/smash/get/diva2:21791/FULLTEXT01.pdf

/smash/get/diva2:21791/FULLTEXT01.pdf
Linköping Studies in Science and Technology
Dissertation No. 1011
ORMOCER Materials
Characterization, LAP- &
Micro-Processing
—
Applied to Optical Interconnects and
High-Frequency Packaging
Steffen Uhlig
Departmentt of Science and Technology
Linköping University, SE-601 74 Norrköping, Sweden
Norrköping 2006
Author:
Titel:
Uhlig, Steffen
ORMOCER Materials Characterization, LAP- & Micro-Processing - Applied
to Optical Interconnects and High-Frequency Packaging
ISBN:
91-85523-99-2
Series:
Linköping studies in science and technology.
Dissertations No. 1011
ISSN:
0345-7524
Printed by LiU-Tryck (Campus Valla, Linköping) in 2006.
Alla människor är till för att öva på,
framförallt sin egen tålamod!
Kaleida Willén, 2005.
Abstract
R
ORMOCER
s are organic-inorganic hybrid polymers. Since their material properties can be tailored precisely during synthesis, they are suitable for a wide range of
applications in dielectric and optical microelectronics. This thesis reports on process
R
s for Sequentially Build-Up (SBU) test vehicles, suitdevelopment of ORMOCER
able for both electrical and optical interconnect. Furthermore, this work includes
materials characterization, such as refractive index studies (system B59:V32), optical loss measurements (systems B59:V32 and B59:B66), and surface characterization
through contact angle measurement and surface energy estimation (systems B59:V32
and B59:B66).
Process development for a high-frequency test vehicle was performed applying a
R
class. Dielectric layers in a
newly developed dielectric material of the ORMOCER
total thickness of 80 μm were build-up on a common FR4 substrate, applying photolithographic processes and moderate process temperatures of below 433 K. The loss
tangent and the permittivity of the material were measured to be 0.024 (loss tangent) and 3.05 (permittivity) over the entire frequency range 10 GHz to 40 GHz. The
compatibility of the material to standard processes of the PCB industry was proven.
Furthermore, a possibility for cost reduction in high-frequency MCM applications was
shown, through the possibility of using low-cost substrates.
The concept of a “flexible manufacture approach” for large-area panel optical backplane interconnects was introduced. Here, a 101.6 mm x 101.6 mm photolithographic
mask is to be stepped-out over a large-area panel substrate (up to 609.6 mm x
609.6 mm). The goal is to be able to create a large amount of continuous and unique
waveguide patterns over the whole area with a small portfolio of masks, thus being
able to minimize excess costs. In practice continuous waveguide patterns were created
over an area of 204.8 mm x 204.8 mm on a large-are panel (609.6 mm x 609.6 mm),
using a large-are mask aligner and a 101.6 mm x 101.6 mm waveguide mask. The optical loss of the waveguides was measured to be 0.6 dB/cm (B59:V32 material system,
λ =850 nm).
In connection to the large-area panel project a re-evaluation on the optical power
budget needed for high bit rate optical interconnects was performed. This work
was mainly based on literature surveys of optical waveguide materials, planar optical
amplifiers, light coupling structures, and planar light-routing structures. It was shown
that optical amplification is necessary at certain places on realistically routed optical
backplanes to boost the optical signal. Therefore, the concept of a flip-chip mountable optical amplifier (FOWA) device, based on planar optical waveguide amplifiers
and Semiconductor Optical Amplifiers, was developed. The device’s design allows an
independent manufacturing to the rest of the board and a mounting at key-positions
using standard pick and place technology. Additionally, it was observed that most of
the amplifier research is focused on the wavelength of 1310 nm and 1550 nm, whereas
optical backplane applications are targeting the 830 nm range.
During SBU processing of waveguide structures was discovered a de-wetting phenomenon of B59 resin on a cured B59:B66 and B59:V32 surface, respectively. Good
wetting behavior could be achieved by adding small amounts of B66 or V32, respectively, to the B59. Surface tension estimations on various compositions of the systems
B59:B66 and B59:V32 could not directly be correlated to the de-wetting phenomenon.
Furthermore, the optical loss properties of B59 were only affected to a minor degree
by adding B66 or V32.
The process route proposed is an efficient alternative to processes including surface activations steps, thus opening possibilities for large-area processing in PCB
industry, where surface activation steps, such as plasma activation or silanization, are
not available.
The process development, materials characterization, and reviews presented
provide a basis for further research on processes for high-performance electro/optical
backplane interconnects with focus on Large-Area Panel processing.
R
, sequential build-up, permittivity, dielectric loss, optiKeywords — ORMOCER
cal loss, process development, large-area panel processing, optical backplane, polymer
waveguide, optical waveguide amplifier
Acknowledgements
I would like to express my sincere gratitude to the following people:
First I would like to thank my supervisor Professor Mats Robertsson for giving me
the opportunity to perform this PhD study. I am grateful for valuable discussions
and constant support during my time being here.
I also wish to thank my colleague Christian Johansson for his contribution to the
practical work described in this thesis. I am thankful for the discussions we had regarding process problems and the dielectric characterizations he performed.
Furthermore, I would like to thank my mentors and co-supervisors, officially assigned
or unofficial and working on voluntary base:
I am indebted to Dr. Dag Andersson, former Director of Studies of the EPROPER
Program, for his support during the past years. Thank you for critical comments
during our discussions and for all your support.
I would like to thank Dr. Nathaniel Robinson for his support, inside and outside the
university life. Without your support I would never have come that far.
I would like to thank Dr. Arne Alping (Expert in Photonic Interconnects, Ericsson
Research, Gothenburg, Sweden) for valuable discussions and the open ear for my research problems.
I would like to thank the administration of the Department for Science and Technology for their support, especially Peter Värbrand, Pia Ruthgård, Mats Fahlmann,
Stan Miklavcic, and Sofie Lindesvik.
Without the help of Acreo AB’s (Norrköping) employees, the practical work described
in this thesis would have been impossible to conduct. I am deep grateful for the support and help of Lars Gustavsson, Bengt Råsander, and Peter Olsson (laboratory
technicians), Joacim Haglund (valuable discussions and training at laboratory equipment), Lennart Granlund (LAP processing, photolithography masks, the mysteries of
“framkalla och bakkalla”), Magnus Svensson (laboratory training), Xin Wang (laboratory training and valuable discussions) and Anurak Sawatdee (laboratory training).
There are countless other Acreo employees, to whom I am thankful for the opportunity to meet!
Ein herzliches Dankeschön gilt auch Dr. Michael Popall, Dr. Ruth Houbertz, Dr.
Lothar Fröhlich und Dipl.-Phys. Gerhard Domann (Fraunhofer Institut für Silicatforschung, Würzburg, Germany). Danke für Euer Vertrauen und Unterstützung und die
R
n für die Prozessentwickkostenlose Bereitstellung von Unmengen an ORMOCERE
I
Acknowledgements
lung. Ruth und Lothar, ich bin Euch zutiefst dankbar für alle “Telefonkonferenzen”
unter den letzten Jahren und euer offenes Ohr für jegliche Art von Porzessproblemen.
Ich möchte mich auch bei Dr. Henning Schröder und seinen Mitarbeitern Günter Lang,
Norbert Arndt-Staufenbiel und Jan Krissler (Fraunhofer Institut für Zuverlässigkeit
und Mikrointegration) für die Unterstützung bedanken.
Special thanks also to my co-workers at the department, especially the “Organic
Electronics Group”, where I became an unofficial member (“wanna-be”)! Thank you
for your support, your company, and for the opportunity to get to know you. Here,
I just like to drag a few of you, which got a special place in my memory and my
heart, into the cone of a spot light: Fredrik Jackobsson, Mari Degerman, Magarita
González, Payman Tehrani, and Martin Évàldßon.1
Roland Emanuelson and Lotta Bohlin are recognized for their efforts to explain the
meaning of the simple word focus to me. The word is growing in me and is starting
to become meaningful. Thank you for showing the door.
Finally, what is life without friends and family? — Very difficult! I would like
to thank my family and relatives back in Germany for their support! I would like
to thank especially Doreen Beska, Ewa & Torsten Engel, Jens Goldammer, Sascha
Kunze, André Müller, Christoph Sommergruber, and all my other friends who are
not mentioned here. You are not forgotten! Thank you Ida van der Woude, Günther
Haase, and Frédéric Cortat for letting me get to know you! It is always a great pleasure for me to pass an evening with playing board games at your place or to met you
somewhere else in my spare time. It is not so often that I get the possibility to speak
free and easily in German up here.
Christina Nilsson, I would like to thank you too. Every effect has a cause. This work
is in fact an effect; you are one of the causes of this.
Other people I met and that supported this work by just dropping one or two important statements (at least for me, you might consider them as unimportant) should be
mentioned here: Marika Immonen, Kerstin Johansen, Kaleida Willén, and Thomas
Flinck. Thank you for the opportunity to meet and to get to know you!
Last but not least, besides the science related work, there were some activities in
music communities from my side during the last years. There I met people that supported (indirectly) this work and broadened my view on other aspects of life. Here,
I would like to thank Johan Melin, Erland Marckwort, My Appelgren, and Rasmus
Ewehag, which showed me how to organize music events but also broadened my view
on electronic music. Furthermore, I would like to thank Kajsa Hellstrand, Mattias
Tervo, and Tommy Boije for all their love and support! Jan Warnstam and Mikael
Langer, thank you guys for introducing me to the deeper meaning of drum’n’bass.
Especially during the final stage of writing of this thesis, the Tuesday evenings with
your web-radio helped to cheer me up.
1 read:
II
Evaldsson
List of papers
Paper I Christian Johansson, Steffen Uhlig, Ola Tageman, Arne Alping, Joacim
Haglund, Mats Robertsson, Michael Popall, and Lothar Fröhlich. Microwave
circuits in multilayer inorganic-organic polymer thin film technology on laminate
substrates. IEEE Transactions on Advanced Packaging, 26(1):81–89, 2003.
—
Related to chapter 8 in this thesis.
Paper II Steffen Uhlig, Lothar Fröhlich, Miaoxiang Chen, Norbert Arndt-Staufenbiel,
Günter Lang, Henning Schröder, Ruth Houbertz, Michael Popall, and Mats
Robertsson. Polymer optical interconnects - a scalable large-area panel processing approach. IEEE Transactions on Advanced Packaging, 29(1):158–170,
February 2006.
—
Related to chapter 6 in this thesis.
Paper III Steffen Uhlig and Mats Robertsson. Limitations to and solutions for
optical loss in optical backplanes. Accepted for publication on Journal of
Lightwave Technology, January 2006.
—
Related to chapter 7 in this thesis.
Paper IV Steffen Uhlig, Gerhard Domann, Ruth Houbertz, Lothar Fröhlich, Henning Schröder, Jan Krissler, Günter Lang, and Mats Robertsson. Preventing of
de-wetting effects for inorganic-organic hybrid polymers applied in sequentially
build-up (SBU) technology without surface pre-treatments. Submitted to IEEE
Transactions on Electronics Packaging Manufacturing
—
Related to section 5.2 in this thesis.
III
List of papers
Author’s contribution to the papers:
Paper I 50% of the process development and 100% of the manufacturing of the SBU
demonstrator.
Paper II All experimental work, except the final design of the Waveguide Mask
and the optical loss measurements. All writing, except section Synthesis of
Materials.
Paper III Complete concept study and all writing.
Paper IV All experimental work, except the optical loss measurements. All writing,
except section Synthesis of Materials.
Related work, not included in this thesis:
Christian Johansson, Steffen Uhlig, Ola Tageman, Arne Alping, Joacim Haglund,
Mats Robertsson, Michael Popall, and Lothar Fröhlich. Microwave circuits in
R
thin film. Proceedings: IMAPS Nordic 2002.
multilayer ORMOCER
Steffen Uhlig, Mats Robertsson, Norbert Arndt-Staufenbiel, Günter Lang, Henning Schröder, Michael Popall* , Lothar Fröhlich, and Ruth Houbertz. LargeR
) for optiarea processing of inorganic-organic hybrid polymers (ORMOCER
cal backplane application. Proceedings of ICG Conference 2004, International
Congress on Glass. Japan, October 2004, Proceedings available on CD-ROM.
*
presenting author
Steffen Uhlig and Mats Robertsson. Flip chip mountable optical waveguide amplifier for optical backplane systems. Poster presentation and proceedings paper
at the 55th Electronic Components and Technology Conference (ECTC’05), Orlando, FL, USA, June 2005, pp. 1880–1887.
Ruth Houbertz* , Herbert Wolter, Peter Dannberg, Jesper Serbin, and Steffen
Uhlig. Advanced packaging materials for optical applications: bridging the
gap between nm-size structures and large-area panel processing. Proceedings of
SPIE — Photonics Packaging and Integration VI, vol. 6126, January 2006.
*
presenting author
IV
Nomenclature
αcr
critical angle of incidence for TIR
βcr
angle, connected to αcr
γ
surface tension in (1 mN/m)
γd
surface tension in (1 mN/m), dispersive part
γp
surface tension in (1 mN/m), polar part
γl
surface tension of a liquid in air in (1 mN/m)
γld
surface tension of a liquid in air in (1 mN/m), dispersive part
γlp
surface tension of a liquid in air in (1 mN/m), polar part
γsl
surface tension on boundary solid-liquid in (1 mN/m)
γsv
surface tension of a liquid in air in (1 mN/m)
d
γsv
surface tension of a liquid in air in (1 mN/m), dispersive part
p
γsv
surface tension of a liquid in air in (1 mN/m), polar part
λ
Wavelength in (1 nm)
θ
contact angle
ξ
angle of acceptance for an optical fiber, connected to αcr
A
area in (1 m2 )
BER
bit-error rate
BR
bit rate in (1/s)
BW
bandwidth in (1 Hz)
d
diameter in (1 m3 )
n
refractive index
NA
numerical apparture (no unit)
V
Nomenclature
T
absolute temperature in (1 K)
Tg
glass transition temperature in (1 K)
Tp
process temperature in (1 K)
V
volume in (1 m3 )
w
width in (1 m)
BER
Bit-Error Rate
CVD
Chemical Vapor Deposition
DMUX
De-Multiplexer
e/o
electro/optical
EMI
electromagnetic interference
FhG-IOF Institut für Angewandte Optik und Feinmechanik
FOWA
Flip-chip mountable Optical Waveguide Amplifier
FPGA
Field-Programmable Gate Array
FSOI
free-space optical interconnect
Gbps
Giga bits per second, bit-/datarate
i/o
input/output
IC
Integrated Circuit
LAP
Large-Area Panel
Mbps
Mega bits per second
MCM
Multi Chip Module
MCM-D
MCM based on deposited thin-fil layers, SBU-technology
MCM-L
MCM based on organic laminate technology, derived from PWB technology
MEMS
micro-electro-mechanical system
MOB
Metal Optical Bench
MQW
Multi Quantum Well
MT
mechanically transferable, connector for optical fiber ribbons
VI
Nomenclature
OE-MCM Opto-Electrical Multi Chip Module, converter unit for photons to electrons
and vice versa
PCB
Printed Circuit Board
PD
Photodiode
PIN
Photo Intrinsic diodes
POF
Polymer Optical Fiber
PWB
Printed Wiring Board
RIE
Reactive Ion Etch
SBU
Sequentially Build-Up
SOA
Semiconductor Optical Amplifier
SOC
System On a Chip
SOP
System On a Package
Tbps
Tera bits per second, see Gbps
TIR
Total Internal Reflection
VCSEL
Vertical Cavity Surface Emitting Laser
VII
Dem geneigten Leser,
der es tatsächlich
bis zur letzten Referenz
auf der letzten Seite durchhält!
Contents
Acknowledgements
I
List of papers
III
Nomenclature
V
1 Introduction
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
R
1.2 ORMOCER
s for optical and RF interconnects . . . . . . . . . . . .
1.3 Scope of thesis — Scientific contribution to the field . . . . . . . . . .
1
1
3
5
2 Optical Backplanes
2.1 The limitations of electrical & optical interconnects . . . . . .
2.2 The PCB industry and the optical backplane market . . . . .
2.3 Motivation — A closer look on bottleneck applications . . . .
2.4 Optical interconnects on wafer-level scale and on board-level .
2.5 Approaches for optical backplanes — An overview . . . . . .
2.6 Optical Backplane — Summary . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7
7
10
11
13
14
23
3 Functional Structures
3.1 General . . . . . . . . . . . . . . . .
3.2 Signal in- and out-coupling elements
3.3 Other approaches . . . . . . . . . . .
3.4 Functional Structures — Summary .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
25
25
26
37
39
4 Materials for Optical Interconnects
4.1 Materials . . . . . . . . . . . .
4.2 Materials processing . . . . . .
4.2.1 Guidelines . . . . . . . .
4.2.2 Deposition methods . .
4.2.3 Patterning techniques .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
41
41
56
56
57
60
5 Characterization
5.1 Optical characterization . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.1 Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.2 Refractive index measurements . . . . . . . . . . . . . . . . . .
67
67
67
69
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
XI
Contents
.
.
.
.
.
72
74
74
76
78
6 Large-Area Panel Processing
6.1 Philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2 The Large-Area Panel (LAP) optical backplane demonstrator . . . . .
6.3 Large-Area Panel processing — Summary . . . . . . . . . . . . . . . .
87
87
91
96
7 Amplification Devices for Optical Interconnects on Optical Backplanes
7.1 Problem Statement — Origin of optical loss on optical links . . . . . .
7.2 Amplification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.1 Waveguide amplifiers . . . . . . . . . . . . . . . . . . . . . . .
7.2.2 Semiconductor Optical Amplifiers (SOA) . . . . . . . . . . . .
7.3 The Flip Chip mountable Optical Waveguide Amplifier (FOWA) approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.3.2 Coupling of light from the pump-light source . . . . . . . . . .
7.3.3 Mounting of the SOA . . . . . . . . . . . . . . . . . . . . . . .
7.3.4 The optical periscope and the low loss coupling . . . . . . . . .
7.3.5 Mounting of the FOWA . . . . . . . . . . . . . . . . . . . . . .
7.3.6 Process issues . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.4 FOWA — Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . .
97
97
104
104
108
5.2
5.1.3 Bit Error Rate (BER) Test .
Surface tension . . . . . . . . . . . .
5.2.1 Theory . . . . . . . . . . . .
5.2.2 Equipment & procedure . . .
R
s
5.2.3 De-Wetting of ORMOCER
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
110
110
111
112
113
113
114
114
8 Dielectrics for RF Interconnects
117
8.1 The SBU microwave demonstrator (DONDO EMW) . . . . . . . . . . 117
8.2 Results & summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
9 Concluding Summary and Outlook
119
10 Bibliography
123
The Papers
Paper I .
Paper II .
Paper III
Paper IV
XII
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
147
149
161
195
237
1 Introduction
1.1 Background
Ulrich and Schaper [1] in 2003 made a statement, which fits well as an introduction
line here: “Open any . . . consumer electronic systems and what will you see? . . . A
circuit board, maybe two, on which are mounted a few integrated circuits and dozens
of tiny discrete devices — capacitors, resistors, and maybe a few inductors.”
The focus of this thesis is on these circuit boards, often plastic like boards, which
come in a green, a brown, or a blue colour. These will be found not only in today’s
consumer electronic products, such as computers, Personal Digital Assistants (PDAs),
mobile phones, audio and video equipment, but also in internet routers and highend computers. The focus of this work is not on the development of boards with
a specific function nor is it on the optimization of electrical communication paths
between components, by running simulations. Rather, it is focused on the process
and materials side of tomorrows high-end Printed Circuit Boards (PCB).
Depending on the size, and the place where they are applied, PCBs are classified
according this hierarchy:
• Backplanes
• Motherboards
• Printed Wiring/Circuit Boards (PWB/PCB).
In electronics industry these terms are used in connection to the packaging levels 1 to
3. The integrated chip, which leaves the semiconductor’s manufactor fab, is assembled alone in a plastic package or together with other chips into a multi chip module
(MCM), system on a chip (SOC) or system on a package (SOP) and thus forms the
packaging level 1. Several MCMs, SOCs, and SOPs can be mounted on a PWB and
form the package level 2. An example of this is a graphics card found in today’s
desktop computers. This card can further be mounted on a motherboard (package
level 3). Finally the motherboard can be mounted onto a backplane, which forms the
backbone of, e.g., internet routers or high-end computers (see Fig. 1.1) [2].
However, on each packaging level the communication between the sub-level components and a higher packaging level has to be ensured with a good signal resolution at
an appropriate speed.
1
1 Introduction
MCM,
1st level package
Chip
Wafer
?
j
Motherboard,
3rd level package
PCB assembly,
2nd level
package
j
Backplane,
3rd level
package
-
Figure 1.1: Packaging levels for products of the electronics industry, going from chip-level
towards rack-level, according to [2]. Consumer products usually stop at the
PCB- or motherboard-level.
2
R
1.2 ORMOCER
s for optical and RF interconnects
Example: Let us have a look at a typical starting-sequence of a computer program:
The user starts a fancy 3d program. The system gets the order to load data into
the main memory (package level 1). Afterwards the graphics card is commanded
to show an image. At first, data have to be transferred from the main memory to
the graphics card (package level 2) and further to the memory on the graphics
card (package level 1). Finally, the processor unit on the graphics card will
render an image and send it to the monitor output. The question arises: Where
is the bottleneck? The communication over the packaging level’s frontiers, i.e.,
chip-to-MCM-to-board-to-MCM-to-chip, is the limiting factor in this chain. Onchip communication will be approaching 10 GHz in the year 2010, whereas the
off-chip clock speed will approach 6 GHz at the same time, as predicted by
the ITRS consortium [3]. Actually, the bus clock-speed (also known as Front
Side Bus) on motherboards has reached 1 GHz [4], which is below one third of
today’s CPU clock speeds, produced by AMD [5] or Intel Corporation [6].
Three major research trends have been observed recently to push the limits of electrical interconnects further and to make products more consumer friendly (e.g., smaller
in size):
• optical backplanes and motherboards [7, 8],
• low-loss dielectric materials for high-speed thin film applications [9], and
• integral/integrated passives (for definitions see both, [2] and [10]).
Much research is thus focused on low-loss dielectrics and integral passives and integrated passives, because their implementation in modern consumer products is easy to
establish. The topic of high-k materials for thin film applications on PCBs is directly
connected to the integral/integrated passives. However, large research activities were
reported from Georgia Tech (Atlanta, USA) in this area [11–15].
The focus of this work is mainly on the first two points in the list above, although
also integral passives have been treated to some extend. Optical interconnects on
motherboard and backplane level can be established either by a guided optical wave
approach or by free space optical interconnects. Here, a planar optical waveguide
approach demonstrating the feasibility of the novel “flexible manufacture approach”
for optical backplanes is presented. Additionally, processing problems of optical backplanes and application limits will be discussed. Furthermore work on process development and material characterization of low-loss dielectric materials will be presented.
R
1.2 ORMOCER
s for optical and RF interconnects
Since around 1990, development and research for a new material class called
R
s is ongoing at the Fraunhofer Institute for Silicate Chemistry
ORMOCER
(FhG-ISC, Würzburg, Germany).
3
1 Introduction
R
Figure 1.2: Reaction scheme of an ORMOCER
resin: from pre-cursors to the resin.
Scheme according to [19].
Applications for this material class can be found in, e.g., large-area protective coatings [16], dental fillings [17], and opto-electronics [18]. The latter is the focus of the
present work.
R
s are hybrid inorganic-organic polymers, where the resin is prepared by
ORMOCER
a sol-gel poly-condensation of alkoxysilanes. A simplified reaction scheme is shown in
Fig. 1.2.
R
s resin, it will behave like
When a photo initiator is added to the ORMOCER
negative photoresist under UV exposure. An initial cross-link of the organic structures due to the exposure will be fulfilled by a thermal curing step at around 423 K.
Additionally, the material is stable up to 543 K and shows good adhesion to various
surfaces such as, oxides, FR4, glass, and Si/SiO2 [19]. The material-class is in certain
aspects superior to its competitors, e.g., methacrylates, BCB, Polyimide, and Epoxy.
Methacrylate-systems have difficulties to withstand high temperatures up to 543 K.
BCB and Polyimide need rather high process temperatures (above 473 K) and introduce high mechanical stress into the substrate, which may causes critical substrate
deformation but also risk for cracking of the polymer layer. Lastly, epoxies are a good
choice, but the dielectric (tan δ ∼0.015) and optical properties cannot compete with
R
s.
ORMOCER
ORMOCER R s with applications in opto-electronics have been reported with low optical loss values (0.2–0.3 dB/cm at 1320 nm, 0.5–0.6 dB/cm at 1550 nm) [20] and
low dielectric loss values in the high-frequency domain (tan δ = 0.0035 for εr = 2.5
up to 40 GHz) [21]. The material is interesting for future MCM-Ds or similar SBUprocessing [2], where dielectric materials with these properties are needed to transmit
electrical signals with low losses across the package. Furthermore, the possibility to
apply certain materials of this class, both as dielectric material [22] and waveguide
core material [23] makes it an interesting choice for hybrid electro/optical backplane
applications, as demonstrated in a small-scale approach [18].
R
In the present work, the focus is on process development of various ORMOCER
materials, and their characterization contact angle measurements, surface tension
estimations, and refractive index measurements. This work was done in close collab-
4
1.3 Scope of thesis — Scientific contribution to the field
oration with the FhG-ISC. Process details are given in sections 5.2, 6.2, and 8.1 and
in the papers attached in the Appendix.
1.3 Scope of thesis — Scientific contribution to the
field
Attached to this thesis are four papers, and a short description of their contribution
to the science field is given below. Furthermore, links are provided pointing at the
appropriate chapter where background information and motivation for the research
can be found.
SBU-Materials — process development and materials characterization: During
the DONDODEM EU-project a new dielectric materials for use in MCM applications was developed. Here, as a final task, process development to create a
four-layer SBU demonstrator with alternating dielectric/metal layers was perR
as dielectric material. One aim
formed, using a UV-patternable ORMOCER
of this task was to characterize the dielectric constant of the material at high
frequencies (1 MHz – 40 GHz) and to show a cost effective alternative for highfrequency packaging. For more information see chapter 8.1 and paper I.
Optical Backplanes — a flexible manufacture approach: Research is performed
worldwide on optical backplanes. An overview is given of the different approaches reported in literature. Based on this material, the concept of the
“flexible manufacture approach” was introduced. A central idea of the approach
is to step-out a set of small high-precision photolithography masks over a largearea panel. The design of the mask-set is to be chosen in such a way, so that it
is possible to create unique, customer specific backplanes. Because the optical
backplane market seems to be more a niche-market than a mass-market, this
approach will give small companies the possibility to act with high flexibility. A
large-area demonstrator was created, showing optical interconnects over 20 cm
in length with an optical loss of 0.6 dB/cm (λ = 850 nm). For more information
see chapter 6 and paper II.
Limitations to and solutions for optical loss in optical backplanes: The
optical
power budget available on realistically routed optical backplanes based on planar
polymer waveguides was critically reviewed. As a result, it was recommended to
do an optical power budget calculation in the following way: firstly, sum all loss
values of functional structures on the link, secondly, determine the length of the
longest link, thirdly, chose the appropriate waveguide material. The outcome
of such a tentative calculation revealed that it is unavoidable to amplify the
signal along the link on realistically routed optical backplanes. Therefore, the
concept of a Flip Chip Mountable Optical Waveguide amplifier was proposed.
Well-known and established technologies for optical interconnects on wafer-levelscale were combined to accomplish this. The device is designed to be mounted
5
1 Introduction
on key-points to establish a suitable signal strength. The concept proposed is
compatible with pick-and-place and reflow processes of the PCB industry. Furthermore, high-speed electrical links and optical links were compared briefly.
The existence of a break-even length beyond which optical interconnects gets
superior on backplanes in technical aspects was revealed. For more information
see chapter 7 and paper III.
Alternative process route to prevent material incompatibilities during processing:
During the large-area panel optical backplane project (see paper II) a material
incompatibility between the core layer and the cladding layer was discovered
during processing of the optical waveguides. A workaround usually employs a
surface activation by silanization or plasma activation procedures, which where
not available for the large-area processing in the department’s cleanroom. An
alternative process route in this case was to make core and cladding material more similar by adding small amounts of refractive index tuning agent to
the (former pristine) core material layer. A straightforward method to detect
de-wetting behaviour was proposed. Furthermore, surface characterizations by
contact angle measurements and surface tension calculations were performed.
For more information see section 5.2 and paper IV.
In the following chapter an introduction and background to the papers attached
will be provided. It will start with an brief overview on optical backplane technologies
(chapter 2), followed by a review on functional structures necessary for light routing
issues in connection to this (chapter 3). Chapter 4 deals with polymer materials
suitable for planar optical waveguide applications, whereas chapter 5 starts of with
characterization techniques for optical properties of these materials. Furthermore,
surface tension measurement aspects, related to paper IV, will also be treated here.
In chapter 6, process description and results of the large-area panel work will be
presented, which is related to paper II found in the Appendix. In chapter 7 (related
to paper III) an introduction to an active optical amplifier device suitable for use
on optical backplanes is presented. Chapter 8 deals with process development and
R
materials, which is related to paper I.
dielectric characterization of ORMOCER
Finally, in chapter 9 the conclusions can be found.
6
2 Optical Backplanes
The terms bandwidth and bit rate are often interchangeably used in the context
of digital optics communication. The bit rate (BR) is equal to the amount of bits
transmitted per one second over a link. The term bandwidth (BW ) is equal to the
frequency range within which a signal can be transmitted without loosing its integrity.
For this thesis the definition BW = BR was chosen, to be consistent with fiberoptic communication terms, although there do exist modulation and signal encoding
technologies to transmit a higher bit rate at a given bandwidth [24].
2.1 The limitations of electrical & optical interconnects
Bottlenecks in today’s electronics are to a large extent associated with the chip-toboard communication and the off-chip communication in general (system bus). In
this context the terms the
The gap in performance between chips and packaging is increasing [2, 8, 25–27].
Industrial applications which have already reached or can reach this frontier very
soon include: internet routers [26], high-end computers [28], communication systems
within spacecraft, airplanes and cars [29–31], and radar [32]. In common applications
such as image processing and virtual reality more bit rate is also needed [25, 33].
Furthermore, biosequence processing, e.g. DNA sequence analysis, needs large bit
rates to perform analysis task in a reasonable time [34]. Applications with symmetric
multiprocessors units, i.e. where several processor units share one memory unit, demand high bit rate communication over a shared bus, where optics show advantages
over electric interconnects [8, 35].
The ITRS report from 2003 predicts a chip-to-board speed for electrical interconnects
up to 3 GHz for the year 2005 and approaching 9.5 GHz for the year 2010 [3]. Problems arise when trying to realise large system interconnect systems and high-speed
communication over long distance links up to the size of the board (usually 60 cm x
48 cm). Additionally, in the reports [25–32, 36] and in the text book of Tummala [2]
it is predicted that the optical communication speed on the board level will rise from
1–10 Gbps in 2004 up to 10–100 Gbps in 2011.
The physical limit of data transmission on Copper-wire-backplanes is reported
to be 2.5 Gbps*m [25, 28, 29, 36], whereas experiments showed that signal transmission at 8.7 Gbps*m [37] up to 15.25 Gbps*m [38] are possible, using duo-binary signal
encoding techniques. A drawback in this case is the high signal loss of 50 dB over
the link [38]. Although it was demonstrated that signal transmission is possible, the
solution presented is doubtful for practical applications. The signal power of -50 dB
7
2 Optical Backplanes
at the output side seems barely applicable in real-life systems.
In general, a high-frequency signal above 10 GHz along a Cu-link on a backplane
is attenuated through skin losses in the wire and dielectric losses in the substrate
[39]. Berger et al. [26] assumed a tolerable electrical power budget of 20 dB over
a 1 m link and calculated bandwidth x length products of 5 to 15 Gbps*m for loss
tangents ranging from 0.02 to 0.004. Other simulations showed that the total signal
attenuation for a 10 GHz electrical link on standard FR4 PCB material can be about
-50 dB over 1 m [39] or 0.5 m [40]. A partial solution to the problem is the use of
low-loss dielectric materials in backplanes, such as Rogers 4000 (tan δ = 0.004 [40],
tan δ <0.0037 (according to data sheet)) instead of FR4 (tan δ = 0.02). A gain in
electrical performance of 100% can be achieved by this choice, whereas the cost will
rise by a factor of 5 [40]. However, a solution for skin loss compensation has not
been presented so far. The current in a Copper wire at high frequencies is pushed
towards the surface of the conductor, thus increasing the resistance of the conductor.
Instead of a solution for the skin loss, different models were proposed in the literature
to integrate theses losses into the design of high frequencies circuits [41, 42].
Contrary to electrical high bit rate interconnects, high bit rate optical interconnects
were demonstrated on board level. A planar polymer waveguide link (length 1 m)
operating at 12.5 Gbps has been reported [43]. The link-loss was measured to be 0.04
dB/cm at 850 nm in this case, thus giving a total signal attenuation of 4 dB over the
link (excluding coupling losses).
Packaging density, signal integrity and immunity against electromagnetic interference (EMI) are crucial problems for electrical interconnects [44], especially for
military, avionic, and automobile applications. Products in these operational fields
should be insensitive to electromagnetic disturbances and should not emit radiation
themselves, which can be guaranteed through appropriate shielding of the devices. To
ensure signal integrity of an electrical link on backplanes some design rules have to be
fulfilled. The width of high-speed (10 Gbps) electrical wires on PCB’s was proposed
to be 500 μm [45]. This forces a distance of 750 μm to the neighboring wire to prevent
cross talk, according to a rule of thumb [46], which gives a pitch of 1.25 mm.
Alternative solutions for EMI and signal integrity of electrical links aspects are (planar) optical waveguides on backplanes [2]. For optical waveguides, dimensions of 70
μm x 70 μm were proposed based on considerations of the capabilities of pick-andplace equipment in the PCB industry [47]. These numbers lead to a distance 20 μm
between neighboring waveguides [48], which gives a pitch of 90 μm. Thus, the packaging density of optical waveguides is much higher than that of electrical links in this
case.
A widely discussed topic is the so called “break-even length” for optical and electrical interconnects (see [26, 45, 49]). The question which arises is: When does optical
interconnects get technically feasible at a given bit rate in connection to a limited
power budget? One answer to this question is based on the power consumption of
high-speed links, as discussed by different research groups for the electrical and the
optical case. Svensson [45] estimated the dissipated power of an optical link to be 4
8
2.1 The limitations of electrical & optical interconnects
W compared to 0.1 W for the electrical equivalent with a given link length of 10 cm
and a total bandwidth of 4 Tbps.
Cho et al. [49] calculated the power dissipation as a function of the link length and
the bit rate and estimated this way a break-even point, beyond which optical links
become feasible. At a given bit rate of 6 Gbps the critical length was estimated to be
43 cm for a dissipated electrical power of 26 mW. Furthermore, it was observed that
this critical length is reduced for higher bit rates and lower BER, therefore giving the
advantage for the optics.
These two examples show that a break-even length in terms of link length can be
determined when appropriate boundary conditions are chosen.
Break-even length calculations are often limited to wiring capacitance and signal attenuation considerations. Pappu and Apsel [50] have chosen a different approach by
taken into account the load capacitance of large-scale electrical signal fan out as well.
Applications of such schemes can be found in, e.g., multibit data paths, clock distribution, and fan out to multiple nodes, where the latency of the distributed electrical
signal is crucial when the system’s size expands. It is believed that optical interconnects on chip level has its eligibility and that break-even length calculations have to
be performed by considering both the load capacitance and the link length in this
case. Calculations for the delay and energy metrics of an optical interconnection fan
out driven by an electrical load showed that the break-even point for electrical fan
out is approximately 250 minimum sized inverters [50].
The economical side cannot be neglected either: The electron has to be converted
into a photon, transmitted over the link and reconverted to an electron on the other
side. This means, not only the design of the link has to be revised, additionally, laser
diodes (preferable Vertical Cavity Surface Emitting Lasers (VCSELs) until now), PDs,
and driver chips have to be included on the board. This introduces excess costs, which
have to be defended in the first place. Here, the question arises: Where is the breakeven point in system costs for resolution of a high bit rate electrical signal compared
to the link alternative based on optics? This question is, at least in public, unsolved
until now, although researcher groups [7, 39, 51] emphasize the importance of low
cost assembly of optical backplanes and the industry is currently looking for adequate
cost models [52].
In summary, the discussion focus’ on two subjects: optical interconnects beyond a
break-even length on backplanes and optical interconnects for clock distribution on
chip-level. When and where optics has to be chosen is strongly dependent on the
application and a decision has to be taken after a careful evaluation of its boundary
conditions in each case.
Solutions, in the form of hybrid electro-optical interconnect systems, e.g. as suggested by Naeemi et al. [40], Griese [53], and Holden [7] have been proposed, e.g. to
get rid of the EMI problem and increase the overall bandwidth of a system. Additionally, a re-evaluation of the frontiers of electrical interconnects in order to increase
the bit rate has been performed [45].
Huang et al. [39] point at that, although the viability of technical solutions for
9
2 Optical Backplanes
optical backplanes has been published, hardly one concept for their use in a computer
system can be found in the literature. Furthermore, the cost of replacing electrical
links with optical links must be reduced to make optics competitive in high-speed
and high-density applications from a market point of view [39]. Lukowicz et al. [33]
reasoned in a similar way and named inadequate and costly packaging technology, and
inadequate system architecture as the reasons for the small impact of optical interconnections on board or rack level in real-life systems. During the last 15 to 20 years
several approaches for optical backplane solutions have been proposed. The launch
of a production line was announced several times, although no breakthrough product
has been spotted on the market to date. However, research groups at universities all
over the world are focusing on the topic optical backplanes, and several companies
have started up research in this area as well, as outlined below.
2.2 The PCB industry and the optical backplane market
It is often stated and was recognized early, e.g. by Hartman et al. [51], that approaches
for optical backplanes should be compatible with standard production methods of the
PCB industry. In particular, the production of an optical backplane shall be regarded
as an add-on step in common PCB production, including the formation of the (optical)
waveguide layers and the positioning of the optical components using standard pickand-place technology. The lack of a successful product on the market can have at
least three reasons:
1. Optical backplanes are niche products and only interesting for a certain consumer group, i.e. high-end users.
2. The PCB industry is too conservative and not willing to change.
3. Technological solutions for optical backplanes are far too expensive to be suitable
for a normal SMT production line.
It is difficult to judge the importance of these factors on the industry’s position on
optical backplanes. The first factor above is related to the market volume. Unfortunately is it hard to do a market analysis on this topic. Information is available,
but only in the form of reports, provided by consulting companies such as Global
Information [54], ElectroniCast [55–58], and BPA Consulting [59]. The only public
information found so far are the reports of Montgomery [60] and Holden [7], and an
announcement on the internet [61]. The information given in these sources is summarized in Fig. 2.1.
Even if the forecasts look promising, it has to be taken into account that the optical
backplane market is strongly dependent on the welfare of the rest of the electronics
industry. For instance, a slow-down was observed on the optical backplane market
over the years 2003 to 2004, which has to be attributed to an overall slow-down in industry/economy and a focus towards high-end copper solutions [61], such as [37, 38].
10
2.3 Motivation — A closer look on bottleneck applications
Market Volume in Million $US
3000
2005−2009 b
2006−2008
c
2500
2003−2008 a
2000
1500
1000
500
0
Figure 2.1: Forecast on the volume of the optical backplane market. The bars are associated with a) Montgomery [60] based on ElectroniCast, from bottom to top:
North America, Europe, Japan/Pacific Area, b) information found on [61],
based on BPA Consulting [59], total market forecast for backplanes including
optical backplanes, c) Holden [7] based on ElectroniCast.
Furthermore, the report of BPA Consulting [59] seems to include the optical backplane market volume in the backplane market volume, whereby summary figures only
are presented in [61]. The volume presented here and in [7] are looking similar, but
there the optical backplane market only was taken into account. From these statements it is not totally clear which market volume can be expected. Looking at the
applications outlined above, it seems that all of them belong to special fields, representing a single technological frontier with a specific need for a solution. Furthermore,
high-end applications are often associated with high production costs. This lets one
conclude that a small amount of the backplane market volume will transform to an
optical backplane market, thus this market is considered to be a niche market. As
long as the acceptance at the customer’s side is limited in this way, the pressure and
demand to the industry is not strong enough to start up production lines. Especially
since the limits of electrical interconnects proposed in the 1980’s can still be stretched
a “few inches”.
2.3 Motivation — A closer look on bottleneck
applications
Motivation, technical approaches, and materials for optical interconnects on boardlevel can be found in numerous papers, of which only some recent review papers will
11
2 Optical Backplanes
be referred to.
The review of Forbes et al. [62] discusses the advantages of optical interconnects
on board level compared to electrical interconnects. Problems of electrical interconnects are the increasing gap between on- and off-chip bandwidth, frequency dependent
losses, cross talk, impedance discontinuities, and power consumption. Their tutorial
focuses on technology and production aspects. A short review regarding solutions for
optical interconnect approaches is given too.
The report of Berger et al. [26] from IBM Research (Zürich, Switzerland) defines
interconnect requirements for future commercial systems, such as Internet switches,
routers, and subsystems. Furthermore, optical interconnects for rack-to-rack and
board-level applications, based on waveguide and free-space optical interconnects, are
reviewed briefly. The report is intended as a roadmap for IBM research in this area
with a timeframe of 2006–2010. The expected limits of Copper-based solutions have
been studied from an applications point of view. Attenuation versus bit rate for
substrates with various dielectric losses (tan δ) has been estimated. With a limit of
tolerable attenuation of 20 dB for 100 cm interconnect length, the maximum bit rate
is 5–15 Gbps for tan δ = 0.02–0.004. Any EMI or crosstalk aspects that may further
limit the applicability of electrical interconnects were not considered.
Holden [7] summarizes the problems electrical interconnects are faced with on high
bit rate backplanes, such as EMI, signal integrity, pin-count density, and signal noise.
Optical interconnects are stated to be a solution for this problem, whereas requirements from the PCB processing point-of-view on these solutions are listed here. These
include technical solutions such as components, processes and design tools, but also
requirements on the materials properties, such as stability of materials to harsh environment conditions. Finally, several backplane approaches, such as the EOCB,
PolyGuide, TOPCat, and Truemode are presented briefly. The report is thought as
a starting point for further research on the area.
The HOLMS project [33], a cooperation between Heriot-Watt University (Scotland), ETH Zürich (Switzerland), THALES Communications (France), Siemens/CLab (Germany), University of Hagen (Germany), SUPELEC (France), University
of Paderborn (Germany), and ILFA AG (Germany), aims at making optical interconnects on board level feasible, by presenting an adequate system architecture and
feasible packaging technologies. It was pointed out that a lot of packaging approaches
exist, but hardly one complete system, which is handling specific tasks, could be
found. For this project, the memory latency in multiprocessor systems was considered to be the bottleneck in (high-end) applications, although doubts have been raised
from other research groups [39]. The HOLMS demonstrator is composed of four CPU
and sixteen memory banks distributed over two PCBs. It is intended that every CPU
should be connected optically to every memory chip. However, the goal of the project
is to demonstrate that “optical interconnection technology is capable (of) . . . providing system-level benefits for real life systems”. A possible target application would
be a large-size image handling equipment, e.g. for medical or radar research.
Ao et al. [34] demonstrated an application for high-speed data string comparison
12
2.4 Optical interconnects on wafer-level scale and on board-level
in biological sequence analysis for DNA analysis. The bottleneck in this case is the
interface between the processor unit and the system memory, where the scoring data
have to be compared with a sequence database. In this case parallel optical interconnects was chosen to overcome the bottleneck and a demonstrator was build to show
the feasibility of the approach.
As mentioned before, various industry and consumer groups welcome commercially
available optical backplanes [29, 30, 32, 63]. Copper wires may be exchanged for
optical waveguide or free-space interconnection methods.
A typical waveguide structure is shown in figure 2.2. Here, the core has a higher refractive index than the cladding to establish the condition of total internal reflection
(TIR). Light is confined within the waveguide and travel along the central axis, as
long as the TIR condition for the incident angle is established [24].
Incident
Light
-
1
q
-
Upper Cladding
Core
Under Cladding
Substrate
Figure 2.2: Waveguide structure containing under cladding, core, and upper cladding layer.
2.4 Optical interconnects on wafer-level scale and on
board-level
Optical/electrical interconnects is discussed in at least two different areas: fiber optical interconnects [8, 64, 65] and optical backplanes [7].
Today, Internet long-haul communication is established using optical fibers, whereas
Ethernet cards in computers still use electrons as input signal. The fiber optical interconnects area deals mostly with silicon bench technology for connecting the optical
fibers to a detector to transform photons into electrons. This technology is well developed and well established on the market. The silicon bench technology enables
high precision V-grooves to align optical fibers to a PD [66]. Furthermore functional
structures can be included on chips such as, De-Multiplexers (DMUX) [24, 67, 68]
and pre-amplifiers [69]. In optical fiber technology, information is simultaneously
13
2 Optical Backplanes
transmitted on several wavelengths around a center wavelength. DMUX split the optical signals of one waveguide, consisting of many different wavelengths, to separated
waveguides with one wavelength each [24, 67, 68]. Furthermore, the incoming signal
has often a rather low strength. Therefore a pre-amplification might be necessary and
can be provided by Semiconductor Optical Amplifiers (SOAs), which are integrated
on the PD chip [65, 69]. Target wavelength bands for these applications are around
1310 nm and 1550 nm, respectively.
In contrast to the fiber technology, optical backplanes are autonomous systems, i.e.
an optical connection to the outer world is the exception rather than the rule. An
application example is found in high-end computers. This encapsulation in communication to the outer world has the side effect that work is mostly focused on wavelengths
around 830 nm, due to the fact that low cost VCSELs and PDs are available and polymer materials have a low-loss optical window for this region.
As discussed further below, the focus of optical backplane interconnects is more on
large-area panel products, which removes some of the process technology possibilities
which exist on chip-level, such as reactive ion etch (RIE) and high-precision photolithography down to the sub-micrometer scale [2, 70]. This not only has an impact
on the processing itself and functional structures (as outlined below), it also reduces
the amount of materials available for optical waveguides, because some of them can
be patterned by RIE only. However, in this thesis the focus is primarily on optical backplanes, although a general overview on the patterning techniques, functional
structures, and waveguide materials suitable for both, wafer-level and large-area processing, will be given too.
2.5 Approaches for optical backplanes — An overview
All over the world, universities, research institutes, and companies have performed
extensive research on optical backplanes. Fig. 2.3 shows where research activities are
found, and the approaches are now described in detail.
In general, there are four main approaches to introducing optical interconnects for
PCB’s and backplanes, in addition to fiber solutions:
• free-space optical interconnect, including the bi-directional approach using holographic systems
• buried waveguides inside PCBs
• optical layer on top of PCBs
• flex-foil based optical interconnect
The definition for free-space optical interconnect (FSOI) is not straightforward.
For every approach in this group, however, the light beam is transmitted without
being confined in a waveguide structure (see Fig. 2.2 and Fig. 2.5 for comparison).
14
2.5 Approaches for optical backplanes — An overview
Optical Waveguide layer on PCB
Flex Foil
Buried
Waveguides
Free-Space
Optical Interconnects
Figure 2.3: Overview on research activities worldwide on the topic optical backplane, without claim on completeness.
Some demonstrators apply micro-mirror micro-electro-mechanical system (MEMS) to
redirect signals between different optical waveguides or fibers [32, 62, 71]. Others use
a transparent building block to guide the light in fan-out operations [33], optical relay
modules [72], or apply real free-space transmission over several centimeters [34].
However, activities in the field of FSOI using micro-mirror MEMS were reported, e.g.
from Halmstad University (Sweden) [32], Heriot-Watt University, Riccarton, Edinburgh (UK) [62] and the University of California, San Diego (USA) [71]. Fig. 2.5
show the principles of this technique, whereby in this case light beams between waveguides are redirected in free-space by employing micro-mirror MEMS.
The report of Agelis et al. [32] (part of the HiSPOT-project [73]) gives a starting point
for further research regarding FSOI. In this report, a model of optical interconnect
employing optical MEMS is presented. In a case study, the feasibility of an optical
interconnection system in a future parallel processing radar system is presented. The
report of Forbes et al. [62] shows a demonstrator for short link FSOI with 4096 data
channels and 250 Mbps data rate per channel. The main purpose of the described
EU-project was to show the compatibility of the multidisciplinary technologies needed
to couple opto-electronic free-space interconnects to silicon VLSI. The report of Esener
and Marchand [71] is primarily a review of their own work on a FSOI demonstrator
with 48 optical channels and data rates up to 800 Mbps. The intention of the review
is to present methods for realizing short path MCM-interconnect. Key parameters are
given for further development in the field, e.g. dimension of optical packages, overall
costs, and effective CAD tools.
Cohen and Jagadish [74] (Australian Department of Defense, Canberra, and Australian National University, Canberra, Australia) presented an alternative method to
micro-mirror MEMS switching. Here, a thermal lens is formed in the resonance cavity
15
2 Optical Backplanes
of a VCSEL. FSOI with a variable angle of ±30◦ was demonstrated and switching
speeds up to 2 GHz are reported to be possible.
The Department of Electrical and Computer engineering at the Colorado State University (USA) [34] demonstrated a bio-sequence processor based on FSOI. A FPGA
board was used to control the bio-sequence analysis chip, equipped with an 8 x 8 matrix of PDs, and the driver unit for the laser source. The optical pathway is composed
of Diffractive Optical Elements, beam splitters, and lens systems. The FPGA was
working at a speed of 166 KHz, an increase up to 20 MHz is possible by increasing
the laser power, whereas further performance increase up to 1 Gbps are dependent
on the availability of high-speed PDs.
Research activities on FSOI were reported from Canada too, here in particular from
McGill University (Montreal, Quebec) and the University of Toronto (Toronto, Ontario) [72]. The central idea was to present an optical node enabling 32 bit interconnect between four multiprocessors. Each node contains a chip bearing 256 transmitters and 256 receivers. The optical link is established over a ring structure between the
processors with optical signal add and drop nodes. The ring is built up of polarization
beam splitter modules, optical relay modules, and micro lenses. The target clock rate
for the optical transmission is 200 MHz, whereas the off-chip electrical frequency is
at 50 MHz. The main result is the high data density of 1250 channels/cm2 achieved.
In a key-structure of the HOLMS project [33, 75] planar integrated optical free-space
systems (PIOFS) are applied to establish a connection between an OE-MCM and
fiber optical links to a neighboring PCB. The PIOFS is a thin transparent layer on
top of the OE-MCM distributing the incoming and out going signals between the
OE-MCM and the fiber connector. The signals are coupled into the layer at a certain
place, and are distributed by back and forth reflection all over the area. This idea is
similar to the bi-directional approach shown in Fig. 2.4.
Similar to this approach is the so-called bi-directional or “shared bus” approach [76],
which was developed by the research group of R.T. Chen at the University of Texas
(Austin, USA). The model for the high-performance bus for multiprocessor systems
was proposed in 1995 [77]. The authors emphasize the compatibility to Futurebus+
and Multibus II. The carrying layer for optical information is a glass sheet. This sheet
works as the physical optical bus layer. Sub-system boards are mounted perpendicular to this sheet. Light is coupled in perpendicular to the glass sheet and is refracted
in opposite directions within the glass layer. Holographic gratings provide the in- and
out- coupling function of light. Fig. 2.4 show the principles of this technique. Alignment, power requirement and data transfer integrity issues were covered in [78, 79].
The alignment problem could be relaxed by applying collimating lenses. Laser input
powers should be between 3.0 μW to 60 μW. Using collimating lenses (f =1 mm)
and a beam pitch of 750 μm, a maximum interconnection distance of 9 cm by a BER
of 10−12 within the glass layer could be achieved [80]. A fully working demonstrator,
containing an optical wave guiding plate, two memory cards, and a processor card is
presented in [81]. High-speed data processing at 1.25 Gbit/s was achieved between
memory and processor. Recently, a demonstrator composed of one processor card and
16
2.5 Approaches for optical backplanes — An overview
4 memory cards was presented, showing the feasibility of the system [76].
The main drawback of FSOI is that it requires shielding the free-space optical path
to avoid, e.g. dust and is therefore less suitable for harsh environments.
Sub-systems
?
)
q
?
?
?
6
k
+
3
Diffractive
Optical
Element
6)
s
Bi-directional bus,
glass substrate
Figure 2.4: Principle of a bi-directional optical interconnect.
Figure 2.5: Free-Space Optical Interconnects (FSOI). Principle. On the left and right hand
sides the waveguide ends are visible. The light beams are re-directed in freespace with the help of optical MEMS. The arrows visualize the redirection of
the beams. The figure is adapted from drawings in [32].
17
2 Optical Backplanes
Buried waveguides for applications in connection to PCBs have been studied
by the Fraunhofer Institute for Reliability and Microintegration (FhG-IZM), Berlin
(Germany) [82], NTT Telecommunications Energy Laboratories, Kanagawa (Japan)
[83], IBM Zürich (Switzerland) [26], Terahertz Photonics Ltd. [84], the ETRI consortium in Korea [85–87], and the University of Dortmund [88] among others. Fig. 2.6
shows a principal sketch of this attempt.
The EOCB project was on-going for several years at the FhG-IZM [28, 89–91]. The
Substrate
Upper Cladding
Core
Under Cladding
Substrate
Figure 2.6: Buried waveguides.
main focus here was to bury/embed waveguide layers into PCBs. The waveguides
were made by hot embossing of polymer foils (see Fig. 2.7). These foils were used in
a PCB lamination process, forming the buried optical layer of the board. The length
of the waveguides was limited to 10 cm due to the dimensions of the shim. In the
article of Ishii et al. [83] a conceptual study is presented regarding NTT’s approach to
optical PCBs. Optical units such as VCSELs and Photo Intrinsic diodes (PINs) are
implemented together with silicon LSI in a single package. These packages are surface
mounted on a PCB board. The authors emphasize this technology, since it utilizes
state-of-the-art technology in the PCB industry. Lenses on the board and on the package collimate the optical beams. Such an interface is referred to as an “OptoBump”.
The light will be coupled by 90◦ beam deflection directly into the optical interlayer
in the PCB [92]. Position and coupling tolerances of ±50 μm by clear eye-diagrams
at 1.25 Gbps were reported. A discussion regarding further development of optical
i/o-packages (transmitter and receiver in separate packages) is given in [93] including
an update in 2003 including measurements on 30 mm long waveguides and a NRZ
signal transmission at 1.25 Gbps [94]. IBM Zürich in 2003 presented a roadmap-like
report giving motivational background for their interest in optical backplanes for Internet router applications [26]. First results of a signal transmission at 12.5 Gbps at
850 nm over a 50 μm x 50 μm waveguide, 1 m in length were published in 2003 [95].
18
2.5 Approaches for optical backplanes — An overview
(a) Bottom cladding material and metal shim
(top) before joining under pressure and heat.
(b) After joining and separation of cladding and
shim.
(c) Grooves in cladding material are filled with
waveguide core material.
(d) Finally, the structure is completed by top
cladding deposition.
Figure 2.7: Embossing method to produce waveguides [82].
19
2 Optical Backplanes
Recently, progress towards a full working demonstrator was reported [43, 96, 97].
In addition to these approaches, the Scottish company Terahertz had an optical backplane based on their TruemodeTM polymers line in their portfolio. The multimode
buried waveguide backplane was equipped with a proper optical in-/out-coupling device as well [84, 98]. Unfortunately, the company went bankrupt in 2002 [99], but
re-appeared shortly after under the name Exxelis [100].
The Electronics and Telecommunications Research Institute (ETRI, Korea) has reported another approach for optical backplanes [85]. The backplane itself contains
buried waveguides with a core size of 60 μm x 60 μm, prepared by the hot embossing
method [91]. The optical loss was measured to be 0.1 dB/cm at 850 nm. The daughter boards (PCBs) are mounted perpendicular to the backplane. The electro/optical
(e/o) signal conversion is performed on a metal optical bench (MOB) on the PCB and
contains parts such as access modules, driver chips, VCSEL arrays, multimode fiber
arrays, deflection prisms, and micro lens arrays. The conversion unit is assembled
together with a so-called “optical slot” which ensures the optical coupling between
PCB and backplane. A clear eye opening could be observed for 8 Gbps NRZ data
transmission between PCB and backplane [85]. An earlier report of this group showed,
that the optical loss in the optical slot is about 7 dB [101]. This is far too high if a
full link shall be established, when considered the reported power budget of 14 dB for
the whole link [85, 101]. For a full point-to-point link two of these optical slots have
to be passed, which already consumes the whole power budget and leaves no room
for further optical losses at all. A coupling solution with relaxed tolerances has been
reported recently, where the optical loss of an “optical slot” was reported to be below
2 dB [87]. Furthermore, the demonstration of a 10 Gbps link was reported from this
research group [86].
The University of Dortmund has reported on an embossing approach [88], where work
was performed related to the OptiCon project [102], which focuses on mass-market
applications. Multimode waveguides with a core cross-section of 70 μm x 70 μm and
12 cm in length were created by an embossing method, showing optical loss of 0.05
dB/cm at 850 nm. An extension to full board dimensions, i.e., 30 cm x 40 cm, is in
progress [103].
In the HOLMS project embedded multimode waveguides will be used too, whereas
the production method is still under discussion (photolithography vs. hot embossing)
[33]. The goal is it to manufacture the waveguide layer first and afterwards to laminate the layer together with the other layers to a complete PCB.
The University of Texas (Austin, USA) and Sanmina-SCI reported on a molding/
embossing approach for the creation of multimode waveguides and micro-mirrors in
flex-foils, which afterwards were to be used in PCB lamination process. The waveguides with a length of 50 mm and a core cross-section of 50 μm x 50 μm showed an
optical loss of 0.6 dB/cm at a wavelength of 850 nm. As core layer material SU-8 was
used, whereas the cladding layer material was TopasTM . A poly(dimethylsiloxane)
(PDMS) shim was used to form the waveguides through embossing [104].
Last but not least, the report of Frese et al. [105] should be mentioned. A short
20
2.5 Approaches for optical backplanes — An overview
Core
?
Upper Cladding
Under Cladding
Substrate
Figure 2.8: Waveguides consisting of under cladding, core and upper cladding layer on top
of a PCB.
review of applications for optical backplanes is given and the report focuses mostly
on a low loss passive optical star coupler for silica fibers with various diameter (loss
better than 3 dB, uniformity better than 2 dB). The LIGA [2] approach for realizing
in-/out coupling structures for buried waveguides was discussed with special attention
to alignment precision.
The topic optical waveguide layer on top of PCB-substrate was treated by
(among others) Daimler Chrysler Research Center, Ulm (Germany) [106], Bellcore,
New Jersey (USA) [51], Ghent University (Ghent, Belgium) [107], Helsinki University of Technology [108], Huazhong University of Science and Technology (Huazhong,
China) [109], Georgia Tech (Atlanta, USA) [110], and by the DONDOMCM EUprogram [18]. A typical outline of this concept is shown in figure 2.8.
In the report of Hartman et al. [51] critical points for this type of implementation
of optical interconnects are indicated. The space for optical circuit elements must be
minimized, optical waveguide layer creation should ideally be a direct extension to
the PCB fabrication process, and materials must be compatible with existing PCB
materials. Furthermore, a trade-off has to be made between coupling loss and the cost
of optical elements. To reach reasonable goals, the requirement of waveguide losses
was suggested to be ≤0.2 dB/cm. Finally, a multimode waveguide demonstrator was
evaluated, which was produced using a projection mask aligner.
R
Robertsson et al. [18] presented a hybrid e/o-MCM substrate. The ORMOCER
material used showed both good optical (optimized optical properties: 0.2 dB/cm at
λ =850 nm, 0.26 dB/cm at λ=1310 nm, blends of optimized materials for optic and
dielectric purpose: 0.68 dB/cm at λ =850 nm, 0.47 dB/cm at λ =1310 nm) and good
dielectric properties (typically εr 3.3, tan δ ≤ 0.004 between 1 kHz and 10 MHz).
Multimode waveguides (width x height x length: 50 μm x 20 μm x 20 mm) with
waveguide mirrors (45◦ ) produced by UV-excimer laser ablation were created. The
optical layers (cladding - core - cladding) were at the same time used as dielectric
21
2 Optical Backplanes
layers in the four metal layer structure of the demonstrator.
In addition to these groups, Moisel et al. [106] are working on their own version of
an optical backplane for avionic applications [29, 111]. Direct laser writing was used
to form multimode waveguides (250 μm x 200 μm [31, 44, 106] and additional 50 μm
x 50 μm [63]) on top of a PCB. The waveguides showed exceptionally low optical
loss (≤0.05 dB/cm at λ =840 nm) over a length of ≥1 m. In and out coupling to
VCSELs and PINs was solved via micro-mirror structures. Bit-error rates (BERs) for
10 Gbit/s are reported to be better than 10−12 [31]. These products were expected to
be available commercially in the year 2003 [63], but so far they have not been spotted
on the market.
The research group of Peter Van Daele at Ghent University (Ghent, Belgium) demonstrated in cooperation with FCI ’s-Hertogenbosch (The Netherlands), and the Department of Applied Physics of the Vrije Universiteeit Brussel (Brussels, Belgium) optical
interconnects on a PCB [107]. Multimode waveguides (length: 5 cm) with the core
dimensions 20 μm x 20 μm and 50 μm x 50 μm were created using photolithography
and laser ablation, respectively. The end facets, for coupling of light from a MTconnector, with an undercut angle of 45◦ were also created by laser ablation. The
laser ablation method is considered a proof of concept in this case for an inexpensive
and flexible manufacture approach. Results from optical loss measurements have not
been reported so far [107].
Another approach was reported from the Packaging Research Center at Georgia Tech
(Atlante, USA) [112]. Waveguides (length x width x height = 14 cm x 50 μm x 7 μm)
were created on top of a PCB, using photo-defining techniques. Their optical loss
was measured to be 0.52±0.11 dB/cm (at 1550 nm) and 0.24±0.08 dB/cm (at 1310
nm). In this approach, the VCSELs and the PDs were embedded into the board.
Blazed gratings, formed by photo-definition applying incoherent optics, were used as
coupling structures.
Research activities have been reported from Finland, especially from VTT Electronics,
Helsinki University of Technology, and Aspocomp Oy [108, 113]. Optical waveguides
were created on top of PCB substrates (10 cm x 10 cm) by photolithography. The
optical loss was measured to be 0.6±0.03 dB/cm at 850 nm [114]. Target bit rate in
this case is a 10 Gbps data transmission over optical waveguides on the board level
[115, 116].
Yoshimura et al. [117] (Fujitsu Computer Packaging Technologies, San Jose, USA)
proposed their idea of optical interconnects using a flex-foil approach. The structural concept is based on flexible fluorinated polyimide foils, containing optical multimode waveguides. MCMs in direct connection to VCSELs or PINs are mounted
directly on this foil. The distance between thinned modules are held as short as possible to ensure high bit rate electrical interconnects. Light is coupled directly into
optical waveguides via 45◦ deflection mirrors. The mirrors are made by direct laser
ablation. These foils are considered to be scalable modules and can be mounted directly on existing PCBs. Optical interconnects can be made in the x-y plane but even
in the z-direction through so called z-links. The driving force behind this project
22
2.6 Optical Backplane — Summary
is the minimization of excess cost introduced through handling/packaging of optical
elements.
2.6 Optical Backplane — Summary
A brief overview on optical backplane approaches was given in this section. Various
techniques and recent results were presented from research centers worldwide. Huang
et al. [39] and Lukowicz et al. [33] emphasized that there is barely a real-life system
available on the market, but rather many different concepts and solutions for key problems. This could be confirmed by scanning through the literature, as shown above.
Exceptions from the overall trend could be spotted with the HOLMS project [33]
and the bio-sequence analysis device [34]. In both cases a specific target application
and an exact bottleneck was presented. A working demonstrator has still to appear
(HOLMS) or was already presented (biosequence analysis). Furthermore it should be
considered carefully when and why it becomes necessary to take the step from electrical interconnects to the optical equivalent, by calculating break-even points, i.e. for
power consumption, signal attenuation, EMI questions, and cost issues.
It is often emphasized that optical backplane approaches should be compatible with
production steps and equipment of today’s PCB industry. This statement is obvious at a first glance, because the approaches can easily be adapted by the industry
and later an up-scaling of production volume to mass-market requirements is easily
done. But does it hold after closer scrutiny? Today, optical backplanes seem to be a
niche-market and no mass-market at all, even if there are demands for high bit rate
applications from different sides [29, 30, 32, 63]. The bottleneck described (memory
latency) is less important for the low-end mass-market products than for high-end
computers. The equipment park for production is available and can be used/adopted
by an optical backplane industry or, rather, an optical backplane manufacturer. It
is doubtful that the PCB industry will simply extend production lines for optical
backplane solutions, due to the fact that there is no obvious mass-market for them as
outlined in the introduction. Furthermore, optical backplanes would put a higher demand on a clean working environment (cleanroom facilities) for PCB manufacturing.
The question remains: Is the customer willing to pay the price for this high-end equipment or is he satisfied with the “not-so-good but good-enough” and much cheaper
equipment? It was proven in other research areas that customers tend to choose the
lower quality and cheaper product instead of the best technical solution (compare,
e.g., high-tech ceramic parts vs. steel equivalents, graphics cards with included sound
chip and without). It is not a path to success to restrict efforts to proving technological concepts and solutions for optical backplanes alone. Instead, the focus should be
more on complete systems, to demonstrate that optical interconnects applied in key
points/communication pathways can increase the data throughput of the whole system. However, without market pull by demanding customers, or a demonstrated cost
effective solution that removes bottlenecks in a real-life application, market success
will not be realized.
23
2 Optical Backplanes
24
3 Functional Structures
3.1 General
For an optical waveguide link, two things have to be consided: first of all the light has
to be coupled from the laser source into the waveguide and on the other side of the link
from the waveguide back to the PD. Different approaches found in literature for such
coupling structures will be presented and discussed briefly in section 3.2. Secondly,
although signals most likely will be distributed over point-to-point links on high bit
rate optical backplanes, routing structures may be necessary in order to guide the
signal from one source to different destinations simultaneously or for switching the
lights pathway. Hereafter, a brief discussion regarding these structures will be given,
although most of them have been demonstrated or are intended for use in connection
to silicon-bench technology.
MEMS switches, enabling 2-dimensional free space interconnects up to a 64 * 64
matrix have been reported. This device is based on silicon-on-insulator technology
and has shown an insertion loss below 4 dB [118]. Furthermore, waveguide splitters
can easily be created by photolithographic processes. A 1-to-8 splitter ideally shows
a loss of 9 dB on one output channel compared to the input power [67], whereas an
excess loss of 0.1 dB was reported for a single Y-waveguide splitter (50 μm x 50 μm,
at 850 nm) [43]. Recently, Polman and van Veggel [119] proposed a device approach
for a 1-to-4 loss-less splitter. Key feature of the device is an Erbium doped waveguide
amplifier (Al2 O3 host, working at 1530 nm), which boosts the incoming signal so that
the output signal can be spliced to 4 individual waveguides without additional losses
in optical power. The amplifier waveguide itself occupies the largest space on the
device, which could be reduced to 1 mm2 by an amplifier waveguide length of 4 cm
at the same time. A net gain of 2.3 dB was measured by a pump power of 10 mW.
However, this approach is mostly intended for use in fiber applications. Although the
concept presented might be applicable on optical backplanes too (see chapter 7.3),
the target wavelength is not suitable for optical backplane applications (850 nm vs.
1530 nm).
Polymer materials can show a thermo-optical (t-o) or electro-optical (e-o) effect.
This can be used for switching the light’s pathway in Y-splitters or -combiners [120].
The t-o effect coefficient is defined as the refractive index change caused by raising
the temperature by 1 K. The localized heat source is used to create this effect in an
appropriate material. These devices offer fully solid state switching without transformation of the optical signal into an electrical signal and back again. For further details
and explanations including sketches see the textbook of Mynbaev and Scheiner [24].
25
3 Functional Structures
Switches of e-o type have a speed advantage (up to 110 GHz switching speed [121]),
but require special materials with high e-o coefficients. The much slower t-o type
switches are much easier to control and to produce. Response time was reported to
be of the order of milliseconds. Examples were stated for Mach-Zehnder interferometers, e.g. with a switching time below 1 ms and a power consumption per trigger-pulse
of 70 mW [67]. A precise temperature control and wavelength control is necessary,
however. Recently, t-o waveguide switches have been reported with an insertion loss
of 1.8 dB and a switching time of less then 7 ms [122]. An optical modulator, based on
a Mach-Zehnder structure and a PMMA-DR1 (e-o active material) optical microring
resonator, with a switching speed well above 1 GHz has been reported [123].
Other important devices are Array Waveguide Gratings (AWGs), which are used
in applications were multiple wavelengths from different waveguides are joined to one
waveguide and vice versa. These devices are also known as multiplexers (MUX) and
de-multiplexers (DMUX) [24]. Recently, Ma et al. [67] gave a brief overview on devices based on passive and active polymer technology. Polymeric material in general
should show low birefringence. Furthermore, temperature independent AWGs based
on polymers are possible because their positive coefficient of thermal expansion can
be matched with their negative t-o effect. Here, the thermal expansion coefficient
is typically defined as the relative change in one physical dimension (e.g. x/y/z) by
raising the temperature by 1 K. Recently, an AWG device working with a penalty of
0.1 dB at a BER of 10−9 at 10 Gbps has been demonstrated [124].
In addition to switches, filters are also of interest. Wavelength tunable filters have
been realized. These devices are a kind of t-o switch; low birefringence of the material
and low crosstalk of the structure are key properties in this case. However, structures
with a material property of dn/dT = −3.1 ∗ 10−4 1/K resulting in a wavelength
resolution of -0.36 nm/K were reported. Finally, Variable Optical Attenuators
(VOA) have also been demonstrated. The purpose of such a device is to equalize
optical power in connection to MUX/DMUX structures and to control the gain of
amplifiers. Regarding non-polymer devices, power consumption is a big issue. Basing
on switching principles described above, these structures can be realized in polymers
with the help of t-o effect. Examples for low power consumption were given for, e.g.
an 8-channel VOA, 1.5 mW, 30 dB attenuation [67].
3.2 Signal in- and out-coupling elements
A brief overview shall be given regarding techniques for approaches for light coupling
structures. These optical elements are crucial to couple light in and out of planar
waveguides. Usually, low cost VCSEL and PIN will be used to transform electrical to
optical signals and vice versa. These components are usually surface mounted, which
implies that the light has to be turned about 90◦ to be coupled into and out of the
waveguide (see Fig. 3.1). One way to deflect the light into or out of the waveguide
plane is to use micro-mirrors with a tilting angle of 45◦ , but there are other approaches
as well.
26
3.2 Signal in- and out-coupling elements
Solder
bump
R
?
?
?
-
*
Deflection
prism
VCSEL
)
Uppercladding
Core
i
Undercladding
Substrate
Figure 3.1: Coupling of light in/out of waveguides. Principle.
The following eight approaches to create these functional structures, which involve
modification of the waveguide ends, could be found in the literature:
1. laser ablation [18, 107]
2. micro machining / high precision cutting technique by 90◦ V-shaped blade [83,
92]
3. reactive ion etching using masking [125]
4. reactive ion etching with tilted samples [126–128]
5. injection molding [105, 129]
6. replication techniques using glass mask and instant curing of waveguide material
to form micro prisms [130] including periscope model [131, 132]
7. collimation lens columns in connection to 45◦ tilted mirrors [114]
8. under-cut of waveguide ends using grey-scale photolithography [113, 115]
Additionally, two grating approaches have been reported in literature, where a modification of the waveguide end facet is not necessary (e.g., polish, chamfering):
1. diffraction / holographic gratings [81]
2. blazed gratings formed by incoherent optics [110]
27
3 Functional Structures
Furthermore, the following four device approaches, enabling low-loss coupling of light,
were found:
1. free-space optical coupling using backside mounting on PCB [26]
2. OptoPin approach [91] and butt-coupling approach [114]
3. Optical Slot [85] and micro lens array approach [114]
4. glued ball lenses on tilted mirror surfaces [114]
Kagami et al. [125] gave a short review on mirror forming techniques including a
discussion of their feasibility. Laser ablation and e-beam techniques were considered to
be highly time consuming, low-throughput approaches, and a high surface roughness
may result. Cutting techniques were considered applicable only in arrayed structures.
Finally, the feasibility of surface tension techniques, applying two polymers of different
surface tension, were discussed.
Recently, Karppinen et al. [114] compared three coupling methods with respect to
their optical loss caused by misalignment relatively to the optical axis. Methods
discussed in this paper were:
• butt-coupling scheme,
• stacked collimate micro lenses in connection to a 45◦ tilted waveguide end, and
• ball lenses glued on a tilted mirror surface.
The butt-coupling scheme is similar to the OptoPin approach explained further below
and shown in Fig. 3.9. In the case of the stacked collimate micro lenses array approach, the beam is focused through a lens-set on a 45◦ tilted waveguide end. For the
last coupling scheme listed above, ball lenses with a diameter of 250 μm were glued on
a tilted mirror surface. The mirror was made of polished glass and had an Aluminum
surface finish. All approaches evaluated are suitable for in- and out-coupling of light
to/from a waveguide.
The first and the last approach have the disadvantage that a hole has to be made
into the backplane, to enable sub-assembly mounting. Furthermore, smooth surfaces
have to be ensured for the waveguide ends to minimize coupling losses. The second
approach needs tilted waveguide ends, which can be produced by, e.g. the blade cutting method described further below.
All three approaches were applied in SMT assembly. Misalignment tolerances were
measured from the transmitter side collecting the light on the other side of the waveguide with a large core fiber. The minimum coupling loss under the condition of perfect
alignment was found to be 3 dB for the butt-coupling approach, whereas the micro
lens and micro ball lens approach both showed 7 dB. Misalignment in all three dimensions relative to the position of the light-source were investigated as well, resulting in
the tolerance levels for 1 dB and 3 dB losses. In Table 3.2.1 the misplacement tolerance for the coupling methods are reproduced, at least for the 3 dB penalty level. The
28
3.2 Signal in- and out-coupling elements
Table 3.2.1: Coupling loss for different light coupling approaches based on SMT technologies according to [114]. In the second column are presented loss values for perfect alignment of the coupling structures. An additional loss of 3 dB occurs
if the coupling structures are misaligned within the tolerances given in the
columns to the right.
Coupling
method
Butt-coupling
Micro lens array
Micro ball lens
Minimum Loss
in (1 dB)
3
7
7
-3 dB loss tolerances (in μm)
Horizontal Vertical Optical axis
±50
±30
±50
±40
±20
±25
+120
+120/-75
-75/+200
butt-coupling sub-assembly still seems to be the method of choice for light coupling,
due to the fact that the alignment tolerance for 6 dB coupling loss are rather large and
at the same time, the coupling loss is still lower compared to the other approaches.
Yoshimura et al. [92] produced 45◦ tilted mirrors on single and multimode waveguides using a simple blade edge cutting technique. A blade with a cutting angle
of 90◦ and very smooth side walls cuts a V-groove in the waveguide structures as
shown in Fig. 3.2. The waveguide end facets act as micro-mirrors and bend the light
about 90◦ . Without any further preparation, these mirrors showed optical loss values
of 0.1 dB for single-mode and 0.27 dB for multimode waveguides at 850 nm. Drawbacks of the method are that the waveguide has to be flipped over and that single
in-/out-couplers are difficult to produce. This process seems to be ideal for buried
waveguide approaches with VCSEL/PIN arrays in connection to waveguide arrays or
ribbons, such as in the EOCB-concept [82]. Successful application of this technique
was demonstrated in [133].
Ishii et al. [94] demonstrated a similar and further developed technique, called OptoBump. Key feature of this approach are two collimating lenses, created by UV-cured
polymer drops, whereby one of them is placed close to the VCSEL/PD and another
close to the TIR mirror on the board. These lenses help relaxing the coupling losses
due to the misalignment of the optical components. The VCSEL is mounted on a
carrier chip and covered by a transparent polymer to prevent back reflections from
the waveguide surface into the laser cavity. On the polymer’s surface is placed the
first collimating lens. The waveguide layer, e.g. a flexible foil, is sandwiched between
two PCBs. A 45◦ TIR mirror is created by a V-shaped 90◦ dicing saw. The PCB
is opened up at the places where the light shall be coupled into the waveguide and
a transparent polymer building block with a collimating lens is mounted there. The
pathway of light is illustrated in Fig. 3.3.
Due to the application of micro optical lenses (400 μm in diameter) in the system
and high precision micro solder bumps the misalignment tolerance could be relaxed
29
3 Functional Structures
to ±100 μm by creating a coupling loss of 1.5 dB at 850 nm wavelength and a core
cross-section of 50 μm x 50 μm. The minimum loss for perfect alignment was reported
to be 0.55 dB [94]. For better sketches see [134].
Kagami et al. [125] focused on surface technologies and individual positioning of
tilted mirrors created by reactive ion etching. An Aluminum layer was deposited
on top of the core layer of waveguides. This layer was patterned with holes of 10 μm
x 10 μm in size. These holes functioned as the etch mask during the applied RIE.
Parameters for the etch-chamber were chosen in such a way that anisotropic etching
occurred (see Fig. 3.4). A passivation layer was formed during the etch process at
the sidewall of the etched trench. If waveguide materials and substrate materials are
used with different CTE, local heating of the substrate can be used to close the edge
mask as shown in the report (CTE difference of two orders of magnitude). The samples were later annealed at approximately the glass transition temperature of the core
material. The etched trenches experienced a re-flow under these conditions. An optimum temperature range could be determined to form 45◦ tilted rectangular shaped
waveguide endings. An Aluminum layer was deposited afterwards to increase the reflection behavior. Optical loss was determined to be 0.1/0.3/0.7 dB at wavelengths of
650/850/1350 nm, respectively. Another approach was reported by Cook et al. [135].
On top of the core layer, a thick layer of photoresist was deposited (18 μm). Proximity exposure resulted in a 34◦ slope of the photoresist at the opening. The slope was
transformed by anisotropic RIE, resulting in a 45◦ slope at the core material. This
surface was later coated with a Cr-Au layer to improve the reflection behavior.
Liu et al. [127] has demonstrated micro-mirrors with the RIE approach too. The
application in this case was MCM intra-connect. The slanted end-facets were produced by tilting the samples about 45◦ in the etch chamber in connection with a
Faraday cage. The waveguides were masked with a metal layer. At the positions
suitable for the mirrors the layer was opened in an area of 50 μm x 50 μm. Precise
and direct loss values were not given. Koike et al. [126] applied this technique too.
Losses of ≤1.5 dB were reported for the mirrors. The RIE technique is applicable on
waver-level scale only due to limitations of large-are plasma etch chambers and the
flatness of large-area substrates.
An injection molding technique to form micro-mirrors was presented by Wiesmann
et al. [129] and Frese et al. [105]. Etch grooves were formed by anisotropic etching of
the (111)-plane of single crystal Silicon. On these masters Nickel was electroplated,
resulting in a shim for further embossing techniques. The Ni-shim was pressed into
the under-cladding polymer waveguide material, forming a tilted sidewall. A metal
deposition step was applied to increase its reflective behavior. Afterwards, waveguide
channels were formed by RIE processes [129]. The channels were filled with the core
material before upper cladding material was applied. Wiesmann et al. [129] reported
losses for the mirror structure of about 1 dB at 1300 nm. Frese et al. [105] optimized
this technology and minimized the losses during in-/out-coupling under the influence
of misalignment of VCSELs and PINs. The structure contains a buried deflection mirror and a cylindrical collimation lens. A top plate contains an additional collimation
30
3.2 Signal in- and out-coupling elements
(a) Cutting of a 45◦ slope at the end of the
waveguide, using a V-blade with 90◦ slope.
(b) Result. Substrate has to be flipped over
to perform in-/out-coupling. The vertical arrow represents the direction of the incident light
beam and the horizontal arrow the direction of
the deflected and in-coupled beam.
Figure 3.2: Blade cutting technique to form 45◦ tilted mirrors. For further information see
text and [92].
31
3 Functional Structures
VCSEL Carrier
Solder Bump
Motherboard
q
q
-
y
VCSEL
i
Transparent polymer block
Micro lens
Waveguide system
Figure 3.3: The OptoBump system according to [134]. The light from the VCSEL is first
to be focused on the upper and the lower lens, and then turned at the tilted
waveguide end by 90◦ through TIR to finally enter the optical waveguide,
composed of cladding and core layers.
Tilted sample inside
the plasma chamber
Sample after etch and
photoresist removal
RIE - Plasma
?
?
?
I
Mask
Polymer
Substrate
R
R
Figure 3.4: Forming of coupling mirrors by Reactive Ion Etching techniques. For further
information see text and [127].
32
3.2 Signal in- and out-coupling elements
Cylindrical
Collimating Lens
Collimating Lens
R
+
Deflection Prism
Cladding
I
Waveguide Core
Figure 3.5: Injection molding technique for micro-mirror creation. The spherical lens on
top focuses the incident beam in the prism. The prism itself deflects the light
beam about 90◦ . Finally, the cylindrical lens focuses the light into the waveguide. For further information see text and [105].
lens (see Fig. 3.5). The lenses are used for focusing the beam. Molding technologies
created the buried structures, whereas the shim was produced by the LIGA technology [2]. Displacement versus loss simulations were performed, investigating a lateral
misalignment of a 400 μm2 photo detector relative to a multimode waveguide with
250 μm x 250 μm. It was shown that at ±250 μm misalignment the signal would drop
about 3 dB. Experimentally, these results were confirmed by applying polymers for
cladding and core materials (Δn =0.032) and lenses with f =5 mm and f =10 mm
respectively (see Fig. 3.5). Simulations and experiments showed good agreement by
displacement about ±500 μm, resulting in a loss value of 3 dB.
Choi et al. [104] reported on a soft molding approach, using PDMS master to create the waveguide structures including the micro-mirrors. Coupling losses originating
form the 45◦ micro-mirrors were not reported so far. A replication approach using
R
was presented by Dannberg et al. [130]. Mithe UV-curing property of ORMOCER
cro prisms were produced but were unfortunately not characterised regarding optical
R
layer and
quality. A photo mask with relief was pressed into the liquid ORMOCER
immediately exposed using a Karl Suss MA6 BA6 mask aligner. Fig. 3.6 visualizes
this idea once more. The periscope approach by Robertsson [131, 132] is based on
this idea. Repetitive molding and surface finishing by metal deposition is used to
build up periscope-like structures on VCSELs or PINs. These optical components
can be easily mounted onto waveguide bearing boards. Even optical vias, opening
the possibilities towards true 3d interconnect of planar waveguides are possible using
this approach (see Fig. 3.7). In this way both optical interconnects in the x-y plane
and even in the z-direction would be possible between several waveguide layers on the
33
3 Functional Structures
(a) Substrate with liqud photo curable polymer on top. Relief glass mask above.
(b) Joint substrate and glass mask. UVlight is applied to cure polymer material instantously. Thus, the mask’s relief structure
is transfered into the polymer surface.
(c) Remaining structure on the substrate after mask separation.
Figure 3.6: Replication techniques, relief mask and mask aligner Karl Suss MA6. For
further information see text and [130].
PCB. Additionally, the alignment tolerance of the periscope can be enhanced by creating curved mirror instead of planar mirror surfaces, thus reflecting a focused beam.
Chang et al. [110] reported on a blazed polymer grating approach. The grating was
produced by incoherent illumination. The grooves showed a height of 2 μm by 250
lines/mm and a tilting angle of 36◦ . The structure can be illuminated at a perpendicular angle and will bend the light about 90◦ and couple it into the waveguide. The
process is currently adapted for low-multimode and single mode waveguides. Loss
values have not been reported so far [110]. Kirk et al. [136] proposed an optical via,
connecting two waveguides on different layers on an optical backplane, by applying
two such blazed grating structures on top of each other. An advantage of this approach is that the alignment problem is relaxed, due to the fact that the gratings are
34
3.2 Signal in- and out-coupling elements
produced by photolithography, which reduces misalignment usually below the 10 μm
mark (see section 6.2 for more information on large-area processing). In Fig. 3.8 a
blazed grating including the optical-via approach is shown.
Immonen et al. [113] demonstrated and characterized the optical loss of under-cut
TIR mirrors (see [115] for better photographs of the tilted waveguide ends), produced
by grey-scale photolithography [137]. A negative type photoresist was applied in this
case. This resist reacts and cross-links during the exposure to light. A grey tone
mask therefore let the polymer react partly from the surface down to the bottom of
the waveguide end and therefore an undercut mirror can be created, well suited for
optical coupling methods.
However, in case of Immonen et al. [113] SU-8-50 (Microchem Corporation) was used
as waveguide core material. The optical loss of the under cut mirrors was measured to
be 1.5 to 1.7 dB at 850 nm. A drawback of this method is, that no metal layer can be
applied on the tilted surface directly to enhance the reflectivity of the mirror. However, due to a large step in refractive index between the polymer waveguide core and
the surrounding air the condition for the TIR should hold, giving sufficient coupling
efficiency. On real backplanes the waveguide system is composed of under cladding,
core and upper cladding layer. During processing it may not be possible to prevent
liquid upper cladding material from penetrating under the under-cut mirror. Thus,
a boundary with a transition from a high refractive index region, e.g. n1 =1.55 (core
layer), towards a slightly lower refractive index region, e.g. n2 =1.54 (cladding layer),
is formed. The condition for TIR cannot be established for this step in refractive
index and an angle of 45◦ (actually, about 83◦ would be necessary). A solution for
this would be to prevent the penetration of cladding material by
• applying a negative/positive resist as cladding and making sure that the mirror
area is not/is definitely exposed, and
• creation of a protective wall with photoresist, e.g. core material, around the
mirror area so that cladding material cannot penetrate under the mirror during
the coating process.
This coupling method could be enhanced by application of lens systems, e.g. formed
by photolithography, to minimize coupling losses due to misalignment of the optical
components.
35
3 Functional Structures
)
)
6
-
-
i
i
Y
Y
Substrate
“Periscope”
Solder
bump
Uppercladding
Core II
Intermediatecladding
Core I
Undercladding
Figure 3.7: Principle structure of optical via / micro periscope. For further information
see text and [131, 132].
Core layer
Blazed Gratings
Cladding layers
?
?
?
- Pathway of light
-
....................................................................................................
....................................................................................................
..
............................................................................
. . Substrate ............................................................................
Figure 3.8: Blazed grating apporach, suitable for optical vias in 3d optical interconnects
on backplanes.
36
3.3 Other approaches
3.3 Other approaches
The OptoPin technique presented within the EOCB project [91] uses PIN and
VCSEL devices on daughter boards mounted to couple light in and out of buried
waveguides. A penetrating cavity in the optical backplane is opened and these daughter boards are mounted perpendicular to the main board. As described above, a similar structure was evaluated by Karppinen et al. [114] regarding coupling losses and
misalignment tolerances.
Another approach was presented by Berger et al. [26]. Here, a cavity in the optical
backplane was opened from the backside and a 45◦ mirror was introduced. The mirror
was used to ensure the 90◦ beam deflection between the buried waveguide and the
VCSEL/PIN. Fig. 3.9 and 3.10 visualize these approaches. Yoon et al. [85] presented
a concept called Optical Slot or “Optical Plug” (see Fig. 3.11). Light is coupled
from the multimode waveguides on a metal optical bench (mounted on the PCB and
containing the VCSEL/PD electronics) to a multimode fiber array. The fiber array
points at a 45◦ micro-mirror. On top of them an array of micro lenses is mounted
which focus the beam onto the polymer waveguide in the optical backplane. Positioning screws allow an alignment in x/y/z directions with ±2 μm tolerance. The optical
coupling loss was measured to be 7 dB as reported earlier [101], which was reduced
to 2 dB recently [87].
Figure 3.9: OptoPin. PCB with burried waveguides is opened. VCSEL, mounted on top
of a daughter board, is mounted in such a way, that direct coupling to the
waveguide is guaranteed. For further information see text and [91].
37
3 Functional Structures
Figure 3.10: Backside mounted mirror. A cavity is opened in a PCB with burried waveguides. From the backside of the board a mirror structure is introduced. A
VCSEL is mounted from the frontside. The mirror serves the 90◦ beam deflection from VCSEL (vertical beam path) to waveguide (horizontal beam
path). For further information see text and [26].
Processing board
Optical Slot
9
j
Multimode Fiber
Lens System
Optical backplane
R
Burried
Waveguide
-
y
Mounting Screws
Turning Mirror
:
Adaptor
Figure 3.11: Optical slot approach [87, 101]. Multimode waveguides are attached to a
metal optical bench. The daughter boards are mounted perpendicular to the
optical backplane.
38
3.4 Functional Structures — Summary
Kim et al. [80] used diffraction gratings in combination with micro lenses on top of
VCSEL/PIN to couple light into/out of a glass plate. Fig. 2.4 shows a sketch of the
beam path in this approach.
3.4 Functional Structures — Summary
In summary, there are many alternatives to produce structures to couple light from
the source into a waveguide and back to a detector. Tilted mirrors on waveguide end
facets can be formed by RIE techniques, blade cutting techniques, embossing techniques or gray scale photolithography. On the other hand, laser ablation techniques
seem to be too expensive and time consuming to be realized on optical backplanes.
Some approaches like RIE techniques or embossing approaches are only viable for
wafer-level interconnects because of equipment limitations for large-area applications.
Others like the OptoPin approach, blade approach, backside mounting of micromirrors, the blazed grating approach, and the periscope approach also seem to be
feasible for optical backplanes. The approach chosen depends on where the waveguides are placed; buried or on top.
The molding approach presented by Frese et al. [105] seems to be a very good choice
for industry because of the large misalignment tolerance. This would allow a fast and
inexpensive pick-and-place process for VCSEL and PIN devices. Molding techniques
may not be applicable directly on a large-area panel. The idea of horizontal and
vertical beam alignment through lens systems can be realized through forming of ball
lenses by surface tension effects [138] and forming of cylinder segments acting as collimation lenses by photolithography.
Furthermore, these lens forming techniques could be applied for the approach of under
cut mirrors [113] also, thus a significant relaxation of the alignment problem in SMT
processing may be achieved.
Additionally, the concept of the OptoBump is promising too, whereby the alignment
of the waveguide layer to the transparent building block has to be relaxed further. So
far, misalignment tolerances for the VCSEL of ±100 μm by a coupling loss of 1.5 dB,
have been reported.
Finally, in Table 3.4.1, a short summary of micro-mirror techniques and corresponding loss values found in the literature is given.
39
3 Functional Structures
Table 3.4.1: Optical losses of coupling structures for light coupling into or out of planar
waveguides (MM - multimode, SM - single-mode, HM - horizontal misalignment, VM - vertical misalignment). Loss above 3 dB by a misalignment of
the optical components of more then 50 μm, were regarded as not suitable for
practical use on optical backplane systems.
Technique
Coupling Loss in (1 dB)
Reference
Reactive Ion Etching
0.1 (λ=650 nm)
0.3 (λ=850 nm)
0.7 (λ=1350 nm)
[125]
[125]
[125]
b
b
b
Laser ablation (moving
aperture)
0.3
[139]
a
90◦ V-blade cutting
0.1 (SM)
0.27 (MM)
[92]
[92]
a,b
a
Grey-scale
photolithography
1.5 . . . 1.7 (λ=850 nm)
[113]
a
Injection molding
3 (±250 μm misalignment)
[105]
a
Micro-mirrors enhanced
by micro lenses
1 (±50 μm misalignment)
8 (±20 μm HM, ± 15μm VM)
[43, 134]
[114]
a
c
Butt-coupling
6 (±50 μm HM, ±40 μm VM)
[114]
c
Micro-mirrors enhanced
by ball lenses
8 (±40 μm HM, ±20 μm VM)
[114]
c
a
b
c
40
acceptable for optical backplanes
suitable for wafer-level scale only
not suitable for optical backplanes, see section 7.1
Remark
4 Materials for Optical Interconnects
4.1 Materials
In addition to solving the light coupling and signal routing aspects of optical interconnects, materials choice is crucial to achieving the desired performance. Important
material properties, which should be considered, are: low optical loss, low birefringence, low dispersion, and mechanical and thermal stability.
The report of Ma et al. [67] gives a good overview of available material systems, their
properties and processing. The report covers
1. conventional optical polymers such as poly-methylmethacrylate (PMMA), polystyrene, polycarbonate, polyurethane, epoxy-materials and
2. novel optical polymers such as dendrimer systems in fluorinated state, polyacrylates (deuterated and halogenated), fluorinated polyimides, PFCB aryl ether
R
R
, Teflon
AF, silicone, fluorinated poly(arylene ether
polymers, BCB, Cytop
sulfide), poly(pentafluorostyrene), and fluorinated hyper branched polymers as
well as
3. non-linear optical polymers were evaluated. Electro-optically active and amplifying materials are also covered.
Advantages and disadvantages with regards to optical properties and processabibilty
are illuminated in their report.
Pitois [140] in her thesis gives an overview about available materials for optical
waveguides. Optical and thermal properties of the following materials are presented:
PMMA, acrylates in fluorinated and deuterated state, polyimides in fluorinated and
halogenated state, acrylates and polysiloxanes (silicone resins).
Zhou [141] reviews polymer material classes useful for optical interconnects such as:
polyimides, acrylic polymers, polyethers and poly(ether ketone)s, PFCB, polycyanoR
rates, polysiloxanes (silicone resins), and commercially available polymers as Cytop
R
and Teflon
AF. The main focus in this work is on the basic chemistry, process parameters in material application, transmission behavior at 1550 nm, and birefringence.
A brief introduction and review of different passive and active polymers is also given.
Eldada [64] presents an overview on polymer and non-polymer materials (semiconductor, sol-gel, material grown by CVD processes) suitable for use in integrated optics.
Polymer materials such as acrylates, polyimides, benzocyclobutenes, arylene ether
sulfides, polysiloxanes and commercially available non-specified materials (OASIC,
IGP) are listed. Key parameters and properties for the materials are presented, such
41
4 Materials for Optical Interconnects
as optical loss at 800/1300/1550 nm, patterning techniques (photo-patterning, RIE
processes), and birefringence behaviour.
In addition to the review articles [64, 67, 141] and the thesis [140] mentioned above
some recent updates on recently developed materials will be given in this section. A
summary is presented in Table 4.1.1.
R
from the FhG-ISC
A very promising material class/category is the ORMOCER
[19, 142]. As mentioned in the literature [67, 140, 141], replacing -CH- groups with
-CF- or -CCl- can reduce optical absorption losses in the tele- and datacom transmission windows. Another method of achieving the same result would be to reduce
the -CH- content in the material by introduction of or exchange with silicone-groups.
R
materials, which were tailored in this way, were meaOptical losses of ORMOCER
sured to be 0.2–0.3 dB/cm at 1320 nm and 0.5–0.6 dB/cm at 1550 nm. The material
was not fluorinated in this case. Furthermore, this material can be used as dielectric
layer in high-frequency electronics packaging (see paper I) and is stable up to a temperature of 543 K.
Ballato et al. [143] discussed the optical loss mechanisms in polymer materials. A
lowering of the attenuation values can be obtained through lowering Tg , by exchanging every -CH with a -CF bond, and lowering of the overall refractive index, with
the drawback of a lowering of the thermo-mechanical stability of the material at the
same time. Furthermore it was pointed out that a chemical, mechanical, and thermal
compatibility between core and cladding material is necessary for a successful creation
of low loss optical waveguides, and therefore essential for acceptance for practical use.
Some important characteristics of these materials are: in general halogenated and/or
deuterated acrylates show very low losses around 840 nm ( [0.01;0.03] dB/cm) and
1310 nm ( [0.06;0.2] dB/cm). In contrast to this, the loss at 1550 nm can be
rather high with values of [0.25;1.7] dB/cm. Commercially available materials such
as TruemodeTM (“highly cross linked multifunctional acrylate”) [84, 99, 100] and
OASIC [67] are reaching these values too, but their chemistry was not specified.
The optical loss for deuterated polysiloxanes were reported to be 0.17 dB/cm at
1300 nm [67]. However, it has to be pointed out that the use of deuterated and
fluorinated/halogenated materials is doubtful, because
• Deuterium is extremely expensive.
• Chemical synthesises involving Fluorine is expensive; additionally, using this
element in organic synthesis will raise questions on environmental impact.
• Organic synthesis involving Halogens will raise questions on environmental impact.
The majority of loss values are in the order of magnitude of 0.01 dB/cm for the 840
nm region and 0.1 dB/cm for the 1310 and 1550 nm region. Of curse, exceptional
values crossing the border to lower and higher values can be found as shown in Table
4.1.1.
Many of the materials are optimized to give good transmission in the wavelength
42
4.1 Materials
domain of λ =1310/1550 nm (traditional wavelength bands for fiber optical interconnects).A large amount of optical loss measurement data is available here. Contrary to
this, optical backplane applications are focusing on the wavelengths around 840 nm,
because less expensive VCSELs and PDs are available. Furthermore, optical losses
tend to be lower around 840 nm compared to 1310/1550 nm and enabling therefore
a relaxation of the power budget limitations of an optical link. Unfortunately, the
amount of optical loss data available for polymer materials in the 840 nm domain is
limited.
Different opinions exist on the acceptance level for optical loss. Hartman et al. [51]
stated that losses below 0.2 dB/cm are sufficient for application on optical backplanes,
whereas Berger et al. [26] stated that 0.1 dB/cm is acceptable and Uhlig and Robertsson [144] concluded that loss values should be below 0.04 dB/cm (see even paper III).
The question of acceptance of loss values is strongly connected to the link length and
the optical power budget available, as outlined in section 7.3.1 and discussed in detail
in paper III. Firstly, the power budget available has to be calculated including all loss
factors caused by structures such as light couplers. Secondly, the length of the longest
link has to be defined, and thirdly, a material with an acceptable loss factor can be
chosen. For on-chip applications the situation is a bit more relaxed and materials
with higher losses might be applicable (e.g., <0.8 dB/cm), due to the fact that link
lengths are usually below 10 cm. On optical backplanes the situation is more critical,
due to the fact that link lengths can easily reach 1 m. Here, an optical loss of <0.08
dB/cm may be necessary for a point-to-point link, if a total coupling loss of 6 dB was
assumed (summary for in-/out-coupling of light).
An overview of optical loss values for polymer materials and the material patterning techniques is given in Table 4.1.1. The patterning technique “photoexposure”
refers to a standard photolithographic process, which can be applied for large-area
processing too. RIE refers to reactive ion etch which is mostly applicable for wafer
scale processes due to equipment limitations. Laser ablation is not listed since nearly
all materials can be laser ablated. A detailed description of patterning techniques is
presented in section 4.2.3.
Fig. 4.1, 4.2, and 4.3 visualizes plots of optical loss versus material class regarding
the wavelengths λ ={830;1310;1550} nm.
Besides the optical properties, the most important technical question for the manufacturer and customer is the maximal applicable process temperature. Low-cost substrate materials such as common FR4 a have rather low maximal process temperature.
These are usually in the range of T =[423;523] K. High performance materials exist
(i.e. Taconics RF-35), which have higher process temperatures (Tg ≥588 K), allowing
different process cycles during PCB manufacturing. For certain component mounting
steps, such as solder bump mounting of components, production process temperatures up to T =573 K can be required. Regarding the listed polymer/polymer–like
materials in this section, some temperature process guidelines are summarized in Table 4.1.2. The so-called glass transition temperature Tg is important in this context,
where amorphous materials soften depending on their degree of cross-linking [145].
43
4 Materials for Optical Interconnects
Above this temperature the mechanical properties, such as Young’s modulus and viscosity, will often change dramatically, which has an influence on the form stability
of the material. Furthermore, harsh temperature cycling around the glass transition
temperature during processing can induce mechanical stress in waveguides, which
leads to birefringence or even scattering losses due to (micro-)cracks in the waveguide
material.
Another important factor in this case would be the decomposition temperature. Most
of the materials listed in Table 4.1.2 can withstand PCB process temperatures (re-flow
processes temperature ≥473 K for short times [2]), whereas other substrate materials
can have even higher process or operating temperatures. Here, it must be ensured,
that critical properties of the waveguide material do not degrade during processing
or operation.
Acrylate-like polymers usually show very low temperature stability, even if new
approaches with multifunctional acryl co-polymers have shifted the frontier for glass
transition to Tg =383 K [146] respectively Tg =[443;453] K [84, 98] and decomposition
temperature to T =573 K [146] respectively T ≥603 K [84, 98]. All other materials,
that have been found in the open literature, have a Tg ≥475 K. Decomposition temperatures were usually reported to be beyond Tg ≥523 K. For some materials only
the process temperatures are given, e.g. the baking temperatures during test-object
creation. These values give at least a hint as to the temperatures these materials can
withstand.
44
Figure 4.1: Optical Loss around λ 840 nm in selected polymers. References and exact values can be found in Table 4.1.1. The
following acronymes were used: OSQ — Organosilsesquioxane, PEI-DR 1 — polyetherimide-disperse red1.
4.1 Materials
45
Figure 4.2: Optical Loss around λ 1310 nm in selected polymers. References and exact values can be found in Table 4.1.1. The
following acronymes were used: OSQ — Organosilsesquioxane, FPAE — Fluorinated polyarylene ether, PBACPC
— Poly(bisphenol A carbonate-co-diphenol carbonate), PEI-DR 1 — polyetherimide-disperse red1, PTS — arrow
structure of polyimide Ta2 O5 -SiO2 , PTP — arrow structure of polyimide Ta2 O5 -polyimide.
4 Materials for Optical Interconnects
46
Figure 4.3: Optical Loss around λ 1550 nm in selected polymers. References and exact values can be found in Table 4.1.1.
The following acronymes were used: FPAE — Fluorinated polyarylene ether, PCE — Polycyanorate ester resin, NB
— Nitroazobenzene, PFEK — Poly(fluorinated ether ketone), PEI-DR 1 — polyetherimide-disperse red1.
4.1 Materials
47
48
Organosilsesquioxane
Polysiloxane
n/a
Acrylates
deuterated
fluorinated
fluorinated and
photochromic
fluorinated and
deuterated
deuterated
(multifunctional
co-polymer)
halogenated
Sub-class
Material class
RIE
Photoexposure
RIE
Photoexposure
Photoexposure
RIE
Photoexposure,
RIE
Photoexposure,
RIE, molding
molding
0.2
0.02
0.02
0.17
0.24
0.07. . . 0.3
0.08
0.06. . . 0.07
0.2
0.5
0.5
0.035. . . 0.05
0.25
0.17
0.01. . . 0.02
0.03
0.2. . . 0.3
0.02. . . 0.18
[67]
[140]
[149]
[147]
[148]
[148]
[140]
[140]
[67]
[140]
[96]
[140]
[146]
[67]
Reference
=
... continued on next page
0.43
0.5
0.35. . . 0.7
0.35
0.27
1.5. . . 1.7
1.35
1.7
0.25. . . 0.7
1.2
0.6. . . 0.8
Optical loss in (1 dB/cm) around
λ 830 nm λ 1300 nm λ 1550 nm
Optical loss values reportet for wavelengths around λ
Photoexposure,
RIE
Photoexposure
Photolocking
Photoexposure
Patterning
Technique
Table 4.1.1: Polymers usable for optical interconnects.
830/1300/1550 nm.
4 Materials for Optical Interconnects
Cyclobenzene
Polyimide
Material class
PFCB
PFCB
halogenated
6FDA/MPDA
silica hybrid
n/a
fluorinated
epoxy
fluorinated/nonfluorinated
n/a
n/a
Sub-class
... continued from last page
Photoexposure
molding
RIBE
n/a
Photoexposure,
no wet etch
n/a
n/a
Photoexposure
RIE
Photoexposure,
RIE
etch
casting,
photoexposure
and RIE
possible
RIE
Patterning
Technique
0.2
0.02. . . 0.04
0.25
0.55
≥ 0.5
0.14
0.45
0.3. . . 0.5
0.3. . . 0.7
0.3
0.14
[67]
[158]
[155]
[155]
[140]
[156]
[157]
[153]
[67]
[140]
[154]
[152]
... continued on next page
0.25
0.25
0.33
3.14
0.4. . . 0.5
3.8
0.5
1.0
0.6
0.5
1.7
[151]
[88]
[150]
≤ 0.06
0.35. . . 0.67
Reference
Optical loss in (1 dB/cm) around
λ 830 nm λ 1300 nm λ 1550 nm
4.1 Materials
49
50
Fluorinated
polyarylene ethers
Material class
Photoexposure
RIE
[166]
[140]
[167]
[165]
[163]
[164]
[67]
[67]
[162]
[161]
[160]
[159]
Reference
... continued on next page
0.28
0.2
0.5
0.2. . . 0.5
0.1. . . 0.3
0.42
1.5
RIE
0.8
RIE
0.42
0.37
0.26
n/a
0.52
RIE
RIE
0.36
n/a
sulfide
sulfone /
EFPAESO
sulfone and sulfide,
co-polymers
ethynyl phenol
phenyl ethynyl
ketones and
sulfones
0.3. . . 0.4
n/a
0.45
RIE
0.3
Optical loss in (1 dB/cm) around
λ 830 nm λ 1300 nm λ 1550 nm
n/a
Patterning
Technique
sulfide
PFCB;
trifluorovinyl aryl
ether; siloxane
PFCB; dendrimers
and trifluoroether
monomers
PFCB; dendrimers
and dye doping
PFCB, trifluoro
aryl ether
BCB
Sub-class
... continued from last page
4 Materials for Optical Interconnects
Dendrimer
fluorinated
poly(ether ketone)
Nitroazobenzene
fluorinated
n/a
n/a
0.36
[67]
[174, 175]
[173]
[140]
[172]
[171]
[170]
[169]
[168]
[168]
Reference
... continued on next page
0.52
0.5
0.6
ether, pehnyl
hydoroquione
0.53
0.5
RIE
0.58
ethynyl
Photoexposure
(laser writing)
n/a
hyperbranched
0.5
RIE
Aromatic
fluoropolyester
0.5
RIE
0.3. . . 0.38
0.5
Optical loss in (1 dB/cm) around
λ 830 nm λ 1300 nm λ 1550 nm
RIE
Patterning
Technique
RIE
sulfone and
styrenes
ketone and
styrenes
ketone
Sub-class
Polycyanorate
ester resins
Material class
... continued from last page
4.1 Materials
51
52
[19]
[178]
0.6
0.2
RIE
Photoexposure
[177]
0.4
arrow structure of
polyimide-Ta2 O5 SiO2
arrow structure of
polyimide-Ta2 O5 polyimide
R
ORMOCER
0.07
0.01
0.03
[177]
[67]
[166]
... continued on next page
0.5. . . 0.6
0.1
2.2
Photoexposure,
RIE
RIE
OASIC
1.7
special/comercial
0.91
RIE
Polyetherimidedisperse
red1
[176]
0.8
[140]
Reference
n/a
0.42
Optical loss in (1 dB/cm) around
λ 830 nm λ 1300 nm λ 1550 nm
Poly(blisphenol A
carbonate-codiphenol
carbonate)
Patterning
Technique
0.39
Sub-class
PFGMA
Material class
... continued from last page
4 Materials for Optical Interconnects
Material class
sol-gel hybrid
polymer doped
with TiO2
perfluoro polymer
TruemodeTM
Sub-class
... continued from last page
0.05
RIE
Photoexposure
0.03. . . 0.05
0.16
0.07
0.45
Optical loss in (1 dB/cm) around
λ 830 nm λ 1300 nm λ 1550 nm
n/a
Patterning
Technique
[180]
[84, 98]
[179]
Reference
4.1 Materials
53
54
523. . . 673
394
453. . . 523
473. . . 613
498
Fluorinated polyarylene ethers
(FPAE)
Polycyanorate ester resins
Hyperbranched aromatic
fluoropolyester
Nitroazobenzene
Fluorinated poly(ether ketone)
Dendrimer
Poly(blisphenol A
carbonate-co-diphenol
carbonate)
Polyetherimide-disperse red1
R
ORMOCER
423
423
733
498
473
433. . . 472
473. . . 573
446
409. . . 423
373. . . 673
393
≥ 567
Cyclobenzene
383
≤383
Glass transition
[Tg ]=K
≤ 453
423. . . 543
333. . . 653
433
Process temperature
[Tp ]=K
Multifunctional acryl
co-polymer
Organo-silsesquioxane
Polysiloxane
Polyimide
Acrylates
Material class
543
523. . . 835
703
673. . . 725
393. . . 673
673
≤683
≥573
603
573
Stable to
[T ]=K
[166]
[19]
[172]
[140, 173]
[67]
[176]
[170]
[171]
[149]
[88, 140, 151, 152]
[67, 140, 153, 154,
156]
[67, 140, 158–160,
162]
[166, 168, 169]
[67, 140]
[147]
[146]
Reference
... continued on next page
Table 4.1.2: Process, glass transition and decomposition temperatures of visualized polymer groups.
4 Materials for Optical Interconnects
Sol-gel hybrid polymer
TruemodeTM
Epoxy
EFPAESO
Fluorinated polyarylene ethers
FPAESI
... continued from last page
Material class
543
543
423
Process temperature
[Tp ]=K
469
427. . . 460
443. . . 510
443. . . 453
Glass transition
[Tg ]=K
603. . . ≥623
548
752
573. . . 785.5
643. . . 762
Stable to
[T ]=K
[179]
[84, 98]
[140]
[164]
[165, 167]
[163, 165]
Reference
4.1 Materials
55
4 Materials for Optical Interconnects
4.2 Materials processing
4.2.1 Guidelines
Back in 1989 were proposed some guidelines for the manufacturing of optical backplanes which partly are still up to date, such as [51]:
1. The waveguide forming process shall be regarded as an add-on step to established PCB production. This implies temperature stability for the materials at
least for short times over T ≥ 473 K.
2. The formation process must be high-speed compatible, to hold production costs
down.
3. The costs for the devices for in- and out-coupling of light should be held as low
as possible. Thus, pick-and-place compatible approaches are favourable over
time consuming alignment of single components by hand.
4. Back then, loss values of 0.2 dB/cm at λ =830 nm seem reasonable. Loss values
below 0.5 dB/cm are regarded as acceptable [51].
The first point of the list can be accepted easily, because it ensures quick access to
a well-established machinery park, which can be used without larger modifications.
Point number two is doubtful, due to the fact that no obvious mass-market is visible
today. Niche-markets with high-end applications only exist today. Furthermore, the
light coupling devices and effort to couple light has to be minimized, which minimizes
the costs in this point as well. Finally, the question on choice of material for optical
interconnects is strongly dependent on the optical power budget available and the
length of the optical link. This issue will be discussed briefly in section 7.1 and in
detail in paper III.
Besides this report, Schröder et al. [89] added some points to the list, by focusing
more on the optical properties of the waveguide material. Basic requirements on a
waveguide material are: Low birefringence, exactly tunable refractive index, stable
under thermal and humidity changes. Furthermore a good adhesion to other materials
in SBU processing, such as, metals and polymers, is required.
The formation of a high quality waveguide depends on the dimension and the shape
of the cross-section, which must be defined precisely. Here, a low surface roughness
on the interface core-cladding is crucial for low-loss waveguides [47, 89]. Defects that
introduce further attenuation in the waveguide are [89]:
• absorption due to impurities (caused by particles)
• Rayleigh-scattering (caused by particles)
• absorption due to absorption bands of the material
• interface roughness
56
4.2 Materials processing
• impurities/bubbles
• inhomogeneous curing
• tilted end face
Research on polymer fiber optic materials was focused on fundamental aspects of
optical loss reduction in in polymer materials. As main factors contributing to the
high optical loss were listed [181]:
• C-H bond overtones,
• electron transitions, and
• structural imperfection.
Thus, research was focused on to exchange the C-H bonds by C-D or C-F bonds.
The list of factors responsible for optical loss in polymer materials was extended with
[182]:
• absorption due to transition metals and organic contaminants
• introducing scattering due to dust, microvoids, inhomogeneity of the core material
The former two lists (see even references [181, 182]) show that the optical loss in polymer materials can be suppressed to a minimum. However, this is dependent on the
basic chemistry of the material itself (elements, chemical bonds, stereo chemistry).
The lists above indicate that the choice of the right material is crucial for a low optical loss in the waveguide. Furthermore, handling and processing of the materials have
to be performed with care, to minimize excess optical loss due to in homogeneities in
the core layer [115].
Optical waveguide processing is not an ad hoc task. Trained personnel and appropriate high-precision equipment for structuring are necessary. Furthermore, a cleanroom facility may be necessary during certain process steps to prevent the waveguide
from exposure to dust.
In the following sections will be given an overview on processing of optical waveguides. Deposition methods for the core- and cladding layers will be discussed, as
well as methods for waveguide structure creation.
4.2.2 Deposition methods
Common methods for deposition of liquid polymer materials in electronics industry
are:
• spin coating [2]
• curtain coating [183]
57
4 Materials for Optical Interconnects
• meniscus coating [2, 184, 185]
• dip coating [2]
• spray coating
• Doctor Blading
Spin coating is a standard method of liquid film deposition in integrated circuit
(IC) production. A drop of liquid will be dispensed on the centre of a substrate. The
rotation of the substrate results in centrifugal forces that will lead to a spread out
of the drop over the whole surface thus generating a smooth film. In that manner
uniform films in thickness of 2 to 20 μm [2] up to 30 μm (own experience) can be
created easily.
A drawback of this method is the large amount of excess liquid needed to achieve the
desired uniform coverage. This became evident during the large-area panel project,
which is explained briefly in chapter 6 and in more detail in paper II. A quadratic
FR4 panel (610 mm x 610 mm) had to be coated by a polymer film of up to 15 μm
in thickness. To calculate the minimum amount necessary to coat the whole surface,
the quadratic panel will be inscribed into a circle, whereby the diagonal of the square
gives the diameter of the circle (see Fig. 4.4). Afterwards the volume of the polymer
film can be calculated according to:
V =h∗A
π
A = d2
√4
d = 2 ∗ 0.61 m
(4.1)
h = 1.5 ∗ 10−5 m
→ V = 8.77 ∗ 10−6 m3
V = 8.77 ∗ 10−3 l
→ V = 8.77 ml.
V is the volume of the film on the substrate (initially, in m3 ), d represents the diameter of the out-scribed circle of the square in m, h is the height of the film.
Theoretically, 8.77 ml liquid polymer is necessary to create a smooth film over the
whole surface of the substrate, but in practice 80 ml were needed to achieve this.
Comparing these two numbers shows that nearly 90% of the required liquid will go
directly into the drain. The excess amount cannot be suppressed as confirmed during
several spin coating cycles from similar projects. The recovery of disposed material
might be possible, but usually requires a re-design of the spin-coating equipment,
additional filter steps to purify the material, and time because of large areas of the
equipment housing are covered by a thin layer of disposed material.
There are more cost effective alternatives to spin coating, such as, meniscus coating,
curtain coating or dip coating [2].
58
4.2 Materials processing
d
610 mm
6
?
610 mm
-
Figure 4.4: Square, circle, and some distances for calculation of theoretical needed amount
of liquid to cover the Large Area Panel (LAP).
Meniscus coating is a technique where the substrate is drawn up side down over a
tube in horizontal direction. On top of the tube are small holes, through which the
liquid polymer is pumped. Unfortunately, there are no calculations or estimations
available in the literature about the amount of material necessary to start up the
process.
According to sketches of experimental set-ups, relatively large amounts of solutions
must be applied to start up the process and keep it running. Furthermore, it is reported that the excess amount of material needed is low during roll-to-roll processing
[184–186]. Therefore, this method is regarded as highly applicable in industry.
The curtain coating technique refers to a method where a curtain of a liquid polymer is formed and the substrate is drawn with a defined speed trough that curtain.
Drawing speed and viscosity of the liquid determine the thickness of the film on the
substrate in both cases. An excess amount of liquid between 1 and 10 liters is necessary to keep the process up and running [187]. Finally, dip coating refers to a
simple dip technique, where the substrate is vertically removed after immersion in the
liquid. The viscosity of the liquid and withdrawing speed of the samples determines
the thickness of the film on the substrate.
Another alternative to the methods for film deposition presented is Doctor Blading. A certain amount of liquid is poured on the substrate. A blade is then drawn
a prescribed distance over the surface creating a defined film thickness. It has to be
pointed out, that only a minimum of excess amount of liquid is necessary to start up
the process.
Spin coating is the method of choice for wafer-scale processes, whereas all other methods are well suited for large-area and roll-to-roll processing.
59
4 Materials for Optical Interconnects
4.2.3 Patterning techniques
The following techniques can be applied to create waveguide patterns:
• Silicon-bench technology & large-area processing:
–
–
–
–
Photolithography [18, 19, 154, 188, 189]
Direct Laser Writing [63, 98, 128, 146, 172, 190]
(Hot) Embossing [89, 91, 184]
Pressure Dispense of waveguides [191]
• Silicon-bench technology only:
– Reactive Ion Etching (RIE) [163, 164, 166, 170, 176]
– Reactive Ion Beam Etching [156]
A detailed description of these methods will be given below.
Ma et al. [67] also mentioned electron beam writing and photo bleaching (shifting
of absorption bands) as suitable patterning methods. Additionally, a comparison of
high frequency reactive ion etching and inductive coupled plasma reactive ion etching
(ICP-RIE) was done. They concluded that the ICP-RIE method leads to lower surface
roughness. These techniques will not be discussed further because of these are applied
on wafer-level-scale only an therefore minor important to the PCB industry [67].
4.2.3.1 Photolithography
A detailed introduction to photolithography is given in [189]. A photolithographic
mask is used to cast a shadow on parts of the substrate during illuminating with a
light source from above (see Fig. 4.5). Photo-active polymers can act as positive or
negative photoresists by exposure light. In case of the positive type resist, chemical
bonds in the material will be broken down during exposure. Afterwards, a solvent
will be used to wash away these areas, whereas the unexposed areas remain on the
substrate. Contrary, in negative type resist cross-linking of polymer structures occurs
by illumination with light. Here, the unexposed areas can be washed away, using a
solvent.
The process is usually as follows:
• deposition of the polymer film
• evaporation of solvent during a prebake step
• exposure
• wet-etch of the exposed/unexposed areas using solvent
• post bake
• hard bake for final cross-linking of the material
60
4.2 Materials processing
?
?
? ? ??? ?
? ?
?
i
Shadow mask
Photoresist
Substrate
Positive
photo resist
Negative
photo resist
N
after development
Figure 4.5: Photolithography for positive and negative type photoresist. Principle.
The structure resolution depends mostly on diffraction frontiers for the applied
light and quality (resolution) of the photolithographic mask. Pattern size is limited
by available exposure equipment. In research institutes, mask aligners, which can
handle wafers of 4-inch (101.6 mm) up to 6-inch (152.4 mm) are common nowadays.
Instead, the semiconductor industry is handling up to 12-inch wafers (304.8 mm) today, creating high precision patterns at 90 nm line width [5, 6]. Processing speed is
rather high. Usually a couple of minutes for alignment and a couple of seconds to
minutes for exposure are necessary [189].
The described mask sizes are not suitable for real backplanes that are 609.6 mm x
609.6 mm. An approach to circumvent this bottleneck is to use a large-area exposure
tool [70, 192].
In section 6.2 and paper II a procedure for stepping out 101.6 mm x 101.6 mm mask
patterns on a 609.6 mm x 609.6 mm substrates is presented. With this technique it
is possible to form continuous patterns over the whole area. In this special case the
process time increases rapidly by shrinking mask size because of the repeated time
consuming alignment of the small masks to the large-area panel.
The authors Houbertz et al. [19] (FhG-ISC, Germany), Streppel et al. [188] (Fraunhofer Gesellschaft, FhG-IOF, Jena, Germany), and Robertsson et al. [18] (Ericsson/Acreo, Sweden) mainly did process development work and characterization of
R
. The material acts as a negative type photoresist. It is recomthe ORMOCER
mended that exposure should be performed in a Nitrogen atmosphere. Air or Oxygen
61
4 Materials for Optical Interconnects
Figure 4.6: Direct Laser writing. The laser stands still and the board, which is coupled to
a computer and CAD system, moves underneath. For further information see
text and [128].
containing atmosphere will create an inhibiting layer, which can reduce the final film
thickness significantly. Houbertz et al. [19] summarized the main properties of the
material and gave important process parameters. Streppel et al. [188] observed a
broadening of the core structures during exposure trials for the 4-layer-waveguidestructure. This was due to back-reflection from the substrate as the layer thickness
was increased. The effect could be minimized by an optical buffer layer, averting the
optical back reflection from the substrate. Robertsson et al. [18] worked with earlier
developments of the material. Single mode waveguides were produced with straight
sidewalls. Multimode waveguides showed slightly inward curved sidewalls.
Kang et al. [154] used a polyimide-like material. In this case the material undergoes a
refractive index change during exposure. This property is used to form the waveguide.
The advantage of this method is that no wet-etch and development step is necessary.
Furthermore, an extra planarization layer will not be needed for multiple layers of
waveguides.
4.2.3.2 Direct Laser Writing
The substrate is mounted on an x-y-z movable stage. A stationary laser is placed perpendicular to the stage and focused on the substrate surface. The stage is connected
to a computer with an integrated CAD system for direct waveguide mask transfer
(see Fig. 4.6).
Chen et al. [128] described their approach of direct laser writing. The x-y positioning resolution was reported to be 0.5 μm. In the z-direction a resolution of 0.01 μm
could be achieved. The scan speed of the stage was reported to be v ≤1 cm/s. Using
a laser source working at λ =325/442 nm, a multimode waveguide with a width of
50 μm could be created. The polyimide waveguides created showed an optical loss of
0.21 dB/cm at λ = 850 nm. All other reports mentioned [63, 98, 146, 172, 190] do not
give any further information concerning the operating conditions and quality of the
62
4.2 Materials processing
waveguides produced with this method. Assuming a tentative waveguide pattern, 10
m in length, and the scan speed given above, the processing time can be calculated to
1000 s or roughly 17 min. These values might be slightly higher in practice, due to repositioning of the state during exposure, because discontinuous waveguide patterns
are likely on realistic optical backplanes. Thus, the processing-time for single and
simple waveguide patterns is quite long, and increases very fast if complex patterns
are necessary.
4.2.3.3 Pressure dispense of waveguides
The methods presented so far have one disadvantage: for the formation of the waveguide layer full coverage of the substrate is necessary. Usually, only a part of the surface
is used for the core section of the optical waveguides. During patterning often more
than 50 % of the core-material layer will be removed. In addition, materials of optical
quality are expensive. Cladding and/or components occupy the rest of the area.
A direct dispense method has been proposed by Keyworth et al. [191]. The substrate
chuck is connected to a control unit, which allows computer or manual guided translation with sub-micrometer precision (see Fig. 4.7). A syringe dispenses the polymer
with an applied pressures of 5 to 20 psi at a writing speed of 1.0 to 1.5 mm/s. The
waveguide structures are immediately cured by a mercury lamp, which is positioned
over the table.
The surface tension and viscosity of the polymer in liquid state define the cross section
of the waveguide. Multimode waveguides 50 μm wide were successfully generated.
The advantage of this method is the possibility to freely position the syringe anywhere on the board. A mask layout can be transferred quickly and easily without
high costs. A drawback of this method is clearly the ill defined sidewall formation
of the waveguides. A rough calculation of processing speed of waveguide pattern, 10
m in length, shows that a process time of at least 10000 s or around 2 h 45 min is
necessary to achieve this. The method seems to be suitable when single waveguides
are necessary, but unsuitable for complex waveguide patterns.
4.2.3.4 (Hot) Embossing
This method has been a long-term topic at the FhG-IZM (Berlin, Germany). A brief
introduction and process description was given by Krabe et al. [91]. Basically, a metal
shim with ridges is pressed at an elevated temperature into a foil, which will later be
the under cladding layer. Afterwards, shim and foil will be carefully separated. The
remaining grooves formed by the ridges are filled with the core material by the doctor
blade method. On top of these layers the upper cladding will be attached. Finally
the whole structure, so called opto foil, will be included in the PCB by a lamination
process (see Fig. 2.7 on page 19).
The electroformed Nickel shim is produced by deep-UV lithography. The surface
roughness of the grooves have been measured to be RM S ≤ 10 nm [89, 90]. The
63
4 Materials for Optical Interconnects
Lamp
Waveguide
Core
Under
Cladding
z
:
Syringe
j
*
Substrate
Figure 4.7: Pressure dispense of a waveguide. Dispenser and mercury lamp are mounted
fixed over the table. The substrate chuck, connected to a CAD system, moves
underneath. The board has moved in negative x-direction. For further information see text and [191].
disadvantage of the method is the limitation to a shim size of 100 mm (year 2001,
[89]). Other authors, like Yoon et al. [85], applied Si-shims, produced by deep RIE
processes, as masters for the embossing. Waveguides with a cross-section of 60 μm
x 100 μm, by sidewall and surface roughness of the bottom area of 10 nm and 2 nm
respectively, were achieved.
Neyer et al. [103] reported on an embossing approach, with work in progress to scale-up
the process to board sizes of 30 cm x 40 cm. An alternative method was presented by
Choi et al. [104], where PDMS shim was applied. The master for the shim was created
by typical photolithography of SU-8. Micro-mirrors, waveguide-ends with a 45◦ tilted
angel, were included as well and cut using a microtome setup. The embossing process
was done under in a vacuum chamber at 90 ◦ C for 10 h. No information regarding
surface roughness was given here. The shim sizes is in this case dependent on the size
of the master and therefore on the size of a master substrate and the capabilities of
the master equipment.
4.2.3.5 Reactive Ion Etching (RIE)
This method works as follows: The under cladding and core layer are deposited and
fully cured. In the following step an etch mask layer is created on top. For this, an
etch-resistant material is deposited. In a subsequent photolithographic process the
waveguide pattern is formed in this layer. These waveguide structures are transferred
to the core layer via a plasma etch process. It is common to use Oxygen plasma. In
the last step, the mask layer is removed and the top cladding will be created (see Fig.
4.8).
As suitable mask materials, chemical vapor deposited (CVD) Silicon Nitride (Si3 N4 )
[163, 164, 176] and Aluminum [156, 170] were reported. The Si3 N4 was patterned
by a CF4 RIE process. Dreyer et al. [170] used a lift-off technique to produce the
64
4.2 Materials processing
O2 -plasma
????
y
i
(Photoresist
removed)
Photoresist
Si3 N4
Polymer
Substrate
:
1
CF4 -plasma
????
y
i
(Si3 N4 removed)
Si3 N4
Polymer
Substrate
:
1
Figure 4.8: Reactive ion etching for waveguide forming. Principle.
waveguide mask. In this case photoresist was patterned with the waveguide structure on top of the polymer layer. Afterwards Aluminum was deposited. Finally, the
photoresist was removed and created, through lift-off of parts of the Aluminum layer,
the desired waveguide pattern. After the mask preparation, all reports mentioned the
application of Oxygen plasma to etch the waveguide structures free. The parameters
required to achieve a smooth surface after dry etch were investigated by Tang et al.
[176]. It was shown that high pressure and low plasma power resulted in small surface
roughness of RM S ≤8 nm. The etch-rate was in the range of 60 to 240 nm/min.
Higher etch rates resulted in higher surface roughness values. All the presented methods in fact have several drawbacks: the sidewalls of the waveguides can be very rough
because their form is mainly defined by the quality of the photolithography of the
etch mask. A scanning electron microscope picture in the report of [156] has shown
this phenomena very clearly. Investigations of the impact of sidewall roughness on
transmission behavior were missing in this case. On the other hand this method does
not seem to be easily applicable in industrial PCB manufacturing. It would be hard
to find a plasma chamber, which is large enough to handle 609.6 mm x 609.6 mm
substrates. In addition, the process time is considerable for such a case. The waveguide pattern first has to be created in an etch protection mask and afterwards it has
to be removed. Assuming an etch rate of 200 nm/min and a waveguide core 50 μm in
height would result in an etch rate of 25 min plus extra buffer time for any required
over etch. A direct photolithographic process or direct laser writing process has an
advantage in this case in that they are faster.
65
4 Materials for Optical Interconnects
4.2.3.6 Reactive Ion Beam Etching
The method proposed by Cornic et al. [156] is very similar to the RIE process. An
Aluminum layer was used as etch mask in this case. The self-developed reactive ion
beam process achieved an etch rate of 200 nm/min. Rough sidewalls were observed,
caused by the photolithography of the Aluminum mask.
4.2.3.7 Patterning techniques — Conclusions
Different methods of forming waveguides were briefly described. Methods involving
RIE are applicable in research areas but not in production in the PCB industry, because of restrictions regarding LAP, etch chamber dimension, suitable mask material,
and etch time. High precision patterns are usually limited to 300 mm but can be
extended by using a large-area mask aligner. The approach shown in chapter 6 can
circumvent the problem of inflexible and expensive large photomasks by a step-out
process of small-size, high-precision waveguide masks. However, pure photolithographic patterning of an optical polymer resist material still has the advantage of
giving smooth sidewalls and the reduction to just one process step compared to the
RIE method. Reasonable resolution can be achieved as well.
On the other hand, direct writing methods are inexpensive but are limited by the
size of the laser spot, positioning, and speed. Additionally no information could be
found to date regarding sidewalls of the waveguide structures created. The only obvious advantage seen is the rather inexpensive transformation of waveguide pattern
to the substrate through direct connection to CAD design tools. The direct dispensing method suffers from a one order of magnitude lower writing speed and a poorly
defined cross section shape.
Hot embossing seems to be a promising method, but is limited to shim sizes of about
10 cm x 10 cm at the moment, whereas work is in progress to reach 30 cm x 40 cm
[103].
However, concerning low cost waveguide formation for a few point-to-point links on
large areas, the direct laser writing method seems to be the preferred choice. Regarding complex waveguide patterns on large areas, step-out patterning in combination
with direct photolithography seems to be the best choice so far.
66
5 Characterization
5.1 Optical characterization
5.1.1 Transmission
One of the main parameters researchers are most interested in is the optical loss value
of a waveguide. This value is usually of the order of magnitude of 10−2 to 100 dB/cm
for polymer waveguide materials. It will become a crucial factor for the limited power
budget on optical backplanes (typically, e.g., 14 dB [85]) if long-distance links around
1 m in length are necessary.
Several methods exist to measure the optical loss, such as single length method, cutback method [156], prism coupler method [193], and the scattering-light measurement
[194]. One of the easiest ways to perform an optical transmisson measurement is the
single length methode, where a waveguide of a certain length is taken and its endfacets are polished roughly in advance. At one of the end facets, light will be coupled
into the guide by a single mode fiber (core diameter around 9 μm) and collected on
the other side by a large-core multimode fiber. An index matching fluid is applied on
the end facets to minimize the coupling losses. For precise measurements piezoelectric
translator elements are necessary for the alignment. The difference between coupled
and recorded light will be noted. After this measurement, the fiber ends are brought
together and the difference between coupled and recorded light will be noted again,
to be able estimate the coupling loss. The difference between the value of the link
with and without the waveguide gives the optical loss in the waveguide.
Another method is the so called cut-back method, as described by Cornic et al.
[156], including a suitable measurement setup. A sketch is shown in Fig. 5.1. The
waveguide was successively shortened by about 1 cm and re-measured in the optical
setup. The slope in the resulting loss versus waveguide length plot is the desired
optical loss of the waveguide in dB/cm. In Fig. 5.2 a hypothetical loss measurement
is shown to clarify this method. Cut-back offers high precision measurements but
requires a larger effort in sample preparation. The sample has to be cut, the end
facets have to be polished, and the set-up has to be re-aligned for each measurement,
which is very time consuming.
The prism coupler method is usually applied to determine the thickness and
refractive index of thin transparent films, as described below. It can be applied
for optical loss measurements on (uncovered) waveguides too. Two prisms have to
be applied at once in this set-up for this method. One of the prisms is fixed in
position with respect to the waveguide and light from a laser source is coupled into
67
5 Characterization
(a) Initial substrate length.
(b) Re-measurement of the system after subsequent shortening of waveguide devices.
Figure 5.1: Cut back method for transmission measurements of optical waveguides. Principle: Light is coupled into the waveguide by a LASER source from the left
hand side. A detector measures the transmitted light power on the right hand
side.
Optical loss over a waveguide, cut−back method
1
0.9
0.8
Optical loss, a.u.
0.7
0.6
0.5
0.4
0.3
0.2
loss measurements
linear
0.1
0
0
1
2
3
4
5
6
7
8
9
10
Length of waveguide, a.u.
Figure 5.2: Hypothetical diagram of an optical loss measurement. The slope of the best-fit
line gives the optical loss in dB/cm.
68
5.1 Optical characterization
the waveguide trough the prism. The second prism is brought into contact with the
waveguide at defined distances and light is coupled out trough it to be collected using
a photo detector. The sensitivity was reported to be 0.02 dB/cm for overall losses
around 1 dB/cm [193]. The light’s signal strength measured can be drawn into a
similar diagram as shown for the cut-back method (see Fig. 5.2). Again, the slope of
the best-fit line is identified with the optical loss of the waveguide.
For the light-scattering measurement method, as described in [194], light from
a laser source is coupled into a waveguide and a camera measures the amount of
scattered light along the channel waveguide. Again, the results from this measurement
are drawn in a diagram similar to the one shown in Fig. 5.2, to derive the loss of
the optical waveguide. This method is useful for large optical loss values below 100
dB/cm, even if losses below 1.0 dB/cm can be detected.
5.1.2 Refractive index measurements
5.1.2.1 Refractive index, Total Internal Reflection (TIR) and Numerical Aperture
An exact refractive index determination is necessary for waveguide fabrication due to
several reasons: The difference in refractive index defines
• the critical angle for the TIR condition of the waveguide [24],
• the angle of acceptance of the waveguide end facet and is therefore connected
to radiation angle of the laserdiode [24], and
• the bandwidth times length product as shown in section 7.1, [195], and paper
III.
In Fig. 5.3 a schematic sketch is shown, visualizing the connection between TIR,
critical angle, and angle of acceptance for the waveguide. Conditions for the TIR,
enabeling the propagation of light inside the waveguide, are:
• transition from a volume with high towards a volume with low refractive index,
and
• the incident angle of light is larger then the critical angle.
If these conditions are satisfied, then the effect of TIR appears, where the refracted
beam disappears and the lightwave is encapsulated in the waveguide.
Starting with Snells law:
(5.1)
n1 · sin α1 = n2 · sin α2
where n1 is the refractive index of waveguide core and n2 of the wvaeguid ecladding,
respectively. α1 is the angle of incident, and α2 the angle of refraction. Under the
condition of TIR α2 = π/2 and with α1 = αcr eq. (5.1) transforms to
π
n1 · sin αcr = n2 sin
2
n2
sin αcr =
.
(5.2)
n1
69
5 Characterization
n2
n3
ξ
αcr
1
n1
q
I
βcr
Figure 5.3: Waveguide schematics explaining the connection between critical angle for TIR
and angle of acceptance. The terms n1 , n2 , and n3 stand for the refractive
index of the waveguide core, the waveguide cladding, and the surrounding
medium, respectively. The term αcr is the critical angle required for TIR, βcr
is connected to αcr to be able to calculate the incident angle of acceptance ξ.
See text for further information.
The relation between αcr and βcr (compare with Fig. 5.3) is
βcr =
π
− αcr .
2
(5.3)
The angle βcr can be related to the acceptance angle for the light that is coupled into
the waveguide. Thus according to Snells law (n3 = nair 1):
n3 · sin ξ = n1 · sin βcr
π
− αcr
= n1 sin
2
= n1 cos αcr
= n1 1 − sin2 αcr
(5.4)
From eq. (5.2) follows
sin ξ = n1
sin ξ =
1−
n21 − n22 .
n2
n1
2
(5.5)
The term n21 − n22 is equal to the numerical apparture N A. If the sine of the angle
of incidence of the coupling light source is lower than the N A then the light will be
70
5.1 Optical characterization
guided along the waveguide due to the satisfied conditions of total internal reflection.
However, in practice it is important to adjust and determin the refractive index for the
optical waveguide system. The prism coupler and refractive index near field pattern
methods can be used to refractive index studies on thin films and short waveguide
sections, respectively, and will be discussed briefly.
5.1.2.2 Prism coupler
The prism coupler method is used to measure the refractive index and the thickness
of thin transparent films. Films of thicknesses between 2 and 10 μm have to be
prepared on good reflecting surfaces, preferably silicon wafers. The film’s surface is
then brought into contact with a prism. The whole setup is mounted on a rotational
table and a laser source points from one side trough the prism and film towards the
substrate surface. Usually the beam gets reflected on the surface and a detector
collects the light. For certain angles the condition of leaky modes is fulfilled. In these
cases, light is coupled into the film and disappears from the detector signal. Studying
two neighboring angles where this effect occurs allows calculation of the refractive
index and the film thickness [196].
The equipment used in our lab works slightly differently [197]: A small air-gap exists
between the film surface and prism. The first detected leaky mode determines the
refractive index of the film, while the other leaky modes are used to calculate the film
thickness. The refractive index resolution is about ±0.0005 with an offset accuracy
of ±0.001. Thickness resolution and accuracy were reported to 0.5 % and 0.3 %
respectively compared to the total film thickness. In practice, film thicknesses of about
5 μm gave the best results. Fig. 5.4 shows a principle sketch of the measurement
set-up.
5.1.2.3 Refractive index near-field pattern
Oberson et al. [20] reported a method for refractive index near-field measurement
for silica waveguides on silicon. “The basic physical properties of optical waveguides
is given by the distribution of the refractive index. From this distribution all local
parameters can be computed, such as the mode profile that determines the coupling
loss.”[20]
In this technique, one end-facet of a short measurement sample (usually 1 cm in
length) is illuminated by a laser beam. On the other side, a detector screen collects
the refracted light. The laser focus can be moved with micrometer precision in the
x/y-plane over the end facet (see Fig. 5.5). For each measurement point, the position
for the refracted beam will be determined on the screen. The refractive index profile
over the waveguide cross section can be calculated by applying Snell’s law (see eq.
(5.7)).
(5.6)
n1 · sin α1 = n2 · sin α2
v1 · s = v2 · s
(5.7)
71
5 Characterization
Prism
Polarized
Incident
Beam
Reflected
Beam
:
n3
z
>
n1
n0
- Leaky Wave
Substrate
Figure 5.4: Prism coupler method for refractive index and thickness measurement on thin
films.
The variables n1,2 represent the refractive index of the two optical media, i.e. air
and the waveguide core. The angles α1,2 are the angles of the light-rays in the corresponding media. Eq. (5.7) represents Snell’s law in vector form, where v1,2 = n1,2 · c
represents the direction of a light ray and its velocity in medium 1/2, n1,2 is again
the refractive index of the medium, c the velocity of light, and s is any vector tangent
to the interface. The resolution in refractive index was reported to 0.0005, and x/y
resolution to 0.1 μm [20].
5.1.3 Bit Error Rate (BER) Test
Derickson [198] described a basic setup for bit error rate tests. With the help of this
test, the quality of an optical data link can be characterized. A random pattern of
bits is coupled into a waveguide, and collected on the other side (see Fig. 5.6).
The BER value is defined as
E(t)
(5.8)
BER =
N (t)
where E(t) is the number of bits that were wrong and N (t) the total number of bits.
First, the optical link containing laser source, pattern generator, and detector unit
is tested, using a number of input powers. This will result in a log/log-plot of BER
versus optical power and, in the optimum case, a best-fit line. Secondly, the waveguide is hooked up to the whole system, and the test is performed once more. If the
waveguide shows no strange effects, this should result in a line parallel to the former
one, permitting the evaluation of BER for the system including the interconnecting
waveguide.
72
5.1 Optical characterization
Detector Screen
Waveguide
9
-
Sample
z
6
Z
Y
Y
6
Laser Source
Z
6
-
X
-X
Figure 5.5: Refractife Index Near Field measurement.
Calibration Sequence
Test Sequence
Bit Sequence
Generator
Fiber Optical Link
Bit Sequence
Generator
? Bit Sequence
Analyser
9
?
DUT
Calibrated
Fiber Optical Link
- Bit Sequence
Analyser
Figure 5.6: Bit Error Rate determination. Measurement setup. DUT — Device Under
Test.
73
5 Characterization
BER requirements for electrical and optical interconnects have been reported by the
IEEE P802.3ap Backplane Ethernet Task Force [199]. On Ethernet based backplanes
BER values below 10−12 are required [200], whereas the values below 10−15 are required for optical interconnect systems [201].
5.2 Surface tension
In SBU processing not only one kind of material will be applied. Often different
kinds of materials have to be deposited on top of each other. Often, a liquid material
has to be deposited by spin coating or similar techniques on a solid surface. Here,
it is important that a good wetting behavior occurs between the liquid deposited
and the underlying surface. One method to test the wetting behavior is to look at
contact angle measurements as described below. Furthermore, the calculation of the
solid’s surface tension is possible from contact angles formed on its surface by several
different test liquids.
5.2.1 Theory
When a drop of liquid is brought into contact with a solid surface then three things
can happen (visualized in Fig. 5.7):
1. total wetting,
2. total de-wetting,
3. intermediate state with:
a) wetting, and
b) de-wetting tendencies.
In thin film processing the total wetting case is desired, whereas the intermediate
case often occurs. The experimentalist has two choices to achieve a sufficiently good
wetting behavior:
1. modify the surface properties by applying a surface activation method, e.g.,
silane coupling agents [202] or plasma activation methods [203], or
2. by changing the surface free energy of the liquid.
The correlation between the contact angle that a drop of a liquid forms on a solid
surface and the surface tension (or surface free energy) of the solid and the liquid can
be expressed by Young’s equation as [204, 205]
γsv = γsl + γlv ∗ cos(θ)
(5.9)
where θ stands for the contact angle formed by the droplet on the surface towards the
surrounding gas phase. The γ-terms are assigned to the surface energy (in mN/m) of
74
5.2 Surface tension
(I)
(II)
(III)
(IV)
Figure 5.7: Tentative contact angles on a solid visualizing a complete hydrophilic, intermediate, and hydrophobic surface of the solid. — (I) complete wetting, (II)
wetting tendency, (III) de-wetting tendency, (IV) complete de-wetting
a specific interface and the indices characterize the boundary: γsv – solid-vapor, γsl
– solid-liquid, and γlv – liquid-vapor (see even Fig. 5.8). The surface tension γlv is
assumed to be equal to γl (surface tension of the liquid in air) at a first approximation
[205].
Furthermore, the surface tension can be written as the sum of a polar part and a
dispersive part (superscripts p and d) [206]:
γ = γp + γd,
(5.10)
which is used in the geometric mean and harmonic mean approach as described further
below.
For the calculation of the surface tension value of a solid several approaches exist. All
of them have in common that the contact angle of a droplet on the surface investigated
and the liquid’s surface tension is necessary to be able to calculate the surface tension.
The geometric and harmonic mean approach require the contact angles of polar and
unpolar liquids on the surface under investigation for calculation of the surface tension.
In case of the geometric mean approach, only one polar and one unpolar liquid is
necessary to be able to solve the equation
p
d +
γld γsv
γlp γsv
,
(5.11)
γl ∗ (1 + cosθ) = 2
γl
γsv
θ
γsl
-
Figure 5.8: Droplet with force diagram of Young’s equation.
75
5 Characterization
wich can be transformed into a linear equation to
p
d +
γsv ∗ x,
y = γsv
(5.12)
where x and y are identified with
x=
and
y=
γlp
γld
(1 + cosθ)γl
.
2 γld
(5.13)
(5.14)
The calculation of the surface tension according to the harmonic mean approach
is performed with the following equation
d d
p
γlp γsv
γl γsv
γl ∗ (1 + cos θ) = 4
+
.
(5.15)
p
d
γlp + γsv
γld + γsv
Unfortunately is it not possible to separate the variables in this case. Other ways
d
p
and γsv
respectively. Several test
have to be found to solve this equation for γsv
liquids, preferably polar and unpolar, are needed to calculate the surface tension.
Furthermore Neumann’s approach can be used to calculate the surface tension,
whereby one test liquid only is necessary. A drawback in this case is that only the
total surface tension can be calculated. A separation into a dispersive part and polar
part of the surface tension is not possible. The equation used is stated as follows:
2
γs
cos(θ) = −1 + 2
∗ exp−β(γl −γs ) .
(5.16)
γl
The parameter β has the value 0.000115 (m2 /mJ)2 . All calculation procedures are
outlined more in detail in paper IV, found in the Appendix.
5.2.2 Equipment & procedure
The experiments described below were performed using a contact angle meter from
KSV. This instrument is equipped with a CCD camera and a frame grabber, capable
of taking pictures at a rate up to 1 picture every 20 ms (see Fig. 5.9). A typical
measurement procedure is as follows: A drop of the test liquid is deposited on the
substrate, and an image of it is recorded by the camera. A computer program, which
is internally based on Axissymmetric Drop Shape analysis (ASDP) algorithms, evaluates the shape of the droplet and returns the contact angle towards the surface [207].
Before a measurement starts, the cleanliness of the equipment is tested by the pendant drop method. A hanging drop is produced at the syringe and from its shape and
76
5.2 Surface tension
Figure 5.9: Photograph of the KSV CAM 200 contact angle meter.
the test liquids density the surface tension can be calculated by the computer program.
If the measured surface tension of the liquid is close to tabulated values (e.g.
±1 mN/m2 ) then the equipment is defined as clean enough for the experimental
series.
Contact angle measurements were performed in the following manner: A drop of the
test liquid was formed at the needle lattice. The needle was then lowered towards the
surface. At initial contact with the surface, the droplet suddenly leaped over and settled down on the surface. At this moment, the frame grabber started automatically.
Initially, the contact angle of a droplet decays until a (quasi) steady state is reached.
This (quasi) stable contact angle is measured and used in all further calculations.
It is stated that a drop volume of between 5 and 8 μl is preferred. Liquids such as
water or formamide (both polar) have no problem to form such amounts in a single
pending drop, whereas di-iodomethane has. In this case three drops were deposited
at one and the same sample position, which joined up to one drop with the desired
amount of liquid. Several measurements were performed to get better statistics. A
standard deviation between 0.3◦ and 2.2◦ is not unusual (based on 10 individual
measurements), which varies with age of the sample and test liquid applied.
Further experiments were performed with the contact angle meter, which are not related to surface tension measurements. In this case the equipment was extended with
a mini-hotplate, capable of heating the sample under controlled conditions (±2 K).
R
resisn on cured
This setup was used to determine the contact angle of ORMOCER
R
surfaces under the influence of heat. It could be shown that the conORMOCER
R
resins for one and the same material differ significantly if
tact angles of ORMOCER
measured at room temperature and at 353 K. At this temperature the solvent evaporates from the resin, which is responsible for the difference in contact angles as shown
below.
77
5 Characterization
R
5.2.3 De-Wetting of ORMOCER
s
5.2.3.1 Motivation
During the lareg-area panel project a de-wetting phenomenon was discovered during
multilayer SBU-processing, as explained more in detail in section 6.2 and papers II
and IV. The goal of the project was to create a waveguide structure composed of an
under cladding, a core layer and an upper cladding layer. Initially, as composition for
the core layer was chosen pure B59, whereas the cladding layers had a composition of
B59:V32=88:12 wt%. The core layer in its resin state showed a de-wetting under processing (pre-bake step, before exposure). This can be prevented by surface activation
of the underlying cladding layer, e.g., by oxygen plasma activation [203] or silanization [202]. However, these methods were not available for large-area processing in our
departments cleanroom. Instead, the material system was changed slightly to make
cladding and core layer more similar to each other in their surface properties. In a
first attempt the core layer was altered by adding 5 wt% of refractive index tuning
agent V32. At the same time, the composition of the cladding layer was changed to
B59:V32=83:17 wt%, to keep the step in refractive index between core and cladding
constant to Δn =0.01.
This approach could prevent de-wetting of the core layer during processing, but
spawned the question: How much can the amount of refractive index tuning agent be
reduced in the core layer by maintaining integrity of the resin film under processing?
The material V32 is in research state only, whereas the refractive index tuning agent
B66 is already established on the market. In initial test with the material system
B59:B66 (B59 as core layer, B59:B66=40:60 wt%, Δn = 0.01) revealed similar behavior. The work’s focus was therefore mainly on the commercially available B66 system.
In the following section only the main results are presented. Further details regarding
the experiments, calculation methods, and results, are found in paper IV.
In particular, experiments such as de-wetting tests, contact angle measurements of
R
R
resin on cured ORMOCER
surfaces, and contact angle measureORMOCER
ments of de-ionized water, di-iodomethane, and formamide were performed and are
explained briefly. The composition of the samples can be found in Table 5.2.1.
5.2.3.2 De-wetting experiments
R
resin occurred during the pre-bake step in wavegThe de-wetting of the ORMOCER
uide processing of the core layer. During the spin coating a smooth film of the core
R
. However,
layer resin could be created on a cladding layer of cured ORMOCER
during the pre-bake at 353 K it lost its integrity. In some cases the shrinkage started
at the border, whereas the final state was a drop of liquid resin in the centre of the
wafer or a scattered surface due to formation of large-area holes. In Fig. 5.10 the
effects described are shown.
78
5.2 Surface tension
Table 5.2.1: Composition of films and test liquids to be used for the experiments. All
compositions listed, except C12, were used in the surface tension related experiments as described in the text.
Index
B59 in (1 wt%)
Composition
B66 in (1 wt%)
V32 in (1 wt%)
C1
C2
C3
C4
C5
C6
100
99.8
99
95
40
0
0
0.2
1
5
60
100
0
0
0
0
0
0
C7
C8
C9
C10
C11
C12
99.9
99
95
88
0
83
0
0
0
0
0
0
0.1
1
5
12
100
17
79
80
(b) Tentative de-wetting of pure B66 on a
B66:B59=60:40 wt% surface after 1 h at 353 K.
Figure 5.10: Sketch of principle of shrinkage. Shrinkage photo/final stage photo.
(a) De-wetting behavior of a wet
film on a cured surface.
Uncured
material
during pre-bake
Wafer
Direction
of material
movement
(c) Tentative de-wetting of pure B59 on
a B59:V32=88:12 wt% surface after 1 h
at 353 K.
5 Characterization
5.2 Surface tension
For the test wafers with a cured layer in cladding layer composition of the system
B59:V32=88:12 wt% and B59:B66=40:60 wt%, respectively, were prepared. On top
of the layer a tentative core layer was deposited, varying in composition from pure
B59 to B59 altered by 5 wt% of refractive index tuning agent. Afterwards the wafers
were placed on a hot plate at 353 K and observed for 1 hour. In the case of the system
B59:V32, the altering of the core layer by 0.1 wt% of V32 was enough to achieve a
stable resin film over the test period. The system B59:B66 needed an amount of 5
wt% of B66 in the core layer to obtain a stable film. All other films failed after time
steps such as 2 min (pure), 10 min (0.1 wt%) and 1 h (1 wt%). This difference in
altering amount can be attributed partly to the difference of the cladding composition
underneath.
R
s
5.2.3.3 Contact angles of ORMOCER
In a further test series the contact angles of the altered core layers of the systems
B59:V32 and B59:B66, respectively, were measured on cured cladding layers of their
material systems, respectively. Pre-tests with the B59:V32 at room temperature
showed no difference in the contact angles measured with respect to their materials
composition (arround 25◦ , see paper IV for details).
This result together with the results from the hot plate experiments described above
lead to the conclusion that the solvent in the resin has an influence on the wetting
behavior and therefore on the contact angle of the cladding layer. Thus, further experiments were conducted on a mini-hotplate (set-point 353 K). The contact angle
differed substantially much between the series this time, as shown in Fig. 5.11. Additionally, contact angles of pure B66 and pure V32 were added.
The contact angles measured for the system B59:B66 show a decay in a duration of
40 to 180 s after the dropplet was settled on the surface (see Fig. 5.11). The decay
time is correlated to the amount of B66. However, steady state contact angles between
20 and 10 degree for the pure B59 and the B59:B66 blends were obtained, whereby
contact angles below 15 degree were obtained only for a composition of B59:B66=95:5
wt%. Pure B66 showed values below the 10 degree line.
The B59:V32 system behaved differently; a clear plateau or steady state contact angle
could not be detected. In this case, the contact angles were measured to be between
0◦ and 23◦ . For the blends of B59:V32 contact angles below 15◦ were observed. Pure
B59 showed a contact angle of around 21◦ .
A comparison with the results from the de-wetting results showed that solutions that
failed these tests showed a contact angle above the 15◦ line, whereas the succeeding
solution fell below that line.
5.2.3.4 Surface tension estimation
Contact angles were measured on surfaces of the systems B59:B66 and B59:B66 in
various compositions (see Table 5.2.1). Liquids applied were de-ionized water, diiodomethane, and formamide. A screen dump of a typical contact angle measurement
81
5 Characterization
Contact angle in degrees
25
20
15
C1
C2
C3
C4
C6
10
5
0
0
50
100
150
200
250
300
350
Time in seconds
R
(a) Contact angles of ORMOCER
resins on a cured C5 surface.
Contact angle in degrees
25
20
15
C1
C7
C8
C9
C11
10
5
0
0
50
100
150
200
250
300
350
Time in seconds
R
(b) Contact angles of ORMOCER
resins on a cured C10 surface.
R
R
Figure 5.11: Contact angle of ORMOCER
solutions on ORMOCER
surfaces under
the influence of heat. The determination of the contact angle had to be
aborted in certain cases due to the fact that the droplet’s frontiers were outside
the observation window and the computer algorithm could not perform the
R
mixtures see Table
calculation. For the compositions of the ORMOCER
5.2.1. The accuracy bars are based on 3 to 5 individual measurements.
82
5.2 Surface tension
Figure 5.12: Screen shot from a contact angle measurement.
is shown in Fig. 5.12.
From the contact angles surface tension values according to the harmonic mean,
geometric mean, and Neumann approach were calculated. A summary of the values
calculated is shown in Fig. 5.13. Detailed values can be found in the Appendix, paper
IV.
In Fig. 5.13 the development of the dispersive, the polar part, and the overall
(summary) of the surface tension in the system B59:B66 and B59:V32, are shown.
For the polar part of the surface tension values around 5 mN/m2 were obtained in
both cases. A tendency to lower values was observed when adding B66 to the B59,
whereas the opposite was observed by adding V32. The dispersive part of the surface
tension for pure B59 was estimated to be 39 mN/m2 and approached 35 mN/m2
when adding B66, and 30 mN/m2 in the V32 case. An overall surface tension was
calculated, including the sum of the polar and the dispersive parts and the results
from the Neumann approach. This value started out at 43 mN/m2 for pure B59 and
approached 36 mN/m2 for booth B66 and V32. Note however, that the calculated
values for each measurement method show a large deviation in their absolute value,
but a similar slope, as found in paper IV.
During the surface tension analysis a correlation to the results obtained in the deR
resin as presented above
wetting or contact angle experiments with ORMOCER
could not be found. No significant difference in the surface tensions between the
surfaces formed by pure B59 and altered by 0.1 wt% V32, which actually showed
83
5 Characterization
45
Surface tension γ in (1 mN/m)
40
35
30
25
γ p for B59:B66
p
γ for B59:V32
d
γ for B59:B66
d
γ for B59:V32
γsum
20
15
10
5
0
100 wt%
B59
compositon
100 wt%
B66/V32
Figure 5.13: Surface tension of the system B59:B66 and B59:V32, overview. The data
represent average values of the surface tension values estimated.
good wetting was detected. The same holds for pure B59 and 5 wt% B66. Even here
no significant difference in the calculated surface tension values were found.
5.2.3.5 Complementary optical characterization
The optical loss (wavelength 1310 nm) of the waveguides (cross section 20 µm x
10 µm) was obtained with the single-length measurement method. The waveguideends were polished roughly and additionally a refractive index matching fluid was
applied to compensate for losses due to remaining end-facet roughness. Light was
coupled into the waveguide at one end using a single mode fiber (core diameter 9 µm)
and collected on the other side connecting a large-core multimode fiber (core diameter
50 µm). The measurement results are summarized in Table 5.2.2.
For the B59:B66 system a significant difference in the optical loss values obtained
between the sample with a pure B59 (0.42 dB/cm) and a B59:B66=95:5 wt% (0.55
dB/cm) waveguide core, respectively, could be observed. Contrary to this, the B59:V32
system showed no clear trend in optical loss values by moving from pure B59 as core
material to the core material modified by 5 wt% of V32. Optical loss values are in
the range between 0.4 and 0.47 dB/cm. Moreover, the measurement values for the
experiment and the reference of a sample with a core composition of B59:V32=95:5
wt% differ significantly, with values of 0.42 and 0.47 dB/cm, respectively.
84
5.2 Surface tension
Table 5.2.2: Optical loss measurements on chosen core materials.
Core
material
Cladding
material
Optical loss at
1310 nm
in (1 dB/cm)
C1
C4
C5
C5
0.42 ± 0.02
0.55 ± 0.10
6
12
4
4.5
C1
C7
C9
C9†
C10
C10
C12
C12
0.44 ± 0.03
0.4 ± 0.01
0.42 ± 0.02
0.47 ± 0.03
6
12
12
5
4
4.5
4.5
4.5
†
Number of
individual
measurments
Waveguide
length in (1 cm)
reference measurement
In all samples light scattering was observed due to particles, which can be attributed
to not optimized waveguide forming processes. The optical loss values presented are
up to a factor of two higher than previously reported values for pure B59 as waveguide
core [19]. The altering of the waveguide core compositions showed significant differences during measurements, although these were not considered as a major effect. The
divergence in the optical loss values (both previously reported and measured here)
can be attributed to the samples, due to light scattering on particles was observed
on them. Additionally, the resolution of the measurement method applied is limited
by factors such as, preparation of the end-facets, in homogeneities of the end facets
(particles), and a numerical aperture mismatch between fiber and planar waveguide.
Therefore, the results of the system B59:V32 suggests that the influence of the V32 on
the quality of the waveguide is below the detection limit of the optical characterization method applied. Contrary to this, as the B66 content in the core layer increases,
the optical loss values increase slightly also according to this measurement setup. It
can be concluded, that the modification of the waveguide core shows a minor effect
on the optical loss properties. Further investigations on optical loss values, such as
the application of the high-resolution cut-back method, in connection to the altering
of the waveguide core with a refractive index tuning agent are left to another study.
5.2.3.6 Contact angle and surface tension — Conclusions
The earlier observed de-wetting problem of pure B59 liquid films during thin film
processing on a particular surface of B59:V32 was circumvented by adding small
amounts of V32 (0.1 wt%) to the B59. Similarly, moderate amounts of B66 (5 wt%)
added to B59 avoided de-wetting on a B59:B66 surface.
For thin film multilayer processing using different material compositions in sequential
85
5 Characterization
polymer films it is thus concluded that small or moderate changes in composition
can significantly reduce differences in surface properties, thus avoiding de-wetting
problems. Material bulk properties can change by applying this process route, which
might be compensated for by fine-tuning of the mixture’s composition. However,
no significant major changes in the optical properties of the B59:B66 system and no
significant differences for the system B59:V32 were observed in our case. Additionally,
no other surface treatment, such as plasma oxidation or silanization, was necessary to
achieve this effect, thus enabling savings in processing steps and cost. Furthermore
this enables large-area processing where suitable equipment for surface activation is
not available.
From the surface energy estimation it was found that the overall surface energy of
the B59 host (cured surface) was reduced when adding B66 or V32. Resins of these
systems clearly show this behavior when the solvent is removed by evaporation. Thus,
the solvent’s influence can partly mask the final surface tension of a resin that is of
practical importance for the de-wetting behavior, i.e. the surface tension of the liquid
resin without solvent.
De-wetting tests under process realistic heating conditions are thus considered as a
fast and good working tool to expose de-wetting behavior, which otherwise can be
hidden in common surface tension tests.
86
6 Large-Area Panel Processing
Paper II, found in the Appendix, presents our contribution to the rather extensive
list of optical motherboard and backplane approaches. Here, the main focus is on
manufacturing aspects, to be able to be competitive on the market and to meet
customer demands. Solutions for key issues such as light coupling have been treated
in chapter 3 and are therefore not considered here.
6.1 Philosophy
As stated above, it is believed that the market for optical motherboards and backplanes is a niche-market only (see section 2.2). A company producing such equipment
is therefore forced to react quite fast on the demands of its customers. Furthermore
two factors have to be taken into account: Firstly, a break-even length for optical
links has to be estimated (see paper III and references therein and additionally [52]).
Beyond this length, optical interconnects are superior to electrical ones both in view
of technical solutions and economical feasibility. Secondly, the waveguide patterns
and the electrical lines have to be created over a large area, up to a board size of,
e.g., 457.2 mm x 609.6 mm. This can be done by direct laser writing applications,
which has proven its feasibility for high-quality waveguide processing for long links
[43, 63]. A drawback in this case is processing time for complex waveguide patterns,
as described in section 4.2.3. An alternative is to use a large-area mask aligner, such
as the ANVIK Hexscan 2050 SME [70]. This machine is capable of exposing panels
up to a size of 609.6 mm x 609.6 mm, using photolithography masks up to 355.6 mm x
355.6 mm in a step-out procedure. With this tool it is possible to perform a step-wise
scaling of backplanes from very small sizes, e.g. 101.6 mm x 101.6 mm up to full-scale
boards.
The idea behind our project was to introduce a method suitable for small-size companies to be able to react fast to the market’s demands. It is based on the principle
that the optical backplane is divided into functional units, 101.6 mm x 101.6 mm
in size. Each of these units contain several chips with specific functions. The photolithography mask set of such a unit provides the infrastructure within the unit and
the interface to the outer world. In that way, communication within and in-between
units is established with one set of masks over the whole board. Below the breakeven length, interconnects between components will be provided by electrical high
bit rate links. For larger distances, e.g., in between components on different units,
optical interconnects will be chosen. The transformation from electrical into optical
signals and vice versa will be handled through VCSEL and PD banks, respectively.
87
6 Large-Area Panel Processing
Figure 6.1: Example of a standardized unit. Each unit can contain different types of
MCM / SOP / SOC, in order to achieve a functional package in each unit.
Furthermore, VCSEL and PD banks are used for electro/optical signal conversion. The electro/optical in-/out-put modul can be used for external signal
handling, e.g., coupling to/from an optical fibre network. The optical waveguide routing should be fixed in such a way that different types of 101.6 mm x
101.6 mm modules can be placed next to each other with guaranteed optical
interconnects in between.
In Fig. 6.1 such a functional unit is shown in detail.
The infrastructure on one unit will be designed in such a way that different types
of units can be placed next to each other, by providing an infrastructure for optical
interconnects over the whole board at the same time. Fig. 6.2 shows this in detail
once more. Here, the functional units named a,b,c, . . . are spread-out all over the
board. Within these units electrical interconnects is predominant, whereas an optical
link between the units a and b is established.
From Fig. 6.2 it is clear that the small photolithography masks have to be aligned
carefully with respect to each other. Ideally, in one of the first process steps alignment marks on the substrate with a large-area mask will be created to support the
alignment of the unit masks. To increase the misalignment tolerance, four different
endings in the overlap were proposed, such as butt-coupling, butt-coupling with overlap, funnel coupling, and a saw tooth coupling as shown in Fig. 6.3.
88
6.1 Philosophy
-
609.6 mm
6
a
a
b
c
a
c
b
a
c
a
c
b
609.6
mm
6
203.2
mm
?
-
?
101.6 mm
Figure 6.2: Theoretically possible 609.6 mm x 609.6 mm optical backplane. The LAP is
tiled into units 101.6 mm x 101.6 mm in size. Different types of units are placed
next to each other, serving different functions (labels a, b, c). One possible
pathway for optical interconnects is shown (arrow).
toward
step#1
toward step#2
Figure 6.3: Four different waveguide end-facets to demonstrate flexible optical interconnects by lithographic step-out processes (from top to bottom): saw–tooth with
overlap, funnel with overlap, straight with overlap, straight tight end-facet.
89
6 Large-Area Panel Processing
In a pre-test for the step-out procedure a misalignment between the waveguide ends
of ≤5 μm was observed. The misplacement can be recalculated to an overlapping
cross section of the two waveguides (see eq. (6.2)).
A=h∗w
Ared = h ∗ wred
h wred
Ared
= ∗
A
h
w
Ared
= xborder
A
wred
→ xborder =
w
(6.1)
(6.2)
In eq. (6.2) A is the cross sectional area of the waveguide, h and w the height and the
width of the waveguide, and Ared and wred the effective intersection dimensions due
to the misalignment of the step-out. The quantity xborder represents the fraction of
the cross section of the junction between the two waveguides due to this misalignment
(compared to perfect alignment). The optical loss L is defined as
L = 10 ∗ log
I
I0
(6.3)
with I as the out-going light and I0 the incoming light. Combining eq. (6.2) and
(6.3) together gives:
wred
I
=
I0
w
= xborder
L = 10 ∗ log xborder
(6.4)
Assuming a waveguide width of 70 μm [47] and a mismatch of 5 μm in between two
waveguide ends would result in an ideal optical loss of 0.3 dB over this connection.
This is not dramatic for the crossing between two units, but has to be considered when
nine borders have to be crossed for a total link length of 1 m. The loss originating
from the step-out mismatches then becomes 2.7 dB in total.
The task for the manufacturer is, to create a portfolio containing small masks to be
able to create a wide range of infrastructure patterns for the optical and electrical
interconnects. One might argue: Why not just create one huge mask and expose the
whole board at once? The answer is rather easy: flexibility and costs of production! If
optical backplanes were mass-produced, with thousands of units shipped a day, then
the production of large-area photolithography masks would be feasible.
Actually, optical backplanes seem to be found in high-end applications only and this
points to a niche-marked of them. Therefore, design changes for the whole board
shall be possible with low excess cost for the mask-set. Large-area masks will here
90
6.2 The Large-Area Panel (LAP) optical backplane demonstrator
presumably increase the projects budget and small masks gain a favour. The customer decides the overall function of the board and the system engineer chooses the
infrastructure based on the portfolio. Finally, the board will be processed according
to these specs. In such a way different types of smaller optical motherboards up to
optical backplanes for rack use containing parallel stripes of optical waveguides, can
be produced.
6.2 The Large-Area Panel (LAP) optical backplane
demonstrator - Processing and Results
Full details for the processing are given in Paper II, found in the Appendix. Here,
the process will described in brief and key-results will be presented.
For the demonstrator a panel of 610 mm x 610 mm size, made of standard FR4 material, was used. All polymer layers were spun on using a large-area spin coater of
type Karl Suss GYRSET RC33 (see Fig. 6.4). The photolithographic process was
performed using the ANVIK Hexscan 2050 SME (see Fig. 6.5), and special equipment
suitable for handling LAP, such as, a convection oven, LAP sputter (KDF 844NT),
and large bowls for the etch-process.
Two types of photolithography masks were used: an 8-inch mask containing alignment
marks and a 4-inch mask containing the waveguide pattern. In a first sub-process the
alignment marks were created on the LAP by stepping out the 8-inch mask over the
whole panel. For this, the panel was first coated with a metal layer (Titanium, sputter deposited) and photolithography processes were applied to create the alignment
marks. It was observed that a specific width of the photoresist pattern of these marks
was 5 μm ±0.5 μm larger then the etched alignment marks, which was attributed to
handling issues during the etch-process (see Fig. 6.6).
Figure 6.4: Large-area panel spin coater Karl Suss GYRSET RC33.
91
6 Large-Area Panel Processing
Figure 6.5: Anvik Hexscan 2050 SME large-area panel mask aligner. In front: 4-inch
waveguide mask, 8-inch alignment marks mask, 610 mm x 610 mm large-area
panel substrate.
After these preparation processes, the waveguide layers of the demonstrator were
built-up on the substrate. The following thickness of the sub layers were achieved:
• buffer layer to prevent penetration of light from the waveguide towards the
substrate (5 μm)
• under cladding (15 μm)
• core layer (10 μm)
• upper clatting (15 μm)
The materials B59 (aka ORMOCORE) (Microresist Technology GmbH, Berlin, Germany; licensed from FhG-ISC) and the refractive index-tuning agent V32 (research
state, FhG-ISC) were chosen for the waveguide layers. In advance, refractive index
studies with these two materials were performed. Mixtures in different compositions
were created and thin films on wafers were produced with them. The refractive index of the films was measured with the prism coupler method (see section 5.1.2.2 for
details). In Fig. 6.7 the resulting diagram is shown.
Initially, pristine B59 was intended to be applied as waveguide core material and a
mixture of B59 and V32 as cladding material. During pre-tests for the LAP-process
de-wetting effects of a liquid spun-on B59 on a cured cladding layer surface were discovered. The effect could be prevented by plasma activation or silinazation processes
of the under laying surface. Both were not available at the department’s clean room
for the large-area processing. Instead, the application of a modified core layer, by
adding 5 wt% of V32 to the B59, on the cladding layer could prevent from the dewetting effect. Thus, the whole refractive index window for the core-cladding-layer
system was shifted to lower values (compare with Fig. 6.7). The final composition of
R
layers applied during the process can be found in Table 6.2.1.
all ORMOCER
92
6.2 The Large-Area Panel (LAP) optical backplane demonstrator
(a) Alignment mark, photoresist on Ti.
(b) Etched alignment mark in Ti on SU-8-5 on
top of a LAP.
Figure 6.6: Photoresist pattern and etched alignment marks in form of an arrow, pointing
at the lower right corner of the picture. The dimension of the tip of the arrow
(120 μm ∗ 120 μm) was the evaluated feature size as described in the text.
The waveguide’s shape is crucial for the optical properties of the waveguide as
described above (see section 4.2). Therefore, a cross section of the full waveguide
cladding stack was analysed by Scanning Electron Microscopy (SEM). This analysis
method has one drawback for surfaces of non-conductive samples: charging occurs;
making it impossible to detect any features with the SEM. Furthermore, the contrast
between the polymer layers might be too low to see anything at all due to the fact that
R
class in both cases). Thus, a test wafer was
they are similar materials (ORMOCER
prepared, applying the same process parameters as for the real waveguide process.
In contrast to the standard processing, a Titanium layer of d=10 nm was deposited
after every polymer layer curing step. Thus, a thin and high-conductive metal layer
surrounded the waveguide core layers and could be clearly observed using SEM (see
Fig. 6.8).
Waveguides in a length of about 20 cm were created during this project by stepping
out the 4-inch masks onto the high-precession 8-inch alignment mark pattern. The
procedure of alignment of the 4-inch masks was very time consuming (1 h for the
test panel in this single case) due to difficulties of the machine’s pattern recognition
system for the alignment marks. With engineering efforts on alignment marks and
the pattern recognition system the time for this process step may be shortened considerably. However, the overall optical loss for the 850 nm wavelength was determined
to be 0.6 dB/cm, and 0.7 dB/cm for 1320 nm. No difference in the optical loss values
depending on the coupling method between the step-outs could be detected. The
measurement values are given in Table 6.2.2.
93
6 Large-Area Panel Processing
1,5600
Refractive index
1,5400
1,5200
1,5000
n @ 632.8 nm
1,4800
n @ 1300 nm
1,4600
1,4400
0
25
50
75
100
B59 in wt%
Figure 6.7: Refractive index of mixtures of V32 and B59.
R
Table 6.2.1: Refractive indices of ORMOCER
systems used. Standard deviation is related to machine setup error.
Material
Composition
in (1 wt%)
Refractive index
at 632 nm
Refractive index
at 1300 nm
V32
B59
pure
pure
1.466 ± 0.001
1.550 ± 0.001
1.456 ± 0.001
1.535 ± 0.001
B59:V32
B59:V32
B59:V32
1:1
83:17
95:5
1.503 ± 0.001
1.535 ± 0.001
1.545 ± 0.001
1.491 ± 0.001
1.521 ± 0.001
1.531 ± 0.001
a
base material, delivered by ISC
base material, delivered by MRT Berlin
c
reflection layer
d
cladding
e
core material
b
94
Remark
a
b
c
d
e
6.2 The Large-Area Panel (LAP) optical backplane demonstrator
Uppercladding
)
Waveguide
core
Undercladding
9
Reflection
layer
Substrate
Figure 6.8: SEM photograph of a cross-section through a waveguide structure containing
a reflection layer, under cladding, core, and upper cladding.
Table 6.2.2: Results of transmission measurements of the FR4 sample. Waveguide length
l = 19.3 cm. One waveguide with a width of w = 8/12/20/30/50 μm measured
in each “border crossing group”.
Type of border crossing
“saw tooth”
“funnel”
“straight, overlap”
“straight, tight”
Optical loss in (1 dB/cm) at
λ =1320 nm
λ =850 nm
0.6 ± 0.2
0.7 ± 0.2
0.6 ± 0.2
0.8 ± 0.1
0.5 ± 0.2
0.6 ± 0.1
0.5 ± 0.1
0.5 ± 0.1
95
6 Large-Area Panel Processing
6.3 Large-Area Panel processing — Summary
The concept of the flexible manufacture approach, enabling fast reacting to the demands of a niche-market of optical backplanes was introduced. A LAP demonstrator was produced, proofing the concept of manufacturing proposed. The LAP was
R
layer in total, without showing any cracks in
coated with a 45 μm ORMOCER
R
the ORMOCER layer or any deformation of the board. The optical loss of the
waveguides was measured to be 0.6 dB/cm at 850 nm and 0.7 dB/cm at 1320 nm,
respectively. These values are around 3 times higher than previously determined values (0.23 dB/cm at 1320 nm [198]), which was attributed to the application of the
refractive index tuning agent in the core layer and the cladding layer, in the first place.
Secondly, the direction of the laser sweep was chosen to be perpendicular to the optical axis of the waveguides. In a visual inspection of samples, produced on Si-wafers,
stripes of inhomogenous exposure were observed. These originate from the method
of exposure of the ANVIK (overlap region of the hexagonal exposure window). The
waveguides produced on the LAP are crossing several of these stripes, which can raise
the optical loss.
In further investigations (see section 5.2.3.6 and paper IV) the waveguides processed
with the Karl Suss Mask Aligner MA6BA6 showed double the loss as previously reported [19]. Thus, non-optimized processes and the exposure by the ANVIK mask
aligner probably increased the losses in the waveguides. Additionally, samples should
be evaluated by the cut-back method instead of the single length method for their
optical losses.
However, the concept of a flexible manufacturing approach was proven for an area of
8-inch in size using a step-out procedure for 4-inch waveguide masks.
Tasks for future projects would be to extend this concept to a full board-size optical
interconnects. Furthermore, the concept has to be proven in real-life mixed systems,
containing both electrical and optical interconnects.
96
7 Amplification Devices for Optical
Interconnects on Optical Backplanes
In this chapter, a brief optical power budget analysis for a link based on planar
polymer waveguides on an optical backplane will be performed. The optical loss values
used are based on the information presented in chapter 4 (Materials and Processing)
and 3 (Functional Structures). Based on the outcome and the resulting limitations
for optical backplanes, the concept of an optical waveguide amplifier will be proposed
to overcome these limitations in the optical power budget. This section constitutes
an introduction to paper III, found in the Appendix.
7.1 Problem Statement — Origin of optical loss on
optical links
The discussion in this section is limited to optical waveguides, consisting of a core
area with a high refractive index and a cladding area with a low refractive index.
Two different types of waves can be generated in optical waveguides; single mode and
multimode. A single mode wave can be excited in a waveguide with a small core
cross-section, e.g. 9 μm in diameter for a wavelength about 1320 nm. If larger crosssections are used, then multiple modes of the lightwave are generated in the core.
In practice, two things are of major interest for optical links: the optical transmission
loss over the link, and the dispersion of high bit rate signals. When a narrow pulse
of light enters a waveguide it will experience pulse spreading after a certain distance.
This means that the time duration of such a pulse gets longer, which causes a problem
in signal resolution if two pulses are sent directly after one another through the
waveguide. In the worst case, two pulses are coupled into the waveguide, but only
one can be resolved on the other side. This is not so critical for the single mode, but
it is for the multimode case. In the latter there is a difference between step index
and graded index multimode structure resulting in the largest pulse increase for step
index multimode structures [24] but no pulse spread for ideal graded index profiles.
In the case of optical backplanes, multimode waveguides are feasible, due to the
fact that the pick-and-place equipment is limited by its placement accuracy during
processing [47]. Furthermore, the fabrication of multimode graded index waveguides
is not feasible on optical backplanes, due to limitations in the planar processing of
the optical layers on backplanes.
Despite the waveguide dispersion dependence on the cross-sectional dimensions and
97
7 Amplification Devices for Optical Interconnects on Optical Backplanes
the numerical aperture of the waveguide, a so-called chromatic dispersion exists too.
Signals will not only propagate on a single wavelength over one waveguide. Several
wavelengths will be transmitted at the same time to be able to put more information
on one waveguide simultaneously. This means that pulses with different wavelengths
travel at different speed through the waveguide, due to different refractive indices.
In summary: an optical waveguide link is limited by optical loss due to material
properties, i.e. direct influence on the optical power budget of the link, and dispersion
for high-speed pulses due to the waveguide design and due to chromatic dispersion,
i.e. limitations in bandwidth for the link.
The discussion of signal dispersion is strongly connected to the step in refractive
index of optical (multimode) waveguides. The characteristic parameter in this case is
the so-called bandwidth x length (BW L) product.1 This term gives an idea on the
bandwidth that can be expected after a certain link length.
Senior [195] presented a formula allowing the estimation of this BW L of an optical
link from its numerical aperture, i.e. its refractive index step, as
BW L = nc ·
n2c
c
− n2cladd
(7.1)
where nc and ncladd stands for the refractive indices of the core and cladding material,
respectively, and c is the vacuum speed of light. In Fig. 7.1, calculations of BW Ls
according to tentative nc and ncladd values are shown. BW L values higher than 20
Gbps*m are only possible if the difference of the refractive indices between core and
cladding is chosen to be below 0.008. At such low refractive index steps the control in
the absolute level of the refractive index becomes crucial. A slight fluctuation in the
refractive index in the cladding layer, for instance, can reduce the theoretical possible
bandwidth dramatically, e.g. from 40 to 30 Gbps*m, and therefore limit the bandwidth of the overall system. At larger steps in refractive index the BW L situation is
more relaxed, as seen in Fig. 7.1 by the lower slope of the curves.
Furthermore, the change of the absolute level of refractive index has no significant
influence on the BW L. Calculations were done with other refractive index windows
than the one for polymers around n =1.5, e.g. with values as n =2, n =2.5, and
n =3, similar to the calculations presented in Fig. 7.1. The difference of these BW L
values to the ones of n =1.5 were calculated, maintaining one and the same step in
refractive index, which resulted in a rise of 0.012, 0.019, and 0.024 Gbps*m, for the
refractive index windows around 2, 2.5 and 3, respectively. This is not a significant
increase in the BW L, since the absolute values are still of the order of magnitude of
Gbps*m and above for the refractive index steps applied.
It was shown that minimizing the refractive index step would increase the BW L of
an optical waveguide system. A drawback of this step is that the small difference
in refractive index will increase the optical loss of the overall link, due to the fact
1 In
chapter 2 on page 7 was pointed out that the terms bandwidth and bit rate are used interchangeably in digital optical interconnects. This concept is followed in the discussion in this
section too.
98
7.1 Problem Statement — Origin of optical loss on optical links
BWL at ncore
12
Bandwidth times length in (bps*m)
10
BWL=ncore*c/(n2core−n2cladd)
11
10
n
=1.5353
core
10
10
ncore=1.5453
1.54
1.535
1.53
1.525
1.52
1.515
Refractive index of cladding ncladd
Figure 7.1: Bandwidth of a multimode waveguide as function of its numerical apperture.
Each line represents a specific refractive index value of the core. Choosing an
appropriate pair of refractive index for core (nc ) and cladding material (ncladd ),
the maximum bandwidth length product can be read directly from the graph.
Empirical model according to Senior [195].
that the number of leaky modes of the waveguide and bending losses originating from
signal routing will increase. The advantages and disadvantages of reducing the step
in refractive index are summarized once more in Table 7.1.1.
Next, the optical power budget available for a link on an optical backplane will
be estimated. This budget depends mainly on two factors, the output power of the
laser diode and the minimum detectable power by the PD. These are both temperature and bandwidth dependent characteristics of the device. Commercially available
laser diodes at 850 nm deliver between -1 and -1.5 dBm at 2.5 and 3.3 Gbps, respectively [208, 209]. Karppinen et al. [114] used a laser-diode providing 3 mW (4.8
dBm) nominal output power at 850 nm for 8 GHz in bandwidth. Further values of
recently published advances in VCSEL design and fabrication report higher values for
both bandwidth and output signal power, as summarized in Table 7.1.2. Especially
for the case of 1310 and 1550 nm the temperature has to be controlled precisely,
otherwise the laser power will drop dramatically. Furthermore, from the table it is
evident that direct modulation of light is possible only up to frequencies as high as
13 GHz [210]. For generation of higher data rates, an external/off-chip modulator is
necessary, whereby Mach-Zehnder interferometers with a modulation bandwidth up
to 24 GHz by an insertion loss of 4 dB [211] and 2 dB at 10 GHz [212], respectively,
99
7 Amplification Devices for Optical Interconnects on Optical Backplanes
Table 7.1.1: Impact of change in step of refractive index on properties of an waveguide
system.
NA
Bandwidth
Bending radius
decrease Δn
increase Δn
decrease
increase
increase
increase
decrease
decrease
have been reported. A drawback of these solutions is that these devices are working
horizontally, whereas VCSELs are working vertically, which opens new challenges for
device manufacturing and packaging technologies.
Commercially available PDs show a sensitivity of -16 dBm at 3.3 Gbps and 850 nm,
which gives a power budget of 14.5 dB at this bandwidth by a laser output power of
-1.5 dBm [208].
Yoon et al. [85] reported on an 8 Gbps optical link on an optical backplane. Their
power budget was reported to be 14 dB. Thus, based on the applied input power
of 2 dBm, this results in a PD sensitivity of -12 dBm at a wavelength of 850 nm.
Karppinen et al. [114] reported on an optical link with 4.7 dBm laser input power
and an noise level for the photo diode of –26.5 dBm at 850 nm, working at 8 GHz in
bandwidth. A margin of 11.5 dB between signal and noise was required to achieve a
BER of 10−12 , although requirements from the industry was reported to be BER of
10−15 [201]. However, a power budget of 18 dB between VCSEL and PD is available
in this case.
However, for all further examinations, we will focus at a working wavelength around
850 nm, and a bandwidth of 8 Gbps with a power budget of 14 dB, since this value
seems to be state-of-the-art.
Above, the optical power budget available between the VCSEL and the PD was estimated. In the further analysis below, a tentative loss scenario for a realistic routed
link on an optical backplane will be sketched.
In chapter 3 losses for functional structures were given, such as for waveguide routing (signal splitters), MUX/DEMUX, and coupling losses. Further losses will be
introduced by bending of the waveguide connected to the step refractive index (for
bandwidth reasons) as outlined above. Furthermore, losses will be introduced through
the waveguide material itself. Here, a loss value of 0.08 dB/cm will be assumed in
the first instance. Table 7.1.3 summarizes typical routing losses once more.
In Fig. 7.2, example scenarios containing a simple optical link on an optical backplane is presented. The power budget for the whole link was set to 14 dB as outlined
above. All other optical losses are chosen according to Table 7.1.3. Fig. 7.2 shows
that the coupling- and routing-dependent losses reduce the optical power at a specific
100
7.1 Problem Statement — Origin of optical loss on optical links
Table 7.1.2: Recent advances of VCSEL at specific wavelengths with corresponding bandwidth and signal output power. cw = continuous wave.
Wavelength
in (1 nm)
Modulation
in (1 GHz)
Output power in
(1 dBm)
Reference
850
cw, 10
cw
13
5.8
4.7
1.7
[213]
[210]
[210]
980
cw
30.3
[214]
1310
10
cw
[2.6; -1.7] b
[5.8; 1.5] c
[215]
[215]
1500
cw
[6; 2.3]
a
b
c
a
c
[216]
no output power given for modulated signal
for T =[293;343] K
for T =[313;358] K
place on the link substantially. This happens for instance right after a 1-to-8 signalsplitter or in connection to coupling of the optical signal. Thus, a large input power
and a very good resolution of the PD is essential for the power budget of the whole
link.
Furthermore, waveguide losses below 0.08 dB/cm are adequate to establish a pointto-point link over 1 m including light coupling, at a given optical power budget of 14
dB. However, the power budget is consumed very fast if further signal routing or a
higher data-rate signal (external modulator, including additional routing) is needed
and/or low-loss waveguide materials are not available. Additionally, it it is shown in
Fig. 7.2 that a large power budget available between VCSEL and PD can relax loss
issues significantly.
However, it is evident from this example, that optical amplifiers on optical backplanes
are needed to be able to provide a sufficient power budget when it comes to large-area
multifunctional approaches.
Pre-amplification at the waveguide input right after the VCSEL, and even postamplification at the output right before the PD, is necessary to get a reasonable
signal resolution. Additional signal amplification on the link is necessary in many
cases to reach sufficient optical power levels after signal routers and multiplexers.
101
7 Amplification Devices for Optical Interconnects on Optical Backplanes
Table 7.1.3: Routing loss on on optical backplanes.
Functional Structure
Optical loss in (1
dB)
Reference / Remark
Optical waveguide
8 (over 1 m)
tentative
Light Coupling Loss
3+3
based on estimations from Table 3.4.1 on page 40, in- and
out-coupling of light to/from the
waveguide
“Border Crossing” /
Flexible manufacture
Approach
2.7
9 borders, see section 6.1, page
90
Waveguide Switch
1.8
[122]
Waveguide Splitter
9
1–8, ideal signal attenuation per
output waveguide
MEMS routing
4
[118]
102
7.1 Problem Statement — Origin of optical loss on optical links
0
Power loss in a tentative backplane (dB)
waveguide loss
−5
incl. in−/out−
coupling of light
−10
incl. border crossing (9x)
−15
−14 dB power budget line
−20
incl. waveguide switcher
incl. MEMS optical routing
−25
0
20
40
60
80
100
Interconnection length (cm)
Figure 7.2: Power budget for tentative optical interconnects on board level.
103
7 Amplification Devices for Optical Interconnects on Optical Backplanes
7.2 Amplification
There are two main approaches for optical amplifiers based on silicon-bench technology:
• waveguide amplifiers, which need a pump light source to be able to amplify an
incoming signal, and
• semiconductor optical amplifiers, which need a coupling to a power source to
be able to amplify the incoming signal.
The amplifier device presented hereafter uses both approaches for light amplification
in separated design set-ups.
7.2.1 Waveguide amplifiers
The principle applied here is the classical laser transition in optical active media [217].
A host material contains active elements, which can be, e.g., Rare Earths (see Table
7.2.1) or organic molecules such as Rhodamine B [218]. A pump light signal causes
an inversion state of electrons in the active elements. The low power signal triggers
a cascade-like transition, whereby the electrons jump from the inversion state to a
ground state. The difference in energy between the inversion and the ground state is
the same as the energy of the emitted photon and the triggering photon. This process
is also called stimulated emission of light, and is in fact the classical laser principle,
which is visualized in Fig. 7.3.
Table 7.2.1: Rare earth elements with suitable emission wavelengths in the optical communication windows [219].
104
Element
Active Ion
Suitable emission wavelength
in (1 nm)
850
1550
Erbium
Er3+
Neodymium
Nd3+
1360
Praseodymium
Pr3+
1300
Thulium
Tm3+
810
1480
1510
7.2 Amplification
Electron
Energy Levels
6
6
Excited
State
66
Up-conversion
Non-radiative Transition / Relaxation
q
Meta-stable Level
-Photon
PUMP
(Absorbtion)
LASER
Stimulated Emission
?
Ground
State
Figure 7.3: Laser principle, energy levels, electron transitions, stimulated emission of photons. The cascade-like stimulated emission of photons is triggered by an external photon, which has the same energy as the energy difference between the
meta-stable inversion state and the ground state of the electrons.
Signal amplification is necessary for the wavelength bands around 850, 1310 and 1550
nm, which can be achieved by, e.g. rare earth elements, such as those presented in
Table 7.2.1. Optical amplification in the 550 to 650 nm wavelength region by Europium [220], Samarium [221], and Rhodamine B (organic dye) [218] has also been
reported.
An overview on recent advances on optical waveguide amplifiers based on silicon bench
technologies is given in Paper III. Below, recent results for waveguide amplifiers based
on Rare Earth elements is listed (see Table 7.2.2). Research seems to be concentrated
on the wavelength regions of 1060 nm and 1530 nm for optical amplification, while
Table 7.2.2: Optical amplifiaction based on silicon bench technology and rare earth doped
waveguides.
Active element
Er
Er
Nd
Amplification
wavelength in (1 nm)
1533
1534
1060
Amplification in (1 dB)
13
18
8
Reference
[222]
[223]
[224]
105
7 Amplification Devices for Optical Interconnects on Optical Backplanes
the 850 nm and 1310 nm bands seem to be neglected. However, fluorescence measurements were reported for all active elements mentioned in connection with planar
waveguides, excluding the 850 nm regions. Working devices for Er were found in the
literature for sol-gel, polymer, and sputtering approaches, whereas for Nd, Pr, and
Tm, the results presented contained fluorescence measurements only, or gain measurements were reported at unsuitable wavelengths.
It was reported in the literature that the length of planar optical amplifier waveguides
is often of the order of centimeters. This originates from the fact that the intrinsic
loss in an amplifier waveguide cannot be compensated through the amplification process after a certain waveguide length. Thus, the optical gain saturates after a certain
waveguide length.
Furthermore, the host material is usually loaded with the active elements up to a
concentration where segregation occurs. This causes quenching effects in light amplification (concurrent up-conversion, non-radiative relaxation), which often occurs if
more than a couple of weight percent of the rare earth element is introduced in the
host material.
The pump signal is attenuated due to the loss properties of the amplifier waveguide.
This light is absorbed and consumed in two main processes: exciting electrons from the
ground state to the inversion state and non-radiative processes, such as up-conversion.
However, a sufficient amount of the pump signal has to be available at every place
of the waveguide to ensure the generation of the inversion state of electrons, which is
the requirement for the laser effect. Additionally, the pump signal will be attenuated
depending on the host material, especially with polymer hosts, and the quality of the
waveguide [225]. Thus, an optimum solution for coupling of the pump light has to
be found, ensuring that sufficient power is available at every place in the waveguide
to create the inversion state. At the same time the effect of up-conversion has to be
minimized, which means a trade-off between launched pump power and optical gain
has to be found.
Several approaches have been reported for coupling the pump light into the amplifier waveguide, such as, directional coupling, counter directional coupling, and bidirectional coupling. Additionally, the gradual coupling method, as proposed by Slooff
et al. [225], and the sideway-excitation, as proposed Fujii [226], have been reported for
silicon-bench technology. The idea behind gradual coupling of pump-light is to place
a second waveguide beside the amplifier waveguide, thus coupling the light gradually
over, from one waveguide to the other. Slooff et al. [225] showed that a theoretical
gain increase from 0.005 dB to 1.6 dB was achieved according to their calculations.
Fujii [226] emphasized in 1999 the heat and efficiency problem with polymer optical
fiber amplifiers. Very high gains have been achieved in graded index fibers by incorporation of rhodamine B in a PMMA host [227]. An optical gain of 28 dB was measured
by applying a pump power of 7 kW. These high pump powers were necessary because
the system must be excited by a pulsed laser. A rather large amount of the pump light
is converted to heat and not used to create the inversion state necessary for the laser
effect. This results in a temperature rise that almost melts the core material [226].
106
7.2 Amplification
The proposal of Fujii [226] shows a new approach. The polymer optical fiber (POF)
is coated with a tube, which acts as a continous wave laser. The POF is pumped
radially and not coaxially as in earlier approaches. Calculations showed that a gain of
20 dB can be reached with only a 5 cm fiber length and 20 W pump power compared
to 1 m fiber length and 7 kW as in the original case. The idea of sideway excitation
may be considered by the implementation of optical amplifiers into optical backplane
systems by using silica on silicon technologies. If tests show that it is insufficient to
obtain high enough optical gain values by applying coaxial excitation then the method
of sideway excitation may be considered. Surface mount technologies as flip-chip in
connection with arrayed VCSELs could be used.
The different approaches for pump-light launch explained above are visualized in Fig.
7.4 once more.
(a) Direct Coupled Pump Light. Signal
light and pump light are entering the active waveguide from the same side.
(b) Counter Directional Coupled Pump
Light. Signal light and pump light are entering the active waveguide from different
sides.
(c) Sideway excitation of the active waveguide.
Figure 7.4: Methods of launching the pump light into an optical waveguide amplifier.
107
7 Amplification Devices for Optical Interconnects on Optical Backplanes
7.2.2 Semiconductor Optical Amplifiers (SOA)
This type of device is based on the same principle as a Fabry-Perot laser diode [228]
SOA-tutorial. SOAs are mainly used in fiber optical interconnections and serve as
post-signal amplifiers for laser sources, compensators for in-line losses of the fiber,
and pre-amplifiers for PDs [229].
The key-structure of the device itself can be either a multi quantum well (MQW),
grown by metal-organic vapor deposition [229–233], or quantum dots [234, 235].
In a quantum well structure, a thin layer (up to 20 nm) with a lower band gap is
embedded within two layers of a higher band gap. Through this, charges are trapped
and an accumulation of electron-hole pairs is created in this layer. The electron-hole
pairs will spontanously recombine, whereby a stimulated emission can be triggered by
an (external) photon. However, the concept of the quantum well leads to an efficient
current-light conversion [24]. In a MQW case, several of these low-band gap layers
are deposited on top of each other, separated by thin high band gap layers. The
efficiency in current-to-light conversion is increased compared to the single quantum
well structure. However, quantum wells are a kind of 1-dimensional structures. They
are built up in the z-direction, while light is traveling through them in the x-direction
(see Fig. 7.5).
This concept is taken one step further in the case of quantum dots: Here, electron-hole
pairs are localized in a three dimensional structure. The first and second dimensions
are defined through the width and the height of the active region, defined by the
method of film deposition. The third dimension is defined through the length of the
quantum dot, which is realized through an etch-process [236].
In the MQW SOA structure, a passive (slab) waveguide structure lies below the
active MQW structure (some μm in height). The low-power signal passes through
the passive waveguide and triggers the laser transition of electrons in the active region
above. Thus, this transition leads to optical amplification of the signal in the passive
region through coupling of light between these waveguides. The light for the quantum
dot SOA device is directly guided through the active area. Material systems used are
Energy
6
AlAs
Quantum Well
Conduction Band
)
GaAs AlAs
i
z-direction
-
Figure 7.5: Bandgap structure for a quantum well.
108
Valence Band
7.2 Amplification
Table 7.2.3: Recent advances on SOA devices reported in literature.
Wavelength
in (1 nm)
Input
AmpliPower
fication
in (1 dBm) in (1 dB)
Saturation
Power
in (1 dBm)
Noise
figure
in (1 dB)
Reference
810
0.8
29
n/a
n/a
[232]
1310
1310
-25
-25
25
[22.5; 19.5]a
11.2
n/a
7.6
n/a
[229]
[231]
[1450; 1570]
1500
1550
[1410; 1500]
n/a
-10
4.1
n/a
15
18.2
15
>25
18.1
30
19.1
>19
4.5
9
n/a
<5
[230]
[233]
[234]
[237]
a
b
c
b
c
T = [298;338] K
at 40 Gbps input signal
at 40 Gbps
InP in combination with InGaAs [230], InGaAsP [229, 233], or AlGaInAs [231]. In
Table 7.2.3, recent advances in SOA devices, working on wavelengths of around 810,
1310, and 1500 nm are summarized.
The length of the active area of MQW SOAs was reported to be 725 μm [229, 231], 1.8
mm [230], and 1 cm [233], whereas the width of the waveguide was generally reported
to be around 4 μm.
Juodawlkis et al. [233] reported a slab waveguide approach for the SOA, resulting
in a fiber-to-fiber gain of 13 dB (single-mode). It was suggested that tapered SOA’s
are useful in multimode applications, whereby the device characteristics are limited
by beam instabilities associated with guiding dynamics. An alternative is the use
of a slab-coupled optical waveguide, where multimode waveguides can be made to
operate in single-mode by coupling them to a slab waveguide [233]. It seems from
the discussion in this paper that advanced device design is necessary to minimize the
insertion loss in a multimode waveguide environment.
109
7 Amplification Devices for Optical Interconnects on Optical Backplanes
7.3 The Flip Chip mountable Optical Waveguide
Amplifier (FOWA) approach
7.3.1 Overview
In the previous sections, estimates for the optical power budget on an optical backplane were presented. In conclusion, it is necessary to boost the optical signal at
certain places on a realistically routed optical backplane. Furthermore, approaches
of amplifiers based on silicon bench technology were briefly discussed. Unfortunately
they are not suitable for direct integration on large-area optical backplanes: In the
case of planar optical waveguide amplifiers, the whole board has to be coated by the
optically active material, and the active waveguide is formed by photolithography,
where nearly 99% of this material will be removed. Thus, the process is not very
efficient. Additionally, the sol-gel material needs process temperatures of around 730
K, which is definitely too high for a standard FR4 board.
The direct implementation of the SOA on a board is not feasible either. Common
SOA devices require high-precision mounting and advanced light coupling optics in
connection to multimode waveguides [233], which makes them unsuitable for direct
use on PCB backplanes. This would lead to positioning problems during the pickand-place processes during manufacturing.
Thus, in this section, the concept of a flip-chip mountable optical waveguide amplifier (FOWA) will be presented instead. The design of the device is chosen in such
a way that positioning with standard pick-and-place technology is possible without
introducing too high optical excess loss. Furthermore, the amplifier waveguide is on
a different plane than the rest of the waveguides on the backplane, due to the facts
that the pump-light source requires a certain vertical distance and solder-bumps for
the amplifier device are feasible only above a certain minimum size (see Fig. 7.10
for tentative measures). Thus, coupling and guiding structures (e.g., periscopes), as
explained in the former section will be applied here too.
In principal, the FOWA device works as follows:
1. The low-power signal is coupled into the periscope structure, and
2. guided inside the periscope from the backplane level to the amplifier level.
3. The low-power signal is coupled into the active region on the amplifier device,
i.e., the amplifier waveguide or the SOA.
4. Additionally, optical pump power is launched to the waveguide amplifier or
electrical pump power to the SOA, respectively.
5. Signal amplification takes place.
6. The amplified signal is coupled back into a second periscope, which guides the
light back to the backplane level, and finally
110
7.3 The Flip Chip mountable Optical Waveguide Amplifier (FOWA) approach
7. the high-power signal is coupled back into the waveguide on the backplane.
In Fig. 7.6 and Fig. 7.7 sketches are shown of the final device including a visualization
of the pathway of light.
Figure 7.6: Perspective view of the FOWA device, including a cylindrical lens with first
periscope (left-hand side), amplifier waveguide, and second periscope (righthand side).
Thus, the following key-features are necessary to be able to build such a device:
1. pump light source (mounting and coupling of light),
2. creation of an amplifier waveguide structure,
3. alternatively, high precision mount of the SOA,
4. forming of an optical periscope,
5. low-loss coupling from and to the optical periscope, and
6. high-precision mount of the FOWA device.
7.3.2 Coupling of light from the pump-light source
In section 3.2 approaches for the coupling of light to and from a waveguide from and
to VCSELs and PDs, respectively, were presented. Dannberg et al. [130] reported a
UV-reaction molding technology, enabling the fabrication of micro prisms for light
coupling purposes with a slope of 45◦ with respect to the horizontal substrate. A
photolithographic relief mask was pressed into a wet layer of an UV-curable polymer
R
in their case) and immediately exposed using a Karl Suss MA6BA6
(ORMOCER
mask aligner. Subsequent development steps in a proper solvent, followed by a baking
step to hard cure the material, completed the process. After deposition of a metal
111
7 Amplification Devices for Optical Interconnects on Optical Backplanes
Undercladding
Periscope
Waveguide with
under-cladding
towards WDM
and amplifier
waveguide
Flip Chip
Amplifier
on carrier
j
Cladding
Core
Cladding
*
. . . .
... .... ... ... ...
.............U............. ?
............................
6
K
- Pathway
of light
Backplane
Cylindrical lens
Figure 7.7: The FOWA device, overview.
layer on the prism slope, e.g. by sputter deposition, light from a flip chip mounted
VCSEL can be coupled directly into a waveguide, as visualized in Fig. 7.8.
On top of such a prism a VCSEL is mounted with the high-precision solder bump
mounting technology [238]. Solder bumps of 50 μm diameter are suitable. Furthermore, a cylindrical waveguide segment, with the function of a collimating lens, will be
used to enhance the coupling efficiency between the prism and the waveguide [105].
Thus, low loss coupling of the pump-light can be ensured.
7.3.3 Mounting of the SOA
Two challenges have to be met when applying the SOA structure for optical amplification: Firstly, high precision mount has to be ensured, enabling low loss coupling
of the low-power signal. Secondly, heat has to be removed from the device in an efficient way (see [233] for further information). Both problems can be solved through
high-precision solder bump mounting [238]. The solder bumps keep the SOA in place,
provide the electrical connections, and transfer the heat away from the device. Also
here 50 μm solder bumps are regarded as suitable.
112
7.3 The Flip Chip mountable Optical Waveguide Amplifier (FOWA) approach
Solder
bump
R
?
?
?
-
*
Deflection
prism
)
i
VCSEL
Uppercladding
Core
Undercladding
Substrate
Figure 7.8: Micro prism with a flip chip mounted VCSEL on top. The vertical beam of the
laser is reflected about 90◦ (vertical-to-horizontal) at the slope of the prism.
The prism was created using a UV-reaction molding technology [130].
7.3.4 The optical periscope and the low loss coupling
Optical periscope structures such as the approach by Robertsson et al. [131, 132]
and the Bragg grating approach from Kirk et al. [136], were presented in section
3.2. Robertsson et al.’s [131, 132] approach is basically an enhancement of Dannberg
et al.’s [36] approach of the prism forming by UV-reaction molding technology. In
this case, the molding and surface finish steps are repeated until the final height and
amount of reflective surfaces of the periscope is reached. Unfortunately, no reports
on working devices employing such periscopes (molding and Bragg grating approach)
have been found in literature. Thus, no comparison regarding optical loss can be
made between the two approaches.
To enhance the coupling from the board level waveguide to the prism, the idea of a
cylinder segment, forming a collimating lens, is applied here once more [105]. One
such lens is built up in front of the first prism, which ensures the low-loss coupling of
light to the amplifier device. Another is created on the optical backplane, next to the
re-entry point of the amplified signal, coming from the periscope. Thus, alignment
requirements for the whole amplifier device can be relaxed.
7.3.5 Mounting of the FOWA
The mounting of the FOWA onto the optical backplane has to be performed with
rather high precision too, although coupling structures, enabling rather large mismatch by still low coupling losses, have been included. Solder bumps 200 μm in
diameter were considered as a good starting point, but it still has to be tested if this
works in practice. The precision relies on the position of the solder pads too.
113
7 Amplification Devices for Optical Interconnects on Optical Backplanes
Cylindrical lens
Funnel
WDM
s
z
Incoming waveguide
on backplane
)
j
i
y
VCSEL with
coupling prism
underneath
-
Periscope
1
Amplifier
Substrate
Solder bump
1
Waveguide on backplane
with cylindrical lens
Figure 7.9: The FOWA device with amplifier waveguide, top view.
7.3.6 Process issues
In the previous sections the key structures of the FOWA device, based on amplifier
waveguide and SOA were briefly discussed. In Fig. 7.9 a top view of the device is presented, and in Fig. 7.10 a cross section, to visualize the measures of the key-structures
discussed above. However, further process issues are presented and discussed in detail
in paper III.
7.4 FOWA — Summary
The power budget available for a realistic routed optical link based on polymer optical
waveguides on backplanes was estimated in this section.
According to the output powers of commercially available laser diodes and the detection level of PDs, the power budget can be set to 14 dB for an optical link on a
backplane. It was concluded that the attenuation in the waveguide material has to be
lower than 0.08 dB/cm at 850 nm to enable an optical link over 1 m. A realistically
routed waveguide link makes signal amplification at certain places on the backplane
necessary, to boost the optical signal to be able to detect it in the PD.
114
7.4 FOWA — Summary
- 60 μm Deflection prism
Solder bump
... ... ... ... ... ... ...
... ...
... ... ... ... ... ..... ..... Carrier ..... .....
... ... ... ... ... ?
.
.
.
.
6
6
50 μm
?
?
6
220 μm
100 μm
?
VCSEL
6
20 μm
?
6
90 μm
?
6
90 μm
?
... ?... ... ... ... ... ... ... ... ... ... ]
.
.
. .. .. .. .. .. .. .. ..
Substrate .... .... .... .... .... .... .... ....
Periscope
Cylindrical
lens
Figure 7.10: Cross-section of the FOWA device visualizing the lateral measures of key
structures. Left-hand side: Solder bumps in size of 220 μm are assumed.
Center and right-hand side: An under-cladding with 20 μm thickness is assumed for the waveguides connected to the deflection prism and the periscope
structure.
Although polymer materials show much lower losses in the 850 nm region compared
to the traditional optical communications bands and work on optical backplanes is
focused on the 850 nm region, no optical amplifier devices have been reported in the
literature. Actually, research is concentrated on the traditional wavelength band for
optical communications, such as 1310 and 1550 nm. Thus, it is suggested to refocus
research in the direction of 850 nm.
Approaches for light amplification in planar waveguides, light coupling structures,
and amplifier waveguide design have been reviewed in this and in former sections.
Selected technologies from these assignments were brought together to shape the presented FOWA concept – a Flip chip mountable Optical Waveguide Amplifier, based on
a waveguide optical amplifier or SOA. The suggested hybrid approach has important
advantages:
• waveguide losses in backplanes and similar systems can be compensated, which
in turn increases the degree of freedom in waveguide routing (length, curvature,
splitting) in these systems,
• the signal amplifier can be produced separately from the board without restrictions related to board-production, allowing efficient use of amplifier area and
amplifier material as well as extreme process temperatures,
• the amplifier can be mounted with standard pick-and-place processes on boards,
115
7 Amplification Devices for Optical Interconnects on Optical Backplanes
with more or less ”arbitrary” waveguide technology, thanks to fault-tolerant
optics, and
• the design allows active cooling of the amplifier waveguide from the backside,
which may be necessary if high pumping powers are used and/or the pump is
to operate continuously.
In summary, the optical power budget of an optical link on a PCB was evaluated
and the need for an amplifier device for optical motherboards and backplanes was
justified. The proposed amplifier concept offers new viable possibilities for high performance and high density optical interconnects for backplanes and other interconnect
substrates.
116
8 Dielectrics for RF Interconnects
8.1 The SBU microwave demonstrator (DONDO
EMW)
The author was also involved in research related to the DONDODEM project during
2001. A high frequency test vehicle was created during this project, using FR4 subR
B59 (aka ORMOCORE) (developed at FhG-ISC (Würzburg,
strates, ORMOCER
Germany) and licensed to Micro Resist Technology (MRT GmbH, Berlin, Germany))
as dielectric material, and Copper. On a 127 mm FR4 substrate, 4 layers of dielectric
material (total thickness 80 μm, 20 μm each) were sequentially built up. Intermediate metal layers were patterned to form staircase and stacked vias, and stacked
capacitors. The last metal layer contained structures to evaluate the dielectric properties of the underlying four dielectric layers. These structures were strip lines, ring
resonators, and stubs.
The process was basically as follows: The circular FR4 had a thickness of about 1
mm and 18 μm Copper on one side. As the first process step Ti was sputter deposited
on top of the substrate to enhance the adhesion of the first B59 layer. Before each
spin coating of the dielectric material an Oxygen plasma treatment was applied to enhance the adhesion of the film to the substrate. The spin deposition was followed by a
pre-bake (T =353 K, 15 min) and exposure of the tacky film in the ANVIK projection
mask aligner [192]. After a post-exposure bake (T =353 K, 15 min), via holes (60 μm)
were developed using an apropriate developer (e.g. hexylacetat:isopropanol=1:1) for
15 s. This step was followed by a thermal curing of the dielectric layer at T =423 K
for 1 h. The thickness of each B59 layer was 20 μm.
Afterwards, a sputter deposition step of Cu/Ti took place, followed by patterning
of the metal layers to create via and capacitor structures and the final pattern on
top. The Ti layer had a thickness of 20 nm and was applied as an adhesion promoter.
Copper was deposited with a thickness of about 1 μm. The B59 and metal deposition
combination was repeated 4 times to reach a final thickness of the dielectric material
of about 80 μm. A picture of the test vehicle is shown in Fig. 8.1.
The evaluation of the test vehicle in the high frequency range gave a permittivity
of εr = 3.05 and tan δ 0.024 in the range of 10 – 40 GHz. Furthermore, stacked
vias showed only half the parasitic inductance compared to staircase vias in this case.
A precise description of the applied evaluation methods regarding the high frequency
characterization was given in [22] (=attached paper I).
117
8 Dielectrics for RF Interconnects
Figure 8.1: DONDODEM high frequency test vehicle.
8.2 Results & summary
R
During this project process development for a new dielectric material (ORMOCER
)
was performed, suitable for high-frequency applications. The test-vehicle contained
four alternating dielectric and metal layers, with a total thickness of 80 μm. The
metal layers were connected by via holes, 60 μm in size.
The process parameters applied allow MCM-L processes at moderate temperatures
(below 180 ◦ C), thus low-cost substrates, such as FR4 can be used, as shown in
R
our case. This makes the ORMOCER
material a feasible alternative to BCB or
polyimides, which need much higher process temperatures. Thus, a reduction of proR
material,
duction costs in high-frequency packaging is possible with the ORMOCER
through the low process temperatures, by maintaining good dielectric properties in
the high-frequency range.
118
9 Concluding Summary and Outlook
R
The focus of this thesis is on ORMOCER
materials. Materials characterization as
well as process development for large-are panel- and micro-processing applied to optical interconnects and high-frequency packaging was performed. Additionally, new
concepts for large-area panel manufacturing and amplifier devices for application on
optical backplanes were developed. The latter concept is based on an extensive critical literature survey and feasibility study on optical interconnects, optical backplanes,
coupling components and optical waveguide amplification technologies.
R
s belong to the material class of organic-inorganic hybrid polymers.
ORMOCER
Due to the possibility of “chemical tailoring” of the material properties in combination with their broad processing possibilities they are suitable for a wide range of
applications in microelectronics, photonics and microoptics. In applications simultaneously demanding good dielectric properties, good and tailorable optical properties
and good ability for SBU structures at low processing temperatures (<433 K) they
are considered to be unique. Photolithographic methods, (UV-)microreplication, laser
direct-writing, and two-photon polymerization can be employed to fabricate complex
2-3 D SBU structures. Thus micro-periscopes forming optical vias and microoptics
for VCSELs, such as deflection prisms and micro lenses are possible to create with
these materials.
Process development for SBU test vehicles (electrical and optical interconnects) has
been performed in connection with materials characterization. These include refractive index studies (system B59:V32), optical loss measurements (systems B59:V32
and B59:B66), and surface characterization through contact angle measurement and
surface energy estimation (systems B59:V32 and B59:B66).
R
B59 was characterized with respect to its dielectric properties.
The ORMOCER
A SBU demonstrator was built, using a FR4 substrate, to prove the compatibility to
PCB industry standard processes (curing temperature 423 K). A total layer thickness
R
was build-up, thus showing a cost effective alternative for
of 80 μm ORMOCER
high-frequency packaging. The permittivity and loss tangent were estimated to be
3.05 and 0.024, respectively, in the frequency range of 10 to 40 GHz (microstrip structures). This material is a good alternative to its competitors in the field (acrylates,
polyimide, BCB), due to its demonstrated good optical properties, dielectric properties, good adhesion to other materials, and moderate curing temperatures. Thus,
R
material due
costs of production can be reduced through using the ORMOCER
to the possibility of the application of low-costs substrates and moderate process
temperatures. At the same time good high-frequency properties are maintained.
119
9 Concluding Summary and Outlook
During process development for the large-area optical backplane demonstrator a
de-wetting phenomenon of the core-layer resin (pristine B59) on top of the cured under cladding (mixture of B59:V32 and B59:B66, respectively) was discovered. This
could be prevented through adding small amounts of the refractive index tuning agent
(0.1 wt% V32 and 5 wt% B66, respectively) to the core layer resin. It was shown in
several hot-plate experiments at 353 K that the solvent in a polymer solution could
hide de-wetting effects, which are exposed when the solvent finally evaporated from
the solution. Surface tension estimations on cured surfaces in various compositions of
core layer material and refractive index tuning agent did not show significant changes
in their values when the materials composition was below or above the de-wetting
break-even point. No significant major changes could be detected in complementary
optical loss measurements, comparing original and altered core materials.
The concept of altering one of the participants can open up process possibilities for
large-area processing by saving process time and process costs. Through this process
route no further surface treatments such as plasma activation or silanization are necessary, which can be unavailable for large-area processing. Finally, de-wetting tests of
liquid polymers under realistic process conditions are considered as a fast and effective working tool to expose de-wetting behavior, which might be a hidden in standard
surface tension tests.
The “flexible manufacture approach” demonstrated can be suitable for small companies, which provide optical backplane solutions for small-scale production lines and
a high flexibility in waveguide routing over the whole board. It is proposed that the
concept can be applied as a kind of portfolio approach, whereby a set of waveguide
masks can be used for multiple and scalable electro/optical backplane solutions. Furthermore, it is suitable for second level and third level packaging. Direct laser writing
provides high production flexibility as well but suffers from large process times if it
comes to complex patterns.
The feasibility of the “flexible manufacture approach” was shown through creation of
a large-area panel (609.6 mm x 609.6 mm) test vehicle by photolithographic step-out
processing. On a FR4 panel a waveguide structure was built-up (under cladding,
R
matecore and upper cladding) in a total thickness of >40 μm, using ORMOCER
rials (system B59:V32). Small photolithographic masks (101.6 mm x 101.6 mm) were
stepped-out over a large-area panel to create a continuous waveguide pattern (204.8
mm x 204.8 mm). A previously observed de-wetting effect of B59 resin (waveguide
core) on a cured B59:V32 surface (waveguide cladding) could be suppressed by adding
small amounts of V32 to the core layer material. Thus, no further surface activation
process-steps were necessary. Additionally, the final structure showed no delamination
or crack formation over the whole large-area panel. The optical loss of the waveguides was measured to be 0.6 dB/cm (B59:V32=95:5 wt%, 850 nm), which is 3 times
higher compared to previously reported values. This effect was attributed mainly to
the refractive index tuning agent applied for the cladding layer and the core layer,
which is non-optimized in the synthesis, and a non-optimized waveguide manufacturing processes. Further optical analysis in the de-wetting study showed that the
120
application of V32 is probably not responsible for the higher losses. It is more likely
that a manufacturing process optimization can decrease the optical loss even further,
which is left to another study.
In connection to the optical backplane project a study involving literature surveys was performed treating materials suitable for planar optical waveguide systems,
optical backplane approaches, optical power budget considerations, and optical amplifiers for planar waveguide systems. Additionally, the competition between optical
interconnects versus electrical interconnects was investigated, leading to the question:
When and under which circumstances are optical interconnects superior to electrical
interconnects? It was found that break-even points in terms of electrical power consumption and link costs exist at a given bandwidth and link length, beyond which
optics gets advantageous.
Furthermore, it was concluded in the study that optical amplification is necessary on
realistically routed optical backplanes at certain places to boost the optical signal.
Therefore the concept of a flip-chip mountable optical amplifier (FOWA) device has
been developed, based on planar optical waveguide amplifiers as well as on Semiconductor Optical Amplifiers. Optical losses can be compensated with the device on
specific places on the board, which in turn increases the degree of freedom in waveguide routing, i.e. in terms of length, curvature, and splitting. The hybrid approach
presented allows independent manufacturing from the rest of the board, thus process
parameters are not limited to PCB processes. Thanks to the fault-tolerant optics
the device can be positioned on the optical backplane using standard pick-and-place
technology.
Additionally, it was observed that most of the amplifier research is focused on the
wavelength of 1310 nm and 1550 nm, whereas optical backplane applications are targeting the 830 nm range. It is therefore proposed to re-direct research activities for
amplifier systems towards the wavelengths around 830 nm. The process development,
materials characterization, and reviews presented provide a starting point for further
research on high-performance electro/optical backplane interconnects from a processing point of view.
In this thesis new approaches for thin film processing applying SBU technology
were presented. In the near future it would be interesting to see which impact these
have on processes of the PCB industry. Furthermore, concepts suitable for applications in connection to optical backplanes have been demonstrated and developed.
In the literature it was emphasized that many different approaches for optical backplane solutions exist but hardly any real solution for a bottleneck application has
been presented. One goal for the future is therefore to take the step forward from
proof-of-concept projects towards bottleneck applications. Thus, target applications
have to be spotted as well as break-even points in terms of link-length and economics
have to be defined for such electro-optical hybrid backplanes. Hopefully, this thesis
can assist in the product’s design and further research and development on the topic.
121
9 Concluding Summary and Outlook
122
10 Bibliography
[1] R. Ulrich and L. Schaper. Putting passives in their place. IEEE Spectrum, 40
(7):26–30, July 2003.
[2] Rao R. Tummala. Fundamentals of Microsystems Packaging. McGraw-Hill
Professional Publishing, 2001.
[3] International Technology Roadmap for Semiconductors — 2003 Edition, Executive summary.
[4] Asus Corporation. Data Sheet for Asus Motherboards for various chip systems and CPUs, October 2005. URL http://se.asus.com/products/mb/
overview_p4.htm.
[5] AMD Press Release. Customers, Partners, And German Government Officials Join AMD At Topping-Off Ceremony For World’s First PurposeBuilt 64-Bit X86 300 Millimeter (mm) Manufacturing Facility, May
2004. URL http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,
,51_104_543~85239,00.html.
[6] INTEL Press Release. Intel To Build New 300 mm Wafer Factory In Arizona,
July 2005.
URL http://www.intel.com/pressroom/archive/releases/
20050725corp.htm.
[7] H.T. Holden. The developing technologies of integrated optical waveguides in
printed circuits. Circuit World, 29(4):42–50, 2003.
[8] Alan F. Benner, Michael Ignatowski, Jeffrey A. Kash, Daniel M. Kuchta, and
Mark B. Ritter. Exploitation of optical interconnects in future server architectures. IBM Journal of Research and Development, 49(4-5):755–775, 2005.
[9] Shichun Qu, Guoping Mao, Fuming Li, Robert Clough, Nelson O’Bryan, and
Robin Gorrell. A new organic composite dielectric material for high performance
IC packages. Proceedings - Electronic Components and Technology Conference,
2:1373–1377, 2005.
[10] Michael Scheffler, Gerhard Tröster, Joaquin Lopez Contreras, Jürgen Hartung,
and Michel Menard. Assessing the cost-effectiveness of integrated passives. Microelectronics International, 17(3):11–15, 2000.
123
10 Bibliography
[11] S.K. Bhattacharya and R.R. Tummala. Next generation integral passives: materials, processes, and integration of resistors and capacitors on pwb substrates.
Journal of Materials Science: Materials in Electronics, 11(3):253–268, April
2000.
[12] S. Ramesh, B.A. Shutzberg, C. Huang, Jie Gao, and E.P. Giannelis. Dielectric
nanocomposites for integral thin film capacitors: materials design, fabrication
and integration issues. IEEE Transactions on Advanced Packaging, 26(1):17–24,
February 2003.
[13] Yang Rao, S. Ogitani, P. Kohl, and C.P. Wong. Novel polymer-ceramic
nanocomposite based on high dielectric constant epoxy formula for embedded
capacitor application. Journal of Applied Polymer Science, 83(5):1084–1090,
January 2002.
[14] Jianwen Xu and C.P. Wong. Low-loss percolative dielectric composite. Applied
Physics Letters, 87(8):082907–1/2/3, 2005.
[15] C.P. Wong, Kyoung-Sik Moon, Suresh Pothukuchi, and Jianwen Xu. Dielectric
properties of polymer composite as integral capacitor. In Proceedings 52nd
Electronic Components and Technology Conference (ECTC’02), pages 1748–
1752. IEEE, June 2002.
[16] Karl-Heinz Haas, Sabine Amberg-Schwab, and Klaus Rose. Functionalized coating materials based on inorganic-organic polymers. Thin-Solid-Films, 351(1–2):
198–203, 1999.
[17] J. Manhart, K.-H. Kunzelmann, H.Y. Chen, and R. Hickel. Mechanical properties of light-cured packable composite resins. Dental Materials, 16:33–40, 2000.
[18] M.E. Robertsson, O.J. Hagel, G. Gustafsson, A. Dabek, M. Popall, L. Cergel,
P. Wennekers, P. Kiely, M. Lebby, and T. Lindahl. O/e-MCM packaging with
new, patternable dielectric and optical materials. In Proceedings 48th Electronic
Components and Technology Conference (ECTC’98), pages 1413–1421. IEEE,
May 1998.
[19] R. Houbertz, G. Domann, C. Cronauer, A. Schmitt, H. Martin, J. U. Park,
L. Fröhlich, R. Buestrich, M. Popall, U. Streppel, P. Dannberg, C. Wächter,
and A. Bräuer. Inorganic–organic hybrid materials for application in optical
devices. Thin Solid Films, 442(1–2):194–200, October 2003.
[20] Philippe Oberson, Bernard Gisin, Bruno Huttner, and Nicolas Gisin. Refracted
near-field measurements of refractive index and geometry of silica-on-silicon
integrated optical waveguides. Applied Optics, 37(31):7268–7272, November
1998.
124
10 Bibliography
[21] L. Fröhlich, R. Houbertz, S. Jacob, M. Popall, R. Müller-Fiedler, J. Graf,
M. Munk, and H. von Zychlinski. Inorganic-organic hybrid polymers as photopatternable dielectrics for multilayer microwave circuits. In 2002 Spring Meeting
Proceedings, volume 726, pages Q11.4.1–Q11.4.6. Materials Research Society,
2002.
[22] Christian Johansson, Steffen Uhlig, Ola Tageman, Arne Alping, Joacim
Haglund, Mats Robertsson, Michael Popall, and Lothar Fröhlich. Microwave
circuits in multilayer inorganic-organic polymer thin film technology on laminate substrates. IEEE Transactions on Advanced Packaging, 26(1):81–89, 2003.
[23] Steffen Uhlig, Lothar Fröhlich, Miaoxiang Chen, Norbert Arndt-Staufenbiel,
Günter Lang, Henning Schröder, Ruth Houbertz, Michael Popall, and Mats
Robertsson. Polymer optical interconnects - a scalable large area panel processing approach. Accepted for publication on IEEE Transactions on Advanced
Packaging, January 2005.
[24] Djafar K. Mynbaev and Lowell L. Scheiner. Fiber–Optics Communications Technology. Prentice Hall, 2001.
[25] Wolfgang Scheel. Von der elektrischen zur optischen Aufbau- und Verbindungstechnik. In BMBF-Projekte zu elektrisch-optischen Leiterplatten, pages 1–11,
2001.
[26] Christoph Berger, Marcel A. Kossel, Christian Menolfi, Thomas Morf, Thomas
Toifl, and Martin L. Schmatz. High-density optical interconnects within largescale systems. In Proceedings of SPIE — VCSELs and Optical Interconnects,
volume 4942, pages 222–235, April 2003.
[27] D.A.B. Miller. Rationale and challenges for optical interconnects to electronic
chips. In Proceedings of the IEEE, volume 88, pages 728–749, June 2000.
[28] Engelbert Strake. Electrical / Optical Circuit Board — EOCB. In BMBFProjekte zu elektrisch-optischen Leiterplatten, pages 13–26, 2001.
[29] Elmar Griese, Jörg-R. Kropp, Jörg Moisel, and Walter Süllau. Projekte zur optischen Aufbau- und Verbindungstechnik im Rahmen des BMBF-Förderschwerpunkt OptoSys. In BMBF-Projekte zu elektrisch-optischen Leiterplatten, pages
13–26, 2001.
[30] Denis Pogarieloff and Michel Andrau. Alcatel space R and D. Alcatel Telecommunications Review, (4):302–306, 2001.
[31] F. Mederer, R. Michalzik, J. Guttmann, H.-P. Huber, B. Lunitz, J. Moisel,
and D. Wiedenmann. 10 Gb/s data transmission with TO-packaged multimode
GaAs VCSELs over 1 m long polymer waveguides for optical backplane applications. Optics Communications, 206(4–6):309–312, June 2002.
125
10 Bibliography
[32] S. Agelis, S. Jacobsson, M. Jonsson, A. Alping, and P. Ligander. Modular
interconnection system for optical PCB and backplane communication. In Proceedings International Parallel and Distributed Processing Symposium., pages
245–250. IEEE, 2002.
[33] P. Lukowicz, J. Jahns, R. Barbieri, P. Benabes, T. Bierhoff, A. Gauthier, M. Jarczynski, G.A. Russell, J. Schrage, W. Sullau, J.F. Snowdon, M. Wirz, and
G. Troster. Optoelectronic interconnection technology in the holms system.
IEEE Journal of Selected Topics in Quantum Electronics, 9(2):624–635, 2003.
[34] T. Ao, K. Hartinger, P. Alluri, and C. Wilmsen. A biosequence processor based
on free-space optical interconnect. IEEE Journal of Selected Topics in Quantum
Electronics, 9(2):565–570, 2003.
[35] Xuliang Han and Ray T. Chen. Improvement of multiprocessing performance
by using optical centralized shared bus. In Randy A. Heyler and Ray T. Chen,
editors, Proceedings of SPIE — Photonics Packaging and Integration IV, volume
5358, pages 80–89, 2004.
[36] D.A.B. Miller. Physical reasons for optical interconnection. International Journal of Optoelectronics, 11(3):155–168, May/June 1997.
[37] J.H. Sinsky, M. Duelk, and A. Adamiecki. High-speed electrical backplane transmission using duobinary signaling. IEEE Transactions on Microwave Theory
and Techniques, 53(1):152–160, January 2005.
[38] A. Adamiecki, M. Duelk, and J.H. Sinsky. 25 Gbit/s electrical duobinary transmission over FR-4 backplanes. Electronics Letters, 41(14):826–827, 2005.
[39] Dawei Huang, T. Sze, A. Landin, R. Lytel, and H.L. Davidson. Optical interconnects: out of the box forever? IEEE Journal of Selected Topics in Quantum
Electronics, 9(2):614–623, 2003.
[40] A. Naeemi, Jianping Xu, A.V. Mule’, T.K. Gaylord, and J.D. Meindl. Optical
and electrical interconnect partition length based on chip-to-chip bandwidth
maximization. IEEE Photonics Technology Letters, 16(4):1221–1223, 2004.
[41] D. DeZutter and L. Knockaert. Skin Effect Modeling Based on a Differential
Surface Admittance Operator. IEEE Transactions on Microwave Theory and
Techniques, 53(8):2526–2538, August 2005.
[42] K.M. Coperich, A.E. Ruehli, and A. Cangellaris. Enhanced skin effect for
partial-element equivalent-circuit (PEEC) models. IEEE Transactions on Microwave Theory and Techniques, 48(9):1435–1442, September 2000.
[43] Gian L. Bona, Bert J. Offrein, Urs Bapst, Christoph Berger, Rene Beyeler,
Russell Budd, Roger Dangel, Laurent Dellmann, and Folkert Horst. Characterization of parallel optical-interconnect waveguides integrated on a printed
126
10 Bibliography
circuit board. In Proceedings of SPIE — Micro-Optics, VCSELs, and Photonic
Interconnects, volume 5453, pages 134–141, 2004.
[44] B. Lunitz, J. Guttmann, H.-P. Huber, J. Moisel, and M. Rode. Experimental
demonstration of 2.5 Gbit/s transmission with 1 m polymer optical backplane.
Electronics Letters, 37(17):1079, August 2001.
[45] C. Svensson. Electrical interconnects revitalized. IEEE Transactions on Very
Large Scale Integration (VLSI) Systems, 10(6):777–788, December 2002.
[46] Stephen C. Hall, Garret W. Hall, and James A. McCall. High-speed digital
system design. John Wiley & Sons, 2000.
[47] E. Griese. Modeling of highly multimode waveguides for time-domain simulation. IEEE Journal of Selected Topics in Quantum Electronics, 9(2):433–442,
2003.
[48] Hiroshi Nishihara, Masamitsu Haruna, and Toshiaki Suhara. Optical Integrated
Circuits. Optical and electro-optical engineering. McGraw Hill, 1987.
[49] Hoyeol Cho, P. Kapur, and K.C. Saraswat. Power comparison between highspeed electrical and optical interconnects for interchip communication. Journal
of Lightwave Technology, 22(9):2021–2033, 2004.
[50] Anand M. Pappu and Alyssa B. Apsel. Analysis of intrachip electrical and
optical fanout. Applied Optics, 44:6361–6372, October 2005.
[51] Davis H. Hartman, Gail R. Lalk, James W. Howse, and Krchnavek Robert R.
Radiant cured polymer optical waveguides on printed circuit boards for photonic
interconnection use. Applied Optics, 28(1):40–47, January 1989.
[52] Adam Singer. NEMI Cost Ananlysis: Optical Versus Copper Backplanes Part
I: Benchmarking Copper, 2004. URL http://thor.inemi.org/webdownload/
newsroom/Presentations/Opto_Copper_APEX04.pdf.
[53] E. Griese. A high-performance hybrid electrical-optical interconnection technology for high-speed electronic systems. IEEE Transactions in Advanced Packaging, 24(3):375–383, August 2001.
[54] Global Information Inc. [Report] High-Speed Optical Data Link Modules:
Market Review and Forecast 2002 (1 Gb/s and Above), January 2003. URL
http://www.gii.co.jp/english/su9189_high_speed_optical.html.
[55] Mindbranch. Backplane transceiver market opportunities, market forecasts, and
market strategies, 2004-2009, August 2004. URL http://www.mindbranch.
com/products/R49-197.html.
127
10 Bibliography
[56] Mindbranch. Internal optical interconnect components & subsystems global
market forecast, August 2001. URL http://www.mindbranch.com/products/
R56-121.html.
[57] Mindbranch. Multifiber transmit links & components forecast, June 2004. URL
http://www.mindbranch.com/products/R56-154_toc.html.
[58] ElectroniCast Corporation. Multifiber transmit links and components, July
2004. URL http://electronicast.ecnext.com/coms2/summary_0254-209_
ITM.
[59] BPA Consulting. Electrical & Optical Backplanes A Worldwide Market and
Technology Review 2005-2010, May 2005. URL http://www.bpaconsulting.
com.
[60] Jeff O. Montgomery. Optical backplanes show dynamic growth potential. Lightwave, 17(3):182, March 2000.
[61] Optical information and communication technology.
Electrical and
Optical Backplanes - An Industry in Transition — Conclusions of
the BPA Consulting market analysis report, December 2004.
URL
http://www.global-electronics.net/id/23264/CMEntries_ID/70994/
cubesig/59c4f7a5b0e1fe44b8d8953b6dadd66c.
[62] M. Forbes, J. Gourlay, and M. Desmulliez. Optically interconnected electronic
chips: a tutorial and review of the technology. Electronics & Communication
Engineering Journal, 13(5):221–232, 2001.
[63] Jörg Moisel, Joachim Guttmann, Hans-Peter Huber, Barbara Lunitz, Manfred
Rode, Reinhold Schoedlbauer, and Richard Bogenberger. Polymer waveguides
for 100cm (40”) optical backplanes. In Proceedings of SPIE — Design and
Fabrication of Planar Optical Waveguide Devices and Materials, volume 4805,
pages 98–105, 2002.
[64] Louay Eldada. Polymer integrated optics: Promise vs. practicality. In Proceedings of SPIE — Organic Photonic Materials and Devices IV, volume 4642,
pages 11–22, 2002.
[65] Jeong Hwan Song, Kyoung-Youm Kim, Jaegeol Cho, Dongkyoon Han, Joohoon
Lee, Yu Sheop Lee, Suntae Jung, Yunkyung Oh, Dong-Hoon Jang, and
Kyung Shik Lee. Thin film filter-embedded triplexing-filters based on directional couplers for ftth networks. IEEE Photonics Technology Letters, 17(8):
1668 – 1670, 2005.
[66] M. Itoh, I. Yoneda, J. Sasaki, H. Honmou, K. Fukushima, and T. Nagahori.
Self-aligned packaging of multichannel photodiode array module using ausn solder bump flip-chip bonding. In EuPac ’96. 2nd European Conference on Elec-
128
10 Bibliography
tronic Packaging Technology and 8th International Conference on Interconnection Technology in Electronics, pages 75–77, Essen, Germany, 1996. Deutscher
Verlag fur Schweisstechnik.
[67] Hong Ma, Alex K.-Y. Jen, and Larry R. Dalton. Polymer-based optical waveguides: Materials, processing, and devices. Advanced Materials, 14(19):1339–
1365, October 2002.
[68] M.R. Poulsen, P.I. Borel, J. Fage-Pedersen, J. Hubner, M. Kristensen, J.H.
Povlsen, K. Rottwitt, M. Svalgaard, and W. Svendsen. Advances in silica-based
integrated optics. Optical Engineering, 42(10):2821–2834, October 2003.
[69] Fengnian Xia, Vinod M. Menon, and Stephen R. Forrest. Photonic integration
using asymmetric twin-waveguide (atg) technology: Part i - concepts and theory.
IEEE Journal on Selected Topics in Quantum Electronics, 11(1):17 – 29, 2005.
[70] One Art Design Anvik Corporation. Anvik corporation — home, May 2003.
URL http://www.anvik.com/.
[71] S. Esener and P. Marchand. Present status and future needs of free-space optical
interconnects. Materials Science in Semiconductor Processing, 3(5/6):433–435,
2000.
[72] Andrew G. Kirk, David V. Plant, Ted H. Szymanski, Zvonko G. Vranesic,
Frank A.P. Tooley, David R. Rolston, Michael H. Ayliffe, Frederic K. Lacroix,
Brian Robertson, Eric Bernier, and Daniel F. Brosseau. Design and implementation of a modulator-based free-space optical backplane for multiprocessor
applications. Applied Optics, 42(14):2465–2481, 2003.
[73] Håkan Forsberg. HiSPOT: The offical homepage of HiSPOT (High Speed Optoelectronics for Optical Interconnects) within CHACH (Chalmers Centre for
High Speed Technology), September 2003. URL http://www.ce.chalmers.
se/~rapid/hispot/.
[74] M.I. Cohen and C. Jagadish. It’s all about speed. IEEE Circuits and Devices
Magazine, 20(1):38–43, 2004.
[75] Manfred Jarczynski and Jurgen Jahns. Planar integrated free-space optics for
optical interconnects and fan-out/in operations. In Proceedings of SPIE —
Photonic Devices and Algorithms for Computing VI, volume 5556, pages 15–26,
Denver, CO, United States, 2004.
[76] Xuliang Han, Gicherl Kim, G. Jack Lipovski, and Ray T. Chen. An optical
centralized shared-bus architecture demonstrator for microprocessor-to-memory
interconnects. IEEE Journal on Selected Topics in Quantum Electronics, 9(2):
512–517, 2003.
129
10 Bibliography
[77] Chunhe Zhao, Tchang-Hun Oh, and R.T. Chen. General purpose bidirectional
optical backplane: high-performance bus for multiprocessor systems. In Proceedings of the Second International Conference on Massively Parallel Processing Using Optical Interconnections, pages 188–195, October 1995.
[78] Gicherl Kim and Ray T. Chen. Characterization of a bi-directional optical
backplane and performance enhancement with multi-bus lines. In Proceedings
of SPIE — Optoelectronic Interconnects VI, volume 3632, pages 85–95, April
1999.
[79] Gicherl Kim and Ray T. Chen. Three-dimensionally interconnected multi-busline bidirectional optical backplane. IEEE Photonics Technology Letters, 11(7):
880–882, July 1999.
[80] Gicherl Kim, Xuliang Han, and Ray T. Chen. An 8-Gb/s optical backplane
bus based on microchannel interconnects: design, fabrication, and performance
measurements. Journal of Lightwave Technology, 18(11):1477–1486, November
2000.
[81] Xuliang Han, Gicherl Kim, and Ray T. Chen. Demonstration of the centralized optical backplane architecture in a three-board microprocessor-to-memory
interconnect system. Optics and Laser Technology, 35(2):127–131, March 2003.
[82] H. Schröder, N. Arndt-Staufenbiel, M. Cygon, and W. Scheel. Planar glass
waveguides for High performance electrical-optical-circuit-boards (EOCB) –
The glass-layer-concept. In Proceedings 53rd Electronic Components and Technology Conference (ECTC’03), pages 1053–1059. IEEE, May 2003.
[83] Y. Ishii, S. Koike, Y. Arai, and Y. Ando. SMT-compatible optical-I/O chip packaging for chip-level optical interconnects. In Proceedings 51st Electronic Components and Technology Conference (ECTC’01), pages 870–875. IEEE, May
2001.
[84] Terahertz Photonics Corp. Truemode backplaneTM enabeling optical backplanes, April 2003. URL http://www.terahertz-photonics.com/materials/
pdf/Truemode\%20Backplane\%20Paper.pdf.
[85] Keun Byoung Yoon, In-Kui Cho, Seung Ho Ahn, Myung Yong Jeong, Deug Ju
Lee, Young Un Heo, Byung Sup Rho, Hyo-Hoon Park, and Byoung-Ho Rhee.
Optical backplane system using waveguide-embedded PCBs and optical slots.
Journal of Lightwave Technology, 22(9):2119–2127, September 2004.
[86] In-Kui Cho, Keun Byoung Yoon, Seung Ho Ahn, Hee-Kyung Sung, Sang Won
Ha, Young Un Heo, and Hyo-Hoon Park. Experimental demonstration of 10
Gbit/s transmission with an optical backplane system using optical slots. Optics
Letters, 30(13):1635–1637, 2005.
130
10 Bibliography
[87] In Kui Cho, Keun Byoung Yoon, Seoung Ho Aha, Jin Tae Kim, Woo Jin Lee,
Kyoung Up Shin, Young Un Heo, and Hyo Hoon Park. Board-to-board optical
interconnection using novel optical plug and slot. In Proceedings of SPIE The International Society for Optical Engineering, volume 5524, pages 359–367,
2004.
[88] S. Kopetz, E. Rabe, W.J. Kang, and A. Neyer. Polysiloxane optical waveguide
layer integrated in printed circuit board. Electronics Letters, 40(11):668–669,
2004.
[89] H. Schröder, J. Bauer, F. Ebling, and W. Scheel. Polymer optical interconnects
for PCB. In First International IEEE Conference on Polymers and Adhesives
in Microelectronics and Photonics, pages 337–343, 2001.
[90] D. Krabe and W. Scheel. Optical interconnects by hot embossing for module
and PCB technology – The EOCB approach –. In Proceedings 49th Electronic
Components and Technology Conference (ECTC’99), pages 1164–1166. IEEE,
June 1999.
[91] Detlef Krabe, Frank Ebling, Norbert Arndt-Staufenbiel, Günter Lang, and Wolfgang Scheel. New technology for electrical/optical systems on module and board
level: the EOCB approach. In Proceedings 50th Electronic Components and
Technology Conference (ECTC’00), pages 970–974. IEEE, May 2000.
[92] R. Yoshimura, M. Hikita, M. Usui, S. Tomaru, and S. Imamura. Polymeric
optical waveguide films with 45◦ mirrors formed with a 90◦ V-shaped diamond
blade. Electronics Letters, 33(15):1311–1312, June 1997.
[93] Y. Ishii, N. Tanaka, T. Sakamoto, and H. Takahara. Fully SMT-compatible
optical-I/O package with microlens array interface. Journal of Lightwave Technology, 21(1):275–280, 2003.
[94] Yuzo Ishii, Tsuyoshi Hayashi, and Hideyuki Takahara. Demonstration of onPCB optical interconnection using surface-mount package and polymer waveguide. In Proceedings 53rd Electronic Components and Technology Conference
(ECTC’03), pages 1147–1152, New Orleans, LA, United States, May 2003.
IEEE.
[95] C. Berger, R. Beyeler, G.-L. Bona, R. Dangel, L. Dellmann, P. Dill, F. Horst,
M.A. Kossel, C. Menolli, T. Morf, B. OfIrein, M.L. Schmatz, T. Toifl, and
J. Weiss. Optical links for printed circuit boards. In The 16th Annual Meeting
of the IEEE Lasers and Electro-Optics Society (LEOS 2003), volume 1, pages
61–62, October 2003.
[96] R. Dangel, U. Bapst, C. Berger, R. Beyeler, L. Dellmann, F. Horst, B. Offrein,
and G.-L. Bona. Development of a low-cost low-loss polymer waveguide technology for parallel optical interconnect applications. In Digest of the LEOS
131
10 Bibliography
Summer Topical Meetings Biophotonics/Optical Interconnects and VLSI Photonics/WBM Microcavities, pages 78–79, 2004.
[97] C. Berger, U. Bapst, G.-L. Bona, R. Dangel, L. Dellmann, P. Dill, M.A. Kossel,
T. Morf, B. Offrein, and M.L. Schmatz. Design and implementation of an optical interconnect demonstrator with board-integrated waveguides and microlens
coupling. In 2004 Digest of the LEOS Summer Topical Meetings: Biophotonics/Optical Interconnects and VLSI Photonics/WBM Microcavities, pages
19–20, 2004.
[98] Terahertz Photonics Corp.
Truemode backplaneTM enabeling optical
backplanes to extend the bandwidth, data sheet, August 2003.
URL
http://www.terahertz-photonics.com/materials/pdf/THz\%20Truemode\
%20Backplane\%20polymer\%20datasheet.pdf.
[99] UK Business Park. Terahertz photonics, June 2004.
ukbusinesspark.co.uk/tes99434.htm.
URL http://www.
[100] Exxelis Limited. Homepage, November 2005. URL http://www.exxelis.com/.
[101] In-Kui Cho, Keun Byoung Yoon, Seung Ho Ahn, Myung Yong Jeong, HeeKyung Sung, Byung Ho Lee, Young Un Heo, and Hyo-Hoon Park. Boardto-board optical interconnection system using optical slots. IEEE Photonics
Technology Letters, 16(7):1754–1756, July 2004.
[102] M. Franke and F.-P. Schiefelbein. Optical interconnects on printed circuit board
level - results based on the German funded project OptiCon. In Proceedings
54th Electronic Components and Technology Conference (ECTC’04), volume 2,
pages 1542–1546. IEEE, 2004.
[103] A. Neyer, S. Kopetz, E. Rabe, W.J. Kang, and S. Tombrink. Electrical-Optical
Circuit Board Using Polysiloxane Optical Waveguide Layer. In Proceedings 55th
Electronic Components and Technology Conference (ECTC’05), pages 246–250,
Orlando, Florida, USA, June 2005.
[104] Chulchae Choi, Yujie Liu, Lei Lin, Li Wang, Jinho Choi, David Haas, Jerry
Magera, and Ray T. Chen. Flexible optical waveguide film with 45-degree
micromirror couplers for hybrid E/O integration or parallel optical interconnection. In Proceedings of SPIE — Photonics Packaging and Integration IV,
volume 5358, pages 122–126, San Jose, CA, USA, 2004.
[105] I. Frese, Stefan Kufner, Maria Kufner, G. Hochmuth, Hans-Dieter Bauer, and
Wolfgang Ehrfeld. Combination of guided wave and free-space micro-optics for
a new optical backplane concept. In Proceedings of SPIE — ROMOPTO 2000:
Sixth Conference on Optics, volume 4430, pages 715–721, June 2001.
132
10 Bibliography
[106] J. Moisel, H.-P. Huber, J. Guttmann, O. Krumpholz, B. Lunitz, M. Rode, and
R. Schoedlbauer. Optical backplane. In 27th European Conference on Optical
Communication, volume 3, pages 254–255, 2001.
[107] G. Van Steenberge, P. Geerinck, S. Van Put, J. Van Koetsem, H. Ottevaere,
D. Morlion, H. Thienpont, and P. Van Daele. MT-compatible laser-ablated
interconnections for optical printed circuit boards. Journal of Lightwave Technology, 22(9):2083–2090, September 2004.
[108] Marika Immonen, Mikko Karppinen, and Jorma Kivilathi. Fabrication and
characterization of optical polymer waveguides embedded on printed wiring
boards. In 3rd International IEEE Conference on Polymers and Adhesives in
Microelectronics and Photonics — Polytronic, pages 91–95, 2003.
[109] Fengguang Luo, Mingcui Cao, Xinjun Zhou, Jun Xu, Zhixiang Luo, Jin
Yuan, and Kai Fu. Optical interconnection technology on electro/optical PCB
(EOPCB). In Proceedings of SPIE — Optoelectronic Devices and Integration,
volume 5644, pages 821–828, Beijing, China, 2005.
[110] Gee-Kung Chang, D. Guidotti, Fuhan Liu, Yin-Jung Chang, Zhaoran Huang,
V. Sundaram, D. Balaraman, S. Hegde, and R.R. Tummala. Chip-to-chip optoelectronics SOP on organic boards or packages. IEEE Transactions on Advanced
Packaging Part B, 27(2):386–397, 2004.
[111] Jörg Moisel, Richard Bogenberger, Joachim Guttmann, Hans-Peter Huber, Oskar Krumpholz, Klaus-Peter Kuhn, and Manfred Rode. Optical backplanes with
integrated polymer waveguides. Optical Engineering, 39(03):673–679, March
2000.
[112] Ping Gui, F.E. Kiamilev, X.Q. Wang, X.L. Wang, M.J. McFadden, M.W. Haney,
and C. Kuznia. A 2-Gb/s 0.5-μm CMOS parallel optical transceiver with fast
power-on capability. Journal of Lightwave Technology, 22(9):2135–2148, 2004.
[113] M. Immonen, Jianbo Wu, and J. Kivilahti. Fabrication of polymer optical
waveguides with integrated micromirrors for out-of-plane surface normal optical
interconnects. In 4th IEEE International Conference on Polymers and Adhesives in Microelectronics and Photonics — Polytronic, pages 206–210, 2004.
[114] Mikko Karppinen, Jukka-Tapani Makinen, Kari Kataja, Antti Tanskanen,
Teemu Alajoki, Pentti Karioja, Marika Immonen, and Jorma Kivilahti. Embedded optical interconnect on printed wiring board. In Proccedings of SPIE —
Micro-Optics, VCSELs, and Photonic Interconnects, volume 5453, pages 150–
164, 2004.
[115] Joni Hietala, Esa Muukkonen, Tuomo von Lerber, and Marika Immonen. Volume Production of On-Board Optical Waveguides. In Proceedings 55th Electronic Components and Technology Conference (ECTC’05), pages 1735–1738,
Orlando, Florida, USA, June 2005.
133
10 Bibliography
[116] Marika Immonen, Jianbo Wu, and Jorma Kivilahti. Influence of Environmental
Stresses on Board-Level Integrated Polymer Optics. In Proceedings 55th Electronic Components and Technology Conference (ECTC’05), pages 1653–1658,
Orlando, Florida, USA, June 2005.
[117] Tetsuzo Yoshimura, James Roman, Yasuhito Takahashi, Michael Lee, Bill Chou,
Solomon I. Beilin, Wen-Chou V. Wang, and Masaaki Inao. Optoelectronic scalable substrates based on film/Z-connection and its application to film optical
link module (FOLM). In Michael R. Feldman, Richard L. Li, W. Brian Matkin,
and Suning Tang, editors, Proc. SPIE: Optoelectronic Interconnects VII; Photonics Packaging and Integration II, volume 3952, pages 202–213, April 2000.
[118] Tze Wei Yeow, K.L.E. Law, and A.A. Goldenberg. SOI-based 2-D MEMS Lswitching matrix for optical networking. IEEE Journal of Selected Topics in
Quantum Electronics, 9(2):603–613, 2003.
[119] Albert Polman and Frank C. J. M. van Veggel. Broadband sensitizers for
erbium-doped planar optical amplifiers: review. Journal Optical Society of
America B, 21(5):871–892, 2004.
[120] C. Wong, editor. Polymers for Electronic and Photonic Applications. Academic
Press Inc., Harcourt Brace Jovanovich, 1993.
[121] Chia-Cheng Chang, Chih-Ping Chen, Che-Chung Chou, Wen-Jang Kuo, and
Ru-Jong Jeng. Polymers for electro-optical modulation. Journal of Macromolecular Science - Polymer Reviews, 45(2):125–170, 2005.
[122] Young-Ouk Noh, Jong-Min Kim, Mi-Sung Yang, Hee-Jin Choi, Hyung-Jong
Lee, Yong-Hyub Won, and Seon-Gyu Han. Thermooptic 2*2 asymmetric digital
optical switches with zero-voltage operation state. IEEE Photonics Technology
Letters, 16(2):446–448, February 2004.
[123] A. Leinse, M.B.J. Diemeer, A. Rousseau, and A. Driessen. A novel high-speed
polymeric EO modulator based on a combination of a microring resonator and
an MZI. IEEE Photonics Technology Letters, 17(10):2074–2076, 2005.
[124] Joon Tae Ahn, S. Park, Jung Yun Do, Jong-Moo Lee, Myung-Hyun Lee, and
Kyong Hon Kim. Polymer wavelength channel selector composed of electrooptic polymer switch array and two polymer arrayed waveguide gratings. IEEE
Photonics Technology Letters, 16(6):1567–1569, June 2004.
[125] M. Kagami, A. Kawasaki, and H. Ito. A polymer optical waveguide with outof-plane branching mirrors for surface-normal optical interconnections. Journal
of Lightwave Technology, 19(12):1949–1955, December 2001.
[126] S. Koike, F. Simokawa, T. Matsuura, and H. Takahara. New optical and electrical hybrid packaging techniques using optical waveguides for optoelectronic
134
10 Bibliography
multichip modules. IEEE Transactions on Components, Packaging, and Manufacturing Technologies, 19(1):124–130, February 1996.
[127] Yujie Liu, Lei Lin, Chulchae Choi, Bipin Bihari, and R.T. Chen. Optoelectronic integration of polymer waveguide array and metal-semiconductor-metal
photodetector through micromirror couplers. Journal of Lightwave Technology,
13(4):355–357, April 2001.
[128] R.T. Chen, Lei Lin, Chulchae Choi, Y.J. Liu, B. Bihari, L. Wu, S. Tang,
R. Wickman, B. Picor, M.K. Hibb-Brenner, J. Bristow, and Y.S. Liu. Fully
embedded board-level guided-wave optoelectronic interconnects. Proceedings of
the IEEE, 88(6):780–793, June 2000.
[129] R. Wiesmann, S. Kalveram, and A. Neyer. Monomode polymer waveguides with
integrated mirrors. In 22nd European Conference on Optical Communication,
volume 2, pages 265–268, 1996.
[130] P. Dannberg, L. Erdmann, A. Krehl, C. Wächter, and A. Bräuer. Integration
of optical interconnects and optoelectronic elements on wafer-scale. Materials
Science in Semiconductor Processing, 3(5–6):437–441, 2000.
[131] Mats Robertsson. Optical interconnects - packaging and performance issues.
Symposium: Short Distance Optical Interconnects — From Backplanes to Intrachip Communication, Chalmers Center for High Speed Technology (CHACH),
March 2003.
[132] Mats Robertsson.
Mirrors for Polymer Waveguides.
WO2004015463, 2004-2-19.
WO-Patent,
[133] M. Hikita, S. Tomaru, K. Enbutsu, N. Ooba, R. Yoshida, M. Usai, T. Yoshida,
and S. Imamura. Polymeric optical waveguide films for short-distance optical
interconnects. IEEE Journal of Selected Topics on Quantum Electronics, 5(5):
1237–1242, September 1999.
[134] Hideyuki Takahara and Robert A. Norwood. Optoelectronic packaging technologies for chip-level optical interconnections. In Proceedings of SPIE — Design
and Fabrication of Planar Optical Waveguide Devices and Materials, volume
4805, pages 125–133, 2002.
[135] J.P.D. Cook, G.O. Este, F.R. Shepherd, W.D. Westwood, J. Arrington,
W. Moyer, J. Nurse, and S. Powell. Stable, low-loss optical waveguides and
micromirrors fabricated in acrylate polymers. Applied Optics, 37(7):1220–1226,
March 1998.
[136] Andrew G. Kirk, Michael Menard, Po Dong, and Eric Bisaillon. Integration
strategies for micro-optical systems. In Proceedings of SPIE — Micro-Optics,
VCSELs, and Photonic Interconnects, volume 5453, pages 83–88, 2004.
135
10 Bibliography
[137] Sean M. Garner, Sang-Shin Lee, Vadim Chuyanov, Antao Chen, Araz Yacoubian, William H. Steier, and Larry R. Dalton. Three-dimensional integrated
optics using polymers. IEEE Journal of Quantum Electronics, 35(8):1146–1155,
1999.
[138] Hsiharng Yang, Ching-Kong Chao, Che-Ping Lin, and Sheng-Chih Shen. Microball lens array modeling and fabrication using thermal reflow in two polymer layers. Journal of Micromechanics and Microengineering, 14(2):277 – 82, 2004/02/.
[139] Optical CrossLinks Inc. Polymer technology overview, January 2001. URL
http://www.opticalcrosslink.com/pdf/WebPolymerTECHNOLOGYPlatform.
pdf.
[140] Claire Pitois. Fluorinated Polymer Optical Waveguides. PhD thesis, Royal Institute of Technology, Department of Polymer Technology, Stockholm, Sweden,
2001.
[141] Ming Zhou. Low-loss polymeric materials for passive waveguide components
in fiber optical telecommunication. Optical Engineering, 41(7):1631–1643, July
2002.
[142] R. Buestrich, F. Kahlenberg, M. Popall, P. Dannberg, R. Müller-Fiedler, and
O. Rösch. ORMOCERs for optical interconnection technology. Journal of SolGel Science and Technology, 20:181, 2001.
[143] John M. Ballato, Jr. Dennis W. Smith, Stephen H. Foulger, Earl Wagener, and
Robert A. Norwood. Theoretical performance of polymer optical fibers, planar
waveguides, and amplifiers. In Proceedings of SPIE — Design and Fabrication
of Planar Optical Waveguide Devices and Materials, volume 4805, pages 1–8,
2002.
[144] Steffen Uhlig and Mats Robertsson. Limitations to and solutions for optical
loss in optical backplanes. Accepted for publication on Journal of Lightwave
Technology, January 2006.
[145] Ulf Gedde. Polymer Physics. Kluwer Academic Publishers, 1 edition, 1999.
[146] F. Tooley, N. Suyal, F. Bresson, A. Fritze, J. Gourlay, A. Walker, and M. Emmery. Optically written polymers used as optical interconnects and for hybridisation. Optical Materials, 17(1–2):235–241, June 2001.
[147] Eunkyoung Kim, Yun Cho Song, Dong-Min Yeu, and Sang-Yung Shin. Low
optical loss perfluorinated methacrylates for a single-mode polymer waveguide.
Chemistry of Materials, 17(5):962 – 966, 2005.
[148] Song Yun Cho and Eunkyoung Kim. Optical loss of photochromic polymer
films. In Proceedings of SPIE — Organic Photonic Materials and Devices V,
volume 4991, pages 282–292, San Jose, CA, United States, 2003.
136
10 Bibliography
[149] Matthew L. Moynihan, Craig Allen, Tuan Ho, Luke Little, Nathan Pawlowski,
Nick Pugliano, James G. Shelnut, Bruno Sicard, Hai Bin Zheng, and Garo
Khanarian. Hybrid inorganic-organic aqueous base compatible waveguide materials for optical interconnect applications. In Proceedings of SPIE — Linear
and Nonlinear Optics of Organic Materials III, volume 5212, pages 50–60, San
Diego, CA, USA, 2003.
[150] Ann W. Norris, Jon V. DeGroot, Takuya Ogawa, Toshinori Watanabe, Tony C.
Kowalczyk, April Baugher, and Robert Blum. High reliability of silicone materials for use as polymer waveguides. In Proceedings of SPIE — Linear and Nonlinear Optics of Organic Materials III, volume 5212, pages 76–82, San Diego,
CA, USA, 2003.
[151] T. Nakai, Y. Ueno, K. Kaneko, S. Tanahashi, and S. Takeda. A siloxane polymer lightwave circuit on ceramic substrate applicable to ultrafast optoelectronic
multi-chip-modules. Optical and Quantum Electronics, 33(7–10):1113–1124,
July 2001.
[152] S. Ponoth, Peter D. Persans, Ram Ghoshal, N. Agarwal, Joel L. Plawsky,
A. Filin, and Q. Z. Fang. Siloxane-based polymer epoxies for optical waveguides.
In Proceedings of SPIE — Applications of Photonic Technology 6, volume 5260,
pages 331–335, 2003.
[153] A. Mochizuki, K. Mune, R. Naitou, T. Fukuoka, and K. Tagawa. New photosensitive polyimide materials and their application to low-loss optical waveguides.
Journal of Photopolymer Science and Technology, 16:243–246, 2003.
[154] Jae-Wook Kang, Jang-Joo Kim, Jinkyu Kim, Xiangdan Li, and Myong-Hoon
Lee. Low-loss and thermally stable te-mode selective polymer waveguide using
photosensitive fluorinated polyimide. IEEE Photon. Technol. Lett., 14(9):1297–
1299, September 2002.
[155] Sung Ho Baek, Jae-Wook Kang, Xiang Dan Li, Myung-Hoon Lee, and Jang Joo
Kim. Zero birefringent polyimide for polymer optical waveguide. Proceedings
of SPIE - The International Society for Optical Engineering, 4991:406 – 413,
2003.
[156] C. Cornic, B. Lucas, A. Moliton, B. Colombeau, and R. Mercier. Elaboration and characterization of a 6FDA/MPDA polyimide-based optical waveguide.
Synthetic Metals, 127(1–3):299–302, March 2002.
[157] Cheng-Tyng Yen, Wen-Chang Chen, Der-Jang Liaw, and Hsin-Yi Lu. Synthesis
and properties of new polyimid-silica hybrid films through both intrachain and
interchain bonding. Polymer, 44(23):7079–7087, 2003.
[158] Shengrong Chen, Jaime Cardenes, Gregory P. Nordin, Chris M. Topping,
Jack J. Jin, Jennifer Thomson, John M. Ballato, Stephen H. Foulger, Jr. Dennis
137
10 Bibliography
W. Smith, and Robert A. Norwood. Direct micro-transfer molding of perfluorocyclobutyl (PFCB) polymer waveguides. In Proceedings of SPIE — Design and
Fabrication of Planar Optical Waveguide Devices and Materials, volume 4805,
pages 55–60, 2002.
[159] Sharon Wong, Hong Ma, Alex K.-Y. Jen, Rick Barto, and Curtis W. Frank.
Highly fluorinated trifluorovinyl aryl ether monomers and perfluorocyclobutane
aromatic ether polymers for optical waveguide applications. Macromolecules,
36(21):8001 – 8007, 2003.
[160] Hong Ma, Sharon Wong, Seok Ho Kang, Jingdong Luo, Marnie Haller, Alex K.Y. Jen, Rick Barto, and Curtis W. Frank. Novel perfluorocyclobutane (PFCB)containing polymers and dendrimers for photonic devices. In Proceedings of
SPIE — Design and Fabrication of Planar Optical Waveguide Devices and Materials, volume 4805, pages 19–26, Seattle, WA, United States, 2002.
[161] Hong Ma, Jingdong Luo, Seok Ho Kang, Sharon Wong, Jae Wook Kang, Alex
K.-Y. Jen, Rick Barto, and Curtis W. Frank. Highly fluorinated and crosslinkable dendritic polymer for photonic applications. Macromolecular Rapid Communications, 25(19):1667–1673, 2004.
[162] Sharon Wong, Hong Ma, Alex K.-Y. Jen, Rick Barto, and Curtis W. Frank.
Perfluorocyclobutane-based polyester(arylene ether)s for applications in integrated optics. Macromolecules, 37(15):5578–5585, 2004.
[163] Jae-Wook Kang, Jae-Pil Kim, Won-Young Lee, Joon-Sung Kim, Jae-Suk Lee,
and Jang-Joo Kim. Low-loss fluorinated poly(arylene ether sulfide) waveguides
with high thermal stability. Journal of Lightwave Technology, 19(6):872–875,
June 2001.
[164] Jae-Pil Kim, Jae-Wook Kang, Jang-Joo Kim, and Jae-Suk Lee. Fluorinated
poly(arylene ether sulfone)s for polymeric optical waveguide devices. Polymer,
44(15):4189–4195, July 2003.
[165] Jae-Wook Kang, Jae-Pil Kim, Jae-Suk Lee, and Jang-Joo Kim. Low-loss polymer optical waveguides using fluorinated poly(arylene ether sulfides and sulfones). Proceedings of SPIE - The International Society for Optical Engineering,
4805:9 – 18, 2002.
[166] Myung-Hyun Lee, Jung Jin Ju, Suntak Park, Jung Yun Do, and Seung Koo
Park. Polymer-based devices for optical communications. ETRI Journal, 24(4):
259–269, August 2002.
[167] Yinghua Qi, Jianfu Ding, Michael Day, Jia Jiang, and Claire L. Callender.
Cross-linkable highly fluorinated poly(arylene ether ketones/sulfones) for optical
waveguiding applications. Chemistry of Materials, 17(3):676–682, 2005.
138
10 Bibliography
[168] Jia Jiang, Claire L. Callender, Chantal Blanchetiere, Sarkis Jacob, Julian P.
Noad, Jianfu Ding, Yinghua Qi, and Michael Day. Optimizing fluorinated
poly(arylene ether)s for optical waveguide applications. Optical Materials, 28
(3), February 2006.
[169] Jia Jiang, Patrick Dumais, Claire L. Callender, Chantai Blanchetiere, Julian P.
Noad, Jianfu Ding, Yinghua Qi, and Michael Day. Polymer photonic devices
using fluorinated poly (arylene ether ketone). In Proceedings of SPIE — Photonics North 2004: Optical Components and Devices, volume 5577, pages 318–325,
Ottawa, Ont., Canada, 2004.
[170] C. Dreyer, M. Bauer, J. Bauer, N. Keil, Huihai Yao, and C. Zawadzki. Polycyanurate ester resins with low loss for use in integrated optics. In First International IEEE Conference on Polymers and Adhesives in Microelectronics and
Photonics, pages 276–283, October 2001.
[171] Seok Ho Kang, Jingdong Luo, Hong Ma, Richard R. Barto, Curtis W. Frank,
Larry R. Dalton, and Alex K.-Y. Jen. A hyperbranched aromatic fluoropolyester
for photonic applications. Macromolecules, 36(12):4355 – 4359, 2003.
[172] N. Srisanit, Xingzhong Yan, Xianjun Ke, and M.R. Jianwen Yangand Wang.
Laser writing of low-loss polymer channel waveguides. In Summaries of Papers
Presented at the Lasers and Electro-Optics. CLEO ’02, volume 1, page 240, May
2002.
[173] Yu Zhao, Fei Wang, Aize Li, Baijun Liu, Zhongwen Wu, Daming Zhang, Shiyong Liu, and Maobin Yi. Cross-linkable fluorinated poly(ether ether ketone)
polymers for optical waveguide devices. Materials Letters, 58(19):2365–2368,
jul 2004.
[174] Claire Pitois, Robert Vestberg, Marlene Rodlert, Eva Malmström, Anders Hult,
and Mikael Lindgren. Fluorinated dendritic polymers and dendrimers for waveguide applications. Optical Materials, 21(1–3):499–506, January 2003.
[175] Claire S. A. Pitois, Anders Hult, Mikael Lindgren, and Robert A. Norwood.
Characterization of fluorinated hyperbranched polymers and dendrimers for
waveguide applications. In Proceedings of SPIE — Design and Fabrication of
Planar Optical Waveguide Devices and Materials, volume 4805, pages 27–34,
Seattle, WA, USA, 2002.
[176] Suning Tang, Bulang Li, Nianhua Jiang, Dechang An, Zhenhai Fu, Linghui
Wu, and R.T. Chen. Ultra-low-loss polymeric waveguide circuits for optical
true-time delays in wideband phased-array antennas. Optical Engineering, 39
(3):643–651, March 2000.
[177] A.K. Chu, I.J. Lan, H.W. Chang, and M.H. Sheng. Low-loss polyimide-Ta2 O5 SiO2 hybrid ARROW waveguides. IEEE Journal of Photonics Technology Letters, 14(1):44–46, January 2002.
139
10 Bibliography
[178] K.B. Yoon. Low-loss multimode waveguides using organic-inorganic hybrid materials. Macromolecular Research, 12:290–292, June 2004.
[179] Xinshi Luo, Congji Zha, and Barry Luther-Davies. Preparation and optical
properties of titania-doped hybrid polymer via anhydrous sol-gel process. Journal of Non-Crystalline Solids, 351(1):29–34, January 2005.
[180] A. Yeniay, Renyuan Gao, K. Takayama, Renfeng Gao, and A.F. Garito. Ultralow-loss polymer waveguides. Journal of Lightwave Technology, 22(1):154–158,
2004.
[181] Toshikuni Kaino, Kaname Jinguji, and Shigeo Nara.
Low loss
poly(methylmethacrylate-d8) core optical fibers. Applied Physics Letters, 42
(7):567–569, 1983.
[182] Jong Kyu Kim and Dong Hack Suh. Estimation on the optic loss for acrylic
polymers by theoretical modification. Journal of Fluorine Chemistry, 125(3):
369–375, March 2004.
[183] John A. Taylor. Reformulating for curtain coating. Adhesives Age, 46(4):26–29,
June/July 2003.
[184] S.K. Bhattacharya, S. Bhatevara, D.A. Sutter, E.W. Kamen, G.S. May, and
R.R. Tummala. An automated workcell for meniscus coating on 24-in packaging
substrates. IEEE Transactions on Components, Packaging and Manufacturing
Technology, 24(4), December 2001.
[185] Alan Krauss and Edward Kamen. Meniscus coating for packaging substrates.
Advanced Packaging, 7(4):40,42–43, May 1998.
[186] Sachin Bhatevara, Swapan Bhattacharya, Robert Madayag, Edward Kamen,
and Gary May. An Automated Workcell For Meniscus Coating On 24 Inch
Packaging Substrates. Technical report, PRC, Georgia Institute of Technology,
2000.
[187] Experimental results from the DONDODEM EU project. partner confidential
report, 2002.
[188] U. Streppel, P. Dannberg, C. Wächter, A. Bräuer, L. Fröhlich, R. Houbertz, and
M. Popall. New wafer-scale fabrication method for stacked optical waveguide
interconnects and 3d micro-optic structures using photoresponsive (inorganicorganic hybrid-) polymers. Optical Materials, 21(1–3):475–483, 2002.
[189] L. F. Thompson, Willson. C. G., and M. J. Bowden, editors. Introduction to
Microlithography. Number 219 in ACS Symposium Series. American Chemical
Society, 1983.
140
10 Bibliography
[190] Jaesun Koo, P.G.R. Smith, R.B. Williams, C. Riziotis, and M.C. Grossel. Uv
written waveguides using crosslinkable pmma-based copolymers. In Summaries
of Papers Presented at the Lasers and Electro-Optics, volume 1, pages 490–491,
May 2002.
[191] B.P. Keyworth, J.N. McMullin, R. Narendra, and R.I. MacDonald. Computercontrolled pressure-dispensed multimode polymer waveguides. IEEE Trans.
Comp., Packag., Manufact. Technol. B, 18(3):572–577, August 1995.
[192] Mats Robertsson, Joacim Haglund, Christian Johansson, Pascal Gibaud, Daniel
Lambert, Michael Popall, Lothar Fröhlich, Arne Tolvgård, Arne Alping, Lars
Linden, Jens Rodekirch, Mark I. Zemel, and Christopher C. Nunes. Large area
patterning of high-density interconnects by novel UV-excimer lithography and
photo-patternable ORMOCERTM -dielectrica. In Proceedings of IMAPS 2001,
13th European Microelectronics and Packaging Conference & Exhibition, pages
276–281, Strasbourg, May 2001.
[193] H. P. Weber, F. A. Dunn, and W. N. Leibolt. Loss measurements in thin-film
optical waveguides. Applied Optics, 12(4):755–757, 1973.
[194] Yasuyuki Okamura, Shinji Yoshinaka, and Sadahiko Yamamoto. Measuring
mode propagation losses of integrated optical waveguides: A simple method.
Applied Optics, 22(23):3892–3894, 1983.
[195] John M. Senior. Optical Fiber Communications: Principles and Practice. Prentice Hall International Series in Optoelectronics, 2nd edition, 1992.
[196] R.Th. Kersten. A new method for measuring refractive index and thickness of
liquid and deposited solid thin films. Optics Communications, 13:327, 1975.
[197] Metricon. Metricon Homepage: Prism Coupler: Model 2010 Basic Description,
September 2003. URL http://www.metricon.com/basic.htm.
[198] Dennis Derickson, editor. Fiber Optic Test and Measurement. Hewlett-Packard
Professional Books. Prentice Hall, 1998.
[199] IEEE P802.3ap Backplane Ethernet Task Force. Homepage, January 2006. URL
http://www.ieee802.org/3/ap/index.html.
[200] IEEE P802.3ap Backplane Ethernet Task Force. IEEE P802.3ap Objectives,
2004. URL http://www.ieee802.org/3/ap/802_3_ap_objectives.pdf.
[201] Arne Alping. IEEE Backplane Ethernet — A Telecom View, May 2004. URL
http://www.ieee802.org/3/ap/public/may04/alping_01_0504.pdf.
[202] Edwin P. Plueddemann. Silane Coupling Agents. Plenum Press, New York, 2nd
edition, 1991.
141
10 Bibliography
[203] N. Inagaki. Plasma surface modification and plasma polymerization. Lancaster
: Technomic, cop., 1996.
[204] J.J.M. Ramos. Does the solid-liquid interfacial tension depend upon equilibration with the liquid vapors?: Comments. Langumir, 13:6607–6609, November
1997.
[205] D.Y. Kwok and A.W. Neumann. Contact angle interpretation: Re-evaluation
of existing contact angle data. Colloids and Surfaces A: Physicochemical and
Engineering Aspects, 161(1):49–62, 2000.
[206] K.S. Birdi, editor. Handbook of Surface and Colloid Chemistry. CRC Press,
1997.
[207] A.W. Neumann and Jan K. Spelt, editors. Applied Surface Thermodynamics.
Marcel Dekker, 1996.
[208] 12 x 2.7 / 3.3 Gbps Parallel Optical Transmitter/Receiver. PicoLight, July 2005.
URL http://www.picolight.com/UserFiles/PDF/05000838R1.pdf.
[209] Agilent AFBR-732B/BE and AFBR- 742B/BE Ultra Short Link Pluggable Parallel Fiber Optic Modules, Transmitter and Receiver — Data Sheet. Agilent, July
2005. URL http://cp.literature.agilent.com/litweb/pdf/5989-1096EN.
pdf.
[210] Y.H. Chang, H.C. Kuo, F.-I. Lai, K.F. Tzeng, H.C. Yu, C.P. Sung, H.P. Yang,
and S.C. Wang. High speed (> 13 GHz) modulation of 850 nm vertical cavity surface emitting lasers (VCSELs) with tapered oxide confined layer. IEE
Proceedings: Optoelectronics, 152(3):170–173, 2005.
[211] F. Gan and F.X. Kartner. High-speed silicon electrooptic modulator design.
IEEE Photonics Technology Letters, 17(5):1007–1009, 2005.
[212] A. Liu, D. Samara-Rubio, L. Liao, and M. Paniccia. Scaling the modulation
bandwidth and phase efficiency of a silicon optical modulator. IEEE Journal of
Selected Topics in Quantum Electronics, 11(2):367–372, 2005.
[213] Fang-I. Lai, Tao-Hung Hsueh, Ya-Hsien Chang, Hao-Chung Kuo, S.C. Wang,
Li-Hong Laih, C.P. Song, and H.P. Yang. 10 Gb/s single-mode vertical-cavity
surface-emitting laser with large aperture and oxygen implantation. Semiconductor Science and Technology, 19(8):L86–L89, 2004.
[214] M. Miller, M. Grabherr, R. Jäger, and K. J. Ebeling. High-Power VCSEL
Arrays for Emission in the Watt Regime at Room Temperature. IEEE Photonics
Technology, 13(3):173–175, 2001.
142
10 Bibliography
[215] Julian Cheng, Chan-Long Shieh, Xiaodong Huang, Guoli Liu, M.V.R. Murty,
C.C. Lin, and D.X. Xu. Efficient CW lasing and high-speed modulation of 1.3μm AlGaInAs VCSELs with good high temperature lasing performance. IEEE
Photonics Technology Letters, 17(1):7–9, 2005.
[216] A. Mereuta, A. Syrbu, V. Iakovlev, A. Rudra, A. Caliman, G. Suruceanu, C.-A.
Berseth, E. Deichsel, and E. Kapon. 1.5 μm VCSEL structure optimization for
high-power and high-temperature operation. Journal of Crystal Growth, 272
(1–4):520–525, 2004.
[217] Eugene Hecht and Alfred Zajac. Optics. Addison–Wesley Publishing Company,
1 edition, 1974.
[218] Y. Koike, T. Ishigure, T. Kobayashi, and E. Nihei. High-speed plastic optical
fibers and amplifiers. In Proceedings of SPIE - The International Society for
Optical Engineering, volume 3419, page 284, 1998.
[219] Christopher S. Koeppen. Photonics studies: rare earth chromophores for polymer optical amplifiers (waveguides). PhD thesis, Univ. of Pennsylvania, 1997.
[220] T. Kobayashi, S. Nakatsuka, T. Iwafuji, K. Kuriki, N. Imai, T. Nakamoto, C.D.
Claude, K. Sasaki, Y. Koike, and Y. Okamoto. Fabrication and superfluorescence of rare-earth chelate-doped graded index polymer optical fibers. Applied
Physics Letters, 71:2421, 1997.
[221] C. Koeppen, S. Yamada, G. Jiang, A.F. Garito, and L.R. Dalton. Rare-earth
organic complexes for amplification in polymer optical fibers and waveguides.
Journal of the Optical Society of America B (Optical Physics), 14:155, 1997.
[222] W.H. Wong, E.Y.B. Pun, and K.S. Chan. Er3+ - Yb3+ codoped polymeric
optical waveguide amplifiers. Applied Physics Letters, 84(2):176–178, January
2004.
[223] D. Barbier, P. Bruno, C. Cassagnettes, M. Trouillon, R.L. Hyde, A. Kevorkian,
and J.M.P. Delavaux. Net gain of 27 dB with a 8.6-cm-long Er/Yb-doped glassplanar-amplifier. In OFC ’98. Optical Fiber Communication Conference and
Exhibit. Technical Digest., volume 2, San Jose, CA, USA, 1998.
[224] G. Karve, B. Bihari, and R.T. Chen. Demonstration of optical gain at 1.06μm
in a neodymium-doped polyimide waveguide. Applied Physics Letters, 77:1253,
2000.
[225] L.H. Slooff, P.G. Kik, A. Tip, and A. Polman. Pumping planar waveguide
amplifiers using a coupled waveguide system. Journal of Lightwave Technology,
19(11):1740–1744, 2001.
[226] K. Fujii. Possibility of plastic optical fiber amplifier by CW-side excitation.
Optics Communications, 171(1,2,3):81–84, 1999.
143
10 Bibliography
[227] A. Tagaya, S. Teramoto, T. Yamamoto, K. Fujii, E. Nihei, Y. Koike, and
K. Sasaki. Theoretical and experimental investigation of Rhodamine B-doped
polymer optical fiber amplifiers. IEEE Journal of Quantum Electronics, 31(12):
2215–2220, 1995.
[228] Donald R. Zimmerman and Leo H. Spiekman. Amplifiers for the masses: EDFA,
EDWA, and SOA amplets for metro ans access applications. Journal of Lightwave Technology, 22(1):63–70, 2004.
[229] Jinyan Jin, Decheng Tian, Jing Shi, and Tongning Li. Fabrication and complete
characterization of polarization insensitive 1310 nm InGaAsP-InP quantum-well
semiconductor optical amplifiers. Semiconductor Science and Technology, 19(1):
120–126, 2004.
[230] K. Morito, S. Tanaka, S. Tomabechi, and A. Kuramata. A Broad-Band MQW
Semiconductor Optical Amplifier With High Saturation Output Power and Low
Noise Figure. IEEE Photonics Technology Letters, 17(5):974–976, 2005.
[231] Hong Ma, Xinjian Yi, and Sihai Chen. 1310 nm AlGaInAs-InP polarization
insensitive multiple quantum well optical amplifier. Infrared Physics and Technology, 45(1):9–12, 2004.
[232] Ping-Hui S. Yeh, I.-Fan Wu, Shijun Jiang, and Mario Dagenais. Monolithic
integration of a semiconductor preamplifier with a semiconductor tapered power
amplifier. In Proceedings of SPIE — The International Society for Optical
Engineering, volume 2148, pages 116–123, 1994.
[233] P.W. Juodawlkis, J.J. Plant, R.K. Huang, L.J. Missaggia, and J.P. Donnelly. High-power 1.5-μm InGaAsP-InP slab-coupled optical waveguide amplifier. IEEE Photonics Technology Letters, 17(2):279 – 281, 2005.
[234] Fujitsu Press Releases. Fujitsu Develops World’s First Semiconductor Optical Amplifier with Signal Waveform Re-shaping Function at 40 Gbps, March
2005.
URL http://www.fujitsu.com/global/news/pr/archives/month/
2005/20050304-01.html.
[235] A.V. Uskov, E.P. O’Reilly, M. Laemmlin, N.N. Ledentsov, and D. Bimberg.
On gain saturation in quantum dot semiconductor optical amplifiers. Optics
Communication, (1–3):211–219, April 2005.
[236] John H. Davis. The Physiscs of low dimension semiconductors, chapter 4.8.
Cambridge University Press, 1998.
[237] Tomoyuki Akiyama, Mitsuru Ekawa, Mitsuru Sugawara, Kenichi Kawaguchi,
Hisao Sudo, Akito Kuramata, Hiroji Ebe, and Yasuhiko Arakawa. An ultrawideband semiconductor optical amplifier having an extremely high penalty-free output power of 23 dBm achieved with quantum dots. IEEE Photonics Technology
Letters, 17(8):1614–1616, 2005.
144
10 Bibliography
[238] Jun’ichi Sasaki, Masataka Itoh, Takemasa Tamanuki, Hiroshi Hatakeyama,
Shotaro Kitamura, Tsuyoshi Shimoda, and Tomoaki Kato. Multiple-chip precise
self-aligned assembly for hybrid integrated optical modules using au-sn solder
bumps. IEEE Trans. Adv. Packag., 24(4):569–575, November 2001.
145
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement