SiGe Electro-Absorption Modulators

for Applications at 1550nm

by
Sarah Bernardis

Laurea in Physics
Universita degli Studi di Padova, 2005

Submitted to the Department of Material Science and Engineering in Partial Fulfillment

of the Requirements for the degree of

Master of Science in Materials Science and Engineering MASACHUSEINr.F

OF TE-.'HLOG.
at the --------
Massachusetts Institute of Technology UN 1 6 00
February 2008

Q 2008 Massachusetts Institute of Technology. All rights reserved.

A AftH ft.S
Signature of Author:
r Department of Mateaias Science and Engineering
January 22, 2008

Certified by:
Lionel C. Kimerling
Thomas Lord Professor of Materials Science and Engineering
Thesis Supervisor

Certified by:
Jurgen Michel
" Department of Materials Science and Engineering
Thesis Supervisor

Acceoted
I by:
J
Samuel M. Allen
POSCO Professor of Physical Metallurgy
Chair, Departmental Committee on Graduate Students
ii
 SiGe Electro-Absorption Modulators for Applications at 1550nm
by
Sarah Bernardis

Abstract
A novel SiGel_, electro-absorption modulator design is experimentally demon-
strated. The device is waveguide integrated, butt-coupled into high index
contrast Si/SiO2 waveguides. 0.75% Silicon concentration in the alloy is op-
timized for 1550nm applications. With its 400nm height, 600nm width, and
50,pm length, the device has a footprint smaller than 30Atm 2. Low effective
driving voltage, <2.5V, is needed to achieve an extinction ratio of 5.2dB in
the broad 1510-1555nm wavelength operation range. At 1550nm, an extinc-
tion ratio of 6.5dB is achieved with an applied effective bias of -2.5V.
High frequency measurements determine the device can reach a 3dB fre-
quency of 1.2GHz. Electrical characterization of the device shows high series
resistance (-15kQ) which is caused by fabrication over-etching during metal
contact deposition. Series resistance reduction to -100Q would allow the de-
vice to reach the predicted 3dB frequency of 100GHz with 10dB extinction
ratio.
A pseudo-linear relation is found between the achieved extinction ratio
and the applied effective bias. The ratio between these two quantities, the
modulation efficiency, can be considered as a new figure of merit of the device.
The slope of this pseudo-linear relation measures 2.2dB/V for extinction
ratio values ranging between 0 and 5.5dB. In terms of modulation depth it
is equivalent to a slope of 40%/V in the range 0.5V-2V. Finally, an ultra-low
power consumption per bit of 34fJ/bit is measured for a capacitance of 11fF
and an effective applied reverse bias of 2.5V.

Thesis Supervisor: L.C. Kimerling

Title: Thomas Lord Professor of Materials Science and Engineering

Thesis Supervisor: J. Michel

Title: Department of Materials Science and Engineering
Acknowledgments

This thesis is like a hand blown glass pumpkin.

Blowing a glass pumpkin requires a starter, a gaffer, and a stemmer. The

starter, the most challenging task among the three, gathers the glass, colors
it, and molds it. The gaffer puts the neckline in and decides the final shape
while the stemmer blows and, as the name suggests, makes the stem. The
stemmer has the task of finishing a pumpkin that encloses the personality of
three people. Three people that worked together to create a unique piece.

In this regard this thesis is like a hand blown glass pumpkin. To my

starters, Prof. L.C. Kimerling and Dr. J. Michel, starters of this project,
always full of ideas and motivation. The gaffer, Dr. J. Liu, sharing his mod-
ulator design and more importantly insites on what/ how/ why/ and more.
I would like to thank you all for giving me the chance of being your stemmer.

Finally, a special "thank you" to J. Cheng, for all the help with the high
speed measurement set-up.
Contents

1 Introduction
1.1 Motivation . ...............
1.2 Modulators: General Overview .....
1.3 Modulator Physical Principles .....
1.3.1 Electro-optic Effect .......
1.3.2 Carrier Injection and Depletion
1.3.3 Thermo-optic Effect ......
1.3.4 All Optical Mechanism .....
1.4 Modulator Optical Structure Design .
1.4.1 Ring Resonators ........
1.4.2 Mach-Zehnder Structure . .
1.4.3 Fabry-Perot Interferometers .
1.5 Summary ................

2 Device Physics, Material's Choice, Design, and Fabrication 35

2.1 Material's Choice Overview ......... .......... 36
2.2 Franz-Keldysh Theory ............ .......... 37
2.3 SiGel_ Alloy .......... .... ......... 39
2.3.1 Absorption Coefficient Optimization . . . . . . . . . . 40
2.4 Device Design ............. ... .. ....... 43
2.4.1 Si/SiO 2 Waveguide ........ ............ 45
2.4.2 SixGelx (x = 0.75%) Electro-absorption Modulator . . 46
2.4.3 Figure of Merit Optimization . . . . . . . . . . . . . . 47
2.5 Device Fabrication ....... .... .. ...... 52
2.6 Summary ........... . . . . ........ 55
3 Experimental Results and Analysis 56
3.1 IV Characteristics ................ ......... 57
3.2 Low Frequency Measurements . ................. 59
3.3 High Frequency Measurements . ................. 66
3.4 Summary ..... .......... .......... 69

4 Conclusions and Future Work 70

4.1 Future work .......... ...... .......... 73
4.1.1 Current Devices ...................... 73
4.1.2 Fabrication Implementation . .............. 74

Bibliography 76
List of Figures

1.1 Example of a 10-layer metal interconnect: projected structure

for the 35nm technology node [1]. Interconnect levels are di-
vided in local, intermediate, and global. ........... . . 2
1.2 Example of an optical link. Light source (on- or off-chip),
modulator, detector, and waveguides are the most important
components of such a link. . .................. . 3
1.3 SEM picture of a 10pm diameter ring modulator coupled to a
waveguide, [6] ........................... .. 15
1.4 Modulation mechanism by thermo-optic induced optical bista-
bility explained by Almeida, [9]. . ............ . . . 17
1.5 Modulation mechanism by carrier induced optical bistability,
[10]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . 17
1.6 Top view of a ring modulator coupled to a waveguide. Inset
shows the ring modulator cross-section with the embedded p-
i-n junction [12]..... .............. ...... . 20
1.7 Ring modulator with an embedded p-i-n-i-p junction struc-
ture. (a) p-i-n-p junction cross-section. (b) Complete device
top view: p-i-n-i-p junction embedded in the double-ridge ring
modulator. Only the outer ring supports an optical mode. [16] 23
1.8 Mach-Zehnder structure showing phase shifters in the two
arms, [23] ............................... 24
1.9 Silicon modulator based on a Mach-Zehnder interferometer
with a pn junction and traveling wave electrodes. 20dB ex-
tinction ratio, f3dB -20GHz, data rate: 30Gbit/s, [20]. ..... 25
1.10 MOS capacitor cross-section from [23]. Extinction ratio of
16dB is reached with 7.7Vpp, 1GHz bandwidth, modulation
speed 4Gbit/s ................... ........ 27
1.11 MOS capacitor cross-section from [26]. Extinction ratio of
3.8dB is reached with 1.4V,, bandwidth 10GHz, and 10Gbit/s
modulation speed ................... ...... 27
1.12 p-i-n Fabry-Perot microcavity with distributed Bragg reflec-
tors. 20p long, 6.9dB extinction ratio with 0.87V, [29] ..... 32
1.13 Improved design of the p-i-n Fabry-Perot microcavity. 10pm
long, 6.9dB extinction ratio with 0.87V, [30] . ......... 32
1.14 3-dimensional full structure of the improved design of the p-i-
n Fabry-P6rot microcavity with distributed Bragg reflectors.
10pm long, 6.9dB extinction ratio with 0.87V, [30] ....... 33
1.15 Fabry-Perot microcavity (<6pm) with distributed Bragg re-
flector structure and embedded p-i-n junction. Low frequency
measurements show 5.93dB modulation depth with 5.6V dc
bias; maximum data transmission speed is 250Mbit/s with
5.87dB extinction ratio with 0.7V(dc) and 2Vpp(ac), [31]. . . . 33

2.1 Elemental Germanium band gap diagram. (a) Bulk Germa-

nium; (b) tensile strained intrinsic Germanium: the direct bad
gap (E'r ) decreases under the effect of tensile strain, [32]. . . 36
2.2 Schematic diagram representing the Franz-Keldysh effect. The
band diagram tilting, due to an applied electric field, generates
a finite tunneling probability for photon energies (hw) smaller
than the band gap (Eg) value. . .................. 37
2.3 Franz-Keldysh effect: absorption coefficient (a) versus photon
energy (hw). The weakly absorbing regime is just below the
band gap value, Eg, ............ ........... 37
2.4 SixGel-, indirect bad gap dependence on Silicon concentra-
tion. For x<15% the relation is linear, [40] .......... . 41
2.5 Figure of merit optimization with respect to Silicon content in
the SiGel_, alloy [36] .......... ............ 43
2.6 3-D representation of the SixGel- modulator, the two Silicon
layers (n+ and p+ regions of the p-i-n junction), and elec-
trodes (W contacts and A1Ti metal layers). Silicon waveguides
(not pictured) couple in and out directly from the SixGel_,.
Design not in scale. ......... ......... .. ..... 44
2.7 2-D longitudinal cross-section of the butt-coupled SixGel_-
modulator with a-Silicon waveguides. Tapered waveguides are
used to enhance coupling efficiency. Design not in scale.... . 45
2.8 High-index contrast Si/Si0 2 waveguide cross-section. Core
dimensions used are indicated in the figure. Figure not in scale. 45
2.9 SixGel_- electro-absorption modulator cross-section. Design
not in scale .......... .................. 47
2.10 Trade-off between the modal overlap in the Silicon waveguide
and the confinement factor in the SixGel_- layer [36]. ..... 49
2.11 Figure of merit optimization with respect to the SiGel_-
modulator layer thickness; various SiGel_- widths are con-
sidered while length is kept constant (50pm) [36]. . ....... 50
2.12 Trade-off between 3dB bandwidth, extinction ratio and SixGel_.
modulator length [36]. .............. ....... 51
2.13 The process start with an SOI substrate. Optical isolation
from the substrate is thus ensured. The oxide layer is 3pm
thick. The crystalline Silicon layer is doped p+ by ion im-
plantation .............. ................ 53
2.14 The c-Silicon p+ Boron doped layer is patterned and mesas
are formed. The mesas are the p+ region of the p-i-n junction
of the device ............................ 53
2.15 An oxide deposition follows. The obtained ondulated layer is
planarized with chemical mechanical polishing (CMP). ..... 53
2.16 Amorphous Silicon is then deposited and patterned. This step
forms the core of the amorphous Silicon waveguides. Another
oxide deposition and planarization form the upper cladding of
this waveguide. .......................... 53
2.17 Once the waveguide structure is defined, standard photolithograpy
steps are performed to open a trench that exposes the p+ c-
Silicon mesa. .......... ................. 53
2.18 As the trench is opened, epitaxial SiXGelx is grown selec-
tively between the input and output of the amorphous Silicon
waveguides. The trench is purposely overfilled then planarized. 54
2.19 The last step is poly-Silicon deposition, n+ Phosphorous im-
plantation, then patterning. Tungsten and A1Ti metal depo-
sition to form vias and contact pads, respectively. ....... . 54
2.20 Fabricated SixGel_- device cross-section. The SiGe modula-
tor is indicated in the middle of the figure. The bottom oxide
layer, the p+ and n+ Silicon layers as well as the metal con-
tacts are indicated. ................... ..... 54
3.1 Device layout. Two sets of devices are shown: in the top set,
modulators are integrated with photodetectors; in the bottom
set, there is one modulator per waveguide. . ........... 57
3.2 IV characteristics of the 50pm long electro-absorption SixGel-,
modulator ................ . . .............. 58
3.3 SEM image of the Tungsten contact. The silicide layer is com-
pletely missing and over-etching of the Silicon layer is clearly
seen ........... .. ................... 59
3.4 Schematic experimental set-up for dc measurements. ...... 60
3.5 Transmittance spectrum of the 50pm electro-absorption SixGelx
modulator at optical "on" state (OV). From the Fabry-Perot
fringes above 1540nm, the absorption loss (2dB) due to indi-
rect band gap absorption is inferred. . ........... . . 62
3.6 Transmittance spectra of the 50pm electro-absorption SiGel_,
modulator for three applied voltages: from lighter to darker
gray as the applied reverse bias increases (black line at OV
inserted as a reference). Above 1540nm, the slope of direct
band gap absorption edge flattens towards longer wavelengths
as the applied reverse bias increases. . .............. 62
3.7 Modulation depth of the 50pm electro-absorption SiGel_.
modulator when 7V reverse bias is applied. Modulation higher
than 78% (6.5dB) can be achieved for a broad wavelength
range as indicated in the figure. . ............ . . . 64
3.8 Performance of the 50pm electro-absorption SiGel_, modu-
lator: (a) as-measured modulation performance as a function
of wavelength. (b) maximum modulation depth as a function
of applied effective reverse bias. Data in (b) is extrapolated
from (a). At 1550nm, a pseudo-linear relation is seen between
the modulation depth and the applied bias in the range 0.5-2V. 65
3.9 Schematic experimental set-up for high frequency measurements. 66
3.10 High frequency response of the 50pm electro-absorption SiGel_i
modulator. (a) Frequency response at 3V reverse bias as a
function of optical input power. (b) Frequency response as a
function of applied reverse bias with 20mW fixed optical input. 68
List of Tables

1.1 Selected published results on Silicon based modulators oper-

ating at 1550nm presented in this chapter. Optical structure
versus bias design. ....................... 12
1.2 Intel's MOS capacitor Mach-Zehnder structure evolution. The
second (Liu [23]) and third column (Liao [27]) are two different
descriptions of the same device. Fourth and fifth represent ex-
perimental (Liao [26]) and modeling (Liao [27]) improvements.
(NA = not applicable) ......... ............. 30

2.1 Si/Si0 2 waveguide calculation results at 1550nm after [36] . . 46

2.2 Optical mode evaluation. TE versus TM fundamental modes
in Si.Gel_x device. Data after [36] . .............. 47
2.3 Parameter dependence on SiGelx thickness. Parameters cal-
culated at 1550nm. After [36]. . .................. 50
2.4 Modal absorption, modal overlap and reflectance influence ex-
tinction ratio and insertion loss. Crosses indicate the depen-
dence of these three parameters on the variables influencing
the figure of merit. H,W, and L are height, width, and length
of the Si.Gel_, layer ........... ........... 52
Chapter 1

Introduction

The aim of this thesis is to provide a comprehensive understanding of the im-

portance of modulators in electronic-photonic integration. The figure of merit
of desirable modulators is characterized by small footprint size, low driving
voltage, and high modulation depth and speed. It is absolutely challenging to
fulfill simultaneously all the above criteria with the additional requirements
of low insertion loss and temperature independent performance. This the-
sis presents a unique design of waveguide integrated modulators with small
footprints, low driving voltages, and a large operational bandwidth over the
entire C-band wavelength range.

1.1 Motivation

Telecommunication and on-chip interconnects are about to reach a bottle-

neck. As fabrication nodes decrease feature size capabilities, transistors can
be shrunk down more, but metal interconnects do not keep up quite as well
with the shrinking process. Two major problems arise as miniaturization oc-
1.1 Motivation

-Passivation

- Dielectric
Etch stop
layer
Dialectic
-diffusion
barrier
la

1,

conductor
-withrmetal
barrier liner

a,
Figure 1.1: Example of a 10-
layer metal interconnect: projected
structure for the 35nm technology
Pre-metal node [1]. Interconnect levels are
-Tungsten divided in local, intermediate, and
ra global.

curs. Below the 150nm node, interconnect RC delay increases as feature size
decreases causing signal desynchronization in clock distribution, resulting in
higher bit error rates. Secondly, heat dissipation due to the on-chip length of
these interconnects causes major problems. Thus, metal interconnects are far
from being ideal for high speed data transmission and are a clear bottleneck
to potential performance improvements of new microprocessors. An example
of metal interconnect is in Figure 1.1.
A viable way of overcoming this challenge would be the integration of
photonics with microelectronics. Data would need to be collected, routed,
modulated, detected in a very small space and in a very efficient way. Optical
interconnects (Figure 1.2) would have the advantage of low power require-
ments, low latencies, high bandwidth and no heat dissipation issues, since
photons do not generate heat as they propagate through waveguides. On-
1.1 Motivation

300MHz to 2.2GHz RF

Mode

Modutator

ed
rms
cal)

Figure 1.2: Example of an optical link. Light source

(on- or off-chip), modulator, detector, and waveguides
are the most important components of such a link.

chip communication between standard CMOS electronic components and

photonic devices would mainly be carried out by a light source, modulator,
detector, and waveguides to connect these three devices. Mainly, the mod-
ulator would imprint Os and is on the on-chip or off-chip generated light
signals and the photodetector would convert these signals back to electrical
ones.
Integration of Silicon photonic interconnects with microelectronics would
represent a revolutionary breakthrough for the next generation of communi-
cation tools as well as computing platforms. Among optical devices, modula-
tors are the ones with highest power consumption. Therefore, they represent
the most critical component for enabling this integration to happen. The
focus of this thesis is a SiGe electro-absorption modulator design with small
1.1 Motivation

footprint and low driving voltage, hence ensuring low power consumption.
Integration of electronics and photonic devices would be greatly enhanced
if the same facilities and technology could be used. Silicon is the main mate-
rial used for electronics but it poses some challenges for photonic components
such as high propagation losses, low electrooptic coefficient, high fiber-to-
waveguide coupling losses. Nanofabrication improvements are slowly solv-
ing each one of these problems. Nowadays, ultra-compact micrometer size,
high-performance active and passive photonic components are being demon-
strated. Thus, Silicon optical interconnects would allow CMOS monolithic
integration and ultra-fast signal processing, allowing for novel chip architec-
tures to be exploited.
Drawback of Silicon is that its direct band gaps are in the range 3-4eV.
An alternative material is Germanium which is also CMOS compatible and
its direct band gap is 0.8eV, corresponding exactly to the wavelength of
1550nm. For wavelengths just smaller than 1550nm, Germanium has good
quantum efficiency in photon absorption. Electron mobility (3900 cm 2 /s.V)
and hole mobility (1900 cm 2 /s-V) are both better than in elemental Silicon.
It is expected that Germanium devices will be faster and will operate with
lower bias with respect to the same Silicon devices.
Being both elemental materials CMOS compatible, their alloy, SiGe is also
CMOS compatible. The optoelectronic properties of this alloy are promising
for modulating applications. Especially, its electro-absorption properties are
of current research interest.
SiGe electro-absorption modulators integrated in high index contrast waveg-
1.2 Modulators: General Overview

uides are investigated in this thesis. These devices are based on the Franz-
Keldysh effect where an applied electric field enhances the absorption coeffi-
cient in the weakly absorbing regime. Device design is optimized for 1550nm
operations.
This thesis will explain why SiGe is used, how to design the electro-
absorption modulator and how it physically works, and finally how to mea-
sure it once it is fabricated. As background knowledge, this first introductory
chapter will summarize published results of other research groups, why their
ideas are great, what it is possible to learn from them, especially how their
designs work. The second chapter will explain the physical principles be-
hind the working device; it will consider material's characteristics that will
be exploited in the device design, the device design itself, its optimization,
and briefly its fabrication. The third chapter will prove experimentally how
incredibly well this SiGe electro-absorption modulator performs. Finally, the
fourth chapter will summarize key points and look into the future.

1.2 Modulators: General Overview

As G Parry wrote in a brief column in Nature, A modulator is analogous to a

camera shutter: it transmits light when it is open, but absorbs light when it is
closed, [2]. This section is meant to explore this picture of the shutter mech-
anism, considering which parameters are of fundamental importance when
evaluating its performance for applications at 1550nm. Physical mechanisms
leading to modulation will be considered in the next section.
1.2 Modulators: General Overview

Several different types of modulators have been demonstrated. The physics

used to achieve modulation as well as the design of the structure highly in-
fluence the performance of the device. In order to understand which devices
work best, a few key common parameters can be compared. These parame-
ters are: device footprint, driving voltage (both dc and ac bias), modulation
depth, switching speed (also rise and fall times), bandwidth, insertion loss,
wavelength sensitivity, and temperature sensitivity.
On-chip optical interconnects need to fit with the always shrinking elec-
tronic counterpart: device footprint needs to be small, possibly reaching the
physically limited dimensions of the wavelength of light being used. As these
devices are monolithically integrated, the driving voltage has limitations im-
posed by the node at which they are fabricated. As an example, at the 90nm
node, only a voltage smaller than 1V will be allowed. In turn, the applied

voltage determines the achievable modulation depth. The modulation depth

is one of the key parameters of the device. It is measured in decibels and
it represents the optical output intensity ratio between the "off" and "on"
states of the modulator. If this ratio is large, the intensity output difference
between the "on" and "off" states of operation is large and the signal-to-noise
ratio is consequently also high. Devices reported in the literature generally
only fulfill a few of these requirements.
In Mach-Zehnder devices, the applied voltage required to achieve a x-
radian phase shift also depends on the length of the arm of the interferome-
ter. The product of these two quantities, voltage and arm length, is usually
considered figure of merit of the device, V,.L,. Goal is to minimize this
1.2 Modulators: General Overview

product.
Device fabrication tolerance represents a challenge when an operational
wavelength is predefined: wavelength sensitivity is one of the major concerns
in devices such as ring modulators. Moreover, temperature sensitive devices,
such as ring resonators, can easily get detuned as a bias is applied. As carriers
flow, the temperature increases and the optical properties change. Even small
temperature changes can drastically compromise device performance.
Another fundamental parameter is the available bandwidth of operation.
It represents the amount of data that can be carried and modulated in a
unit time. It is calculated to be the difference between the upper and lower
frequencies at which the modulation depth falls to 50% of its maximum value.
Gathering these concepts together into a quantitative expression, a gen-
eral figure of merit, independent of device design, could be given by the
ratio:

Figure of Merit = Bandwidth / Power - Device Footprint

In the case of an electro-absorption modulator, a simple figure of merit is

the ratio between the modulation depth (Aa) and the insertion loss at the
optical "on" state (aon):

Electro-absorption Modulator = •

since the former should be as high as possible and the latter, instead, as
low as possible. On the optical "on" state, the modulator is supposed to be
completely transparent to light. Insertion loss refers to optical power loss due
1.3 Modulator Physical Principles

to material absorption, sidewall scattering due to surface roughness, and loss

due to optical mode mismatch between the waveguide and the modulator
structure. High insertion losses deteriorate the optical output. Insertion loss
and modulation depth are mathematically defined as:

Insertion Loss = -10 log{e - °on-

' L

Modulation Depth = -10log { - }

where a,o refers to the absorption coefficient at the optical "on" state, •off
to the optical "off" state, and L is the device length.

1.3 Modulator Physical Principles

Light modulation in Silicon based devices implies a change in the optical field
due to some applied signal. This signal is typically electrical, but other para-
metric changes are possible. These parametric changes modify the refractive
index seen by the propagating optical mode. Using an electrical signal, the
most efficient way of implementing optical modulation in Silicon is through
carrier injection or depletion.
Free carriers can be injected either optically with a pump laser or electron-
ically with a p-i-n diode (usually under forward bias). Other CMOS devices,
such as metal-oxide-semiconductor (MOS), have been used especially in car-
rier inversion mode. Devices that work at a few GHz have been demonstrated
both with Mach-Zehnder and ring resonator structures. Drawback of carrier
injection is that the recombination time is of the order of nanoseconds, thus
limiting the device speed. A reverse bias could be applied to sweep out the
1.3 Modulator Physical Principles

carriers faster, but their transit time would still be a limiting factor. Also,
carrier injection happens only under high current, leading to higher power
consumption. In the case of rings, also, there is a tradeoff between high qual-
ity factor and speed. The higher the quality factor, the higher the photon
lifetime in the cavity, and the higher the delay in modulation. The lower
the quality factor, the lower the modulation depth (i.e. the contrast between
resonant and non-resonant wavelength becomes smaller).
In the following remaining of this section, mechanisms exploited to mod-
ulate Silicon based devices at 1550nm are briefly explained.

1.3.1 Electro-optic Effect

A change in the real part of the refractive index with an externally applied
electric field is called electro-refraction,while a change in the imaginary part
of the refractive index is called electro-absorption.

Electro-refraction Effects Even though other integrated optical tech-

nologies exploit electro-refraction, it is not useful in Silicon. The two main
electro-refraction effects are Pockels and Kerr effects. The Pockels effect, a
change in refractive index linearly proportional to the applied electric field,
is polarization dependent and it is not present in Silicon due to its crystal-
lographic structure. The Kerr effect, a second-order electric field effect, is
present in Silicon but it is extremely weak. Besides, this effect is extremely
small compared to carrier injection effects.
1.3 Modulator Physical Principles

Electro-absorption Effect This effect is known as Quantum Confined

Stark Effect in quantum well devices and as Franz-Keldysh Effect in bulk
material. The refractive index is rapidly modified by field effects. In the
case of the Franz-Keldysh effect, it induces a change in both the real and
imaginary parts of the refractive index. When reverse biasing a p-i-n junc-
tion, a high electric field is generated in the intrinsic layer. This field induces
changes in both real and imaginary parts of the refractive index. However,
the electro-absorption effect is stronger of the two. It distorts the energy
bands of the semiconductor, shifting the band gap energy, which in turn
modifies the absorption properties of the material (i.e. the imaginary part of
the refractive index), particularly at wavelengths close to the band gap.
As the electric field is turned "on" and "off", the material becomes trans-
parent or opaque to specific wavelengths. Devices that rely on this effect
can theoretically reach modulating speeds of tens of GHz as the response of
the material to the electric field happens in the picosecond time range. The
reverse bias applied to the junction allows for smaller current to flow in the
device with respect to devices that work under forward bias, as the ones that
exploit free carrier injection: device comparison shows lower power required
to achieve the same modulation depth.

1.3.2 Carrier Injection and Depletion

Optical losses are enhanced by material absorption which is due to the pres-
ence of free carriers. The imaginary part of the refractive index is related
to the loss coefficient. Hence, modification of the carrier concentration im-
1.3 Modulator Physical Principles

plies a change in the refractive index. Free carrier injection decreases the
refractive index inducing intensity modulation. Injection, depletion, accu-
mulation, and inversion are all mechanisms that modify the free carrier con-
centration in carrier dispersion devices. Electrical structures such as p-i-n
diodes, metal-oxide-semiconductors field-effect transistors, and bipolar-mode
field-effect transistors allow this mechanism to be exploited.

1.3.3 Thermo-optic Effect

If the real part of the refractive index is temperature dependent, phase mod-
ulation of the transmitted light can be induced by local variation of the
temperature. Also, while the thermo-optic effect induces a red shift (i.e. it
increases the refractive index), the carrier injection or depletion effect causes
a blue shift (i.e. the refractive index decrease). Interference of the two mech-
anisms should be considered during device design and operation. Drawback
of this mechanism is slow frequency modulation (MHz). For this reason,
devices relying only on this effect will not be considered in the following.

1.3.4 All Optical Mechanism

All optical modulation relies on optical retardation. One beam of light (pump
beam) is used to control a second beam of light (probe beam). An index
change is produced and it induces a phase shift in the propagating wave.
This phase shift is converted to an intensity shift through the structure of
the device (i.e. using an interferometric structure). There are two ways of
achieving all optical modulation: either by thermo-optic non-linear effect
1.4 Modulator Optical Structure Design

All-Optical Modulation Bias Induced Modulation

All-Optical Optical Bistability p-i-n MOS
Ring Resonator [6]-[8] [9], [10] [11]-[16] [17], [18]
Mach-Zehnder [19] [20]-[22] [23]-[27]
Fabry-Perot [28] [291-[31]

Table 1.1: Selected published results on Silicon based modulators operating

at 1550nm presented in this chapter. Optical structure versus bias design.

induced by the large Silicon thermo-optic coefficient; or by carrier induced

effects. These two effects in conjunction with device structure (Mach-Zehnder
and ring resonator) are described in the following section.
All optical modulation enables for ultra-fast signal processing, overcoming
the optics-to-electronics conversion limitation.

1.4 Modulator Optical Structure Design

While examining the principles of operation of all different combinations of
optical and electrical structures as of now demonstrated, recently published
results (2004-2007) will be explained to provide a hands-on feeling of what
these structures are capable of. A variety of different structures for appli-
cations at 1550nm have been simulated and sometimes also fabricated and
tested as schematized in Table 1.1.
The three main optical structures are ring resonators, and Mach-Zehnder
or Fabry-Perot interferometers. There can be no bias involved as in the
case of all optical modulation, or these devices can have p-i-n embedded
1.4 Modulator Optical Structure Design

structures, or MOS configuration. Different matching of optical structure

and electrical structure allows for different achievable performance.

1.4.1 Ring Resonators

Before listing different ring configurations that are found in the literature,
a few key concepts that are common to the geometrical structure itself are
here summarized. Reference papers on this topic are [3]-[5].
Rings are wavelength selective resonant structures. Light is evanescently
coupled from an adjacent waveguide to the ring cavity. The ring circumfer-
ence determines the resonant wavelength: only light with wavelength equal
to an integer multiple of the ring circumference will be trapped reducing
transmission at the through waveguide port. The amount of light evanes-
cently coupled from the waveguide to the ring is of fundamental importance
and it depends on the gap between the two structures. Another very im-
portant parameter is the lifetime of photons in the resonator. As a general
rule of thumb, a high cavity quality factor implies strong light confinement
and hence strong wavelength dependence on the index of refraction of the
ring, which in turn implies high modulation. High switching speed is, on the
other hand, achieved with low photon lifetime, i.e. low quality factor. High
transmittance is detected when there is low total insertion loss.
A ring is a multi-pass device: at resonance, light travels multiple times in-
side the device in round trips, increasing the optical path length and leaving
the physical device length unmodified. Less carriers are needed to achieve the
targeted modulation depth because light has multiple interactions with them:
1.4 Modulator Optical Structure Design

less power is required to drive carriers in and out of the device. Also, ring
resonators are optimized to modulate light of a specific wavelength. Light
of all other wavelengths is not affected by the ring structure. This makes
ring resonators extremely interesting for wavelength division multiplexing
applications. Drawback of the ring resonator is that it is highly tempera-
ture sensitive. Oxidation induced strain between the Silicon core and oxide
cladding is suggested to diminish temperature variation issues, [10].
Typically, rings are either used for all optical modulation or they are
externally biased to obtain free carrier effects. All optical modulation exploits
non-linear optic effects in two main ways: the first one uses two beams of
light, the first beam changes the optical properties of the material, while the
second one is being modulated; the second method involves optical bistability
effects induced by one- or two-photon absorption mechanisms. Both methods
are experimentally demonstrated ([9] and [10]).

All Optical Modulation All optical modulation relies on two beams of

light: the first one controls the properties of the waveguide material, which,
in turn, directly affect the second one. A control beam and probe beam are
coupled into the ring through two different resonances. The control beam
generates free carriers by (one- or) two-photon absorption effect, triggering
the plasma dispersion effect which reduces the refractive index of the ring
waveguide. A small change in refractive index is sufficient to detune the
resonance of the probe beam. In fact the probe resonance experiences a blue-
shift which induces modulation of its output power. As the control beam is
1.4 Modulator Optical Structure Design

Figure 1.3: SEM picture of a

10Lm diameter ring modula-
tor coupled to a waveguide,
[6]

switched off, the probe resonance shifts to its original position thanks to the
free carrier recombination.
Almeida [6] (and Lipson [7]) exploits all optical modulation due to two-
photon absorption in a ring with 10tm diameter (Figure 1.3). The probe
wavelength used is 1535nm while the pump wavelength is 1550nm with 10ps
pulse duration. Probe and pump cavity photon lifetime are 1.8ns and 2.8ps,
respectively. Just below resonance, at 1535nm, the measured modulation
depth is 94% (12.2dB); on resonance, 1535nm, 91% (10.45dB); and above
resonance (1554nm and 1550nm), 75% and 97%. The relaxation time is 450ps
and it is determined by the recombination of photoexcited free carriers on
the unpassivated sidewalls of the structure.
Two-photon absorption is used also in another ring device by Xu [8]. It is
optimized for high speed measurements as wavelength conversion rather than
modulation. "On" resonance, light is trapped inside the ring and light that
is back-coupled to the through waveguide is destructively interfering with
directly transmitted light. When the resonator shifts from "on" resonance
1.4 Modulator Optical Structure Design

to "off" resonance, little coupling between the input waveguide and the ring
resonator happens: light mainly propagates through the waveguide; only
the small amount of light that couples to the ring experiences a temporal
refractive index change in the ring which in turn generates a wavelength
shift. As this wavelength-shifted light back-couples to the through waveguide,
no destructive interference is measured. Instead, overshoots and damped
oscillations due to beating effects are noticed. The former is determined
by the cavity photon lifetime, while the wavelength shift determines the
frequency of the latter. The ring has 10pm diameter, resonance wavelength
is 1550nm, cavity photon lifetime is 9.1ps; wavelength conversion happens
with a speed of 0.9Gbit/s. Implementing the structure with a p-i-n junction,
the expected carrier lifetime is 30ps with a wavelength conversion rate of
10Gbit/s.
Manolatou [5] simulates two-photon absorption in two rings of 2 and
5pm radiuses. Resonant wavelengths are 1561 and 1558nm, respectively.
Modulation depth of -90% can theoretically be achieved with both rings:
in the case of the 2pm one, the pump energy is 1.25pJ, where the required
energy to generate the free carrier concentration is 0.042pJ; while in the case
of the 5pm ring, the pump energy is 3pJ and the energy required to generate
the needed free carrier concentration is only 0.1pJ.

Optical Bistability due to Thermo-optic Effect Measuring the

transfer function of the ring (i.e. fixing the input wavelength and detecting
the transmitted power versus input power), a resonance shift is seen due to
1.4 Modulator Optical Structure Design 17

IL

0
0
91
o. o.s5
0

10 P,
03-
Input Power (pW)
Fig. 3. Hysteresis curve for the quasi-TE mode of a SOI f2
ring resonator excited at Ao = 1562.0 nm. Markers define 0.12
curve points related to the transmission spectra in the
insets: 0 unshifted spectrum on the linear region before
hysteresis loop, 0 shifted spectrum due to high-input in- 0 2
- gi --
4
.
Q2 . I
14 . IsI
8e 10
creasing power, O resonance shifted to pump wavelength. Power (mW)
Input

Figure 1.4: Modulation mechanism by Figure 1.5: Modulation mecha-

thermo-optic induced optical bistabil- nism by carrier induced optical
ity explained by Almeida, [9]. bistability, [10].

thermo-optic effects. Transmission is affected because the resonance shift

depends on the optical power circulating in the ring. A hysteresis loop is
created as this circulating optical power is determined by the wavelength
difference between the optical source and the shifted resonance. Using a
ring with a 10pm diameter and a 100kHz square modulated 1561nm pump
wavelength Almeida [9] experimentally show that 8dB modulation can be
achieved with 250pW of power required for switching. Instead, if constant
optical power of 550pW is supplied through the pump wavelength, less than
45 1pW are required to achieve 10dB modulation on a probe wavelength of
1580nm.

Optical Bistability due to Carrier-Induced Effect Carrier-induced

bistability relies on generation of free carriers by two-photon absorption. Free
1.4 Modulator Optical Structure Design 18

carriers reduce the refractive index inducing a blue-shift in the ring resonator
spectrum.
If the starting optical pulse has shorter wavelength than the resonant
wavelength of the ring, free carrier concentration in the ring is low. How-
ever, the optical pulse will produce free carriers inducing resonance blue-
shifting. The blue-shift itself will enhance free carrier generation causing
more blue-shifting. This induces a positive feedback loop that feeds itself
until the wavelength of the input optical pulse becomes longer than the res-
onant wavelength of the ring. The blue-shift process stabilizes when the
induced blue-shift becomes large enough and the wavelength of the input
optical pulse becomes longer than the ring resonant wavelength, the posi-
tive feedback becomes negative. A hysteresis loop is seen when this positive
feedback loop becomes negative (Figure 1.5).
Carrier induced optical bistability is reported by Xu [10] at a wavelength
of 1532nm in a ring with 10/pm diameter and quality factor of 14,000. The
effect is achieved with power smaller than 10mW. The reported transmission
drop at resonance is of 14dB.
Thermo-optic and carrier induced bistability effects are compared in the
same paper [10]. Experimental observations show that for the same input
optical pulse width (23ns), thermo-optic induced bistability prevails on car-
rier induced bistability. Also, in order to obtain the same results using the
two mechanisms separately, the required input optical power required to trig-
ger the thermooptic mechanism is one order of magnitude smaller than that
required to trigger carrier induced effects (10mW).
1.4 Modulator Optical Structure Design

Externally Applied Bias When an externally applied bias is used to

induce free carrier effects, two structures are widely used in the literature:
p-i-n junctions and MOS structures [11].
p-i-n junctions are better to achieve high speed modulation. In terms of
optical and electrical confinement, the optical mode traveling in the waveg-
uide has greater overlap with the region where the index of refraction changes
due to the applied bias (electrical region) [12]. Clear drawback is that if ex-
traction of carriers under reverse bias happens in the picosecond range, carrier
injection under forward bias happens in the nanosecond range. A decreased
injection time can be achieved with a resonant p-i-n structure where the rise
and fall times are enhanced.

p-i-n and p-i-n-i-p Configurations Free carriers are injected in the

intrinsic region of the device through a p-i-n junction embedded in the ring
structure. An electric field is generated where the waveguide lies. The refrac-
tive index is modified as well as the ring resonant wavelength. This produces
signal modulation of the transmitted light. Carrier injection is obtained for-
ward biasing the junction, while to sweep out electron-hole pairs faster and
achieve higher modulation speed, the junction is reverse biased. High speed
modulation can be achieved when optical transmission reaches its maximum
before the electrical junction reaches steady-state and optical rise time is
shorter than electrical rise time.
A milestone paper on ring performance is a Nature paper by Xu [12].
Their research team fabricated a 12pm diameter ring (Figure 1.6) with qual-
1.4 Modulator Optical Structure Design

-----------------
Output I

-- -- -- - Waeguile
,%
""-.
• ....
• !i,.........
...
. ...

M+IC

Figure 1.6: Top view of a ring mod-

ulator coupled to a waveguide. In-
set shows the ring modulator cross-
-------
L-------.--- -I . ____ _ _
Rin - section wit h thne embedded p-i-n
input junction [12].

ity factor of 39,350 and a cavity photon lifetime of 33ps. Static measurements

show 15dB extinction ratio with a forward bias of 0.94V and reverse bias of
-0.3V. Device speed is 0.4Gbit/s when a non-return to zero input signal is
used and a peak to peak voltage (Vp) of 3.3Vp (-1.85V, +1.45V); if a re-
turn to zero signal is used, then the speed is 1.5Gbit/s with 6.9Vpp (-2.8V,
+4.1V). In this latter case, 200ps rise time and 150ps fall time are measured.
Of the same research group, further experimental results are presented by
Preble [13]. A ring of the same diameter (12pm) is used under reverse bias
conditions to show that by using a pump wavelength of 1528nm and a probe
wavelength of 1559nm, 6.8dB extinction ratio can be achieved with -10V bias
and 19pJ pump energy. This device is predicted to reach 5GBit/s with 2.2pJ
pump energy. Both Xu [12] and Preble [13] work with an embedded p-i-n
junction that does not fully encompass the ring. Speed is limited by the
1.4 Modulator Optical Structure Design

difficulty of carrier extraction in the region of the ring in the proximity of

the waveguide where there is no junction.
The next ring generation presented by the same group, Xu [14] improves
the junction encompassing the ring: a nearly closed circular junction is
formed around the ring with an additional n+ doping added outside the
waveguide region, allowing for efficient carrier extraction under reverse bias.
Furthermore, Xu proves the feasibility for wavelength division multiplexing
applications. Four cascaded rings are fabricated on the same waveguide.
Radiuses are 4.98pm, 5.00pm, 5.02pm, 5.04pm and the achieved extinction
ratios are 13dB for the first two rings and less than 10dB for the second two.
Modulation speed is of the order of 5GBit/s, while it is theoretically claimed
it should reach 10GBit/s.
According to Xu [15], a forward biased, carrier injection based p-i-n ring
modulator can achieve high bit rates (>10OGbit/s) only if the trade-off be-
tween the optical rise time and charge storage time is broken. The device
used to prove this concept is the same as in [14]. Applying a high forward
bias for a time shorter than the time needed for charge saturation to occur
and then decreasing it to a lower value, allows to have a short rise time
and a short charge storage time as less carriers are injected in the intrinsic
region. Steady state charge and charge storage time depend on the lower
voltage while the optical rise time depends on the higher voltage. If high
forward bias pulses of -4V are applied to each transition edge of the pseudo-
random 12.5GHz bit sequence used to probe the modulator with non-return
to zero signal (Vp p-8V), 12.5Gbit/s modulation with -9dB extinction ratio
1.4 Modulator Optical Structure Design

is achieved. Another advantage of such a biasing scheme is that, ultimately,

power consumption is also reduced thanks to the lower carrier injection into
the junction.
Manipatruni models in [16] a high field transport device relying on a novel
doping profile device based on a symmetric p-i-n-i-p scheme: two mirroring
p-i-n diodes (Figures 1.7(a) and 1.7(b)), where one ridge is used as optical
waveguide and the other one as a "charge reciprocating ring". The unipolar
behavior of the design allows for the decoupling of the time response of
electrons and holes, avoiding jitter effects (due to electron and hole different
recombination times). As a forward bias is applied to one junction, the charge
density needed to reverse bias the second p-i-n junction limits the injected
charge. Instead, when a reverse bias is applied onto the first junction, the
first junction will be reverse biased and it will inject carriers into the second
junction accordingly. The simulated device has dimensions <10pm. If a
Q=5,000 is assumed, as well as an insertion loss of 3dB, and a critical coupling
condition ring loss of 8dB/cm, under an ac bias with 5Vpp, the extinction
ratio is 12dB with modulation speed 40Gbit/s (NRZ). Rise and fall times are
10ps and 15ps, respectively. The estimated power consumption of the device
is of 2.25fJ/bit/device length.

Metal-Oxide-Semiconductor Configuration Barrios [17] (and Lip-

son [18]), simulates accumulation, depletion and inversion regimes of a MOS
ring structure with 14pm radius. Only the first two regimes are considered
adequate for optical device performance. The device reaches 5.75dB extinc-
1.4 Modulator Optical Structure Design 23

I. *1
250
nm

600 nm i 450nm :600 nrbI*rib~on0 >nm

5

r3OO
nn :300 nn
Intrinsic silicon E Pdoped silicon E Ndoped silicon []silcon dioxide flMetal contacts

(a)

Waveguide - Intrinsic silicon

Thin slab -
P doped silicon
Charge reciproating ring -ring N doped silicon
SLightly Pdoped silicon

Figure 1.7: Ring modulator with an embedded p-i-n-i-p junction structure.

(a) p-i-n-p junction cross-section. (b) Complete device top view: p-i-n-i-p
junction embedded in the double-ridge ring modulator. Only the outer ring
supports an optical mode. [16]

tion ratio and 59% transmittivity in the accumulation regime with Vgate of OV
in the "off" state and Vgate of 5V in the "on" state. In the depletion regime
again with OV of Vgate("off" state) and -5V Vgate("on" state), transmittivity

reaches 23%.

1.4.2 Mach-Zehnder Structure

A Mach-Zehnder interferometer is composed of a waveguide branching out

in two arms that then reunify (Figure 1.8). If a phase modulator is built in
in one arm (or in both), the wave passing through one arm experiences a
phase lag with respect to the other arm; interference results upon waveguide
1.4 Modulator Optical Structure Design

Phase shifter
NA

ight

Figure 1.8: Mach-Zehnder struc-

LO ture showing phase shifters in the
'\Phs two arms, [23].

reunification. Destructive interference happens if the relative phase shift of

the waves traveling in the two arms varies by exactly 7r radians. Phase mod-
ulation is translated to optical intensity modulation by switching between
constructive and destructive interference; the transmitted intensity can be
varied between "Os" and "ls".
A Mach-Zehnder device is a single-pass device. This automatically implies
that a longer device is required to induce the significant modulation depth.
The small effective index variation achievable in Silicon requires longer device
length to induce an appreciable amount of modulation. As a general rule, the
longer the arm of the modulator, the lower the required applied bias. Silicon
based applications at 1550nm normally require device lengths of the order of
millimeters. Despite device dimensions, the main advantage of this structure
with respect to ring resonators and Fabry-Perot cavities is that is has low
sensitivity to temperature variations and broader operation wavelength.

All optical A 2mm long Mach-Zehnder modulator with either balanced

or unbalanced arms is simulated by Almeida with pump and probe wave-
lengths of -1550nm [19]. The device relies on the plasma dispersion effect
and the free carrier lifetime is estimated to be 1.4ns with a modulation time
1.4 Modulator Optical Structure Design

Traveling-wave electrodes

Figure 1.9: Silicon modulator

based on a Mach-Zehnder inter-
ferometer with a pn junction and
traveling wave electrodes. 20dB
extinction ratio, f3dB -20GHz,
data rate: 30Gbit/s, [20].

of 20ps; no modulation depth is reported. The same device is also fabri-

cated: arm length of the balanced arm design is 4mm; its modulation time
is 64ps. Slightly shorter is the device with unbalanced arms, 3mm with 51ps
modulation time. Both devices show less than 5dB extinction ratio.

p-n and p-i-n Configurations A pn junction high speed Silicon mod-

ulator based on the carrier depletion effect is designed by Liu in [20], [21].
It is a single mode device working at 1550nm. As Figure 1.9 depicts, the
p-doped Silicon rib waveguide has an n-doped Silicon cap layer. Rib width
and height dimensions are 0.6-0.5pm 2 . While the p-doping profile is uniform
(-1.5-10 17cm-3), the n-profile is graded with lowest concentration at the pn
interface (from -1.5.10 17cm - 3 to -3.10 18 cm- 3 ). Traveling wave electrodes
are used to decrease the RC limitation. This device has -20dB extinction
ratio with an insertion loss of -7dB. The V,-L figure of merit is -4V.cm.
Using a 1mm long phase shifter per arm, 6.6Vp ac signal, 3.5V dc bias, the
roll-off bandwidth measures '20GHz and data transmission rate is 30Gbit/s.
An ultra-compact Mach-Zehnder interferometer (<200pm long) with a
p-i-n embedded junction on an SOI substrate is proposed by Green in [22].
1.4 Modulator Optical Structure Design

The waveguide has width and height of 550-220nm 2 and it lies on a 35nm
Silicon layer that acts as a conductive path between the intrinsic waveguide
and the heavily doped p+ and n+ regions (both having doping concentration
of _-11020 cm- 3 ). Thanks to the waveguide dimensions, both photons and
injected free carriers are simultaneously confined in the intrinsic region of
the junction. The overlap between the optical mode and the free carriers
ensures large modulation of the modal effective index. The small waveguide
dimensions also oblige large injected free carrier changes to take place mainly
inside the small waveguide core cross-section. At A=1550nm, 16GHz RC fre-
quency cut-off is measured (200fF reverse bias capacitance). Metal contacts
unintentionally placed cause 12dB of on-chip loss which in turn limits the
extinction ratio to 6-10dB. Under these conditions and with a V,=1.8V, the
figure of merit is V,.L=0.36V.mm. V,-L is given by a 200pm long phase
shifter and a An~4-10 - 3 modal effective index change, and an injection of
free electron-hole pair density of -1.5.10 18 cm - 3 . The high frequency behav-
ior is investigated with a 1.2Vpp non-return to zero pattern, and 3.5V peak
amplitude pre-emphasis pulses. When a bias of 0.3V is applied to a 100pm
long phase shifter operating at 5Gbit/s, a dc current of 2.17mA is gener-
ated corresponding to a consumption of dc power of 230uW (due to 49Q
forward resistance) and of an RF power of 41mW. The same measurement
is repeated with no applied bias on a 200pm long phase shifter at 10Gbit/s.
Results are: 287mA dc current corresponding to 287pW and 51mW dc and
RF power consumptions, respectively. The RF power consumption requires
an energy/bit equal to 5pJ/bit.
1.4 Modulator Optical Structure Design

Mletal contact
I I
MtPocy-Si
Oxic
L 0.9 Anm
'11
1.4 tpm
I Xc10

Y- Ixf Buried
oxide

Figure 1.10: MOS capacitor cross- Figure 1.11: MOS capacitor

section from [23]. Extinction ratio of cross-section from [26]. Extinc-
16dB is reached with 7.7Vp, 1GHz tion ratio of 3.8dB is reached with
bandwidth, modulation speed 4Gbit/s. 1.4Vp, bandwidth 10GHz, and
10Gbit/s modulation speed.

Metal-Oxide-Semiconductor Configuration The advantage of a Mach-

Zehnder with a MOS configuration under accumulation regime is that is is

a majority carrier device. The device bandwidth is not limited by carrier
recombination as it is the case for p-i-n junctions.
A Mach-Zehnder interferometer with a MOS structure embedded in a
Silicon rib waveguide (width-rib height: 2.5-2.3 ptm 2 ) is presented in [23], [25],
and [24]. A cross-section of the MOS structure is pictured in Figure 1.10. A
16 cm -3 )
potential drop (applied forward bias, VD) across the n-doped (1.10
Silicon layer and the p-doped (-5.10 16 cm - 3 ) poly-crystalline Silicon layer
forms two charge accumulation layers near the gate oxide: an electron charge
layer at the n-doped Silicon-oxide interface and a hole charge layer at the p-
doped Silicon oxide interface. The refractive index is modified only in the
charge layers and this ultimately triggers an optical absorption change.
Modeled and measured results of an individual phase shifter reveal linear
1.4 Modulator Optical Structure Design 28

proportionality both between the driving voltage and the experienced phase
shift, and between the phase shift and the phase shifter length. The obtained
figure of merit is: V,.L -8V-cm.
dc measurements are performed with a 10mm asymmetric Mach-Zehnder
phase shifter with one MOS capacitor per arm. At 1540pm, an extinction
ratio of 16dB is measured with a peak-to-peak voltage of -7.7Vpp. Insertion
loss is estimated as ~15.3dB (4.3dB interface coupling loss, 6.7dB on-chip
loss). A Mach-Zehnder with a 2.5mm phase shifter with only one MOS
capacitor is used for ac measurements. The modulation speed is studied by
using either a r.m.s sinusoidal source (0.18V) or a digital pulse pattern (3.5V)
at 1558pm. Enhancement of device sensitivity is obtained by applying a 3V
dc bias. The measured modulation speed exceeds 1GHz (2.5GHz is claimed
in [27]). 1Gbit/s data transmission is demonstrated with a 3V dc bias and a
3.5V pulse pattern and a pseudo-random electrical data input.
Improvements suggested by Liu [23] are taken into account in [26]. The
poly-crystalline Silicon is replaced by crystalline Silicon grown by lateral epi-
taxial overgrowth (ELO), thus reducing optical losses. As phase modulation
efficiency depends both on the waveguide size and charge layer thickness,
waveguide dimensions are down scaled (width-rib height: 1.6-1.6 pm 2 ) as
well as the gate oxide thickness (from 12 nm to 10.5nm); doping concentra-
tions are increased to achieve higher bandwidth (n-doping -2.10 17 cm - 3 and
p-doping 1.10 18 cm- 3 ). V..L becomes 3.3V.cm and a bandwidth exceeding
10GHz is later claimed by the same author in [27].
High speed measurements are obtained by applying -3.3V dc bias to n-
1.4 Modulator Optical Structure Design

doped Silicon slab, a varying dc bias to the low speed section of the modulator
arm (to reach quadrature), and 1.4V ac pulses to the p-Silicon. Data trans-
mission is as high as 10Gbit/s with 3.8dB extinction ratio. These values are
still limited by the drive circuitry.
Theoretical improvements are suggested by Liao in [27]: modeling the
transient response of the phase modulator, 10GHZ bandwidth is predicted
with a graded doping profile. Physical length details of this structure is given
in Table 1.2 as well as a comparison with the previous two MOS devices ([23],
[26]).
1.4 Modulator Optical Structure Design 30

Liu [23] Liao [27] Liao [26] Liao [27]

Publication Year 2004 2006 2005 2006
Result Type Experimental Experimental Modeling
Material c-Si(n); poly-Si(p) c-Si(n); ELO-Si(p)
Waveguide Dimensions 2.5-2.3pm 2 1.6-1.6pm 2 1. 1ym 2
Gate Oxide Thickness 12nm 10.5nm 6nm
n-Doping 1.7.1016cm 3 2.1017cm - Graded
p-Doping 5.10 6 cm - 3 1-10 18cm - 3 Graded
Vj.L, ~8V-cm 7.8V-cm 3.3V-cm
Transmission Loss/L, Not Specified 10dB/L, 9dB/L,
Frequency Response:
Phase Modulator Length 10mm 2.5mm 15mm
3dB Bandwidth ~1GHz >2.5GHz 10.2GHz >10GHz
On-chip Loss 6.7dB 4dB 10dB 2dB
Total Insertion Loss -15.3dB Not Specified
Data Transmission Performance:
Total Device Length Not Specified 15mm 15mm
Arm Length Not Specified 13mm 13mm
Phase Modulator Length 2.5mm 3.45mm 3.45mm
Low-Speed Section Length NA 4.75mm 4.75mm
Extinction Ratio 16dB 1.3dB 4.2dB(3.8dB) -
dc Bias on n-Si 3V -3.3V -3.3V -
ac Bias on Phase Modulator 3.5V 1.4V 1.4V -
Bias on Low-Speed Section NA Not Specified Not Specified -
Rise Time Not Specified 82ps 57ps(55ps) -
Bit Rate 1Gbit/s 4Gbit/s 6Gbit/s(10Gbit/s) -

Table 1.2: Intel's MOS capacitor Mach-Zehnder structure evolution. The

second (Liu [23]) and third column (Liao [27]) are two different descriptions
of the same device. Fourth and fifth represent experimental (Liao [26]) and
modeling (Liao [27]) improvements. (NA = not applicable)
1.4 Modulator Optical Structure Design

1.4.3 Fabry-P rot Interferometers

All Optical Modulation A fabry-Perot structure relying on all optical

modulation is demonstrated in [28]. The input optical signal is modulated
by variation of the optical pumping source. The cavity acts as an amplifier,
hence allowing for a non-local optical power source.

p-i-n Configuration A free carrier dispersion effect based Fabry-Perot

cavity with distributed Bragg reflector microcavities is simulated by Barrios
[29]. The device length is 20pm and it is optimized to work at 1550nm.
It is a single mode device with optical confinement ensured by distributed
Bragg reflector microcavities and the Fabry-Perot cavity itself, and with
electrical confinement ensured by a side trench etched p-i-n structure. Static
response of the device is characterized by 6.9dB extinction ratio and 87%
transmittivity obtained with a bias of 0.87V, an optical input power of 25pW,
and 116A/cm 2 drive current. The dynamic response is simulated with 300ns
pulses for both the "on" state (0.87V) and the "off" state (-5V). Calculated
rise time is 14ns, 1.25ns fall time, and 16ns switching time. Simulations show
no significant thermo-optic effect.
An improved similar structure is modeled in [30]. Again based on the
free carrier dispersion effect, this time it is a straight rib waveguide intensity
modulator with a one dimensional photonic crystal structure embedded in a
cavity 10pm long. Static response shows theoretical extinction ratio of 6.9dB
achieved with 0.87V and an input optical power of 1.53/pW/ipm and a driving
current of 1.67ptA/utm. The dynamic response is measured as in the previous
1.4 Modulator Optical Structure Design 32

z
Wt, wIb
Vout
9,-0d• U--t *-

(h.rh X

i;
wol ugimn

Figure 1.12: p-i-n Fabry-P6rot Figure 1.13: Improved design of

microcavity with distributed the p-i-n Fabry-Perot microcav-
Bragg reflectors. 20p long, 6.9dB ity. 10pm long, 6.9dB extinction
extinction ratio with 0.87V, [29] ratio with 0.87V, [30]

device: 300ns pulses for both "on" state and "off" state voltages of 0.87V and
-5V, respectively. The rise time is 1.11ns and the fall time is 0.18ns, while
the switching time is 1.3ns. The major achievements of this newly proposed
structure are the switching time improvement and the dimension shrinkage.
Schmidt presents in [31] an ultra-small Fabry-Perot cavity embedded in
a high index contrast Si/SiO 2 waveguide with five periodic holes on either
side of the cavity forming a distributed Bragg reflector. Holes have 220nm
diameter and 420nm spacing. The cavity is 2.51pm long, while the total
device length including the holes is <6p/m. Based on free carrier dispersion,
the cavity is the intrinsic layer of a p-i-n junction and the p and n lie on
the sides of the cavity, as depicted in Figure 1.15. With 5.6V dc bias, a
modulation depth of 5.93dB is achieved at A=1568nm. The insertion loss
is 5.1dB, coupling and waveguide losses combined are 23.5dB. Qexperimental
measures - 253, while Qtheoretical -780. Such a difference is due to sidewall

scattering and device design sensitivity to fabrication. When 0.7V dc bias

is applied in conjunction with an ac swing of 2.0Vp,, the extinction ratio
 1.5 Summary 33
1.5 Summary 33

""

Figure 1.14: 3-dimensional full Figure 1.15: Fabry-Perot microcav-

structure of the improved design ity (<6pm) with distributed Bragg
of the p-i-n Fabry-Perot microcav- reflector structure and embedded p-
ity with distributed Bragg reflectors. i-n junction. Low frequency mea-
10pm long, 6.9dB extinction ratio surements show 5.93dB modulation
with 0.87V, [30] depth with 5.6V dc bias; max-
imum data transmission speed is
250Mbit/s with 5.87dB extinction
ratio with 0.7V(dc) and 2Vpp(ac),
[31].
is 5.87dB and the transmission speed is 250Mbit/s. Rise and fall times are
2.27ns and 1.96ns, respectively. It is claimed that if Qexperimental could equal

Qtheoreticat, the modulation speed would be -1Gbit/s, with a dc bias of only

0.3V.

1.5 Summary

As modulators are becoming key building blocks in the electronic-photonic

integration, the need to fulfill small footprint, low driving voltage, large mod-
ulation depth and high speed requirements becomes fundamental. This first
1.5 Summary

chapter summarizes light modulation physical principles. Electro-optic, car-

rier injection and depletion, thermo-optic and all-optical mechanisms are
reviewed. These mechanisms are then combined to optical structures. Non-
biased devices as well as p-i-n structures or MOS configurations are explored
through a partial literature review of Silicon based devices operating operat-
ing at 1550nm.
Ring modulators, Mach-Zehnder interferometers and Fabry-Perot cavities
are the three main structures used for Silicon based modulators operating at
1550nm. The key advantage of ring modulators is their small footprint, but
they are highly temperature sensitive and they are not broad band devices.
Instead, Mach-Zehnder devices have larger wavelength range of operation;
they are tolerant to small temperature differences but they have large foot-
print. Published results on Fabry-Perot based modulators always exploit
the use of photonic cavities which are highly sensitive to fabrication and
temperature variations.
Chapter 2

Device Physics, Material's

Choice, Design, and Fabrication

Key guiding parameter to keep in mind while reading this chapter is the figure
of merit of the device being studied. The optimization of such parameter,
Ac/an,, (i.e. the ratio between the modulation depth and the insertion loss)
is of key importance while considering the physical mechanism leading to
modulation and the choice of the material of the device.
This chapter explains the theoretical background and the alloy compo-
sition choice on which the final working device relies. The modulator butt-
coupled to high index contrast waveguides and based on the Franz-Keldysh
effect to produce modulation is a novel device. Theoretical calculations of

~ butt-coupled modulator to Si/SiO 2 waveguides predict 10dB

this SiGel_
modulation depth with a 3dB frequency higher than 100GHz. The only ex-
pected speed limitation is the RC delay: in fact, the Franz-Keldysh effect
operates in the picosecond time frame.
2.1 Material's Choice Overview

(al E lb E
'-I

I,
L
Figure 2.1: Elemental Ger-
k manium band gap diagram.
(a) Bulk Germanium; (b) ten-
sile strained intrinsic Germa-
nium: the direct bad gap (E r )
dpcrp~.ass ulndcr t-hp fF~ef.t nf
bulk Ge tensile strained i-Ge tensile strain, [32].

2.1 Material's Choice Overview

In unstrained, crystalline Silicon, the linear electro-optic effect (Pockels ef-

fects) is nonexistent; the non-linear electro-optic effect (Kerr) and the electro-
absorption effects (Franz-Keldysh) are weak. Moreover, its direct band gaps
lie in the range 3-4eV. Not having all these challenges, Silicon would be per-
fect for monolithic integration on a CMOS platform because it would enable
the integration of photonic devices with microelectronics.
Bulk Germanium, with its direct band gap value (0.8eV) and its higher
electron (3900 cm 2 /s-V) and hole (1900 cm 2 /s-V) mobilities with respect to
Silicon, is a more favorable candidate for faster devices that operate at lower
power.
During epitaxial growth, Germanium is fully relaxed on Silicon; at room
temperature, thermal mismatch generates tensile strain. As a result, the
band gap energy of the material decreases (Figure 2.1) and becomes opti-
mized for the wavelength window of 1633-1650nm, [32]. The addition of a
2.2 Franz-Keldysh Theory 37
2.2 Franz-Keldysh Theory 37

| . ....

Eh hn

Figure 2.2: Schematic diagram Figure 2.3: Franz-Keldysh ef-

representing the Franz-Keldysh fect: absorption coefficient (ca)
effect. The band diagram tilt- versus photon energy (hw). The
ing, due to an applied electric weakly absorbing regime is just
field, generates a finite tunnel- below the band gap value, E,.
ing probability for photon ener-
gies (hw) smaller than the band
gap (Eg) value.

small amount of Silicon increases the Germanium band gap and the direct
band gap (Er) can be tailored to be equivalent to the wavelength of 1550nm.

Enhanced Franz-Keldysh effect in tensile strained SixGel-, epitaxially grown

on Silicon wafers is observed [32]. Depending on the strain, the direct band
gap wavelength range varies between 1550nm and 1610nm; while the indirect
band gap is -1900nm and it is less sensitive to strain.

2.2 Franz-Keldysh Theory

The absorption coefficient depends on the applied electric field and on the
photon energy. As an electric field is applied, the conduction and valence
2.2 Franz-Keldysh Theory

bands tilt. Their respective evanescent wave function tails extend into the
band gap generating a finite probability of tunneling (Figure 2.2).
For photon energies with hw < E r (E, being the direct band gap value),
the absorption coefficient increases with the applied field as the band tilt
steepens. While for photon energies with hw > E'r the absorption coeffi-
cient oscillates depending on the constructive or destructive interference of
the wave functions of the conduction band with the one of the valence band.
These Franz-Keldysh oscillations are caused by the overlap of the wave func-
tions inside the conduction and valence bands under a specific electric field
and varying photon energies as plotted in Figure 2.3.
Referring to the figure of merit of the electro-absorption modulator being
optimized, in the weakly absorbing regime, when the photon energy hw is
only slightly smaller than the value of the direct band gap E,, the insertion
loss is reduced because the indirect gap has only a weak absorption contri-
bution. The modulation depth, Aa, is maximized if hw is close to the band
edge (- Eg). For this reason, photon energies slightly below the direct band
gap are expected to optimize the figure of merit.
The key idea of the Franz-Keldysh effect is the addition of a uniform elec-
tric field term in the time-independent Schrodinger equation for an electron
hole pair. A few approximations can be performed: first, the uniformity of
the field across the material is assumed thanks to the geometry and dimen-
sion of the device being studied. Coulombic interactions of the electron hole
pair can be neglected as the electric field effect dominates on the exciton
effects. This is true only in the case of photon energies smaller than the
2.3 SiGel_- Alloy

band gap value, and it is indeed the photon energy range of interest since
the Franz-Keldysh effect of the weakly absorbing regime is the one exploited
in these electro-absorption modulators. Another important approximation
while solving the Schr6dinger equation is the assumption that only electrons
and holes that reside near the conduction and valence band edges take part
in the band to band transitions.
The solution to the Schrodinger equation written with these approxima-
tions, the total wave function, is then inserted in the absorption coefficient,
a(w), derived from Fermi's Golden Rule. The applied electric field directly
modifies the absorption coefficient of the material.
A detailed mathematical formulation of the solution to the Schr6dinger
equation and absorption coefficient calculation can be found in Chuang [35]
and summarized in Liu [36].

2.3 SixGel_- Alloy

As previously mentioned, tensile strained Germanium has a band gap equiv-

alent to the wavelength range 1633-1650nm depending on the strain level.
The addition of Silicon in forming SixGel_- alloy offers the freedom of band
gap adjustment to a predetermined value. In the case of the SiGelx alloy,
as the applied electric field tilts the band structure, both the direct and the
indirect band gap values are affected. By increasing the Silicon content, the
direct and indirect bands get further apart and the material tends to behave
as indirect deteriorating its optical properties.
2.3 SixGel_- Alloy 40

2.3.1 Absorption Coefficient Optimization

The theoretical modeling of the electro-absorption properties of the material

is done using the generalized Franz-Keldysh theory proposed by Shen and
Pollak in [33]. As the Franz-Keldysh effect is found to be strongest in the
direct band gap, the indirect band gap contribution is neglected [34].
Their proposed theory requires the a priori knowledge of four material
parameters: the direct band gap value, effective electron and hole masses,
the real part of the refractive index, and the optical transition matrix element.
The four material parameter are evaluated as follows:

1. The direct band gap value depends on the Silicon concentration in the
alloy and on the thermally induced strain.

Direct Band gap Calculation with no Tensile Strain:

Theoretical absorption calculations in the weakly absorbing regime
are performed by first understanding the approximate Silicon concen-
tration needed, then including strain effects in the calculations.

When it is assumed there is no strain, the SixGel_- direct energy gap

can be inferred by linear interpolation between the pure Silicon (4.06eV
at room temperature) and the pure Germanium with the following [38]:

E! (SixGexi) - 0.80 + 3.26 - x

According to this formula a Silicon dilute concentration of <2% is re-

quired to achieve the targeted wavelength of 1550nm. In the same way,
2.3 SiGel_- Alloy

?·1
eV

1.0

0.9

0.7 Figure 2.4: SixGel-x indirect

bad gap dependence on Silicon
0.6
concentration. For x<15% the
x relation is linear, [40].

linear interpolation in the regime 0-15% (Figure 2.4) can be used to

calculate the indirect SizGel_- band gap, [39]:

EL(SixGeix) = 0.66 + 1.2 -x

Band gap Calculation including Tensile Strain:

The first consideration regards the shape of the device. While a blan-
ket layer of SixGe_-x grown on a Silicon substrate would have isotropic
in-plane strain, due to the trench-growth, elongated shape of the device,
it is found that the material is fully relaxed in the transverse direction
and it is strained in the longitudinal one (i.e. the light propagation di-
rection) [36]. Trench growth is used for two reasons. First it improves
coupling losses. Secondly, a single-mode device is obtained only with
a cross-section smaller than 1-1pom2 and device length of the order of
tens of microns to ensure high modulation depth.
2.3 SixGelx Alloy

Experimental and theoretical selective trench growth details with

Borgstrom and Wulff construction as well as strain explanation and
detailed calculations can be found in Ref. [36].

Once the tensile strain is determined, formulae for the light hole, heavy
hole and split off band gaps can be found in the literature [37]. They
imply the knowledge of deformation potentials and elastic constants,
which can be determined by linear interpolation between Silicon and
Germanium.

2. The refractive index can also be found by linear interpolation according

to [41]:
n(SixGel_x) = 4.10 - 0.64 -x

3. Effective electron and hole masses of pure Germanium are used, as the
SiýGel_x ones with x<0.02 are experimentally undistinguishable from
pure Germanium [42].

4. The transition matrix element for the direct band gap of SixGelx is
extrapolated from k-p theory.

Once these four material parameters are numerically defined, the Franz-
Keldysh effect can be modeled and optimized with respect to Silicon concen-
tration.

Simulation iterations with increasing Silicon (0.5%-1.1%) concentration

show that the band gap shifts from 1570nm to 1530nm [36]. As Silicon
is added, the material starts behaving as if it had an indirect band gap,
2.4 Device Design 43

a-=(100kV/cm)~(1okV/cm)
3.-4 Ai
3.2
3.0 ,..-. nm
,-2.8

S2.6

2.0-
18-
1.6J Figure 2.5: Figure of merit
-4 optimization with respect
0.5 0.6 0.7 0.8 0.9 1.0 1.1 to Silicon content in the
Si content (%) SixGelx alloy [36].

consequently deteriorating its optical properties. The physical reason is that

the Franz-Keldysh effect due to the direct band gap becomes insignificant,
while the one due to the indirect band gap absorption prevails.
As is seen in Figure 2.5, at the optical "off" state with an applied electric
field of 100kV/cm, the figure of merit •a/a,,o at 1550nm is maximized for

Silicon concentrations in the range 0.7-0.8%, where the value reaches -3.0.
This translates to an extinction ratio of 12dB, assuming an insertion loss of
4dB. The maximum device length allowed in order to achieve an insertion
loss <4dB is 70pm and it is found by studying the absorption coefficient as
a function of applied electric field in the range 0-130kV/cm [36].

2.4 Device Design

The device experimentally analyzed in the next chapter is designed by Liu

[36]. The following is intended to highlight the main points of the design and
fabrication.
2.4 Device Design

X Z
AICu Metal ICu Metal

n+ poly-Silicon VContacts

SiGe Modulator
)+ c-Silicon on So1

Figure 2.6: 3-D representation of the SixGelx modulator, the two Silicon
layers (n+ and p+ regions of the p-i-n junction), and electrodes (W contacts
and A1Ti metal layers). Silicon waveguides (not pictured) couple in and out
directly from the SixGel_,. Design not in scale.

The complete design includes the SiGel_- modulator as well as the

waveguides used to couple light in and out of the device. The novelty of
the structure relies on the butt-coupling scheme. A schematic representation
of the structure is in Figure 2.6 and the waveguide-modulator coupling is
shown in Figure 2.7. Tapered crystalline Silicon waveguides are evanescently
coupled to tapered amorphous Silicon waveguides. The amorphous Silicon
waveguides butt-couple light into the modulator.
The key idea of the butt-coupling design is that the center of the input

(output) waveguide terminates (begins) in the center of the SiGel_- cross-

section perpendicular to the light propagation direction.
Waveguide and modulator designs will be considered first; attention will
then be focused on the complete structure in order to maximize the figure of
merit.
2.4 Device Design 45

Figure 2.7: 2-D longitudinal

cross-section of the butt-
coupled Si.Gexi modulator
with a-Silicon waveguides.
Tapered waveguides are used
a-Si WG SiGe Modulator to enhance coupling efficiency.
=== c-Si WG Design not in scale.

500nm

c-Silicon 2 00 n m Figure 2.8: High-index contrast

Si/SiO 2 waveguide cross-section.
Core dimensions used are indi-
Oxide
cated in the figure. Figure not in
scale.

2.4.1 Si/SiO 2 Waveguide

A single mode Si/SiO 2 high index contrast waveguide (Silicon refractive in-
dex: 3.48; SiO2 index: 1.46) is used as bus to couple light into the modu-
lator. The width of the Silicon core is 500nm, and its thickness is 200nm.
These dimensions ensure single mode operation. An oxide overcladding on
the waveguide is assumed for simulation purposes and light leakage to the
substrate is avoided thanks to the incorporation of a thick oxide layer (3/tm)
underneath the Silicon layer. A waveguide cross-section is shown in Figure
2.8.
Both TE (where the electric field polarization is parallel to the substrate)
and TM (where the electric field is perpendicular to the substrate) modes
are evaluated. A higher confinement factor is found in the case of TE polar-
2.4 Device Design

Si/SiO 2 Waveguide Parameters (1550nm) TE TM

Effective Index 2.443 1.737
Confinement Factor 0.825 0.524
Bending Loss (2p4m Bending Radius) 2.1dB/cm 1346dB/cm

Table 2.1: Si/SiO 2 waveguide calculation results at 1550nm after [36]

ization, which ensures lower bending losses and hence higher on-chip packing
densities. For the purpose of on-chip integration this concept is of funda-
mental importance and hence the device structure will be optimized for TE
polarization. Table 2.1 shows a comparison between simulated parameters
for TE and TM fundamental modes.

2.4.2 SizGel-, (x = 0.75%) Electro-absorption Modu-

lator

The following structure is considered for simulation purposes (Figure 2.9):

a 200nm thick crystalline Silicon layer with 2-10 19 /cm 3 p+ doping concen-
tration as undercladding; a 400nm high, and 600nm wide SiGelx intrinsic
3
region; and a polycrystalline Silicon top cladding 200nm high with 2-10 19 /cm
n+ doping concentration. As in the case of the waveguide, leakage to the
substrate is prevented with a 3 pm SiO 2 undercladding. Maximum device
length is 70pm, to obtain an insertion loss lower than 4dB. Table 2.2 shows
simulation results of such a device.
2.4 Device Design

SixGel_, Parameters (1550nm) Optical State TE TM

Effective Index 3.71 3.728
Confinement Factor 0.886 0.881
Imaginary Effective Index "on" state, OV 0.00210 0.00210
Modal Absorption Coefficient "on" state, OV 170/cm 170/cm
Imaginary Effective Index "off" state, -3.3V 0.00767 0.00733
Modal Absorption Coefficient "off" state, -3.3V 622/cm 594/cm

Table 2.2: Optical mode evaluation. TE versus TM fundamental modes in

SixGel_, device. Data after [36]

200nm n+ poly-Si
400nm SiGe
- 200nm p+ c-Si
- 3/m SiO 2 Figure 2.9: SixGel_, electro-
1
Czi n
ociliS

~hCCL~:4
Substrate
absorption modulator cross-
section. Design not in scale.

2.4.3 Figure of Merit Optimization

At the beginning of this chapter, AAc/cor, i.e. the ratio between the mod-
ulation depth and the insertion loss, is mentioned. In terms of material
parameters this expression becomes:

ExtintitionRatio _ absorption(V # 0) - Kabsorption(V = 0)

InsertionLoss Iabsortpion(V = 0) + 0co7uplng

All these parameters are here evaluated while the design device is being
optimized.

Insertion Loss The insertion loss is given by the sum of the absorption
loss (labsorption) at the optical "on" state and the coupling loss (lcoupling):
2.4 Device Design

Absorption Loss It depends on the SixGelx modal absorption coef-

ficient when no bias is applied, a,,mod(V=0):

Absorption Loss = absorption(V = 0) -10-log e-amod(O)L},

where L is the modulator length. In turn, the modal absorption coefficient is

related to the absorption coefficient of the Si.Gel_- and neighboring Silicon
cladding layers and on the amount of optical power in them.

Coupling Loss Total coupling loss is given by modal dispersion (in

the direction perpendicular to the direction of propagation), and reflection
losses (in the propagation direction) between the Silicon waveguide and the
SixGel-, device.
Coupling losses between the waveguide and the modulator and back to the
waveguide can be decreased with the butt-coupling design. The mode overlap
is enhanced when the waveguide and the modulator are butt-coupled but it is
also a function of device dimensions. The overlap can be studied as a function
of device height while keeping the SixGel_- width fixed (600nm). Lower mode
mismatch is found for thinner SixGel_x thickness, which leads, consequently,
to higher coupling efficiency. On the other hand, the confinement factor is
also a function of device height: it increases as the device height becomes
thicker. The intersection of these two trends (Figure 2.10) leads to a trade-
off between an optimized confinement factor in the SixGel_x layer and a
maximized mode overlap between the Si/SiO 2 and the SixGelx.
While considering the overall device performance, general rules are: as the
mode overlap improves (i.e. thinner SixGel_- layer), the confinement factor
2.4 Device Design 49

(c)
I,/

0.95. IL 0.9
0
0.90-
0.8 U_
- 0.85-
a)
0 0.80 i 0.7 E
W=600 nm a Figure 2.10: Trade-off between
*
S the modal overlap in the Silicon
S0.75 0.6
- 200 300 400 500 600 0 waveguide and the confinement
GeSi thickness H (nm) factor in the SixGel_- layer [36].

worsens. Since only the light that is physically inside the SiGel_, layer
can be modulated, a higher confinement factor is desirable. Also, as the
SiGel_- layer becomes thinner, at the optical "on" state (OV) the built-in
electric field increases, drastically decreasing the absorption coefficient. The
absorption coefficients of the optical "on" and "off" states become less far
apart. Therefore, the achievable modulation depth decreases. A comparison
between a 600nm and a 200nm thick layer is reported in Table 2.3. The
estimated total coupling loss is as low as -IdB.

Extinction Ratio It is given by the difference between the absorption loss

of the SiGel_, as a bias is applied to the device, Iabsorption(V f 0), and the

absorption coefficient when no bias is applied, labsorption(V = 0):

Extinction Ratio = Aatmod = Kabsorption(V # 0) - 6absorption(V = 0).

Scanning the SixGel_

1 thickness while keeping the width constant, for
a device length of 50/m, an optimal thickness is found. This thickness is
400nm, independently of the device width, as shown in Figure 2.11.
2.4 Device Design 50
2.4 Device Design 50

SixGel-x Thickness 600nm 200nm

Confinement Factor 0.926 0.625
Built-in Electric Field, "on" state (OV) 7.5kV/cm 22.5kV/cm
Absorption Coefficient 140/cm 254/cm
Aa/con 3.5 1.5
Reflection Loss 0.2dB 0.13dB
Modal Loss 0.9dB 0.43dB
Total Coupling Losses 1.1dB 0.56dB

Table 2.3: Parameter dependence on SixGel_, thickness. Parameters calcu-

lated at 1550nm. After [36].

Z.2-

2.0.
• 1.8-
. 1.6.

1.2. Figure 2.11: Figure of merit op-

1.0.
-0- w=oo nm timization with respect to the
--- W=700 nm I
i Opm SiGel_, modulator layer thick-
0.8- --- W-800 nm
ness; various SiGelx widths
200 300 400 500 600 are considered while length is
Ge0.9925Si 0.07 Thickness H (nm) kept constant (50pim) [36].
2.4 Device Design 51

Insertion Loss (dB)

4 6 8 10 12
IVLU" - .. '. w
N
3 110- 3030 O
S100.
790. 20
80o
C
M 70- lo i
Figure 2.12: Trade-off between
m 604
C) 501 j 3dB bandwidth, extinction ratio
- 1 T_ __ _1__L~ 1_~~~1
20 40 60 80 100 120 140 160 and Sixe-x modulator length
Device Length (pm) [36].

For a thinner device layer, the confinement factor is too small and the
modulation depth is reduced; for a thicker layer, the modal overlap decreases
and the built-in electric field decreases. Another plot of interest (Figure 2.12),
is device length optimization versus both 3dB frequency and extinction ratio.
The 3dB frequency is highest for shorter devices where it can theoretically
reach -100GHz. As the Franz-Keldysh effect takes place in the picosecond
time scale, the only speed limitation would be the RC delay of the device.
A trade-off between device length, which would ensure high extinction ratio,
but also high insertion loss and low 3dB bandwidth is unavoidable.
Concluding remark is that the optimization of the figure of merit of the
electro-absorption modulator highly depends on material properties, con-
finement factors (FSiGe, Fc-Si, Fpoly-Si), absorption coefficients (CSiGe, Oac-Si,

apoly-si) and modulator physical dimensions (height, H, width, W, and length,

L) of the device. The dependence of these parameters is schematically shown

in Table 2.4.
2.5 Device Fabrication

Parameters influencing FOM cSiGe apoly,c-Si SiGe Fpoly,c-Si H W L

aCmod(0), Camod(V) X X x x x x
Modal overlap x x x x
Reflectance x x x

Table 2.4: Modal absorption, modal overlap and reflectance influence extinc-
tion ratio and insertion loss. Crosses indicate the dependence of these three
parameters on the variables influencing the figure of merit. H,W, and L are
height, width, and length of the SixGel_- layer.

2.5 Device Fabrication

The SiGel_. butt-coupled waveguide-modulator structure fabrication pro-

cess is explained with the aid of Figure 2.13-2.19. Fabrication is performed

on a 180nm line at BAE Systems. A cross-section of the final device is shown

in Figure 2.20.
2.5 Device Fabrication

Figure 2.13: The process start with

an SOI substrate. Optical isolation
Oxid- p+ c-Silicon from the substrate is thus ensured.
Oxide The oxide layer is 3pm thick. The
Silicon Substrate e crystalline Silicon layer is doped p+
by ion implantation.

Figure 2.14: The c-Silicon p+

p+ c-Silicon Boron doped layer is patterned and
Oxide mesas are formed. The mesas are
Silicon Substrate the p+ region of the p-i-n junction
of the device.

* Oxide
p+ c-Silicon Figure 2.15: An oxide deposition
Oxide follows. The obtained ondulated
1 e
L22YI C2ZIjIUC
Quilian
I I| I Q kh+fr
+I|*•1-.
layer is planarized with chemical
- -"
-r Itll
"" 1I" f

mechanical polishing (CMP).

Figure 2.16: Amorphous Silicon is

Oxide then deposited and patterned. This
a-Silicon step forms the core of the amor-
- Oxide phous Silicon waveguides. Another
Oxide - p+ c-Silicon oxide deposition and planarization
1 C:iliSrn
o Shusratc
, form the upper cladding of this
.'II1•,1 11I ".J. de .7LI
waveguide.

Oxvirle

coll, I ,U u
iliS
S
b

CALI
aCae
t
t
1 k
a-Silicon
Oxide
p+ c-Silicon
Figure 2.17: Once the waveguide
structure is defined, standard pho-
tolithograpy steps are performed to
open a trench that exposes the p+
c-Silicon mesa.
2.5 Device Fabrication 54
2.5 Device Fabrication 54

SiGe
Oxide Figure 2.18: As the trench is
a-Silicon opened, epitaxial SiGelx is grown
~ Oxide selectively between the input and
Oxide p+ c-Silicon output of the amorphous Silicon
$Rifimn qh--qt The trench is pur-
A
2111vv rate waveguides.
posely overfilled then planarized.

n+ poly-Silicon
SiGe
Figure 2.19: The last step is poly-
Oxide

Oxide
M O
F
a-Silicon
Oxide
p+ c-Silicon
Silicon deposition, n+ Phospho-
rous implantation, then patterning.
Tungsten and A1Ti metal deposi-
tion to form vias and contact pads,
A
L IIV22CHYIZJL
5 ilirrnn
| .|| +
Cllhcmta
•" || I

respectively.

Figure 2.20: Fabricated SixGel_-

device cross-section. The SiGe
modulator is indicated in the mid-
dle of the figure. The bottom oxide
layer, the p+ and n+ Silicon lay-
ers as well as the metal contacts are
indicated.
2.6 Summary

2.6 Summary

This chapter reviews the design of the SiGel-, electro-absorption modulator

butt-coupled to high index contrast Si/SiO 2 waveguides. It starts with a
theoretical review of the Frank-Keldysh effect. This theory is then applied
to evaluate material absorption properties of the alloy used for the design of
the modulator. A SiGel_- alloy with a Silicon concentration of x=0.75% and
with 0.2% tensile strain is chosen in order to optimize the device performance
at 1550nm. Performance optimization with respect to the figure of merit of
the modulator (the ratio between the extinction ratio and the insertion loss)
involves also the physical dimensions of the device. The final design has
height and width of 400nm and 600nm, respectively, while the length ranges
between 30 and 500pm. Last, the fabrication process flow is illustrated.
The device experimentally presented in the next chapter has a length of
50pm. The expected 3dB frequency is -100GHz, with an extinction ratio of
-10dB and an insertion loss smaller than 5dB.
Chapter 3

Experimental Results and

Analysis

The first SixGelx electro-absorption butt-coupled modulators are measured

in [36]. Improvements in the fabrication process flow as well as in the coupling
design are the enabling factors that permitted to perform the measurements
here described. Improvements still need to be undertaken. Metallization-
related issues are the major contributors to low yield: devices that are both
electrically and optically active are a rarity. Nonetheless, devices that have
diode behavior and are optically transparent represent a breakthrough in
modulator design and performance.
The device layout used is depicted in Figure 3.1. Two sets of devices are
pictured divided by a waveguide with no devices on it. In the top device
series, there are modulators integrated with photodetectors. For modulators
with lengths shorter than 90pm, photodetectors are all 120tm long, while
for modulators 120 and 150,um long, photodetectors are 150tm long. The
second series, instead, shows one modulator per waveguide; each modulator
3.1 IV Characteristics

Figure 3.1: Device layout. Two sets of devices are shown: in the top set,
modulators are integrated with photodetectors; in the bottom set, there is
one modulator per waveguide.

has a different length (30-500pm). Important feature of this layout is that

modulators and photodetectors are integrated on the same waveguide.
In the experimental description that follows, only the performance re-
garding modulators will be presented. As in the laboratory, electrical testing
is performed first, followed by optical measurements, the same will be done
here: the first section is on current-voltage (IV) results, while the second
section is on low frequency (dc) measurements, followed by high frequency
measurements.

3.1 IV Characteristics

Electrical probing is performed in the range -3.3, 5V as shown in Figure 3.2.

Two wafers are probed (1776 devices per wafer divided in 33 dies): one single
device showing diode behavior is found on the first wafer, all other devices
3.1 IV Characteristics 58
58
3.1 IV Characteristics

50um EA SiGe Modulator: IV Charatersitics

5.
a
0

Figure 3.2: IV character-

istics of the 50um long
-2 0 2 4
Applied Voltage [V] electro-absorption SixGelx
modulator.

behave as open circuits. The second wafer has six dies that have devices
showing diode behavior even though they have high series resistance.
Among the electrically active devices that have diode behavior, the ones on
dies with highest yield and lowest resistance are used to measure modulation
depth. Other devices have lower current flow (in the nanoAmpire range for
a forward bias of 3V).
As the standard diode model does not fit accurately the IV data, fabri-
cation issues need to be closely considered in understanding the IV behavior
of the devices. The high series resistance is due to over-etching during fab-
rication of the metal contacts. Figure 3.3 shows that the silicide layer is
missing underneath the Tungsten. However, the measured series resistance
is of the order of 20-30kQ and an expected speed of 1-1.5GHz can be inferred
for devices with a length of 30-50pm.
 3.2 Low Frequency Measurements

Figure 3.3: SEM image of the

Tungsten contact. The sili-
cide layer is completely miss-
ing and over-etching of the
Silicon layer is clearly seen.

3.2 Low Frequency Measurements

The experimental set-up used for dc measurements is schematically pictured

in Figure 3.4. A Newport continuous wavelength laser source is used for
wavelengths in the range 1470-1570nm. Light from the fiber is directly cou-
pled to the waveguide; transmitted light is also directly coupled to a fiber.
This type of coupling drastically reduces scattering losses compared to free
space coupling. Light is then routed to a photodetector connected to a power
meter which feeds back into a lock-in amplifier. The lock-in amplifier is con-
nected to a pulse generator that determines the applied electrical bias on the
modulator through a bias tee.
Preliminary results on a first device show 30% modulation depth around
1560nm. A linear response between 5.2 and 6.5V, with a linearity factor of
13% modulation depth per applied volt, is measured. A maximum of 18%
modulation depth is obtained in this linear regime.
The modulation depth is given by the ratio between the "on" and "off"
3.2 Low Frequency Measurements 60
3.2 Low Frequency Measurements 60

Figure 3.4: Schematic ex-

perimental set-up for dc
measurements.

optical transmission states. Transmission scans also provide information on

the Franz-Keldysh oscillations. The knowledge of both modulation depth
and Franz-Keldysh oscillations allows the estimate of the electric field truly
present inside the intrinsic region of the device. This extrapolated value is
lower than the one expected from the technically applied bias. The cause
of this difference is the high series resistance in the device. The device is
expected to reach a modulation depth as high as 90% as soon as the origin
of the over-etching during fabrication is clarified and avoided.
A second modulator is considered. It is 50pm long and it lies on the
bottom device set of the layout previously pictured (Figure 3.1). Among all
electrically active devices on the same die, it is the only optically transparent
one. Even simple waveguides, normally used as test structures to measure
waveguide loss and insertion loss, are opaque. No transmitted light can be
measured. An explanation to this is that the metal filling used for chemical-
mechanical polishing on top of the structure is actually interfering with light
propagation drastically increasing waveguide losses, which are estimated to
be of the order of 30dB/cm.
As the optical "on" state is considered (Figure 3.5), low transmittance be-
low 1540nm is a result of the onset of direct band gap absorption. Above this
3.2 Low Frequency Measurements

wavelength, the SixGelx absorption is weak and it is only due to the indirect
band gap. An estimate of this absorption is inferred from the magnitude of
the Fabry-Perot fringes, -2dB.
Around 1550nm, the total insertion loss is determined to be ~3.7dB:
r2dB absorption loss due to indirect band gap and 1.7dB coupling loss be-
tween the waveguides and the modulator.
As a reverse bias is applied (Figure 3.6) and the electric field inside the
SixGel_- layer increases, the absorption edge flattens towards longer wave-
lengths. Absorption above 1540nm significantly increases while the trans-
mittance decreases.
If absorption is temperature induced, the absorption edge undergoes a
shift while keeping the slope constant. As is seen in Figure 3.6, the slope
decreases as the applied reverse bias increases, indicating that the experi-
mentally measured absorption is a result of the Franz-Keldysh effect rather
than thermal effects.
The technically applied reverse bias spans the range OV-7V. As mentioned
for the previous device, the electric field generated inside the intrinsic layer
of the device corresponds to a lower effective applied bias. This effective bias
is estimated using the Franz-Keldysh oscillations of the first device and the
IV measurements of the second device. The first device is used as the second
one does not show Franz-Keldysh oscillations and, hence, it is not possible
to calculate the effective electric field. Under these assumptions, the highest
technically applied reverse bias of 7V corresponds to an effective reverse bias
of -2.5V.
 3.2 Low Frequency Measurements 62
3.2 Low Frequency Measurements 62

Transmittivity at Optical on-State

" J

Wavelength [nm]

Figure 3.5: Transmittance spectrum of the 50ipm electro-absorption SixGel_-

modulator at optical "on" state (OV). From the Fabry-Perot fringes above
1540nm, the absorption loss (2dB) due to indirect band gap absorption is
inferred.

Transmittivity at Optical off-State

1500 1520 1540 1560

Wavelength [nm]

Figure 3.6: Transmittance spectra of the 50pm electro-absorption SixGel_-

modulator for three applied voltages: from lighter to darker gray as the
applied reverse bias increases (black line at OV inserted as a reference). Above
1540nm, the slope of direct band gap absorption edge flattens towards longer
wavelengths as the applied reverse bias increases.
3.2 Low Frequency Measurements 63

The achievable modulation depth is wavelength dependent. Figure 3.7

shows 70% modulation depth (5.2dB extinction ratio) in the 1535-1552nm
wavelength range, where field-induced direct band edge tilting is observed.
High extinction ratio, >7dB (i.e. 80% modulation depth), is obtained in the
wavelength range 1510-1533nm with an effective reverse bias of 2.5V (7V).
The maximum extinction ratio measured is 14dB (96% modulation depth)
at 1523nm; at the wavelength of 1550nm, it measures 6.5dB (78%). The
improved performance is due to device design with large overlap between
the optical field and the electric field. Modulation depths decreases as lower
reverse bias is applied as shown in Figure 3.8(a).
At 1550nm, the maximum modulation depth as a function of reverse effec-
tive bias discloses a pseudo-linear behavior with a slope coefficient of 40%/V

(achieved modulation depth per applied volt) in the range 0.4V< Veffective
<2.5V, Figure 3.8(b). Equivalently, in terms of extinction ratio, the slope is
2.2dB/V in the extinction ratio range 0-5.5dB. The ratio between the extinc-
tion ratio and the effective applied bias, the effective modulation, could be
considered as a new figure of merit of the device. Also shown is the theoreti-
cal calculation obtained by inverting the mathematical relationship between
the modulation depth and the absorption coefficient which is a function of
the applied bias. The theoretical calculation does not predict a linear trend:

ModulationDepth = 1 - e - Aa(V).L

Further investigation is needed to understand the experimental behavior.

Linear behavior is sought as it would mimic the linear behavior of Mach-
3.2 Low Frequency Measurements

Modulation Depth at 7V Reverse Bias

Figure 3.7: Modulation

depth of the 50,/m
_o electro-absorption
0C
S
SixGej_x modulator
o when 7V reverse bias
is applied. Modula-
tion higher than 78%
(6.5dB) can be achieved
of brLIr o 1ad l t-
SA VV CV enIgI
_CI U

Wavelength [nm] range as indicated in

the figure.

Zehnder devices. This feature would allow the device to become marketable
in analog applications.
Another fundamental parameter of this modulator is the power consump-
tion per bit of information processed and it can be calculated as !c-V 2 , where
c is the device capacitance (11fF) and V is the applied effective reverse bias.
For example, at 1500nm, with an effective bias of -2.5V, it is as low as
34fJ/bit. Hence, by reducing the capacitance, the driving voltage decreases,
as also the energy required per bit.
 3.2 Low Frequency Measurements

Modulation Depths as a Function of Wavelength

Wavelength [nm]

Modulation Depth as a Function of Applied Effective Reverse Bias

1UU II

0
aO
o 8

60

40 1550nm
-Theoretical Prediction
2 0o
t) * Experimental Data
- Linear Fit
0
0 1 2 3 4 5
Effective Reverse Bias [V]

(b)
Figure 3.8: Performance of the 50pm electro-absorption SiGel_- modulator:
(a) as-measured modulation performance as a function of wavelength. (b)
maximum modulation depth as a function of applied effective reverse bias.
Data in (b) is extrapolated from (a). At 1550nm, a pseudo-linear relation is
seen between the modulation depth and the applied bias in the range 0.5-2V.
3.3 High Frequency Measurements 66
3.3 High Frequency Measurements 66

Figure 3.9: Schematic

experimental set-up
for high frequency
measurements.

3.3 High Frequency Measurements

Measurements are performed with a set-up that is schematically drawn in

Figure 3.9. A Newport cw laser (1470-1570nm) generates the optical input.
Light propagates through a fiber which passes through a polarizer and then
through an optical amplifier. The input light is coupled to the on-chip waveg-
uide from the fiber. The dc voltage supply is connected to the device through
a bias tee. The RF signal is generated by an Agilent network analyzer and
it is delivered to the device through the same bias tee. From the through
waveguide, the transmitted light is collected into a fiber which terminates
into a photodetector. The photodetector is connected to a transimpedance
amplifier which in turn is connected to the network analyzer. As the device
has a high series resistance, a 50Q terminated probe is used to contact the
sample to overcome the impedance mismatch.
This system is used to evaluate the dynamic response of the SixGel-_
electro-absorption modulator in the high frequencies 10MHz-10GHz range.
The absolute RF power (as measured by the network analyzer) can be
3.3 High Frequency Measurements 67

evaluated either as a function of optical input power or as a function of ap-

plied reverse bias. Figure 3.10(a) shows the frequency response as a function
of the optical input power ranging from 5mW to 25mW while the applied
reverse bias is kept constant at a value of -3V. Figure 3.10(b) shows the ab-
solute RF power measured as a function of applied reverse bias (0-5V) while
keeping the optical input power constant at 20mW. The 3dB frequency at
3V reverse bias is 1.2GHz. The relatively low 3dB frequency is due to excess
series resistance from the electrical contact fabrication processes. In principle
the device should have a much higher bandwidth on the order of 100GHz.
It can be seen from Figure 3.10(a) and 3.10(b) that at low frequencies
an increased optical input power as well as an increased applied reverse bias
increases the detected output power. With the set-up as explained, parasitic
inductance could account for the abnormal low-frequency performance. Also,
further investigations are necessary to determine whether the high series
resistance is frequency dependent and/or applied bias dependent.
3.3 High Frequency Measurements

1E7 1E8 1E9 1E10

Frequency [Hz]

1E7 1E8 1E9 1E10

Frequency [Hz]

(b)
Figure 3.10: High frequency response of the 50pm electro-absorption
SiGel_, modulator. (a) Frequency response at 3V reverse bias as a func-
tion of optical input power. (b) Frequency response as a function of applied
reverse bias with 20mW fixed optical input.
3.4 Summary

3.4 Summary

A 50pm long SixGel_- electro-absorption modulator butt-coupled to high

index contrast Si/SiO 2 waveguides is experimentally probed. Both low fre-
quency and high frequency measurements are presented. Fabrication over-
etching issues, occurred during metal contact deposition, generate a 15kQ
series resistance. The calculated 3dB frequency of this device is of the order
of -100GHz; because of the high series resistance, the expected bandwidth
is of 1-1.5GHz. From high frequency measurements, 1.2GHz 3dB bandwidth
is determined with an applied reverse bias of 3V. It is experimentally demon-
strated that modulation is a result of the electro-absorption effect rather than
thermal effects: the slope of the direct band edge tilts towards longer wave-
lengths as the applied reverse bias increases. 70% modulation depth with low
effective driving voltage, - 2 .5Veffective, is obtained in the wavelength range

1510-1555nm; while modulation depth higher than 80% is measured in the

range 1510-1535nm. At 1550nm, the modulation depth is 78%. The pseudo-
linear behavior between the achieved modulation depth and the applied ef-
fective voltage suggests that the modulation efficiency, given by the ratio
between the extinction ratio and the effective applied bias, could become a
new figure of merit of the device. In this case, the modulation efficiency is
2.2dB/V in the extinction ratio range of 0-5.5dB, or equivalently, 40%/V in
the effective applied voltage range between 0.4V and 2V. Finally, the power
consumption is as low as 34fJ/bit.
Chapter 4

Conclusions and Future Work

Modulators are fundamental building blocks that will enable electronic- pho-
tonic integration. The aim of this thesis is to experimentally prove the per-
formance of a novel SiGel_, electro-absorption modulator design, waveguide
integrated, butt-coupled to Si/SiO 2 high-index contrast waveguides. Design,
fabrication, and experimental results are presented.
As reviewed in the first chapter, the electronic-photonic integration de-
pends on the ability of modulators to fulfill requirements regarding their
footprint, driving voltage, modulation depth and speed, as well as inser-
tion loss and temperature insensitive performance. A summary of published
Silicon based modulators optimized for applications at 1550nm is given; at-
tention is focused on the evaluation of combinations of optical and electrical
structures. Ring modulators have small footprint but they are highly tem-
perature sensitive and they are not broad band devices. In most cases, due to
fabrication tolerance, trimming and/or thermal tuning is required to achieve
performance at a predetermined wavelength. Mach-Zehnder devices have
larger wavelength range of operation; they are tolerant to small temperature
differences but they have large footprint. Published results on Fabry-Perot
based modulators always exploit the use of photonic cavities which are highly
sensitive to fabrication and temperature variations.
The second chapter describes the device background both theoretically
and with simulations: from the physical modulation mechanism, to the choice
of the material, as well as device dimension calculations. The fabrication pro-
cess flow is also mentioned. The chapter starts with the physical principle on
which the modulator is based, the Franz-Keldysh effect. Once the theoreti-
cal principle is defined, the best material choice is evaluated and hence the
SiGe alloy composition is calculated in terms of the optimization of the de-
vice figure of merit. An alloy composition with 0.75% Silicon and with 0.2%
tensile strain is determined. For a modulator 50pm long, simulations pre-
dict 100GHz 3dB frequency, with 10dB extinction ratio and insertion losses
smaller than 5dB.
The experimental device performance is evaluated in the third chapter.
IV characterization reveals a high series resistance of the order of 15kQ that
theoretically limits the 3dB frequency to only 1-1.5GHz. dc measurements
prove the modulation is a result of electro-absorption effects rather than
thermal effects as the slope of the direct band gap tilts towards longer wave-
lengths with increasing applied reverse bias. 70% modulation depth with low
effective driving voltage, -2 .5Veffective, is obtained in the wavelength range
1510-1555nm; while modulation depth higher than 80% is measured in the
range 1510-1535nm. At 1550nm, the modulation depth is 78%. Pseudo-
linear behavior between achieved modulation depth and applied effective
voltage suggests that the modulation efficiency, given by the ratio between
the extinction ratio and the effective applied bias, could become a new fig-
ure of merit of the device. Pseudo-linear behavior with a slope of 2.2dB/V
is found in the extinction ratio range of 0-5.5dB, or equivalently, a slope
of 40%/V in the effective applied voltage range between 0.4V and 2V. From
high frequency measurements the 1.2GHz 3dB bandwidth is determined with
an applied reverse bias of 3V. Abnormal low-frequency behavior is shown;
further investigations are required to explain these results.
The only SiGe modulators as of now published in the literature are by
Jonghthammanurak at MIT, [43], and by the Miller's group at Stanford,
[44]-[46] . The first one is a Mach-Zehnder device, while the second one is
a multi-quantum well device embedded in a Fabry-P6rot cavity. The former
one has large footprint, the latter one one works at wavelengths in the range
1440-1460nm and will not be able to reach 1550nm.
Concluding, the SiGe waveguide-integrated modulator optimized for per-
formance at 1550nm described in this thesis represents a breakthrough in
modulator research: it has small footprint (30pm 2 ), driving effective voltage
lower than 2 .5Veffective, large modulation depth (>70%) in a broad wave-

length range (1510-1555nm), power consumption as low as 34fJ/bit, and

1.2GHz 3dB bandwidth.
4.1 Future work

4.1 Future work

Suggestions regarding future work can be divided in two distinct phases: the
first one refers to the available working devices; the second one would require
further fabrication runs.

4.1.1 Current Devices

As mentioned while explaining experimental results, high series resistance is

the major issue degrading both electrical and optical performances. Yield
is clearly a significant limitation: devices that show open circuit behavior
are by far the majority with respect to devices showing diode behavior (one
diode-like device out of 1776 in the case of the first probed wafer). This is
primarily due to metal contact fabrication-related issues that cause a high
series resistance, tens of kQ while normal contact resistance should be of
the order of 10-100Q. Optical performance is degraded as the expected 3dB
frequency (>100GHz) is impossible to reach (the maximum 3dB frequency
measured is 1.2GHz at -3V applied bias).
High frequency measurements show abnormal behavior at low frequen-
cies: it is of interest to investigate whether this behavior is due to the high
series resistance, to impedance mismatch in the measurement set-up used,
to resonances in the system, or to a combination of these factors. Also, it is
unknown whether the series resistance is either applied voltage dependent or
optical input power dependent, or even frequency dependent. Devices with
different series resistance values could be used to exclude or clarify these
4.1 Future work

possibilities.
Linearity is obstinately sought to mimic Mach-Zehnder devices in order to
be useful in analog applications. Unfortunately, for the device being consid-
ered, the theoretical equation describing the relation between the maximum
modulation depth (measured at a specific wavelength) and the applied re-
verse bias is not linear. The pseudo-linear relation that is found (40%/V in
the range 0.5V< Veffective <2.5V) is only an experimental observation which
should be confirmed by expanding the measurements to all devices that are
in dies showing similar electrical diode-like behavior. Simulations of similar
devices could also reveal if it is possible to create a device that transforms
the relation between the modulation depth achieved and the applied reverse
bias to a truly linear relation, where "truly linear" implies linearity in both
theory and experimental observations.

4.1.2 Fabrication Implementation

The measured 3dB bandwidth of the 50pm SixGel_- electro-absorption mod-

ulator is 1.2GHz. Theoretical calculations predicted this value to be higher
than 100GHz as the only limitation should be RC with an extinction ratio
higher than 10dB. These values are not reached due to major fabrication is-
sues during metal contact fabrication. The origin of the over-etching beneath
the metal contacts causing high series resistance should be understood and
avoided.
Also, metal fill used for chemical-mechanical polishing drastically deterio-
rates optical propagation in waveguides. Waveguide loss is estimated to be as
4.1 Future work 75

high as 30dB/cm thanks to the metal layer being too close to the waveguide
layer. This layer should be either etched after the polishing is performed,
or it should be placed further away from the waveguide layer. Most of the
electrically active devices would then also be optically transparent.
These improvements would increase yield, and would allow measuring
devices with ~100GHZ 3dB frequency, extinction ratio >10dB, low driving
voltage and small footprint.
Bibliography

[1] Peercy PS, "The Drive to Miniaturization", Nature 406, 1023-1026

(2000)

[2] Parry G, "Silicon's New Shine", Nature 437, 1244 (2005)

[3] Lipson M, "Overcoming the Limitations of Microelectronics Using Si

Nanophotonics: Solving the Coupling, Modulation and Switching Chal-
lenges", Nanotechnology 15, S622-S627 (2004)

[4] Lipson M, "Guiding, Modulating, and Emitting Light on Silicon - Chal-

lenges and Opportunities", Journal of Ligthwave Technology 23, 4222-
4238 (2005)

[5] Manolatou C and Lipson M, "All-Optical Modulators Based on carrier

Injection by Two-Photon Absorption", Journal of Lightwave Technol-
ogy 24, 1433-1439 (2006)

[6] Almeida VR, Barrios CA, Panepucci R, and Lipson M, "All-Optical

Control of Light on a Silicon Chip", Nature 431, 1081-1084 (2004)

[7] Lipson M, "Switching Light on a Silicon Chip", Optical Materials 27,

731-739 (2005)

[8] Xu Q, Almeida VR, and Lipson M, "Micrometre-Scale All-Optical

Wavelength Conveter on Silicon", Optics Express 14, 9430-9435 (2006)

[9] Almeida VR and Lipson M, "Optical Bistability on a Silicon Chip",

Optics Letters 29, 2387-2389 (2004)

[10] Xu Q and Lipson M, "Carrier-Induced Optical Bistability in Silicon

Ring Resonators", Optics Letters 31, 341-343 (2006)
BIBLIOGRAPHY 77

[11] Lipson M, "Compact Electro-Optic Modulators on a Silicon Chip",

IEEE Journal of Selected Topics in Quantum Electronics 12, 1520-
1526 (2006)

[12] Xu Q, Schmidt B, Pradhan S,and Lipson M, "Micrometre-Scale Silicon

Electro-Optic Modulator, Nature 435, 325-327 (2005)

[13] Preble SF, Xu Q, Schmidt BS, and Lipson M, "Ultra-Fast all-Optical

Modulation on a Silicon Chip", Optics Letters 30, 2891-2893 (2005)

[14] Xu Q, Schmidt B, Shakya J, and Lipson M, "Cascaded Silicon Micro-

Ring Modulators for WDM Optical Interconnection", Optics Express
14, 9431-9435 (2006)

[15] Xu Q, Manipatruni S,Schmidt B, Shakya J, and Lipson M, "12.5Gbit/s

Carrier-Injection-Based Silicon Micro-Ring Silicon Modulators", Op-
tics Express 15, 430-436 (2007)

[16] Manipatruni S,Xu Q, and Lipson M, "PINIP Based High-Speed High-

Extinction Ratio Micron-Size Silicon Electro-Optic Modulator", Optics
Express 15, 13035-13042 (2007)

[17] Barrios CA and Lipson M, "Modeling and Analysis of High-Speed

Electro-Optic Modulation in High Confinement Silicon Waveguides
Using Metal-Oxide-Semiconductor Configuration", Journal of Applied
Physics 96, 6008-6015 (2004)

[18] Lipson M, "Switching light on a silicon chip", Optical Materials 27,

731-739 (2005)

[19] Almeida VR, Xu Q, and Lipson M, "Ultrafast Integrated Semiconduc-

tor Optical Modulator Based on the Plasma-Disperion Effect", Optics
Letters 30, 2403-2405 (2005)

[20] Liu A, Liao L, Rubin D, Nguyen H, Ciftcioglu B, Chetrit Y, Cohen R,

Izhaky N, Basak J, and Paniccia M, "Recent Advances in High Speed
Silicon Optical Modulator", Proceedings of SPIE 6477, 647710 (2007)

[21] Liu A, Liao L, Rubin D, Nguyen H, Ciftcioglu B, Chetrit Y, Izhaky N,

and Mario Paniccia , "High-speed optical modulation based on carrier
depletion in a silicon waveguide", Optics Express 15, 660-668 (2007)
BIBLIOGRAPHY

[22] Green WMJ, Rooks MJ, Sekaric L, Vlasov YA, "Ultra-Compact, Low
RF Power, 10Gb/s Silicon Mach-Zehnder Modulator", Optics Ex-
pressl5, 17106-17113 (2007)
[23] Liu A, Jones R, Liao L, Samara-Rubio D, Rubin D, Cohen 0, Nico-
laescu R, and Paniccia M, "A High-Speed Silicon Optical Modulator
Based on a Metal-Oxide-Semiconductor Capacitor", Nature 427, 615-
618 (2004)
[24] Liu A and Paniccia M, "Advances in Silicon Photonic Devices for
Silicon-Based Optoelectronic Applications", Physica E 35, 223-228
(2006)
[25] Izharky N, Morse MT, Koehl S, Cohen O, Rubin D, Barkani A, Sarid
G, Cohen R, and Paniccia M, "Development of CMOS-Compatible
Integrated Silicon Photonics Devices", IEEE Journal of Selected Topics
in Quantum Electronics 12, 1688-1698 (2006)
[26] Liao L, Samara-Rubio D, Morse MT, Liu A, Hodge D, Rubin D, Keil
UD, and Franck T, "High Speed Silicon Mach-Zehnder Modulator",
Optics Express 13, 3129-3135 (2005)
[27] Liao L, Samara-Rubio D, Liu A, Rubin D, Keil UD, Franck T,
Hodge D, and Paniccia M, "High Speed Metal-Oxide-Semiconductor
Capacitor-Based Silicon Optical Modulators", Japanese Journal of Ap-
plied Physics 45, 6603-6608 (2006)
[28] Lipson M and Kimerling LC, "All Optical Switch for Optical Integrated
Circuits", United States Patent, Patent No.: US 6,684,003 B2 (2004)
[29] Barrios CA, Almeida VR, and Lipson M, "Low-Power-Consumption
Short-Lenght and High-Modulation-Depth Silicon Electro-Optic Mod-
ulator", Journal of Lightwave Technology 21, 1089-1098 (2003)
[30] Barrios CA, Almeida VR, Panepucci R, and Lipson M, "Electroop-
tic Modulation of Silicon-on-Insulator Submicrometer-Size Waveguide
Devices", Journal of Lightwave Technology 21, 2332-2339 (2003)
[31] Schmidt B, Xu Q, Shakya J, Manipatruni S, and Lipson M, "Compact
Electro-Optic Modulator on SIlicon-on-Insulator Substrates Using Cav-
ities with Ultra-Small Modal Volume", Optics Express 15, 3140-3148
(2007)
BIBLIOGRAPHY

[32] Liu J, Sun X, Pan D, Wang X, Kimerling LC, Koch TL, and Michel J,
"Tensile-Strained, n-Type Ge as a Gain Medium for Monolithic Laser
Integration on Si", Optics Express 15, 11272-11277 (2007)

[33] Shen H and Pollak FH, "Generalized Franz-Keldysh Theory of Elec-

tromodulation", Physics Review B 42, 7097-7102 (1990)

[34] Frova A, Handler P, Germano FA, and Aspnes DE, "Electro-Absorption

Effects at the Band Edges of Silicon and Germanium", Physics Review
145, 575-583 (1966)

[35] Chuang SL, "Physics of Optoelectronic Devices", Wiley Series in Pure

and Applied Optics, Chapter 13, Goodman JW Series Editor (1995)

[36] Liu J, "GeSi Photodetectors and Electro-absorption Modulators for Si

Electronic-photonic Integrated Circuits", PhD Thesis, MIT (2007)

[37] Pollak FH and Cardona M, "Piezo-Electroreflectance in Ge, GaAs, and

Si", Physics Review 172, 816 (1968)

[38] Pickering C, Carline RT, Robbins DJ, Leong WY, Barnett SJ, Pitt
AD, and Cullis AG, Journal of Applied Physics 73, 239 (1993)

[39] Weber J and Alonso MI, "Near-band-gap photoluminescence of Si-Ge

alloys", Physics Review B 40, 5683(1989)

[40] Madelung O, Landolt-Bornstein, Properties of Group IV elements and

III-V, II-VI, and I-VII Compounds, Numerical Data and functional
relationships in science and technology Vol 17a, chapter 1.6, Springer
(Berlin) 1982

[41] Humlicek J, Properties of strained and relaxed Silicon and Germanium,

EMIS Datareviews Series, No 12, INSPEC, London (1995), chapter 4.

[42] Madelung O, Landolt-B6rnstein, Properties of Group IV elements and

III-V, II-VI, and I-VII Compounds, Numerical Data and functional
relationships in science and technology Vol 17a, chapter 1.6, Springer
(Berlin) 1982

[43] Jonghthammanurak S, "Ge-rich SiGe Materials for Field-Effect Mod-

ulator Applications", PhD Thesis, MIT (2008)
BIBLIOGRAPHY 80

[44] Kuo YH, Lee YK, Ren Y, Roth JE, Kamins TI, Miller DAB, and Harris
JS, "Strong Quantum-Confined Stark Effect in germanium Quantum-
Well Structures on Siliocn", Nature 437, 1334-1336 (2004)

[45] Roth JE, Fidaner O, Schaevitz RK, Kuo YH, Kamins TI, Harris JS,
and Miller DAB, "Optical Modulator on Silicon Employing Germanium
Quantum Wells", Optics Express 15, 5851-5859 (2007)

[46] Okyay AK, Pethe AJ, Kuzum D, Latif S, Miller DAB, and Saraswat
KC, "SiGe Optoelectronic Metal-Oxide Semiconductor Field-Effect
Transistor", Optics Letters 32, 2022-2024 (2007)