong><br />
conversion and U-bend loss taken into account.
114 6. MMI-<strong>based</strong> components<br />
Polarisation conversion in U-bends was unknown at the time of designing the demultiplexer. It<br />
has been measured using a number of test waveguides, which contain an increasing number of<br />
U-bends. The results are shown in figure 6.16. It can be seen that the amount of <strong>polarisation</strong><br />
conversion exceeds 50%, which is, if we only consider one <strong>polarisation</strong>, identical to an excess<br />
waveguide loss of 3 dB. Taking into account the <strong>polarisation</strong> conversion as shown in figure<br />
6.16, and a U-bend loss of 0.6 dB per U-bend, the response has been simulated and is shown in<br />
figure 6.15d. The predicted excess loss is 5.5 dB and the cross talk decreases to approximately<br />
-10 dB. This corresponds well with the measured response of figure 6.15a.<br />
Polarisation conversion [%]<br />
80<br />
60<br />
40<br />
20<br />
0<br />
0 1 2 3 4 5 6<br />
Number of U-bends<br />
Figure 6.16 Measured <strong>polarisation</strong> conversion in the U-bends.<br />
Third experiment: MMI-MZI demultiplxer with adapted U-bends<br />
For a third experiment we use the same configuration as for the previous experiment (see figure<br />
6.5b). However, the 1.4 μm narrow bends are replaced by bends with a width in the whispering<br />
gallery (WG) regime. In this way, <strong>polarisation</strong> conversion is avoided as discussed in section<br />
3.9. As deeply etched waveguides are used, the width and radius are chosen at 3 μm and 50 μm<br />
respectively, which will give negligibly low <strong>polarisation</strong> conversion.<br />
At the input and output of the MMI-couplers 3 μm wide waveguides are used in order to relax<br />
production tolerances. In the array they are tapered down over a length of 50 μm to a width of<br />
1.4 μm in order to ensure monomode operation. Application of such narrow waveguides leads<br />
to increased coupling losses at junctions to the broad curved waveguides. Additonally, the<br />
losses at a junction between two oppositely curved waveguides increase as well, due to the<br />
highly asymmetric modal field distribution in the broad waveguide bend. Tapers cannot be<br />
used here, as then the U-bends cannot be placed close to one other anymore, whereby avoiding<br />
the usage of this geometry.<br />
Figure 6.17a shows the calculated transmission at a junction from a 1.4 μm wide straight to a<br />
curved waveguide, for two widths of the bend. In figure 6.17b the calculated transmission<br />
between two oppositely curved waveguide is shown, from a 1.4 μm wide bend to both a<br />
1.4 μm and a 3.0 μm wide bend respectively. As the array arm with three U-bends (the largest<br />
number for a 4 ×<br />
4 demultiplexer, see figure 6.5b) contains two straight-curved junctions and<br />
six curved-curved junctions, the importance of a low junction loss becomes evident. When<br />
using narrow bends (1.4 μm), the total junction loss amounts to 0.4 dB, which is sufficiently
6.2 MMI-MZI demultiplexer 115<br />
low. In the case of broad bends (3.0 μm), the total of junction losses is approximately 3.2 dB<br />
and the loss per U-bend is therefore in the order of 1.1 dB excluding radiation losses. Referring<br />
to figure 6.13, the cross talk level is then expected to be -20 and -23 dB respectively, and the<br />
insertion loss will increase by 1.8 dB.<br />
Transmission [dB]<br />
0.0<br />
-0.5<br />
-1.0<br />
-1.5<br />
1.4-1.4 μm 1.4-3.0 μm<br />
-2.0<br />
-0.5 0.0 0.5 1.0<br />
Offset [μm]<br />
(a) (b)<br />
-2.0<br />
-0.5 0.0 0.5 1.0 1.5 2.0<br />
Offset [μm]<br />
The performance has been calculated for this MMI-MZI demultiplexer and is shown in figure<br />
6.18a. The insertion loss of the demultiplexer is approximately 1.8 dB, which is an increase of<br />
1.5 dB with respect to the ideal case, i.e. the U-bends are considered to be without loss (see<br />
figure 6.6). The calculated cross talk level lies between -20 and -25 dB, which corresponds<br />
well with the predicted values of -20 and -23 dB (figure 6.13).<br />
Transmission [dB]<br />
The losses at a junction between two oppositely curved waveguides should be minimised, as<br />
they mainly determine the total junction losses for the longest array arm. This can be done by<br />
inserting a short bi-modal waveguide section, as shown schematically in figure 6.19. The<br />
principle is explained as follows. The high junction loss is mainly due to the asymmetry of the<br />
waveguide mode in the bend, which is inherent to WG modes. At the transition from the WG<br />
Transmission [dB]<br />
0.0<br />
-0.5<br />
-1.0<br />
-1.5<br />
1.4-1.4 μm 1.4-3.0 μm<br />
Figure 6.17 Calculated transmission between a straight to a curved (a), and<br />
between two oppositely curved waveguides (b), for two narrow waveguides (solid<br />
lines), and for a 1.4 μm and a 3.0 μm wide waveguide (dashed lines).<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
1<br />
3 1<br />
4<br />
2<br />
2 4<br />
-30<br />
1520 1525 1530 1535 1540 1545 1550<br />
Wavelength [nm]<br />
-30<br />
1520 1525 1530 1535 1540 1545 1550<br />
Wavelength [nm]<br />
(a) (b)<br />
Transmission [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
1<br />
4 3<br />
2 1 2 4<br />
Figure 6.18 Calculated response of the third MMI-MZI demultiplexer without<br />
(a) and with (b) U-bends with bimodal sections for TE <strong>polarisation</strong>.<br />
3
116 6. MMI-<strong>based</strong> components<br />
bend to a straight waveguide not only the fundamental mode is excited, but also the first-order<br />
mode. If the length of the straight waveguide is chosen in such a way that at the end the modes<br />
are shifted π radians in phase with respect to each other, a mirrored replica of the input field is<br />
obtained, which will couple with low loss to the field in the oppositely curved waveguide. In<br />
this way, the total U-bend loss can be decreased to a value in the order of 0.4 dB (instead of the<br />
expected 0.8 dB for two junctions, see figure 6.17b), if the transition section has a length of<br />
approximately 11 μm. Additionally, due to the short length of the inserted waveguide, the<br />
dimensions of the U-bend are hardly influenced.<br />
Figure 6.19 Schematic diagram of the bi-modal transition waveguide for the<br />
junction loss reduction between two oppositely curved waveguides.<br />
In figure 6.18b the calculated response for a device with bi-modal sections in the U-bends is<br />
shown. The U-bend loss is approximately 0.4 dB, leading to a predicted increase of the<br />
insertion loss of 0.6 dB (see figure 6.13). The calculated response shows an insertion loss of<br />
1.1 dB, which is 0.8 dB higher than for the ideal case, whereas an increase of 0.6 dB was<br />
predicted. The cross talk is expected to be better than -30 dB, which can be verified from the<br />
figure.<br />
A MMI-MZI demultiplexer with optimised U-bends has been produced. The measured<br />
response of a device without AR-coating is shown in figure 6.20a and b, for TE and TM<br />
<strong>polarisation</strong> respectively. Using a <strong>polarisation</strong> filter, it was verified that no <strong>polarisation</strong><br />
conversion occurred in the array arms. The on-chip losses are 3.5-5.5 dB for TE <strong>polarisation</strong><br />
and 2.5-4.5 dB for TM <strong>polarisation</strong>, which include almost 1.0 dB propagation loss. The cross<br />
talk was measured to be better than -7.5 dB (-11 dB best case) for TE <strong>polarisation</strong> and better<br />
than -9 dB (-14 dB best case) for TM <strong>polarisation</strong>. This may be due to width deviations leading<br />
to errors in the phase transfer. A TE-TM shift of 1.9 nm was observed.<br />
=<br />
=<br />
+<br />
+
6.3 Spatial mode filter 117<br />
On-chip loss [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
3<br />
1 2 4 3 1<br />
-30<br />
1525 1530 1535 1540 1545 1550 1555<br />
Wavelength [nm]<br />
6.2.4 Conclusion<br />
In this section a novel type of phased-array demultiplexer has been presented and a detailed<br />
operation and tolerance analysis has been given. It can be concluded that, although MMI<br />
couplers are commonly considered to be tolerant devices, the MMI-<strong>based</strong> demultiplexer is not.<br />
It is sensitive to width variations of the MMI coupler, which makes it difficult to match the<br />
<strong>wavelength</strong> response of a device to fixed system <strong>wavelength</strong>s. Additionally, the application of<br />
U-bends gives rise to power imbalance in the array arms, which results in increased cross talk<br />
levels. Therefore, an improved design may be found in the layout as depicted in figure 6.5c<br />
[55,76]. This design has the advantage of all array arms having the same number of junctions<br />
and bends, but, on the other hand, the device size being of larger dimension.<br />
In contrast with conventional phased-array <strong>demultiplexers</strong>, the (narrow) bandwidth of the low<br />
cross talk window cannot be enlarged without changing the channel spacing, which makes the<br />
device less attractive for use in WDM systems. However, this narrow bandwidth is no problem<br />
for a multi-<strong>wavelength</strong> laser as suggested by Herben et al. [55].<br />
Summarising, we can conclude, that for the MMI-MZI <strong>demultiplexers</strong> to become a success,<br />
the production process has to be accurately controlled, as they are very sensitive to width<br />
deviations (which increase with increasing MMI-width) and to array guide loss imbalance.<br />
Their application may therefore be restricted to low numbers of channels (up to 4).<br />
6.3 Spatial mode filter<br />
2<br />
-30<br />
1525 1530 1535 1540 1545 1550 1555<br />
Wavelength [nm]<br />
(a) (b)<br />
Devices are mainly designed to operate optimally for the fundamental waveguide mode. Firstorder<br />
waveguide modes will disturb the device operation (they cause, for instance, the “ghost”<br />
images in a phased-array demultiplexer as discussed in section 4.4) and therefore excitation of<br />
these modes should be avoided. Especially for the fibre-chip coupling it is difficult to maintain<br />
low first-order excitation over a long period of time. Therefore, a - preferably compact - spatial<br />
mode filter is desired. Mode filtering can be done by an adiabatic asymmetric Y-junction [16],<br />
which is, however, difficult to produce due to the stringent demands on the lithographic<br />
resolution necessary to obtain a sharp intersection angle. Additionally, this device is relatively<br />
On-chip loss [dB]<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-25<br />
3<br />
1<br />
2 4 3 1 2<br />
Figure 6.20 Measured response of the third MMI-MZI demultiplexer for TE (a)<br />
and TM <strong>polarisation</strong> (b).
118 6. MMI-<strong>based</strong> components<br />
long. A compact solution to this problem is the application of a single MMI coupler, as<br />
developed simultaneously both at TU Delft and ETH Zurich <strong>independent</strong>ly of each other.<br />
Leuthold et al. [74] published the first device, which was produced in a low-contrast<br />
waveguide structure. The device measured 340 μm by 12 μm, had a first-order mode rejection 1<br />
of 18 dB and an excess loss of 0.2 dB. Shortly thereafter it was followed by the device<br />
described in this thesis [36], which is more compact and has a higher first-order mode<br />
rejection, due to the fact that it was produced in a high-contrast waveguide structure.<br />
6.3.1 Operation principle<br />
The principle of such a MMI-<strong>based</strong> spatial mode filter lies in the fact that symmetric and<br />
asymmetric modes applied at the input, are mapped differently onto output waveguides as<br />
shown schematically in figure 6.21. The operation principle can be explained as follows.<br />
According to theory [13,14,121], using a centre-fed MMI coupler N1-fold images of the input<br />
field are obtained at LMMI,1 = 3Lπ ⁄ 4N 1 . This length equals the length of a N 2 × N 2 power<br />
splitter-combiner LMMI,2 = 3Lπ ⁄ N 2 with N1 = 1 and N2 = 4. The result is that, if a zero-order<br />
mode is applied at the input waveguide, it will be imaged at the centre output waveguide.<br />
For first-order mode excitation, we consider the input waveguide as a combination of two<br />
waveguides that touch each other, but neither of them lies at the centre of the MMI coupler.<br />
The first-order field is then considered as the sum of two separate zero-order fields, which are<br />
in counter phase, i.e. one of which has an additional phase term . From table 6.1 (columns<br />
with i = 2 and i = 3, inputs i = 1 and i = 4 are not used) it can be directly seen that constructive<br />
interference will occur into the outer output waveguides 1 and 4, and destructive interference<br />
into the centre output waveguides 2 and 3. Due to the balanced power distribution over both<br />
input fields, which is inherent to first-order fields, there is no residual power coupled into the<br />
centre output waveguides, as long as the length and width are made as designed. Simulation<br />
results obtained with a standard 2D Beam Propagation Method (BPM) analysis show the<br />
operation of the spatial mode filter and are depicted in figure 6.22.<br />
W/2<br />
L MMI<br />
e jπ –<br />
j = 1<br />
j = 2<br />
j = 3<br />
j = 4<br />
Figure 6.21 Schematic diagram of a MMI-<strong>based</strong> spatial mode filter, including<br />
mapping characteristics for zero-order mode (filled) and first-order mode (open).<br />
1. The first-order mode rejection is defined as the amount of power emerging at the centre output<br />
waveguide when a first-order mode is applied at the input waveguide.<br />
W
6.3 Spatial mode filter 119<br />
6.3.2 Tolerance analysis<br />
(a) (b)<br />
Figure 6.22 Simulated filter operation obtained with BPM analysis: zero-order<br />
mode excitation (a) and first-order mode excitation (b).<br />
As the mode filters are specially designed for usage in the raised-strip waveguide <strong>based</strong><br />
phased-array demultiplexer, the waveguide structure as depicted in figure 4.3 is used for the<br />
tolerance analysis. The guiding strip is 2.2 μm thick and 3.0 μm wide in order to obtain zero<br />
birefringence. The width of the outer waveguides is chosen at 2.0 μm for optimum coupling<br />
efficiency. In order to obtain the smallest possible MMI coupler, the gaps between the<br />
waveguides were chosen to be equal to the lithographic resolution of 0.6 μm. In order to obtain<br />
low insertion loss, the width of the MMI section was chosen at 8 μm, resulting in a small offset<br />
of the two outer output waveguides. By calculating the propagation constants of the<br />
fundamental and first-order mode in the MMI section, the length of the section is then 134 μm.<br />
Bandwidth of first-order mode rejection and insertion loss<br />
As the mode filter consists of a single MMI coupler, the bandwidth equals the range 2 δλ centred<br />
around the design <strong>wavelength</strong> λc (see equation 6.9). The waveguide structure is extremely<br />
good guiding and the effective mode width equals approximately the waveguide width. Equation<br />
6.9 can then be rewritten as:<br />
2 δλ Z( α)<br />
Nπd 2<br />
0<br />
≅ Z( α)<br />
2w2<br />
≈ (6.20)<br />
-----------------λ<br />
2 c<br />
4W MMI<br />
--------------λ<br />
2 c<br />
W MMI<br />
whereby N = 4 has been used (as it operates as a 4 ×<br />
4 coupler) and w is the waveguide width<br />
of the narrow waveguides. Filling in the values, the 1-dB bandwidth follows as λc /10, which<br />
demonstrates the potentially wide application range of the device.<br />
Figure 6.23 shows the response of the mode filter versus the <strong>wavelength</strong> deviation Δλ from the<br />
design <strong>wavelength</strong> of 1535 nm. The symbols 0_0 and 0_1 (or 1_0 and 1_1) denote the<br />
transmission to the centre output and to one of the outer waveguides respectively, when the<br />
fundamental (or first-order) mode is applied at the input. It can be seen that a first-order mode
120 6. MMI-<strong>based</strong> components<br />
rejection (denoted by 1_0) better than -30 dB is obtained over a wide <strong>wavelength</strong> range. It is<br />
noted that due to the fact that the power of a first-order mode applied at the input is equally<br />
divided over the two outer waveguides at the output, the -3 dB level denotes zero excess loss.<br />
Transmission [dB]<br />
0<br />
-1<br />
-2<br />
-3<br />
-4<br />
-5<br />
-125<br />
-50 -40 -30 -20 -10 0 10 20 30 40 50<br />
Δλ [nm]<br />
Width tolerance of first-order mode rejection and insertion loss<br />
Analogous to the <strong>wavelength</strong> tolerance, the expression for the width is rewritten as follows:<br />
The 1-dB width tolerance calculated using this equation equals 0.45 μm, so for this device a<br />
lateral dimensional control of ± 0.2 μm, which is a realistic value for a standard lithographic<br />
process, is sufficient for low-loss operation. In addition, if the width is controlled within<br />
± 0.2 μm of the design value, the first-order mode rejection remains below -30 dB. Figure 6.24<br />
shows the response versus the deviation ΔW of the width with respect to the designed value.<br />
0_0<br />
1_1<br />
0_1<br />
1_0<br />
0<br />
-25<br />
-50<br />
-75<br />
-100<br />
Figure 6.23 Response of the mode filter versus <strong>wavelength</strong> variation.<br />
Transmission [dB]<br />
0<br />
-2<br />
-4<br />
-6<br />
-8<br />
2<br />
de Suppression Rejection ratio [dB]<br />
δW ≤ Z( α)<br />
-----------------<br />
(6.21)<br />
2W MMI<br />
-25<br />
-50 [dB]<br />
0_0<br />
1_1<br />
0_1 -75<br />
1_0 Rejection<br />
-100<br />
-10<br />
-125<br />
-1.0 -0.5 0.0<br />
ΔW [μm]<br />
0.5 1.0<br />
Figure 6.24 Response of the mode filter versus width variation.<br />
0<br />
Suppression ratio [dB]
6.3 Spatial mode filter 121<br />
Layer thickness tolerance of first-order mode rejection and insertion loss<br />
The device is very tolerant to a variation of the guiding layer thickness. When the layer<br />
thickness is varied within ± 0.2 μm, the transverse index varies within ± 0.1 %. This can be<br />
converted to a deviation of the length using equation 6.8, which leads to a value in the order of<br />
± 0.1 μm.<br />
6.3.3 Experimental results<br />
The spatial mode filter as described in the previous section has been produced in cooperation<br />
with Philips Optoelectronics Centre. The centre input and output waveguide is 3.0 μm wide<br />
and the outer output waveguides are 2.0 μm wide. The MMI coupler with dimensions of<br />
134 × 8<br />
μm 2 , is the smallest reported so far [36]. The measured response is shown in figure<br />
6.25. In order to be able to excite the zero-order mode as well as the first-order mode in a<br />
controllable manner, a lateral offset was applied in the input waveguide. This offset introduces<br />
an additional excess loss, which is depicted as the dashed line in the graph. The excess loss of<br />
the filter was measured to be 0.5 ± 0.2 dB, which is slightly higher than expected. The firstorder<br />
mode rejection is better than -30 dB, which corresponds well with predictions.<br />
Rejection ratio [dB]<br />
-20<br />
-25<br />
-30<br />
-35<br />
6.3.4 Conclusion<br />
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4<br />
Offset [μm]<br />
Figure 6.25 Measured filter response: first-order rejection ratio (triangles) and<br />
excess loss (squares) as a function of the offset applied at the input waveguide.<br />
The lines denote the simulated rejection ratio (solid) and excess loss (dashed).<br />
An analysis has been presented of a mode filter <strong>based</strong> on a single MMI coupler, which is easy<br />
to produce as only one tolerant etching step is required. Calculations clearly show the<br />
advantages of the device: compact size, low-loss operation, high first-order mode rejection,<br />
large optical bandwidth and tolerance to width variations. The device can therefore be used in<br />
a wide variety of applications. Experiments demonstrate the low loss operation (0.5 dB) and<br />
the high first-order mode rejection (better than -30 dB).<br />
0<br />
-1<br />
-2<br />
-3<br />
-4<br />
-5<br />
Excess loss [dB]
122 6. MMI-<strong>based</strong> components<br />
6.4 Optical imaging of MMI patterns<br />
Accurate methods for measuring two-dimensional intensity patterns in integrated optical<br />
devices are not available at present. Techniques which image the light scattered at<br />
inhomogeneities in the waveguide layer, suffer from the large local inhomogeneity of the<br />
scattering sources. Another method is positioning identical devices at different distances<br />
relative to a cleaved end-face, as shown schematically in figure 6.26. This method only<br />
provides us with one-dimensional intensity scans at different positions in the devices, provided<br />
that the excitation conditions can be reproduced. In this section we present a method of<br />
imaging the field patterns in integrated optical devices using two-step cooperative<br />
upconversion in erbium ions incorporated in the waveguides.<br />
Cooperative upconversion is a process whereby two excited Er 3+ ions exchange energy,<br />
promoting one of them to a higher energy level [22]. Two sequential upconversion processes<br />
lead to emission of green light (519 and 545 nm). As the process depends on the concentration<br />
of excited Er 3+ ions, which in turn depends on the intensity of the field distribution, the<br />
emission of green light is a (roughly fourth-power) replica of the intensity distribution in the<br />
waveguide. This work has been carried out in cooperation with the FOM-Institute for Atomic<br />
and Molecular Physics, and has been presented earlier [30,59].<br />
Figure 6.26 Schematic diagram of positioning identical devices at different<br />
distances relative to a cleaved end face.<br />
6.4.1 Upconversion mechanism<br />
Figure 6.27 shows the energy level diagram of Er 3+ and a schematic graph of how cooperative<br />
upconversion proceeds. The various energy levels of the Er 3+ ion are numbered 0 to 6. Figure<br />
6.27a illustrates the first-order process between two Er 3+ ions in the first excited state, whereby<br />
one of them transfers its energy to the other, promoting the latter to the third excited state. This<br />
process depends quadratically on the concentration of Er 3+ in the first excited state, as two ions<br />
are involved. The ions in the third excited state decay rapidly and non-radiatively to the second<br />
excited state. As the lifetime of level 2 is relatively long (0.25 ms) a significant population in<br />
the second excited state is built up.<br />
Subsequently, a second upconversion process can take place (figure 6.27b), whereby two ions<br />
in the second excited state interact in a similar fashion, thereby populating the sixth excited<br />
state. Radiative decay from this state to the ground state causes emission at 519 nm. In<br />
addition, non-radiative decay can occur to the fifth excited state; radiative decay of this state to<br />
the ground state causes emission at 545 nm. As two subsequent upconversion steps are
6.4 Optical imaging of MMI patterns 123<br />
involved, the emission of this visible green light is roughly proportional to the fourth power of<br />
the concentration of Er 3+ in the first excited state, which, in turn, linearly depends on the<br />
intensity of the infrared light. It thus becomes possible to image the intensity distribution of<br />
1.48 μm pump light directly by monitoring the green emission. The measurement resolution is<br />
then limited by the diffraction limit for 519 and 545 nm light. Therefore, field intensities can<br />
be determined with a measuring resolution of approximately 3 times (!) better than if the pump<br />
light, propagating in the waveguide, were imaged directly.<br />
Energy [eV]<br />
2.4<br />
2.3<br />
1.9<br />
1.6<br />
1.3<br />
0.8<br />
520<br />
545<br />
660<br />
798<br />
979<br />
1530<br />
Wavelength [nm]<br />
⇒<br />
0 0<br />
6.4.2 Measurement principle<br />
Figure 6.28 shows the measurement set-up, where a simple microscope objective placed above<br />
the waveguide is used for imaging the green light intensity distribution as the 1485 nm light is<br />
guided through the waveguide.<br />
⇒<br />
(a) (b)<br />
Figure 6.27 Energy level diagram of Er 3+ showing cooperative upconversion.<br />
SiO 2<br />
Al 2 O 3<br />
SiO 2 substrate<br />
Figure 6.28 Principle of the measurement setup.<br />
microscope objective<br />
0.3 μm<br />
0.6 μm<br />
6<br />
5<br />
4<br />
3<br />
2<br />
1
124 6. MMI-<strong>based</strong> components<br />
Accurate measurements can be made by digitizing the output on a CCD camera. The fourth<br />
power dependence between the green and the infrared light is favourable for attaining high<br />
measurement accuracy. The short <strong>wavelength</strong> of the upconverted light makes it possible to<br />
measure with a resolution considerably below the diffraction limit for the infrared light, which<br />
allows for high accuracy imaging with medium quality objectives.<br />
6.4.3 MMI-coupler simulation<br />
Self-imaging is a property of multimode waveguides, which means that an input field is<br />
reproduced in single or multiple images at periodic intervals in the MMI-section as shown in<br />
figure 6.29. As discussed earlier, at distances L = LMMI ⁄ N an N-fold image of the input field<br />
is obtained if the length of the MMI-coupler is chosen according to LMMI = 3Lπ ⁄ 4 .<br />
LMMI ------------<br />
5<br />
LMMI ------------<br />
3<br />
LMMI<br />
------------<br />
2<br />
L MMI<br />
Figure 6.29 Schematic diagram of the self-imaging principle of a MMI-coupler.<br />
Figure 6.30a shows the result of a simulation performed with the Beam Propagation Method<br />
(BPM), using a 21 μm wide MMI-section centre-fed by a 2 μm wide input waveguide. The<br />
light coming out of the input waveguide is seen to be diverging in the MMI-section and being<br />
reflected by the side walls. The reflections cause the interference patterns which produce at<br />
certain distances the single and multiple images typical for MMI devices.<br />
(a)<br />
(b)<br />
Figure 6.30 Multimode interference: calculated patterns using BPM (a), and a<br />
microscope image, visible as green light (b).
6.5 Discussion 125<br />
6.4.4 Experiment<br />
A MMI-coupler was produced in an aluminium oxide ridge waveguide structure [117],<br />
implanted with erbium [57] to a peak concentration of 1.3 at.%. The device consists of a 2 μm<br />
wide input waveguide, centre-fed to a 21 μm broad MMI-section. Light from a 1485 nm high<br />
power laser was coupled into the input waveguide using a fibre taper. The power in the<br />
waveguide was approximately 4 mW, calculated by substracting the estimated fibre-chip<br />
coupling loss from the source power. Figure 6.30b shows the microscope image of the<br />
multimode interference pattern, directly after the transition from the input waveguide to the<br />
MMI-section. This image appears as green light due to a two-step cooperative upconversion<br />
process as explained earlier. As can be seen, the measured intensity profile corresponds very<br />
well with the calculated profile in figure 6.30a.<br />
6.4.5 Conclusion<br />
For the first time the optical interference pattern in multimode interference couplers operating<br />
at 1485 nm has been made visible using an optical microscope. This is done by imaging the<br />
green light (519 and 545 nm), which is generated by upconversion at high pumping power<br />
levels in waveguides with a high erbium concentration. As the intensity of the green light is<br />
roughly proportional to the fourth power of the pump signal intensity, it becomes possible to<br />
image the 1485 nm intensity distribution with a resolution limited by the diffraction limit of<br />
green light. For the 545 nm green light used here, this means that the measuring resolution is<br />
approximately 3 times (!) better than if the pump light propagating in the waveguide was<br />
imaged directly. This method is demonstrated with a MMI-coupler produced in an aluminium<br />
oxide ridge waveguide structure implanted with erbium. The calculated intensity profiles<br />
correspond very well with the measured data. This method offers an unique opportunity for<br />
accurate measurements of lateral field patterns.<br />
6.5 Discussion<br />
This Chapter was dedicated to novel applications of MMI couplers. First a phased-array<br />
demultiplexer <strong>based</strong> on MMI couplers has been presented and a detailed operation together<br />
with a tolerance analysis has been given. Although MMI couplers are commonly considered to<br />
be tolerant devices the MMI-<strong>based</strong> <strong>demultiplexers</strong> with U-bends have very stringent demands<br />
on the production process. An improved design may be found in a less compact design with<br />
identically shaped array arms (as depicted in figure 6.5c). In this way production tolerances are<br />
relaxed, but even then variations in array arm losses have a rather large impact on the cross talk<br />
performance. As the requirements on the width become more stringent with increasing width,<br />
the width should be kept as narrow as possible, limiting the number of channels. Successful<br />
application of MMI-<strong>based</strong> <strong>demultiplexers</strong> is therefore only expected for small numbers of<br />
channels (up to 4).<br />
Secondly, a mode filter <strong>based</strong> on a single MMI coupler has been discussed. The device is easy<br />
to produce, as only one tolerant etching step is required. Because of its small size, it combines<br />
a high first-order mode rejection with a large optical bandwidth and is tolerant to width<br />
variations. It can therefore be used in a wide variety of applications.<br />
Finally, the optical interference pattern in MMI couplers operating at 1485 nm is made visible<br />
for the first time using an optical microscope. This is done by imaging the green light, which is<br />
generated by upconversion at high pumping power levels in aluminium oxide ridge waveguides
126 6. MMI-<strong>based</strong> components<br />
with a high erbium concentration. The calculated interference correspond very well with the<br />
measured patterns and this method offers therefore an unique opportunity for accurate<br />
measurements of lateral field patterns.
Chapter 7<br />
Discussion and conclusions<br />
In this thesis the design and production of <strong>polarisation</strong> <strong>independent</strong> PHASAR <strong>demultiplexers</strong><br />
has been described. The operation principle of PHASAR <strong>demultiplexers</strong> has been analysed and<br />
elaborate attention has been paid to a number of different design requirements, such as<br />
<strong>polarisation</strong> independence, flattened <strong>wavelength</strong> response, low-loss operation, etc. As a matter<br />
of fact, in this way a sort of “PHASAR demultiplexer design manual” has been obtained.<br />
Furthermore, an overview of the most important applications has been given.<br />
Polarisation handling is an important issue for <strong>wavelength</strong> <strong>demultiplexers</strong> and therefore this<br />
subject has been dealt with in a comprehensive way. A novel type of <strong>polarisation</strong> converter,<br />
<strong>based</strong> on waveguide bends with ultra-small radii fabricated in a deeply etched waveguide structure,<br />
has been discussed. Simulations and experiments show that high <strong>polarisation</strong> conversion<br />
can be obtained. On the other hand, the device is sensitive to variations of the width and side<br />
wall angle of the waveguide. Another important result of the analysis of deeply-etched<br />
waveguide bends is, that by employing broad waveguide bends, <strong>polarisation</strong> conversion can be<br />
minimised. As the side wall angle of the waveguide can be reproduced by using specific<br />
settings of the etching process, a more detailed analysis of the effect of the side wall on<br />
<strong>polarisation</strong> conversion may therefore be a suggestion for future work. Another suggestion<br />
could be the insertion of a <strong>polarisation</strong> converter <strong>based</strong> on ultra-short bends in the middle of a<br />
PHASAR demultiplexer.<br />
Several methods to make <strong>demultiplexers</strong> <strong>polarisation</strong> <strong>independent</strong> are discussed in this thesis.<br />
An extensive analysis of the <strong>polarisation</strong> <strong>independent</strong> raised-strip waveguide has been<br />
presented. Important advantages of this type of waveguide are: low fibre-chip coupling losses,<br />
compact device dimensions and broadband <strong>polarisation</strong> independence. A method for reducing<br />
the insertion loss has been introduced and demonstrated, and the occurence of so-called<br />
“ghost” images has been investigated. Finally, a packaged device integrated with<br />
photodetectors has been described. This is the first compact PHASAR receiver in an industrystandard<br />
Butterfly package.<br />
Other methods for obtaining <strong>polarisation</strong> independence have also been described. The<br />
compensation of the <strong>polarisation</strong> dispersion of the array waveguides, for instance, is discussed<br />
extensively, including a tolerance analysis. This method has been applied to our waveguide<br />
structure and experimental results are presented. The production is relatively simple, as a<br />
selective wet-chemical etchant can be used for the <strong>polarisation</strong> dispersion compensating
128 7. Discussion and conclusions<br />
section in the PHASAR. Another suggestion for future work is the analysis of the <strong>polarisation</strong><br />
dependent insertion loss of the device, which was obtained using this method.<br />
Additionally, two methods which use <strong>polarisation</strong> conversion and splitting respectively, are<br />
introduced and discussed with emphasis on cross talk tolerance, as no experimental results are<br />
available. Obviously, future work is suggested on the application of these methods.<br />
Novel components <strong>based</strong> on Multimode Interference (MMI) are also presented. The design of<br />
a MMI-<strong>based</strong> demultiplexer is being discussed, together with a tolerance analysis with<br />
experimental results being presented. It is found that the usage of this type of <strong>demultiplexers</strong> is<br />
restricted to a low number of channels (up to 4). In addition, the bandwidth of the low cross<br />
talk window is rather small, which, however, is not a problem for a multi-<strong>wavelength</strong> laser.<br />
Furthermore, a novel type of spatial mode filter <strong>based</strong> on a single MMI coupler is discussed.<br />
The discussion and tolerance analysis show the advantages of this device: compact size, large<br />
optical bandwidth and tolerance to width variations. Experiments demonstrate the low-loss<br />
operation combined with a high first-order mode rejection.<br />
Finally, the interference patterns of the infrared pump light in a MMI coupler with a high<br />
erbium concentration are made visible by imaging the green light, generated by upconversion<br />
at high pumping power levels. In this way, the field intensity of the infrared light can be imaged<br />
with a resolution below the theoretical diffraction limit (almost three times!). This is demonstrated<br />
with a MMI-coupler produced in an aluminum oxide ridge waveguide structure<br />
implanted with erbium.
Appendix A<br />
Overview of phased-array<br />
<strong>demultiplexers</strong> produced<br />
In this appendix an overview is given of phased-array <strong>demultiplexers</strong> produced and some key<br />
parameters. For the laboratories (third column) the following abbreviations are used:<br />
Akzo Akzo Nobel Central Research, Arnhem, The Netherlands<br />
Alcatel Alcatel Alstholm Recherche, Marcoussis, France<br />
Lucent Lucent Technologies (f.k.a. AT&T Bell Laboratories), Holmdel, USA<br />
Bellcore Bellcore, Red Bank, USA<br />
BNR BNR Europe Limited, Harlow, United Kingdom<br />
CNET France Telecom CNET, Bagneux, France<br />
DUT Delft University of Technology, Delft, The Netherlands<br />
Maryland University of Maryland, Baltimore, USA<br />
NTT Nippon Telegraph and Telephone Corporation Optoelectronics<br />
Laboratories, Ibaraki, Japan<br />
OKI OKI Electric Industry Corporation, Tokyo, Japan<br />
POC Philips Optoelectronics Centre, Eindhoven, The Netherlands<br />
Siemens Siemens A.G. Research Laboratories, Munich, Germany<br />
Furthermore, the following symbols are used:<br />
N Number of channels<br />
Δf Channel spacing<br />
CT Cross talk<br />
Δλpol TE-TM shift<br />
The loss is defined as the total loss of the device including the chip coupling losses. However,<br />
these losses are not always mentioned in the publications and therefore the superscript<br />
notations (*) and (**) are used to denote the on-chip loss and the excess loss measured relative<br />
to a straight waveguide respectively. In order to obtain a <strong>polarisation</strong> <strong>independent</strong> device<br />
(Δλ pol = 0), a number of methods can be used (see section 2.2.4). These methods are denoted<br />
by the following superscript numbers: (1) order trick, (2) λ/2 plate, (3) <strong>polarisation</strong> dispersion<br />
compensation and (4) <strong>polarisation</strong> <strong>independent</strong> waveguides.
130 Appendix A. Overview of phased-array <strong>demultiplexers</strong> produced<br />
Table A.1 Overview of PHASARs produced and some key parameters.<br />
Ref. Year Laboratory Material N<br />
Δf<br />
[GHz]<br />
Loss<br />
[dB]<br />
CT<br />
[dB]<br />
size<br />
[mm 2 ]<br />
162 89 DUT Al 2 O 3 4 213 0.6-3.2**
Table A.1 Overview of PHASARs produced and some key parameters.<br />
Ref. Year Laboratory Material N<br />
19 95 Alcatel <strong>InP</strong> 4 250 5-6*
132 Appendix A. Overview of phased-array <strong>demultiplexers</strong> produced
Appendix B<br />
Dispersion of the phased array<br />
The phased array is designed at a particular (freely chosen) central <strong>wavelength</strong> λ c in vacuo in<br />
such a way that the optical path length difference ΔL between adjacent array waveguides is<br />
equal to an integer number of <strong>wavelength</strong>s measured inside the array waveguide:<br />
ΔL m λc = ⋅ --------<br />
whereby N eff is the effective index of the waveguide mode. From figure 2.1b it can be seen that<br />
the tilt angle dΘ results from a phase shift dΦ between adjacent waveguides and can be<br />
described as:<br />
β FPR,λc<br />
dΘ<br />
=<br />
whereby is the propagation constant in the Free Propagation Regio (FPR) at the central<br />
<strong>wavelength</strong>:<br />
N FPR,λc<br />
with being the effective index in the FPR calculated at the central <strong>wavelength</strong>.<br />
The dispersion of the array is described as the displacement ds of the focal spot in the image<br />
plane (see figure 2.1b) per unit frequency change:<br />
N eff<br />
dΦ ⁄ β dΦ ⁄ β FPR,λc<br />
FPR,λc<br />
asin⎛<br />
---------------------------- ⎞ ≈ ----------------------------<br />
⎝ ⎠<br />
β FPR,λc<br />
d a<br />
=<br />
N FPR,λc<br />
R a<br />
2π<br />
⋅ ----λc<br />
d a<br />
(B.1)<br />
(B.2)<br />
(B.3)<br />
ds dΘ<br />
----- R<br />
d f a ⋅ ------- --------------------d<br />
f daβ FPR,λc<br />
dΦ<br />
= = ⋅ -------<br />
(B.4)<br />
d f
134 Appendix B. Dispersion of the phased array<br />
with:<br />
dΦ<br />
-----d<br />
f<br />
whereby N g is the group index of the waveguide.<br />
Combining these two equations leads to:<br />
whereby Δα = da ⁄ Ra is the divergence angle between the array waveguides in the fan-in and<br />
fan-out sections.<br />
Finally, we find for the dispersion:<br />
If we rearrange this equation we find:<br />
with m' being defined as:<br />
d----d<br />
f<br />
2πf ⎛-------- ⋅ N<br />
c eff ⋅ ΔL⎞<br />
2π<br />
----- N<br />
⎝ ⎠ c eff f dN ⎛ eff<br />
+ ------------ ⎞ 2π<br />
= = ⋅ ⋅ ΔL = ----- ⋅ N<br />
⎝ d f ⎠ c g ⋅ ΔL (B.5)<br />
dy<br />
---d<br />
f<br />
λ c<br />
------------------------ Ra ----- 2π<br />
⋅ ⋅ ----- ⋅ N<br />
c g ⋅ ΔL ----------------- ΔL 1<br />
= = ⋅ ------- ⋅ ---- (B.6)<br />
Δα<br />
2πN FPR,λc<br />
Δy<br />
Δy<br />
In this equation m' is the order of the demultiplexer, for which the material and waveguide<br />
dispersion have been taken into account in contrast with the fixed (integer) order m of the<br />
array. It is also noted that the term N eff ⁄ N in equation B.8 accounts for the transition<br />
FPR,λc<br />
from the array waveguides to the FPR.<br />
From equation B.5 it an be seen that the response of the phased array is periodical. After each<br />
phase change of 2π, the field will be imaged at the same position Δy. This period in the<br />
frequency domain is called the Free Spectral Range (FSR) and is shown in figure 2.2. It is<br />
found as the frequency shift for which the phase shift ΔΦ equals 2π, from which we find:<br />
By substituting B.1 equation, we end up with:<br />
d a<br />
N g<br />
----------------- ΔL<br />
-------<br />
Δα<br />
Δf<br />
= ⋅ ⋅ -----<br />
N eff<br />
N FPR,λc<br />
N FPR,λc<br />
f c<br />
f c<br />
N g<br />
N FPR,λc<br />
λ c<br />
1<br />
----------------- -------<br />
Δα<br />
Δf<br />
= ⋅ ⋅ ----- ⋅ m' ⋅ --------<br />
f<br />
m' m 1 -------- dN eff<br />
= ⋅ ⎛ + ⋅ ------------ ⎞<br />
⎝ d f ⎠<br />
N eff<br />
2π<br />
----- ⋅ Δ f<br />
c FSR ⋅ N g ⋅ ΔL = 2π<br />
Δ f FSR<br />
=<br />
f c<br />
---m'<br />
N eff<br />
f c<br />
(B.7)<br />
(B.8)<br />
(B.9)<br />
(B.10)<br />
(B.11)
Appendix C<br />
Waveguide mode effective width<br />
The diffraction properties of the phased array are conveniently expressed in terms of the<br />
effective mode width w e , defined as the width of a uniform intensity distribution with the same<br />
maximum intensity and power content as the modal field:<br />
w e<br />
∫<br />
+∞<br />
E( y)<br />
2 dy<br />
-∞<br />
= -----------------------------<br />
2<br />
Emax (C.1)<br />
Substitution of the expression for the (TE-polarised) modal field [160] yields the following<br />
expression for w e :<br />
w e<br />
=<br />
wwg -------- ⎛ 2<br />
1 + --⎞<br />
1<br />
⋅ ≈ w ⎛<br />
2 ⎝ v⎠<br />
wg 0.5 + ---------------- ⎞<br />
⎝ V – 0.6⎠<br />
(C.2)<br />
whereby wwg is the waveguide width and V and v are the normalised V-parameter and the<br />
normalised transverse attenuation constant respectively. The far right expression, which is<br />
found empirically by curve fitting in the range 1 < V < 10 , gives us a simple expression for<br />
estimating the effective width.<br />
Substitution of the Gaussian distribution E( y)<br />
Eo y into equation A.1 yields<br />
the following relation between we and wo :<br />
2 2<br />
= exp(<br />
– ⁄ wo) wo = we ⋅ 2 ⁄ π<br />
(C.3)<br />
Using equations A.2 and A.3 the modal field is easily “translated” into an equivalent Gaussian<br />
field.
136 Appendix C. Waveguide mode effective width
Appendix D<br />
Cross talk penalty of the tunable laser<br />
D.1 Measurement with a tunable laser<br />
For the measurement of the <strong>wavelength</strong> response of the <strong>demultiplexers</strong>, the set-up as depicted<br />
in figure D.1 is used. The HP 8168A tunable laser source is connected to a <strong>polarisation</strong><br />
controller, which in turn is connected to a lensed fibre, used for focusing the light onto the<br />
cleaved facet of the chip. The light is coupled into the waveguide and after propagating through<br />
the device, it reaches the other end of the chip. There, the light is being picked up by a microscope<br />
objective, which has its focal plane aligned with the facet and is focused onto a pinhole<br />
with an aditional lens. In this way, the pinhole acts as a spatial filter, so only the output of a single<br />
waveguide is detected by the germanium photo diode.<br />
HP 8168A<br />
<strong>polarisation</strong><br />
controller<br />
lensed<br />
fibre<br />
microscope<br />
objective<br />
Ge-diode<br />
If we now take a look at the response of the tunable laser when operating at a <strong>wavelength</strong> of<br />
1540 nm (see figure D.2), we can see that the side mode suppression ratio is below -40 dB (the<br />
output power of the laser is 300 μW, or -5.2 dBm), but the laser also has a broad spontaneous<br />
emission spectrum. When measuring the response of devices with a periodic <strong>wavelength</strong><br />
response, such as the phased-array demultiplexer, then not only the power of the peak<br />
<strong>wavelength</strong> is detected by the (<strong>wavelength</strong> inselective) Ge-diode, but also periodic parts of the<br />
spectrum. This is due to the fact that light of other diffraction orders also couple into the<br />
receiver waveguide of the device. As a consequence, especially the measurement of the cross<br />
talk is being influenced, as at a <strong>wavelength</strong> outside the pass-band of the device the signal is<br />
suppressed to a level of the same order of magnitude as the level of the spontaneous emission.<br />
chip<br />
Figure D.1 Schematic diagram of the measurement setup.<br />
lens<br />
pinhole
138 Appendix D. Cross talk penalty of the tunable laser<br />
At the transmission peak <strong>wavelength</strong> the signal level is hardly influenced, because it is a few<br />
orders of magnitude higher than the spontaneous emission.<br />
Response [dBm]<br />
0<br />
-20<br />
-40<br />
-60<br />
-80<br />
1400 1450 1500<br />
Wavelength [nm]<br />
1550 1600<br />
Figure D.2 Measured response of the HP 8168A tunable laser, operating at a<br />
<strong>wavelength</strong> of 1540 nm.<br />
The amount of spontaneous emission detected by the Ge-diode depends on the FSR of the<br />
measured demultiplexer and can be calculated by overlapping the response of the tunable laser<br />
with the periodic response of the demultiplexer and integrating over the complete <strong>wavelength</strong><br />
range. It is obvious that a large FSR leads to a low cross talk penalty, as then only a small part<br />
of the spontaneous emission spectrum is detected by the Ge-diode.<br />
D.2 Measurement with an EDFA as source<br />
For the measurement we now use the set-up as depicted in figure D.3. Instead of the tunable<br />
laser, the Amplified Spontaneous Emission (ASE) of an Erbium Doped Fibre Amplifier<br />
(EDFA) is used as source. As the light from the EDFA is circularly polarised, we now need a<br />
<strong>polarisation</strong> filter. The measured ASE spectrum is shown in figure D.4a. The microscope<br />
objective and the Ge-diode are replaced by a lensed fibre and a HP 71451A optical spectrum<br />
analyser respectively.<br />
EDFA<br />
cleaved<br />
fibre<br />
microscope<br />
objective<br />
<strong>polarisation</strong><br />
filter<br />
microscope<br />
objective<br />
chip<br />
lensed spectrum<br />
fibre analyser<br />
Figure D.3 Schematic diagram of the measurement setup with EDFA and<br />
spectrum analyser.
D.3 Conclusions 139<br />
As an example we use the <strong>polarisation</strong> <strong>independent</strong> raised-strip-waveguide-<strong>based</strong> PHASAR as<br />
presented in section 4.2.3, the measurement curve (using the tunable laser) of which is shown<br />
in figure D.4b (dotted line, compare with figure 4.10). This device has a FSR of approximately<br />
32 nm and a cross talk better than -25 dB. The response measured using the EDFA is<br />
callibrated with respect to the ASE spectrum and is shown in figure D.4b (solid line). Both<br />
curves are normalised to a peak maximum of 0 dB for good comparison. The improvement of<br />
the cross talk level (from -25 dB to -32 dB) can be observed clearly.<br />
ASE [dBm]<br />
0<br />
-20<br />
-40<br />
-60<br />
1500 1520 1540 1560 1580 1600<br />
Wavelength [nm]<br />
D.3 Conclusions<br />
-60<br />
1540 1545 1550<br />
Wavelength [nm]<br />
1555 1560<br />
(a) (b)<br />
From the results presented here, it can be concluded that the best way to measure the response<br />
of periodic devices is to use an EDFA and a spectrum analyser. In this way the spontaneous<br />
emission of the tunable laser is filtered out completely and better cross talk figures are<br />
obtained. Additionally, this method has the advantage that the measurement takes less time<br />
(only a few seconds), during which the fibre-chip coupling efficiency will hardly vary.<br />
Furthermore, the chips do not need to be anti-reflection coated, because of the limited slit<br />
width (0.1 nm) of the spectrum analyser.<br />
Transmission [dB]<br />
0<br />
-20<br />
-40<br />
5 dB<br />
Figure D.4 ASE of an EDFA (a) and the response of the PHASARs (b).<br />
32 dB
140 Appendix D. Cross talk penalty of the tunable laser
Appendix E<br />
Excitation coefficient of the first-order mode<br />
With reference to figure 4.14, an estimation is made of the excitation coefficient of the firstorder<br />
mode at the junction between the free propagation region and the array waveguides. A<br />
Gaussian approximation for the waveguide field is used and it is assumed that the gap between<br />
the array waveguides equals zero. In figure E.1 both the far field of a transmitter waveguide<br />
(dashed line) and the array waveguide fields (solid line) at the junction between the free<br />
propagation region and the array waveguides are depicted.<br />
intensity<br />
array aperture<br />
Figure E.1 Fields at the junction between the FPR and the array waveguides: far<br />
field of a transmitter waveguide (dashed) and array waveguide fields (solid).<br />
The maximum slope of a Gaussian field is 1/e 2 . If the number of array waveguides is chosen in<br />
such a way that almost all power of the transmitter far-field is captured (see figure 2.3b), the<br />
maximum slope of the far-field equals 1/(Nae 2 ) along a single array waveguide, with Na being<br />
the number of array waveguides. (It should be noted that for larger gaps the slope is even less<br />
steep.) We consider a single array waveguide field as denoted by the circle in the figure. We<br />
can consider the far-field being built up locally of a uniform part and a linearly decreasing part.<br />
The uniform part contributes to excitation of the fundamental mode in the array waveguide.<br />
The linearly decreasing part, on the other hand, causes excitation of the first-order mode.<br />
The excitation coefficient of the first-order mode can be calculated using the overlap integral of<br />
the field distribution of the first-order mode and a saw tooth with slope 1/(Nae 2 ). The result of<br />
this calculation is shown in figure E.2 as a function of Na . From this figure it can be seen that<br />
for realistic values of Na ( N a ><br />
10 ) the first-order excitation is well below -50 dB and may<br />
therefore be neglected.
142 Appendix E. Excitation coefficient of the first-order mode<br />
Excitation coefficient [dB]<br />
-20<br />
-40<br />
-60<br />
-80<br />
0 20 40 60 80 100<br />
Na Figure E.2 First-order mode excitation coefficient versus the number of array<br />
guides N a .
Appendix F<br />
Phase correction in MMI-MZI <strong>demultiplexers</strong><br />
As discussed in Chapter 6, a correction on the lengths of the array guides of a MMI-MZI<br />
demultiplexer has to be made in order to obtain the proper phase distribution at the input of the<br />
second MMI-coupler for the central <strong>wavelength</strong>. In this appendix it will be proven that this<br />
correction is neglegibly small and will not influence the dispersion of the array.<br />
The maximum length correction δL max will not exceed a half (central) <strong>wavelength</strong> in the<br />
waveguide:<br />
δL max<br />
whereby N eff is the mode index of the waveguide. This will cause a maximum phase error:<br />
whereby Δf ch is the channel spacing. If we neglect material and waveguide dispersion, we can<br />
write for the far right term in the above equation:<br />
By substituting equation F.3 and F.1 into F.2, we find for the maximum phase error:<br />
λ c<br />
1<br />
c<br />
≤ -- ⋅ -------- = ------------------<br />
2 N eff 2 f cN eff<br />
dβ<br />
δϕmax = δLmax ⋅ Δβ ≈ δLmax ⋅ Δ f ch ⋅ ----d<br />
f<br />
dβ<br />
----- = β<br />
d f<br />
1 1<br />
---- -------- dN eff<br />
⋅ ⎛ + ⋅ ------------ ⎞<br />
⎝ d f ⎠<br />
f c<br />
N eff<br />
δϕmax π Δ f ch<br />
≤ ⋅ -----------<br />
β<br />
≈ ---f<br />
c<br />
As Δ f ch ⁄ f c « 1,<br />
the maximum phase error δϕmax «<br />
1 and therefore the influence on the<br />
dispersion of the array is negligibly small.<br />
f c<br />
(F.1)<br />
(F.2)<br />
(F.3)<br />
(F.4)
144 Appendix F. Phase correction in MMI-MZI <strong>demultiplexers</strong>
References<br />
[1] R. Adar, C.H. Henry, C. Dragone, R.C. Kistler, and M.A. Milbrodt, “Broad-band array<br />
multiplexers made with silica waveguides on silicon”, J. Lightw. Tech., 11 (2), pp. 212-<br />
219, February 1993.<br />
[2] P. Albrecht, M. Hamacher, H. Heidrich, D. Hoffmann, H.P. Nolting, and C.M. Weinert,<br />
“TE/TM mode splitters on InGaAsP”, IEEE Phot. Tech. Lett., 2 (2), pp. 114-115,<br />
February 1990.<br />
[3] M.R. Amersfoort, M.K. Smit, Y.S. Oei, B.H. Verbeek, P. Demeester, F.H. Groen, and<br />
E.G. Metaal, “Small-size, low-loss, 4-channel phased-array <strong>wavelength</strong> division<br />
(de)multiplexer on <strong>InP</strong>”, in Procs. 6th Eur. Conf. on Int. Opt. (ECIO’93), Neuchatel,<br />
Switzerland, Postdeadline Papers, pp. 2-5, April 18-22, 1993.<br />
[4] M.R. Amersfoort, C.R. de Boer, Y.S. Oei, B.H. Verbeek, P. Demeester, F.H. Groen, and<br />
J.W. Pedersen, “High performance 4-channel PHASAR <strong>wavelength</strong> demultiplexer<br />
integrated with photodetectors”, in Procs. 19th Eur. Conf. on Opt. Comm. (ECOC’93),<br />
Montreux, Switzerland, Postdeadline Papers, pp. 49-52, September 12-16, 1993.<br />
[5] M.R. Amersfoort, C.R. de Boer, B.H. Verbeek, P. Demeester, A. Looyen, and J.J.G.M.<br />
van der Tol, “Low-loss phased-array <strong>based</strong> 4-channel <strong>wavelength</strong> demultiplexer<br />
integrated with photodetectors”, IEEE Phot. Tech. Lett., 6 (1), pp. 62-64, January 1994.<br />
[6] M.R. Amersfoort, C.R. de Boer, F.P.G.M. van Ham, M.K. Smit, P. Demeester, J.J.G.M.<br />
van der Tol, and A. Kuntze, “Phased-array <strong>wavelength</strong> demultiplexer with flattened<br />
<strong>wavelength</strong> response”, Electron. Lett., 30 (4), pp. 300-302, February 1994.<br />
[7] M.R. Amersfoort, Phased-array <strong>wavelength</strong> <strong>demultiplexers</strong> and their integration with<br />
photodetectors. PhD thesis, Delft University of Technology, Delft, The Netherlands,<br />
1994. ISBN 90-407-1041-4.<br />
[8] M.R. Amersfoort, P.A. Besse, H. Melchior, M.K. Smit, and C. van Dam, “Wavelength<br />
demultiplexer with MMI’s”, EP-P Patent request 95200182.4, Priority date: January 26,<br />
1995.<br />
[9] M.R. Amersfoort, J.B.D. Soole, H.P. Leblanc, N.C. Andreakakis, A. Rajhel, and C.<br />
Caneau, “8x2 nm polarization-<strong>independent</strong> WDM detector <strong>based</strong> on compact arrayed<br />
waveguide demultiplexer”, in Integrated Photonics Research (IPR’95), Dana Point,<br />
USA, Postdeadline Papers, pp. PD3-1 - PD3-2, February 23-25, 1995.<br />
[10] M.R. Amersfoort, J.B.D. Soole, H.P. Leblanc, N.C. Andreadakis, A. Rajhel, and C.<br />
Caneau, “Passband broadening of integrated arrayed waveguide filters using multimode<br />
interference couplers”, Electron. Lett., 32 (5), pp. 449-451, February 1996.<br />
[11] M.R. Amersfoort, J.B.D. Soole, H.P. Leblanc, N.C. Andreakakis, A. Rahjel, C. Caneau,
146 References<br />
M.A. Koza, R. Bhat, C. Youtsey, and I. Adesida, “Polarization-<strong>independent</strong> <strong>InP</strong>-arrayed<br />
waveguide filter using square cross-section waveguides”, in Optical Fiber<br />
Communication (OFC’96), San Jose, USA, pp. 101-102, February 25-March 1, 1996.<br />
[12] M.R. Amersfoort, C.E. Zah, B. Pathak, F.Favire, P.S.D. Lin, A. Rahjel, N.C.<br />
Andreakakis, R. Bhat, C. Caneau, and M.A. Koza, “Wavelength accuracy and output<br />
power of multi<strong>wavelength</strong> DFB laser arrays with integrated star couplers and optical<br />
amplifiers”, in Integrated Photonics Research (IPR’96), Boston, USA, pp. 478-481,<br />
April 29-May 2, 1996.<br />
[13] M. Bachmann, P.A. Besse, and H. Melchior, “General self-imaging properties in NxN<br />
multi-mode interference couplers including phase relations”, Appl. Opt., 33 (18), pp.<br />
3905-3911, June 1994.<br />
[14] P.A. Besse, M. Bachmann, and H. Melchior, “Phase relations in multi-mode interference<br />
couplers and their application to generalized integrated Mach-Zehnder optical switches”,<br />
in Procs. 6th Eur. Conf. on Int. Opt. (ECIO’93), Neuchatel, Switzerland, pp. 2.22-2.23,<br />
April 18-22, 1993.<br />
[15] P.A. Besse, M. Bachmann, H. Melchior, L.B. Soldano, and M.K. Smit, “Optical<br />
bandwidth and fabrication tolerances of multimode interference couplers”, J. Lightw.<br />
Tech., 12 (6), pp. 1004-1009, June 1994.<br />
[16] W.K. Burns and A.F. Milton, “Mode conversion in planar dielectric separating<br />
waveguides”, IEEE J. Quantum Electron., QE-11 (1), pp. 32-39, 1975.<br />
[17] H. Bissessur, F. Gaborit, B. Martin, P. Pagnod-Rossiaux, J.-L. Peyre, and M. Renaud, “16<br />
channel phased array demultiplexer on <strong>InP</strong> with low <strong>polarisation</strong> sensitivity”, Electron.<br />
Lett., 30 (4), pp. 336-337, February 1994.<br />
[18] H. Bissessur, F. Gaborit, B. Martin, G. Ripoche, and P. Pagnod-Rossiaux, “Tunable<br />
phased-array <strong>wavelength</strong> demultiplexer on <strong>InP</strong>”, Electron. Lett., 31 (1), pp. 32-33,<br />
January 1995.<br />
[19] H. Bissessur, F. Gaborit, B. Martin, and G. Ripoche, “Polarisation-<strong>independent</strong> phasedarray<br />
demultiplexer on <strong>InP</strong> with high fabrication tolerance”, Electron. Lett., 31 (16), pp.<br />
1372-1373, August 1995.<br />
[20] H. Bissessur, B. Martin, R. Mestric, and F. Gaborit, “Small-size, <strong>polarisation</strong>-<strong>independent</strong><br />
phased-array <strong>demultiplexers</strong> on <strong>InP</strong>”, Electron. Lett., 31 (24), pp. 2118-2120, November<br />
1995.<br />
[21] H. Bissessur, P. Pagnod-Rossiaux, R. Mestric, and B. Martin, “Extremely small<br />
polarization <strong>independent</strong> phased-array <strong>demultiplexers</strong> on <strong>InP</strong>”, IEEE Phot. Tech. Lett., 8<br />
(4), pp. 554-556, April 1996.<br />
[22] P. Blixt, J. Nilsson, T. Carlnäs, and B. Jaskorzynska, “Concentration-dependent<br />
upconversion in Er 3+ -doped fiber amplifiers: experiments and modelling”, IEEE Phot.<br />
Techn. Lett., 3 (11), pp. 996-998, November 1991.<br />
[23] C.A. Brackett, “Dense <strong>wavelength</strong> division multiplexing networks: principles and<br />
applications”, IEEE J. Select. Areas Comm., Vol. 8, pp. 948-964, 1990.<br />
[24] C.A. Brackett, A.S. Acampora, J. Sweitzer, G. Tangonan, M.T. Smith, W. Lennon, K.C.<br />
Wang, and R.H. Hobbs, “A scalable multi<strong>wavelength</strong> multihop optical network: a<br />
proposal for research on all-optical networks”, J. Lightw. Tech., 11 (5/6), pp. 736-753,<br />
May/June 1993.<br />
[25] T. Brenner, C.H. Joyner, M. Zirngibl, and C. Doerr, “Compact waveguide grating array<br />
demultiplexer”, in Integrated Photonics Research (IPR’96), Boston, USA, Postdeadline<br />
Papers, pp. PDP3-1 -PDP3-3, April 29-May 2, 1996.<br />
[26] K.S. Chiang, “Dispersion characteristics of strip dielectric waveguides”, IEEE trans.
References 147<br />
MTT, 39 (2), pp. 349-352, February 1991.<br />
[27] P.C. Clemens, G. Heise, R. Marz, H. Michel, A. Reichelt, and H.W. Schneider, “Optical<br />
phased array in SiO 2 /Si with adaptable center <strong>wavelength</strong>”, in Procs. 7th Eur. Conf. on<br />
Int. Opt. (ECIO’95), Delft, The Netherlands, pp. 505-508, April 3-6, 1995.<br />
[28] P.C. Clemens, G. Heise, R. Marz, A. Reichelt, and H.W. Schneider, “Wavelengthadaptable<br />
optical phased array in SiO 2 -Si”, IEEE Phot. Tech. Lett., 7 (10), pp. 1040-1041,<br />
October 1995.<br />
[29] C. van Dam, L.C.N. de Vreede, M.K. Smit, J.L. Tauritz, and B.H. Verbeek, “Optical chip<br />
design with a microwave CAD-system”, in Procs. 10th Eur. Conf. on Circuit Theory and<br />
Design, Copenhagen, Denmark, Vol. III, pp. 1316-1323, September 2-6, 1991.<br />
[30] C. van Dam, J.W.M. van Uffelen, M.K. Smit, G.N. van den Hoven, and A. Polman,<br />
“Optical imaging of multimode interference patterns with a resolution below the<br />
diffraction limit”, in Procs. 7th Eur. Conf. on Int. Opt. (ECIO’95), Delft, The<br />
Netherlands, pp. 125-128, April 3-6, 1995.<br />
[31] C. van Dam, M.R. Amerfsoort, G.M. ten Kate, F.P.G.M. van Ham, M.K. Smit, P.A.<br />
Besse, M. Bachmann, and H. Melchior, “Novel <strong>InP</strong>-<strong>based</strong> phased-array <strong>wavelength</strong><br />
demultiplexer using a generalized MMI-MZI configuration”, in Procs. 7th Eur. Conf. on<br />
Int. Opt. (ECIO’95), Delft, The Netherlands, pp. 275-278, April 3-6, 1995.<br />
[32] C. van Dam, H. Heidrich, M. Hamacher, C. Steenbergen, M. Smit, and C. Weinert,<br />
“Curved waveguide as polarization converter”, Philips patent nr. 15.544, Appl. nr.<br />
95202957.7, November 2, 1995.<br />
[33] C. van Dam, A.A.M. Staring, E.J. Jansen, J.J.M. Binsma, T. van Dongen, M.K. Smit, and<br />
B.H. Verbeek, “Loss reduction for phased-array <strong>demultiplexers</strong> using a double etch<br />
technique”, in Integrated Photonics Research (IPR’96), Boston, USA, pp. 52-55,<br />
April 29-May 2, 1996.<br />
[34] C. van Dam, L.H. Spiekman, F.P.G.M. van Ham, F.H. Groen, J.J.G.M. van der Tol, I.<br />
Moerman, W.W. Pascher, M. Hamacher, H. Heidrich, C.M. Weinert, and M.K. Smit,<br />
“Novel compact <strong>InP</strong>-<strong>based</strong> <strong>polarisation</strong> converters using ultra short bends”, in Integrated<br />
Photonics Research (IPR’96), Boston, USA, pp. 468-471, April 29-May 2, 1996.<br />
[35] C. van Dam, L.H. Spiekman, F.P.G.M. van Ham, F.H. Groen, J.J.G.M. van der Tol, I.<br />
Moerman, W.W. Pascher, M. Hamacher, H. Heidrich, C.M. Weinert, and M.K. Smit,<br />
“Novel compact <strong>polarisation</strong> converters <strong>based</strong> on ultra short bends”, IEEE Phot. Tech.<br />
Lett., 8(10), pp. 1346-1348, October 1996.<br />
[36] C. van Dam, A.A.M. Staring, E.J. Jansen, J.J.M. Binsma, T. van Dongen, M.K. Smit, and<br />
B.H. Verbeek, “Compact spatial mode filter <strong>based</strong> on a MMI coupler”, in Procs. 1996<br />
Symp. IEEE/LEOS Benelux Chapter, Enschede, The Netherlands, pp. 112-115, November<br />
28, 1996.<br />
[37] C. van Dam, A.A.M. Staring, E.J. Jansen, J.J.M. Binsma, T. van Dongen, M.K. Smit, and<br />
B.H. Verbeek, “Elimination of ghost images in the response of PHASAR<strong>demultiplexers</strong>”,<br />
in Procs. 8th Eur. Conf. on Int. Opt. (ECIO’97), Stockholm, Sweden, pp.<br />
268-271, April 2-4, 1997.<br />
[38] S. Day, J.P. Stagg, D. Moule, S.J. Clemens, C. Rogers, S. Ohja, T. Clapp, J. Brook, and<br />
J. Morley, “The elimination of sidelobes in the arrayed waveguide WDM”, in Integrated<br />
Photonics Research (IPR’96), Boston, USA, pp. 48-51, April 29-May 2, 1996.<br />
[39] R.J. Deri, N. Yasuoka, M. Makiuchi, A. Kuramata, and O. Wada, “Polarisation-dependent<br />
modal properties of diluted multiple-quantum-well optical waveguides”, Electron. Lett.,<br />
25 (17), pp. 1162-1164, August 1989.<br />
[40] R.J. Deri, E.C.M. Pennings, A. Scherer, A.S. Gozdz, C. Caneau, N.C. Andreakakis, V.
148 References<br />
Shah, L. Curtis, R.J. Hawkins, J.B.D. Soole, and J.I. Song, “Ultracompact monolithic<br />
integration of balanced, polarization diversity photodetectors for coherent lightwave<br />
receivers”, IEEE Phot. Tech. Lett., 4 (11), pp. 1238-1240, November 1992.<br />
[41] E. Desurvire, J.R. Simpson, and P.C. Becker, “High-gain erbium-doped traveling-wave<br />
fiber amplifier”, Opt. Lett., 12 (11), pp. 888-890, November 1987.<br />
[42] C.R. Doerr, M. Zirngibl, and C.H. Joyner, “Chirping of the waveguide grating router for<br />
free-spectral-range mode selection in the multifrequency laser”, IEEE Phot. Tech. Lett.,<br />
8 (4), pp. 500-502, April 1996.<br />
[43] C.R. Doerr, M. Zirngibl, and C.H. Joyner, “Chirping of the waveguide grating router for<br />
free-spectral-range mode selection in the multi-frequency laser”, in Integrated Photonics<br />
Research (IPR’96), Boston, USA, pp. 132-135, April 29-May 2, 1996.<br />
[44] C. Dragone, “An NxN optical multiplexer using a planar arrangement of two star<br />
couplers”, IEEE Phot. Tech. Lett., 3 (9), pp. 812-815, September 1991.<br />
[45] C. Dragone, C.A. Edwards, and R.C. Kistler, “Integrated optics NxN multiplexer on<br />
silicon”, IEEE Phot. Tech. Lett., 3 (10), pp. 896-899, October 1991.<br />
[46] F. Fiedler and A. Schlachetzki, “Optical parameters of <strong>InP</strong>-<strong>based</strong> waveguides”, Solid<br />
State Electron., 30 (1), pp. 73-83, January 1987.<br />
[47] M. Fukui, K. Oda, H. Toba, K. Okamoto, and M. Ishii, “10 channel x 10Gbit/s WDM add/<br />
drop multiplexing/transmission experiment over 240 km of dispersion-shifted fibre<br />
employing unequally-spaced arrayed-waveguide-grating ADM filter with fold-back<br />
configuration”, Electron. Lett., 31 (20), pp. 1757-1759, September 1995.<br />
[48] B. Glance, I.P. Kaminow, and R.W. Wilson, “Applications of the integrated waveguide<br />
grating router”, J. Lightw. Tech., 12 (6), pp. 957-962, June 1994.<br />
[49] F.E. Goodwin, “A 3.39-μm infrared optical heterodyne communication system”, IEEE J.<br />
Quantum Electron., QE-3 (11), pp. 524-531, November 1967.<br />
[50] K. Hattori, T. Kitagawa, M. Oguma, Y. Ohmori, and M. Horiguchi, “Erbium-doped<br />
silica-<strong>based</strong> waveguide amplifier integrated with a 980/1530 nm WDM coupler”,<br />
Electron. Lett., 30 (11), pp. 856-857, May 1994.<br />
[51] J.M. Heaton, R.M. Jenkins, D.R. Wight, J.T. Parker, J.C.H. Birbeck, and K.P. Hilton,<br />
“Novel 1-to-N way integrated optical beam splitters using symmetric mode mixing in<br />
GaAs/AlGaAs multimode waveguides”, Appl. Phys. Lett., 61 (15), pp. 1754-1756, 1992.<br />
[52] M. Heiblum and J.H. Harris, “Analysis of curved optical waveguides by conformal<br />
transformation”, IEEE J. Quantum Electron., QE-11 (2), pp. 75-83, February 1975.<br />
[53] H. Heidrich, P. Albrecht, M. Hamacher, H.P. Nolting, H. Schroeter-Janssen, and C.M.<br />
Weinert, “Passive mode converter with a periodically tilted <strong>InP</strong>/GaInAsP rib waveguide”,<br />
IEEE Phot. Tech. Lett., 4 (1), pp. 34-36, January 1992.<br />
[54] B.R. Hemenway, M.L. Stevens, D.M. Castagozzi, D. Marquis, S.A. Parikh, J.J. Carney,<br />
S.G. Finn, E.A. Swanson, I.P Kaminow, C. Dragone, U. Koren, T.L. Koch, R. Thomas,<br />
C. Özveren, and E. Grella, “A 20-channel <strong>wavelength</strong>-routed all-optical network<br />
deployed in the Boston Metro area”, in Optical Fiber Communication (OFC’95), San<br />
Diego, USA, Postdeadline Papers, pp. PD8-1 - PD8-5, February 26-March 3, 1995.<br />
[55] C.G.P. Herben, X.J.M. Leijtens, C. van Dam, and M.K. Smit, “Chirped MMIdemultiplexer<br />
for use in multi<strong>wavelength</strong> lasers”, in Procs. 1996 Symp. IEEE/LEOS<br />
Benelux Chapter, Enschede, The Netherlands, pp. 91-94, November 28, 1996.<br />
[56] Y. Hida, Y. Innoue, and S. Imamura, “Polymeric arrayed-waveguide grating multiplexer<br />
operating around 1.3 μm”, Electron. Lett., 30 (12), pp. 959-960, June 1994.<br />
[57] G.N. van den Hoven, E. Snoeks, A. Polman, J.W.M. van Uffelen. Y.S. Oei, and M.K. Smit,<br />
“Photoluminescence characterization of Er-implanted Al 2 O 3 films”, Appl. Phys. Lett., 62
References 149<br />
(24), pp. 3065-3067, 1993.<br />
[58] G.N. van den Hoven, R.J.I. Koper, A. Polman, C. van Dam, J.W.M. van Uffelen, and<br />
M.K. Smit, “Net optical gain at 1.53 μm in Er-doped Al 2 O 3 waveguides”, Appl. Phys.<br />
Lett., 68 (14), pp. 1886-1888, April 1996.<br />
[59] G.N. van den Hoven, A. Polman, C. van Dam, J.W.M. van Uffelen, and M.K. Smit,<br />
“Direct imaging of optical interference in Er-doped Al 2 O 3 waveguides”, Opt. Lett., 21<br />
(8), pp. 576-578, April 1996.<br />
[60] Y. Innoue, Y. Ohmori, M. Kawachi, S. Ando, T. Sawada, and H. Takahashi, “Polarization<br />
mode converter with polyimide half waveplate in silica-<strong>based</strong> planar lightwave circuits”,<br />
IEEE Phot. Tech. Lett., 6 (5), pp. 626-628, May 1994.<br />
[61] Y. Inoue, A. Himeno, K. Moriwaki, and M. Kawachi, “Silica-<strong>based</strong> arrayed-waveguide<br />
grating circuit as optical splitter/router”, Electron. Lett., 31 (9), pp. 726-727, April 1995.<br />
[62] O. Ishida and H. Takahashi, “Loss-imbalance equalisation of arrayed-waveguide grating<br />
add-drop multiplexer”, Electron. Lett., 30 (14), pp. 1160-1162, July 1994.<br />
[63] O. Ishida, H. Takahashi, and Y. Inoue, “FDM-channel selection filter employing an<br />
arrayed-waveguide grating multiplexer”, Electron. Lett., 30 (16), pp. 1327-1328, August<br />
1994.<br />
[64] O. Ishida, H. Takahashi, S. Suzuki, and Y. Innoue, “Multichannel frequency-selective<br />
switch employing an arrayed-waveguide grating multiplexer with fold-back optical<br />
paths”, IEEE Phot. Tech. Lett., 6 (10), pp. 1219-1221, October 1994.<br />
[65] O. Ishida, T. Hasegawa, M. Ishii, S. Suzuki, and K. Iwashita, “4x4, 7-FDM-channel<br />
reconfigurable network hub emplying arrayed-waveguide-grating (AWG) multiplexers”,<br />
in Optical Fiber Communication (OFC’95), San Diego, USA, Postdeadline Papers, pp.<br />
PD9-1 - PD9-5, February 26-March 3, 1995.<br />
[66] C.F. Janz, M.R. Paiam, B.P. Keyworth, and J.N. Broughton, “Bent waveguide couplers<br />
for (de)multiplexing of arbitrary broadly-separated <strong>wavelength</strong>s using two-mode<br />
interference”, IEEE Phot. Tech. Lett., 7 (9), pp. 1037-1039, September 1995.<br />
[67] R.M. Jenkins, J.M. Heaton, D.R. Wight, J.T. Parker, J.C.H. Birbeck, G.W. Smith, and K.P.<br />
Hilton, “Novel 1xN and NxN integrated optical switches using self-imaging multimode<br />
GaAs/AlGaAs waveguides”, Appl. Phys. Lett., 64, pp. 684-686, February 1994.<br />
[68] C.H. Joyner, M. Zirngibl, and J.P. Meester, “A multifrequency waveguide grating laser<br />
by selective area epitaxy”, IEEE Phot. Tech. Lett., 6 (11), pp. 1277-1279, November<br />
1994.<br />
[69] C.H. Joyner, M. Zirngibl, and J.C. Centanni, “An 8 channel digitally tunable transmitter<br />
with electroabsorption modulated output by selective area epitaxy”, in Integrated<br />
Photonics Research (IPR’95), Dana Point, USA, Postdeadline Papers, pp. PD2-2 - PD2-4,<br />
February 23-25, 1995.<br />
[70] C.H. Joyner, M. Zirngibl, and J.C. Centanni, “An 8-channel digitally tunable transmitter<br />
with electroabsorption modulated output by selective-area epitaxy”, IEEE Phot. Tech.<br />
Lett., 7 (9), pp. 1013-1015, September 1995.<br />
[71] F.P. Kapron, D.B. Keck, and R.D. Maurer, “Radiation losses in glass optical<br />
waveguides”, Appl. Phys. Lett., 17, pp. 423-425, November 1970.<br />
[72] K.C. Kao and G.H. Hockham, “Dielectric fiber surface waveguides for optical<br />
frequencies”, Proceedings IEE, 113 (7), pp. 1151-1158, July 1966.<br />
[73] R.M. Knox and P.P. Toulios, “Integrated circuits for the millimeter through optical<br />
frequency range”, in Procs. Symp. on Submillimeter Waves, New York, USA, pp. 497-<br />
516, March 31-April 2, 1970.<br />
[74] J. Leuthold, R. Hess, J. Eckner, P.A. Besse, and H. Melchior, “Spatial mode filter realized
150 References<br />
with multimode interference couplers”, Opt. Lett., 21 (11), pp. 836-838, June 1996.<br />
[75] X.J.M. Leijtens, L.H. Spiekman, C. van Dam, L.C.N. de Vreede, M.K. Smit, and J.L.<br />
Tauritz, “CAD-tool for integrated optics”, in Procs. 7th Eur. Conf. on Int. Opt.<br />
(ECIO’95), Delft, The Netherlands, pp. 463-466, April 3-6, 1995.<br />
[76] L.O. Lierstuen and A. Sudbø, “8-channel <strong>wavelength</strong> division multiplexer <strong>based</strong> on<br />
multimode interference couplers”, IEEE Phot. Tech. Lett., 7 (9), pp. 1034-1036,<br />
September 1995.<br />
[77] W. Lin, H. Li, Y.J. Chen, M. Dagenais, and D. Stone, “Dual-channel-spacing phasedarray<br />
waveguide grating multi/<strong>demultiplexers</strong>”, IEEE Phot. Tech. Lett., 8 (11), pp. 1501-<br />
1503, November 1996.<br />
[78] W. Lui, K. Magari, N. Yoshimoto, S. Oku, T. Hirono, K. Yokayama, and W.P. Huang,<br />
“Design and analysis of bending waveguide <strong>based</strong> semiconductor <strong>polarisation</strong> rotators”,<br />
in Procs. 8th Eur. Conf. on Int. Opt. (ECIO’97), Stockholm, Sweden, pp. 30-33, April 2-<br />
4, 1997.<br />
[79] T.H. Maiman, “Stimulated optical radiation in ruby”, Nature, 187, pp. 493-494, August<br />
1960.<br />
[80] E.A.J. Marcatili, “Dielectric rectangular waveguide and directional coupler for integrated<br />
optics”, Bell Syst. Tech. J., 48, pp. 2071-2102, September 1969.<br />
[81] R. März, Integrated optics: design and modeling. Artech House Inc., 1994.<br />
[82] R.J. Mears, L. Reekie, I.M. Jauncy, and D.N. Payne, “Low-noise erbium-doped fibre<br />
amplifier operating at 1.54 μm”, Electron. Lett., 23 (19), pp. 1026-1028, September 1987.<br />
[83] R. Mestric, H. Bissessur, B. Martin, and A. Pinquier, “1.31-1.55-μm phased-array<br />
demultiplexer on <strong>InP</strong>”, IEEE Phot. Tech. Lett., 8 (5), pp. 638-640, May 1996.<br />
[84] R. Mestric, M. Renaud, B.G. Martin, and F. Gaborit, “Extremely compact 1.31-1.55 μm<br />
phased array duplexer on <strong>InP</strong> with -30 dB crosstalk over 100 nm”, in Procs. 22nd Eur.<br />
Conf. on Opt. Comm. (ECOC’96), Oslo, Norway, pp. 3.131-3.134, September 15-19,<br />
1996.<br />
[85] National Council of Intellectual Property Law Associations and the Patent and Trademark<br />
Office of the United States Department of Commerce, National Inventors Hall of Fame.<br />
Internet site http://www.invent.org/inventure.html.<br />
[86] Y.S. Oei, C. van Dam, F.P. van Ham, L.H. Spiekman, B.H. Verbeek, F.H. Groen, E.G.<br />
Metaal, and J.W. Pedersen, “Improved RIE technique for controlled roughness and<br />
anisotropy in <strong>InP</strong>-<strong>based</strong> devices”, in Procs. 18th State-of-the-art Program on Compound<br />
Semiconductors (SOTAPOCS XVIII), Honolulu, U.S.A., Volume 93-27, pp. 134-141,<br />
May 16-21, 1993.<br />
[87] K. Okamoto and H. Yamada, “Arrayed-waveguide grating multiplexer with flat spectral<br />
response”, Opt. Lett., 20 (1), pp. 43-45, January 1995.<br />
[88] K. Okamoto, K. Moriwaki, and S. Suzuki, “Fabrication of 64x64 arrayed-waveguide<br />
grating multiplexer on silicon”, Electron. Lett., 31 (3), pp. 184-186, February 1995.<br />
[89] K. Okamoto, K. Takiguchi, and Y. Ohmori, “16ch optical add/drop multiplexer using<br />
silica-<strong>based</strong> arrayed-waveguide gratings”, in Optical Fiber Communication (OFC’95),<br />
San Diego, USA, Postdeadline Papers, pp. PD10-1 - PD10-5, February 26-March 3, 1995.<br />
[90] K. Okamoto, K. Takiguchi, and Y. Ohmori, “16-channel add/drop multiplexer using<br />
silica-<strong>based</strong> arrayed-waveguide gratings”, Electron. Lett., 31 (9), pp. 723-724, April<br />
1995.<br />
[91] K. Okamoto, M. Ishii, Y. Hibino, and H. Toba, “Fabrication of unequal channel spacing<br />
arrayed-waveguide grating multiplexer modules”, Electron. Lett., 31 (17), pp. 1464-1465,<br />
August 1995.
References 151<br />
[92] K. Okamoto, M. Ishii, Y. Hibino, and Y. Ohmori, “Fabrication of variable bandwidth<br />
filters using arrayed-waveguide gratings”, Electron. Lett., 31 (18), pp. 1592-1594, August<br />
1995.<br />
[93] K. Okamoto, M. Okuno, A. Himeno, and Y. Ohmori, “16-channel optical add/drop<br />
multiplexer consisting of arrayed-waveguide gratings and double-gate switches”,<br />
Electron. Lett., 32 (16), pp. 1471-1472, August 1996.<br />
[94] K. Okamoto, K. Syuto, H. Takahashi, and Y. Ohmori, “Fabrication of 128-channel<br />
arrayed-waveguide grating with 25 GHz channel spacing”, Electron. Lett., 32 (16), pp.<br />
1474-1476, August 1996.<br />
[95] H. Okayama and M. Kawahara, “Waveguide array grating demultiplexer on LiNbO 3 ”, in<br />
Integrated Photonics Research (IPR’95), Dana Point, USA, pp. 296-298, February 23-25,<br />
1995.<br />
[96] H. Okayama, M. Kawahara, and T. Kamijoh, “Reflective waveguide array demultiplexer<br />
in LiNbO 3 ”, J. Lightw. Tech., 14 (6), pp. 985-990, June 1996.<br />
[97] H. Onaka, H. Miyata, G. Ishikawa, K. Otsuka, H. Ooi, Y. Kai, S. Kinoshita, M. Seino, H.<br />
Nishimoto, and T. Chikama, “1.1 Tb/s WDM transmission over a 150 km 1.3 μm zerodispersion<br />
single-mode fiber”, in Optical Fiber Communication (OFC’96), San Jose,<br />
USA, Postdeadline Papers, pp. PD19-2 - PD19-5, February 25-March 1, 1996.<br />
[98] M.R. Paiam, C.F. Janz, R.I. MacDonald, and J.N. Broughton, “Compact planar 980/1550<br />
nm <strong>wavelength</strong> multi/demultiplexer <strong>based</strong> on multimode interference”, IEEE Phot. Tech.<br />
Lett., 7 (10), pp. 1180-1182, October 1995.<br />
[99] W. Pascher and R. Pregla, “Vectorial analysis of bends in optical strip waveguides by the<br />
method of lines”, Radio Science, 28 (6), pp. 1229-1233, 1993.<br />
[100] R. Pregla, “The method of lines for the analysis of dielectric waveguide bends”, J. Ligthw.<br />
Tech., 14 (4), pp. 634-639, April 1996.<br />
[101] J.W. Pedersen, J.J.G.M. van der Tol, E.G. Metaal, Y.S. Oei, F.H. Groen, and I. Moerman,<br />
“Mode converting polarization splitter on InGaAsP/<strong>InP</strong>”, in Procs. 20th Eur. Conf. on<br />
Opt. Comm. (ECOC’94), Firenze, Italy, pp. 661-664, September 25-29, 1994.<br />
[102] E.C.M. Pennings, J. van Schoonhoven, J.W.M. van Uffelen, and M.K. Smit, “Reduced<br />
bending and scattering losses in new optical ‘double-ridge’ waveguide”, Electron. Lett.,<br />
25 (11), pp. 746-747, May 1989.<br />
[103] E.C.M. Pennings, Bends in optical ridge waveguides: modeling and experiments. PhD<br />
thesis, Delft University of Technology, Delft, The Netherlands, 1990. ISBN 90-<br />
9003413-7.<br />
[104] E.C.M. Pennings, R.J. Deri, and R.J. Hawkins, “Simple method for estimating usable<br />
bend radii of deeply etched optical rib waveguides”, Electron. Lett., 27 (17), pp. 1532-<br />
1533, August 1991.<br />
[105] E.C.M. Pennings, R.J. Deri, R. Bhat, T.R. Hayes, and N.C. Andreakakis, “Ultracompact,<br />
all-passive optical 90°-hybrid on <strong>InP</strong> using self-imaging”, IEEE Phot. Tech. Lett., 6 (5),<br />
pp. 701-703, 1992.<br />
[106] E.C.M. Pennings, M.K. Smit, and G.D. Khoe, "Micro-optic versus waveguide devices -<br />
an overview", in Procs. 5th Micro-Optics Conf. (MOC’95), Hiroshima, Japan, pp 248-<br />
255, October 18-20, 1995.<br />
[107] E.C.M. Pennings, M.K. Smit, A.A.M. Staring, and G.D. Khoe, Integrated-optics versus<br />
micro-optics - a comparison", in Integrated Photonics Research (IPR’96), Boston, USA,<br />
pp. 460-463, April 29-May 2, 1996.<br />
[108] B.M.A. Rahman, F.A. Fernandez, and J.B. Davies, “Review of the finite element methods<br />
for microwave and optical waveguides”, Proc. IEEE, 79 (10), pp. 1442-1448, October
152 References<br />
1991.<br />
[109] Lord Rayleigh, “The problem of the whispering gallery”, The London, Edinburgh, and<br />
Dublin philosophical magazine and journal of science, Series 6 (20), pp. 1001-1004,<br />
1910.<br />
[110] A. Rigny, C. Ramus, A. Bruno, Y. Rafle, H. Sik, G. Post, M. Carre, and A. Carenco,<br />
“Taper-assisted <strong>polarisation</strong> compensation in efficiently fibre-coupled <strong>InP</strong><br />
demultiplexer”, Electron. Lett., 32 (20), pp. 1885-1886, September 1996.<br />
[111] R. van Roijen, E.C.M. Pennings, M.J.N. van Stralen, T. van Dongen, B.H. Verbeek, and<br />
J.M.M. van der Heijden, “Compact <strong>InP</strong>-<strong>based</strong> ring lasers employing multimode<br />
interference couplers and combiners”, Appl. Phys. Lett., 64 (14), pp. 1753-1755, April<br />
1994.<br />
[112] F. Rottmann, A. Neyer, W. Mevenkamp, and E. Voges, “Integrated-optic <strong>wavelength</strong> multiplexers<br />
on lithium niobate <strong>based</strong> on two-mode interference”, J. Ligthw. Tech., 6 (6), pp.<br />
946-952, June 1988.<br />
[113] Y. Shani, C.H. Henry, R.C. Kistler, and K.J. Orlowsky, “Four-port integrated optic<br />
polarization splitter”, Appl. Opt., 29 (3), pp. 337-339, 1990.<br />
[114] Y. Shani, C.H. Henry, R.C. Kistler, R.F. Kazarinov, and K.J. Orlowsky, “Integrated optic<br />
adiabatic polarization splitter on silicon”, Appl. Phys. Lett., 56 (2), pp. 120-121, January<br />
1990.<br />
[115] Y. Shani, R. Alferness, T. Koch, U. Koren, M. Oron, B.I. Miller, and M.G. Young,<br />
“Polarization rotation in asymmetric periodic loaded rib waveguides”, Appl. Phys. Lett.,<br />
59 (11), pp. 1278-1280, September 1991.<br />
[116] M.K. Smit, “New focussing and dispersive planar component <strong>based</strong> on an optical phased<br />
array”, Electron. Lett., 24 (7), pp. 385-386, March 1988.<br />
[117] M.K. Smit, Integrated optics in silicon-<strong>based</strong> aluminum oxide. PhD thesis, Delft<br />
University of Technology, Delft, The Netherlands, 1991. ISBN 90-9004261-X.<br />
[118] M.K. Smit and C. van Dam, “PHASAR-<strong>based</strong> WDM-devices: principles, design and<br />
applications”, IEEE J. Sel. Topics Quant. Electron., 2 (2), pp. 236-250, June 1996.<br />
[119] L.B. Soldano, M. Bouda, M.K. Smit, and B.H. Verbeek, “New small-size single-mode<br />
optical power splitter <strong>based</strong> on multi-mode interference, in Procs. 18th Eur. Conf. on Opt.<br />
Comm. (ECOC’92), Berlin, Germany, pp. 465-468, September 27-October 1, 1992.<br />
[120] L.B. Soldano, A.H. de Vreede, M.K. Smit, B.H. Verbeek, E.G. Metaal, and F.H. Groen,<br />
“Mach-Zehnder interferometer <strong>polarisation</strong> splitter in InGaAsP/<strong>InP</strong>”, IEEE Phot. Tech.<br />
Lett., 6 (3), pp. 402-405, March 1994.<br />
[121] L.B. Soldano and E.C.M. Pennings, “Optical multi-mode interference devices <strong>based</strong> on<br />
self-imaging: principles and applications”, J. Lightw. Tech., 13 (4), pp. 615-627, April<br />
1995.<br />
[122] J.B.D. Soole, M.R. Amersfoort, H.P. Leblanc, N.C. Andreadakis, A. Rajhel, and C.<br />
Caneau, “Polarisation-<strong>independent</strong> monolithic eight-channel 2 nm spacing WDM<br />
receiver detector <strong>based</strong> on compact arrayed waveguide demultiplexer”, Electron. Lett., 31<br />
(15), pp. 1289-1291, July 1995.<br />
[123] J.B.D. Soole, M.R. Amersfoort, H.P. Leblanc, N.C. Andreadakis, A. Rajhel, C. Caneau,<br />
M.A. Koza, R.Bhat, C. Youtsey, and I. Adesida, “Polarisation-<strong>independent</strong> <strong>InP</strong> arrayed<br />
waveguide filter using square cross-section waveguides”, Electron. Lett., 32 (4), pp. 323-<br />
324, February 1996.<br />
[124] J.B.D. Soole, M.R. Amersfoort, H.P. Leblanc, N.C. Andreakakis, A. Rajhel, C. Caneau,<br />
R. Bhat, and M.A. Koza, “Use of multimode interference couplers to broaden the<br />
passband of dispersive integrated WDM filters”, in Integrated Photonics Research
References 153<br />
(IPR’96), Boston, USA, pp. 44-47, April 29- May 2, 1996.<br />
[125] J.B.D. Soole, M.R. Amersfoort, H.P. Leblanc, N.C. Andreadakis, A. Rajhel, C. Caneau,<br />
R. Bhat, M.A. Koza, C. Youtsey, and I. Adesida, “Use of multimode interference couplers<br />
to broaden the passband of <strong>wavelength</strong>-dispersive integrated WDM filters”, IEEE Phot.<br />
Tech. Lett., 8 (10), pp. 1340-1342, October 1996.<br />
[126] L.H. Spiekman, Y.S. Oei, E.G. Metaal, F.H. Groen, I. Moerman, and M.K. Smit,<br />
“Extremely small multimode interference couplers and ultrashort bends on <strong>InP</strong> by deep<br />
etching”, IEEE Phot. Tech. Lett., 6 (8), pp. 1008-1010, August 1994.<br />
[127] L.H. Spiekman, F.P.G.M. van Ham, A. Kuntze, J.W. Pedersen, P. Demeester, and M.K.<br />
Smit, “Polarization-<strong>independent</strong> <strong>InP</strong>-<strong>based</strong> phased-array <strong>wavelength</strong> demultiplexer with<br />
flattened <strong>wavelength</strong> response”, in Procs. 20th Eur. Conf. on Opt. Comm. (ECOC’94),<br />
Firenze, Italy, pp. 759-762, September 25-29, 1994.<br />
[128] L.H. Spiekman, Y.S. Oei, E.G. Metaal, F.H. Groen, P. Demeester, and M.K. Smit,<br />
“Ultrasmall waveguide bends: the corner mirrors of the future?”, IEE Proc.<br />
Optoelectron., 142 (1), pp. 61-65, February 1995.<br />
[129] L.H. Spiekman, A.H. de Vreede, F.P.G.M. van Ham, A. Kuntze, J.J.G.M. van der Tol, P.<br />
Demeester, and M.K. Smit, “Flattened response ensures polarization independence of<br />
InGaAsP/<strong>InP</strong> phased array <strong>wavelength</strong> demultiplexer”, in Procs. 7th Eur. Conf. on Int.<br />
Opt. (ECIO’95), Delft, The Netherlands, pp. 517-520, April 3-6, 1995.<br />
[130] L.H. Spiekman, A.A.M. Staring, C. van Dam, E.J. Jansen, J.J.M. Binsma, M.K. Smit, and<br />
B.H. Verbeek, “Space-switching using <strong>wavelength</strong> conversion at 2.5 Gb/s and integrated<br />
phased array routing”, in Procs. 21st Eur. Conf. on Optical Communications (ECOC’95),<br />
Brussels, Belgium, Postdeadline Papers, pp. 1055-1058, September 17-21, 1995.<br />
[131] L.H. Spiekman, M.B.J. Diemeer, T.H. Hoekstra, and M.K. Smit, “First polymeric phased<br />
array <strong>wavelength</strong> demultiplexer operating at 1550 nm”, in Integrated Photonics Research<br />
(IPR’96), Boston, USA, pp. 36-39, April 29-May 2, 1996.<br />
[132] L.H. Spiekman, A.A.M. Staring, J.J.M. Binsma, E.J. Jansen, T. van Dongen, P.J.A. Thijs,<br />
M.K. Smit, and B.H. Verbeek, “A compact phased array <strong>based</strong> multi-<strong>wavelength</strong> laser”,<br />
in Integrated Photonics Research (IPR’96), Boston, USA, pp. 136-138, April 29-May 2,<br />
1996.<br />
[133] L.H. Spiekman, M.R. Amersfoort, A.H. de Vreede, F.P.G.M. van Ham, A. Kuntze, J.W.<br />
Pedersen, P. Demeester, and M.K. Smit, “Design and realization of <strong>polarisation</strong><br />
<strong>independent</strong> phased array <strong>wavelength</strong> <strong>demultiplexers</strong> using different array orders for TE<br />
and TM”, J. Lightw. Tech., 14 (6), pp. 991-995, June 1996.<br />
[134] L.H. Spiekman, Compact integrated optical components for telecommunication<br />
networks. PhD thesis, Delft University of Technology, Delft, The Netherlands, 1996.<br />
ISBN 90-9009718-X.<br />
[135] A.A.M. Staring, L.H. Spiekman, C. van Dam, E.J. Jansen, J.J.M. Binsma, M.K. Smit, and<br />
B.H. Verbeek, “Space-switching 2.5 Gbit/s signals using <strong>wavelength</strong> conversion and<br />
phased array routing”, Electron. Lett., 32 (4), pp. 377-379, February 1996.<br />
[136] A.A.M. Staring, L.H. Spiekman, J.J.M. Binsma, E.J. Jansen, T. van Dongen, P.J.A. Thijs,<br />
M.K. Smit, and B.H. Verbeek, “A compact 9 channel multi-<strong>wavelength</strong> laser”, IEEE<br />
Phot. Tech. Lett., 8 (9), pp. 1139-1141, September 1996.<br />
[137] A.A.M. Staring, C. van Dam, J.J.M. Binsma, E.J.Jansen, A.J.M. Verboven, L.J.C.<br />
Vroomen, J.F. de Vries, M.K. Smit, and B.H. Verbeek, “Packaged PHASAR-<strong>based</strong><br />
<strong>wavelength</strong> demultiplexer with integrated detectors”, in Procs. 23rd Eur. Conf. on Opt.<br />
Comm. (ECOC’97), Edinburgh, Scotland, Paper We2A-2, September 22-25, 1997.<br />
[138] C.A.M. Steenbergen, C. van Dam, T.L.M. Scholtes, A.H. de Vreede, L. Shi, J.J.G.M. van
154 References<br />
der Tol, P. Demeester, and M.K. Smit, “4-channel <strong>wavelength</strong> flattened demultiplexer<br />
integrated with photodetectors”, in Procs. 7th Eur. Conf. on Int. Opt. (ECIO’95), Delft,<br />
The Netherlands, pp. 271-274, April 3-6, 1995.<br />
[139] C.A.M. Steenbergen, C. van Dam, A.H. de Vreede, L. Shi, J.W. Pedersen, P. Demeester,<br />
and M.K. Smit, “Integrated 1 GHz 4-channel <strong>InP</strong> phasar <strong>based</strong> WDM-receiver with Si<br />
bipolar frontend array”, in Procs. 21st Eur. Conf. on Opt. Comm. (ECOC’95), Brussels,<br />
Belgium, pp. 211-214, September 17-21, 1995.<br />
[140] C.A.M. Steenbergen, M.O. van Deventer, L.C.N. de Vreede, C. van Dam, M.K. Smit, and<br />
B.H. Verbeek, “System performance of a 4 channel phasar WDM receiver operating at<br />
1.2 Gb/s”, in Optical Fiber Communication (OFC’96), San Jose, USA, pp. 310-311,<br />
February 25-March 1, 1996.<br />
[141] C.A.M. Steenbergen, C. van Dam, A. Looijen, C.G.P. Herben, M. de Kok, M.K. Smit,<br />
J.W. Pedersen, I. Moerman, R.G.F. Baets, and B.H. Verbeek, “Compact low loss 8x10<br />
GHz <strong>polarisation</strong> <strong>independent</strong> WDM receiver”, in Procs. 22nd Eur. Conf. on Opt. Comm.<br />
(ECOC’96), Oslo, Norway, pp. 1.129-1.132, September 15-19, 1996.<br />
[142] S. Suzuki, Y. Innoue, and Y. Ohmori, “Polarisation-insensitive arrayed-waveguide<br />
grating multiplexer with SiO 2 -on-SiO 2 structure”, Electron. Lett., 30 (8), pp. 642-643,<br />
April 1994.<br />
[143] S. Suzuki, A. Himeno, Y. Tachikawa, and Y. Yamada, “Multichannel optical <strong>wavelength</strong><br />
selective switch with arrayed-waveguide grating multiplexer”, Electron. Lett., 30 (13),<br />
pp. 1091-1092, June 1994.<br />
[144] Y. Tachikawa, Y. Inoue, M. Kawachi, H. Takahashi, and K. Inoue, “Arrayed-waveguide<br />
grating add-drop multiplexer with loop-back optical paths”, Electron. Lett., 29 (24), pp.<br />
2133-2134, November 1993.<br />
[145] Y. Tachikawa and M. Kawachi, “Lightwave transrouter <strong>based</strong> on arryed-waveguide<br />
grating multiplexer”, Electron. Lett., 30 (18), pp. 1504-1506, September 1994.<br />
[146] Y. Tachikawa, M. Ishii, Y. Inoue, and T. Nozawa, “Integrated-optic arrayed-waveguide<br />
grating multiplexers with loop-back optical paths”, in Procs. 7th Eur. Conf. on Int. Opt.<br />
(ECIO’95), Delft, The Netherlands, pp. 267-270, April 3-6, 1995.<br />
[147] Y. Tachikawa and K. Okamoto, “32 <strong>wavelength</strong> tunable arrayed-waveguide grating laser<br />
<strong>based</strong> on special input/output arrangement”, Electron. Lett., 31 (19), pp. 1665-1666,<br />
September 1995.<br />
[148] Y. Tachikawa and Y. Inoue, “Tunable WDM-channel group drop filters <strong>based</strong> on smart<br />
device configuration”, Electron. Lett., 31 (23), pp. 2029-2030, November 1995.<br />
[149] Y. Tachikawa, Y. Inoue, M. Ishii, and T. Nozawa, “Arreyed-waveguide grating<br />
multiplexer with loop-back optical paths and its applications”, J. Lightw. Tech., 14 (6),<br />
pp. 977-984, June 1996.<br />
[150] K. Takada, Y. Inoue, H. Yamada, and M. Horiguchi, “Measurement of phase error<br />
distributions in silica-<strong>based</strong> arrayed-waveguide grating multiplexers by using Fourier<br />
transform spectroscopy”, Electron. Lett., 30 (20), pp. 1671-1672, September 1994.<br />
[151] H. Takahashi, S. Suzuki, K. Kato, and I. Nishi, “Arrayed-waveguide grating for<br />
<strong>wavelength</strong> division multi/demultiplexer with nanometre resolution”, Electron. Lett., 26<br />
(2), pp. 87-88, January 1990.<br />
[152] H. Takahashi, I. Nishi, and Y. Hibino, “10 GHz spacing optical frequency division<br />
multiplexer <strong>based</strong> on arrayed-waveguide grating”, Electron. Lett., 28 (4), pp. 380-382,<br />
February 1992.<br />
[153] H. Takahashi, Y. Hibino, and I. Nishi, “Polarization-insensitive arrayed-waveguide<br />
grating <strong>wavelength</strong> multiplexer on silicon”, Opt. Lett., 17 (7), pp. 499-501, April 1992.
References 155<br />
[154] H. Takahashi, Y. Hibino, Y. Ohmori, and M. Kawachi, “Polarization-insensitive arrayedwaveguide<br />
<strong>wavelength</strong> demultiplexer with birefringence compensating film”, IEEE Phot.<br />
Tech. Lett., 5 (6), pp. 707-709, June 1993.<br />
[155] H. Takahashi, K. Oda, H. Toba, and Y. Inoue, “Transmission characteristics of arrayed<br />
waveguide NxN <strong>wavelength</strong> multiplexer”, J. Lightw. Tech., 13 (3), pp. 447-455, March<br />
1995.<br />
[156] J.J.G.M. van der Tol, J.W. Pedersen, E.G. Metaal, Y.S. Oei, H. van Brug, and I. Moerman,<br />
“Mode evolution type polarization splitter on InGaAsP/<strong>InP</strong>”, IEEE Phot. Tech. Lett., 5<br />
(12), pp. 1412-1414, December 1993.<br />
[157] J.J.G.M. van der Tol, J.W. Pedersen, E.G. Metaal, Y.S. Oei, F.H. Groen, and I. Moerman,<br />
“Efficient short passive polarization converter”, in Procs. 7th Eur. Conf. on Int. Opt.<br />
(ECIO’95), Delft, The Netherlands, pp. 319-322, April 3-6, 1995.<br />
[158] J.J.G.M. van der Tol, J.W. Pedersen, E.G. Metaal, J.J.W. van Gaalen, Y.S. Oei, and F.H.<br />
Groen, “New metal free polarization splitter on InGaAsP/<strong>InP</strong>”, in Integrated Photonics<br />
Research (IPR’96), Boston, USA, pp. 456-459, April 29-May 2, 1996.<br />
[159] T. Uitterdijk, H. van Brug, F.H. Groen, H.J. Frankena, C.G.M. Vreeburg, and J.J.G.M.<br />
van der Tol, “Integrable polarization insensitive InGaAsP/<strong>InP</strong> Mach-Zehnder switch”, in<br />
Integrated Photonics Research (IPR’96), Boston, USA, pp. 486-489, April 29-May 2,<br />
1996.<br />
[160] H.G. Unger, Planar optical waveguides and fibres. Oxford Engineering Science Series,<br />
Clarendon Press, 1977.<br />
[161] A.R. Vellekoop and M.K. Smit, “Low-loss planar optical <strong>polarisation</strong> splitter with small<br />
dimensions”, Electron. Lett., 25, pp. 946-947, 1989.<br />
[162] A.R. Vellekoop and M.K. Smit, “A polarization <strong>independent</strong> planar <strong>wavelength</strong><br />
demultiplexer with small dimensions”, in Procs. Eur. Conf. on Opt. Integr. Systems at<br />
Amsterdam (ECIOSA’89), Amsterdam, The Netherlands, paper D3, September 25-28,<br />
1989.<br />
[163] A.R. Vellekoop and M.K. Smit, “Four-channel integrated-optic <strong>wavelength</strong><br />
demultiplexer with weak polarization dependence”, J. Lightw. Tech., 9 (3), pp. 310-314,<br />
March 1991.<br />
[164] B.H. Verbeek, A.A.M. Staring, E.J. Jansen, R. van Roijen, J.J.M. Binsma, T. van Dongen,<br />
M.R. Amersfoort, C. van Dam, and M.K. Smit, “Large bandwidth <strong>polarisation</strong><br />
<strong>independent</strong> and compact 8 channel PHASAR demultiplexer/filter”, in Optical Fiber<br />
Communication (OFC’94), San Jose, USA, Postdeadline Papers, pp. 63-66, February 20-<br />
25, 1994.<br />
[165] B. Verbeek and M.K. Smit, “Phased array <strong>based</strong> WDM devices”, in Procs. 21st Eur.<br />
Conf. on Opt. Comm. (ECOC’95), Brussels, Belgium, pp. 195-202, September 17-21,<br />
1995.<br />
[166] C.G.M. Vreeburg, C.R. de Boer, Y.S. Oei, B.H. Verbeek, R.T.H. Rongen, H. Vonk, M.R.<br />
Leys, and J.H. Wolter, “Strained <strong>InP</strong>/InGaAs quantum well layers for <strong>wavelength</strong><br />
<strong>demultiplexers</strong>”, in Procs. 7th Eur. Conf. on Int. Opt. (ECIO ‘95), Delft, The Netherlands,<br />
pp. 283-286, April 3-6, 1995.<br />
[167] C.G.M. Vreeburg, T. Uitterdijk, Y.S. Oei, M.K. Smit, F.H. Groen, J.J.G.M. van der Tol,<br />
P. Demeester, and H.J. Frankena, "Compact integrated <strong>InP</strong>-<strong>based</strong> add-drop multiplexer",<br />
in Procs. 22nd Eur. Conf. on Opt. Comm. (ECOC’96), Oslo, Norway, pp. 5.67-5.70,<br />
September 15-19, 1996.<br />
[168] R.G. Walker, “Simple and accurate loss measurement technique for semiconductor<br />
optical waveguides”, Electron. Lett., 21 (13), pp. 581-583, June 1985.
156 References<br />
[169] I.H. White, K.O. Nyairo, P.A. Kirkby, and C.J. Armistead, “Demonstration of a 1x2<br />
multichannel grating cavity laser for <strong>wavelength</strong> division multiplexing (WDM)<br />
applications”, Electron. Lett., 26 (13), pp. 832-834, June 1990.<br />
[170] H. Yamada, K. Takada, Y. Inoue, Y. Hibino, and M. Hroiguchi, “10 GHz-spaced arrayedwaveguide<br />
grating multiplexer with phase-error-compensating thin-film heaters”,<br />
Electron. Lett., 31 (5), pp. 360-361, March 1995.<br />
[171] H. Yamada, K. Takada, Y. Inoue, Y. Ohmori, and S. Mitachi, “Statically-phasecompensated<br />
10 GHz spaced arrayed-waveguide grating”, Electron. Lett., 32 (17), pp.<br />
1580-1582, August 1996.<br />
[172] D. Yap, L.M. Johnson, and G.W. Pratt, Jr., “Passive Ti:LiNbO 3 channel waveguide TE-<br />
TM mode splitter”, Appl. Phys. Lett., 44 (6), pp. 583-585, March 1984.<br />
[173] M.G. Young, U. Koren, B.I. Miller, M. Chien, T.L. Koch, D.M. Tennant, K. Feder, K.<br />
Dreyer, and G. Raybon, “Six <strong>wavelength</strong> laser array with integrated amplifier and<br />
modulator”, Electron. Lett., 31 (21), pp. 1835-1836, October 1995.<br />
[174] C.E. Zah, F.J. Favire, B. Pathak, R. Bhat, C. Caneau, P.S.D. Lin, A.S. Gozdz, N.C.<br />
Andreakakis, M.A. Koza, and T.P. Lee, “Monolithic integration of multi<strong>wavelength</strong><br />
compressive-strained multiquantum-well distributed-feedback laser array with star<br />
coupler and optical amplifiers”, Electron. Lett., 28 (25), pp. 2361-2362, December 1992.<br />
[175] F. Zamkotsian, H. Okamoto, K. Kishi, M. Okamoto, Y. Yoshikuni, K. Sato, and K. Oe,<br />
“An <strong>InP</strong>-<strong>based</strong> optical multiplexer for 100-GHz pulse-train generation”, IEEE Phot.<br />
Tech. Lett., 7 (5), pp. 502-504, May 1995.<br />
[176] M. Zirngibl, C. Dragone, and C.H. Joyner, “Demonstration of a 15x15 arrayed waveguide<br />
multiplexer on <strong>InP</strong>”, IEEE Phot. Tech. Lett., 4 (11), pp. 1250-1253, November 1992.<br />
[177] M. Zirngibl, C.H. Joyner, L.W. Stulz, Th. Gaiffe, and C. Dragone, “Polarisation<br />
<strong>independent</strong> 8x8 waveguide grating multiplexer on <strong>InP</strong>”, Electron. Lett., 29 (2), pp. 201-<br />
202, January 1993.<br />
[178] M. Zirngibl and C.H. Joyner, “12 frequency WDM laser <strong>based</strong> on a transmissive<br />
waveguide grating router”, Electron. Lett., 30 (9), pp. 701-702, April 1994.<br />
[179] M. Zirngibl, C.H. Joyner, L.W. Stulz, U. Koren, M.-D. Chien, M.G. Young, and B.I.<br />
Miller, “Digitally tunable laser <strong>based</strong> on a waveguide grating multiplexer and an optical<br />
amplifier”, IEEE Phot. Tech. Lett., 6 (4), pp. 516-518, April 1994.<br />
[180] M. Zirngibl, C.H. Joyner, and L.W. Stulz, “Demonstration of 9x200 Mbit/s <strong>wavelength</strong><br />
division multiplexed transmitter”, Electron. Lett., 30 (18), pp. 1484-1485, September<br />
1994.<br />
[181] M. Zirngibl, B. Glance, L.W. Stulz, C.H. Joyner, G. Raybon, and I.P. Kaminow,<br />
“Characterization of a multi<strong>wavelength</strong> waveguide grating router laser”, IEEE Phot.<br />
Tech. Lett., 6 (9), pp. 1082-1084, September 1994.<br />
[182] M. Zirngibl, C.H. Joyner, and L.W. Stulz, “WDM receiver by monolithic integration of<br />
an optical preamplifier, waveguide grating router and photodiode array”, Electron. Lett.,<br />
31 (7), pp. 581-582, March 1995.<br />
[183] M. Zirngibl, C.H. Joyner, and P.C. Chou, “Polarisation compensated waveguide grating<br />
router on <strong>InP</strong>”, Electron. Lett., 31 (19), pp. 1662-1664, September 1995.<br />
[184] M. Zirngibl, C.H. Joyner, C.R. Doerr, L.W. Stulz, and H.M. Presby, “A 18 channel multi<br />
frequency laser”, in Integrated Photonics Research (IPR’96), Boston, USA, pp. 128-131,<br />
April 29-May 2, 1996.
List of symbols<br />
List of symbols and acronyms<br />
αi starting angle of array waveguide i<br />
αϕ angular propagation loss<br />
β propagation constant in waveguide<br />
βFPR propagation constant in FPR<br />
βϕ angular propagation constant<br />
βt transformed angular propagation constant<br />
β0 propagation constant of fundamental mode<br />
β1 propagation constant of first-order mode<br />
C coupling coefficient<br />
c speed of light<br />
D dispersion<br />
da pitch of the array waveguides in the array aperture<br />
dr pitch of the receiver waveguides in the image plane<br />
d0 waist of Gaussian field<br />
Δα divergence angle of the array waveguides in the array aperture<br />
Δβ propagation constant difference<br />
Δfch channel spacing<br />
ΔfFSR free spectral range<br />
Δfpol TE-TM shift<br />
ΔfL L-dB bandwidth<br />
ΔΦ phase shift between adjacent array waveguides<br />
Δϕj,i phase difference between port j+1 and port j when exciting port i<br />
ΔL path length difference between adjacent array waveguides<br />
Δλ01 modal dispersion shift<br />
ΔNpol <strong>polarisation</strong> dispersion<br />
Δn index contrast<br />
Δspol <strong>polarisation</strong> shift along the image plane<br />
Er radial component of the electrical field<br />
Ex transverse component of the electrical field<br />
Eϕ angular component of the electrical field<br />
ε small loss term<br />
η coupling efficiency
158 List of symbols and acronyms<br />
fc central (design) frequency<br />
ϕj,i phase transfer from port i to port j<br />
ϕpol <strong>polarisation</strong> angle<br />
Hi height of array waveguide i<br />
i integer number<br />
j integer number<br />
j imaginary unit –<br />
1<br />
k integer number<br />
k0 wave number in vacuo<br />
kx wave number in the transverse direction<br />
ky wave number in the lateral direction<br />
L distance between the focal points<br />
LMMI length of MMI coupler<br />
La section length<br />
Lc coupling length<br />
Lo central channel insertion loss<br />
Lp propagation loss in the array<br />
Lu non-uniformity<br />
Lπ beat length<br />
li path length of array waveguide i<br />
λc central (design) <strong>wavelength</strong><br />
m order of the array<br />
m' order of the beam<br />
N number of channels<br />
Na number of array waveguides<br />
Neff effective index in waveguide<br />
Neff,t transformed effective index<br />
NFPR effective index in FPR<br />
Ng group index<br />
Nridge effective index in the ridge region<br />
Nstraight effective index in a straight waveguide<br />
NTE effective index for TE <strong>polarisation</strong><br />
NTM effective index for TM <strong>polarisation</strong><br />
Nx transverse effective index in the ridge<br />
N1 transverse effective index in the ridge<br />
N2 transverse effective index in the region next to the ridge<br />
n index distribution<br />
n<strong>InP</strong> refractive index of <strong>InP</strong> substrate<br />
nsub substrate index<br />
nt transformed index distribution<br />
nQ refractive index of the quaternary layer<br />
P power<br />
PTE power in TE field<br />
PTM power in TM field<br />
R radius of curvature<br />
Ra length of FPR<br />
Rco cut-off radius<br />
Ri radius of curvature of array waveguide i
List of symbols and acronyms 159<br />
Rt transformed radius of curvature<br />
r radial coordinate<br />
Si s<br />
straigth section length (including Ra ) of array waveguide i<br />
position along image plane<br />
T transmission matrix<br />
Tc transmission matrix of a curved waveguide<br />
Tcc transmission matrix of a junction between two curved waveguides<br />
Ts transmission matrix of a straight waveguide<br />
Tsc transmission matrix of a junction between a straight and a curved waveguide<br />
Tu transmission matrix of a U-bend<br />
θ dispersion angle<br />
θa angular width of the array aperture<br />
θο angular width of the far field<br />
U field distribution<br />
u transformed coordinate<br />
V lateral V-parameter (normalised film parameter)<br />
w waveguide width<br />
we effective width<br />
wwg waveguide width<br />
W waveguide width<br />
WMMI width of MMI coupler<br />
Wm cut-off width of the lateral mode m<br />
ΔW waveguide width deviation<br />
List of acronyms<br />
ACTS Advanced Communications Technologies and Services<br />
ADM Add-Drop Multiplexer<br />
AR Anti-Reflection<br />
ASE Amplified Spontaneous Emission<br />
AW Array Waveguides<br />
BPM Beam Propagation Method<br />
CAD Computer-Aided Design<br />
CATV Cable Television<br />
CCD Charge-Coupled Device<br />
CT Cross Talk<br />
DBR Distributed Bragg Reflector<br />
DEMUX Demultiplexer<br />
DFB Distributed Feedback<br />
DH Double Hetero<br />
EIM Effective Index Method<br />
EDFA Erbium-Doped Fibre Amplifier<br />
FEM Finite Element Method<br />
FPR Free Propagation Region<br />
FSR Free Spectral Range<br />
MAGIC Multistripe Array Grating Integrated Cavity<br />
MDS Microwave Design System
160 List of symbols and acronyms<br />
MMI Multimode Interference<br />
MOCVD Methal-Organic Chemical Vapour Deposition<br />
MOL Method Of Lines<br />
MQW Multiple Quantum Wells<br />
MSI Medium-Scale Integration<br />
MUX Multiplexer<br />
MW Multi-<strong>wavelength</strong><br />
MZI Mach-Zehnder Interferometer<br />
NA Numerical Aperture<br />
OCC Optical Cross Connect<br />
OMVPE Organo-Metallic Vapour-Phase Epitaxy<br />
PECVD Plasma-Enhanced Chemical Vapour Deposition<br />
PHASAR Phased Array<br />
POC Philips Optoelectronics Centre<br />
Pt Platinum<br />
Q Quaternary material (= InGaAsP)<br />
RACE Research and technology development in Advanced Communications<br />
technologies in Europe<br />
RIE Reactive Ion Etching<br />
RX Receiver<br />
SEM Scanning Electron Microscope<br />
SOA Semiconductor Optical Amplifier<br />
SSI Small-Scale Integration<br />
TE Transverse Electric<br />
TM Transverse Magnetic<br />
TOBASCO Towards Broadband Access Systems for CATV Optical Networks<br />
TR Transition Region<br />
TX Transmitter<br />
WDM Wavelength Division Multiplexing<br />
WG Whispering Gallery<br />
X Switch
Summary<br />
Wavelength (de)multiplexers are key components in optical networks. Research on integrated<br />
optic (de)multiplexers has been focused on phased-array (PHASAR) <strong>based</strong> devices, which<br />
appear to be robust and production tolerant because they can be produced in conventional<br />
waveguide technology and require a relatively simple manufacturing process. In this thesis, a<br />
comprehensive analysis of the operation and design of PHASAR <strong>demultiplexers</strong> is presented,<br />
in which special attention has been paid to specific design requirements such as <strong>polarisation</strong><br />
independence and low-loss operation. PHASAR <strong>demultiplexers</strong> presented in this thesis are<br />
produced on an <strong>InP</strong> substrate, as this material allows for small device dimensions and for<br />
monolithic integration with active components such as for instance detectors and optical<br />
amplifiers.<br />
The major part of this thesis is focused on methods for making PHASARs <strong>polarisation</strong><br />
<strong>independent</strong>. As the state of <strong>polarisation</strong> of the transmitted light is unknown after propagating<br />
through fibre and as it varies in time, it is important that the transfer of PHASAR<br />
<strong>demultiplexers</strong> are insensitive to such variations. One way of making a PHASAR <strong>polarisation</strong><br />
<strong>independent</strong> is by inserting <strong>polarisation</strong> converters in the middle of the array. It was found that<br />
waveguide bends with a very small radius, produced in a deeply etched waveguide structure,<br />
have <strong>polarisation</strong>-converting properties, an extensive analysis of which is given in this thesis.<br />
This work resulted in a novel type of <strong>polarisation</strong> converter with high conversion, low loss, and<br />
of compact size.<br />
Another way to make <strong>demultiplexers</strong> <strong>polarisation</strong> <strong>independent</strong> is by using <strong>polarisation</strong><br />
<strong>independent</strong> waveguides. Experiments, performed in cooperation with Philips Optoelectronics<br />
Centre (POC), demonstrate that with this type of waveguide, compact devices can be made<br />
with low on-chip loss and good cross talk values. Additionally such a device has been<br />
integrated with photodetectors and packaged, which led to the first optical receiver mounted in<br />
a compact, industry-standard butterfly package.<br />
A third method for making PHASARs <strong>polarisation</strong> <strong>independent</strong> is by compensating the<br />
<strong>polarisation</strong> dispersion of the array waveguides by inserting a waveguide section with a<br />
different <strong>polarisation</strong> dispersion. Although this method is sensitive to width and layer<br />
variations of the waveguide structure, it is relatively easy from a production point of view.<br />
Experiments show that application of this method leads to <strong>polarisation</strong> independence at the<br />
cost of a small increase of the device length (less than 1 mm).<br />
Furthermore two other methods to obtain <strong>polarisation</strong> independence are discussed. The first<br />
makes use of a compact <strong>polarisation</strong> converter (as discussed earlier) inserted in the middle of<br />
the PHASAR demultiplexer. The second method uses a <strong>polarisation</strong> splitter at the input of the
162 Summary<br />
PHASAR demultiplexer. Both methods are described in this thesis.<br />
If the free propagation regions of the PHASAR demultiplexer are replaced with multimode<br />
interference (MMI) couplers, a special type of phased-array demultiplexer is obtained. In this<br />
way, the number of array waveguides can be reduced to a number equal to the number of <strong>wavelength</strong><br />
channels. Additionally, due to the uniform splitting ratio and low insertion loss of the<br />
MMI couplers, low insertion loss of the device is expected. A number of experiments have<br />
been performed, one of which uses a deeply etched waveguide structure. This allows for the<br />
use of very small waveguide bends (radius 50 μm) and leads to the smallest demultiplexer ever<br />
reported.<br />
A single MMI coupler can also be used as a mode splitter. Such a device can be applied for<br />
filtering higher order modes. The advantage of such a MMI-<strong>based</strong> mode splitter is the ease of<br />
manufacture and its small size. Analysis and experiments presented in this thesis show that a<br />
MMI-<strong>based</strong> mode splitter is very tolerant with respect to variations in <strong>wavelength</strong>, width, and<br />
layer thickness.<br />
Finally, the field distribution of the infrared light propagating in a waveguide has been made<br />
visible for the first time, with a resolution below the theoretical diffraction limit. This was<br />
possible using a two-step upconversion mechanism of erbium atoms incorporated in an<br />
aluminum oxide waveguide structure. If two sequential upconversion processes take place,<br />
visible green light is emitted. As the process depends on the concentration of excited erbium<br />
ions (which in turn depends on the intensity of the field distribution) the emission of green<br />
light is a replica of the intensity distribution in the waveguide. The field distribution of the<br />
infrared light in a MMI coupler has been measured. It has been demonstrated that the measured<br />
data match very well with calculated intensity profiles.
Samenvatting<br />
Golflengte (de)multiplexers spelen een sleutelrol in optische netwerken. Onderzoek op het<br />
gebied van golflengte (de)multiplexers heeft zich met name gericht op componenten gebaseerd<br />
op een phased-array (PHASAR). Deze lijkt robuust en fabricage-tolerant te zijn omdat die in<br />
een conventionele golfgeleidertechnologie gemaakt kan worden met behulp van een relatief<br />
eenvoudig fabricageproces. In dit proefschrift wordt een uitvoerige analyse van de werking en<br />
het ontwerp van PHASAR <strong>demultiplexers</strong> beschreven, waarbij speciale aandacht besteed<br />
wordt aan specifieke ontwerp-eisen zoals polarisatie-onafhankelijkheid en lage verliezen. Alle<br />
PHASAR <strong>demultiplexers</strong> in dit proefschrift zijn vervaardigd op <strong>InP</strong> substraat, hetgeen de<br />
mogelijkheid biedt tot de realisatie van kleine component-afmetingen, en tot monolithische<br />
integratie met actieve componenten, zoals detectoren en optische versterkers.<br />
Het leeuwendeel van dit proefschrift is gericht op methoden om PHASAR’s polarisatieonafhankelijk<br />
te maken. Aangezien de polarisatietoestand van het verzonden licht na<br />
propagatie door de glasvezel onbekend is, en tevens varieert in de tijd, is het belangrijk dat de<br />
response van PHASAR <strong>demultiplexers</strong> daarvoor niet gevoelig is. Een methode om PHASAR’s<br />
polarisatie-onafhankelijk te maken is door polarisatie-omzetters in het midden van de reeks<br />
golfgeleiders te plaatsen. Het blijkt dat golfgeleiderbochten met een kleine bochtstraal in een<br />
diepgeëtste golfgeleiderstructuur, polarisatiedraaiende eigenschappen hebben, die uitgebreid<br />
in dit proefschrift worden beschreven. Dit werk leidde tot de realisatie van een nieuw type<br />
polarisatie-omzetter met een hoge omzetting, laag verlies en compacte afmetingen.<br />
Een andere methode om PHASAR’s polarisatie-onafhankelijk te maken is door gebruik te<br />
maken van polarisatie-onafhankelijke golfgeleiders. Experimenten, in samenwerking met<br />
Philips Optoelectronics Centre (POC) uitgevoerd, laten zien dat op deze manier compacte<br />
componenten gemaakt kunnen worden, die lage verliezen en weinig kanaaloverspraak hebben.<br />
Daarnaast is een demultiplexer geintegreerd met detectoren en in een behuizing gemonteerd,<br />
hetgeen leidde tot de eerste optische ontvanger die gemonteerd is in een compacte, industriestandaard<br />
butterfly behuizing.<br />
Een derde methode voor het verkrijgen van polarisatie-onafhankelijke PHASAR’s is door het<br />
compenseren van de polarisatiedispersie van de reeks golfgeleiders door middel van het<br />
plaatsen van een sectie met een andere polarisatiedispersie. Alhoewel deze methode gevoelig is<br />
voor breedte- en diktevariaties in de golfgeleider, kan het op een vanuit het oogpunt van<br />
fabricage eenvoudige wijze toegepast worden. Experimenten laten zien dat polarisatieonafhankelijkheid<br />
gerealiseerd kan worden ten koste van slechts een kleine toename van de<br />
componentlengte (minder dan 1 mm).<br />
Verder worden twee additionele methoden voor polarisatie-onafhankelijkheid besproken. De
164 Samenvatting<br />
eerste maakt gebruik van een compacte polarisatie-omzetter (zoals eerder beschreven), die in<br />
het midden van de PHASAR demultiplexer geplaatst wordt. Bij de tweede methode worden de<br />
polarisaties gescheiden aan de ingang van de PHASAR demultiplexer met behulp van een<br />
polarisatiescheider. Beide methoden worden in dit proefschrift beschreven.<br />
Indien de vrije-propagatie regio’s van de PHASAR demultiplexer vervangen worden door multimode<br />
interferentie (MMI) koppelaars, wordt een speciale variant van de phased-array demultiplexer<br />
verkregen. Op deze manier kan het aantal array-golfgeleiders gereduceerd worden tot<br />
het aantal golflengtekanalen. Verder wordt laag verlies van het gehele component verwacht,<br />
gezien de uniforme vermogensdeling en de lage verliezen van de MMI koppelaars. Een aantal<br />
experimenten zijn uitgevoerd, waarvan één in een diep geëtste golfgeleider- structuur. Hierdoor<br />
kunnen zeer kleine bochten gebruikt worden (bochtstraal 50 μm), hetgeen er toe leidde dat de<br />
kleinste demultiplexer ooit gerapporteerd kon worden gerealiseerd.<br />
Een MMI koppelaar kan ook gebruikt worden voor het scheiden van modes. Hiermee kunnen<br />
dan hogere orde modes weggefilterd worden. Het voordeel van dergelijke scheiders is dat ze<br />
eenvoudig vervaardigd kunnen worden met compacte afmetingen. Analyse en experimenten in<br />
dit proefschrift tonen aan dat modescheiders gebaseerd op een MMI koppelaar ongevoelig zijn<br />
voor variaties in de golflengte, breedte en dikte.<br />
Tenslotte is voor de eerste keer de infrarode lichtverdeling in een golfgeleider zichtbaar<br />
gemaakt met een resolutie die onder de theoretische diffractielimiet ligt. Dit bleek mogelijk<br />
door een twee-staps opconversie mechanisme te gebruiken van erbium-atomen, die in een<br />
aluminiumoxide golfgeleiderstructuur ingebracht zijn. Indien er twee van zulke conversies<br />
achtereenvolgend plaatsvinden, wordt er zichtbaar groen licht uitgezonden. Aangezien het<br />
proces afhankelijk is van de concentratie aangeslagen erbium-atomen, dat op zijn beurt weer<br />
afhankelijk is van de veldverdeling in de golfgeleider, is de hoeveelheid uitgezonden groen<br />
licht een replica van de veldverdeling. De veldverdeling van het infrarode licht in een MMI<br />
koppelaar is gemeten, en er is aangetoond dat de meetresultaten goed overeenstemmen met<br />
berekende veldverdelingen.
Dankwoord<br />
Waarschijnlijk is het dankwoord één van de meest moeilijke hoofdstukken om te schrijven,<br />
omdat je het risico loopt dat je iemand, die bijgedragen heeft aan het tot stand komen van dit<br />
proefschrift, vergeet te bedanken. Omdat zoveel mensen een bijdrage geleverd hebben, zal het<br />
moeilijk zijn om dit te voorkomen en daarom wil ik bij deze een ieder bedanken voor de<br />
samenwerking. Een aantal mensen wil ik echter in het bijzonder noemen.<br />
De meeste dank ben ik verschuldigd aan Meint Smit, die met zijn creativiteit en opbouwende<br />
kritiek een grote bijdrage heeft geleverd aan het tot stand komen van dit proefschrift. Bart<br />
Verbeek wil ik bedanken voor het opstarten van het onderzoek. Hans Blok, mijn promotor,<br />
zorgde met zijn invalshoek voor een verhelderende kijk op zaken, hetgeen ook de leesbaarheid<br />
van dit proefschrift verhoogde. Siang Oei, met wie ik gedurende een jaar op plezierige wijze de<br />
kamer gedeeld heb, was onmisbaar om mij van alles over de InGaAsP technologie te leren.<br />
Alle devices die in dit proefschrift zijn besproken, konden alleen maar gerealiseerd worden<br />
dankzij een bijdrage van de technici en daarom wil ik ze speciaal bedanken. Tom Scholtes en<br />
Liang Shi deponeerden siliciumnitridelagen, waarop Aad de Vreede en Koos van Uffelen de<br />
lithografie verzorgden, hetgeen niet altijd even eenvoudig was vanwege mijn eisen ten opzichte<br />
van de maatvoering. Samen met Ed Metaal (KPN Research) was Frans van Ham een goede<br />
afmaker met zijn onnavolgbaar etswerk, alhoewel je het klieven van de samples maar beter niet<br />
door hem kunt laten verzorgen voordat hij een kop koffie op heeft. Natuurlijk was dit alles<br />
onmogelijk zonder de kwaliteitsmaskers gemaakt door Otto Meijer (Philips Research Masker<br />
Centrum) en later door Fokke Groen. Ab Kuntze en Adrie Looyen zorgden voor de “finishing<br />
touch” met hun anti-reflectie coatings. Laatstgenoemde wil ik nog speciaal bedanken samen<br />
met Aad van der Lingen voor het masker schrijven en goud sputteren voor de traliedemultiplexer.<br />
Lucas Soldano heeft mij de principes van de MMI koppelaar bijgebracht. Martin Amersfoort<br />
leerde mij alles over fotodiodes, en samen hebben wij de eerste polarisatie-onafhankelijke<br />
PHASAR ontworpen gebaseerd op de strip golfgeleiderstruktuur. Kees Steenbergen, Leo<br />
Spiekman en Kees Vreeburg, met wie ik veel waardevolle discussies heb gevoerd, waren fijne<br />
collega’s om mee samen te werken, ondanks het feit dat de eerste twee stiekem getrouwd zijn.<br />
Wellicht zal laatstgenoemde het ook doen? Leo de Vreede heeft me vanaf het begin, al toen ik<br />
nog aan het afstuderen was, geholpen met MDS, en later werd hij verlost van die taak door<br />
Xaveer Leijtens. Met hem als kamergenoot was het prettig werken, niet in de laatste plaats<br />
door het verminderen van het aantal keren koffie halen. Patrice Le Lourec leverde een<br />
waardevolle bijdrage door het PHASAR simulatie programma verder te ontwikkelen en in<br />
MDS te implementeren. Verder was het een genoegen om Chretien Herben, die mij van
166 Dankwoord<br />
experimentele data voorzag, Peter Harmsma en Annett Siefke als collega te hebben.<br />
Mijn speciale dank gaat uit naar Toine Staring, Hans Binsma en Edwin Jansen van het Philips<br />
Optoelectronics Centre voor de vruchtbare samenwerking die leidde tot de succesvolle<br />
realisatie van de op stripgolfgeleiders gebaseerde polarisatie-onafhankelijke phased-array<br />
<strong>demultiplexers</strong>. Jos van der Tol en Jørgen Pedersen van KPN Research worden hartelijk<br />
bedankt voor de discussies en suggesties. Tenslotte dank ik Gerlas van den Hoven, Albert<br />
Polman en Pieter Kik van Amolf voor de samenwerking die heel wat publikaties opgeleverd<br />
heeft, en - ook niet onbelangrijk - veel mooie foto’s van naar zichtbaar licht geconverteerde<br />
veldverdelingen van infrarood licht in golfgeleiders (zie onder andere de omslag).<br />
Een flink aantal studenten hebben mij geholpen, van wie ik er twee in het bijzonder wil<br />
noemen. Loek van der Helm was de eerste die ik mocht begeleiden. Samen hebben we de<br />
eerste (en, helaas, ook de laatste) tralie-demultiplexer van het lab ontworpen. Michiel ten Kate<br />
heeft een goede bijdrage geleverd door zijn grondige aanpak van de metingen aan de MMI-<br />
MZI <strong>demultiplexers</strong>, waardoor hij mij behoorlijk wat werk uit handen genomen heeft. Verder<br />
waren daar nog Marco Boeren, Bart van Geest, Marco Kroonwijk, Robert Chandler, Marcel<br />
Schaar, Martin Bouda en Richard Verhaar, die ik heb mogen begeleiden en mij veel werk uit<br />
handen hebben genomen.<br />
Mia van der Voort en Wendy van Schagen wil ik bedanken voor hun efficiente “office<br />
management”, en, natuurlijk, de gezelligheid die hun aanwezigheid met zich meebracht. Ook<br />
hebben ze altijd de uitstapjes van de vakgroep tot in de puntjes georganiseerd.<br />
In het bijzonder wil ik Renée bedanken, omdat zij er altijd voor mij is. Zij zorgde voor een<br />
rustige thuishaven en bracht vele avonden in eenzaamheid door, waardoor ze het op die manier<br />
mogelijk maakte dat ik dit proefschrift kon voltooien. In de nabije toekomst zal ik haar, en<br />
Marloes en Menno, de vrije tijd goedmaken die ik niet met hen heb kunnen doorbrengen.<br />
Annemieke (“Micky”) bedank ik omdat ze dit proefschrift volledig heeft doorgeworsteld op<br />
zoek naar eventuele fouten en mogelijke verbeteringen ter verhoging van de leesbaarheid. Ook<br />
ben ik mijn ouders dankbaar voor het feit dat zij mij gestimuleerd hebben te leren en mijn<br />
horizon te verbreden. Tenslotte dank ik al mijn vrienden: ook al hebben ze niets bijgedragen<br />
aan dit proefschrift, ze zorgden in ieder geval voor de nodige ontspanning.
Biography<br />
Cor van Dam was born in The Hague, the Netherlands, on 10th May 1967. After completing<br />
his secondary education at “Maerlant Lyceum” in 1985, he started studying electrical<br />
engineering at the Delft University of Technology, receiving his master’s degree in December<br />
1990. His thesis research was carried out in the Integrated Optics group of the Laboratory of<br />
Telecommunication and Remote Sensing Technology and concerned the development of an<br />
CAD-tool for the design and analysis of photonic integrated circuits, using a microwave design<br />
system (Hewlett-Packard’s MDS). In January 1991 he was appointed to the scientific staff of<br />
the laboratory where he started his PhD research, the result of which lies in front of you. Since<br />
March 1997, he has been working as a postdoctoral research scientist in the area of mobile and<br />
satellite communications at the Physics and Electronics Laboratory of TNO (TNO-FEL) in<br />
The Hague.