PDF 2
PDF 2
PDF 2
Departamento de comunicaciones
Junio 2013
Acknowledgements
Agradecer a Mariam por su inestimable ayuda, cualquier duda que he tenido la has
hecho tuya para poder darme la mejor solución posible. Soy muy afortunada por haberte
tenido todo este tiempo de compañera de trabajo. Pero mi agradecimiento a ti va más
allá de lo laboral. No tengo palabras para describir tu calidad como persona y como
amiga. Has sido uno de mis mayores apoyos en los momentos difíciles y siempre has
estado ahí de manera incondicional sin esperar nada a cambio. Nunca podré llegar a
agradecerte como te mereces todo lo que has hecho por mí. Ni que decir tiene que
empezaste siendo sólo una compañera de trabajo pero hoy día eres una gran amiga y
puedo afirmar que para toda la vida. Te quiero muchísimo “mori”.
Albita y Marta, mis amores, toda una vida juntas contra viento y marea. Sois parte de mí
y por tanto parte de esta tesis. Muchas veces os amenazaba con que si no os portabais
bien no apareceríais en estos agradecimientos. Sabéis que eso es imposible, tengo
mucho que agradeceros. Cuando uno tiene que llevar a cabo un trabajo como es esta
tesis y todo se vuelve negro, la realización de dicho trabajo se vuelve muy complicada.
Pero vosotras siempre habéis estado ahí para darme esa luz que se necesita para seguir
adelante. A cualquier hora, en cualquier momento y cualquier lugar, sea como fuere,
una llamada ha bastado para que vinierais corriendo a brindarme vuestro hombro. Eso
no se olvida nunca. Sois amigas, confidentes, hermanas y todo por lo que alguien debe
dar gracias todos los días de tener a su lado. Gracias por todo chicas, os he querido os
quiero y os querré toda mi vida.
Por supuesto que no me olvido del resto de “Femeninas Emprendedoras”. Ana, Patri y
Elia. Esas quedadas semanales han sido durante la realización de esta tesis. Esas noches
de “marujeo” y de “terapia” han sido de lo mejor de estos 4 años, siempre las recordaré
con una sonrisa todas y cada una de ellas aunque no voy a hablar en pasado porque
espero que sigamos haciéndolas. Risas aparte, también he tenido la suerte de contar con
vuestro apoyo y cariño en los peores momentos, siempre pendientes de mis
“problemillas” y de mi estado de ánimo. Sois buenas personas, buenas amigas y espero
teneros siempre a mi lado. Os quiero mil chicas. Pedrito, literalmente, una página de
este tesis es “teua”, gracias, eres un crack. Rubén, decirte que estás lejos pero estás
cerca, mi mayor regalo de la carrera es haberte conocido.
También quiero agradecer a Javi García, Josema y Pak por los grandes y buenos
momentos vividos en estos últimos meses de desesperación durante la fase final de
escritura de la tesis la cual me alegro que haya coincidido en tiempo con la vuestra. Por
fin somos doctores chicos y parecía que no iba a llegar nunca. El humor que le habéis
puesto y las risas que me habéis hecho sacar son de mucho agradecer. Os deseo de
corazón muchísima suerte a todos allá donde vayáis.
Eva, Jose, Belén, Pepa, mis “Wasay”. Que sepáis que esa “secta” es de lo mejor que me
ha pasado en estos 2 últimos años. Como sé que vosotros no sois mucho de leer en
inglés he hecho los agradecimientos (acknowledgements) en castellano para que no os
quejéis y porque sé que me lo habríais estado recordando durante mucho tiempo.
Vuestra “coles”, “cables”, “chispas”, “cani”, “tocineta”,…, o sea yo, os quiere agradecer
vuestros ánimos y apoyo durante esta tesis, y deciros que soy feliz de tener esa segunda
familia llamada “Wasay”. Os quiero.
Gracias a mis tíos Seve, Choni y Antonio por vuestro apoyo incondicional y vuestro
continuo interés en esta tesis.
La tesis consiste en 3 capítulos principales así como una sección de conclusiones del
trabajo conseguido. El capítulo uno está destinado a dar una descripción general de los
beneficios del uso de la fotónica de silicio, mostrando sus retos y oportunidades, así
como a dar una visión profunda de todos los aspectos relacionados con la modulación
electro-óptica. El capítulo dos está dedicado a desarrollar moduladores de silicio de altas
prestaciones para aplicaciones digitales. Específicamente, se presentan nuevas
estructuras ópticas diferentes a las convencionales con el objetivo de mejorar el
rendimiento de la modulación o al menos algunos parámetros críticos en la modulación.
El tercer capítulo se dedica a las aplicaciones analógicas. Se describe el concepto de la
fotónica de microondas, así como diferentes investigaciones llevadas a cabo en el
ámbito analógico para su aplicación en el campo de la fotónica integrada de
microondas, todas ellas usando moduladores electro-ópticos de silicio compatibles con
los procesos de fabricación CMOS, lo que valida el potencial de la fotónica de silicio
como un prometedor enfoque para permitir el desarrollo de aplicaciones de la fotónica
integrada de microondas. Por último, las conclusiones sobre el trabajo realizado se
proporcionan en el Capítulo 4.
i
ii
Resum
La tesi consisteix en 3 capítols principals així com una secció de conclusions del treball
aconseguit. El capítol un está destinat a donar una descripció general dels beneficis de
l'us de la fotònica de silici, mostrant els seus reptes i oportunitats, així com a donar una
visió profunda de tots els aspectes relacionats amb la modulació electro-òptica. El
capítol dos está dedicat a desenvolupar moduladors de silici d'altes prestacions per à
aplicacions digitals. Específicament, es presenten noves estructures òptiques diferents a
les convencionals amb l'objectiu de millorar el rendimient de la modulació o al
menys alguns parámetres crítics en la modulació. El tercer capítol es dedica a les
aplicacions analògiques. Es descriu el concepte de la fotònica de microones, així com
diferents investigacions dutes a terme en l'àmbit analògic per à la seua aplicació en el
camp de la fotònica integrada de microones, totes elles usant moduladors electro-òptics
de silici compatibles amb els processos de fabricació CMOS, el que valida el potencial
de la fotònica de silici com un prometedor enfocament per à permetre el
desenvolupament d'aplicacions de la fotònica integrada de microones. Per últim, les
conclusions sobre el treball realitzat es proporcionen en el Capítol 4.
iii
iv
Abstract
Silicon photonics is one of the most exciting and fastest growing photonic technologies
in recent years. The salient feature of this technology is its compatibility with the
mature silicon IC manufacturing based on complementary metal-oxide semiconductor
(CMOS) processes widely used in microelectronic industry. Another motivation is the
availability of high-quality silicon-on-insulator (SOI) planar waveguide circuits that
offer strong optical confinement due to the high index contrast between silicon (n=3.45)
and SiO2 (n=1.45). This opens up miniaturization and very large scale integration of
photonic devices allowing photonic integrated circuits for a wide range of applications
and markets, from optical telecommunications to bio-photonic devices or precise fibre
sensors. Optical modulators are key building-blocks for high speed signal transmission
and information processing in any photonic interconnection solution. The work
developed in this thesis, as part of the objectives of the European project HELIOS in
which it is framed, is essentially focused on realizing compact and efficient modulators
integrated on silicon chips.
The thesis consists of three main chapters as well as the concluding section on the
work accomplished. Chapter one is aimed at giving a general description of the benefits
of using silicon photonics, showing its challenges and opportunities as well as at giving
a deeply overview of all issues related to the electro-optic modulation. Chapter two is
devoted to develop silicon modulators with high features for digital applications.
Specifically, new optical structures different to the conventional ones are presented with
the aim of enhancing the modulation performance or at least several critical parameters
in the modulation. Chapter three is dedicated to the analog applications. The concept of
microwave photonics is described as well as different researches carried out in the
analog scope for application in the field of integrated microwave photonics, all of them
using CMOS-compatible electro-optic silicon modulators which validate the potential of
silicon photonics as a promising approach for enabling the development of integrated
microwave photonics applications. Finally, conclusions on the work realized are
provided in Chapter 4.
v
vi
Contents
Acknowledgment
Resumen ......................................................................................................................i
Resum ..........................................................................................................................iii
Abstract .......................................................................................................................v
Contents .......................................................................................................................vii
1. Introduction ..........................................................................................................1
1.1. Silicon photonics .............................................................................................1
1.2. Electro-optic modulation in silicon .................................................................6
1.2.1. Optical modulators ...............................................................................7
1.2.2. Modulation in silicon: the plasma dispersion effect.............................8
1.2.3. Phase modulation: electrical structures ................................................9
1.2.4. Intensity modulation: Mach-Zehnder Interferometer (MZI) ................11
1.3. Development process of photonic structures...................................................14
1.3.1. Design...................................................................................................14
1.3.2. Mask generation ...................................................................................15
1.3.3. Fabrication ............................................................................................16
1.3.4. Characterization ...................................................................................17
1.4. Objectives and outline of the thesis .................................................................17
vii
2.3.2.2. Basic operation principle of slow-light effect..........................49
2.3.2.3. Corrugated slow-light waveguide design ................................51
2.3.3. Experimental slow-light MZI (SL-MZI) modulators ...........................55
2.3.3.1. High-speed SL-MZI modulator ...............................................55
2.3.3.2. Low-drive voltage SL-MZI modulator ....................................59
2.3.4. Method for obtaining propagation losses .............................................62
2.3.4.1. Rib waveguides ........................................................................62
2.3.4.2. Corrugated waveguides............................................................68
2.4. Active RF characterization set-ups..................................................................71
2.4.1. Data transmission measurements .........................................................71
2.4.2. Small signal measurements ..................................................................73
Acronyms .....................................................................................................................157
Bibliography ................................................................................................................159
viii
ix
x
Chapter 1
Introduction
Among all these main application areas, the field of ICTs is witnessing a
development speed unprecedented in history. Internet traffic grows by 40-60% per year,
almost unchanged for >10 years, driven by new user applications and user access data
rates by 42% per year [OCL11]. So, as traffic demands are growing much more rapidly
than revenues, service providers must continually seek ways to carry the traffic more
efficiently. The idea of using light beams to replace wires has been dominating during
1
Chapter 1: Introduction
the last decades all long-distance communications but the trend nowadays is
progressively taking over in networks over increasingly shorter distances: from optical
fiber communications to optical interconnect computing; and silicon photonics seems to
be the most effective technology for such purpose.
Signal transmission and processing by means of optical beams rather than electrical
currents or radio waves has been a very interesting topic since the 1960s [KAO66]. Low
loss, high data rate capabilities, noise rejection and electrical isolation are just a few of
the important features that make fiber optic technology ideal for the transport of
information in communication networks [GAM00]. In current telecom and datacom
networks, the optical fibers are used to transport the information in the form of light and
are usually installed in the intermediate links between the core networks and between
networks and small sub-networks, whilst copper wire or coaxial cable is commonly
used in the access network which connects subscribers to their immediate service
provider. Replacing at least part of these links with fiber shortens the remaining copper
segments and allows them to run much faster. Following this approach, new emerging
technologies such as FTTH (Fiber-To-The-Home) started to arouse the interest of
telecom operators many years ago. The key objective of FTTH technology is to be able
to offer to the customers a higher speed and broadband access, by bringing the fiber link
directly to their homes. However, even in FTTH technology, the network equipment
mainly remains electronic circuitry, and the optical signals are just used to transport the
data, nor for data processing. So, in the network nodes, the routing and signal
processing is performed electrically, and the electrical signal is converted back into light
and sent on the output fiber and transport it towards the next node. These optoelectronic
conversions cause an increase of the latency of the whole transport network, and the
benefits of high velocity capability of using optical fibers is unfortunately wasted. This
fact represents a real bottleneck in current communication networks. The most efficient
way for making the most of optical fiber advantages is to replace microelectronic
circuitry with optical circuits capable for all-optical signal processing. Fortunately,
besides the main function of guiding light, optical fibers can also be developed as
attractive photonic components performing desired optical signal processing, through
corresponding modifications on the fiber core. So, Photonics has seen a big
development in the last years at the request of this communication market. Nowadays,
the capacity of optical communication on long hauls is reaching some Tbit/s over
thousands of kilometres. And all these are thanks to the progress in optical fibre
fabrication, the use of erbium-doped-fiber-amplifiers (EDFA) and Raman amplifiers,
modulators and single frequency lasers. However, these optical fiber-based devices
result in high cost and large discrete elements. The key to make more advanced and
cheaper components is the integration of these optical components on a single chip.
On the other hand, as the first semiconductor transistor was demonstrated in 1943,
and the later invention of integrated circuit (IC), microelectronics has kept developing
in a very fast pace. The size of complementary metal-oxide-semiconductor (CMOS)
transistors, as a key factor concerning the computer performance, has been continuously
2
1.1. Silicon photonics
scaled following the Moore's Law, which states that the number of transistors on an
integrated circuit doubled approximately every two years [MOO65]. However,
conventional copper electronic interconnects are not suitable for the ultra-high
bandwidth on-chip communication in future so, a significant performance limitation in
ICs has become the metal interconnect. For that, a big driving force is the need to
replace electronic interconnects in today’s microprocessors, which has become the
bottleneck that limits the operating speed of microelectronic devices. Optical or
photonics interconnects seems to be the better solution to overcome such bottleneck due
to its larger bandwidth, lower power consumption, smaller interconnect delays, and
better resistance to electromagnetic interference. When one speak about optical
interconnects, one may think in four levels of interconnects which can be easily
identified: rack-to-rack, board-to-board, chip-to-chip and on-chip. These optical
interconnects are the main motivation to look for silicon photonics.
Figure 1-1: Optical interconnects, “This approach to signal transfer is moving from longer-
distance applications, such as linking separate computers, to joining chips within a computer”
[SAV02].
So, the key to face the actual trend is to make advanced and cheap components
towards the integration of optical components on a single chip. When the optical
components are integrated on chip, we talk about photonic integrated circuits (PICs).
Following same approach as in microelectronic devices, the commitment towards a
process of photonic device integration and miniaturization has attracted the interest of
industry which has allowed the emergence of integrated photonics. So, the main
objective is to find a technology in which a variety of PICs can be easily fabricated in a
generic foundry process. This concept is well known in silicon microelectronics. As
happened with silicon microelectronics, photonic technology to the mass market may
happen only if one can bring high volume manufacturing at low cost. It was possible
3
Chapter 1: Introduction
Changing the field but without departing from the scope of the photonic, the area of
microwave photonics (MWP) has evolved in parallel to the development of optical
communication, as optical components have become available at microwave
frequencies and beyond. We refer to MWP as the study of photonic devices operating at
microwave frequencies and their applications in microwave and optical systems. The
interdisciplinary field of MWP covers research within a frequency span from MHz to
THz. The flexibility of microwave technology in combination with the low loss, large
bandwidth and the insensitivity to electromagnetic disturbances of photonics, create a
solid foundation for a vast number of MWP applications. These applications span from
common antenna TV, to antenna remoting of communication systems and radars and to
signal generation and signal processing. Further, a particular aspect that recently gains
significant interests is the use of PIC technology in the microwave photonics (MWP)
field, technology known as integrated microwave photonics (IWMP), for enhanced
functionalities and robustness as well as the reduction of size, weight, cost and power
consumption. IMWP is an emerging field, in which analog radio frequency (RF) signals
are generated, distributed, processed and analyzed using the strength of photonic
techniques. It is a technology that enables various functionalities which are not feasible
to achieve only in the microwave domain. The past years have also seen significant
increase of silicon photonics technology for MWP systems.
After presenting all challenges and opportunities of silicon photonics, one can
imagine that tremendous amount of progress has been reported in the last decade in
individual photonic components on silicon, including hybrid lasers, fast and low-power
silicon electro-optic modulators, and high-speed photodetectors or even transceivers,
4
1.1. Silicon photonics
since significant global companies such as IBM or Intel, have included the silicon
photonics technology among his main lines of research. As a matter of fact, the
integration of modulators and photodetectors (the two key interfaces bridging the
electrical and optical domains) on the same chip, enabling a fully PIC with high speed
electronics, is one crucial step towards a fully functional optical interconnect system.
Figure 1-2: (a) Intel Silicon photonics-based 50 Gbit/s link (5 channels @12.5Gbps) with a bit
of scaling: 1 Terabit/s link forecasted (25 channels @40Gbit/s) (b) Intel’s packaged 40Gbit/s
modulator [INT], (c) angled view of a portion of an IBM chip showing blue optical waveguides
transmitting high-speed optical signals and yellow copper wires carrying high-speed electrical
signals [IBM].
In this thesis, the previously introduced SOI technology is employed to develop one
of these key building blocks. This device, as already abovementioned, is the electro-
optic modulator which links the electrical and optical domain. So, as proof of concept,
this thesis intends to be a contribution to the field of silicon photonics through the
demonstration of high-speed modulation in silicon devices to achieve more complex
building blocks and thus going one step further towards a fully integration of
modulators devices in the CMOS-compatible silicon photonics platform for digital and
analog applications for future inter/intra-chip interconnect applications.
This thesis is framed in the European project called HELIOS [HEL] (pHotonics
ELectronics functional Integration on CMOS). The HELIOS consortium has developed
innovative means to combine a photonic layer with a CMOS circuit, using
microelectronics fabrication processes. HELIOS has gathered 19 partners among the
5
Chapter 1: Introduction
major European CMOS Photonics and Electronics players and potential users. It has
driven the European RTD in CMOS Photonics and has paved the way for industrial
development [HEL].
This section introduces all the issues related to electro-optic modulation effect. The
definition of optical modulators, as well as different ways to classify them, is
introduced. Moreover, the basic principles and performance analysis of a Mach-Zehnder
Interferometer (MZI) modulator are discussed in order to get a comprehensive picture of
the mechanisms taking place in silicon to achieve phase and amplitude modulation
which are also described. Finally, a description of the complete development process of
is reviewed.
6
1.2. Electro-optic modulation in silicon
An optical modulator is a device that is used to modulate (that is, vary the
fundamental characteristics of) a light beam propagating either in free space or in an
optical waveguide. These devices can alter different beam parameters, allowing them to
be categorized as either amplitude, phase or polarization modulators. The modulating
signal may come in several forms such that modulation may be carried out either via
electro-optical, magneto-optical, mechano-optical or acousto-optical effects. The prefix
placed before “optical” defines the nature of the modulating signal [BRI11B]. In
addition, modulators can be also classified as either electro-refractive or electro-
absorptive modulators. Applying an electric field to a material may change both real and
imaginary coefficients of the refractive index. A change in the real coefficient of the
refractive index (Δn) with an applied electric field is known as electro-refraction,
whereas a change in the imaginary coefficient of the refractive index (Δα) is known as
electro-absorption. Another main classification of modulators can be defined by the
signal type of the modulating signal.
Figure 1-3: (a) Analog (b) digital amplitude modulation in a MZI modulator.
On one hand, we can talk about analog optical modulation, where the optical carrier is
gradually modulated in phase or intensity via a continuous electrical signal and on the
other hand, digital optical modulation, which converts a continuous light waveform
(containing no information) into a discontinuous carrier that carries information in a
binary form, that is, ‘1’s and ‘0’s referred to as ‘bits’. These binary states correspond to
7
Chapter 1: Introduction
the case when light is respectively switched on and off. Specifically, in this thesis
electro-optical modulators are developed for both digital and analog applications, so the
modulating signal will be respectively, a set of bits and a continuous RF signal. Figure
1-3 illustrates an example of analog and digital amplitude modulations of a MZI
modulator.
18 18
8.5 10 Ne 6.0 10 Nh (1.2)
where ΔNe and ΔNh are respectively the variations in concentration of electrons and
holes in cm-3.
Let us consider now that light coming up from a continuous wave laser can be
confined and guided by total internal reflection (mechanism through which light can be
confined and hence guided within the central layer of the silicon waveguide) in a single
mode silicon waveguide. As for electronic devices, silicon needs to be doped with
elements from the third and fifth columns of the periodic classification of elements to
form a semiconductor junction capable of injecting, depleting or accumulating charge
carriers (electrons and holes). This enables to accurately control the electrical response.
Such control is required to realize diodes, transistors, detectors, and others devices.
Conveniently using this existing technology, it becomes possible to combine an optical
structure guiding the propagating mode with an electrical structure in charge of
injecting, depleting or accumulating a plasma of free carriers overlapping the mode in
question. The next section deals with this issue.
8
1.2. Electro-optic modulation in silicon
2
neff LACT (1.3)
where ∆neff is the effective index change, λ is the wavelength and LACT is the length of
the active region. By embedding such optical phase modulation in a MZI structure, it is
possible to convert the phase modulation into intensity modulation as we will discuss in
the section 1.2.4.
The speed of the optical modulation is directly related to the intrinsic speed of charge
carrier motion, but however may be limited by other effects. Since the first GHz
modulator [LIU04] was demonstrated in 2004, a variety of designs were reported, each
of them involving specific charge carrier motion in an appropriate electrical structure.
So, electrical manipulation of the charge density with the propagation light is achievable
through mechanism such as carrier injection, accumulation or depletion. An overview
on each of them is hereinafter given.
The first electrical mechanism used for achieving high-speed modulation was the
carrier accumulation. In such case, carrier density modulation is obtained by driving a
metal-oxide-semiconductor (MOS) capacitor (Figure 1-4(a)) into the accumulation
mode of operation. A thin insulating layer of oxide, SiO2 (p-type doped) is used to
isolate two halves of the waveguide (slightly n-type doped) to create a capacitor
structure. When a positive voltage is applied between the contacting pads, electrons and
holes accumulate at the gate oxide layer, that is, at the Si/SiO2 interface. This change in
charge density causes the optical mode propagating to overlaps with those accumulated
electrical charges, leading to a variation in its effective index of refraction. Carrier-
accumulation modulators are not limited by the relatively long minority carrier lifetime
in silicon, which is the case of carrier-injection modulators, but are instead dominated
by the device resistance and capacitance. Thus, these modulators generally suffer from
high capacitance and the lack of modulation efficiency due to the low amount of carrier
involved.
9
Chapter 1: Introduction
On the other hand, carrier injection-based modulators are usually based on forward
biased PIN diode (Figure 1-4(b)). The basic PIN structure architecture consists of a
silicon optical waveguide embedded into highly p- and n-type doped silicon regions.
When the junction is forward biased, minority charge carrier motion through the
intrinsic silicon region, where the mode propagates, is produced. So, forward-biasing
the device causes free electrons and holes to be injected into the intrinsic waveguide
region. The optical mode is confined into the intrinsic silicon region to avoid high losses
due to heavily doped p- and n-type regions and overlaps with the current of charge
carriers injected through the intrinsic region.
There is a trade-off between the intrinsic silicon region width, meaning higher
modulation speed and optical losses induced by heavily doped regions. Because of the
slow carrier/generation recombination process modulation are in principle limited to 1
GHz. To counteract such a fundamental limitation, a pre-emphasis driving scheme
consisting in applying short peaks of current at the beginning of each electrical bits to
accelerate the carrier generation process and followed by a reverse bias operation to
extract carriers more efficiently, may been employed which adds complexity to the
overall device. Moreover, the advantage of PIN diode architectures is the high amount
of charge carriers involved in the electrical process leading to strong change in silicon
refractive index. Therefore, smaller modulation lengths at the expense of non negligible
power consumption unless low driving voltages are used. So, special care must be taken
to minimize the insertion losses and power consumption.
Finally, depletion modulators are based on the change in space charge region
depletion width of a PN junction, that is, slightly doped p- and n-type regions in the
waveguide to form a PN diode (Figure 1-4(c)). The diode’s depletion width, and
therefore the free-carrier density in the waveguide, varies with the applied reverse bias
in such a way that, the depletion area of the diode becomes larger with increasing
reverse bias voltage.
Figure 1-4: Cross section of typical phase shifters to implementing the (a) carrier accumulation
mechanism through a MOS structure, (b) carrier injection mechanism through a PIN structure,
(c) carrier depletion mechanism through a PN structure, used to electrically manipulate the free-
carrier concentrations in plasma dispersion based silicon optical modulators.
Due the fact that depletion modulators are based on the change in space charge region
depletion width of a PN junction and hence involve only majority carriers, no
generation/recombination process takes place since the modulator behaves as a RC
10
1.2. Electro-optic modulation in silicon
circuit with specific rise and fall time constant. This effect is intrinsically fast but
involves a lower amount of carriers than in the case of the injection regime, which
implies that carrier depletion-based modulator are generally less efficient than their
injection counterparts. Despite this drawback, the carrier depletion mechanism is the
technique that manipulates free-carrier densities in a modulator avoiding the speed
limitation posed by the minority carrier lifetime. So, modulation speeds in excess of 30
GHz can be reached by using carrier depletion modulators, but at reduced efficiency as
mentioned above. For all, the challenge remains to improve efficiency while
maintaining other performance metrics. Special care in decreasing the access (or series)
resistance and decreasing the junction capacitance must be taken into account.
Apart from these three main electrical structures, with the increasing develop in the
fabrication processes, evolution of them or new electrical structures have been proposed
in the last years. For example, a PIPIN (used in this thesis) structure which allows
achieving low loss optical modulator working at the same cut-off frequency than a PN
diode [RAS10]. A high-speed silicon MZI modulator was proposed, in [XU12], based
on interleaved PN junctions. This doping profile enabled both high modulation
efficiency and low doping-induced loss. Moreover, in [AKI12A, BAB12 and AKI12B]
the authors presented high-speed operation of PIN diode-based silicon MZI modulators
that have side-wall gratings on both sides of the waveguide core. In [THO10], a new
design of high speed carrier depletion silicon optical modulator was introduced which
featured the use of a self-aligned fabrication process to form the PN junction.
Let us briefly review the basic operation of the interferometer acting as electro-optic
modulator: the incoming light is split into two waveguides under the influence of
conducting electrodes, as shown in Figure 1-7. The electro-optical effect (the plasma
dispersion effect in the case of a silicon phase shifter) induces a change in the refractive
index of one or both interferometer arm and phase-modulates the light propagating into
that arm (modifying the values of Ф1 and/or Ф2) according to the electric voltage
applied to each electrode. By combining the two paths with different phase
modulations, this phase modulation is turned into an intensity modulation in such a way
that the power at the output can be defined as:
where Pin is the power input and ∆Ф=Ф2-Ф1 is the phase difference between both
arms.
The ideal working principle of this amplitude modulator is based on the fact that the
active part of the first arm shifts the phase of the incoming wave by π (by applying a
suitable voltage Vπ) with respect to the second arm. As a result, the splitted waves
interfere destructively, leading ideally to no optical signal at the output. The modulator
is in turned off and a “0” is transmitted (Figure 1-6(a)). On the contrary, if no voltage
is applied to the phase shifter, the phase of the wave remains unchanged and hence the
two waves interfere constructively at the output leading to no alteration of the optical
signal. The modulator is turned on and a “1” is transmitted (Figure 1-6(b)).
Figure 1-6: Ideal MZI amplitude modulator consisting of two phase shifters integrated into both
arms of the device. The operation principle is shown.
12
1.2. Electro-optic modulation in silicon
The efficiency of the modulator active region is defined by the VπLπ product, where
Vπ is the bias voltage applied to the junction, and Lπ is the length of the active region
that leads to a π-phase variation. The better efficiency is obtained when VπLπ is minimal,
that is, when the active region length is the shortest and/or the applied bias voltage is the
lowest. Thus, this VπLπ product can also be read as a footprint indication.
Two basic types of MZI modulators with different configurations can be defined:
Single drive: this type of modulator has a single access to both branches. By
introducing an electrical signal in such single access, it will affects in the same
way to both electrodes but with opposite sign. This modulator can only works in
a push-pull configuration, which is obtained by applying data and bias voltage in
one arm and inverted data and inverted bias voltage in the other arm.
Dual drive: this kind of modulators has two accesses, one for each of the two
branches. So it is possible to work in asymmetric configuration, besides in a
push-pull configuration, where the modulating signal and the bias voltage are
applied to only one of the interferometric branches, either to the same or to
different branches.
Finally, there are several figures of merit that are used to characterize a modulator
[REE10]:
13
Chapter 1: Introduction
1.3.1. Design
When talking about light propagation in optical integrated waveguides, one must be
thinking in two physical approaches which are presented in such mechanism: on one
hand the ray optics approach or geometrical optics, which enables light propagation
along an optical waveguide by total internal reflection [REE04] (in silicon this
mechanism is highly effective due to the high index contrast effect); and on the other
hand, the electromagnetic approach (by means of Maxwell’s equations), which allows
to derive the guided modes and the field patterns. Maxwell’s equations enable to make a
rigorous study of the mode profiles propagating along the considered waveguide.
First, the designer must try to model the performance of the device under study. The
modeling of photonic structures may not always be carried out in an analytic way,
especially when one deals with high index contrast structures with non uniform
refractive index profiles all over the propagation length. In this case, the device under
study must be simulated to analyze its performance.
Figure 1-7: (a) BeamPROP simulation of the fundamental mode in a silicon waveguide of
220nm height and 450nm width, (b) FullWAVE simulation on a waveguide-ring resonator
structure with a distance between them of 200nm in the coupling region. The power density
trough the ring and the waveguide is represented.
Several techniques exist to perform these simulations. Among them, some of those that
have been employed to simulate the behaviour of our photonic modulators are: the finite
time difference domain (FDTD) method, which consists in discretizing Maxwell’s
equations via finite differences, the plane wave expansion (PWE) method, which, in
14
1.3. Development process of photonic structures
Besides from the abovementioned optical modelling, also the electrical modelling
must be taken into account when working with the electro-optic modulators based on
the plasma dispersion effect. In this case, an extra semiconductor device simulation is
needed to simulate the electrical structures which compound the phase shifters of the
electro-optical modulators under study. Specifically, in this thesis, ATLAS simulation
tool from SILVACO has been used. It predicts the electrical behavior of specified
semiconductor structures and provides insight on the internal physical mechanisms
associated with device operation. DC, alternating current (AC) and transient behavior of
a given device may be modelled via solving the discretized Poisson’s equation (derived
from Maxwell’s law), which basically relates variations in electrostatic potential to local
charge densities.
All the simulations carried out to finally designing the electro-optical modulators
used in this thesis have been developed under the framework of the work package 3 of
HELIOS project partnered by the University of Surrey (UNIS), the Université Paris-Sud
(UPS) and the Nanophotonics Technology Center in the Universitat Politècnica de
Valencia (UPVLC).
15
Chapter 1: Introduction
Figure 1-8: Final GDSs mask layers of the fabricated modulators used in this thesis, (a) MZIs
and slow-wave MZI modulators and (b) micro-ring and ring-assisted MZI modulators.
1.3.3. Fabrication
Today’s trend in silicon photonics is demonstrating that any new device design may be
mass produced in CMOS foundries. Such entities consist of advanced nanofabrication
tools and highly experienced engineers. Working in a controlled environment gives the
designer the opportunity to be aware of the reality of fabrication constraints and
capabilities.
16
1.4. Objectives and outline of the thesis
Figure 1-9: (a) Photograph of an 8-in SOI wafer with hundreds of photonic components and
circuits and (b) picture of a single silicon chip fabricated in CEA-LETI consisting of 26 electro-
optical modulators.
1.3.4. Characterization
The main task of this thesis is focused at this step of the device processing. The
characterization in the laboratory of all electro-optic silicon-based modulators
developed and presented along the thesis has been carried out entirely at the NTC
facilities. Several types of characterization processes have been performed: passive, RF,
high speed, small signal. All of them are deeply explained in each section, as well as,
the characterization set-up and the equipment used and obviously the obtained results.
17
Chapter 1: Introduction
18
Chapter 2
Digital applications
19
Chapter 2: Digital applications
(MOS) structure. The advantage of a MOS structure is its process similar to a transistor.
This is precisely the desired property for future photonic and electronic integration. The
drawback is that the modulation based on a MOS structure is limited by the RC delay.
Furthermore, a modulator with low capacitance has a wide bandwidth but requires a
higher radio-frequency (RF) voltage to achieve the same amount of charge. Therefore,
is a trade-off between speed and voltage (and efficiency). The first modulator based on
a MOS structure was proposed and demonstrated in February 2004 by Intel [LIU04].
Moreover, this modulator was the first one based on the plasma dispersion effect. The
length of the device was 2.5mm and results showed that the optical output signal
faithfully reproduced the 1 Gbit/s electrical input data stream. The modulator efficiency
was 8V.cm and it presented a high insertion losses (IL) of 13.5dB (4.3dB per interface
coupling loss to and from the waveguide and 6.7dB on-chip loss). Later, in 2005, the
same authors achieved an increased speed of 10Gbit/s with 3.8dB of extinction ratio
(ER) [LIA05]. The modulator was 3.5mm which a better modulation efficiency of
3.3V.cm although higher IL of 19dB (10dB of on-chip loss and 9dB of coupling loss).
Further, NEC Corporation group recently demonstrated a high speed of 12.5Gbit/s
operation in case of 120µm phase-shift length, high optical modulation efficiency of
0.5-0.67V.cm and lower IL of 7.5dB, by using a projection MOS junction structure
designed to maximize the overlapping between the optical field and the carrier
modulation region by using a compact rib waveguide with highly-efficient optical
confinement [FUJ10].
Otherwise, injection modulators are usually based on forward biased PIN diode;
highly doped p- and n-regions are separated by an ‘intrinsic region’ in which the
waveguide is formed and forward-biasing the device causes free electrons and holes to
be injected into the ‘intrinsic’ waveguide region. The advantage of this type of structure
is that it can provide electrical injection of carriers to further increase the refractive
index change respect to other structures, which in turn provide much lower VπLπ.
However, there is a main drawback associated with this type of structure. It is the
modulation speed typically determined by the free carrier’s lifetime. Therefore, pre-
emphasis (PE) (which consists on apply short peaks of current at the beginning of each
electrical bit to accelerate the carrier generation process and followed by a reverse bias
operation to extract carriers more efficiently) is a commonly used technique for
successfully mitigating this fundamental limitation enabling broadband operation. In
such configuration, a 200μm MZI modulator operating at 10Gbit/s was reported in 2007
by IBM research group [GRE07]. The short modulation length is due to the low device
modulation efficiency of 0.036V.cm. However, a large measured on-chip loss of 12dB
was presented in the work. In January 2012 a PIN-diode-based silicon Mach-Zehnder
modulator with side-wall gratings on both sides of the waveguide core has been
presented by the Photonics Electronics Technology Research Association (PETRA) and
others research groups in Japan [AKI12A]. A VπLπ 0.29V.cm was obtained at
12.5Gbit/s using the fabricated modulator and the PE technique. The ultra short
modulator (250μm) showed low IL of 4.2dB compared with the IBM modulator. Latter,
in May 2012, the authors achieved a higher modulation speed of 25Gbit/s in a shorter
20
2.1. Introduction: state-of-art of plasma dispersion effect-based MZI
modulator with a phase shifter length of 100μm by optimization of the doping profile.
The modulator showed similar IL and lower dynamic ER [BAB12]. Finally, the same
groups enhanced again these results, in August 2012, showing an operation speed of
50Gbit/s in a 250μm-length modulator with similar results of both ER and IL. They
designed a doping profile of the PIN diodes to operate the MZI at a higher speed than in
their previous works [AKI12B].
Since 2004, Mach-Zehnder modulators using reverse-biased PN diodes have been the
most commonly investigated for near-future applications. In these diodes lightly doped
p- and n- type are in contact in the waveguide forming the PN diode. This mechanism is
called depletion and the depletion area of the diode becomes larger with increasing
reverse bias voltage. Among the previous mentioned, carrier-depletion on a reverse-
biased PN junction has been shown to be the best option for high speed operation and
this kind of modulator has already exhibited as high-speed operation as 50Gbit/s.
However, typically the density of carriers, mostly provided by donors and acceptors is
low and a large doping may introduce more optical loss. This change can only provide
carriers refractive index change around Δn~10-4 resulting in a large efficiency VπLπ.
This great efficiency requires a length of several millimetres of the phase shifter,
therefore a good design of travelling wave electrode is required to achieve a high
modulation speed. In 2007, Intel demonstrated the highest modulation speed until that
moment; it was 40Gbit/s with 30GHz of bandwidth in a 1mm travelling device with a
low ER of 1dB and a large VπLπ of 4V.cm. Moreover, the total on-chip optical loss was
less than 4 dB [LIA07]. From that moment, many works have been published improving
these parameters. We are going to review the modulation performance of the reported
carrier depletion MZIs with speeds up to 40GHz of the most important research groups.
In 2001, researchers of the Advanced Technology Institute in University of Surrey
demonstrated 40Gbit/s optical modulation in silicon is demonstrated with a large ER for
the first time. Phase modulators with an efficiency of 2.7V.cm were incorporated into
MZIs with 3.5mm and 1mm length arms. ERs of up to 10dB were demonstrated at
40Gbit/s from the 3.5mm MZI, with a corresponding optical loss of approximately
15dB [THO11]. On the other hand, the most recent modulator presented by this research
group achieves a modulation speed of 50Gbit/s also in a 1mm travelling device
[THO12]. The bandwidth showed by the modulator is 40GHz and a dynamic ER of
3.1dB. With the progress in the fabrication processes, in recent years some works have
presented with optimized electrical structures based on reverse biased carrier depletion.
For example, in April 2012, the research group Institut d’Electronique Fondamentale
(IEF) showed a reverse biased PIPIN modulator operating at 40Gbit/s. Two devices
were presented, one with an ER of 3.2dB and an optical loss of 4.5 dB, and another one
with an ER of 6.6dB and an optical loss of 6dB. The bandwidth of the larger was
20GHz and both showed an efficiency of 3.5V.cm [ZIE12]. Also in 2012, Bell Labs. &
Alcatel-Lucent research group reported a 2 mm-long device, whose switching speed
reached 50 Gbit/s with an ER of 4.7 dB [DON12]. Finally, very recently, the State Key
Laboratory on Integrated Optoelectronics in China demonstrated in February 2013 a
very high speed of 60Gbit/s in a 750µm MZI modulator with doping optimization of the
21
Chapter 2: Digital applications
PN junction [XIA13]. The next Table 2-1 summarizes the above mentioned metrics of
the silicon-based MZI modulators reported.
Dyn. Insertion
Electrical Research VπLπ BW Data rate [Gbit/s] Length
[ref] ER losses
mechanism group [V.cm] [GHz] @Vpp [µm]
[dB] [dB]
[LIU04] 8 2.5 1 2500 15.3
Carrier Intel
Accumulation [LIA05] 3.3 10 10@3.5 3.8 3450 19
(MOS)
NEC Corp. [FUJ10] 0.5-0.67 - 12.5@3.5 3-6 120 7.5
IBM [GRE07] 0.036 - 10@7 - 200 12
Table 2-1: The most important parameters of the last published MZI modulators based on the
plasma dispersion effect, (*1) for the 4700µm length, (*2) BW6dB, (*3) predicted, (*4) at
50Gbit/s.
Therefore, it is clear that there is a clear trade-off in conventional MZM between the
drive voltage, modulation speed and IL. On one hand, high modulation speeds are
achieved by minimizing the modulation length, which also contributes to reduce IL, but
increases the required drive voltage. On the other hand, low drive voltages are achieved
by increasing the modulation length which penalizes the IL and modulation speed.
Overall, although tremendous efforts have been carried out over the last years, the trade-
off between, IL, high speed and drive voltage inherent to silicon-based modulators
seems to persist.
22
2.2. Micro-ring resonator structures
This is the reason for that new optical structures different to the MZI one have been
proposed in the last years improve the modulation efficiency or device footprint. Part of
the work to overcome such issues is to find a way to increase the interaction between
charges carriers (refractive index variation) and the propagating optical mode. Two of
the most promising solutions for such purpose are to take advantage of slow-wave and
resonant structures modulation enhancement, which allow for more compact and less
power consumption electro-optic modulators.
In this way, the objective of this chapter is to develop silicon modulation structures
in order to enhance the performance of conventional MZI modulators and to
demonstrate that the incorporation of optical phase modulators into slow-wave and
resonant structures can offer advantages in both reduced device footprint and power
consumption in terms of required drive voltage meanwhile high speed operation is
maintained and offer the prospect of compact optoelectronic devices integrated with
CMOS circuits on the same silicon substrate.
external voltage and device size. Modulation based on the intensity and phase response
of a ring resonator is considered and analyzed. In the latter, phase modulation is
converted to an intensity modulation by using a Mach-Zehnder interferometer structure,
which provides a more flexible structure as the ring resonator just needs to be over-
coupled, thus not requiring critical coupling, which in turn offers a higher robustness
and the potential for achieving a higher modulation speed.
This method for modulating was also used by Kotura Inc. & Sun Labs, which in May
2010 obtained a data rate of 12.5Gbit/s with a ring radius of 5μm and a driving voltage
of 3Vpp to achieve ~6dB of extinction ratio. The modulator consisted on an asymmetric
lateral PN junction [DON10A]. Later, in October 2010, they presented a silicon
racetrack resonator modulator with radius of 10.4μm and a straight coupling length of
5μm which achieves over 8dB of modulation depth at also 12.5 Gbit/s with a lower
driving voltage of 1V. The design improvement was the PN junction embedded along
the entire circumference of the resonator, including the coupler part of the resonator and
the bus waveguide. The modulator efficiency was of 0.71V.cm [DON10B].
Also in 2010, Sun Labs & Luxtera demonstrated, for the first time, a flip-chip
integrated all-CMOS photonic modulator with driver using a carrier-depletion ring
modulator and CMOS driver circuits. It showed a 3dB ER and was operated error-free
at a data rate of 5Gbps. Small signal measurements of the CMOS ring modulator
showed a 3dB bandwidth in excess of 15GHz at 1V of reverse bias, indicating that
further increases in transmission rate is possible while retaining compatibility with
CMOS drive voltages [ZHE10].
In 2011, the Oracle Labs group published a work reporting a very efficient silicon
ring modulator with a ring radius of 7.5μm. Its high-speed section, employing PN
junction working in carrier-depletion mode, enabled a very high speed operation of
25Gbit/s modulation and an ER >5dB with only 1Vpp driving [LI11].
24
2.2. Micro-ring resonator structures
Table 2-2: State-of-art of ring resonators-based silicon electro-optical modulators, (*) Racetrack
add-drop configuration with a 300.8μm and 67.4 μm circumference, (PE) Pre-emphasis.
The basic configuration, which consists of a ring resonator (resonant at a wavelength λi)
coupled to a straight waveguide, is shown in the Figure 2-1. Key parameters are the
separation between waveguide and ring in the coupling region (d), the ring radius (R)
and the losses in the ring in dB/cm (α). If the wavelength, for example, λi satisfies the
resonant condition, that is,
neff L m i (2.1)
25
Chapter 2: Digital applications
the coupling of the wave with wavelength λi will be enhanced and all others will be
suppressed. As a result, only λi will resonate inside the ring, while the rest of the
wavelengths will pass through and output. In this way, a ring resonator modulator can
be consider as a Notch filter. Here neff is the effective index of the bending waveguide,
L=2πR is length of the ring, and m is an integer.
Figure 2-1: Ring resonator in Notch filter configuration and the most important design
parameters.
Figure 2-2: Model of a single ring resonator with coupled to one waveguide.
Defining that a single unidirectional mode of the resonator is excited, the coupling is
lossless, single polarization is considered, and taking into account that the different
kinds of losses in the ring resonator are incorporated in the attenuation constant, the
interaction can be described by the matrix relation:
Et1 t Ei1
*
(2.2)
Et 2 t* Ei 2
The complex mode amplitudes are normalized, so that their squared magnitude
corresponds to the modal power. The coupler parameters t and κ depend on the specific
coupling mechanism used. The “∗” denotes the conjugated complex value of t and κ,
respectively. The matrix is symmetric because the networks under consideration are
reciprocal. Therefore:
2
t2 1 (2.3)
26
2.2. Micro-ring resonator structures
In order to simplify the model, Ei1 is chosen to be equal to 1. Then the round trip in the
ring is given by:
Ei 2 e j Et 2 (2.4)
where α is the loss coefficient of the ring (zero loss: α=1) and θ=ωL/c, being L the
length of the ring which is given by L=2πR, where R is the radius of the ring measured
from the center of the ring to the center of the waveguide, c the phase velocity of the
ring mode (c=c0/neff) and the fixed angular frequency ω=k.c0, c0 refers to the vacuum
speed of light. The vacuum wave-number k is related to the wavelength λ through:
k=2π/λ. Using the vacuum wave-number, the effective refractive index neff can be
introduced easily into the ring coupling relations by:
2 neff
k neff (2.5)
L kc0 L 2 neff 2 R R
k neff 2 R 4 2 neff (2.6)
c c
*
Et 2 (2.9)
1 t *e j
This leads to the transmission power Pt1 in the output waveguide, which is:
2 2
2 t 2 t cos t
Pt1 Et 2
(2.10)
2
1 t 2 t cos t
where t=|t|exp(jφt), |t| representing the coupling losses and φt the phase of the coupler.
2 2
2
1 t
Pi 2 Ei 2 2 2
(2.11)
1 t 2 t cos t
27
Chapter 2: Digital applications
A special case happens when α=|t| in Equation (2.12), when the internal losses are
equal to the coupling losses, so the transmitted power equals zero, which is known in
literature as critical coupling, due to destructive interference. By using the above
equations, the behaviour of a basic ring resonator filter configuration consisting of only
one waveguide and one ring can be taken. This model can be extended to suit the
requirement of various types of ring resonator configurations.
Ring resonator filters can be described by some figures of merit which will be
explained in the next paragraphs. The first figure is free spectral range (FSR), which
represents the distance between resonance peaks. By using the propagation constant
from Equation (2.5), neglecting the wavelength dependency of the effective refractive
index, a simple approximation can be obtained for the FSR
neff
k (2.14)
which leads to the FSR=Δλ expressed in Equation (2.15), which is the difference
between the vacuum wavelengths corresponding to two resonant conditions.
1 2
2
FSR (2.15)
L neff L
Note that the equation above is for the resonant condition next to a resonance found for
the used propagation constant. If the wavelength dependence of the effective index is
not neglected, we will obtain a modified version of Equation (2.14):
k
ng (2.16)
neff
ng neff (2.17)
28
2.2. Micro-ring resonator structures
The group refractive index can be used instead of the effective index in order to obtain
accurate values when avoiding approximation is important. The modified FSR ∆λ is
then given by:
2
FSR (2.18)
ng L
The next parameter of importance is the resonance width which is defined as the full
width at half maximum (FWHM) or 3 dB bandwidth (2δλ) of the resonance line-shape.
Using Equations (2.6) and (2.14) and assuming real coupling coefficients, no coupling
losses and without phase term, after some operations, we obtain that:
2
1 t2
2 (2.19)
Lneff t
The expression which is commonly used can be obtained by assuming weak coupling
and λ>>δλ:
2 2
FWHM 2 (2.20)
Lneff
The finesse F of the ring resonator filter, which is defined as the ratio of the FSR and
the width of a resonance for a specific wavelength (FWHM) is another parameter which
can now be directly calculated from the parameters defined previously is:
FSR t
F (2.21)
FWHM 2 1 t2 2
A parameter which is closely related to the finesse is the quality factor Q of a resonator,
which is a measure of the sharpness of the resonance. It is defined as the ratio of the
operation wavelength and the resonance width:
neff L t neff L
Q F (2.22)
1 t2
One key figure of merit of resonators is their bandwidth, which is inversely proportional
to the quality factor defined in Equation (2.22). While a narrow bandwidth (high Q) is
suitable for enhancing light–matter interactions such as lasing [REI06] and sensing
[ARM06], a wide bandwidth (low Q), on the other hand, allows for optical signals of a
broader spectrum and higher speed [LEE06] and is more tolerant to variations in the
environment.
Once the most important parameters of a MRR have been presented we are going to
explain the basic performance of a MRR based modulator in silicon. The refractive
29
Chapter 2: Digital applications
index in the ring can be modified by means of the plasma dispersion effect. The change
of the real part of the refractive index induces a change in the effective refraction index
which shifts the spectral response of the ring. This property can be used to implement
an intensity modulator. Basic performance of an intensity modulator based on a single
ring resonator is shown in Figure 2-3. A continuous wave (CW) of a certain wavelength
is injected into the ring resonator. Minimum transmission is achieved when the
wavelength is located at a resonance. Furthermore, this transmission is ideally zero
when the ring operates at critical coupling, i.e. when the coupled power is equal to the
power lost in the ring (critical coupling condition) as it has been explained previously:
The spectral response of the ring resonator is shifted when the effective index is
modified. Thus, the output power increases up to a maximum value when the frequency
shift is enough to move the resonance out of the working wavelength. Therefore, to
optimize the performance of the intensity modulator, we are interested in the lowest
transmission at resonance to maximize the extinction ration and the narrowest spectrum
to minimize the required frequency shift, i.e. the required external voltage.
Figure 2-4 shows the full-width at half maximum (FWHM) and transmission of a
ring resonator as a function of ring losses and power coupling and taking into account
different radii. In the upper plots, power coupling is chosen for each radius to operate at
critical coupling while in the lower plots losses are 10dB/cm. The operating wavelength
is 1550nm and the effective index is 2.76.
It can be seen that the FWHM increases as the losses (Figure 2-4(a)) and coupling
(Figure 2-4(b)) increase. Furthermore, it is also interesting to notice that the FWHM is
lower for higher radii and that it converges to a constant value. This constant value is
determined by the FSR, which is higher for higher radius, so the FWHM cannot be
increased more than the FSR value.
On the other hand, the transmission as a function of losses is the same for the three
different radii, as shown in Figure 2-4(c), because the power coupling is changed in
each case to operate at critical coupling considering 10dB/cm losses. The different
power coupling value for each radius is shown in the transmission plot as a function of
the power coupling in Figure 2-4(d). In both cases, it can also be seen that the variation
30
2.2. Micro-ring resonator structures
of the transmission with power coupling and losses is more abrupt, and therefore
undesired for the intensity modulator, when the ring is smaller and the ring is under
coupled, i.e. when the coupled power is lower than the power lost in the ring.
Figure 2-4: FWHM and transmission of a ring resonator as a function of ring losses and power
coupling and taking into account different radii. In the upper plots, power coupling is chosen for
each radius to operate at critical coupling while in the lower plots losses are 10dB/cm.
In summary, we should work with big radii and close to the critical coupling point
for optimizing the performance of the intensity modulator, i.e. maximum extinction
ratio and minimum frequency shift. Furthermore, a ring with a higher radius is more
robust against coupling variations than a ring with a lower radius.
The modulator consists of a MRR embedded in a PIPIN diode structure and coupled
to a straight waveguide, as shown in Figure 2.5(a), fabricated on a silicon-on-insulator
(SOI) wafer. The ring and input waveguide of dimensions of 420nm x 390nm have been
partially dry etched to leave a 100nm thick slab as shown Figure 2-5(b). As it can be
observed, a p-doped slit has been inserted in the intrinsic region in order to act as a
source of holes. When a reverse bias is applied on the diode, holes are swept out of the
central p-doped region by carrier depletion, leading to a refractive index change. Figure
2-5(c) shows a scanning electron microscope (SEM) picture of some of the fabricated
structures and Figure 2-5(d) depicts a zoomed-in view of a MRR. Doping
31
Chapter 2: Digital applications
concentrations in the -p and -n parts of the diode are 8.1017 and 1.1018cm-3 respectively.
Within the 100nm-wide p-doped slit, doping concentration reaches 3.1017cm-3. The
lumped electrodes, placed 0.5µm away from the rib edges to minimize the overlap
between the optical mode and so the optical losses, are formed by highly doped p+ and
n+ regions with concentrations of 1.1019cm-3 in order to decrease the access resistance
between the active region and the metal contacts. For the same reason, the waveguide
was only etched in an 800nm-wide region on each side of the rib [RAS11A, RAS11B].
Figure 2-5: (a) Schematic of the MRR silicon-based electro-optic modulator, (b) cross section of
the device active region with a PIPIN diode (c) top view of some of the fabricated MRR
modulators and (d) close up view of the 40µm radius MRR.
2
L neff (2.23)
where λ is the operation wavelength, L is de ring length (L=2πR) and Δneff is the
effective index change due to the plasma dispersion effect.
The optical modulator is fabricated on a 200mm SOI wafer with a 2 μm-thick buried
oxide layer and a 400nm-thick crystalline silicon film. Deep-UV optical lithography and
ion implantation is used to obtain the highly doped p+ and n+ regions. For hard mask
purpose, SiO2 is then deposited by LPCVD. The waveguides and the slits are patterned
using deep-UV lithography. After hard mask etching by reactive ion etching, Boron
implantation (p in the slit) is performed to obtain the doped layer in the middle of the
waveguide and n and p doped regions are then obtained by lithography and ion
implantation, followed by thermal annealing. Finally, metal stack is deposited onto the
32
2.2. Micro-ring resonator structures
wafer, and the electrodes are patterned and etched down to the SiO2 cap layer. The
process used is fully compatible with SOI CMOS technology and could be transferred
in high-volume microelectronic manufacturing.
We first carried out passive measurements with the aim of designing the optimum
parameters of the MRR structures around the operating wavelength of 1550nm. Optical
measurements have been performed with an external cavity laser (ECL) having tuneable
range from 1500 to 1580nm. Light was butt-coupled to the chip via a lensed fiber
(details of horizontal coupling and passive characterization can be found in appendix
A).
Figure 2-6: (a) Experimental and (b) simulated ER as a function of the ring to waveguide
separation.
33
Chapter 2: Digital applications
Therefore, although there is a slight discrepancy for the optimum separation for the
ring with R=40µm probably due to fabrication deviations, in general a good agreement
between experimental and simulation results is achieved. Finally, a MRR with big
radius (R=40µm) and waveguide to ring separation of 200nm point has been chosen for
optimizing the performance of the intensity modulator, i.e. maximum extinction ratio
and minimum frequency shift required. Furthermore, a ring with a higher radius is more
robust against coupling variations than a ring with a lower radius.
Once the optimum MRR parameters were designed based on previous passive
measurements, the ring modulator was characterized under static and high speed
regimes. Static measurements for varying applied voltages were realized to estimate the
DC ER as well as the modulation efficiency deduced from the achieved phase shift.
High speed tests were carried out to evaluate the MRR modulator performance. In both
cases the diode was electrically biased via 40 GHz electric probes. The spectrum of the
MRR with 40µm radius shows a free-spectral range of 2.5nm and a 9dB drop in
transmission at the resonant wavelength around 1556.17nm. The 3-dB bandwidth of
such resonance ∆λ is 0.075nm measured from the spectrum, corresponding to a quality
factor of around Q~20750 (from Equation (2.22)).
Figure 2-7: Transmission spectrum as a function of the wavelength for different applied bias
voltages.
Figure 2-7 shows transmission spectrum under different bias voltages. It can be
observed that a DC ER of 4dB is achievable at a bias voltage of 5V while it improves
up to 7dB when the bias voltage increases to 10V.
34
2.2. Micro-ring resonator structures
On the other hand, the voltage-induced wavelength shift (∆λ) in the ring resonator
modulator output spectrum can be used to extract the effective index variation (∆neff) as
follows:
neff (2.24)
FSR·LA
where LA is the active length that, in our case, is slightly smaller than the MRR
perimeter (LA=207µm<LMRR=251µm), λ is the operation wavelength and FSR is the free
spectral range.
Figure 2-8 shows the effective index variation as a function of applied bias voltage for
the measured MRR.
A modulation efficiency VπLπ of 2.8V.cm has been calculated from such effective
index variation by means of Equation (2.25):
L (2.25)
2 neff
Finally, high speed performance of the silicon ring modulator was characterized by
measuring the data transmission capability by using the set-up shown in Figure 2-9. The
ring modulator with 40µm radius was driven with a nonreturn-to-zero (NRZ)
pseudorandom bit sequence (PRBS) of length 231-1 delivered by a bit pattern generator
(BPG), which is connected to an external clock. The electrical signal is amplified
through a high-speed RF amplifier to achieve the desired voltage swing (Vpp=8V) and
combined to a -4V DC bias voltage (in order to ensure reverse bias operation) using a
bias-Tee. The modulating signal is applied by a high-speed ground-signal-ground
(GSG) RF probes to the lumped electrodes, which are not required to be terminated
externally by a 50Ω load. The output modulated optical signal is then optically
35
Chapter 2: Digital applications
amplified by an erbium-doped fiber amplifier (EDFA), filtered via an optical filter (OF)
and finally photo-detected by a digital communication analyzer (DCA). Details of
model equipments and pictures of the real set-up are given in section 2.4.
Figure 2-9: Schematic of the characterization setup for the data transmission measurements.
Figure 2-10 shows the obtained eye diagrams. The ER is 4.5dB at bit rates of 5Gbit/s
and 10Gbit/s and 3.5dB at 12.5Gbit/s.
Figure 2-10: Optical eye diagrams with an ER of 4.5dB at 5Gbit/s and 10Gbit/s and with an ER
of 3.5dB at 12.5Gbit/s.
36
2.2. Micro-ring resonator structures
In conclusion, we have presented a high speed 40µm radius ring resonator modulator
in which the modulation is achieved by carrier depletion effect in a PIPIN diode
structure. The modulator exhibited an ER of 4.5dB at a data transmission of 5Gbit/s and
10Gbit/s and 3.5dB at 12.5Gbit/s with a drive voltage of 8Vpp.
The phase response of the MRR can also be used to achieve phase modulation. This
phase modulation is converted to an intensity modulation by using a MZI structure, as it
is shown in Figure 2-11(a).
Figure 2-11: (a) Schematic of the RAMZI modulator with the parameters of the MRR and (b)
Basic performance of a phase modulator based on a MRR.
37
Chapter 2: Digital applications
To explain the basic performance of the RAMZI optical structure, which is depicted
in Figure 2-11(b), it is assumed that the input signal is a CW located at the MRR
resonance. The side-coupled MRR induces a π-phase shift to the signal that travels
through one arm of the MZI which will interfere destructively with the signal that
propagates through the other arm of the MZI. This π-phase shift is thus responsible for
minimum transmission at the MZI output. When the spectral response of the MRR
resonator is shifted due to the effective index variation and the frequency shift is enough
so that the operation wavelength is now out of the resonance, the MRR will not
introduce any additional phase shift and the signals at both arm ends will interfere
constructively, leading to maximum transmission at the MZI output.
The ER of the modulator is ideally infinite if the attenuation at both arms of the MZI
is the same. However, it is worth noting that the intensity response of the MRR
resonator will also vary when the effective index is modified (assuming losses are not
negligible, which is usually the case). This will introduce additional attenuation to the
signal travelling through the arm with the MRR at resonance, which could in turn
deteriorate the ER of the modulator due to the power unbalance between both arms of
the MZI. This power unbalance will be lower as the power coupling factor, i.e. the
fraction of power coupled from the waveguide into the MRR, increases. Therefore, the
MRR does not need to be operated at the critical coupling point and just over-coupled.
However, it should be realized that too high power coupling factors could increase the
IL as the change of the MRR phase response for a given effective index variation will
be smaller.
Figure 2-12: Transmission at the MZI output as function of the difference in attenuation between
arms when the MZI is operating at (a) minimum transmission (π-phase shift) and (b) maximum
transmission (0-phase shift).
Figure 2-12 shows the transmission at the MZI output as function of the difference in
attenuation between arms when the MZI is operating at (a) minimum transmission (π-
phase shift) and (b) maximum transmission (0-phase shift). First of all, it can be seen
that at minimum transmission, the performance of the MZI is much more sensible to
attenuation differences than at maximum transmission. As the attenuation increases, the
minimum transmission is higher while the maximum transmission is lower so the ER
38
2.2. Micro-ring resonator structures
will decrease. For instance, if the attenuation difference is 3dB, the ER is degraded to
9.57dB.
The RAMZI modulator consists of a MRR embedded in the same PIPIN diode structure
(Figure 2-13(b)) with the same doping levels than previous MRR modulator but this
time coupled to one arm of a MZI, as shown in Figure 2-13(a). The MZI consists of two
arms of equal lengths of 215µm with 3dB star coupler splitters and combiners. Figure 2-
13(c) shows an optical microscope image of the fabricated device. The overall footprint
of the modulator without probe pads is 680µm x 125µm.
Figure 2-13: (a) Schematic of the fabricated ring-assisted MZI modulator, (b) cross section of
the device active region, (c) top view of fabricated modulator.
The fact that the MRR does not need to be operated at the critical coupling factor
(which we have already explained in the previous point) implies that modulator
robustness against fabrication deviations could be significantly improved with respect to
standalone MRR based intensity modulators. Fabrication deviations could lead to both
variations on the MRR losses as well as variations in the designed separation between
the waveguide and the MRR and therefore in the power coupling factor. Figure 2-14
shows the theoretical ER as a function of the coupling factor and the ring losses for the
standalone MRR based intensity modulator and taking into account different effective
index variations. The MRR radius is 40μm. The same plot is shown in Figure 2-15 for
the RAMZI modulator. IL, i.e. losses at on-state, are lower than 5dB in the values
shown in Figures 2-14 and 2-15. Results in such figures were calculated by using the
transmission Equation (2.10) of the ring resonator.
Comparing the results shown in Figure 2-14 and 2-15 the enhanced performance
achieved by the RAMZI modulator can be clearly seen. High ER values in the
standalone MRR modulator are only achieved when there is a correspondence between
the MRR losses and the coupling factor, i.e. the critical coupling condition, and
therefore undesired variations in the MRR losses or coupling factor significantly
39
Chapter 2: Digital applications
degrade the ER even though the effective index variation is increased. The ER in the
RAMZI modulator is much less sensitive to variations in losses and coupling factor and
is always improved for coupling values above the critical coupling condition when the
effective index variation increases.
Figure 2-14: Theoretical ER as a function of the coupling factor and the ring losses for the MRR
based intensity modulator for different effective index variation. IL are lower than 5dB for the
values shown, where the ER is higher than 5dB (dark blue), 10dB (light blue), 15dB (yellow)
and 20dB (brown).
Figure 2-15: Theoretical ER as a function of the coupling factor and the ring losses for the
RAMZI modulator for different effective index variation. IL are lower than 5dB in the values
shown, where the ER is higher than 5dB (dark blue), 10dB (light blue), 15dB (yellow) and 20dB
(brown). The dotted red line shows the critical coupling condition.
40
2.2. Micro-ring resonator structures
The fabricated RAMZI modulator has a 1µm coupling length and 150nm waveguide
to MRR separation dimensions. The power coupling factor has been calculated to lie
around 0.3 for such coupling length and waveguide to MRR separation based on 2D-
FDTD simulations and the effective index method. Such value has been chosen as a
trade-off between large robustness and high extinction ratio-low IL performance.
Figure 2-16: Transmission spectrum of the RAMZI modulator as a function of the wavelength
for different applied bias.
The spectrum of the RAMZI exhibits a FSR of 2.4nm and a 10dB drop in
transmission at the resonant wavelength around 1550.6nm. Figure 2-16 shows the
RAMZI transmission spectrum for varying bias voltages. The voltage-induced
wavelength shift (∆λ) in the RAMZI modulator output spectrum can be used to extract
the effective index variations (∆neff) using the Equation (2.24) with LA=208µm
(<LMRR=253µm). Figure 2-17(a) depicts the deduced effective index variations as a
function of applied bias voltage. The VπLπ product has also been calculated from such
effective index variations through Equation (2.25). Its evolution as a function of the
41
Chapter 2: Digital applications
applied voltage is shown in Figure 2-17(b). Nearly constant VπLπ around 4.3V.cm for
reverse bias voltages ranging from 2V to 10V is found.
Figure 2-17: (a) Modulation efficiency as a function of applied bias voltage and (b) Effective
index variation as a function of applied bias voltage.
Figure 2-18: Simulated and experimental RAMZI ER and IL versus wavelength under an
applied voltage of 8V.
The high speed performance of the RAMZI modulator has been characterized by
measuring both its 3dB frequency roll-off and data transmission capability. Figure 2-19
shows the experimental set-up for measuring the high speed optical frequency response.
42
2.2. Micro-ring resonator structures
Light is coupled from an ECL to the chip via a lensed fiber after polarization control.
The small electrical signal delivered by the RF port 1 of the vectorial network analyzer
(VNA) is coupled a -4V DC bias via a bias-Tee and applied to the lumped electrodes
through high speed GSG probes to the RAMZI modulator. The output modulated signal
is amplified with an EDFA and photo-detected by a photodiode connected to the port 2
of the VNA in order to extract the electro-optical S21 transfer function. Details of model
equipments and pictures of the real set-up are given in section 2.4.
Figure 2-19: Experimental setup for measuring the high speed optical frequency response of the
RAMZI modulator.
The resulting data is given in Figure 2-20, which shows that the RAMZI modulator has
a 3-dB roll-off frequency as high as 19GHz so the fabricated device displays a
promising figure in terms of frequency of operation.
Figure 2-20: Electro-optical response of the RAMZI modulator. It is show a 3dB roll-off
frequency of 19GHz.
43
Chapter 2: Digital applications
It should be noticed that over-coupling the MRR makes the modulator faster when
cavity photon lifetime is the limiting factor of the modulation speed. In our case, the
cavity photon lifetime is around 15ps and the measured 3-dB modulation bandwidth is
19GHz, which are in good agreement with the theoretical results reported in [BEA10].
To characterize the high speed data transmission performance, the RAMZI modulator
was driven in a lumped configuration in the same way than the MRR modulator (set-up
in Figure 2-9). A -4V DC bias was applied to the modulator to ensure reverse bias
operation over the entire voltage swing. Figure 2-21 shows the obtained eye diagrams.
The ER is 6.7dB at a bit rate of 10Gbit/s and 2.6dB at 20Gbit/s. IL at on-state, which
include losses at the input/output couplers and propagation losses outside the MRR, are
around 6dB. Despite the lower ER at 20 Gbits/s, which is attributed to the limited
bandwidth (12GHz) of our driver amplifier, the 3-dB roll-off frequency of 19GHz
shown in Figure 2-20, indicates that our modulator is capable of transmitting data above
20Gbits/s.
Figure 2-21: Eye diagram of the RAMZI modulator with an ER of 6.7dB at 10Gbit/s and 2.6dB
at 20Gbit/s.
Overall, we have reported a high speed silicon RAMZI modulator fabricated with a
standard CMOS-compatible process. The RAMZI modulator features the potential for
high speed performance and higher robustness against fabrication deviations than its
standalone MRR based intensity modulator counterpart as the MRR does not require
being at the critical coupling operation point to achieve enhanced performance in terms
of ER and IL. In our device, modulation is achieved by carrier depletion in a PIPIN
diode structure over a RF/optical interaction length of only ~200µm featuring a 3-dB
electro-optical bandwidth as high as 19GHz and a data transmission rate up to 20Gbit/s.
44
2.3. Slow-wave structures
So, the chapter is structured as follows. First of all, a review of state-of-the art of
modulators based on slow-wave structures is provided. Second, the basic theory of the
slow-light effect is briefly introduced as well as the basic design procedure of the slow-
wave waveguide used latter for the experimental modulators. The proposed slow-wave
corrugated waveguide was originally proposed in [BRI11B] for enhancing modulation
and this point is a simple summary of the design procedure followed in such work. The
next point is devoted to the characterization of two silicon-based slow-light Mach-
Zehnder Interferometer (SL-MZI) modulators, one showing a high speed of 40Gbit/s
[BRI12B] and the other a low-drive voltage operation [BRI12A].
Finally a method for measuring the propagation losses, since it is one of the main
drawbacks derived by the use of the slow-light structures, is developed. Such method is
proposed and demonstrated experimentally to be independent of the fiber-waveguide
coupling losses, so that identical coupling conditions for each measurement were not
required. The method is based on the analysis of the transmission spectra of asymmetric
MZIs, which enables the calculation of the propagation losses as well as the power
imbalance of the branching structure of a MZI. The main contribution of the author is
the proposed method for measuring the propagation losses, which was originally
proposed and developed for conventional rib waveguides [GUT12B] and then it was
extended for corrugated waveguides [BRI12A] in collaboration with A. Brimont.
45
Chapter 2: Digital applications
The first silicon electro-optical modulator based on photonic crystals (PhCs) was
demonstrated in 2005 [JIA05] and showed low speed operation (300KHz) under a drive
voltage of 7.5mV and for an active length of 80μm. The modulation was achieved via
free holes injected through a PIP diode integrated into a MZI. Two years later, the same
group followed up on these results [GU07] and reported high-speed operation using a
PhC-based MZI having similar dimensions. The active element is made up of a PIN
diode in one arm of a MZI. The authors reported a modulation bit rate of 1Gbit/s with
ER of ~3dB. Another PhC-based modulator embedded in a capacitor was also
demonstrated [CHE09]. The device features a figure of merit of 0.18V.cm, which means
that for a 6V applied voltage, a π-phase shift is achieved over a length of 300µm. This
reduced interaction length is approximately one order of magnitude lower than for
conventional rib waveguide MOS and depletion modulators. The modulator was driven
at 1.6Gbits/s with 3dB ER and with ng~50. In 2011, pre-emphasized 10Gbits/s operation
with 7.2 dB ER in a 200 µm-long PhC modulator embedded in PN junction has been
reported [NGU11] in spite of a 3dB roll-off frequency of 3GHz. The device feature
modulation efficiencies of VπLπ~0.056V.cm and 0.016V.cm when respectively operated
in both depletion and injection regimes and only in the injection regime at a group index
of ~18. In 2012, the same authors reported the demonstration of 40 Gbit/s modulation in
a sub-100μm MZI with an acceptable drive voltage of 5.3Vpp [NGU12B]. Meanwhile,
researches of this group, with a high experience in slow-light phenomenon [BRI09,
BRI10], demonstrated the first carrier depletion-based slow- wave modulator as a means
to enhance the modulation efficiency [BRI11A] The device is based on a 500 µm long
periodic laterally corrugated waveguide (LCWG) embedded in a self aligned PN
junction [THO10]. It combines the attractive properties of slow-light propagation in a
nanostructured periodic waveguide together with a high speed semiconductor PN diode,
and exhibits error free modulation up to 20 Gbit/s with 6.3dB ER at group index of
ng~11. The longer phase shifter features 10Gbit/s error free operation with 9.5dB ER at
the same group index. These results are supported by modulation rate capabilities
reaching 40 Gbit/s (ER~3dB). Modulation efficiencies are VπLπ ~1.27V.cm is obtained
for a group index of 11 and as low as ~0.45V.cm when the group index increases up to
~22, which is respectively 2 and 5 times greater than in a 400nm reference rib
waveguide modulation. The following Table 2-3 summarizes the all these results of the
slow-wave modulators reported.
46
2.3. Slow-wave structures
As we has commented in the Introduction chapter, in the SOI technology the high
refractive index contrast between silicon (n=3.5) and silica (n=1.45) enables the
realization of integrated waveguides with submicron cross-sections. The extreme light
confinement in SOI optical waveguides paved the way toward the realization of ultra-
dense photonic integrated circuits on a single silicon chip but may result in higher
propagation losses due to the increased interaction of the optical mode with sidewall
roughness [LEE00]. This issue is, in fact, the main limitation of slow-wave based
devices as increasing the group index and therefore decreasing the group velocity will
lead to higher losses.
From 1980 to 1990 these were usually high; about 5-20dB/cm [SOR86], based on
ridge waveguides modulators, and have been reduced to less than 1dB/cm [GAR05]
based on buried waveguides modulators after 2000. Because high losses are prohibited
for the majority of applications, smoothing techniques have been successfully
developed to reduce light scattering [LEE01]. During years, numerous methods have
been developed to estimate this critical parameter. Some of them, such as the two-prism
and three-prism methods [WEB73, WON80, OKA83], consist in measuring the light
scattered above the waveguide under study. The advantage of these techniques is that no
constant coupling conditions are required. However, the quality of the wafer surface is
crucial, which limits therefore the measurement accuracy. Other methods are based on
measuring the absorbed optical power [TIT93], but in this case, the coupling efficiency
is critical. Moreover, the well-known cut-back method, which is based on a comparison
of the transmission through waveguides of different lengths, has been widely employed
owing to its ease of use. However, it requires identical coupling conditions, which is
difficult to achieve in practice [VLA04, REE92]. As an alternative, the Fabry–Perot
method and modified versions have been proposed to estimate the propagation losses
independently of the coupling losses. Nevertheless, this method is based on the use of
reflective facets, which are avoided in practical devices where high efficiency coupling
techniques are precisely used to reduce the facet reflections and improve coupling
[REG85, ROE11].
47
Chapter 2: Digital applications
L 1 n vg
(2.26)
2 n c
This equation shows that for a given perturbation of the refractive index (δn) the size of
the device scales linearly with the group velocity of light (vg). The group velocity of
light, defined by vg=c/ng, where ng is the group index, provides an estimation on how
much the group velocity of a wavepacket propagating through a dispersive material is
reduced compared to the speed of light in a vacuum c. Another important parameter is
the overlap factor (σ) which stands for the fraction of the total energy of the mode that is
stored in the region where is being applied. Therefore, for instance, for δn=10-3, n=3.47
and λ=1550nm, size device can be decreased below 100μm if group velocities lower
than 0.04c (ng=25) can be achieved and σ=1.
Following, we are going to define some basic concepts which are necessary to know
and understand the slow-light effect:
Bands diagram: The structures are characterized by periodic bands diagram (or
dispersion relation) which is a 2D or 3D representation of the relationship
between the frequency (normalized units ωΛ/2πc being 'Λ' period) and the wave
vector (k normalized units/π). Another typical definition of the bands diagram is
a representation of the frequency or energy of the allowed modes in a periodic
structure as a function of wave vector.
Brillouin Zone: The importance of the Brillouin zone stems from the Bloch
wave description of waves in a periodic medium, in which it is found that the
solutions can be completely characterized by their behaviour in a single
48
2.3. Slow-wave structures
Brillouin zone. So, due to the periodicity of the structure, the band diagram is
repeated with a period Gz=2π/Λ, therefore, it suffices to represent the wave
vectors in the space -Gz/2<k<Gz/2 (-π/Λ<k<π/Λ) and this region is called
Brillouin zone.
Once established the theoretical basis, we already are able to explain the basic operation
of the slow light effect. The form in which SOI technology produces slow light “on-
chip” is via the use of in 1, 2 or even 3 dimensional periodic structures, where the
dielectric periodicity is usually achieved via ordered nano-structuring [VLA05]. Indeed,
slow-wave structures, due to their periodic nature, present the great advantage of
slowing down the propagation of light. So, incorporating these slow-wave structures in
optical modulators can provide advantages in the reduced length of the device as well as
in the power consumption, all this determined by such reduction in the group velocity.
The period, size and shape of the repeated pattern of such periodic nano-structure define
the frequency at which the speed of light theoretically drops inside the waveguide, thus
providing complete liberty for the designer. The nature of slow light, which consists in
the coupling in phase and amplitude between successive back and forth reflections
experienced by the propagating wave at each unit cell, can intuitively be described with
the ray optics theory in mind. Here, we focus on a 1 dimensional (1D) laterally
corrugated periodic structure (LCWG), although the slow wave propagation principle
remains the same for the other types of PhCs. A typical band diagram of a deep-etched
1D LCWG is depicted in Figure 2-22(a). There are two slow light propagation
mechanisms: by omnidirectional reflection or backscattering. Here, we are interested in
the last mechanism which is produced in the band edge (k=0.5*2π/Λ, Λ≈λ/2).
Figure 2-22: (a) Band diagram of a 1D laterally corrugated waveguide. The blue solid lines
represent the first three modes of the optical structure of which only those located below the
light line (green dashed) propagate through the optical structure theoretically without losses, (b)
Slow light propagation by backscattering [BRI11B].
49
Chapter 2: Digital applications
From such physical insight, one may think that slow light occurs only at the edge of
the Brillouin zone in a reduced bandwidth which is one of the two main drawbacks
associated to the slow-light effect. However, recent works on 2D PhCs with line defect,
that is a 2D PhCs with a single line of missing hole, [FRAN06, KUB07, SET07] refuted
this hypothesis showing that slow light may also be achieved inside the Brillouin zone
over a wide bandwidth by increasing the coupling between modes of different orders
naturally interacting with each other. These so-called engineered PhCs are of great
interest for electro-optic modulation applications owing to their artificially increased
optical bandwidth.
The possibility of obtaining low group velocity values is one of the most attractive
features of the structures possessing periodicity in the propagation direction. A direct
application of this low group velocity is the possibility to create compact delay lines,
which is a fundamental element to the creation of more complex optical devices.
Besides the increasing delay experienced by the signal when propagates through the
structure, the reduction in the group velocity of the guided mode also causes a high
interaction between the electromagnetic energy of the guided mode and the structure
itself. Such increased field-material interaction will increase the efficiency of any effect
in the structure, either as the positive character as the non-linear effects in the material,
amplification, phase changes, etc., or negative as the structure heating, absorption
material, losses due to sidewall roughness, etc. So, other of the most important negative
effects derived by the use of slow-light structures is the associated propagation losses
due to roughness and positional disorder when the group velocity of the guided mode is
low. Despite the great advance in the last years of the integrated photonic devices
fabrication processes, there is a limitation in the minimum achievable roughness. Such
roughness is mainly responsible for the propagation losses in the photonic structures.
For the case of slow-wave structures, losses will also be influenced by the low group
velocity. In this case there will also be a greater interaction between the guided mode
and fabrication deviations. It is clear that, to reduce propagation losses will be necessary
to perform a high accuracy fabrication process with the aim of decreasing the
imperfections created in the sidewall waveguide. One way to achieve greater precision
50
2.3. Slow-wave structures
in the manufacturing will be using the simplest possible designs in which the limitations
of the available fabrication process are not as critical.
In summary, both PhCs and LCWGs are subject to optical propagation losses usually
inherent to the fabrication deviations (roughness, disorder...) and specific waveguide
geometry. In the fast light regime, state of the art devices were reported to feature
propagation losses as low as 4.1±0.9dB/cm [FAO06] and 0.92±0.14dB/cm [GNA08] for
PhCs and 1D deep-etched LCWG, respectively. Still, some doubts about the loss
dependency against group velocity subsist when entering in the slow light regime. Some
authors have suggested that the propagation losses have a proportional relation with
(1/vg) in the case of PhCs [KUR05]. However, other studies conclude that this
relationship is not so simple and losses are scaled approximately by a factor (1/vg)2
[HUH05] near the band edge, that is, for very low group velocity (i.e. below c/100).
And even further studies indicate that these losses are proportional to (1/vg)1/2 [FAO07].
Interestingly, the need for such very low group velocity is not mandatory required for
modulation applications due the trade-off arising between useful optical bandwidth (i.e.
the wavelength/frequency interval where the expected performance of the device is
maintained), loss and low group velocity. Namely, the group velocity dependence upon
frequency usually features a steep slope limiting the useful operating bandwidth. For
lower group velocities, i.e. as we move closer to the band edge, this restriction becomes
more severe, which clearly impairs potential applications.
51
Chapter 2: Digital applications
undesired effects above mentioned are eliminated or at least reduced. It must be stated
here that, the reason for using a shallow-etched version of the LCWG instead the deep-
etched one (without silicon slab) is because silicon modulators require contacting
regions located on both side of the waveguide to allow contact with the electrodes to
electrically tune the refractive index.
Figure 2-23: Schematic of the shallow-etched LCWG with slab height (h) considered in this
thesis to enhanced modulation [BRI11B].
As previously seen, enabling light to propagate with slow group velocity in the form
of Bloch modes is a matter of inducing multiple reflections inside any periodically
patterned waveguide. Thereby, the wide variety of architectures potentially available
should be only limited by the technological constraints, i.e. the resolution of the
lithography process. In this way, the choice of the waveguide parameters relies mainly
on the critical dimensions (CD), that is, the minimum dimensions that may be resolved
by the lithography process. Since these LCWGs are expected to be fabricated with deep-
UV lithography process, the corrugations dimensions must be selected with CD of about
100nm in order to address the fabrications issues. Then, this CD sets a lower limit on
the minimum dimension of the corrugated slow-wave waveguide.
The design method is then to set the period (Λ) of the transversal corrugations and
varying We and W dimensions of these elements with the aim of matching the slow
wave region wavelengths around 1.55μm. For such task, the Bragg condition is used.
So, we need to estimate the period at which the Bragg condition is satisfied. This
condition, for the corrugated waveguide under study, is defined by:
Bragg
Λ m (2.27)
(neff ,W neff ,We )
Where Λ is the Bragg grating period, neff,We and neff,W the effective index of the optical
mode propagating into the wide (We) and narrow sections (W), m the mode order and
52
2.3. Slow-wave structures
λBragg is the Bragg wavelength which, in our case, must be equal to 1.55μm. So, taking
into account this equation the Bragg grating period and therefore the repetition period of
the transversal corrugations in the slow-light waveguides can be obtained.
Figure 2-24: Contour plot of the Bragg wavelength @ the photonic band edge versus width (We)
and length (Wi) of the transversal corrugations for the fundamental mode of with Λ=300nm,
W=300nm and h=100nm [BRI11B].
53
Chapter 2: Digital applications
Finally, taking into account everything mentioned along the point, the final passive
structure, which is shown in the Figure 2-25, consists on a rib single mode waveguide of
W=300nm x H=220nm with a thick slab of h=100nm with corrugated elements of
We=650nm of width and Wi=100nm of length which are repeated periodically every Λ
=310nm. These parameters are summarized in Table2-4 and are used in the slow-light
silicon-based modulators demonstrated in the following section.
Table 2-4: Summary of the final shallow-etched LCWGs parameters of the corrugated
waveguide chosen for the slow-wave electro-optic modulator.
54
2.3. Slow-wave structures
On one hand, a high speed operation with high efficiency and ER at quadrature and
with a relatively low on-chip IL is shown [BRI12B] enhancing previous results reported
in [BR111A]. On the other hand, the use of slow-light is exploited to reduce the driving
voltage [BRI12A]. So, we take advantage of the high modulation efficiency achievable
in slow-light modulators in order to lower the drive voltage down to values that fulfil
current CMOS requirements, again enhancing previous results reported in [BR111A].
Moreover the variation of the IL with a group index is also analyzed and discussed in
other point dedicated entirely to present a method for calculating the propagation losses,
first in MZIs with conventional rib waveguides [GUT12B] and second with corrugated
one [BRI12A]. As a result, relatively compact single-drive phase shifters featuring low
drive voltage, moderate IL, and high-speed performance are demonstrated assuring the
benefits of slow-light as a means to enhance the performance of silicon modulators
based on the plasma dispersion effect.
Figure 2-26(a) shows the schematic of the slow-wave modulator. Both arms of the MZI
consist of two identical 500 μm long slow-wave phase shifters to balance the optical
power and thus maximize the notch depth. An extra 900 μm long rib waveguide section
has been added on one arm of the MZI to allow the operating point of the modulator to
be precisely set.
55
Chapter 2: Digital applications
Figure 2-27: Scanning electron microscope images of (a) the corrugated waveguide, (b) the
contacting AlCu pads and (c) the travelling wave coplanar electrodes.
Figure 2-28(b) depicts a close-up portion of the slow-light region (ng~8) of the MZI
modulator under reverse bias voltages varying between 0V, −3V, and −5V. The
maximum transmission of the MZI is situated around −6 dB at 0 V, corresponding to a
net on-chip loss of 6dB, from which 5dB arise from the slow-wave phase shifter
56
2.3. Slow-wave structures
(coupling loss and propagation loss included) and 1 dB is due to the tapered multimode
interference (MMI) structures (0.5dB/MMI). As can be observed, a DC ER (referred to
as ER in Figure 2-28(b)) of around 7dB may be achieved at the quadrature point,
designated “A” and located at 1547.7nm wavelength. At this point, the maximum
transmission is situated at −9dB, which means that the IL penalty at the “1 1evel” is
9dB, including the 1dB loss arising from the two MMIs. The latter is extracted from the
difference between the two curves shown in Figure 2-28(a). With an additional loss
penalty of 2dB (point “B”), leading to an overall IL of 11dB, the ER increases up to
~11dB. Additionally, the effective index change as a function of the applied reverse was
determined experimentally and supported by numerical calculations based on Soref and
Bennett’s empirical equations [SOR87].
Figure 2-28: (a) Respective normalized spectra of the phase shifter (red curve, top) and the MZI
(blue curve, bottom), (b) close-up view of the MZI spectrum for biases of 0, −3, and −5 V
[BRI12B].
Figure 2-29 confirms that both experimental and numerical calculations are in good
agreement, regardless of the small variations that may appear owing to the error in
determining the notch shift across wavelength for varying reverse bias voltage. A clear
enhancement of the modulation efficiencies (VπLπ) between the fast (ng~4.4) and slow
(ng~8) light regimes may be observed. It should be noticed that even in the fast light
regime, the slight increase in group index compared to a conventional rib waveguide
(ng~3.8) already provides a small but noticeable enhancement. Overall, for a group
index of ~8, the modulation efficiency is improved by a factor of ~2 compared to a
conventional silicon modulator with identical doping conditions. In the slow-light
regime (ng~8), experimental modulation efficiencies of 0.58V.cm and 0.85V.cm for
reverse biases of 1V and 5V were respectively calculated. By contrast, in the fast light
regime (ng~4.4), modulation efficiencies of 0.95V.cm and 1.62V.cm were extracted for
the same corresponding reverse biases.
57
Chapter 2: Digital applications
Figure 2-29: Effective index change as a function of reverse bias for group index values of ∼4.4
(fast light; red curve) and ∼8 (slow light; blue curve). Modulation efficiencies VπLπ are pointed
by the arrows [BRI12B].
To evaluate the high-speed performance and the data transmission capability of the
slow-wave modulator, we used the same set-up as in previous section 2.2 for the MRR
modulators (it is shown in Figure 2-9) with the differences that the coupling is vertical
instead horizontal (details in Appendix A) and the electrodes are in a travelling wave
instead lumped configuration, so the modulating signal is applied to the travelling-wave
electrodes, which are required to be terminated externally by a 50Ω load coupled to a
DC-block. Details of model equipments and pictures of the real set-up are given in
section 2.4.So, to evaluate the high-speed performance of the slow-wave modulator,
40Gbit∕s eye diagram acquisitions at quadrature and 2dB below quadrature were carried
out under a 5Vpp drive voltage, as respectively depicted in Figures 2-30(a) and 2-30(b).
The measured extinction ratios are 6.6dB and 8.5dB at quadrature and 2dB below
quadrature, respectively.
Figure 2-30: 40Gbit/s eye diagram (a) at quadrature and (b) 2dB below quadrature [BRI12B].
58
2.3. Slow-wave structures
One should be aware that the group index at which the modulator is operated is slightly
lower (ng~8) than that used in [BRI11A] (ng~11). Besides the lower optical losses, the
choice of using such a group index value is motivated by the possibility of achieving a
better electro-optical velocity matching and thus a higher modulation bandwidth.
Although additional effects such as the microwave attenuation and impedance mismatch
also contribute to the bandwidth reduction, the velocity mismatch between the electrical
and optical signals plays a major role, especially as the group index increases
[BRI11A]. However, it is important to remark that to sustain the performance in terms
of extinction ratio, the phase shifter length maybe reduced according to the increase in
group index. In this way, the higher velocity mismatch is partly counteracted by the
shorter electro-optical interaction length.
On the other hand, the optical bandwidth of our slow-light modulator, defined as the
wavelength range over which the group index variations remain within 10% of the
target value [BRI10], is around 1.3nm [BRI11A]. This value is far inferior to that of
conventional MZI modulators (~20-30nm). The last challenge that will definitely allow
slow-light to win its spurs is showing a sustained performance over a broad optical
bandwidth using engineered slow-light structures [BRI10]. Interestingly, the relatively
low optical loss value positions our slow-light modulator among the state of the art of
current 40Gbit∕s modulators from an IL and footprint point of view.
As in the previous point, the device under study is again an asymmetric carrier depletion
MZI which converts the optical phase modulation in the slow-light waveguide into
amplitude modulation. Unlike the previous one, this time one arm of the MZI consists
of a 1mm-long slow-light phase shifter, whereas the other arm contains a rib waveguide
having the same length and doping conditions. The rest of parameters, both corrugated
waveguide dimensions and doping concentrations, are the same than section 2.2.3.1.
Figure 2-31(a) shows the MZI slow-light modulator spectra under varying reverse-
bias voltage. The bias voltage is varied between +0.5 and -2.5 V. The modulation
efficiency increases with the group index, but using higher group indices will also have
a negative impact on the IL, as we will be demonstrate in the next section dedicated to
59
Chapter 2: Digital applications
Figure 2-31: (a) Close-up view of the slow-light region for bias voltages varying from +0.5 to -
2.5V. The dotted circle is the quadrature point, (b) Experimental (blue solid line) and simulated
(red dashed line) effective index change versus applied voltages at 1546.9nm wavelength
(measured at the quadrature point). Measured modulation efficiencies VπLπ for each
corresponding applied voltage and taking +0.5 V as a reference are also shown [BRI12A].
Such Figure 2-31 shows the experimental and simulated effective index change and
modulation efficiency for varying reverse-bias voltage at the quadrature point. As can
be observed, the modulation efficiency varies strongly for increasing reverse-bias
voltage due to the nonlinear behaviour of the PN junction. Modulation efficiencies VπLπ
as low as 0.42, 0.55, 0.62, 0.7, and 0.8V.cm have been respectively extracted for 0.5, 1,
1.5, 2, and 3V applied voltages at 1546.9nm wavelength. Although for bias voltage
beyond -0.5V both experimental and simulated effective index changes are in good
agreement, a slight discrepancy takes place below -0.5V. We believe this might be due
to the fact that we assume an ideal uniform doping profile in our electrical simulations,
on one hand, and/or that the actual target doping concentrations are slightly lower than
expected [GAR09]. One might also put forward the fact that the slow mode distribution
might be altered from the simulations to experiment, hence interacting in a slightly less
efficient way with charge carriers.
On the other hand, data transmission measurements were realized in the same way by
using the same set-up (Figure 2-9) that in previous section 2.3.3.1. To analyze the high-
speed performance of the slow-light modulator under low drive voltages, eye diagram
acquisitions at quadrature (λ=1546.9nm) from 5 to 25Gbit/s were carried out under
drive voltage ranging from 1 to 4V. The results are summarized in Figures 2-32, 2-33
and 2-34.
60
2.3. Slow-wave structures
Figure 2-32: Eye diagrams at 5 Gbit/s for varying peak-to-peak drive voltage [BRI12A].
Figure 2-33: Eye diagrams at 25 Gbit/s for varying peak-to-peak drive voltage [BRI12A].
Figure 2-34: Eye diagrams at 10 Gbit/s for varying peak-to-peak drive voltage [BRI12A].
61
Chapter 2: Digital applications
Interestingly, and as a result of the high modulation efficiencies, the eye diagram in the
range 5-10Gbit/s remains open for very low drive voltages (1, 1.5, and 2V), and the
corresponding ERs lie between 3dB and 4.3dB. DC bias voltages are 0, 0.25, and 0.5V,
respectively. For drive voltages ranging from 3 to 4V (respective DC biases are 1 and
1.5V), the eye diagrams open up, showing that ultrahigh-speed operation at 25 Gbit/s is
achieved, as shown in Figure 2-33. The relatively low ERs are due to the power
unbalance between the two arms of the MZI, which produces a null height of only
10dB. Using an asymmetric MZI featuring two identical corrugated waveguides in each
arm would increase the null height and lower the “0” level. This would notably increase
the dynamic ERs for drive voltage beyond 2V.
62
2.3. Slow-wave structures
through the branching elements. The splitter and combiner components may either be
Y-junctions or MMI.
Figure 2-35: Schematic view of the MZI structure. Splitter and combiner portions are
represented by a generic box.
Ideally, the branching elements should split and recombine the input power equally
in both arms. However, this is not always true due to fabrication process deviations and
may therefore have a significant impact on the component performance, depending on
the application [CHA01]. Therefore, in our analysis, we initially consider the most
general case of having branching elements with different performance at the input and
output of the MZI structure. At the branching element acting as a power splitter, that is,
located at the MZI input, the amount of power coupled to the shorter arm is denoted ain
while the amount of power coupled to the longer arm is denoted bin. In a similar way,
we can denote aout and bout the power coupling coefficients that govern the output
branching element, which acts as a power combiner.
Starting from of the transfer function of a MZI and assuming the same propagation
losses for both arms, we can obtain the maximum electrical field at the MZI output as:
L L L
Emax i ain aout e bin bout e (2.28)
where Ei is the input electric field including coupling losses, α(Np/cm) is the
propagation losses in the MZI arms, L(cm) is the length of the shorter MZI arm and
ΔL(cm) is the length difference between the MZI arms.
63
Chapter 2: Digital applications
To obtain the expression for the minimum electric field, there will be an uncertainty
because we do not know a priori if the output power from the longer arm is lower or
higher than that from the shorter arm as it will depend on the branching elements,
propagation losses and length difference between the MZI arms. Therefore, for the sake
of simplicity, we can define the minimum electric field at the MZI output as:
L L L
Emin i ain aout e bin bout e (2.29)
It should be pointed out that this value may be either positive or negative. Combining
Equations (2.28) and (2.29), we can obtain the following linear equation:
E dB L cm . dB / cm R dB (2.30)
where
Emax Emin
E dB 20 log10 (2.31)
Emax Emin
ain aout
20 log10 , Emin 0
bin bout
R dB (2.32)
bin bout
20 log10 , Emin
ain aout
and R(dB) is denoted as the branching ratio. As can be noticed in Equation (2.32), the
exact expression of the branching ratio will depend on the sign of the minimum electric
field obtained from Equation (2.29). However, in both cases, the branching ratio value
derived from Equation (2.30) will indicate the degree of asymmetry of the branching
elements. For perfectly symmetric branching elements, R(dB)=0 while as the
asymmetry in the branching elements increases R(dB) will give rise to values greater or
lower than zero. Hence, considering a set of MZIs with different optical length
differences (ΔL), we can represent the single parameter E(dB) as a function of ∆L for
each MZI. Then, in a similar way than for the cut-back method, we linearly interpolate
through these points to finally estimate the propagation losses α(dB/cm) of the MZI as
the slope of the interpolated line. In addition, the intersection with the Y-axis (∆L=0) of
the interpolated line will give the branching ratio.
The estimation of the propagation losses and the branching ratio will be independent
of the coupling loss, which is demonstrated by the fact that Ei is not present in Equations
(2.30)-(2.32). Furthermore, it should be pointed out that if the branching ratio is known,
for instance from previous measurements, then only one MZI structure will be required
to estimate the propagation losses, which will be easily derived just by looking at the
maxima and minima of the interference transmission spectrum and applying Equation
(2.30).
64
2.3. Slow-wave structures
Figure 2-36: SEM picture of fabricated structures: (a) MZI, (b) Y-junction detail.
To perform the experiment we fabricated three MZIs having optical length differences
of ΔL=1, 2, 3mm. The optical characterization was carried out using a tunable laser as a
coherent light source, which was coupled to a fiber PC. The performance of the device
has been measured using vertical coupling through grating couplers over a bandwidth
span of 20nm from λ=1540nm to 1560nm. The measured transmission spectra are
shown in Figure 2-37.
Figure 2-37: Detail of the measured transmission spectra of the MZI’s with (a) ∆L=1mm, (b)
∆L=2mm, (c) ∆L=3mm. The spectral range is reduced for clarity.
65
Chapter 2: Digital applications
As already mentioned, when the splitting and combining ratio is 50%, the MZI is
balanced. In this case, we can obtain a complete destructive interference when the
phases in both arms are opposite that leads ideally to an infinite extinction ratio.
However, the imbalance of the splitter and combiner as well as the excess loss of one
arm with respect to the other gives rise to a reduction of the extinction ratio, as occurs in
Figure 2-37. The excess loss in one arm is directly responsible for the ER variation as
the length difference between the MZIs changes. On the other hand, we attribute the ER
variation with wavelength to the effect of stitching. It is well known that stitching
originates due to offsets in the pattern at different writing fields. In our case, a writing
field of 150µm x150µm has been used in the EBL process. Consequently, MZIs with
longer lengths are more likely to suffer offsets in the pattern (stitching errors) and
exhibit therefore more ER variability as it can be clearly seen in such Figure 2-37.
Applying the method described, we obtained that the propagation losses, shown in
Figure 2-38(b), are on average 8dB/cm with a standard deviation of 3.5dB/cm for the
measured wavelength range. It can be noticed here that such relatively high losses can
be attributed to stitching and sidewall roughness. Roughness loss reduction can be
achieved with techniques such as oxidation smoothing of the sidewalls [LEE01, SPA05]
and stitching may be minimized via substrate-tilt correction procedure [GNA07A,
GNA07B]. On the other hand, Figure 2-38(a) depicts the branching ratio as a function
of the wavelength. It can be seen that the branching elements are not perfectly
symmetric, which results in an average branching ratio of 0.83dB with a standard
deviation of 0.82dB. The standard deviation is high due to the undesired ER variation
with wavelength. Therefore, assuming identical branching elements at the input and
output of the MZI structures (a=ain=aout and b=bin=bout), as well as power conservation
(a+b=1), it is possible to extract a and b coefficients from Equation (2.32) with
acceptable accuracy and hence completely characterize the branching element
performance.
Figure 2-38: (a) Branching ratio and (b) propagation losses versus wavelength obtained from
method.
66
2.3. Slow-wave structures
The uncertainty given by the two possible solutions can be resolved by comparing
the experimental ERs of the MZI transmission spectra with the theoretical ER values,
both shown in Table 2-5 and calculated from Equations (2.28) and (2.29). These values
take into account the propagation loss and branching ratio that were extracted from the
proposed method. It can be seen that the ER values from solution a~0.52 are in very
good agreement with the average ER values derived from the transmission spectra of
MZIs thus demonstrating the robustness of the proposed method and indicating that
more power is coupled into the shorter arm of the MZI.
Table 2-5: Average theoretical ERs and experimental ERs derived and from the transmission
spectra.
Overall, the proposed method introduces interesting improvements over the cut-back
method as it allows for reliably calculating the propagation losses as well as the
branching asymmetry of MZI structures. The method, based on analyzing the
interference fringes of the MZI spectra, avoids the requirement of constant fiber-to-chip
coupling conditions, which is difficult to achieve in practice. Moreover, the proposed
technique may also be of great interest to systematically estimate the performance of
MZI-based structures, such as filters, switches or modulators, in which the power
imbalance introduced by the branching elements as well as the propagation losses, are
67
Chapter 2: Digital applications
The previous method is proposed and extended to accurately measure the IL of the
slow-light phase shifter [BRI12A]. In order to estimate the IL versus group index with a
single measure, the method exploits again the MZI structure consisting of a slow-light
waveguide in one of its arms, as illustrated in Figure 2-39(a). It should be noticed that
the calculation of the group index variations has been proposed with similar procedures
[GOM07, NOT01]; however, to the best of our knowledge, a method for also estimating
the IL from the MZI spectrum has not yet been reported. Figure 2-39(b) shows the
optical transmission spectrum of the MZI structure. The FSR changes across the
spectrum as a result of the group index dependence upon wavelength, which increases
as we approach the band edge of the slow-light mode.
Figure 2-39: (a) Schematic of the MZI structure used in the proposed method to estimate the IL
in the slow-light phase shifter, (b) normalized optical transmission spectrum of the MZI
structure [BRI12A].
Therefore, the FSR variations across wavelength can be used to extract the group index
from
2
FSR (2.34)
n L nref nsw L
where λ is the central wavelength of the FSR; nsw, nref, and n are the respective group
indices in the slow-light, 300nm wide reference rib, and 400nm wide rib waveguides; L
is the length of the slow-light and reference rib waveguides, which is the same; and ΔL
stands for the length difference between the MZI arms considering only the 400nm wide
rib waveguide. In our case, L=1mm, ΔL=0.9mm, nref~3.5, and n~3.7. The selection of
the sign in the right-hand side of Equation (2.34) depends upon the considered
68
2.3. Slow-wave structures
wavelength range and is related to the fact that the FSR exhibits different behaviours as
we move from the fast-light region toward the PBG, as can be observed in Figure 2-
39(b). In the fast-light regime (far enough from the PBG), the separation between the
interference fringes remains approximately constant, which is consistent with a nearly
unvarying group index value, and the negative sign must be used in the Equation (2.34).
However, as we enter the slow-light regime, that is, as we move closer to PBG, the FSR
increases up to a maximum value, which means that the phase difference between the
MZI arms is minimized. According to (2.34), the FSR is infinite when the denominator
is zero, which means that there is no phase difference between the MZI arms. It can be
straightforwardly obtained that this occurs for a group index of 6.8. From this point
forward, the positive sign must be used in the Equation (2.34) because higher group
indices imply that the effective optical path length of the shorter arm becomes larger
than that of the longer arm, thus causing the FSR to decrease.
The influence of loss variations versus group index in the slow-light waveguide is
also clearly observed in Figure 2-39(b). Initially, the losses in the reference arm are
higher than in the fast-light region (i.e., the region where the group index is
approximately constant) of the slow-light waveguide, hence reducing the contrast ratio
of the MZI interference fringes. As the group index increases, the losses in the slow-
light waveguide increase accordingly, reducing further the power imbalance between
arms. This, in turn, contributes to improve the contrast ratio. The maximum contrast
ratio is achieved when the total losses in each of the MZI arms are similar, which takes
place for wavelengths around 1552nm. At wavelengths lower than 1552nm, the power
imbalance increases back again, reducing therefore the contrast ratio, though this time
caused by higher losses in the slow-light waveguide. Therefore, the IL in the slow-light
phase shifter can be estimated from the variation of the contrast ratio with wavelength.
The proposed method for estimating the IL starts from the evaluation of the maximum
and minimum electrical field at the MZI output, which can be written as
L L L
Emax E0 mL c p e sw
mL L e ref
(2.35)
L L L
Emin E0 mL c p e sw
mL L e ref
(2.36)
Emax Emin
E 20log10 Cte C p ref sw L (2.37)
Emax Emin
69
Chapter 2: Digital applications
where Cte is a constant value, which depends only on the possible splitter/combiner
branching asymmetry and the αΔL losses. Thereby, we can obtain an estimation of both
coupling and propagation losses within the slow-light waveguide at a given wavelength
just by considering the maxima and minima or, in other words, the contrast ratio of the
MZI interference fringes, which makes the proposed method independent on the
external alignment and the fiber-to-the-chip coupling losses.
The IL, which include both coupling and propagation losses, can be derived as
follows
E E , region
IL Cp sw L fw L (2.38)
E E , region 2
where αfw and E are, respectively, the propagation losses and mean value of E(λ) in the
fast-light region, where the group index is approximately constant and where the
coupling losses are assumed to be negligible. The two different expressions for regions
1 and 2 depicted in Figure 2.39(b) assume that, in region 1, the output power from the
longer arm is lower than that from the shorter arm, and vice versa in region 2, where the
sign must be changed in the Equation (2.36) as also occurs in (2.34). The propagation
losses, αfw, can be estimated from Equation (2.37) giving rise to a value around
3dB/mm.
Figure 2-40: (a) Group index and (b) IL variation versus wavelength. (c) Slow-light phase
shifter IL versus group index. [BRI12A]
Figure 2.40 shows (a) the group index and (b) IL dependence on wavelength, which are
derived from Equations (2.34) and (2.38), respectively. Both results can be combined in
order to obtain the IL versus group index of the slow-light phase shifter structure, as
depicted in Figure 2-40(c). It can be seen that the IL increase for increasing group
index, yielding to a trade-off between high modulation efficiency and low IL, which
relies on the group index operating point [BRI11A].
DIN12]. Nevertheless, IL as low as ~2dB for a group index of ~28 have been reported
in an MZI modulator with 200μm long slow-light phase shifters based on a 2D-PhC
structure although considering lower implantation doses [NGU12A]. Overall, strategies
to reduce the IL should be focused on optimizing the overlap between the active region
and the slow-light mode propagating through the periodic optical waveguide. As a
matter of fact, a doping compensation method in which the doping profile is optimized
to reduce absorption losses without degrading the modulation efficiency and speed has
been recently proposed and demonstrated for conventional silicon modulators [TU11].
Such a method could also be applied for the proposed slow-light modulator. Finally,
improving the coupling technique to the slow-light mode would also be especially
desirable as the group index increases.
In conclusion, a simple method has been proposed for experimentally estimating the
IL versus group index of a slow-light photonic waveguiding structure. Only a single
measure is required, and the loss estimation is independent of the alignment and fiber-
to-the-chip coupling losses.
Light emitted by an ECL laser is injected into the modulator via grating couplers if
vertical coupling (slow-wave modulators) or lensed fiber and inverted tapers if
horizontal coupling (MRR-based modulators). The transmitted power is optimized
through the use of a polarization controller (PolaRITE™ General Photonics). The
description of both coupling techniques as well as the passive characterizations set-ups
are given in Appendix A.
71
Chapter 2: Digital applications
Data transmission measurements are realized by driving the modulator with a NRZ
PRBS of length 231-1 delivered by a BPG (SHF BPG 44E) connected to an external
clock (R&S SMR RF signal generator). The electrical signal is generally amplified
through a high speed RF amplifier (Picosecond Pulse lab model 5865) to achieve a
decent voltage swing (Vpp) and combined to a DC bias supplied by a DC Power Source
(Agilent E3647A) using a bias-Tee (SHF BT-110) . For consistency with S21 electro-
optical measurements, the modulating signal is applied through a high speed GSG input
probes to the electrodes (travelling wave in slow-wave modulators and lumped
electrodes in the MRR-based modulators) which are terminated externally by a 50 Ω
resistance coupled to a DC block (SHF DCB-110) by using a GSG output probes to
avoid reflexions (in the case of travelling wave configuration).
Figure 2-41: Schematic of the characterization setup and pictures of each component used in
along the chapter to measure the data transmission capability of the silicon-based
modulators.
For completeness, pictures of the entire real set-up used in the Nanophotonics
Technology Center facilities are shown in Figure 2-42.
72
2.4. Active RF characterization set-ups
Figure 2-42: Pictures of the characterization setup for the eye diagram acquisition.
So, Figure 2-43 shows the experimental set-up (and real pictures of each component)
for measuring the high speed optical frequency response. Light is coupled from an ECL
to the chip via a lensed fiber (details in appendix A) after polarization control
(PolaRITE™ General Photonics). The small electrical signal delivered by the port 1 of a
RF vectorial network analyzer (Agilent PNA-X N5247A) is coupled a DC bias voltage
supplied by a DC Power Source (Agilent E3647A) using a bias-Tee (SHF BT-110) and
applied to the lumped electrodes through high speed GSG probes to the RAMZI
modulator. The output modulated signal is amplified with an EDFA (Amonics AEDFA-
18-B-FA) and photo-detected by a high speed photodiode (u2t 50GHz XPDV2120R)
connected to the port 2 of the RF VNA in order to extract the electro-optical S21 transfer
73
Chapter 2: Digital applications
function. The signal generator is swept from 100MHz to 50GHz and the photo-receiver
output is normalized by the input drive voltage for all frequencies
Figure 2-43: Schematic of the characterization setup and pictures of each component used in
section 2.3.3.2 to measure 3dB frequency roll-off of the silicon-based RAMZI modulator.
74
Chapter 3
Analog applications
Analyzing the previous definitions one can easily to get an idea of the basic concept
of a MWP link, shown in Figure 3-1(a), which consists of: analog electrical signals at
microwave frequencies are transported over a photonic link composed by optical fiber.
In its simplest form, an MWP link consists of an electrical-to-optical (E/O) conversion
of the RF signal by a modulation device and an optical-to-electrical (O/E) reconversion
of the optical signal by a photodetector once it has been transported. Further, when
processing signal (as time delay, phase shift, filtering or frequency conversion) in the
optical domain between the two conversions is added, we are referring to a MWP
system, which is shown in Figure 3-1(b).
75
Chapter 3: Analog applications
from the consumer point of view, broadband wireless access networks, for example,
attaining in-home fiber networks supporting wireless communications at data rates of
multiple Gbit/s. The increasing demand in speed, bandwidth, and dynamic range as well
as small and lightweight devices are some of the reasons for that MWP has attracted
great interest. So, MWP enables the realization in microwave systems that are becoming
complex or even not possible in RF domain, exploiting the unique capabilities of
photonics in terms of size, power, cost, high data-transfer capacity and immunity to
electromagnetic interference.
Figure 3-1: Schematic of the fundamental concept of (a) an MWP link, (b) a MWP system and
(c) an IMWP chip.
Despite the potential of MWP, there are several factors that still distance the MWP
solutions to replace traditional microwave solutions. The dynamic range and the
reliability are still insufficient and the cost and power consumption are still too high. On
one hand, the use of discrete components as lasers, modulators and detectors, leads to
the system to be bulky, expensive and high power consumer. On the other hand, the
interconnections with fiber pigtails reduce the robustness of the system. The most
efficient solution so far to overcome these mentioned drawback factors, is the RF and
photonics integration [MAR12], that is, the use of photonic integrated circuits (PICs) to
address MWP functionalities, which makes arise a new technology so-called Integrated
Microwave Photonics (IMWP). With the incorporation of MWP components in
photonic circuits, and therefore the use of the IMWP technology, one can achieve a
reduction in: footprint, inter-element coupling losses, packaging cost as well as power
dissipation. This concept, shown in Figure 3-1(c), is very much in line with the recent
trend, already commented in the introduction of this thesis, of large scale PIC
technology for the future photonic network-on-chip (NoC) in scaled chip
multiprocessors systems [COL11, SMI12, LEE10] in digital approach, but it must be
taken into account that, due to the stringent requirements in the management of analog
signals, PIC technology for integrated MWP should show higher performance than the
76
3.1. Introduction: Microwave Photonics
one expected from the digital applications. For this reason less attention has been paid
to analog applications, and so far, MWP has addressed lower-volume market, and
therefore lower volume PIC productions [MAR12].
One of the available platforms frequently used to demonstrate IMWP features is the
silicon-on-insulator (SOI) technology which allows much MWP functionalities to be
addressed in a single chip. SOI integration based devices is therefore a hot topic in the
area of IMWP as its state of art is mature and compatible with CMOS technology,
which provides it the capability to monolithically integrate optical and microelectronic
components. Moreover it provides immunity to electromagnetic interference, besides
that the miniaturization of the devices due to its high index contrast between Si and
SiO2 which allows large-scale integration [JAL06].
Among the numerous building blocks required to enable a fully functional PIC chip
along with high speed electronics, electro-optic modulators are essential. These are just
the devices employed in MWP links for converting the RF signals into the optical
format, that is, they are the responsible of E/O conversions and therefore of paramount
importance for achieving high performance links. As it has already been explained in
the first chapter of the thesis, the leading approach for developing electro-optical
modulators in silicon photonics is currently via the plasma dispersion effect which
consists of varying the free-carrier concentration in a doped silicon waveguide [REE10,
SOR87]. Significant milestones in modulation devices have been achieved during recent
years through the plasma dispersion effect in silicon although mostly for digital
applications [BRI12A, THO12, BRI12B, AKI12B, ZIE12] as it has been reviewed in
chapter 2. But, until recently, less attention has been paid to analog applications due to
the fact that IMWP is a much younger emerging discipline which is experiencing its
impulse in the last years.
77
Chapter 3: Analog applications
This chapter presents different researches carried out in the analog scope for
application in the field of IMWP, all of them using CMOS-compatible electro-optic SOI
modulators which validate the potential of silicon CMOS photonics as a promising
approach for enabling large scale photonic integration driving the development of
integrated microwave photonics applications.
One of the impediments for MWP link to reach higher performance and replace the
traditional microwave solutions is the distortion added by the E/O and O/E conversions
and moreover, if the electro-optic conversion is carried out by a nonlinear-material
modulator as silicon-based. Besides, from the perspective of system performance,
silicon modulators cannot yet provide the stringent requirements of MWP in the
management of analog signals; hence most of the advantages in silicon photonics for
MWP have been focused on exploiting the optical nonlinearities [MAR12] but also on
enhancing the linearity of the silicon devices. For these reasons, an analytical model to
obtain distortion closed expressions of the most widely used modulator structure that is
the Mach-Zehnder Interferometer (MZI), is theoretically developed and experimentally
validated in the section 3.2. Further, the model allows the designer to set a proper
choice of modulator parameters to linearize the device or to enhance the nonlinearities
giving a potential tool to predict the nonlinear modulator performance and the
parameters needed for a full study of a microwave photonic link [GUT13B].
The following sections 3.3 and 3.4 review two of the above mentioned MWP
applications in which we have used a silicon-based electro-optic modulator, explaining
more deeply in each case the basic principles and showing our experimental
measurements. First, the section 3.3 is devoted to describe and demonstrate
experimentally a method for photonic-mixing of microwave signals exploiting the
nonlinearities produced by a silicon-based slow-light MZI modulator enhanced via
slow-light propagation, which is used to successfully demonstrate an up-conversion
process [GUT12C]. Finally, in section 3.4, a photonic-based integrated filter for the
processing of RF signals is implemented by a silicon-based simple MZI modulator
[GUT13A]. In both sections, a review of the basic theory and, additionally, the state-of-
the-art of SOI devices used in the corresponding application are provided and the
experimental results obtained showing the potential of using CMOS silicon-based
modulators for IMWP technology. Finally, in section 3.5 the set-ups used for the RF
characterization are explained in detail describing the equipment models and showing
pictures of the real set-ups in the Nanophotonics Technology Center facilities.
78
3.2. Analytical model for silicon-based MZIs nonlinear distortion
As mentioned in the introduction of this chapter, the heart of any MWP system is the
MWP link which basically consists of a modulation device for E/O conversion
connected by an optical fiber to a photodetector that does the O/E conversion. These
conversions in the MWP link add nonlinear optical distortions to the output RF signal
which must be reduced or exploited depending on the application to carry out in the
system. On one hand, high linearity is required for photonic RF signal transmission, so
the nonlinear distortions due to these E/O-O/E conversions are a limiting factor for this
purpose. In this case, the distortion terms closest to the frequency of the transmitted
signal have to be reduced or ideally eliminated. Two important performance metrics for
measuring the linearity of an analog optical link is the spurious-free dynamic range
(SFDR) and carrier-to-distortion ratio (CDR). The former is defined as the signal-to-
noise ratio of a carrier when the largest spur signal is at the noise floor level.
A lot of methods to improve the commercial LiNbO3 modulators linearity have been
published. The general idea in these linearization schemes is to introduce certain pre-
distortion to compensate for the existing ones with the cost of increased system
79
Chapter 3: Analog applications
Table 3-1: Reported results in photonic transmission of RF signals in the scope of MWP field,
(*) In the next section 3.3.1 it will be explained the difference between harmonic distortion
(HD) and intermodulation distortion (IMD).
80
3.2. Analytical model for silicon-based MZIs nonlinear distortion
In contrast with the plain analog signal transport, some nonlinear applications such as
mixing or frequency conversion have been demonstrated in the IMWP field [GUT12C].
In these cases, the nonlinear operation of the modulator is exploited, taking advantage
of the distortion produced by the nonlinearities. In the section 3.3 we will center in this
kind of application. The purpose of naming them at this point is to highlight the fact that
the study of the distortion properties of silicon electro-optic modulators is an important
question for all silicon IMWP applications.
The most common way to characterize the modulator linearity is to perform a tone
modulation which can be classified into two categories: single-tone or multi-tone
modulation. Obviously, they differ in the number of carrier frequencies included in the
modulating signal and they produce different types of distortion. Harmonic distortion
can be defined as a single-tone (only a single frequency carrier is used as modulating
signal) distortion product caused by device nonlinearity.
Figure 3-2: (a) Single-tone and (b) two-tone output spectrum diagram of a nonlinear device
showing the frequency and the names of the generated distortion products.
81
Chapter 3: Analog applications
It is well known from theory and many published works [COX04, MAR09] that linear
material-based modulators such LiNbO3, in which the distortion is produced only by the
nonlinear modulator transfer function, which is periodic and usually have the same
performance with the bias point. This periodic function is a nonlinear cosine-squared
and at quadrature bias point (QB), all even order distortion terms ideally vanish (in
practice they are minimized), while fundamental signal and odd order terms maximize;
by contrast, at maximum (MATB) or minimum transmission bias point (MITB), the odd
order distortion terms are suppressed (again ideally) and the even ones are maximized.
Figure 3-3 shows a spectrum diagram example for a single-tone test in the three key
points of the modulator transfer function for a linear-material modulator.
Figure 3-3: Output spectrum diagram example for a single-tone test modulation in a linear-
phase modulator for the three key points of the transfer function: MATB, QB and MITB.
Although the work of in this thesis is focused in integrated SOI modulators, the
distortion of a commercial LiNbO3 electro-optic modulator has been characterized, with
the aim of corroborating the distortion behavior explained above. The device used was
82
3.2. Analytical model for silicon-based MZIs nonlinear distortion
40Gbit/s modulator from Corning OTI Inc. and the setup used for the measurements is
shown in Figure 3-4.
Figure 3-4: Schematic of the setup used to carry out the distortion characterization of a
commercial LiNbO3 modulator.
The laser used in this measurement was an external cavity laser (ECL) with a
wavelength of 1550 nm and an optical output power of 8dBm. The laser is pigtailed
with a standard single mode fiber and the polarization is optimized via the use of a
polarization controller (PC). The MZM bias voltage was controlled with a voltage
supply and the RF signal was supplied to the modulator using a RF signal generator.
The modulated optical power from the modulator was then detected using a
photodetector. The detector output was connected to a bias-Tee and the DC output is
connected to a multimeter, while the RF output was connected to an RF spectrum
analyzer (detail of all components and measurement devices in section 3.5).
Figure 3-5: Theoretical [MAR09] and experimental average photocurrent as a function of the
bias voltage of a commercial LiNbO3 MZI modulator.
DC MZM characterization was done by means of varying the modulator bias voltage.
The average detected photocurrent was measured for every bias voltage using a
multimeter obtaining the typical modulator V/I curve as shown in Figure 3.5. In such
83
Chapter 3: Analog applications
Figure are also shown the three key bias operation points: MATB, QB and MITB.
Moreover, the measurements results were compared with the simulation results from
analytical expressions developed in [MAR09]. From the average photocurrent
measurements, we found that the value of the half-wave voltage Vπ was 4.5V.
Figure 3-6: Theoretical and experimental fundamental and 2nd and 3rd harmonic distortion
powers as a function of the bias voltage of a commercial LiNbO3 MZI modulator.
We can easily identify in the above figure by comparing with the previous I/V curve
shown in Figure 3.5, the QB point (around 2.25V and 6.75V) as the bias voltage that
gives minimum HD2 power and the MITB (around 4.6V) and MATB (around 9.3V)
points as bias voltages for which the fundamental and the HD3 powers are minimized,
corroborating the distortion feature explained at the beginning of this section for the
linear-material electro-optic modulators.
84
3.2. Analytical model for silicon-based MZIs nonlinear distortion
the Soref equations [SOR87]. This nonlinear dependency of the index change with the
applied voltage also produces intermodulation and harmonic distortion terms. So, this is
the main motivation for the development of a model with analytical expressions which
can help the designer to predict the distortion performance in a silicon-based MZI
modulator.
The starting point of the model is the effective index change with the applied DC
voltage. In a silicon-based modulator under plasma dispersion effect, in this case carrier
depletion effect, one can derive such effective index change from the depletion width of
the PN en junction. For an asymmetrically doped junction, meaning that the n-doping
concentration is higher than the p-doping, the depletion width can be represented as:
2 Vb Vapp
WD (3.1)
eN A
Where WD is the depletion width, ε is the permittivity of silicon, e is the electron charge,
NA is the acceptor concentration, Vb is the built-in voltage (contact potential) and Vapp is
the applied voltage. On the other hand, the effective index modulation (dn) due to the
bias voltage change (dVapp) can be described by the following Equation (3.2) [VAC10,
CHU95]:
dn
2
CJ2 (3.2)
dVapp e A
A
CJ (3.3)
WD
Where n is the effective refractive index of the waveguide, δ is a empirical constant that
accounts for both confinement effect (overlap between the optical mode and the spatial
distribution of free carrier density change in the waveguide phase shifter) and the
coefficient of charge density induced refractive index change, CJ is the junction
capacitance and A is the area of the capacitor. From (3.1) to (3.3) we find by integration
that the effective index change with the applied voltage can be approximated by a
logarithmic curve as shown in Equation (3.4). As we have already mentioned, this
dependency has a nonlinear form for a silicon-based modulator.
Vapp
neff k ln 1 (3.4)
Vb
where k=δNA/2 (dimensionless) and Vb (in volts) are constants. In our case, we will use
85
Chapter 3: Analog applications
these parameters to fit the curve with the DC performance experimental results of the
effective index variation. This effective index change induces a phase shift that
accomplishes the Equation (3.5) already described in Chapter 1.
2
neff LACT (3.5)
where λ is the wavelength, ∆neff is the effective index change and LACT is the phase
shifter length.
Furthermore, in case of having a length difference between two arms of the MZI also
induces an additional phase shift proportional to this length difference. Thereupon, we
can regard that the total induced phase shift is due to two factors: the device structure,
related with modulator transfer function, which has been called ∆ФBias and allows
driving the modulator to the appropriate operation bias point; and the silicon in the
active area, which has been named as ∆ФSi, and it is described in Equation (3.7). So, the
final phase shift expression can be written as:
86
3.2. Analytical model for silicon-based MZIs nonlinear distortion
Also, the phase shift due to the silicon including the Equation (3.4) of the effective
index change into the Equation (3.7) results:
2 Vapp
Si LACT neff ln 1 (3.12)
Vb
where we have defined the design parameter σ (in radians) as shown in the Equation
(3.13):
2
LACT k (3.13)
On the other hand, in a two-tone test, the input RF signal takes the form:
where VRF1 and VRF2 are the RF signals amplitudes and ω1=2πf1 and ω2=2πf2 are the
angular modulating frequencies of the two input tones. Thus, the total applied voltage to
the electro-optic modulator is the sum of one DC term and one RF term, as:
If we make a compilation of the above equations and we substitute each equation into
each other, we realize that the final expression for the silicon phase shift results:
VRF1 VRF2
ln X DC 1 cos 1t cos 2 t
Vb X DC Vb X DC
where:
VDC
X DC (3.17)
Vb
87
Chapter 3: Analog applications
VRF1
X RF1 (3.18)
Vb X DC
VRF2
X RF2 (3.19)
Vb X DC
For simplicity, the model has been developed under the assumption that the two tone
input signals are of equal amplitude and the small signal approximation VRF<<Vb, so:
Finally, we can write the silicon phase shift as described in the Equation (3.22):
Si YDC ln 1 Y (3.22)
where Y and YDC are given by Equation (3.23) and (3.4) respectively:
Y X cos t
1 cos 2 t (3.23)
VDC
YDC ln X DC ln (3.24)
Vb
Now, the silicon phase shift can be expanded into the natural logarithm Taylor series as
follows:
n 1 Yn 1 2 1 3
ln 1 Y 1 Y Y Y ... (3.25)
n 1 n 2 3
The silicon phase shift resulted of this Taylor series will consist of a DC component
which does not depend on ω, and a set of harmonics and intermodulation distortion
terms as it is shown in the Equation (3.26):
88
3.2. Analytical model for silicon-based MZIs nonlinear distortion
1 2 3 3
X X X cos 1t
2 4
1 2 1 3
Si X cos 2 1t X cos 3 1t (3.26)
4 12
1 2 1 3
X cos 1 2 t X cos 2 1 2 t
2 4
Including this expression of the silicon phase shift into the Equation (3.11) of the output
optical power we can perform another Taylor expansion of the cosine and sine which
are as follow:
n
1 x3 x5
sin x x2n 1
x ... (3.27)
n 0 2n 1 ! 3! 5!
n
1 2n x2 x4
cos x x ... (3.28)
n 0 2n ! 2! 4!
The result is again a set of harmonic and intermodulation products. The amplitudes of
all these components will be sums of linear combinations of X, YDC and σ. The desired
expressions that describe the modulator amplitudes nonlinearities are finally obtained
and described by the Equations (3.29)-(3.31) in which the sub-index ‘c’ is referring to
cosine and ‘s’ to sine:
1 1
PFUND. 1 cos Bias FUNDc X sin Bias FUNDs X (3.29)
2 2
1 2 1
PIMD 2 1 2 cos Bias IMD 2 c X sin Bias IMD 2 s X2 (3.30)
2 2
1 3 1
PIMD 3 2 1 2 cos Bias IMD 3c X sin Bias IMD 3 s X3 (3.31)
2 2
As we can observe, the fundamental, 2nd-order and 3rd-order amplitudes are the terms
with linear, quadratic and cubic dependence on the modulating signal VRF, respectively.
Finally, the cosine phase expressions are:
2
FUNDc YDC (3.32)
1 1 2
IMD 2 c YDC (3.33)
2 2
1 3 2
IMD 3c YDC (3.34)
4 8
89
Chapter 3: Analog applications
1 2 3
FUNDs YDC (3.35)
2
1 1 2 1 3
IMD 2 s YDC YDC (3.36)
2 4 2
1 1 2 3 1 3
IMD 3 s YDC YDC (3.37)
4 8 8 8
The IMD amplitudes differ from the HD amplitudes even though they are basically
generated by the same mechanism. It has been found that the IMD2 amplitude is twice
that of the HD2 amplitude while the IMD3 amplitude is three-times the HD3 amplitude.
However, the powers of these components rather than the amplitudes are usually
measured. If we regard the amplitude to be either a current or a voltage, then the power
considered here is an electrical or an RF power. Thus we can easily identify that the
powers of the distortion terms are proportional to the square of their amplitudes. So, we
can deduce that the power of the HD2 terms expressed in decibels are approximately 6
dB lower compared to the IMD2 powers, and the HD3 powers are approximately 9.5 dB
lower relative to the IMD3 powers. Therefore, we can affirm that:
In the above equations, higher order terms than 3 have been simplified (in the original,
there are terms up to order 7). The deviation suffered by this simplification can be
neglected. As proof, the next Figure 3-7 represents the 2nd and 3rd harmonic and
intermodulation products of a 1mm-long silicon-based MZI, in which the dotted line
corresponds to the distortion obtained with the original equations without simplification
and the solid line to the same obtained from the simplified ones. As it can be observed,
the highest deviation of only 0.05nm is produced in the fundamental term which is an
acceptable value. Moreover, in that figure one can confirm the relation between
harmonic and intermodulation products (Equations 3.38 and 3.39).
90
3.2. Analytical model for silicon-based MZIs nonlinear distortion
Figure 3-7: Distortion as a function of the wavelength in a 1mm-long silicon-based MZI with
180µm of length difference between their arms, for a V DC=1V. Dotted line corresponds to the
distortion obtained with the original equations and the solid line to the same obtained
from the simplified ones.
Figure 3-8: Third-order intermodulation product power as a function of the phase shifter length
in a ∆ФBias=π/2 modulator with a wavelength laser of 1550nm.
91
Chapter 3: Analog applications
For example, starting from a nonlinear effective index change curve as a function of DC
bias voltage with k=0.52.10-3 and Vb=8V in the Equation (3.4), in a quadrature bias
modulator (modulator is designed such that the bias phase shift between the two arms at
VDC=0 and VRF=0 is π/2), one can linearize the modulator cancelling the 3rd-order
nonlinearity knowing the optimum phase shifter length. Figure 3-8 shows the lengths of
the phase shifter which yield the most linear performance for several DC voltages. It
can be observed, for example, that for a DC voltage of 1V the modulator phase shifter
length must be 825.5µm.
Figure 3-9: Contour of 3rd order (a) and 2nd (b) intermodulation products power as a function of
bias phase and DC bias voltage for a MZI with a phase shifter length of 1mm and λ=1550nm.
92
3.2. Analytical model for silicon-based MZIs nonlinear distortion
appreciated between simulation and our analytical model, but curves follow the same
trend. In any case, the analytical expressions always overestimate the simulation results.
For a fixed quadrature modulator with an active length of 1mm and a fixed 1550nm
wavelength, we can obtain from Figure 3-9 that approximately 1.7V is the optimum DC
voltage for cancelling the 3rd order intermodulation term. Moreover we can observe in
such figure another point of view of the model application: for a DC voltage around
8.9V the fundamental term vanishes and the 2nd order intermodulation product is
maximized. This point could be useful for a frequency up-conversion process. The input
signal can be set in the fundamental frequency and the signal information can be
received in the 2nd order term ensuring a good level of power in such signal.
It should be noticed that the free carrier concentration variation due to the plasma
dispersion effect alters the imaginary part of the refractive index thus introducing
absorption losses in the material. Such absorption losses usually depend on the applied
voltage in a nonlinear form, which also needs to be taken into account in the nonlinear
distortion analysis.
First, the effect of constant optical losses in the arms of the MZI on the nonlinear
distortion has been analyzed by means of simulations. It has been obtained that having
an unbalanced power distribution at the arms of the MZI only affects to the absolute
value of the RF output power but the transmission response remains unaltered so,
quality parameters as CDR or SFDR are not affected.
93
Chapter 3: Analog applications
On the other hand, the dependency of absorption losses with voltage can also be
approximated by a logarithmic form. Simulation results show that in this case the
transmission response of the intermodulation products is red-shifted with respect to the
response obtained with the developed model, i.e. the case of having no loss. Therefore,
the variation of absorption losses (∆α=αmax-αmin) with voltage will reduce the accuracy
of the model. However, the effect introduced by losses can be compensated in a real
case by adjusting the bias (Фbias) for the target VDC.
Figure 3-11: Absorption losses variation with the DC voltage. The fitted curve is obtained from
a logarithmic nonlinear approximation. It is shown both the maximum variation of absorption
losses between 0V and 10V and the absorption losses variation with the voltage for 2V pp of RF
modulation voltage.
In the specific case of a 1mm silicon MZI modulator which will be used to validate the
analytical model, the measured voltage-dependent absorption losses are very low so the
effect is almost negligible. It has been obtained that the maximum variation is
∆α=0.45dB between 0 and 10V, that is, for ∆V=10V. So, if the RF modulation voltage
is around 2Vpp, the variation of absorption losses with voltage will be of the order of
0.11dB, as shown in Figure 3-11.
Figure 3-12 shows the simulated RF output power of the IMD3 intermodulation product
as a function of the bias phase and DC voltage. It can be seen that the effect of losses
are almost negligible for the case of ∆α=0.45dB but discrepancies are more noticeable
as losses increases. However, if, for example, the IMD3 is minimized for VDC=2V and a
Фbias=0.5π. The effect of losses could be compensated just by slightly increasing Фbias to
0.65π for a worst-case scenario of ∆α=2dB. Therefore, the developed model remains as
a powerful tool for analyzing nonlinear distortion in silicon MZI modulators.
94
3.2. Analytical model for silicon-based MZIs nonlinear distortion
Figure 3-12: Simulated RF output power of the IMD3 intermodulation product as a function of
the bias phase and DC voltage for different values of absorption losses variation. The absorption
losses compensation through Фbias parameter in a specific case for modulation linearization it is
also shown.
95
Chapter 3: Analog applications
Figure 3-13: Schematic of the fabricated silicon-based electro-optic MZI modulator. The MZI is
asymmetric with a length difference of 180µm and the phase shifter in the shorter arm is 1mm
of length. MMI are used for divide the input light and combine the output modulated signal.
Figure 3-14: (a) Schematic of the PN junction located at the middle of the active waveguide
phase shifter, (b) Optical microscope image of the fabricated modulator with a clear view of the
traveling-wave ground–signal–ground (GSG) coplanar, electrodes, (c) Detail of the GSG
electrode.
As it was already mentioned, the model starts from the effective index change
measurement, so, first of all, static and DC performance measurements have been
carried out. The spectrum of the silicon-based MZI exhibits a free-spectral range (FSR)
of 2.42nm and a 10dB drop in transmission at the resonant wavelength around 1552nm.
Figure 3-15 shows the MZI normalized transmission spectrum for varying DC voltages
from 0V to 10V.
96
3.2. Analytical model for silicon-based MZIs nonlinear distortion
Figure 3-15: Normalized DC transmission spectrum of the MZI modulator for different applied
DC voltages.
As the modulator is asymmetric we have had to take into account this asymmetry in the
bias phase as follows:
where ng is the group index and ∆L is the length difference between both MZI arms.
Moreover, the minus sign is because the active area is located in the shorter arm and
thus the resonance shifts to lower wavelengths with the DC voltage. Finally, we have
added an offset term (Фoffset) which is required to fit the theoretical spectrum with the
experimental one (in DC measurements). So, for a fixed arm difference, the bias point
of the modulator can be tuned either through acting on the DC voltage or on the
operation wavelength and vice-versa. Taking into account these Equations (3.40) and
(3.41) and including them into the fundamental and intermodulation distortion described
in Equations (3.29)-(3.31), one can obtain the optimum length difference and DC
voltage for cancelling the desired intermodulation distortion. From the measured FSR,
the group index has been calculated to be 3.9. Moreover, we obtain an offset phase of
0.36π to match the theoretical and experimental DC measurements and this Фoffset value
will be used in the Equation (3.40) to the model experimental validation for our
fabricated modulator. On the other hand, the voltage-induced wavelength shift (Δλ) in
the MZI modulator output spectrum and the index group can be used to extract the
effective index variations (Δneff), through the next equation (3.42):
97
Chapter 3: Analog applications
L
neff ng (3.42)
LACT
Figure 3-16 depicts the calculated and fitted effective index variation as a function of
applied DC voltage. The fitted curve has been obtained from the nonlinear
approximation described in the equation (3.4).
Figure 3-16: Effective index change with the voltage. The measured data are obtained from the
measured voltage-induced wavelength shift through equation (3.42). The fitted curve is obtained
from nonlinear approximation described in the equation (3.4).
98
3.2. Analytical model for silicon-based MZIs nonlinear distortion
Figure 3-17: Schematic of the experimental set-up for the single-tone test carried out to
experimental validation of the model.
As mentioned previously, due to the asymmetric MZI structure and the material in the
active area, the bias point of the modulator can be tuned either through acting on the DC
voltage or on the operation wavelength in contrast with conventional LiNbO3
modulators in which the bias point is driven only with a DC voltage. So, for a given
wavelength, linear or nonlinear optimum operation can be achieved by adjusting the DC
voltage and vice-versa. Thereby, representations of the distortion as a function of DC
voltage and wavelength can help the designer to obtain the optimum combination
wavelength-DC voltage point.
Figure 3-18 shows a contour of fundamental (PFUNDAM.) in (a), 2nd harmonic (PHD2) in
(b) and 3rd harmonic (PHD3) in (c), obtained from analytical expressions for our
fabricated device. So, from the above Figure 3-18 we can obtain that for 1V of VDC the
HD2 is minimized at around 1551.25nm and HD3 at 1551.75nm. To validate these
results, we first tested the model by measuring the harmonic distortion utilizing a fixed
DC voltage of 1V (Figure 3-19) and tuning the operating wavelength. As it can be
observed in such figure, the measurements show the same trends and are in very good
agreement with the analytic model corroborating the results obtained from Figure 3-18.
Moreover, it must be highlighted that the experimental curves are affected by the band-
pass of the optical filter shape. With this graphic one can realize that fundamental and
3rd-order harmonic distortion, have not the same performance with the bias parameter
(in this case, the operation wavelength) as it occurred in linear-material modulators and
it was corroborated with Figure 3-6 for a LiNbO3 modulator.
99
Chapter 3: Analog applications
Figure 3-18: Contour of the fundamental (a), the 2nd (b) and the 3rd order (c) harmonic products
powers as a function of DC voltage and wavelength for our fabricated MZI.
Figure 3-19: Measured values and obtained curves from the analytical expressions of the
fundamental and harmonic distortion performance as a function of the wavelength for a DC bias
voltage of 1V. The RF input power of the single-tone is 10dBm (2Vpp).
The level of distortion using the carrier-to-distortion parameter can also be evaluated.
Defining it as the ratio between the harmonic distortion product power and the received
power fundamental tone:
100
3.2. Analytical model for silicon-based MZIs nonlinear distortion
PHD 2
CDR2 (3.43)
PFUNDAM .
PHD3
CDR3 (3.44)
PFUNDAM .
Figure 3-20 shows the experimental CDR (a), theoretical CDR2 (b) and CDR3 (c) as a
function of the wavelength for 1V of DC voltage. From that figure we can see that
according to the analytical expressions, for 1V maximum CDR, around 20dB for CDR2
and around 1dB for CDR3, appears at 1552.4nm as expected from Figure 3-19 in which
the fundamental is minimized. In contrast, minimum CDR, around -56dB for CDR2 and
-83.5dB for CDR3 (obviously, the slight error displacement in HD3 appears again) are
obtained at 1551.25nm where HD2 is reduced and at 1551.55nm where HD3 is reduced,
respectively. So, we can affirm that the analytically estimated CDR is found to be in
good agreement with the experimental one.
Figure 3-20: Carrier-to-distortion ratio (a) from the experimental measurements and (b) CDR2
and (c) CDR3 from the analytical model. The RF input power of the single-tone is 10dBm
(2Vpp).
101
Chapter 3: Analog applications
Figure 3-21: Fundamental, HD2 and HD3 powers as a function of the modulating RF input
power for (a) λ=1552.1nm and (b) λ=1552.4nm. Curves are obtained from the analytical model
and VDC=1V.
Finally, the modulation depth was measured. It is well known that over-modulating a
modulation device, which occurs when the RF modulating signal swing is too high,
produces high distortion levels which strongly affect the signal output creating
undesired harmonics. So, the modulation depth as a function of the RF input power was
obtained through the output temporal signals to corroborate that the measured
harmonics in our fabricated device were only due to the modulator nonlinear feature and
not due to over-modulation. This can be observed in Figure 3-22(b) for λ=1552.1nm.
No modulation depth higher than the unit is obtained in the whole range of RF input
powers measured. Figure 22(a) inset shows an example of the measured temporal
signal. In particular it corresponds to the case of 12dBm of RF input power which
results in a modulation depth of around 50%.
Figure 3-22: (a) Measured temporal signal for the case of 12dBm of RF input power and (b)
modulation depth as a function of the RF input power for the case of λ=1552.1nm and V DC=1V.
It is observed that no over-modulation is produced.
102
3.3. Frequency mixing application
103
Chapter 3: Analog applications
Many works have been published about frequency mixing techniques using
commercial LiNbO3 modulators, but silicon photonics mixers had not been reported so
far. In [GOP93] and [KAR08] microwave optical mixing in different configurations of
MZI modulators (single MZI and a cascade interferometric modulator pair) was
investigated. The authors demonstrated that the modulators could be employed for
down-converting anywhere over a 15GHz [GOP93] and 18GHz [KAR08] frequency
span, but both with large conversion losses (CL). The up-converting RF frequency
feature of a single MZI was simulated and tested at 28GHz in [MART00] this time with
lower CL. We can now consider, for example, applications of microwave photonic
mixing to RoF mobile communications and WLANs, where the desired mixing
frequency at the output of the system can be in the 60GHz range. In this field, much
study has been done on optical transmission of UWB signals. For a UWB signal at high
frequency, one first attempt at up-conversion was made at around 40GHz with cascaded
modulators but with again large CL [NGU07]. Later, generation of a 60GHz UWB
signal over fiber by up-conversion using cascaded EOMS was demonstrated by the
same authors in [NGU09] with a suitable error vector magnitude value of 12.6%.
Recently, a down-conversion frequency process with a new RoF architecture by using a
combination of two optical phase modulators (OPM) and optical filtering has been
demonstrated high frequency-conversion features [PAG11]. A thoroughly overview of
optical techniques used for frequency conversion or mixing can be found in [CAB10],
which gives the advantages and drawbacks of all techniques presented as well as the
research results. The next Table 3-2 summarizes the CL results obtained in these works
(it must be remembered that all of them are based on LiNbO3 commercial modulators).
Table 3-2: Reported results of the conversion losses in microwave photonic frequency
conversions by using commercial EOMs.
104
3.3. Frequency mixing application
MWP mixer) and a frequency down-conversion from 1.05GHz to 50MHz with low
conversion losses of 10dB was reported. In silicon, two down-converting links were
demonstrated based on a compact and fully tunable silicon optical filter [YU13, TU10]
but no silicon photonic mixer as device was used.
In a mixing process if we want to produce an output frequency that is lower than the
second input signal frequency, then the process is called down-conversion and the RF is
the input and the IF is the output. On the other hand, if we want to produce an output
signal that is at higher frequency than the second input signal, it is referred to as up-
conversion and the IF is the input and RF is the output. Figure 3-23 illustrates a
frequency domain representation of down-conversion (a) and up-conversion (b)
processes. As Figure 3-23 depicts, IF/RF signals tend to be information bearing signals
(as denoted by the broadened spectra surrounding the RF and IF center frequencies).
During the frequency conversion, the information carried by the RF (IF) signal is
frequency translated to the IF (RF) output.
Figure 3-23: Definitions of frequency down-conversion (a) and frequency up-conversion (b)
mixing processes.
105
Chapter 3: Analog applications
To produce a new frequency (or news frequencies) a nonlinear device is required and
in principle any nonlinear device can be used to make a mixer circuit. These
nonlinearities will be responsible for the generation of many additional frequencies
other than the pure sum and difference frequencies. So that, electro-optic modulator can
be a simple but efficient method of generating mixing products and converting
frequency in a microwave photonic link by taking advantage of their nonlinear transfer
function (and the nonlinearity produced by the material in the case of silicon-based
modulator) which brings several harmonic and intermodulation products as it has been
studied in the previous section 3.2.
106
3.3. Frequency mixing application
Figure 3-24: Schematic of the slow-wave MZI modulator used for mixing. The MZI is
asymmetric with a length difference of 180µm and the phase shifter in the shorter arm is 1mm
of length. MMI are used for divide the input light and combine the output modulated signal.
Figure 3-25: Measured transmission spectrum of the slow-wave silicon-based modulator. The
photonic band-gas and both slow-light and fast-light resonances chosen from the experiment are
clearly differenced.
107
Chapter 3: Analog applications
Due to the asymmetric MZI structure used for the experiment, the bias point can be
tuned either through acting on the DC voltage or on the operation wavelength in the
same way that in the asymmetric MZI used in previous section 3.2. In this case, the
experimental measurements were carried out by tuning the wavelength to drive the
modulator to the appropriate bias operating point and utilizing a fixed 2V DC voltage.
Thus, if we observe the measured modulator transmission spectrum, illustrated in the
Figure 3-25, we can distinguish the operating wavelength in the slow light region for
QB and MITB, close to the photonic band-gap due to the of slow-light effect, where the
measured group index is around 11, from the operating wavelength in the fast-light
region also for quadrature and minimum transmission, away from the band-gap, and
where the measured group index is around 4.5.
To study and measure the frequency up-conversion, the IF and OL frequencies are
set to 1GHz and 9.25GHz respectively. The experimental setup used is shown in Figure
3-26. The input light emitted by an ECL is coupled from standard single mode fibers to
the chip via grating couplers (vertical coupling explained in Appendix A). The
polarization is optimized via the use of a PC. Two signal generators operating at
fIF=1GHz and fOL=9.25GHz are power combined and the two tone input is coupled to a
DC bias of 2V using a bias-Tee and applied through high speed GSG probes to one port
of the slow-light MZI modulator while the other port is terminated with a 50Ω load. The
output modulated signal is amplified with an EDFA, filtered by a tunable OF and photo-
detected by a high speed photodiode. Finally, a vectorial signal analyzer is used to
measure the received signal. Electrical amplification is performed with a low noise
amplifier featuring 26dB RF gain and 6dB noise figure (details in section 3.5).
Figure 3-26: Schematic of the experimental set-up used for measuring the frequency up-
conversion mixing process.
108
3.3. Frequency mixing application
To evaluate the mixing process performance of our silicon modulator, we defined the
CL as the ratio between the photo-detected electrical power at fIF and the photo-detected
electrical power of the considered intermodulation term at fRF=NfIF+fOL (N=1 for IMD2
and N=2 for IMD3) as in [MART00].
Pf IF Pf IF
CL (3.45)
Pf RF PNf IF fOL
Figure 3-27: Conversion losses at QB and MITB bias points for both slow and fast-light regions
at IMD2 intermodulation frequency. OL input power was set to be +20dBm.
On the other hand, lower conversion losses were also obtained in the slow light region
(SLR) when compared to the fast light region (FLR), ~2-4dB more at MITB and ~1dB
at QB, as depicted in Figure 3-28. As a result, the IMD2 frequency (10.25GHz) and
MITB operation point was chosen to up-convert a vector modulated signal.
109
Chapter 3: Analog applications
Figure 3-28: Conversion losses at QB and MITB bias points for both slow and fast-light regions
at IMD2 intermodulation frequency. OL input power was set to be +20dBm.
Moreover, the SFDR, defined as the range of output powers over which no
intermodulation signal is above the noise level implying that no disturbance of the
desired signal is produced, was also measured. For our mixing functionality, we
consider that the signal to which we will later up-convert a modulated signal is the
fIF+fOL product (IMD2). Therefore, if one qualitatively observes the output frequency
spectrum diagram of a mixer (Figure 3-29), it is realized that the critical combination
product in that case is 2fIF+fOL (IMD3) since it is the closest and therefore the most
difficult to filter.
Figure 3-29: Output frequency spectrum diagram of a mixer in which the OL frequency is very
higher than IF frequency.
As illustrated in Figure 3-30, our SFDR is the output power difference between the
intersection of the IMD3 power signal at 11.25GHz with the noise floor and the IMD2
110
3.3. Frequency mixing application
power signal at 10.25GHz. SFDR values of 62.8dB.Hz2/3 and 60.2dB.Hz2/3 for the slow
and fast light regimes have been measured respectively.
Figure 3-30: IMD2 and IMD3 powers as a function of the IF input power at MITB. SFDR for
both (a) slow and (b) fast light region are represented. Noise floor level has been measured at -
141dBm/Hz. OL input power was set to be +20dBm.
These results confirm again the enhanced performance brought by the slow-light regime
of the modulator. In both cases, a third order intercept point around ~25dBm has been
measured. For completeness, we note that intermodulation terms grow as the square and
the cube of the input power for IMD2 and IMD3, respectively. It must be stated here
that, obviously, lower SFDR values have obtained with respect the reported values in
Table 3-2 with commercial LiNbO3, due to the high system noise floor obtained, but by
contrast excellent conversion losses are obtained, so, there is a good trade-off between
them, besides the advantages of our silicon-based photonic mixer in terms of CMOS
compatibility and fully integration capability.
Apart from the SFDR, another performance figure typical to characterize MWP links
is the error vector magnitude (EVM) which is measured at the receiver and gives a good
indication of the signal quality. The EVM parameter is used in most wireless system
standards such as IEEE802.11g/b, UMTS, WiMAX, etc. to calculate the performance of
a transmitter/modulator or receiver/demodulator implementing the respective standard.
Following the optical transmission, the lower EVM, the lower deterioration the signal
will have suffered. An error vector is a vector in the I-Q plane between the ideal
transmitted constellation point and the point received by the receiver. In other words, it
is the difference between actual received symbols and ideal transmitted symbols as
shown in Figure 3-31. The average power of the error vector, normalized to signal
power, is the EVM.
111
Chapter 3: Analog applications
Figure 3-31: Definition the error vector magnitude in the I/Q plane.
Some examples of EVM requirements can be found. ZigBee (IEEE 802.15.4) tolerate
an EVM of up to -9dB (35%) [INS07]. The IEEE 802.16e-2005 standard prescribes that
a Mobile WiMAX transmitter must have a minimum EVM of -15dB (17.7%) for a
QPSK½ modulation and -20.5dB (9.4%) for 16_QAM½. [IEEE05] Finally, EVM
threshold to ensure communication between Ultra Wideband (UWB) devices is -14.4
dB (19%) as [ECMA368]. As example, the silicon-based MZM examined in [VAC10]
allows converting 802.11 QAM modulated OFDM signals to optical domain achieving
an EVM of -30dB (3%) a data rate of 54Mb/s.
Figure 3-32: Image obtained from the signal analyzer of the received constellation at 10.25GHz
in the slow-light region for 17dBm of IF input power. It is highlighted the average value of the
EVM parameter and the frequency.
112
3.3. Frequency mixing application
For the EVM measurements and for the test scenario presented here, we choose an
I/Q modulated signal at the IF frequency of f IF=1GHz under a digital modulation
scheme of QPSK with 20Mbps up-converted to a frequency of fIF+fOL=10.25GHz with
the electro-optic modulator biased at MITB point. The experimental setup is the same to
that depicted in Figure 3-26 with the IF signal generator and vectorial signal analyzer
properly programmed for such measurements. Figure 3-32 depicts an image of the
received constellation at 10.25GHz obtained from the vector signal analyzer.
Finally, Figure 3-33 shows the measured EVM (in dB in the left axis and in % in the
right axis) in both slow and fast light regions. It is clear from the figure that enhanced
performance in terms of EVM is obtained in the slow light regime over the entire range
of IF input powers. Outstanding results have been obtained with a very low optimum
EVM of 8.5% and 12% for both slow and fast light regimes, respectively, allowing a
very clean constellation. This suggests that the received signal is of excellent quality.
Furthermore, the curve meets the expected behavior and decreases for low IF input
power (due to the electrical noise) since an optimum point is utilized. It should be
noticed that from this optimum IF input power point (17dBm), the curve increase due to
the IF power source saturation.
Figure 3-33: EVM as a function of IF Input Power for both fast and slow light propagation.
Inset shows received constellation in the slow light region case at the optimum point of IF input
power (+17dBm).
We have shown in this section a very efficient method of generating mixing products
in a silicon modulator exploiting the nonlinearity produced by both the MZI transfer
function and the plasma dispersion effect in silicon. Furthermore, the conversion
efficiency has been enhanced via slow-light propagation in the active region of the
modulator appearing the slow-light effect as a potential technology for building
113
Chapter 3: Analog applications
microwave devices. Optimum performance has been achieved at MITB with low
conversion losses around 7dB and large SFDR of 62.8dB.Hz2/3. As a matter of
evaluation, a 20Mbps QPSK signals at 1GHz frequency has been successfully up-
converted to 10.25GHz without the need for high-frequency complex RF components.
A minimum EVM of about 8% has been achieved, meeting well enough the EVM
requirements for an acceptable quality of the received constellation. Overall, this simple
photonic silicon slow-light based mixer scheme can implement a promising
functionality for future broadband microwave-millimeter-wave communication
applications in the IWMP scope.
An MWP signal processor is a photonic subsystem designed with the aim of carrying
equivalent tasks to those of an ordinary microwave filter within an RF system or link. In
addition, with the development of the increasingly complex analog photonic link (APL)
[UR11], more advanced functionalities are required to be explored such as high
selectivity and tunable filtering. So, significant efforts have been focused on the
implementation of flexible filters free from bandwidth constraints which generally are
required to be reconfigurable filters with narrow bandwidth and wide FSR. Although
the electronic-based designs report high performance they usually cannot offer
wideband responses with high reconfigurability. Moreover, the possibility of integrating
MWP filters in a single chip is highly attractive since it will open the way to structures
with reduced footprint, lower power consumption, higher stability, reliability and cost
reduction due to the possibility of mass production. Thereby, the silicon-based MWP
signal processors are believed to be a promising solution to these challenges in the
complex APL. In this way, silicon photonics is currently one of the most preferred
114
3.4. Microwave photonic filtering application
All-optical signal processing in SOI lattice filters can support the fully reconfigurable
optical transfer function by cascading many identical unit cells in different ways. A unit
cell is a basic building block used to construct higher-order filters. Thereby, the
versatility and operability of the filter depends on the functionality of the unit cell that
can be configured to have any first order optical transfer function with any filter taps
configuration. In the last five years, several integrated SOI-based microwave photonics
filters have been reported [RAS07, FEN10, DON10, ALI11, IBR11, YU13] carrying
out the task of filtering in the optical domain.
115
Chapter 3: Analog applications
Figure 3-34: General concept diagram of (a) traditional RF filter and (b) microwave photonic
filter.
116
3.4. Microwave photonic filtering application
signal processing where a number of weighted and delayed samples of the RF signal are
produced in the optical domain and combined upon detection. According to the number
of samples N, the filter can be classified as either a finite impulse response (FIR) filter
which combines at their output a finite set of delayed and weighted taps of the input
signal or an infinite impulse response (IIR) filter which is based on recirculating cavities
to provide an infinite number of delayed and weighted replicas of the input signal. In
this thesis, we will demonstrate a silicon-based FIR filter so we will center in this kind
of filters.
Provided that the nonlinearity in the system is small and negligible, the entire system
can be considered as linear and time-invariant, in which the output y(t) can be written
as:
N 1
y(t ) an x(t nT ) (3.46)
n 0
where x(t) is the RF input signal, T is the time delay difference between two adjacent
taps and an is the coefficient or weight of the n-th tap (if all taps have the same absolute
value the filter is uniform). Applying the Fourier transform, the FIR filter transfer
function is then obtained,
N 1
j nT
H( ) an e (3.47)
n 0
where ω=2πfRF. The equation (3.47) shows that the filter spectrum is periodic with a
period given by fRF=1/T and known as free spectral range (FSR). The selectivity of any
of the pass-bands or resonances is given by the -3dB bandwidth and the filter rejection
is obtained through the relation between the main to secondary sidelobe ratio (MSSR).
Figure 3-35 shows an example of a microwave photonic filter spectrum with the most
important parameters.
Figure 3-35: Typical response of a uniform 5-tap ([1 -1 1 -1 1]) photonic microwave filter
showing the most important parameters.
117
Chapter 3: Analog applications
Moreover, it can be pointed that the MWP filter flexibility, which is achieved by acting
over the different parameters in equation (3.47), is evaluated in terms of tunability,
reconfigurability and selectivity. The first one refers to the possibility to dynamically
change the position of filter resonances (the central filter frequency). To provide
tunability it is necessary to alter the value of the sampling period T and therefore the
FSR. The reconfigurability refers to the possibility to dynamically change the values of
coefficients an which results in a reduction of the sidelobes and therefore in a filter
shape change (changing the MSSR). Finally, the filter selectivity is dictated by the
number of samples (N=2 provides a notch filter and N>2 a band-pass filter) which can
be increased by cascading the filter unit cell. The effect of the relevant parameters on
the filter response is illustrated in the next Figure 3-36.
Figure 3-36: Examples of filter flexibility changing sample parameters to obtain tunability (T),
selectivity (N) and reconfigurability (|an|)
Broadly, the usual implementation of microwave photonic FIR filters can follow two
approaches. The first one is based on a single optical source in combination with
multiple delay-lines implemented, for example, by a set of optical fibers or waveguides.
This simple scheme does not allow tuning as this would require changing the T value
118
3.4. Microwave photonic filtering application
and the weight of the n-th tap (an) is changed by inserting loss or gain devices in the
fiber. The other one, followed in this thesis, is based on the combination of a dispersive
delay line and different optical carriers where the T value is changed by tuning the
wavelength separation among the carriers, thereby allowing tunability and an is readily
adjusted by changing the optical power emitted by the optical sources. The last, is
considered a wavelength-division-multiplexing architecture.
The silicon-based MZI modulator is that we already used for the distortion analytical
model validation in section 3.2 and it is shown again in Figure 3-37(a). Static
performance measurements have been firstly carried out to characterize the modulator.
Figure 3-37(b) shows the effect of varying DC voltages from 0V to 10V on the
normalized transmission spectrum of the modulator. It is shown that a large static
extinction ration close to 30 dB is achieved at wavelengths around 1551.5 nm. The
voltage-induced wavelength shift, Δλ, is used to calculate the effective index variation
from the Equation (3.42) with LA=1mm as the modulation length. We obtained that the
effective index variation changes from 0 to 3.10-4 when the voltage increases from 0V
to 10V. The VπLπ product, commonly used to evaluate the efficiency of the modulator,
has also been calculated from such effective index variations obtained from the voltage-
induced wavelength. A nearly constant VπLπ value around 2.25 V.cm has been obtained.
The insertion loss of the modulator is around 12dB which includes the MMI losses.
119
Chapter 3: Analog applications
Figure 3-37: (a) Schematic of the silicon-based 1mm MZI electro-optic modulator for
implementing the proposed photonic RF filter. The length difference between both arms is
180μm. Inset shows an optical microscope image of the traveling-wave ground–signal–ground
(GSG) coplanar electrodes, (b) Normalized transmission spectrum variation with the DC
voltage.
Figure 3-38 shows the block diagram of the proposed dispersion-based photonic
microwave filter. This filter architecture already mentioned in the previous point is
well-known for optical processing of RF signals. The principle of operation is based on
modulating a multi-wavelength optical source with the input microwave signal (RFin).
The modulated signal is then propagated through a dispersive medium and photo-
detected to obtain the filtered microwave signal (RFout). Hence, each optical source
implements a filter sample that is selectively delayed by employing the dispersive
medium. The amplitude of each sample can be changed by using different techniques
but the simplest way is controlling the output power of the optical sources as it has been
already commented in the theory section. In such a way, a FIR filter can be
implemented with transfer function expressed by Equation (3.47).
The main advantage of this filter architecture is a very stable response due to
operation in the incoherent regime. Furthermore, the filter transfer function can be
reconfigured by dynamically changing the relative power of the optical wavelengths and
it can be tuned by controlling the dispersive medium [VID03]. However, positive and
negative filter coefficients are needed to allow a full versatility in the transfer function
thus enabling flat-pass bands or high out-of-band suppression [CAO03, VID06]. The
equivalent of positive and negative coefficients can be obtained by using opposite
slopes of the modulation transmission response. The principle of the technique used is
120
3.4. Microwave photonic filtering application
based on the π-phase inversion which happens in the Mach-Zehnder modulators. The
relative sign experienced by the microwave signal depends on the slope of the
modulator transmission spectrum. If two optical carriers modulated by a microwave
signal are placed at wavelengths that correspond with opposite slopes in the modulator
response, as depicted in Figure 3-39, the microwave signals after photo-detection will
show the same average power but a π-phase shift. In this way, positive and negative
coefficients for a FIR filter can be obtained while keeping the stability inherent to
incoherent operation if the input microwave signal is modulated over both slopes.
Figure 3-39: Concept of the π phase inversion suffered by a microwave modulating signal over
the optical transfer function of a MZM as a function of the wavelength for a fixed DC voltage.
This technique has been employed with only one single LiNbO3 Mach-Zehnder
modulator by exploiting the Vπ dependence with wavelength [VID06]. However, this
dependence was very weak and optical sources with a very large separation (1300 and
1550 nm) were required. That makes the filter implementation complex, introduces high
loss at 1300 nm (more than 20 dB) since the modulator is not optimized for operation at
both 1550 and 1300 nm, prevents the use of EDFAs and limits further integration with
other subsystems such as on-a-chip dispersive delay lines [SAN12].
In this work, we propose this approach for achieving positive and negative
coefficients but this time based on exploiting the asymmetry of a silicon-based MZM.
The asymmetry of the modulator is achieved by introducing a length difference of only
180 µm between the MZM arms. Such asymmetry introduces a strong dependence with
wavelength in the modulator response. If, for instance, two optical carriers modulated
by a microwave signal are placed at wavelengths that correspond with opposite slopes
in the modulator response, the microwave signals after photo-detection will show a π-
phase shift. So, by properly choosing the wavelength of the optical carriers, it is
possible to control the sign of the corresponding filter coefficient. For our experiment,
we choose a 4V of DC voltage and over the transmission spectrum for this voltage we
121
Chapter 3: Analog applications
choose the wavelengths such that correspond with opposite slopes in the 4V modulator
response. For the two-tap experiment we used two optical lasers emitting at
λ1=1550.75nm and λ2=1552.30nm while for the three-tap one we used an array of lasers
emitting at λ1=1547.51nm, λ2=1549.13nm and λ3=1550.75 nm (Figure 3-40). The
L=10km fiber coil provides a dispersion of D=17ps/km.nm which, for a wavelength
separation between the two optical carriers of 1.55nm for the two-tap case, corresponds
to a time delay between adjacent samples of T=263.5ps (T=D.L.Δλ) or in other words a
free spectral range of FSR=1/T=3.8GHz.
Figure 3-40: Normalized modulator transmission spectrum for a fixed 4V applied DC voltage
showing the chosen wavelengths for (a) two-tap experiment and (b) three-tap experiment.
The experimental setup used for the experiment is shown in Figure 3-41.
Figure 3-41: Experimental setup used to characterize the photonic RF filter performance.
122
3.4. Microwave photonic filtering application
A set of optical CW carriers are modulated by the silicon modulator. The RF signal
from a vector network analyzer is combined with the DC voltage using a bias-Tee and
applied to the modulator through a broadband GSG probe. The output of the electrode is
terminated by a DC-block and a 50Ω load using another GSG probe. After modulation,
the signals are amplified with an EDFA, filtered by a tunable OF and launched to a
dispersive medium (a reel of 10 km of single-mode fiber). Chromatic dispersion in the
fiber will introduce a time delay between the optical carriers. Then the optical signal at
the output of the fiber coil is photo-detected and the delayed electrical signal is injected
into the vector network analyzer. If a sweep of the frequency of the modulating
electrical signal is carried out, the photonic microwave filter response is measured
through the S21.
Figure 3-42 shows the normalized filter response between 0-15 GHz for a 4V applied
DC voltage when two optical carriers (λ1=1550.75nm and λ2=1552.30nm) enter the
modulator with the same output power at the input of the photodiode. These
wavelengths are placed at opposite slopes of the modulator response as it is shown in
the Figure 3-40(a), resulting in filter coefficients [a1 a2]=[-1 1]. A very good agreement
between theory, described by Equation (3.47), and measurements can be seen in such
figure, demonstrating that two coefficients of opposite sign have been obtained.
Figure 3-42: Experimental two-tap filter response with amplitude distribution [1 -1]. The solid
line represents the experimental results and the dotted the theoretical estimation.
Figure 3-43 depicts the filter response when three optical carriers (λ1=1547.51nm,
λ2=1549.13nm and λ3=1550.75nm) are fed to the modulator with 4V applied DC
voltage. The filter response agrees again very well with the theoretical response
corresponding to the amplitude distribution [a1 a2 a3]=[-1 1 -1] (Figure 3-40(b)) for
frequencies up to 10 GHz.
123
Chapter 3: Analog applications
Figure 3-43: Experimental three-tap filter response with amplitude distribution [1 -1 1]. The
solid line represents the experimental results and the dotted the theoretical estimation.
By changing the number of taps (that is, increasing the number of optical carriers) we
have demonstrated the selectivity of the filter. Also, since the output powers (amplitude
of each tap) of the lasers can be adjusted independently at high speed, the shape of the
filter transfer function can be changed therefore showing the reconfigurability allowed
by the filter. Further, in this case tunability by changing the T delay between optical
carriers through the change of de wavelength difference between optical carriers is not
possible due that they must be properly set in the wavelengths which correspond to QB
bias (3dB respect to the maximum) of the MZI response. However, an alternative way
to achieve filter discrete tuning consists on changing the length of the fiber path
thorough a switched dispersion matrix (SDM) as in [VID03]. Therefore, since the filter
tuning performance depends on the length of the fiber paths, it is also possible to choose
the fiber lengths allowing discrete tuning over the whole filter spectral response margin.
Furthermore, this additional component (SDM) could be integrated together with the
proposed silicon modulator for building a CMOS-compatible PIC.
124
3.5. Active RF Characterization set-ups
In this point, the set-up used for the distortion products (both harmonic and
intermodulation) measurements is reviewed in detail. For such task it is necessary to
visualize and measure the frequency spectrum of the signals. For that, the main
component used is the vectorial signal analyzer (VSA) which allows us to display
features of the spectrum of the signal showing all distortion products and their power
levels (for the SFDR, CDR and CL measurements), the constellation diagram of
received signals and error vector magnitude measurements. This is the main set-up used
in section 3.2 for the MZI distortion analytical model validation and the analog
applications of frequency mixing by using a slow-wave MZI in section 3.3, although
slight differences are shown between them. We are going to explain in detail the set-up
used for the analytical distortion model validation since it is the most complex of both.
Light emitted by a CW laser is injected into the modulator via grating couplers
(vertical coupling). The transmitted power is optimized through the use of a polarization
controller (PolaRITE™ General Photonics). The description of coupling techniques as
well as the passive characterizations set-ups is given in Appendix A.
A single tone test was performed so, a RF signal operating at fRF is delivered by a RF
signal generator (R&S SMR-50 with pulse modulation capability). If two-tone is
required, as in section 3.3, two RF signals operating at fIF and fOL delivered by two RF
signal generators are power combined by an electrical hybrid (Aeroflex Weinschel
1575). The single (or two) tone input is coupled a DC bias voltage supplied by a DC
Power Source (Agilent E3647A) via a bias-Tee (SHF BT-110) and applied through high
speed ground-signal-ground (GSG) input probes to the travelling wave electrodes
terminated externally by a 50 Ω load coupled to a DC block (SHF DCB-110) through
high speed GSG output probes to avoid reflexions (in section 3.3 neither GSG output
probes nor DC block are used since the electrode configuration is lumped electrode).
125
Chapter 3: Analog applications
With the vector signal analyzer we display the spectrum of the electrical received
signal and we can measure the fundamental and distortion products levels which allow
us to obtain the SFDR, the CDR and the CL (in the case of the mixing application).
Further, for the EVM measurements we must obtain the received constellations of an
I/Q modulated signal. In such case, we properly programme the RF signal generator and
vector signal analyzer.
The next Figure 3-44 depicts again the schematic of the set-up but this time showing a
real picture of each component used.
Figure 3-44: Schematic of the characterization setup and pictures of each component used in
section 3.2 for distortion analytical model validation to evaluate the distortion products. The
setup used in section 3.3 for the frequency mixing applications respectively is part of this one.
126
3.5. Active RF Characterization set-ups
For completeness, pictures of the entire real set-up used in the Nanophotonics
Technology Center facilities are shown in Figure 3-45.
Figure 3-45: Pictures of the characterization setup used in section 3.2-3.3 for the distortion
products measurements (and modulation depth in section 3.2 through temporal signals using the
DCA).
On the other hand, the set-up used for small signal characterization has been used in
section 3.4. Such characterization was necessary to demonstrate the filtering application
(by using a silicon-based MZI). Small signal measurements consist basically on the
determination of the electrical and electro-optical frequency response of a given
electrical or electro-optical device via the calculation of the scattering parameters. In
our case, we were interested on measure the S21 parameter to visualize the filter
response. For such purpose, the main component used is the vector network analyzer.
The experimental setup used for the experiment is shown in Figure 3-46 (and real
pictures of each component). A set of optical CW carriers emitted by a set of CW lasers
are combined by an optical star coupler, amplified by an EDFA (Keopsys BT2-C-18-
BO-FA) and injected into the modulator via grating couplers (the description of vertical
coupling is given in Appendix A). The transmitted power emitted by each CW laser is
optimized through the use of a polarization controller (PolaRITE™ General Photonics).
The small electrical signal delivered by the port 1 of the RF vector network analyzer
127
Chapter 3: Analog applications
Figure 3-46: Schematic of the characterization setup and equipment pictures used in section 3.4
for obtaining the filter response thought the S21 parameter measurement.
For completeness, pictures of the entire real set-up used in the Nanophotonics
Technology Center facilities are shown in Figure 3-47.
128
3.5. Active RF Characterization set-ups
Figure 3-47: Pictures of the characterization setup used in section 3.4 for the filter response
measurements (through the S21 parameter).
129
130
Chapter 4
The objective of the chapter centered in digital applications was to develop silicon
modulation structures in order to enhance the performance of conventional MZI
modulators and to demonstrate that the incorporation of optical phase modulators into
slow-wave and resonant structures can offer advantages in both reduced device footprint
and power consumption meanwhile high speed operation is maintained. This will offer
the prospect of compact optoelectronic devices integrated with CMOS circuits on the
same silicon substrate. This objective has been successfully accomplished based on the
obtained results, which can be summarized as:
131
Chapter 4: Conclusions and future outlook
As future work in the digital approach, it is proposed the CMOS integration of drivers
and DC circuitry and the assembling and packaging of the silicon modulators. Respect
to slow-light modulators the optimization of the travelling wave electrodes for the
modulator operation in the slow-light regime and investigate alternative designs to
increase the optical bandwidth with use of engineered slow-light structures are some of
the future directions proposed. On the other hand, for the RAMZI modulator, a possible
enhancement can be adding an additional ring placed in the other arm of the MZI. This
structure will be useful to overcome the difference of attenuation between both arms of
the MZI and so maximize the extinction ratio.
The chapter devoted to analog applications presents different researches carried out
in the analog scope for application in the field of IMWP, all of them using CMOS-
compatible electro-optic silicon-based MZI modulators which validate the potential of
silicon CMOS photonics and SOI technology as a promising approach for enabling
large scale photonic integration driving the development of integrated microwave
photonics applications. The work achieved in this scope can be summarized as follows:
132
Chapter 4: Conclusions and future outlook
As possible direction for further research in this scope it can be suggested that it would
be interesting to demonstrate the performance enhancement techniques presented here at
a higher frequency range, i.e. at microwave or millimeter-wave frequency range, since
nowadays the trend in microwave photonics systems are moving toward high
frequencies.
Overall, we believe that the results presented in this work may have both scientific
and commercial importance as proof of concept. It explores silicon photonic modulation
devices through extensive empirical validation and characterization, as part of the
requited photonic building blocks for use in analog and digital communication systems.
133
134
Appendix A
Conventional single-mode SOI nanophotonic waveguides are about 220 nm thick and
500 nm wide. However, standard single-mode fibers (SMF) are very large size
compared to nanophotonic SOI waveguides. Typical core diameter size of a standard
SMF is about 8μm. Thus, coupling light from an optical fiber to a nanophotonic
waveguide becomes a difficult task. The high difference in the dimensions between the
fiber and the waveguide causes a high mismatch between their optical modes. Due to
this mismatch, a coupling structure is needed to adapt a wide fiber to a narrow silicon
waveguide improving the coupling loss. To solve the fiber-chip coupling problem, two
main strategies have appeared in literature, depending on the physical direction in which
the fiber is coupled to the chip: vertical and horizontal coupling techniques.
135
Appendix A: Horizontal & vertical Coupling and passive characterization
facet quality and the use of lensed fibers may improves the coupling efficiency. A
schematic of configuration is shown in Figure A-1.1:
Figure A-1.1: Standard fiber-to-chip horizontal coupling technique by using lensed fibers and
inverted taper.
In the Figure A-1.2, a schematic of the basic set-up with horizontal coupling for passive
characterization (optical measurements) is shown.
Figure A-1.2: Schematic of the passive characterization set-up with horizontal coupling. Picture
of each component is also shown.
136
A.1. Horizontal coupling
The next pictures in the Figure A-1.3 are (a) the entire set-up, (b) the holders where the
sample is positioned and the 3-axes translation stages and (c) a detail of the fiber-chip-
objective configuration.
Figure A-1.3: Pictures of the horizontal coupling set-up used in the Nanophotonics Technology
Center laboratories during the thesis development.
137
Appendix A: Horizontal & vertical coupling and passive characterization
Figure A-2.1: Standard fiber-to-chip vertical coupling technique by using grating couplers.
As excitation source, an external cavity laser (ECL Photonetics) is used. After the laser,
in the same way that horizontal coupling, the input polarization state is firstly set by
adjusting an external polarization controller (PC). After the polarizer, the light is
injected into the chip with the selected polarization, through vertical coupling with a
single mode fiber tilted under an angle of 13 degrees relative to the surface normal of
the grating. Next we place an optical power meter (THORLABS PM320E) to
monitoring the received power, which were connected to a computer by a GPIB bus, in
order to measure it. The next Figure A-2.2 shows again the basic set-up with vertical
coupling for passive characterization (optical measurements) and, for completeness,
Figure A-2.3 shows pictures of the entire setup are also shown
138
A.1. Vertical coupling
Figure A-2.2: Schematic of the passive characterization set-up with vertical coupling. Picture of
each component is also shown.
Figure A-2.3: Pictures of the vertical coupling set-up used in the Nanophotonics Technology
Center laboratories during the thesis development.
139
140
List of publications
JOURNAL PUBLICATIONS
141
List of publications
CONFERENCE PUBLICATIONS
142
List of publications
OTHER PUBLICATIONS
143
List of publications
144
List of figures
Figure 1-3: (a) Analog (b) digital amplitude modulation in a MZI modulator ............7
Figure 1-4: Cross section of typical phase shifters to implementing the (a)
carrier accumulation mechanism through a MOS structure, (b) carrier
injection mechanism through a PIN structure, (c) carrier depletion mechanism
through a PN structure, used to electrically manipulate the free-carrier
concentrations in plasma dispersion based silicon optical modulator. .........................10
Figure 1-6: Ideal MZI amplitude modulator consisting of two phase shifters
integrated into both arms of the device. The operation principle is shown ..................12
Figure 1-8: Final GDSs mask layers of the fabricated modulators used in this
thesis, (a) MZIs and SL-MZI modulators and (b) MRR and RAMZI
modulators ....................................................................................................................16
145
List of figures
Figure 1-9: (a) Photograph of an 8-in SOI wafer with hundreds of photonic
components and circuits and (b) picture of a single silicon chip fabricated in
CEA-LETI consisting of 26 electro-optical modulators...............................................17
Figure 2-1: Ring resonator in Notch filter configuration and the most
important design parameters .........................................................................................26
Figure 2-2: Model of a single ring resonator with coupled to one waveguide .............26
Figure 2-6: (a) Experimental and (b) simulated extinction ratio as a function
of the ring to waveguide separation ..............................................................................33
Figure 2-8: Effective index variation as a function of applied bias voltage .................35
Figure 2-9: Schematic of the characterization setup for the data transmission
measurements ...............................................................................................................36
Figure 2-10: Optical eye diagrams with an extinction ratio of 4.5dB at 5Gbit/s
and 10Gbit/s and with an extinction ratio of 3.5dB at 12.5Gbit/s. ...............................36
Figure 2-11: (a) Schematic of the RAMZI modulator with the parameters of
the MRR and (b) Basic performance of a phase modulator based on a MRR .............37
Figure 2-13: (a) Schematic of the fabricated ring-assisted MZI modulator, (b)
cross section of the device active region, (c) top view of fabricated modulator ..........39
Figure 2-14: Theoretical ER as a function of the coupling factor and the ring
losses for the MRR based intensity modulator for different effective index
variation. Insertion losses are lower than 5dB for the values shown, where the
146
List of figures
ER is higher than 5dB (dark blue), 10dB (light blue), 15dB (yellow) and
20dB (brown)................................................................................................................40
Figure 2-15: Theoretical ER as a function of the coupling factor and the ring
losses for the RAMZI modulator for different effective index variation.
Insertion losses are lower than 5dB in the values shown, where the ER is
higher than 5dB (dark blue), 10dB (light blue), 15dB (yellow) and 20dB
(brown). The dotted red line shows the critical coupling condition. ............................40
Figure 2-19: Experimental setup for measuring the high speed optical
frequency response of the RAMZI modulator ..............................................................43
Figure 2-21: Eye diagram of the RAMZI modulator with an extinction ratio
of 6.7dB at 10Gbit/s and 2.6dB at 20Gbit/s .................................................................44
Figure 2-23: Schematic of the shallow-etched LCWG with slab height (h)
considered in this thesis to enhanced modulation [BRI11B] .......................................52
Figure 2-24: Contour plot of the Bragg wavelength @ the photonic band edge
versus width (We) and length (Wi) of the transversal corrugations for the
fundamental mode of with Λ=300nm, W=300nm and h=100nm [BRI11B] ...............53
147
List of figures
Figure 2-28: (a) Respective normalized spectra of the phase shifter (red curve,
top) and the MZI (blue curve, bottom), (b) close-up view of the MZI
spectrum for biases of 0, −3, and −5 V [BRI12B]........................................................57
Figure 2-29: Effective index change as a function of reverse bias for group
index values of ~4.4 (fast light; red curve) and ~8 (slow light; blue curve).
Modulation efficiencies VπLπ are pointed by the arrows [BRI12B] ............................58
Figure 2-30: 40Gbit/s eye diagram (a) at quadrature and (b) 2dB below
quadrature [BRI12B] ....................................................................................................58
Figure 2-31: (a) Close-up view of the slow-light region for bias voltages
varying from +0.5 to -2.5V. The dotted circle is the quadrature point, (b)
Experimental (blue solid line) and simulated (red dashed line) effective index
change versus applied voltages at 1546.9nm wavelength (measured at the
quadrature point). Measured modulation efficiencies VπLπ for each
corresponding applied voltage and taking +0.5 V as a reference are also
shown [BRI12A] ..........................................................................................................60
Figure 2-32: Eye diagrams at 5Gbit/s for varying peak-to-peak drive voltage
[BRI112A] ....................................................................................................................61
Figure 2-33: Eye diagrams at 25Gbit/s for varying peak-to-peak drive voltage
[BRI12A] ......................................................................................................................61
Figure 2-34: Eye diagrams at 10Gbit/s for varying peak-to-peak drive voltage
[BRI12A] ......................................................................................................................61
Figure 2-35: Schematic view of the MZI structure. Splitter and combiner
portions are represented by a generic box ....................................................................63
Figure 2-36: SEM picture of fabricated structures: (a) MZI, (b) Y-junction
detail .............................................................................................................................65
Figure 2-37: Detail of the measured transmission spectra of the MZI’s with
(a) ∆L=1mm, (b) ∆L=2mm, (c) ∆L=3mm. The spectral range is reduced for
clarity ............................................................................................................................65
Figure 2-38: (a) Branching ratio and (b) propagation losses versus wavelength
obtained from method ...................................................................................................66
Figure 2-39: (a) Schematic of the MZI structure used in the proposed method
to estimate the insertion losses in the slow-light phase shifter, (b) normalized
optical transmission spectrum of the MZI structure [BRI12A] ....................................68
Figure 2-40: (a) Group index and (b) insertion losses variation versus
wavelength. (c) slow-light phase shifter insertion loss versus group index
[BRI12A] ......................................................................................................................70
148
List of figures
Figure 2-42: Pictures of the characterization setup for the eye diagram
acquisition.....................................................................................................................73
Figure 3-1: Schematic of the fundamental concept of (a) an MWP link, (b) a
MWP system and (c) an IMWP chip ............................................................................76
Figure 3-2: (a) Single-tone and (b) two-tone output spectrum diagram of a
nonlinear device showing the frequency and the names of the generated
distortion products. .......................................................................................................81
Figure 3-4: Schematic of the setup used to carry out the distortion
characterization of a commercial LiNbO3 modulator...................................................83
Figure 3-6: Theoretical and experimental fundamental and 2nd and 3rd
harmonic distortion powers as a function of the bias voltage of a commercial
LiNbO3 MZI modulator ................................................................................................84
Figure 3-9: Contour of 3rd order (a) and 2nd (b) intermodulation products
power as a function of bias phase and DC bias voltage for a MZI with a phase
shifter length of 1mm and λ=1550nm ..........................................................................92
149
List of figures
Figure 3-11: Absorption losses variation with the DC voltage. The fitted
curve is obtained from a logarithmic nonlinear approximation. It is shown
both the maximum variation of absorption losses between 0V and 10V and
the absorption losses variation with the voltage for 2Vpp of RF modulation
voltage. .........................................................................................................................94
Figure 3-14: (a) Schematic of the PN junction located at the middle of the
active waveguide phase shifter, (b) optical microscope image of the fabricated
modulator with a clear view of the traveling-wave ground–signal–ground
(GSG) coplanar, electrodes, (c) detail of the GSG electrode .......................................96
Figure 3-16: Effective index change with the voltage. The measured data are
obtained from the measured voltage-induced wavelength shift through
equation (3.42). The fitted curve is obtained from nonlinear approximation
described in the equation (3.4) .....................................................................................98
Figure 3-17: Schematic of the experimental set-up for the single-tone test
carried out to experimental validation of the model .....................................................99
Figure 3-18: Contour of the fundamental (a), the 2nd (b) and the 3rd order (c)
harmonic products powers as a function of DC voltage and wavelength for
our fabricated MZI........................................................................................................100
Figure 3-19: Measured values and obtained curves from the analytical
expressions of the fundamental and harmonic distortion performance as a
function of the wavelength for a DC bias voltage of 1V. The RF input power
of the single-tone is 10dBm (2Vpp) ..............................................................................100
150
List of figures
Figure 3-22: (a) Measured temporal signal for the case of 12dBm of RF input
power and (b) modulation depth as a function of the RF input power for the
case of λ=1552.1nm and VDC=1V. It is observed that no over-modulation is
produced .......................................................................................................................102
Figure 3-24: Schematic of the slow-wave MZI modulator used for mixing.
The MZI is asymmetric with a length difference of 180µm and the phase
shifter in the shorter arm is 1mm of length. MMI are used for divide the input
light and combine the output modulated signal ............................................................107
Figure 3-26: Schematic of the experimental set-up used for measuring the
frequency up-conversion mixing process .....................................................................108
Figure 3-27: Conversion losses at QB and MITB bias points for both slow
and fast-light regions at IMD2 intermodulation frequency. OL input power
was set to be +20dBm...................................................................................................109
Figure 3-28: Conversion losses at QB and MITB bias points for both slow
and fast-light regions at IMD2 intermodulation frequency. OL input power
was set to be +20dBm...................................................................................................110
Figure 3-30: IMD2 and IMD3 powers as a function of the IF input power at
MITB. SFDR for both (a) slow and (b) fast light region are represented. Noise
floor level has been measured at -141dBm/Hz. OL input power was set to be
+20dBm ........................................................................................................................111
Figure 3-31: Definition the error vector magnitude in the I/Q plane ...........................112
Figure 3-32: Image obtained from the signal analyzer of the received
constellation at 10.25GHz in the slow-light region for 17dBm of IF input
power. It is highlighted the average value of the EVM parameter and the
frequency ......................................................................................................................112
151
List of figures
Figure 3-33: EVM as a function of IF Input Power for both fast and slow light
propagation. Inset shows received constellation in the slow light region case
at the optimum point of IF input power (+17dBm) ......................................................113
Figure 3-34: General concept diagram of (a) traditional RF filter and (b)
microwave photonic filter .............................................................................................116
152
List of figures
Figure 3-45: Pictures of the characterization setup used in section 3.2-3.3 for
the distortion products measurements (and modulation depth in section 3.2
through temporal signals using the DCA) ....................................................................127
Figure 3-47: Pictures of the characterization setup used in section 3.4 for the
filter response measurements (through the S21 parameter) .........................................129
153
154
List of tables
Table 2-1: The most important parameters of the last published MZI
modulators based on the plasma dispersion effect. ......................................................22
Table 2-5: Average theoretical ERs and experimental ERs derived and
from the transmission spectra .......................................................................................67
155
156
Acronyms
AC Alternating Current
APL Analog Photonic Link
BPG Bit Pattern Generator
CDR Carrier-to-Distortion
CL Conversion Losses
CW Continuous Wave
CMOS Complementary Metal Oxide Semiconductor
DC Direct Current
EBL Electron-Beam Lithography
ECL External Cavity Laser
EDFA Erbium-Doped Fiber Amplifier
E/O Electrical-to-Optical
ER Extinction Ratio
EVM Error Vector Magnitude
FIR Finite Impulse Response
FLR Fast-Light Region
FSR Free Spectral Range
GDS Graphic Database System
GSG Ground-Signal-Ground
HD2 2nd-order Harmonic Distortion
HD3 3rd-order Harmonic Distortion
IC Integrated Circuit
IIP3 3rd-order Input Intercept Point
IIR Infinite Impulse Response
IL Insertion Losses
157
Acronyms
158
Bibliography
[BAB08] T. Baba, “Slow light in photonic crystals,” Nature Photonics, vol. 2, no.
8, pp. 465–473, Aug. 2008.
159
Bibliography
160
Bibliography
[CHU95] Shun Lien Chuang, Physics of optoelectronic devices, John Wiley. 1995.
161
Bibliography
[COX04] Charles Howard Cox, “Analog optical links: theory and practice,”
Cambridge, 2004.
[ECMA368] International Standard, High rate ultra wideband PHY and MAC
Standard, 3rd Edition, December 2008.
162
Bibliography
[GAM00] W. A. Gambling, “The Rise and Rise of Optical Fibers,” IEEE Journal of
Selected Topics in Quantum Electronics, vol. 6, pp. 1084–1093, 2000.
163
Bibliography
[GU07] L. Gu, W. Jiang, X. Chen, L. Wang, and R. T. Chen, “High speed silicon
photonic crystal waveguide modulator for low voltage operation,”
Applied Physics Letters, vol. 90, no. 7, p. 071105, 2007.
164
Bibliography
[IEEE05] IEEE Standard for Local and Metropolitan Area Networks: IEEE Std.
802.16e-2005.
165
Bibliography
166
Bibliography
167
Bibliography
[MAR09] D. Marpaung, “High dynamic range analog photonic links : design and
implementation.” University of Twente, Enschede, The Netherlands, 27-
Aug-2009.
168
Bibliography
169
Bibliography
Gbit/s silicon optical modulator," Optics Express, vol. 19, no. 7, pp.
5827-5832, Mar. 2011.
[RSO00] Rsoft Design Group, Inc., 2000 Executive Group Blvd. Ossining, NY
10562, “www.rsoftdesign.com".
170
Bibliography
[SMI12] M. Smit, J. van der Tol, and M. Hill, “Moore’s law in photonics,” Laser
& Photonics Reviews, vol. 6, no. 1, pp. 1–13, Jan. 2012.
171
Bibliography
[TU11] X. Tu, T.-Y. Liow, J. Song, M. Yu,, and G. Q. Lo, “Fabrication of low
loss and high speed silicon optical modulator using doping compensation
method,” Optics Express, vol. 19, no. 19, p. 18029, Aug. 2011.
172
Bibliography
[XIA13] X. Xiao, H. Xu, X. Li, Z. Li, T. Chu, Y. Yu, and J. Yu, "High-speed,
low-loss silicon Mach-Zehnder modulators with doping optimization,"
Optics Express, vol. 21, no. 4, pp. 4116-4125, 2013.
[XU12] H. Xu, X. Xiao, X. Li, Y. Hu, Z. Li, T. Chu, Y. Yu, and J. Yu, "High
speed silicon Mach-Zehnder modulator based on interleaved PN
junctions," Opt. Express 20, 15093-15099, 2012.
173
Bibliography
174