WAVELENGTH MULTIPLEXING BY SPATIAL BEAM SHIFTING
IN MULTILAYER THIN-FILM STRUCTURES
A DISSERTATION
SUBMITTED TO THE DEPARTMENT OF ELECTRICAL ENGINEERING
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
Martina Gerken
March 2003
ii
© Copyright by Martina Gerken 2003
All Rights Reserved
iii
I certify that I have read this dissertation and that, in my opinion, it is fully adequate in
scope and quality as a dissertation for the degree of Doctor of Philosophy.
____________________________________
David A. B. Miller, Principal Advisor
I certify that I have read this dissertation and that, in my opinion, it is fully adequate in
scope and quality as a dissertation for the degree of Doctor of Philosophy.
____________________________________
Olav Solgaard
I certify that I have read this dissertation and that, in my opinion, it is fully adequate in
scope and quality as a dissertation for the degree of Doctor of Philosophy.
____________________________________
Shanhui Fan
Approved for the University Committee on Graduate Studies.
iv
v
AbstractWavelength Division Multiplexing (WDM) systems allow the transmission of multiple
channels over a single fiber by encoding each channel with a different optical
wavelength. Compact and cost-effective wavelength multiplexing and demultiplexing
devices are needed for combining the different WDM channels on the transmitter side
and splitting them at the receiver. This dissertation investigates the use of a single
multilayer thin-film stack with high spatial dispersion for multiplexing or demultiplexing
multiple WDM channels by spatial beam shifting. The thin-film stack is designed such
that multiplexed light incident at an angle experiences a wavelength-dependent effective
group propagation angle in the stack. This translates to a wavelength-dependent spatial
beam shift and demultiplexing at the output surface.
We introduce four different types of thin-film stacks with high spatial dispersion:
Periodic stacks using the “superprism effect” in one-dimensional photonic crystals,
chirped stacks exploiting a wavelength-dependent penetration depth, resonator stacks
with dispersion due to stored energy, and numerically optimized non-periodic stacks
utilizing a mixture of the two previous dispersion effects. The experimental results of a
200-layer periodic stack and a 66-layer non-periodic stack are discussed and compared.
Because of its greater design freedom, the non-periodic stack gives both a linear shift
with wavelength, and a larger usable shift than the thicker periodic stack.
Multiple bounces off the stack can be performed to increase the spatial beam shift. Using
eight bounces off the 66-layer stack, a nearly linear 100-µm shift is achieved between 827
and 841 nm. This shift is sufficient to separate four channels by their Gaussian beam
widths. We discuss that the number of separable channels is proportional to the total
beam shift. Investigating over 600 different stacks, we develop a heuristic model
predicting the maximum shift for a given stack thickness, material system, and incidence
angle. From this model we find that the multiplexing of eight to sixteen WDM channels
using a single thin-film stack with high spatial dispersion seems well possible.
vi
vii
AcknowledgementsFirst and foremost, I wish to thank my research advisor, Prof. David Miller, without
whom this work would never have happened or been possible. He got me started on
investigating the spatial dispersion of periodic thin-film stacks for wavelength
multiplexing. He extensively helped me during my first quarter at Stanford to figure out
the basic physics. In my further research he helped me whenever I asked, and gave me
the freedom to try out new ideas on my own. His continuous support as well as his
continuous challenge for more physical intuition helped me to advance in my research
and strive for more insight. Thank you very much!
Thanks to Prof. Rafael Piestun for discussing ideas with me and providing physical
insight during the initial research on the superprism effect in photonic crystals. Bianca
Keeler and I worked together on demonstrating the spatial dispersion of periodic stacks.
Her experimental expertise enabled us to quickly obtain first experimental proof of the
superprism effect. Prof. Franz Kärtner gave me the opportunity to spend the summer of
2001 in his research group at the University of Karlsruhe. I appreciated the discussions
on temporal and spatial dispersion, and gained more insight into the design of double-
chirped stacks. Thanks also to my reading committee, Prof. Shanhui Fan and Prof. Olav
Solgaard, as well as the chair of my defense committee, Prof. Brian Wandell, for their
interest in my work and their helpful questions and comments.
On the fabrication side, I am indebted to several people whose help allowed me to
transfer this work from a mere theoretical exercise to an actual device. I would like to
thank Chien-Chung Lin for quickly growing our first periodic stack that allowed us to
collect proof-of-principle experimental results. Many thanks to Petar Atanackovic and
Glenn Solomon for teaching me how to grow thin-film stacks in an MBE. Their vast
knowledge of fabrication issues allowed me to better judge what can be fabricated and
what is not reasonable. I wish to thank Andrew Clark, Anmin Zheng, and Phil Anthony
of JDS Uniphase for taking the time in their busy schedule to fabricate both the 200-layer
periodic stack and the 66-layer non-periodic stack presented in this thesis. I especially
viii
appreciated discussions with Andrew on how to design stacks with less stringent
fabrication tolerances. Finally, thanks to Helen Kung and Tom Carver for providing me
with reflective coatings for the thin-film stacks.
Thanks to everyone in the Miller group: Diwakar Agarwal, Hatice Altug, Sameer
Bhalotra, Aparna Bhatnagar, Ray Chen, Henry Chin, Christof Debaes, Volkan Demir,
Onur Fidaner, Noah Helman, Yang Jiao, Bianca Keeler, Gordon Keeler, Helen Kung, Jon
Roth, Vijit Sabnis, Liang Tang, Ryohei Urata, Michael Wiemer, Micah Yairi, and Wei
Zhou. I appreciated your help as well as our various discussions about research and life in
general. Thanks to each one of you for making research fun and bringing delicious food
to the group meetings! Thanks also to the “Mezzanine people” for the great work
environment. On the administrative side, I would like to thank Ingrid Tarien for
guaranteeing that everything runs smoothly in the Miller group.
I greatly appreciated the support of a Sequoia Capital Stanford Graduate Fellowship. This
fellowship not only provided the major part of my financial support at Stanford, but also
allowed me to meet students from other research groups and other departments as well as
the donors of the fellowship. This gave me a broader perspective and enriched my stay at
Stanford.
Before coming to Stanford, I had the fortune to work for Greg Faris at SRI International.
He taught me the art of optical experiments and how to perform successful research.
Much of my decision to pursue a doctoral degree was based on my great research
experience at SRI. Many thanks to Greg in particular as well as everyone else at the
Molecular Physics Lab.
I am grateful to my family and friends for their support in my research endeavors and for
reminding me that there are things in life other than research. Finally, I want to thank my
husband Ingo. He never tires of reading my research writings nor of listening to my
practice talks. He never complains about my strange work hours and always encouraged
me in my research. Thank you for your constant support!
ix
Table of Contents
Chapter 1 Introduction........................................................................................... 1
1.1 Current MUX/DEMUX Devices ............................................................................ 2
1.2 Thin-Film Spatial MUX/DEMUX.......................................................................... 5
References ...................................................................................................................... 6
Chapter 2 Designing Dispersive Structures......................................................... 8
2.1 “Superprism Effect” in Photonic Crystals .............................................................. 9
2.2 Periodic versus Non-Periodic Structures .............................................................. 10
2.3 Designs for Temporal Dispersion Compensation ................................................. 12
2.3 Thin-Film Filter Design Techniques..................................................................... 13
References .................................................................................................................... 14
Chapter 3 Superprism Effect in 1-D Photonic Crystals .................................. 18
3.1 Superprism Effect Theory..................................................................................... 18
3.2 Experimental Results ............................................................................................ 21
3.3 Finite Number of Periods and Finite Beam Width ............................................... 23
3.4 Improved Superprism Structures .......................................................................... 32
References .................................................................................................................... 36
Chapter 4 Physics of Spatial and Temporal Dispersion................................... 38
4.1 Relating Spatial and Temporal Dispersion ........................................................... 39
4.2 Relationship between Dispersion and Stored Energy........................................... 42
4.3 Sample Structures Verifying the Proportionality of Stored Energy,
Group Delay, and Spatial Shift ............................................................................. 44
4.4 Influence of Material Dispersion .......................................................................... 53
References .................................................................................................................... 59
x
Chapter 5 Chirped Stacks .................................................................................. 61
References .................................................................................................................... 68
Chapter 6 Resonator Stacks............................................................................... 69
References .................................................................................................................... 80
Chapter 7 Numerically Optimized Stacks ........................................................ 82
7.1 Designing Stacks using Numerical Optimization ................................................. 83
7.2 Experimental Results for an Optimized Stack ...................................................... 88
7.3 Comparison of Periodic and Non-Periodic Designs ............................................. 90
7.4 Experiment: 4-Channel Wavelength Demultiplexer............................................. 92
References .................................................................................................................... 94
Chapter 8 Maximum Number of Channels ...................................................... 95
8.1 Number of Volume Modes ................................................................................. 100
8.2 Verification of Volume Mode Model ................................................................. 106
8.3 Number of Surface Modes .................................................................................. 108
8.4 Crosstalk between Modes ................................................................................... 111
8.5 Number of Surface Modes including Bounces ................................................... 114
8.6 Verification of Surface Mode Model.................................................................. 119
8.7 Designing for a Maximum Number of Modes.................................................... 123
8.8 Discussion of the 4-channel DEMUX in 7.4 ...................................................... 127
References .................................................................................................................. 132
Chapter 9 Maximum Shift ............................................................................... 133
9.1 Automatic Generation of Designs....................................................................... 136
9.1.1 Automating the Refinement Process........................................................... 137
9.1.2 Automatic Refinement of Fixed Start Designs ........................................... 138
9.1.3 Automatic Generation of Starting Designs ................................................. 141
9.2 Deriving a Heuristic Shift Model........................................................................ 143
xi
9.3 Constancy of Dispersion×Wavelength-Range-Product ...................................... 149
9.4 Model for the Maximum Channel Number ........................................................ 151
References .................................................................................................................. 152
Chapter 10 Applications of Spatially Dispersive Stacks.................................. 154
10.1Wavelength Multiplexing and Demultiplexing .................................................. 154
10.2Step-Like Beam Shifting..................................................................................... 160
10.3Temporal vs. Spatial Dispersion ......................................................................... 167
10.4Beam Steering..................................................................................................... 168
10.5Other Applications Using Beam Shifting ........................................................... 169
References .................................................................................................................. 170
Chapter 11 Conclusions...................................................................................... 171
Appendix A Bloch Calculation........................................................................... 175
References .................................................................................................................. 177
Appendix B Coordinate Transformation (K, β, ω)→(K, θ, ω)......................... 178
B.1 Group Propagation Angle in Terms of (K, β, ω) ................................................ 179
B.2 Group Propagation Angle in Terms of (K, θ, ω) ................................................ 181
References .................................................................................................................. 182
Appendix C Transfer Matrix Calculation......................................................... 183
References .................................................................................................................. 195
Appendix D Simulating Beams by Fourier Decomposition............................. 196
References .................................................................................................................. 205
Appendix E Numerical Optimization Methods ................................................ 207
References .................................................................................................................. 216
xii
Appendix F Beam Cone in a Dispersive Stack ................................................. 218
References .................................................................................................................. 223
Appendix G Composition of Structures ............................................................ 224
xiii
List of Tables
Chapter 6
Table 6.1. Reflection coefficients for different λB/4-layer combinations
(λB=980nm) of the substrate S (ns=1.52), high index material H
(nH=2.06, dH= λB/4nH), and low index material L (nL=1.456,
dL= λB/4nL) at 860nm, 54° incidence angle, and s-polarization................ 76
Chapter 8
Table 8.1. Wavelength, low angle, and high angle of the different modes for the
example in Fig. 8.9.................................................................................. 107
Table 8.2. Wavelength, group propagation angle in the structure for the input
angle plus the half cone angle (Angle+), group propagation angle
minus the half cone angle (Angle-), position where the plus-angle
crosses the back interface (Position+), and position where the minus-
angle crosses the back interface (Position-) for the different modes for
the example in Fig. 8.16.......................................................................... 120
Table 8.3. Wavelength, group propagation angle in the structure for the input
angle plus the half cone angle (Angle+), group propagation angle
minus the half cone angle (Angle-), position where the plus-angle
crosses the back interface (Position+), and position where the minus-
angle crosses the back interface (Position-) for the different modes for
the example in Fig. 8.21.......................................................................... 125
xiv
Chapter 9
Table 9.1. Comparison of linear shift characteristics obtained with the double-
chirped stacks in Chapter 5, the coupled-cavity stack in Chapter 6,
and the numerically optimized stack in Chapter 7.................................. 134
Chapter 10
Table 10.1. Relationship between spot size, angular range, and wavelength range
at 1540 nm, 45º incidence angle, and ∂ω/∂β=0.3c . ............................... 164
Appendix E
Table E.1. Comparison of refinement methods. ........................................................ 215
Appendix G
Table G.1. Layer Composition of the structures appearing in Chapters 3 to 5.......... 227
Table G.2. Layer Composition of the structures appearing in Chapters 6 to 10........ 230
xv
List of Illustrations
Chapter 1
Fig. 1.1. Schematic of a Wavelength Division Multiplexing (WDM) system............... 2
Fig. 1.2. Traditional MUX/DEMUX technologies. (a) Prism. (b) Grating.................... 3
Fig. 1.3. State-of-the-art MUX/DEMUX technologies. (a) Arrayed waveguide
grating (AWG). (b) Fiber Bragg grating (FBG). (c) Thin-film filter
(TFF). .......................................................................................................... 4
Fig. 1.4. Thin-film spatial MUX/DEMUX. ................................................................... 5
Chapter 2
Fig. 2.1. Operating schematics of four different types of thin-film structures that
can be used for demultiplexing multiple wavelengths by spatial beam
shifting. The structure in (a) is periodic; (b) – (d) are non-periodic
structures. (a) Superprism effect in a one-dimensional photonic crystal
(combined here with a simple reflection off of the right face). (b)
Wavelength-dependent penetration depth. (c) Wavelength-dependent
number of roundtrips in the structure. (d) Combination of wavelength-
dependent penetration depth and number of roundtrips. ............................ 9
Chapter 3
Fig. 3.1. Schematic of the superprism effect................................................................ 18
Fig. 3.2. 1-D photonic crystal operated in transmission (a) or reflection (b)............... 20
Fig. 3.3. Schematic of the experimental setup. ............................................................ 21
xvi
Fig. 3.4. Experimentally observed intensity on a CCD trace as a function of
position and wavelength for a 100-period one-dimensional photonic
crystal........................................................................................................ 22
Fig. 3.5. Comparison of the theoretically expected and experimentally observed
shift of the beam center position as a function of wavelength for a
100-period one-dimensional photonic crystal........................................... 23
Fig. 3.6. Reflectance and shift as a function of wavelength predicted using the
transfer matrix method for plane waves.................................................... 24
Fig. 3.7. Group propagation angle as a function of wavelength calculated using
Bloch theory with incidence angles of 40° (solid), 36.5° (dash), and
43.5° (dash-dot). For a Gaussian beam with a spot size of 4.7 µm at
880 nm and a center incidence angle of 40°, the intensity has
decreased to 1/e2 for beam components at incidence angles of 36.5°
and 43.5°. .................................................................................................. 27
Fig. 3.8. Intensity reflected from the 100-period stack as a function of wavelength.
Red shows the experimental data and blue the simulation results
obtained using the transfer matrix method and Fourier decomposition
of a Gaussian beam. .................................................................................. 29
Fig. 3.9. Normalized intensity (a), spot size along interface (b), and beam shift (c)
for the stack reflection and the shifting beam as obtained from a
Gaussian beam fit to experiment and simultaion...................................... 31
Fig. 3.10. (a) Layer thickness as a function of layer number for a λ/4-Bragg stack
and for an impedance matched Bragg stack. (b) Reflectance for both
stacks having a backside gold coating applied. (c) Group propagation
angle with wavelength. Simulations are performed for plane waves
using the transfer matrix technique........................................................... 35
xvii
Chapter 4
Fig. 4.1. Bragg Stack (App. G, 3-2) with Gold Coating on Backside: Reflectance,
group delay, group velocity in the x-direction, and spatial shift as a
function of wavelength (solid lines – exact calculations, dotted lines –
approximations). ....................................................................................... 46
Fig. 4.2. Impedance Matched Bragg Stack (App. G, 3-3) with Gold Coating on
Backside: Reflectance, group delay, group velocity in the x-direction,
and spatial shift as a function of wavelength (solid lines – exact
calculations, dotted lines – approximations)............................................. 47
Fig. 4.3. Double-Chirped Stack (App. G, 5-6): Reflectance, group delay, group
velocity in the x-direction, and spatial shift as a function of
wavelength (solid lines – exact calculations, dotted lines –
approximations). ....................................................................................... 48
Fig. 4.4. Gires-Tournois Resonator (App. G, 6-1): Reflectance, group delay,
group velocity in the x-direction, and spatial shift as a function of
wavelength (solid lines – exact calculations, dotted lines –
approximations). ....................................................................................... 49
Fig. 4.5. Coupled-Cavity Stack (App. G, 6-2): Reflectance, group delay, group
velocity in the x-direction, and spatial shift as a function of
wavelength (solid lines – exact calculations, dotted lines –
approximations). ....................................................................................... 50
Fig. 4.6. Numerically Optimized Stack (App. G, 7-2) with Gold Coating on
Backside: Reflectance, group delay, group velocity in the x-direction,
and spatial shift as a function of wavelength (solid lines – exact
calculations, dotted lines – approximations)............................................. 51
Fig. 4.7. Four-Step Design (App. G, 10-1): Reflectance, group delay, group
velocity in the x-direction, and spatial shift as a function of
wavelength (solid lines – exact calculations, dotted lines –
approximations). ....................................................................................... 52
xviii
Fig. 4.8. (a) Refractive index and (b) extinction coefficient of GaAs as a function
of wavelength............................................................................................ 54
Fig. 4.9. 10-µm layer of GaAs sandwiched between GaAs on the incidence side
and a backside gold coating: Reflectance, group delay, group velocity
in the x-direction, and spatial shift as a function of wavelength (solid
lines – exact calculations, dotted lines – approximations)........................ 55
Fig. 4.10. Change of the refractive index n, the group refractive index ng, the shift
along the x-direction sx and the group delay τ in GaAs. .......................... 57
Fig. 4.11. 500-nm layer of GaAs sandwiched between an incident material with
n=2 and a backside gold coating: Reflectance, group delay, group
velocity in the x-direction, and spatial shift as a function of
wavelength (solid lines – exact calculations, dotted lines –
approximations). ....................................................................................... 59
Chapter 5
Fig. 5.1. (a) shows the Bragg wavelength as a function of the position in the
structure for five different 60-layer SiO2/Ta2O5 double-chirped mirror
designs. In (b) the theoretical spatial shift as a function of wavelength
is plotted for an incidence angle of 45° and p-polarized light. An
approximately linear shift is observed for all five designs. The
dispersion increases with decreasing chirp in the Bragg wavelength.
The maximum dispersion is achieved with a single-chirped Bragg
stack (f=0). ................................................................................................ 64
Fig. 5.2. (a) Physical layer thicknesses for a 200-layer SiO2/Ta2O5 double-chirped
structure. (b) Theoretically calculated shift as a function of
wavelength at 40° incidence angle and p-polarization. The circles
indicate the wavelengths and shifts corresponding to the diagrams in
(c). (c) E-field parallel to the interface of the forward propagating
wave as a function of the position in the structure for four different
xix
wavelengths – 780 nm, 830 nm, 880nm, and 930 nm. The vertical
lines indicate the position of the interfaces between layers. Light is
incident from the left, and the structure extends from 0 µm to 28 µm. .... 67
Chapter 6
Fig. 6.1. (a) Physical layer thicknesses for a 33-layer SiO2/Ta2O5 Gires-Tournois
resonator structure. (b) Theoretically calculated shift as a function of
wavelength at 45° incidence angle and p-polarization. The circles
indicate the wavelengths and shifts corresponding to the diagrams in
(c). (c) E-field parallel to the interface of the forward propagating
wave as a function of the position in the structure for four different
wavelengths – 815 nm, 827 nm, 830nm, and 832 nm. The vertical
lines indicate the position of the interfaces between layers. Light is
incident from the left, and the structure extends from 0 µm to 8 µm. ...... 70
Fig. 6.2. Results for approximating the desired phase characteristics by a fourth
order allpass polynomial. .......................................................................... 73
Fig. 6.3. (a) Periodic phase derivative of the transfer function as a function of the
normalized frequency. (b) The expected shift for a 4-cavity structure
is plotted as a function of wavelength for three different cavity optical
thicknesses. ............................................................................................... 74
Fig. 6.4. Shift as a function of wavelength at 54° incidence angle and s-
polarization for the design with approximate reflectors (blue) and the
final refined design (black). Below the graph, the quarter wave layer
sequences for both designs are given. Remember that the quarter
wave layers are for a wavelength of 980 nm and 0° incidence angle. ...... 77
Fig. 6.5. (a) Physical layer thicknesses for a 33-layer SiO2/Ta2O5 4-cavity
structure. (b) Theoretically calculated shift as a function of
wavelength at 54° incidence angle and s-polarization. The reflectance
of the structure is 100%. The circles indicate the wavelengths and
xx
shifts corresponding to the diagrams in (c). (c) E-field parallel to the
interface of the forward propagating wave as a function of the
position in the structure for four different wavelengths – 842 nm, 846
nm, 850nm, and 854 nm. The vertical lines indicate the position of the
interfaces between layers. Light is incident from the left, and the
structure extends from 0 µm to 15.4 µm................................................... 79
Chapter 7
Fig. 7.1. Generation of the starting design. (a) Quarter wave Bragg stack. (b)
Impedance matched stack. (c) Half-wave layers added to thinnest
layers. ........................................................................................................ 85
Fig. 7.2. (a) Shift as a function of wavelength and (b) reflectance as a function of
wavelength for the start design, the refined design, and the refined
design with backside gold coating at 45° incidence angle and p-
polarization. .............................................................................................. 86
Fig. 7.3. (a) Physical layer thicknesses for a 66-layer, numerically optimized
SiO2/Ta2O5 structure. (b) Theoretically calculated shift as a function
of wavelength at 54° incidence angle and p-polarization. The
reflectance of the structure is improved to nearly 100% by a gold
layer on the very right. The circles correspond to the diagrams in (c).
(c) E-field parallel to the interface of the forward propagating wave as
a function of the position in the structure for four different
wavelengths – 821 nm, 828 nm, 835nm, and 842 nm. The vertical
lines indicate the ....................................................................................... 87
Fig. 7.4. Experimentally observed intensity on a CCD trace as a function of
position and wavelength for a 66-layer numerically optimized stack
for an incidence angle of 54° and p-polarization. ..................................... 89
Fig. 7.5. Experimentally observed and theoretically calculated spatial dispersion
of a 66-layer SiO2/Ta2O5 dielectric stack with a total thickness of 13.4
xxi
µm on a quartz substrate for an incidence angle of 54° and p-
polarization. .............................................................................................. 90
Fig. 7.6. Comparing the performance of a periodic and a non-periodic structure. ...... 91
Fig. 7.7. (a) Experimentally observed intensity on a CCD trace as a function of
position and wavelength for 8 bounces off a 66-layer numerically
optimized stack. (b) Experimentally observed and theoretically
calculated shift as a function of wavelength. ............................................ 92
Fig. 7.8. 4-channel wavelength demultiplexer. ............................................................ 93
Chapter 8
Fig. 8.1. (a) Volume modes. (b) Surface modes. ......................................................... 96
Fig. 8.2. Physical layer thicknesses of the periodic and non-periodic design and
the double-chirped mirror starting design................................................. 97
Fig. 8.3. Reflectance of the two designs at 40° incidence angle.................................. 99
Fig. 8.4. Zoomed-in reflectance of the two designs at 40° incidence angle. ............... 99
Fig. 8.5. Calculated group propagation angle. ........................................................... 100
Fig. 8.6. Calculated Shift along the interface for a single bounce. ............................ 100
Fig. 8.7. Schematic of three volume modes for focussing on the front surface......... 101
Fig. 8.8. Increase in the number of volume modes per 1 nm wavelength interval as
a function of the stack dispersion for an input half cone ∆θin =1°.......... 104
Fig. 8.9. Volume mode calculation for the non-periodic design for an incidence
angle of 40°, an input half cone of 0.6°, and a wavelength interval
from 1525 nm to 1565 nm. (a) shows the position of the obtained
modes in terms of wavelength and propagation angle within the
crystal. (b) depicts a cartoon of the obtained modes............................... 107
xxii
Fig. 8.10. Number of volume modes as obtained from simulation and the expected
number of modes calculated using (8 - 6) as a function of the input
half cone in (a) and the input spot size in (b).......................................... 108
Fig. 8.11. Schematic of three non-overlapping surface modes on the back surface. . 109
Fig. 8.12. Schematic of beam shift after two bounces for two neighboring modes... 112
Fig. 8.13. Gaussian profile of two neighboring modes that are separated by 5 nm.
The vertical green lines represent the position of the blue channel.
The solid lines are for the center frequency and the dotted lines
represent the shift with 50 GHz signal modulation. ............................... 112
Fig. 8.14. Crosstalk as a function of signal bandwidth for the non-periodic design
and a channel spacing of 5 nm................................................................ 113
Fig. 8.15. Theoretical maximum number of modes in the interval from 1525 nm to
1565 nm as a function of incident angle assuming a dispersion of
60µm/40nm, ns=1.52 and a crosstalk of –40dB (c1=3.8). ...................... 119
Fig. 8.16. Surface mode calculation for the non-periodic design for an incidence
angle of 40°, focussing on the back side at 1545 nm with a spot size
of 20 µm (corresponding to ∆θin = 1.4°), a wavelength interval from
1525 nm to 1565 nm, and 13 bounces. To obtain a crosstalk around –
40dB, a spacing of 3.8*20 µm/cos(40°)=100 µm is chosen between
modes. (a) shows the position of the obtained modes in terms of
wavelength and propagation angle within the crystal. (b) depicts a
cartoon of the modes obtained graphing the two 1/e E-field rays for
each mode. .............................................................................................. 120
Fig. 8.17. Number of surface modes as obtained from simulation, the expected
number of modes calculated using (8 - 14), and the maximum number
of modes obtained from (8 - 23) for the non-periodic design as a
function of the number of bounces within the structure. ........................ 122
Fig. 8.18. Physical layer thicknesses of the optimal non-periodic design. ................ 123
xxiii
Fig. 8.19. Reflectance of the optimal designs at 40° incidence angle........................ 124
Fig. 8.20. Group propagation angle and shift of the optimal designs at 40°
incidence angle........................................................................................ 124
Fig. 8.21. Surface mode calculation for the optimal design for an incidence angle
of 40°, focussing on the back side at 1545 nm with a spot size of 20
µm (corresponding to ∆θin = 1.4°), a wavelength interval from 1525
nm to 1565 nm, and 13 bounces. To obtain a crosstalk around –40dB,
a spacing of 3.8*20 µm/cos(40°)=100 µm is chosen between modes.
(a) shows the position of the modes obtained, in terms of wavelength
and propagation angle within the crystal. (b) depicts a cartoon of the
obtained modes graphing the two 1/e-E-field rays for each mode. ........ 125
Fig. 8.22. Number of surface modes as obtained from simulation, the expected
number of modes calculated using (8 - 14), and the maximum number
of modes obtained from (8 - 23) for the optimal design as a function
of the number of bounces within the structure........................................ 126
Fig. 8.23. (a) Schematic of a 4-channel demultiplexer. (b) Scaled drawing of the
same 4-channel demultiplexer. ............................................................... 128
Fig. 8.24. Increasing the number of channels by increasing the number of
bounces. .................................................................................................. 129
Fig. 8.25. Increasing the number of channels by decreasing the spot size. ............... 130
Fig. 8.26. Reducing the device size by reducing the substrate thickness. ................. 131
Chapter 9
Fig. 9.1. (a) Shift as a function of dispersion for the designs in Table 9.1. (b) Shift
divided by stack thickness for the different designs. .............................. 135
Fig. 9.2. 20-period SiO2-Ta2O5 design for a 40º incidence angle. The shift was
specified to increase from 5 µm to 25 µm over a 70 nm operating
range........................................................................................................ 139
xxiv
Fig. 9.3. 20-period SiO2-Ta2O5 design for a 40º incidence angle. The shift was
specified to increase from 5 µm to 35 µm over a 70 nm operating
range........................................................................................................ 139
Fig. 9.4. 20-period SiO2-Ta2O5 design for a 40º incidence angle. The shift was
specified to increase from 5 µm to 45 µm over a 70 nm operating
range........................................................................................................ 140
Fig. 9.5. Example of a 20-period starting design for refractive indices of 1.45 and
2.09, an incidence angle θ of 40º, and an operating wavelength of
1550nm. .................................................................................................. 142
Fig. 9.6. Dependency of the shift on the number of periods...................................... 144
Fig. 9.7. Dependency of the shift on the incidence angle. ......................................... 144
Fig. 9.8. Dependency of the shift on the average refractive index............................. 145
Fig. 9.9. Dependency of the shift on the refractive index contrast. ........................... 145
Fig. 9.10. Dependency of the shift on the physical stack thickness........................... 146
Fig. 9.11. Dependency of the shift on the optical stack thickness. ............................ 146
Fig. 9.12. Dependency of the shift on the relative error. ........................................... 147
Fig. 9.13. Dependency of the shift on the filling ratio. .............................................. 147
Fig. 9.14. Two poor models for comparison. (a) Assuming the shift to be
inversely proportional to the refractive index. (b) Assuming the shift
to be inversely proportional to the refractive index cubed. .................... 148
Fig. 9.15 Normalized wavelength range as a function of the specified dispersion.... 150
Chapter 10
Fig. 10.1. Operating principle of a frequency filter. .................................................. 156
Fig. 10.2. Demultiplexing architecture based on cascading filters. For 64 channels
63 filters are needed. ............................................................................... 156
xxv
Fig. 10.3. Operating principle of a wavelength interleaver........................................ 157
Fig. 10.4. Demultiplexing architecture based on cascading interleavers. For 64
channels 63 interleavers are needed........................................................ 157
Fig. 10.5. Demultiplexing architecture based on a combination of interleavers and
filters. For 64 channels 63 devices are needed. ...................................... 158
Fig. 10.6. Demultiplexing architecture based on a combination of interleavers and
spatial MUXes. For 64 channels only 15 devices are needed................. 159
Fig. 10.7. Intensity as a function of position and wavelength for 8 bounces off a
66-layer numerically optimized stack at 54º incidence angle and p-
polarization. ............................................................................................ 160
Fig. 10.8. Schematic of a (a) Gaussian passband shape compared to a (b) flat-top
passband shape (transmission refers to the energy transfer from input
to output)................................................................................................. 161
Fig. 10.9. Flat-top passband shape (a) corresponds to a step-like beam shift with
wavelength (b). ....................................................................................... 162
Fig. 10.10. Shift (a) and reflectance (b) after 8 bounces off a 100-layer
numerically optimized stack at 45º incidence angle and p-polarization. 163
Fig. 10.11. Shift as a function of wavelength for a plane wave at 45º incidence
angle, and Gaussian beams with 15 µm, 30 µm, and 50 µm spot size. .. 165
Fig. 10.12. Shift as a function of wavelength for four different incidence angles
for a 66-layer stack.................................................................................. 166
Fig. 10.13. (a) Intensity as a function of position and wavelength for 8 bounces
off a 66-layer numerically optimized stack at 48º incidence angle and
p-polarization. (b) Comparison between experiment and plane wave
theory. ..................................................................................................... 167
Fig. 10.14. Systems using stacks with high spatial dispersion. (a) Combination of
two stacks with opposite dispersion can be used to obtain spatial
dispersion without temporal dispersion. (b) System providing
xxvi
temporal dispersion without spatial dispersion. This system could also
be used to manipulate channels of different wavelengths
independently.......................................................................................... 168
Fig. 10.15. Wavelength-dependent beam steering. .................................................... 169
Appendix A
Fig. A.1. Labeling of periodic stack........................................................................... 175
Appendix C
Fig. C.1. Naming conventions for a multilayer stack. ............................................... 186
Fig. C.2. Conventions for the positive directions of k, E, and H in the case of p-
polarization (TM) and s-polarization (TE). ............................................ 186
Appendix D
Fig. D.1. Gaussian beam with w0=4.7µm and λ=890nm. (a) Calculated from exact
formula (D - 4). (b) Obtained from Fourier decomposition (D - 22).
Note that the ξ- and the ζ-axes have different scales.............................. 198
Fig. D.2. Direction cosines α and γ for propagation in the x-z-plane. ....................... 200
Fig. D.3. Amplitude of Fourier components normalized by π00wE for a
Gaussian beam with w0=4.7µm and λ=890nm. ...................................... 202
Fig. D.4. Propagation of a Gaussian beam with w0=4.7µm and λ=890nm in free
space........................................................................................................ 202
Fig. D.5. Propagation of a Gaussian beam with w0=4.7µm and λ=890nm in free
space at an angle of 40º with respect to the z-axis.................................. 203
Fig. D.6. Intensity distribution for a Gaussian beam with w0=4.7µm and λ=890nm
incident onto a 40-µm slab of material with refractive index n=2.5 at
an angle of 40º with respect to the z-axis. .............................................. 205
xxvii
Appendix E
Fig. E.1. Target reflectance as a function of wavelength for an EDFA gain-
flattening filter. ....................................................................................... 208
Fig. E.2. Reflectance of two 200-layer start designs. The merit function and the
total thickness of both designs are given to the right.............................. 209
Fig. E.3. Golden section search refinement. .............................................................. 210
Fig. E.4. Secant method refinement. .......................................................................... 211
Fig. E.5. Conjugate gradient algorithm refinement.................................................... 212
Fig. E.6. BFGS algorithm refinement. ....................................................................... 212
Fig. E.7. Damped least squares refinement................................................................ 214
Fig. E.8. Hooke & Jeeves pattern search refinement. ................................................ 214
Fig. E.9. Refined design using a combination of the different methods.................... 215
Appendix F
Fig. F.1. The change in (∂β/∂ω)|K=const as a function of wavelength. ........................ 221
Fig. F.2. Angular range ∆θstruc for an input angular range of ∆θin = 0.5º directly
calculated and calculated from (8 – 17) as being proportional to the
dispersion. ............................................................................................... 222
Fig. F.3. Angular range ∆θstruc for an input angular range of ∆θin = 0.05º directly
calculated and calculated from (8 - 17) as a function of wavelength. .... 223
xxviii
xxix
… physical intuition is often just mathematical consequences
with which we have lived long enough
to make them part of our “world picture.”
– Amnon Yariv
A. Yariv, IEEE J. Select. Topics Quantum Electron., 6/6, 1486 (2000).
xxx
1
Chapter 1
Introduction
With the emergence of high-speed computer connections to the Internet, data traffic has
been increasing rapidly. To support this growth, transmission systems of higher and
higher capacity are needed. Today’s transmission systems use optical fiber to connect
different locations, since optical fiber has a much higher data capacity than electrical
wires. As the installation of new optical fiber is, however, very expensive, different
multiplexing techniques have been developed that allow the transmission of multiple
channels over a single fiber, thus increasing the data capacity that can be transmitted over
the existing fiber.1 These techniques include Time Division Multiplexing (TDM) and
Wavelength Division Multiplexing (WDM).
TDM systems interleave several lower bitrate datastreams for high bitrate transmission
over the optical fiber. In WDM systems each channel is encoded with a different optical
wavelength. Since light of different wavelengths does not interact (at least to first order),
the channels can be overlapped for transmission over a single fiber as shown
schematically in Fig. 1.1. Today Dense WDM (DWDM) – named so because of their
dense channel spacing of 100 GHz or less – with typically 64 channels is employed in
2
long-haul systems. Coarse WDM (CWDM), with channel spacings around 20 nm and
approximately four to sixteen channels, is used in metro systems.
λ1 λ8
λ1
λ8
λ1
λ8
MUX
DEMUX
Fig. 1.1. Schematic of a Wavelength Division Multiplexing (WDM) system.
Crucial components of a WDM system are the wavelength multiplexer (MUX),
performing the spatial overlap of the different wavelength channels on the transmission
side, and the wavelength demultiplexer (DEMUX), spatially separating the different
wavelength channels at the receiver. The MUX/DEMUX should have low loss, low
crosstalk between channels, low polarization dependent loss (PDL), as well as some
tolerance for the drift of the channel wavelengths. From a systems perspective it should
furthermore be compact, temperature-stable, and, especially in the case of CWDM, cost-
effective. In this thesis a novel wavelength MUX/DEMUX based on the spatial
dispersion of multilayer thin-film structures is introduced.
1.1 Current MUX/DEMUX DevicesPrisms2 and gratings3 are traditional MUX/DEMUX devices. Both prisms and gratings
allow for the separation of many different wavelengths using just a single device. Prisms
rely on the wavelength-dependent refractive index of the prism material as well as the
special prism geometry schematically shown in Fig. 1.2(a) to disperse different
wavelengths into different angular directions. Due to their insufficient dispersion, prisms
are not used in today’s WDM systems.
Gratings (Fig. 1.2(b)) use interference effects between the reflections off different grating
grooves to separate beams of different wavelengths. Depending on the number of
grooves, gratings can have sufficient dispersion for current WDM systems. Gratings have
3
low loss and crosstalk, are temperature-stable and cost-effective. As the beam steering of
gratings is continuous, a change in the center wavelength of a channel results in loss and
crosstalk. Current research is investigating the combination of gratings with micro-optics
to eliminate this effect.4 Low-frequency gratings (<400 lines/mm) are polarization
independent. Unfortunately, such gratings require quite large focussing optics leading to
large component packages. High-frequency gratings can be used in combination with
polarization control to reduce the component size.5
(a) (b)
Fig. 1.2. Traditional MUX/DEMUX technologies. (a) Prism. (b) Grating.
Besides diffraction gratings, state-of-the-art MUX/DEMUX devices include arrayed
waveguide gratings (AWG), fiber Bragg gratings (FBG), and thin-film filters (TFF)
schematically shown in Fig. 1.3.6 AWGs 7,8,9,10 (Fig. 1.3(a)) split the light of all
wavelengths onto an array of waveguides of different lengths. After propagation through
the waveguides, the exiting light interferes at a wavelength-dependent position resulting
in wavelength multiplexing or demultiplexing. AWGs are particularly interesting for high
channel count systems, as they are compact and allow the (de)multiplexing of many
channels using a single device. AWGs need to be temperature stabilized, and are
expensive in fabrication and operation. Their high cost is only justified for high channel
count systems.
For cost-effective lower channel count systems, thin-film filters or fiber Bragg gratings
are typically used, due to their lower initial cost and the possibility to upgrade the channel
count progressively. Fiber Bragg gratings10 (Fig. 1.3(b)) are optical fibers that have a
refractive index modulation imposed along the direction of propagation. This grating
structure is designed such that one channel is transmitted, while all other channels are
reflected. The reflected channels are separated from the input light with a circulator.
4
Fiber Bragg gratings are low loss and polarization independent. As each fiber Bragg
grating only demultiplexes one channel, a separate grating is needed for each channel
leading to cascading losses and a more complex module assembly as the number of
channels is increased.
(a)
(b)
Bragg grating(c)
L H L H L...
Fig. 1.3. State-of-the-art MUX/DEMUX technologies. (a) Arrayed waveguide
grating (AWG). (b) Fiber Bragg grating (FBG). (c) Thin-film filter (TFF).
Thin-film filters11 (Fig. 1.3(c)) are stacks of alternating high and low index materials.
They are designed such that one channel is transmitted and all other channels are
reflected. The typical size of a thin-film filter is 1 mm by 1 mm. Again a separate device
is needed for each channel. Thin-film filters are assembled with micro-optics and used
slightly off normal incidence to separate the incoming from the reflected light. They are
low loss, low crosstalk, polarization independent, tolerant towards a drift of the channel
wavelength, and very temperature-stable. Thin-film filters are manufactured cost-
effectively on a wafer-scale. Today thin-film filters have the largest market share for
applications in WDM systems. Their main drawback is that they can only demultiplex a
5
single channel per device, again resulting in larger modules and cascading losses as the
number of channels increases. In the next section I will introduce how a single thin-film
stack can be used for multiplexing or demultiplex multiple channels, overcoming this
drawback.
1.2 Thin-Film Spatial MUX/DEMUXHere a novel thin-film device is introduced that multiplexes or demultiplexes multiple
WDM channels simultaneously. While traditional thin-film filters use the amplitude
reflection and transmission properties of multilayer stacks, this device employs
wavelength-dependent spatial beam shifting. Fig. 1.4 shows a schematic of the proposed
device.
zy
x
Thin-FilmStack
Mirror
Mirror
Substrate
Out 1Out 2Out 3
In Focussing lens
Out 4
Fig. 1.4. Thin-film spatial MUX/DEMUX.
The multiplexed light is incident from the top left corner at an angle onto the thin-film
stack. The stack is designed such that different wavelengths propagate at different
effective group propagation angles. Thus, beams incur a wavelength-dependent shift in
the structure. Multiple bounces can be performed off or through the stack to increase the
6
spatial separation of beams of different wavelengths. This spatial MUX/DEMUX
provides the advantages of thin-film filters (except polarization independence), while at
the same time eliminating the need for a separate thin-film device for each channel.
The key component of this spatial MUX/DEMUX is the multilayer thin-film stack with
high spatial dispersion. This thesis investigates how to design thin-film structures with
high spatial dispersion and how they can be used in WDM systems. Chapter 2 suggests
four different approaches for obtaining highly dispersive thin-film stacks and reviews
relevant literature. Chapters 3, 5, 6, and 7 discuss the four different design approaches in
detail. Chapter 4 explores the relationship between spatial dispersion, temporal
dispersion, and stored energy in a structure. This is important for the design algorithms of
Chapters 5 and 6 and gives at the same time physical insights into the origins of spatial
dispersion. Chapter 8 investigates the maximum number of channels that can be obtained
using a given dispersive thin-film stack and a device architecture as depicted in Fig. 1.4.
Chapter 9 examines how much spatial dispersion can be obtained depending on the stack
materials, the stack thickness, and the incidence angle. Finally, Chapter 10 discusses how
thin-film stacks with high spatial dispersion can be used in DWDM and CWDM systems
as well as possible other applications. Chapter 11 summarizes the results obtained.
References
[1] R. Ramaswami and K. N. Sivarajan, Optical Networks: A Practical Perspective,
Morgan Kaufmann Publishers, San Francisco, CA (1998).
[2] E. Hecht, Optik, Addison-Wesley (Deutschland) GmbH, Bonn (1989).
[3] C. Palmer and E. Loewen, Diffraction Grating Handbook, Richadrson Grating
Laboratory, Rochester, New York (2000).
[4] C.X. Yu, D.T. Neilson, C.R. Doerr and M. Zirngibl, “Dispersion-Free (De)mux
with Very High Figure-of-Merit,” Optical Fiber Communication Conference 2002,
Anaheim, CA (2002). Talk WS1.
[5] W. T. Boord, T. L. Vanderwert, R. DeSalvo, “Bulk diffraction gratings play
increasing role in optical networking,” Lightwave (March 2001).
7
[6] “MUX-DEMUX components market to decline 22 %,” WDM Sol. (July 04, 2001).
[7] M. K. Smit, “New Focusing and Dispersive Planar Component Based on an Optical
Phased Array,” Electron. Lett., 24/7, 385-386 (1988).
[8] H. Takahashi, S. Suzuki, K. Kato, I Nishi, “Arrayed-Waveguide Grating for
Wavelength Division Multi/Demultiplexer With Nanometer Resolution,” Electron.
Lett., 26/2, 87-88 (1990).
[9] C. Dragone, “An N x N Optical Multiplexer Using A Planar Arrangement of Two
Star Couplers,” IEEE Photon. Techn. Lett., 3/9, 812-815 (1991).
[10] C.K. Madsen and J.H. Zhao, Optical Filter Design and Analysis - A Signal
Processing Approach, John Wiley & Sons, Inc. (1999).
[11] H. A. MacLeod, Thin-Film Optical Filters, Institute of Physics Publishing (2001).
8
Chapter 2
Designing Dispersive Structures
In this thesis four different approaches for designing thin-film multilayer structures with
high spatial dispersion are researched. Spatial dispersion refers to a change in the beam
exit position with wavelength. This chapter introduces the four different approaches
schematically shown in Fig. 2.1, and reviews relevant literature for each approach. The
different approaches are discussed in more detail in Chapters 3 to 7. For each structure,
polychromatic light is incident at an angle from the top left corner. All structures are
operated in reflection and demultiplex light by a wavelength-dependent shift along the x-
axis. After exiting the structures, beams of different wavelength propagate in parallel
once again. The stack in Fig. 2.1(a) is periodic; (b) – (d) are non-periodic stacks.
The first approach is to use the superprism effect in a one-dimensional photonic crystal as
shown in Fig. 2.1(a). Different wavelengths propagate at different group velocity angles
within the structure and are thus spatially shifted along the x-axis. The second approach
is to use a non-periodic structure that reflects different wavelengths at different positions
along the z-axis as depicted in Fig. 2.1(b). Since the structure is operated at an angle, this
wavelength-dependent penetration depth leads to a spatial shift along the x-axis. Thirdly,
9
a wavelength-dependent number of roundtrips, as in the structure in Fig. 2.1(c), also
results in a wavelength-dependent beam shift along the x-axis, because of the operation at
an angle. Finally, the structure in Fig. 2.1(d) utilizes a combination of wavelength-
dependent penetration depth and wavelength-dependent number of roundtrips to
demultiplex polychromatic light. The next sections review the relevant literature for each
approach.
Mirror
Periodic stack
θin θin
Resonator stack
θin
Chirped stack General stack
θin
zy
x
(a) (b) (c) (d)
Fig. 2.1. Operating schematics of four different types of thin-film structures that
can be used for demultiplexing multiple wavelengths by spatial beam shifting. The
structure in (a) is periodic; (b) – (d) are non-periodic structures. (a) Superprism
effect in a one-dimensional photonic crystal (combined here with a simple
reflection off of the right face). (b) Wavelength-dependent penetration depth. (c)
Wavelength-dependent number of roundtrips in the structure. (d) Combination of
wavelength-dependent penetration depth and number of roundtrips.
2.1 “Superprism Effect” in Photonic CrystalsPolychromatic light incident at an angle onto one of the surfaces of a prism is dispersed
within the prism, i.e., light rays of different wavelengths propagate at different angles in
the prism. Rays exiting the prism have a wavelength-dependent propagation angle due to
the prism geometry. Conventional prisms rely on material dispersion. Since the change in
refractive index with wavelength is rather weak for transparent materials, this limits the
obtainable dispersion.
10
Previous research has shown that photonic crystal structures can be used to obtain much
higher spatial dispersion.1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19 Photonic crystals are artificial
structures composed of one-dimensional, two-dimensional, or three-dimensional periodic
arrangements of different materials.20 Because of the wavelength scale feature sizes of
photonic crystals, these structures exhibit a behavior that is very distinct from that of bulk
materials. Wavelength regimes with high dispersion have been observed in theory and
experiment for one-dimensional, two-dimensional, and three-dimensional photonic
crystals. As these artificial structures exhibit much higher dispersion than the material
dispersion of conventional prisms, this phenomenon has been termed the “superprism
effect”.
Realizing that a periodic multilayer thin-film stack behaves as a one-dimensional
photonic crystal,9,21,22,23,24 we expect that such structures exhibit the superprism effect and
can be used for wavelength multiplexing and demultiplexing. In Chapter 3 the superprism
effect in a 100-period one-dimensional thin-film structure containing two dielectric layers
per period is explored. We operate the dielectric stack in reflection, performing two
passes through the stack as depicted in the schematic in Fig. 2.1(a). Seen from the side,
polychromatic light is incident from the top left corner onto the periodic dielectric stack.
Just outside the stop-band (the high reflection spectral region), different wavelengths
propagate at different group propagation angles within the thin-film structure. Therefore,
beams of different wavelengths exit the dielectric stack at different positions along the x-
direction. In connection with ultrafast optics this effect has been called a “spatial chirp.”25
We experimentally observe this spatial shift with excellent agreement with theory. After
exiting the dielectric stack, the beams propagate parallel once again.
2.2 Periodic versus Non-Periodic StructuresFor application purposes, it is desirable to have the ability to design the spatial dispersion
with wavelength to given specifications. A linear spatial dispersion with wavelength or
frequency is, for example, of practical interest for multiplexing or demultiplexing
devices. Unfortunately, the design space of a periodic thin-film structure is very limited.
11
Only the period length, the materials, the distribution of the materials in a period, and the
incidence angle can be chosen. Considering this limited number of degrees of freedom, it
is not surprising that all periodic thin-film structures with two layers per period exhibit a
similar nonlinear spatial shift with wavelength.
To achieve a linear spatial shift, we need to increase the degrees of freedom. One
possible approach would be to increase the number of layers per period. For a sufficiently
large number of periods and a period length comparable to the wavelength, the optical
properties within such a structure do not depend on the position in the structure, and the
concept of effective optical properties is valid.2,4 Therefore, increasing the number of
layers per period is the method of choice if the application requires a constant phase
velocity or group velocity within the volume of the structure, as, for example, for phase
matching purposes. For other applications only the properties of the exiting light matter.
In the case of a wavelength demultiplexer only the properties of the light along the
exiting surface of the structure are important. As long as beams of different wavelengths
are spatially separated at the exit surface, it does not matter what happened to the beams
within the structure. This realization allows us to consider a whole new class of highly
dispersive structures – non-periodic structures with a high spatial dispersion along the
exit surface.19,26 Non-periodic structures have a much higher number of degrees of
freedom. In addition to the materials and the incidence angle, we can choose all the layer
thicknesses independently. With such a high number of degrees of freedom, we expect a
much higher design freedom as well.
The difficult task is now to devise a design that fulfills the desired specifications, such as,
e.g., a linear shift with wavelength and, ideally, to deduce design principles and physical
understanding that enable future designs. Considering that a 200-layer structure has more
than 200 degrees of freedom, searching the whole design space for an optimal structure is
not a feasible solution. The next two sections discuss three possible design approaches for
non-periodic structures with high spatial dispersion.
12
2.3 Designs for Temporal Dispersion CompensationWhile non-periodic thin-film structures with high spatial dispersion have not been
investigated previously, several thin-film design methods have been developed for
temporal dispersion compensation purposes in femtosecond-laser cavities27,28,29,30,31 and
optical fibers.32,33,34 In Chapter 4 we examine the relationship between spatial and
temporal dispersion, and show that methods used for designing structures with temporal
dispersion can be modified to obtain spatial dispersion.19 In Chapters 5 and 6 two
analytical approaches for designing thin-film stacks with high spatial dispersion based on
techniques developed for temporal dispersion compensation are introduced.
In Chapter 5 non-periodic thin-film structures are analyzed that exhibit a wavelength-
dependent penetration depth or turning point. An example of such a structure is a chirped
mirror, i.e., a dielectric stack with a position-dependent period length. Different
wavelengths are reflected at different positions along the z-axis. Operated at normal
incidence, such a chirped mirror exhibits temporal dispersion. An analytical algorithm for
designing structures with temporal dispersion has been developed previously based on
coupled-mode theory.27,28,29 Here we modify this algorithm for operating the structure at
an angle. At an angle the wavelength-dependent penetration depth corresponds to a
wavelength-dependent spatial shift along the x-axis at the exit surface as schematically
shown in Fig. 2.1(b). Therefore, such a structure can be used to demultiplex beams of
different wavelengths by spatial beam shifting.
In Chapter 6 resonator structures with one or multiple cavities are discussed. Examples of
single-resonator structures are Fabry-Perot and Gires-Tournois resonators.33 In a
resonator, light “bounces” back and forth. The effective number of bounces (or
roundtrips) differs significantly between wavelengths that are on-resonance and ones that
are off-resonance. If we operate such a resonator structure at an angle, this wavelength-
dependent number of roundtrips results in different amounts of beam shifting along the x-
axis as indicated figuratively in Fig. 2.1(c). Note that for a single resonator structure all
wavelengths are reflected at the same physical depth within the stack. Thus, the
maximum beam shift possible cannot be deduced from geometrical arguments. As the
13
incident light traverses a resonator multiple times before exiting, such structures exhibit
stored energy. In Chapter 4 we will analyze the relationship between stored energy and
beam shifting. More complex coupled-resonator structures with multiple cavities can be
designed to obtain superior dispersion characteristics. Coupled-cavity structures have
previously been designed for temporal dispersion compensation using digital lattice
techniques.32,33,34 Chapter 6 explains how this analytical technique can be modified to
design structures with spatial disperison.
Both analytical design procedures discussed in Chapters 5 and 6 are based on a reduction
of the degrees of freedom to a smaller subset. For the design of a 200-layer structure, for
example, not all possible 200-layer structures are considered, just a subset. It is shown
that both methods can be used to design thin-film structures exhibiting a linear spatial
shift along the x-axis with wavelength. Furthermore, the limitations of these two different
analytical design procedures are investigated.
2.3 Thin-Film Filter Design TechniquesThe fourth approach for designing multilayer thin-film stacks with high spatial dispersion
is based on modifying traditional thin-film filter design techniques. Thin-film filters with
a wide range of reflection and transmission properties have been designed for a long
time. The design process is typically divided into two stages – synthesis and
refinement.35,36,37,38 During the filter synthesis a multilayer stack is generated that closely
matches the desired performance.39 During refinement, the performance of a start design
is further improved.40 In Chapter 7 we show that thin-film filter design techniques can be
adapted to design thin-film stacks with high spatial dispersion. A synthesis procedure is
described to obtain a good starting design. Numerical optimization techniques are then
used to obtain the desired spatial dispersion. In this procedure all the degrees of freedom
are kept, and a locally optimal structure is achieved. We find that dispersive thin-film
structures obtained by numerical optimization combine effects based on a wavelength-
dependent turning point with effects based on a wavelength-dependent number of
roundtrips as depicted schematically in Fig. 2.1(d). We verify in theory and experiment
that a linear shift with wavelength can be obtained using such a structure and that it is
14
larger than can be obtained by a simple single reflection off of any particular depth in the
structure.
References
[1] R. Zengerle, “Light propagation in singly and doubly periodic planar waveguides,”
J. Mod. Opt., 34/12, 1589-1617 (1987).
[2] R. Zengerle and O. Leminger, “Frequency Demultiplexing Based on Beam Steering
in Periodic Planar Optical Waveguides,” J. Opt. Commun., 11/1, 11-12 (1990).
[3] J. P. Dowling and C. M. Bowden, “Anomalous index of refraction in photonic
bandgap materials,” J. Mod. Opt., 41/2, 345-351 (1994).
[4] S.-Y. Lin, V. M. Hietala, L. Wang, and E.D. Jones, “Highly dispersive photonic
band-gap prism,” Opt. Lett., 21/21, 1771-1773 (1996).
[5] H. Kosaka, T. Kawashima, A. Tomita, M. Notomi, T. Tamamura, T. Sato, S.
Kawakami, “Superprism phenomena in photonic crystals,” Phys. Rev. B, 58/16,
R10 096-R10 099 (1998).
[6] S. Enoch, G. Tayeb, D. Maystre, “Numerical evidence of ultrarefractive optics in
photonic crystals,” Opt. Comm., 161, 171-176 (1999).
[7] H. Kosaka, T. Kawashima, A. Tomita, M. Notomi, T. Tamamura, T. Sato, S.
Kawakami, “Superprism phenomena in photonic crystals: Toward microscale
lightwave circuits,” J. Lightw. Techn., 17/11, 2032-2038 (1999).
[8] B. Gralak, S. Enoch, G. Tayeb, “Anomalous refractive properties of photonic
crystals,” J. Opt. Soc. Am. A, 17/6, 1012-1020 (2000).
[9] B. E. Nelson, M. Gerken, D. A. B. Miller, R. Piestun, C.-C. Lin, J. S. Harris, Jr.,
“Use of a dielectric stack as a one-dimensional photonic crystal for wavelength
demultiplexing by beam shifting,” Opt. Lett. 25/20, 1502-1504 (2000).
[10] E. Silvestre, J. M. Pottage, P. St. J. Russell, P. J. Roberts, “Design of thin-film
photonic crystal waveguides,” Appl. Phys. Lett., 77/7, 942-944 (2000).
[11] A. N. Naumov, R. B. Miles, P. Barker, A. M. Zheltikov, “Ultradispersive prisms
and narrow-band tunable filters combining dispersion of atomic resonances and
photonic band-gap structures,” Laser Phys., 10/2, 622-626 (2000).
15
[12] M. Notomi, K. Yamada, A. Shinya, J. Takahashi, C. Takahashi, I. Yokohama,
“Extremely large group-velocity dispersion of line-defect waveguides in photonic
crystal slabs,” Phys. Rev. Lett., 87/25, 253902-1 - 253902-4 (2001).
[13] T. Ochiai and J. Sanchez-Dehesa, “Superprism effect in opal-based photonic
crystals,” Phys. Rev. B, 64, 245113-1 - 245113-7 (2001).
[14] W. Park and C. J. Summers, “Extraordinary refraction and dispersion in two-
dimensional photonic-crystal slabs,” Opt. Lett. 27/16, 1397-1399 (2002).
[15] T. Baba and M Nakamura, “Photonic Crystal Light Deflection Devices Using the
Superprism Effect,” IEEE J. Quantum Electron., 38/7, 909-914 (2002).
[16] L. Wu, M. Mazilu, T. Karle, T. F. Krauss, “Superprism Phenomena in Planar
Photonic Crystals,” IEEE J. Quantum Electron., 38/7, 915-918 (2002).
[17] K. B. Chung and S. W. Hong, “Wavelength demultiplexers based on the
superprism phenomena in photonic crystals,” Appl. Phys. Lett., 81/9, 1549-1551
(2002).
[18] T. Baba and T. Matsumoto, “Resolution of photonic crystal superprism,” Appl.
Phys. Lett., 81/13, 2325-2327 (2002).
[19] M. Gerken and D. A. B. Miller, “Multilayer thin-film structures with high spatial
dispersion,” Appl. Opt. 42/7 (2003), 1330-1345.
[20] J.D. Joannopoulos, R.D. Meade, and J.N. Winn, Photonic Crystals-Molding the
Flow of Light, Princeton University Press (1995).
[21] C. Elachi, “Waves in Active and Passive Periodic Structures: A Review,” Proc. of
the IEEE, 64/12, 1666-1698 (1976).
[22] P. Yeh, A. Yariv, C.-S. Hong, “Electromagnetic propagation in periodic stratified
media. I. General theory,” J. Opt. Soc. Am., 67/4, 423-438 (1977).
[23] A. Yariv and P. Yeh, “Electromagnetic propagation in periodic stratified media. II.
Birefringence, phase matching, and x-ray lasers,” J. Opt. Soc. Am., 67/4, 438-448
(1977).
[24] Y. Fink, J. N. Winn, S. Fan, C. Chen, J. Michel, J. D. Joannopoulos, E. L. Thomas,
“A Dielectric Omnidirectional Reflector,” Science, 282, 1679-1682 (1998).
[25] I. Walmsley, L. Waxer, C. Dorrer, “The role of dispersion in ultrafast optics,” Rev.
of Scientific Instr., 72/1, 1-29 (2001).
16
[26] M. Gerken, B. E. Nelson, and D. A. B. Miller, “Thin-Film Wavelength
Demultiplexer Based on Photonic Crystal and Group Velocity Effects,” OSA
Conference on Integrated Photonics Research 2002, Vancouver, BC (July 17-19,
2002).
[27] R. Szipöcs, K. Ferencz, C. Spielmann, F. Krausz, “Chirped multilayer coatings for
broadband dispersion control in femtosecond lasers,” Opt. Lett. 19/3, 201-203
(1994).
[28] P. Tournois and P. Hartemann, “Bulk chirped Bragg reflectors for light pulse
compression and expansion,” Opt.Commun. 119, 569-575 (1995).
[29] N. Matuschek, F.X. Kärtner, and U. Keller, “Exact Coupled-Mode Theories for
Multilayer Interference Coatings with Arbitrary Strong Index Modulations,” IEEE
J. Quantum Electron., 33/3, 295-302 (1997).
[30] N. Matuschek, F.X. Kärtner, and U. Keller, “Theory of Double-Chirped Mirrors,”
IEEE J. Select. Topics Quantum Electron., 4/2, 197-208 (1998).
[31] N. Matuschek, F.X. Kärtner, and U. Keller, “Analytical Design of Double-Chirped
Mirrors with Custom-Tailored Dispersion Characteristics,” IEEE J. Quantum
Electron., 35/2, 129-137 (1999).
[32] G. Lenz and C. K. Madsen, “General optical all-pass filter structures for dispersion
control in WDM systems,” J. Lightw. Techn., 17/7, 1248-1254 (1999).
[33] C.K. Madsen and J.H. Zhao, Optical Filter Design and Analysis - A Signal
Processing Approach, John Wiley & Sons, Inc. (1999).
[34] M. Jablonski, Y. Takushima, K. Kikuchi, “The realization of all-pass filters for
third-order dispersion compensation in ultrafast optical fiber transmission systems,”
J. Lightwave Techn., 19/8, 1194-1205 (2001).
[35] P. Baumeister, “Design of multilayer filters by successive approximations,” J. Opt.
Soc. Am., 48/12, 955-958 (1958).
[36] J. A. Dobrowolski, “Completely automatic synthesis of optical thin film systems,”
Appl. Opt., 4/8, 937-946 (1965).
[37] H. A. MacLeod, Thin-Film Optical Filters, Institute of Physics Publishing (2001).
[38] A. Thelen, Design of Optical Interference Coatings, McGraw-Hill, Inc. (1989).
17
[39] Li Li and J. A. Dobrowolski, “Computation speeds of different optical thin-film
synthesis methods,” Appl. Opt., 31/19, 3790-3799 (1992) and references herein.
[40] J. A. Dobrowolski and R.A. Kemp, “Refinement of optical multilayer systems with
different optimization procedures,” Appl. Opt., 29/19, 2876-2893 (1990) and
references herein.
18
Chapter 3
Superprism Effect in 1-D Photonic Crystals
The first section of this chapter discusses the theory of the superprism effect in one-
dimensional photonic crystals. The second section presents experimental results
demonstrating the superprism effect in a 200-layer periodic thin-film stack with excellent
agreement between theory and experiment. In the last two sections limitations of the
periodic design and possible improvements are discussed.
3.1 Superprism Effect TheoryIn this section we explain how to calculate the superprism effect, i.e. the wavelength-
dependent group propagation angle θgr and the exit position sx as depicted in Fig. 3.1.
zy
x
θin θgr
∆sx
λ1, λ2
λ1
λ2
sx1
Fig. 3.1. Schematic of the superprism effect.
19
Close to the transmission stop-band edge, one-dimensional periodic structures exhibit a
rapid change in the phase1 and group2,3 velocity properties with wavelength. As we are
here interested in the propagation direction of beams, we need to consider the group
propagation angle θgr, which corresponds to the direction of energy flow.4 First, we
consider the case of a wide beam whose behavior can be approximated by a plane wave
calculation. The propagation of a plane wave is described by a single wavevector k. We
can place the coordinate system such that this wavevector k only has components in the
x- and z-directions and is thus given by k = β x + K z. The effect of a finite beam size is
discussed in Section 3.3 and in Appendix D.
For a periodic structure with a sufficient number of periods, the dispersion properties of
the structure can be modeled by Bloch theory (see Appendix A).4,5 In this approach an
infinite structure is assumed and the dispersion relation between the wavevector K in the
z-direction, the wavevector β in the x-direction, and the frequency ω is obtained using
periodic boundary conditions. The group velocity in the photonic crystal is given by
(3 - 1).
( ) zxkv kg gzgx vv +=∇= ω (3 - 1)
The group velocities vgx in the x-direction and vgz in the z-direction are given by (3 - 2)
and (3 - 3).4
constKgxv
=∂∂
=βω (3 - 2)
constgz K
v=∂
∂=
β
ω (3 - 3)
The group propagation angle θgr is calculated using (3 - 4).
( )
= −
gz
gxgr v
vK 1tan,, βωθ (3 - 4)
All angles are taken with respect to the z-axis. Thus, θgr is 0° if the beam propagates
along the z-axis. In the case that K and β are given as functions of the frequency ω and
20
the incidence angle θ, we can transform (3 - 4) to (3 - 5) by performing a coordinate
transformation and carefully calculating the partial derivatives as shown in Appendix B.
( ) ( ) ( )
∂∂
∂∂
−= −
θωθβ
θωθ
ωθθ,,tan, 1 K
gr (3 - 5)
This group propagation angle is of course identical to the one obtained by determining
the normal n in a wavevector diagram plotting contours of constant frequency.2,3,4
( ) ( ) zxnθ
ωθβθ
ωθ∂
∂+
∂∂
−=,,K (3 - 6)
Finally, the exit position in reflection, sx, along the surface of the dielectric stack in the
x-direction is given by (3 - 7).
( )gz
gxgrx v
vLLs 2tan2 == θ (3 - 7)
L is the total stack thickness and θgr can be calculated either by (3 - 4) or by (3 - 5). In a
bulk material the group propagation angle θgr changes only slowly with wavelength.
Therefore, beams of different wavelengths exit the material at approximately the same
position. Close to the stop band in a periodic dielectric stack though, the group
propagation angle changes rapidly with wavelength. Due to this superprism effect, beams
of different wavelengths exit the material at different positions and are spatially
demultiplexed.3
Periodic thin-film stack
Operation in transmission Operation in reflection
Mirror
Periodic thin-film stack
θinθin
(a) (b)
Fig. 3.2. 1-D photonic crystal operated in transmission (a) or reflection (b).
21
Fig. 3.2 shows that the stack can be operated either in transmission or in reflection. All
stacks presented here are operated in reflection as this doubles the spatial shift and allows
easily for multiple bounces off the stack. The next section compares the theoretically
calculated and the experimentally obtained shift with wavelength for a 100-period
dielectric stack.
3.2 Experimental ResultsHere experimental results are discussed for a 100-period dielectric stack consisting of
alternating layers of SiO2 (n=1.456 at 880 nm) and Ta2O5 (n=2.06) with a total stack
thickness of 30 µm on a quartz substrate (n=1.52). All the layers have a physical
thickness of 150 nm. Appendix G lists the layer composition of this stack. The stack is
operated in reflection as shown in Fig. 3.2(b) except that no mirror is applied to the
backside and there is thus some transmission loss. A schematic of the experimental setup
is shown in Fig. 3.3.
Tunable Laser
CCDOscilloscope TVPC
Spherical lenses
Substrate
θin
Multilayer thin-film stack
Cylindrical lens
Fig. 3.3. Schematic of the experimental setup.
Light from a tunable laser is focussed onto the tilted sample under test. The tunable laser
beam is incident onto the dielectric stack through the substrate. In a telescope style setup
a magnified image of the exiting light is focussed onto a CCD camera. Due to the light
propagation through a tilted plate, the focal point in the plane of the light beams is
22
different from the one perpendicular to the plane. The cylindrical lens compensates for
this difference, such that both beam directions are focussed on the CCD and the beam
appears approximately circular. The CCD trace is observed on the TV and oscilloscope as
well as stored on a computer for further data evaluation.
Wavelength in nm
Posi
tion
in µ
m
Quartz substrate
100-period SiO2-Ta2O5thin-film stack
Shiftingbeam
Frontreflection
Fig. 3.4. Experimentally observed intensity on a CCD trace as a function of
position and wavelength for a 100-period one-dimensional photonic crystal.
Fig. 3.4 shows the intensity observed along one CCD trace as a function of position and
wavelength for an incidence angle of 40°, p-polarization, and a spot size (Gaussian beam
radius) of 4.7 µm. The spot size is the incident Gaussian spot size in vacuum
perpendicular to the direction of beam propagation.6 Two exiting beams are observed (as
well as a weak reflection off the air-substrate-interface, which is not shown). One beam is
caused by a reflection off the front of the dielectric stack due to the impedance mismatch
and does not change position as a function of wavelength. The intensity of this beam
increases relative to the second beam as the reflectivity of the stack increases closer to the
stop-band edge. The second beam is the real signal beam that propagates through the
dielectric stack twice as seen in Fig. 3.2(b).
Fig. 3.5 plots the center position of the beams as a function of wavelength. The center
positions are obtained by fitting the data to a Gaussian beam shape, adjusting position,
width, and amplitude. Excellent agreement between the experimentally observed shift
23
compared to the theoretically expected shift along the x-direction is obtained for the
shifting beam. As expected, the front reflection is not shifting with wavelength.
870 875 880 885 890
0
5
10
15
Shifting beam (experiment)Front reflection (experiment)Shifting beam (Bloch theory)
Wavelength in nm
Wavelength in nm
Shift
in µ
m
Fig. 3.5. Comparison of the theoretically expected and experimentally observed
shift of the beam center position as a function of wavelength for a 100-period one-
dimensional photonic crystal.
3.3 Finite Number of Periods and Finite Beam WidthEvaluating the experimental results in section 3.2, we find two main differences between
the experimental results and the Bloch calculation modeling. The first difference is that
we observe two reflected beams instead of just one. The second difference is that Bloch
theory predicts an increase of the spatial shift up to the stop band edge at 901 nm with the
group propagation angle approaching 90° and the shift thus approaching infinity, while
we observe only an increase in the shift up to around 892 nm. At larger wavelengths the
shifting beam appears very distorted and no clear peak can by determined. Here it is
discussed how these two differences are based on the finite number of periods and the
finite beam width.
As Bloch theory models the propagation of plane waves in an infinite medium, it cannot
predict the front reflection off the interface between substrate and stack nor can it be used
to calculate the transmission loss. These are due to the finite extent of the real stack. In
24
order to correctly predict the reflected and the transmitted amplitudes, a different type of
calculation is needed. Here we use a transfer matrix method to calculate the relationship
between the incident, reflected, and transmitted field of plane waves for arbitrary thin-
film stacks. Appendix C describes the calculation details of this method. Using the
transfer matrix method, the reflectance and shift are calculated as a function of
wavelength for the 100-period stack discussed in the previous section. The plane wave
light is incident from the substrate side. Fig. 3.6 plots the results. Observe that the shift
calculated for very low reflected amplitudes is not correct due to numerical inaccuracies.
870 875 880 885 890 895 900 905 9100
0.2
0.4
0.6
0.8
1
Rp
λ2
nm
870 875 880 885 890 895 900 905 9100
20
40
60
80
sp
um
λ2 λ2,
Wavelength in nm
Ref
lect
ance
Wavelength in nm
Shift
in µ
m
Fig. 3.6. Reflectance and shift as a function of wavelength predicted using the
transfer matrix method for plane waves.
25
The obtained shift is very different from what is expected from Bloch theory. Instead of a
smoothly increasing shift with wavelength, we see oscillations that become more and
more rapid approaching the bandedge at 902 nm. These oscillations are due to a
combination of the finite nature of the stack and the plane wave calculation. Part of the
plane wave energy is reflected off the interface between substrate and stack. Another part
propagates through the stack, is reflected off the stack–air interface, propagates through
the stack again and is transmitted into the substrate. Other parts of the plane wave
perform even more bounces in the stack. Considering now at the total reflected light, we
see that it is composed of light that traveled different paths. Therefore, the interference of
these different parts causes oscillations in the reflectance and the shift. The stack behaves
like a Fabry-Perot resonator with transmission on resonance.
In order to separate out the part that experiences the superprism effect, we need to
eliminate the interference between parts of the light that have traveled different paths.
This can be done by reducing the spot size. Once the spot size is sufficiently small, the
different light paths lead to a splitting of the incident beam into different exiting beams as
seen in section 3.2. Before reducing the spot size, we observed strong interference effects
in our experiments. To correctly predict the transmitted and reflected beam parts of a
tightly focussed beam, we need to simulate the transfer of spatially limited beams through
thin-film stacks. In this research, I implemented two methods for calculating the
propagation of beams – the Finite Difference Time Domain (FDTD) technique7 and a
discrete Fourier decomposition technique.8
FDTD is based on replacing the derivatives in Maxwell’s equations by differences.7
Choosing the time and position steps correctly, the propagation of any type of
electromagnetic (EM) wave can be calculated. The advantage of FDTD is that it can be
used for calculating the propagation through arbitrary structures. The disadvantage is that
the calculation can become very time consuming. The position steps should be chosen at
approximately a tenth of the smallest feature size. If all features are larger than the
wavelength λ of the light, this corresponds to λ/10. But since the multilayer thin-film
stacks of interest here can have layer thicknesses smaller than λ /10 (e.g. as layers as thin
26
as 20 nm can be fabricated today) the position step size has to be chosen even smaller.
This also results in a smaller time step size. Such calculations therefore quickly become
time consuming and the accuracy requirements have to be stringent to prevent
propagation of calculation inaccuracies.
A multilayered stack is a linear system for sufficiently small intensities. Furthermore, it is
space-invariant along the layers, i.e. the fields only depend on the difference between
input and the exit position, not on the absolute position along the stack. The propagation
of arbitrary electromagnetic waves through such linear, space-invariant systems can be
calculated by decomposing the input field into elementary components, these can be
propagated individually, and the output field is obtained by summing the propagated
individual components.8 This technique is discussed in detail in Appendix D. By
performing a Fourier transform on the input field, the field is decomposed into plane
wave components that have identical wavelength, but different propagation directions.
These plane waves can be propagated using the transfer matrix method discussed in
Appendix C. Finally, superposing the individually propagated plane waves, the total field
is obtained.
Considering that a spatially limited beam can be described as a superposition of plane
waves with different incidence angles, we can understand why we only observe an
increase in the shift up to around 892 nm, even though Bloch theory predicts an increase
of the spatial shift up to the stop band edge at 901 nm with the group propagation angle
approaching 90° and the shift thus approaching infinity. The reason for this discrepancy
is that Bloch theory is only exactly correct for an infinitely wide beam as only one
incidence angle is included in the calculation.
For a Gaussian beam with a spot size (1/e2-intensity radius) of 4.7 µm at 880 nm and an
incidence angle of 40°, the intensity has decreased to 1/e2 for beam components at
incidence angles of 36.5° and 43.5° (see Appendix D). As seen in Fig. 3.7, the stop band
edge is a function of the incidence angle. Therefore, different parts of the beam are
shifted by different amounts leading to beam widening and distortion. These distortions
27
limit the usable portion of the theoretically predicted shift and clarify why we only
observed a beam shift up to 892 nm. At longer wavelengths part of the beam is already
within the stop band, distorting the beam shape severely.
860 870 880 890 900 91020
30
40
50
60
70
80
90
40 deg36.5 deg43.5 deg
Wavelength in nm
Gro
up p
ropa
gatio
n an
gle
in d
eg
Fig. 3.7. Group propagation angle as a function of wavelength calculated using
Bloch theory with incidence angles of 40° (solid), 36.5° (dash), and 43.5° (dash-
dot). For a Gaussian beam with a spot size of 4.7 µm at 880 nm and a center
incidence angle of 40°, the intensity has decreased to 1/e2 for beam components at
incidence angles of 36.5° and 43.5°.
This also explains why we will aim for a linear shift with wavelength in the following
chapters – for a linear shift with wavelength the beam shape at the exit surface is
independent of wavelength. The beam gets stretched at the output surface, but not
distorted. This stretching corresponds to a change in the focal point and is eliminated by
choosing the focusing lens correspondingly. In the case that the beam is focused in both
the x-direction and the y-direction, some polarization mixing occurs, since some of the
beam components have wavevector components in the y-direction. This has to be taken
into account for determining the exiting beam shape. Since the wavelength multiplexing
device discussed here does not require strong focusing in the y-direction, we limit
28
ourselves here to elliptical beams that are strongly focused in the x-direction and little
focused in the y-direction. In this case we can conclude that a constant dispersion for the
polarization of interest corresponds to an undistorted beam shape. Unfortunately, due to
the limited number of degrees of freedom available, a one-dimensional periodic dielectric
stack with two layers per period cannot be designed to exhibit a linear shift with
wavelength.
In order to correctly predict the front reflection off the 100-period stack and the reflected
shifting beam, the propagation of the incident Gaussian beam with w0=4.7µm is
calculated using the Fourier decomposition technique discussed in Appendix D. Fig. 3.8
plots the resulting reflected intensities and the experimental data for four different
wavelengths. The left beam is the front reflection and the right beam the shifting beam.
Very good agreement for both amplitude and position of the beams is obtained.
Furthermore, the broadening of the shifting beam as the bandedge is approached is
modeled correctly in the simulation. The extra oscillations in the experimental data are
Airy disks caused by beam clipping.
For a more detailed comparison, Fig. 3.9(a) plots the amplitudes of the stack reflection
and the shifted beams as a function of wavelength obtained by performing a Gaussian
beam fit to both experiment and simulation. Fig. 3.9(b) shows the spot size along the
interface. Finally, Fig. 3.9(c) graphs the shift obtained in experiment and theory. Both the
experimental and the theoretical spot sizes as well as the center positions of the shifting
beam are not calculated correctly for wavelengths longer than 892 nm as the intensity is
too low and the Gaussian beam fitting is therefore not working correctly. For
wavelengths shorter than 892 nm, we see that Bloch theory modeling predicts the same
shift as the transfer matrix method in combination with Fourier decomposition, but the
beam modeling is necessary to correctly predict amplitude and shape of the shifting beam
as seen in Fig. 3.8.
29
40 20 0 20 40 60 80
0
0.2
0.4
40 20 0 20 40 60 80
0
0.2
0.4
40 20 0 20 40 60 80
0
0.2
0.4
40 20 0 20 40 60 80
0
0.2
0.4
Position in µm
Nor
mal
ized
Inte
nsity
λ = 870 nm
λ = 880 nm
λ = 890 nm
λ = 894 nm
Experimental DataTransfer Matrix Simulation
Fig. 3.8. Intensity reflected from the 100-period stack as a function of wavelength.
Red shows the experimental data and blue the simulation results obtained using the
transfer matrix method and Fourier decomposition of a Gaussian beam.
30
870 875 880 885 890 895 9000
0.2
0.4
0.6
0.8
1
Wavelength in nm
Nor
mal
ized
inte
nsity
Wavelength in nm
Nor
mal
ized
Inte
nsity
(a)
870 875 880 885 890 895 9000
5
10
15
Wavelength in nm
Spot
size
in m
icro
n
Wavelength in nm
Spot
Siz
e in
µm
(b)
31
870 875 880 885 890 895 9000
5
10
15
20
25
Shifting beam (experiment)Stack reflection (experiment)Shifting beam (simulation)Stack reflection (simulation)Shift using Bloch calculation
Wavelength in nm
Shift
in m
icro
n
Wavelength in nm
Shift
in µ
m
(c)
Fig. 3.9. Normalized intensity (a), spot size along interface (b), and beam shift (c)
for the stack reflection and the shifting beam as obtained from a Gaussian beam fit
to experiment and simultaion.
In conclusion, we have seen in this section that the finite number of periods in the one-
dimensional photonic crystal results in reflections off the interfaces with the surrounding
media. These reflections result in interference effects. To see the superprism effect, the
incident beam has to be focussed tightly enough to prevent field overlap between the
different reflections. If this is achieved, the shifting beam portion is modeled correctly by
Bloch theory except for the wavelength region close to the bandedge. Since a focussed
beam can be seen as a superposition of plane waves with different propagation directions,
different parts of the beam see a different dispersion curve. This is particularly noticeable
close to the bandedge as some components are still propagating in the stack while others
32
are within the bandgap. Therefore, the finite beam width leads to beam distortions close
to the bandedge.
3.4 Improved Superprism StructuresIn the previous sections we have seen that we can use the photonic crystal superprism
effect to spatially offset beams of different wavelength. We also found, though, that part
of the incident light is lost by reflection off the front of the dielectric stack and by
transmission through the stack. The transmission loss can easily be eliminated by
applying a reflective coating to the backside of the periodic stack. This could for example
be a gold coating or a further dielectric mirror stack.
The reflection off the front of the stack is due to an impedance mismatch between the
substrate and the periodic dielectric stack that is caused mainly by the sudden periodicity
and not as much by the difference in refractive index.9 The same problem occurs in
corrugated waveguide structures and in fiber Bragg gratings. In the case of corrugated
waveguides the reflection can be eliminated by a tapering of the surface relief,2 while in
the case of apodized fiber Bragg gratings the refractive index contrast is slowly
increased.10,11 To prevent the reflection off the front of the thin-film stack, we could use a
“tapered” Bragg stack.9 In such a Bragg stack the periodicity is slowly “turned on” by
increasing the amount of high index material in each period.
Fig. 3.10(a) shows the layer thickness as a function of the layer number for a 60-layer
λ/4-Bragg stack and for a 60-layer impedance matched Bragg stack. The composition of
the stack is given in Appendix G. For the periodic λ/4-Bragg stack the layer thicknesses
dHB of the high index material and dLB of the low index material are calculated using
(3 - 8) and (3 - 9).
( ) 2sin14
−
=
HH
BHB
nn
dθ
λ (3 - 8)
33
( ) 2sin14
−
=
LL
BLB
nn
dθ
λ (3 - 9)
In these equations λB is the Bragg wavelength, nH and nL are the refractive indices, and θ
is the propagation angle in vacuum. In order to achieve impedance matching, the first PDC
periods of the stack are chirped as a function of the period p as given in (3 - 10) and
(3 - 11). In our example stack, the first 25 periods are chirped, while the layer thicknesses
of the last five periods are calculated using (3 - 8) and (3 - 9).
( )( )
05.1
2sin14
−
=DC
HH
BHBIM P
p
nn
pdθ
λ (3 - 10)
( )( ) 2
05.1
sin14
42
−
−
=
LL
DC
BB
LBIM
nn
Pp
pdθ
λλ
(3 - 11)
The Bragg wavelength is 850 nm for all periods for both stacks. The refractive indices are
1.45 and 2.09, the incidence angle in vacuum is 45°, and only p-polarization is
considered. The increasing ratio between high and low index material is clearly visible
for the impedance-matched stack. Fig. 3.10(b) shows the reflectance of the two different
stacks. Both stacks have a gold coating applied to the backside to eliminate transmission
loss. The reflectance is close to unity for both stacks. Fig. 3.10(c) shows the group
propagation angle as a function of wavelength for the two stacks and compares it to the
group propagation angle obtained from Bloch theory. Both reflectance and group
propagation angle are calculated using the transfer matrix method for plane waves.
Since the shifting beam and the reflection off the front of the stack interfere for the
periodic Bragg stack, we see large oscillations in the calculated group propagation angle.
The group propagation angle obtained for the impedance matched Bragg stack on the
other hand oscillates only slightly and follows the Bloch calculation curve well. Thus, we
34
see that the impedance matching eliminates the front reflection successfully. If no
oscillations occur in the plane wave calculation, this signifies that all of the intensity
propagates as a single beam through the structure. See Chapter 10 for a more detailed
discussion on this topic. The impedance matched stack is not ideally periodic anymore
and therefore does not track the Bloch curve close to the bandedge.
10 20 30 40 50 600
100
200
300
400
Bragg stackImpedance matched Bragg stack
Period Number
Bra
gg W
avel
engt
h in
nm
Layer number
Laye
r thi
ckne
ss in
nm
(a)
850 900 950 1000 1050 1100 1150 12000
0.2
0.4
0.6
0.8
1
Gold coated Bragg stackGold coated impedance matched Bragg stack
Wavelength in nm
Ref
lect
ance
(b)
35
850 900 950 1000 1050 1100 1150 12000
20
40
60
80
100
Bloch calculationGold coated Bragg stackGold coated impedance matched Bragg stack
Wavelength in nm
Gro
up p
ropa
gatio
n an
gle
in d
eg
(c)
Fig. 3.10. (a) Layer thickness as a function of layer number for a λ/4-Bragg stack
and for an impedance matched Bragg stack. (b) Reflectance for both stacks having
a backside gold coating applied. (c) Group propagation angle with wavelength.
Simulations are performed for plane waves using the transfer matrix technique.
We see that impedance matching can be used to eliminate front reflections and ensure
low loss for the shifting beam. Furthermore, the dispersion just outside the reflection
band is still due to the superprism effect, since it is well described by Bloch theory for
wavelengths larger than 950 nm. Interestingly, we note that the impedance matched stack
also has high dispersion between 910 nm and 950 nm. This dispersion is not due to the
superprism effect, but due to the wavelength-dependent penetration depth caused by the
chirping of the stack.
From Fig. 3.10(c) we see that this might be a very interesting effect as the total change in
the group propagation angle is larger than the one obtained with the superprism effect.
Furthermore, the group propagation angle changes approximately linearly with
wavelength between 910 nm and 950 nm. This is interesting for practical devices as it
36
allows for equidistant channel spacing. Such a constant dispersion cannot be obtained for
periodic one-dimensional photonic crystals. All periodic structures have a similar non-
linear shift with wavelength that cannot be changed much due to the limited number of
degrees of freedom that a periodic stack has. In Chapter 5 to 7 we will investigate non-
periodic stacks that have degrees of freedom and can be designed to have both low loss
and constant dispersion. Before turning to such structures, Chapter 4 investigates the
common features of multilayer thin-film stacks with spatial dispersion.
References
[1] J. P. Dowling and C. M. Bowden, “Anomalous index of refraction in photonic
bandgap materials,” J. Mod. Opt., 41/2, 345-351 (1994).
[2] R. Zengerle, “Light propagation in singly and doubly periodic planar waveguides,”
J. Mod. Opt., 34/12, 1589-1617 (1987).
[3] B. E. Nelson, M. Gerken, D. A. B. Miller, R. Piestun, C.-C. Lin, J. S. Harris, Jr.,
“Use of a dielectric stack as a one-dimensional photonic crystal for wavelength
demultiplexing by beam shifting,” Opt. Lett. 25/20, 1502-1504 (2000).
[4] A. Yariv and P. Yeh, “Electromagnetic propagation in periodic stratified media. II.
Birefringence, phase matching, and x-ray lasers,” J. Opt. Soc. Am., 67/4, 438-448
(1977).
[5] P. Yeh, A. Yariv, C.-S. Hong, “Electromagnetic propagation in periodic stratified
media. I. General theory,” J. Opt. Soc. Am., 67/4, 423-438 (1977).
[6] A. E. Siegman, Lasers, University Science Books, Sausalito, CA (1986).
[7] A. Taflove and S. C. Hagness, Computational Electrodynamics: The Finite-
Difference Time-Domain Method, Artech House (2000).
[8] J. W. Goodman, Introduction to Fourier Optics, The McGraw-Hill Companies, Inc.
(1996).
[9] N. Matuschek, F.X. Kärtner, and U. Keller, “Theory of Double-Chirped Mirrors,”
IEEE J. Select. Topics Quantum Electron., 4/2, 197-208 (1998).
[10] F. Oullette, “Dispersion concellation using linearly chirped Bragg grating filters in
optical waveguides,” Opt. Lett. 12, 847-849 (1987).
37
[11] B. J. Eggleton, G. Lenz, N. Litchinitser, D. B. Patterson, R. E. Slusher,
“Implications of fiber grating dispersion for WDM communication systems,” IEEE
Phot. Techn. Lett., 9/10, 1403-1405 (1997).
38
Chapter 4
Physics of Spatial and Temporal Dispersion
In the Chapter 2 we introduced four physical effects resulting in high spatial dispersion in
multilayer thin-film stacks– the superprism effect in periodic structures, dispersion due to
a wavelength-dependent penetration depth, dispersion due to multiple effective roundtrips
in the structure, and a combination of penetration depth and roundtrips. In Chapter 3 we
investigated the superprism effect in one-dimensional photonic crystals in detail and
proved its existence in experiment. Before looking at non-periodic stacks with high
spatial dispersion in Chapters 5 to 7, we will derive here the relationship among spatial
dispersion, temporal dispersion, and stored energy.
These relationships show how all four different dispersion phenomena introduced in
Chapter 2 are related to a wavelength-dependent amount of stored energy in the structure.
This is very interesting as it allows the derivation of some general connections between
different structures with spatial dispersion. Before looking at stored energy dependencies,
the relationship between spatial and temporal dispersion is investigated. We will show
that spatial and temporal dispersion are approximately proportional. This is important
39
from both a systems point of view as well a for design considerations. We finish the
chapter with looking at the effects of material dispersion.
4.1 Relating Spatial and Temporal DispersionIn this section we will explore the relationship between spatial and temporal dispersion
characteristics, and explain why methods used for designing structures with temporal
dispersion can be applied in a modified form to obtain spatial dispersion. Spatial
dispersion manifests itself in a change of the beam exit position sx as a function of
wavelength as shown in Fig. 3.1. Temporal dispersion refers to the change in the group
delay τgroup as a function of wavelength. We are considering here the shift and the group
delay in reflection. A similar study can be performed for the case of transmission. The
group delay in reflection is calculated by dividing the shift sx along the x-direction by the
effective speed of light vgx in this direction. As shown in (4 - 1) we can rewrite this
expression in terms of the more commonly used dependency on the phase upon reflection
φrefl using (3 - 7) and (4 - 2), where L is the total thickness of the stack.
const
refl
constconstKx
gx
xgroup
KLsvs
=== ∂
∂=
∂∂
=∂∂
==ββ ω
φ
ωωβ
τ 2 (4 - 1)
KLrefl 2=φ (4 - 2)
In the case of a periodic structure, the relationship among the wavevector K in the z-
direction, the wavevector β in the x-direction, and the frequency ω is obtained from
Bloch theory as shown in Appendix A. For a non-periodic structure we can use the
transfer matrix technique to relate K, β, and ω. This is demonstrated in Appendix C. For
a non-periodic stack and a finite periodic stack, K is the effective wavevector for the
reflected light. Since the structure is not periodic, K is not constant throughout the stack.
The group velocities vgx and vgz are in this case also effective quantities, i.e., they
represent the total effect of the stack, but are not constant within the stack. Therefore, by
calculating the group delay τgroup, we obtain the total time elapsed from entering the stack
to exiting the stack, but we cannot determine how much delay the light incurred in each
part of the stack.
40
(4 - 1) shows that the shift experienced by a beam of light is related to the group delay in
reflection by the group velocity vgx along the layers. Here we are interested in the
dispersive properties of multilayer stacks, i.e. the change of the shift and the group delay
with wavelength. To relate spatial and temporal dispersion, we therefore need to
investigate the change of the group velocity vgx with wavelength. First, we will calculate
the group velocity in a WKB-type approximation.
The WKB-approximation (also called the semiclassical or quasiclassical approximation
in quantum mechanics) states that if the local wavelength λ(z), which is linked to the
local wavevector K(z), changes slowly with z, the accumulated phase change can be
calculated by integrating the wavevector K(z) from the start position z1 to the end position
z2.1,2 This result is exactly true for uniform media as well as infinite periodic media,
where the wavevector K is obtained from Bloch theory and independent of z as discussed
in Appendix A. The resulting phase is given by (4 - 2). The WKB-approximation has
previously been applied to calculate the accumulated phase change of a chirped Bragg
stack.3 Here we use the WKB-approximation to obtain an approximate phase upon
reflection φappr for a general multilayer stack.
As the local wavevector K(z) is not a priori known in this case, we will use the
wavevector corresponding to a uniform medium with refractive index ni for the ith layer.
Even though this is strictly only a good approximation for low index contrast stacks, we
will see in Section 4.3 that it is quite good for many structures of interest. Replacing the
WKB-integral by a sum, the approximated phase upon reflection φappr is then given by
(4 - 3), where di the layer thickness of the ith layer in the stack.
( ) ∑
−
−≈
iiiappr dn
c2
2
2, βω
ωβφ (4 - 3)
Using (4 - 2) and (4 - 3) we therefore obtain the approximate dispersion relation (4 - 4)
for the approximate wavevector Kappr in the z-direction.
41
( ) ∑
−
−=
iiiappr dn
cLK 2
21, βω
ωβ (4 - 4)
Again, this is an effective wavevector for the overall effect of the stack, not a local
wavevector. The derivatives of the approximate dispersion relation with respect to ω and
β are (4 - 5) and (4 - 6). Here it is assumed that the refractive indices ni are independent
of frequency, which is quite correct for the typical dielectrics used in multilayer thin-film
stacks. If the refractive indices do depend on frequency, the derivatives have to be
modified accordingly as discussed in Section 4.4.
( )∑
−
−=∂
∂
= i
i
ii
const
appr
cn
dc
n
LK
22
2
1,
ωβω
ωβ
β
(4 - 5)
( )∑
−
=∂
∂
= i
i
i
const
appr
cn
dc
LK
22
1,
ωβ
ωβ
β
ωβ
ω
(4 - 6)
Using the rules for taking implicit derivatives,4 we get expression (4 - 7) for vgx,appr.
∑
∑
−
−
≈∂∂
∂∂−=
∂∂
==
iii
iiii
appr
appr
constKapprapprgx cnd
cndn
cKK
vappr
22
222
2,
1
ωβ
ωβ
βω
β
ω
ωβ (4 - 7)
Substituting β = ω/c sin(θ), where θ is the incidence angle in vacuum, (4 - 7) is rewritten
as (4 - 8).
( )
( )
( )∑
∑
−
−
=
∂∂
==
iii
iiii
constKapprapprgx nd
ndn
cv 22
222
, sin
sin
sin11
θ
θ
θωβ (4 - 8)
42
We see that the resulting approximate expression for vgx,appr only depends on the
incidence angle and not on the frequency. Thus, vgx,appr is independent of frequency
within this approximation. We found numerically that vgx,appr is approximately constant
for many structures we evaluated, even though the constant value might be different from
the one calculated using (4 - 8). (4 - 8) can still be used to roughly approximate vgx
though. In section 4.3, vgx is plotted for different structures of interest validating that the
group velocity along the layers is approximately constant with frequency.
This is very interesting as we see from (4 - 1) that the change in the shift with frequency,
i.e. the spatial dispersion, and the change in the group delay with frequency, i.e. the
temporal dispersion, are proportional, if vgx is independent of frequency.5 This result
provides both physical insight and has practical consequences. As spatial and temporal
dispersion are approximately proportional, existing structures with temporal dispersion
can be modified to obtain structures with spatial dispersion. This is demonstrated in
Chapters 5 and 6. Furthermore, we see that a spatial shift with wavelength corresponds at
the same time to a temporal delay. If this is not desired, two structures with opposite
spatial dispersion can be used in series as discussed in more detail in Chapter 10. With
the correct arrangement this doubles the spatial shift and removes the temporal delay
between wavelengths. If on the other hand the temporal delay is desired, the light could
be backreflected through the same structure, canceling the spatial shift and doubling the
temporal delay.
4.2 Relationship between Dispersion and StoredEnergy
In section 4.1 we discussed that temporal dispersion and spatial dispersion are
approximately proportional for many structures of interest. Here we will investigate the
relationship among spatial dispersion, temporal dispersion, and stored energy in
multilayer thin-film stacks. As detailed in Appendix C, a multilayer thin-film stack with
uniform plane waves as incident light can be seen as a two-port system, one port on either
side of the stack. The transfer function r(β,ω) relates the reflected light to the incident
43
light, while the transfer function t(β,ω) relates the transmitted light to the incident light.
We are particularly interested here in multilayer stacks that reflect all of the light, such
that there is no transmission loss. For this case the magnitude of r(β,ω) is unity. As seen
from (4 - 1) the derivative of the phase change upon reflection φrefl with frequency
determines the group delay τgroup. Relationship (4 - 9) between stored energy and delay
has previously been derived for linear, time-invariant, lossless electrical networks6 and
microwave circuits7,8 from Tellegen’s theorem.
PW
PWW totme
group =+
=τ (4 - 9)
In this equation We is the electrical energy stored in the system, Wm is the stored magnetic
energy, and P is the incident power. Since multilayer dielectric stacks are lossless and
linear for sufficiently low energies, we expect the group delay to be proportional to the
stored energy divided by the incident power. Appendix C describes how to calculate the
time-averaged electromagnetic energy density wavg(z) as a function of the position z in the
layer. It is also detailed how the incident irradiance I⊥,inc perpendicular to the layer
interfaces is calculated. (4 - 10) and (4 - 11) give the equations that relate wavg(z) and I⊥,inc
to the total stored electromagnetic energy as well as to the incident power per unit area
Ax,y on the interface.
( )∫=L
avgxytot dzzwAW0
(4 - 10)
incxy IAP ,⊥= (4 - 11)
Neglecting material dispersion, we substitute the equations for wavg(z) and I⊥,inc from
Appendix C. The group delay is calculated from the E-field in the multilayer stack using
(4 - 12) for p-polarization and (4 - 13) for s-polarization.
( )( )[ ] ( )[ ]
( ) 0
2
0,,
0
2
,,,,,,20
2
,
cos21
expexpcos2
1
Z
nE
dzzzkjEzzkjEn
inc
incxf
L
iizibxiizifxi
i
TMgroup
θ
θ
ε
τ∫ −−−−
= (4 - 12)
44
( )[ ] ( )[ ]
( )0
2
0,,
0
2
,,,,,,02
,cos
21
expexp21
Z
nE
dzzzkjEzzkjEn
incincyf
L
iizibyiizifyi
TEgroupθ
ετ
∫ −+−−= (4 - 13)
Here, Ex,f,i, Ex,b,i, Ey,f,i, and Ey,b,i are the forward and backward propagating E-field
components along the interfaces at positions zi. The formulae differ between p- and s-
polarization due to the different relationship between E|| and the total E-field amplitude.
ninc is the refractive index and θinc is the propagation angle of the incident medium (in our
case mostly the substrate). Z0 is the vacuum impedance of 377Ω. (4 - 12) and (4 - 13) are
only exactly correct for lossless structures with zero transmission, but they are also good
approximations of the group delay for structures that have some transmission loss or
absorption. In this case (4 - 12) and (4 - 13) give an upper bound for the group delay. A
similar calculation can be performed for structures with unity transmission.6,7
Once the group delay is obtained from the stored energy, the expected shift can be
calculated using (4 - 1). Section 4.3 demonstrates the relationship between stored energy,
group delay, and shift for several example structures. This relationship is very interesting,
as it is common to all four types of dispersive structures introduced in Chapter 2 and
discussed in more detail in Chapters 3, 5, 6, and 7. We see that even though the design
algorithms and stack sequences are very different for the different structures, the origin of
the dispersion always lies in a changing amount of stored energy with frequency.9
Therefore, any linear, time-invariant, lossless stack that exhibits a change in the stored
energy with frequency will also exhibit temporal and spatial dispersion.
4.3 Sample Structures Verifying the Proportionality ofStored Energy, Group Delay, and Spatial Shift
In sections 4.1 and 4.2 we discussed that spatial shift, group delay, and stored energy are
all approximately proportional for lossless structures with unity reflectance. Here we
show that this is indeed a good approximation for many structures of interest. For all the
structures investigated in this section the layer composition is given in Appendix G. The
45
figure captions list the structure numbers for easy lookup in Appendix G. Except for the
two improved superprism structures, all structures are discussed in detail in later chapters.
Fig. 4.1 and Fig. 4.2 show results obtained for the two improved superprism structures
discussed in section 3.4. For both the gold coated Bragg stack and the impedance
matched Bragg stack the plane wave reflectance, the group delay, the group velocity in
the x-direction normalized by the vacuum speed of light c, and the spatial shift are
calculated for p- and s-polarization as a function of wavelength. The solid lines show the
exact results obtained from transfer matrix calculations for plane waves. We are not
considering focussed beams here. As discussed in Appendix D, the behavior of focussed
beams is obtained by a decomposition into plane wave components. Thus, if the
relationship among spatial shift, group delay, and stored energy is correct for plane
waves, they will also be applicable to focussed beams which are a superposition of plane
wave components.
The group delay is additionally calculated from the E-field using the relationship between
group delay and stored energy in the stack given in (4 - 12) and (4 - 13). As the
reflectance is close to unity for both stacks, we see that this approximation gives a very
good estimate of the group delay. The approximate group velocity vgx,appr is obtained
from (4 - 8). There is some deviation from the exact solution as the index contrast is not
low – here it is 2.06 to 1.456. Finally, the spatial shift is approximated by multiplying the
group delay obtained from the stored energy by vgx,appr. We see that for both structures the
spatial shift is estimated well verifying the approximate proportionality among group
delay, stored energy, and spatial shift. We conclude that the superprism effect is indeed
based on wavelength-dependent energy storage, i.e. for a larger change in the stored
energy with wavelength, a larger change in the propagation direction with wavelength is
obtained and vice versa.
46
700 800 900 10000.8
0.85
0.9
0.95
1
p-pols-pol
Wavelength in nm
Ref
lect
ance
700 800 900 10000
0.5
1
p-pol : exactp-pol : energy approx.s-pol : exacts-pol : energy approx.
Wavelength in nm
Gro
up d
elay
in p
s
700 800 900 10000.15
0.2
0.25
0.3
0.35
Wavelength in nm
vgx
/ c
700 800 900 10000
20
40
60
80
100
Wavelength in nm
Shift
in u
m
Fig. 4.1. Bragg Stack (App. G, 3-2) with Gold Coating on Backside: Reflectance,
group delay, group velocity in the x-direction, and spatial shift as a function of
wavelength (solid lines – exact calculations, dotted lines – approximations).
47
850 900 950 1000 10500.97
0.98
0.99
1
p-pols-pol
Wavelength in nm
Ref
lect
ance
850 900 950 1000 10500
0.05
0.1
0.15
0.2
0.25
p-pol : exactp-pol : energy approx.s-pol : exacts-pol : energy approx.
Wavelength in nm
Gro
up d
elay
in p
s
850 900 950 1000 10500.22
0.24
0.26
0.28
0.3
0.32
0.34
Wavelength in nm
vgx
/ c
850 900 950 1000 10500
5
10
15
20
Wavelength in nm
Shift
in u
m
Fig. 4.2. Impedance Matched Bragg Stack (App. G, 3-3) with Gold Coating on
Backside: Reflectance, group delay, group velocity in the x-direction, and spatial
shift as a function of wavelength (solid lines – exact calculations, dotted lines –
approximations).
Fig. 4.3 shows the results obtained for a double-chirped stack. This type of stack is
explained in detail in Chapter 5. It uses a wavelength-dependent penetration depth to
generate spatial dispersion as introduced in Fig. 2.1(b). Assuming a constant E-field
amplitude up to the penetration depth and zero field beyond that depth, we see from the
integral (4 - 12) or (4 - 13) that a larger penetration depth corresponds to more stored
48
energy within the stack. Therefore, the proportionality between group delay and stored
energy is no surprise.
800 850 900 9500.8
0.85
0.9
0.95
1
p-pols-pol
Wavelength in nm
Ref
lect
ance
800 850 900 9500
0.1
0.2
0.3
0.4
0.5
p-pol : exactp-pol : energy approx.s-pol : exacts-pol : energy approx.
Wavelength in nm
Gro
up d
elay
in p
s
800 850 900 9500.22
0.24
0.26
0.28
0.3
0.32
Wavelength in nm
vgx
/ c
800 850 900 9500
10
20
30
40
Wavelength in nm
Shift
in u
m
Fig. 4.3. Double-Chirped Stack (App. G, 5-6): Reflectance, group delay, group
velocity in the x-direction, and spatial shift as a function of wavelength (solid lines
– exact calculations, dotted lines – approximations).
Fig. 4.4 and Fig. 4.5 graph the results for two resonator stacks that directly use a
wavelength-dependent stored energy to obtain dispersion as introduced in Fig. 2.1(c).
These two stacks are discussed in detail in Chapter 6.
49
810 820 830 840 8500.94
0.96
0.98
1
p-pols-pol
Wavelength in nm
Ref
lect
ance
810 820 830 840 8500
0.2
0.4
0.6
0.8
p-pol : exactp-pol : energy approx.s-pol : exacts-pol : energy approx.
Wavelength in nm
Gro
up d
elay
in p
s
810 820 830 840 8500.27
0.28
0.29
0.3
0.31
0.32
Wavelength in nm
vgx
/ c
810 820 830 840 8500
20
40
60
80
Wavelength in nm
Shift
in u
m
Fig. 4.4. Gires-Tournois Resonator (App. G, 6-1): Reflectance, group delay, group
velocity in the x-direction, and spatial shift as a function of wavelength (solid lines
– exact calculations, dotted lines – approximations).
50
845 850 8550.9
0.92
0.94
0.96
0.98
1
p-pols-pol
Wavelength in nm
Ref
lect
ance
845 850 8550
0.1
0.2
0.3
0.4
0.5
p-pol : exactp-pol : energy approx.s-pol : exacts-pol : energy approx.
Wavelength in nm
Gro
up d
elay
in p
s
845 850 8550.195
0.2
0.205
0.21
0.215
0.22
Wavelength in nm
vgx
/ c
845 850 8550
10
20
30
40
Wavelength in nm
Shift
in u
m
Fig. 4.5. Coupled-Cavity Stack (App. G, 6-2): Reflectance, group delay, group
velocity in the x-direction, and spatial shift as a function of wavelength (solid lines
– exact calculations, dotted lines – approximations).
Fig. 4.6 plots the reflectance, the group delay, the group velocity in the x-direction, and
the spatial shift for a numerically optimized thin-film stack. This type of stack is
discussed in detail in Chapter 7. We argued with Fig. 2.1(d) that such a stack uses a
combination of wavelength-dependent penetration depth and stored energy to generate
spatial dispersion. Here we have discussed that a wavelength-dependent penetration
depth can also be seen as a type of energy storage effect. Indeed, we can also
approximate the spatial shift quite well from the stored energy for this structure.
51
810 820 830 840 8500.985
0.99
0.995
1
p-pols-pol
Wavelength in nm
Ref
lect
ance
810 820 830 840 8500
0.1
0.2
0.3
0.4
p-pol : exactp-pol : energy approx.s-pol : exacts-pol : energy approx.
Wavelength in nm
Gro
up d
elay
in p
s
810 820 830 840 8500.24
0.26
0.28
0.3
0.32
0.34
Wavelength in nm
vgx
/ c
810 820 830 840 8500
10
20
30
40
Wavelength in nm
Shift
in u
m
Fig. 4.6. Numerically Optimized Stack (App. G, 7-2) with Gold Coating on
Backside: Reflectance, group delay, group velocity in the x-direction, and spatial
shift as a function of wavelength (solid lines – exact calculations, dotted lines –
approximations).
Finally, Fig. 4.7 shows the results obtained for a four-channel step design discussed in
Chapter 10. As in the other figures we see that the difference between the shift calculated
using the exact transfer matrix calculation and the approximated shift is mainly caused by
the error in the group velocity approximation. Transmission loss leads to an
underestimating of the group delay.
52
1500 1520 1540 1560 15800.96
0.97
0.98
0.99
1
p-pols-pol
Wavelength in nm
Ref
lect
ance
1500 1520 1540 1560 15800
0.2
0.4
0.6
0.8
p-pol : exactp-pol : energy approx.s-pol : exacts-pol : energy approx.
Wavelength in nm
Gro
up d
elay
in p
s
1500 1520 1540 1560 15800.22
0.23
0.24
0.25
0.26
0.27
0.28
0.29
Wavelength in nm
vgx
/ c
1500 1520 1540 1560 15800
10
20
30
40
50
60
Wavelength in nm
Shift
in u
m
Fig. 4.7. Four-Step Design (App. G, 10-1): Reflectance, group delay, group
velocity in the x-direction, and spatial shift as a function of wavelength (solid lines
– exact calculations, dotted lines – approximations).
In this chapter we have discussed the proportionality among spatial shift, group delay,
and stored energy for many structures of interest. The key result is that all the different
types of dispersive structures discussed in this work ultimately rely on a wavelength-
dependent stored energy. Therefore, the dispersion in all the structures has the same
physical origin. The art is to develop design algorithms for generating multilayer thin-
film stacks with the desired dispersion characteristics, i.e. with the desired change in the
stored energy as a function of wavelength. Fig. 4.3 to Fig. 4.6 already give an impression
53
of the different types of dispersion characteristics that can be obtained. We see several
structures with a linear spatial shift over different wavelength intervals and almost unity
reflectance. As discussed in Chapter 3, these characteristics are very desirable from a
practical point of view. The next three chapters will discuss three different design
algorithms that were used to generate these structures. We will also emphasize that even
though the structures use very different layer combinations to generate the dispersion, the
dispersion is in each structure based on a changing amount of stored energy with
wavelength. Before proceeding, we will finish this chapter by looking at the influence of
material dispersion on spatial dispersion and temporal dispersion in Section 4.4.
4.4 Influence of Material DispersionIn sections 4.1 to 4.3 we investigated the relationship between spatial shift, group delay,
and stored energy neglecting material dispersion, i.e., assuming that the refractive indices
ni are independent of frequency. The dispersion in the stacks in section 4.3 is due to the
multilayer structure and we refer to it as “structural” dispersion to distinguish it from
material dispersion. Here we discuss the relationship between material dispersion,
temporal dispersion, and spatial dispersion.
For frequency-dependent refractive indices, we need to use the partial derivative (4 - 14)
instead of (4 - 5) in the derivation of the approximate group velocity vgx,appr along the
layers.
( ) ( ) ( ) ( )
( )∑
−
+
−=∂
∂
= i
i
iii
i
const
appr
cn
nnn
cd
LK
22
2
dd
1,
ωβ
ω
ωω
ωωω
ω
ωβ
β
(4 - 14)
This results in the approximate group velocity vgx,appr given in (4 - 15).
( ) ( )
( ) ( ) ( ) ( ) ( )
( ) ( )∑
∑
−
−
+
=
iii
iii
iii
apprgx nd
ndn
nn
cv 22
222
, sin
sind
d
sin1
,1
θω
θωωω
ωωω
θθω(4 - 15)
54
We see that vgx,appr is no longer independent of frequency. Therefore, the group delay
τgroup and the shift sx in (4 - 1) are no longer proportional.
Furthermore, we need to include the material dispersion in the calculation of the stored
energy as given in (4 - 16), which replaces (C - 42) in Appendix C.10
( ) ( ) ( )( ) ( )
( ) ( ) ( ) ( ) ( )
∂∂
++=
∂∂
+=
2220
220
,,,2,,41
,,,41,,
zxEznznznzxH
zxEzzxHzxwavg
ωω
ωωωµ
ωωεω
µω
(4 - 16)
Let us consider the effect of the material dispersion on the stored energy, the group delay,
and the spatial shift for GaAs. The refractive index and the extinction coefficient of GaAs
are given in Fig. 4.8.11
500 1000 1500 20003
3.5
4
4.5
Wavelength in nm
Ref
ract
ive
Inde
x
500 1000 1500 20000
0.2
0.4
0.6
Wavelength in nm
Extin
ctio
n C
oeff
icie
nt
(a) (b)
Fig. 4.8. (a) Refractive index and (b) extinction coefficient of GaAs as a function of
wavelength.
Fig. 4.9 plots the reflectance, the group delay, the group velocity in the x-direction, and
the spatial shift as a function of wavelength for a 10-µm layer of GaAs for θ=45°. The
incidence material is also GaAs and the backside is assumed to be coated with a material
with constant refractive index nAu=0.16-5i (value for gold at λ=880nm). As there are no
material interfaces up to the reflective coating, there is no structural dispersion. All of the
dispersion is due to material dispersion. Fig. 4.9 plots the approximate group delay
55
calculated using (4 - 9) and (4 - 16) for the stored energy. To obtain a real result, the real
part of the group delay is taken at the end of the calculation. We see that the group delay
is still proportional to the stored energy for unity reflectance in the case of material
dispersion. For wavelengths smaller than 880 nm, GaAs has significant absorption and
the group delay cannot be approximated from the stored energy.
500 1000 1500 20000
0.5
1
p-pols-pol
Wavelength in nm
Ref
lect
ance
500 1000 1500 20000
0.2
0.4
0.6
p-pol : exactp-pol : energy approx.s-pol : exacts-pol : energy approx.
Wavelength in nmG
roup
del
ay in
ps
500 1000 1500 20000.01
0.02
0.03
0.04
0.05
0.06
Wavelength in nm
vgx
/ c
500 1000 1500 20000
1
2
3
4
5
6
Wavelength in nm
Shift
in u
m
Fig. 4.9. 10-µm layer of GaAs sandwiched between GaAs on the incidence side and
a backside gold coating: Reflectance, group delay, group velocity in the x-direction,
and spatial shift as a function of wavelength (solid lines – exact calculations, dotted
lines – approximations).
56
The group velocity vgx,appr is approximated by the value of (4 - 15) at 1240 nm. Fig. 4.9
shows that this is not a good approximation and that the group velocity has significant
frequency dependency. The approximated shift does not agree with the exactly calculated
shift. While the exactly calculated shift increases with wavelength, the approximation
predicts a decreasing shift with wavelength. We conclude that temporal dispersion and
spatial dispersion are not proportional if the dispersion is due to a frequency-dependent
refractive index. In fact, we find that for a uniform dispersive material with refractive
index nu(ω), the group velocity in the x-direction vgx,u(ω,θ) is given by (4 - 17).
( ) ( )( ) ( ) ( )
( )( )( )ω
θωθ
ωω
ωωω
θθω
ug
u
uuu
ugx nc
nnn
cv,2
,,sin
dd
sin, =+
= (4 - 17)
The group refractive index ng,u(ω) is defined by (4 - 18).12
( ) ( ) ( )ωω
ωωωd
d,
uuug
nnn += (4 - 18)
The propagation angle in the uniform medium θu is calculated from the incidence angle
in vacuum θ using Snell’s law. The group velocity in the z-direction vgz,u(ω,θ) is given by
(4 - 19).
( ) ( )( )( )ω
θωθθω
ug
uugz n
cv
,,
,cos, = (4 - 19)
Therefore, we can calculate the group delay τgroup,u for propagation of light over a
distance of 2L in the uniform medium as shown in (4 - 20).
( ) ( )( )
( )( )
( )( )
2
,,
,,
sin1
2),cos(
2,
2,
−
===
ωθ
ω
θωθ
ω
θωθωτ
u
ug
u
ug
ugzugroup
nc
nLc
nLv
L (4 - 20)
Approximating the square root in the denominator of (4 - 20) by unity, we see that the
group delay τgroup,u changes proportionally to the group index ng,u(ω) and is thus a group
velocity effect. The spatial shift along the x-direction sx,u is calculated in (4 - 21).
57
( ) ( ) ( ) ( )( ) ( )
( ) ( )( )
2,,,
sin1
sin2,tan2,,,
−
===
ωθ
ω
θθωθθωθωτθω
uu
uugxugroupux
nn
LLvs (4 - 21)
Again approximating the square root in the denominator of (4 - 21) by unity, we find that
the spatial shift sx,u is inversely proportional to the refractive index nu(ω) and is thus a
phase velocity effect. Fig. 4.10 plots the relative change of the refractive index nu, the
group refractive index ng,u, the shift along the x-direction sx,u and the group delay τgroup,u
in GaAs. It is clearly visible that the group delay is a group effect, while the spatial shift
is a phase effect in the case of material dispersion.
600 800 1000 1200 1400 1600 1800 20000.8
1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
n/n0ng/ng0s0/sGD/GD0
Wavelength in nm
Rat
ios
Re( n (λ) ) / Re( n (1300 nm) )Re( ng (λ) ) / Re( ng (1300 nm) )Re( sx (1300 nm) ) / Re( sx (λ) )Re( τ (λ) ) / Re( τ (1300 nm) )
Fig. 4.10. Change of the refractive index n, the group refractive index ng, the shift
along the x-direction sx and the group delay τ in GaAs.
For the case of structural dispersion, we found that the group delay and the spatial shift
are proportional and that both are group velocity effects. The difference between material
dispersion and structural dispersion lies in the fact that material dispersion is isotropic,
58
i.e., the refractive index changes equally with frequency for all propagation directions (of
course, there are also anisotropic materials, which we do not consider here13). Therefore,
the group velocities in the x-direction and in the z-direction change proportionally and the
changes cancel for the spatial shift. In the case of structural dispersion, on the other hand,
we introduce a rapid variation of the group velocity in the z-direction with frequency due
to the layered structure, but there is no variation of the refractive index along the layers in
the x-direction.
Fig. 4.11 shows an example of the combined effect of structural dispersion and material
dispersion. A 500-nm layer of GaAs is sandwiched between an incident material with
n=2 and a backside gold coating (assumed to have a constant refractive index of
nAu=0.16-5i) for θ=45°. Using (4 - 16) for the stored energy, good agreement between the
exact and the approximate group delay result. With part of the dispersion being due to
structural dispersion, the shift is again better approximated by multiplying the group
delay by a constant group velocity value.
In conclusion, we see that material dispersion and structural dispersion have a different
influence on the temporal dispersion and the spatial dispersion. In the case of structural
dispersion, spatial shift, group delay, and stored energy are proportional for unity
reflectance. In the case of material dispersion, on the other hand, stored energy and group
delay are proportional, but the spatial shift is a phase effect and only occurs due to the
frequency-dependent refraction at the incident surface. For the dielectric stacks used in
this thesis the material dispersion is very low and can be neglected. All of the observed
dispersion is due to structural dispersion.
59
1000 1200 1400 1600 1800 20000.85
0.9
0.95
1
p-pols-pol
Wavelength in nm
Ref
lect
ance
1000 1200 1400 1600 1800 20000
0.01
0.02
0.03
p-pol : exactp-pol : energy approx.s-pol : exacts-pol : energy approx.
Wavelength in nm
Gro
up d
elay
in p
s
1000 1200 1400 1600 1800 20000.03
0.04
0.05
0.06
0.07
0.08
Wavelength in nm
vgx
/ c
1000 1200 1400 1600 1800 20000
0.1
0.2
0.3
0.4
0.5
Wavelength in nm
Shift
in u
m
Fig. 4.11. 500-nm layer of GaAs sandwiched between an incident material with
n=2 and a backside gold coating: Reflectance, group delay, group velocity in the x-
direction, and spatial shift as a function of wavelength (solid lines – exact
calculations, dotted lines – approximations).
References
[1] R. Shankar, Principles of Quantum Mechanics, New York (1994).
[2] A. B. Migdal, Qualitative Methods in Quantum Theory, Addison-Wesley
Publishing Co., Inc. Redwood City, CA (1989).
60
[3] N. Matuschek, F.X. Kärtner, and U. Keller, “Theory of Double-Chirped Mirrors,”
IEEE J. Select. Topics Quantum Electron., 4/2, 197-208 (1998).
[4] I.N. Bronstein, K.A. Semendjajew, G. Musiol, H. Muehlig, Taschenbuch der
Mathematik; Verlag Harri Deutsch, Thun und Frankfurt am Main, pp. 232-237
(1993).
[5] M. Gerken and D. A. B. Miller, “Multilayer thin-film structures with high spatial
dispersion,” Appl. Opt. 42/7 (2003), 1330-1345.
[6] P. Penfield, Jr., R. Spence, and S. Duinker, Tellegen’s Theorem and Electrical
Networks, M.I.T. Press, Cambridge (1970).
[7] C. Ernst, V. Postoyalko, N. G. Khan, “Relationship Between Group Delay and
Stored Energy in Microwave Filters,” IEEE Trans. Microwave Theory and Techn.
49/1 (2001).
[8] C. Ernst and V. Postoyalko, “Comments on “Relationship Between Group Delay
and Stored Energy in Microwave Filters”,” IEEE Trans. Microwave Theory and
Techn. 49/9 (2001).
[9] M. Gerken and D. A. B. Miller, “The relationship between the superprism effect,
group delay, and stored energy in 1-D photonic crystals and photonic
nanostructures,” MRS Spring Meeting, San Francisco, CA (April 21-25, 2003).
Paper J2.7.
[10] H. A. Haus, Waves and Fields in Optoelectronics, Prentice-Hall, Inc., Englewood
Cliffs, New Jersey (1984).
[11] S. L. Chuang, Physics of Optoelectronic Devices, John Wiley & Sons, Inc., New
York (1995).
[12] G. Grau and W. Freude, Optische Nachrichtentechnik, Springer-Verlag, Berlin
(1991).
[13] A. Yariv and P. Yeh, Optical Waves in Crystals, John Wiley & Sons, New York
(1984).
61
Chapter 5
Chirped Stacks
Temporal dispersion compensation plays an important role in femtosecond lasers and in
fiber communications. It has previously been observed that double-chirped mirrors are
well suited for femtosecond laser cavities as they have a broad reflection bandwidth and
can at the same time be used to compensate for temporal dispersion in the cavity.1 In a
Bragg mirror, light is only strongly reflected for wavelengths near the Bragg wavelength.
In a simple-chirped mirror, the Bragg wavelength is slowly changed as a function of the
position in the stack. If the period length increases coming from the incidence side,
longer wavelengths penetrate deeper into the structure and thus accumulate more delay
upon reflection. For the opposite chirp, shorter wavelengths penetrate deeper and the
opposite dispersion is obtained.
As mentioned in Chapter 3 the sudden turn-on of the periodicity leads to a reflection off
the front of the dielectric stack. In section 3.4 it was shown that this reflection can be
suppressed by adding an impedance matching chirp to the stack. A stack that has both a
chirp in the wavelength to obtain dispersion and an impedance matching chirp to prevent
62
loss is called a double-chirped mirror. An analytical design of such mirrors is possible
using an exact coupled-mode theory2 and a WKB-type approximation.3
Remembering that temporal dispersion and spatial dispersion are approximately
proportional as discussed in Chapter 4, we expect that the same type of algorithm can be
used to design a structure with spatial dispersion. First a double-chirped stack with
temporal dispersion is designed as described in [3]. A standard Bragg reflector may
follow the double-chirped mirror section to increase the reflectivity to unity. This normal-
incidence design is then modified for oblique incidence by taking into account that the
Bragg wavelength λB is given by (5 - 1).
( ) ( )
−+
−=
22sin1sin12
LLL
HHHB n
dnn
dn θθλ (5 - 1)
nH and nL are the refractive indices, and dH and dL are the thicknesses of the high and low
index materials respectively. θ is the incidence angle in vacuum. Fig. 5.1 compares the
performance of five different 60-layer structures designed in this manner. The layer
sequences for these stacks are given in Appendix G. Fig. 5.1(a) shows the chirped Bragg
wavelength as a function of the period number p calculated using (5 – 2) and using five
different factors f. The chirp law (5 – 2) is chosen such that a linear shift as a function of
wavelength is obtained.3
( )pf
nmpB 02541.01800
−=λ (5 - 2)
Using this chirped Bragg wavelength, the layer thicknesses of the single-chirped
structures are calculated using (5 - 3) and (5 - 4).
( ) ( )( ) 2
sin14
−
=
HH
BHSC
nn
ppdθ
λ (5 - 3)
63
( ) ( )( ) 2
sin14
−
=
LL
BLSC
nn
ppdθ
λ (5 - 4)
In order to achieve impedance matching, the first PDC periods are double chirped as given
in (5 - 5) and (5 - 6). In our calculations, 25 of the 30 periods are double-chirped.
( ) ( )( )
05.1
2sin14
−
=DC
HH
DCBHDC P
p
nn
Ppdθ
λ (5 - 5)
( )
( ) ( ) ( )
( ) 2
2
sin14
sin12
−
−−
=
LL
HHHDC
B
LDC
nn
nnpdp
pdθ
θλ
(5 - 6)
In Fig. 5.1(b) the spatial shifts as a function of wavelength are plotted for the five
designed structures. For clarity the shift is only plotted up to the wavelength at which it
stops increasing. We see that all five structures have an approximately linear shift with
wavelength as desired.
Structures with a larger chirp in the Bragg wavelength exhibit less dispersion but show
that dispersion over a larger wavelength range. The maximum shift is approximately
constant for the given allowed ripple. There is a tradeoff between the maximum shift and
the ripple4 in the shift for a given number of layers. That is, we could design a structure
with less ripple and less shift, or a structure with more ripple and more shift depending on
the application requirements.
Because the dispersion increases with decreasing chirp in the Bragg wavelength, the
maximum dispersion is achieved with a simple Bragg stack (though the ratio between
high and low index material in a period is still chirped for impedance matching). Note
that this linear shift appears within the stop-band of the single-chirped Bragg stack. It is
64
not identical to the non-linear shift observed just outside the stop-band as discussed in
Chapter 3.
10 20 30750
800
850
900
950
1000
1050
Period Number
Bra
gg W
avel
engt
h in
nm
850 900 950 1000 10500
5
10
Wavelength in nm
Shift
in u
m
Wavelength in nmSh
ift in
µm
Period Number
Brag
g W
avel
engt
h in
nm f=0.5
f=0.33f=0.2f=0.1f=0
Period
(a) (b)
Fig. 5.1. (a) shows the Bragg wavelength as a function of the position in the
structure for five different 60-layer SiO2/Ta2O5 double-chirped mirror designs. In
(b) the theoretical spatial shift as a function of wavelength is plotted for an
incidence angle of 45° and p-polarized light. An approximately linear shift is
observed for all five designs. The dispersion increases with decreasing chirp in the
Bragg wavelength. The maximum dispersion is achieved with a single-chirped
Bragg stack (f=0).
The double-chirped mirror design algorithm cannot be used to obtain a higher dispersion
for the given number of layers. This limitation of the dispersion is not a fundamental
physical limit for layered dielectric structures. As we will see in Chapter 6, structures
with larger dispersion can be designed. The limitation is rather due to the fact that, in this
algorithm, the degrees of freedom are limited, and only a subset of all possible structures
is considered. Specifically, the algorithm only allows for a monotonic change in the
65
Bragg wavelength. We can easily imagine that relaxing this requirement would allow for
other structures that might have a larger dispersion.
The calculations in Fig. 5.1 are done using the transfer matrix method given in Appendix
C applied to the center ray of the beam. This plane-wave approximation correctly predicts
the behavior of a Gaussian beam if the shift does not exhibit any rapid changes with
wavelength. Rapid oscillations in the shift are indicative of interference effects between
different beams.
An example of a structure with strong oscillations is the periodic structure discussed in
Chapter 3. The plane-wave approximation predicts strong oscillations in the shift with
wavelength, but with sufficient focussing, we actually see two separate beams – one that
is stationary corresponding to a reflection off of the front of the stack and one that
changes position with wavelength. These distinct beams cannot be predicted by the
plane-wave approximation and the full beam-behavior has to be modeled using, e.g., the
Fourier decomposition technique in Appendix D.
For the double-chirped structures discussed in this section on the other hand, the plane-
wave approximation can be used because all reflections except the desired shifted beam
are suppressed sufficiently well and no rapid changes appear in the dispersion. In
particular, the reflection off the front surface of the stack has been suppressed by the
impedance matching chirp.
To visualize the origins of dispersion, we plot the E-field amplitude of the forward
propagating wave as a function of the position in the structure. Since the structures we are
discussing here are nearly 100% reflecting and lossless, the amplitudes of the forward
and backward propagating waves are approximately equal for any specific wavelength.
Interference effects between the forward and backward propagating waves form a rapidly
oscillating standing wave pattern. The amplitude of the forward propagating wave
outlines the envelope of this standing wave pattern.
66
In Fig. 5.2(a) an example of a 200-layer double-chirped structure is shown. Appendix G
gives the layer composition for this stack. The structure is again designed to exhibit a
linear shift with wavelength which is well achieved as seen in Fig. 5.2(b). In Fig. 5.2(c)
we see that the field penetrates deeper into the structure with increasing wavelength,
resulting in both temporal and spatial dispersion.
Note a surprising behavior shown by the shift with this stack. Suppose we propagated a
beam through a piece of material with the same thickness as our dielectic stack (28 µm),
and a refractive index of 1.6, corresponding to the average index of our dielectric stack
material, and reflected it off of the very back of that material, corresponding to the largest
penetration possible. Then, from a simple calculation based on Snell’s law, we would
obtain a shift of 25 µm. But our shift here is 35 µm. Thus, the structure exhibits
additional dispersion that cannot be explained by the penetration depth. This extra
dispersion is due to stored energy.
In Fig. 5.2(c), larger than unity E-field amplitudes correspond to energy buildup. We
observe that only the shortest wavelength plot does not exhibit any energy buildup. For
the other wavelengths the field forms an Airy-type standing wave pattern as is usual for a
linearly changing potential. As was explained in Chapter 4, stored energy results in
dispersion explaining the larger than expected total shift in Fig. 5.2(b). Fig. 4.3 shows
that the observed shift can be approximated well from the total stored energy in the
structure.
Note also that this structure stores energy without the use of a front mirror, but solely due
to the shape of the potential. In other words, the structure stores energy without requiring
a resonant cavity in the conventional sense; certainly we cannot describe the energy
storage in terms of a cavity with two fixed mirrors. The position of the energy buildup
changes continuously with wavelength. This effect might be interesting for future active
and passive devices, since it has the energy build-up of a simple resonant effect without a
narrow resonant response.
67
0
2
4
0
2
4
0
2
4
0 5 10 15 20 25 300
2
4
780 800 820 840 860 880 900 920 9405
0
5
10
15
20
25
30
35
40
Wavelength in nm
Shift
in u
m
50 100 150 2000
100
200
300
SiO2 layersTa2O5 layers
Layer Number
Phys
ical
Thi
ckne
ss in
nm
780 nm
830 nm
880 nm
930 nm
(a)
(b)
(c)
Layer numberPhys
ical
thic
knes
s in
nm
Wavelength in nm
Shift
in µ
m
Depth in µm
E-fie
ld p
aral
lel t
o in
terfa
ce in
a.u
.
Fig. 5.2. (a) Physical layer thicknesses for a 200-layer SiO2/Ta2O5 double-chirped
structure. (b) Theoretically calculated shift as a function of wavelength at 40°
incidence angle and p-polarization. The circles indicate the wavelengths and shifts
corresponding to the diagrams in (c). (c) E-field parallel to the interface of the
forward propagating wave as a function of the position in the structure for four
different wavelengths – 780 nm, 830 nm, 880nm, and 930 nm. The vertical lines
indicate the position of the interfaces between layers. Light is incident from the left,
and the structure extends from 0 µm to 28 µm.
In conclusion we demonstrated in this section that the analytical design method3
developed for generating double-chirped mirrors for temporal dispersion compensation in
femtosecond laser cavities can be modified for designing structures with spatial
dispersion. This method is very useful for broadband designs with rather low dispersion,
but it is not suitable for designing structures with high dispersion. In Chapter 6 we will
68
examine a different design method that is particularly useful for narrowband designs with
high dispersion.
References
[1] N. Matuschek, F.X. Kärtner, and U. Keller, “Theory of Double-Chirped Mirrors,”
IEEE J. Select. Topics Quantum Electron., 4/2, 197-208 (1998).
[2] N. Matuschek, F.X. Kärtner, and U. Keller, “Exact Coupled-Mode Theories for
Multilayer Interference Coatings with Arbitrary Strong Index Modulations,” IEEE
J. Quantum Electron., 33/3, 295-302 (1997).
[3] N. Matuschek, F.X. Kärtner, and U. Keller, “Analytical Design of Double-Chirped
Mirrors with Custom-Tailored Dispersion Characteristics,” IEEE J. Quantum
Electron., 35/2, 129-137 (1999).
[4] M. Sumetsky, B. J. Eggleton, C. M. de Sterke, “Theory of group delay ripple
generated by chirped fiber gratings,” Opt. Express, 10/7, 332-340 (2002).
69
Chapter 6
Resonator Stacks
In this chapter we investigate resonator type structures, i.e., structures with a wavelength-
dependent amount of stored energy. As first introduced in Fig. 2.1(c) the wavelength-
dependent stored energy corresponds to a wavelength-dependent number of effective
round-trips through the resonator cavity, which leads to a wavelength-dependent spatial
shift of the exiting light. This intuitive picture was confirmed by the calculations in
Chapter 4 showing that the change in stored energy with wavelength is proportional to
temporal and spatial dispersion.
Simple examples of resonator structures are Fabry-Perot and Gires-Tournois
resonators.1,2,3,4 A Fabry-Perot resonator consists of a cavity between two partial
reflectors usually (but not necessarily) of equal reflectivity. At the resonant wavelength
the transmittance of the filter is unity (for equal reflectivities) and a maximum amount of
energy is stored in the cavity. A Gires-Tournois structure consists of one partial reflector,
a cavity, and a 100% reflector. Technically, such a resonator is just a special case of a
Fabry-Perot resonator. For the Gires-Tournois resonator the reflectance is unity at all
wavelengths though again the amount of stored energy is maximum at the resonant
70
wavelength. Moving away from the resonant wavelength, the amount of stored energy
decreases. The transmittance and reflectance for the respective structures can only ever
be unity in the absence of absorption. For structures with absorption, the loss is highest at
the resonant wavelength because the beam executes the maximum number of roundtrips.
10 20 300
2000
4000
SiO2 layersTa2O5 layers
Layer Number
Phys
ical
Thi
ckne
ss in
nm
0
2
4
6
0
2
4
6
0
2
4
6
2 0 2 4 6 8 100
2
4
6
810 815 820 825 830 835 840 845 8500
5
10
15
20
25
30
35
Wavelength in nm
Shift
in u
m
Layer number
Phys
ical
thic
knes
s in
nm
Wavelength in nm
Shift
in µ
m
Depth in µm
E-fie
ld p
aral
lel t
o in
terfa
ce in
a.u
.
815 nm
827 nm
830 nm
832 nm
(a)
(b)
(c)
Fig. 6.1. (a) Physical layer thicknesses for a 33-layer SiO2/Ta2O5 Gires-Tournois
resonator structure. (b) Theoretically calculated shift as a function of wavelength at
45° incidence angle and p-polarization. The circles indicate the wavelengths and
shifts corresponding to the diagrams in (c). (c) E-field parallel to the interface of
the forward propagating wave as a function of the position in the structure for four
different wavelengths – 815 nm, 827 nm, 830nm, and 832 nm. The vertical lines
indicate the position of the interfaces between layers. Light is incident from the left,
and the structure extends from 0 µm to 8 µm.
Fig. 6.1 gives an example of a Gires-Tournois structure. This structure consists of 33
layers. The layer sequence is listed in Appendix G. A 3.6-µm SiO2 cavity is surrounded
71
by a four-period quarter wave stack on the incident side and a 12-period quarter wave
stack serving as a near unity reflector. The stack has interfaces to air on both sides. In
Fig. 6.1(a) the physical layer thicknesses are graphed. Fig. 6.1(b) plots the resulting shift
as a function of wavelength for this structure. The calculations in Fig. 6.1 are done using
the transfer matrix method given in Appendix C applied to the center ray of the beam.
This plane-wave approximation correctly predicts the behavior of a Gaussian beam if the
shift does not exhibit any rapid changes with wavelength as discussed in Chapter 5. The
non-linear shift as a function of wavelength in Fig. 6.1(b) leads to beam distortions as
observed in Section 3.3.
In Fig. 6.1(c) the E-field amplitude parallel to the interface of the forward propagating
wave is plotted as a function of the position in the structure. Comparing the shift in
Fig. 6.1(b) with the energy buildup in Fig. 6.1(c), we can clearly see that a larger energy
buildup corresponds to a larger spatial shift as expected. The field penetrates the stack to
the same distance for all wavelengths. Therefore, the shift cannot be explained by a
geometrical penetration depth argument. Fig. 6.1(b) reveals an impressive total shift of 34
µm, which is much larger than the total stack thickness of 8 µm.
Resonators allow for high dispersion over a narrow wavelength range. Unfortunately, the
degrees of freedom in the design of a single-cavity resonator structure are very limited.
For example, in the case of the Gires-Tournois structure we can only choose the
reflectivity of the partial reflector and the cavity thickness. Not surprisingly, the shape of
the dispersion for all Gires-Tournois structures is very similar to the one depicted in
Fig. 6.1. In order to match desired dispersion characteristics more closely, the number of
cavities in the structure can be increased. An analytical procedure for designing such
coupled-cavity filters using a digital lattice technique has previously been established
both for microwave filters5,6,7 and thin-film structures.3,8
We are particularly interested in the design of allpass structures, i.e., structures with
constant unity reflectance and wavelength-dependent phase properties. Allpass filters
have previously been designed for temporal dispersion compensation.2,9 Here we will
72
show with an example that the analytical procedure can be modified to design structures
with spatial dispersion. Our goal is again to design a structure exhibiting a linear spatial
shift along the x-direction with wavelength. The example structure consists of five
reflectors with four cavities in between. The last reflector has unity reflectance. Thus, the
reflectance of the structure is 100% neglecting any absorption. The design procedure has
three stages – first the desired transfer function is approximated by an allpass polynomial,
next the reflection coefficients of the reflectors between the cavities are derived using an
order reduction technique, and finally the reflectors and cavities are implemented in a
physical structure.
Initially, all cavity lengths are assumed to be equal with a cavity round-trip time T. The
transfer function of an allpass filter with N cavities is then given by (6 - 1), where
z-1 = e-jωT. ANR(z) is the reverse polynomial of AN(z) of order N.
)()(
)(
0
0
zAzA
za
zazzH
N
RN
N
n
nn
N
n
nn
NAP ==
∑
∑
=
−
=− (6 - 1)
In the first step of the design process, a polynomial AN(z) is determined such that the
phase of HAP(z) approximates the shape of the desired phase response. The allpass
polynomial coefficient a0 is set to unity, which corresponds to the last reflector having a
reflection coefficient of unity amplitude. The resulting transfer function HAP(z) exhibits a
periodic behavior with the period given by the free spectral range FSR=1/T.3 Here we
want to design a structure with a linear shift as a function of wavelength. In Chapter 3 it
is shown that the shift along the exit interface is calculated by (6 – 2).
const
constK
gz
gxx K
Lvv
Ls=
=
∂∂
∂∂==
βω
βω22 (6 - 2)
L is the total stack thickness, vgx the group velocity in the x-direction, vgz the group
velocity in the z-direction, K and β are the wavevectors in the z- and in the x-direction,
and ω the frequency. K is related to the phase upon reflection φrefl by (6 – 3).
73
reflLK φ
21
= (6 - 3)
From Chapter 4 we know that the group velocity vgx along the layers is approximately
constant. Therefore, it follows from (6 – 2) and (6 – 3) that the derivative of the phase
φrefl should change approximately linearly with respect to frequency in order to obtain a
shift that is linear with frequency. We will initially design for a linear change of the shift
with frequency instead of wavelength. Due to the narrow bandwidth of the generated
designs, we can easily change this to wavelength in the last optimization step of the
procedure.
0 0.1 0.2 0.3 0.4 0.5
Desired characteristicDesign after 1. least-squares runFinal design
Normalized frequency
Phas
e de
rivat
ive
in a
.u.
Normalized Frequency
Phas
e de
rivat
ive
in a
.u.
Fig. 6.2. Results for approximating the desired phase characteristics by a fourth
order allpass polynomial.
The crosses in Fig. 6.2 show the desired phase derivative as a function of frequency. As
the resulting function is periodic with the free spectral range, the frequency normalized to
the free spectral range FSR can be used for simplicity. The fourth order allpass
approximation shown in Fig. 6.2 is obtained using a least squares algorithm.9,10 This is a
numerical algorithm for finding an allpass polynomial of the form (6 – 1) to approximate
the desired phase characteristics. The final coefficients of the polynomial are given in
(6 – 4).
74
=
153.0661.0453.1779.11
a (6 - 4)
Alternatively, an analytic algorithm based on the calculation of the cepstral coefficients
can be used for finding the allpass polynomial coefficients.11 This technique essentially
uses a truncated Fourier decomposition of the desired transfer function. Here I used the
first, numerical technique as I did not find out about the second technique until after I had
completely designed this example. The second technique is probably superior as it is
analytical and always generates realizable reflection coefficients, i.e. reflection
coefficients with amplitudes smaller or equal to unity.
850 900 9500
5
10
15
20
25
30
Lc=1.7 umLc=2.9 umLc=5.8 um
Wavelength in nm
Shift
in u
m
Wavelength in nm
Shift
in µ
m
0 2 4Normalized frequency
Phas
e de
rivat
ive
in a
.u.
Normalized Frequency
Phas
e de
rivat
ive
in a
.u.
(a) (b)
Fig. 6.3. (a) Periodic phase derivative of the transfer function as a function of the
normalized frequency. (b) The expected shift for a 4-cavity structure is plotted as a
function of wavelength for three different cavity optical thicknesses.
75
Fig. 6.3(a) plots the phase derivative of the approximate transfer function with the
coefficients a given in (6 - 4). Once an approximate transfer function is obtained, the
expected shift can estimated from (6 - 2) after choosing a cavity round-trip time T. The
effective group velocity vgx in the cavity is approximated from (4 - 8) choosing an
incidence angle of 45° and a refractive index of 2.09 for the cavity. Fig. 6.3(b) shows the
influence of the optical thickness of the cavity on the expected spatial shift with
wavelength. As the same transfer function HAP(z) is used for all three cavity lengths, the
shape of the shift with wavelength is the same. The total shift increases proportionately
with the cavity length, while at the same time the operating range decreases inversely
with the length. Thus, the dispersion is proportional to the square of the cavity length.
After choosing an appropriate cavity length to obtain the desired dispersion, we need to
realize the partial reflectors as thin-film structures. The reflection coefficients ri (i=1 to 4,
r5=1) of the reflectors are deduced using an order reduction technique. Starting with order
m=N, the order m of the polynomial Am(z) in (6 – 2) is reduced in each step by 1. The
coefficients of the polynomial Am(z) are am,n (n=0 to m) as defined in (6 - 1).The
algorithm for each order reduction step is given in (6 - 5) and (6 - 6).12
mmmN ar ,1 −=−+ (6 - 5)
2,
1,
1 1)()(
)(mm
mmmmm
m azzAazA
zA−
−=
−−
− (6 - 6)
For our 4-cavity sample structure, we want to realize an allpass filter HAP(z) with the
polynomial A4(z)=1+1.779z-1+1.453z-2+0.661z-3+0.153z-4 as seen from (6 - 4). In the first
step of the order reduction we set m=N=4. From (6 - 5) we calculate the reflectivity of the
first partial reflector as r4+1-4=r1 =-a4,4 =-0.153. Then we use the order reduction (6 - 6) to
obtain the polynomial of order m=4-1=3 to be A3(z)=1+1.718z-1+1.26z-2+0.399z-3. Using
again (6 - 4) we find the second partial reflector to have r4+1-3=r2 =-a3,3 =-0.399.
Continuing the order reduction, we calculate the reflection coefficients for all four partial
reflectors to have the values given in (6 – 7).
76
−−−−−
=→
=
1859.0683.0399.0153.0
153.0661.0453.1779.11
ra (6 - 7)
All reflection coefficients have to have amplitudes smaller or equal to unity. Otherwise,
the stack cannot be realized with a physical structure. Once the necessary reflection
coefficients are determined, the reflectors can be designed using standard thin-film design
procedures.4 Another approach is to approximate the reflectors by available quarter-wave
structures and correct for the error in the reflectivity by modifying the individual cavity
lengths.9 This approach is chosen here.
Table 6.1 lists the reflection coefficients for different quarter wave layer combinations.
As typical deposition systems monitor the growth of quarter wave stacks at 0° incidence
angle, we choose 980 nm quarter wave layers. These stacks will be reflective at 54°
incidence angle around 860 nm. Light is incident from the quartz substrate and the cavity
material is the high index material Ta2O5. Using Table 6.1 the first reflection coefficient
r1=-0.153 is for example approximated by a single low index quarter wave layer between
the substrate and the high index material. Similarly the other reflectors are chosen.
Layers r
SH 0.19
SLH -0.241
HLH -0.418
H(LH)2 -0.711
Layers r
H(LH)3 -0.869
H(LH)4 -0.943
H(LH)5 -0.974
H(LH)10 -0.997
Table 6.1. Reflection coefficients for different λB/4-layer combinations
(λB=980nm) of the substrate S (ns=1.52), high index material H (nH=2.06,
dH= λB/4nH), and low index material L (nL=1.456, dL= λB/4nL) at 860nm, 54°
incidence angle, and s-polarization.
For our 4-cavity example we choose an optical cavity round-trip length of 5.8 µm, Ta2O5
as the cavity material, s-polarized light, and an incidence angle of 54°. This corresponds
77
to a physical thickness of 2.6 µm for each cavity. After designing the reflectors it is
important to subtract the optical thickness of each reflector from the corresponding
optical cavity length, since they constitute part of the round-trip time. The necessity to
implement the reflector within the optical thickness of the cavity sets a lower limit on the
cavity thickness and thus the maximum achievable operating wavelength range. On the
other hand, the cavity length cannot be chosen too large, as field interference is necessary
for the operation of the device. If the distances between the reflectors are too large, a
pulse is split into several pulses and an optical rattler is obtained.13
842 844 846 848 850 852 8540
5
10
15
20
25
Desired shiftWavelength in nm
Shift
in u
m
Wavelength in nm
Shift
in µ
m
Approx. Sub L [H]24.88 L [H]24.88 LHL [H]22.98 (LH)2L [H]21.07 (LH)9L AirRefined Sub L [H]24.84 L [H]24.94 LHL [H]22.94 (LH)2L [H]20.98 (LH)9L Air
Fig. 6.4. Shift as a function of wavelength at 54° incidence angle and s-polarization
for the design with approximate reflectors (blue) and the final refined design
(black). Below the graph, the quarter wave layer sequences for both designs are
given. Remember that the quarter wave layers are for a wavelength of 980 nm and
0° incidence angle.
Fig. 6.4 plots the shift as a function of wavelength for the design with the approximate
reflectors in blue. Furthermore, the resulting layer sequence in terms of quarter wave
layers is listed as well. Note the different cavity lengths due to subtracting the optical
78
thicknesses of the reflectors from the cavity thicknesses. Due to the approximate
reflectors, the shift is not linear any more. As the final step of the design the cavity
thicknesses are numerically optimized to achieve a linear shift as a function of
wavelength. The optimization was performed using the conjugate gradient algorithm with
the Hestenes-Stiefel formula (see Chapter 7 and Appendix E for details on numerical
optimization techniques).14 Fig. 6.4 plots the resulting linear shift of the refined design
and gives the final layer sequence. The resulting SiO2/Ta2O5 thin-film structure has 33
layers and a total thickness of 15.4 µm. The layer composition is given in Appendix G.
Fig. 6.5 investigates the origin of the dispersion in more detail. In Fig. 6.5(a) the physical
layer thicknesses are graphed. The four cavities separated by quarter wave reflectors are
visible. The layer structure is also given in Appendix F. Fig. 6.5(b) plots the resulting
shift as a function of wavelength for this structure using a plane wave transfer matrix
calculation. A linear shift is obtained over a narrow wavelength range demonstrating that
this technique enables the design of narrowband structures with high dispersion.
In Fig. 6.5(c) the E-field amplitude parallel to the interface of the forward propagating
wave is plotted as a function of the position in the structure. For longer wavelengths a
larger amount of energy build-up occurs in the structure. As discussed for the single-
cavity resonators, this wavelength-dependent energy storage causes spatial and temporal
dispersion. We also see that for a structure with more than one cavity, part of the
dispersion can be attributed to a wavelength-dependent penetration depth.
79
(a)
(b)
(c)
842 844 846 848 850 852 8540
5
10
15
20
25
Wavelength in nm
Shift
in u
m
10 20 300
1000
2000
3000
SiO2 layersTa2O5 layers
Layer Number
Phys
ical
Thi
ckne
ss in
nm
Layer number
Phys
ical
thic
knes
s in
nm
0
2
4
0
2
4
0
2
4
0 5 10 150
2
4
842 nm
846 nm
850 nm
854 nm
Wavelength in nm
Shift
in µ
m
Depth in µm
E-fie
ld p
aral
lel t
o in
terfa
ce in
a.u
.
Fig. 6.5. (a) Physical layer thicknesses for a 33-layer SiO2/Ta2O5 4-cavity structure.
(b) Theoretically calculated shift as a function of wavelength at 54° incidence angle
and s-polarization. The reflectance of the structure is 100%. The circles indicate the
wavelengths and shifts corresponding to the diagrams in (c). (c) E-field parallel to
the interface of the forward propagating wave as a function of the position in the
structure for four different wavelengths – 842 nm, 846 nm, 850nm, and 854 nm.
The vertical lines indicate the position of the interfaces between layers. Light is
incident from the left, and the structure extends from 0 µm to 15.4 µm.
In conclusion, we demonstrated in this section that coupled-cavity allpass filters can be
designed to exhibit linear spatial dispersion using a digital lattice filter technique.
Furthermore, the technique can be employed to design structures with a non-monotonic
shift as a function of wavelength. The major practical limitation we found in this
approach is the necessity to design reflectors with very low reflectivity if a larger number
of cavities is desired. For example, a linear 8-cavity design calls for five reflectors with
80
less than 10% reflectivity, of which three should have less than 2% reflectivity.
Therefore, it becomes quite difficult to design appropriate thin-film reflectors for
structures with a higher number of cavities. As the only way to achieve a larger free
spectral range without decreasing the total spatial shift is to decrease the cavity length
and thus increase the number of cavities, this technique is in practice limited to the design
of narrowband structures. Again this is a limitation of this specific design technique;
using a different design technique, structures with dispersion over a broad wavelength
range can be designed as discussed in Chapter 5. Only a subset of all possible structures
is considered in this chapter – structures with large cavities separated by reflectors. In the
next chapter a third method for designing non-period multilayer stacks will be discussed
that is not limited to a particular type of structure, e.g., chirped or resonator. This more
general method based on numerical optimization permits more degrees of freedom and
therefore more flexibility in the design.
References
[1] I. Walmsley, L. Waxer, C. Dorrer, “The role of dispersion in ultrafast optics,” Rev.
of Scientific Instr., 72/1, 1-29 (2001).
[2] G. Lenz and C. K. Madsen, “General optical all-pass filter structures for dispersion
control in WDM systems,” J. Lightw. Techn., 17/7, 1248-1254 (1999).
[3] C.K. Madsen and J.H. Zhao, Optical Filter Design and Analysis - A Signal
Processing Approach, John Wiley & Sons, Inc., New York (1999).
[4] H. A. MacLeod, Thin-Film Optical Filters, Institute of Physics Publishing,
Philadelphia (2001).
[5] G. Matthaei, E. M. T. Jones, L. Young, Microwave filters, impedance-matching
networks, and coupling structures, Artech House (1980).
[6] I. Hunter, Theory and design of microwave filters, The Institution of Electrical
Engineers (2001).
[7] J.-S. Hong, M. J. Lancaster, Microstrip filters for RF/microwave applications,
Wiley-Interscience (2001).
81
[8] E. M. Dowling and D. L. MacFarlane, “Lightwave Lattice Filters for Optically
Multiplexed Communication Systems,” J. Lightwave Techn., 12/3, 471-486 (1994).
[9] M. Jablonski, Y. Takushima, K. Kikuchi, “The realization of all-pass filters for
third-order dispersion compensation in ultrafast optical fiber transmission systems,”
J. Lightwave Techn., 19/8, 1194-1205 (2001).
[10] M. Lang and T. I. Laakso, “Simple and Robust Method for the Design of Allpass
Filters Using Least-Squares Phase Error Criterion,” IEEE Trans. Circ. and Syst. II,
41/1, 40-48 (1994).
[11] K. Rajamani and Y.-S. Lai, “A novel method for designing allpass digital filters,”
IEEE Signal Proc. Lett., 6/8, 207-209 (1999).
[12] A. H. Gray, Jr. and J. D. Markel, “Digital Lattice and Ladder Filter Synthesis,”
IEEE Trans. Audio and Electroacoustics, AU-21/6, 491-500 (1973).
[13] V. Narayan, E. M. Dowling, D. L. MacFarlane, “Design of multimirror structures
for high-frquency bursts and codes of ultrashort pulses,” IEEE J. Quantum
Electron., 30/7, 1671-1680 (1994).
[14] E.K.P. Chong and S.H. Zak, An Introduction to Optimization, John Wiley & Sons,
Inc. (1996).
82
Chapter 7
Numerically Optimized Stacks
Chapter 4 demonstrates that the spatial dispersion of a multilayer stack is proportional to
the wavelength-dependent stored energy in that stack. Therefore, the task of designing the
spatial dispersion properties of a stack corresponds to designing for an appropriate
amount of stored energy at different wavelengths. Chapters 5 and 6 discussed two
methods for designing the dispersion characteristics of thin-film structures. The chirped
stacks in Chapter 5 result in a wavelength-dependent penetration depth with more energy
stored for further penetration. The resonator stacks in Chapter 6 store energy using
cavities that are separated by reflectors. Both types of stacks can be designed using a
mainly analytical design algorithm. Both algorithms allow for the design of a linear
spatial shift with wavelength in contrast to the periodic structures discussed in Chapter 3.
The algorithm for chirped stacks in Chapter 5 results in broadband designs, while the
coupled-cavity structures designed in Chapter 6 using digital signal processing techniques
are narrowband. The different bandwidths of the two structure types results from the fact
that each method considers another subset of all possible structures –chirped structures
and coupled-cavity structures respectively. This limitation to a particular type of structure
83
limits the available degrees of design freedom and thus the achievable dispersion
characteristics. In the case of the chirped stacks in Chapter 5, we limited ourselves, e.g.,
to stacks consisting of pairs of layers with a monotonically changing combined optical
thickness. It is clear that this limitation affects the dispersion characteristics that can be
achieved, limiting the algorithm for chirped stacks to broadband designs. A similar
argument holds for the resonator stacks in Chapter 6.
In this chapter we investigate how a thin-film structure can be designed without limiting
the degrees of freedom by just considering a specific type of structure. For fabrication
purposes it normally does not matter if the structure is double-chirped, coupled-cavity or
something completely different (though simple periodic structures are somewhat easier to
make because monitoring during growth is easier to interpret). More likely the total
thickness of the structure, the minimum and maximum thickness of individual layers, and
the choice of materials set the limits to what can be fabricated. Therefore, we would
ideally investigate all possible structures that can be fabricated, and find the one that most
closely matches our desired dispersion characteristics. Unfortunately, the computation
time of this approach becomes prohibitive for more than a couple of layers. This is a
common problem in the design of thin-film filters with specified reflectance and
transmittance characteristics.1,2,3,4,5,6 Here we explore how numerical techniques
developed for the design of thin-film filters can be applied to the design of thin-film
structures with spatial dispersion.
7.1 Designing Stacks using Numerical OptimizationThe design of thin-film filters is normally divided into two steps. In the first step a start
design is synthesized that approximately fulfills the required characteristics.5 In the
second step, numerical optimization procedures are used to gradually improve the
performance of the start design.6 The performance of a design is measured by a merit
function MF – a single number comparing the current design characteristics with the
desired design characteristics.7 The definition of the merit function we use is given in
(7 - 1). Qi is the current value of a quantity of interest, QiT the target value of that
84
quantity, ∆Qi the acceptable deviation, N is the number of sampling points, and p the p-
norm used.8 In the case of p = 2, the merit function is the root-mean-square difference
between the current values and the target values of the quantities of interest.
pN
i
p
i
iTi
QQQ
NMF
/1
1
1
∆−
= ∑=
(7 - 1)
The calculation of the merit function is by no means limited to reflectance or
transmittance values. In the same manner we can specify a desired spatial shift as a
function of wavelength and judge the performance of the current design by sampling the
shift at different wavelengths. Therefore, the same numerical refinement techniques used
to design thin-film filters can be applied to the design of thin-film structures with spatial
dispersion. Any one of the analytical methods discussed in Chapters 3, 5, and 6 can be
used to generate a starting design. In our experience the most important property of the
starting design is to eliminate strong ripple, i.e., to provide impedance matching. Rapid
oscillations in the shift with wavelength are difficult to remove by numerical
optimization. We had more success in starting with a low but constant dispersion and
increasing it during optimization. To ensure a constant dispersion, we normally calculate
the merit function MF employing a p-norm with p between 6 and 10. This large p-norm
ensures an approximately identical error in the shift for all wavelengths.8 Thus, the
resulting shift is, for example, linear but offset from the originally specified shift. The
sampling points have to be chosen close enough in wavelength to prevent oscillations
between the points. A spacing of around 2 nm at 850 nm appears to work well.
We implemented six different numerical optimization techniques – golden section
search,8 secant method,8 conjugate gradient algorithm,8 Broyden-Fletcher-Goldfarb-
Shanno (BFGS),8 damped least squares method,6 and Hooke&Jeeves pattern search.9 A
brief discussion of these techniques and a comparison of the results obtained is given in
Appendix E. The first two methods are one-dimensional search methods, i.e., the
different parameters are optimized sequentially. The other methods vary all parameters
simultaneously. All methods search for the local minimum of the merit function, where a
merit function with p=2 is used for the damped least squares method. A lower non-local
85
optimum may be found though by, e.g., choosing the interval size of the golden section
search large, or by taking large steps in the Hooke&Jeeves pattern search technique. See
Appendix E for more details on this topic. Our design algorithm uses the different
numerical optimization techniques sequentially. This is successful as a different design
method may find a lower minimum if one method is “stuck” in a shallow local optimum.
Numerical optimization allows the design of structures with dispersion characteristics
that cannot be achieved using the analytical techniques discussed in the previous sections.
Layer Number
Phy
sica
l Thi
ckne
ss in
nm 10 20 30 40 50 600
150
300
450
600
10 20 30 40 50 600
150
300
450
600
10 20 30 40 50 600
150
300
450
600
SiO2Ta2O5
(a)
(b)
(c)
Fig. 7.1. Generation of the starting design. (a) Quarter wave Bragg stack. (b)
Impedance matched stack. (c) Half-wave layers added to thinnest layers.
As an example we designed a 66-layer, 23.7-µm thick, thin-film structure with a linear
shift over a 40-nm wavelength range around 1550 nm (corresponding approximately to a
typical Er-fiber amplifier bandwidth for the C-band). The composition of this designed
stack is given in Appendix G. This type of medium-wide wavelength range is difficult to
86
achieve with the double-chirped structures in Chapter 5 or the coupled-cavity structures
in Chapter 6. We used an impedance matched Bragg stack as the start design. Fig. 7.1(a)
shows the initial quarter-wave Bragg stack, while the impedance matching can be seen in
Fig. 7.1(b). Light is incident from the left side onto the stack. Half-wave layers are added
to the thinnest layers of the structure to facilitate fabrication as seen in Fig. 7.1(c).
Fig. 7.2 plots the shift and reflectance as a function of wavelength for the start design
given in Fig. 7.1(c). The start design has unity reflectance and nearly no spatial
dispersion.
1500 1520 1540 1560 1580 16000
10
20
30
40
50
Wavelength in nm
Shift
in R
efle
ctio
n in
um
1500 1520 1540 1560 1580 16000
0.2
0.4
0.6
0.8
1
Wavelength in nm
Wavelength in nm
Shift
in µ
m
Wavelength in nm
Ref
lect
ance
Start designRefined designRefined design with gold mirrorTargeted linear shiftTargeted linear shift (shifted)
(a) (b)
Fig. 7.2. (a) Shift as a function of wavelength and (b) reflectance as a function of
wavelength for the start design, the refined design, and the refined design with
backside gold coating at 45° incidence angle and p-polarization.
We refined the design specifying a linearly increasing spatial shift for 21 different
wavelengths as seen in Fig. 7.2(a). Using a p-norm of 8 the linearly increasing shift is
obtained with only a small amount of ripple. Due to the high p-norm the shift of the
refined design is offset from the initially specified shift. This is not important, as only the
difference in the shift between wavelengths, i.e., the constant dispersion, matters. It can
be seen from the shifted target line that the desired dispersion is obtained. Fig. 7.2(b)
shows that the reflectance of the refined design drops around 1580 nm. In order to
achieve a high reflectance for all wavelengths, a gold coating can be applied to the last
87
layer. To prevent loss, the last layers of the structure could also be specified as a Bragg
stack. In our design algorithm we can set which layers are to be changed and what the
minimum and maximum layer thicknesses are. This guarantees that the design can be
fabricated. The performance of the stack with backside gold coating is also plotted in
Fig. 7.2. A linear shift with wavelength over the entire EDFA C-band is obtained with
high reflectance for all wavelengths.
810 815 820 825 830 835 840 8450
5
10
15
20
25
Wavelength in nm
Shift
in u
m
20 40 600
100
200
300
400
500
SiO2 layersTa2O5 layers
Layer Number
Phys
ical
Thi
ckne
ss in
nm
0
2
4
6
0
2
4
6
0
2
4
6
2 0 2 4 6 8 10 12 140
2
4
6
821 nm
828 nm
835 nm
842 nm
(a)
(b)
(c)
Layer number
Phys
ical
thic
knes
s in
nm
Wavelength in nm
Shift
in µ
m
Depth in µm
E-fie
ld p
aral
lel t
o in
terfa
ce in
a.u
.
Fig. 7.3. (a) Physical layer thicknesses for a 66-layer, numerically optimized
SiO2/Ta2O5 structure. (b) Theoretically calculated shift as a function of wavelength
at 54° incidence angle and p-polarization. The reflectance of the structure is
improved to nearly 100% by a gold layer on the very right. The circles correspond
to the diagrams in (c). (c) E-field parallel to the interface of the forward
propagating wave as a function of the position in the structure for four different
wavelengths – 821 nm, 828 nm, 835nm, and 842 nm. The vertical lines indicate the
88
position of the interfaces between layers. Light is incident from the left, and the
structure extends from 0 µm to 13.4 µm.
For experimental testing, this design was scaled from 1550 nm to 830 nm as we have a
tunable laser available around 830 nm. In the fabricated design the minimum layer
thickness is 57 nm, and the maximum thickness is 453 nm. Six calibration layers were
added in the front, which is the substrate side. The composition of stack is given in
Appendix G. In Fig. 7.3(a) the physical layer thicknesses are plotted. Fig. 7.3(b) shows
the shift as a function of wavelength for the 66-layer structure with a backside gold
coating. A highly linear shift is obtained. To investigate the origins of the observed
dispersion, in Fig. 7.3(c) the E-field parallel to the interfaces is plotted as a function of
the position in the stack for four different wavelengths. We see that the penetration depth
increases for longer wavelengths and more energy is stored at the same time. Thus, the
dispersion of this general structure is based on both phenomena – a wavelength-
dependent penetration depth and resonant energy storage.
7.2 Experimental Results for an Optimized StackFor the fabrication of the numerically optimized stack the deposition rate was determined
using the first six quarter-wave calibration layers and the remaining design was fabricated
by timed deposition. No active monitoring was used. Systematic deposition errors, i.e.
systematic relative errors in the layer thicknesses, occur if the calibration predicts the
wrong deposition rate. Such systematic errors in the layer thickness lead to a change in
the wavelength as well as the dispersion profile. First experiments with the fabricated
stack at 45° incidence angle did not show a linear shift as a function of wavelength. By
varying the layer thicknesses systematically and comparing the simulation results to the
experimental results, we found that the SiO2-layers of the fabricated stack were 4.3% too
thick, while the Ta2O5-layers were 0.3% too thick. Changing all the layer thickness of the
design by these factors, we obtain good agreement between experiment and simulation.
Appendix G gives the layer thicknesses for the scaled, fabricated stack. The fabricated
stack has a linear shift for an incidence angle of 54°. This is why all experimental results
are for 54° instead of the designed 45°. Random absolute errors, which are for example
89
caused by the shutter closing time, should not be larger than approximately ±1 nm in
order to obtain a working design. Random relative errors need to be less than ±0.3 % of
the layer thickness. Because we see good agreement between experiment and theory once
the systematic errors are taken into account, the random errors seem to be sufficiently
small.
The experiment is again conducted using the setup shown in Fig. 3.3. Fig. 7.4 shows the
intensity observed along a CCD trace as a function of wavelength. Contrary to the
experiment discussed in Chapter 3, only one beam exits the structure, which demonstrates
the successful suppression of all other reflections by impedance matching. The linear
beam shifting with wavelength is clearly visible.
Wavelength in nm
Posi
tion
in µ
m
Quartz substrate
66-layer SiO2-Ta2O5thin-film stack
Fig. 7.4. Experimentally observed intensity on a CCD trace as a function of
position and wavelength for a 66-layer numerically optimized stack for an
incidence angle of 54° and p-polarization.
To analyze the spatial shift more closely, a Gaussian beam profile is fitted to the
experimental data at each wavelength adjusting the beam amplitude, beam center
position, and beam width. The experimentally observed shift of the peak as a function of
wavelength is plotted in Fig. 7.5. The linear shift between 820 nm and 840 nm is clearly
visible. As the shift does not exhibit any rapid changes in this wavelength range, the
plane-wave approximation predicts the experimentally observed shift accurately. For
90
wavelengths larger than 840 nm, only the beam simulation correctly predicts the
observed shift. The plane wave calculation is not accurate due to the rise and fall of the
shift at 843 nm. Different components observe a different shift as discussed in Chapter 3.
The slight difference between the shift in Fig. 7.3(b) and Fig. 7.5 is due to the fact that
the first calculation includes the backside gold coating and the second does not. The
overall excellent agreement between the experimentally observed shift and the
theoretically expected shift confirms the concepts discussed above for obtaining spatial
dispersion.
820 825 830 835 840 8450
5
10
15
20
25
Plane-Wave TheoryExperimentBeam Simulation
Wavelength in nm
Shift
in u
m
Fig. 7.5. Experimentally observed and theoretically calculated spatial dispersion of
a 66-layer SiO2/Ta2O5 dielectric stack with a total thickness of 13.4 µm on a quartz
substrate for an incidence angle of 54° and p-polarization.
7.3 Comparison of Periodic and Non-Periodic DesignsIn Chapters 3, 5, 6 and 7 we discuss four different ways of designing structures with high
spatial dispersion. In Chapter 3 we demonstrated the strong spatial dispersion of periodic
thin-film structures close to the stop-band edge both in theory and experiment.
Unfortunately, periodic structures with two layers per period do not offer enough degrees
of freedom to design structures with desired dispersion characteristics, e.g., a linear shift
91
with wavelength. In Chapter 5 we showed that the wavelength-dependent penetration
depth of double-chirped structures can be used to obtain spatial dispersion, and that these
structures are particularly useful for broadband designs. In this section we also found that
non-resonant energy storage has an important contribution to the shift, and that the
obtained shift cannot be explained from pure geometrical reasoning. The coupled-cavity
structures discussed in Chapter 6 employ a wavelength-dependent amount of stored
energy to obtain dispersion, and are very effective for narrowband designs. Finally, in
this chapter we demonstrated in theory and experiment that structures using a
combination of a wavelength-dependent turning point and stored energy can be designed
using numerical optimization methods.
1520 1525 1530 1535 1540 1545 1550 1555 1560 1565
0
10
20
30
40
Shift
in u
m
200-layer periodic (experiment)200-layer periodic (theory)66-layer non-periodic (experiment)66-layer non-periodic (theory)
Wavelength in nm
Shift
in µ
m
Fig. 7.6. Comparing the performance of a periodic and a non-periodic structure.
Fig. 7.6 compares the results obtained for the periodic structure from Chapter 3 and the
non-periodic structure from this chapter. The results are scaled to the 1550-nm
wavelength range for better comparison with current commercial interests for
telecommunication. The non-periodic structure exhibits a total shift comparable to that of
the periodic structure. But the same shift is obtained with just a third of layers, and the
observed shift is linear with wavelength, which is much more desirable for practical
applications. Furthermore, the non-periodic structure is essentially loss-less as all loss
92
reflections (e.g., front surface reflection) are suppressed. Fig. 7.6 clearly demonstrates the
superior dispersion characteristics of the non-periodic stack and shows why it is worth
the effort to explore new types of structures besides the photonic crystal superprism
effect.
7.4 Experiment: 4-Channel Wavelength DemultiplexerSo far we have discussed the spatial shift obtained after a single bounce off the stack. The
shift of approximately 20 µm in Fig. 7.5 is not sufficient though to separate multiple
channels of different wavelengths except if these are focussed very tightly. As discussed
in Chapter 1 several bounces can be performed off the stack in order to increase the
spatial separation between beams of different wavelengths. As we are operating the stack
through the substrate, a gold coating can easily be deposited on the opposite substrate
side allowing for multiple bounces as seen in Fig. 7.8.
Wavelength in nm
Posi
tion
in µ
m
828 830 832 834 836 838 840 8420
50
100
150
TheoryExperiment
Wavelength in nm
Shift
in u
m
Wavelength in nm
Shift
in µ
m
(a) (b)
Fig. 7.7. (a) Experimentally observed intensity on a CCD trace as a function of
position and wavelength for 8 bounces off a 66-layer numerically optimized stack.
(b) Experimentally observed and theoretically calculated shift as a function of
wavelength.
In Fig. 7.7(a) the intensity along one CCD trace is plotted as a function of wavelength for
eight bounces off the numerically optimized stack introduced in section 7.2 without the
93
backside gold coating applied. Fig. 7.7(b) compares the experimentally observed shift to
the shift calculated using the plane-wave calculation and multiplying by eight. We see
that the shift is indeed eight times as large for eight bounces off the stack as expected.
0 20 40 60 80 100 120 140 160
Position in µm
Am
plitu
de in
a.u
.
827 .3 nm
831.0 nm
834.7 nm
840.6 nm840.6 nm834.7 nm831.0 nm827.3 nm
zy
x
827.3 nm831.0 nm834.7 nm840.6 nm
Multiplexedlight
1 mm 13.4 µm
≈10
mm
…
Fig. 7.8. 4-channel wavelength demultiplexer.
Fig. 7.8 demonstrates that this shift can be used to separate four beams by their Gaussian
beam width.10 The first three beams have a channel spacing of 3.7 nm. The spacing
between the third and fourth channels, 5.9 nm, is chosen to be larger because of the
reduced dispersion. The amplitudes of the beams are decreasing with wavelength because
we have not yet applied the gold coating to the stack side. The dielectric stack itself has a
lower reflectance for larger wavelengths. The slight beam distortions visible in Fig. 7.8
are due to beam clipping in the experiment and would not appear in a properly designed
module. In Chapter 8 the question of how many channels can be separated using a given
stack with spatial dispersion is investigated, i.e., could the stack designed here separate
more than four channels and is there an ultimate limit to the number of channels that can
be separated.
94
References
[1] P. Baumeister, “Design of multilayer filters by successive approximations,” J. Opt.
Soc. Am., 48/12, 955-958 (1958).
[2] J. A. Dobrowolski, “Completely automatic synthesis of optical thin film systems,”
Appl. Opt., 4/8, 937-946 (1965).
[3] H. A. MacLeod, Thin-Film Optical Filters, Institute of Physics Publishing,
Philadelphia (2001).
[4] A. Thelen, Design of Optical Interference Coatings, McGraw-Hill, Inc., New York
(1989).
[5] Li Li and J. A. Dobrowolski, “Computation speeds of different optical thin-film
synthesis methods,” Appl. Opt., 31/19, 3790-3799 (1992) and references herein.
[6] J. A. Dobrowolski and R.A. Kemp, “Refinement of optical multilayer systems with
different optimization procedures,” Appl. Opt., 29/19, 2876-2893 (1990) and
references herein.
[7] J. A. Dobrowolski, F.C. Ho, A. Belkind, V.A. Koss, “Merit functions for more
effective thin film calculations,” Appl. Opt., 28/14, 2824-2831 (1989).
[8] E.K.P. Chong and S.H. Zak, An Introduction to Optimization, John Wiley & Sons,
Inc. (1996).
[9] T.E. Shoup and F. Mistree, Optimization Methods with Applications for Personal
Computers, Prentice-Hall, Inc. (1987).
[10] M. Gerken and D. A. B. Miller, “Thin-Film (DE)MUX based on group-velocity
effects,” ECOC 2002, Paper 11.3.3, Copenhagen, Denmark (September 8-12,
2002).
95
Chapter 8
Maximum Number of Channels
Chapters 3 to 7 discussed how to design multilayer stacks with high spatial dispersion.
This chapter now investigates how many channels can be multiplexed or demultiplexed
using a given multilayer stack. In this chapter, spatially separable channels are referred to
as “modes” of the structure. First we need to define what is meant by a mode of the
structure. As we are investigating structures with a continuous shift as a function of
wavelength, all wavelengths can propagate through the structure. A mode is therefore not
defined by the fact that it propagates through the structures whereas other modes do not.
Hence this definition is unlike that of modes in a waveguide. A mode is also not defined
by the fact that it can be physically distinguished from other modes. Since different
wavelengths are associated with the different modes, they can always be distinguished.
Our definition of a mode is related to the spatial extent of a channel without considering
its wavelength. Rigorously, perhaps, our modes are beam forms that are spatially
orthogonal. Beams that do not overlap at all are certainly spatially orthogonal and other
orthogonal patterns might also be possible, e.g., as in the communication modes of [1].
Beams that mostly do not overlap can be considered approximately orthogonal, and do
96
lead to the right overall counting of such modes as discussed in [1]. This definition is
based on the idea that we want to use the dispersion of multilayer structures to spatially
separate channels of different wavelengths, i.e. we want to measure the power of a
demultiplexed channel without considering its wavelength. The question of the number of
modes of a given stack is therefore equivalent to the question of how many separate
channels can we demultiplex assuming wavelength-insensitive detectors.
Two possible types of spatial modes are considered here as depicted in Fig. 8.1. “Volume
modes” are defined as being modes that have mutually exclusive propagation cones
within the volume of the stack as shown in Fig. 8.1(a). “Surface modes” may overlap
within the volume of the stack, but are separated on the exit surface as seen in Fig. 8.1(b).
(a) (b)
Fig. 8.1. (a) Volume modes. (b) Surface modes.
We chose these two mode definitions for their practical interest – the volume mode
picture corresponds to focusing the beam on the input surface, while the surface mode
picture corresponds to focusing on the exit surface. Other mode definitions could be
chosen equally well. Note that ultimately the number of resolvable modes does not
depend on the position of the focal point, but only on the structure.
The number of volume modes and the number of surface modes are calculated and
compared here for two different example designs – a periodic stack and a numerically
optimized stack that has a group propagation angle changing linearly with wavelength.
Both stacks are composed of 200 alternating layers of SiO2 (n=1.45) and Ta2O5 (n=2.09)
97
on a quartz substrate (n=1.52). The layer sequences are given in Appendix G. In the first
design all layers are 261 nm thick. The backside of the stack is gold-coated to increase
the reflectivity. This design uses the superprism effect discussed in Chapter 3 to obtain
spatial dispersion. This design will be called the “Periodic Design.” The second design is
a numerically optimized design. The method for designing double-chirped mirror
structures discussed in Chapter 5 was used to generate a start design. It was optimized
using the techniques described in Chapter 7 and Appendix E. This design will be called
the “Non-periodic Design.”
Fig. 8.2 plots the physical thickness of the layers for the two different designs. The blue
squares show the identical thickness of all layers for the periodic design. The cyan circles
are the layer thicknesses of the synthesized second design before refinement. It can be
seen how the thicknesses of the high refractive index layers increase gradually, while the
thicknesses of the low refractive index layers decrease. Furthermore, the thickness of a
period also increases with layer number. The magenta crosses represent the layer
thicknesses of the refined non-periodic design.
0 20 40 60 80 100 120 140 160 180 2000
200
400
600
Periodic DesignDouble-Chirped Mirror DesignRefined Non-periodic Design
Layer #
Phys
ical
Thi
ckne
ss in
nm
Fig. 8.2. Physical layer thicknesses of the periodic and non-periodic design and the
double-chirped mirror starting design.
98
Fig. 8.3 and Fig. 8.4 show the reflectance of the periodic and the non-periodic design at
40° incidence angle. It can be seen that the non-periodic structure is designed to have
unity reflectance without the need of a backside gold coating. For the periodic design
Fabry-Perot like oscillations are visible in the reflectance. They are due to the larger
absorption by the gold coating for wavelengths that are on resonance because theses have
a larger stored energy in the structure corresponding to multiple effective bounces off the
gold coating as discussed in Chapter 2.
Fig. 8.5 graphs the group propagation angle for the two different designs. Since the
periodic design has many periods it can be calculated using either the Bloch dispersion
relation (Appendix A) or the transfer matrix dispersion relation (Appendix C). It is quite
obvious that the results obtained from these calculations differ. The transfer matrix
calculation shows strong oscillations, which are caused by the interference of the shifting
beam with a front reflection as discussed in Chapter 3.
The Bloch calculation on the other hand results in a smooth curve, since an infinite
medium is assumed such that interface reflections are not included. Assuming that the
beam is focused sufficiently to prevent the interference effects, we can use the Bloch
calculation values for the group propagation angle. The non-periodic stack is designed to
suppress the front reflection and has a smooth change in the group propagation angle with
wavelength. Furthermore, this device structure was designed to have a linearly changing
group propagation angle with wavelength compared to the non-linear characteristics of
the periodic design. Fig. 8.6 graphs the calculated shift along the front interface assuming
a single bounce for both structures.
99
1520 1530 1540 1550 1560 15700
0.2
0.4
0.6
0.8
1
Transfer Matrix Calculation
Periodic Design
Wavelength in nm
Refle
ctan
ce
1520 1530 1540 1550 1560 15700
0.2
0.4
0.6
0.8
1
Transfer Matrix Calculation
Non-periodic Design
Wavelength in nm
Refle
ctan
ce
Fig. 8.3. Reflectance of the two designs at 40° incidence angle.
1520 1530 1540 1550 1560 15700.95
0.96
0.97
0.98
0.99
1
Transfer Matrix Calculation
Periodic Design
Wavelength in nm
Refle
ctan
ce
1520 1530 1540 1550 1560 15700.97
0.98
0.99
1
Transfer Matrix Calculation
Non-periodic Design
Wavelength in nm
Refle
ctan
ce
Fig. 8.4. Zoomed-in reflectance of the two designs at 40° incidence angle.
100
1520 1530 1540 1550 1560 15700
20
40
60
80
Transfer Matrix CalculationBloch Calculation
Periodic Design
Wavelength in nm
Gro
up p
ropa
gatio
n an
gle
in d
eg
1520 1530 1540 1550 1560 15700
20
40
60
80
Transfer Matrix Calculation
Non-periodic Design
Wavelength in nm
Gro
up p
ropa
gatio
n an
gle
in d
egFig. 8.5. Calculated group propagation angle.
1520 1530 1540 1550 1560 15700
200
400
600
800
Transfer Matrix CalculationBloch Calculation
Periodic Design
Wavelength in nm
Shift
in u
m
1520 1530 1540 1550 1560 15700
20
40
60
80
100
Transfer Matrix Calculation
Non-periodic Design
Wavelength in nm
Shift
in u
m
Fig. 8.6. Calculated Shift along the interface for a single bounce.
8.1 Number of Volume ModesIn this section we are going to consider the number of volume modes that can be
separated by a structure. Here we do not consider what happens for multiple bounces and
101
intermediate propagation through the substrate. We define volume modes as modes that
propagate in mutually exclusive propagation cones. For simplicity we will assume a
device as shown in Fig. 8.7.
L
d
Fig. 8.7. Schematic of three volume modes for focussing on the front surface.
Plane waves of different wavelength propagate at different angles within the device. As
shown in Appendix D, a beam of light can be decomposed into plane-wave components
with different propagation directions. Thus, a beam of light consists of a range of
propagation angles. In principle a beam has components at all angles, but many of these
might have very small amplitudes and can thus be neglected. Here we consider the
angular range to be delimited by those components whose intensity has dropped to 1/e2
from the center component. For less crosstalk between channels a smaller value may be
chosen, which only results in a scaling of the results given here.
Assuming the input beam of multiplexed wavelengths is focused on the front of the
device and that the input diameter of the beam d is small compared to the width of the
device L, modes are non-overlapping in the volume and at the output if their angular
ranges are non-overlapping within the structure. If, for example, the first mode has an
angular content within the device of 10º to 20º and the next mode occupies the angular
range from 20º to 30º, these modes will not overlap at the output and are thus separate
modes.
Following this definition of separable modes, the number of modes within a small
frequency range ∆ω is given by (8 – 1).
102
( ) ( ) ( )( )instruc
groupgroupinesN
θωθθ
ωωθθωωθθθωωθ
∆∆
∆−−∆+=∆∆∆
,~,~2~,~2~,~
,,~,~mod (8 - 1)
In this equation ∆θstruc is the angular range of a mode within the structure for a given
angular range ∆θin of the input beam in vacuum. θ is the incidence angle of the beam,
θgroup is the group propagation angle in the stack, and ω is the frequency. As seen in
Appendix B, the dispersion relation can be expressed either in terms of the incidence
angle and the frequency ( )ωθ ~,~ or in terms of the wavevector along the layers β and the
frequency ( )ωβ , . In order to distinguish these two variable sets we use the tilde for the
first case. The variable transformation between ( )ωθ ~,~ and ( )ωβ , is discussed in more
detail in Appendix B. Defining the dispersion Dispω as shown in (8 – 2).
( ) ( )ω
ωθθωθω ~
~,~~,~
∂
∂= groupDisp (8 - 2)
The total number of modes between ω1 and ω2 is given by (8 – 3). Note there has to be
one mode even without any dispersion, since one beam can always simply propagate
through the structure.
( ) ( ) ( )( )∫∫ ∆∆
+=∆∆+=∆2
1
2
1
~,~,~~,~
1,~,~,~1,,,~mod21mod
ω
ω
ωω
ω
ωθωθθ
ωθθωωθθωωθ dDispdNN
instrucinesines (8 - 3)
To evaluate this number of modes, we need to find an expression for ∆θstruc. ∆θstruc can be
estimated from ∆θin using (8 – 4), where the difference is approximated by a differential.
( ) ( )in
groupinstruc θ
θ
ωθθθωθθ ∆
∂
∂=∆∆ ~
~,~,~,~ (8 - 4)
In order to relate ∆θstruc to the dispersion Dispω, the group velocity angle θgroup is
developed into a Taylor series as shown in Appendix F. This results in relationship (8 - 5)
between the angular range ∆θstruc of a beam within a dispersive stack, the input angular
range ∆θin, and the dispersion of the stack Dispω.
103
( ) ( ) in
avg
instruc
cDisp
nθ
ωβ
θ
θωωθ
θ
θθωθθ ω ∆
∂∂
−+
−=∆∆
~sin
~cos~~,~~sin
~cos,~,~22
(8 - 5)
As seen in Chapter 4 and Appendix F, ∂β/∂ω is approximately constant with wavelength.
The only rapidly varying term with wavelength in (8 - 5) is the dispersion Dispω. Thus,
the angular range of a mode ∆θstruc can be estimated as the input angular range ∆θin
multiplied by the sum of a constant term added to a term that is proportional to the
dispersion. The validity of this approximation is limited to stacks that have a constant
angular dispersion or to small input angular ranges as differences are replaced by
differentials. Substituting expression (8 - 5) for ∆θstruc into (8 - 3), we obtain the number
of volume modes within a given wavelength range as given in (8 - 6).
( )
( )
∫ ∆
∂∂
−+
−
+=∆2
1
~1
~sin
~cos~~sin~,~
~cos
11,,,~
22
21mod
ω
ω
ω
ωθ
ωβ
θ
θω
θωθ
θ
θωωθ d
cnDisp
Nin
avg
ines
(8 - 6)
(8 - 6) reveals that the number of modes is inversely proportional to the input angular
range. Thus, half the input angular range means double the number of modes, and double
the input range half the number of modes. This result was verified in simulations
presented in a section 8.2.
The interesting part of (8 - 6) is the denominator of the first fraction. It can be seen that
for small dispersions, ( )
− θωθθ ω
~sin~,~~cos 22avgnDisp dominates and the number of
modes increases linearly with increasing dispersion. For large dispersion on the other
hand,
∂∂
−ωβ
θθω c~sin~cos~ dominates and the number of modes is independent of the
dispersion. This result might be expected, since a larger dispersion not only leads to a
104
larger beam shift, but also to beam broadening. Thus within this model of volume modes,
above a certain dispersion nothing can be gained by increasing the dispersion further.
In Fig. 8.8 the number of volume modes per wavelength interval of 1 nm is graphed as a
function of the dispersion calculated using (8 - 6). An input half cone of ∆θin =1° is used.
In the left graph the group velocity vgx=∂ω/∂β of the periodic structure is used and in the
right graph vgx of the non-periodic design is used. The total number of volume modes is
obtained by multiplying the modes per nm by the wavelength range over which the
dispersion is achieved. As seen from (8 - 6) one more mode needs to be added. The non-
periodic design has, e.g., a dispersion of 0.7 deg/nm over a 40 nm wavelength range.
Thus, we expect from Fig. 8.8 a total of (0.093modes/nm)·40nm + 1 = 3.7 modes for
∆θin=1°. Section 8.2 verifies this estimate. It is clearly visible from Fig. 8.8 that the
number of volume modes does not increase significantly above a certain dispersion.
0 0.5 1 1.5 20
0.05
0.1
0.15
0.2vgx of Periodic Design
Dispersion in deg/nm
Mod
es p
er n
m
0 0.5 1 1.5 20
0.02
0.04
0.06
0.08
0.1vgx of Non-periodic Design
Dispersion in deg/nm
Mod
es p
er n
m
Fig. 8.8. Increase in the number of volume modes per 1 nm wavelength interval as
a function of the stack dispersion for an input half cone ∆θin =1°.
∂β/∂ω can be estimated very roughly by the value obtained for a homogeneous layer of
refractive index navg. In this case βhom is given by (8 - 7).
22
hom Kcnavg −
=
ωβ (8 - 7)
105
The derivative with respect to ω for a constant K is then calculated as shown in (8 - 8).
θβ
ω
δωδβ
sin
2
hom2
2hom
cn
cn avgavg
constK
===
(8 - 8)
Thus, the higher average refractive index of the periodic design leads to a higher ratio
∂β/∂ω and to a higher number of modes per wavelength interval of 1 nm.
Knowing how to estimate the number of modes using (8 - 6), the question is how to
maximize the number of modes for a given stack. Here we will consider the case that we
have designed a structure that has a certain dispersion behavior and is operated at a given
incidence angle. Even though the incidence angle appears as a variable in (8 - 6), it
cannot really be changed easily as the dispersion curve is only designed and valid for a
certain angle. Thus, to change the input angle, we would have to design a new structure.
The same argument is true for the operating range of the structure, which is also
determined by the design. The only parameter we can change is the angular range of the
input beam (as long as we only probe a part of the dispersion curve that is well behaved).
In (8 - 6) the thickness of the structure does not appear except that the calculation is only
correct for d<<L as shown in Fig. 8.7. (8 - 6) predicts more modes for a structure with a
smaller input angular range. Thus, we would want to choose the input angular range as
small as possible to obtain the maximum number of volume modes. But a smaller input
angular range results in a larger beam waist, and the device has to have a larger thickness
for the condition d<<L to be true. Therefore, the only conclusion we can draw from
(8 - 6) is that the angular range should be as small as possible and the thickness has to
grow correspondingly. For a given physical thickness the optimal number of modes
cannot be determined using the volume-mode calculation as we leave the range of
validity d<<L of this calculation, when decreasing the angular range and thus increasing
the spot size d.
106
8.2 Verification of Volume Mode ModelIn section 8.1 we have derived (8 - 6) for calculating the number of volume modes. In
this section we will test this model by calculating the distinguishable volume modes for
the non-periodic stack in a numerical simulation and comparing the resulting number of
modes to the theoretical model. Assuming that the width d of the modes at the input
surface is negligible compared to the length of the structure L (d<<L), the number of
modes supported by the structure is numerically obtained as follows. First the angular
range of the mode with the lowest wavelength is calculated from the input angular range.
Fig. 8.9 gives an example of such a calculation.
The C-band of an Erbium-doped fiber amplifier (EDFA) from 1525 nm to 1565 nm is
chosen as wavelength interval. The lowest mode has a wavelength of 1525 nm. From
Table 8.1 we see that this mode occupies an angular range from 8.24° to 12.26°. Now we
know that the next higher mode has to have at least a low angle of 12.26°. Using a
numerical search algorithm, we find the next possible mode to have a wavelength of
1535 nm and an angular range in the structure of 12.26° to 19.74°. We continue to find
modes in this manner until the high wavelength of 1565 nm in this example is reached.
Table 8.1 gives the numerical data for the five modes found. In Fig. 8.9(b) a cartoon of
the modes is depicted.
Using the numerical simulation technique described above, we can now test the
theoretical number of modes expected from (8 - 6) against the number of modes actually
obtained. Using (8 - 6) the number of modes for the non-periodic design is calculated as a
function of the input angular range ∆θin. The integral is performed by calculating ten
intermediate points. Only the dispersion Dispω is assumed to be a function of frequency.
The resulting number of modes is plotted as the crosses in Fig. 8.10. The number of
modes is rounded off to the previous integer, since fractional modes are physically not
possible. The squares plotted in Fig. 8.10 represent the number of modes obtained from a
numerical simulation as described above.
107
1520 1530 1540 1550 1560 15705
10
15
20
25
30
35
Wavelength in nm
Prop
agat
ion
angl
e in
cry
stal
in d
eg
(a) (b)
Fig. 8.9. Volume mode calculation for the non-periodic design for an incidence
angle of 40°, an input half cone of 0.6°, and a wavelength interval from 1525 nm to
1565 nm. (a) shows the position of the obtained modes in terms of wavelength and
propagation angle within the crystal. (b) depicts a cartoon of the obtained modes.
Wavelength in nm Low angle in ° High angle in °15251535154415541565
8.2412.2619.7526.0034.50
12.2619.7425.9934.5043.79
Table 8.1. Wavelength, low angle, and high angle of the different modes for the
example in Fig. 8.9.
Fig. 8.10 nicely shows the agreement between simulation and theoretical number of
modes obtained for the volume mode model (8 - 6). Also we see again that the number of
modes is inversely proportional to the input angular range as predicted before. Thus, we
would want to choose a small angular range, e.g. 0.2°, which corresponds to a spot size
(Gaussian beam radius) of 140 µm. But if d is about twice the spot size, the thickness L
of the stack would have to be several times this value for the condition d<<L to be
fulfilled.
108
0 1 2 30123456789
101112131415
Simulated number of modesExpected number of modes
Input half cone in deg
Num
ber o
f mod
es
0 50 100 1500123456789
101112131415
Input spot size in um
Num
ber o
f mod
es
(a) (b)
Fig. 8.10. Number of volume modes as obtained from simulation and the expected
number of modes calculated using (8 - 6) as a function of the input half cone in (a)
and the input spot size in (b).
8.3 Number of Surface ModesIn the last section we have shown that the volume-mode calculation can be used to
estimate the number of volume modes for a given structure, if the condition d<<L is
fulfilled. The volume-mode calculation cannot, however, be used to estimate the
maximum number of modes, since we leave the range of validity of this calculation. In
this section, a different approach for obtaining the number of modes supported by a
structure is introduced. For wavelength demultiplexing, separable modes have to be
spatially non-overlapping at the output interface, but not necessarily within the structure.
This results in a mode picture as given in Fig. 8.11.
109
L
∆x
Fig. 8.11. Schematic of three non-overlapping surface modes on the back surface.
Here we assume that the backside is the output surface of interest, i.e. the position where
we place the detectors or output waveguides. Any other plane could be chosen equally
well, as this is a linear system. A separate lens may be necessary to focus the light onto
the output devices in that case. Assuming a monotonically increasing group propagation
angle θgroup with frequency, the number of modes within a frequency range ω1 to ω2 is
given by the total shift in this range divided by the surface ∆x occupied by one mode.
Furthermore, we have already one mode without any dispersion, which is added in
(8 - 9).
( ) ( )( ) ( )( )( )1
~,~tan~,~tan,~,~,~ 12
21mod +∆
−=∆
xL
xN groupgroupes
ωθθωθθωωθ (8 - 9)
In order to calculate the number of modes within a frequency range ∆ω, we start from
expression (8 - 10) for the number of modes within an angular range ∆θgroup.
( )( ) ( )
x
LxN
groupgroup
groupgroup
groupes ∆
∆−−
∆+
=∆∆∆2
~,~tan2
~,~tan,,~,~
mod
θωθθ
θωθθ
θωθ
(8 - 10)
Using identity (8 - 11), (8 – 10) can be approximated by (8 - 12).
( )βα
βαβα
coscossintantan ±
=± (8 - 11)
110
( ) ( )( )ωθθ
θθωθ ~,~cos
,,~,~2mod
group
groupgroupes x
LxN
∆
∆=∆∆∆ (8 - 12)
Finally, plugging in the definition of the dispersion Dispω as defined in (8 - 2), we obtain
the number of modes within a frequency range ∆ω as given in (8 - 13).
( ) ( )( )( )ωθθ
ωωθωωθ ω
~,~cos
~,~,,~,~
2modgroup
es xDispLxN
∆∆
=∆∆∆ (8 - 13)
If ∆x or Dispω are not the same for all the modes, but a function of frequency, (8 - 13) can
be integrated to obtain the number of modes within a frequency range given in (8 - 14).
( ) ( )( )( )∫ ∆
+=∆2
1
~~,~cos
~,~1,,,~
221mod
ω
ω
ω ωωθθ
ωθωωθ d
xDispLxN
groupes (8 - 14)
Now we consider again the question, how to maximize the number of modes obtained for
a given dispersion curve and given incidence angle. We see that we get more modes the
larger the thickness L of the stack is and the smaller the surface ∆x occupied by one mode
is. In the volume-mode calculation, we assumed that the input beam is focused on the
front surface of the structure. But this is not necessary. From (8 - 14) we see that we
obtain a larger number of modes if the beam is focused on the output surface.
For a general structure, only one mode will be focused exactly at the backside of the
structure occupying a surface depending on the spot size w0. Other modes will be out of
focus and require a larger area. Given a certain structure, the focus has to be chosen in
such a way that the largest number of modes can be fitted on the back surface. In
principle, a structure could be designed in such a way that all modes focus at the same
distance L from the front surface. This will allow for the largest number of modes
possible, since moving any mode out of focus will increase its size ∆x and therefore
lower the total number of modes.
Concluding this argument, we see from (8 - 14) that a larger thickness L of the stack and
a smaller surface ∆x occupied by each mode lead to a larger number of modes. Again we
did not find a real optimal beam size. The number of modes is larger the smaller the beam
111
waist is. One limit to decreasing the beam waist further is given by the fact that a smaller
beam waist means a larger input angular range and thus a larger area of the dispersion
curve is probed. Since the dispersion curve is only valid over a certain wavelength range
and thus over a certain angular range, this limits the size of the input beam. In the next
section ∆x is related to the spot size w0 using crosstalk considerations. From w0 we can
calculate what angular range a beam has and what range of the dispersion curve will be
probed. In section 8.5 we investigate how performing several bounces within a given
structure influences the number of modes.
8.4 Crosstalk between ModesIn this section we will calculate the crosstalk between modes as a function of the channel
spacing ∆x and the Gaussian spot size w0. The crosstalk is largest between two
neighboring modes and is determined by the overlap of the mode fields. Fig. 8.12 shows
a schematic of two neighboring modes with two bounces in the device, introducing the
nomenclature used. In the following calculations it is assumed that the beams have a
Gaussian profile along the direction of the beam shift and are elongated in the other
direction. Fig. 8.13 shows the Gaussian profile of two neighboring modes as it is seen
along the exit interface in Fig. 8.12. Here the dispersion profile of the non-periodic
design, as given in Fig. 8.5, is assumed. The solid lines are for the center wavelengths,
which are separated by 5 nm and the dotted lines show how far the beams would shift due
to a modulation bandwidth of 50 GHz. The vertical green lines represent the spatial width
∆x of the blue mode.
112
Ls
2wt
2w0
2w1
d1
ns
Fig. 8.12. Schematic of beam shift after two bounces for two neighboring modes.
120 140 160 180 200 220 240 2600
0.2
0.4
0.6
0.8
1
Position in um
Pow
er in
a.u
.
Fig. 8.13. Gaussian profile of two neighboring modes that are separated by 5 nm.
The vertical green lines represent the position of the blue channel. The solid lines
are for the center frequency and the dotted lines represent the shift with 50 GHz
signal modulation.
113
The crosstalk can be calculated by integrating the power of the red mode between the
green lines and dividing it by the power of the blue mode within this spatial section. For
the case that all beams are focused on the exit interface and have a spot size w0 at that
position, all channels have an equal spatial width ∆x and the center positions xc1 and xc2
of two channels are also separated by ∆x. For a negligible shift due to signal modulation,
the crosstalk can be calculated using (8 - 15).
( )
( )5.0
5.02erf2
5.12erf
2exp
2exp
1
12
22
1
12
21
−
∆
∆
=
−−
−−
=
∫
∫∆+
∆+
t
txx
x t
c
xx
x t
c
wx
wx
dxw
xx
dxw
xx
Crosstalk (8 - 15)
In order to incorporate the additional crosstalk due to the beam shift with signal
modulation, integration over frequency as shown in (8 - 16) is performed.
( )( )
( )( )∫ ∫
∫ ∫∆+
∆−
∆+
∆+
∆−
∆+
−−
−−
=mod
mod
mod
mod
5.0
5.0
1
12
22
5.0
5.0
1
12
21
2exp
2exp
λλ
λλ
λλ
λλ
λλ
λλ
c
c
c
c
ddxwxx
ddxwxx
Crosstalkxx
x t
c
xx
x t
c
(8 - 16)
Fig. 8.14 graphs the crosstalk as a function of the modulation frequency. As expected the
crosstalk increases with the signal modulation frequency, but up to 100 GHz we can
expect a quite low crosstalk of around –40 dB.
0 100 200 300 400 50050
40
30
20
10
Signal bandwidth in GHz
Cro
ssta
lk in
dB
Fig. 8.14. Crosstalk as a function of signal bandwidth for the non-periodic design
and a channel spacing of 5 nm.
114
Using these crosstalk considerations, we can thus determine the necessary spatial
separation ∆x of two modes. In the next section the maximum number of modes possible
is determined.
8.5 Number of Surface Modes including BouncesAs discussed in section 8.3 the highest number of modes is obtained if all modes are
focused on the output surface. It turns out that the condition of all modes focussing on the
output surface is approximately equivalent to having a shift that changes linearly with
wavelength. This can be seen as follows. First we redefine the dispersion as given in
(8 - 17), where s is the shift along the interface upon reflection defined in (8 - 18) with
respect to the total thickness of the multilayer stack L.
( ) ( )( )ωλ
ωθωθλ ~
~,~~,~
∂∂
=sDisps (8 - 17)
The definition of the dispersion with respect to the wavelength λ has the advantage that
the dispersion is without dimension in this case and does not change upon scaling the
device. Taking the derivative of the shift in (8 - 18) we obtain (8 - 19).
( )groupLs θtan2= (8 - 18)
( )groupgroup
Ld
dsθθ 2cos
2= (8 - 19)
Thus, Dispω in (8 - 2) and Dispsλ in (8 - 17) are related by (8 - 20).
( ) ( ) ( ) ( )
−=
cDisp
LDisp s
group
πωλ
ωθθ
ωθ λω 2
~~,~2
cos~,~ 22
(8 - 20)
A beam with an angular range of 2∆θstruc for a given input angular range 2∆θ focuses
approximately at a distance Lf as given in (8 - 21). w1 is the spot size at the input interface
as seen in Fig. 8.12, which is identical for all modes.
( )struc
groupf wL
θ
θ
∆≈
2
1
cos(8 - 21)
115
The focal distance has to be equal to the thickness of the structure if the modes are to
focus at the output surface. Therefore, we obtain condition (8 - 22) for ∆θstruc.
( )L
w groupstruc 2
cos2
1
θθ =∆ (8 - 22)
If ∆θstruc is to fulfill (8 - 22) for the given input angular range, we see from (8 - 19) by
replacing the differential by a difference that ∆s has to be constant for that same input
range. In Appendix F we derived in (F - 14) that a change in the input angular range ∆θ is
approximately proportional to a change in the frequency ∆ω. Therefore we deduce that
the change in the shift should be constant over the given frequency interval, which will at
the same time result in a constant shift for the given input angular range. This condition is
equivalent to having a constant spatial dispersion. As the frequency ranges considered are
small compared to the absolute frequency, a change in frequency is approximately
proportional to a change in wavelength. The simplest type of structure that fulfills
condition (8 - 22) for focussing all modes at the same position is a structure that has a
linear shift as a function of wavelength. In Chapter 10 we will see, though, that other
structures can fulfill the equation as well. The change in the shift only has to be constant
over the input angular range ∆θ corresponding to a specific frequency range ∆ω at the
position of the different channels. It is still fulfilled if there are discontinuities between
the center positions of the different channels such as in the step-design in Chapter 10.
Next let us consider the maximum number of channels we obtain if all channels are
focussed at the output plane and therefore have an equal spatial extent ∆x and are equally
spaced in wavelength. For the following calculations we will consider structures with a
linear shift as a function of wavelength and thus a constant dispersion Dispsλ = cDisp. For
operation in transmission, cDisp is the dispersion in transmission, while for operation upon
reflection, it is the dispersion after one bounce. Thus, to obtain the total dispersion we
need to multiply by the number of bounces Nb. With these considerations the number of
surface modes (8 - 14) is rewritten as (8 - 23).
( ) ( )∫ −∆
+=∆
+=∆2
1
1221mod 11,,,λ
λ
λλλλλx
cNd
xcN
xcN DispbDispbDispes (8 - 23)
116
In this calculation the channel spacing ∆λ is given by (8 - 24).
Dispb cNx∆
=∆λ (8 - 24)
Also, the number of modes agrees with the picture of dividing the total bandwidth by the
bandwidth of one mode as given in (8 - 25).
λλλ
∆−
+= 12mod 1esN (8 - 25)
So far we only considered the propagation through the multilayer stack and not the
propagation through the substrate. For the broadening of Gaussian beams the total
propagation length is relevant. In the following it is assumed that the substrate is thick
compared to the stack. Thus, the total propagation distance Lp is determined by the
substrate thickness Ls and is given by (8 - 26), where θs is the propagation angle in the
substrate.
( )s
sbp
LNL
θcos2
= (8 - 26)
All beams have the same spot size w1 at the input and ∆x is a constant as discussed above.
While the propagation through the substrate determines the beam size, the layers only
cause the beam shift and may offset the focal plane or lead to distortions as discussed
previously. As seen in section 8.4, ∆x is determined by crosstalk considerations. It
follows from (8 - 15) that for a constant ratio c1=∆x/wt the crosstalk is constant. c1=3.2
corresponds to approximately -30dB crosstalk and c1=3.8 to -40dB. ∆x is related to ∆λ as
given in (8 - 27), where cDisp is the dispersion of a single bounce and Nb is the number of
bounces.
t
bDisp
t wNc
wxc
λ∆=
∆=1 (8 - 27)
The beam propagates as a Gaussian beam within the structure. The beam broadening
limits the number of bounces. In order to keep the field of the different bounces
separated, (8 - 28) follows from Fig. 8.12.
117
21
1 ≥wd (8 - 28)
In (8 - 28) d1 is the separation between the center of the different bounces and w1 is the
spot size along the interface at the entering surface. d1 is calculated with (8 - 29), where
θs is the propagation angle in the substrate.
( )ssLd θtan21 = (8 - 29)
The beam size of a Gaussian beam is given by (8 - 30) and (8 - 31).2
( )2
0 1
+=
Rzzwzw (8 - 30)
λπ nwzR
20= (8 - 31)
Using these equations, w1 can be approximated for z>> zR as in (8 - 32). The subscript
“s“ refers to the values within the substrate.
( )( ) ( )
( )( ) ( )sss
sb
s
s
ss
ssb
s
s
s
s
sb
nwLNw
nwLNw
LNww
θπλ
θλ
πθ
θθθ
cos2
coscos/21
coscoscos2
2
2
21 ≈
+=
= (8 - 32)
Substituting (8 - 29) and (8 - 32) into (8 - 28) we obtain (8 - 33).
( )
( ) ( )
( ) ( ) 2cossin
cos2
cos
tan2 2
2
≥=λ
πθθ
θπλ
θ
θ
b
stss
sss
sb
s
s
ss
Nnw
nwLNw
L (8 - 33)
Thus, we get as condition (8 - 34) for the spot size.
( ) ( ) sss
bt n
Nwπθθ
λ2cossin
2≥ (8 - 34)
This is a very important equation, since it defines how far we can reduce the spot size. In
section 8.2 we saw that we get more modes the smaller we choose the spot size. Now we
have a condition for the smallest spot size. Thus, the maximum number of modes is
118
obtained if the equal sign in (8 - 34) is fulfilled. Substituting (8 - 34) into (8 - 27), we get
condition (8 - 35) for ∆λ.
( ) ( ) sssDispbDisp
t
ncc
Ncwc
πθθλ
λ 211
cossin2
≥=∆ (8 - 35)
Finally, using (8 - 25) we obtain the maximum number of surface modes possible on the
wavelength interval from λ1 to λ2 as shown in (8 - 36).
( ) ( ) ( )λ
πθθλλ
λλλ
1
21212
2cossin
11c
ncN sssDisp
m
−+≤
∆−
+= (8 - 36)
Or, rewriting (8 - 36) in terms of the incident angle in air, we reach (8 - 37).
( )( ) ( )
1
2
2
12
2
sin1sin1
cn
cN s
Disp
m
−
−+≤
θθπ
λλλ for z>> zR (8 - 37)
(8 - 37) assumes that the channels are focussed at the output surface. In principle we
could perform double the number of bounces such that the channels are focussed after
half the bounces and the exit beam size is identical to the incident beam size. In this case
we obtain twice as many channels, but we need to use a second lens on the output side to
refocus the channels.
It is interesting to note in (8 - 37) that increasing the number of bounces or decreasing the
spot size cannot increase the number of surface modes. For a given crosstalk, the beam
size determines the number of bounces to be performed. The first factor in expression
(8 - 37) shows that only the relative wavelength interval matters. All designs are scalable
to a different wavelength range and have, after the scaling, the same number of modes as
before. As expected, either a higher dispersion or a larger crosstalk (smaller c1) leads to a
larger number of modes. Furthermore, a larger index of refraction in the substrate also
increases the number of modes. Fig. 8.15 graphs the dependence of the maximum number
of modes on the incident angle. It can be seen that the number of modes is approximately
independent of the angle, if the incident angle is larger than 40°.
119
0 20 40 60 800
5
10
15
Incidence angle in deg
Max
imum
num
ber o
f mod
es
Fig. 8.15. Theoretical maximum number of modes in the interval from 1525 nm to
1565 nm as a function of incident angle assuming a dispersion of 60µm/40nm,
ns=1.52 and a crosstalk of –40dB (c1=3.8).
8.6 Verification of Surface Mode ModelSimilarly to the verification of the volume mode model in section 8.2 we will here test
the surface mode model by numerical simulation. The number of surface modes can be
calculated numerically by determining the number of non-overlapping channels along the
exit surface of the structure. First we determine, which wavelength is going to be focused
at the exit interface in distance L. In the example of Fig. 8.16 the center wavelength 1545
nm of the interval is chosen to be focused with a spot size of 20µm. Now we calculate
how large the spot size w1 at the input interface has to be for that beam to be focused.
Together with the incidence angle and the focal spot size, this completely determines the
Gaussian beam at the input interface.
The first mode is chosen to be at the lowest wavelength, here at 1525 nm. For this mode
the 1/e E-field rays are calculated and the cross points with the exit interface are
determined. Table 8.2 lists the calculated angles for the two 1/e-E-field rays and the
position where they cross the back interface. The spacing separating the first mode from
the next mode depends on the desired crosstalk characteristics. To obtain a crosstalk of
around –40dB between neighboring modes, c1=3.8 is chosen. For the definition of c1 see
(8 - 27).
120
1520 1530 1540 1550 1560 157010
20
30
40
50
Wavelength in nm
Prop
agat
ion
angl
e in
cry
stal
in d
eg
(a) (b)
Fig. 8.16. Surface mode calculation for the non-periodic design for an incidence
angle of 40°, focussing on the back side at 1545 nm with a spot size of 20 µm
(corresponding to ∆θin = 1.4°), a wavelength interval from 1525 nm to 1565 nm,
and 13 bounces. To obtain a crosstalk around –40dB, a spacing of 3.8*20
µm/cos(40°)=100 µm is chosen between modes. (a) shows the position of the
obtained modes in terms of wavelength and propagation angle within the crystal.
(b) depicts a cartoon of the modes obtained graphing the two 1/e E-field rays for
each mode.
Wavelength in nm Angle+ in ° Angle- in ° Position+ in µm Position- in µm15251542154815541563
16.9629.5834.5938.7044.91
6.8112.9517.2621.9527.23
382.3 711.6 864.51004.01250.0
610.4 748.7 849.9 965.71106.0
Table 8.2. Wavelength, group propagation angle in the structure for the input angle
plus the half cone angle (Angle+), group propagation angle minus the half cone
angle (Angle-), position where the plus-angle crosses the back interface (Position+),
and position where the minus-angle crosses the back interface (Position-) for the
different modes for the example in Fig. 8.16.
If a mode is focused on the backside, the rays cross at that position and the spot size
along the interface is wt=w0/cos(θ) resulting in a separation between modes given by
121
∆x = c1* wt. For a mode that is not focused on the backside, a larger surface area ∆x is
needed. Here the surface area is increased by the amount between the two rays. This is
not exactly right, but a sufficiently good approximation.
The second mode can now be calculated by searching numerically for a mode that is
separated by the spacing width c1* wt from the first mode. In our example the next mode
is found to have a wavelength of 1542 nm. Table 8.2 lists the numerical data for the five
modes obtained. The distance between the high point of one mode and the low point of
the next mode is always separated by the spacing of 100 µm. In Fig. 8.16 (b) the modes
are shown graphically. Even though the structure might be operated in reflection or with
multiple bounces, the beam path is unfolded in the forward direction. As different
wavelengths propagate in parallel in the substrate, the substrate changes the focal position
approximately equally for all beams. Thus, it does not need to be considered. It is clearly
visible that the three center modes in Fig. 8.16 (b) are quite well focused while the outer
modes are running out of focus.
In (8 - 14) we obtained for the number of surface modes the formula repeated below.
( ) ( )( )( )∫ ∆
+=∆2
1
~~,~cos
~,~1,,,~
221mod
ω
ω
ω ωωθθ
ωθωωθ d
xDispLxN
groupes (8 - 14)
Now we can compare the number of modes predicted by (8 - 14) against the number of
modes obtained by simulation. For the calculation of the integral in (8 - 14) ten
intermediate points are used and a sum is formed. ∆x is calculated using the focal
distance given in (8 - 21) and the spot size at the input w1. Since the input angular range
∆θin =1.4° is rather large, the difference is not replaced by a differential to determine the
angular range within the structure ∆θstruc. The maximum number of surface modes for a
given dispersion is calculated using (8 - 37), also repeated below, where ∆x is assumed to
be the constant ∆x = c1* wt independent of wavelength.
( )( ) ( )
1
2
2
12
2
sin1sin1
cn
cN s
Disp
m
−
−+≤
θθπ
λλλ (8 - 37)
122
0 1 2 3 4 5 6 7 8 9 10 11 12 13 140
1
2
3
4
5
6
7
8
9
Simulated number of modesExpected number of modesMaximum number of modes
Bounces
Num
ber o
f mod
es
Fig. 8.17. Number of surface modes as obtained from simulation, the expected
number of modes calculated using (8 - 14), and the maximum number of modes
obtained from (8 - 23) for the non-periodic design as a function of the number of
bounces within the structure.
Fig. 8.17 graphs the expected number of modes, the maximum number of modes, and the
simulated number of modes as a function of the number of bounces. The expected
number of modes agrees well with the simulated number of modes verifying (8 - 14) for
modeling the number of surface modes. The number of modes obtained with this design
remains significantly below the maximum number of modes though. This is due to the
fact that not all the modes are focused. Some of the surface space is used for unfocused
modes instead of for new modes as visible in Fig. 8.16. In order to obtain the maximum
number of modes, all the modes have to be focused on the backside, i.e. the beam exit
position should change linearly as a function of wavelength. Such a design is introduced
in the next section. This is one of the reasons we were especially interested in designing
stacks with a linear shift as a function of wavelength in Chapters 3 to 7.
123
8.7 Designing for a Maximum Number of ModesIn the preceding sections we concluded that the surface mode picture is valid to calculate
the number of Gaussian modes supported by a structure. The total number of modes is
limited by the condition that the field of consecutive bounces should not overlap and is
given by (8 - 37) repeated below.
( )( ) ( )
1
2
2
12
2
sin1sin1
cn
cN s
Disp
m
−
−+≤
θθπ
λλλ (8 - 37)
To obtain the maximum number of modes, the shift of the structure has to be linear with
wavelength. In this case all modes can be focused on the exit surface of the structure. An
example structure with a linear shift was designed by refining a double-chirped start
design. The new structure has again 200 layers and a thickness of 49 µm. Fig. 8.18 shows
the physical thickness of the layers as a function of the layer position within the stack. As
this design has the optimal linear shift as a function of wavelength, we will call it the
“optimal design.” Its layer composition is given in Appendix G.
0 20 40 60 80 100 120 140 160 180 2000
200
400
600
Layer #
Phys
ical
Thi
ckne
ss in
nm
Fig. 8.18. Physical layer thicknesses of the optimal non-periodic design.
Fig. 8.19 gives the reflectance of the new design as a function of wavelength at 40°
incidence angle. The reflectance drops somewhat for the longer wavelengths and might
be increased with a gold coating. Fig. 8.20 graphs the group propagation angle and the
124
shift as a function of wavelength. The calculations were performed using the transfer-
matrix method. The shift of this design is nicely linear with a dispersion of
cDisp = 1.4 µm/nm.
1520 1540 15600
0.2
0.4
0.6
0.8
1
Wavelength in nm
Ref
lect
ance
Fig. 8.19. Reflectance of the optimal designs at 40° incidence angle.
1520 1540 156010
20
30
40
50
Wavelength in nm
Gro
up p
ropa
gatio
n an
gle
in d
eg
1520 1540 15600
20
40
60
80
100
120
Wavelength in nm
Shift
in u
m
Fig. 8.20. Group propagation angle and shift of the optimal designs at 40°
incidence angle.
125
1520 1530 1540 1550 1560 157025
30
35
40
45
50
Wavelength in nm
Prop
agat
ion
angl
e in
cry
stal
in d
eg
(a) (b)
Fig. 8.21. Surface mode calculation for the optimal design for an incidence angle of
40°, focussing on the back side at 1545 nm with a spot size of 20 µm
(corresponding to ∆θin = 1.4°), a wavelength interval from 1525 nm to 1565 nm,
and 13 bounces. To obtain a crosstalk around –40dB, a spacing of 3.8*20
µm/cos(40°)=100 µm is chosen between modes. (a) shows the position of the
modes obtained, in terms of wavelength and propagation angle within the crystal.
(b) depicts a cartoon of the obtained modes graphing the two 1/e-E-field rays for
each mode.
Wavelength in nm Angle+ in ° Angle- in ° Position+ in µm Position- in µm1525153215381544154915561563
28.3332.6935.8138.9141.6944.1446.68
12.5816.8521.3025.1528.9232.8836.80
684.9 815.3 916.51025.01132.01233.01347.0
708.0 809.2 919.71021.01126.01246.01375.0
Table 8.3. Wavelength, group propagation angle in the structure for the input angle
plus the half cone angle (Angle+), group propagation angle minus the half cone
angle (Angle-), position where the plus-angle crosses the back interface (Position+),
and position where the minus-angle crosses the back interface (Position-) for the
different modes for the example in Fig. 8.21.
126
Using (8 - 37) the maximum number of modes possible with this design in the
wavelength interval from 1525 nm to 1565 nm is obtained to be Nm=8. For a spot size of
20 µm at the focus, this requires 13.7 bounces within the structure. Since only an integer
number of bounces can be performed, we will use 13 bounces. Fig. 8.21 shows the result
of the numerical simulation of this structure and Table 8.3 gives the data corresponding
to the modes. The cartoon in Fig. 8.21(b) depicts that all modes are focused on the back
interface of the structure. Instead of the theoretical limit of eight modes, seven modes are
obtained with this structure. This is due to the reduction of the number of bounces from
13.7 to 13.
0 1 2 3 4 5 6 7 8 9 10 11 12 13 140
1
2
3
4
5
6
7
8
9
Simulated number of modesExpected number of modesMaximum number of modes
Bounces
Num
ber o
f mod
es
Fig. 8.22. Number of surface modes as obtained from simulation, the expected
number of modes calculated using (8 - 14), and the maximum number of modes
obtained from (8 - 23) for the optimal design as a function of the number of
bounces within the structure.
In Fig. 8.22 the simulated, the expected, and the maximum number of modes are given as
a function of the number of bounces within the structure. It can be seen that for 13
bounces only seven modes are predicted. The quality of the design can also be seen in the
fact that the expected and the maximum number of modes are nearly identical.
Furthermore, the number of modes obtained in the simulation also agrees with the
127
maximum number of modes. Thus, we accomplished our goal of designing a structure
that supports the maximum number of modes for the given dispersion. Chapter 9 will
discuss if it is possible to design a structure with the same bandwidth that has a higher
dispersion and therefore supports more modes. This structure has not been fabricated and
we conclude this chapter with a discussion of the experimental results obtained for the
66-layer design with a linear shift as a function of wavelength introduced in Chapter 7.
8.8 Discussion of the 4-channel DEMUX in 7.4After the rather abstract discussion of the factors limiting the number of channels in the
last sections, we will here repeat some key results using the 66-layer non-periodic stack
from section 7.4 as an example. In section 7.4 we experimentally demonstrated the
demultiplexing of four channels by their Gaussian beam width using a stack with a linear
shift as a function of wavelength. In the experiment 8 bounces and a spot size of
w0=10µm were used. In Chapters 1 to 7 only schematic drawings of the device have been
considered such as the one shown in Fig. 8.23(a) depicting the center ray of the beam. In
reality the beam propagates as Gaussian beams through the stack as shown in the scaled
drawing in Fig. 8.23(b). In the drawing the beams are delimited by the angular
components of 1/e2 intensity. As shown in the previous sections, the widening of the
Gaussian beam limits the number of bounces that can be performed.
From (8 - 37) we obtain that we should be able to demultiplex 7 channels using the given
66-layer stack. In order to obtain all seven channels, we could modify the experiment in
two ways – increase the number of bounces of decrease the incident spot size. In section
8.5 we have seen that we can deduce the number of bounces necessary to obtain the
maximum number of channels from the spot size. We calculate that for a spot size of
w0=10µm, 18 bounces should be performed.
128
zy
x
827.3 nm831.0 nm834.7 nm840.6 nm
Multiplexedlight
1 mm 13.4 µm
≈10
mm
…
240 µm
400 µm
Zoomed-in:
10 mm
1.013 mm
Gold coating
Substrate
Dielectric stack(13.4 µm thick)
Lens
(a) (b)
Fig. 8.23. (a) Schematic of a 4-channel demultiplexer. (b) Scaled drawing of the
same 4-channel demultiplexer.
Fig. 8.24 shows on the left the beam propagation with 8 bounces and on the right with 18
bounces. The beams are depicted by their 1/e2-intensity rays. The top ray is shown as a
dashed line and the bottom ray as a solid line. The center ray is not depicted. Due to the
larger number of bounces the device size is increased, but we can now demultiplex 7
channels. We see that the number of bounces is limited by field overlap between bounces
at the input side. In order to prevent loss through the input window in the gold coating,
the bounces should not overlap. A second method for demultiplexing seven channels with
the given stack is to reduce the input spot size to w0=4.5µm as shown in Fig. 8.25. Again
the number of bounces is limited by field overlap at the input window in the gold coating.
Due to the smaller focus size, the Gaussian beam broadens quicker and therefore only 8
bounces are possible. But on the other hand a smaller total shift is required to separate the
channels due to their smaller spatial extend.
129
240 µm
400 µm
Zoomed-in:
10 mm
1.013 mm
w0=10 µm, 8 bounces,4 channels
w0=10 µm, 18 bounces,7 channels
240 µm
400 µm
Zoomed-in:
25 mm
1.013 mm
Rays of1/e2-intensity
Overlap limitsnumber of bounces
Fig. 8.24. Increasing the number of channels by increasing the number of bounces.
Fig. 8.24 and Fig. 8.25 clearly show that the field of different bounces shouldn’t overlap
to prevent loss and that the number of bounces and the input spot size are coupled. As
discussed theoretically in section 8.5 from a given spot size follows the maximum
number of bounces that can be performed without field overlap. Equivalently, for a given
number of bounces, the smallest possible spot size follows. As long as an optimal pair of
spot size and number of bounces is chosen, the maximum number of channels is
obtained. This is the reason that the number of bounces and the spot size do not appear in
the model (8 - 37) for the maximum number of channels possible. Note that we could
double the number of bounces for the same spot size if we use a second focusing lens at
the output, thus doubling the number of channels possible.
130
240 µm
400 µm
Zoomed-in:
10 mm
1.013 mm
w0=10 µm, 8 bounces,4 channels
10 mm
1.013 mm
w0=4.5 µm, 8 bounces,7 channels
240 µm
400 µm
Zoomed-in:
Rays of1/e2-intensity
Overlap limitsreduction inspot size
Fig. 8.25. Increasing the number of channels by decreasing the spot size.
One might think that it is possible to increase the number of bounces if the substrate
thickness is reduced, because this reduces the total propagation distance. As seen from
Fig. 8.26 that is not true because the distance between the bounces is reduced by the same
factor. Therefore the substrate thickness has no influence on the number of separable
channels. A thinner substrate has the advantage, though, that the total device size shrinks
without decreasing the number of separable channels.
Fig. 8.26 demonstrates that reducing the substrate thickness from Ls=1mm to Ls=80µm
reduces the device length from 10 mm to 1.2 mm. The substrate thickness is limited by
overlap of the exiting light of the least shifted channel with the last bounce of the most
shifted channel as seen in Fig. 8.26. For a substrate thinner than 80µm either the red
channel would experience loss if we extended the gold coating or the blue channel would
have loss as part of the light already exits the structure before the last bounce.
Furthermore, this “blue” light would appear as crosstalk in the “red” channel. Therefore,
131
for the most compact device the substrate thickness should be chosen just thick enough to
prevent overlap (mechanical stability permitting).
70 µm
150 µm
Zoomed-in:
10 mm
1.013 mm
w0=4.5 µm, Ls =1 mm,7 channels
240 µm
Zoomed-in:
400 µm
1.2 mm
93.4 µm
w0=4.5 µm, Ls=80 µm,7 channels
Rays of1/e2-intensity
Overlap limitsreduction insubstrate thickness
Fig. 8.26. Reducing the device size by reducing the substrate thickness.
In conclusion, Chapter 8 has derived two models for calculating the number of modes
that a given multilayer stack can demultiplex. The volume mode model (8 - 6) is
concerned with modes that are separated throughout the whole volume of the structure by
mutually exclusive angular content of the modes. More interesting for the case of a
wavelength demultiplexing device is the number of possible surface modes, i.e. the
maximum number of channels we can spatially separate on one output surface. Using
(8 - 14) we can calculate how many modes we can expect at best for a given dispersion
profile. Furthermore, we derived that the best dispersion profile is given by a linear shift
as a function of wavelength, since in this case all modes are focussed on the same output
plane. Assuming that we have such an ideal linear shift as a function of wavelength, the
optimal number of channels can be calculated using (8 - 37). We see that (8 - 37) does
not depend on the number of bounces or the spot size as long as an optimal combination
132
of the two is chosen. In order to obtain a larger number of modes, either the dispersion or
the bandwidth of the stack has to be increased as the number of modes is proportional to
the dispersion-bandwidth product. In Chapter 9 we will investigate how large of a
dispersion-bandwidth product – and therefore number of channels – can be obtained for a
given material system, stack thickness, and incidence angle.
References
[1] D. A. B. Miller, "Communicating with waves between volumes - evaluating
orthogonal spatial channels and limits on coupling strengths," Appl. Opt. 39, 1681
– 1699 (2000).
[2] A. E. Siegman, Lasers, University Science Books, Sausalito, CA (1986).
133
Chapter 9
Maximum Shift
In Chapter 8 we discussed how many wavelength channels we can expect to spatially
demultiplex using a given thin-film stack. We found that the largest number of channels
is obtained if the stack has a linear shift as a function of wavelength. In this case all
channels focus at the same output plane and the maximum number of channels Nchannels is
given by (9 - 1).
( ) ( )
crosstalk
schannels c
nsN2
sin1sin1
2
2
−
∆+≤
θθπ
λ(9 - 1)
Here we replaced the dispersion×wavelength-range product by the total shift ∆s as given
by (9 - 2).
λ∆=∆ Dispcs (9 - 2)
In (9 - 1) θ is the incidence angle in vacuum, ns the refractive index of the substrate, and
ccrosstalk a constant fixed by the crosstalk between channels. ccrosstalk=3.2 corresponds to
approximately -30dB and ccrosstalk=3.8 to -40dB crosstalk between adjacent channels as
derived in section 8.4. A linear shift corresponds to a constant dispersion cDisp=∂s/∂λ. λ is
134
the center wavelength of the device and ∆λ is the wavelength range over which the
device has a linear shift.
Since a given design typically only works for a particular wavelength λ, incidence angle
θ, and substrate refractive index ns, and shows a specific total shift ∆s, the only way we
can influence the number of channels given in (9 - 1) is by trading off the crosstalk.
Allowing for a larger crosstalk, we obtain more channels, for less crosstalk, less channels.
Thus, if we want to demultiplex a particular number of channels, we need to design a
stack with a minimum total shift ∆s that can be calculated from (9 - 1). A larger total shift
∆s corresponds to more channels. Table 9.1 compares the structures we have obtained
using the different design methods discussed in Chapters 3 to 7. As we could not obtain a
linear shift as a function of wavelength for the superprism effect, it is not listed here.
Double-Chirped Coupled-Cavity Optimized
Number of Layers 200 60 33 66Stack Thickness L 28 µm 9 µm 16 µm 13 µm
Wavelength Range ∆λ 170 nm 40→200nm 12 nm 22 nm
Total Shift ∆s 35 µm 10 µm 22 µm 22 µm
Spatial DispersioncDisp=∆s/∆λ 0.2 µm/nm 0.05→0.25
µm/nm 1.8 µm/nm 1.0 µm/nm
Table 9.1. Comparison of linear shift characteristics obtained with the double-
chirped stacks in Chapter 5, the coupled-cavity stack in Chapter 6, and the
numerically optimized stack in Chapter 7.
As discussed previously, the different design methods result in different spatial dispersion
values. It is clear from Table 9.1 that the double-chirped designs have a low dispersion,
the coupled-cavity design has a high dispersion, and the optimized design has an
intermediate dispersion. Due to the different bandwidth of the designs, the dispersion is
not a good figure of merit. The number of channels that can be demultiplexed depends on
the total shift, i.e. the product of the dispersion with the wavelength range of the design.
Therefore, we consider the total shift as the figure of merit for how good a design is. In
135
Table 9.1 we observe that the 200-layer double-chirped design has the largest total shift
and would therefore demultiplex the largest number of channels. The five 60-layer
double-chirped designs have on the other hand the lowest total shift, while the coupled-
cavity design and the optimized design exhibit an intermediate amount of shift. Checking
now the stack thickness values for the different design, we see that the 200-layer double-
chirped design is also the thickest design, while the 60-layer double-chirped stacks are
the thinnest designs. This might explain why the 200-layer design exhibits the largest
shift. To test this hypothesis, we divide the shift by the stack thickness for each design.
Fig. 9.1(a) shows the shift as a function of the dispersion for the different designs and
Fig. 9.1(b) plots the shift normalized by the stack thickness for the different designs.
0.01 0.1 1 100
10
20
30
40
Chirped DesignsCavity DesignOptimized Design
Dispersion [um/nm]
Wav
elen
gth
Ran
ge/T
hick
ness
[nm
/um
]
0.01 0.1 1 100
0.5
1
1.5
2
Dispersion [um/nm]
Wav
elen
gth
Ran
ge/T
hick
ness
[nm
/um
]
Dispersion in µm/nm
Shift
in µ
m
Dispersion in µm/nm
Shift
/ St
ack
Thic
knes
s
(a) (b)
Fig. 9.1. (a) Shift as a function of dispersion for the designs in Table 9.1. (b) Shift
divided by stack thickness for the different designs.
Inspecting Fig. 9.1(b) we see that all double-chirped designs perform equally well if the
varying thickness is taken into account. Therefore, we can now hypothesize that the total
shift is proportional to the stack thickness.
Suppose now we are given the task to design a structure that demultiplexes 100 channels
with a wavelength spacing of 100 GHz or 0.8 nm around 1550 nm. From Fig. 9.1(b) we
conclude that we probably need to have a specific minimum stack thickness to achieve
136
this goal. The interesting questions now are what materials should we use for the design,
how many layers are necessary, and is such a design possible at all.
In the next sections I will try to answer these questions by developing a heuristic model
for estimating the achievable total shift ∆s of a structure with certain material and
operational properties. The approach used will be to generate large numbers of designs
with different parameters and then compare the results to extract common features. In
section 9.1 I will discuss how to generate appropriate designs automatically. In section
9.2 the model derived from the more than 700 generated designs will be presented.
Section 9.3 discusses the numerical results. In section 9.4 the maximum number of
channels is calculated as a function of the material refractive indices, the stack thickness,
the incidence angle, the crosstalk, and the center wavelength. Using this heuristic model
our design question of a 100-channel demultiplexer will be analyzed.
9.1 Automatic Generation of DesignsIn Chapter 7 we discussed that numerical optimization techniques allow the design of the
most general stacks, as they are not a priori limited to certain types of structures such as
chirped stacks or resonator stacks. Therefore, we will use numerical techniques in this
chapter. As shown in Chapter 7 the procedure for numerically obtaining a stack with the
desired dispersion properties is to synthesize a starting design followed by repeated
iterations of numerical refinement. An impedance matched Bragg stack is for example a
good starting design as discussed in Chapter 7. This starting design was then optimized
by numerical refinement techniques. The implemented refinement techniques include the
Golden-Section-Search (GS) and the Secant-Method (SEC), as well as the Conjugate-
Gradient-Algorithm (CG), the Broyden-Fletcher-Goldfarb-Shanno-Algorithm (BFGS),
the Damped-Least-Squares-Method (DLSQ), and the Hooke-&-Jeeves-Pattern-Search
(HJ) technique.1,2,3,4,5,6,7 While the first two are one-dimensional search algorithms, the
last four methods use techniques varying all design parameters simultaneously. All
techniques except DLSQ use merit functions to measure the performance of a design and
achieve performance improvements. During the refinement process the various methods
137
are used in varying order and for different numbers of iterations. The resulting designs
are manually tested every couple of iterations to ensure a good progress. Such a design
process takes on the order of four to five days as a single iteration of a specific algorithm
has a run time on the order of hours.
9.1.1 Automating the Refinement Process
It is obvious that this design process cannot be used to generate a sufficient number of
designs to derive a heuristic model within a couple of months. To accomplish this, we
have to both decrease the design time and automate the design process. Decreasing the
design time leads to less optimized designs. Thus, it is important to find a good balance
between optimization time and design error. By trial and error I found that using the
algorithm given in (9 - 3) results in reasonable designs.
Refinement algorithm (9 - 3)
1. 2 iterations HJ
2. 2 iterations CG
3. 1 iteration SEC
4. 2 iterations HJ
5. 1 iteration GS
6. 2 iterations BFGS
The algorithm is performed either one or two times for a given set of parameters using a
p-norm of six (see Chapter 7). The DLSQ method is not used as it is not based on a merit
function and only allows for second-degree i.e. square error correction.
The runtime trefine in minutes needed to refine one design can be estimated by (9 - 4),
where the proportionality constant is specific to the type of computer used. Here we
implemented all algorithms as MathCad programs and used a Pentium III 750 MHz
computer with 256 MB RAM. In (9 - 4), i is the number of times the algorithm in (9 - 3)
is performed, m is the number of periods, and p the number of points used for the merit
function.
138
[ ] pmitrefine2
6501min = (9 - 4)
From (9 - 4) we can calculate that refining a design with 100 periods using 15 points in
the merit function and one run of algorithm (9 - 3) will take approximately 230 minutes.
For 50 periods, we would need 60 minutes and, for 20 periods, 10 minutes. The number
of points necessary for the merit function depends on the wavelength interval of interest.
The spacing of the points has to be fine enough in wavelength to prevent resonances in
the design. For a 1550-nm design the spacing between points should be at least 5 nm or
0.3% of the wavelength. Thus, for a 75 nm operating range, 15 points are needed.
9.1.2 Automatic Refinement of Fixed Start Designs
For approximately the first half of the designs I manually generated a starting design
using the theory of double-chirped mirror structures.8 Then I used the automated
refinement process described in the previous section to refine the starting design to
different performances. Mostly, I fixed the operating range and specified a different total
shift for the different designs. Fig. 9.2 to Fig. 9.4 show three designs generated in such a
manner. The blue crosses represent the start design, the black circles the specified shift,
and the red crosses the refined design. The solid black line is the actual operating range
of the designs. On the right the actual operating parameters of each design as well as the
fitting error are given.
For the design in Fig. 9.2 the merit function was set such that the shift increases from 5
µm to 25 µm over a wavelength range of 70 nm. We see that the refined design agrees
well with desired characteristics.
For Fig. 9.3 and Fig. 9.4 the same starting design was used, but the shift was specified to
increase to 35 µm and 45 µm respectively. The final designs obtained do not agree well
with the desired characteristics. They start out with a shallower slope than desired, but do
show the right slope for higher wavelengths. Therefore, only part of the wavelength range
is actually usable.
139
1520 1540 1560 15800
5
10
15
20
25
30
Start designRefined designSpecified shiftvalid shift
Wavelength in nm
Shift
in u
m
∆λ 63= nm
cDisp 0.286= um/nm
shift 18= um
errorlsq 0.68= um
errorrel 3.779%=
Fig. 9.2. 20-period SiO2-Ta2O5 design for a 40º incidence angle. The shift was
specified to increase from 5 µm to 25 µm over a 70 nm operating range.
1520 1540 1560 15800
10
20
30
40
Start designRefined designSpecified shiftvalid shift
Wavelength in nm
Shift
in u
m
∆λ 50.167= nm
cDisp 0.429= um/nm
shift 21.5= um
errorlsq 2.604= um
errorrel 12.112%=
Fig. 9.3. 20-period SiO2-Ta2O5 design for a 40º incidence angle. The shift was
specified to increase from 5 µm to 35 µm over a 70 nm operating range.
140
1520 1540 1560 15800
10
20
30
40
50
Start designRefined designSpecified shiftvalid shift
Wavelength in nm
Shift
in u
m
∆λ 32.667= nm
cDisp 0.571= um/nm
shift 18.667= um
errorlsq 1.59= um
errorrel 8.518%=
Fig. 9.4. 20-period SiO2-Ta2O5 design for a 40º incidence angle. The shift was
specified to increase from 5 µm to 45 µm over a 70 nm operating range.
Algorithm for finding the usable wavelength range of a design (9 - 5)
1. Assume that the maximum specified wavelength is the end
wavelength of the range.
2. Assume that the design has the specified dispersion in the usable
range.
3. Calculate the shift at 60 points in the specified wavelength range.
4. Calculate the standard deviation from the specified line, weighting
the result by the inverse of the square root of the number of points.
5. Repeat 4, sequentially neglecting one more point at the beginning
and shifting the specified line to the new starting point until only 6
points are left.
6. Find the starting point that results in the smallest weighted standard
deviation. This is the minimum wavelength of the range.
7. Plot the resulting fit for the wavelength range found.
141
For most of the designs obtained in the way described here, the usable wavelength range
lies on the high wavelength side of the specified operating range. Under the assumption
that the usable wavelength range extends all the way to the maximum specified
wavelength, I implemented a search algorithm to find the smallest wavelength for which
the design still performs sufficiently well. This algorithm is given in (9 - 5).
The weighing by the number of points in steps 4 and 5 is done to favor larger wavelength
intervals. Otherwise the algorithm finds only parts of the actual wavelength range. That is
also the reason why at least 6 points are used for the evaluation. Fewer points can lie on a
perfect line by chance leading to a too small found wavelength range. This algorithm has
been applied to the designs in Fig. 9.2 to Fig. 9.4. The best weighted fit is given by the
black solid line clearly visible in Fig. 9.3 and Fig. 9.4. The determined wavelength
interval ∆λ, as well as the specified dispersion cDisp and the resulting shift are given to the
right of the graphs. Also given are the standard deviation errorlsq and the relative standard
deviation errorrel, i.e. the error divided by the total shift. The relative error represents the
quality of the design.
9.1.3 Automatic Generation of Starting Designs
In the previous paragraphs it has been demonstrated that we can automatically generate
different designs from a single starting design and find the usable wavelength ranges of
the designs. To test the performance of a more general type of design, it is necessary to
vary also the starting design. We want to be able to specify the refractive indices of the
two materials used, the number of layers, the incidence angle, the polarization, and the
operating wavelength and then automatically generate an appropriate starting design. It
turns out that an impedance matched Bragg stack is a good starting design. The Bragg
wavelength under a non-zero incidence angle is given by (9 - 6) and (9 - 7).
( )HHHLLLB dndn θθλ coscos2 += (9 - 6)
=
HLHL n ,
,sinarcsin θ
θ (9 - 7)
142
A single chirp of the structure performs the impedance matching. Thus, the layer
thicknesses dL and dH for the low and high refractive index layers can be calculated as a
function of the period number i using (9 - 8) and (9 - 9).
<≤−
−<≤
−−
=
mimforn
miforn
mi
d
LL
B
LL
B
Li
4cos4
41cos4
52
05.1
θλ
θ
λ
(9 - 8)
<≤−
−<≤
−=
mimforn
miforn
mi
d
HH
B
HH
B
Hi
4cos4
41cos4
5
05.1
θλ
θ
λ
(9 - 9)
Fig. 9.5 gives an example of a 20-period starting design for refractive indices of 1.45 and
2.09, an incidence angle θ of 40º, and an operating wavelength of 1550nm.
Fig. 9.5. Example of a 20-period starting design for refractive indices of 1.45 and
2.09, an incidence angle θ of 40º, and an operating wavelength of 1550nm.
Now we have completely automated the design process including the generation of the
starting design and the refinement to a final design.
0 5 10 15 20 25 30 35 400
200
400
600
Layer #
Phys
ical
Thi
ckne
ss [n
m]
143
9.2 Deriving a Heuristic Shift ModelUsing the automated design process described in section 9.1 a total of 760 different
designs was generated including 190 designs with 20 periods, 49 with 30 periods, 379
with 50 periods, 22 with 60 periods, and 120 with 100 periods. Each design had a
different set of design and material parameters. The refractive indices, the incidence
angle, the number of layers, and the specified dispersion characteristics were varied. All
designs assume a center wavelength of 1550 nm and p-polarization. Different center
wavelengths could be obtained by simple scaling, while the effect of polarization still
needs to be analyzed at a later time. The pure computation time for these designs was
approximately 1100 hours or 45 days. In this section I will present the results obtained
from these designs. 137 designs were filtered out as they had a relative error larger than
25%, leaving 623 results to be analyzed.
As discussed in the introduction to this chapter, we want to determine the total shift ∆s as
a function of the design and material parameters. Fig. 9.6(a) to Fig. 9.11(a) plot the
resulting shift as a function of the different design parameters. In all the graphs red points
represent designs of 20 or 30 periods, blue points 50 or 60 periods, and magenta points
100 periods, as seen from Fig. 9.6. By considering the resulting shift of all these different
structures, the heuristic model (9 - 10) is obtained for the total shift. In this equation ∆n is
the refractive index difference between the two used material, navg is the average
refractive index, L the total stack thickness and θ the incidence angle in vacuum.
( )θλ sin16 2avg
Disp nnLcs ∆
=∆=∆ (9 - 10)
Intuitively, it seems reasonable that the maximum shift possible should be proportional to
the relative width of the Bragg stopband for the given parameters and to the distance
traveled in the design. These assumptions were used to find a model fitting the data. If
this model is correct, normalizing the observed shift by the right hand side of (9 – 10)
should lead to a unity result independent of the material and design parameters.
Fig. 9.6(b) to Fig. 9.11(b) plot the normalized shift as a function of the different design
parameters.
144
0 20 40 60 80 1000
200
400
600
800
m=20/30m=50/60m=100
Average Refractive Index
Shift
[um
]
0 20 40 60 80 1000
0.5
1
1.5
2
2.5
m=20/30m=50/60m=100
Average Refractive Index
Nor
mal
ized
Shi
ft [a
.u.]
Number of Periods
Shift
in µ
m
Number of Periods
Shift
/ M
odel
Fig. 9.6. Dependency of the shift on the number of periods.
20 40 600
200
400
600
800
m=20/30m=50/60m=100
Average Refractive Index
Shift
[um
]
20 40 600
0.5
1
1.5
2
2.5
m=20/30m=50/60m=100
Average Refractive Index
Nor
mal
ized
Shi
ft [a
.u.]
Incidence Angle in deg
Shift
in µ
m
Incidence Angle in deg
Shift
/ M
odel
Fig. 9.7. Dependency of the shift on the incidence angle.
From Fig. 9.7 we see that the maximum obtainable shift increases with incidence angle.
The maximum shift decreases with the average refractive index as shown in Fig. 9.8. It
increases with the refractive index contrast ratio and the stack thickness as seen from
Fig. 9.9 to Fig. 9.11.
145
1 1.5 2 2.5 3 3.5 40
200
400
600
800
m=20/30m=50/60m=100
Average Refractive Index
Shift
[um
]
1 1.5 2 2.5 3 3.5 40
0.5
1
1.5
2
2.5
m=20/30m=50/60m=100
Average Refractive Index
Nor
mal
ized
Shi
ft [a
.u.]
Average refractive index
Shift
in µ
m
Average refractive index
Shift
/ M
odel
Fig. 9.8. Dependency of the shift on the average refractive index.
0 10 20 30 40 50 60 700
200
400
600
800
m=20/30m=50/60m=100
Average Refractive Index
Shift
[um
]
0 10 20 30 40 50 60 700
0.5
1
1.5
2
2.5
m=20/30m=50/60m=100
Average Refractive Index
Nor
mal
ized
Shi
ft [a
.u.]
Refractive Index Contrast in %
Shift
in µ
m
Refractive Index Contrast in %
Shift
/ M
odel
Fig. 9.9. Dependency of the shift on the refractive index contrast.
146
0 20 40 600
200
400
600
800
m=20/30m=50/60m=100
Average Refractive Index
Shift
[um
]
0 20 40 600
0.5
1
1.5
2
2.5
m=20/30m=50/60m=100
Average Refractive Index
Nor
mal
ized
Shi
ft [a
.u.]
Physical Stack Thickness in µm
Shift
in µ
m
Physical Stack Thickness in µm
Shift
/ M
odel
Fig. 9.10. Dependency of the shift on the physical stack thickness.
0 10 20 30 40 50 600
200
400
600
800
m=20/30m=50/60m=100
Average Refractive Index
Shift
[um
]
0 10 20 30 40 50 600
0.5
1
1.5
2
2.5
m=20/30m=50/60m=100
Average Refractive Index
Nor
mal
ized
Shi
ft [a
.u.]
Optical Stack Thickness in µm
Shift
in µ
m
Optical Stack Thickness in µm
Shift
/ M
odel
Fig. 9.11. Dependency of the shift on the optical stack thickness.
Fig. 9.12 shows the shift as a function of the relative fitting error. It can be seen that there
is no particular dependency of the shift on the fitting error.
147
0 10 20 300
200
400
600
800
m=20/30m=50/60m=100
Average Refractive Index
Shift
[um
]
0 10 20 300
0.5
1
1.5
2
2.5
m=20/30m=50/60m=100
Average Refractive Index
Nor
mal
ized
Shi
ft [a
.u.]
Relative Fitting Error in %
Shift
in µ
m
Relative Fitting Error in %
Shift
/ M
odel
Fig. 9.12. Dependency of the shift on the relative error.
0 20 40 60 80 1000
200
400
600
800
m=20/30m=50/60m=100
Average Refractive Index
Shift
[um
]
0 20 40 60 80 1000
0.5
1
1.5
2
2.5
m=20/30m=50/60m=100
Average Refractive Index
Nor
mal
ized
Shi
ft [a
.u.]
Filling Ratio in %
Shift
in µ
m
Filling Ratio in %
Shift
/ M
odel
Fig. 9.13. Dependency of the shift on the filling ratio.
Fig. 9.13 plots the shift as a function of the wavelength interval filling ratio, i.e. the ratio
of the usable wavelength interval to the total specified wavelength interval. The large
number of designs with a filling ratio of approximately 50 % is due to the fact that I
actually specified the filling ratio to be 50 % after guessing a shift model. The 50 %
148
filling ratio is desirable for this type of test as small filling ratios have a larger error and
large filling ratios do not test the design limits as well.
( )inavg
Disp nnLcs θλ sin16 1
∆=∆=∆
1 1.5 2 2.5 3 3.5 40
0.5
1
1.5
m=20/30m=50/60m=100
Average Refractive Index
Nor
mal
ized
Shi
ft [a
.u.]
Average refractive index
Shift
/ M
odel
1 1.5 2 2.5 3 3.5 40
2
4
6
8
m=20/30m=50/60m=100
Average Refractive IndexN
orm
aliz
ed S
hift
[a.u
.]
( )inavg
Disp nnLcs θλ sin16 3
∆=∆=∆
Average refractive indexSh
ift /
Mod
el
(a) (b)
Fig. 9.14. Two poor models for comparison. (a) Assuming the shift to be inversely
proportional to the refractive index. (b) Assuming the shift to be inversely
proportional to the refractive index cubed.
After looking at Fig. 9.6(b) to Fig. 9.11(b) the reader may not be fully convinced that
(9 - 10) is indeed a good model for the shift. As a comparison Fig. 9.14 plots the
normalized shift for two wrong modes. Fig. 9.14(a) assumes the shift to be inversely
proportional to the refractive index, while Fig. 9.14(b) assumes the shift to be inversely
proportional to the refractive index cubed instead of the correct inverse proportionality to
the refractive index squared. In Fig. 9.14(a) a decrease in the shift as a function of the
average refractive index is clearly visible. In Fig. 9.14(b) an increase of the shift with
average refractive index is seen. Thus, a shift that is inversely proportional to the square
of the average refractive index seems indeed to be the best model. As we are basically
guessing a model here based on a numerical data set, there is room for error and
149
improvement. Maybe the average refractive index is really inversely proportional to the
refractive index to the power 2.1. As this result is close to the guessed one, we cannot
conclusively decide either way from the given data. An analytical model is needed to
answer this question satisfactorily. But since no analytical model is available to date, we
will use the derived heuristic model in the mean time.
9.3 Constancy of Dispersion×Wavelength-Range-Product
In 9.2 we numerically derived a model for the maximum total shift that can be obtained
for a given parameter set. The heuristic model in (9 - 10) tells us that for a given set of
refractive indices, a given stack thickness, and a given incidence angle, the total shift ∆s,
which is equal to the dispersion×wavelength-range-product cDisp∆λ, is constant.
Therefore, if we specify a specific dispersion, the wavelength range over which we can
achieve this dispersion is limited and vice versa. This fact can be seen very impressively
if we plot the wavelength range normalized by the expected cDisp∆λ-product as a function
of the dispersion for all the generated designs. In Fig. 9.15 this normalized wavelength
range is plotted as a function of the dispersion. The red diamonds represent designs with
less than 10 % fitting error, the pink plusses 10 % to 25 % fitting error, and the blue
points show the performance of the designs listed in Table 9.1. If our model is correct
that the dispersion×wavelength-range-product cDisp∆λ is constant, dividing the observed
wavelength range by the theoretical value of cDisp∆λ should result in 1/cDisp., which is just
inversely proportional to the dispersion itself. The expected curve is plotted as the solid
black line in Fig. 9.15.
Fig. 9.15 clearly shows that there seems to be a physical limit to the size of the
dispersion×wavelength-range-product. It cannot be pure chance that we did not find any
design with a higher normalized wavelength range. Designs can always be worse as the
automated design process may have failed. The high number of designs close to the
maximum dispersion×wavelength-range-product shows though that our design procedure
works quite well. The designs discussed in Chapters 5 to 7 all lie slightly below the
150
model. This may be due to a tradeoff between ripple in the shift and total shift. As the
designs in this chapter are generated automatically, they may not be as flat as the designs
of the previous chapters. Fig. 9.15 also shows very impressively that all three design
algorithms discussed in Chapters 5 to 7 yield structures of similar performance.
Therefore, the algorithm that most easily generates a good design should be chosen. For
broadband designs this is most likely the double-chirped mirror design technique, while
for narrowband designs digital lattice techniques resulting in coupled-cavity structures
are interesting. If none of these analytical techniques is applicable as for structures with
an intermediate bandwidth, numerical optimization can be used.
0.01 0.1 1 10 1001 .10 3
0.01
0.1
1
10
100
Dispersion [um/nm]
Nor
mal
ized
Wav
elen
gth
Ran
ge [a
.u.]
Dispersion [µm/nm]
Wav
elen
gth
Ran
ge /
(cD
isp∆
λ )th
eory
[nm
/µm
]
Model10 % < relative error < 25 %relative error < 10 %Chirped DesignsCoupled-Cavity DesignOptimized Design
Fig. 9.15 Normalized wavelength range as a function of the specified dispersion.
151
9.4 Model for the Maximum Channel NumberNow we can use the heuristic model (9 - 10) for the maximum shift to estimate the
maximum number of channels given the refractive indices of the two materials used, the
stack thickness, the incidence angle, the polarization, and the operating wavelength.
Substituting (9 - 10) into (9 - 1), we obtain (9 - 11).
( ) ( )
−
∆+≤ 2
22
2sin1sin81
s
inin
avgcrosstalkchannels nn
nc
LN θθπ
λ(9 - 11)
It is interesting to note that for a given set of materials, incidence angle, and adjacent
channel crosstalk, everything on the right hand side except for the ratio of the stack
thickness L to the wavelength λ is fixed. For a crosstalk –40dB, a 45º incidence angle,
and alternating SiO2 and Ta2O5 layers, the number of channels can be estimated by
(9 - 12).
λLN OTaSiO 53.0145,522 +≤°− (9 - 12)
For a 50 µm stack and an operating wavelength of 1550 nm, we obtain a maximum of 18
channels from (9 - 12). For a 13 µm stack and an operating wavelength of 850 nm, we
expect 9 channels. The 13-µm stack discussed in Chapter 8 had a theoretical limit of 7
channels, which is not far off from the nine channels estimated here. Keep in mind that
this shift model is an approximation and thus serves as a rough guide to how many
channels we might expect. For comparison we consider the number of channels obtained
for a –40dB crosstalk, a 45º incidence angle, and alternating GaAs and AlAs layers in
(9 - 13).
λLN AlAsGaAs 21.0145, +≤°− (9 - 13)
We see that we expect 2.5 times fewer channels using GaAs-AlAs as compared to SiO2-
Ta2O5 for the same stack thickness. If a design is transferred from SiO2-Ta2O5 to GaAs-
AlAs though, the thickness scales by the inverse of the refractive index. Therefore, the
same design would be a factor of two thinner in GaAs-AlAs, resulting in about 5 times
fewer channels.
152
It is interesting to note that the number of channels scales as the thickness divided by the
wavelength. Thus, scaling a design to a different wavelength in the same material system
does not change the number of channels, just as we would expect. In this model the
thickness not the number of layers seems to matter. That is not correct though, since the
layer thickness should be on the order of a quarter wavelength. Therefore, fixing a stack
thickness also fixes the number of layers needed.
Finally, let us consider the design question posed at the beginning. Can we design a
structure that demultiplexes 100 channels with a wavelength spacing of 100 GHz or
0.8 nm around 1550 nm? Setting the number of channels in (9 – 12) equal to 100, we see
that we would need a stack that is approximately 270 µm thick. This corresponds to
approximately 1000 layers, which is well out of the range of what is possible today. If we
have the freedom to change the material system from SiO2-Ta2O5 to something different,
we see from (9 - 11) that we would want a high contrast ratio and a low average
refractive index. Furthermore, changing to a larger incidence angle could also help to
obtain more channels.
References
[1] Li Li and J. A. Dobrowolski, Appl. Opt., 31/19, pp.3790-3799 (1992).
[2] A. Thelen, Design of Optical Interference Coatings, McGraw-Hill, Inc. (1989).
[3] C.K. Madsen and J.H. Zhao, Optical Filter Design and Analysis - A Signal
Processing Approach, John Wiley & Sons, Inc. (1999).
[4] J. A. Dobrowolski, F.C. Ho, A. Belkind, and V.A. Koss, Appl. Opt., 28/14,
pp.2824-2831 (1989).
[5] J. A. Dobrowolski and R.A. Kemp, Appl. Opt., 29/19, pp.2876-2893 (1990).
[6] E.K.P. Chong and S.H. Zak, An Introduction to Optimization, John Wiley & Sons,
Inc. (1996).
[7] T.E. Shoup and F. Mistree, Optimization Methods with Applications for Personal
Computers, Prentice-Hall, Inc. (1987).
153
[8] N. Matuschek, F.X. Kärtner, and U. Keller, “Analytical Design of Double-Chirped
Mirrors with Custom-Tailored Dispersion Characteristics,” IEEE J. Quantum
Electron., 35/2, 129-137 (1999).
154
Chapter 10
Applications of Spatially Dispersive Stacks
In this chapter we discuss possible applications for thin-film stacks with high spatial
dispersion. Section 10.1 considers using such stacks for dense wavelength division
multiplexing (DWDM) or coarse wavelength division multiplexing (CWDM). While in
the previous chapters we have targeted a linearly changing beam position as a function of
wavelength, section 10.2 investigates the use of step-like beam shifting. While a linear
beam shift with wavelength corresponds to a Gaussian passband shape, this step-like
beam shifting results in a passband shape that is flat-top. This is interesting for system
considerations. Section 10.3 suggests the use of spatially dispersive stacks for
wavelength-dependent beam steering.
10.1 Wavelength Multiplexing and DemultiplexingIn Chapter 1 we discussed that wavelength division multiplexing (WDM) can be
employed to increase the data capacity of optical fibers. In WDM systems each channel is
encoded with a different optical wavelength. The different wavelength channels are
spatially overlapped using a wavelength multiplexer (MUX) for transmission over a
single fiber as shown schematically in Fig. 1.1. On the receiver side a wavelength
155
demultiplexer (DEMUX) is used to spatially separate the wavelength channels. Today
Dense WDM (DWDM) with typically hundreds of channels is employed in long-haul
systems, and Coarse WDM (CWDM) with approximately four to sixteen channels is used
in metro systems. Here we will first consider the use of the spatial MUX/DEMUX
discussed in this work for DWDM systems and afterwards for CWDM systems.
The International Telecommunications Union (ITU) has standardized a frequency grid for
DWDM. It is centered at 193.1 THz (1552.52 nm) with a channels spacing of 100 GHz
(0.8 nm).1 The center frequency is chosen at 1.55 µm as optical fibers have minimum
propagation loss at this wavelength. Erbium-doped fiber amplifiers (EDFA) can be used
to simultaneously amplify all wavelength channels between approximately 1520 nm to
1570 nm in the optical domain. Therefore, the so-called conventional band (C-band) from
1520 nm to 1570 nm has initially been used for the transmission of up to 64 channels
over a single fiber. Recently, the number of DWDM channels has been increased further
by using smaller channel spacings of 50 GHz or even 25 GHz and by extending the
wavelength range to the short band (S-band) from 1450 nm to 1525 nm and the long band
(L-band) from 1560 nm to 1610 nm. As there is no standard definition for the position of
these bands, all the intervals given are only approximate.
Here we will consider a typical DWDM system with 64 100-GHz channels in the C-band.
Today the most common choice of a multiplexing/demultiplexing device for such a
system is probably an arrayed waveguide grating (AWG). As discussed in Chapter 1
AWGs allow a high channel count at low loss in a compact device. A single AWG can be
used to multiplex or demultiplex all 64 channels, justifying the rather high initial cost of
the AWG. As seen in the experimental results in Chapter 7 as well as the discussion on
the maximum number of expected channels in Chapters 8 and 9, it is not likely that the
spatial MUX discussed in this work can be used to multiplex or demultiplex 64 channels.
A different fabrication technique or device architecture (such as an in-plane device)
would be needed to achieve a sufficient stack thickness and refractive index contrast for
demultiplexing 64 channels. Likely, the thin-film spatial MUX can demultiplex around 8
channels using today’s fabrication technology. Therefore, a single such device is not
156
sufficient for a 64-channel system. As shown below several devices can be combined
though for a 64-channel system.
The main disadvantage of AWGs is that they cannot be upgraded modularly. Therefore, a
company setting up a DWDM system with eight channels to be upgraded to 64 channels
gradually, might choose thin-film filters (TFF) or fiber Bragg gratings (FBG) instead of
an AWG. Both these devices are essentially filters and operate by transmitting one
channels while reflecting all other channels as depicted in Fig. 10.1. Therefore, in order
to multiplex or demultiplex 64 channels, 63 different TFFs or FBGs need to be cascaded
as shown in Fig. 10.2.
Frequency100 GHz
Frequency
Frequency
100 GHzFilter
Fig. 10.1. Operating principle of a frequency filter.
Channel 1
…
100 GHzFilter
64 channels with100 GHz spacing
Channel 2100 GHzFilter
Channel 3100 GHzFilter
100 GHzFilter
…
Fig. 10.2. Demultiplexing architecture based on cascading filters. For 64 channels
63 filters are needed.
157
The architecture as shown in Fig. 10.2 has two main disadvantages. First, all filters need
to be different, as they are targeted for a different wavelength. This increases the
fabrication cost. It is less expensive to fabricate many identical devices. Second, the last
channel has to pass through all the filters and is thus heavily attenuated, while the first
channel is hardly attenuated at all. Both these disadvantages can be alleviated by using a
log2N-architecture, where N is the number of channels.2 This architecture is based on the
use of interleavers instead of filters. As shown in Fig. 10.3, an interleaver separates out
every other channel. Thus, every channel passes through a maximum of log2N stages
distributing the attenuation evenly. An example of a log2N-architecture is given in
Fig. 10.4. Even though this architecture is advantageous compared to the one in Fig. 10.2,
we see that still 63 devices are needed to separate out 64 channels.
Frequency100 GHz
100 GHzInterleaver
Frequency
Frequency200 GHz
200 GHz
Fig. 10.3. Operating principle of a wavelength interleaver.
100 GHzInterleaver
200 GHzInterleaver
200 GHzInterleaver
400 GHzInterleaver
400 GHzInterleaver
800 GHzInterleaver
800 GHzInterleaver
1600 GHzInterleaver
1600 GHzInterleaver
3200 GHzInterleaver
3200 GHzInterleaver
Channel 1
Channel 2
…
…
…
… …
64 channels with100 GHz spacing Channel 3
Channel 4
Fig. 10.4. Demultiplexing architecture based on cascading interleavers. For 64
channels 63 interleavers are needed.
158
100 GHzInterleaver
200 GHzInterleaver
200 GHzInterleaver
400 GHzInterleaver
400 GHzInterleaver
…
…
Channel 1…
800 GHzFilter
Channel 3800 GHzFilter
Channel 5800 GHzFilter
800 GHzFilter
…
64 channels with100 GHz spacing
Channel 2
…
800 GHzFilter
Channel 4800 GHzFilter
Channel 6800 GHzFilter
800 GHzFilter
…
Fig. 10.5. Demultiplexing architecture based on a combination of interleavers and
filters. For 64 channels 63 devices are needed.
As shown in Fig. 10.5, any combination of the architectures introduced in Fig. 10.2 and
Fig. 10.4 may also be used for demultiplexing, but any architecture based on filters or
interleavers will need 63 devices to separate out 64 channels. The reason why any of the
N-channel architectures using filters or interleavers needs N-1 devices is that both these
devices have only two outputs – the reflected and the transmitted beam. The big
advantage of the spatial MUX discussed in this work is that multiple channels can be
demultiplexed using a single device. Instead of being a 3-port device like filters or
interleavers (one input, two outputs), the spatial MUX can, for example, have nine ports
(one input, eight outputs). Using it in an architecture as shown in Fig. 10.6, we see that
only 15 instead of 63 devices are needed for a 64-channel system. Since only 15 devices
are necessary, the optical assembly is less complicated than using filters or interleavers
and the overall module size is reduced. The size of a single spatial MUX is larger though
than the size of a single TFF. Since the number of stages is reduced using a spatial MUX,
the loss may be lowered compared to other architectures. The typical loss for commercial
8-channel 100-GHz multiplexers based on thin-film filters is around 4 to 5 dB. Our
device could probably do better than that.
159
100 GHzInterleaver
200 GHzInterleaver
200 GHzInterleaver
400 GHzInterleaver
400 GHzInterleaver
…
…
64 channels with100 GHz spacing
Channel 3Channel 4Channel 5
SpatialMUXwith
6.2 nmchannelspacing Channel 6
Channel 7Channel 8
Channel 1Channel 2
Channel 3Channel 4Channel 5
SpatialMUXwith
6.2 nmchannelspacing Channel 6
Channel 7Channel 8
Channel 1Channel 2
Fig. 10.6. Demultiplexing architecture based on a combination of interleavers and
spatial MUXes. For 64 channels only 15 devices are needed.
Concluding, we see that the spatial MUX could be used beneficially for DWDM
applications if modularity of the system is desired. Due to the similarity with a TFF, it
can be expected that the cost of a spatial MUX is substantially lower than that of an
AWG, while the simultaneous multiplexing of several channels decreases the module
complexity and cost compared to TFFs of FBGs. The only disadvantage of the spatial
MUX is that it is polarization sensitive. A combination of two spatial MUXs can be used
to spatially shift the two different polarizations separately. One spatial MUX could be
designed to shift the p-polarization and not effect the s-polarization and the second MUX
could work vice versa. The incoming light does not need to be split into the two
polarizations, but can be propagated through the two stacks in series. Alternatively,
multilayer coatings could be deposited on both sides of the substrate, each side affecting
only one polarization.
CWDM systems are just emerging for increasing the data capacity of local or metro
optical networks. CWDM systems have typically four to sixteen channels with channel
spacings around 20 nm in the wavelength ranges around 0.8 µm, 1.3 µm, or 1.55 µm. For
these systems cost is the most important factor. As the cost of multiplexing and
160
demultiplexing devices is distributed over only a couple of channels, the MUX/DEMUX
devices have to be cost-effective both in initial cost as well as in operation. Typically,
TFFs or FBGs are used in an architecture as the one shown in Fig. 10.2. For CWDM
systems the spatial MUX described in this thesis is particularly interesting, as a single
spatial MUX is sufficient for multiplexing or demultiplexing all CWDM channels.
Furthermore, the spatial MUX can be fabricated cost-effectively using well-known thin-
film fabrication technology. As shown in the previous chapters, it is no problem to design
a spatial MUX with a 20-nm channel spacing. A factor important for CWDM that we
have not discussed so far is the passband shape. In the next section we will discuss this
topic and introduce an improved spatial MUX for CWDM applications.
10.2 Step-Like Beam ShiftingIn the previous chapters we have discussed stacks that have a linear shift as a function of
wavelength. These stacks allow for the highest number of channels, as all the different
channels focus on the same output plane. Fig. 10.7 repeats the experimental result for 8
bounces off a 66-layer numerically optimized stack discussed in section 7.4.
Posi
tion
alon
g x-
axis
in µ
m
Out 4
Out 3
Out 2
Out 1
Wavelength in nm
λc1=827.3 nmλc2= 831.0 nm
λc3= 834.7 nmλc4= 840.6 nm
Fig. 10.7. Intensity as a function of position and wavelength for 8 bounces off a 66-
layer numerically optimized stack at 54º incidence angle and p-polarization.
161
Fig. 10.7 also shows the spatial position of the four wavelength channels that can be
spatially demultiplexed using this structure. The field of the channels at 827.3 nm,
831.0 nm, 834.7 nm, and 840.6 nm is spatially separated and these channels are therefore
demultiplexed. Even though the field at 831.0 nm is exactly within the output Out 2, as
soon as the wavelength of the channel drifts to longer or shorter wavelengths, part of the
field is in Out 3 or Out 1 respectively. This leads to loss in Out 2 and to crosstalk in Out 3
or Out 1. Since the beams have a Gaussian beam shape, a linear beam shift as a function
of wavelength corresponds to a Gaussian passband shape as plotted schematically in
Fig. 10.8(a). In a transmission system the laser wavelengths may drift, e.g., due to
temperature changes. In DWDM systems channel wavelengths are normally stabilized
and do not drift much. In CWDM systems on the other hand, the lasers are not stabilized
for cost reasons and may drift significantly. Therefore, it is very important for good
system performance that the multiplexing devices have a flat-top passband shape as
shown in Fig. 10.8(b).
1500 1520 1540 1560 1580
Channel 1Channel 2Channel 3Channel 4
Wavelength in nm1500 1520 1540 1560 1580
Wavelength in nm
Wavelength in nm
0
-30
-60
Tran
smis
sion
in d
B
Wavelength in nm
0
-30
-60
Tran
smis
sion
in d
B
Gaussian passband Flat-top passband
(a) (b)
Fig. 10.8. Schematic of a (a) Gaussian passband shape compared to a (b) flat-top
passband shape (transmission refers to the energy transfer from input to output).
162
A flat-top passband shape allows for laser drift, as there is no loss over a range of
wavelengths not just a single wavelength as in the case of Gaussian passbands. Fig. 10.9
shows that in order to implement a flat-top passband shape for the spatial MUX, we need
to design a structure that has a step-like shift as a function of wavelength. Such a step like
shift results in a range of wavelengths to be shifted to the same exit position. Therefore, if
the output is located at this position, there will be no loss for any wavelength within the
range.
1500 1520 1540 1560 1580
0
50
100
150
200
Desired shiftWavelength in nm
Shift
in u
m a
fter 8
bou
nces
Wavelength in nm
Shift
in µ
m
1500 1520 1540 1560 1580
Channel 1Channel 2Channel 3Channel 4
Wavelength in nm
Wavelength in nm
0
-30
-60
Tran
smis
sion
in d
B
(a) (b)
Fig. 10.9. Flat-top passband shape (a) corresponds to a step-like beam shift with
wavelength (b).
Fig. 10.10(a) demonstrates that we can indeed design a multilayer thin-film stack that has
a step-like shift as a function of wavelength.3 The results are for a 100-layer SiO2/Ta2O5
stack at 45º incidence angle and p-polarization. The stack composition is given in
Appendix G. The stack was designed using the numerical optimization techniques
described in Chapter 7. The last 20 layers were fixed as a quarter-wave dielectric mirror,
while the other 80 layers were varied to approximate the desired step-like shift shown in
Fig. 10.10(a). The reflectance seen in Fig. 10.10(b) could be increased further by adding
more mirror layers. The resulting stack has a total thickness of 33.1 µm.
163
1500 1520 1540 1560 15800
0.2
0.4
0.6
0.8
1
Wavelength in nm
Ref
lect
ance
afte
r 8 b
ounc
es
1500 1520 1540 1560 1580
0
50
100
150
200
DesignDesired shift
Wavelength in nm
Shift
in u
m a
fter 8
bou
nces
Wavelength in nm
Shift
in µ
m a
fter 8
bou
nces
Ref
lect
ance
afte
r 8 b
ounc
es
Wavelength in nm
(a) (b)
Fig. 10.10. Shift (a) and reflectance (b) after 8 bounces off a 100-layer numerically
optimized stack at 45º incidence angle and p-polarization.
Fig. 10.10 proves that we can use the techniques developed in the earlier chapters for a
varied range of dispersion profiles. The resulting 4-channel step-like design with 20-nm
channel spacing is particularly interesting for CWDM applications.
It is also advantageous for cascading multiple devices to prevent a shrinking of the low
loss passband range. In Chapter 8 we discussed that a linear shift as a function of
wavelength results in all beams being focussed on the same output plane. The channels
are also all focussed at the same position if for each channel wavelength and for all
angular components of the incident beam the spatial dispersion has the same constant
value. From Fig. 10.10 we see that the dispersion is zero around the center wavelength of
each channel. Therefore, if the input angular range is sufficiently small, all beams will
focus at the same position. Appendix D discusses that a beam can be seen as a
composition of plane waves with different incidence angles. Only a limited range of these
angular components has a significant intensity. Thus, the input angular range of ±∆θ can
be approximated by the drop of the intensity to 1/e2 as given in (10 - 1).
0wπλ
θ ≈∆ (10 - 1)
164
As shown in Appendix F, a change in the input angle θ is approximately proportional to a
change in frequency ω, since the right hand side of (10 - 2) is approximately constant.
ωβ
θ
θωθω
∂∂
−−=
∆∆
c~sin
~cos~(10 - 2)
Table 10.1 gives three examples of the relationship between spot size, angular range, and
wavelength range. The group velocity along the layers ∂β/∂ω is set to 0.3 times the speed
of light. This is approximately the right value as seen from different investigated
structures in Chapter 4.
Spot Size w0 Angular Range ∆θ Wavelength Range ∆ω
15 µm
30 µm
50 µm
1.9º
0.9º
0.6º
14 nm
7 nm
4 nm
Table 10.1. Relationship between spot size, angular range, and wavelength range at
1540 nm, 45º incidence angle, and ∂ω/∂β=0.3c .
From Table 10.1 we see that for a beam with a spot size of 50 µm, the different beam
components probe approximately ±4 nm of the design around the incident frequency.
Remember that the incident light is assumed to be monochromatic and that this probing is
due to the different angular components. As the design in Fig. 10.10 is quite flat over a
±4 nm wavelength range around the center wavelength, we expect the beams to be
focussed at the same output plane. Towards the edge of the band of zero dispersion, they
will become broadened though. Therefore, not the whole range that is flat for the plane
wave calculation in Fig. 10.10 may be usable. Due to the different angular components, a
beam is distorted if the dispersion is not constant. This again shows why stacks with a
linear shift as a function of wavelength are so interesting. Features of the shift in addition
to linear dispersion will be averaged out depending on the angular range of the beam.
Depending of the shape of the dispersion curve, this averaging can have very distorting
consequences for the beam profile. You can basically imagine that for a strong ripple in
the dispersion, the different beam components are not transferred to the correct position
at the output, resulting in beam distortion. In a very extreme case the ripple of the Bragg
165
stack depicted in section 3.3 corresponds to half of the beam components being shifted,
while the other components are reflected off the front of the stack. This splitting of the
angular content of the beam into two sets, one shifted and one not shifted results in two
separate output beams as discussed in Chapter 3 and broadening of these output beams,
as their angular content is reduced. Careful beam simulations for the Bragg stack show
that the reflected beam is broader below the stop band than within the stop band. A
dispersion profile that has a constant component, and additionally only components that
vary slowly with wavelength, will transfer the input beam to the output with very little
distortion.
1500 1510 1520 1530 1540 1550 1560 1570 1580
25
30
35
40
45
50
Plane WaveBeam 15 umBeam 30 umBeam 50 um
Wavelength in nm
Shift
in u
m a
fter 8
bou
nces
Wavelength in nm
Shift
afte
r one
bou
nce
in µ
m
Fig. 10.11. Shift as a function of wavelength for a plane wave at 45º incidence
angle, and Gaussian beams with 15 µm, 30 µm, and 50 µm spot size.
166
As a 50-µm beam probes a ±4 nm wavelength range, features smaller than this
wavelength range will be averaged out. For a 15-µm beam features smaller than ±14 nm
will be averaged out. Considering this, we cannot expect the design in Fig. 10.10 to work
well with a 15-µm beam. Fig. 10.11 plots the shift as a function of wavelength for plane
waves and three different beam sizes. Indeed we see that for a 15-µm beam the step-like
shift is completely averaged out, while it is well visible for the 30-µm and 50-µm beam
size. This shows that in constructing a spatial MUX with a step-like shift we have to take
care to use a sufficiently large spot size.
800 820 840 860 880 900 9200
5
10
15
20
25
40 deg45 deg50 deg55 deg
Wavelength in nm
Spat
ial S
hift
alon
g x-
dire
ctio
n in
um
Wavelength in nm
Shift
in µ
m
Fig. 10.12. Shift as a function of wavelength for four different incidence angles for
a 66-layer stack.
As the 4-channel step design has not yet been fabricated, we conclude this section with
some experimental results obtained from the 66-layer stack discussed in Chapter 7. In
Chapter 7 we used the stack at 54º incidence angle and observed a linear shift as a
function of wavelength. By luck (and consideration, of course) we found that the same
design can serve as a 3-channel step design. As a change in incidence angle and a change
in wavelength are only approximately proportional, we observe a change in the shift
profile as well as the wavelength position if we change the incidence angle. Fig. 10.12
167
plots the shift as a function of wavelength for four different incidence angles. At 55º we
see the linear shift, but at around 50º the shift appears step-like.
Indeed, performing the experiment we clearly see in Fig. 10.13(a) the step-like spatial
shift at an incidence angle of 48º demultiplexing three channels.3 The intensity at each
wavelength is normalized to unity, as this structure was not designed for high reflectance
at 48º and shows significant loss. Nevertheless, the agreement between the
experimentally observed shift and the theoretically calculated shift as seen in
Fig. 10.13(b) is promising.
Wavelength in nm
Posi
tion
in µ
m
830 840 850 860 870 88050
0
50
100
150
200
ExperimentTheory
Wavelength in nm
Shift
in u
m
Wavelength in nm
Shift
in µ
mOut 3
Out 2
Out 1
(a) (b)
Fig. 10.13. (a) Intensity as a function of position and wavelength for 8 bounces off
a 66-layer numerically optimized stack at 48º incidence angle and p-polarization.
(b) Comparison between experiment and plane wave theory.
10.3 Temporal vs. Spatial DispersionAs discussed in Chapter 4, stacks with spatial dispersion exhibit temporal dispersion at
the same time. For short pulse operation, it may be important to eliminate this temporal
dispersion. Fig. 10.14(a) shows how a combination of two stacks with opposite
dispersion can be used in series to cancel the temporal dispersion, while doubling the
spatial dispersion. On the other hand, backreflection through the same stack can be used
168
if a system with temporal dispersion, but no spatial dispersion, is desired as shown in
Fig. 10.14(b). Replacing the mirror by a modulator array or another optical element, the
different wavelength components can be modulated independently. This system can
easily be extended to a 2-D array by using the dimension out of the plane of the page. In
fact, all of the designs here have the advantage that multiple beams in the y-direction can
be multiplexed in parallel – a potentially significant benefit.
Stack 2Stack 1 Stack 1 Mirror,modulatorarray, or otheroptical element
(a) (b)
Fig. 10.14. Systems using stacks with high spatial dispersion. (a) Combination of
two stacks with opposite dispersion can be used to obtain spatial dispersion without
temporal dispersion. (b) System providing temporal dispersion without spatial
dispersion. This system could also be used to manipulate channels of different
wavelengths independently.
10.4 Beam SteeringAnother potential application for stacks with high spatial dispersion is wavelength-
dependent beam steering. Beams of different wavelength leave the device at different exit
positions, but propagate in parallel contrary to the case of a prism or a grating. The spatial
dispersion of the stack may be converted to an angular dispersion by use of a simple lens
as shown in Fig. 10.15. From Fig. 10.15 we see that the dispersive stack could therefore
be used for wavelength-dependent angular beam steering, i.e. by changing the
wavelength of the beam, the propagation direction is changed. Since the stack can be
designed to exhibit a large spatial shift over a narrow wavelength range, in principle very
little tuning of the laser wavelength may be necessary. Thus, we obtain a beam steering
device that is both compact and can be tuned rapidly, as it is tuned electrically not
mechanically. The experimental setup discussed in section 3.2 uses this beam steering
169
effect as the focal plane is imagined onto a CCD camera approximately 1m away. As
long as there is no beam clipping on the front side gold coating, the focal plane does not
have to be exactly at the output plane, since the device is linear. The same is true for
alignment along the x-direction. Since the y-direction anyway has no influence on the
device operation, the device is therefore tolerant to an offset in any direction.
Furthermore, since a change in the incident angle has, to first order, an equivalent effect
to a change in the incident wavelength, a deviation from the specified incidence angle is
also not critical. It can be corrected by wavelength tuning without resulting in loss. The
high tolerance to alignment errors is a big advantage of this type of device, compared,
e.g., to waveguide devices that have to fulfill mode-matching criteria to prevent loss.
zy
x
Lens
Fig. 10.15. Wavelength-dependent beam steering.
10.5 Other Applications Using Beam ShiftingIn contrast to thin-film filters or fiber Bragg gratings, the stacks with spatial dispersion
discussed here have continuous dispersion. Therefore, their application is not limited to
discrete channels. As seen in section 10.3 the spatial dispersion of the device can be
converted to an angular dispersion using a lens and the continuous spatial shift is
170
converted to a continuous angular sweep. Therefore, this device can in principal replace
prisms or gratings used to obtain angular dispersion in many applications including many
other spectrometer applications. The advantage of stacks with spatial dispersion is that
their dispersion profile can be designed to best suit the application. In this work we have
seen that a wide range of dispersion profiles can be obtained using the design techniques
described in Chapters 5 to 7. This unprecedented flexibility in designing the spatial
dispersion properties is interesting for existing systems and might lead the way to other
novel systems.
References
[1] R. Ramaswami and K. N. Sivarajan, Optical Networks: A Practical Perspective,
Morgan Kaufmann Publishers, San Francisco, CA (1998).
[2] C.K. Madsen and J.H. Zhao, Optical Filter Design and Analysis - A Signal
Processing Approach, John Wiley & Sons, Inc. (1999).
[3] M. Gerken et al, IEEE LEOS 2002 Annual Meeting, Glasgow, Scotland (November
10-14, 2002). Paper ThV 3.
171
Chapter 11
Conclusions
This work investigated multilayer thin-film stacks with high spatial dispersion in terms of
the physical origin of spatial dispersion, methods for designing stacks with spatial
dispersion, and applications of stacks with spatial dispersion. We found that the spatial
shift a beam experiences upon reflection off a multilayer stack is proportional to the
stored energy within the stack. Therefore, designing a stack with spatial dispersion is
equivalent to designing a structure that has a wavelength-dependent amount of stored
energy. Four different types of structures that have a wavelength-dependent amount of
stored energy and thus exhibit spatial dispersion have been discussed – one-dimensional
photonic crystal structures, chirped stacks, resonator stacks, and numerically optimized
stacks.
Periodic Bragg stacks are one-dimensional photonic crystals and exhibit the superprism
effect, i.e. a spatial beam shifting as a function of wavelength, just outside a reflection
band. Unfortunately, periodic structures with two layers per period do not offer enough
degrees of freedom to design structures with desired dispersion characteristics, e.g., a
linear shift with wavelength. One way to overcome this limitation is to increase the
172
number of layers per period. This method should be used if a structure with constant
effective properties is desired. Here, on the other hand, we were more interested in the
aggregate transfer function from entering to exiting the structure. That is we wanted to
design a structure exhibiting a spatial beam shift along the exit surface with wavelength.
In that case, it does not matter what happens to the beam within the structure. Taking this
into account, we are no longer limited to periodic structures.
Three different algorithms were developed for designing the spatial dispersion
characteristics of non-periodic stacks. Considering that spatial and temporal dispersion
are approximately proportional, we modified two analytical methods used for temporal
dispersion compensation to obtain spatial dispersion. We showed that a coupled-mode
theory approach could be modified to design double-chirped structures with spatial
dispersion over a broad wavelength range. Narrowband coupled-cavity structures can be
designed using digital lattice techniques. Both these techniques can be used to design, for
example, stacks exhibiting a linear shift as a function of wavelength. Depending on the
desired wavelength range the best algorithm is chosen. Both chirped stacks and resonator
stacks are still limited to a specific stack configuration though. Chirped stacks always
have a monotonically increasing period length, while resonator stacks consist of cavities
separated by reflectors. This limits the available degrees of freedom and explains why
chirped stacks result in useful broadband designs, but are not good for narrowband
designs, and resonator stacks vice versa.
To make full use of all available degrees of freedom, we should not limit ourselves to a
particular type of stack, but consider all possible stack configurations. The problem in
this case is to find the best configuration out of all possibilities. For more than a couple of
layers, not all possible stack configurations can be investigated due to the prohibitively
large number of parameters. This is a problem well known from the design of thin-film
filters with desired reflectance and transmittance characteristics. As in the case of thin-
film filter design, numerical optimization methods can be used to find a good design,
even though this may not be the “best” possible design. Since numerically optimized
thin-film stacks are not limited to a particular stack configuration, more complex
173
dispersion profiles can be obtained than in the case of chirped or resonator stacks. The
drawback is the increased computation time necessary for the numerical optimization
procedure compared to the analytical design used for chirped and resonator stacks. In
general we conclude that a larger number of degrees of freedom leads to more design
flexibility and therefore superior dispersion characteristics.
We discussed in detail the use of multilayer stacks with spatial dispersion as wavelength
multiplexing and demultiplexing devices. Considering a device geometry with the
dispersive stack on one side of the substrate and a reflective gold coating on the other
side, several bounces can be performed through the substrate. Each bounce off the stack
increases the spatial beam shift and therefore the spatial separation between different
wavelength channels. Gaussian beam broadening limits the number of bounces. We find
that the largest number of channels is obtained if the stack has constant dispersion, since
in this case all beams focus on the output plane.
The number of channels that can be demultiplexed with a given stack is proportional to
the total shift obtained with one bounce off the stack. Therefore, the total shift of
different stacks should be compared to judge their performance. The dispersion is not a
good figure of merit, since the total number of channels is proportional to the dispersion
multiplied by the wavelength range over which the dispersion is observed. Thus, low
dispersion over a broad wavelength range may be better than high dispersion over a very
narrow wavelength range. The dispersion×wavelength-range-product is equal to the total
shift. Investigating over 600 different structures, we found that there seems to be a
physical limit to how much shift can be obtained for a given stack thickness, material
system, and incidence angle. The total shift appears proportional to the stack thickness,
the refractive index contrast, and the sine of the incident angle, while it is inversely
proportional to the square of the average refractive index. Interestingly, all three design
methods for non-periodic stacks result in a similar amount of total shift that is close to the
observed physical limit.
174
Thin-film stacks are not limited in their use to wavelength multiplexing or demultiplexing
components in WDM systems. As they have a continuous dispersion, dispersive thin-film
stacks could for example replace prisms or gratings in applications requiring a
wavelength-dependent element. In contrast to prisms or gratings, their dispersion
characteristics can be freely designed, which is a potential benefit for current applications
and may pave the way for new system architectures. The spatial dispersion of multilayer
stacks can also be converted to an angular dispersion by a lens in sequence with the stack.
Therefore, the device can, for example, be used for wavelength-dependent beam steering.
Future integrated photonic systems might use the concepts discussed here to obtain
spatial dispersion. The ideas considered here for one-dimensional thin-film structures can
also be transferred to two- or three-dimensional structures. Two-dimensional photonic
crystal structures fabricated by lithography, for example, are by no means limited to
periodic structures. As seen here for one-dimensional structures, breaking the periodicity
actually offers more design freedom and interesting new physics. The difficult task is to
design two- or three-dimensional structures with the desired characteristics. The methods
discussed here for designing non-periodic stacks may be a good starting point for
investigating the design of higher dimensional systems.
175
Appendix A
Bloch Calculation
For an infinite periodic medium with two layers per period the dispersion relation is
given by (A - 1).1 In this equation, K is the amplitude of the wavevector in the z-direction
of the periodicity, while β is the amplitude of the wavevector parallel to the layers in the
x-direction. c in the speed of light in vacuum and ω is the angular frequency of the light.
na and nb are the refractive indices, and la and are lb the layer thicknesses of the two
different materials as shown in Fig. A.1.
na nb na nb... na
la lb
Λz
y
x...nb
Fig. A.1. Labeling of periodic stack.
176
( )
( )
−
−
∆−
−
−
−
=+
bbaa
bbaaba
lnc
lnc
lnc
lnc
llK
22
22
22
22
sinsin,
coscos)(cos
βω
βω
ωβ
βω
βω
(A - 1)
β is calculated by (A - 2) and is the same in all layers. nin is the refractive index of the
incidence medium and θin is the incidence angle of the light.
( )ininnc
θω
β sin= (A - 2)
∆(β,ω) is given by (A - 3) for TE-polarization and by (A - 4) for TM-polarization.
( )
−
−
+
−
−
=∆2
2
22
22
22
21,
βω
βω
βω
βω
ωβ
b
a
a
b
TE
nc
nc
nc
nc
(A - 3)
( )
−
−
+
−
−
=∆2
22
22
2
22
2
22
2
21,
βω
βω
βω
βω
ωβ
ba
ab
ab
ba
TM
nc
n
nc
n
nc
n
nc
n(A - 4)
For plotting a K-β wavevector diagram or other calculations requiring the absolute value
of K, the dispersion relation (A - 1) has to be solved for K as a function of ω and β. This
is problematic due to the periodicity of the cosine. To determine the absolute value of K,
we have to consider carefully how many times π should be added. This can be done by
first calculating K far way from any stopband using the angle obtained from Snell’s law
as the group propagation angle. Slowly approaching the stopband, the correct number of
π is obtained by guaranteeing that there are no discontinuities in K. For many calculations
though, only the derivative of K with respect to ω or β is of interest and the constant term
can be neglected.
177
The Bloch calculation can also be used to model the dispersion properties of a finite bulk
periodic medium neglecting all boundary effects.
References
[1] A. Yariv and P. Yeh, “Electromagnetic propagation in periodic stratified media. II.
Birefringence, phase matching, and x-ray lasers,” J. Opt. Soc. Am., 67/4, 438-448
(1977).
178
Appendix B
Coordinate Transformation (K, β, ω)→(K, θ, ω)
In this appendix, the details of the coordinate transformation from variables (K, β, ω) to
(K, θ, ω) are given, where K is the wavevector in the z-direction perpendicular to the
layers, β is the wavevector in the x-direction along the layers, ω is the frequency, and θ is
the incidence angle in vacuum. Two main equations often used in this thesis are (B - 1)
and (B - 2) for calculating the group velocities vgx in the x-direction and vgz in the z-
direction.
constKgxv
=∂∂
=βω (B - 1)
constgz K
v=∂
∂=
β
ω (B - 2)
The calculation of these equations is straigthforward using the variables (K, β, ω), since it
is clear how to hold the third variable constant when taking the partial derivatives. On the
other hand, we often want to know how the group velocities change with θ and ω. As β
is the same in all layers, we can easily calculate it from (B - 3).
179
( ) ( )θω
ωθβ sin,c
= (B - 3)
c is the speed of light in vacuum. Remember that θ is in vacuum and the corresponding
refractive index is therefore unity. K(θ, ω) is calculated by replacing the independent
variables (β, ω) by (θ, ω) in the dispersion relation. It is then straightforward to plot
K-β-wavevector diagrams1,2 using K(θ, ω) and β(θ, ω) as parametric functions and
matching pairs. It is much more challenging though to calculate the partial derivatives
(B - 1) and (B - 2) correctly for given values of θ and ω. As the coordinate transform
from (K, β, ω) to (K, θ, ω) has caused quite some confusion and calculation mistakes, I
am giving my derivation for calculating the group propagation angle in detail here.
Section B.1 shows how to calculate the group propagation angle using equations in terms
of (K, β, ω). Section B.2 looks at the case using the independent variables (K, θ, ω).
B.1 Group Propagation Angle in Terms of (K, β, ω)First let us assume that we have the dispersion relation given as an implicit function
relating (K, β, ω) as shown in (B - 4). This relationship can, e.g., be obtained by
subtracting one side of (A - 1) from the other side.
0),,( =ωβKf (B - 4)
Now we take the derivative of f(K,β,ω) with respect to ω as shown in (B - 5).3,4
( ) ( ) ( ) 0,,,,,,=
∂∂
∂∂
+∂∂
∂∂
+∂
∂ωβ
βωβ
ωωβ
ωωβ KfK
KKfKf (B - 5)
Partial derivatives (indicated by ∂) always assume that all other variables are held
constant, e.g., ∂f(K,β,ω)/∂ω assumes that K and β are held constant. The constant
variables are sometimes indicated after a vertical line as shown in (B - 1) and (B - 2).
This is not necessary if we know what the variables of the function are. It is very helpful
though, if equations are shown without a context. If (B - 2) were given without the
clarification β=const, we might think that we could take the partial derivative of our
180
function K(θ,ω) with respect to ω to obtain the group velocity in the z-direction. But in
fact, this would give a completely wrong result, since we held θ constant instead of β.
To calculate ∂β/∂ω, we set K constant in (B - 5) and obtain (B - 6).1 Then (B - 1) is just
the reciprocal of (B - 6).
( )( ) βωβ
ωωβωβ
∂∂∂∂−
=∂∂
= ,,,,
KfKf
constK
(B - 6)
Note here that we have explicitly justified that ( )constKconstK ==
∂∂=∂∂ ωββω 1 .
Similarly, ∂Κ/∂ω in (B - 7) is obtained by setting β constant in (B - 5). (B - 2) is the
reciprocal of (B - 7).
( )( ) KKf
KfK
const ∂∂∂∂−
=∂∂
= ωβωωβ
ω β ,,,, (B - 7)
Finally, the group propagation angle θgr is calculated by (B - 8).
( )
( )( )( )
( )
( )( )
∂∂∂∂
=
∂∂∂∂−
∂∂∂∂−
=
=
−−
−
KKfKf
KfKf
KKfKf
vv
Kgz
gxgr
ωββωβ
βωβωωβ
ωβωωβ
ωβθ
,,,,tan
,,,,
,,,,
tan
tan,,
11
1
(B - 8)
Solving f(K,β,ω) for K, we can rewrite (B - 4) as given in (B - 9).
0),(),,( =−= ωβωβ gKKf (B - 9)
The function g(β,ω) only depends on β and ω. In the case of a periodic stack as discussed
in Appendix A, this separation is for example achieved by taking the inverse cosine of
(A - 1). Now we can calculate the partial derivatives in (B - 8) as shown in (B - 10) and
(B - 11).
( ) ( )β
ωββ
ωβ∂
∂−=
∂∂ ,,, gKf (B - 10)
181
( ) 1,,=
∂∂
KKf ωβ (B - 11)
Substituting (B - 10) and (B - 11) into (B - 8) and writing g(β,ω)≡K(β,ω) since
g(β,ω)=K(β,ω) from (B - 9) we obtain (B - 12).
( ) ( )
∂
∂−= −
βωβ
ωβθ,tan, 1 K
group (B - 12)
Thus, if we have solved the dispersion relation to obtain K(β,ω) we can use (B - 12) to
calculate the group propagation angle θgr.
B.2 Group Propagation Angle in Terms of (K, θ, ω)In the last section we discussed the calculation of the group propagation angle θgr in terms
of the independent variables (K, β, ω). Here we will perform the coordinate
transformation to (K, θ, ω). We assume that K(β,ω) is given and we want to transform it
to θ and ω as shown in (B - 13).
( ) ( )ωθωβ ~,~~, KK → (B - 13)
To distinguish the new variables from the old ones, the new variables are given with a
tilde on top. The quantities K and K~ will be identical at the end of any calculation in a
given physical situation, but K and K~ are functions of different variables, and similarly
for θ, θ~ and ω, ω~ . The new independent variables are calculated in terms of the old
variables as given in (B - 14) and (B - 15).
( )
= −
ωβ
ωβθc1sin,~ (B - 14)
( ) ωωβω =,~ (B - 15)
We use the chain rule to calculate the group propagation angle θgr after the coordinate
transformation.4 Thus, expression (B - 12) for the group velocity angle becomes (B - 16).
182
( ) ( )( )
( )( )
∂∂
∂∂
+∂
∂∂
∂−= −
ωθβω
ωωθ
ωθβθ
θωθ
ωθθ ~,~~
~~,~~
~,~~
~~,~~
tan~,~~ 1 KKgroup (B - 16)
From (B - 15) it is clear that the second term in the parantheses is zero. Thus the group
propagation angle is given by (B - 17).
( ) ( )( )
( )( )
∂∂
−=
∂∂
∂∂
−= −−
θωθωθ
ωθβθ
θωθ
ωθθ ~cos~~~,~~
tan~,~~
~~,~~
tan~,~~ 11 cKKgroup (B - 17)
We see that we actually have to take the derivative of K with respect to θ not to ω as we
might first expect after looking at (B - 1) and (B - 2).
References
[1] A. Yariv and P. Yeh, “Electromagnetic propagation in periodic stratified media. II.
Birefringence, phase matching, and x-ray lasers,” J. Opt. Soc. Am., 67/4, 438-448
(1977).
[2] B. E. Nelson, M. Gerken, D. A. B. Miller, R. Piestun, C.-C. Lin, J. S. Harris, Jr.,
“Use of a dielectric stack as a one-dimensional photonic crystal for wavelength
demultiplexing by beam shifting,” Opt. Lett., 25/20, 1502-1504 (2000).
[3] I.N. Bronstein, K.A. Semendjajew, G. Musiol, H. Muehlig, Taschenbuch der
Mathematik; Verlag Harri Deutsch, Thun und Frankfurt am Main, pp. 232-237
(1993).
[4] G. Merziger and T. Wirth, Repetitorium der Höheren Mathematik, Binomi,
Springe, pp. 389-391, p.531 (1993).
183
Appendix C
Transfer Matrix Calculation
Here we explain how to calculate the propagation of plane waves through arbitrary
multilayer stacks using a transfer matrix approach.1 Appendix D discusses the
propagation of beams of light. The propagation of electromagnetic waves is governed by
Maxwell’s equations given in (C - 1) to (C - 4).1,2,3
( ) ( )tt ,, rBrE &−=×∇ (C - 1)
( ) ( ) ( )ttt ,,, rDrJrH &+=×∇ (C - 2)
( ) ( )tt ,, rrD ρ=∇ (C - 3)
( ) 0, =∇ trB (C - 4)
E is the electric field strength, D is the electric displacement, J is the electric current
density, H is the magnetic field strength, B is the magnetic flux density, and ρ is the
electric charge density. r indicates the position in space and t the time. Bold letters
indicate vectors and matrices, italic letters are scalars. Assuming a source-free medium,
the electric charge density is zero everywhere and there are no source currents.
184
Furthermore, we assume simple media,3 i.e., linear, time-invariant, isotropic and
homogeneous media. In this case, the material equations are given by (C - 5) to (C - 7),
with ε, µ, and κ being simple scalars.
( ) ( ) ( )tt ,, rErrD ε= (C - 5)
( ) ( ) ( )tt ,1, rBr
rHµ
= (C - 6)
( ) ( ) ( )tt ,, rErrJ κ= (C - 7)
Taking the curl of (C - 1) and (C - 2) and substituting the other two Maxwell’s equations
as well as the material equations, we obtain (C - 8) and (C - 9).
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )tttt ,,,, rErrrErrrBrE &&&& εµκµ −−=×−∇=×∇×∇ (C - 8)
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )ttttt ,,,,, rHrrrHrrrDrJrH &&&& εµκµ −−=+×∇=×∇×∇ (C - 9)
With the vector identity (C - 10)
( ) AAA 2∇−∇∇=×∇×∇ (C - 10)
as well as (C - 3), (C - 4), and ρ =0, (C - 8) and (C - 9) are rewritten as (C - 11) and
(C - 12).
( ) ( ) ( ) ( ) ( ) ( ) ( )ttt ,,,2 rErrrErrrE &&& εµκµ +=∇ (C - 11)
( ) ( ) ( ) ( ) ( ) ( ) ( )ttt ,,,2 rHrrrHrrrH &&& εµκµ +=∇ (C - 12)
Equations (C - 11) and (C - 12) are called the telegraph equations or wave equations.
Equations (C - 11) and (C - 12) are not independent. Solving for the E-field, the H-field
can be obtained from (C - 2) and vice versa. Therefore, we limit ourselves to solving for
the E-field and relate it to the H-field at the end.
As we are interested in the propagation of light, we now consider waves with harmonic
time dependencies. Such waves are most easily dealt with using vector phasors as given
185
in (C - 13). Remember, though, that the real part of the phasor represents the actual
physical field.
( ) ( ) ( )tt ωjexp, rErE = (C - 13)
Substituting (C - 13) into (C - 11), we obtain (C - 14).
( ) ( ) ( ) ( ) ( ) ( ) 0,j, 22 =−+∇ tt rErrrrrE ωκµωεµ (C - 14)
The amplitude of the complex wavevector k is defined by (C - 15).
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )cn
ck r
rrrrrrrrrrrr ω
εωκµ
εµω
ωκµωεµ =−=−=0
2 jj (C - 15)
In this reformulation, c is the speed of light in vacuum and n is the complex refractive
index. The wavevector itself is defined pointing in the direction of propagation, which is
indicated by the unitvector ek in (C - 16).
( ) ( ) ( )rkerrk k= (C - 16)
We obtain the differential equation (C - 17) for the E-field.
( ) ( ) ( ) 0,, 22 =+∇ tkt rErrE (C - 17)
For the case that k is not a function of r, (C - 17) has two solutions, a forward
propagating plane wave given in (C - 18) and a backward propagating plane wave given
in (C - 19). The E-field is pointing perpendicular to the direction of propagation for both
solutions.
( ) ( )[ ] kekrrE ⊥−= tjEt ff ωexp, (C - 18)
( ) ( )[ ] kekrrE ⊥+= tjEt bb ωexp, (C - 19)
These solutions are uniform plane waves, as their amplitude does not change along the
planes of constant phase.
Now we discuss the case of a multilayered stack as the one shown in Fig. C.1. Within
each layer i the wavevector ki is constant and (C - 18) and (C - 19) are therefore solutions
describing the propagation of light in that layer. We will call the forward propagating
186
amplitude in layer i Ef,i and the backward propagating amplitude Eb,i. As shown in
Fig. C.1 the coordinate system is oriented such that the layer interfaces are perpendicular
to the z-axis. Without any loss of generality, we now assume that the propagation
direction of the light lies in the x-z-plane and there is thus no y-dependency of the fields.
z
x
n0 n1 n2 n3 nN+1nNnN-1
…
d1 d2 d3 dNdN-1 E||b,N+1=0
E||f,N+1=Etrans=1
E||f,0= Einc
E||b,0= Erefl
E||b,1
E||f,1E||b,2
E||f,2E||b,N
E||f,NE||b,3
E||f,3
z1=0
z2 z3 z4 zN-1 zN zN+1=L
Fig. C.1. Naming conventions for a multilayer stack.
ni ni+1
kf,i
z
x
kb,i
Ef,i
Eb,i
kf,i+1
Ef,i+1
Hf,i
kb,i+1
Eb,i+1
Hb,i+1
Hb,i
Hf,i+1
p-polarization (TM)
θi θi+1
ni ni+1
kf,i
z
x
kb,ikf,i+1
Ef,i
Eb,i
Ef,i+1
Hf,i
Hb,i
Hf,i+1
s-polarization (TE)
θi
kb,i+1
Eb,i+1
Hb,i+1θi+1
Vector pointing out of pageVector pointing into page
Fig. C.2. Conventions for the positive directions of k, E, and H in the case of p-
polarization (TM) and s-polarization (TE).
The E-field can have components in any direction perpendicular to the propagation
direction. The mathematics is most easily performed by splitting the field into the
components in the plane of incidence (p-polarization, TM) and perpendicular to the plane
of incidence (s-polarization, TE). Since, we are only considering linear media here, the
187
two components can be treated separately and the total field is obtained in the end by
adding the two components again. The H-field is perpendicular to both the E-field and
the direction of propagation. Fig. C.2 gives the conventions for the positive directions for
k, E, and H for both TE- and TM-polarization.1 Any other convention could be chosen
equally well as long as the same convention is used throughout.4
The forward and backward propagating E-field in layer i are related to each other and to
the forward and backward propagating E-field in layer i+1 by the boundary conditions at
the interface between i and i+1. Following Maxwell’s equations, the tangential
components of the E-field and the H-field have to be continuous across the interface.3
This condition can only be fulfilled over the whole interface if the wavevector component
along the interface is constant. This constant wavevector component is called β. As we
confined the wavevector to the x-z-plane, β points in the x-direction and only the z-
component changes as a function of the position. As β is the same everywhere in the
structure, it is convenient to cast equations in terms of β. β can easily be calculated by
(C - 20)
( )iinc
θω
β sin= (C - 20)
using any known refractive index and propagation angle combination. Normally, the
refractive index n and the propagation angle θ of the incident medium or vacuum are
used. Snell’s law given in (C - 21)
( ) ( )11 sinsin ++= iiii nn θθ (C - 21)
follows immediately from the fact that β is constant across all interfaces. As discussed
previously, we will perform all calculations for the E-field. The H-field amplitude is
calculated from the E-field amplitude by (C - 22).
( ) ( )( )
( ) ( )0ZnE
ZEH rr
rrr == (C - 22)
In this equation Z(r) is the impedance at position r, Z0=376.7Ω is the impedance in
vacuum, and n(r) the refractive index at position r. The direction of the H-field can be
188
determined from Fig. C.2. Since the E-field component E|| parallel to the interface is
constant across the interface, we express the total E-fields in terms of E||. (C - 23) to
(C - 26) give the equations for the forward and backward propagating E- and H-fields in a
multilayer stack for the case of TM-polarization. Ex,f,i and Ex,b,i are the E-field
components parallel to the interface at the left interface position zi as shown in Fig. C.1.
They are complex numbers with the phase representing the phase of the E-field at
position zi. The index i is determined from the position z in the stack.
( ) ( ) ( )( )[ ] ( ) ( ) zx eeE iiiizi
ifxfTM zzkxtj
Etzx θθβω
θsincosexp
cos,, ,
,, −−−−= (C - 23)
( ) ( ) ( )( )[ ] ( ) ( ) zx eeE iiiizi
ibxbTM zzkxtj
Etzx θθβω
θsincosexp
cos,, ,
,, +−+−= (C - 24)
( ) ( ) ( )( )[ ] yeH iizii
ifxfTM zzkxtj
ZE
tzx −−−= ,,
, expcos
,, βωθ
(C - 25)
( ) ( ) ( )( )[ ] ( )yeH −−+−= iizii
ibxbTM zzkxtj
ZE
tzx ,,
, expcos
,, βωθ
(C - 26)
The total E-field and H-field are calculated by adding the forward and the backward
propagating fields. As the H-field has only a y-component, it is easiest to calculate the
total H-field and obtain the E-field from (C - 22). The total H-field is given in (C - 27).
( ) ( ) ( )( )[ ]( ) ( )[ ] ( )[ ] ye
HHH
iizibxiizifxii
bTMfTMTM
zzkjEzzkjEZ
xtj
tzxtzxtzx
−−−−−
=
=+=
,,,,,,
,,
expexpcos
exp
,,,,,,
θβω (C - 27)
We see from (C - 27) that the H- and E-field are separable into a part that only depends
on x and one that only depends on z. Thus, we can consider the x-dependency and the z-
dependency separately and multiply the results. This is NOT the same though as
calculating the field for normal incidence and multiplying it by the part with x-
dependency. This approach would not fulfill the boundary conditions. Ex,f,i and Ex,b,i
depend on the angle of incidence and the frequency as seen below.
189
For the case of TE-polarization Ey,f,i and Ey,b,i are the E-field components parallel to the
interface. (C - 28) to (C - 31) give the equations for the forward and backward
propagating E- and H-fields for TE-polarization.
( ) ( )( )[ ] yeE iizifyfTE zzkxtjEtzx −−−= ,,, exp,, βω (C - 28)
( ) ( )( )[ ] yeE iizibybTE zzkxtjEtzx −+−= ,,, exp,, βω (C - 29)
( ) ( )( )[ ] ( ) ( ) zx eeH iiiizi
ifyfTE zzkxtj
ZE
tzx θθβω sincosexp,, ,,
, +−−−−= (C - 30)
( ) ( )( )[ ] ( ) ( ) zx eeH iiiizi
ibybTE zzkxtj
ZE
tzx θθβω sincosexp,, ,,
, +−+−= (C - 31)
Again, the total E-field is calculated by adding the forward and backward propagating
fields as shown in (C - 32).
( ) ( ) ( )( )[ ] ( )[ ] ( )[ ] ye
EEE
iizibyiizify
bTEfTETE
zzkjEzzkjExtjtzxtzxtzx
−+−−−=
=+=
,,,,,,
,,
expexpexp,,,,,,
βω(C - 32)
Equations (C - 27) and (C - 32) together with (C - 22) allow us to calculate the E-field
and the H-field at any position in the multilayer stack assuming that the fields parallel to
the interface E||,f,i and E||,b,i are known for each layer i. Next we discuss how to calculate
these fields using using a standard transfer matrix method.1 Assuming that no light is
incident from the right side onto the N-layer stack shown in Fig. C – 1 and that the
transmitted field has unity magnitude, the incident and reflected amplitudes on the left
side of the stack can be calculated using (C - 33).
=
−− 0
1EE
,113,222,111,0refl||,
inc||,NNN DPDPDPD K (C - 33)
In (C - 33) Pi and Di,i+1 are the propagation and interface matrices as obtained from the
interface boundary conditions from Maxwell’s equations. Pi(β,ω) is given in (C - 34)
with ni being the refractive index of layer i, di the layer thickness, β is the amplitude of
190
the wavevector parallel to the layers in the x-direction, ω is the angular frequency of the
light, and c the speed of light in vacuum.
( )
−
−
−
=
ii
ii
i
dnc
j
dnc
j
22
22
exp0
0exp
,
βω
βω
ωβP (C - 34)
The interface matrices Di,i+1(β,ω) given in (C - 35) depend on the reflection coefficients
ri,i+1(β,ω) and the transmission coefficients ti,i+1(β,ω) between adjacent layers which are
given by equations (C - 36) and (C - 37). These formulae differ from the Fresnel
formulae, as these are the reflection and transmission coefficients for the component of
the field parallel to the boundary, not for the total field.1,5
( ) ( )( )
( )
=
+
+
++ 1,
,1,
1,1,
1,
1,1, ωβ
ωβ
ωβωβ
ii
ii
iiii r
rt
D (C - 35)
( ) ( ) ( )( ) ( )ωβωβ
ωβωβωβ
,,,,
,1,,
1,,1,
+
++ +
−=
ieffieff
ieffieffii nn
nnr (C - 36)
( ) ( )( ) ( )ωβωβ
ωβωβ
,,,2
,1,,
,1,
++ +
=ieffieff
ieffii nn
nt (C - 37)
The effective refractive indices for the cases of TE- and TM-polarization are given by
(C - 38) and (C - 39) respectively.1
( )2
,, 1,
−=
iiiTEeff n
cnnω
βωβ (C - 38)
( )2,,
1
,
−
=
i
iiTMeff
nc
nn
ωβ
ωβ (C - 39)
If the expression under the square root becomes negative as, e.g., in the case of total
internal reflection, the sign of the effective refractive indices has to be chosen carefully to
191
prevent unphysical exponentially growing field amplitudes, though the method does then
model such situations correctly.
The fields parallel to the interface E\\,f,i and E\\,b,i in intermediate layers i are calculated by
just performing a subset of the matrix multiplications in (C - 33) as shown in (C - 40).
=
−−++++ 0
1EE
,112,111,ib,||,
if,||,NNNiiiiii DPDPDP K (C - 40)
Alternatively, the intermediate fields can be stored while performing calculation (C - 33)
from right to left. Now the E-field and the H-field at any position in the multilayer stack
can be calculated using equations (C - 27) and (C - 32) together with (C - 22). The
electromagnetic energy density w is given by (C - 41).3
( ) ( ) ( )( ) ( ) ( )( )
( ) ( )( ) ( ) ( )( )22
22
,,Re,,Re
,,Re21,,Re
21,,
tzxHztzxEz
tzxHztzxEztzxw
µε
µε
==
+=(C - 41)
Normally, we are not interested in the instantaneous variations of the energy density, but
instead in the time-averaged values. The time-averaged energy density wavg is given by
(C - 42). Note that the extra factor of ½ is due to the time averaging of a harmonic wave.
( ) ( ) ( ) ( ) ( ) 22 ,21,
21, zxHzzxEzzxwavg µε == (C - 42)
In the case of TM-polarization, wavg is given by (C - 43), where we have used (C - 27) for
the total H-field.
( )( )
( )[ ] ( )[ ] 2
,,,,,,20
2
, expexpcos2
1, iizibxiizifxi
iTMavg zzkjEzzkjEnzxw −−−−=
θε (C - 43)
For TE-polarization (C - 44) is obtained by substituting the total E-field from (C - 32)
into (C - 42).
( ) ( )[ ] ( )[ ] 2
,,,,,,02
, expexp21, iizibyiizifyiTEavg zzkjEzzkjEnzxw −+−−= ε (C - 44)
192
Next we will consider the energy transport across interfaces as this gives the reflectance
and the transmittance of the stack. The Poynting vector S in (C - 45) gives the
instantaneous rate of energy flow across a unit area, where the direction of propagation is
the normal vector to this unit area. As multiplying two fields is a nonlinear operation, we
cannot do the multiplication in phasor notation, but have to take the real parts.1
( ) ( )( ) ( )( )tzxtzxtzx ,,Re,,Re,, HES ×= (C - 45)
Again, we are more interested in the time-averaged value of S, which is called the
irradiance I. For a harmonic wave, the irradiance I can be calculated using (C - 46).1
( ) ( ) ( )( )∗×= tzxtzxzx ,,,,Re21, HEI (C - 46)
In order to calculate the transmittance of a multilayer stack, the ratio of the transmitted
irradiance to the incident irradiance has to be taken. In general, this CANNOT be done by
simply calculating the irradiances of the forward and backward propagating fields from
(C - 46) and taking the ratios, since these irradiances are not parallel if the refractive
indices of the incident and transmitting media are different. The tilt of the irradiances
changes the unit area.4 The correct transmittance could be calculated by obtaining the
irradiances from (C - 46) and accounting for the changed unit area at the end in the
calculation of the transmittance.5 Alternatively, the unit area can be fixed along the
interfaces and all irradiances can be calculated with respect to this unit area, i.e., the
irradiances I⊥ perpendicular to the interfaces are used.1 Here we use the second approach.
I⊥ is calculated from the fields along the layers as given in (C - 47).
( ) ( ) ( )( )∗⊥ = tzxHtzxEzxI ,,,,Re
21, |||| (C - 47)
Using (C - 23) to (C - 26), the incident irradiance I⊥,TM,inc for TM-polarization is
calculated to be (C - 48).
( ) ( ) 0
2
0,,,, cos2
1,Z
nEzxI
inc
incxfincTM θ
=⊥ (C - 48)
193
ninc is the refractive index in the incident medium and θinc the incidence angle. The
reflected irradiance I⊥,TM,refl and the transmitted irradiance I⊥,TM,trans are given by (C - 49)
and (C - 50).
( ) ( ) 0
20,,
,, cos21,
ZnE
zxIinc
incxbreflTM θ
=⊥ (C - 49)
( ) ( ) 0
2
1,,,, cos2
1,Z
nEzxI
trans
transNxftransTM θ
+⊥ = (C - 50)
ntrans is the refractive index in the transmitting medium and θtrans the propagation angle of
the transmitted light. For the case of TE-polarization, the incident irradiance I⊥,TE,inc, the
reflected irradiance I⊥,TE,refl, and the transmitted irradiance I⊥,TE,trans are given by (C - 51)
to (C - 53). These equations are obtained using (C - 28) to (C - 31).
( )( )
0
2
0,,,,
cos21,
ZnE
zxI incincyfincTE
θ=⊥ (C - 51)
( )( )
0
2
0,,,,
cos21,
ZnE
zxI incincybreflTE
θ=⊥ (C - 52)
( )( )
0
2
1,,,,
cos21,
ZnE
zxI transtransNyftransTE
θ+⊥ = (C - 53)
To calculate the transmittance, TTM for the case of TM-polarization and TTE for the case
of TE-polarization, we take the ratio of the transmitted and the incident irradiances. The
results are given in (C - 54) and (C - 55).
( ) ( )( )
( )( )transinc
inctrans
xf
NxfTM n
nE
ET
θθ
ωβ
ωβωβ
coscos
,,
,2
0,,
1,, += (C - 54)
( ) ( )( )
( )( )incinc
transtrans
yf
NyfTE n
nE
ET
θθ
ωβ
ωβωβ
coscos
,,
,2
0,,
1,, += (C - 55)
194
Finally, taking the ratio of the reflected and incident irradiances, we see that the
reflectance is given by (C - 56) for both polarization states.
( ) ( )( )
2
0,||,
0,||,
,,
,ωβ
ωβωβ
f
b
EE
R = (C - 56)
For the calculation of beams, the amplitude reflection coefficient r(β,ω) and transmission
coefficient t(β,ω) of the entire stack are also of interest. r(β,ω) and t(β,ω) are also called
the E-field transfer functions in reflection and transmission. Viewing the complete stack
as a system, these functions describe how the system acts on a given incident field, i.e.
r(β,ω) describes how the stack changes the amplitude and phase of an incident plane
wave at position z=0 upon reflection, while t(β,ω) characterizes how the amplitude and
phase of the transmitted field at position z=L compare to the incident field at z=0. As is
detailed in Appendix D, an incident beam of light can be decomposed into its plane wave
components. Since a multilayer dielectric stack is a linear, space-invariant (in the x- and
the y-direction) system, these components can be propagated through the stack
individually with their propagation determined by r(β,ω) and t(β,ω). Finally, the reflected
and transmitted total field is obtained by recombining the individually propagated
components. r(β,ω) and t(β,ω) are obtained by dividing the total reflected and
respectively transmitted E-field amplitude by the total incident E-field amplitude as
shown in (C - 57) to (C - 59). Again TM- and TE-polarization have to be distinguished
for the transmission coefficient.
( ) ( )( )ωβ
ωβωβ
,,
,0,||,
0,||,
f
b
EE
r = (C - 57)
( ) ( )( )
( )( )trans
inc
xf
NxfTM E
Et
θθ
ωβ
ωβωβ
coscos
,,
,0,,
1,, += (C - 58)
( ) ( )( )ωβ
ωβωβ
,,
,0,,
1,,
yf
NyfTE E
Et += (C - 59)
Finally, in order to calculate the spatial shift upon reflection, we are interested in the
wavevector K(β,ω) within the stack in the z-direction. This then determines the effective
195
group propagation angle. Since the stack is not periodic, K(β,ω) is only an effective
wavevector. The phase change upon reflection can be calculated using (C - 60).
( ) ( )( ) ( )( )ωβωβωβφ ,arg,arg, 0,||,0,||, fbrefl EE −= (C - 60)
The sign of the phase change depends on the positive vector directions chosen in
Fig. C.2. This has to be taken into account in determining if there is a maximum or a
minimum of the standing wave at the interface. Assuming a total stack thickness L, the
wavevector K(β,ω) in the z-direction is related to the phase change upon reflection as
given in (C - 61).
( ) ( )L
K refl
2,
,ωβφ
ωβ = (C - 61)
(C - 61) is the dispersion relation among K, β, and ω for a finite periodic or non-periodic
stack calculated using the transfer matrix method.
References
[1] H. A. MacLeod, Thin-Film Optical Filters, Institute of Physics Publishing,
Philadelphia (2001).
[2] D. A. Mlynski, Elektrodynamik, Skript der Vorlesung Elektrodynamik, Karlsruhe
(1994).
[3] U. S. Inan and A. S. Inan, Engineering Electromagnetics, Addison-Wesley
Longman, Inc., Menlo Park, CA (1999).
[4] E. Hecht, Optik, Addison-Wesley (Deutschland) GmbH, Bonn (1989).
[5] M. Born and E. Wolf, Principles of Optics, Cambridge University Press,
Cambridge, UK (1999).
196
Appendix D
Simulating Beams by Fourier Decomposition
In this appendix we discuss how to calculate the propagation of beams through linear,
space-invariant systems using a Fourier transform technique. The E-field in space is
given by the complex function E(x,y,z)=Ef(x,y,z)+Eb(x,y,z), where Ef(x,y,z) is the
complex amplitude of the forward propagating wave and Eb(x,y,z) is the complex
amplitude of the backward propagating wave at position (x,y,z).1 The goal is to calculate
E(x,y,z) at any point in space for a given incident field Ef(x,y,0) at position z=0 (any other
constant z could be chosen as well). We are assuming here for simplicity that a scalar
theory can be used to solve Maxwell’s equations and that the H-field can be calculated
from the E-field.1,2 For simple media a direct analytical relationship might exist between
E(x,y,z) and Ef(x,y,0). Let us, for example, consider the case of a uniform plane wave
propagating in the z-direction in free space. Since the plane wave is uniform, i.e., its
amplitude does not change along the plane of constant phase, Ef(x,y,0) is a constant as
given in (D - 1).
( ) 00,, EyxE fPW = (D - 1)
The forward propagating field at any point is then given by (D - 2).
197
( ) ( ) ( )zkiyxEzyxE fPWfPW 0exp0,,,, = (D - 2)
k0 is the wavevector as given in (D - 3).
ck πω2
0 = (D - 3)
Since no reflections occur in free space, EbPW(x,y,z)=0 everywhere and the total E-field is
equal to the forward propagating E-field. In Appendix C it is shown how the forward and
backward propagating waves can be calculated for an arbitrary layered stack assuming
that the incident field is a uniform plane wave.
Now we consider the propagation of beams of light, i.e. incident fields Ef(x,y,0) that are
not plane waves. In particular we will discuss the propagation of the fundamental
Gaussian beam. The Gaussian beam is the ideal diffraction limited wave and often
approximates laser beams well.3 It has rotational symmetry around the axis of
propagation. (D - 4) gives the field equation for the Gaussian beam in polar coordinates.
The direction of propagation is renamed ζ, since in this thesis the z–axis is the direction
perpendicular to the layers of a multilayer stack and we shortly will want to look at
Gaussian beams incident on such stacks with non-normal incidence such that ζ and z are
not the same direction.
( ) ( ) ( ) ( ) ( )( ) ( )( )tkR
rkw
rwwEtrEGB ωζζη
ζζζζ −
−
−= 0
20
2
20
0 iexpiexp2
iexpexp,, (D - 4)
(D - 5) to (D - 8) define the functions used in the definition of the Gaussian beam. z0 is
the Rayleigh range.
2000 2
1 wkz = (D - 5)
( ) 20
2
0 1z
ww ζζ += (D - 6)
( )
+= 2
201
ζζζ
zR (D - 7)
198
( )
=
0
arctanzζ
ζη (D - 8)
ξ-po
sitio
n in
µm
ζ-position in µm
ξ-po
sitio
n in
µm
ζ-position in µm
0 0.25 0.5 0.75 1Intensity color code
(a) (b)
Fig. D.1. Gaussian beam with w0=4.7µm and λ=890nm. (a) Calculated from exact
formula (D - 4). (b) Obtained from Fourier decomposition (D - 22). Note that the ξ-
and the ζ-axes have different scales.
Fig. D.1(a) shows a plot of a Gaussian beam calculated using (D - 4). Now let us assume
the incident field at a plane z=0 has the Gaussian beam profile EfGB(r,0,0). A multilayered
stack extends from z=0 to z=L. We could calculate the field EfGB(x,y,z) directly at all
positions and times using a Maxwell solver, such as the finite difference time domain
(FDTD) technique.4 Since this approach is, however, quite computationally intensive, we
are choosing a different approach that uses the fact that a multilayered stack is a linear,
space-invariant system.2 For sufficiently small intensities, the system is linear. The
system is space-invariant, since the z-direction transfer function only depends on the
difference between the input position (xin,yin) and the exit position (xout,yout), not on the
absolute position along the stack. For linear, space-invariant systems, the input field can
be decomposed into elementary components, these can be propagated individually, and
the output field is obtained by summing the propagated individual components.
199
One possible decomposition of the incident field E(x,y,z=0) is obtained by performing the
two-dimensional spatial Fourier transform given by (D - 9).2
( ) ( ) ( )( ) yxyfxfzyxEzffA yxyxFxy dd2iexp0,,0;, ∫ ∫∞
∞−
∞
∞−
+−=== π (D - 9)
The inverse Fourier transform (D - 10) expresses the field as a function of the Fourier
components AFxy(fx,fy;z=0).
( ) ( ) ( )( ) yxyxyxFxy ffyfxfzffAzyxE dd2iexp0;,0,, ∫ ∫∞
∞−
∞
∞−
+=== π (D - 10)
Comparing the integrand of (D - 10) with the equation for plane waves (D - 2), we see
that the Fourier transform decomposes the E-field into plane waves with amplitudes
AFxy(fx,fy;z=0) and wavevectors kx in the x-direction and ky in the y-direction given by
(D - 11) and (D - 12).
xx fk π2= (D - 11)
yy fk π2= (D - 12)
Here we limit ourselves to propagation of the beam in the x-z-plane and to calculations
for y=0. The one-dimensional Fourier transform and inverse transform are then given by
(D - 13) and (D - 14).
( ) ( ) ( ) xxfzxEzfA xxFx d2iexp0,0; ∫∞
∞−
−=== π (D - 13)
( ) ( ) ( ) xxxFx fxfzfAzxE d2iexp0;0, ∫∞
∞−
=== π (D - 14)
The spatial frequency fx is related to the directional cosine α, depicted in Fig. D.2, by
(D - 15).
λα
=xf (D - 15)
200
z
xk
θin=acos(γ)
acos(α)
Fig. D.2. Direction cosines α and γ for propagation in the x-z-plane.
Thus, we see that the integration in (D - 14) corresponds to the summation of plane
waves with amplitudes AFx(fx;z=0) propagating in different directions in the x-z-plane.
We can calculate the change of the plane wave amplitudes AFx(fx;z) as a function of z
using e.g. the transfer matrix method. The transfer function relating AFx(fx;z) to AFx(fx;0)
is called H(fx;z) as shown in (D - 16).
( ) ( ) ( )0;;; xFxxxFx fAzfHzfA = (D - 16)
Therefore, we can now calculate E(x, z) at any point in the x-z-plane by propagating the
individual Fourier components and summing them up after propagation as given in
(D - 17).
( ) ( ) ( )λα
αλπ
λα
π d2iexp;d2iexp;, ∫∫∞
∞−
∞
∞−
== xzAfxfzfAzxE FxxxxFx (D - 17)
For practical purposes we will replace the continuous Fourier transform in (D - 17) by a
discrete Fourier transform as shown in (D - 18), where αi is the directional cosine of the
ith-component and ∆α is the angular separation of the different components.
( )λα
αλπ
λα ∆
= ∑
ii
iFx xzAzxE 2iexp;, (D - 18)
Due to the discrete nature of the Fourier transform (D - 18), the resulting E-field is
repetitive in space. The repetition distance ∆x is given by (D - 19).
αλ
∆=∆x (D - 19)
If the E-field of interest is limited in space or can be approximated by a spatially limited
function extending over a spatial distance smaller than ∆x, the discrete Fourier transform
201
(D - 18) combined with a spatial filter of width ∆x correctly reconstructs the original
field. Solving (D - 19) for ∆α the minimum required angular sampling distance is
obtained. This is the Whittaker-Shannon sampling theorem well known from information
theory.2,5
Next we will apply the Fourier decomposition technique to calculating the propagation of
Gaussian beams. Instead of considering a rotationally symmetric Gaussian beam as given
in (D - 4), we will limit ourselves to a beam that has a Gaussian beam profile
perpendicular to the direction of propagation in the x-z-plane and is infinite in the y-
direction. Such an elongated beam is interesting for our application because spatial
wavelength demultiplexing using a multilayer thin-film stack only requires focusing in
the x-direction. Simultaneous focusing in the y-direction leads to unwanted polarization
mixing effects. Using again the ξ-ζ-coordinate system, where ζ is again the propagation
direction of the beam and ξ is the direction perpendicular to the propagation direction,
(D - 20) gives the E-field at position ζ=0.
( )
−=== 2
0
2
0 exp0,0,w
EtEGBapprξ
ζξ (D - 20)
Taking the Fourier transform of (D - 20), the Fourier decomposition amplitudes (D - 21)
are calculated.
( ) ( )222000 exp0,0, xxFGBappr fwwEtfA ππζ −=== (D - 21)
Using relation (D - 15) and the definition of the incident angle given in Fig. D.2, Fig. D.3
plots the Fourier amplitudes as a function of the plane wave propagation angle for the
components of a Gaussian beam with w0=4.7µm and λ=890nm. It can be seen that the
amplitude has decreased to 1/e for beam components at incidence angles of ± 3.5°.
Substituting (D - 16) and (D - 21) into (D - 18), we obtain the propagation of the
approximate Gaussian beams to be given by (D - 22).
( )λα
ξαλπ
λα
ππζλα
ζξ∆
−
= ∑
ii
iiGBappr wwEHE 2iexpexp;,
222
000 (D - 22)
202
H(fx;ζ) is the transfer function along the ζ-axis. As seen from (D - 2) the transfer function
of plane waves in free space is given by (D - 23).
( ) ( )zkizfH xfPW 0exp; = (D - 23)
10 8 6 4 2 0 2 4 6 8 100
0.10.2
0.30.40.5
0.60.7
0.80.9
1
Propagation Angle in deg
Nor
mal
ized
Am
plitu
de
Fig. D.3. Amplitude of Fourier components normalized by π00wE for a Gaussian
beam with w0=4.7µm and λ=890nm.
0 0.5 120
0
20
40
60
Incident (zeta=0)Position in microns
0 0.25 0.5 0.75 1Intensity color code
Normalized Intensity
ξ -po
sitio
n in
µm
ζ-position in µm
Fig. D.4. Propagation of a Gaussian beam with w0=4.7µm and λ=890nm in free
space.
203
Fig. D.4 plots the propagation of a Gaussian beam with w0=4.7µm and λ=890nm in free
space using 41 Fourier components spaced by dα =0.0075. This corresponds to a spatial
repeat distance of ∆x=119µm. The right hand graph in Fig. D.4 shows the intensity
distribution of the forward propagating light along the ξ–axis for ζ=0. Fig. D.1(b) shows
a more extended plot of the same beam. Comparing the beam in Fig. D.1(b) obtained
from the Fourier decomposition to the one in Fig. D.1(a) calculated from the exact
formula (D - 4), we see that they agree well within the Rayleigh range. Outside the
Rayleigh range, the beam calculated from the exact formula has a lower amplitude. This
is due to the widening of the beam in both the x- and the y-direction as the rotationally
symmetric widening beam approaches a spherical wave. The intensity of the beam is
constant for integration over the entire x-y-plane at a fixed z-value. (D - 22) on the other
hand considers a beam that is only focussed in the x-direction and infinite in the y-
direction. In this case the intensity is constant for integration along the x-axis for a fixed
y- and z-value.
F t
0 0.5 120
0
20
40
60
Incident (z=0)Position in microns
0 0.25 0.5 0.75 1Intensity color code
Normalized Intensity
x-po
sitio
n in
µm
z-position in µm
z
x
ζ
ξ
Fig. D.5. Propagation of a Gaussian beam with w0=4.7µm and λ=890nm in free
space at an angle of 40º with respect to the z-axis.
204
Next let us perform a coordinate transformation to calculate the field of a Gaussian beam
propagating at an angle θin with respect to the z-axis as depicted in Fig. D.5. ξ and ζ are
related to x and z as given in (D - 24) and (D - 25).
( ) ( )zx inin θθξ sincos −= (D - 24)
( ) ( )zx inin θθζ cossin += (D - 25)
Substituting (D - 24) and (D - 25) into (D - 20) and taking the Fourier transform at z=0,
we obtain the Fourier amplitudes given in (D - 26).
( ) ( )( )
−
−=
==
22
2
000
sincos
expcos
0,0,λ
θαπ
θπ
θλα in
ininFGBappr
wwEtzA (D - 26)
Substituting (D - 26) into (D - 18), we finally obtain (D - 27) for the propagation of a
Gaussian beam at an angle.
( ) ( ) ( )( )
λα
ξαλπ
λθα
πθ
πθλ
α d2iexpsincos
expcos
;,2
22
000∑
−
−
=
ii
ini
inin
iGBappr
wwEzHzxE
(D - 27)
Fig. D.5 plots the propagation of a Gaussian beam with w0=4.7µm, λ=890nm, and
θin=40º in free space using 41 Fourier components spaced by dα =0.0075. Fig. D.6 shows
the propagation of the same beam incident onto a 40-µm slab of material with refractive
index n=2.5. The total field is plotted, summing the forward and backward propagating
E-fields at each position. The transfer function H(fx;z) relating the incident plane waves at
z=0 to the forward and the backward propagating waves at each position z is obtained
using the transfer matrix method discussed in Appendix C. In the left hand plot of
Fig. D.6, both the refracted beam within the slab and the transmitted beam are clearly
seen. The reflections off the first air – slab interface and off the second slab – air interface
are dimly visible. We can also see the standing wave interference pattern between
forward and backward propagating light at the interfaces. On the right hand side of
Fig. D.6, the reflected light, i.e., the backward propagating intensity at z=0, is plotted
along the x-axis for a unity amplitude incident beam. The first and second reflections can
205
be clearly distinguished and it can be seen that they are approximately equal in
amplitude. Also shown is the transmitted light, i.e., the forward propagating intensity at
z=L=40µm.
0 0.5 1 1.5 2Intensity color code
0 0.5 120
0
20
40
60
Incident (z=0)Reflected (z=0)Transmitted (z=L)
Position in microns
Normalized Intensityz-position in µm
x-po
sitio
n in
µm
n=2.5n=1 n=1
Fig. D.6. Intensity distribution for a Gaussian beam with w0=4.7µm and λ=890nm
incident onto a 40-µm slab of material with refractive index n=2.5 at an angle of
40º with respect to the z-axis.
In conclusion we have described how to calculate the propagation of a beam of light
through an arbitrary multilayered stack by employing a Fourier decomposition of the
beam into plane wave components, propagating these components individually, and
obtaining the total field by summing the individual components.
References
[1] U. S. Inan and A. S. Inan, Engineering Electromagnetics, Addison Wesley
Longman, Inc., Menlo Park, CA (1999).
[2] J. W. Goodman, Introduction to Fourier Optics, The McGraw-Hill Companies, Inc.
(1996).
206
[3] A. E. Siegman, Lasers, University Science Books, Sausalito, CA (1986).
[4] A. Taflove and S. C. Hagness, Computational Electrodynamics: The Finite-
Difference Time-Domain Method, Artech House (2000).
[5] A. V. Oppenheim, R. W. Schafer, J. R. Buck, Discrete-Time Signal Processing,
Prentice Hall, New Jersey (1999).
207
Appendix E
Numerical Optimization Methods
The design of thin-film filters is typically divided into two stages – (i) synthesis of a start
design with characteristics that somewhat match the desired ones, and (ii) refinement of
this start design to gradually improve its performance. The performance of a structure is
measured by a merit function MF – a single number comparing the current design
characteristics with the desired design characteristics.1 The definition of the merit
function used is given in (E - 1).
pN
i
p
i
iTi
QQQ
NMF
/1
1
1
∆−
= ∑=
(E - 1)
Qi is the current value of a quantity of interest, QiT the target value of that quantity, ∆Qi
the acceptable deviation, N is the number of sampling points, and p the p-norm used.2 In
the case of p = 2, the merit function is, e.g., the root-mean-square difference between the
current values and the target values of the quantities of interest.
Several synthesis methods are available to obtain a good start design. Graphical methods
are possible if the design has only a few layers.3,4 Analytical techniques are particularly
desirable as they normally allow the fast generation of a start design. Such techniques
208
include the synthesis based on coupled-mode theory5 as discussed in Chapter 5, digital
signal processing techniques6 introduced in Chapter 6, as well as techniques based on an
inverse Fourier transform.3,7 As the Fourier transform technique is only applicable to the
synthesis of transmittance profiles, it was not implemented here. A numerical synthesis
method that can be used for any type of design is the flip-flop method.8 This method is
based on the idea that thin layers of just two materials can approximate any index profile.
The desired total thickness of the stack is divided into a large number of thin layers with
alternating refractive indices. The refractive index of these layers is sequentially flipped,
keeping the index flipped if this improves the performance. At the end neighboring layers
of the same index are combined. Finally, intuition and experience are very important in
generating a good start design.
As an example we discuss here the design of an erbium doped fiber amplifier (EDFA)
gain equalizing filter. Since an EDFA does not amplify all wavelengths equally, a gain-
equalizing filter is needed to achieve a flat gain spectrum. Operating the filter in
reflection, the desired reflectance as a function of wavelength is plotted in Fig. E.1.
Fig. E.2 shows the performance of two 200-layer start designs – one generated using the
flip-flop technique and one a guessed start structure. In the following the performance of
different refinement techniques in improving these structures will be compared.
1530 1535 1540 1545 1550 1555 1560 15650
0.2
0.4
0.6
0.8
1
Target reflectanceWavelength in nm
Ref
lect
ance
Fig. E.1. Target reflectance as a function of wavelength for an EDFA gain-
flattening filter.
209
1530 1540 1550 15600
0.2
0.4
0.6
0.8
1
Target reflectanceMartina's guessFlipflop synthesis
Wavelength in nm
Ref
lect
ance
Merit function
MFMG 54.5=
MFFF 17.4=
Thickness
dMG 36.8um=
dFF 36.0um=
Fig. E.2. Reflectance of two 200-layer start designs. The merit function and the
total thickness of both designs are given to the right.
We implemented six different numerical optimization techniques – golden section
search,2 secant method,2 conjugate gradient algorithm,2 Broyden-Fletcher-Goldfarb-
Shanno (BFGS),2 damped least squares method,9 and Hooke&Jeeves pattern search.10 All
six techniques are local optimization techniques, i.e., if used in their range of validity
they will find the closest local minimum in the merit function. Due to the large number of
degrees of freedom available in the design of a 200-layer structure, this next local
minimum may be very shallow and several minima may be located close together. Lower
minima can therefore be found by using the numerical techniques somewhat outside their
range of validity.
Let us look at the example of the golden section search technique. This technique is a
one-dimensional search method that uses an golden section interval reduction to reduce
the number of points that have to be calculated in each search step.2 The golden section
search technique only finds the local minimum if it is guaranteed that there is only one
local minimum within the initial interval. Using the interval reduction, this local
minimum is found. If we now use this technique on an interval that has several local
210
minima, there is no guarantee which one it will find. But it is likely that a lower
minimum is found than if we choose a small interval. Therefore, using local optimization
techniques out of their original range of validity often results in better designs.
Choosing for example a large initial interval for the golden section search can result in
structures that are quite different from and superior to the original structure. As the
golden section search is a one-dimensional technique, the different layer thicknesses are
optimized sequentially keeping the refractive indices constant. Fig. E.3 plots the
performance of the two start designs refined using the golden section search technique.
Figs. E.3 to E.8 are obtained by running the refinement technique in question for one
iteration on each start design. As different methods are programmed differently, the
length of the calculation time is given in addition to the final merit function. Longer
runtimes of some optimization techniques might further improve the performance.
1530 1540 1550 15600
0.2
0.4
0.6
0.8
1
Target reflectanceMartina's guessFlipflop synthesis
Wavelength in nm
Ref
lect
ance
Merit function
MFMG 13.5=
MFFF 14.9=
Thickness
dMG 67.7um=
dFF 41.5um=
Calculation time
tMG 101min=
tFF 132min=
Fig. E.3. Golden section search refinement.
Newton’s method2 is an important technique in numerical optimization. In one
dimension, this method approximates the function at the given position by a parabola.
The next estimate for the local minimum is then the minimum of the parabola. This
method allows fast optimization, but may not converge if the second derivative of the
211
function is positive for some values. The secant method is basically Newton’s method
with the second derivative necessary for the calculation of the parabola replaced by a
difference of first derivatives.2 This technique is also a one-dimensional search method
and the layers are optimized sequentially. Fig. E.4 plots the performance improvements
achieved using the secant method.
1530 1540 1550 15600
0.2
0.4
0.6
0.8
1
Target reflectanceMartina's guessFlipflop synthesis
Wavelength in nm
Ref
lect
ance
Merit function
MFMG 12.4=
MFFF 13.2=
Thickness
dMG 36.4um=
dFF 35.9um=
Calculation time
tMG 66min=
tFF 74min=
Fig. E.4. Secant method refinement.
Next we turn to multivariable search techniques. These techniques optimize all layer
thicknesses simultaneously. One such method is the steepest descent method.2 At a given
point of the function, the gradient of steepest descent is calculated. Then a one-
dimensional line search is performed to find the minimum in this direction and the
procedure is repeated from this point on. Due to the nature of the steepest descent
algorithm, the directions of search are always perpendicular leading to a slow
approximation of the local minimum. A much quicker convergence is obtained using the
multidimensional Newton’s method.2 Here a quadratic Taylor expansion of the function
is used and the minimum of the expansion is taken as the next point. As for the one-
dimensional case, this algorithm does not guarantee convergence. The conjugate gradient
algorithm is an intermediate between the steepest descent and Newton’s method. The
search direction is calculated as a linear combination of the previous direction and the
212
current gradient such that all directions are mutually conjugate with the Hessian.2 For this
algorithm convergence is guaranteed and the rate is faster than for the steepest descent
method. Fig. E.5 plots the results obtained using the conjugate gradient algorithm.
1530 1540 1550 15600
0.2
0.4
0.6
0.8
1
Target reflectanceMartina's guessFlipflop synthesis
Wavelength in nm
Ref
lect
ance
Merit function
MFMG 40.7=
MFFF 15.3=
Thickness
dMG 36.8um=
dFF 35.9um=
Calculation time
tMG 17min=
tFF 50min=
Fig. E.5. Conjugate gradient algorithm refinement.
1530 1540 1550 15600
0.2
0.4
0.6
0.8
1
Target reflectanceMartina's guessFlipflop synthesis
Wavelength in nm
Ref
lect
ance
Merit function
MFMG 13.6=
MFFF 11.5=
Thickness
dMG 36.7 um=
dFF 35.8 um=
Calculation time
tMG 76min=
tFF 127 min=
Fig. E.6. BFGS algorithm refinement.
213
The Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm is a quasi-Newton method
based on the idea of modifying the update formula such that descent is guaranteed.2
Fig. E.6 graphs the performance improvements obtained using the BFGS algorithm.
All the algorithms discussed so far – golden section search, secant method, conjugate
gradient algorithm, and BFGS algorithm – use a merit function to judge the performance
of a design. For the design of a specific reflectance profile, the performance of the design
is calculated using the merit function in (E - 1) and the design is improved such that this
merit function value decreases. In contrast to this, the damped least squares algorithm
directly calculates the influence of the layer thicknesses on the reflectance values. By
calculating the derivative of each reflectance value with respect to each layer thickness,
the best (in the least squares sense) combination of layers is obtained.9 In order to prevent
unphysical layer thicknesses, dampening vectors are introduced that “damp” the change
of the layer thickness in the undesired direction. As seen in Fig. E.7 the results obtained
with this algorithm here are very unsatisfactory. Keep in mind that we are comparing
here the implementation of algorithms as well as the algorithms themselves. Therefore,
the implementation of the damped least squares algorithm might not be very good. It did
perform better though for simpler structures than the example chosen here.
All prior techniques except for the one-dimensional golden section search required the
calculation of derivatives. As these derivatives are often not available analytically, they
have to be calculated numerically, adding to the total computation time. The Hooke &
Jeeves pattern search technique is a very powerful multivariable search algorithm that
does not require the calculation of derivatives.10 The algorithm is divided into two phases
– an exploration phase and a pattern move. During the exploration phase the parameters
are sequentially increased and decreased by a small step. If an improvement in the merit
function is achieved, the step is recorded in a pattern vector. For the pattern move the
pattern vector is multiplied by an acceleration factor and is added to the parameter vector.
If there is no improvement after a pattern move and a subsequent exploration, the step
size is reduced. Fig. E.8 plots performance of both start designs after the application of
one iteration of the Hooke & Jeeves pattern search technique.
214
1530 1540 1550 15600
0.2
0.4
0.6
0.8
1
Target reflectanceMartina's guessFlipflop synthesis
Wavelength in nm
Ref
lect
ance
Merit function
MFMG 55.7=
MFFF 17.9=
Thickness
dMG 36.7um=
dFF 36.0um=
Calculation time
tMG 10min=
tFF 18min=
Fig. E.7. Damped least squares refinement.
1530 1540 1550 15600
0.2
0.4
0.6
0.8
1
Target reflectanceMartina's guessFlipflop synthesis
Wavelength in nm
Ref
lect
ance
Merit function
MFMG 8.8=
MFFF 13.7=
Thickness
dMG 36.5um=
dFF 35.9um=
Calculation time
tMG 96min=
tFF 67min=
Fig. E.8. Hooke & Jeeves pattern search refinement.
Table E.1 compares the performance of the different implemented designs in terms of the
final merit function, the final thickness, and the calculation time for each refinement
algorithm and for both start designs. The green boxes show the best results, while the red
215
numbers indicate poor performance. It can be seen that all methods except for the
damped least squares algorithm improved the performance of the start designs. The larger
thicknesses of the golden section search designs are due to the large intervals used in the
implementation. A connection between calculation time and merit function is visible as
well. Interestingly, both start designs result in equally good refined designs.
MFMG MFFFdMG
(µm)dFF
(µm)tMG
(min)tFF
(min)
Starting designs 54.5 17.4 36.8 36.0
Golden Section 13.5 14.9 67.7 41.5 101 132
Secant Method 12.4 13.1 36.4 35.9 66 74
Conjugate Gradient 40.7 15.3 36.8 35.9 17 50
BFGS 13.6 11.5 36.7 35.8 76 127
Damped Least Sq. 55.7 17.9 36.7 36.0 10 18
Hooke & Jeeves 8.8 13.7 36.5 35.9 96 67
Table E.1. Comparison of refinement methods.
1530 1540 1550 15600
0.2
0.4
0.6
0.8
1
Target reflectanceCombination of methods
Wavelength in nm
Ref
lect
ance
Merit function
MFMG 3.6=
Thickness
dMG 42.4um=
Fig. E.9. Refined design using a combination of the different methods.
216
In conclusion we find that no one algorithm is the clear winner. All algorithms have their
merits and we actually achieved the best performance improvements using the different
optimization algorithms sequentially. Fig. E.9 shows the performance of a gain flattening
filter designed using a combination of the different techniques.
For this work numerical optimization is a means to achieve multilayer thin-film stacks
with superior dispersion characteristics. To this end, I programmed this selection of
standard optimization procedures. As seen in Chapter 9, the resulting designs seem to be
close to the limit of what is physically achievable. Therefore, there was no need to
implement further, maybe more powerful, optimization techniques such as the needle
technique or genetic algorithms.2 In the pursuit of the best design, it might be well worth
the effort to implement these or other modern techniques.
References
[1] J. A. Dobrowolski, F.C. Ho, A. Belkind, V.A. Koss, “Merit functions for more
effective thin film calculations,” Appl. Opt., 28/14, 2824-2831 (1989).
[2] E.K.P. Chong and S.H. Zak, An Introduction to Optimization, John Wiley & Sons,
Inc. (1996).
[3] Li Li and J. A. Dobrowolski, “Computation speeds of different optical thin-film
synthesis methods,” Appl. Opt., 31/19, 3790-3799 (1992) and references herein.
[4] H. A. MacLeod, Thin-Film Optical Filters, Institute of Physics Publishing,
Philadelphia (2001).
[5] N. Matuschek, F.X. Kärtner, and U. Keller, “Analytical Design of Double-Chirped
Mirrors with Custom-Tailored Dispersion Characteristics,” IEEE J. Quantum
Electron., 35/2, 129-137 (1999).
[6] G. Lenz and C. K. Madsen, “General optical all-pass filter structures for dispersion
control in WDM systems,” J. Lightw. Techn., 17/7, 1248-1254 (1999).
[7] P.G. Verly, Appl. Opt., 35/25, pp.5148-5153 (1996).
[8] W. H. Southwell, Appl. Opt., 24/4, pp.457-460 (1985).
217
[9] J. A. Dobrowolski and R.A. Kemp, “Refinement of optical multilayer systems with
different optimization procedures,” Appl. Opt., 29/19, 2876-2893 (1990) and
references herein.
[10] T.E. Shoup and F. Mistree, Optimization Methods with Applications for Personal
Computers, Prentice-Hall, Inc. (1987).
218
Appendix F
Beam Cone in a Dispersive Stack
Here the angular range ∆θstruc of a beam within a dispersive stack is discussed as a
function of the input angular range ∆θin. As shown in Appendix D, a beam of light can be
decomposed into plane-wave components with different propagation directions. Thus, an
incident beam of light consists of a range of incidence angles. Here we will show that a
change in the incidence angle has a similar effect on the group propagation angle θgroup as
a change in the incident frequency. Therefore, we expect that the angular range of a beam
in a multilayer stack depends on the dispersion of the stack. Here we will quantify this
dependency.
In principle, a beam has components at all angles, but many of these might have very
small amplitudes and can thus be neglected. Here we choose a beam to be delimited by
the 1/e2 intensity components. All calculations are equally correct for a different
delimiting value, as long as the same value is chosen for the input angular range ∆θin and
the angular range ∆θstruc in the stack. Approximating differences by a differentials, ∆θstruc
can be estimated for a given ∆θin using (F – 1).
219
( ) ( )in
groupinstruc θ
θ
ωθθθωθθ ∆
∂
∂=∆∆ ~
~,~,~,~ (F - 1)
We see that the angular range ∆θstruc depends on the change of the group propagation
angle θgroup with the incident angle θ. Without dispersion, i.e., for a frequency-
independent constant average refractive index navg, the group propagation angle θgroup is
given by Snell’s law as seen in (F – 2).
( )
= −
=avg
Dispgroup nθ
ωθθ~sinsin~,~ 1
0, (F - 2)
An approximation of navg can be obtained by calculating the total optical thickness of the
structure and dividing the result by the physical thickness of the structure. For this case
the angular range ∆θstruc is calculated as.
( ) ( )in
avg
inDispgroup
inDispstrucn
θθ
θθ
θ
ωθθθωθθ ∆
−=∆
∂
∂=∆∆ =
= ~sin
~cos~
~,~,~,~
22
0,0, (F - 3)
We see that that the size of the beam cone in the stack is proportional to the incident
beam cone for a given center incidence angle θ.
Next we will consider the beam cone in a dispersive stack with the dispersion of the stack
Dispω defined in (F – 4).
( ) ( )ω
ωθθωθω ~
~,~~,~
∂
∂= groupDisp (F - 4)
Intergrating (F – 4), we obtain the group propagation angle in the dispersive stack as
given in (F – 5).
( ) ( )∫ ∂= ωωθωθθ ω~~,~~,~ Dispgroup (F - 5)
In order to calculate (F – 1) from (F – 5), we remember that the partial derivatives ω~∂
and θ~
∂ are related by (F – 6).1
220
( )( ) θ
ωωωθθ
θωθθ
θω
θ
~~
~~,~
~~,~
~~
∆∆
≈∂∂
∂∂−=
∂∂
= group
group
constgroup
(F - 6)
Therefore, we can rewrite (F – 1) as (F – 7).
( ) ( ) ( )ω
ω
ωθθθ
θ
ωθθθωθθ ~
~~,~
~~,~
,~,~∆
∂
∂−=∆
∂
∂=∆∆ group
ingroup
instruc (F - 7)
Substituting (F – 5) into (F – 7) we obtain (F – 8).
( ) ( ) ( ) ωωθωω
ωωθθωθθ ω
ω ~~,~~~
~~,~,~,~
∆+−=∆∂
∂∂−=∆∆ ∫ CDisp
Dispinstruc (F - 8)
Since (F – 8) has to reduce to (F – 3) in the case of Disp=0, we can calculate the value of
the integration constant C. Substituting C and (F – 6) into (F – 8), we obtain the
relationship between the beam cone ∆θstruc in a dispersive stack and the incident beam
cone ∆θin as (F – 9).
( ) ( ) ( ) ( )( ) in
group
groupinDispstrucinstruc Disp θ
ωωθθ
θωθθωθθωθθθωθθ ω ∆
∂∂
∂∂+∆∆=∆∆ = ~~,~
~~,~~,~,~,~,~,~
0, (F - 9)
For a given dispersive stack, the beam cone Disp=0 can be estimated by calculating the
beam cone in a wavelength regime far away from any dispersion. Next we will examine
the second part of the sum in (F – 9) closer. Using the variable transformation (F – 10)
and (F – 11), we obtain (F – 12) and (F – 13).
( )θω
β~sin
~
c= (F - 10)
ωω ~= (F - 11)
( ) ( ) ( )
( )θ
ωβ
ωβθθω
ω
ωβθ
θβ
β
ωβθ
θ
ωθθ
~cos~,
~,
~,
~~,~
cgroup
groupgroupgroup
∂
∂=
∂∂
∂
∂+
∂∂
∂
∂=
∂
∂
(F - 12)
221
( ) ( ) ( )
( ) ( )ω
ωβθθβ
ωβθωω
ω
ωβθ
ωβ
β
ωβθ
ω
ωθθ
∂
∂+
∂
∂=
∂∂
∂
∂+
∂∂
∂
∂=
∂
∂
,~sin,
~,
~,
~~,~
groupgroup
groupgroupgroup
c
(F - 13)
Substituting (F - 12) and (F - 13) into (F - 6) we obtain (F - 14).
( )( ) ( )
( ) constgroup
groupgroup
group
group
cc
c
=∂∂
−=
∂∂
∂∂+
=∂∂
∂∂
θωβ
θ
θω
βωβθ
ωωβθθ
θω
ωωθθ
θωθθ
~sin
~cos~
,,~sin
~cos~
~~,~
~~,~(F - 14)
In Chapter 4 we have seen that (∂β/∂ω)|K=const is approximately constant with wavelength.
Fig. F.1 plots the change in (∂β/∂ω)|K=const for the two structures discussed in Chapter 8.
1520 1530 1540 1550 1560 15702
0
2
4
6
8
10
Transfer Matrix Calculation
Periodic Design
Wavelength in nm
Chan
ge in
%
1520 1530 1540 1550 1560 15700
1
2
3
Transfer Matrix Calculation
Non-periodic Design
Wavelength in nm
Chan
ge in
%
Fig. F.1. The change in (∂β/∂ω)|K=const as a function of wavelength.
Without proof, we are assuming now that relation (F - 15) is a good approximation.
( )( )
( )( ) βωβ
ωωβωβ
βωββωωβ
ωβ
θ ∂∂∂∂
−=∂∂
≈∂∂
∂∂∂−=
∂∂
== ,,
,,
22
2
KK
KK
constKconstgroup
(F - 15)
As seen in Chapter 4, a change in the group propagation angle is tightly linked to a
change in the wavevector K in the case of structural dispersion. This could be the
222
underlying reason that (F - 15) is a good approximation for many structures of interest. A
rigorous proof still remains to be performed.
Substituting (F - 3) and (F - 14) into (F - 9), we finally obtain an interesting
approximation for ∆θstruc as a function of frequency. The result is given in (F - 16).
( ) ( ) in
K
avg
instruc
cDisp
nθ
ωβ
θ
θωωθ
θ
θθωθθ ω ∆
∂∂
−+
−≈∆∆
~sin
~cos~~,~~sin
~cos,~,~22
(F - 16)
The only rapidly varying term with wavelength in (F - 15) is the dispersion Dispω. Thus,
the angular range of a mode ∆θstruc can be estimated as the input angular range ∆θin
multiplied by the sum of a constant term added to a term that is proportional to the
dispersion. The validity of this approximation is shown with Fig. F.2.
1520 1530 1540 1550 1560 15700
2
4
6
Bloch CalculationApproximation ~Dispersion + const
Periodic Design
Wavelength in nm
Ang
ular
rang
e in
deg
1520 1530 1540 1550 1560 15701
2
3
4
5
6
Transfer Matrix CalculationApproximation ~Dispersion + const
Non-periodic Design
Wavelength in nm
Ang
ular
rang
e in
deg
Fig. F.2. Angular range ∆θstruc for an input angular range of ∆θin = 0.5º directly
calculated and calculated from (8 – 17) as being proportional to the dispersion.
In Fig. F.2 the angular range within the structure is graphed as a function of wavelength
for an input angular range of 0.5º. The black and pink curves are obtained by directly
calculating the group propagation angles for the two extreme input angles. The cyan
223
curves on the other hand are calculated using (F - 15). The deviations of the two curves
are mainly due to the fact that (F - 15) estimates the angular range by just looking at the
central incidence angle. Fig. F.3 shows the same calculation for an input angular range of
0.05º. In this case the agreement between the curves is very good even for extreme
points, thus confirming the hypothesis that the error in Fig. F.2 is mainly due to replacing
differences by differentials.
1520 1530 1540 1550 1560 15700
0.2
0.4
0.6
Bloch CalculationApproximation ~Dispersion + const
Periodic Design
Wavelength in nm
Ang
ular
rang
e in
deg
1520 1530 1540 1550 1560 15700
0.1
0.2
0.3
0.4
0.5
0.6
Transfer Matrix CalculationApproximation ~Dispersion + const
Non-periodic Design
Wavelength in nm
Ang
ular
rang
e in
deg
Fig. F.3. Angular range ∆θstruc for an input angular range of ∆θin = 0.05º directly
calculated and calculated from (8 - 17) as a function of wavelength.
References
[1] I.N. Bronstein, K.A. Semendjajew, G. Musiol, H. Muehlig, Taschenbuch der
Mathematik; Verlag Harri Deutsch, Thun und Frankfurt am Main, pp. 232-237
(1993).
224
Appendix G
Composition of Structures
Here the layer compositions of the discussed structures are given. Table G.1 lists the
layer thicknesses as a function of the layer number and the layer material for the stacks
discussed in Chapters 3 to 5. The first row names the structures in the format
Chapter-Stack number. The stacks are numbered by order of appearance. The second row
describes the type of structure. The third row lists the wavelength range of interest. As
many structures have a periodic behavior with frequency, a different wavelength range
could be chosen just as well. The wavelength range given here corresponds to the one
discussed in the text and can be seen as a rough guide of where to look if plotting the
dispersion characteristics of the different stacks. Next the incidence angle in vacuum and
the polarization used in designing the stack, as well as the total stack thickness, are given.
The remaining rows list the stack composition starting with the substrate material. All
layer thicknesses are in nanometers. The substrate material for the fabricated stacks is
quartz (n=1.52 at 880 nm). Some of the simulations have been performed assuming an
incidence material such as SiO2 or air. All stacks considered have alternating layers of
SiO2 (n=1.456 at 880 nm) and Ta2O5 (n=2.06 at 880 nm). As in some stacks the first
layer is Ta2O5, while it is SiO2 in other stacks, designs that have SiO2 as the first layer are
225
shifted down by one row. Table G.2 gives the composition of the stacks discussed in
Chapters 6 to 10.
3-1 3-2 3-3 5-1 5-2 5-3 5-4 5-5 5-6200-layerPeriodic
60-layerPeriodic
60-layerImpedance
MatchedPeriodic
60-layerDouble-Chirped
f=0.5
60-layerDouble-Chirpedf=0.33
60-layerDouble-Chirped
f=0.2
60-layerDouble-Chirped
f=0.1
60-layerSingle-Chirped
f=0
200-layerDouble-Chirped
860 - 901 935 - 1050 900 - 1050 800-1100 800-1010 800-950 800-900 800-880 780-950
40° 45° 45° 45° 45° 45° 45° 45° 40°
p-pol. p-pol. p-pol. p-pol. p-pol. p-pol. p-pol. p-pol. p-pol.
30.0 8.3 9.0 9.4 9.1 8.8 8.7 8.5 28.1
Quartz SiO2 SiO2 Quartz Quartz Quartz Quartz Quartz Quartz1 Ta2O5 150.0 108.0 3.72 SiO2 150.0 167.9 330.0 312.2 312.0 311.9 311.8 311.7 291.83 Ta2O5 150.0 108.0 7.6 2.1 1.9 1.8 1.8 1.7 0.54 SiO2 150.0 167.9 323.9 307.4 307.1 306.7 306.4 306.1 290.65 Ta2O5 150.0 108.0 11.7 6.5 6.1 5.8 5.6 5.4 1.56 SiO2 150.0 167.9 317.6 302.4 301.9 301.3 300.8 300.3 289.47 Ta2O5 150.0 108.0 15.8 11.2 10.4 9.9 9.5 9.2 2.58 SiO2 150.0 167.9 311.2 297.4 296.6 295.8 295.1 294.3 288.19 Ta2O5 150.0 108.0 19.9 15.9 14.8 14.0 13.6 13.1 3.510 SiO2 150.0 167.9 304.8 292.2 291.2 290.1 289.3 288.3 286.811 Ta2O5 150.0 108.0 24.2 20.7 19.2 18.3 17.7 17.1 4.612 SiO2 150.0 167.9 298.2 287.1 285.8 284.5 283.4 282.2 285.513 Ta2O5 150.0 108.0 28.4 25.5 23.8 22.6 21.8 21.1 5.714 SiO2 150.0 167.9 291.6 281.9 280.3 278.7 277.4 276.1 284.215 Ta2O5 150.0 108.0 32.7 30.4 28.3 26.9 26.0 25.1 6.816 SiO2 150.0 167.9 285.0 276.7 274.8 272.9 271.4 269.9 282.817 Ta2O5 150.0 108.0 37.0 35.4 32.9 31.3 30.2 29.2 7.918 SiO2 150.0 167.9 278.3 271.5 269.2 267.1 265.4 263.7 281.519 Ta2O5 150.0 108.0 41.3 40.3 37.5 35.6 34.4 33.3 9.020 SiO2 150.0 167.9 271.6 266.3 263.7 261.3 259.3 257.4 280.121 Ta2O5 150.0 108.0 45.6 45.3 42.2 40.1 38.7 37.4 10.122 SiO2 150.0 167.9 264.8 261.1 258.1 255.4 253.2 251.1 278.823 Ta2O5 150.0 108.0 50.0 50.3 46.8 44.5 43.0 41.6 11.324 SiO2 150.0 167.9 258.1 255.9 252.5 249.5 247.1 244.7 277.425 Ta2O5 150.0 108.0 54.4 55.4 51.5 49.0 47.3 45.7 12.426 SiO2 150.0 167.9 251.2 250.8 246.9 243.5 241.0 238.4 276.027 Ta2O5 150.0 108.0 58.8 60.4 56.2 53.4 51.6 49.9 13.528 SiO2 150.0 167.9 244.4 245.7 241.3 237.6 234.8 232.0 274.629 Ta2O5 150.0 108.0 63.2 65.5 61.0 57.9 55.9 54.1 14.630 SiO2 150.0 167.9 237.6 240.6 235.7 231.6 228.6 225.6 273.231 Ta2O5 150.0 108.0 67.6 70.6 65.7 62.4 60.3 58.3 15.832 SiO2 150.0 167.9 230.7 235.5 230.0 225.6 222.4 219.2 271.833 Ta2O5 150.0 108.0 72.1 75.8 70.5 67.0 64.7 62.6 16.934 SiO2 150.0 167.9 223.8 230.5 224.4 219.6 216.1 212.7 270.435 Ta2O5 150.0 108.0 76.5 80.9 75.3 71.5 69.1 66.8 18.136 SiO2 150.0 167.9 216.8 225.6 218.8 213.6 209.8 206.2 269.037 Ta2O5 150.0 108.0 81.0 86.1 80.1 76.1 73.5 71.1 19.238 SiO2 150.0 167.9 209.9 220.7 213.2 207.6 203.6 199.7 267.639 Ta2O5 150.0 108.0 85.5 91.2 84.9 80.7 77.9 75.4 20.440 SiO2 150.0 167.9 202.9 215.8 207.6 201.5 197.3 193.2 266.241 Ta2O5 150.0 108.0 90.0 96.4 89.7 85.2 82.3 79.6 21.542 SiO2 150.0 167.9 195.9 211.0 202.0 195.5 191.0 186.7 264.843 Ta2O5 150.0 108.0 94.5 101.6 94.5 89.8 86.7 83.9 22.744 SiO2 150.0 167.9 188.9 206.3 196.4 189.4 184.6 180.2 263.445 Ta2O5 150.0 108.0 99.0 106.8 99.4 94.4 91.2 88.2 23.946 SiO2 150.0 167.9 181.9 201.6 190.9 183.4 178.3 173.6 262.047 Ta2O5 150.0 108.0 103.5 112.0 104.2 99.1 95.6 92.5 25.048 SiO2 150.0 167.9 174.9 197.0 185.3 177.3 171.9 167.0 260.549 Ta2O5 150.0 108.0 108.0 117.3 109.1 103.7 100.1 96.9 26.250 SiO2 150.0 167.9 167.9 192.5 179.8 171.2 165.6 160.4 259.151 Ta2O5 150.0 108.0 108.0 122.5 114.0 108.3 104.6 101.2 27.452 SiO2 150.0 167.9 167.9 191.1 177.5 168.4 162.5 157.1 257.753 Ta2O5 150.0 108.0 108.0 125.7 116.7 110.8 106.9 103.4 28.654 SiO2 150.0 167.9 167.9 192.9 178.4 168.9 162.7 157.1 256.355 Ta2O5 150.0 108.0 108.0 126.9 117.4 111.1 107.0 103.4 29.756 SiO2 150.0 167.9 167.9 194.8 179.4 169.4 162.9 157.1 254.8
Substrate
Description
Incidence Angle in Vacuum
PolarizationTotal Stack Thickness
in µm
Operatingwavelengthrange in nm
226
3-1 3-2 3-3 5-1 5-2 5-3 5-4 5-5 5-657 Ta2O5 150.0 108.0 108.0 128.2 118.0 111.5 107.2 103.4 30.958 SiO2 150.0 167.9 167.9 196.8 180.4 169.9 163.2 157.1 253.459 Ta2O5 150.0 108.0 108.0 129.4 118.7 111.8 107.3 103.4 32.160 SiO2 150.0 167.9 167.9 198.7 181.4 170.4 163.4 157.1 251.961 Ta2O5 150.0 130.7 119.4 112.1 107.5 103.4 33.362 SiO2 150.0 250.563 Ta2O5 150.0 34.564 SiO2 150.0 249.165 Ta2O5 150.0 35.766 SiO2 150.0 247.667 Ta2O5 150.0 36.868 SiO2 150.0 246.269 Ta2O5 150.0 38.070 SiO2 150.0 244.771 Ta2O5 150.0 39.272 SiO2 150.0 243.373 Ta2O5 150.0 40.474 SiO2 150.0 241.975 Ta2O5 150.0 41.676 SiO2 150.0 240.477 Ta2O5 150.0 42.878 SiO2 150.0 239.079 Ta2O5 150.0 44.080 SiO2 150.0 237.581 Ta2O5 150.0 45.282 SiO2 150.0 236.183 Ta2O5 150.0 46.484 SiO2 150.0 234.685 Ta2O5 150.0 47.686 SiO2 150.0 233.287 Ta2O5 150.0 48.888 SiO2 150.0 231.789 Ta2O5 150.0 50.090 SiO2 150.0 230.391 Ta2O5 150.0 51.292 SiO2 150.0 228.893 Ta2O5 150.0 52.594 SiO2 150.0 227.395 Ta2O5 150.0 53.796 SiO2 150.0 225.997 Ta2O5 150.0 54.998 SiO2 150.0 224.499 Ta2O5 150.0 56.1100 SiO2 150.0 223.0101 Ta2O5 150.0 57.3102 SiO2 150.0 221.5103 Ta2O5 150.0 58.5104 SiO2 150.0 220.1105 Ta2O5 150.0 59.7106 SiO2 150.0 218.6107 Ta2O5 150.0 61.0108 SiO2 150.0 217.2109 Ta2O5 150.0 62.2110 SiO2 150.0 215.7111 Ta2O5 150.0 63.4112 SiO2 150.0 214.3113 Ta2O5 150.0 64.6114 SiO2 150.0 212.8115 Ta2O5 150.0 65.8116 SiO2 150.0 211.3117 Ta2O5 150.0 67.1118 SiO2 150.0 209.9119 Ta2O5 150.0 68.3120 SiO2 150.0 208.4121 Ta2O5 150.0 69.5122 SiO2 150.0 207.0123 Ta2O5 150.0 70.7124 SiO2 150.0 205.5125 Ta2O5 150.0 72.0126 SiO2 150.0 204.1127 Ta2O5 150.0 73.2128 SiO2 150.0 202.6129 Ta2O5 150.0 74.4
227
3-1 3-2 3-3 5-1 5-2 5-3 5-4 5-5 5-6130 SiO2 150.0 201.2131 Ta2O5 150.0 75.7132 SiO2 150.0 199.7133 Ta2O5 150.0 76.9134 SiO2 150.0 198.2135 Ta2O5 150.0 78.1136 SiO2 150.0 196.8137 Ta2O5 150.0 79.4138 SiO2 150.0 195.3139 Ta2O5 150.0 80.6140 SiO2 150.0 193.9141 Ta2O5 150.0 81.8142 SiO2 150.0 192.4143 Ta2O5 150.0 83.1144 SiO2 150.0 191.0145 Ta2O5 150.0 84.3146 SiO2 150.0 189.5147 Ta2O5 150.0 85.5148 SiO2 150.0 188.1149 Ta2O5 150.0 86.8150 SiO2 150.0 186.6151 Ta2O5 150.0 88.0152 SiO2 150.0 185.2153 Ta2O5 150.0 89.3154 SiO2 150.0 183.7155 Ta2O5 150.0 90.5156 SiO2 150.0 182.3157 Ta2O5 150.0 91.7158 SiO2 150.0 180.8159 Ta2O5 150.0 93.0160 SiO2 150.0 179.4161 Ta2O5 150.0 94.2162 SiO2 150.0 177.9163 Ta2O5 150.0 95.5164 SiO2 150.0 176.5165 Ta2O5 150.0 96.7166 SiO2 150.0 175.0167 Ta2O5 150.0 98.0168 SiO2 150.0 173.6169 Ta2O5 150.0 99.2170 SiO2 150.0 172.2171 Ta2O5 150.0 100.5172 SiO2 150.0 170.7173 Ta2O5 150.0 101.7174 SiO2 150.0 169.3175 Ta2O5 150.0 103.0176 SiO2 150.0 167.8177 Ta2O5 150.0 104.2178 SiO2 150.0 166.4179 Ta2O5 150.0 105.5180 SiO2 150.0 165.0181 Ta2O5 150.0 106.7182 SiO2 150.0 164.4183 Ta2O5 150.0 107.4184 SiO2 150.0 164.6185 Ta2O5 150.0 107.6186 SiO2 150.0 164.8187 Ta2O5 150.0 107.7188 SiO2 150.0 165.1189 Ta2O5 150.0 107.9190 SiO2 150.0 165.3191 Ta2O5 150.0 108.1192 SiO2 150.0 165.6193 Ta2O5 150.0 108.2194 SiO2 150.0 165.8195 Ta2O5 150.0 108.4196 SiO2 150.0 166.1197 Ta2O5 150.0 108.5198 SiO2 150.0 166.3199 Ta2O5 150.0 108.7200 SiO2 150.0 166.6201 Ta2O5 108.9
Table G.1. Layer Composition of the structures appearing in Chapters 3 to 5.
228
6-1 6-2 7-1 7-2 8-1 8-2 8-3 10-1 10-133-layerGires-
Tournois
33-layerCoupled-
Cavity
66-layerNumericallyOptimized(Designed)
66-layerNumericallyOptimized,
(Fabricated)
200-layerPeriodic
200-layerNon-PeriodicLinear Angle
200-layerNon-PeriodicLinear Shift
100-layerFour-Step
Design
66-layerThree-Step
Design
810-850 842-854 1510-1580 815-845 1520-1570 1520-1570 1520-1570 1500-1580 810-850
45° 54° 45° 54° 40° 40° 40° 45° 48°
p-pol. s-pol. p-pol. p-pol. p-pol. p-pol. p-pol. p-pol. p-pol.
8.0 15.4 23.7 13.4 52.2 48.2 48.9 33.1 13.4
Air Quartz Quartz Quartz Quartz Quartz Quartz Quartz Quartz1 Ta2O5 337.0 188.0 261.2 360.0 404.8 188.02 SiO2 167.0 168.3 485.8 276.6 261.2 0.8 473.7 656.8 276.63 Ta2O5 109.8 2954.0 337.0 188.0 261.2 415.5 20.3 837.2 188.04 SiO2 167.0 168.3 485.8 276.6 261.2 20.0 465.0 643.9 276.65 Ta2O5 109.8 2967.0 337.0 188.0 261.2 378.6 11.5 439.5 188.06 SiO2 167.0 168.3 485.8 276.6 261.2 40.9 453.6 672.6 276.67 Ta2O5 109.8 118.9 372.7 208.0 261.2 411.2 14.5 492.8 208.08 SiO2 167.0 168.3 668.3 380.6 261.2 6.2 482.4 673.5 380.69 Ta2O5 109.8 2728.0 346.9 194.0 261.2 482.2 27.5 456.9 194.010 SiO2 3577.7 168.3 668.5 380.6 261.2 25.5 449.0 556.9 380.611 Ta2O5 109.8 118.9 391.0 218.0 261.2 568.6 19.0 418.2 218.012 SiO2 167.0 168.3 562.1 320.3 261.2 9.9 483.2 531.8 320.313 Ta2O5 109.8 118.9 447.7 250.0 261.2 442.7 21.0 357.3 250.014 SiO2 167.0 168.3 518.7 295.4 261.2 11.8 476.7 576.5 295.415 Ta2O5 109.8 2495.0 497.3 278.0 261.2 548.2 22.0 409.2 278.016 SiO2 167.0 168.3 484.8 276.6 261.2 13.8 502.1 462.0 276.617 Ta2O5 109.8 118.9 812.1 453.0 261.2 361.2 22.6 404.8 453.018 SiO2 167.0 168.3 502.2 286.0 261.2 15.7 511.8 560.1 286.019 Ta2O5 109.8 118.9 810.7 453.0 261.2 578.4 24.0 420.4 453.020 SiO2 167.0 168.3 462.3 263.1 261.2 0.2 520.1 462.1 263.121 Ta2O5 109.8 118.9 421.1 235.0 261.2 487.3 15.9 416.0 235.022 SiO2 167.0 168.3 546.2 311.0 261.2 56.2 496.9 504.3 311.023 Ta2O5 109.8 118.9 439.0 245.0 261.2 577.3 18.6 441.7 245.024 SiO2 167.0 168.3 507.9 289.1 261.2 21.5 494.3 485.7 289.125 Ta2O5 109.8 118.9 446.8 249.0 261.2 480.7 21.5 437.4 249.026 SiO2 167.0 168.3 485.6 276.6 261.2 23.5 471.1 408.3 276.627 Ta2O5 109.8 118.9 454.7 254.0 261.2 478.2 25.5 453.0 254.028 SiO2 167.0 168.3 459.4 262.1 261.2 25.5 541.5 453.0 262.129 Ta2O5 109.8 118.9 462.7 258.0 261.2 475.8 37.5 457.7 258.030 SiO2 167.0 168.3 452.1 257.9 261.2 27.5 539.9 432.0 257.931 Ta2O5 109.8 118.9 470.6 263.0 261.2 509.9 48.8 475.4 263.032 SiO2 167.0 168.3 440.8 251.7 261.2 29.5 526.0 392.5 251.733 Ta2O5 109.8 118.9 106.8 60.0 261.2 544.1 38.7 491.1 60.034 SiO2 167.0 168.3 422.6 241.3 261.2 68.1 502.8 463.2 241.335 Ta2O5 103.9 58.0 261.2 505.1 28.8 475.9 58.036 SiO2 401.1 228.8 261.2 33.5 478.1 387.1 228.837 Ta2O5 101.9 57.0 261.2 502.6 30.0 83.8 57.038 SiO2 387.9 221.5 261.2 35.5 466.3 415.1 221.539 Ta2O5 110.0 61.0 261.2 463.6 32.4 89.6 61.040 SiO2 385.7 219.4 261.2 37.5 474.2 405.9 219.441 Ta2O5 118.1 66.0 261.2 461.2 36.1 95.5 66.042 SiO2 363.1 207.0 261.2 39.6 472.4 393.8 207.043 Ta2O5 126.2 70.0 261.2 493.6 40.2 101.9 70.044 SiO2 350.4 199.7 261.2 41.6 491.5 409.3 199.745 Ta2O5 134.4 75.0 261.2 438.9 54.3 107.1 75.046 SiO2 348.2 198.6 261.2 61.0 509.1 422.5 198.647 Ta2O5 142.5 80.0 261.2 558.0 67.7 113.0 80.048 SiO2 325.2 185.1 261.2 82.2 466.6 334.1 185.149 Ta2O5 150.7 84.0 261.2 578.4 47.5 118.8 84.050 SiO2 286.1 163.3 261.2 84.3 442.4 371.1 163.351 Ta2O5 158.9 89.0 261.2 448.8 46.9 124.6 89.052 SiO2 301.1 171.6 261.2 49.7 459.3 362.6 171.653 Ta2O5 167.2 93.0 261.2 428.9 46.9 130.5 93.054 SiO2 344.6 196.6 261.2 51.8 456.4 307.2 196.655 Ta2O5 175.4 98.0 261.2 443.8 47.9 136.3 98.056 SiO2 289.0 164.3 261.2 53.8 433.9 365.6 164.3
Total Stack Thickness
in µmSubstrate
Description
Operatingwavelengthrange in nm
Incidence Angle in Vacuum
Polarization
229
6-1 6-2 7-1 7-2 8-1 8-2 8-3 10-1 10-157 Ta2O5 175.4 98.0 261.2 441.3 50.8 142.2 98.058 SiO2 253.5 144.6 261.2 55.9 452.3 345.6 144.659 Ta2O5 175.4 98.0 261.2 457.9 53.7 148.1 98.060 SiO2 361.3 205.9 261.2 57.9 450.2 347.6 205.961 Ta2O5 175.4 98.0 261.2 455.4 67.6 154.0 98.062 SiO2 346.7 197.6 261.2 79.2 429.6 317.6 197.663 Ta2O5 195.4 109.0 261.2 433.8 71.4 160.0 109.064 SiO2 307.6 175.8 261.2 81.2 456.0 308.6 175.865 Ta2O5 186.3 104.0 261.2 319.6 65.2 165.9 104.066 SiO2 100.3 57.2 261.2 81.6 433.2 299.5 57.267 Ta2O5 261.2 488.2 66.9 171.968 SiO2 261.2 66.2 431.1 290.569 Ta2O5 261.2 426.2 68.2 177.970 SiO2 261.2 68.3 418.1 281.471 Ta2O5 261.2 368.0 69.0 177.972 SiO2 261.2 53.0 425.0 248.373 Ta2O5 261.2 384.7 70.1 177.974 SiO2 261.2 55.1 422.1 281.575 Ta2O5 261.2 436.1 71.7 178.076 SiO2 261.2 92.0 439.7 291.577 Ta2O5 261.2 487.6 84.0 178.078 SiO2 261.2 41.8 428.3 248.279 Ta2O5 261.2 377.1 76.3 178.080 SiO2 261.2 96.2 426.9 305.781 Ta2O5 261.2 482.6 89.0 200.382 SiO2 261.2 98.2 434.6 304.583 Ta2O5 261.2 426.0 81.6 200.384 SiO2 261.2 82.9 410.8 304.585 Ta2O5 261.2 406.1 84.1 200.386 SiO2 261.2 85.0 408.2 304.587 Ta2O5 261.2 403.5 86.2 200.388 SiO2 261.2 87.1 405.2 304.589 Ta2O5 261.2 401.1 88.0 200.390 SiO2 261.2 89.2 413.1 304.591 Ta2O5 261.2 398.5 89.4 200.392 SiO2 261.2 91.3 400.4 304.593 Ta2O5 261.2 395.9 90.8 200.394 SiO2 261.2 93.5 406.5 304.595 Ta2O5 261.2 376.0 92.1 200.396 SiO2 261.2 95.6 394.7 304.597 Ta2O5 261.2 390.8 93.8 200.398 SiO2 261.2 97.7 400.4 304.599 Ta2O5 261.2 459.8 105.7 200.3100 SiO2 261.2 82.4 389.1 304.5101 Ta2O5 261.2 424.1 107.8102 SiO2 261.2 121.1 385.7103 Ta2O5 261.2 419.9 90.2104 SiO2 261.2 123.2 394.0105 Ta2O5 261.2 382.4 113.0106 SiO2 261.2 125.3 371.6107 Ta2O5 261.2 380.0 105.9108 SiO2 261.2 90.9 389.2109 Ta2O5 261.2 394.8 109.1110 SiO2 261.2 93.0 358.3111 Ta2O5 261.2 373.2 112.1112 SiO2 261.2 112.5 365.3113 Ta2O5 261.2 353.2 114.9114 SiO2 261.2 114.6 363.2115 Ta2O5 261.2 368.0 117.2116 SiO2 261.2 116.8 370.9117 Ta2O5 261.2 365.6 119.4118 SiO2 261.2 118.9 368.0119 Ta2O5 261.2 380.4 121.3120 SiO2 261.2 121.1 365.0121 Ta2O5 261.2 360.5 122.9122 SiO2 261.2 123.2 352.2123 Ta2O5 261.2 357.9 124.1124 SiO2 261.2 125.3 349.2125 Ta2O5 261.2 355.5 125.3126 SiO2 261.2 127.5 345.8127 Ta2O5 261.2 352.8 126.6128 SiO2 261.2 129.6 333.0129 Ta2O5 261.2 404.4 128.1
230
6-1 6-2 7-1 7-2 8-1 8-2 8-3 10-1 10-1130 SiO2 261.2 149.2 329.3131 Ta2O5 261.2 328.6 130.1132 SiO2 261.2 133.9 327.6133 Ta2O5 261.2 273.8 132.2134 SiO2 261.2 136.1 354.9135 Ta2O5 261.2 396.7 134.6136 SiO2 261.2 138.2 342.7137 Ta2O5 261.2 286.1 147.3138 SiO2 261.2 122.9 327.3139 Ta2O5 261.2 374.3 159.8140 SiO2 261.2 142.5 326.7141 Ta2O5 261.2 317.7 151.9142 SiO2 261.2 144.7 304.1143 Ta2O5 261.2 278.7 144.4144 SiO2 261.2 129.4 323.2145 Ta2O5 261.2 403.2 146.7146 SiO2 261.2 149.0 320.8147 Ta2O5 261.2 310.2 139.2148 SiO2 261.2 151.1 318.6149 Ta2O5 261.2 290.1 141.8150 SiO2 261.2 153.3 316.2151 Ta2O5 261.2 468.8 144.5152 SiO2 261.2 155.5 314.0153 Ta2O5 261.2 283.4 146.9154 SiO2 261.2 157.6 311.5155 Ta2O5 261.2 300.1 149.1156 SiO2 261.2 142.4 278.0157 Ta2O5 261.2 314.9 151.5158 SiO2 261.2 161.9 347.1159 Ta2O5 261.2 275.9 163.1160 SiO2 261.2 200.6 272.7161 Ta2O5 261.2 273.3 155.2162 SiO2 261.2 166.3 333.1163 Ta2O5 261.2 270.8 157.2164 SiO2 261.2 168.4 298.6165 Ta2O5 261.2 268.2 168.9166 SiO2 261.2 170.6 296.0167 Ta2O5 261.2 265.8 192.4168 SiO2 261.2 172.8 291.7169 Ta2O5 261.2 263.3 173.6170 SiO2 261.2 138.5 279.8171 Ta2O5 261.2 260.7 175.6172 SiO2 261.2 177.1 288.2173 Ta2O5 261.2 223.4 177.6174 SiO2 261.2 142.8 285.6175 Ta2O5 261.2 255.7 169.9176 SiO2 261.2 144.9 261.9177 Ta2O5 261.2 253.2 171.7178 SiO2 261.2 147.2 280.5179 Ta2O5 261.2 250.8 163.8180 SiO2 261.2 149.3 246.9181 Ta2O5 261.2 286.3 176.0182 SiO2 261.2 187.0 267.0183 Ta2O5 261.2 286.7 177.3184 SiO2 261.2 187.4 277.5185 Ta2O5 261.2 304.4 177.8186 SiO2 261.2 187.6 277.7187 Ta2O5 261.2 324.1 178.1188 SiO2 261.2 187.9 278.3189 Ta2O5 261.2 230.2 178.4190 SiO2 261.2 188.3 278.7191 Ta2O5 261.2 271.0 188.8192 SiO2 261.2 188.4 290.3193 Ta2O5 261.2 290.7 178.9194 SiO2 261.2 188.8 289.8195 Ta2O5 261.2 212.5 179.4196 SiO2 261.2 187.2 267.3197 Ta2O5 261.2 52.6 169.6198 SiO2 261.2 29.1 273.9199 Ta2O5 261.2 164.8 201.0200 SiO2 261.2 170.5 277.5201 Ta2O5 161.5
Table G.2. Layer Composition of the structures appearing in Chapters 6 to 10.