EB1690
EB1690
EB1690
OF MICRORESONATORS
IN OPTICS AND
PHOTONICS
Founding Editor
Brian J. Thompson
University of Rochester
Rochester, New York
Edited by
ANDREY B. MATSKO
Central figure: Savchenkov, A.A., Matsko, A. B., Grudinin, I., Savchenkova, E. A., Strekalov, D., and Maleki, L., Optical
vortices with large orbital momentum: generation and interference. Opt. Express, 14, 2888-2897 (2006).
Top and bottom figures: Savchenkov, A.A., Matsko, A. B., Ilchenko, V. S., Strekalov, D., and Maleki, L. Direct observation
of stopped light in a whispering-gallery-mode microresonator. Phys. Rev., A 76, 023816 (2007).
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2009 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been
made to publish reliable data and information, but the author and publisher cannot assume responsibility for the valid-
ity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may
rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or uti-
lized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopy-
ing, microfilming, and recording, or in any information storage or retrieval system, without written permission from the
publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://
www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923,
978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For orga-
nizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
Preface...............................................................................................................................................xi
Editor...............................................................................................................................................xv
Contributors................................................................................................................................. xvii
ix
© 2009 by Taylor & Francis Group, LLC
The race for compactness and scalability of optical and photonic devices calls for the devel-
opment of efficient micro- and nano-optical elements. The optical resonators are impor-
tant here because they can be used in the optical signal processing systems as modulators,
filters, delay lines, switches, sensors, and so on. The number of the different types of
resonators increases every day, and the basic research of their properties is gradually and
steadily substituted with applied research. Such practical issues as efficient packaging
and robust coupling, as well as integration of the resonators into complex optical systems,
become especially important when one tries to bridge the gap between the fundamental
research and practical implementation.
There are many scientific books and reviews discussing the properties of optical microre-
sonators, and I believe that at this stage it is important to have a collection reviewing the
basic directions in the development of the practical applications of the microresonators,
which is my goal with this book. Though it is practically impossible to cover the whole
field with several contributions, I hope that this collection will provide readers with the
flavor of the applied studies in the field and will convince them that systems containing
microresonators will soon become as common and widespread as electronic devices con-
taining quartz oscillators. I also hope that this book will attract the attention of a general
audience dealing with R&D in broadly defined physics/electrical engineering areas to the
fascinating world of the microresonators. The chapters are written by brilliant scientists
and engineers working in the field and can be understood by any graduate student in the
field.
Traditional mirrored optical resonators are utilized in all branches of optics where, for
example, multiple recirculation of optical power is required to maintain laser oscillation,
to increase the effective path length in spectroscopic or resolution in interferometric mea-
surements, and to enhance wave mixing interactions. Crucial properties of the resonators,
such as high quality (Q) factor and finesse, can be achieved with the highest reflectivity
and low-loss mirrors. Despite their versatility, these resonators have remained fairly com-
plex devices. They are prone to vibration instabilities because of relatively low-frequency
mechanical resonances. Stability and small modal volume are of great importance for
practical applications; however, miniaturization of conventional Fabry–Perot resonators is
either complicated and expensive, or yields rather low Q-factors.
This book contains several reports on the progress in the rapidly growing field of mono-
lithic micro- and nano-resonators. Such resonators do not have localized mirrors as such.
The light is confined inside these resonators due to their morphology. The monolithic reso-
nators are characterized by the unique combination of properties unreachable in other
resonator structures. They have tiny volumes along with huge finesse and Q factors. The
modal spectrum of the resonators can be efficiently engineered. These properties make the
resonators extremely efficient in multiple applications.
The first chapter in this book, authored by Takasumi Tanabe et al. (NTT Corporation,
Japan), is devoted to photonic crystal-based resonators (nanocavities). Among various
microresonators, photonic crystal nanocavities have the smallest mode volume (V) and
nearly the highest Q/V value. High Q/V devices are attracting considerable attention
because they enable multiple quantum and nonlinear optics applications. Recent progress
on these ultrahigh-Q photonic crystal nanocavities is also discussed in the first chapter.
xi
© 2009 by Taylor & Francis Group, LLC
Various designs of photonic crystal nanocavities, fabrication and the characterization tech-
nologies are reviewed. In addition, various applications like light buffering, slow light
propagation, all-optical switching, and bistable memory operation are discussed by the
authors.
The second chapter, authored by Charles Santori et al. (Hewlett-Packard Laboratories,
USA), is devoted to the discussion of applications of a particular type of distributed feed-
back microresonators called “pillar microcavities”. These microcavities are well suited for
efficient coupling of dipole emitters to a single mode in free space and thus are suitable for
generation of photons on demand. The design, fabrication and characterization of single-
photon devices based on single InAs quantum dots coupled to pillar microcavities formed
from AlAs/GaAs distributed-Bragg-reflector mirrors are described in this chapter. Several
applications including quantum cryptography and entanglement formation through two-
photon interference are presented. Future applications that could be developed as the
devices improve are also discussed.
Chapters 3 through 11 deal with the resonators in which the closed trajectories of light
are supported by a variety of total internal reflection in curved and polygonal transpar-
ent dielectric structures. The circular optical modes in such resonators, frequently called
whispering gallery modes (WGMs), can be understood as closed circular beams supported
by total internal reflections from the boundaries of the resonators. High values of Q-factor
can be achieved in WGMs of very small volume, in certain cases as small as cubic wave-
length, with appropriately designed dielectric interface and with use of transparent mate-
rials. Applications of the microresonators made of various materials, including silicon,
fused silica, fluorite, lithium niobate, and polymers are discussed in these chapters.
These resonators have cylindrical, spherical, spheroidal, toroidal, ring, and other shapes and
topologies. When the reflecting boundary has high index contrast, and radius of curvature
exceeds several wavelengths, the radiative losses, similar to bending losses of a waveguide
with high refractive index contrast, become very small, and the Q factor of the resonators
becomes limited only by and material attenuation scattering caused by geometrical imperfec-
tions (e.g. surface roughness).
Fabrication of the open dielectric resonators can be simple and inexpensive, and they
lend themselves to integration. The unique combination of very high Q (as high as 1011)
and very small volume has attracted interest in the applications of the resonators in fun-
damental science and engineering. Small size also results in excellent mechanical stability
and easy control of the resonator parameters. The authors describe applications of the
resonators for filtering and modulating light, for detecting chemical and biological sub-
stances. Various lasers and oscillators based on the resonators are also discussed.
Namely, Lute Maleki et al. (OEwaves Inc., USA) discuss application of crystalline WGM
resonators in filtering and laser stabilization in Chapter 3. Applications of polygonal-
shaped microdisk resonators are studied in Chapter 4, authored by Andrew W. Poon
et al. (The Hong Kong University of Science and Technology, People’s Republic of China).
Applications of electro-optic polymer ring resonators for millimeter-wave modulation
and optical signal processing are reviewed by William H. Steier et al. (University of
Southern California, USA) in Chapter 5. Chapter 6, authored by Melanie Lebental et al.
(Ecole Normale Superieure de Cachan, France), is devoted to the discussion of properties
of organic micro-lasers. Practical applications of optical microfiber loop and coil resona-
tors are described in Chapter 7 by Misha Sumetsky (OFS Laboratories, USA). Chapter 8,
authored by Xudong Fan (University of Missouri, Columbia, USA), deals with optofluidic
ring resonator biological and chemical sensors. An application of crystalline microreso-
nators for fabrication of a non-electronic wireless receiver with immunity to damage by
Andrey B. Matsko
Andrey B. Matsko (MS, 1994 and PhD, 1996, Moscow State University, Russia) has been
a principal engineer with OEwaves Inc. since 2007. He joined the company after six year
employment as a senior/principal member of technical staff at Jet Propulsion Laboratory
(JPL) and four year post-doctoral training at the Department of Physics, Texas A&M
University. He has numerous publications in the field and holds several patents. His cur-
rent research interests include, but are not restricted to, applications of whispering gallery
mode resonators in quantum and nonlinear optics, and photonics; coherence effects in
resonant media; and quantum theory of measurements. He is a member of the Optical
Society of America and a member of the Program Committee of Photonics West: Laser Resonators
and Beam Control Conference. He received JPL’s Lew Allen Award for excellence in 2005.
xv
© 2009 by Taylor & Francis Group, LLC
Gary Betts
Photonic Systems Inc. Ronald Holzwarth
Burlington, Massachusetts Max Planck Institut für Quantenoptik
Garching, Germany
Eugene Bogomolny
Laboratoire de Physique Théorique et Nick K. Hon
Modéles Statistiques Photonic Device Laboratory
Université Paris-Sud Department of Electronic and Computer
Orsay, France Engineering
The Hong Kong University of Science and
Bart Bortnik Technology
Department of Electrical Engineering Clear Water Bay, Hong Kong, People’s
University of California Los Angeles Republic of China
Los Angeles, California
Rick C. J. Hsu
Hui Chen Broadcom Corporation
Photonic Device Laboratory Irvine, California
Department of Electronic and Computer
Engineering
Yu-Chueh Hung
The Hong Kong University of Science and
Department of Electrical Engineering
Technology
University of California Los Angeles
Clear Water Bay, Hong Kong, People’s
Los Angeles, California
Republic of China
xvii
© 2009 by Taylor & Francis Group, LLC
Akihiko Shinya
Optical Science Laboratory Kerry J. Vahala
NTT Basic Research Laboratories California Institute of Technology
NTT Corporation Pasadena, California
Atsugi, Japan
Jelena Vučkovic’
Siyka I. Shopova Edward L. Ginzton Laboratory
Biological Engineering Department Stanford University
University of Missouri Stanford, California
Columbia, Missouri
Contents
1.1 Introduction.............................................................................................................................2
1.2 Small Optical Cavities Fabricated on 2D Photonic Crystal Slabs....................................3
1.2.1 2D and 3D Photonic Crystals....................................................................................3
1.2.2 Ultrasmall Cavity: Photonic Crystal Nanocavity..................................................6
1.3 Designing High-Q Photonic Crystal Nanocavities............................................................7
1.3.1 Design of High-Q Photonic Crystal Nanocavity....................................................7
1.3.2 Waveguide-Coupled High-Q Photonic Crystal Nanocavity................................8
1.3.3 Various Types of High-Q Photonic Crystal Cavities.............................................9
1.3.3.1 Line Defect Cavities with Modulated End-Holes....................................9
1.3.3.2 Point Defect Hexapole Cavity with Rotational Symmetry
Confinement............................................................................................... 12
1.3.3.3 Width-Modulated Line Defect Cavity with Mode-Gap
Confinement............................................................................................... 16
1.3.3.4 Other Photonic Crystal Nanocavities..................................................... 17
1.3.4 Discussion of Structural Error and Q.................................................................... 19
1.3.5 Fabrication of Photonic Crystal Slabs.................................................................... 20
1.4 Characterization of Ultrahigh-Q Photonic Crystal Nanocavities.................................. 20
1.4.1 Spectral Domain Measurement.............................................................................. 20
1.4.1.1 Spectrum Measurement with Frequency
Tunable Laser.............................................................................................. 21
1.4.1.2 Spectrum Measurement using Electro-Optic
Frequency Shifter....................................................................................... 21
1.4.2 Time Domain Measurement................................................................................... 23
1.4.3 Technical Issues Related to Obtaining Accurate Q.............................................. 25
1.5 Applications of High-Q Photonic Crystal Nanocavities................................................. 26
1.5.1 Caging Light and Slow Light.................................................................................. 26
1.5.1.1 Caging Light Using Ultrahigh-Q Photonic
Crystal Nanocavity.................................................................................... 26
1.5.1.2 Slow Light with Photonic Crystal Nanocavity...................................... 26
1.5.2 Compact Optical Add-Drop Filter..........................................................................30
1.5.3 All-Optical Switching.............................................................................................. 31
1
© 2009 by Taylor & Francis Group, LLC
1.1 Introduction
Light is fast and thus can carry large amounts of data in a very short time, which makes
photonic technology a promising communication tool. In fact, photonic technologies, such
as optical fiber, that support a high transmittance speed1 are becoming more important
in our lives. However, these technologies have had limited practical applications in data
transmission, and signal processing has yet to find a commercial use. Although all-optical
signal processing has been widely studied for several decades,2 it has been difficult to
employ in practical systems, often because the required operating energy was too large.3,4
This is due to the fundamental nature of light. In other words, light is fast but difficult to
store or confine in a small space. This makes photonic approaches difficult to handle. In
contrast, there is a growing demand for a practical all-optical signal processor because
today the network system bandwidth is often limited by the speed of the electronics used
at network nodes.
Optical nonlinearities, which can change such material characteristics as refractive
index or absorption, are key phenomena in terms of achieving optical signal processing.5–7
But their coefficients are usually small. As a result, a high input power is required, which
makes the device impractical. However, the input power can be significantly reduced if we
can achieve strong light confinement.
Photonic crystals8,9 are attracting considerable attention because of their strong light
confinement and small structure. It has already been shown that photonic crystals can
confine light in a very small space if we construct ultrahigh-Q (quality factor) photonic
crystal nanocavities.10,11 Since the photon density in these small cavities is extremely high,
various optical nonlinearities occur at a small input energy, which enables the fabrication
of practical all-optical switches12 and logic gates.13 Indeed all-optical switching at an input
energy of less than 100 fJ has been demonstrated using a two-dimensional (2D) silicon
photonic crystal nanocavity.12
In this chapter, we describe recent progress on 2D photonic crystal nanocavities, and
introduce some applications. These applications include add-drop filters for wavelength
division multiplexing, optical buffers, slow light, all-optical switching, bistable memory
operation, and logic gates such as flip-flop and pulse retiming circuits.
c
ω= k (1.1)
n
Г–K
Г–M
Si
SiO2
Figure 1.1
Scanning electron microscope image of a 2D photonic crystal fabricated on a silicon slab. Hexagonal arrays of
air holes were fabricated by using electron beam lithography and dry etching. The diameter of the air hole is
200 nm and the lattice constant is 400 nm. The silicon slab is about 200 nm thick.
0.4
Frequency (λ/a)
0.3
PBG (0.255–0.305)
0.2
0.1
LL(air)
0
Г M K G
Г–K M
K
Г–M Г
1st BZ
Figure 1.2
Energy band diagram of 2D photonic crystals with hexagonal air holes. LL: light line. PBG: photonic band gap.
BZ: Brillouin zone.
where c is the light velocity and n is the refractive index of the cladding. Equation 1.1 is
shown as a line, known as a light line, in a band diagram. Photons that are beyond this line
satisfy the total reflection condition; hence they are vertically confined. The photonic crys-
tal waveguide is theoretically lossless for a frequency component that is below the light
line. According to Equation 1.1, the slope of the light line becomes steeper as the refractive
Transmittance (dB)
PBG
–20
–30
Figure 1.3
Transmittance spectrum of 2D photonic crystal in Γ–M and Γ–K directions.
0.2 Even
(b)
0.15
0.1
Even
0.05
0
0 0.1 0.2 0.3 0.4 0.5
kx
Figure 1.4
(a) Schematic image of a line defect silicon photonic crystal waveguide fabricated on a silicon-on-insulator wafer.
(b) Scanning electron microscope image of a fabricated air-bridge type silicon photonic crystal waveguide.
(c) Band diagram of a line defect photonic crystal waveguide.
index contrast between the slab and cladding increases. Indeed, Figure 1.4c shows light
lines for air cladding (n ~ 1) and SiO2 cladding (n ~ 2.4), where it can be seen that a larger
light component will be confined in the air cladding case. Obviously, a steeper light line
is preferred because it yields greater tolerance as regards the structural design. For this
reason, an air-bridge structure, where the photonic crystal slab is sandwiched by air, is
often preferred to a SiO2 cladding structure.
The lowest reported propagation loss for an air-bridge photonic crystal waveguide is 2
dB/cm,19,20 which is sufficiently small if we take the total size of the device into account.
Since on-chip photonic devices will be much smaller than a centimeter in size, the total
loss is less than 1 dB.
(a)
0 λ
(b) (c) N·
n
x x
0 λ 0 λ
n 2n
Figure 1.5
(a) Schematic illustration of an optical cavity with a length of Nλ/n. (b) Optical cavity with a length of λ/n. (c)
Smallest optical cavity with a length of λ/(2n).
relatively difficult. On the other hand, photonic crystal nanocavities can couple light
through photonic crystal waveguides, which are fabricated on the same chip. As a result, a
large number of cavities can be connected in tandem or in parallel though photonic crystal
waveguides. The surface of the 2D photonic crystal is flat, which makes the mechanical
strength high. As a result, the optical coupling strength between the cavities and the
waveguides is stable, once they are determined at the designing stage.
Small size, ultrahigh-Q and the ability of making integrated photonic circuit makes
photonic crystal nanocavities attractive for various studies.
for the air-cladding slab. Therefore, the strategy for achieving low vertical loss, hence for
achieving an ultrahigh-Q in a 2D photonic crystal nanocavity system, is to find a structure
where the Fourier transform of the optical mode yields very few components inside the
light cone. Examples of the Fourier space distribution of the optical mode of the point
defect photonic crystal are shown in Figure 1.6.
In terms of k-space design, the ideal optical mode profile is a sinc function, because it
has a square shaped function in the k-space.29 However it is not possible to find a structure
that can exhibit a perfect sinc shaped optical mode, because sinc function requires
infinite endpoints to define. Akahane et al. used a more convenient strategy to obtain
a high-Q mode, namely they used a Gaussian function as a figure of merit.30 Since the
0.1
0.01
E-3
E-4
Figure 1.6
The k-space distributions of three different modes of a point defect photonic crystal nanocavity. The Q values
calculated by 3D FDTD are shown in the panel.
Fourier function of a Gaussian curve is also Gaussian and its overlap with the light cone
is relatively small, a Gaussian shape is a good practical shape for obtaining a high-Q. So
far this strategy appears to work very well; in fact, a theoretical Q of about 2 × 107 has been
obtained based on this strategy.29
Qunloaded = Qv (1.2)
where Qv is the Q determined by the out-of-slab radiation loss. On the other hand, the Q
of the cavity coupled to the waveguides is called the loaded Q, and its value depends on
the out-of-slab radiation loss and the coupling between the waveguides. The loaded Q can
be written as
loaded = Qv + Q h
Q - 1 - 1 - 1 (1.3)
where Qh is the Q value determined by the coupling strength with the waveguides. Figure 1.7
shows this schematically. Note that the unloaded Q is identical to Qv for an isolated cavity
because Qh- 1 is null. By employing this equation and simple coupled mode theory,31,32 we
can derive the relationship between Q and the transmittance. The transmittance T of a
cavity/waveguide system is given as,33
2 2
Q Q -1
T = loaded = -1 h -1 (1.4)
Qh Qv + Qh
Qv : vertical radiation
Qh: coupling with WG
PhC WG
PhC WG
PhC cavity
Figure 1.7
Schematic illustration of photonic crystal nanocavity coupled to photonic crystal waveguides.
For a given Qv, Equations 1.3 and 1.4 tell us that the loaded Q and the transmittance are
in a tradeoff relationship. However, if we can design a high Qv, it should be possible to
achieve a high transmittance and a high loaded Q simultaneously. Therefore it is extremely
important to find designs that can exhibit a high Qv.
Figure 1.8
Schematic illustration of a line defect photonic crystal nanocavity with shifted end-holes.
10–2 104
10–3
Light cone
Qv ~ 2.7×103 Qv ~ 2.9×104 103
10–4 0.1 0.2 0.3
End hole shift (× a)
Figure 1.9
(a) Fourier space distribution (k-space) of the spatial mode of the L2 cavity with different end-hole shifts. The
light cone is indicated by the dotted circle. a is the lattice constant. (b) Isolated Q (Qv) of the L2 cavity with
different end-hole shifts. Q reaches a maximum value of 2.9×104 at an end-hole shift of ~0.2a. The slab thickness
is 0.5a, and the hole diameter is 0.55a.
angle of 60 degrees as shown in Figure 1.11, which makes it easier for the cavity field
to overlap the waveguide ends when the waveguides are placed in a shoulder coupled
direction. In addition, the optical field of the waveguide termination exhibits gradual
decay in the 60-degree direction.38 As a result the coupling between the cavity and the
waveguides always becomes smaller as the cavity/waveguide distance is increased for
shoulder coupling. This is more complicated with in-line coupling.
Figure 1.12 shows the calculated and measured spectra of the L3 cavity shown in
Figure 1.10a and b. The resonance at 1550.36 nm yields a theoretical Qv of ~ 3 × 104 com-
pared with 1547.68 nm and Q = 1.84 × 104 for the experiment. Since the transmittance for
the fabricated sample was very small (T = ~ 2.1%) the measured Q should be almost identical
to the Qv of the cavity. The mode volume calculated using 3D FDTD is 7.2 × 10 -2 µm3.
Figure 1.13 is the measured spectrum of the L4 cavity shown in Figure 1.10b. It has two
resonant modes, which we call mode C and mode S. The Q values are Qc = 1.15 × 104 for
mode C and Q S = 2.3 × 104 for mode S. The corresponding photon lifetimes are τc = 9.3 ps
and τS = 19.1 ps.
(a) (b)
Г–M
Г–K
c-hole r-hole
t-hole
Figure 1.10
Scanning electron microscope image of fabricated silicon photonic crystal nanocavities. The lattice constant for
both samples is a = 420 nm with a hole diameter of d = 0.55a. (a) An L3 nanocavity is fabricated with c-, t-, and
r-hole diameters of 0.25a, 0.57a, and 0.3a, respectively. The r-holes are shifted 40 nm in the t-hole direction. (b)
Four point defect nanocavity with c-, t-, and r-hole diameters of 0.45a, 0.57a, and 0.3a, respectively. The r-hole
shift is 60 nm in the t-hole direction.
Output
waveguide
Mode profile
Input
waveguide
Figure 1.11
A silicon photonic crystal L4 cavity coupled to an input/output waveguide in the Γ–M direction. The inset is the
mode profile shown on a log scale.
Historically, a line-defect cavity with shifted end holes has significantly increased the Q
of 2D photonic crystal nanocavity systems and given them various possible applications.30
Therefore, although there are now a number of other types of cavities that can exhibit a
much higher Q, this type of cavity has been widely employed for various applications
(a) (b)
1×10–3 101
1×10–4
Transmittance (a.u.)
Transmittance (%)
1×10–5 100
1×10–6
1×10–7
10–1
1×10–8
1×10–9
1×10–10 10–2
1350 1400 1450 1500 1550 1600 1546.5 1547.0 1547.5 1548.0 1548.5 1549.0
Wavelength (nm) Wavelength (nm)
Figure 1.12
(a) Calculated transmission spectrum of an L3 cavity, which has the design parameters of the cavity shown
in Figure 1.10. (b) The measured transmittance spectrum of the fabricated silicon photonic crystal L3 cavity
shown in Figure 1.10. (Reprinted with permission from Tanabe, T. et al., Appl. Phys. Lett., 87(15), 151112, 2005. ©
American Institute of Physics.)
101
Mode S
Mode C
Transmittance (%)
100
10–1
10–2
1530 1531 1532 1567 1568 1569
Wavelength (nm)
Figure 1.13
Transmittance spectrum of the silicon photonic crystal L4 cavity shown in Figure 1.10b. The Q values of mode
C and mode S are Q C = 1.15 × 104 and Q S = 2.3 × 104, respectively. (Reprinted with permission from Tanabe, T. et al.,
Appl. Phys. Lett., 87(15), 151112, 2005. © American Institute of Physics.)
Cavity
ck
Figure 1.14
Schematic image of a point defect photonic crystal nanocavity. The innermost holes are shifted slightly towards
the outside to optimize the Q of the hexapole mode.
Hz Energy of EM field
Dipole
Hexapole
Quadrupole
Figure 1.15
Magnetic and electro-magnetic fields of the dipole, hexapole and quadrupole modes seen in a point defect
photonic crystal nanocavity.
The explanation for the ultrahigh Q obtained in the hexapole mode is different from that
for line-defect type photonic crystal nanocavities. Figure 1.17 shows the mode distribution
and phase property of the magnetic field. The beauty of its spatial rotational symmetry
originates in the matching between the symmetry of the optical mode and the pattern of
the holes in the hexagonal photonic crystal lattice. In addition, because the phase of the
2.0×106
1.5×106
Unloaded Q
1.0×106
5.0×105
1.22 1.23 1.24 1.25 1.26 1.27
Shift ck (× a)
Figure 1.16
Calculated theoretical Q of the hexapole mode for different hole shifts ck, where ck is as shown in Figure 1.14.
(Reprinted with permission from Tanabe, T. et al., Appl. Phys. Lett., 91(2), 021110, 2007. © American Institute of
Physics.)
– +
+ –
– +
Figure 1.17
Profile of hexapole mode. The + and - represent the phase of the Hz magnetic field (z: perpendicular to
the slab).
nearby Hz component is reversed (shifted by π), as indicated in Figure 1.17, the optical far
field is cancelled out. As a result, the out-of-slab radiation is reduced by the destructive
interference effect of the far-field pattern, which was initially discussed by Johnson et
al.45 Hence, the hexapole mode can exhibit an ultrahigh Q. We call this light confinement
mechanism “rotational symmetric confinement”.
The rotational symmetries yield different interesting characteristics of the hexapole
mode when it is coupled with waveguides. Figure 1.18a and b, respectively, show the
in-line coupling and side coupling of the hexapole mode with the waveguides. Owing
(a) (b)
Figure 1.18
(a) Butt-coupled structure of separation=7 with Hz distribution. (b) Side-coupled structure of separation=7 with
Hz distribution. (From Kim, G.-H. et al., Opt. Express, 12, 6624–6631, 2004. © Optical Society of America. With
permission.)
(a) (b)
1.24
V (λ/n)3
1.22
1.20
1.18
QH
107
QV
106 QT
Q
105
104
5 7 9 11 13 15
Min/20 0 Max/20 Separation
Figure 1.19
(a) Shoulder-coupled structure of separation=7 with Hz distribution. (b) Modal volume (V) and quality factors
(Qs) of hexapole mode. (From Kim, G.-H. et al., Opt. Express, 12, 6624–6631, 2004. © Optical Society of America.
With permission.)
to the symmetries of the optical mode, the light hardly couples with the waveguides.
This characteristic enables the cavity to be extremely well isolated from closely posi-
tioned waveguides or cavities, which offers the possibility of the dense packing of cavity-
waveguide systems. To couple the hexapole mode with the waveguides, the waveguides
are placed in a shoulder-coupled configuration as shown in Figure 1.19. This configuration
enables efficient coupling of the optical mode with the waveguides. When the separation
is 7, Qloaded is almost identical to Qh. According to Equation 1.4, the transmittance is nearly
100%. For a separation of 9, QT is 1.9 × 105 and the transmittance is 88%. In fact, the transmit-
tance of the fabricated sample was almost 100% when the separation was 9.46
(a) (b)
Output WG
Transmittance (a.u.)
Hex cavity
4.8 pm
Input WG Г–K
Г–M
1547.48 1547.52 1547.56
Wavelength (nm)
Figure 1.20
(a) Scanning electron microscope image of a hexapole cavity with photonic crystal waveguide ends fabricated
on a silicon photonic crystal. WG: waveguide, Hex: hexapole. (b) Transmittance spectrum of the hexapole mode.
The solid line is the fitted Lorenz curve. (Reprinted with permission from Tanabe, T. et al., Appl. Phys. Lett., 91(2),
021110, 2007. © American Institute of Physics.)
Figure 1.20 shows a scanning electron microscope image and the transmittance spectrum
of the fabricated sample. The obtained spectrum exhibits an extremely high Q of 3.2 × 105,
which is the highest experimental Q yet reported for photonic crystal nanocavities, except
for those with mode-gap confinement.44 This cavity is a good candidate for achieving an
ultrahigh Q in photonic crystals.
A1 A2
B B
Line defect
B B
C C C A A C
C C B B
C B B C C A A C
Line defect
B B B B
C A A C B B
B B
C A A C
B B
C B B C
C C
C C W = W0.9
A3
Z
Line defect
A A
X W = W0.98
Hole shift A A
dA
dB
dC
W = W0.9
Figure 1.21
Different designs of width-modulated line defect cavity (A1, A2, A3). The width of the baseline defect is
0.98 × a 3 (W0.98) and 0.90 × a 3 (W0.90). Holes marked with A, B, and C are shifted slightly towards the
outside of dA, dB, and dC, respectively. Typical values for the A1 cavity are dA = 9 nm, dB = 6 nm, dC = 3 nm, a = 420
nm, r = 108 nm, t = 204 nm, where r is the air hole radius and t is the slab thickness. (Reprinted with permission
from Kuramochi, E. et al., Appl. Phys. Lett., 88(4), 041112, 2006. © American Institute of Physics.)
Our first generation mode-gap confined silicon photonic crystal nanocavity is shown in
Figure 1.22a. The obtained Q was not very high (Figure 1.22b and c), because the nanocavity
was fabricated on SiO2 cladding and the parameters were not optimized.35
Figure 1.21 shows the latest design, which exhibits a theoretical Q of 1.2 × 108 and a mode
volume of 1.51(λ/n)3 for the cavity A1.47 The calculated spatial mode profile is shown in
Figure 1.23a. A scanning electron microscope image is shown in Figure 1.23b, where the
circled region is the cavity. Although it is hard to distinguish, the holes at the cavity are
slightly shifted towards the outside. The cavity exhibits an extremely high Q of 1.2 × 106,
which is one of the highest Q values achieved by photonic crystal nanocavities.10 This Q
value corresponds to a photon lifetime of about 1 ns, which opens the possibility for various
applications such as optical buffering and cavity quantum-electro dynamics. Note that
there is an approximately two orders discrepancy between the theoretical and experimental
values. This is caused by the material absorption and various types of scattering, such as
absorption at the surface or scattering due to fabrication error. Of these we speculate that the
variation in the hole diameter plays an important role in determining the experimental Q.
(a) (b) 1
Nb = 2
0.6 W 0.8
Transmittance
0.6
0.4 W 0.4
0.2
0.6 W
0
1480 1490
(c) 0.15
Nb = 3
Transmittance
0.1
0.05
0
1480 1490
Wavelength (nm)
Figure 1.22
Resonant-tunneling filter using the mode-gap in the width-varied waveguides. The samples are fabricated on
silicon-on-insulator photonic crystal slabs, where the undercladding is SiO2. (a) Structural design. (b) Measured
transmittance spectrum around the resonant wavelength for a barrier width of N b = 2. Q = 408 and T = 86%.
(c) Measured transmittance spectrum for N b = 3, Q = 1350 and T = 12%. (From Notomi, M. et al., Opt. Express, 12,
1551–1561, 2004. © Optical Society of America. With permission.)
1.3 pm
Figure 1.23
(a) Mode profile of width-modulated line defect cavity with mode-gap confinement shown in Figure 1.21 A1. (b)
Scanning electron microscope image of the fabricated sample on silicon photonic crystal. The circle indicates
the cavity region. (c) Transmittance spectrum of the cavity shown in (b).
107
106
Qv
105
Waveguide-width modulated cavity
Hexapole mode cavity
L4 line defect cavity
104
0 1 2 3 4 5
σ (nm)
Figure 1.24
Calculated Q values for three different types of cavities (waveguide-width modulated cavity, hexapole mode
cavity, and L4 line-defect cavity), with respect to radius randomness.
(a) EB resist
(b)
(c) Air (n = 1)
SOI (Si, n = 3.46)
SiO2 (n = 1.48) Air (n = 1)
Si substrate
Figure 1.25
Schematic diagram of the air-bridged silicon photonic crystal slab. (a) Photonic crystals are patterned on resist
by electron beam lithography. (b) Photonic crystal is formed on silicon by dry etching. (c) The underlying SiO2
layer was removed by wet-etching.
those obtained experimentally. Hence, a standard deviation of 2–3 nm provides a good fit
with the experimental results. Although such a small radius variation is very difficult to
estimate with a scanning electron microscopic measurement, we believe that the value is
not far from the actual fabrication accuracy. Therefore we speculate that the limitation of
the experimental Q results from the randomness in the air-holes radius. Further detailed
discussions can be found elsewhere.20
Ecav
Q=ω× (1.5)
Pout
where ω is the resonant angular frequency of the cavity, Ecav is the energy stored in the
c avity, and Pout is the output power (i.e. optical energy loss per unit time) from the cavity.
We can derive a different expression from Equation 1.1, where Q is described in the spectral
domain.
ω λ
Q= ≈ (1.6)
∆ω ∆λ
Here, ∆ω and ∆λ are the linewidth of the resonant spectrum in terms of frequency and
wavelength, respectively. λ is the resonant wavelength of the cavity. Obviously, Equation
1.6 suggests that the cavity Q can be determined by measuring the transmittance linewidth
of the cavity when the cavity is coupled with the waveguides. Indeed the Q values of
photonic crystal nanocavities have been characterized in the spectral domain. However, it
should be noted that spectral domain measurement is now not an easy task because the Q
of photonic crystal nanocavities has increased rapidly, and therefore, ultrahigh wavelength
resolution is needed to measure the ultra-narrow transmittance linewidth of a nanocavity
accurately. For a telecom wavelength, the linewidth for a cavity with a Q of 1.2 × 106 is about
1.3 pm as shown in Figure 1.22. Therefore a sub-pm wavelength resolution is required.
Absorption (linear)
70 pm
10–1 > 20 dB
10–2
Power (a.u.)
Figure 1.26
Optical output spectrum from single-sideband modulator at a radio frequency of 10 GHz. Inset is the spect
roscope image of the absorption of acetylene (13C2H2) gas at room temperature obtained with the single-sideband
modulator setup to demonstrate the accurate spectrum measurement. (Reproduced by kind permission of the
IET from Tanabe, T. et al., Electron. Lett., 43, 187–188, 2007.)
Frequency (GHz)
2 1 0 –1 –2
0
10
10 0 Q = 1.23 × 106
10 –1
τ = 1.02 ns
Transmittance (dB)
10 –2
1564.75 1564.76 1564.77 1564.78 1564.79
Wavelength (nm)
10 –1
10 –2
Figure 1.27
Measured spectrum obtained using first-order single-sideband light scan. Inset is the measured spectrum when
a –1st-order single sideband is used for the scan. The dots are the measured plot and the solid line represents
the fitted Lorenz function. (Reproduced by kind permission of the IET from Tanabe, T. et al., Electron. Lett., 43,
187–188, 2007.)
Q = ωτ (1.7)
where τ is the photon lifetime of the cavity. It is not very easy to measure τ when Q is
small; but the τ measurement becomes more accurate as Q increases. Indeed a Q of 1.2 × 106
corresponds to a photon lifetime of ~ 1 ns at telecom wavelengths. The ring-down method
is a direct way to characterize the τ and Q values of a cavity. First, the cavity is charged
with an input CW light, and then the input is suddenly turned off. τ can be obtained by
observing the decaying optical signal at the output waveguide. A schematic diagram of the
measurement is shown in Figure 1.28. Note that the signal light must be kept sufficiently
low to prevent the cavity exhibiting nonlinearity such as two-photon absorption or free-
carrier absorption, which may modify the cavity Q. Therefore we employ time-correlated
single photon counting for the ultrahigh-Q measurement, as this allows us to measure an
extremely weak signal light with a time resolution of ~ 70 ps.10
The measured signal is shown in Figure 1.29. The decay is a smooth exponential curve,
and the fitted decay is 1.01 ns, which is the photon lifetime of the cavity. This value cor-
responds to a Q of 1.2 × 106, which is in good agreement with the value obtained from a
spectral-domain measurement. Time domain measurement provides a direct view of the
photon trapping by the photonic crystal nanocavity.
Another aspect of the time domain measurement can be clarified by performing a
ring-down measurement on a side-coupled cavity.52 Figure 1.30a and b show a schematic
diagram and the transmittance spectrum of a side-coupled cavity. The transmittance
Input waveform
Monitored region
Power
∞ exp(−t/τ)
Time
Output waveform
Figure 1.28
Schematic diagram of the ring-down measurement. The dotted line represents a rectangular input pulse and
the solid line represents the output from a cavity. (From Tanabe, T. et al., Opt. Express, 15, 7826–7839, 2007. ©
Optical Society of America. With permission.)
τ = 1.01 ns
Q = 1.2×106
0.01
0 1000 2000 3000
Time (ps)
Figure 1.29
Output of a ring-down waveform from the waveguide-width modulated silicon photonic crystal nanocavity
shown in Figure 1.22b.
10–1
Power (a.u.)
τ = 210 ps
Cavity 10–2
10–3
Path B
Figure 1.30
(a) Schematic diagram of a side-coupled cavity with a photonic crystal waveguide. (b) Transmission spectrum
of the side-coupled cavity. Sometimes it is difficult to obtain the Q from the spectral bandwidth because of the
background Fabry–Pérot oscillation. (c) The output waveform of a side-coupled cavity when a rectangular pulse
is applied.
spectrum exhibits a dip at the resonance as shown in Figure 1.30b. Side-coupled cavity
configurations are useful for constructing multi-channel add-drop filters or photonic
DRAMs.20,53 The transmitted waveform of a 6-ns square input pulse is shown in Figure
1.30c, along with the output pulse. The wavelength of the input light is adjusted to the
cavity resonance; therefore, the light cannot propagate though the device in a steady state.
However, the pulse transmits at the rising and falling edges of the square pulse. This
can be explained as follows. The light propagates until the cavity is charged and forms
interference. Interference occurs between the light that travels straight from the input
waveguide toward the output waveguide (path A) and the output light from the side-
coupled nanocavity (path B). When the light is turned off, path A is immediately cut off.
As a result the light cannot interfere and the path B light is observed as a ring-down
waveform. The discharging signal is observed after the input has been turned off, which
constitutes an intuitive demonstration of light trapping in a photonic crystal nanocavity.
(a)
106
Qtime
1.5
1.0
105 106
Qspec
Figure 1.31
Qspec is the Q value obtained from the transmittance spectrum bandwidth and Qtime is the Q value obtained from
the decay of the ring-down waveform. (a) Square dots show the Qtime measured using time correlated single
photon counting and round dots show the Qtime measured using a digital sampling oscilloscope. The dotted line
indicates the ideal case. (b) Qtime/Qspec with respect to Qspec. (From Tanabe, T. et al., Opt. Express., 15, 7826–7839,
2007. © Optical Society of America. With permission.)
agreement, which confirms the reliability of the measurements. Figure 1.31 also suggests
that a detector speed of ~ 70 ps is sufficient to measure a cavity that has a Q of >105, but a
faster response is required to obtain an accurate Q for those cavities with a value of <105. A
detailed discussion of accurate measurement in the spectral and time domains is provided
elsewhere.52
Pump
W/o pump pulse
1
Power (a.u.)
W/ pump pulse
0.1
Figure 1.32
Example of the dynamic tuning of the photon lifetime of a silicon photonic crystal nanocavity. (Reprinted by
permission from Macmillan Publishers Ltd from Tanabe, T. et al., Nature Photon., 1, 49–52, 2007.)
(a) 1 (b) φ
2 0.8 π/2
T(ω)
τ = photon lifetime Tilt = 2τ
0.6
arg(T(ω)) = Pulse delay
0.4
–2/τ 0 2/τ ω–ω0
0.2
–2 –1 1 2
ω–ω0
Figure 1.33
(a) Schematic illustration of the transmittance intensity of a Fabry–Pérot cavity. The photon lifetime is determined
by the transmittance bandwidth. (b) Schematic illustration of the phase property of a Fabry–Pérot cavity. The
pulse delay is given by the tilt of the dispersion curve.
Next, we discuss the relationship between the group velocities of a single cavity and a
CROW. A schematic explanation is shown in Figure 1.34. Recall that the maximum delay
is ~ 2τ. When we determine the device length L of a waveguide coupled single photonic
crystal nanocavity as shown in Figure 1.34b, the smallest group velocity is given as vg ≈ L/2τ.
Now, let us consider a case where the cavities are tandem connected. If the horizontal
coupling strength is the same as that for a single cavity; roughly speaking, the pulse delay
is multiplied by the number of cavities. In other words, if a CROW consists of N cavities, the
maximum delay will be N times larger than that of a single cavity as shown in Figure 1.34c.
However, the resulting velocity is the same as that for a single cavity, because the device
length is N times greater (Figure 1.34d). This means that the group velocity demonstrated
in the single cavity should correspond to the minimum value that can be obtained in an
ideal CROW system.
Figure 1.35 shows the experimental pulse transmittance results for three different input
pulses using a single nanocavity with a Q of 1.2 × 106 and an L of 7.6 µm. The delay of a
pulse at the nanocavity is measured by comparing the output from the photonic crys-
tal nanocavity with that of a reference waveguide. A Gaussian-like input optical pulse is
generated by using an electrical pulse shaper and a 40-GHz electro-optic modulator. A
maximum delay of 1.4 ns was obtained. The obtained delay corresponds to a group velocity
of 5.4 km/s, and this value is the smallest for any dielectric based slow-light material.
Figure 1.35 shows that different input pulse widths yield different delays. Additionally,
the pulse broadens when it propagates through a high-Q cavity, and this is due to the nar-
rowing of the spectral bandwidth. Note that bandwidth narrowing does not explain the
pulse delay.
To investigate the pulse delaying phenomenon in detail, a numerical model was studied
using simple coupled mode equations.32 The input and the output of the cavity are given as,
ω–ω0 –2/τ 0
Number of cavities: N
2/τ vg(min) ~ NL / 2Nτph
Figure 1.34
(a) Schematic illustration of the dispersion curve of a single cavity. (b) Schematic illustration of a single
nanocavity coupled to waveguides. (c) Schematic illustration of the dispersion curve of a CROW consisting of N
cavities. (d) Schematic illustration of a CROW consisting of N cavities.
Ref.
Power (a.u.)
WG PhC
cavity
–4 –2 0 2 4 6 8 –4 –2 0 2 4 6 8 –4 –2 0 2 4 6 8
Time (ns) Time (ns) Time (ns)
Figure 1.35
Output pulses from the reference waveguide (Ref. WG) and silicon photonic crystal nanocavity for different
input pulse widths. ∆τin is the input pulse width and the τd is the pulse delay.
da
= jω 0 - ( 2 τ ) a + ( 2 τ ) pin ( t )
-1 -1/2
(1.8)
dt
pout ( t ) = ( 2 τ ) a (t )
-1/2
(1.9)
where,
2
Pin ( t ) = pin ( t ) (1.10)
2
Pout ( t ) = pout ( t ) (1.11)
Pin, Pout, and ω0 are the input pulse power, output pulse power, and center angular frequency
of the cavity resonance, respectively. We assumed a Gaussian shaped input pulse and
normalized the time with the cavity photon lifetime τ. The calculated Pin and Pout values
for different input pulse widths are shown in Figure 1.36. Different input pulse widths
yield different delays. These calculations revealed that, for a practical input pulse width,
the pulse exhibits a delay of about 1.6 times the photon lifetime. This is consistent with the
experimental results.
The calculation results are summarized in Figure 1.37. As the width of the input pulse
increases, the pulse delay gradually increases and then levels off at 2τ, which is consistent
with the discussion in Figure 1.33b. However, the calculation also suggests that the delay
is about 1.6τ for a practical pulse width, which agrees with the experiment. In Figure 1.33,
the ratio of the pulse delay versus the output pulse width is given as the pulse shift, which
reaches a maximum value of 0.43 when the input pulse width is ~ τ. Unfortunately, this
result shows that single cavity is not very practical as an optical buffer. However, the delay
can be significantly increased when a number of cavities are connected in tandem to form
a CROW as shown in Figure 1.34. And regarding the group velocity, that demonstrated in
a single cavity corresponds to the minimum value that can be obtained in an ideal CROW
system. Indeed, photonic crystal based CROWs are currently being studied to obtain a
large group delay with an extremely small footprint.59
0 0 0
Power (a.u.)
–2 –1 0 1 2 3 4 5 6 –2 –1 0 1 2 3 4 5 6 –2 –1 0 1 2 3 4 5 6
0 0 0
–2 –1 0 1 2 3 4 5 6 –2 –1 0 1 2 3 4 5 6 –2 –1 0 1 2 3 4 5 6
Time (τph)
Figure 1.36
Calculated input (dotted line) and output (solid line) pulses from a photonic crystal nanocavity at different
input widths. ∆τin is the input pulse width and the τd is the pulse delay.
2.0
Output pulse width ∆τout (τ)
1.5
0.5
Pulse shift Θ
0.0
0 1 2 3 4 5 6
Input pulse width (τ)
Figure 1.37
Three curves are shown with respect to the FWHM of the input pulse ∆τin. The FWHM of the output pulse ∆τout
(normalized with τ), the delay τd between the input and output pulse peaks (normalized with τ), and the pulse
shift (Θ = τd/∆τout). The square dots are the experimental pulse delay results.
(a) (b)
–10
–20
Figure 1.38
(a) Scanning electron microscope image and schematic diagram of a compact wavelength division multiplexer
composed of L3 silicon photonic crystal nanocavities connected in parallel along a photonic crystal waveguide.
The arrow indicates the flow of the input and output light, and the circles indicate the position and alignment
of the cavities. The resonant wavelength of every cavity is shifted slightly in order to achieve wavelength mul-
tiplexing. (b) The output spectrum of the device.
with a photonic crystal nanocavity can transmit towards the drop port. The fabricated
sample is shown in Figure 1.38a, where the system is composed of five L3 cavities with
different resonant wavelengths. The device is only ~ 18 µm in length. In order to obtain
the desired resonant wavelength, the shift at each cavity end is tuned. In addition,
the width of the photonic crystal waveguide is fine tuned to control the transmittance
bandwidth of the through port. The system design is discussed in detail elsewhere.60 The
transmittance spectrum of the device is shown in Figure 1.38b, where five wavelength
components are separated in the through port spectrum. Since the device is ultra-small,
this demonstration shows that photonic crystal based filters can be integrated in a
photonic circuit on-chip.
(a) (b)
Cavity
resonance
S2 S1 S2 S1
Figure 1.39
Schematic illustration of the mode-S spectrum of the cavity resonance and the signal light of a switching
experiment. S1 and S2 represent the signal light with different detuning. The wavelength of S1 is detuned
slightly from the cavity resonance (to a longer wavelength) and S2 is close to the initial cavity resonance. (a)
When the control pulse is off. The transmittance of the S1 light is low (off) and that of S2 is high (on). (b) When
the control pulse is injected and thermo-optic effect is present. The transmittance of S1 is high (on) and that of
S2 is low (off).
10
S1
NOT
Power (a.u.)
1
S2
AND
0.1
0 1 2 3 4 5
Time slot
Figure 1.40
Switching demonstration based on thermo-optic effect using silicon photonic crystal nanocavity. Each time slot
corresponds to approximately 1 ms.
Pump detuning: 0 nm
Probe detuning: 0.38 nm
Probe output (log scale)
–1
–2
τ = 100 ns
–3
–200 –100 0 100 200 300 400 500 600
Time (ns)
Figure 1.41
Temporal response of the probe output. The 64-µW pump light was suddenly turned off at 0 ns. The transmit-
tance of the CW probe light exhibits a 100-ns decay.
Another important characteristic that must be considered for a switching device is the
operating speed. The operating speed of thermo-optic effect based devices is considered to
be relatively slow because it is determined by the effective heat relaxation time. However,
this is not always the case with ultrasmall devices. To obtain an effective heat relaxation
time for silicon photonic crystal nanocavity based switches, we performed a time-resolved
measurement. We injected a 400-ns rectangular pulse with the mode-S resonance and
measured the transmittance of a CW probe signal at a wavelength of 0.38-nm detuned
from the mode-C resonance. The measured results are shown in Figure 1.41 where the
probe light exhibits an exponential like decay. The exponential fitted decay is 100 ns,
which we determined as the effective thermal relaxation time. This value is surprisingly
small for a thermo-optic based device. The fast switch off is achieved because of the small
size. Although the thermal diffusion constant is considered small in silicon, the value is
sufficiently large for a nanocavity device, which is about 0.1 µm3 in size. As a result, the
heat diffuses efficiently outside the cavity, which makes the effective thermal relaxation
time very fast.
10–1
50 ps
δ = –0.45 nm
10–2
Figure 1.42
Transmittance of the signal light for different detuning values δ. A ~100-fJ control pulse is injected into the
silicon photonic crystal nanocavity at t = 0 ps.
Figure 1.43
Switching contrast as a function of different input control pulse energies. Uin is the energy of the control pulse
at the input photonic crystal waveguide and Ueff is the effective energy that couples with the cavity estimated
from the spectrum shape and the transmittance. (a) is when δ = -0.3 nm and (b) is when δ = 0 nm.
only ~ 4%. Therefore, by optimizing the shape of the input pulse spectrum and the trans-
mittance of the nanocavity with the waveguides, we expect to reduce the energy to a few 10
fJ. This shows that a control pulse of a few fJ is sufficient to operate this switching device.
Indeed, the absorbed energy calculated from the cavity resonance shift is estimated to be
only 2 fJ. This demonstration opens the possibility for the development of integrated all-
optical switches and logic gates on a silicon chip.20,63
∂ne ( r , t ) 1 1
= ∇ ⋅ Jn (r , t ) - ne ( r , t ) - ne _ eq ( r , t ) (1.12)
∂t q τ slab
∂nh ( r , t ) 1 1
∂t
= ∇ ⋅ J p (r , t ) -
q τ slab
[ nh (r , t ) - nh _ eq (r , t )] (1.14)
The subscript e represents electrons and h represents holes. n is the carrier density and
ne_eq and nh_eq are the carrier densities under thermal equilibrium. µ and D are the carrier
mobility and diffusion. τslab, q, φ, and φex are the effective carrier lifetime for a 200-nm
thick slab, electric charge, internal potential and external potential, respectively. These
four equations describe the dynamics of the electrons and holes in the silicon substrate,
where the diffusion and electron-hole recombination are taken into account. The interface
between silicon and air is represented by the following equation.
where S is the surface recombination speed. Since the electrons and holes create internal
potential, a Poisson equation is considered.
1
∇2φ (r , t ) = [ nh ( r , t ) - ne (r , t )] (1.17)
ε 0ε r
where n'(t) are carrier densities normalized with the field profile of the cavity.
Before calculating the 2D photonic crystal nanocavity switches, we calculated the
cross-sectional carrier dynamics of a silicon waveguide to determine the surface velocity
S by comparing the obtained result with those obtained in the experiments. The cross-
sectional optical mode profile of a 400 × 200-nm silicon waveguide was obtained with the
beam propagation method and was employed to determine the initial electron and hole
density distributions. The physical parameters of silicon were drawn from references, as
De = 35 cm 2/s, Dh = 12 cm 2/s, µe = 1450 cm 2/Vs and µh = 500 cm2/Vs.65 And the bulk lifetime
of silicon was set at 0.5 µs. Using these values, we calculated the effective carrier lifetimes
for various S values, and the results are summarized in Figure 1.44. According to our
experiment, the effective carrier lifetime of a silicon waveguide based switch is about
105
104
Switching recovery (ps)
103
102
101
100
101 102 103 104 105 106 107 108 109
Surface recombination velocity (cm/s)
Figure 1.44
Calculated switching recovery time of a 200 × 400-nm silicon waveguide for different surface recombination
velocities.
1.07 ns.46 This value was obtained by injecting a CW signal light and short control pulses
from the same input and monitoring the signal transmittance. Since reflection at the input
and the output waveguide facets creates a Fabry–Pérot oscillator, the transmittance of the
signal light is modified when the carrier-plasma dispersion effect is present. Different
studies also show that the effective carrier lifetime of a silicon waveguide system is about
1 ns.61,66 According to Figure 1.44, this corresponds to a surface velocity of S = 8 × 103 cm/s.
This value is similar to previously reported silicon and air surface values.67,68 Hence, we
can now use this value to calculate the carrier dynamics in photonic crystal nanocavity
switches. The initial distributions of electrons and holes are given by the power of
the electrical field profile calculated by using the FDTD calculations. The calculated
initial electron density distribution for an L4 cavity is shown in Figure 1.45a, when the initial
electron density was 1 × 1017 cm-3, which results in a wavelength shift of about - 0.15
nm. The carrier densities after 8 and 24 ps are shown in Figure 1.45b and c, where rapid
diffusion is present. Figure 1.46 shows the wavelength shift, which corresponds to the
1 1 1
y (µm)
y (µm)
0 0 y (µm) 0
–1 –1 –1
0 ps 8 ps 24 ps
–2 –2 –2
–2 –1 0 1 2 –2 –1 0 1 2 –2 –1 0 1 2
x (µm) x (µm) x (µm)
Figure 1.45
(a) Initial electron distribution of an L4 silicon photonic crystal nanocavity. (b) Electron density after 8 ps.
(c) After 24 ps.
10–1
Wavelength shift (nm)
S = 0 cm/s
S = 8×103 cm/s
10–2
0 100 200 300 400
Time (ps)
Figure 1.46
Calculated wavelength shift for different surface recombination velocities S.
switching recovery time, for two different S values of 8 × 103 and 0 cm/s. The difference
between the two lines is not very significant, and this is particularly so at the initial stage
(t < 20 ps). It suggests that surface recombination does not play a significant role in deter-
mining the effective lifetime in photonic crystal nanocavity switches, but that diffusion
is an important factor. It should be noted that the exponential fitted decay was 0.26 ns,
which is a much smaller than that of a silicon waveguide. The short lifetime is the result
of the unique structure of the photonic crystal nanocavity switches, where the carriers
can diffuse efficiently outside the cavity region. Although the diffusion coefficient of
silicon is considered to be relatively small (i.e. slow diffusion), its value is noticeably
large when it is compared with the ultrasmall volume of the nanocavity switches. In
other words, it is confirmed by calculation that the very small effective carrier lifetime
demonstrated using a photonic crystal nanocavity results from the small mode volume
of the device.
(a) (c)
1 λctrl = 1536.7 nm λsig = 1570.2 nm
Transmittance (a.u.)
S2 out BPF
S1 out EDFA
t 28G Oscilloscope
Figure 1.47
(a) Transmittance spectrum of the silicon photonic crystal nanocavity switch. (b) Schematic illustration of
the pulse train modulation experiment. (c) Schematic diagram of the experimental setup. PPG: Pulse pattern
generator, LD: Laser diode, WDM: Wavelength division multiplexer, PCS: Photonic crystal nanocavity switch,
EDFA: Erbium doped fiber amplifier, BPF: Band pass filter, LiNbO3: LiNbO3 optical modulator.
Signal clock
Control pulse
+0.06 nm
–0.2 nm
Figure 1.48
Signal clock (first row) is modulated with a random 5-GHz return-to-zero control pulse train (second row). The
transmittances of the signal pulse trains (third and fourth rows) for two different wavelengths are shown in
the panel.
return-to-zero optical pulse train is prepared and injected into the photonic crystal switch.
Then the output was measured with a sampling oscilloscope. Injecting a control pulse into
the cavity produces carriers, which cause Mode-S to shift toward a shorter wavelength.
This results in a high transmittance for S2. Logically, this is an AND operation. On the
other hand, for S1, pulses are transmitted only when no control pulses arrive at the switch.
The experimental results are shown in Figure 1.48. Although the thermal accumulation
results in a slightly smaller contrast after a sequence of identical control bits, the modu-
lated signal recovers quickly before the next optical pulse arrives. Overall, the signal clock
was successfully modulated. The launched energy of each control optical pulse was about
500 fJ.
(a) (b)
0 5.0×1014 cm–2 w/ annealing
(1) 110 ps
–5
(2) 14 –2
2.0×10 cm w/ annealing
–10 (3) (1) No ion–implantation
Propagation loss (dB)
Power (a.u.)
(–1.9 dB/cm)
(3) 1.0×1011 cm–2 w/o annealing 11 –2
3.0×10 cm w/o annealing
(–11.8 dB/cm)
–20
(4) 3.0×1011 cm–2 w/o annealing 840 ps
(6) (–33.3 dB/cm)
–25 (5) 1.0×1012 cm–2 w/o annealing
3.0×1010 cm–2 w/o annealing
(–80.1 dB/cm)
(7) (6) 2.0×1014 cm–2 w/ annealing 850 ps
–30 (–16.0 dB/cm)
(7) 5.0×1014 cm–2 w/ annealing
Figure 1.49
(a) Propagation loss of a silicon waveguide as a function of ion dose and processing conditions. (b) Measured
carrier lifetime. (Reprinted with permission from Tanabe, T. et al., Appl. Phys. Lett., 90(3), 031115, 2007. © American
Institute of Physics.)
at an ion dose of 3.0 × 1011 cm-2. At the cost of the propagation loss, this lifetime value is
not small. In an attempt to reduce the propagation loss, we annealed the waveguide for 15
min at 900°C in a nitrogen atmosphere. This crystallizes the silicon and creates dislocation
loops at the center of the slab.66 As a result, the propagation loss was reduced to -16 dB/cm
for an ion dose of 2 × 1014 cm-2. Surprisingly, as shown in Figure 1.49, the carrier lifetime
was only 150 ps, which proved that recombination via a dislocation loop is very effective.
We employed this condition for the ion implantation of a photonic crystal nanocavity.
The fabrication process used for the ion implanted photonic crystal nanocavity is shown
in Figure 1.50. First, photonic crystals are fabricated via electron-beam lithography and dry
etching. Next, the wafer is covered with a resist mask and the photonic crystal nanocavity
regions are selectively ion implanted. Finally, annealing is performed followed by the air-
bridge process.
The photonic crystal nanocavity that we employed for this experiment was a point defect
hexapole cavity. First, we compared the Q values of the photonic crystal nanocavity with
and without ion implantation. The Q value was 3.6 × 104 for the sample without ion implan-
tation and 2.1 × 104 for the sample with ion implantation. The Q value was smaller for the ion
implanted sample but the difference is not significant for switching applications. Indeed
the switching contrast with respect to the input pulse energy is shown in the inset of
Figure 1.51 where the modulation depth is not significantly different. The recorded transmit-
tance waveforms of the signal light when a control pulse was injected at 0 ps are shown in
Figure 1.51a and b for the ion implanted sample and an as-fabricated sample. The switching
speed of the implanted sample is only 1/3 that of the original device, while the required
switching energy remains almost the same. This is one demonstration of our approach for
increasing the switching speed of a silicon-based photonic crystal nanocavity switch.
EB resist
SiO2 (n = 1.48)
Si wafer
(1) (2)
Ar ion implantation
Air-bridge
(3) (4)
Figure 1.50
Fabrication process used for ion-implanted silicon photonic crystal nanocavity. (1) A photonic crystal is patterned
using electron beam lithography. (2) A photonic crystal is fabricated by dry etching. (3) Ar ions are selectively
implanted in the photonic crystal nanocavity region. (4) An air-bridge is formed by wet etching. (Reprinted
with permission from Tanabe, T. et al., Appl. Phys. Lett., 90(3), 031115, 2007. © American Institute of Physics.)
Transmittance (a.u.)
220 ps
1.0
Non-ion-implanted 0.2 110 ps Ion-implanted
0.8
Modulation depth
Ion-implanted
0.6
0.1 50 ps
0.4
0.2 0.1
0.0
100 1000
Non-ion-impl. Input energy (f J)
0.01
–200 0 200 400 600 800 –200 0 200 400 600 800
Time (ps) Time (ps)
Figure 1.51
(a) Transmittance from ion-implanted and non-ion-implanted silicon photonic crystal nanocavity switch when
the wavelength of the signal light is at the resonance of the nanocavity. The control pulse is injected at t = 0 ps.
Inset is the switching contrast versus the input control pulse energy for ion-implanted and non-ion-implanted
devices. (b) The transmittance waveform when the signal light is slightly detuned towards a shorter wavelength.
(Reprinted with permission from Tanabe, T. et al., Appl. Phys. Lett., 90(3), 031115, 2007. © American Institute of
Physics.)
dependent shift. If the wavelength of the input light is slightly detuned in the same direction
as the shift of the cavity resonance, positive feedback occurs at a certain input power when
it is increased. As a result, the input/output power characteristic of the cavity exhibits opti-
cal bistability. Optical bistability results in a sudden jump from the off (low transmittance)
to the on (high transmittance) state. Once the cavity is on, it remains on until the input
power is reduced significantly because the cavity is exhibiting positive feedback. Optical
bistability is believed to be essential if we are to implement logic gates all-optically.
Because photonic crystal nanocavities have a large Q/V, they are expected to be capa-
ble of low power operation, which has long been considered difficult to achieve with
conventional devices. Here, we employed a four point defect cavity with shifted end holes
to demonstrate optical bistability. First, we studied bistable operation based on the thermo-
optic effect. The input wavelength of the laser light is scanned from a shorter to a longer
wavelength at different input powers. The recorded output power with respect to the input
wavelength is shown in Figure 1.52a. At an input power greater than 25 µW, a sudden drop
is observed at the long wavelength edge, which is one proof of optical bistability. Since
the thermo-optic effect shifts the cavity resonance to a longer wavelength, we know that
the measured bistability is the result of the thermo-optic effect caused by two photon
absorption carriers. In addition, the input versus output power for a CW laser at differ-
ent detuning values is shown in Figure 1.52b. When the input light is slightly detuned
to a longer wavelength, the input/output curve exhibits hysteresis, which is additional
evidence of optical bistability. The threshold power of optical bistability for this cavity was
only ~ 40 µW, which is significantly smaller than previous reported values obtained using
larger cavities. It is even possible to achieve a bistable threshold power of just a few µW by
using a local-width-modulated line-defect photonic crystal nanocavity with an extremely
high Q.10 Since the operation is based on the thermal effect, the operating speed is limited
by the thermal relaxation time, which is typically ~ 100 ns as discussed previously.
Since carrier-plasma dispersion is present in silicon, it is also possible to obtain optical
bistability using this nonlinearity. We performed a time-resolved measurement to observe
250 µW
Output power Pout (µW)
40 pm
101 160 µW 60 pm
100 µW 101
80 pm
60 µW
25 µW
100 10 µW
2.5 µW 100
250 µW 100 pm
10–1 120 pm
140 pm
160 pm
10–1 180 pm
–2
10
1567.5 1568.0 1568.5 1569.0 1569.5 101 102
Wavelength (nm) Input power Pin (µW)
Figure 1.52
Optical bistability experiment based on the thermo-optic effect for an L4 silicon photonic crystal nanocavity
with a Q value of 33,400 (i.e. spectrum FWHM of 47 pm). (a) Spectrum measured using a wavelength tunable
laser diode at different input powers swept from a shorter to a longer wavelength. The displayed value is the
power at the photonic crystal waveguide. (b) The input/output power characteristics as a function of wave-
length detuning from the cavity resonance. (From Notomi, M. et al., Opt. Express., 13, 2678–2687, 2005. © Optical
Society of America. With permission.)
(a) (b)
3.95 mW 1.58 mW 100 1.00 mW 0.40 mW
2.50 1.00 0.63 0.25
100
Power out (mW)
Power out (a.u.)
10–1
10–1
10–2
0.63 mW 0.25 mW 0.16 mW 0.06 mW
0.40 0.16 0.10 0.04
10–2 10–3
–200 0 200 400 600 800 –200 0 200 400 600 800
Figure 1.53
(a) Transmittance of a step input pulse at different input powers. (b) Calculated results. (From Tanabe, T. et al.,
Opt. Lett., 30, 2575–2577, 2005. © Optical Society of America. With permission.)
100
Power out (a.u.)
(1)
10–1
(2)
10–2 –1
10 100 101
Power in (mW)
Figure 1.54
Experimental demonstration of an optical transistor (transphasor) by injecting 200-ps wide input pulses at
different input powers.
du ( t ) u(t )
= P ( t )T [ δ + ∆n ( t ) λ 0 ] - (1.19)
dt τ ph
dN ( t ) 2 λ cβ N (t )
= 2 2 u(t ) -
2
(1.20)
dt hn V τc
dUT ( t ) N ( t ) UT ( t )
= Eg - (1.21)
dt τc τT
where u(t), N(t), and UT(t) are the photon energy in the cavity, the two-photon absorbed
carrier density, and the thermal energy density, respectively. n, β, and Eg are the refractive
index, two photon absorption coefficient, and bandgap energy of silicon, respectively. c, h,
and λ are the velocity of light, Planck’s constant, and the wavelength of the pump light,
respectively. τc and τT are the effective carrier lifetime and thermal relaxation time and
they are set at 80 ps and 100 ns, respectively, where the values are estimated from the
switching experiment. Equation 1.19 describes the photon energy stored in the cavity,
–0.2
–0.1
Detuning (nm)
0.0
0.1
0.2
Figure 1.55
2D mapping of transmittance waveform for a step input light at different detuning values. (From Tanabe, T.
et al., Opt. Lett., 30, 2575–2577, 2005. © Optical Society of America. With permission.)
Equation 1.20 illustrates the generation and relaxation of two-photon absorption carriers,
and Equation 1.21 is the generation and relaxation of heat. Equation 1.22 is the refractive
index modulation induced by the carrier plasma dispersion and the thermo-optic effect.
Equations 1.20 through 1.22 are solved in a self-consistent way, where the refractive index
obtained using Equation 1.22 is fed back to the first term on the right hand side of Equation
1.20. The transmitted waveforms are calculated in Figure 1.53b for a step input at different
input powers. The physical parameters are taken from references and the experiment;
hence there are no freely adjustable parameters. We observed a nonlinear jump, overshoot
and the slowing down effect and their behavior agrees well with the experiment.
Optical bistability is the key to achieving all-optical logic gates such as optical memory
or optical transistors. It can constitute a basic element of all-optical random access memory
(RAM)-based buffer.70 A numerical and experimental demonstration of optical memory
operation based on optical bistability achieved with the carrier-plasma effect is shown in
Figure 1.56. A step input is injected into the cavity and the power is adjusted slightly below the
bistable threshold power (off state). Injecting a set pulse enables the generation of two-photon
absorption carriers, which turns the cavity on. The calculated transmittance is shown in
Figure 1.56a. The calculation shows that the cavity can maintain the on state for more than
1 ns after the input pulse is injected. The experimental study shown in Figure 1.56b was
performed based on the calculation. The transmittance of the step input is indicated with (1).
When a set pulse is injected, it switches the bistable mode from off to on. As predicted by the
numerical study, the cavity maintains the on state after the set pulse is injected, as shown on
line (2). A negative pulse is applied to reset the cavity (3). Since the cavity stores on and off
state information, it can be regarded as a bistable optical memory device.
10–3
Power in/out (µW)
(2)
102 10–4
Output of (on) 0 ON ON
10
OFF
Output of (off)
1 OFF
10
10–1
(3)
(1)
–2
10
100
0 400 800 1200 1600 0 400 800 1200
Time (ps) Time (ps)
Figure 1.56
(a) Calculation of bistable memory operation with a silicon photonic crystal nanocavity. A 200-ps wide set pulse
is added to the bias step input at 500 ps to switch the cavity to the ON state. (b) Experimental demonstration of
bistable memory operation with an L4 photonic crystal nanocavity. The set pulse is added at 400 ps to set the
cavity to the ON state. A negative reset pulse is added at 800 ps. (1) The output of a step input. (2) The output
when a set pulse is added. (3) The output when a set pulse and a reset pulse are added. (From Tanabe, T. et al.,
Opt. Lett., 30, 2575–2577, 2005. © Optical Society of America. With permission.)
highly sophisticated logic gate circuits, such as memory, retiming or regenerative circuits.
In photonics, a flip-flop has been fabricated by combining two optical bistable devices.4
However, the reported device is not suitable for on-chip integration owing to its bulky
configuration and the large energy it requires for operation. The optical bistability
demonstrated using silicon photonic crystal nanocavities in the previous section will pro-
vide a good platform for constructing an on-chip optical flip-flop device. Based on the idea
of coupling two optical cavities to enable cross-feedback, which is the key to obtaining
flip-flop operation, we propose the design shown in Figure 1.57a. This photonic crystal
nanocavity system permits SR flip-flop operation, where the outputs can be switched by
injecting set or reset pulses into the inputs. S, R, and B are the input waveguides for the set
pulse, reset pulse and bias CW light, respectively. Q and Q are the output waveguides. A
simplified conceptualized diagram of the system is shown in Figure 1.57b. If we compare
this photonic system with that of the electrical SR flip-flop circuit, we can find various sim-
ilarities. We performed a 2D FDTD calculation based on this structural design. The time
sequence is shown in Figure 1.57d, where the set (CS) and reset (CR) pulses are applied with
Q Q Q Q
WG2 WG2
CR CS
B S
R
Q B Q
Drain
(d) (e)
B 22.5 ps (44 GHz)
CS
Output
Q
CR
Q Q
Q 0 20 40 60 80 100 120
Time (ps)
100
10–1
10–2
10–3
10–4
t=0 Set pulse Q=1 Reset pulse Q=0 Set pulse Q=1
Figure 1.57
All optical SR flip-flop consisting of two bistable cavities coupled to waveguides. (a) Schematic illustration of
the structural design based on hexagonal air-hole 2D photonic crystals. Two L7 cavities are connected through
waveguides. The end holes of the L7 cavities are shifted by 0.30a with hole diameters of 0.24a. (b) Simplified
conceptualized diagram of the system. (c) Equivalent electronic circuit diagram of the SR flip-flop. (d) Time
sequence of three inputs (bias, and set clock pulse, and reset clock pulse) and two outputs. (e) Calculated output
waveform and optical mode profiles of the optical SR flip-flop by using 2D FDTD. (From Notomi, M., et al., Opt.
Express., 15, 17458–17481, 2007. © Optical Society of America. With permission.)
different timings. We assumed Kerr nonlinearity for the calculation. The cavities CVS and
CVR have two resonant modes. The wavelength of one of the resonant modes has the same
wavelength between the two cavities. This mode is employed for the bias input. When we
inject a set pulse, the cavity CVS is turned on, and this enables the bias light to travel toward
port Q. On the other hand, when we inject a reset pulse, the cavity CVR is turned on, which
enables the bias light to flow toward port Q. The calculated results are shown in Figure 1.57e
along with the intensity profiles of the device. The calculated results show that a successive
SR flip-flop operation at 44 Gb/s is possible by using this system. It should be noted that
an optimized system will enable us to achieve a faster operating speed. We expect that
this demonstration of an on-chip optical flip-flop by using a photonic crystal platform will
constitute a significant step towards the development of all-optical integrated circuits.
PC
Pout1 Pout2
Data (λ2)
PA
Clock (λ1)
Pin1
Pin2 C1 Clock (λ3)
Transmittance
(Pout 3)
Pin3 C2 0.6
W0 0.4
PB
0.2
a 0
PD P 0 200 400 600
out3
Time (ps)
Figure 1.58
All-optical pulse retiming circuit based on two bistable cavities. (a) Schematic illustration of the photonic
circuit. The lattice constant a is 400 nm and the hole diameter is 0.55a. (b) Time sequence and calculated output
waveform obtained with 2D FDTD. AlGaAs material is assumed for the calculation (χ(3)/ε0 = 4.1 × 10- 19 (m2/V2)).
(From Notomi, M., et al., Opt. Express., 15, 17458–17481, 2007. © Optical Society of America. With permission.)
calculation. The instantaneous driving power is 60 mW for all three inputs. The calculated
result shows that the timing jitter of the data is completely removed and the input is
converted to a return-to-zero output data stream with a speed of 50 Gb/s. It is surprising
that such a complicated logic operation can be achieved by a photonic circuit composed
solely of two nanocavities. Note that the structure can be further optimized to reduce the
operating power or to enhance the operating speed.
1.6 Summary
The photonic crystal nanocavity system will enable us to confine photons in a very small
space for an extremely long period. Since the linewidth of the transmittance spectrum
and the footprint of the device are extremely small we are able to fabricate compact add-
drop filters. If we focus on the temporal characteristics of the nanocavity system, we find
that the system is capable of trapping photons and this constitutes an on-chip photonic
buffer memory. Even though it was not discussed in this chapter, we can perform various
operations to the trapped photons because the demonstrated photon lifetime is now of
the order of nanoseconds, which enable novel manipulation of light.20,71,72 An ultrahigh-Q
nanocavity system will also result in the very slow propagation of pulses, and this will
open the possibility of fabricating slow-light on-chip devices. In addition to the above
mentioned applications that utilize the linear property of the cavity system, photonic
crystal nanocavities are becoming more important for various light–matter interaction
studies, owing to the high photon density achievable at a small input power. As a result
of efficient light–matter interaction, active behavior such as all-optical switching or logic
operation is becoming possible. In this chapter, we demonstrated all-optical switching,
bistable memory operation, and various logic gate operations. It is extremely important
that such operation is achieved at a significantly small input power. These demonstrations
pave the way to the realization of on-chip photonic circuits.
References
1. M. Jinno, Y. Miyamoto, and Y. Hibino. 2007. Optical-transport networks in 2015. Nature Photon.,
1, 157–159.
2. H. M. Gibbs. 1985. Optical bistability: Controlling light with light. Academic Press, Orlando, FL.
3. S. D. Smith, A. C. Walker, F. A. P. Tooley, and B. S. Wherrett. 1987. The demonstration of restoring
digital optical logic. Nature, 325, 27–31.
4. H. Tsuda and T. Kurokawa. 1990. Construction of an all-optical flip-flop by combination of two
optical triodes. Appl. Phys. Lett., 57, 1724–1726.
5. J. Takeda, K. Nakajima, S. Kurita, S. Tomimoto, S. Saito, and T. Suemoto. 2000. Time-resolved
luminescence spectroscopy by the optical Kerr-gate method applicable to ultrafast relaxation
processes. Phys. Rev. B, 62, 10083–10087.
6. E. Centeno and D. Felbacq. 2000. Optical bistability in finite-size nonlinear bidimensional
photonic crystals doped by a microcavity. Phys. Rev. B, 62, 7683–7686.
7. M. Soljacic and J. D. Joannopoulos. 2004. Enhancement of nonlinear effects using photonic
crystals. Nature Mat., 3, 211–219.
32. H. Haus. 1984. Waves and fields in optoelectronics. Prentice-Hall, Englewood Cliffs, NJ.
33. M. Notomi, A. Shinya, E. Kuramochi, S. Mitsugi, H.-Y. Ryu, T. Kawabata, T. Tsuchizawa, T.
Watanabe, T. Shoji, and K. Yamada. 2004. Photonic-band-gap waveguides and resonators in
SOI Photonic crystal slabs. IEICE Trans. Electron., E87-C, 398–408.
34. S. Mitsugi, A. Shinya, E. Kuramochi, M. Notomi, T. Tsuchizawa, and T. Watanabe. 2003.
Resonant tunneling wavelength filters with high Q and high transmittance based on photo-
nic crystal slabs. In Proceedings of 2003 IEEE LEOS Annual Meeting. Institute of Electrical and
Electronics Engineers, New York, pp. 214–215.
35. M. Notomi, A. Shinya, S. Mitsugi, E. Kuramochi, and H. Ryu. 2004. Waveguides, resonators and
their coupled elements in photonic crystal slabs. Opt. Express, 12, 1551–1561.
36. S. Mitsugi, A. Shinya, E. Kuramochi, M. Notomi, T. Tsuchizawa, and T. Watanabe. 2004.
Wavelength adjustable photonic crystal wavelength filters with high-Q and high-transmittance.
In Technical Digest of PECS-V, p. 153.
37. Y. Akahane, T. Asano, B.-S. Song, and S. Noda. 2003. Fine-tuned high-Q photonic-crystal
nanocavity. Opt. Express, 13, 1202–1214.
38. A. Shinya, S. Mitsugi, E. Kuramochi, T. Tanabe, G. Kim, G. Kira, S. Kondo, K. Yamada, T.
Watanabe, T. Tsuchizawa, and M. Notomi. 2005. Ultrasmall resonant tunneling/dropping
devices in 2D photonic crystal slabs. Proc. SPIE, 5727, 72–85.
39. T. Yoshie, A. Scherer, J. Hendrickson, G. Khitrova, H. M. Gibbs, G. Rupper, C. Ell, O. B. Shchekin
and D. G. Deppe. 2004. Vacuum Rabi splitting with a single quantum dot in a photonic crystal
nanocavity. Nature, 432, 200–203.
40. A. Badolato, K. Hennessy, M. Atatüre, J. Dreiser, E. Hu, P. Petroff, and A. Imamoğ lu. 2005.
Deterministic coupling of single quantum dots to single nanocavity modes. Science, 308,
11580–1161.
41. A. Shinya, S. Mitsugi, E. Kuramochi, and M. Notomi. 2006. Ultrasmall multi-port channel drop
filter in two-dimensional photonic crystal on silicon-on-insulator substrate. Opt. Express, 14,
12394–12400.
42. H.-Y. Ryu, M. Notomi, and Y.-H. Lee. 2003. High-quality-factor and small-mode-volume hexa-
pole modes in photonic-crystal-slab nanocavities. Appl. Phys. Lett., 83, 4294–4296.
43. G.-H. Kim, Y.-H. Lee, A. Shinya, and M. Notomi. 2004. Coupling of small, low-loss hexapole
mode with photonic crystal slab waveguide mode. Opt. Express, 12, 6624–6631.
44. T. Tanabe, A. Shinya, E. Kuramochi, S. Kondo, H. Taniyama, and M. Notomi. 2007. Single point
defect photonic crystal nanocavity with ultrahigh quality factor achieved by using hexapole
mode. Appl. Phys. Lett., 91, 021110.
45. S. G. Johnson, S. Fan, A. Mekis, and J. D. Joannopoulos. 2001. Multipole-cancellation mecha-
nism for high-Q cavities in the absence of a complete photonic band gap. Appl. Phys. Lett., 78,
3388–3391.
46. T. Tanabe, K. Yamada, K. Nishiguchi, A. Shinya, E. Kuramochi, H. Inokawa, M. Notomi, T.
Tsuchizawa, T. Watanabe, H. Fukuda, H. Shinojima, and S. Itabashi. 2007. Fast all-optical
switching using ion-implanted silicon photonic crystal nanocavities. Appl. Phys. Lett., 90,
031115.
47. E. Kuramochi, M. Notomi, S. Mitsugi, A. Shinya, T. Tanabe, and T. Watanabe. 2006. Ultrahigh-Q
photonic crystal nanocavities realized by the local width modulation of a line defect. Appl. Phys.
Lett., 88, 041112.
48. K. Inoshita and T. Baba. 2003. Lasing at bend, branch and intersection of photonic crystal
waveguides. Electron. Lett., 39, 844–846.
49. B.-S. Song, S. Noda, T. Asano, and Y. Akahane. 2005. Ultra-high-Q photonic double-heterostruc-
ture nanocavity. Nature Mat., 4, 207–210.
50. T. Tawara, T. Ito, T. Tanabe, K. Tateno, E. Kuramochi, M. Notomi, and H. Nakano. 2006. Highly
selective ZEP/AlGaAs etching for photonic crystal structures using Cl2/HI/Xe mixed plasma.
Jpn. J. Appl. Phys., 45, L917–L919.
51. T. Tanabe, M. Notomi, and E. Kuramochi. 2007. Measurement of an ultra-high-Q photonic crys-
tal nanocavity using a single-side-band frequency modulator. Electron. Lett., 43, 187–188.
52. T. Tanabe, M. Notomi, E. Kuramochi, and H. Taniyama. 2007. Large pulse delay and small
group velocity achieved using ultrahigh-Q photonic crystal nanocavities. Opt. Express, 15,
7826–7839.
53. Y. Tanaka, J. Upham, T. Nagashima, T. Sugiya, T. Asano, and S. Noda. 2007. Dynamic control of
the Q factor in a photonic crystal nanocavity. Nature Mat., 6, 862–865.
54. Q. Xu, P. Dong, and M. Lipson. 2007. Breaking the delay-bandwidth limit in a photonic structure.
Nature Phys., 3, 406–410.
55. M. Notomi, K. Yamada, A. Shinya, J. Takahashi, C. Takahashi, and I. Yokohama. 2001. Extremely
large group-velocity dispersion of line-defect waveguides in photonic crystal slabs. Phys. Rev.
Lett., 87, 253902.
56. S.-C. Huang, M. Kato, E. Kuramochi, C.-P. Lee, and M. Notomi. 2007. Time-domain and spec-
tral-domain investigation of inflection-point slow-light modes in photonic crystal coupled
waveguides. Opt. Express, 15, 3543–3549.
57. A. Yariv, Y. Xu, R. K. Lee, and A. Scherer. 1999. Coupled-resonator optical waveguide: a pro-
posal and analysis. Opt. Lett., 24, 711–713.
58. F. Xia, L. Sekaric, and Y. Vlasov. 2007. Ultracompact optical buffers on a silicon chip. Nature
Photon., 1, 65–71.
59. M. Notomi, E. Kuramochi, and T. Tanabe. 2008. Large-scale arrays of ultraigh-Q coupled nano-
cavities. Nature Photon., 2, 741–747.
60. A. Shinya, S. Mitsugi, E. Kuramochi, and M. Notomi. 2006. Ultrasmall multi-port channel drop
filter in two-dimensional photonic crystal on silicon-on-insulator substrate. Opt. Express, 14,
12394–12400.
61. V. Almeida, C. Barrios, R. Panepucci, and M. Lipson. 2004. All-optical control of light on a
silicon chip. Nature, 431, 1081–1084.
62. M. Notomi, A. Shinya, S. Mitsugi, G. Kira, E. Kuramochi, and T. Tanabe. 2005. Optical bistable
switching action of Si high-Q photonic-crystal nanocavities. Opt. Express, 13, 2678–2687.
63. A. Shinya, S. Mitsugi, T. Tanabe, M. Notomi, I. Yokohama, H. Takara, and S. Kawanishi. 2006.
All-optical flip-flop circuit composed of coupled two-port resonant tunneling filter in two-
dimensional photonic crystal slab. Opt. Express, 14, 1230–1235.
64. T. Tanabe, H. Taniyama, and M. Notomi. 2008. Carrier diffusion and recombination in photonic
crystal nanocavity optical switches. IEEE J. Lightwave Technol., 26, 1396–1403.
65. K.-H. Hellwege. 1982. Landolt-Börnstein Numerical Data and Functional Relationships in Science
and Technology Group III: Crystal and Solid State Physics. Volume 17a. Springer-Verlag, Berlin.
66. K. Yamada, T. Tsuchizawa, T. Watanabe, H. Fukuda, H. Shinojima, T. Tanabe, and S.-I. Itabashi.
2006. All-optical signal processing using nonlinear effects in silicon photonic wire waveguides.
In The 19th Annual Meeting of the IEEE Lasers & Electro-Optics Society. Institute of Electrical and
Electronics Engineers, New York.
67. O. Palais and A. Arcari. 2003. Contactless measurement of bulk lifetime and surface
recombination velocity in silicon wafers. J. Appl. Phys., 93, 4686–4690.
68. H. Mäckel and A. Cuevas. 2003. Determination of the surface recombination velocity
of unpassivated silicon from spectral photoconductance measurement. In 3rd World Conference
on Photovoltaic Energy Conversion. Institute of Electrical and Electronics Engineers, New
York.
69. T. Tanabe, K. Yamada, K. Nishiguchi, E. Kuramochi, A. Shinya, H. Inokawa, S. Kawanishi, and
M. Notomi. 2006. Fast all-optical pulse train modulation by silicon photonic crystal nanocavi-
ties. In Proceedings of the 19th Annual Meeting of the IEEE Lasers and Electro-Optics Society. Institute
of Electrical and Electronics Engineers, New York.
70. K. Kitayama, S. Arakawa, S. Matsuo, M. Murata, M. Notomi, R. Takahashi, and Y. Itaya. 2007.
All-optical RAM-based buffer for packet switch. In Proceedings of Photonic in Switching, 2007.
Institute of Electrical and Electronics Engineers, New York.
71. M. Notomi and S. Mitsugi. 2006. Wavelength conversion via dynamic refractive index tuning of
a cavity. Phys. Rev. A, 73, 051803.
72. M. Notomi, H. Taniyama, S. Mitsugi, and E. Kuramochi. 2006. Optomechanical wavelength
and energy conversion in high-Q double-layer cavities of photonic crystal slabs. Phys. Rev. Lett.,
97, 023903.
Jelena Vučkovic’
Stanford University
Matthew Pelton
Argonne National Laboratory
Glenn S. Solomon
National Institute of Standards and Technology and University of Maryland
Edo Waks
Institute for Research in Electronics and Applied Physics
Contents
2.1 Introduction...........................................................................................................................54
2.2 Design and Fabrication........................................................................................................ 58
2.2.1 Design of Pillar Microcavities................................................................................. 59
2.2.2 Growth of QDs..........................................................................................................65
2.2.3 Fabrication of Pillar Structures............................................................................... 68
2.2.3.1 First Generation.......................................................................................... 68
2.2.3.2 Second Generation..................................................................................... 69
2.2.3.3 Third Generation........................................................................................ 72
2.3 Device Characterization...................................................................................................... 72
2.3.1 Modifying Single QD Spontaneous Emission...................................................... 73
2.3.1.1 First and Second Generation.................................................................... 74
2.3.1.2 Third Generation........................................................................................ 79
2.3.2 Photon Statistics........................................................................................................80
2.3.2.1 Mechanism for Single-Photon Generation in QDs................................ 81
2.3.2.2 Experimental Results with Pillar DBR Devices.....................................83
2.3.3 Efficiency.................................................................................................................... 86
2.3.4 Photon Indistinguishability.................................................................................... 89
2.3.5 Strong Coupling........................................................................................................ 95
53
© 2009 by Taylor & Francis Group, LLC
2.4 Applications........................................................................................................................... 98
2.4.1 BB84 Quantum Key Distribution........................................................................... 99
2.4.2 Entanglement Generation without a “True” Interaction................................... 102
2.4.3 Single-Mode Teleportation.................................................................................... 108
2.4.4 Coherent Single-Photon Emission and Trapping............................................... 114
2.4.4.1 Coherent Photon Generation in Ideal Systems.................................... 115
2.4.4.2 Performance of Practical Systems.......................................................... 119
2.4.4.3 Performance as a Single-Photon Source............................................... 120
2.4.4.4 Mathematical Details of the Theory...................................................... 120
2.5 Conclusions.......................................................................................................................... 125
References...................................................................................................................................... 126
2.1 Introduction
This chapter describes devices designed to emit single-photons at controlled time intervals,
as needed for various quantum information applications. Before describing how such a
device might be constructed, let us define exactly what we mean by a source of single-
photons. Such a light source is quite different from a classical source such as a laser or
a light emitting diode, where the number of photons per pulse varies randomly, usually
following a Poisson distribution. However, a realistic single-photon source will never be
perfect, sometimes emitting zero or multiple photons, and thus the distinction between a
classical photon source and a realistic single-photon source is subtle. We therefore need
quantitative measures to distinguish between light sources based on photon statistics.
From the viewpoint of system applications, other characteristics are also important. The
most important parameters describing a single-photon source are:
1. The second-order coherence function g 0(2) [1,2] which places an upper bound on
the two or multi-photon probability per pulse, pm = p ( n ≥ 2 ) ≤ (1/2) g 0( ) n , where
2 2
〈n〉 is the average photon number per pulse. A photon source with 〈n〉 = 1 and with,
g 0( 2 ) = 1 , g 0( 2 ) < 1 , or g 0( 2 ) = 0 is called a Poisson, sub-Poisson or ideal single-photon
source, respectively. Since a realistic single-photon source has 0 < g 0( 2 ) < 1, it is a
sub-Poisson photon source operating in the regime of n < 1. When a single-photon
source is used in quantum key distribution systems, therobustness against a spe-
cific eavesdropping attack, called the photon number splitting attack, requires a
small value of g 0( 2 ).
2. An external quantum efficiency ηtotal, which defines an overall out-coupling effi-
ciency of internally generated single-photons into a desired single waveguide
mode.
3. Maximum repetition frequency fmax of single-photon generation, which deter-
mines an upper bound on the clock frequency of a system. These two parameters,
ηtotal and fmax, must be reasonably high for practical applications. In the case of a
laser, the stimulated emission of photons enhances these parameters. For a single-
photon source, however, stimulated emission is not an option and we need a new
trick to achieve high values of ηtotal and fmax. One of the most useful techniques
available for this purpose is to use cavity quantum electrodynamical effects [3] to
enhance spontaneous emission.
4. The overlap v(∆t = 0) ≡ v(0) between two single-photon wavefunctions [4], which
describes the degree of quantum indistinguishability of two single-photons when
the time delay Dt between them is zero. This overlap can be computed either in the
space-time or momentum-frequency domains, and an ensemble average is taken
over all possible quantum states generated by the source. To use a single-photon
source for such quantum information systems as a Bell state analyzer, quantum
teleportation and linear optics quantum computation, single-photons must be
identical quantum particles with v(0) = 1. However, a realistic single-photon source
has a finite timing jitter and dephasing for the generated electromagnetic field,
which reduces v(0) below one.
5. An emission wavelength uncertainty, Dl, in different single-photon sources. When
a single-photon source is used in a large scale photonic system, massively parallel
generation of single-photons is needed. In such a case, the emission wavelength of
single-photon sources must be identical.
(a) (b)
r r
Relaxation
a Pumping
Emission
Pumping Emission
e
b g
Figure 2.1
Three-level schemes states used for generating single-photons. (a) Configuration for incoherent generation of
single-photons by spontaneous emission. (b) Configuration for coherent generation of single-photons by Raman
scattering.
γL
κL
κc a
Figure 2.2
A single “atom”-cavity system and physical parameters.
ωa (2.1)
g =℘ ,
2 εV
where ℘ is the dipole moment, wa is the frequency of the dipole transition, e is the permit-
tivity of the semiconductor, and V is the cavity mode volume. For the present discussion
we have assumed that the dipole is located at the cavity field maximum and that the dipole
is perfectly aligned with the electric field. The spontaneous emission rate of the dipole in
the bulk material (without the cavity) is [12],
nω 3a℘2
Γ= , (2.2)
3 πε 0 c 3
Γ 1 λ3
fosc = = ω a Γ. (2.3)
3Γ ceo ( 4 πK )2 n
Here, K = e/√4εme, where e and me are the electron charge and mass, l is the wavelength
in vacuum, and Γ ceo = nω 2a e 2 /(6 πε 0me c 3 ) is the spontaneous emission rate from a classical
electron oscillator [13]. We can then re-write the coupling rate as,
g = K 3 fosc /V . (2.4)
V=
∫∫∫ ε(r) |E(r) | d r .
2 3
(2.5)
max(ε(r ) | E(r ) |2 )
For the maximum possible interaction between a cavity photon and a QD dipole, the elec-
tric field of a single photon should be as large as possible at the location of the QD. This
occurs when we confine the photon to the smallest possible space. Another important
parameter is the rate at which photons leak out of the cavity. This is given by k = kC + kL,
where kC is the rate at which photons couple to a down-stream waveguide mode and kL
is the rate at which photons are lost to other spurious modes or absorption in the cavity
mirrors. These determine the length of time over which the atom and cavity photon can
interact, and the coupling efficiency between a cavity and an external optical channel.
There are a number of figures of merit to be considered from these constants, which we
now consider and discuss in the context of semiconductors. First is the quality factor of
the cavity, Q = w cav/κ for cavity frequency w cav. The small size of semiconductor microcavi-
ties make them subject to surface defects, which prevents their Q from being as high as
in larger cavities. For single-photon generation and quantum networking, the ratio kC/k
is also critical to ensure that all single photons that leak out of the cavity end up in the
desired waveguide mode.
The Purcell factor is defined as the enhancement factor of the spontaneous emission
decay rate in the weak-coupling limit (g << κ),
kg 2 / Γ
F( ∆ ) = , (2.6)
∆ + k2 /4
2
where ∆ = w a –w cav is a detuning parameter between the QD or atomic transition frequency
w a and cavity resonant frequency w cav, and Γ is the spontaneous emission rate with-
out a cavity QED effect. The out-coupling efficiency into a desired waveguide mode is
given by
ΓL 1
ηtotal = 1 - × , (2.7)
ΓF(0) 1 + k L / k c
ω ± = ω a - i(k + Γ ) / 4 ± g 2 - (k - Γ )2 / 16 . (2.8)
When g > |κ – Γ|/4, the system shows a splitting into two eigenmodes separated by a
frequency called the vacuum Rabi splitting. This is the strong-coupling regime.
Strong coupling is not needed for the coherent generation and capture of single-photons,
but it makes such schemes far simpler. In the weak coupling regime (g < |κ – Γ|/4), large
detunings and very fast, critically shaped pulses must be used for the coherent gen-
eration and trapping of single-photons. The critical figure of merit for system applica-
tions of single-photon sources is not g/4|κ – Γ|, that is, whether we are in the strong or
weak coupling regime, but rather the cooperativity parameter (or spontaneous emission
coupling efficiency), C = 4g 2/Γκ. An immediate physical indication that this parameter
may take on a high value in existing semiconductor microcavity systems is the Purcell
effect, in which the rate of radiation of an electric dipole is enhanced by the modified
local density of photon states inside a cavity. This may be seen from Equation 2.4 by
taking g, Γ << κ, in which case the + solution is ωa – i(Γ/2)(1 + C), indicating that the radia-
tive damping of the atom is enhanced by the factor 1 + C. C may be quickly estimated
as C = 3/4π2 × (λ/n)3 × Q/V. Semiconductor microcavity systems show a key advantage in
achieving high values of C in comparison to larger cavities that might be used with
trapped atoms or ions.
Pillar microcavities containing InAs quantum dots (QDs) combine small mode vol-
umes with efficient coupling to free-space and thus can make excellent single-photon
sources. The only major drawback in this system is the low operation temperature ( < 50
K) required to avoid excessive broadening and nonradiative losses in the QD transitions.
We will begin by describing the design, crystal growth and lithography involved in mak-
ing these devices at Stanford University. Next, we describe optical characterization of
pillar microcavity structures, first demonstrating spontaneous emission modification in
single InAs QDs, and then testing the degree to which these structures, excited by short
optical pulses, can satisfy the requirements of reduced two-photon probability, high col-
lection efficiency, and photon indistinguishability. We also describe quantum-optical
characterization of pillar microcavities provided by the Forchel group at the Univerity of
Würzburg which have high enough Q factors to reach the strong-coupling regime [14]. We
then describe proof-of-principle demonstrations of simple applications for these devices,
including quantum cryptography and experiments based on two-photon interference,
and provide a detailed theory of coherent trapping and emission in a three-level Λ system
coupled to a microcavity, which has implications for future applications in the area of
quantum networking.
structures, the goals being to obtain narrow pillars with straight sidewalls and minimal
surface roughness.
MPT DBR
layers
nh s
b a Spacer
nh
MPB nl DBR
layers
nh
Figure 2.3
Left: Parameters for a pillar microcavity. The pillar is rotationally symmetric around the vertical axis. Right:
Electric-field intensity for the fundamental (HE11) mode in a pillar microcavity. The electric field is predomi-
nantly polarized in the plane of the distributed Bragg reflector layers. Since a pillar with perfect rotational sym-
metry supports two orthogonally polarized modes, any polarization in this plane (linear, circular, elliptical) is
possible.
elliptical. Since epitaxial QDs have two electric dipole transitions polarized in this plane,
the cavity and QD polarizations align perfectly, in theory. However, since asymmetry in
the pillar shape can produce a polarization splitting of the cavity modes, and asymmetry
in the QD can produce a polarization splitting in the dipole transitions, more complicated
situations are possible, as will be mentioned in the next section. For the pillar modes, there
is a small deviation from the in-plane polarization at larger distances from the pillar axis.
All theoretical analyses presented in this section are performed by the Finite-Difference
Time-Domain (FDTD) method; for the detailed description of this method, please refer to
Pelton et al. [15].
The rule of thumb generally used for designing pillars is to make mirror layers one-
quarter wavelength thick, and to choose the optical thickness of the spacer equal to the
target wavelength. In the case of a planar DBR cavity (with diameter D → ∞), this choice of
parameters leads to the maximum reflectivities of the mirrors and the maximum Q-factor
of the cavity mode: the cavity operates at the Bragg wavelength, for which the partial reflec-
tions from all high- and low-refractive-index interfaces add up exactly in phase. However,
since the strength of the cavity QED phenomena is proportional to the ratio of the cavity
Q-factor to the mode volume V, we will try to design pillars in such a way that this ratio is
maximized. In order to do so, we will explore structures with small diameters D, and will
attempt to improve their Q factors.
In general, Q reduces with a decrease in D, due to the combination of two loss mecha-
nisms: longitudinal loss through DBR mirrors, and transverse loss due to imperfect TIR con-
finement in the transverse direction. Let us address the longitudinal loss first. The decrease
in the post diameter D implies a change in the dispersion relation of the 1D photonic crys-
tal, and the size and position of its bandgap, as illustrated in Figure 2.4. In this figure, it is
0.2
GaAs/AlAs
0.19 GaAs/AlGaAs
0.18
a/λ
0.17
0.16
0.15
Figure 2.4
Bandgap edges, calculated using the FDTD method (points), of the fundamental (HE11) mode in a cylindri-
cal one-dimensional photonic crystal in the GaAs/AlAs or GaAs/Al xGa1 - xAs material systems. The lines are
guides to the eye. The GaAs/AlAs photonic crystal has the following parameters: nh = 3.57, nl = 2.94, b = 85 nm,
and a = 155 nm. The GaAs/Al xGa1 - x As photonic crystal has the following parameters: nh = 3.57, nl = 3.125, b = 80
nm, and a = 150 nm. (See Figure 2.3 for definition of parameters.) The bandgap edges for D → ∞ are positioned at
a/λ equal to 0.1445 and 0.1634 for the GaAs/AlAs photonic crystal, and at a/λ equal to 0.1431 and 0.1565 for the
GaAs/Al xGa1–xAs photonic crystal. (From Vučkovic’, J., Pelton, M., Scherer, A., and Yamamoto, Y., Phys. Rev. A, 66
(2002) 023808. © American Physical Society. With permission.)
assumed that the high- and low-refractive-index regions of the photonic crystal consist of
GaAs and AlAs, with refractive indices of nh = 3.57 and nl = 2.94, and thicknesses of 70 and
85 nm, respectively, or that they consist of GaAs and Alx Ga1 - x As, with refractive indices of
nh = 3.57 and nl = 3.125, and thicknesses of 70 and 80 nm, respectively. The y-axis gives a/λ,
where a is the mirror periodicity (155 or 150 nm; see figure caption) and l is the vacuum
wavelength corresponding to the bandgap edge. When the diameter D decreases, the fre-
quencies of the bandgap edges increase, and the size of the bandgap decreases. For struc-
ture diameters larger than 2 mm, bandgap edges can be approximated by their values at
D → ∞. Therefore, as D decreases, the blue shift of the cavity mode wavelength l increases
relative to the target wavelength at which the 1D cavity operates [15]. Simultaneously, the
size of the photonic bandgap decreases, implying that the cavity mode is less confined in
the longitudinal direction than in the planar cavity case.
The cavity mode is strongly localized in real space, and consequently delocalized in
Fourier space (k-space), meaning that it consists of a wide range of wave-vector compo-
nents. Some of these components are not confined in the post by TIR; i.e., they are posi-
tioned above the light line, where they can couple to radiative modes, leading to transverse
loss. A cavity mode which is strongly confined in the longitudinal direction by high-reflec-
tivity mirrors is delocalized in Fourier space and suffers large transverse loss. Similarly,
a mode that is delocalized in the longitudinal direction is more localized in Fourier space
and suffers less transverse loss. Therefore, when optimizing the quality factor of three-
dimensional (3D) pillars, there is a tradeoff between these two loss mechanisms.
In the middle of a large bandgap, the longitudinal confinement is strongest, but the
Q-factor is limited by transverse loss. By shifting the resonant wavelength away from the
mid-gap (e.g., by tuning the thickness of the cavity spacer), one can delocalize the mode in
real space, localizing it more strongly in Fourier space, reducing the contribution of wave-
vector components above the light line, and thereby decreasing the transverse radiation
loss. Eventually, as the mode wavelength approaches the bandgap edges, the loss of lon-
gitudinal confinement starts to dominate and Q drops. Therefore, in the pillars with high
reflectivity mirrors and finite diameter, it is expected that the maximum Q will be located
away from the mid-gap position. Moreover, since the mode wavelength can be tuned from
the mid-gap towards any of the two bandgap edges, two local maxima of Q (i.e., a double
peak behavior in Q versus mode wavelength) are expected. Besides detuning the mode
wavelength from the mid-gap, we can also suppress the transverse loss by relaxing the
mode slightly in the longitudinal direction, i.e., by reducing the reflectivities of photonic
crystal mirrors and decreasing the bandgap size. This can be achieved by shrinking the
cavity diameter, or by changing the photonic crystal parameters (e.g., by reducing the
refractive-index contrast).
In this section, we study both of these approaches to Q optimization: tuning the mode
wavelength away from the mid-gap by changing the spacer thickness, and tuning the mir-
ror reflectivities by changing photonic crystal parameters or cavity diameter [16]. We also
show that the employment of very high reflectivity mirrors cannot lead to high-Q cavities
with small diameters, as the transverse radiation loss is high, resulting from very strong
mode localization in the longitudinal direction. Finally, we comment on the degradation
of the pillar quality factor as a result of fabrication imperfections.
Let us first study the same set of pillar parameters as in our earlier work [15], chosen
in such a way that the cavity would operate at the Bragg wavelength for D → ∞: a = 155
nm, b = 85 nm, s = 280 nm, nh = 3.57 and nl = 2.94, MPT = 15 and MPB = 30. Such a refractive
index contrast corresponds to the GaAs/AlAs material system used in the experiments
presented in this chapter. In an attempt to maximize the ratio of the cavity Q-factor to the
mode volume, we here analyze only the behavior of the HE11 mode for the cavity diameter
D below 0.5 mm.
In order to tune the mode frequency within the bandgap, we tune the spacer thickness
s. Results for l, Q, V, and Q/V are shown in Figures 2.5 and 2.6. From Figure 2.4, we see
that the bandgaps in these structures extend from 875 to 969 nm, from 850 to 920 nm, and
from 790 to 850 nm, for structure diameters of 0.5, 0.4, and 0.3 mm, respectively. As we have
noted previously, when D decreases, the bandgap edges shift towards lower wavelengths,
and the size of the bandgap decreases. The cavity mode wavelength is blue-shifted in this
process, as can be seen in Figure 2.5.
The mode volume V is minimized when the mode wavelength is located near the middle
of the bandgap, as shown in Figure 2.6. For the structures with D equal to 0.4 and 0.3 mm,
the maximum Q-factor also occurs close to the mid-gap. Different behavior is seen for the
980 10000
D = 0.5 µm D = 0.5 µm
960 D = 0.4 µm 9000 D = 0.4 µm
D = 0.3 µm D = 0.3 µm
940 8000
920 7000
900 6000
λ (nm)
880 5000
Q
860 4000
840 3000
820 2000
800 1000
780 0
220 240 260 280 300 320 340 360 220 240 260 280 300 320 340 360
s (nm) s (nm)
Figure 2.5
Wavelength λ and quality factor Q of the fundamental mode in a pillar with a = 155 nm, b = 85 nm, nh = 3.57, nl = 2.94,
MPT = 15 and MPB = 30. The cavity diameter D and the spacer thickness s are tuned. (From Vučkovic’, J., Pelton, M.,
Scherer, A., and Yamamoto, Y., Phys. Rev. A, 66 (2002) 023808. © American Physical Society. With permission.)
4 6000
D = 0.5 µm D = 0.5 µm
D = 0.4 µm D = 0.4 µm
3.5 D = 0.3 µm 5000 D = 0.3 µm
3 4000
(Q/V)(λ/nh)3
V(nh/λ)3
2.5 3000
2 2000
1.5 1000
1 0
220 240 260 280 300 320 340 360 220 240 260 280 300 320 340 360
s (nm) s (nm)
Figure 2.6
Mode volume V and ratio of quality factor Q to V for the HE11 mode in a pillar with a = 155 nm, b = 85 nm, nh = 3.57,
nl = 2.94, MPT = 15 and MPB = 30. The cavity diameter D and the spacer thickness s are tuned. (From Vučkovic’, J., Pelton,
M., Scherer, A., and Yamamoto, Y., Phys. Rev. A, 66 (2002) 023808. © American Physical Society. With permission.)
structure with D equal to 0.5 mm, which has a local minimum of Q at mid-gap and exhibits
a double-peak behavior.
The double-peak behavior is due to the mechanism described above. In the middle of
the bandgap, where the longitudinal mode confinement is strongest and the mode volume
is minimum, the radiation loss in the transverse direction is high, and the Q-factor is
degraded. By shifting the resonant wavelength away from the mid-gap, the mode is delo-
calized in real space, leading to a reduction in the transverse radiation loss (e.g., at the posi-
tions of the two peaks in Q). Eventually, as the mode wavelength approaches the bandgap
edges, the loss of longitudinal confinement starts to dominate, Q drops, and the mode vol-
ume increases. To support this explanation, we analyzed the same structure, with D = 0.5
mm, but with the number of mirror pairs on top (MPT) increased from 15 to 25 (not shown).
At mid-gap, Q did not increase significantly with MPT. The mode there is already strongly
confined in the longitudinal direction, and the addition of extra pairs does not change
the longitudinal loss. The modal Q-factor is determined by the radiation loss in the trans-
verse direction, which is independent of MPT. On the other hand, the Q’s at the two peaks
increased with MPT. At these points, the mode is not confined as well in the longitudinal
direction, and longitudinal loss can be reduced by adding more mirror pairs.
As an even stronger demonstration of our explanation for the double-peak behavior, we
separate the radiation loss into the loss above the top pillar surface (La), and the loss below
it (Lb). The total Q is a combination of two newly introduced quality factors, Qa and Qb,
which are inversely proportional to La and Lb, respectively:
1 1 1
= + . (2.9)
Q Qa Qb
It follows from their definition that Qa and Qb are measures of the longitudinal and
transverse loss, respectively. We analyze two sets of structure parameters, corresponding
to the local maximum or minimum in Q. For s = 270 nm and D = 0.5 mm (local maximum),
we calculate Qa ≈ 14,500 and Qb ≈ 13,910, while, for s = 290 nm and D = 0.5 mm (local mini-
mum), we calculate Qa ≈ 16,000 and Qb ≈ 5100. These results show that the local minimum
in Q is due to an increase in the transverse loss, manifested as a drop in Qb.
Let us now address the single-peak behavior of Q as a function of cavity spacer thick-
ness, when D is equal to 0.4 or 0.3 mm. Structures with smaller diameters have smaller
bandgaps, as illustrated in Figure 2.4, and the cavity modes are more delocalized in the
longitudinal direction, relative to the structure with D = 0.5 mm. The defect modes must
therefore be more localized in Fourier space, and will thus suffer less radiation loss in the
transverse direction. This implies that the Q-factors are determined mostly by the longitu-
dinal loss. They reach their maxima at the mid-gap, where the mode volume is minimum,
and the longitudinal confinement is strongest.
The maximum Q/V ratio of almost 6000 (where V is measured in cubic wavelengths in
the high-refractive index material) is achieved for the structure with D = 0.4 mm, as shown
in Figure 2.6. For this structure, the Q-factor is close to 9500, and the mode volume is
1.6(λ/nh)3. For D = 0.4 mm, a variation in the thicknesses of the mirror layers allows us to
achieve a small increase in the Q-factor, to 10,500, and in the Q/V ratio, to 6500. This result
is obtained for a = 155 nm, b = 75 nm, and s = 290 nm. Further improvement of the Q/V ratio
could possibly be achieved by employing more sophisticated methods such as genetic
algorithms [17] and inverse design [18], in which more pillar parameters (e.g., individual
mirror layers) are tuned.
In reality, imperfections in the growth and fabrication processes used to construct the
pillars inevitably lead to reduced performance. Possible problems include deviations in
the layer thickness, residual absorption, surface roughness, and nonvertical sidewalls. As
discussed below, we find experimentally that the Q values of pillar structures are sig-
nificantly reduced compared with the planar cavities before etching, and we attribute this
reduction mainly to vertical undercut and insufficient etch depth through the lower DBR,
which lead to an increase in the transverse losses and diffraction in the lower DBR, respec-
tively, and cause a degradation of the quality factor of the fundamental mode, as shown in
Figure 2.7 [15]. The problems of the insufficient etch depth and the vertical undercut have
been gradually minimized in subsequent generations of our pillars, thereby leading to
improved experimental Q-factors and better cavity QED results. Details on the microcav-
ity fabrication and characterization are given in the following sections.
We noted above that a potential route to maximizing Q for small pillar diameters is the
construction of a photonic crystal with a small refractive-index perturbation. As the per-
turbation gets smaller, the cavity mode becomes more delocalized in real space, and conse-
quently more localized in Fourier space. This, in turns, leads to reduction in the transverse
radiation loss. Furthermore, the cavity resonance can be located at lower frequencies,
1800
1600 MPT=10
MPT=11
1400 MPT=12
MPT=15
1200
1000
Q
800
600
400
200
0
250 300 350 400 450 500 550 600
Diameter on the bottom (nm)
Figure 2.7
Left: Electric field intensity for the fundamental mode in a pillar with realistic wall profile. Right: The effect of
vertical undercut on the Q-factor of the fundamental mode. For a pillar with a fixed etch depth (3535 nm, cor-
responding to etching through 21 mirror pairs and a spacer), with the diameter on top of 600 nm, and a straight
etch through the top nine mirror pairs, the Q-factor drops dramatically with an increase in undercut (i.e., with
the reduction of the pillar diameter on the bottom). Different curves correspond to different numbers of mirror
pairs on top of the spacer.
where the density of free-space radiation modes is smaller. In order to compensate for
the increased longitudinal loss, we need to put more MPT of these structures. We studied
pillars in the GaAs/AlxGa1 - xAs material system with lower index contrast (nh = 3.57 and
nl = 3.125) and showed that even though Q larger than 14000 can be achieved for D = 0.5
mm, V also increases, and the maximum Q/V ratio is similar to that calculated for the
GaAs/AlAs system. Furthermore, this Q/V ratio requires more top mirror pairs than in the
GaAs/AlAs system, thereby implying more challenging fabrication [16].
On the other hand, if we employ a material system with a high refractive-index contrast,
such as GaAs/AlOx (nh = 3.57 and nl = 1.515), a larger bandgap (and thus a better longitudi-
nal confinement of the cavity mode) could be achieved, but the mode suffers more radia-
tion loss in the transverse direction, which limits its Q-factor to values below 1000, even
with pillar diameters D of up to 1.3 mm [16].
2.2.2 Growth of QDs
There are several choices of optically active, discrete emitters in semiconductor systems,
including deep-level states such as color centers in diamond, quantum well (QW) interface
fluctuations, and colloidal and epitaxial QDs. It is possible, in principle, to integrate all of
these emitters in microcavities. One of the most advanced planar microcavity systems is
the epitaxial system utilizing groups III and V semiconductors such as GaAs and AlAs;
this system is compatible with epitaxial QDs and QW fluctuations. Epitaxial QDs generally
have deeper confining potentials and are well separated from the GaAs band edge. Even
though they have size variations, their shape and polarization are more predictable than
interface fluctuations.
Semiconductor crystal growth techniques, especially the molecular-beam epitaxy tech-
nique used here, can produce near monolayer control of the deposition thickness. This is
used to make the high-quality planar microcavity structures described in this chapter, as
well as QWs. A natural approach to fabricating atomic-like discrete emitters was to extend
this control used to make QWs to 2D and 3D confinement with quantum wires and QDs
through post-growth processing. These results have been encouraging, but not resound-
ing, because the lithography techniques used to provide lateral confinement typically do
not have the monolayer resolution that is available through epitaxial growth techniques.
Furthermore, because of interfacial damage, it is difficult to directly pattern active regions
using processing.
In homoepitaxial crystal growth (epitaxial layer and substrate are same material) of
cubic materials, the (100) growth surface is the lowest energy facet plane [19]. If the growth
temperature and flux rates are appropriate, adatom attachment on the growth surface is
to kink and ledge sites or flat island nucleation regions. Crystal growth thus proceeds by
the lateral growth of kinks and ledges, or by the expansion of flat, monolayer-scale high
islands. Ideally, as one monolayer is filled, new, monolayer-high nucleation sites are cre-
ated, and the 2D (100) growth surface propagates. In contrast, during heterogeneous crystal
growth (epitaxial layer and substrate are different materials), as more adatoms are depos-
ited onto the growing surface, this growth surface can go through structural changes that
can be as dramatic as the complete loss of epitaxial coherency, or as small as changes in
surface reconstruction, surface roughness, or the abruptness of a heterointerface [20–23].
In the case of InAs on GaAs, the lattice mismatch is 7.2%. The critical thickness for relaxa-
tion by dislocation generation is expected to be between a few monolayers [24] and 15
monolayers (4.5 nm) [25], depending on the theoretical model and the crystal growth con-
ditions. Because of the similar InAs and GaAs crystal structures, at least one chemisorbed
monolayer of InAs can be stable on the GaAs substrate. Below the critical thickness, or after
a chemisorbed layer, a metastable phase can exist. This is phenomenologically known as
the Stranski–Krastanow (SK) growth regime [26]. In this growth regime, the lattice-mis-
matched growth produces surface islands. The islands, which become the QDs, increase
the total energy with respect to the bulk crystal, because they create additional interface
surface area. However, the islanding surface allows for partial strain energy relaxation due
to the surface curvature and distance from the hetero-interface. This growth regime is a
transitional growth mode between the compliant, planar growth regime that characterizes
ideal MBE growth, and plastically relaxed growth, since, as islands grow and merge, the
surface area can no longer expand to accommodate the increasing strain energy.
Since the early 1990s, this technique has been used to make QDs in the InAs/GaAs and
Ge/Si semiconductor systems by overgrowing the island layer with the host semiconduc-
tor [27]. The resulting QDs are narrow-bandgap regions commensurate with a host crystal
and, for InAs, contain on the order of 20,000–50,000 atoms. The electron and hole potential
confinement and small size are such that discrete electron and hole many-body states can
be observed with transitions dominated by radiative processes.
This QD fabrication process produces random arrays of islands with slightly varying
sizes, and ensembles which are either ordered or containing identical QDs have not been
demonstrated to date. This is due to the secondary role of the energies that drive the QD
formation process in epitaxial systems. The nonuniform QD size and shape lead to a nonu-
niform QD spectral distribution. By making dilute QD ensembles and accessing single
QDs through lithographic patterning, the ensemble broadening can be an advantage,
because single-QD states can be spectrally distinguished.
For MBE growth on planar or vicinal surfaces, the InAs QD size and density are controlled
predominantly by the amount of InAs deposition and the surface temperature [28]. Other
process controls that also contribute are the InAs growth rate and the V/III flux ratio [29].
All of these parameters control the kinetic limitations to the islanding process, including In
adatom surface diffusion, In desorption, and alloy mixing. The InAs QDs used for the stud-
ies described in this chapter were deposited in a MBE system using an As2 flux that is pro-
duced by thermally decomposing the dominant vapor species, As4, present above heated
solid arsenic. In addition, the As source has an adjustable valve, so that the V/III flux ratio
can be precisely controlled. The V/III beam equivalent pressure ratio is kept between 8 and
20 for all layers. The GaAs growth rate was kept low, between 0.1 and 0.2 µm/hr, providing
long In adatom surface diffusion lengths and, thus, a more equilibrium QD distribution.
The amount of material deposited has a significant impact on the QD distribution. Below
approximately 1.8 monolayer (mls) of InAs, the surface is rough, but there is no island
formation, since the lattice mismatch strain is accommodated by InAs lattice compression
[28,30]. This region is called the wetting layer.
At a surface temperature of 480°C, with 2 mls of deposited InAs, the average island
diameter is 15 nm. When the equivalent of three planar mls of InAs have been deposited,
the additional material is accommodated by an increase in the island size and density.
The island size increases because more material has been deposited and migrates to the
islands. The increase in the island density increases as more InAs islands nucleate [28].
Between 1.8 and 3 mls the QD density changes from 0 to 1000 mm-2. In order to incorporate
single QDs into pillar microcavities, a density of less than 100 mm - 2 is desired and less than
10 mm - 2 is best.
When an additional 1 ml of InAs is deposited, so that the total InAs deposited is 4 mls,
the islands begin to coalesce and this process no longer provides effective bulk strain
relaxation through surface curvature. Plastic relaxation through dislocation generation
begins to occur. Thus, there is a narrow range of InAs deposition between slightly larger
than 1.8 mls (0.5 nm) to approximately 3 mls (0.9 nm) where useful QDs can be made.
Furthermore, there is a correlation between the density of QDs and the average QD size:
the greater the density, the larger the size.
A cross-sectional scanning tunneling microscopy (XSTM) image obtained by cleaving
the QD sample in vacuum is shown in Figure 2.8. The cleaved surface is (110), and in this
image tunneling occurs between the tip and empty conduction band states [31]. The QDs
are covered with epitaxial GaAs. While the QDs nucleate on a planar wetting layer region,
the image indicates that the wetting layer diffuses throughout the QD region, so that the
QDs lie within a rough QW region formed by the wetting layer. This is most likely the result
of strain-enhanced diffusion. Diffusion of In away from the QDs appears to be limited.
The surface temperature affects the QD size and density through the surface diffusion
and desorption rates. Below 460°C, the In adatom surface diffusion is limited and island
formation is inhibited. Conversely, with our low growth rates, by 540°C, the desorption
rate of In is larger than the In input to the surface and no QDs are present. However, at
temperatures above 500°C, alloying with the surrounding GaAs occurs and, coupled with
desorption, can lead to a desired arrangement, where large QD sizes can be made with low
density. This is the technique used here, where the amount of deposited InAs was varied
from 1.9 to 2.2 mls, and the growth temperature was between 510 and 530°C.
By this method, we produced QDs with densities ranging from ∼10 to 100 µm–2, allow-
ing isolation of a small number of QDs inside of pillar structures. At the same time,
the photoluminescence wavelengths ranged from 850 to 950 nm, making quantum
optics experiments much easier, since silicon avalanche-photodiode photon counters,
with high efficiency and low dark counts, could be used. Measurements performed to
determine other properties of these QDs are described below, but here we summarize
their main properties. The radiative lifetimes usually varied from ∼0.5 to 1.3 ns. If all of
the electron–hole recombination is radiative, this corresponds to an oscillator strength
fosc ∼ 2.6–6.7 (see Equation 2.3). However, it is difficult experimentally to determine the
radiative efficiency except when the photoluminescence lifetime is observed to decrease
due to microcavity coupling. In this case a radiative efficiency into the cavity mode can
be accurately estimated, and as discussed below, this efficiency can be quite high. The
homogeneous linewidth (the linewidth of a single QD) is sample-dependent, and prob-
ably is affected by factors such as the etching process and the optical excitation method.
Using an interferometric measurement (described below) we have observed coherence
lengths up to 275 ps, corresponding to a homeogeneous linewidth of 1.16 GHz, or 0.003
nm. To observe such a linewidth, measurements must be performed at liquid-helium
temperatures. There is little temperature dependence below 10 K, but by 50 K broaden-
ing and decreased photoluminescence efficiency can typically be observed with a grat-
ing spectrometer (0.02 nm resolution). The lowest-energy exciton level (one electron–hole
pair) consists of four states due to the four possible spin combinations for one electron
Figure 2.8
Cross-sectional scanning-tunneling-microscope image of a vacuum-cleaved QD sample. The cleave plane is
(110). Tunneling is to empty band states. The wetting layer is observed to diffuse in the growth direction and
throughout the QD region. Diffusion from the disk-shaped QDs is limited. (From Wu, W., J. R., Tucker, G. S.,
Solomon, and Harris. J. S., Appl. Phys. Lett., 71 (1997), 1083–1085.)
(spin ± 1/2) and one heavy hole (spin ± 3/2). Two of these states, the “bright” excitons,
have allowed radiative recombination, and are degenerate in cubic symmetry. Their
dipole moments for spontaneous emission to the empty QD state are polarized orthogo-
nally to each other, and parallel to the growth plane. In actual QDs, asymmetry typically
produces an energy splitting of these two bright-exciton states into components with
orthogonal, linear polarizations along special axes determined by the asymmetry [32].
In the interferometric measurements we have observed splittings ranging from ~10 to 50
meV, or 2.4 to 12 GHz. The inhomogeneous linewidth (ensemble linewidth determined by
size variations among the QDs) is four orders of magnitude larger than the homogeneous
linewidth and is ~30 nm.
1400
1200
1000
Quality factor
800
600
400
200
0 1 2 3 4 5 6 7
Diameter (µm)
Figure 2.9
Quality factor of the fundamental modes of first-generation pillar microcavities. Points indicate experimentally
measured values, and solid lines indicate calculated values.
Figure 2.10
Electric field intensity for the fundamental mode of (a) an ideal and (b) a realistic first-generation pillar micro-
cavity, each with a top diameter of 0.5 µm, calculated by the finite-difference time-domain method. The intensity
is represented by a grey scale. Half a longitudinal cross-section of each post is shown. Lines represent interfaces
between different materials. (From Pelton, M., Vučkovic’, J., Solomon, G. S., Scherer, A., and Yamamoto, Y., IEEE
J. Quantum Electronics, 38 (2002), 170–177.)
Figure 2.14a. This, in turn, was achieved by lowering the concentrations of Cl 2 and BCl3,
increasing the applied rf power, and decreasing the process pressure, thereby reducing
the chemical component and increasing the physical component of the etch. As well, the
etch process was divided into three stages; in each subsequent stage, the partial pres-
sure of Cl2 and the process pressure were decreased, while the applied RF power was
increased to maintain a roughly constant DC bias. Finally, sample heating was reduced
by improving heat transfer between the sample and its holder and by cooling the holder
to approximately 3°C.
An effect of these changes in the etch process, though, is to make the etch less selective
with respect to the mask material, so that the gold masks used in the first-generation proc-
ess are completely eroded long before the etch is complete. The gold is therefore replaced
with nickel, which is more resistant to plasma etching [37], and the metal thickness is
increased. This, in turn, requires the use of a bilayer electron-beam resist, consisting of a
top layer of high-molecular weight PMMA and a lower layer of a co-polymer of MMA and
methacrylic acid (P(MMA-MAA)) [38]. Exposure and development of this resist leads to an
undercut profile, allowing for the liftoff of Ni layers up to 100 nm thick and with feature
sizes down to 150 nm.
Figure 2.11 shows a scanning-electron microscope image of a pillar microcavity pro-
duced with this second-generation process. The etch proceeds far into the lower DBR
stack. Undercutting is still present, but is greatly reduced compared to the first-generation
posts, and occurs mostly below the QD active layer. Light in the second-generation posts
thus does not experience significant diffraction loss in the lower DBR, and measured qual-
ity factors for small pillar diameters are higher, as shown in Figure 2.12.
Despite the improvement, the quality factor still degrades as the post diameter decreases.
This has previously been attributed to scattering by roughness at the post sidewalls [39],
but FDTD calculations suggest that the loss is actually due to the tapered shape of the
posts. The taper leads to coupling between the fundamental confined mode and other
modes, including higher-order waveguide modes and radiation modes. The quality fac-
tors of the pillar microcavity modes could therefore still be increased by further improv-
ing the etch process.
Figure 2.11
Scanning-electron microscope image of a second-generation pillar microcavity with a top diameter of 0.6
µm and a height of 4.2 µm. (From Pelton, M., Santori, C., Vučkovic’, J., Zhang, B., Solomon, G. S., Plant, J. and
Yamamoto, Y., Phys. Rev. Lett., 89 (2002), 233602. © American Physical Society. With permission.)
1300
1200
1100
1000
Quality factor 900
800
700
600 Second generation posts
First generation posts
500
400
300
200
0 1 2 3 4 5
Post diameter (µm)
Figure 2.12
Measured quality factors for first-generation and second-generation pillar microcavities.
1µm
Figure 2.13
Scanning electron micrograph showing a fabricated array of third-generation pillar microcavities.
Γ if = (2 π / )ρ(ω c ) | b | H | a |2 ,
where ρ(wc) is the photon field density of states at the transition frequency wc , |b〉 is
the final state, |a〉 is the initial state, and H is the atom–vacuum field dipole interaction
Hamiltonian.
Fermi’s golden rule describes the weak-coupling regime, where the atomic excitation is
irreversibly lost to the field. The spontaneous emission decay rate into a particular cavity
mode is enhanced by increasing ρ(wc) at this cavity resonance. The enhancement factor is
simply the ratio of the density of field states at the atomic frequency inside the cavity to
that in free space [34]. For a Lorentzian emission line and a Lorentzian cavity mode reso-
nance the enhancement is given by the Purcell factor, F, in Equation 2.6. This equation can
be rewritten as
3Qλ 3i ∆λ 0 ( ∆λ 0 + ∆λ i )
F= cos 2 θ,
4π V 4( λ 0 - λ i )2 + ( ∆λ 0 + ∆λ i )2
2
where Q is the cavity quality factor, and V is the cavity mode volume. The second term
is the convolution of the Lorentizans of the cavity mode and emitter emission, with ∆λ0
and ∆λi the linewidths of the cavity and emitter, at center wavelengths λ0 and λi, respec-
tively. This term represents deturning of the emitter and cavity and accounts for varying
linewidths. The third term accounts for the relative angle θ between the emitter dipole
and cavity mode field. We assume no cavity mode degeneracy. For the ideal case where a
nondegenerate emission state with a negligible linewidth is perfectly dipole-aligned and
on resonance at the antinode of an ideal cavity, also with negligible linewidth, this reduces
to 3Qλ3/4π2V. Thus, at a certain wavelength, one would like a cavity with a high Q and a
small V, in order to achieve large enhancement. As the modeling described above has dem-
onstrated, pillar microcavities are capable of achieving these conditions simultaneously,
and are therefore well suited for spontaneous-emission enhancement.
(a) (b)
0.6
0.5
Intensity (a.u.)
Lifetime (ns)
top
DBR
0.4
QDs
0.3
bottom
DBR
894 896 898
1 µm Wavelength (nm)
Figure 2.14
A single quantum dot (QD) post microcavity. (a) Scanning-electron microscope image of a first-generation pillar
microcavity. The diameter at the top of the post is 0.5 µm. (b) A PL spectrum (solid) and the spectral dependence
of the spontaneous emission lifetime (filled circles: experimental, dashed line: theory) at 4 K of a tapered post
microcavity with a less than 0.5 µm cavity diameter. (From Solomon, G. S., Pelton, M., and Yamamoto, Y., Phys.
Rev. Lett., 86 (2001) 3903–3906. © American Physical Society. With permission.)
60
50
Intensity (a.u.)
40
Energy (meV)
30
20
1.330 1.334 1.339 1.343
Energy (eV)
10
0
0 1 2 3 4 5 6
Diameter (µm)
Figure 2.15
Measured and calculated energy shifts of mode centers as a function of post diameter for microcavity posts
with QD active regions. Measured values determined from 4 K PL, calculations assuming the fields are separa-
ble into transverse and lateral components. The insert shows the 4 K luminescence from a 5 µm diameter post,
where the confined modes are indicated by arrows. (From Solomon, G. S., Pelton, M., and Yamamoto, Y., Phys.
Rev. Lett., 86 (2001) 3903–3906. © American Physical Society. With permission.)
Without a cavity, the QDs have a spontaneous emission lifetime of 1.3 ns, and with the
planar microcavity this is reduced to 1.1 ns, as shown in Figure 2.16a. In the planar micro-
cavity case, there is a continuous distribution of cavity resonant modes from the cut-off
wavelength, λ c = s 932 nm to a wavelength corresponding to the stopband edge. Since
the number of modes increases with decreasing wavelength from the cutoff wavelength,
λc, the spontaneous emission lifetime is continually decreased from the cutoff wave-
length to shorter wavelengths. For the 3D post microcavities, the modes are discrete, so
that the minimum spontaneous emission lifetime occurs at the center of each mode. This
1.5
1.25
Lifetime, τ (ns)
Intensity (a.u.)
1.20 1.2
1.15
0.9
Cavity
1.10 resonance
0.6
1.05
931.8 932.0 932.2 928 930 0 0.5 1.0 1.5
Wavelength (nm) Wavelength (nm) Time (ns)
Figure 2.16
Spontaneous emission lifetime of QDs in microcavity posts. The average QD lifetime at spectral positions in
the planar cavity is shown in (a), and for a post diameter of 2.5 µm in (b). The intensity is also shown in (b). In
(c) the time-dependent spontaneous emission is shown for InAs QDs in bulk GaAs and a single InAs QD reso-
nant with the fundamental mode of a 0.5-µm diameter microcavity post. (From Solomon, G. S., Pelton, M., and
Yamamoto, Y., Phys. Rev. Lett., 86 (2001) 3903–3906. © American Physical Society. With permission.)
unmistakable mark of the 3D confinement is seen in Figure 2.16b for the microcavity with
a post diameter of 2.5 mm and also in Figure 2.14b for a post diameter of 0.5 mm. For the 2.5
mm diameter post microcavity, a continuous distribution of QDs emit photons into the sin-
gle cavity mode. The reduction of spontaneous emission lifetime is largest when the QDs
are on resonance, but the modification disappears off-resonance, near the mode edge. In
Figure 2.16c the time-dependent spontaneous emission is shown for a single QD in reso-
nance with a 0.5 mm diameter microcavity post (τ = 0.28 ns), measured by a streak camera
after the pulsed excitation from a 200 fs Ti:Al2O3 laser. The rise time (≈ 50 ps) is determined
by the carrier capturing time by the QD and the decay time (≈ 0.30 ns) is the radiative
lifetime. The spontaneous emission lifetime at the center of each fundamental mode is
plotted against the post diameter in Figure 2.17, together with the theoretical prediction.
The reduced spontaneous emission lifetime in a small post microcavity is primarily due
to the reduced mode volume. This offsets the decrease in Q, which is approximately 300 in
the 0.5 µm diameter post.
In the above analysis, we neglected the spatial dependence of the electric field intensity,
|E(r)|2, inside the
micropost cavity. Because the QDs are not identical, it is likely there
is variation in ℘ E(r). One might suppose that the spectral QD emission distribution is
coincidentally the same as the spatial QD distribution; that is, QDs detuned spectrally
from the cavity are also not in the center of the post, and thus the QDs happen to be spa-
tially detuned from the electric field maximum the same way they spectrally detuned.
This would still be a cavity QED effect, but would be related to ℘ E(r) and not Q/V.
Because the QD ensemble is random, such ordering 2 is unlikely. As well, the good agree-
ment between theory assuming
a constant |℘⋅ E | and experiment (Figure 2.14) justifies
neglecting variations in |℘⋅ E |2. The theory contains no free parameters; the cavity Q used
in the theory is obtained independently from decay rate measurements, using the filtered
spontaneous emission under high pump conditions. Coincidental alignment of |℘ E(r)|2
and the spectral detuning for each QD will not satisfy our criterion.
0.8
0.7
Lifetime (ns)
0.6
0.5
0.4
0.3
0.2
0 1 2 3 4 5 6
Post diameter (µm)
Figure 2.17
The spontaneous emission lifetime (•) at the center of the fundamental mode is plotted as a function of the post
diameter. The lifetime for the 0.5 µm diameter post is for a single QD. Lines are the result of an approximate
model for QDs at the center of the disk (dashed) or averaged radially over the disk (solid).
Intensity (a.u.)
Figure 2.18
The 4 K photoluminescence spectrum of a micropost cavity, where the ground state excitonic emission from
two QDs is indicated by arrows.
0 ns
4K
Time (ns)
QD2
2.2 ns QD1
0 ns
50K
Time (ns)
QD2
2.2 ns
910 908 906 904 902 900
Wavelength (nm)
Figure 2.19
Streak camera image of temporal and spectral distribution from a 0.6 µm diameter micropost cavity. The ground
state excitonic emission from a few QDs is observed. The QDs of interest are labeled. Upper: 4 K sample tem-
perature; lower: 50 K sample temperature.
the observed lifetime changes. The detuning curve is found by measuring the detuning of
each QD exciton state from the cavity mode. The enhancement for the two QDs are deter-
mined by Lorentzian fits, and are similar: QD1 = 4.1, and QD2 = 3.9. Several factors could
account for the small difference. Since the spatial position of each QD is unknown, the field
5 5
QD1
QD2
4 4
3
3
2
2
1
QD1 1
QD2
0
0 10 20 30 40 50 60 70 –2 –1 0 1 2 3
Temperature (K) QD detuning (nm)
Figure 2.20
Spontaneous emission decay rate measured for two QDs as a function of sample temperature on the left and
spectral detuning on the right. The detuning is the difference in wavelength from each QD state and the cavity
mode.
amplitude at each QD is unknown, but could vary. The dipole strength and orientation of
each QD could also vary. The fitted linewidths of the two Lorentzians, on the other hand,
are the same within the errors, since they couple to the same cavity. The corresponding
quality factor is 775 ± 160, in agreement with the results shown in Figure 2.12.
1 6
Q=1270
0.8 5
PL intensity (a.u.)
4
0.6
1/τ (1/ns)
0.7 nm
3
0.4
2
0.2
1
0
918 920 922 0 0.5 1 1.5 2
λ (nm) |λQD−λc| (nm)
0.8
0.6
PL intensity (a.u.)
τoff = 976 ps
0.4
0.2
τon=188 ps
0
0 0.5 1 1.5 2
Time (ns)
Figure 2.21
Top-left: Background emission filtered by a pillar cavity. Top-right: Decay rate of the emission line as a function
of the absolute value of its detuning from the cavity resonance (|λQD − λc|). The dot emission wavelength was
tuned by changing the sample temperature within the 6–40 K range. Bottom: time-dependent photolumines-
cence from the emission line on-resonance with the cavity, as opposed to this same emission line off-resonance.
(Reprinted with permission from Vučkovic’, J., Fattal, D., Santori, C., Solomon G., and Yamamoto, Y., Appl. Phys.
Lett., 82 (2003) 3597. © American Institute of Physics. With permission.)
fivefold spontaneous emission rate enhancement (Purcell factor F) is observed for the dot
coupled to a cavity, as opposed to the dot off of resonance (top-right plot of Figure 2.21).
pm 1 - (1 + n )e - n . (2.10)
n (n - 1)
g 0( 2 ) = 2
. (2.11)
n
It is straightforward to show from Equation 2.11 that for any possible photon number dis-
(2) 2
tribution, pm ≤ (1/2) g 0 n , as noted in the Introduction [2].
In this section, we first discuss the mechanism for multi-photon suppression in optically
excited QDs. We then show experimental results for QDs in pillar microcavities, including
a multi-photon suppression factor up to 50 compared with an equivalent Poisson source.
We also describe some unexpected blinking and long-timescale antibunching effects we
have observed, which are related to the optical excitation method.
1’’ 2 1 1’ 32 1 32 1
0
0.5
Time (ns)
1.5
Figure 2.22
Streak camera images of emission from a single QD in a 0.2 µm mesa under pulsed, 708 µm laser excitation with
powers (a) 27 µW, (b) 108 µW, and (c) 432 µW. For larger powers, multi-excitonic emission (lines 2, 3) occurs first,
followed by single-exciton emission (line 1). (From Santori, C., Solomon, G. S., Pelton, M., and Yamamoto, Y.,
Phys. Rev. B, 65 (2002), 073310. © American Physical Society.)
The degree of two-photon suppression that can be obtained by this method also depends
on the excitation method. The main issue is that, if carriers are injected too slowly, it is pos-
sible for a second electron–hole pair to be injected after a first exciton has recombined, and
in this case two photons will be emitted at the single-exciton frequency. If we consider only
a two-level system, a simple rate-equation model for incoherent excitation predicts that the
degree of two-photon supression (the two-photon probability of this device divided by
that for a Poisson source) is approximately Γtlaser, where G is the single-exciton spontane-
ous emission rate, and tlaser is the excitation pulse duration. If an excitation laser pulse is
used to excite above the GaAs bandgap, then, even if the pulse is very short, the effective
tlaser can be on the order of 50 ps, due to the carrier capture process. In this case, the degree
of two-photon suppression is limited to a factor of approximately 10. This becomes worse
when a microcavity is used to shorten the spontaneous emission lifetime. We have found
that by tuning the excitation laser on resonance with a higher-energy transition within
the QD, much better performance is possible. In this case, electron–hole pairs are created
directly inside the QD. This excitation scheme is shown in Figure 2.23a. Initially, the laser
excites the QD to an excited level X*. Experimentally, such a resonance is found through
a photoluminescence excitation (PLE) measurement, in which the intensity of the single-
exciton ground-state emission is measured as a function of laser frequency [53]. In our
samples, such resonances could typically be found at a wavelength approximately 20 nm
shorter than that of the ground-state single-exciton transition. The transition from X* to the
lowest-energy exciton state X is apparently very fast, on the order to 10 ps or less, followed
by emission of a single-photon from the X transition. It is possible sometimes for the laser
to create a second electron–hole pair, raising the system to level XX*; in this case, spectral
filtering is required to reject the biexciton photon. For these experiments, we used a mod-
elocked Ti–Sapphire laser with ~2 ps pulse duration and corresponding linewidth below
0.5 nm. If a biexcitonic energy shift of a few meV is also present in the excited transition
(X* to XX*) [53], we have an additional mechanism for two-photon suppression, since the
(a) (b)
Excitation Relaxation
XX*
XX
α
X* X X
X
β
(Emitted r Detect
photon) α e–, h+ e–, h+
0 0 0
Figure 2.23
(a) Level diagram for resonant excitation scheme. (b) Diagram including single carrier injection pathways.
detuning of the laser from the XX* transition is substantially larger than the laser linew-
idth [9,54].
τ = t2 – t1 Coincidence
Computer
electronics
t2
Cryostat
APD
t1
φ λ
APD
Laser
Figure 2.24
Schematic diagram of photon correlation setup. φ: adjustable half-wave plate followed by a polarizer. λ: spectral
filter. (From Santori, C., Fattal, D., Vučkovic’, J., Solomon, G. S., Waks, E., and Yamamoto, Y., Phys. Rev. B, 69 (2004),
205324. © American Physical Society.)
1100 8
1000 6
900
counts
4
800 2
700 0
−5 0 5
600
Counts
τ (ns)
500
400
300
200
100
0
−50 0 50 100 150
τ (ns)
Figure 2.25
Histogram of the relative delay τ = t2–t1 for photons detected by counters 2 and 1, obtained from a QD coupled
to a DBR pillar microcavity. The inset shows the small peak at τ = 0, barely detectable above the background,
indicating a large two-photon suppression factor. (Reprinted with permission from Vučkovic’, J., Fattal, D.,
Santori, C., Solomon, G., and Yamamoto, Y., Appl. Phys. Lett., 82 (2003) 3598. © American Institute of Physics.
With permission.)
distributed source. This high degree of two-photon suppression is due partly to the nar-
row (0.1 nm) spectral filter used in this measurement. For other devices measured on this
sample, the two-photon probability was typically suppressed by a factor of 10–20.
Another feature in the data which must be explained is the increase of the peak areas
around τ = 0. Such behavior suggests a “blinking” of the QD photon source: given that we
detect a photon in pulse j, the probability of detecting another photon in the next pulse
j + 1 is substantially higher than the probability of detecting a photon at a much later time.
We observe this behavior for QDs under resonant excitation [9,4,55]. A variety of memory
effects have been reported in the optical emission of single semiconductor QDs. Some
occur on millisecond or longer timescales, including blinking [56–58], two-color blinking
(a) (b)
50 µW
200 µW
2 500 µW
1200 µW
1
Normalized peak area
1
200 kcps
100 kcps
50 kcps
25 kcps 0
0
0 5 10 0 5 10
g(2) Peak number g(2) Peak number
Figure 2.26
Normalized peak areas from photon correlation histograms plotted versus peak number (circles), and two-
sided exponential fits (lines) using Equation 2.12. (a) Measurement of a single-photon device using above-band
excitation, measured at four different excitation powers, resulting in the indicated count rates on one detector
(shifted vertically for clarity). (b) Results for the device from Figure 2.25 under resonant excitation at four dif-
ferent excitation powers as indicated.
[59], and spectral diffusion [60,61]. However, the blinking observed here occurs on much
faster timescales, from less than 10 ns to at least 800 ns. We have also observed a second
type of memory effect using weak above-band excitation, in which the peaks in the photon
correlation measurement increase slowly away from τ = 0 [55,62,63].
Several photon correlation measurements from single-photon devices under different
excitation conditions are shown in Figure 2.26. In this case, only the normalized peak areas
Am are plotted versus peak number, with peak 0 corresponding to τ = 0. The normalization
corresponds approximately to dividing the raw peak area by the expected peak area for
a Poisson-distributed source, based on the measured singles count rates, with small cor-
rections for detector dark counts. In all cases, the peak areas are well fitted by a simple
two-sided exponential function:
where g1 and τb are fitting parameters that characterize the amplitude and timescale of the
memory effect, respectively. The experimentally observed sign of g1 depends on whether
we use resonant or above-band excitation.
These memory effects are most likely caused by charge fluctuation of the QD. Charge
fluctuation can explain both the blinking (resonant excitation) and the long-timescale
antibunching (above-band excitation) in terms of the level diagram in Figure 2.23b. These
behaviors have simple qualitative explanations. When a first photon is detected from
pulse j, the QD is empty immediately afterward. For single-carrier injection, which is
possible under above-band excitation, two injections must occur before another photon
can be emitted, and thus it is unlikely that another photon will be emitted from pulse
j + 1 if the injection rate is small. The opposite argument applies for the injection of entire
electron–hole pairs (resonant excitation). If a photon is detected from pulse j, it is espe-
cially likely that another photon will be emitted from pulse j + 1, since only a single
additional pair needs to be injected. At much later times, the QD will sometimes be
charged, in which case it will appear dark at the particular wavelength being detected.
It is possible to describe this behavior with a simple probabilistic model, which provides
adequate fits to our data [55].
It may seem surprising that the charge of a QD should ever change under resonant exci-
tation, for which the excitation energy is tuned below the GaAs bandgap and below the
InAs wetting layer band edge. Perhaps it could be related to the “wettinglayer tail” observed
in PLE spectra and attributed to continuum states associated with the combined wetting
layer-QD system [64], or to Auger processes, which would allow an electron to escape from
the QD. For a single-photon device, this charge fluctuation means a factor of approximately 2
reduction in total efficiency, which is a problem for some applications.
2.3.3 Efficiency
In the absence of a microcavity, few of the photons emitted by a single QD escape from
the high-refractive-index semiconductor substrate containing the dot, and the fraction of
photons that can be collected is very low. If the QD is in a pillar microcavity, on the other
hand, the majority of the photons can be emitted into a single cavity mode, and the major-
ity of those photons can subsequently escape from the cavity in a free-space mode with a
nearly Gaussian spatial profile. A single QD in a pillar microcavity can thus serve as an
efficient source of single-photons.
We demonstrated efficient single-photon generation using the second-generation micro-
cavity structures [65]. We selected a particular post with a top diameter of 0.6 mm that
exhibits a clear single-dot emission line, indicated by the arrow in Figure 2.27. The meas-
ured photon–photon correlation function for an average pump power of 10.9 mW is shown
in Figure 2.28. The side peaks are separated by the laser repetition rate and all have nearly
equal areas, while the central peak is nearly absent, reflecting strong suppression of the
multi-photon probability.
Due to a low emission rate from QDs in this particular sample, adjacent peaks overlap,
and it is necessary to fit the data in order to determine the area of the central peak. In order
to reduce the number of free parameters in the fit, we measured the recombination rate
of the QD as 1/Γ = 2.2 ± 0.6 ns, and the response time of the autocorrelation setup, using
pump-laser light that is scattered off of the pillar, as τresp = 473 ± 29 ps. Each peak in the
photon correlation data from the QD is thus described by a two-sided exponential with a
decay constant τtotal ≈ 1/Γ + τresp = 2.7 ± 0.7 ns. Figure 2.28 shows the result of the fit, which
involves only two free parameters: the area of the central peak and the area of all the other
peaks. The ratio of these two areas provides an estimate of g 0( 2 ) and gives an upper bound
on the probability that two or more photons are present in the same pulse, as explained
above.
Results for various pump powers are summarized in Figure 2.29. Normalized central
peak areas as low as 0.10 were obtained, reflecting a tenfold reduction in the multi-photon
probability as compared to a source with Poissonian statistics. The multi-photon prob-
ability increases with pump power, suggesting that other states, apart from the desired
QD emission, are being excited by the pump laser and are contributing a background of
unregulated photons. As explained above, resonant excitation and spectral filtering can
greatly decrease this background.
1600
1400
1200
Intensity (a.u.)
1000
800
600
400
200
0
840 850 860 870
Wavelength (nm)
Figure 2.27
Photoluminescence from a single QD in a pillar microcavity, for excitation with photon energy above the GaAs
bandgap.
60
Coincidences
30
0
–40 0 40
Delay (ns)
Figure 2.28
Measured autocorrelation for photons from a single QD in a pillar microcavity with an incident pump power
of 10.9 µW (points), and corresponding fit (line). The measurement integration time is 600 sec. (From Pelton, M.,
Santori, C., Vučkovic’, J., Zhang, B., Solomon, G.S., Plant, J., and Yamamoto, Y., Phys. Rev. Lett., 89 (2002), 233602.
© American Physical Society.)
In order to measure the efficiency of our single-photon source, we calibrate the detec-
tion efficiency of the measurement apparatus, by again scattering pump-laser light off the
pillar. The scattered power is measured immediately after the collection lens, and is com-
pared to the total photon count rate at the detectors, yielding an overall detection efficiency
of 3.02 ± 0.16%. The collection lens in front of the sample, in turn, captures only part of the
light that escapes from the post; this capture fraction can be estimated by calculating the
effective divergence angle of light emerging from the post. Both the approximate method
0.30
0.25
0.20
(2)
0
g
0.15
0.10
0 10 20 30 40 50 60
Average pump power (µW)
Figure 2.29
Normalized area of the central autocorrelation peak for photons from a QD in a pillar microcavity as a func-
tion of pump power (points), as well as a linear fit (line), as a guide for the eye. (From Pelton, M., Santori, C.,
Vučkovic’, J., Zhang, B., Solomon, G.S., Plant, J., and Yamamoto, Y., Phys. Rev. Lett., 89 (2002), 233602. © American
Physical Society.)
outlined above and rigorous FDTD simulations indicate that the lens used, which has a
numerical aperture of 0.5, collects approximately 22% of the light in the emitted beam.
Using these values for the efficiencies of the collection lens and the measurement appara-
tus, then, it is possible to determine the mean photon number per pulse, 〈n〉, by comparing
the measured photon count rate and the measured laser repetition rate.
The efficiency, ηtotal, of the source is then determined by assuming that the emitted light
consists of a statistical mixture of perfectly antibunched single-photons and a background
of photons with Poissonian statistics. The coupling of this state into the traveling-wave
mode leaving the top of the pillar microcavity is modeled as an attenuation by a linear
loss, giving
ηtotal = n 1 - g 0( 2 ) . (2.13)
Results are shown in Figure 2.30. The solid line is a fit according to the saturation
equation
where P is the pump power, Psat is the saturation pump power, and ηmax is the saturated
efficiency. The maximum efficiency according to this fit is 37.6 ± 1.1%. This number repre-
sents the external quantum efficiency of the device, and is approximately two orders of
magnitude higher than the external efficiency for a QD in bulk GaAs.
The external efficiency is limited by two factors (see Equation 2.7). The first is the coupling
coefficient, ηqd,cav, or the fraction of photons emitted by the QD that are captured by a single
cavity mode. It depends on the cavity-enhanced emission rate Γ and the emission rate Γ0 in
40
30
25
20
15
0 20 40 60 80 100
Pump power (µm)
Figure 2.30
Measured efficiency of the single-photon source as a function of pump power (points), together with a satura-
tion fit (line). (From Pelton, M., Santori, C., Vučkovic’, J., Zhang, B., Solomon, G. S., Plant, P., and Yamamoto, Y.,
Phys. Rev. Lett., 89 (2002), 233602. © American Physical Society.)
the absence of a cavity: ηqd,cav = 1 - (Γ0 - Γc)/Γ, where Γc/Γ0 is the fraction of light that would be
coupled into the cavity mode in the limit Q → 0. In order to determine Γ0, we measured the
recombination lifetimes for excitons in other QDs on the same sample that are out of reso-
nance with the cavity mode. The measured off-resonant lifetime is 25.4 ± 1.4 ns, correspond-
ing to a coupling coefficient ηqd,cav = 83 ± 23%.
The second factor in determining device efficiency is the light extraction efficiency,
ηcav,out, or the fraction of light in the cavity that escapes into a single traveling-wave mode.
This, in turn, can be determined by comparing the measured quality factor, Q = 628 ± 69,
for the pillar microcavity and the measured quality factor, Q 0 = 1718 ± 13, for the planar
microcavity. Q 0 is almost entirely determined by the leakage rate through the top DBR;
assuming that this rate is unchanged when the cavity is etched into posts, any difference
between Q and Q 0 will be due to etch-induced loss. The extraction efficiency is then sim-
ply given by ηcav,out = Q/Q 0 = 36.6 ± 4.0%. Combining the estimated ηqd,cav and ηcav,out gives an
expected external quantum efficiency of ηtotal = 30 ± 9%. This agrees, within the error, with
the measured efficiency at saturation.
emitter, the emitter must therefore be excited in a deterministic way and interact little with
its surrounding environment while the photon is emitted.
Here, we describe an experiment testing the indistinguishability of successive photons
emitted by the same semiconductor QD in a DBR microcavity [4,70]. The experiment is
based on “two-photon interference”: when identical single-photons enter a 50–50 beam-
splitter from opposite sides, quantum mechanics predicts that both must exit in the same
direction if their wave packets overlap perfectly, an effect which can be attributed to the
Bose–Einstein statistics of photons. This effect was first demonstrated experimentally in
the Hong–Ou–Mandel experiment [71], using pairs of highly correlated photons produced
by parametric downcoversion. Theoretically, this behavior is also expected for single,
independently generated photons [72], but, until recently, sources of single-photons have
not had the efficiency or spectral characteristics required to realize such an experiment.
Besides the demonstration described here, two-photon interference has now also been
demonstrated using photons from single atoms [73] and from QDs in photonic crystal
cavitities [74]. Recently two-photon interference has been observed using photons from
two separate trapped ions [75], a major advance for quantum networks.
For this experiment, we used the third generation sample described above, cooled to 3–7
K. Many pillars with only one or two QDs on resonance with a fundamental cavity mode
were found. To generate single-photons, we focused 3 ps Ti–Sapphire laser pulses every
13 ns onto these pillars from a steep angle. The laser was tuned to an excited-state absorp-
tion resonance of the QD. The QD emission was collected, and a single polarization was
selected. The emission was then spectrally filtered with a resolution of approximately 0.1
nm using a diffraction grating, and coupled into a single-mode fiber. By this method, we
obtained bright single-photon sources with excellent two-photon suppression and negli-
gible background emission. We chose three QDs for this study, denoted as dots 1, 2, and
3, with emission wavelengths of 931, 932, and 937 nm, respectively. As explained above,
the parameter g 0( 2 ) quantifies two-photon suppression as the probability to generate two
photons in the same pulse, normalized by an equally bright Poisson-distributed source.
We estimate g 0( 2 ) = 0.053, 0.067, and 0.071 for dots 1, 2, and 3, respectively. However, for the
experiment described below, the important parameter is the probability to generate two
photons in the same pulse, for either of two consecutive pulses, divided by the probability
to generate one photon in each pulse. We estimate this quantity to be g 0′ = 0.039, 0.027, and
0.025 for QDs 1, 2, and 3, respectively. The difference between g 0( 2 ) and g 0′ is due to the blink-
ing in our source, as discussed above.
The spectral properties of QDs can be highly variable, depending on the crystal growth
and fabrication details. In many cases, quite broad lines are observed, which can be attrib-
uted to spectral diffusion of the exciton transions [76]. However, some reports have indi-
cated coherence times [77] or time-averaged linewidths [76,78,79] close to the radiative
limit, showing that decoherence can be slow. Measurements of the temporal and spec-
tral properties of devices 1, 2, and 3, averaged over many excitation pulses, are shown in
Figure 2.31. In Figure 2.31a, the average emission intensity is plotted versus time following
a resonant excitation pulse, measured using a streak camera. By fitting decaying exponen-
tial functions, we estimate the spontaneous emission lifetimes τs of devices 1, 2, and 3 to
be 89, 166, and 351 ps, respectively. This variation is due largely to differences in how well
each QD couples to its microcavity. In Figure 2.31b, a Michelson interferometer was used
to measure the coherence length of the time-averaged emission. The curves show how the
interference fringe contrast varies with path-length difference, and give the magnitude of
the Fourier transform of the intensity spectrum. When we did not select a single polari-
zation, we sometimes observed oscillatory behavior due to polarization splitting of the
emission lines [32]. For devices 2 and 3 with splittings of 13 and 17 meV, respectively, we
were able to eliminate this effect by selecting a particular linear polarization. For device
1, the 45 meV splitting could not easily be eliminated. One possible explanation is that the
QD emission couples to just one cavity mode having a polarization rotated ~45° relative
to the splitting axes of the QD. We estimate the 1/e coherence lengths τc (divided by c) for
devices 1, 2, and 3 to be 48, 223, and 105 ps, respectively. Device 2 is closest to being Fourier
transform-limited, with 2τs/τc = 1.5. When this ratio is equal to 1, perfect two-photon inter-
ference is expected.
The main elements of the two-photon interference experiment are shown in Figure 2.32.
Now, the device is excited twice every 13 ns by a pair of equally intense pulses with 2 ns
separation. Two pulses, each containing zero or one photon, emerge from the single-mode
fiber. They are split into two arms by a beamsplitter, with one arm (2 ns + Dt) longer than the
other. The beams recombine at a different place on the same beamsplitter. The two outputs
are collected by photon counters, and a photon correlation histogram is generated of the
relative delay time τ = t2 - t1 for two-photon coincidence events, where t1 and t2 are the meas-
ured detection times at photon counters 1 and 2, respectively. A series of histograms with
varying Dt obtained in this way from device 2 are shown in Figure 2.33. Five peaks appear
within the central cluster, corresponding to three types of coincidence events. For peaks “a”
and “e” at t = 4 ns, the first photon follows the short arm of the interferometer, the second
photon follows the long arm, and one photon goes to each counter. For peaks “b” and “d” at
t = 2 ns, both photons follow the same arm. For the central peak “c” at t = 0, the first photon
follows the long arm, and the second photon follows the short arm, so that the two photons
make their second pass through the beamsplitter at the same time. Only in this case can
two-photon interference occur, and for perfect two-photon interference, peak “c” vanishes.
When the source successfully delivers a pair of photons, the two-photon state can be
written as:
|ψ =
∫ ds x(s)∫ dt y(t)a (s)a (t + 2 ns) | 0 ,
† †
(2.15)
(b)
(a)
1
0.8
Fringe contrast
Intensity (a.u.)
0.4
dot 2: 166 ps
0.2
dot 1: 89 ps
0
0 0.2 0.4 0.6 0.8 1 –0.6 –0.4 –0.2 0 0.2 0.4 0.6
Time after pulse (ns) Path-length difference (ns)
Figure 2.31
Time-resolved average spontaneous emission intensity following a laser pulse, measured by streak camera. (b)
Results from a single-photon interference experiment showing fringe contrast versus path-length difference.
Measurements were made using the setup in Figure 2.32 with the path-length difference varied about zero. The
average output intensity was measured while a piezo-electric transducer modulated the path-length difference
by a few wavelengths. The fringe contrast was estimated using the minimum and maximum vales from the
resulting sinusoidal oscillations. (From Santori, C., Fattal, D., Vučkovic’, J., Solomon, G. S., and Yamamoto, Y.,
Nature (London), 419 (2002) 594–597.)
Photon counters
t2 t1 Nonpolarizing
beamsplitter
2 ns + ∆t
2 ns
Single-mode
fiber Retroreflectors
Figure 2.32
Two-photon interference experiment: Every 13 ns, two pulses, separated by 2 ns and each containing 0 or 1
photon, arrive through a single-mode fiber. The pulses pass through an interferometer with 2 ns path-length
difference, and the outputs are detected by photon counters. When the first photon follows the long arm, and
the second photon follows the short arm, two-photon interference is possible. Time interval analysis electronics
allows these events to be distinguished from the other possibilities. (From Santori, C., Fattal, D., Vučkovic’, J.,
Solomon, G. S., and Yamamoto, Y., Nature (London), 419 (2002) 594–597.)
80
∆t = −353 ps
60
40
20
0
80
Coincidences (min)
∆t = 0 ps
60
40 c
20
a b d e
0
80
∆t = 347 ps
60
40
20
0
−10 −5 0 5 10
τ (ns)
Figure 2.33
Time correlation histograms obtained by exciting device 2 with two pulses 2 ns apart and sending the result-
ing photon emission into the interferometer with path-length difference 2 ns + ∆t. The central peak at τ = 0 cor-
responds to events when photons from opposite arms arrive at the beamsplitter at the same time and exit in
opposite directions.
where a†(t) is the photon creation operator at time t, x(s) and y(t) define the photon wave
packets, and |0〉 is the vacuum state (use of such a photon creation operator in the time
domain requires a proper definition and certain assumptions [72]). We assume that the
photon wave packets are much shorter than 2 ns. In the limit of low collection efficiency,
the mean areas of peaks “a”–”e” are:
where N is the number of repetitions, η(2) is the combined two-photon generation and
detection efficiency, and RI and TI are the beamsplitter intensity coefficients of reflection
and transmission, respectively. As defined above, the parameter g 0′ characterizes the two-
photon emission probability (following a single pulse), with g 0′ = 0 for an ideal single-photon
source, and g 0′ = 1 for a Poisson-distributed source. The parameter v1 is the interference
fringe contrast measured when an ideal monochromatic calibration source is sent into the
interferometer, and accounts for optical surface imperfections. The parameter v(∆t) = 〈|∫dtx(t)
y*(t + ∆t)2|〉 in the expression for peak “c” is the mean overlap between the wave packets of
the two photons for interferometer path-length difference (2 ns + Dt). An ensemble average
is performed over all possible two-photon states generated by the source.
In Figure 2.33, the reduced area of peak “c” at Dt = 0 compared to peaks “b” and “d” can
only be explained in terms of two-photon interference. As the delay Dt is changed, we
observe that peak “c” becomes taller and also begins to split into two separate peaks as Dt
becomes comparable to the detector time resolution (~200 ps). Let us define the quantity
M(∆t) = Ac/(Ab + Ad) in terms of the peak areas in Equations 2.17 through 2.19. This quantity
is equal to the conditional probability, given that two photons collide at the beamsplit-
ter, that the photons exit in opposite directions, in the limit g 0′ → 0 . We measured M(Dt)
at many path length offsets Dt, and the results are shown in Figure 2.34. For all three
devices, we observe reductions in the coincidence probability near Dt = 0, by factors of 0.61,
0.69, and 0.62 for devices 1, 2, and 3, respectively. The remaining coincidences are partly
due to independently measured optical imperfections in our setup, R I/TI = 1.1 and v1 = 0.92.
Without these imperfections, the coincidence reduction factors would be v(0) = 0.72, 0.81,
and 0.74 for devices 1, 2, and 3, respectively. We fitted the data in Figure 2.34 to the function
M(∆t) = 0.5[1 − a exp(–|∆t|/τm)], where the fitting parameters a and τm characterize the depth
and the width of the coincidence dip, respectively. The fits, shown as solid lines, match the
data well. For an ideal spontaneous-emission source and perfect interferometer we would
expect a = 1 and τm equal to the spontaneous emission lifetime. The fitted values of τm we
obtain are 80, 187, and 378 ps for devices 1, 2, and 3, respectively, in good agreement with
the spontaneous emission lifetimes τs given above. For devices 1 and 3, this result is sur-
prising given the short coherence lengths τc also given above. This suggests that for QDs
0.5
(a)
0.25
dot 1 : 80ps
0
Opposite output probability
0.5
(b)
0.25
dot 2 : 187ps
0
0.5
(c)
0.25
dot 3 : 378ps
0
–0.4 –0.2 0 0.2 0.4
Delay offset (ns)
Figure 2.34
For devices 1 (a), 2 (b), and 3 (c), the probability that two photons that collide at the beamslitter exit in opposite
directions, versus interferometer delay offset ∆t. The data are fitted to a two-sided exponential function as
described in the text. (From Santori, C., Fattal, D., Vučkovic’, J., Solomon, G. S., Yamamoto, Y., Nature (London)
419 (2002) 594–597.)
1 and 3, the primary spectral broadening mechanism occurs on a timescale much longer
than 2 ns. Such a “spectral diffusion” effect could occur, for example, due to charge fluc-
tuations in the vicinity of the QD [76].
The best results are from device 2, with an estimated two-photon overlap of 0.81. We
expect that the remaining imperfection is caused by two types of decoherence. The first is
related to the excitation process. When the QD is first excited by a laser pulse, the generated
electron–hole pair is initially in an excited state, and must relax to its lowest state through
phonon emission before a photon can be emitted at the proper wavelength. Since this tim-
ing information is lost to the environment through the phonon, we can think of this proc-
ess in terms of a classical random variable defining the start of the photon wavepacket. The
second type of decoherence occurs after the QD is in the lowest single-exciton state, while
the photon is being emitted. This includes decoherence caused by phonons [80,81], as well
as spectral diffusion processes as discussed above. A simple model predicts that the two-
photon overlap is equal to,
Γ r
v(0) = , (2.21)
Γ + 2γ r + Γ
where r is the relaxation rate of the upper exciton level into the lowest-energy single-
exciton level, G is the excited-state population decay rate through spontaneous emission,
and γ accounts for additional decoherence processes which create a broadened Lorentzian
spectrum. This model predicts that, initially, the photon state purity can be improved by
increasing Γ through the Purcell effect, so that Γ >> 2γ. However, a maximum is reached
when Γ = 2γ r , after which the excitation process becomes the dominant source of deco-
herence. For reasonable parameter values 1/g = 1 ns and 1/r = 10 ps, this gives τr = 1/
(Γmax) ≈ 70 ps. This is not much shorter than the shortest lifetimes already observed with
the present devices. Therefore, to see a large improvement over the present results we
expect that a coherent excitation scheme is required. Schemes based on a vacuum-stimu-
lated Raman process [10,11,82] have been explored for atomic single-photon sources, and
could be applied to QDs, as well [83]. An improved coherent excitation scheme which can
be applied to QDs coupled to microcavities is described in the Section 2.4.4. The degree
of two-photon interference observed here is not yet high enough for realistic applica-
tion to linear-optical quantum computing [68], but is high enough to demonstrate some
simpler schemes such as generation of polarization entanglement through two-photon
interference and post-selection, and single-mode teleportation. These demonstrations
are also discussed below.
k+Γ (k - Γ )2
ω± = ω0 - i ± g2 - , (2.22)
4 16
and strong coupling requires g > |κ - Γ|/4. For typical QDs and semiconductor cavities,
κ >> Γ, so the strong coupling condition reduces to g > κ/4. Since the coupling strength
g = ℘ ω / 2 ε 0V and the cavity decay rate κ = w0/Q, the figure of merit for a microcavity
intended to achieve strong coupling is given by Q/√V (rather than Q/V, which maximizes
the Purcell enhancement). Strong coupling between a single (In,Ga)As QD and a semicon-
ductor microcavity has been achieved using pillar microcavities [14], microdiscs [84], and
photonic crystal resonators [85,86].
An electron microscope image of a pillar microcavity fabricated at the University of
Würzburg, similar to those used to first demonstrate strong coupling [14], is shown in
Figure 2.35a. The device contains Bragg mirrors consisting of 26 and 30 pairs of GaAs/
AlAs layers above and below a GaAs cavity. At the central antinode of the cavity is a layer
of InGaAs QDs with an indium content of about 40% and a density of 1010 cm–2. A typical
device contains on the order of 100 QDs, but their spectral density is low enough that sin-
gle dots can be seen interacting with the cavity. Further details on device fabrication can
be found in Löffler et al. [88]. Pillars with diameters from 1 to 4 µm were fabricated. The
highest figure of merit Q/√V was found in the 1.8 m pillars, which exhibited quality fac-
tors from 10,000 to 20,000 with a mode volume V = 0.43 µm–3.
(a) (b)
C X
PL intensity (a.u.)
937.1 nm
725 nm
Figure 2.35
(a) SEM image of a 1.2 µm diameter pillar cavity. (b) Photoluminescence spectra of a 1.8 μm pillar with above-
band pumping (725 nm) and resonant pumping (937.1 nm). Cavity emission is labeled C, QD exciton emission
is labeled X. (From Press, D., Götzinger, S., Reitzenstein, S., Hofmann, C., Löffler, A., Kamp, M., Forchel, A., and
Yamamoto, Y., Phys. Rev. Lett., 98 (2007), 117402. © American Physical Society. With permission.)
(a) (b)
956.9
X X
C
Wavelength (nm)
18 K 956.8
C
C
956.7 X
PL intensity (a.u.)
956.6
8 10 12 14 16 18
Temperature (K)
(c)
0.08
10.5 K C
0.06 C
FWHM (nm)
X 0.04
C X
X
8K
956.4 956.6 956.8 957 0.02
Wavelength (nm) 8 10 12 14 16 18
Temperature (K)
Figure 2.36
(a) Temperature dependent photoluminescence from a strongly coupled pillar under resonant excitation (pump
wavelength 936.25–936.45 nm). Resonance is at 10.5 K. (b,c) Central wavelength and FWHM of longer wave-
length (circles) and shorter wavelength (squares) lines as a function of temperature. (From Press, D., Götzinger,
S., Reitzenstein, S., Hofmann, C., Löffler, A., Kamp, M., Forchel A., and Yamamoto, Y., Phys. Rev. Lett., 98 (2007),
117402. © American Physical Society. With permission.)
behavior similar to that observed here has also been demonstrated using a QW embed-
ded in a planar microcavity [89], in which many QW excitons contribute to the splitting at
resonance. We may distinguish between the quantum case involving a single QD exciton
and the semiclassical case involving a collection of emitters by measuring the autocorrela-
tion function of the photons emitted from the resonantly coupled system g r( 2,r) (τ) . Due to
the extremely short (~15 ps) lifetime of the coupled QD-cavity system, the system must be
excited with short pulses to resolve antibunching behavior using much slower detectors.
Strong antibunching is observed in the photons emitted by the resonantly coupled
QD-cavity system, as shown in Figure 2.37a. The measured value of the autocorrelation
function g r( 2,r) (0) = 0.18 < 0.5 proves that a single quantum emitter dominates photon
emission.
When the QD was red detuned by 0.4 nm from the cavity, a surprisingly large amount
of cavity emission persisted even though the resonant pump laser selectively excited only
the chosen QD (see Figure 2.37e). Photon statistics were collected from the cavity and QD
emission separately. The QD emission was antibunched as expected, with g (x2, x) (0) = 0.19
(Figure 2.37b). Interestingly, the cavity emission was also antibunched with g c( 2,c) (0) = 0.39 ,
demonstrating that most of the cavity emission originated from a single quantum emitter
(Figure 2.37c). Finally the cross-correlation function between the cavity and QD emission
was measured to be g (x2,c) (0) = 0.22 (Figure 2.37d), conclusively proving that both emission
peaks originate from the single QD emitter.
(a) (b)
g (2)
r,r (τ)
g (2)
x,x(τ)
(e)
0.18 0.19
C
PL Intensity (a.u.)
X
Coincidences
(c) (d)
g (2)
c,c (τ) g (2)
x,c(τ)
Figure 2.37
(a) Autocorrelation function of strong coupling system at resonance. (b–e) QD is detuned by 0.4 nm from cav-
ity. (b) Autocorrelation function of QD emission. (c) Autocorrelation function of cavity emission. (d) Cross-
correlation function of QD and cavity. (e) Photoluminescence spectrum. Shaded regions indicate pass bands of
spectral filter used for correlation measurements.
The strong cavity emission cannot be attributed solely to radiative coupling to the QD at
such a large detuning. Another mechanism, possibly mediated by the absorption or emis-
sion of acoustic phonons, must be responsible for efficiently channeling excitations from
the QD to the cavity. Further studies are needed to fully understand this coupling.
2.4 Applications
Many applications have been proposed for single-photon devices in the area of quantum
information, ranging from quantum key distribution, which uses the laws of quantum
mechanics to enable fundamentally secure communication, to large-scale quantum com-
putation, which uses “quantum parallelism” to allow for an exponential speed-up in cer-
tain problems over a classical computer. Here, we describe several applications relevant
to single-photon devices based on pillar microcavities. For three of these applications—
quantum key distribution [90], formation of polarization entanglement [91], and dual-
rail teleportation [92]—we have already performed proof-of-principle demonstrations, as
described below. The last part of this section presents a theory of coherent generation
and trapping of photons in a three-level system coupled to a microcavity [93]. We expect
this scheme to be the next step in improving the performance of QD-based single-photon
devices, and we hope that such techniques will eventually enable construction of quantum
networks for scalable quantum computation.
tph
〈 n〉 =
∫ 〈a (t)a(t)〉dt,
0
†
(2.23)
where tph is the time duration of the photon pulse, and a†(t) and a(t) are the photon creation
and annihilation operators, respectively, at time t; and (2) the autocorrelation g 0( 2 ), given by
tph tph
g (2)
0 =
∫ ∫
0 0
〈 a † (t)a † (t ′)a(t ′)a(t)〉dtdt ′
. (2.24)
〈n〉2
The communication rate of BB84 has been extensively studied, both for attenuated laser
sources [98] and single-photon sources [2]. Figure 2.38 plots the calculated communica-
(2)
tion rate as a function of channel loss for several different values of 〈n〉 and g 0 . The left
(2)
panel plots the rate for several different values of g 0 when 〈n〉 = 1. An attenuated laser
would feature a communication rate characterized by g 0( 2 ) = 1, while the curve g 0( 2 ) = 0 is
an ideal single-photon source. Note that the Poisson light bit rate decreases faster than
(a) (b)
100 100
n =1
n = 10−1
10–2 10–2 n = 10−2
n = 10−3
Bits per pulse
10–6 10–6
10–8 10–8
g∆ = 1
g ∆ = 10–2∆ g∆ = 0
g = 10–4
10–10 10–10
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50
Channel loss (dB) Channel loss (dB)
Figure 2.38
Security improvement for sources with supressed g 0( 2 ) (indicated in the figure as g∆). (a) Communication rate as
a function of channel loss for different values of g 0( 2 ). (b) Communication rate for g 0( 2 ) = 10-2 and different values
of the average flux 〈n〉.
the ideal single-photon device. This is because the single-photon device does not suffer
from photon splitting attacks. Thus, the rate decrease is only due to the increasing channel
loss. For Poisson light, as the channel loss increases the effect of the multi-photon states is
enhanced, forcing us to reduce the average number of photons. Intermediate devices with
0 < g 0( 2 ) < 1 feature two types of behaviors. At low channel losses they behave similarly to
the ideal device where the bit rate decreases in proportion to the channel transmission. At
higher loss levels the multi-photon states start to make a significant contribution and the
behavior gradually switches over to that of Poisson light. Note that each curve features a
cutoff channel loss, beyond which secure communication is no longer possible. A smaller
g 0( 2 ) implies that more loss can be tolerated.
The right panel of Figure 2.38 plots the communication rate for different values of 〈n〉
with g 0( 2 ) = 10-2. At low channel losses, 〈n〉 has a significant effect on the communication rate,
but at higher loss levels most of the curves meet with the ideal curve. Only the extremely
lossy device with efficiency of 10 - 3 fails to rejoin the ideal curve, and features a smaller
cutoff loss. This implies that at sufficiently high loss levels the flux of the source does not
have an impact on the communication rate, provided the flux exceeds a critical level. For
the simulation parameters in Figure 2.38, this critical level occurs somewhere between 10 –2
and 10 –3. Furthermore, the cutoff loss level, which defines the maximum communication
distance, is unaffected.
We demonstrated the improvement in the performance of a quantum-cryptography sys-
tem using our third-generation pillar microcavities. The performance of the particular
structure used was characterized as described in the previous section. The left panel of
Figure 2.39 shows the mean photon number per pulse for the micropillar device used. The
emission rate initially increases linearly with pump power, but eventually saturates at a
nominal rate of 0.07 photons per pulse. This average includes all losses in the measurement
system. The autocorrelation of the QD is shown in the right panel of Figure 2.39, obtained
for a pump intensity indicated by the point A in the left panel of the figure.
A full quantum-cryptography system requires active control and synchronization of
the photon polarization state. Figure 2.40 illustrates a system used to achieve this con-
trol. The microcavity-QD structure is excited by a pulsed laser whose repetition rate
defines the clock cycle of the experiment. An electro-optic modulator is used to prepare
(a) (b)
200
0.07
0.06
150
0.05
0.04
100
n
0.03 A
0.02
50
0.01
0.14
0.00
0
0 200 400 600 800 1000 1200 –80 –60 –40 –20 0 20 40 60 80
Pump power (uW) Time delay (ns)
Figure 2.39
Characterization of QD. (a) Photon emission rate measurement as a function of excitation power.
(b) Autocorrelation taken at excitation power indicated by point A in left panel.
Alice Counter
D0
He Spec. slit λ/4 plate
λ/2 plate 50/50
Cryo. Channel
Pinhole Sm fiber BSP
PBS EOM PBS D1
Flip
mirror D4 PBS D2
Dot Grating
Laser pulse D3
Data Amp.
TIA
Gen.
TIA
Bob
Figure 2.40
Apparatus for implementation of BB84 with a QD coupled to a micro-pillar cavity.
the polarization state of each photon before it enters the channel. A data generator, whose
signal is amplified by a high power amplifier, drives the modulator. The data generator is
synchronized to the laser pulses, and produces a random, four-level signal, corresponding
to the four different polarization states in the BB84 protocol. Bob’s detection apparatus is
composed of a 50–50 beamsplitter, which partitions each photon randomly to one of two
polarization analyzers. Both Alice and Bob share a common clocking signal from the data
generator. Each of Bob’s detection events is recorded by a time-interval analyzer (TIA),
together with a time stamp of the event relative to the common clock. A detection is also
used to generate a logic pulse (containing no information about the detection result) which
triggers a second TIA in Alice’s apparatus. This TIA records the polarization state that was
prepared, along with a time stamp that can be used for later comparison with Bob’s data.
The left panel of Figure 2.41 shows a correlation of the data recorded by Alice and Bob.
The central diagonal of the bar graph corresponds to error events, which are predominantly
due to effects such as phase drift and imperfections in the polarization optics. The overall
Bit Error Rate (BER) of the system is measured to be 2.5%. These errors can be handled in
(a) (b)
104
Laser (sim)
0.5 103 QD (sim)
Figure 2.41
(a) Data correlation between Alice and Bob shows a 2.5% bit error rate. (b) Experimental comparison between
a laser and QD coupled to a micro-pillar cavity. The QD achieves a 5dB improvement in channel loss over the
attenuated laser.
the communication system using two way error correction methods [99]. Such error correc-
tion algorithms are very efficient and work within about 25% of the Shannon limit.
In the right panel of Figure 2.41 one can see a comparison between the performance of
the microcavity QD to a standard attenuated laser in the presence of channel losses. At low
loss levels, the communication rate with the attenuated laser is higher, because the laser
starts out with a macroscopically large number of photons, which can be attenuated to any
desired average. This is in contrast to the QD, which is limited by the device efficiency and
losses in subsequent optics. However, at higher channel losses, the laser emits too many
multi-photon states, causing a more rapid decrease in communication. At around 16 dB,
the QD begins to outperform an attenuated laser. Above 23 dB of loss, secure communi-
cation is no longer possible with the laser, while the QD can withstand channel losses of
about 28 dB. This demonstrates the security advantage of the source in the presence of
channel losses.
In the final phase of the communication, the secret key is used as a one time pad to
exchange the message. Figure 2.42 shows how this is done. A picture of Memorial Church
at Stanford University serves as the message. A 20 kB secret key is exchanged over the
communication system. Alice uses her copy of the key to do a bitwise exclusive OR opera-
tion with every bit of the message. The resulting encrypted message looks like white noise
to anyone who does not posses a copy of the key, as shown in Figure 2.42b. This panel plots
the pixel value histogram of the original and encrypted message. The original message has
clear structure in the histogram, while the encrypted message features a flat pixel value
histogram indicative of white noise. Bob decodes the encrypted message by performing a
second bitwise exclusive OR using his copy of the key, faithfully reproducing the original
message.
(a)
200
1000 150
100
500
50
0 0
0 50 100 150 200 250 0 50 100 150 200 250
Pixel value Pixel value
Figure 2.42
(a) One time pad encryption protocol using exchanged 20 kB key. (b) Pixel histogram of original and encrypted
message. Encrypted message shows flat pixel distribution, as expected from white noise.
states, for instance, can be harnessed to produce interesting effects that would otherwise
require the use of extremely (maybe prohibitively) large optical nonlinearities. In this
section, we present the results of an experiment using quantum interference to produce
post-detected polarization entangled photons, for which we obtained a violation of Bell’s
inequality. In the next section, we will present an experiment in which the quantum state
of a single optical mode could be teleported to another optical mode via the same quantum
interference effect.
Entanglement, the nonlocal correlations allowed by quantum mechanics between dis-
tinct systems, is a central concept in quantum information science [100]. These nonlocal
correlations are often understood as the result of prior interactions between the quantum
mechanical systems, something like a memory of those interactions. In light of relatively
recent experiments in the field of quantum information [66]), this is too limited a view.
Entanglement can be induced between noninteracting particles, provided they are quan-
tum mechanically indistinguishable. Pioneering work by Shih and Alley [67], followed by
Ou and Mandel [101], used a post-selection procedure to induce entanglement between two
identical photons produced in a nonlinear crystal. More recently, entanglement swapping
experiments [102,103] used two independent entangled photon pairs to induce entangle-
ment between photons of different pairs which never interacted.
Our experiment makes use of the same underlying principle for entanglement genera-
tion. However, starting with two single-photons guarantees that at most a single-photon
pair can be generated at a time. The entanglement formation relies on two crucial features
of a micro-pillar QD single-photon source: (1) its ability to suppress multi-photon pulses
[9], and (2) its ability to generate consecutively two photons that are quantum mechani-
cally indistinguishable [4].
When two spatio-temporally identical photons of orthogonal polarizations arrive
together (“collide”) at a nonpolarizing beam splitter (NPBS), a quantum interference effect
ensures that, if detected simultaneously at different output ports of the NPBS, they should
be entangled in polarization [101]. More precisely, when the two optical modes correspond-
ing to the output ports, denoted “c” and “d,” of the NPBS have a simultaneous single occu-
pation, their joint polarization state is expected to be the EPR-Bell state:
1
| Ψ- 〉 = (| H 〉c | V 〉d - | V 〉c | H 〉d ).
2
The input port modes of the NPBS, denoted as “a” and “b”, are related to the output modes
“c” and “d” by the 50–50% NPBS unitary matrix according to:
1
a H /V = (c H /V + dH /V ),
2
1
bH /V = (c H /V - dH /V ),
2
where subscripts “H” and “V” specify the polarization (horizontal or vertical) of a given
spatial mode. The quantum state corresponding to single-mode photons with orthogonal
polarizations at port “a” and “b” can be written as:
1 † †
aH† bV† | vac〉 = (c H c V - d †Hd †V - c †Hd †V + c †Vd †H ) | vac〉
2
As pointed out by Popescu et al. [104], this state already features nonlocal correlations
and violates Bell’s inequality without the need for post-selection, by using photo-detectors
that can distinguish photon numbers 0, 1 and 2. If, however, we decide to discard the
events when two photons go the same way (recording only coincidence events between
modes “c” and “d”), we obtain the post-selected state:
1 † † † †
(cH dV - cV dH ) | vac〉 = | Ψ - 〉
2
From QD
A 2 ns
HWP
2 ns
H
V pol,. control
H H
NPBS 2 NPBS 1
Single mode
fibers
Figure 2.43
Experimental setup for entanglement generation. Single-photons from the QD microcavity device are sent
through a single mode fiber, and have their polarization rotated to H. They are split by a first NPBS (1). The
polarization is changed to V in the longer arm of the Mach-Zehnder configuration. The two path of the interfer-
ometer merge at a second NPBS (2). The output modes of NPBS 2 are matched to single mode fibers for subse-
quent detection. The detectors are linked to a time-to-amplitude converter for a record of coincidence counts.
wavefunctions overlap at the second nonpolarizing beam-splitter (NPBS 2). In all other
cases (not of interest), the single-photon pulses “miss” each other by at least 2 ns which is
greater than their width (100–200 ps). Two single-photon counter modules (SPCMs) in a
start-stop configuration were used to record coincidence counts between the two output
ports of NPBS 2, effectively implementing the post-selection (if photons exit NPBS 2 by
the same port, then no coincidences are recorded by the detectors). Single-mode fibers
were used prior to detection to facilitate the spatial mode-matching requirements. They
were preceded by quarter-wave and polarizer plates to allow the analysis of all possible
polarizations.
The detectors were linked to a time-to-amplitude converter and multi-channel analyzer
to record histograms of coincidence events versus detection time delay τ. A typical histo-
gram is shown on Figure 2.44, with the corresponding post-selected events. For given ana-
lyzer settings (θ1, θ2), we denote by C(θ1, θ2) the number of post-selected events normalized
by the total number of coincidences in a time window of 100 ns. This normalization is inde-
pendent of (θ1, θ2), since the input of NPBS 2 are two modes with orthogonal polarizations.
C(θ1, θ2) measures the average rate of coincidences throughout the time of integration.
The single count rate at the output of the single-mode fiber was 9400 s–1, from which we
infer a total quantum efficiency of 0.13% (detection loss included). The total pair produc-
tion rate for QD 1 was 12 s –1 after fiber, so that useful pairs were generated with a rate of
1.5 s–1 (we lose a factor 8 due to the post-selection and by excluding “bad-timing” events).
High suppression of two-photon pulses and high mean overlap (indistinguishability)
between consecutive photons were observed. The overlap was measured in a photon
bunching experiment [4], as described above.
Several methods can be used to prove the success of entanglement generation. The tra-
ditional method is a Bell inequality test, that proves the nonlocal nature of the state of
two quantum systems if they are sufficiently entangled. Another more complete way is to
35
Integration window
30
25
Recorded coincidences
20
Peak area
15
176 165
10 81 92
76
5
0
−13 −2 0 2 13
Time delay at NPBS 2 (ns)
Figure 2.44
Zoom on a typical correlation histogram, taken on QD1. Coincidences with delay τ between detectors A and B
were actually recorded for –50 ns<τ<50 ns. The integration time was 2 min, short enough to guarantee that the
QD is illuminated by a constant pump power. The central region –1 ns<τ<1 ns corresponds to the post-selected
events: the corresponding photons overlapped at NPBS 2 where they took different exit ports. (From Fattal, D.,
Inoue, K., Vučkovic’, J., Santori, C., Solomon. G. S., and Yamamoto, Y., Phys. Rev. Lett., 92 (2004) 37903. © American
Physical Society.)
reconstruct the full density matrix describing the joint polarization state of two photons, a
procedure known as quantum state tomography. The presence of entanglement can then be
read from the density matrix by several mathematical methods.
A Bell-inequality test was performed for post-selected photon pairs from QD1. Following
Clauser et al. [105], if we define the correlation function E(θ1, θ2) for analyzer settings θ1 and
θ2 as
Table 2.1
Normalized Coincidence Counts C(θ1, θ2) 103 for Various
Polarizer Angles used in the Bell Inequality Test
θ2\θ1 0° 45° 90° 135°
2 g 0( 2 )
RI
- v(0)
1 TI
ρmodel =
(2)
RI
+ RTII + 4 g 0 TI
TI
- v(0)
RI
(2)
2g 0
R I and TI are the intensity reflection and transmission coefficients of NPBS 2 (R I/TI ∼ 1.1
in our case). Using the independently measured values for g 0( 2 ) and v(0), we obtain
excellent quantitative agreement of our model to the experimental data, with a fidelity
Tr ( ρexp model ρexp ) as high as 0.997.
1/2 ρ 1/2
The negativity of the state ρmodel is proportional to ( v(0) - 2 g 0( 2 ) ), which means that entan-
glement exists as long as v(0) > 2 g 0( 2 ). This simple criterion can be applied to any single-pho-
ton source for which the intensity autocorrelation and photon overlap values are known.
It indicates the extent to which that source will be able to generate entangled photons
through two-photon interference and post-selection.
0.5 0.5
0.25 0.25
0 0
HH HH
HV HV
VV VV
VH VH VH VH
VV HV VV HV
HH HH
EXPERIMENT
IDEAL
Real component Imaginary component
0.5 0.5
0.25 0.25
0 0
HH HH
HV HV
VV VV
VH VH VH VH
VV HV VV HV
HH HH
Figure 2.45
Reconstructed polarization density matrix for the post-selected photon pairs emitted by our device. The small
diagonal HH and VV components are caused by finite two-photon pulses suppression (g 0( 2 ) > 0). Additional
reduction of the off-diagonal elements originates from the imperfect indistinguishability between consecu-
tively emitted photons. (From Fattal, D., Inoue, K., Vučkovic’, J., Santori, C., Solomon, G. S., and Yamamoto, Y.,
Phys. Rev. Lett., 92 (2004) 37903. © American Physical Society.)
mode of the optical field. A QD in a micro-pillar cavity was used to generate a stream
of nearly identical photons, each defining a quantum bit of information in the so-called
“dual-rail” representation. Upon interference and measurement of two such qubits, a tar-
get and an ancilla, the state of the target qubit is observed to change physical support
(optical modes) with very high fidelity to the original state. In particular, the coherence
between two different optical modes encoding the target qubit was transferred with high
fidelity to the output modes. The observed fidelity is 80%, in agreement with the residual
distinguishability between consecutive photons from the source. An improved version
of this teleportation scheme using more ancillas constitutes a building block for scalable
quantum computing using linear optics and photo-detectors.
Photons are almost ideal carriers of quantum information, since they have little interac-
tion with their environment and are easy to manipulate individually with linear optics.
The main challenge of optical quantum information processing is the design of controlled
interactions between photons, necessary for the realization of nonlinear quantum gates.
Photons do not naturally “feel” the presence of other photons, unless they propagate in a
medium with high optical nonlinearity. The amount of optical nonlinearity required to per-
form controlled operations between single-photons is however not easily accessible. Large
optical nonlinearities having noticeable effect at the single-photon level might become
eventually available with the development of cavity QED or electromagnetically induced
transparency (EIT) techniques, but at present these possibilities remain quite remote.
Probabilistic quantum-logic gates can be implemented with linear optics only [68,108,109],
but as such, they are not suitable for scalable quantum computation. In a seminal paper,
Gottesman and Chuang [110] suggested that quantum gates could be applied to photonic
qubits through a generalization of quantum teleportation [111]. In such a scheme, the infor-
mation about the gate is contained in the state of ancilla qubits. The implementation of a
certain class of gates can then be reduced to the problem of preparing the ancilla qubits in
some wisely chosen entangled state. Such a problem can be solved “off-line” with linear-
optics elements alone, provided the photons used are quantum mechanically indistin-
guishable particles [66]. Following this idea, Knill, Laflamme and Milburn (KLM) [68]
proposed a scheme for efficient linear-optics quantum computation (LOQC) based on the
implementation of the controlled-sign gate (C-z gate) through teleportation. Since the C-z
gate effectively acts on only one of the two modes composing the target qubit, a simplified
procedure can be used where a single optical mode is teleported, instead of the two modes
composing the qubit. This procedure will be referred to as single mode teleportation, to dis-
tinguish it from the usual teleportation scheme. In its basic version using one ancilla qubit,
it succeeds half of the time. In its improved version using an arbitrarily high number of
ancillas, it can succeed with a probability arbitrarily close to one [68,112].
Single-mode teleportation, in its simplest form, involves two qubits, a target and an
ancilla, each defined by a single-photon occupying two optical modes (Figure 2.46). The
target qubit can, a priori, be in an arbitrary state
|ψ t 〉 = ζ 0 |0〉L + ξ 0 |1〉L ,
where the logical |0〉L and |1〉L states correspond to the physical states |1〉1|0〉2 and |0〉1|1〉2,
respectively, in a dual-rail representation. The ancilla qubit is prepared with a beam-splitter
(BS a) in the coherent superposition
1 1
| ψa〉 = (| 0〉L + | 1〉L ) = (| 1〉3 | 0〉 4 + | 0〉3 | 1〉 4 ).
2 2
Target ‘1’
|0 |0
‘2’ ‘C’
|1 C
BS 1 Output
|1 D
‘3’ ‘D’
|0 BS a |1
Ancilla ‘4’
Figure 2.46
Schematic of single mode teleportation. Target and ancilla qubits are each defined by a single-photon occupying
two optical modes (1–2 and 3–4). When detector C records a single-photon, the state in modes 1–4 reproduces
the initial state of the target. In particular, the coherence between modes 1-2 of the target can be transferred to
a coherence between modes 1–4. (From Fattal, D., Diamanti, E., Inoue, K., and Yamamoto, Y., Phys. Rev. Lett., 92
(2004) 37904. © American Physical Society.)
One rail of the target is mixed with one rail of the ancilla with a beam-splitter (BS 1), for
subsequent detection in photon counters C and D. The state after mixing can be written in
terms of modes C, D, 1 and 4 as
1 1
|ψ t 〉12 |ψ a 〉34 = |10〉CD |ψ t 〉14 + |01〉CD (Z|ψ t 〉14 ),
2 2
ζ0 ξ
+ | 00〉CD | 11〉14 + 0 (| 02 〉+ | 20〉)CD | 00〉14 .
2 2
For a given realization of the procedure, if only one photon is detected at detector C, then
the state of the output qubit (modes 1 and 4) is |ψt 〉, so the teleportation was successful.
Similarly, if only one photon is detected at detector D, then the output state is Z|ψt 〉, so in
this case we have to apply the Pauli operator Z (phase shift of π) to the output modes to
retrieve the initial state |ψt 〉 [113]. We did not implement this active feedforward here, in
order to simplify the experiment. For this reason, our teleportation procedure succeeds
with probability 1/4 (as compared to 1/2 had we used feedforward).
It is interesting and somewhat enlightening to describe the same procedure in the frame-
work of single-rail logic. In this framework, each optical mode supports a whole qubit,
encoded in the presence or absence of a photon, and the single mode teleportation can be
viewed as entanglement swapping. Indeed, for the particular values ζ = ξ = 1/√2, modes 1
and 2 find themselves initially in the Bell state |ψ + 〉12, while modes 3 and 4 are in a similar
state |ψ + 〉34. A partial Bell measurement takes place using BS 1 and counters C/D, which if
it succeeds leaves the system in the entangled state |ψ + 〉14, so that entanglement swapping
occurs. In the rest of the chapter, we choose to consider the scheme in the dual-rail picture,
since it is a more robust, and hence realistic, way of storing quantum information, albeit at
the expense of using two modes per qubit.
The success of the teleportation depends mostly on the transfer of coherence between
the pairs of modes (1–2) and (1–4). If the target qubit is initially in state |0〉L = |1〉1 |0〉2, then the
ancilla photon cannot end up in mode 4, because of the postselection condition, so that the
output state is always |1〉1 |0〉4 as wanted. The same argument applies when the target qubit
is in state |1〉L. Hence, the success of the teleportation is granted when the target qubit is
not in a superposition state. However, when the target qubit is in a coherent superposition
of |0〉L and |1〉L, the output state might not retrieve the full initial coherence. A good way to
test the coherence transfer is by changing the optical path length Δ on mode (1). If the tel-
eportation procedure does not randomize the phase between mode (1) and mode (4), then
changing Δ in a controlled manner changes the well defined phase between modes (1–4),
which can be observed by interfering modes (1) and (4) in an auxiliary setup. If, however,
the teleportation randomizes the phase between modes (1)–(4), then changing the path
length Δ will not have any effect on the interferometric signal.
The experimental setup is shown in Figure 2.47. Two photons emitted consecutively by
the photon source are captured in a single-mode fiber. In the dual-rail representation, we
refer to the first photon as the ancilla, and to the second photon as the target (see Figure
2.46). The ancilla qubit, initially in state |0〉L, is delayed in free space to match the target
qubit temporally at BS 1. The delay must be adjusted to within a fraction of the photon’s
temporal width (~200 ps or 6 cm in space). Note that the mode matching is significantly
easier here than in similar experiments using photons from PDC, where the optical path
length must be adjusted with a tolerance of only a few microns [114].
The ancilla is prepared in the superposition state |ψanc 〉 = 1/√2 (|0〉L + |1〉L ) with a beam-
splitter “BS a”. The target qubit is prepared in a similar maximum superposition state
(with “BS t”). The path length Δ of mode (1) is changed in a controlled manner with a
piezo-actuated mirror. The “partial Bell measurement” responsible for the teleportation
is done at BS 1 by mixing the optical modes (2) of the target qubit and (3) of the ancilla
qubit, with subsequent detection in counter C. A Mach-Zehnder type setup is used to
measure the coherence between the two modes (1) and (4) of the output qubit. It is com-
posed of a 50-50 beam-splitter BS 2 mixing modes (1) and (4), with subsequent detection
From QD
Delay
Piezo
loop Ancilla Target
1 BS t
∆
2 SM-fiber
BS 1
C Polarizer
3
BS a
D
BS 2
A
4
Figure 2.47
Experimental setup for single-mode teleportation. All the beam-splitters (BS) shown are 50–50 nonpolarizing
BS. The teleportation procedure works when the ancilla photon is delayed, but the target is not. After prepa-
ration, the target photon occupies modes 1 and 2, and the ancilla occupies modes 3 and 4. Modes 2 and 3 are
mixed at BS 1 and subsequently measured by detectors C and D, this step being the heart of the teleportation.
When C records a single-photon, another single-photon occupies modes 1–4 (output qubit). The phase coher-
ence between modes 1–4 in the output state is measured by mixing those modes at BS 2 and recording single
counts at detector A or B. Note that since an event is recorded only if A and C or B and C clicked, more than one
photon could not have reached detector C.
20 20
Detector A Detector B
18 18
16 16
14 14
Coincidence counts
12 12
10 10
8 8
6 6
4 4
2 2
0 0
−4 −2 0 2 4 −4 −2 0 2 4
Time delay A/C (ns) Time delay A/C (ns)
Figure 2.48
Typical correlation histograms taken simultaneously between detectors A/C and B/C. The central region indi-
cated by the dashed lines correspond to the postselected events, when target and ancilla photons had such a
timing that it is impossible to distinguish between them based on the time of detection. As the path length ∆
varies, so does the relative size of the central peaks for detector A and B. The sum of count rates for the central
peaks of detector A and B was 800/s, independently of φ as shown in Figure 2.49. (From Fattal, D., Diamanti, E.,
Inoue, K., and Yamamoto, Y., Phys. Rev. Lett., 92 (2004) 37904. © American Physical Society.)
Detector A + Detector B
0.2 Detector B
0.15
0.1
0.05
Detector A
0
0 1 2 3 4 5 6 7 8
Voltage applied to piezo (V)
Figure 2.49
Verification of single mode teleportation. Coincidence counts between detector A/C and B/C are plotted for
different voltages applied to the piezo transducer, i.e. for different path length ∆ of mode (1). The observed
modulation of the counts implies that the initial coherence contained in the target qubit was transferred to a
large extent to the output qubit. The reduced contrast (~60%) is principally due to imperfect indistinguishability
between the target and ancilla photons. (From Fattal, D., Diamanti, E., Inoue, K., and Yamamoto, Y., Phys. Rev.
Lett., 92 (2004) 37904. © American Physical Society.)
the post-selected window (-1 ns < τ < 11 ns) was normalized by the total number of counts
recorded in detectors A and B in the broader window - 5 ns < τ < 5 ns, corresponding to
all events where one photon went through BS 1 and the other through BS 2 (but only one
quarter of the time with right timing). Complementary oscillations are clearly observed at
counter A and at counter B, indicating that the initial coherence was indeed transferred to
the output qubit. In other words, mode (2) of the target qubit was “replaced” by mode (4) of
the ancilla without a major loss of coherence.
Were the initial coherence fully conserved during the single mode transfer, the count
rate at detectors A and B would be proportional to cos2(π∆/λ) and sin2(π∆/λ), respectively
(λ being the single-photon wavelength), giving a perfect contrast as the path length Δ is
varied. More realistically, part of the coherence can be lost in the transfer, resulting in a
degradation of the contrast. Such a degradation is visible in Figure 2.49. It arises mainly due
to a residual distinguishability between ancilla and target photons. Slight misalignments
and imperfections in the optics also result in an imperfect mode matching at BS 1 and BS
2, reducing the contrast further. Finally, the residual presence of two-photon pulses can
reduce the contrast even more, although this effect is negligible here. The overlap v(0) = |〈ψt
|ψanc 〉|2 between target and ancilla wave-packets [4], the two-photon suppression factor g 0( 2 ),
as well as the nonideal mode matching at BS 1 and BS 2—characterized by the first-order
interference visibilities v1, v2—were all measured independently, as described above. The
results are v(0) ~ 0.75, g 0( 2 ) ~ 2%, v1 ~ 0.92 and v2 ~ 0.91. The contrast C in counts at detector A
or B when we vary the phase φ should then be
v(0) ⋅ v1 ⋅ v2
C= ∼ 0.62.
1 + g 0( 2 ) / 2
This predicted value compares well with the experimental value of Cexp ~ 0.60.
The fidelity of teleportation is F = (1 + C)/2 ~ 0.8. This high value is still not enough to
meet the requirements of efficient LOQC [68]. In particular, the quantum indistinguish-
ability of the photons must be further increased to meet these requirements. As discussed
above, the QD exciton from which the photon is generated dephases on a time scale of a
few nanoseconds [115], which degrades the photon indistinguishability. Through further
increases in the spontaneous emission rate, one can reduce the QD radiative lifetime well
below this dephasing time. However, jitter in the photon emission time will eventually
prevent any further reduction of the QD lifetime. Time jitter happens as a consequence of
the incoherent character of our excitation method, and is currently on the order of 10 ps.
Time jitter can, in principle, be completely suppressed using a coherent excitation tech-
nique described in the next section (see e.g., Kuhn et al. [11] for such a scheme with single
atoms). It therefore seems vital to develop similar techniques with single QDs if they are
to be used for optical QIP.
for ideal systems, then study the effect of losses in practical systems. The mathematical
details of the model are presented last.
√κ a Cavity mode
Л system
|r
α(t)
1 iΦ
(Arbitrary single photon pulse) ‒ Ω( t)e (t) g0
2
(Control pulse) (Cavity)
|e |g
Figure 2.50
Composition of a “node”: three-level atom or QD in a Λ configuration placed in a single mode optical micro-
cavity. The g–r transition frequency is close to the incident photon and cavity resonance, and has a vacuum
Rabi frequency g with the cavity mode. The e–r transition couples to a classical laser pulse called the “control
pulse”. This transition does not couple to the cavity, due, for instance, to polarization or frequency mismatch.
The cavity mode, mathematically represented by annihilation operator a, is coupled to an external radiation
mode, which gives it a finite decay rate k. It is assumed that the cavity can exchange energy solely with that
particular external mode.
The main result of this section is that this absorption process can be done perfectly if
and only if the photon pulse satisfies the following condition for all times:
t
| ψ (t) |2 1 k
∫
E(t) ≡ | ψ ( s) |2 ds -
-∞
k
- 2 | ψ - ψ (t) |2 > 0
kg 2
(2.25)
This relation tells us that the single-photon bandwidth cannot be much greater than
k, and that g cannot be much smaller than k. However in general, it does not require an
often-assumed adiabaticity condition (|ψ / ψ|<< g), nor does it require the regime of strong
coupling (g >> κ). We also learn that the detuning of the photon pulse relative to the cav-
ity frequency can be as large as g. The proof of Equation 2.25 is given at the end of this
section.
When this relation holds, we can always design a control pulse Ω(t) that enforces a
destructive interference between two photon amplitudes: one reflected by the front mirror
of the cavity and the other absorbed and re-emitted by the system (see Figure 2.51***). As a
result, there is no net reflection of the incident photon. The absolute value of the required
control pulse needs to be
The control pulse must, in general, have a chirp to compensate both for a possible chirp
in the incident photon pulse and for an eventual finite detuning δ. The exact expression
for the chirp is rather complicated, and can be found in the mathematical details provided
below. However if the photon envelope has no chirp, and if δ = 0, the control pulse does not
need to have a chirp and is given by the following simple expression (also derived in Yao
et al. [118] although not in closed form):
Figure 2.51
Physical picture of the photon trapping process. The incoming photon wave-packet can escape the node in two
distinct ways: by direct reflection on the cavity front mirror (solid line) or by absorption and re-emission from
the atom-cavity system (dashed line). If the control pulse applied to the Λ system is well designed, the prob-
ability amplitudes for these two events interfere destructively, and the photon stays trapped in the atom-cavity
system with a probability of one.
Because of time-reversal symmetry, the solution to the photon trapping problem also
gives a solution to the photon emission problem (Figure 2.52). This means that ultra-fast
single-photon pulses can be produced deterministically, provided they obey the time-re-
versed relation (Equation 2.25). Relation 2.25 generally implies that arbitrary single-photon
pulses can be perfectly emitted or trapped as long as they are not detuned from the cavity
resonance w 0 by more than g, and their bandwidth γp does not exceed min (κ, (g2/κ)). This
is true even in the weak coupling regime (g << κ). In the strong or intermediate coupling
regimes, these processes can be made as fast as the cavity decay rate k, in which case
the control pulse becomes highly nonadiabatic and must be precisely designed to match
the corresponding photon envelope (Figure 2.53). In the large detuning case, which was
studied earlier by Cirac et al. [69], relation 2.25 can only be satisfied in the strong coupling
regime, and the control pulse shape is given by
We write the result in this form, since (gΩ(t))/(2δ) is the effective Raman coupling
between the two ground states |e〉 and |g〉 after adiabatic elimination of level |r〉.
0.2
α(t)
0
–50 –10 0 10 50
Classical pulse to be applied
2
Ω(t) (κ)
0
–50 –10 0 10 50
Populations
1
|e
0.5
Trapping Generation
|g
|r
0
–50 –40 –30 –20 –10 0 10 20 30 40 50
Time (1/κ)
Figure 2.52
Control pulse Ω(t) to apply to trap a single-photon pulse of given amplitude ψ(t). The bottom plot shows the
simulated evolution of the state of the node with time. The parameters used in the simulation are g = k, δ = 0.
Note that the control pulse is highly nonadiabatic ( Ω/Ω ~ Ω ~ g). Looking at the plots from right to left (time-
reversal) gives the solution to the photon emission problem. Modifying the control pulse in a time region where
the photon amplitude is negligibly small does not significantly affect the dynamics. Hence, a control pulse that
should be always “on” in the remote past can actually be turned on only a little while before the photon ampli-
tude starts to rise (as indicated by the dashed line).
0
α(t)
–0.5
–50 –10 0 10 50
Control pulse applied
2
Ω(t) (κ)
–2
–50 –10 0 10 50
Node populations
1
|e
0.5
Trapping Generation
|g
|r
0
–50 0 50
Time (1/κ)
Figure 2.53
Generation and trapping of a single-photon pulse with oscillating amplitude in a nonadiabatic regime.
Additionally, if the incident photon has no chirp, the control pulse must have a necessary
t) given by
chirp Φ(
2
(t) = - | Ω(t) | .
Φ (2.29)
4δ
Relations 2.28 and 2.29 agree with and extend the results of Cirac et al. [69], where identi-
cal nodes and time-reversal invariant photon shapes were assumed. The necessary chirp
of the control pulse can be interpreted as a cancellation of the AC stark-shift induced by
level |r〉 or level |e〉 via the classical control pulse.
Another limit of interest is the adiabatic regime, where the incident (emitted) photon
pulse is slow enough so that γp << min [κ, (g2/κ)]. This is the regime in which existing coher-
ent single-photon sources are operated. If we further assume δ = 0, the control pulse has no
chirp and takes the following simple form:
Ω(t) g ψ (t)
= . (2.30)
2 k t
∫ -∞
|ψ ( s)| ds 2
Note that the denominator is the square root of the total energy carried by the photon up
to time t and absorbed by the cavity. Therefore, the intensity |Ω(t)|2 of the control pulse is
proportional to the instantaneous rate at which the single-photon energy enters the cavity.
This result is intuitive: it means that any energy accumulated in level |g〉 from the radia-
tion mode must be immediately transferred to level |e〉; otherwise, it will be re-radiated
and lost. Equation 2.30 can be expressed equivalently as
k Ω(t) k t Ω( s) 2
ψ (t) =
g 2
exp 2
2 g -∞ 2
∫ds ,
(2.31)
which gives the photon shape if we know the (adiabatic) control pulse. This result explains
previous experimental observations and simulations of coherent photon emission via the
STIRAP technique [11,117]. In the adiabatic regime, the emitted photon pulse follows the con-
trol pulse except for a slight correction given by the exponential term in Equation 2.31, caus-
ing the emitted photon pulse to rise more slowly and decay faster than the control pulse.
kL
Lc ∼ . (2.32)
k
Spontaneous decay from level |r〉 at rate Γr further reduces the efficiency and can degrade
the quantum-mechanical overlap between two photons emitted consecutively [4]. The
reduction of efficiency L due to Γr can be estimated as
Γ k 1 ∞
L ∼ 2r
g
+
4 k -∞
∫
|ψ ( s)|2 ds ,
(2.33)
and for most valid photon shapes does not exceed (Γrκ)/(2g2). In the photon-emission case,
if level |r〉 decays spontaneously to level |e〉, emitted photons will have fluctuating shapes
and their mean overlap v(0) will be reduced. (If ψ1(t) and ψ2(t) are the amplitudes of two
2
photons, their overlap v(0) is defined as the quantity ∫ ∞ ψ 1 (t)ψ ∗2 (t)dt ). It can be shown,
-∞
however, that
Γ rk
v(0) > 1 - , (2.34)
2 g2
and this bound is very loose. Dephasing of level |r〉 at rate Γd can also reduce the overlap
between consecutive photons (by a factor as large as 2Γd/γp) and, to a much lesser extent,
the overall efficiency of the scheme. Note that the effect of losses and noise due to the
transit through level |r〉 are greatly reduced in the strong coupling regime. Interestingly,
even in the intermediate or weak coupling regimes, these effects can be small if the Purcell
factor (4g2)/(Γrκ) is much greater than 1. This is the case of QD-cavity systems fabricated
today by many groups [14,84,85,119].
Therefore, a large vacuum Rabi frequency g is not required for efficient, coherent photon
generation or trapping, even in the presence of realistic experimental imperfections. This
coherent scheme can reduce the impact of dephasing and finite lifetime of level |r〉, and
achieves a high generation efficiency and quantum indistinguishability.
The success of this scheme also depends on the ability to initialize the system into states
|e〉 or |g〉 and on the coherence lifetime of these states. In one possible implementation,
using a neutral QD, states |g〉 and |e〉 are assigned to the bright exciton states, and |r〉 is the
biexciton state. In this case initialization requires a π pulse to excite the QD from its empty
(ground) state to one of the bright exciton states. The coherence lifetime of the bright exciton
states is short (a few hundred picoseconds at best), causing serious degradation. A better
implementation is to use a charged QD or a charged impurity such as a shallow donor.
In singly charged QDs, a three-level Λ system (a system with an excited state coupled by
optical transitions to two ground states) can be obtained by applying a magnetic field per-
pendicular to the growth direction of the QD [52]. In this case |e〉 and |g〉 are assigned
to the two spin sublevels of the single electron, and |r〉 is one of the trion states having
two electrons and one hole. State preparation with fidelity up to 99.8% has recently been
reported in this system using optical pumping [120]. In III-V semiconductors, the electron
spin coherence lifetime is expected to be ~1–10 ns, limited by the random effective magnetic
field produced by nuclear spins in the lattice. Since the nuclear spins fluctuate on a long
timescale, a much longer electron spin coherence lifetime of T2 ~ 1 μs should be possible
using spin echo techniques; see Yao et al. [121] for a recent discussion of electron spin coher-
ence in QDs. These lifetimes are long enough to have little effect on the photon generation
and trapping schemes described here. However, this electron spin T2 sets the limit on how
long quantum information can be stored.
0.05
0
–200 –150 –100 –50 0 50 100 150 200
Control pulse applied
0.8
0.6
Ω(t) (κ)
0.4
0.2
0
–200 –150 –100 –50 0 50 100 150 200
Node populations
1
|e
0.5 |r
Trapping Generation
|g
0
–200 –150 –100 –50 0 50 100 150 200
Time (1/κ)
Figure 2.54
Deterministic photon trapping/generation in the weak-coupling regime. The parameters are g/k = 0.1 and δ = 0.
The photon pulse cannot be as fast as in the intermediate or strong coupling cases, and the upper limit on the
photon bandwidth is g2/k, instead of k. Level |r〉 is also significantly more populated, causing increased loss by
spontaneous emission (~22 Γr/k here if the loss is small) and degradation of consecutive photon overlap. These
undesired effects could still be small in a realistic QD-cavity system with κ~(1 ps)–1, g~(10 ps)–1 and Γr~(1 ns)–1,
for instance, where the quantum efficiency would be 98% and overlap greater than 95%.
–0.2
–0.4
–100 –80 –60 –40 –20 0 20 40 60 80 100
Classical pulse applied
3
2
Ω(t) (κ)
1
0
–1
–100 –80 –60 –40 –20 0 20 40 60 80 100
Node populations
1
|e
0.5
Trapping Generation
|g |r
0
–100 –80 –60 –40 –20 0 20 40 60 80 100
Time (1/κ)
Figure 2.55
Generation and trapping of a composite single-photon pulse for differential phase shift QKD. The pulse is a
superposition of two elementary pulses well separated in time and with a π phase shift between them.
The photon pulse ψ(t) is incoming on a one-sided cavity with decay rate k. We assume
the photon is in a spatial traveling mode perfectly matched with a single confined mode of
the cavity, described by annihilation operator a. Inside the cavity, there is a Λ-type quan-
tum system with two metastable states |g〉 and |e〉 and an excited state |r〉. The transition
|g〉 − |r〉 is coupled to the vacuum field of the confined cavity mode, with vacuum Rabi
frequency g. The |r〉 − |e〉 transition is coupled to the control pulse Ω(t). We will assume an
ideal system where none of the levels decay by spontaneous emission.
The output field of the cavity is related to the input field and the confined field by the
following relation [123] in the Heisenberg picture :
This relation makes the implicit assumption that the coupling of the cavity mode to dif-
ferent longitudinal modes of the waveguide is the same in a frequency range k around
the cavity resonance. This condition might not be satisfied for photonic crystal structures
unless designed carefully (e.g., one should avoid coupling near the waveguide band edge).
The Heisenberg equation of motion for the cavity mode a is
da k
= i[ H Λ , a] - a + k ain (2.36)
dt 2
The first term represents coherent evolution of the confined field due to its coupling to
the Λ system. The second and third terms describe damping and associated “noise” due to
coupling of the cavity photon to a continuum of radiative modes outside of the cavity.
We treat the coupled atom-cavity system as an entity called a “node”, interacting with
the external radiation field. We define a nonnormalized wave-function ψ(t) for the excited
node, i.e., for a node that has absorbed a photon:
where |X, n〉 represents the state X of the Λ system and the number n of photon inside the
cavity. The state ψ(t) evolves according to:
k
G = - ig r - G + k ψ (t), (2.38)
2
Ω
r = - iδ r - igG - i e , (2.39)
2
Ω∗
e = - i r, (2.40)
2
where we used the rotating-wave approximation. The carrier frequency for the photon
pulse is taken to be w c, the resonant frequency of the cavity. We assumed that the two-
photon resonance condition is satisfied, and δ is the common detuning of the laser and
cavity field from the excited level |r〉. From these equations one can show that
t t
This quantity represents the energy absorbed in the atom-cavity system from the incom-
ing photon pulse up to time t. Note that it depends on the detuning implicitly through the
amplitude G(t). We therefore always have the intuitive result that
If and only if Equation 2.43 is satisfied, the system absorbs the photon perfectly, and is
left in state |e, 0〉. Indeed, as t → + ∞, ψ (t) → 0, therefore G(t) → 0. Then, in virtue of rela-
∞
tion 2.38, r(t) → 0. Then, e(t) → ∫- ∞|ψ ( s)|2 ds = 1.
The condition (Equation 2.43) for perfect trapping can be seen as an impedance match-
ing condition and has a straightforward physical interpretation. If we apply Equation 2.35
to the initial state |Ψ0 〉 of the system, and re-adopt the state-varying (Schrödinger) picture,
we find that:
+∞
Limp =
∫|
-∞
k G( s) - ψ ( s) |2 ds . (2.45)
In designing a control pulse, we are looking for a classical pulse Ω(t) that imposes the
impedance matching condition 2.43 on the system. If such a classical pulse exists, it must
satisfy the following equations:
ψ = k G, (2.46)
k
G = - ig r + G , (2.47)
2
Ω
r = - iδ r - igG - i e , (2.48)
2
Ω∗
e = - i r. (2.49)
2
t
Γr
t
∞
∫
| e | = | ψ | - 2 | ψ | | ψ |2
2
kg
22
t
∫ ∫
|ψ|2 Γ r k |ψ - (k/2)ψ|2
- 1 + 2
- , (2.50)
k 2g kg 2
r + δ) -|G|2 Φ
|r|2 (Φ g
e=
Φ 2
. (2.51)
|e|
This solution exists if and only if the right hand side of Equation 2.50 is positive at all
times, which leads to the criterion (Equation 2.25) announced earlier. If this is true, then
|Ω(t)| is bounded on any compact time interval and therefore is a maximal solution of
the above system of differential equations for all times in virtue of the Cauchy–Lipschitz
theorem [124].
Finally, we present solutions for the control pulse Ω(t) in two special cases. In the large-
detuning limit, ∆ >> κ, g, |d log(ψ)/dt|, |Ω|, |d log(Ω)/dt|. In this case level |r〉 has negli-
gible population and can be adiabatically eliminated from the dynamics, reducing the
three-level problem to a simpler two-level one, with an effective laser induced coupling
Ωeff(t) = gΩ(t)/∆. In this case, the trapping control pulse must be
A solution consistent with the condition δ >> Ω can be found only in the strong coupling
regime where g >> κ. In the strong coupling regime this solution simplifies to
2
= -|Ω| ,
Φ (2.54)
4δ
if the single-photon pulse has no chirp (constant phase). This last relation can be simply
interpreted as a compensation of the AC Stark shift created by the far detuned control
pulse on the e–r transition, as was already noted by Cirac et al. [69]. In this large-detuning
regime, fast photon pulses can still be trapped or emitted, the bandwidth being limited by
the cavity decay k only. However, strong coupling is necessary.
In the zero-detuning case the control pulse must have amplitude
Ω(t) 2 - (k/2) ψ + g 2 ψ 2
ψ
= t . (2.55)
∫
2
kg 2 2
ψ (t)| -|ψ - (k/2)ψ|2
|ψ ( s)| ds - g |ψ 2 2
-∞
If the single-photon pulse has no chirp, then neither does Ω(t). Also, we no longer need
strong coupling for the denominator to be positive, and perfect trapping can occur even
if g ~ k.
A special case of interest is that of a slow photon pulse ( | ψ / ψ | k ) in the strong-cou-
pling regime g >> k. In this case we obtain a simple result:
2
Ω(t) g 2 | ψ (t) |2
∼ t
. (2.56)
2
∫
-∞
2
| ψ ( s) | ds.k
This can be differentiated to solve for the photon amplitude in terms of the control pulse:
k Ω(t)
2 k + ∞ Ω( s) 2
|ψ (t)| = 2
g 2
2
exp 2
2g
t
2
ds ,
∫ (2.57)
which is convenient for comparing the envelopes of the control and photon pulses. These
results explain previous experimental observations and simulations of coherent photon
emission via the STIRAP technique [11,117]. In the adiabatic setting used in these experi-
ments, the emitted photon pulse “follows” the control pulse except for a slight correction
given by the exponential term in Equation 2.57, causing the emitted photon pulse to rise
more slowly and decay faster than the control pulse.
2.5 Conclusions
We have shown that single QDs in pillar microcavities can serve as high-quality sources of
single-photons. Through advances in design and processing, we have been able to achieve
strongly modified spontaneous-emission rates, efficient photon collection, and strong cou-
pling between QD excitons and cavity photons. This high performance has enabled basic
applications such as improved quantum cryptography, entanglement generation, and
quantum teleportation. Further advances in such microcavity single-photon sources are
expected to lead to a range of quantum technologies.
References
1. R. J. Glauber. The quantum theory of optical coherence. Physical Review, 130: 2529–2539, 1963.
2. E. Waks, C. Santori, and Y. Yamamoto. Security aspects of quantum key distribution with sub-
Poisson light. Phys. Rev. A, 66(4): 042315, 2002.
3. E. M. Purcell. Spontaneous emission probabilities at radio frequencies. Phys. Rev., 69: 681,
1946.
4. C. Santori, D. Fattal, J. Vučkovic’, G. S. Solomon, and Y. Yamamoto. Indistinguishable photons
from a single-photon device. Nature (London), 419(6907): 594–597, 2002.
5. A. Imamoğ lu and Y. Yamamoto. Noise suppression in semiconductor p-i-n junctions: Transition
from macroscopic squeezing to mesoscopic Coulomb blockade of electron emission processes.
Phys. Rev. Lett., 70(21): 3327–3330, 1993.
6. A. Imamoğ lu and Y. Yamamoto. Turnstile device for heralded single-photons: Coulomb block-
ade of electron and hole tunneling in quantum confined p-i-n heterojunctions. Phys. Rev. Lett.,
72(2): 210–213, 1994.
7. J. Kim, O. Benson, H. Kan, and Y. Yamamoto. A single-photon turnstile device. Nature (London),
397(6719): 500–503, 1999.
8. P. Michler, A. Kiraz, C. Becher, W. V. Schoenfeld, P. M. Petroff, L. Zhang, E. Hu, and A. Imamoğ lu.
A quantum dot single-photon turnstile device. Science, 290: 2282–2285, 2000.
9. C. Santori, M. Pelton, G. Solomon, Y. Dale, and Y. Yamamoto. Triggered single-photons from a
quantum dot. Phys. Rev. Lett., 86(8): 1502–1505, 2001.
10. M. Hennrich, T. Legero, A. Kuhn, and G. Rempe. Vacuum-stimulated Raman scattering based
on adiabatic passage in a high finesse optical cavity. Phys. Rev. Lett., 85: 4872–4875, 2000.
11. A. Kuhn, M. Heinrich, and G. Rempe. Deterministic single-photon source for distributed quan-
tum networking. Phys. Rev. Lett., 89: 067901, 2002.
12. M. O. Scully and M. S. Zubairy. Quantum Optics. Cambridge University Press, Cambridge, UK,
1997.
13. E. Siegman. Lasers. University Science Books, Mill Valley, CA, 1986.
14. J. P. Reithmaier, G. Se˛k, A. Löffler, C. Hofmann, S. Kuhn, S. Reitzenstein, L. V. Keldysh, V. D.
Kulakovskii, T. L. Reinecke, and A. Forchel. Strong coupling in a single quantum dot-semicon-
ductor microcavity system. Nature, 432: 197–200, 2004.
15. M. Pelton, J. Vučkovic’, G. S. Solomon, A. Scherer, and Y. Yamamoto. Three-dimensionally con-
fined modes in micropost microcavities: Quality factors and Purcell factors. IEEE J. Quantum
Electron., 38(2): 170–177, 2002.
16. J. Vučkovic’, M. Pelton, A. Scherer, and Y. Yamamoto. Optimization of three-dimensional
micropost microcavities for cavity quantum electrodynamics. Phys. Rev. A, 66: 023808, 2002.
17. J. Goh, I. Fushman, D. Englund, and J. Vučkovic’. Genetic optimization of photonic bandgap
structures. Optics Express, 15: 8218–8230, 2007.
18. D. Englund, I. Fushman, and J. Vučkovic’. General recipe for designing photonic crystal cavi-
ties. Optics Express, 13: 5961–5975, 2005.
19. W. A. Tiller. The Science of Crystallization: Microscopic Interfacial Phenomena. Cambridge University
Press, Cambridge, UK, 1991.
20. H. Gao. A boundary perturbation analysis for elastic inclusions and interfaces. Int. J. Solids
Structures, 28: 703–725, 1991.
21. L. B. Freund and F. Jonsdottir. Instability of a biaxially stressed thin film on a substrate due to
material diffusion over its free surface. J. Mech. Phys. Solid., 41(7): 1245–1264, 1993.
22. B. G. Orr, D. Kessler, C. W. Synder, and L. Sander. A model of strain-induced roughening and
coherent island growth. Europhys. Lett., 19(1): 33–38, 1992.
23. D. Vanderbilt and L. K. Wickham. Elastic energies of coherent germanium islands on silicon.
Mat. Res. Symp. Proc., 202: 555, 1991.
24. R. People and J. C. Bean. Calculation of critical layer thickness versus lattice mismatch for
GeSi/Si strained-layer heterostructures. Appl. Phys. Lett., 47: 322–324, 1985.
25. J. W. Matthews and A. E. Blakeslee. Defects in epitaxial multilayers. I. Misfit dislocations.
J. Cryst. Growth, 27(1): 118–125, 1974.
26. N. Stranski and L. Krastanow. Akad. Wiss. Wien Math. Naturwiss. Kl. Abt. 2B Chemie, 146: 797,
1937.
27. G. W. Bryant and G. S. Solomon. Optics of Quantum Dots and Wires. Artech House Inc., Boston,
MA, 2005.
28. G. S. Solomon, J. A. Trezza, and J. S. Harris Jr. Substrate temperature and monolayer coverage
effects on epitaxial ordering of InAs and InGaAs islands on GaAs. Appl. Phys. Lett., 66: 991–993,
1995.
29. G. S. Solomon, J. A. Trezza, and J. S. Harris Jr. Effects of monolayer coverage, flux ratio, and
growth rate on the island density of InAs islands on GaAs. Appl. Phys. Lett., 66: 3161–3163,
1995.
30. D. Leonard, K. Pond, and P. M. Petroff. Critical layer thickness for self-assembled InAs islands
on GaAs. Phys. Rev. B, 50: 11687–11692, 1994.
31. W. Wu, J. R. Tucker, G. S. Solomon, and J. S. Harris. Jr. Atom-resolved scanning tunneling micro-
scopy of vertically ordered InAs quantum dots. Appl. Phys. Lett., 71: 1083–1085, 1997.
99. G. Brassard and L. Salvail. Lecture notes in computer science. In T. Hellseth, editor. Advances in
Cryptology–EUROCRYPT ’93. Vol. 765, 410–423. Springer, Berlin, Germany 1994.
100. H. Bennett and D. P. DiVicenzo. Quantum information and computation. Nature (London), 404:
247–255, 2000.
101. Z. Y. Ou and L. Mandel. Violation of Bell’s inequality and classical probability in a two-photon
correlation experiment. Phys. Rev. Lett., 61: 50–53, 1988.
102. J.-W. Pan, D. Bouwmeester, H. Weinfurter, and A. Zeilinger. Experimental entanglement swap-
ping: Entangling photons that never interacted. Phys. Rev. Lett., 80: 3891–3894, 1998.
103. J.-W. Pan, M. Daniell, S. Gasparoni, G. Weihs, and A. Zeilinger. Experimental demonstration
of four-photon entanglement and high-fidelity teleportation. Phys. Rev. Lett., 86: 4435–4438,
2001.
104. S. Popescu, L. Hardy, and M. Zukowski. Revisiting Bell’s theorem for a class of down-
conversion experiments. Phys. Rev. A, 56: R4353, 1997.
105. J. F. Clauser, M. A. Horne, A. Shimony, and R. A. Holt. Proposed experiment to test local
hidden-variable theories. Phys. Rev. Lett., 23: 880, 1969.
106. A. G. White, D. F. V. James, P. H. Eberhard, and P. G. Kwiat. Nonmaximally entangled states:
Production, characterization, and utilization. Phys. Rev. Lett., 83: 3103–3107, 1999.
107. A. Peres. Separability criterion for density matrices. Phys. Rev. Lett., 77: 1413–1415, 1996.
108. T. C. Ralph, A. G. White, W. J. Munro, and G. J. Milburn. Simple scheme for efficient linear
optics quantum gates. Phys. Rev. A, 65: 12314, 2002.
109. T. B. Pittman, M. J. Fitch, B. C. Jacobs, and J. D. Franson. Experimental controlled-not logic gate
for single-photons in the coincidence basis. Phys. Rev. A, 68: 032316, 2003.
110. D. Gottesman and I. L. Chuang. Demonstrating the viability of universal quantum computa-
tion using teleportation and single-qubit operations. Nature (London), 402: 390, 1999.
111. C. H. Bennett, G. Brassard, C. Crépeau, R. Jozsa, A. Peres, and W. K. Wootters. Teleporting an
unknown quantum state via dual classical and Einstein-Podolsky-Rosen channels. Phys. Rev.
Lett., 70: 1895–1899, 1993.
112. J. D. Franson, M. M. Donegan, M. J. Fitch, B. C. Jacobs, and T. B. Pittman. High-fidelity quantum
logic operations using linear optical elements. Phys. Rev. Lett., 89: 137901, 2002.
113. S. Giacomini, F. Sciarrino, E. Lombardi, and F. De Martini. Active teleportation of a quantum
bit. Phys. Rev. A, 66: 030302, 2002.
114. E. Lombardi, F. Sciarrino, S. Popescu, and F. DeMartini. Teleportation of a vacuum-photon
qubit. Phys. Rev. Lett., 88: 070402, 2002.
115. D. Fattal, C. Santori, J. Vučkovic’, G. S. Solomon, and Y. Yamamoto. Indistinguishable single-
photons from a quantum dot. Phys. Stat. Sol (b), 238: 305, 2003.
116. H. J. Briegel, W. Dur, J. I. Cirac, and P. Zoller. Quantum repeater: The role of imperfect local
operations in quantum communication. Phys. Rev. Lett., 81: 5932–5935, 1998.
117. M. Keller, B. Lange, K. Hayasaka, W. Lange, and H. Walther. Continuous generation of single-
photons with controlled waveform in an ion-trap cavity system. Nature (London), 431: 1075–
1078, 2004.
118. W. Yao, R.-B. Liu, and L. J. Sham. Theory of control of the spin-photon interface for quantum
networks. Phys. Rev. Lett., 92: 30504, 2005.
119. D. Englund, D. Fattal, E. Waks, G. Solomon, B. Zhang, T. Nakaoka, Y. Arakawa, Y. Yamamoto,
and J. Vučkovic’. Controlling the spontaneous emission rate of single quantum dots in a 2D
photonic crystal. Phys. Rev. Lett., 95: 013904, 2005.
120. M. Atatüre, J. Dreiser, A. Badolato, A. Högele, K. Karrai, and A. Imamoğ lu. Quantum-dot spin-
state preparation with near-unity fidelity. Science, 312: 551–553, 2006.
121. W. Yao, R.-B. Liu, and L. J. Sham. Theory of electron spin decoherence by interacting nuclear
spins in a quantum dot. Phys. Rev. B, 74: 195301, 2006.
122. K. Inoue, E. Waks, and Y. Yamamoto. Differential phase shift quantum key distribution. Phys.
Rev. Lett., 89: 37902, 2002.
123. D. F. Walls and G. J. Milburn. Quantum Optics. Springer-Verlag, Berlin, 1994.
124. F. Laudenbach. Calcul differentiel et integral. Editions de l’Ecole Polytechnique, France, 2000.
Contents
3.1 Introduction......................................................................................................................... 134
3.2 Fabrication Technique........................................................................................................ 135
3.3 Coupling Techniques.......................................................................................................... 137
3.3.1 Critical Coupling..................................................................................................... 137
3.3.2 Prism......................................................................................................................... 139
3.3.3 Angle-Cut Fiber....................................................................................................... 142
3.3.4 Fiber Taper............................................................................................................... 142
3.3.5 Planar Coupling...................................................................................................... 143
3.4 Modal Structure and Spectrum Engineering................................................................. 144
3.4.1 The Spectrum and the Shape of the Resonator.................................................. 145
3.4.2 White Light Resonators.......................................................................................... 147
3.4.3 Single-Mode Resonators........................................................................................ 151
3.4.4 Elliptical Resonators............................................................................................... 155
3.5 Quality Factor and Finesse of Crystalline Resonators.................................................. 157
3.5.1 Fundamental Limits............................................................................................... 162
3.5.2 Technical Limits...................................................................................................... 164
3.6 Filters and Their Applications............................................................................................. 166
3.6.1 First-Order Filters................................................................................................... 167
3.6.2 Periodical Poling and Reconfigurable Filters...................................................... 169
3.6.3 Third-Order Filters................................................................................................. 171
3.6.4 Fifth-Order Filters................................................................................................... 174
3.6.5 Sixth-Order Filters.................................................................................................. 174
3.6.6 Tuning of the Multi-Resonator Filter................................................................... 175
3.6.7 Resonator Coating Technique............................................................................... 178
3.6.8 Insertion Loss.......................................................................................................... 178
3.6.9 Vertically Coupled Resonators.............................................................................. 180
3.6.10 Microwave Photonics Applications...................................................................... 184
3.6.10.1 Opto-Electronic Oscillator...................................................................... 184
3.6.10.2 Microwave Photonic Receivers.............................................................. 185
3.7 Frequency Stability of WGM Resonators........................................................................ 187
3.7.1 Fundamental Thermodynamic Limits................................................................ 189
3.7.1.1 Thermorefractive Fluctuations: Steady State....................................... 189
3.7.1.2 Thermorefractive Fluctuations: Spectrum........................................... 190
133
© 2009 by Taylor & Francis Group, LLC
3.1 Introduction
Open dielectric resonators with whispering gallery modes (WGMs) are unique because of
their ability to store light in microscopically small volumes for long periods of time. The
small size of the resonator allows densely packaged optical elements, and the intrinsically
high quality (Q) factors of WGMs result in reduced attenuation in photonic circuits.
Evidently, the first indirect observation of WGMs in optics was realized in crystalline
WGM lasers when laser action was studied in Sm:CaF2 crystalline resonators several mil-
limeters in size [1]. Microsecond-long transient laser operation also was observed with a
ruby ring of several millimeters in diameter at room temperature [2]. Transient oscillations
were attributed to pulsed laser excitation of WGMs with Qs of 108–109.
WGMs were first directly observed by elastic light scattering from spherical dielec-
tric particles in liquid resonators [3,4]. It was recognized that these modes could help in
measurement of the size, shape, refractive index, and temperature of nearly spherical
particles [5,6]. WGMs were also used to determine the diameter of an optical fiber [7],
and the strong influence of WGMs on fluorescence [8–10] and Raman scattering [11–13]
was recognized. Laser action in a free droplet was first studied by Tzeng et al. [14] and
Lin et al. [15]. An extensive study of WGM resonators with a focus on their applications
was started when high-Q optical modes were observed in solidified droplets of fused
amorphous materials [16,17], and after an efficient scheme for coupling with the modes
was developed [18–20].
The use of amorphous WGM resonators for practical applications is challenging because
of several reasons. (i) The highest quality factor of WGMs in the resonator is determined by
Rayleigh scattering occurring due to residual surface roughness and inhomogeneities of
the resonator host material [21]. This is generally the case, even though a resonator formed
by surface tension forces has nearly a defect-free surface characterized by molecular-scale
inhomogeneities. (ii) Amorphous resonators are not environmentally stable. For instance,
the quality factor of fused silica resonators degrades with time due to diffusion of atmo-
spheric water into the material. (iii) It is difficult to produce both an optically transparent
and a highly nonlinear amorphous dielectric material.
Despite multiple problems including those mentioned above, amorphous resonators are
widely studied, and various approaches have been developed to mitigate problems associ-
ated with them. Many applications of these resonators have been proposed and some of
them are discussed in other chapters of this book.
The primary goal of our research reviewed in this chapter is to return to the early studies
of crystalline WGM resonators [1,2] and to develop technologies for controllable fabrication
of the resonator, so that various useful properties of crystals that extend beyond the amor-
phous counterparts can be put to advantage. In what follows we show that this goal, for
most parts, has been met. To meet these objectives, we have developed efficient methods of
coupling to the high-Q modes of the resonators. Finally, we have used a crystalline WGM
resonator for practical optical and photonic applications.
[26,30–32]. The high-Q obtained by mechanical polishing was not a miracle, though. Total
internal reflection resonators (another type of open dielectric resonators resembling WGM
resonators) made out of lithium niobate have Q ≈ 7 × 108 (optical path length 1.6 cm, index
of refraction 2.19, finesse F = 6000, at 2 mm) [33]. The JPL group explicitly demonstrated the
capability of mechanical polishing, ultimately producing calcium fluoride WGM resona-
tors with Q > 1011 and showing that the material, not the polishing quality, determines the
achievable quality factor [34].
Mechanical polishing, along with subsequent annealing, is routinely used to produce
ultra-smooth crystalline surfaces. For instance, atomically smooth surface (0.25 nm) on
LiNbO3 and LiTaO3 substrates were produced by thermal annealing of mechanically
polished samples with 1–2 nm roughness [35]. Even better smoothness (0.1–0.2 nm) can
be produced with quartz crystals using soft material polishers, fine abrasive powders,
and suitable working environments, and by carefully reducing the mechanical action
in the polishing process [36]. Mechanical polishing is not the only technique that can
lead to achieving good surfaces. Single point diamond turning is also a very promising
approach. For example, surfaces with 3 nm roughness have been achieved with diamond
turning of CaF2 [37].
We adopted various known techniques to fabricate WGM resonators out of different
types of crystals. The basic procedure is the following. We start with a crystalline wafer
and drill the wafer with a hollow drill to fabricate a cylindrical preform. The preform is
then attached to a holder and is either polished or diamond turned. There is no unique
recipe for production of a perfect surface. We have achieved ultra-high quality factors by
selecting the polishing speed, and the size of the grain in a diamond slurry. As a rule, all
the algorithms change from a crystal to a crystal, and from a crystal cut to a crystal cut.
To date, the basic achievement is the absence of the Rayleigh scattering in our resonators,
which is the basis of the claim that the surface roughness does not restrict the quality fac-
tor of crystalline resonators.
Hand polishing works well for fabrication of larger resonators. For resonators with
diameters as small as tens of micrometer, diamond turning technique is more suitable.
The diamond turning technique we used for fabrication of crystalline microresonators
is based on a home-made apparatus [38], and allows precise engineering of geometrical
features of the resonator such as surface curvature or profile. We have fabricated magne-
sium fluoride and calcium fluoride resonators to validate the technique. The quality of the
surface of resonators produced with this apparatus is quite high after a final polishing,
and is characterized with less than 0.2 nm surface roughness. The quality factor of the
microresonators fabricated with CaF2 exceeds 108.
The entire fabrication process of microresonators generally includes two steps. The first
step is diamond turning, which employs computer control of a precision lathe. Structures
obtained in this step are engineered to about 40 nm precision and have optical Q fac-
tors of up to 107 (surface quality limited). For instance, a chain of three CaF2 resonators
presented on Figure 3.1 was fabricated with this approach. Resonators are separated by
approximately 70 mm and are coupled to each other for efficient energy transfer. If higher
Q factors are needed, additional optical polishing has to be performed. This polishing step
naturally modifies the structure that is initially obtained by the diamond turning process.
Special diamond turning parameters must be employed to avoid brittle machining and
to achieve the required surface smoothness. Parameters such as speed of rotation of the
workpiece, feed and cutting angle, diamond cutter geometry and sharpness, as well as
the type of lubricant are very important in achieving ductile regime of machining [39].
Special attention should also be paid to vibration isolation of the turning process. In our
Figure 3.1
A stack of three WGM resonators fabricated by diamond turning technique on a cylindrical preform made out
of calcium fluoride.
diamond turning setup we use a home-made air bearing to provide the required stiffness
and repeatability of rotation of the workpiece. Optical polishing is performed by applica-
tion of polycrystalline diamond abrasives. Polishing is performed in several steps with
decreasing diamond grit size followed by a cleaning process. It was found that small par-
ticles are extremely difficult to remove from the surface of the material, and the cleaning
process is most critical in achieving ultrahigh Q factor. Cleaning was performed in a clean
room environment with the use of organic solvents.
A A
Aout Aout1
α α/2
A in A in
Figure 3.2
Models of WGM resonators.
in RF engineering (see e.g. [41,42]), has been recently applied to optical WGM resonators
[43]. Indeed, unlike lossless Fabry–Perot (FP) resonators where intrinsic loss is identi-
cal to the coupling loss, the strength of evanescent wave coupling to WGM resonators is
independent of their intrinsic loss. Criticality implies that the coupling strength between
waveguide and resonator must match the loss of any nature to provide a 100% energy
exchange at resonance.
Let us model a solid state WGM resonator as a ring cavity containing a linear absorber local-
ized in space (Figure 3.2a). To find the intracavity field A we write the steady state condition
where a << 1 is the (frequency independent) total amplitude absorption of the light in the
resonator per round trip, Tc << 1 is the coupling factor, L = 2pR is the circumference of the
resonator, R is the radius of the resonator, A is the slow amplitude operator of the intracav-
ity light, and n is the refractive index of the material. Equation 3.1 indicates that the steady
state of the field amplitude is achieved by compensation of the round trip loss of the exter-
nal optical pumping Ain.
The amplitude of the output pump light Aout can be found from equation
and it is equal to
1 - Tc - exp[iω Ln / c]exp[-a]
Aout = Ain . (3.3)
1 - exp[iω Ln / c]exp[-a] 1 - Tc
ac T c Ln Ln Ln
γ= , γc = c , ω = (ω - ω 0 ) + 2 πl, 2 πl = ω 0 ,
Ln 2L n c c c
where ω0 is the mode frequency, l is the mode number. In the vicinity of the resonance we
have
γ c - γ - i(ω 0 - ω )
Aout = - Ain . (3.4)
γ c + γ + i(ω 0 - ω )
Equation 3.4 shows that all the power is absorbed in the resonator (Aout = 0) if γc = γ. This is
the condition of the criticality of coupling.
In the case of a WGM resonator with two couplers (“add-drop” configuration, Figure
3.2b) the criticality condition looses its meaning. In fact, the intracavity field in the case of
two couplers is calculated from
where Tc1 << 1 and Tc2 << 1 are the coupling factors, for two partially transparent mirrors.
Output fields are described by equations
γ c 1 - γ c 2 - γ - i(ω 0 - ω )
Aout 1 = - Ain , (3.8)
γ c 1 + γ c 2 + γ + i(ω 0 - ω )
2 γ c 1γ c 2
Aout 2 = - Ain , (3.9)
γ c 1 + γ c 2 + γ + i(ω 0 - ω )
in the vicinity of the resonance. We have taken into account that eiω Ln/2c ≈ − 1 in this case. It
is easy to see now that the filter is nearly ideal if γc1 = γc2 >> γ. The criticality of the coupling
does not make sense here.
3.3.2 Prism
Prism coupling was initially developed to send light to optical waveguides [44–46]. Prism-
waveguide coupling efficiency exceeding 90% has been demonstrated elsewhere [47,48].
Prism coupling to WGMs has been investigated both theoretically and experimentally
[18–20,49–51]. The best efficiency of prism coupling to WGMs in microspheres reported to
date is ~ 80% [49]. A coupling efficiency exceeding 97% was achieved in elliptical lithium
niobate resonators [52]. In this chapter we report on achieving 99% overall nonlocal cou-
pling with a toroidal WGM resonator having 4.2 mm in diameter.
The technique of prism coupling is based on three main principles. First, the input beam
is focused inside the prism at an angle that provides phase matching between the eva-
nescent wave of the total internal reflection spot, and the WGM, respectively. Second, the
beam shape is tailored to maximize the modal overlap in the near field. And third, the gap
between the resonator and the prism is optimized to achieve critical coupling. The loaded
quality factor for the prism coupler in the case of the basic WGM sequence of microresona-
tors is given by the expression (see [20]):
π(nr2 - 1)m3/ 2 d
Qc = exp 4 π nr2 - 1 , (3.10)
2 nr 2
n -n
2 λ
p r
WGM
resonator
Diamond
prism
Collimators
Optical fibers
Figure 3.3
Schematic of prism coupling with a WGM resonator.
where nr is the refractive index of the resonator, np is the refractive index of the prism, d
is the shortest distance from the prism to the resonator, l is the wavelength of the light
in the vacuum, and m is the azimuthal number of the mode. It is easy to see that np > nr is
expected.
The schematic diagram of a prism coupler is shown in Figure 3.3. The output of a single
mode fiber is collimated and focused on the surface of the prism. The prism is located in
the close proximity of the WGM resonator, and the collimated beam is focused on the con-
tact spot between the resonator and the prism. The refractive index of the prism np is larger
than the refractive index of the resonator nr, and the phase velocity of the collimated beam
along the surface of the prism is c/np sin φ, where c is the speed of light in vacuum, and φ is
the angle of incidence. The phase velocity of the optical WGM is equal to c/nr. This trivial
approximation is good enough in practice. The modes are phase-matched when
sin φ nr / np . (3.11)
The phase matching of the modes is not the only factor required for efficient coupling. For
example, when the focused beam is not adjusted with the prism–resonator contact point,
then there is no coupling at all since there is no aperture matching. Aperture matching is
the second important factor for achieving an efficient coupling. Aperture matching with a
Gaussian beam occurs when
r
cos φ , (3.12)
R
where r is the vertical radius of curvature of the resonator in the vicinity of the contact
spot, and R is the horizontal radius of curvature of the resonator next to the contact spot.
Equations 3.11 and 3.12 show that phase and aperture matching are generally impossible
to obtain in the case of coupling with the basic WGM sequence in a microsphere. This is
why the best achieved coupling with a microsphere is on the level of 80% [49].
It is possible though to improve the coupling by varying both the incidence angle and
some other parameter. Two choices are available for this. One choice is to change the verti-
cal radius of curvature of the resonator, r, i.e. make a toroidal resonator. The resonators rim
must be shaped such that
nr 2
r = R 1 - . (3.13)
np
r
nr = np 1 - . (3.14)
R
This can be done by selecting modes located deeper in the resonator since
( )
1/ 3 -1
m
nr > nreff nr 1 + a q , (3.15)
2
where aq is the qth root of the Airy function, Ai(–z), which is equal to 2.338, 4.088, and
5.521 for q = 1, 2, and 3, respectively. Practically speaking, these approaches are both used
together. At the onset, the resonator is made so that its curvature parameters satisfy the
condition, Equation 3.13. Possible fabrication errors can be fixed by selection of a deeper
mode.
It is possible to realize aperture matching with aspherical (astigmatic) optics. An
axio-symmetric Gausian beam can be transformed to an elliptical beam that has per-
fect overlap with modes of an arbitrary WGM resonator. This method is now under
development.
The third factor limiting the coupling efficiency is intermodal mixing. WGM modes in
larger resonators overlap and interact with each other. The mode of interest emits into
other modes limiting the maximum achievable optical coupling (in the sense of the criti-
cal coupling). To suppress this effect the number of modes should be reduced by, for
example, shaping of the resonator [53]. However, it is possible to achieve excellent cou-
pling even in a toroidal multimode resonator. For example, we have measured 99% non-
local coupling in a toroidal WGM resonator having 4.2 mm in diameter and 120 mm in
thickness Figure 3.4.
0.08
Voltage on a linear detector (a.u.)
0.07
0.06
0.05
0.04
0.03
0.02
0.01
99% optical coupling
0.00
200 400 600 800
Frequency (MHz)
Figure 3.4
Spectrum of a WGM resonator taken using a prism coupler. The coupling efficiency approaching 99% is
achieved.
WGM
resonator
Single-mode
fiber
Figure 3.5
Angle-polished fiber coupler for WGMs.
WGM
resonator
Tapered fiber
Figure 3.6
Tapered fiber coupler for WGMs.
allowing simple focusing and alignment of the input beam, as well as collecting the output
beam. It saves the fundamental mode and filters other waveguide modes. The efficiency of
tapered fiber couplers reaches 99.99% for coupling fused silica resonators [65]. Similar cou-
plers were also used in photonic crystal resonators [66,68]. Unfortunately, fiber tapers are
unsupported waveguides without cladding and are very fragile. They are only useful for
resonators with refractive indices similar to silica (1.4–1.45), and cannot generally be used
with higher index glass and crystalline resonators. A good coupling achieved with con-
ventional fiber tapers, though, has been demonstrated with very small silicon resonators
[71]. The coupling is possible in this case because phase matching is not very important if
the size of the resonator is comparable with the wavelength.
The phase velocity depends on the taper’s diameter and the refractive index. The diam-
eter of the taper gradually decreases towards the tip, and so does the effective refractive
index. By placing the resonator at a proper point of the taper, it is possible to achieve
the ultimate phase matching and critical coupling. Practically speaking, this technique is
limited by available fibers since the effective refractive index of the taper can be changed
only between values of refractive index of the air/vacuum and the refractive index of the
material the fiber is made of. When the resonator is made of lithium niobate or diamond,
for example, it is not trivial to make a phase-matched taper, though the planarized silicon
waveguide would likely do the trick.
the minimum the potential moves into the cavity [91,93]. The deeper the minimum, the
better the potential can be described by the oscillatory function.
Both methods of spectral engineering mentioned above can be applied to crystalline
WGM resonators. The index of refraction of resonators can be modified by doping of the
host material. This, however, is a complex task, and the modification of the resonator shape
to obtain a desired WGM spectrum appears much simpler. In what follows we provide
several examples of how shaping the WGM resonators changes their spectra.
where k = ω/c is the wave number, ε(r) is the coordinate dependent dielectric susceptibility,
E is the electric field of the mode, and r = (z, r, φ) is the radius vector.
We assume that the resonator is axially symmetric and that its radius R(z) weakly
depends on z, the axis of symmetry. Using the adiabatic approximation along the z axis
we present the scalar solution of Equation 3.16
∫
± i k z dz + mφ r
Ψ (r ) ∼ e
Φ , (3.17)
R( z)
where
( )
1/ 3
1 m
km ,q m + aq (3.19)
nR( z) 2
is the radial wave number defined by the equation for the radial field distribution Φ
∂2 Φ 1 ∂ Φ 2
2 +
∂ x x ∂x
+ km ,q -
m 2
x
Φ = 0. ( ) (3.20)
The light is confined along the z axis if the corresponding component of the wave vector
(kz) has two zeros, z1 and z2 (turning points), and is positive in the interval z2 > z > z1. Then
the Wentzel–Kramers–Brillouin (WKB) quantization along the z axis leads to the equation
that defines the eigenvalues km,q,p (see e.g. [90])
z2
1
∫ - km2 ,q = π p + .
1/ 2
2
k (3.21)
m , q , p 2
z1
Such separation of variables as well as the expression for the resonator spectrum (km,q,p) is
generally valid if [90,94]
dR
kz R << 1, z2 > z > z1 . (3.22)
dz
Condition 3.22 restricts the class of resonators that can be analyzed with this technique.
However, we can claim that for that class of resonators a selection of the shape of the sur-
face R(z) allows modification of the resonator spectrum. For instance, it helps to evaluate
the shape of a WGM resonator with an equidistant spectrum [90].
The originally proposed spherical WGM resonators are “over-moded”, with a com-
plex quasi-periodic spectrum and unequal mode spacings. A significant reduction in the
mode spectral density is achieved in highly oblate spheroidal microresonator (microtorus)
[88,89]. Our experiments also show that changing the shape of a millimeter-sized crystal-
line dielectric resonator results in the rarefication of the observed WGM spectrum without
destroying the quality factor.
As an example, we have measured [95] WGM spectra of disc resonators possessing
spherically (Figure 3.7a), toroidally (Figure 3.7b), and conically shaped rims (Figure
3.7c). The toroidal geometry, with its transverse curvature diameter nearly equal to the
thickness of the disk, resulted in a significant rarefaction of the spectrum compared to
a spherical layer (separate modes are visible in Figure 3.7b as opposed to conglomer-
ates in Figure 3.7a). However, with this geometry it was difficult to further increase the
transverse curvature to eliminate all but one WGM per free spectral range (FSR): disks
smaller than 100 mm in thickness were hard to polish into the toroidal shape. As an
alternative to further improving the spectral regularity, we produced a conical resonator
that was subsequently polished at the rim into a very sharp bend toroid with osculating
curvature diameter of ~ 20 mm (Figure 3.8). The spectrum proved to contain only one
major mode (distinct successive peaks) (Figure 3.7c). A high Q of 5 × 107 was obtained
with this resonator.
It is useful to present here an analytical expression for the spectrum of an arbitrary
spheroid found in [96]
5 10 15 20 25 30 0 5 10 15 20 0 5 10 15 20
Laser frequency detuning (GHz)
Figure 3.7
Spectrum of a disk LiNbO3 resonator with (a) spherically shaped rim, (b) with torridally shaped rim, (c) with
conically shaped rim.
Figure 3.8
Conically shaped resonator spectrum of which is shown in Figure 3.7c. (From Savchenkov, A.A., Matsko, A.B.
and Maleki, L., “White-light whispering gallery mode resonators”, Opt. Lett. 31, 92–94 (2006)).
1/ 3
l 2 p(R - r ) + R χn 3a 2q l -1/ 3
nk p ,q ,l R l + a q + - +
2 2r n2 - 1 20 2
where R and r are equatorial and polar radiuses, kp,q,l is the wavenumber, p = l − |m| = 0, 1, 2,. . .
and q = 1, 2, 3,. . . are integer mode indices, and c = 1 for quasi-TE and χ = 1/n2 for quasi-TM
modes. The basic sequence of WGMs is given by p = 0 and q = 1.
Equation 3.23 does not show any actual rarefaction of the spectrum of a spheroidal reso-
nator. It shows rather the removal of degeneracy by quantum numbers m and l. Hence, the
results reported in Figure 3.7 come from the selectivity of coupling in a particular resona-
tor, not from the real spectrum rarefaction. An additional reason for the clean spectrum
of oblate spheroids is the reduction of the quality factors of the high-order modes. In what
follows we discuss possible ways for obtaining real “single-mode” resonators beginning
with a description of their opposite counterparts, “white light” resonators.
in between the resonator mirrors to cancel the variation of wavelength with frequency
[99,100]. This would make the resonator simultaneously resonant for all frequencies, i.e.
result in “whitening” the spectrum of the resonator. Such resonators were suggested for
improving the sensitivity for gravitational wave detection.
The total achievable spectral width of the existing white light resonators is restricted by
the spectral characteristics of the resonator mirrors. Usually the mirrors have good reflec-
tivity within several nanometers of a selected optical frequency.
We theoretically proposed and experimentally demonstrated essentially “white light”
WGM resonators in a frequency range exceeding a full octave [101]. The basic characteris-
tics of the resonator are (i) power build-up at virtually any frequency inside the transpar-
ency window of the material it is made of; (ii) nearly the same Q-factor (ring-down time) at
any frequency in the transparency window, and (iii) nearly critical coupling with cleaved
fibers at any frequency in that window. These resonators can be used in cavity ring-down
spectroscopy as an alternative for FP white-light cavities. The white light WGM resonators
made of photorefractive materials also can be used in electro-optical modulators.
Let us explain the basic idea of the fabrication of the white light WGM resonator. The spec-
trum of WGMs in a finite cylinder of radius R can be approximated by Equation 3.21. The
frequency distance between modes in any selected radial mode family is characterized with
the FSR divided by the ratio of the resonator thickness and wavelength. The spectrum km,q,p
of a thick enough WGM resonator is very dense because changing the quantum number P
does not change the mode frequency significantly. The Q-factors of various modes generally
do not depend on the mode numbers, so all the modes have nearly identical Qs. Hence, a
thick cylindrical or a prolate toroidal resonator can be considered as a white-light cavity.
We fabricated a high-Q CaF2 WGM resonator (Q > 109) and loaded it heavily. The resona-
tor had 0.5 mm thickness and 0.5 cm diameter. A proper coupling technique, e.g. a cleaved
fiber (seen in Figure 3.9) readily results in the excitation of majority of the modes. The
coupling efficiency was better than 40% for all frequencies at which the lasers could be
tuned. The resonator was tested with 1320 nm (4 kHz linewidth), 780 nm (10 kHz line
width) lasers, and 532 nm (incoherent) light. The spectrum of the resonator is essentially
Fiber coupler
Eout
Resonator
Ein
Figure 3.9
White light fiber coupled whispering gallery mode resonator. The resonator is thick (500 μm) which results
in the coexistence of various mode families in the same geometrical volume in the vicinity of the rim of the
resonator. The continuous spectrum of the resonator is achieved because the fiber coupler is slightly tilted with
respect to the center of the rim of the resonator to maximize interaction with all the modes. (Reprinted from
Savchenkov, A. A., Matsko, A. B., and Maleki, L., Opt. Lett. 31, 92–94, 2006. With permission from The Optical
Society of America (OSA).)
0 20 40
0.5 780 nm
0.2
0.1
0.0
0 200 400
Frequency detuning (GHz)
Figure 3.10
Spectra of the white light resonator at 1320 nm and 780 nm. Critical coupling corresponds to Pout/Pin = 0, no
coupling – to Pout/Pin = 1. (Reprinted from Savchenkov, A. A., Matsko, A. B., and Maleki, L., Opt. Lett. 31, 92–94,
2006. With permission from OSA.)
0.5
Input
Output
0.0
0 300 0 300 0 300
Time (ns)
Figure 3.11
Normalized power of the input and output pulses of light taken at various carrier frequencies. (Reprinted from
Savchenkov, A. A., Matsko, A. B., and Maleki, L., Opt. Lett. 31, 92–94, 2006. With permission from OSA.)
continuous if the frequency difference between modes is much smaller than their spectral
width (Figure 3.10). The average distance between modes in a radial mode family was less
than 20 MHz. We simultaneously excited several radial mode families.
To measure the quality factor at 1320 and 780 nm we used a ring-down technique with
a τ1 = 5 ns shutter. For the green light we used an acousto-optic shutter with τ1 = 31 ns
response time. Characteristic response times of the photodiode and circuits were less
than 1 ns. The shape of the light pulse on the entrance and exit of the resonator are shown
in Figure 3.11. The average ring-down time was τ0 = 65 ns, which corresponds to Q ~ 108.
Variations of the Q-factor with the carrier frequency of the lasers were less than 50%.
Therefore, the resonator has all the necessary “white light” properties in a frequency
interval larger than an octave.
Ring-down tails in Figure 3.11 have a small amplitude modulation. The modulation
results from a slight loss of orthogonality between the modes of the resonator resulting
from the coupling technique, and the beating between the modes. Changing the shape of
the resonator allows suppression of the beats. Mode beats also disappear if the coupling
with the resonator increases (Q-factor decreases).
Let us now discuss possible applications of this white light resonator. One application is
a wide bandwidth electro-optic modulator. An approach to implement coupling between
light and microwave fields in a WGM resonator was recently proposed. In that study, an
efficient resonant interaction of several optical WGMs and a microwave mode was achieved
by engineering the shape of a microwave resonator coupled to a micro-toroidal optical
cavity. Based on this interaction a new kind of electro-optic modulator (as well as photonic
microwave receiver) was suggested and realized [31,98]. The modulation frequency in the
modulator is determined by the FSR of the resonator.
The problem of the application is in the necessity of lasers with very narrow linewidth,
much smaller than the spectral width of the WGMs in the electro-optic resonator. The laser
must be locked to a selected WGM to ensure continuous modulation. Application of white-
light cavities would relax those requirements, as mentioned above. As an example, let us con-
sider two lasers with different properties used in WGM electro-optical modulators: (i) a laser
with a narrow linewidth but large frequency drift and (ii) a laser with a large linewidth.
Narrow linewidth lasers interact with essentially a single WGM mode at each instant of
time. The modes naturally belonging to different mode families change with time, depend-
ing on the drift of the laser. The microwave field is applied to the WGM cavity using a
proper strip-line microwave resonator placed along the rim of the cavity. The microwave
field couples WGMs of the same mode family, and does not couple modes from differ-
ent families. For instance, if the carrier frequency of the laser light coincides with a WGM
characterized with wave vector km,q,p, the microwave radiation can excite modes with
km-m1,q,p and km+ m1 ,q , p only, where m1 is an integer number. Those modes, irrespective of which
family, are nearly equidistant and their volumes and Q-factors are nearly equal, especially if
the WGM resonator is large enough. Therefore, at each instant of time the WGM resonator
can be considered as being “single mode”, and the modulator operates unaffected by the
frequency drift of the laser. There is no need to lock the laser to a mode of the resonator.
In the second case, a laser with a large linewidth simultaneously interacts with several
modes belonging to different mode families in the white-light cavity EOM. Again, the
modulation process “shifts” the photons from the carrier frequency to the frequency of
the sidebands based on the same mode family interaction. If the different mode families
were not equidistant, or they had different Q-factors, the noise in the sidebands would
increase. However since the modes have identical properties, we expect no deterioration
of the modulation.
Another application of the white light WGM resonators is in cavity ring-down spectros-
copy, as mentioned above. WGM resonator-based chemical and biosensors were previ-
ously proposed and widely discussed [102–115]. Measurement of the material absorption
as a function of frequency is one of the ways to detect specific chemical or biological sub-
stances. If the resonator surface is covered with a layer of the material to be studied, the
quality factor decreases, and the ring-down time of the resonator τ changes as
-1
V ξ
τ = τ 0 1 + ef (3.24)
Vr ξ r
where τ0 is the ring-down time of the resonator in vacuum, Vef is the effective volume in the
added substance, Vr is the volume of the mode in the resonator, ξ is linear absorption per
round trip of the material under study, and ξr is absorption of the material of the resonator
per round trip. We neglected scattering in this expression.
White light WGM resonators, in the same way as white light FP resonators, can be useful
for spectroscopy purposes. White light resonators allow tuning the laser interacting with
the resonator instead of tuning the resonator, which can significantly simplify the experi-
ments and the requirements for the laser characteristics could also be relaxed.
of the refractive index of the fiber material. The core of the fiber has a larger index of
refraction than the cladding material that surrounds it. The difference of the indices is
small, so only one mode propagates inside the core, while the others decay into the clad-
ding (Figure 3.12A). For instance, the condition for the single mode operation of a planar
waveguide is [118]
λ
dco < , (3.25)
2 ∆ε
where dco is thickness of the core, λ is wavelength of light in vacuum, and ∆ε is the differ-
ence between the susceptibilities of the core and the cladding material, respectively. As a
result, the core may have a reasonably large diameter. Note that the core of a single-mode
fiber is a multi-mode fiber if the cladding is removed.
Let us consider the WGM resonator as a multi-mode gradient waveguide (Figure 3.12B)
[119]. The resonator becomes an ideal single-mode-family resonator only if the waveguide
is thin enough. Following this trivial approach the WGM resonator should be designed
as an approximately half-wavelength-thick torus to support a single mode family. Recent
experiments confirmed this conclusion and nearly single-mode resonators have been dem-
onstrated [89].
There is another, nontrivial, approach to the problem based on the analogy with single-
mode optical fiber. A WGM resonator made of any transparent material with any size can
be transformed into a single-mode resonator if the appropriate geometrically defined “core
and cladding” are developed (Figure 3.12C).
To show that WGM modes can be described using a waveguide formalism, we start with
the usual wave equation (Equation 3.16). Higher-order WGMs (i.e. those with the wave-
length much smaller than the resonator size) of both TE and TM modes are localized in
A B
C
Eout
Ein
Figure 3.12
(A) A structure and mode localization in the optical fiber. Only one propagating mode, shown by the solid line,
survives in the core of the fiber, while others (dotted line) penetrate into the cladding and dissipate. (B) Mode
localization in the whispering gallery mode resonator. The resonator corresponds to a multimode gradient fiber
where the index of refraction is set by the resonator shape, not by the change of the refractive index of the reso-
nator host material, which is constant. Both the fundamental and auxiliary modes survive. (C) Mode localiza-
tion in a low-contrast whispering gallery mode resonator. Only a single-mode family survives. The other modes
penetrate into the cylinder rod and dissipate. The coupling to the single mode family is achieved with, e.g. a
cleaved fiber coupler. (Reprinted from Savchenkov, A., Grudinin, I. S., Matsko, A. B., Strekalov, D., Mohageg, M.,
Ilchenko, V. S., and Maleki, L., Opt. Lett. 31, 1313–1315, 2006. With permission from OSA.)
A B
3 µm
30 µm
3 µm
Figure 3.13
(A) Scanning electron microscope image of the resonator whose spectrum is shown in Figure 3.14. The resonator
has nearly Gaussian shape with 2.5 μm height and 5 μm full width at the half maximum. (B) The image of the
profile of the resonator shown in (A). (C) Intensity map of the field in the resonator shown in (A) simulated by
numerical solution of Equation 3.26. (Reprinted from Savchenkov, A., Grudinin, I. S., Matsko, A. B., Strekalov, D.,
Mohageg, M., Ilchenko, V. S., and Maleki, L., Opt. Lett. 31, 1313–1315, 2006. With permission from OSA.)
the vicinity of the equator of the resonator. For this problem, cylindrical coordinates can be
conveniently used. Applying the technique of separation of variables, and assuming that
the resonator radius changes as R = R0 + L(z) (R0 >> |L(z)|) in the vicinity of the equator, we
transform Equation 3.16, for TE mode family, to
∂ 2E ∂ 2E 2 L( z) m2
+ + k ε 1 + 2 - E = 0, (3.26)
∂r 2 ∂z 2 R0 r 2
where m is the angular momentum number of the mode (we assumed that m >> 1), E(r,z)
is the scalar field amplitude, ε is the susceptibility of the resonator host material. This
equation is similar to the gradient waveguide equation. It is easy to see that, for instance,
modes of a spherical WGM resonator coincide with modes of a gradient waveguide with
parabolic distribution of refractive index in the z direction. Hence, it is the geometry of the
surface that should be modified to produce an ideal single-mode WGM resonator. A “core”
for the WGM “waveguide” can be realized by the proper design of the resonator surface in
the vicinity of the equator. The rest of the resonator body plays the role of the “cladding”
(Figure 3.1C).
Consider a resonator consisting of a cylindrical drum and a small, ring like protrusion,
L(z) = L0 for d ≥ z ≥ 0, on its surface. The drum’s effective susceptibility does not depend on
the z coordinate and is equal to ε. The effective susceptibility of the ring, ε(1 + 2L0/R0) (see
Equation 3.26), is slightly larger. Therefore, the ring is the core that confines the light in the
z direction, while the drum is the cladding. The condition for “single mode” operation of
the resonator using Equation 3.25 is:
d 2 L0ε 1
1> > . (3.27)
λ R0 2
0 (a) TM
–10
0 (b)
–10
–20
–30
0 5 10 15 20 25 30
Frequency (GHz)
Figure 3.14
(A) The spectrum of a nearly ideal single-mode resonator obtained using a 980 nm laser as well as the cleaved
fiber couplers (Figure 3.12C). The mode number is approximately m = 3 × 104. The low-Q modes on the spectrum
background belongs to the drum. (B) The spectrum of the multimode resonator made on the same drum (2.5 μm
height and 15 μm width) and detected using the same coupling technique. The resonator is partially visible in
the lower right corner of Figure 3.13A. (Reprinted from Savchenkov, A., Grudinin, I. S., Matsko, A. B., Strekalov,
D., Mohageg, M., Ilchenko, V. S., and Maleki, L., Opt. Lett. 31, 1313–1315, 2006. With permission from OSA.)
This condition (Equation 3.27) stays valid for a resonator with an arbitrary large radius.
Both the width and the height of the ring can be much larger than the wavelength of light.
The ratio L0/R0 plays the same role as the ratio ∆∈/∈ in an optical fiber.
To demonstrate the single-mode operation experimentally, we built such a resonator by
using a two-step fabrication process that is described in one of the previous sections of
this chapter. We fabricated a CaF2 rod of 5 mm in diameter. WGMs in such a rod have
extremely dense spectra. After that we fabricated a small ring with dimensions of the
order of several microns on the surface (Figure 3.2).
We used an angle-polished fibers to couple light into the resonator. The spectrum
shown was obtained with two angle-polished fiber couplers, one for the input and one
for the output. The fiber couplers were made with standard single mode fiber with clad-
ding diameter of about 150 µm and core of about 5 µm. The photodetector signal rep-
resents transmission of the resonator. The two families of modes shifted by a few GHz
were observed, one for each polarization. One of such families is shown in Figure 3.14A.
Logarithmic detector measurements have shown that there were no other modes down
to –30 dB level, only the two mode families. It should be noted that, given the geometrical
parameters of the resonator and the couplers, the single mode regime of this resonator
is not a function of coupling but an intrinsic property. Moreover, the logarithmic mea-
surements were performed in a loaded configuration, that is, when the couplers are in
contact with the resonator. We also confirmed that this was a single-mode resonator by
performing numerical simulation of its parameters which clearly demonstrated that only
the fundamental modes survive. Intrinsic Q factors of both families of modes shown in
Figure 3.14A are on the order of 107 (loaded Q is 2 × 106, Figure 3.15). The particular reso-
nator’s Q-factor was limited by residual surface roughness.
With this approach, one can make a single-mode optical WGM resonator of any size. A
resonator the size of an apple requires a ring with dimension of tens of microns for single
0
150 MHz FWHM
–5
Single mode
–20
–25
–30
0 2 4 6 8 10
Frequency (GHz)
Figure 3.15
A Lorentzian fit of a WGM spectrum of which is shown in Figure 3.14A.
Figure 3.16
Elliptical LiNbO3 resonators. (Reprinted from Mohageg, M., Savchenkov, A., and Maleki, L. Opt. Express. 15,
4869–4875, 2007. With permission from OSA.)
mode propagation of one micron wavelength light. This is counter-intuitive because the
size of the single-mode channel is much larger than the wavelength. This type of experi-
ment may be interesting as a fundamental research investigation. For practical applica-
tions small resonators are of more interest. In this context, our approach leads to a novel
means for engineering of microcavity spectra.
The type of the dielectric media used for the proposed resonators is not important, as
long as it is solid and transparent. Generally, since the behavior of the system is defined
only by its geometry, it is not even important for what kind of waves the structures is a
resonator. Microwave, acoustical or mechanical chains of resonators of this type have the
same features and basis of design as these optical counterparts.
3.4.4 Elliptical Resonators
Mechanical fabrication that allows engineering the shape of the rim of a toroidal resonator,
also enables producing elliptical and other kinds of geometrically shaped symmetric reso-
nators. These resonators can retain high-Q WGMs, and also can possess unstable low-Q
modes. Elliptical high-Q lithium niobate WGM resonators (see Figure 3.16) were studied
elsewhere [52]. It was shown that the elliptical geometry of the resonator rim simplifies the
coupling procedure. Coupling efficiency exceeding 97% was demonstrated.
Unlike modes in race track resonators, WGMs in elliptical resonators are stable. Their
approximate spectrum can be found by means of the eikonal method [52]. Assuming a
closed resonator we find for the high-order modes
where E(1, ε) is the complete elliptic integral of the second kind. Equation 3.28 has two
parameters that define the geometry of the ellipse: eccentricity ε and semimajor axis
length R.
Obtaining a rigorous expression for the modal spectra of elliptical resonators is rather
involved. The determination of WGM spectrum in a spherical resonator is an easy task
because the Helmholtz equation can be solved exactly by separation of variables. This is
not the case for an elliptical resonator. A detailed mathematical analysis of the WGMs in
elliptical resonators with large size compared to the resonant wavelength can be found
in [121,122]. Waveguide coupling to elliptical WGM resonators were studied numerically
[123]. It is worth noting that laser emission from ellipsoidal resonators was studied experi-
mentally [124].
The fabrication procedure of the crystalline elliptical resonators is straightforward.
A piece of crystal in the shape of a baseball diamond was cut from a Z-cut wafer of congru-
ent LiNbO3. The lengths of the axes of the diamond were selected to be slightly larger than
the target dimensions of the desired ellipse. The edges of this structure were mechani-
cally ground until the sharp edges of the diamond shape were smooth in a continuous
curve resembling an ellipse. At this point, mechanical polishing techniques were applied
to shape the edge curvature and polish the edge surfaces.
Three optical resonant cavities of elliptical cross section are shown in Figure 3.16. The
eccentricity of the first cavity is 0.57. The lengths of the semimajor and semiminor axes are
4.4 and 3.6 mm, respectively. The second cavity has an eccentricity of 0.71, with semimajor
and semiminor axis lengths of 4.6 and 6.5 mm, respectively. The thickness of both resona-
tors is 0.5 mm. The third cavity has an eccentricity of 0.32, with semimajor and semiminor
axis lengths of 3.4 and 3.6 mm, respectively.
Design of the resonator geometries is impacted by the coupling mechanism. Evanescent
coupling of light into dielectric resonators requires phase matching between the wave-
number in the coupling medium and the wavenumber of the resonant mode. As is shown
above, it means that the refractive index of the coupling medium should be larger than the
refractive index of the resonator host material to couple to WGMs. Further, the coupling
material should have a comparable optical transparency window to that of the host mate-
rial to avoid additional round trip losses.
LiNbO3 is a birefringent dielectric, and the coupling conditions for TE and TM modes in
this material are slightly different. For example, to couple to TE modes of the resonator, the
free space beam should have an incidence angle of 22° with respect to the plane tangent to
the prism–resonator interface. A circular symmetric beam entering the prism at this angle
will be elliptical in cross section at the interface plane. For TE modes, the spot size along
the polar axis of the resonator is larger than the spot size measured along the azimuthal
axis of the resonator by a factor of 2.7. There will be a drop in the overall coupling effi-
ciency when the ratio achieved during fabrication is not equal to this value.
The ratio is fixed for resonators with circular symmetry, and is subject to the precision of
the fabrication apparatus, knowledge of the refractive indices of the resonator and coupler
materials, and knowledge of the wavelength of light being used. Realization of a critical
curvature ratio at some point along the perimeter of the resonator is more probable for an
elliptical resonator. The curvature ratio varies along the perimeter as the radius of curva-
ture changes in those resonators.
Experiments with elliptical lithium niobate resonators were performed with 1550 nm
light evanescently coupled to the resonators by means of a diamond prism [52]. The laser
light was scanned over several FSRs of the resonators. Light reflected from the cavity-
prism interface was collected by a 3 mm diameter light pipe and sent to an InGaAs
photodetector. One of the assumptions of the approximation used to model WGMs in
elliptical resonators is that the modes of the same modal indices travel the same optical
path regardless of the input positions. That is, the FSR of the resonator is constant as the
input position is varied. The position of the incident beam, governed by the location of
the diamond prism along the cavity perimeter was varied between semimajor axes in
30° steps. It was found that throughout the various input angles, the same frequencies of
light are resonant with the cavity and the same FSR was observed [52]. This observation
confirms the theoretical assumption.
The FWHM of the mode was measured, and the value was used to calculate the Q-factor
of this mode, yielding 1.5 × 108, which corresponds to the highest quality factor measured
with congruent lithium niobate resonators (see next section). The measured FSR agrees
well with the calculated values from Equation 3.28.
A coupling efficiency of 97.3% into the third resonator (Figure 3.16) was achieved. It was
found that the coupling efficiency varied when the contact position between the prism and
resonator was changed. The coupling efficiency refers to the fraction of light transmitted
from within the coupling medium to a specific resonator mode and out through an output
path symmetric with the input path. Light that couples from the mode to a different output
channel, i.e. another WGM, is accounted for as additional loss. Fresnel losses at the air–
prism interface are also present in the system. They could be mitigated by an appropriate
antireflection coating of the prism. Taking Fresnel losses and loss associated with instru-
mentation into account we found that the best laser-to-detector transmission efficiency
through the resonator remains in excess of 91%.
buildup of optical power inside the resonator and the Purcell factor [129] are proportional
to finesse. Sometimes quality factor is more important. For example, the inverse threshold
power of intracavity hyperparametric oscillation is proportional to Q2 [130], and efficiency
of parametric frequency mixing is proportional to Q3 [131]. Therefore, it is important to
know both the maximally achievable finesse and quality factor values of a resonator.
The knowledge of the resonator quality factor is important for the applications in mate-
rial science. Not much data is available for the optical absorption coefficients, since there
is no technique capable of measuring this coefficient with high precision when absorption
becomes very weak. Interestingly, the WGM resonators may be used to measure internal
absorption with good precision if the Q factor is only limited by the internal absorption.
For example, the absorption coefficient reported by CaF2 crystal producer is 2 × 10 − 5 cm − 1.
According to the expression for the internal absorption limited Q factor, this corresponds
to Q = 4.2 × 109 at wavelength of 1064 nm. On the other hand, with this piece of crystal we
have measured Q = 5.3 × 1010, which gives a more accurate upper estimate for the absorption
coefficient 1.6 × 10 −6 cm −1.
WGM resonators can have larger finesse compared with FP resonators. For instance,
fused silica resonators with finesse 2.3 × 106 [132] and 2.8 × 106 [133] have been demon-
strated. Crystalline WGM resonators reveal even larger finesse values, F = 6.3 × 106 [134],
because of low attenuation of light in the transparent optical crystals. The large values of
F and Q result in the enhancement of various nonlinear processes. Low threshold Raman
lasing [135,136], opto-mechanical oscillations [137], frequency doubling [131], and hyper-
parametric oscillations [138,139] based on these resonators have been recently demon-
strated. Theory predicts the possibility of achieving nearly 1014 for Q-factors of optical
crystalline WGM resonators at room temperature [131,134], which correspond to a finesse
level higher than 109.
Experiments, thus far, have shown numbers that are a thousand times lower. The dif-
ference between the theoretical prediction and the experimental values is due to material
imperfections. To bridge this gap, a technique to substantially reduce the optical losses
caused by the imperfections was developed [34]. A specific multi-step asymptotic process
was utilized for removing imperfections in the host material of the resonator. This tech-
nique was previously used to reduce microwave absorption in dielectric resonators [24].
One step of the process consists of mechanical polishing performed after high temperature
annealing. Several repeated subsequent steps lead to a significant reduction of the optical
attenuation and, as a result, the increase of Q-factor and finesse of the resonator. With this
approach we demonstrated a CaF2 WGM resonator with F > 107 and Q > 1011.
There are several practical issues related to the study of properties of ultra-high-Q reso-
nators. The study is challenging because the spectral width of the corresponding WGM
is expected to be less than a kHz. Direct observation of such a narrow line could be done
with a narrowband tunable laser with a high level of frequency stability. Such lasers are
not available in many laboratories. Another problem is related to the parasitic contribu-
tions from nonlinear effects. Because the threshold of stimulated Raman scattering (SRS)
is very low in high-Q resonators [136] the direct observation of the intrinsic spectral width
of a WGM with a continuous wave laser radiation is a problem. Thermal oscillations and
drifts of the position of the optical resonance also make a time-averaged observation
problematic.
A nonlinear ring-down technique that was recently demonstrated [140] helps avoiding
these limitations. In this approach, the laser frequency is swept rapidly across the reso-
nance line. A part of the light is accumulated in the mode during the sweep and is reemit-
ted in the direction of the detector. The reemitted radiation interferes with the laser light,
and beatnote signals can subsequently be observed. The power of the laser light is gener-
ally much larger than the power of the reemitted light. Hence, the envelope of the decaying
oscillations follows the decay of the amplitude of the reemitted light. This decay is twice as
long as the intensity decay. The optical Q-factor of the cavity is expressed as Q = ωτ/2.
We selected a fluorite WGM resonator with optical loss limited by the material attenua-
tion, not by the surface scattering [141]. The resonator was 4.5 mm in diameter, had 0.5 mm
in thickness, and 32 µm in diameter of sidewall curvature. The resonator was placed into
the center of a 3-feet long air-filled transparent tube made of annealed fused silica. The
tube was installed into a 20 cm long horizontal tube furnace. The heated furnace core had
approximately one inch in diameter and three inches in length. We increased the tempera-
ture of the furnace core from room temperature to 650°C during a 3 hour period, and kept
the temperature stabilized for 1 day. The core was then cooled back during 3 hours, and
the fluorite resonator was subsequently repolished. We repeated this process three times
keeping the same annealing duration but gradually decreasing the size of the grain of the
diamond slurry used for polishing. This approach resulted in a significant increase of the
ringdown time at the end of the process.
The measured ringdown spectrum did not change substantially after the first anneal-
ing stage. However, the ringdown time increased significantly after the third stage. The
measured ringdown signals are shown in Figure 3.17. A five-fold increase of the optical
ringdown time is clearly observed. It is also worth noting that the measured quality factor
is several times larger when compared with the quality factor of calcium fluoride resona-
tors observed previously at 1.55 µm (see [134,141]).
The power in the mode of the annealed resonator was clearly above the threshold of
the nonlinear loss related to SRS [140]. The measurement method used in the experiment
has a low dynamic range, so it was not possible to observe the nonlinear decay with the
high-amplitude signals. For that reason the ringdown time was measured using a smaller
optical power. In such a measurement two major precautions must be taken: (i) the laser
wavelength must interact with only one mode, which means that no WGM doublets are
0.01
(a) 1 (b) 0.03
Amplitude (a.u.)
Amplitude (a.u.)
0.00 0.00
Relative amplitude
Relative amplitude
–0.01 –0.03
0 50 100 150
0 200 400 Time (µs)
0.1 Time (µs)
0.1
τ=110±3µs τ=20±1µs
Figure 3.17
(a) Ring-down signal after one annealing step. (b) Ring-down signal after three annealing steps. The exponen-
tial fit (solid curve) is nonlinear in the logarithmic scale because the exponents have constant offsets. Keeping
in mind that Q=ωτ/2, and ω=2πc/λ, λ=1.55µm, we find that the values of the quality factors after the first and
third annealing steps are Q=1.2 × 1010 and Q=6.7 × 1010, respectively. It is important to note that the initial value
of the Q-factor corresponds to the earlier observations [134], while the final value is the apparent improvement
at the given wavelength. (Reprinted from Savchenkov, A. A., Matsko, A. B., Ilchenko, V. S., and Maleki, L., Opt.
Express 15, 6768–6773, 2007. With permission from OSA.)
allowed within the investigated frequency span. (ii) the sweep of the frequency of the laser
must be controlled and the local frequency of the beatnote signal must never increase; such
an increase can occur if the laser is swept back to the starting frequency rapidly. In this
way the single excitation of a single WGM is ensured.
The best ringdown signal measured with all the precautions is shown in Figure 3.18, left.
The oscillation period of the beatnote was evaluated in order to ensure that the measured
line is not a result of several consecutive excitations of the optical mode caused by dither-
ing of the laser frequency (Figure 3.18, right). This shows that the carrier frequency indeed
moves gradually from the resonance and the WGM is not excited twice.
Let us discuss the fitting procedure to the theoretical curve in more detail. The positions
of the maxima and minima of the beatnote peaks, as well as the zero crossings of the beat-
note signal, were identified for the evaluation of the oscillator period. The time coordinate
of each peak was subtracted from the coordinate of the adjacent peak and the averaged
time for the two adjacent peaks was obtained. The same procedure was repeated for the
zero-crossing points. In this way the time dependence of the period of the beat note signal,
shown in Figure 3.18 right, was deduced. The dependence is linear at the tail of the curve.
The initial period has a different time dependence because the frequency of the WGM
changes much faster immediately after the frequency of the pump laser is tuned away
from the WGM. The change is determined by multiple nonlinear processes, e.g. WGM
frequency shift due to after-interaction cooling of the resonator.
Using linear approximation for the beatnote period we found the period of the waveform
to fit the experimental data. We solved the equation φ(t + Period(t)/2) − φ(t) = π approximat-
ing it by a linear differential equation φ (t) = 2 π / Period(t). Using the experimental results
the Period(t) = ξ − zt is obtained, where ξ = 71.1 µs, and z = 0.11 (time t is measured in micro-
seconds). As a result, the expression
(a) (b)
80 Positive peaks
Negative peaks
0.001 Crossings
Oscillation period (µs)
Linear fit
Amplitude (a.u.)
0.000 60
Period (µs)=71.1–0.11 × Time (µs)
–0.001
40
–0.002
20
0.0 0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
Time (ms) Time (ms)
Figure 3.18
(a) Ringdown signal after the third annealing step taken with low-power laser radiation (solid line). The theo-
retical fit of the signal is shown by the dotted line. (b): Evaluated change of the period of the signal. The solid
line stands for the linear fit of the time dependence of the period of the ringdown signal. (Reprinted from
Savchenkov, A. A., Matsko, A. B., Ilchenko, V. S., and Maleki, L., Opt. Express 15, 6768–6773, 2007. With permis-
sion from OSA.)
follows for the phase. The final oscillation waveform presented in Figure 3.18, right is given
by cos[φ(t)].
The time dependence of the beatnote amplitude was extracted in the following way.
The amplitude of a minimum of the oscillating beatnote signal was subtracted from the
adjacent maximum. The value shows the relative oscillation amplitude. This value was
related to the moment of time equally separated from the maximum and the minimum
time coordinates. The result of the evaluation is shown in Figure 3.19. Using the first three
points of the dependence the initial decay time τ1 = 130 µs was found. Using the part of the
signal in the interval between 0.15 and 0.35 ms the final decay time, τ2 = 510 µs was found.
Finally, we used the expression for the nonlinear decay rate discussed by Matsko et al.
[142] to fit the amplitude decay rate. The resultant fit of the beatnote waveform is shown in
Figure 3.18, left. In this way, we can conclude that the intrinsic linear quality factor of the
CaF2 WGM resonator approaches 3 × 1011 at 1.55 µm (20-fold improvement compared with
the initial quality factor).
Let us calculate the finesse of such a resonator. The expression for the finesse, as well
as the quality factor, of WGM resonators naturally includes index of refraction of the
resonator host material, unlike the case of an empty FP resonator: Q = 2πaFn0/λ, and
F = c/[2an0(γ0 + γc)]. Here a is the radius of the resonator, n0 is the index of refraction of the
resonator host material, γ0 and γc are the intrinsic and coupling amplitude decay rates,
respectively. The values of light intensity inside Iin and outside I0 the resonator are related
as Iin/I0 = 2γc F/[π(γ0 + γc)]. Using the experimental data (2a = 0.45 cm, n0 = 1.42) the finesse of
the resonator is found to be F = (2.1 ± 0.6) × 107.
The annealing process discussed above improves the transparency of the material
because an increased temperature results in the enhancement of the mobility of defects
induced by the fabrication process, and also reduces any residual stress birefringence [143].
The increased mobility leads to the recombination of defects and their migration to the
surface [24]. It is worth noting that this annealing technique is similar to the previously
developed procedures [144,145].
1.2
1
0.8
Relative amplitude
0.6
0.4
τ1=130±10 µs
0.2 τ2=510±140 µs
Figure 3.19
Change of the signal amplitude shown in Figure 3.18 left, with time. Initial and final quality factors are
Q = (7.9 ± 0.5) × 1010 and Q = (3 ± 1) × 1011, respectively. Solid line stands for the theoretical fit of the ringdown signal
found using formalism presented by Matsko et al. [142]. A small oscillation of the decaying signal may result
from the residual scattering in the material becoming observable at the given value of quality factor.
The significant improvement of the Q-factor and finesse of a fluorite WGM resonator
discussed above does not reach the fundamental limit. The straightforward annealing of
a WGM resonator leads to Q > 1011 at 1.55 µm while our earlier theoretical prediction gives
Q 1013 at this wavelength [141]. To improve the quality of annealing we suggest using a
larger furnace with much lower thermal gradients. Thermal gradients within a sample
play the leading role in the defect redistribution [143]. The amplitude of the gradient can be
reduced not only by increasing the volume of the oven but also by increasing the thermal
conductivity and optimizing the shape of the container the resonator is placed into dur-
ing the annealing procedure. A fluorite container is the best choice for annealing fluorite
WGM resonators.
In this section, we demonstrated that a proper thermal annealing procedure combined
with mechanical polishing allows producing WGM resonators with vastly improved
finesse and quality factors. We expect that the technique described here will help to
improve quality factors and finesses of WGM resonators made of other transparent crys-
talline materials.
scattering dominates. The quantitative ratio of Rayleigh and Brillouin scattering can be
estimated using the Landau–Plachek relationship (see e.g. [149]):
a Ri βT
-1 (3.30)
aB βS
where βT and βs are the isothermal and isobaric compressibilities, respectively. This ratio
is generally less than unity in crystals and is proportional to temperature in the low tem-
perature limit [150]
βT C
- 1 = P - 1 ~ T4 ,
βS CV T →0 (3.31)
where Cp and Cv are the specific heat capacities at constant pressure and constant volume,
respectively. Rayleigh scattering can be neglected in a perfect crystal at low temperature,
as a result of Equation 3.31).
Temperature dependence of spontaneous Raman and Brillouin scattering mechanisms
is determined by the number of participating phonons, given by a Bose population factor
[151]. The corresponding light attenuation coefficient accounting for both Stokes and anti
Stokes components in a transparent solid may be estimated as follows (see also [149])
-1
λ 4 Ω 1
a B ,R ≈ a B 0 ,R 0 0 exp B ,R - 1 + (3.32)
λ kBT 2
where λ is the wavelength of light, αB0,R0 are the scattering parameters given by the proper-
ties of a particular crystal corresponding to λ0, ΩB is the Brillouin frequency shift for 90°
scattering and ΩR is the Raman frequency shift.
We assume that the crystal has a single phonon branch. Using data evaluated for CaF2
at T = 300 K in [148]: Ω B / kBT 1/ 300, Ω R / kBT 1/ 4, λ0 = 0.532 µm, αB = 2.411 × 10 −7 cm −1,
a R 4.34 × 10-8 cm -1 we obtain a B0 8 × 10-10 cm −1, a R0 1.1 × 10-8 cm -1. It is easy to see
that Brillouin scattering is significantly suppressed at low temperature and the attenu-
ation of light in a perfect crystal is determined by the spontaneous Raman scattering.
The room temperature attenuation, on the other hand, is given by spontaneous Brillouin
scattering.
It is shown by Logunov and Kuchinsky [148] that the existing calcium fluoride crystals
have transparency close to the fundamental limit (Equation 3.32) in the UV. However, such
a low attenuation is yet to be demonstrated in experiments with visible and infrared light.
The measured transparency is two to three orders of magnitude lower than the funda-
mental limit in those frequency bands [38,141,134] because of the extrinsic and intrinsic
impurities of the material.
We now discuss the optical properties of fluorite WGM resonators. The maximum qual-
ity factor of the resonator is given by Qmax = 2πn(λ)/[λα(λ)], where n(λ) is the refractive index
of the material, λ is the wavelength of light in vacuum, and α(λ) is the total loss coefficient of
the bulk dielectric material. The corresponding wavelength dependence of the attenuation
as well as Q factor for CaF2 are shown in Figure 3.20. The approximation of the maximum
quality factor of CaF2 WGM resonators found elswhere [141,134] using existing experimen-
tal results nearly coincides with the fundamental limit presented in Figure 3.20.
The wavelength dependence of the index of refraction of the material at T = 300 K was
found using the four-term Sellmeier equation [152]. Blue and red absorption wings may
1016
10–4 Attenuation at T=300 K
10–5 Attenuation at T=0 K
Attenuation α (cm–1)
1014
Maximum Q factor
10–6
10–7
1012
10–8
10–9
1010 Q factor at T=300 K
10–10 Q factor at T=0 K
10–11 108
0 1 2 3 4 5 0 1 2 3 4 5
Wavelength λ (µm) Wavelength λ (µm)
Figure 3.20
Attenuation in ideal CaF2 and Q factor of ideal crystalline WGM resonators at room and nearly absolute zero
temperature. Contributions from spontaneous Brillouin, Rayleigh and Raman scattering as well as blue and
red wing absorption are added. (Reprinted from Grudinin, I. S., Matsko, A. B., and Maleki, L., Opt. Express 15,
3390–3395, 2007. With permission from OSA.)
The intrinsic absorption of the resonator material is the second important effect restrict-
ing the value of the measured Q factor. The absorption is a function of crystal growing
process and is defined by the optical properties of a crystal at a given wavelength. This
parameter is particularly small in crystalline materials, since they can be produced with
high purity (see previous section). Recent developments in UV lithography have stimu-
lated the production of extremely pure CaF2 monocrystals. However, even in those purest
crystals the “reference-sheet” absorption rarely falls below 10 ppm/cm, which corresponds
to quality factors on the order of 1010.
The surface roughness is made small by the polishing technique, as confirmed by mea-
surements with an AFM. In its contribution to surface scattering factor, the surface rough-
ness is accompanied by the subsurface material damage. This damage is represented by
the microfractures and embedded particles and is usually confined to within a shallow
surface layer of the material. The thickness of this layer is determined by the polishing
and grinding process. Studies with atomic force microscope (AFM) have shown the small
amount of diamond particles embedded into the resonator surface. This could be useful
in quantum optics, since quantum dots containing diamonds or other nanoparticles can
be embedded in this way to create a coupled system without noticeable degradation of the
Q factor.
For example, we have fabricated a calcium fluoride resonator with 100 mm in diameter
and optical Q factor of 4 × 108. The surface roughness does not restrict the value of the qual-
ity factor. The measurements performed with an AFM showed that the surface roughness
in small resonators is about the same as in the large disks with Q factor of about 109 and is
practically negligible (s = 0.33 nm) [38]. Let us estimate the maximum Q factor that could
be achieved in a resonator with measured surface roughness. We will use the expression
derived by Gorodetsky [21]
3λ 3 R
Q≈ (3.33)
8 nπ 2 B 2 s 2
where l is the wavelength, R is the radius of the resonator, n is the refraction index of
the resonator material, B is the correlation length, and s is the roughness. The formula
was obtained as an approximation which takes the correlation length to be much smaller
than the wavelength of light (sB2 << λ 3). It was found that Q = 4 × 1012 for λ = 1 μm, R = 50
μm, n = 1.43, B = 2 nm (was not calculated), and s=0.33 nm [38]. The estimated value of the
Rayleigh scattering limited Q factor scales proportionally to the radius for smaller resona-
tors and is still much larger than any Q factor observed in WGM resonators.
More rigorous studies of the surface quality of ultra-high Q WGM resonators were per-
formed in [134]. It was taken into account that the noise introduced by AFM is small (the
AFM empty image standard deviation was 0.07 nm). The surface roughness was measured
to be 0.19 nm and 0.15 nm in two different samples. Calculations were made with a stan-
dard expression for the unbiased variance. These values are significantly smaller than
those typical of fused silica resonators, and demonstrate the capability of the fabrication
technique. It is possible to achieve even better results if advanced polishing techniques are
used.
In order to use the analysis, the correlation length B also has been carefully evaluated.
The correlation length is defined as the decay length of the exponential autocorrelation
function. We are interested in the correlation length along the direction of light propa-
gation, which in our case is coincident with the direction of polishing traces. Computer
algorithms of the image rotation generally alter the correlation function. In addition, the
autocorrelation function depends on the direction of computation for the type of the sur-
face discussed here. In order to overcome these problems, a two-dimensional correlation
function may be computed with the use of a Wiener–Khinchin theorem and fast Fourier
transform algorithms. It was found that B 40 nm, 20 times larger than the value assumed
in [38]. The maximum corresponding Q is on the order of 7 × 1011 and is still higher that the
quality factors measured experimentally.
It was also found that when an excimer grade CaF2 is used, it is possible to achieve a
Q factor higher than in any other open resonator and any other WGM resonator without
annealing. The 5.5 mm in diameter CaF2 WGM resonator was fabricated with Q factor of
(5.31 ± 0.04) × 1010 measured with a Nd:YAG laser at wavelength of 1064 nm. This is probably
due to specific impurity content of the resonator host material.
To conclude this section, we note that a record high optical Q factor of a fluorite WGM
resonators fabricated with a diamond grinding technique was demonstrated. Even higher
Qs for WGM resonators made out of this and other materials were predicted after an anal-
ysis of possible loss mechanisms in crystals and the influence of nonlinear processes.
The extreme Q factors of crystalline resonators open new opportunities for optical filter-
ing and spectroscopy. A single mode cavity made with lithium niobate could operate as a
tunable fiber coupled etalon, more compact and with a higher finesse than a FP resonator.
We discuss such filters in the following section.
[161]. Fabrication of a fifth-order narrow band tunable WGM filter was reported in
Savchenkov et al. [162].
Existing WGM filters still do not solve all the problems of the microwave photonics. For
example, a recently demonstrated lithium niobate WGM filter [161] has a third-order filter
function, 30 MHz linewidth, and is tunable over a wider range of frequency (>20 GHz).
Agile tunability accompanied by the third-order filter function, narrow linewidth, and
low loss, make this filter exceptionally useful for various microwave photonics applica-
tions. However, the comparatively low-order filter function represents a limitation for the
application of these filters in microwave systems that require efficient sidemode rejection,
and high transmission/rejection contrast. While increasing the number of coupled WGRs
theoretically yields the required high-order high-contrast filter function, the technical
problems associated with the device fabrication pile up exponentially. Some of these prob-
lems were analyzed and solutions were proposed in Savchenkov et al. [162]. In what follows
we review recent achievements in fabrication of filters using crystalline WGM resonators.
ne2 n2
∆vTE = v0 r33EZ , ∆vTM = v0 o r13EZ , (3.34)
2 2
where v0 = 2 × 1014 Hz is the carrier frequency of the laser, r33 = 31 pm/V and r13 = 10 pm/V
are the electro-optic constants, ne = 2.28 and no = 2.2 are the refractive indexes of LiNbO3,
EZ is the amplitude of the electric field applied along the cavity axis. We worked with TM
modes because they have larger quality factors than TE modes. If the quality factor is not
Figure 3.21
Scheme of single resonator filters with different coupling elements. The resonator with prism coupling element
is a stop-band filter. The resonator with two angle-polished fiber couplers is a pass-band filter.
Figure 3.22
Filter setup with two prism couplers (passband configuration). The lithium niobate disc resonator is coated
with metal.
0 Volts
40 Volts
80 Volts
120 Volts
5 10 15 20 25
Laser frequency tuning (GHz)
Figure 3.23
Tunability of the resonator spectrum, 14.7 GHz spectral shift per 100 V of applied voltage. Resonator FSR is
9.5 GHz.
very important, it is better to work with TE modes because their electro-optic shifts are
three times as much as those of TM modes for the same value of the applied voltage.
Theoretically, ∆vTE and ∆vTM do not depend on the resonator properties and are related
to the fundamental limitations of optical resonator-based high speed electro-optic modu-
lators [169]. The results for different resonators measured in our experiment are not com-
pletely identical, a result that stems from the imperfections of the cavity metal coatings as
well as a partial destruction of the coating during the polishing procedure.
The measured insertion loss for the pass band filter shown in Figure 3.22 was less than
5 dB (Figure 3.24). The insertion loss occurs primarily due to inefficient coupling to the
Loaded Q-factor
107
IL0 = 4.45 dB
Figure 3.24
Insertion loss versus quality factor for a single resonator filter. The loading determined by the distance between
the coupling prisms and the resonator was changed by temperature. The crosses stand for the experimental
point and the line is the theoretical fit of the experimental results.
mode. We believe that antireflection coating of the coupling prisms or use of special grat-
ings placed on high-index fibers may reduce the losses significantly.
For the pass band filter shown in Figure 3.22, we analyzed the operation of the filter in
the switching regime. We sent microwave pulses to the filter and observed a change in
light transmission as the microwave pulses entered the filter. An example of such a mea-
surement is shown in Figure 3.25.
1.1
1.0
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
–300 –200 –100 0 100 200 300 –500 0 500
Laser frequency offset (MHz) Time (ms)
Figure 3.25
Left: the transmission resonance of the filter. The filter has 13.8 MHz FWHM and 4.3 dB insertion loss at the reso-
nance. Right: 200 ns pulse front response in the switching regime. .
The filter operates at 1.55 mm wavelength, though the wavelength of operation is lim-
ited only by the absorption loss of lithium niobate and can be anywhere from about 1.0 to
1.7 mm. The reproducible value of the finesse of the filter (F) exceeds F = 300, but in some
experiments finesse F = 1000 was achieved.
The filter is a circular resonator made from electro-optic material (Figure 3.26). As dis-
cussed above, a homogeneous change of an applied electric field results in a homogeneous
change of the refractive index for a single-domain crystalline resonator and, as a conse-
quence, in a frequency shift of the whole resonator spectrum. We propose to manipulate
the domain structure in the material to produce an inhomogeneous electro-optic effect in
the resonator in such a way that targetted mode families experience frequency shifts with
respect to other mode families.
The maximum frequency shift of the TE and TM mode can be found from
-1
n2
∆ω TE = ω0 e
2 ∫ r33 | ETE | EZ dr | ETE |2 dr ,
V
2
∫ (3.35)
V
-1
n2
∆ω TM = ω 0 o
2 ∫
V
r13 | ETM | EZ dr | ETM |2 dr ,
V
2
∫ (3.36)
where w0 is the carrier frequency of the laser, r33 and r13 are the electro-optic coefficients,
sign of which is determined by the direction of the domain of the crystal, which could be
either + z or − z, with respect to the homogeneous DC bias electric field Ez; no and ne are the
refractive indices of lithium niobate, ETE and ETM are the amplitudes of the electric fields of
the modes. The integration is taken over the resonator volume V.
Various modes of a WGM resonator have various spatial dependences (Figure 3.26b).
The tunability of the modes can be changed in different ways by choosing spatial depen-
dence of the electro-optic coefficients in a correct manner through poling the crystal.
A schematic diagram of the filter configuration is shown in Figure 3.26. Several 2.6 mm
diameter disk-shaped resonators of LiNbO3 were fabricated, at 120 mm thickness. The rims
(a) (b)
A-A
E2
A
A R
Figure 3.26
Schematic of the metalized disc resonator used in the reconfigurable filter. The resonator is made from a peri-
odically poled lithium niobate wafer. (a) Picture of the disc resonator. The inverted domain structure is shown
with gray circles. (b) Crossection of the resonator close to the disc rim. Inverted domain structure is shown by
striped areas. Two geometrical structures of electro-magnetic modes of the resonator (radius dependence of
the mode power distribution) are shown under the resonator crossection. The mode depicted by solid line is
barely shifted by the applied voltage, while the mode shown by dotted line moves. (Reprinted from Mohageg,
M., Savchenkov, A., Strekalov, D., Matsko, A., Ilchenko, V., and Maleki, L., Electron. Lett. 41, 91–92, 2005. With
permission from Institute of Electrical Engineers (IEE).)
of the cylindrical disks were polished to resemble the surface of a sphere. A ferroelectric
domain structure reminiscent of a set of rings concentric with the axis of the disk resona-
tor was generated by dragging a 1 mm diameter electrode across the surface of the crystal
while applying a 2.5 kV bias between the electrode and the bottom of the crystal, causing
a permanent change in the structure of the material polarization. The poling process took
place at room temperature and was visualized in situ by reflecting light from the bottom
surface of the crystal. Domain walls, that is, barriers between polarizations parallel to + z
and -z directions, are clearly visualized as dark bands in the reflected light.
The top and bottom surfaces of the polished and poled disk resonator were placed into
contact with metal electrodes. These electrodes were connected to a 0–150 V regulated DC
power supply. A probe beam of 1.55 mm scanned over 20 GHz was coupled into the reso-
nator through a diamond prism. This allowed observing the absorption spectrum of the
poled disk resonator and the motion of the modes as the voltage bias across the resonator
axis was increased.
The first disk was poled with a ring-shaped domain pattern 5 mm edge-to-edge at the
disk rim and a 35 mm thick ring-shaped domain 20 mm away. The probe was coupled into
the WGMs of the resonator with quality factor of 107. When the probe beam was coupled
into the high-Q modes of the first disk and voltage was applied, the radical mode was
observed to change frequency with respect to the rest of the spectra at a rate of 21 MHz/V
(Figure 3.27). The Q-factor of the resonance was maintained constant through the most of
the motion. The mode motion was observed over 10 V of bias change.
10
4
3
8
Transmission (a.u.)
4
1 5
2
2 2
3 4 1
3 5
0 4
–0.5 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5
Relative frequency (GHz)
Figure 3.27
Spectrum of the WGM resonator and its tunability. The resonances 3 and 4 move through each other with bias
voltage change. When the frequencies of the modes coincide, the mode interact due to residual light scattering
in the disc resonator. Relative frequency shift between modes 3 and 4 versus applied bias voltage. (Reprinted
from Mohageg, M., Savchenkov, A., Strekalov, D., Matsko, A., Ilchenko, V., and Maleki, L., Electron. Lett. 41, 91–92,
2005. With permission from IEE.)
a DC voltage of ~ ± 150 V to an electrode. The basic operation principles of the filter are
discussed in Savchenkov et al. [161].
The FSR of each resonator is approximately 10 GHz, and, therefore, the filter can be
tuned practically at any optical frequency in the transparency window of lithium niobate.
To clarify this statement, let us consider a single resonator first. If we are able to tune its
spectrum by the FSR frequency, the resonator spectrum would overlap with itself, which
means that it is always possible to find and to shift a resonator mode to coincide with an
arbitrary external signal frequency. We do not need to use the same mode for filtering.
Now, if we consider several interacting resonators, and tune the frequency of a mode of
the smallest resonator (the largest FSR) by the FSR value, and hence the modes of the other
resonators by values exceeding their FSR, then a high-order filter passband may be con-
structed at any desired frequency.
We note the following distinctive features/advantages of this type of filter over conven-
tional filters: (i) agile tunability accompanied by a high-order filter function, (ii) narrow
linewidth, and (iii) low fiber-to-fiber loss. Combination of these three features makes this
a unique filter for a wide range of applications in optics and microwave photonics.
Tunable single-resonator filters are characterized by their finesse which is equal to the
ratio of the filter’s FSR and the bandwidth. The three-resonator filter has a significantly
more sparse spectrum as compared with a stand alone WGM resonator. This feature is
due to the so called Vernier effect [170] and is similar to the coupled fiber-ring resonators
[171,172] which are noted for a rare spectrum due to a single sequence of modes. Finesse
of the system is very large as the FSR exceeds 1 THz, though we were unable to measure it
accurately because of the limited tuning range of the laser. The tuning speed of the filter is
limited by the wiring layout and is approximately equal to 10 ns, while the actual shifting
time of the spectrum is determined by the filter’s bandwidth and does not exceed 30 ns.
The resonators were cut out from Z-cut metalized crystalline preforms. Each resonator
perimeter edge was hand polished to a toroidal shape. The repeatable value of the loaded
quality factor of the main sequence of the resonator modes was Q 1.6 × 107 , which
corresponds to 15 MHz bandwidth of the mode. We studied several disks with nearly
identical characteristics.
The resonators were arranged in the horizontal direction with home-made flex manipu-
lators (Figure 3.28). There was no need in a vertical adjustment because the surface of the
stage the resonators were placed on was optically polished and all resonators had the
same thickness (with nanometer accuracy). The gaps between the resonators and between
the prisms and the resonators were 50–100 nm, which corresponds to the evanescent field
scale. We did not measure the gaps directly but adjusted them to have the appropriate
system response.
Light was sent into, and retrieved out of, the first and last resonators in the array of the
filter via coupling prisms. The repeatable value of fiber-to-fiber insertion loss, primarily
due to inefficient coupling to the resonator modes, was less than 6 dB across the entire cas-
caded structure. The maximum transmission was obtained when light was resonant with
the resonators modes. The antireflection coating of the coupling prisms or use of special
gratings placed on high-index fibers may reduce the losses significantly. Tuning the filter
was realized by applying a voltage to the top and bottom electrodes fabricated with gold.
The gold coating is absent on the resonator perimeter edge where WGMs are localized.
The differences in the size of the resonators is important in the fabrication of the multi-
resonator filter. Our aim was to produce spectral lines in all three resonators with a similar
width to allow the realization of a complex line structure. If the interacting modes of reso-
nators have different widths, then as they are tuned to approach each other, the height of
the narrower resonance will simply track the shape of the wider ones, which is of no use
for the filter application. The size of each resonator is important since cavities of similar
size have similar optical coupling and similar FSRs. So similar optical modes with the
same efficiency can be excited in each of them. The performance of the filter demonstrates
that indeed similar resonators with approximately the same parameters were fabricated.
Figure 3.29 depicts the spectrum obtained in the experiment with three cascaded LiNbO3
resonators. To highlight the filter performance the theoretical third-order Butterworth fit
of the curve is also plotted. Obviously, the three-cavity filter has much faster roll off com-
pared with the Lorentzian line of the same full width at half maximum. On the other
Prism
Metal-coated
resonator
Collimator
Figure 3.28
A packaged assembly of the three resonator filter.
–10
–30
–40
–50
–60
–70
9.90 9.95 10.00 10.05 10.10
Microwave frequency (GHz)
Figure 3.29
Transmission curve of the filter and its fit with Butterworth profile function γ 6/[(v – v 0)6 + γ 6], where γ = 5.5 MHz, v 0
determines the center of the filter function and primarily depends on the resonators’ geometrical dimensions.
Voltages applied to the resonators vary from near zero to 10 V to properly adjust modes of each individual reso-
nator and construct the collective filter function as shown.
hand, the filter function does not look exactly like a third-order bandpass because of small
differences between the cavity Q-factors and dimensions of the resonators.
The experimentally measured electro-optic tuning of the filter’s spectral response, and
tuning of its center frequency with an applied voltage, is shown in Figure 3.30. The filter
exhibits a linear voltage dependence in the ±150 V tuning range, i.e. the total tuning span
exceeds the FSR of the resonator. Changing the tuning voltage from 0 to 10 V shifted the
spectrum of the filter by 1.3–0.8 GHz for TM mode polarization.
(a) 0 (b)
0
–10
–20 –20
–30
–40
–40
–50 –60
–60
9.85 9.90 9.95 10.00 10.05 10.10 10.15 8 9 10 11 12
Microwave frequency (GHz) Frequency detuning (GHz)
Figure 3.30
Demonstration of the tunability of the three-resonator filter. Left: Tuning of the filter and its Butterworth fit
with 16.6 MHz pass band. Right: Broadband tuning of the filter. It is possible to see spurious pass bands not
completely suppressed with the Vernier effect.
0
–20
–20 –40
Transmission (dB)
–60
–40
–80
10 12
–60
–80
0 20 40 60 80
Frequency detuning (GHz)
Figure 3.31
Demonstration of the five-resonator filter. (Reprinted from Savchenkov, A. A., Matsko, A. B., Ilchenko, V. S.,
Yu, N., Maleki, L., Millimeter and Submillimeter Waves and Workshop on Terahertz Technologies, 2007. The
Sixth International Kharkov Symposium, 1, 79–84. With permission from Institute of Electrical and Electronics
Engineers (IEEE).)
a stable the shape for the passband. An example of the six-pole filter function with 13 MHz
pass band is shown in (Figure 3.32).
(a) (b)
–20
–30
Transmission (dB)
–40
–50
–60
–70
Figure 3.32
Demonstration of the sixth-order filter.
optical modes. For instance, the resonators have identical spectra if the error in their radius
is much less than ∆R = R/Q. This is approximately 0.01 nm for a Q = 108 resonator of 0.2 cm
in diameter. Moreover, even if such resonators are fabricated, their relative temperature
should be stabilized to better than T = n/κQ = 2 mK to avoid relative drift of their spectra,
where k ∼ 10-5 K −1 and n = 2.3 are the combined thermal expansion/thermorefractive coef-
ficient and the extraordinary index of refraction of the lithium niobate, respectively. Such a
thermal stabilization is generally not convenient. Aside from these stringent requirements
the Q-factors of the resonators must also be identical within a few percent margins to real-
ize a proper filter function.
A partial solution to this is provided by the electro-optical properties of the material.
It is possible to trim frequencies of each (nonidentical) resonator with a DC electric bias,
or voltage, applied to the resonator. The value of the bias should follow the changes of
the frequencies of the resonator and the laser. This scheme works well for a single WGM
filter that requires only one bias [160]. The case of the three-WGR filter [161] is more chal-
lenging because it requires three independent biases; it is difficult to independently track
frequencies of the three WGMs involved, especially the frequency of the central resonator.
The growing complexity associated with filters having even a higher order makes their
realization quite challenging.
We propose two schemes that simplify tuning the independent resonator frequencies.
For the three-WGR filter the method involves optical probing the frequencies of two end
resonators when the coupling between the resonators is small (tunable gaps are large, see
Figure 3.33A). The frequency of the central resonator is selected by subsequently reduc-
ing the gaps between the resonators, and ramping the bias of the central resonator until
the proper transmission function is realized. A similar technique works with the five-
resonator filter (Figure 3.33B). In the first step a required second-order transmission func-
tion is realized with two uncoupled end pairs of the resonators. The central resonator is
then tuned after the coupling is restored.
Nonetheless, this technique has several disadvantages: (i) it requires dark periods in the
operation of the filter, (ii) it does not work with filters having higher order than five, and
(iii) it requires a mechanical change of the shape of the filter (tuning the gaps between the
WGM resonators). We have found a more efficient and robust technique to address these
A (a) (b)
Laser Electric bias Electric bias
Collimator
Coupling
prism WGR WGR WGR WGR WGR WGR WGR WGR
Detector
Tracking
detector
Computer
B
Voltage on the detector (a.u.)
Tracking
Coupling WGM detector
prism resonator
Detector
Polymer
Coupling coating
gap
Figure 3.33
(A) Possible configurations for tuning the multiresonator WGM filters: (a) three resonator filters, (b) five resona-
tor filters, and (c) multiple resonator filters. Scheme (c) involves tracking photodiodes. (B) Demonstration of the
efficiency of the tracking photodiode method. The visibility of a WGM doublet as a function of the resonator
load, regulated by the gap between the coupling prism and the resonator surface. This was observed by means
of direct detection of light transmission through the resonator as well as by measurement with the tracking
photodiode. (Reprinted from Savchenkov, A. A., Matsko, A. B., Ilchenko, V. S., Yu, N., Maleki, L., Millimeter
and Submillimeter Waves and Workshop on Terahertz Technologies, 2007. The Sixth International Kharkov
Symposium, 1, 79–84. With permission from IEEE.)
multipole filter (see, e.g. Figure 3.33C). The spectrum of each resonator in the chain is
tracked by a photodiode, and is actively controlled by the microprocessor with feedback of
the bias voltage applied to each resonator. There is practically no limitation on the number
of WGRs in the filter with such a computer controlled scheme.
3.6.8 Insertion Loss
Let us understand the nature of the insertion loss of the multiresonator filter. Using the
formalism presented by Madsen and Zhao [174] we describe the field accumulated in each
resonator with complex coefficients of transmission (Tj) and reflection (Rj). The coupling
between resonators and between resonators and prism couplers is described by constants
kj (Figure 3. 35). The transmission and reflection coefficients are connected by a recurrent
relation
T (37 )
j+1
1 Tj
= Φ j , (3.37)
i k j exp[ -(iδ j + a j )]
R
j+1 R j
(a) (b)
Colloidal
H WGM h graphite
Figure 3.34
Application of the colloidal graphite for fabrication of DC electrodes mounted on crystalline WGM filters.
(Reprinted from Savchenkov, A. A., Matsko, A. B., Ilchenko, V. S., Yu, N., Maleki, L., Millimeter and Submillimeter
Waves and Workshop on Terahertz Technologies, 2007. The Sixth International Kharkov Symposium, 1, 79–84.
With permission from IEEE.)
T6 T0
R5 T4 R3 T2 R1
T5 R4 T3 R2 T1
R6 R0
κ5 κ4 κ3 κ2 κ1 κ0
Figure 3.35
Parameters used to describe the resonator chain. (Reprinted from Savchenkov, A. A., Matsko, A. B., Ilchenko, V.
S., Yu, N., Maleki, L., Millimeter and Submillimeter Waves and Workshop on Terahertz Technologies, 2007. The
Sixth International Kharkov Symposium, 1, 79–84. With permission from IEEE.)
where δj and αj represent the phase build-up and absorption per round trip in the resona-
tor. For the sake of simplicity we assume that δj = δ and αj = 0. Then the transfer function for
the five-resonator chain is
T0 ( iδ+a )/ 2 R
H 11 = e , H 21 = 6 e( iδ+a ). (3.38)
T6 T6
It is easy to find that Pr/Pin = |H21|2 and Pout/Pin = |H11|2. The relative power in resonator #1
is given by P1/Pin = |T5/T6|2, in resonator #2 is P2/Pin = |T4/T6|2, etc.
To further simplify the problem we assume that k0 = k5 = ka, k1 = k4 = kb, and k2 = k3 = kc. The
resonator chain becomes a fifth-order Butterworth filter with power transfer function
1 k 2a k2
| H 11 |2 10
, if k b , k c a (1 - k a ), and k a is arbitrary. (3.39)
1 + 2 7 / 5 δ / k a 8 16
a c k c 2 π an
γa= , γ c = a , δ = ∆ω ,
2π a n 2π a n c
where a is the radius of each resonator, and n is the refractive index of the material.
In reality α ≠ 0 and the filter is always absorptive (|H11|2 < 1 for δ = 0). We calculate
|H11 (δ = 0)|2 keeping the relative coupling between the resonators the same as when evalu-
ated for the lossless case. The resultant absorption increases approximately exponentially
with the increase of relative intrinsic absorption Figure 3.36. The resonators must be overcou-
pled to obtain small losses. In our experiment with the five-pole filter γa/γc ~ 0.25. The observed
12 dB fiber-to-fiber insertion loss results from the absorption in the chain as well as other
miscellaneous losses (reflection from the prism surface, not perfect phase matching, etc).
100
Transmission (Pout/Pin)
10–1
10–1
10–4
10–2
10–7
10–3 0 2 4 6
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Inverse loading (γa/γc)
Figure 3.36
Dependence of the transmission coefficient of the filter on the resonator loading. (Reprinted from Savchenkov,
A. A., Matsko, A. B., Ilchenko, V. S., Yu, N., Maleki, L., Millimeter and Submillimeter Waves and Workshop on
Terahertz Technologies, 2007. The Sixth International Kharkov Symposium, 1, 79–84. With permission from
IEEE.)
Figure 3.37
Schematic of the resonator waveguide. (Reprinted from Matsko, A. B., Savchenkov, A. A., and Maleki, L., Opt.
Lett. 30, 3066–3068, 2005. With permission from OSA.)
the resonators that are within several microns of each other. The evanescent field of a reso-
nator in the chain can be studied from the results of simulations made for the single-mode
resonator (see Figure 3.13). The dimensions of the evanescent field inside the post exceed
several micron. If several such resonators are placed close together on the same post, they
will communicate through the “intrapost” evanescent field.
The propagation of light in the waveguide can be described with the usual wave
Equation 3.16. We are interested in the case of high-order WGMs localized in the vicinity
of the equator of the resonator. For the sake of simplicity we consider the TE mode family
and change variables in Equation 3.16 as E = Ψe±imφ/√r, m is the angular momentum num-
ber of the mode. We consider a low contrast structure, assuming that the resonator radius
changes as R = R0 + L(z) and R0 >> |L(z)|. Then Equation 3.16 is transformed to
∂2 Ψ ∂2 Ψ 2 L( z) m2
+ + k ε 1 + 2 - Ψ = 0, (3.40)
∂r 2 ∂z 2 R0 r 2
∂2 Ψ z 2 L( z)
2 + 2k ε Ψ z = - k z2 Ψ z , (3.41)
∂z R0
∂2 Ψ r 2 2 m2
+ k ε - k - Ψ r = 0, (3.42)
r 2
z
∂r 2
k 2 εL0 / R0 k z
sinh lg 2 k 2 εL0 / R0 - k z2 sin [ k z (l - lg )] ,
+ 2 2
2 k εL0 / R0 - k z
where we assumed that k(2εL0/R0)1/2 > kz (this assumption comes from the condition that
a single localized WGM resonator has at least one confined mode). As follows from
Equation 3.43 kz is determined in allowed bands separated by forbidden band gaps (see
in Figure 3.38).
Let us characterize the frequency spectrum of the entire system. We start from the
expression for the frequency of WGMs derived from Equation 3.44.
ω2
ε = km2 ,q + k z2 .
c2 (3.44)
In the case of small interaction between the resonators, when k(2εL0/R0)1/2 >> kz and klg(2εL0/
R0)1/2 >> 1 we estimate from Equations 3.45 and 3.46 that each WGM mode in the resonator
chain transforms to a frequency band with center.
2
ωc 1 πp
ε k ν ,q + (3.45)
c 2 km ,q l - lg
and width
∆ω 4π 2 p 2 R0
2 L0
ε 2 exp -lg km ,q . (3.46)
c km ,q (l - lg )3 2 L0 R0
The spectrum of the modes due to their radial confinement, being eigenvalues of Equation
3.42, is described by
( )
1/ 3
1 m ε
km ,q m + aq - , (3.47)
R0 2 ε-1
© 2009 by Taylor & Francis Group, LLC
6
12 0
Vg
5
β×10–2 (cm–1)
8
4 Vg = 5×10–4 c
kz×10–3 (cm–1) 4
3 Vg 0
0
0 4 8 12
Frequency detuning (GHz)
2
0
0 2 4 6 8 10 12
β×10–2 (cm–1)
Figure 3.38
A solution of Equation 3.45 for period l = 25 mm (lg = 5 mm), wavelength l = 1.5 mm, radius R0 = 0.3 cm, bump height
L 0 = 30 mm, and ε0 = 2.2. Inset: Structure of the first band evaluated using Equation 3.46. (Reprinted from Matsko,
A. B., Savchenkov, A. A., and Maleki, L., Opt. Lett. 30, 3066–3068, 2005. With permission from OSA.)
where aq is the qth root of the Airy function, Ai(-z) (q is a natural number). This is, in fact, a
spectrum of high-order WGMs of an infinite cylinder with radius R0. We now have all the
necessary elements to describe the waveguide.
One of its parameters is the group velocity, which can be evaluated using wave number
b(1.43) and frequency w(1.44). First of all, even without calculation and using the analogy
with photonic band-gap structures, it is clear that the group velocity of light propagating
in the system, determined as Vg = dω/dβ, approaches zero if light is tuned to the band edge.
However, at that point the dispersion of the group velocity is also large, so such a tuning is
impractical. The dispersion is nearly linear in the center of the band-gap and group veloc-
ity there can be estimated as
Vg ε ∆ω ε l 2 πp 2l R0 2 M0
≈ = 2 exp -lg km,q . (3.48)
c 2 πc km,q (l - lg )3 2 L0
R0
It is easy to see that the group velocity decreases exponentially as the resonators are pulled
apart, or the depth of trenches separating them increases. The result could be obtained
even without analytically solving the problem. Indeed, because light propagates due to
coupling between adjacent resonators, the smaller the coupling, the slower is the light
propagation.
In reality, the resonator material possesses intrinsic losses due to scattering and sur-
face inhomogeneities, as well as absorption. Because of these losses the modes have a
finite ring-down time t. The maximum group delay in the set of coupled resonators can-
not exceed this ring-down time without a significant absorption of light. The minimum
group velocity is then Vgmin ≈ l/2πτ, which corresponds to the propagation of light through
a single uncoupled resonator. In the case of strong coupling between the resonators, when
∆ωπ >> 2π, light interacts with many resonators and its propagation can be studied using
the formalism presented above.
Finally, we made numerical simulations of light propagation in a finite delay line made
of WGM resonators and shown in (Figure 3.39, left). We assumed that initially light is con-
fined in the bottom resonator. After that, we switch the interaction between the resonators
and observe pulse propagation toward the top of the resonator stack, a subsequent reflec-
tion, and backwards propagation (Figure 3.39, right). The ring-down time t was chosen to
be 1 ms, which is theoretically achievable in calcium fluoride WGM resonators [159].
Comparing group velocity of light in this waveguide with that of CROWs [177], we con-
clude that, under equivalent conditions, our waveguide always results in a group velocity
lower than the group velocity in CROWs. The ratio of the group velocities, approximately
equal to the ratio of thickness and diameter of the resonator, can easily give an order of
magnitude of difference.
1 ms
Figure 3.39
Left: A set of coupled WGM rings placed on a single rod. Right: A map obtained with 2D numerical simulations
describing ultra-slow propagation of an optical pulse. The group velocity is less than 20 cm/s. (Reprinted from
Matsko, A. B., Savchenkov, A. A., and Maleki, L., Opt. Lett. 30, 3066–3068, 2005. With permission from OSA.)
(a) EOM
Photodiode
Ein Delay
line
EM
Amplifier
Filter
Phase mod. LN MR
Tunable laser
Figure 3.40
(a) Generalized scheme of an OEO. (b) A tunable microwave photonic filter that includes a lithium niobate WGM
resonator.
studied theoretically by Strekalov et al. [181]. An OEO with a lithium niobate WGM modu-
lator was demonstrated by Matsko et al. [182].
The fundamental noise performance of an OEO is determined by the energy storage
time, or quality factor Q, of the optical storage element. Q-factor can be very large. A long
fiber delay can easily have several microsecond long storage times, corresponding to the
equivalent microwave Q of about a million at 10 GHz frequency. This is a high value com-
pared to conventional dielectric microwave resonators used in oscillators. A high-Q opti-
cal resonator can play the role of the optical storage element, in addition to the role of the
electronic filter. While an optical fiber is a true time delay and introduces group as well
as phase delays. An optical resonator produces group delay only, and phase delay of the
microwave signals due to these resonant elements is negligibly small.
We have made an OEO with a tunable microwave photonic filter as shown in Figure
3.40b. The OEO includes a DFB laser and 130 m delay line. The photonic filter has a tun-
able first-order pass band shown in Figure 3.41, left. As a result of the filter tunability we
have achieved a smooth tuning of the oscillation frequency in the 8–9 GHz range, without
degradation of the performance of the device Figure 3.41, right.
We also demonstrated an OEO with a lithium niobate WGM filter inserted into the opti-
cal loop Figure 3.42, left. The filter is used to suppress generation of spurious modes in the
OEO loop. The noise power spectrum of the oscillator is shown in Figure 3.42, right.
–25 –20
S21 (dB)
–30 –30
–40
–35
–50
–40
–60
–45
9.8 9.9 10.0 10.1 10.2 8 9
Frequency (GHz) Microwave frequency (GHz)
Figure 3.41
Left: Tunable 11 MHz photonic filter. Right: Demonstration of the frequency shift of the OEO.
(b)
(a) Filter compensation Tuning –40
delay
YAG Laser –60
Phase noise (dBc/Hz)
Phase
modulator –80
–100
Single pole LN Main OEO
VCP optical filter fiber delay –120
–140
RF output –160
Broadband RF amplifier Broadband PD 10–1 100 101 102 103 104
Frequency offset (GHz)
Figure 3.42
Left: Scheme of the OEO with optical WGM filter in the optical part of the oscillator. Right: Phase noise of the
photonic oscillator with optical power –6 dBm and WGM resonator quality factor 2 × 108.
E in(t - τ d )e - iω t γ
Eout (t) ≈ , (3.49)
γ + i(ω - ω 0 + φ in(t))
Transmitter
Broadband EO
modulator
CCD Mixer
Optical
Tuning
delay
MW
amplifier
MW generator
10 GHz Tunable WGM filter
PD
MW generator
10 GHz
TV
Figure 3.43
Schematic diagram of the video signal transmission experiment. The solid thin line corresponds to optical fibers,
solid thick lines to microwave waveguides, and dashed lines to electric circuits. (Reprinted from Ilchenko, V. S.,
Savchenkov, A. A., Matsko, A. B., and Maleki, L., Proc. SPIE 4969, 195–206, 2003. With permission from Society
of Optical Engineers (SPIE).)
where E (t - τ ) is the slow field amplitude, t stands for the group delay, φ (t) results from
in d d in
the phase diffusion of the pump laser. The filter works well when the phase diffusion of
the pump laser is small γ φ in(t) ; it transforms the phase fluctuations of the laser into the
amplitude fluctuations for large phase diffusion.
It is important to note that to characterize and retrieve encoded information, the filter
output should be mixed with a monochromatic light on a photodetector. The filter contains
a high-Q cavity that introduces group delay td into the signal. This group delay results in
an additional source of frequency-to-amplitude noise conversion when the output signal
from the filter is mixed with the light that did not pass through the filter. This happens
unless the scheme is balanced. We inserted the filter into a Mach-Zehnder configuration
with a fiber delay line Lf to compensate for the group delay. The delay line length was
equal to Lf = nodF/2nf = 1.2 m , where n f 1.5 is the refractive index of the fiber material and
F = 300 is the cavity finesse. Such a compensation is not required if the laser linewidth is
much smaller than the width of the cavity resonance. In our case optical characterization
of the filter was achieved using a semiconductor diode laser with a 30 MHz FWHM line,
which is quite large. The laser power in the fiber was approximately 2.5 mW.
expansion coefficients, and are both thermally and mechanically isolated to achieve
stability. The geometrical dimensions of this type of resonator is typically about 10 cm or
larger. Lasers locked to these resonators demonstrate Hertz and even subHertz linewidth,
with frequency stabilities as high as ~ 10 −15 at 1 s.
WGM resonators can also be used as frequency etalons [183,184]. The basic advantages
of WGM resonators compared with FP ones are: (i) the small size; (ii) a large wavelength
range in which WGM resonators have high Q-factor; and (iii) the low sensitivity of WGM
resonators to external mechanical noise. This latter property is the result of the unique
orthogonality relations between optical and acoustical WGM modes, in addition to high-Q
factors and low density of the WGM spectra. All these features make utilization of the
WGM resonators attractive in place of FP resonators for a laser stabilization applications.
On the other hand, WGM resonators also have several disadvantages compared with
FP resonators. (i) Reference FP resonators contain a specific mirror spacer material which
has a low thermal expansion coefficient. WGM resonators cannot contain such a material.
The polished rim surface of the WGM resonator plays the role of the resonator mirrors and
the host material is the only resonator “mirror spacer” material possible. Such a material
generally has a large thermal coefficient of expansion. This problem can be mediated by
efficient thermal isolation of the resonator, which is simpler than the isolation of FP resona-
tors because of the small size and structure rigidity of the WGM resonators. (ii) Reference
FP resonators are empty, usually evacuated, and are comparably large. These properties
reduce the fundamental thermodynamic fluctuations, which are important even if the
resonator is kept at a constant temperature. WGM resonators are small and material-filled,
and hence suffer more strongly from the thermodynamic fluctuations [185]. The fluctua-
tions can be partially suppressed by passive and active frequency stabilization schemes.
(iii) The optical nonlinearity of WGM resonators is much larger than the nonlinearity of FP
resonators, because of the same reasons as indicated in the previous item. The circulating
optical power level should be limited in a WGM resonator to mediate this problem. The
power fluctuations of a laser interrogating the resonator should also be small.
The goal of this section, based on the materials published in [183,184], is elucidating the
properties of WGM frequency fluctuations, resulting from the basic fundamental thermo-
dynamic as well as quantum optic principles. We evaluate the frequency spectra of ther-
morefractive, thermal expansion, and thermoelastic fluctuations, together with the steady
state WGM frequency uncertainty resulting from those fluctuations. We also study the
photothermal and ponderomotive fluctuations originating from the measurement proce-
dure and find their frequency spectra.
The stability of a passively stabilized millimeter-sized WGM resonator made of a cer-
tain class of crystalline materials is primarily determined by thermorefractive fluctuations
[183]. Those fluctuations have been predicted and successfully observed in fused silica
microspheres [185]. The frequency stability limit of a cylindrical WGM resonator having
a thickness of one hundred microns and a diameter of several millimeters is on the order
of one part in 10 −12 at one second integration time [183]. Thermorefractive fluctuations
increase inversely proportional to the mode volume, and the predicted stability is limited
because of small volumes of the WGMs.
It was shown, however, that a proper selection of the resonator host material is essen-
tial for stabilization of the WGM frequency [184]. Photorefractive fluctuations can be sup-
pressed in some materials like magnesium fluoride if the proper operation temperature is
selected. But in those resonators thermal expansion fluctuations become dominant in the
frequency stability limit. It is possible to design active schemes to further stabilize fluctua-
tions that result from the residual thermal expansion, using specific thermal expansion
properties of some crystals that are not homogeneous along different crystal axes. In this
way, the achieved frequency stability could be higher than the stability determined by the
fundamental thermodynamic limit.
We also discuss some methods for achieving the desired frequency stability. The sim-
ple passive temperature stabilization is not practical because it must sustain fluctuations
smaller than submicroKelvin level. To solve this problem we make use of methods similar
to technologies developed for the stabilization of quartz crystalline rf oscillators [186]. The
scheme includes (i) compensation of the temperature drifts of the resonator by connect-
ing it with special elements having appropriate linear or nonlinear thermal expansion
(c.f. oven controlled crystal radio frequency oscillator) and (ii) usage of two WGM families
having different thermo-optical constants for measurement and compensation of the reso-
nator temperature fluctuations. These methods can be very efficient. It will be argued that
the relative stability of two modes in the same resonator separated by an octave can be
better than one part in 10-14, at one second integration time.
∆n
( ∆T )m = a n ( ∆T )m , (3.51)
n
kBT 2
〈( ∆T )m2 〉 = , (3.52)
CpVmρ
where kB is the Boltzmann’s constant, T is the absolute temperature, r is the density of the
resonator host material, Vm is the mode volume (assumed to be much less than the volume
of the resonator), Cp is the specific heat capacity at constant pressure of the resonator host
material. In what follows we assume that Cp = CV = C for a crystalline material. The mode
volume of a WGM belonging to the basic mode sequence of a spherical resonator is
λ 7 /6 11/6
Vm = 3.4 π 3/ 2 R , (3.53)
2 πn
The value of the mode volume of a toroidal/cylindrical WGM resonator varies depending
on the resonator thickness and can exceed the value for the spherical resonator.
∂u
- D∆u = F(r, t),
∂t (3.54)
where D = κ/(ρC), κ is the thermal conductivity coefficient, and C is the specific heat capac-
ity. The thermal source is normalized such that the quadratic deviation of the mode tem-
perature coincides with Equation 3.52.
We are interested in the temperature fluctuations averaged over the mode volume
u(t) =
∫ u(r , t) | Ψ(r) | dr ,
2
(3.55)
where |Ψ(r)|2 is the normalized spatial power distribution for the light localized in a WGM,
∫|Ψ(r)|2 dr = 1.
The spectral power density of the random process u –(t) and the quadratic deviation of the
–
average mode temperature 〈u(0) 〉 are given by
2
Su (Ω) =
∫
-∞
- iΩτ
u ∗(t)u(t + τ) e dτ , (3.56)
∞
dΩ
( ∆T )2m = u(0)2 =
∫ S (Ω) 2 π .
u (3.57)
-∞
Let us consider a resonator formed on the surface of an infinitely long cylinder of radius
R by a cylindrical protrusion of radius R + ∆R (R >> ∆R) and thickness L. We solve Equation
3.54 and numerically evaluate the spectral density for a calcium fluoride resonator with
R = 0.3 cm and L = 0.01 cm (see Figure 3.44, left)).
We also find the fluctuations for a thin cylindrical resonator of thickness L and radius R,
R >> L. A simple analytical approximation is possible in this case. We present the correla-
tion of the fluctuation forces as
kBT 2 DL
F(r , t)F(r ′ , t ′) 16 π δ(r - r ′)δ(t - t ′); (3.58)
ρCVm
solving Equation 3.54, we derive an approximate expression for the spectral density of the
thermorefractive noise
3/ 2 2 -1
kBT 2 R 2 R 2 | Ω | 1 R2 Ω
Su (Ω) = 1 + + . (3.59)
ρCVm 12 D D 9 3 6 D 8m1/ 3
The spectral density of the frequency noise, given by Sδ ω /ω (Ω) = a 2nSu (Ω), is approximately
the same for the case of a thin resonator and for a resonator being a part of an infinite cyl-
inder (Figure 3.44 left).
We consider a calcium fluoride resonator of radius R = 0.3 cm and thickness L = 0.01
cm driven with l = 1.55 mm light. We find m 2 πRn / λ ≈ 2.7 × 10 4 and R / m2 / 3 1.1 × 10-3 .
The thermal diffusivity for CaF2 is equal to D = 3.6 × 10 −2 cm2/s, hence characteristic
frequencies for the process are D/R 2 = 0.4 s −1, Dm4/3/R 2 = 3.2 × 105 s −1, and D/L2 =
360 s −1. To find the factor kBa n2T 2 /ρCVm 4 × 10-24 we have taken αn = − 0.8 × 10 −5 K −1, and
Vm = 2 πRL × R / m2 / 3 6 × 10-6 cm3.
(a) (b)
10–13
10–13
S1/2 (Hz–1/2)
S1/2 (Hz–1/2)
10–16
10–16
δω/ω
δω/ω
10–19 10–19
10–22 10–22
100 103 106 109 1012 100 103 106 109 1012
Frequency (Hz) Frequency (Hz)
Figure 3.44
Left: Spectral density for a calcium fluoride resonator with R = 0.3 cm and L = 0.01 cm. The resonator is formed
on the surface of an infinitely long cylinder of radius R by a cylindrical protrusion of radius R + DR (R >> ∆R)
and thickness L. Right: The fluctuations for a thin cylindrical resonator of thickness L and radius R, R >> L.
(Reprinted from Matsko, A. B., Savchenkov, A. A., Yu, N., and Maleki, L., J. Opt. Soc. Am. B 24, 1324–1335, 2007.
With permission from OSA.)
We find the Allan variance of the WGM frequency by integrating the evaluated spectral
density of fluctuations using the expression from Audoin [188]
∞
2 sin 4 (Ωτ / 2)
2
s ( τ) =
π ∫
0
Sδ ω /ω (Ω)
(Ωτ / 2)2
dΩ , (3.60)
where we took into account that Sδω/ω(Ω) is a double sided spectral density. The evaluated
Allan variance for the resonator is shown in Figure 3.45. To understand the change of slope
of the dependence shown in Figure 3.45 we note that integration (averaging) in Equation
3.60 occurs in the vicinity of W = 0 when t → ∞, where spectral density is approximately
constant and σ2(τ) ~ 1/τ. In the case of t → 0 the integration in Equation 3.60 occurs in a
wide band centered at frequency W → ∞, so that Sδω/ω(Ω) ~ 1/Ω2 and σ2(τ) ~ 1/τ. The mono-
tonic function s2(t) naturally has a maximum at some specific value of t. The increase of
Allan variance with time for small t is not counterintuitive because the thermorefractive
fluctuations result in thermal drift of the WGM frequency. The maximum value of the drift
is restricted and longer integration results in averaging down of the fluctuations.
An advantage of crystalline WGM resonators compared with other solid state resonators
is that WGM resonators can be made out of various materials with various thermorefrac-
tive constants. For instance, it is important to mention the unique properties of MgF2. This
crystal has vanishing extraordinary (at ~ 74°C) and ordinary (at ~ 176°C) thermorefractive
coefficients [189]. Tuning the temperature of a MgF2 WGM resonator to the vicinity of zero
thermorefractive coefficient an = 0 allows suppression of the fundamental thermorefrac-
tive noise 〈(∆ω TR)2〉 1/2/ω → 0. Technical thermorefractive noise is also compensated because
temperature stability on the order of 2 mK required to reach ∆ne/ne = 10 −14 is feasible.
The basic conclusion of this section is that the thermorefractive noise does not limit the
stability of WGM resonators made out of certain materials, even with moderate tempera-
ture stabilization and no sophisticated compensation. However, the other noise sources
have to be dealt with to reach higher stability.
10–12
10–13
Allan variance (σ)
10–14
10–15
10–16
10–17
10–1 102 105 108
Time (s)
Figure 3.45
Thermorefractive Allan variance of the frequency of a mode of a cylindrical calcium fluoride WGM resonator
with R = 0.3 cm and L = 0.01 cm. (Reprinted from Savchenkov, A. A., Matsko, A. B., Ilchenko, V. S., Yu, N., and
Maleki, L., J. Opt. Soc. Am. B 24, 2988–2997, 2007. With permission from OSA.)
〈( ∆Vr )2 〉 β (3.61)
= kBT T ,
Vr2 Vr
and βT = − [(1/V) (∂V/∂p)]T is the isothermal compressibility of the resonator host material.
Fluctuation of the radius of a spherical resonator due to fluctuations of its volume is
∆R ∆Vr
∆V = . (3.62)
R 3Vr
kBT 2
〈( ∆T )2r 〉 = , (3.63)
CpVrρ
∆R
∆T = a l ( ∆T )r , (3.64)
R
〈( ∆R)2 〉 βT 2 k BT
2
= k BT + a l . (3.65)
R2 9Vr CpVrρ
It is important to note that the noise depends on the shape of the resonator. For instance, the
thermoelastic fluctuations of a thin cylindrical resonator mounted on a thin stem should
be considered in a different way compared with the above derivation. The above estimate
is valid only if there is at least one continuous (quasi-continuous) dimension in the system,
for instance, if the resonator is formed by a small protrusion on a long cylindric rod.
the resonator of the lowest order. The contributions from other modes of higher order is
comparably small. We assume that in the vicinity of the WGM localization the radius of
the resonator changes in accordance with
∂( ∆R)
+ (- iΩ 0 + Γ 0 )∆R = FR (t), (3.66)
∂t
where the oscillation frequency is taken to be equal to the eigenfrequency of the lowest
order radial mode of a spherical liquid resonator of radius R
πvs (3.67)
Ω0 = ,
R
where vs is the speed of sound. The decay rate of the acoustic mode can be very small. The
minimal value of the rate is thermodynamically limited [190]
Ω 20kTa l2ρ
Γ0 ≥ . (3.68)
9C 2
This is a very small value. The realistic value of the quality factor (Ω0/2Γ0) of the acoustic
mode is expected to exceed 5 × 104 [191].
We select the fluctuational force FR(t) such that it obeys
βT R 2
〈 FR∗ (t)FR (t ′)〉 = Γ 0 kBT δ(t - t ′); (3.69)
9Vr
and obtain an expression for the spectral density of the radius fluctuation:
βT Γ0
S∆R / R = kBT . (3.70)
9Vr (Ω - Ω 0 )2 + Γ 20
〈( ∆ω TE1 )2 〉 2R2 / π 2D
S∆R / R = 2 . (3.71)
ω 1 + (ΩR 2 / Dπ 2 )2
10–15
S1/2 (Hz–1/2)
δω/ω
10–18
10–21
Figure 3.46
The spectral density of the frequency noise Sδω/ω(Ω)=S∆R/R(Ω) due thermoelastic noise. (Reprinted from Matsko,
A. B., Savchenkov, A. A., Yu, N., and Maleki, L., J. Opt. Soc. Am. B 24, 1324–1335, 2007. With permission from
OSA.)
This expression basically tells us that the frequency dependence of the spectral density is
determined by the slowest thermal diffusion time associated with the thermal diffusion
along the radius of the resonator. Using Equation 3.60 we calculate the Allan variance of
the frequency of the WGM resulting from the fundamental thermal expansion fluctua-
tions of a z-cut magnesium fluoride resonator of radius R = 0.3 cm and thickness L = 0.01
cm (Figure 3.47). The thermal diffusivity for MgF2 is equal to D = 7.2 × 10 −2 cm2/s and the
characteristic frequency for the process is D/R 2 = 0.8 s −1. Equation 3.71 gives the top bound-
ary of the low frequency spectral density.
In reality, the resonator can be placed on a metal plate possessing a high thermal conduc-
tivity, so eventually the time constant R2/π2D should be replaced with L2/π2D; this would
reduce the value of the low frequency spectral density significantly. For example, the ther-
mal diffusivity of aluminium is D = 0.97 cm2/s at 300 K. Copper has a bit larger value:
D = 1.15 cm2/s at 300 K. Placing the resonator on a polished copper plate (or squeezing the
resonator between two copper plates) would result in more than an order of magnitude
reduction of phase noise at zero frequency, leading to a reduction of the corresponding
Allan variance below one part in 10 −14 at one second integration time. It is important to note
that placing a resonator onto a copper plate will reduce the quality factor of the mechani-
cal modes of the resonator. This reduction will lead to an enhancement of the influence of
thermoelastic fluctuations on the frequency stability.
We find the solution to the problem in the optimal shaping of the resonator allowing to
increase the resonator volume without changing the characteristic time constant of the
process. The resonator should have a nearly spherical shape, or the shape of a cylinder
with equal radius and height. The light should travel in a small protrusion [53] that does
not influence the thermal and mechanical modes of the resonator. For instance, a nearly
spherical single mode MgF2 WGM resonator of radius R = 0.3 cm will have an Allan vari-
ance less than one part per 10 −14 at one second integration time. This improvement will not
increase the thermoelastic fluctuations.
10–15 10–14
10–15
10–18
1/2
10–16
10–21
10–17
100 103 106 109 100 103 106 109
Frequency (Hz) Time (s)
Figure 3.47
Power spectral density and Allan variance of the thermal expansion defined frequency fluctuations of a mode of
a cylindrical magnesium fluoride WGM resonator with R = 0.3 cm and L = 0.01 cm. (Reprinted from Savchenkov,
A. A., Matsko, A. B., Ilchenko, V. S., Yu, N., and Maleki, L., J. Opt. Soc. Am. B 24, 2988–2997, 2007. With permission
from OSA.)
∂u (3.72)
- D∆u = FP (r , t),
∂t
where Fp(r, t) is the fluctuational force describing noise due to the absorption of the pho-
tons in the resonator
ω m,q, p
〈 FP (r , t)FP (r ′ , t ′)〉 = 〈 Pabs 〉 | Ψ(r ) |2 δ(r - r ′)δ(t - t ′), (3.73)
ρ2C 2
wm,q,p is the angular frequency of the corresponding WGM, 〈Pabs〉 is the expectation value
for the absorbed power, |Ψ(r)|2 is the power distribution in a WGM (∫ | Ψ(r ) |2 dr = 1).
We find that the spectral density of the temperature fluctuations can be approximated
by
ω m ,q , p 〈 Pabs 〉 π 6R 3/ 2 / 64L3/ 2
Su (Ω) ≈ . (3.74)
ρ2C 2 Vr2 Ω 2 + π 6D2 / 16L3R
A comparison of the exact numerical simulation and the approximation of the spectral
density is presented in Figure 3.48.
δω 1 ∆R
= 1 + K ε , (3.75)
ω 2 R
where factor Kε = − Eε −1∂ε/∂p ranges from 1 to 10 [190], E is Young’s modulus of the material,
and ε = n2 is the electric susceptibility of the material.
For the sake of simplicity we take into account only one mechanical mode having the
lowest mechanical frequency and assume that the probe light is resonant with the cor-
responding WGM. The last condition is required to avoid mechanical instability or addi-
tional rigidity added to the mechanical system by light. The mechanical oscillations of the
resonator surface are described by the equation
10–17
S1/2 (Hz–1/2)
10–20
δω/ω
10–23
Figure 3.48
Spectral density of the photothermal fluctuations of a WGM frequency (Sδ1/ω2/ ω = a nSu1/ 2) calculated for a fluorite
resonator with R=0.3 cm and L=0.01 cm interrogated with coherent 3.55 mm light of 1 mW power assuming that
the light is absorbed in the resonator. Solid (dashed) line stands for the simulation (analytical calculations).
(Reprinted from Matsko, A. B., Savchenkov, A. A., Yu, N., and Maleki, L., J. Opt. Soc. Am. B 24, 1324–1335, 2007.
With permission from OSA.)
where m* ≈ ρVr is the effective mass of the oscillator, dP(t) is the variation of the optical
power in the corresponding WGM. Calculating dP(t) using general Langevin formalism
we derive an expression for the spectral density of the fluctuations:
2
2 π ω m, q , p 〈 P 〉 2 γ R 1
S∆R / R = ∗ , (3.77)
m cR τ0 γ 2R + Ω 2 (Ω 20 - Ω 2 )2 + Γ 20 Ω 2
where gR is the spectral width of the mode. Because generally Ω0 >> γR, Γ0 we find an expres-
sion for the square deviation of the radius of the resonator resulting from the fluctuations
of the radiation pressure:
1/ 2
〈∆R 2 (t)〉1/ 2 2 π ω m ,q , p 〈 P〉 γR (3.78)
∗ 4 1 + .
R m cR τ 0Ω 0 Γ0
Let us estimate the value for a fluorite resonator with R = 0.3 cm and L = 0.01 cm interro-
gated with coherent 1.55 mm light of 1 mW power. We also assume that γR = 2π × 104 rad/s
and Γ0 = 100 rad/s. We find the averaged power inside the resonator 〈P〉 = 1 mW /τ 0 γ R 250
W, square deviation of the radius (〈∆R 2 (t)〉 / R 2 )1/ 2 5.5 × 10-16 , and low frequency spectral
density (S∆R/R(0))1/2 = 3 × 10 −18 Hz −1/2. The corresponding spectral density of frequency fluc-
tuations for Kε = 4 is shown in Figure 3.49.
Our calculations show that the radiation pressure fluctuations are comparably weak in a
sufficiently large WGM resonator, and can be neglected in the majority of cases when the
resonator is interrogated with low power light.
RF2 ∼ ω ( a lo - a le ) ∆TR ,
∆ω (3.79)
where ∆ω RF2 is the variation of the frequency difference between the two modes deter-
mined by the temperature fluctuations of the resonator, w is the optical frequency, and
alo (ale) is the thermal expansion coefficient for X and Y (Z) directions. The third mode,
TE, is excited in the XY plane. The frequency difference between this mode and the TM
mode in the same plane contains information about the temperature in the WGM channel.
Both modes are influenced by the thermal expansion in the same way. Using results of the
10–15
10–18
S1/2 (Hz–1/2)
10–21
δω/ω
10–24
10–27
Figure 3.49
Spectral density of the ponderomotive backaction fluctuations of frequency (Sδ1/ω2/ ω = (1 + K ε / 2)S∆1/R2/ R ) calculated
for a fluorite resonator with R = 0.3 cm and L = 0.01 cm interrogated with 1 mW coherent light. (Reprinted from
Matsko, A. B., Savchenkov, A. A., Yu, N., and Maleki, L., J. Opt. Soc. Am. B 24, 1324–1335, 2007. With permission
from OSA.)
∆ω TM a lo ∆ω
RF 2
∼ . (3.80)
ω a lo - a le ω
(iii) relative drift of the laser frequency resulting from the fluctuations of the driving
currents was not stabilized.
Very recently, we have demonstrated injection locking of DFB lasers to crystalline
WGM resonators. Initial measurements of the beat note of two lasers locked to calcium
fluoride and lithium tantalate resonators show that the laser linewidth is less than 2 kHz
(Figure 3.50). Further studies of the locking efficiency are required.
3.8 Conclusion
In this chapter, we have discussed the basic linear properties of crystalline optical WGM
resonators. We have elucidated advantages of these resonators in practical application such
as optical and microwave filtering. We have highlighted some important applications of the
resonators in metrology for optical and microwave frequency stabilization, where a long pho-
ton storage time helps to suppress phase and frequency deviation of oscillators. The results of
our R&D efforts clearly show the ample opportunities in applications of the resonators.
Because of the volume restrictions, we were unable to cover other applications of the
microresonators, covered in the other chapters of the book. For instance, high-Q and long
recirculation of light in compact WGM resonators offers interesting new capabilities in
spectroscopy and sensing, where the change in Q or resonance frequency of WGMs can
serve as a measure of absorption in the surrounding medium, or in a small (down to single
molecule) quantity of deposited substance on resonator surface. The resonator can also be
used for measurement of change in ambient parameters, such as temperature, pressure,
motion, etc.
Beat note power (a.u.)
–3 –2 –1 0 1 2 3
Frequency detuning (kHz)
Figure 3.50
The beating signal of two lasers one of which is locked to a lithium tantalate resonator with loaded bandwidth
approximately 3 MHz and calcium fluoride resonator with loaded bandwidth approximately 400 KHz. Laser
power in the fiber at the exit of the resonator is 300 mW and 100 mW for LiTaO3 and CaF2 resonators, respectively.
The resolution bandwidth of the measurement is 300 Hz and the video bandwidth is 30 kHz.
References
1. C. G. B. Garrett, W. Kaiser, and W. L. Bond. Stimulated emission into optical whispering gallery
modes of spheres. Phys. Rev. 124, 1807–1809, 1961.
2. P. Walsh and G. Kemeny. Laser operation without spikes in a ruby ring. J. Appl. Phys. 34,
956–957, 1963.
3. A. Ashkin and J. M. Dziedzic. Observation of resonances in the radiation pressure on dielectric
spheres. Phys. Rev. Lett. 38, 1351–1354, 1977.
4. P. Chylek, J. T. Kiehl, and M. K. W. Ko. Optical levitation and partial-wave resonances. Phys.
Rev. A 18, 2229–2233, 1978.
5. A. Ashkin and J. M. Dziedzic. Observation of optical resonances of dielectric spheres by light
scattering. Appl. Opt. 20, 1803–1814, 1981.
6. P. Chyek, V. Ramaswamy, A. Ashkin, and J. M. Dziedzic. Simultaneous determination of refrac-
tive index and size of spherical dielectric particles from light scattering data. Appl. Opt. 22,
2302–2307, 1983.
7. J. F. Owen, P. W. Barber, B. J. Messinger, and R. K. Chang. Determination of optical-fiber diam-
eter from resonances in the elastic scattering spectrum. Opt. Lett. 6, 272–274, 1981.
8. R. E. Benner, P. W. Barber, J. F. Owen, and R. K. Chang. Observation of structure resonances in
the fluorescence-spectra from microspheres. Phys. Rev. Lett. 44, 475–478, 1980.
9. S. C. Hill, R. E. Benner, C. K. Rushforth, and P. R. Conwell. Structural resonances observed in
the fluorescence emission from small spheres on substrates. Appl. Opt. 23, 1680–1683, 1984.
10. S. C. Hill, R. E. Benner, C. K. Rushforth, and P. R. Conwell. Sizing dielectric spheres and cylin-
ders by aligning measured and computed resonance locations: algorithm for multiple orders.
Appl. Opt. 24, 2380–2390, 1985.
11. R. Thurn and W. Kiefer. Raman-microsampling technique applying optical levitation and radi-
ation pressure. Appl. Spectr. 38, 78–83, 1984.
12. S.-X.Qian, J. B. Snow, and R. K. Chang. Coherent Raman mixing and coherent anti-Stokes
Raman scattering from individual micrometer-size droplets. Opt. Lett. 10, 499–501, 1985.
13. J. B. Snow, S.-X.Qian, and R. K. Chang. Stimulated Raman scattering from individual water and
ethanol droplets at morpholody-dependent resonances. Opt. Lett. 10, 37–39, 1985.
14. H. M. Tzeng, K. F. Wall, M. B. Long, and R. K. Chang. Laser emission from individual drop-
lets at wavelengths corresponding to morphology-dependent resonances. Opt. Lett. 9, 499–501,
1984.
15. H. B. Lin, A. L. Huston, B. J. Justus, and A. J. Campillo. Some characteristics of a droplet
whispering-gallery-mode laser. Opt. Lett. 11, 614–616, 1986.
16. L. Collot, V. Lefevre-Seguin, M. Brune, J.-M. Raimond, and S. Haroshe. Very high-Q whispering-
gallery mode resonances observed in fused silica microspheres. Europhys. Lett. 23, 327–334,
1993.
17. M. L. Gorodetsky, A. A. Savchenkov, and V. S. Ilchenko. Ultimate Q of optical microsphere
resonators. Opt. Lett. 21, 453–455, 1996.
18. S. Shiller and R. L. Byer. High-resolution spectroscopy of whispering gallery modes in large
dielectric spheres. Opt. Lett. 16, 130–132, 1991.
19. D. R. Rowland, and J. D. Love. Evanescent wave coupling of whispering gallery modes of a
dielectric cylinder, IEE Proceedings. J. Optoelectronics 140, 177–188, 1993.
20. M. L. Gorodetsky and V. S. Ilchenko. High-Q optical whispering gallery microresonators–
precession approach for spherical mode analysis and emission patterns with prism couplers.
Opt. Comm. 113, 133–143, 1994.
21. M. L. Gorodetsky, A. D. Pryamikov, and V. S. Ilchenko. Rayleigh scattering in high-Q micro-
spheres. J. Opt. Soc. Am. B 17, 1051–1057, 2000.
22. D. G. Blair and I. N. Evans. High Q microwave properties of a sapphire ring resonator. J. Phys.
D 15, 1651–1656, 1982.
23. V. B. Braginsky. V. P. Mitrofanov and V. I. Panov. Systems with small dissipation. Chicago University
Press, Chicago, IL, 1985.
24. V. B. Braginsky, V. S. Ilchenko, and K. S. Bagdassarov. Experimental observation of fundamental
microwave absorption in high quality dielectric crystals. Phys. Lett. A 120, 300–305, 1987.
25. W. L. Bond. Making crystal spheres. Rev. Sci. Instr. 25, 401–401, 1954.
26. D. A. Cohen and A. F. J. Levi. Microphotonic components for a mm-wave receiver. Solid State
Electron. 45, 495–505, 2001.
27. T. Baer. Continuous-wave laser oscillation in a Nd:YAG sphere. Opt. Lett. 12, 392–394, 1987.
28. V. S. Ilchenko, X. S. Yao, and L. Maleki. Microsphere integration in active and passive photonics
devices. Proc. SPIE 3930, 154–162, 2000.
29. V. S. Ilchenko and L. Maleki. Novel whispering-gallery resonators for lasers, modulators, and
sensors. Proc. SPIE 4270, 120–130, 2001.
30. D. A. Cohen and A. F. J. Levi. Microphotonic millimetre-wave receiver architecture. Electron.
Lett. 37, 37–39, 2001.
31. D. A. Cohen, M. Hossein-Zadeh, and A. F. J. Levi. Microphotonic modulator for microwave
receiver. Electron. Lett. 37, 300–301, 2001.
32. D. A. Cohen, M. Hossein-Zadeh, and A. F. J. Levi. High-Q microphotonic electro-optic modula-
tor. Solid State Electron. 45, 1577–1589, 2001.
33. D. K. Serkland, R. C. Eckardt, and R. L. Byer. Continuous-wave total-internal-reflection optical
parametric oscillator pumped at 1064 nm. Opt. Lett. 19, 1046–1048, 1994.
34. A. A. Savchenkov, A. B. Matsko, V. S. Ilchenko, and L. Maleki. Optical resonators with ten mil-
lion finesse. Opt. Express 15, 6768–6773, 2007.
35. G. Lee. Realization of ultrasmooth surface with atomic scale step structure on LiNbO3 and
LiTaO3 substrates. Opt. Express 10, 556–560, 2002.
36. J. L. Yuan, P. Zhao, J. Ruan, Z. X. Cao, W. H. Zhao, and T. Xing. Lapping and polishing process for
obtaining super-smooth surfaces of quartz crystal. J. Mat. Process. Technol. 138, 116–119, 2003.
37. J. W. Yan, K. Syoji, and J. Tamaki. Crystallographic effects in micro/nanomachining of single-
crystal calcium fluoride. J. Vac. Sci. Technol. 22, 46–51, 2004.
38. I. S. Grudinin, A. B. Matsko, A. A. Savchenkov, D. Strekalov, V. S. Ilchenko, and L. Maleki. Ultra
high Q crystalline microcavities. Opt. Commun. 265, 33–38, 2006.
39. J. W. Yan, J. Tamaki, K. Syoji, T. Kuriyagawa. Single-point diamond turning of CaF2 for nano-
metric surface. Int. J. Adv. Manuf. Technol. 24, 640–646, 2004.
40. S. T. Chu, B. E. Little, V. Van, J. V. Hryniewicz, P. P. Absil, F. G. Johnson, D. Gill, O. King,
F. Seiferth, M. Trakalo, and J. Shanton. Compact full C-band tunable filters for 50 GHz channel
spacing based on high order micro-ring resonators. Optical Fiber Communication Conference
2004, presentation PDP9.
41. C. H. Townes and A. L. Shawlow. Microwave spectroscopy. McGraw-Hill, New York, 1955.
42. J. P. Gordon. Variable coupling reflection cavity for microwave spectroscopy. Rev. Scientific
Instruments 32, 658–661, 1961.
43. A. Yariv. Critical coupling and its control in optical waveguide-ring resonator systems. IEEE
Photon. Tech. Lett. 14, 483–485, 2002.
44. R. Ulrich. Theory of the prism-film coupler by plane wave analysis. J. Opt. Soc. Am. 60, 1337–
1350, 1970.
45. P. K. Tien and R. Ulrich. Theory of prism-film coupler and thin film light guides. J. Opt. Soc. Am.
60, 1325–1336, 1970.
46. A. V. Chelnokov, and J.-M. Lourtioz. Optimized coupling into planar waveguides with cylin-
drical prisms. Electron. Lett. 31, 269–271, 1995.
47. D. Sarid, P. J. Cressman, and R. L. Holman. High-efficiency prism coupler for optical wave-
guides. Appl. Phys. Lett. 33, 514–515, 1978.
48. D. Sarid. High efficiency input-output prism waveguide coupler: an analysis. Appl. Optics 18,
2921–2926, 1979.
49. M. L. Gorodetsky and V. S. Ilchenko. Optical microsphere resonators: optimal coupling to high-
Q whispering-gallery modes. J. Opt. Soc. Am. B 16, 147–154, 1999.
94. B. Z. Katsenelenbaum, L. Mercader del Rio, M. Pereyaslavets, M. Sorolla Ayza, and M. Thumm.
Theory of nonuniform waveguides: the cross-section method. Institute of Electrical Engineers,
London, 1998.
95. V. S. Ilchenko, A. A. Savchenkov, A. B. Matsko, and L. Maleki. Tunability and synthetic line-
shapes in high-Q optical whispering gallery modes. Proc. SPIE 4969, 195–206, 2003.
96. M. L. Gorodetsky and A. E. Fomin. Geometrical theory of whispering-gallery modes. IEEE J.
Sel. Top. Quantum Electron. 12, 33–39, 2006.
97. G. Meijer, M. G. H. Boogaarts, R. T. Jongma, D. H. Parker, and A. M. Wodtke. Coherent cavity
ring down spectroscopy. Chem. Phys. Lett. 217, 112–116, 1994.
98. V. S. Ilchenko, A. A. Savchenkov, A. B. Matsko, and L. Maleki. Whispering-gallery-mode
electro-optic modulator and photonic microwave receiver. J. Opt. Soc. Am. B 20, 333–342,
2003.
99. A. Wicht, K. Danzmann, M. Fleischhauer, M. Scully, G. Müller, and R.-H. Rinkleff. White-light
cavities, atomic phase coherence, and gravitational wave detectors. Opt. Commun. 134, 431–439,
1997.
100. G. S. Pati, M. Salit, K. Salit, and M. S. Shahriar. Demonstration of a tunable-bandwidth white-light
interferometer using anomalous dispersion in atomic vapor. Phys. Rev. Lett. 99, 133601, 2007.
101. A. A. Savchenkov, A. B. Matsko, and L. Maleki. White-light whispering gallery mode resona-
tors. Opt. Lett. 31, 92–94, 2006.
102. W. von Klitzing, R. Long, V. S. Ilchenko, J. Hare, and V. Lefevre-Seguin. Frequency tuning of
the whispering-gallery modes of silica microspheres for cavity quantum electrodynamics and
spectroscopy. Opt. Lett. 26, 166–168, 2001.
103. S. Blair and Y. Chen. Resonant-enhanced evanescent-wave fluorescence biosensing with cylin-
drical optical cavities. Appl. Optics 40, 570–582, 2001.
104. R. W. Boyd and J. E. Heebner. Sensitive disk resonator photonic biosensor. Appl. Optics 40,
5742–5747, 2001.
105. E. Krioukov, D. J. W. Klunder, A. Driessen, J. Greve, and C. Otto. Sensor based on an integrated
optical microcavity. Opt. Lett. 27, 512–514, 2002.
106. S. Arnold, M. Khoshsima, I. Teraoka, S. Holler, and F. Vollmer. Shift of whispering-gallery
modes in microspheres by protein adsorption. Opt. Lett. 28, 272–274, 2003.
107. F. Vollmer, S. Arnold, D. Braun, I. Teraoka, and A. Libchaber. Multiplexed DNA quanti. cation
by spectroscopic shift of two microsphere cavities. Biophys. J. 85, 1974–1979, 2003.
108. C. Y. Chao and L. J. Guo. Design and optimization of microring resonators in biochemical sens-
ing applications. J. Lightwave Technol. 24, 1395–1402, 2006.
109. S. Y. Cho and N. M. Jokerst. A polymer microdisk photonic sensor integrated onto silicon. IEEE
Photon. Technol. Lett. 18, 2096–2098, 2006.
110. N. M. Hanumegowda, I. M. White, and X. D. Fan. Aqueous mercuric ion detection with micro-
sphere optical ring resonator sensors. Sensors Actuators B 120, 207–212, 2006.
111. A. M. Armani, R. P. Kulkarni, S. E. Fraser, R. C. Flagan, and K. J. Vahala. Label-free, single-
molecule detection with optical microcavities. Science 317, 783–787, 2007.
112. H. Zhu, I. M. White, J. D. Suter, P. S. Dale, and X. Fan. Analysis of biomolecule detection with
optofluidic ring resonator sensors. Opt. Express 15, 9139–9146, 2007.
113. F. Xu, P. Horak, and G. Brambilla. Optical microfiber coil resonator refractometric sensor. Opt.
Express 15, 7888–7893, 2007; Opt. Express 15, 9385–9385, 2007.
114. M. Sumetsky. Optimization of optical ring resonator devices for sensing applications. Opt. Lett.
32, 2577–2579, 2007.
115. A. T. Rosenberger. Analysis of whispering-gallery microcavity-enhanced chemical absorption
sensors. Opt. Express 15, 12959–12964, 2007.
116. A. A. Savchenkov, A. B. Matsko, D. Strekalov, V. S. Ilchenko, and L. Maleki. Mode filtering in
optical whispering gallery resonators. Electron. Lett. 41, 495–497, 2005.
117. M. Kuznetsov, M. Stern, and J. Coppeta. Single transverse mode optical resonators. Opt. Express
13, 171–181, 2005.
118. R. G. Hunsperger. Integrated Optics: Theory and Technology. Springer Verlag, Berlin, 2002.
119. S.-T. Ho, D. Y. Chu, J.-P. Zhang, S. Wu, and M.-K. Chin. In Optical processes in microcavities
R. K. Chang and A. J. Campillo (Eds). Word Scientific, Singapore, 1996, p. 339.
120. M. V. Berry. Regularity and chaos mechanics, illustrated in classical by three deformations of a
circular “billiard”. Eur. J. Phys. 2 91–102, 1981.
121. A. Ancey, A. Folacci, and P. Gabrielli. Exponentially improved asymptotic expansions for reso-
nances of an elliptic cylinder. J. Phys. A: Math. Gen. 33, 3179–3208, 2000.
122. S. Ancey, A. Folacci, and P. Gabrielli. Whispering-gallery modes and resonances of an ellip-
tic cavity. J. Phys. A: Math. Gen. 34, 1341–1359, 2001; J. Phys. A: Math. Gen. 34, 2657–2657,
2001.
123. S. V. Boriskina, T. M. Benson, P. Sewell, and A. I. Nosich. Tuning of elliptic whispering-gallery-
mode microdisk waveguide filters. J. Lightwave Technol. 21, 1987–1995, 2003.
124. L. Zhang, Y. -X. Wang, F. Zhang, and R. O. Claus. Observation of whispering-gallery and direc-
tional resonant laser emission in ellipsoidal microcavities. J. Opt. Soc. Am. B 23, 1793–1800,
2006.
125. G. Rempe, R. J. Thompson, H. J. Kimble, and R. Lalezari. Measurement of ultralow losses in an
optical interferometer. Opt. Lett. 17, 363–365, 1992.
126. C. J. Hood, H. J. Kimble, and J. Ye. Characterization of high-finesse mirrors: Loss, phase shifts,
and mode structure in an optical cavity. Phys. Rev. A 64, 033804, 2001.
127. H. R. Bilger, P. V. Wells, and G. E. Stedman. Origins of fundamental limits for reflection losses
at multilayer dielectric mirrors. Appl. Opt. 33, 7390–7396, 1994.
128. H.-J. Cho, M.-J. Shin, and J.-C. Lee. Effects of substrate and deposition method onto the mirror
scattering. Appl. Opt. 45, 1440–1446, 2006.
129. E. M. Purcell. Spontaneous emission probabilities at radio frequencies. Phys. Rev. 69, 681–681,
1946.
130. A. B. Matsko, A. A. Savchenkov, D. Strekalov, V. Ilchenko, and L. Maleki. Optical hyperpara-
metric oscillations in a whispering-gallery-mode resonator: Threshold and phase diffusion.
Phys. Rev. A 71, 033804, 2005.
131. V. S. Ilchenko, A. A. Savchenkov, A. B. Matsko, and Lute Maleki. Nonlinear optics and crystal-
line whispering gallery mode cavities. Phys. Rev. Lett. 92, 043903, 2004.
132. D. W. Vernooy, V. S. Ilchenko, H. Mabuchi, E. W. Streed, and H. J. Kimble. High-Q measure-
ments of fused-silica microspheres in the near infrared. Opt. Lett. 23, 247–249, 1998.
133. T. J. Kippenberg, S. M. Spillane, and K. J. Vahala. Demonstration of ultra-high-Q small mode
volume toroid microcavities on a chip. Appl. Phys. Lett. 85, 6113–6115, 2004.
134. I. S. Grudinin, V. S. Ilchenko, and L. Maleki. Ultrahigh optical Q factors of crystalline resonators
in the linear regime. Phys. Rev. A 74, 063806, 2006.
135. S. M. Spillane, T. J. Kippenberg, and K. J. Vahala. Ultralow-threshold Raman laser using a
spherical dielectric microcavity. Nature 415, 621–623, 2002.
136. I. S. Grudinin and L. Maleki. Ultralow-threshold Raman lasing with CaF2 resonators. Opt. Lett.
32, 166–168, 2007.
137. T. J. Kippenberg, H. Rokhsari, T. Carmon, A. Scherer, and K. J. Vahala. Analysis of radiation-
pressure induced mechanical oscillation of an optical microcavity. Phys. Rev. Lett. 95, 033901,
2005.
138. T. J. Kippenberg, S. M. Spillane, and K. J. Vahala. Kerr-nonlinearity optical parametric oscilla-
tion in an ultrahigh-Q toroid microcavity. Phys. Rev. Lett. 93, 083904, 2004.
139. A. A. Savchenkov, A. B. Matsko, D. Strekalov, M. Mohageg, V. S. Ilchenko, and L. Maleki. Low
threshold optical oscillations in a whispering gallery mode CaF2 resonator. Phys. Rev. Lett. 93,
243905, 2004.
140. A. A. Savchenkov, A. B. Matsko, M. Mohageg, and L. Maleki. Ringdown spectroscopy of
stimulated Raman scattering in a whispering gallery mode resonator. Opt. Lett. 32, 497–499,
2007.
141. A. A. Savchenkov, V. S. Ilchenko, A. B. Matsko, and L. Maleki. Kilohertz optical resonances in
dielectric crystal cavities. Phys. Rev. A 70, 051804R, 2004.
142. A. B. Matsko, A. A. Savchenkov, and L. Maleki. Ring-down spectroscopy for studying proper-
ties of CW Raman lasers. Opt. Commun. 260, 662–665, 2006.
143. B. Wang, R. R. Rockwell, and J. List. Linear birefringence in CaF2 measured at deep ultraviolet
and visible wavelengths. J. Microlithography, Microfabrication, and Microsystems 3, 115–121, 2004.
144. V. Deuster, M. Schick, Th. Kayser, H. Dabringhaus, H. Klapper, and K. Wandelt. Studies of the
facetting of the polished (100) face of CaF2. J. Cryst. Growth 250, 313–323, 2003.
145. Q. -Z. Zhao, J. -R. Qiu, X. -W. Jiang, C. -J. Zhao, and C. -S. Zhu. Fabrication of internal diffraction
gratings in calcium fluoride crystals by a focused femtosecond laser. Opt. Express 12, 742–746,
2004.
146. I. S. Grudinin, A. B. Matsko, and L. Maleki. On the fundamental limits of Q factor of crystalline
dielectric resonators. Opt. Express 15, 3390–3395, 2007.
147. M. E. Lines. Scattering losses in optic fiber materials. I. A new parametrization. J. Appl. Phys. 55,
4052–4057, 1984; Scattering losses in optic fiber materials. II. Numerical estimates. J. Appl. Phys.
55, 4058–4063, 1984.
148. S. Logunov and S. Kuchinsky. Experimental and theoretical study of bulk light scattering in
CaF2 monocrystals. J. Appl. Phys. 98, 053501, 2005.
149. I. L. Fabelinskii. Molecular scattering of light. Plenum Press, New York, 1968.
150. D. C. Wallace. Thermodynamics of crystals. Dover, New York, 1998.
151. R. Loudon. The Raman effect in crystals. Adv. Phys. 50, 813–864, 2001.
152. M. Daimon and A. Masumara. High-accuracy measurements of the refractive index and its
temperature coefficient of calcium fluoride in a wide wavelength range from 138 to 2326 nm.
Appl. Opt. 41 5275–5281, 2002.
153. E. Palik. Handbook of optical constants of solids. Academic, New York, 1998.
154. M. Lines. Ultralow-loss glasses, Ann. Rev. Mater. Sci. 16, 113–135, 1986.
155. A. Laufer, J. Pirog, and J. McNesby. Effect of temperature on the vacuum ultraviolet transmit-
tance of lithium fluoride, calcium fluoride, barium fluoride, and sapphire. J. Opt. Soc. Am. 55,
64–66, 1965.
156. L. L. Boyer, J. A. Harrington, M. Hass, and H. B. Rosenstock. Multiphonon absorption in ionic
crystals. Phys. Rev. B 11, 1665–1680, 1975.
157. S. Venugopalan and A. K. Ramdas. Effect of uniaxial stress on the raman spectra of cubic crys-
tals: CaF2, BaF2, and Bi12GeO20. Phys. Rev. B 8, 717–734, 1973.
158. R. W. Boyd. Nonlinear optics. Academic Press, New York, 1992.
159. A. A. Savchenkov, V. S. Ilchenko, A. B. Matsko, and L. Maleki. KiloHertz optical resonances in
dielectric crystal cavities. Phys. Rev. A 70, 051804(R), 2004.
160. A. A. Savchenkov, V. S. Ilchenko, A. B. Matsko, and L. Maleki. Tunable filter based on whisper-
ing gallery modes. Electron. Lett. 39, 389–391, 2003.
161. A. A. Savchenkov, V. S. Ilchenko, A. B. Matsko, and L. Maleki. High-order tunable filters based
on a chain of coupled crystalline whispering gallery mode resonators. IEEE Photonics Technol.
Lett. 17, 136–138, 2005.
162. A. A. Savchenkov, A. B. Matsko, V. S. Ilchenko, N. Yu, and L. Maleki. Microwave photonics
applications of whispering gallery mode resonators physics and engineering of microwaves.
Millimeter and Submillimeter Waves and Workshop on Terahertz Technologies, 2007. MSMW
’07. The Sixth International Kharkov Symposium, 1, pp. 79–84, 2007.
163. B. E. Little, S. T. Chu, H. A. Haus, J. Foresi, and J. P. Laine. Microring resonator channel drop-
ping filters. J. Lightwave Tech. 15, 998–1005, 1997.
164. J. V. Hryniewicz, P. P. Absil, B. E. Little, R. A. Wilson, and P. -T. Ho. Higher order filter response
in coupled microring resonators. IEEE Phot. Tech. Lett. 12, 320–322, 2000.
165. S. T. Chu, B. E. Little, W. Pan, T. Kaneko, and Y. Kukubun. Cascaded microring resonators
for crosstalk reduction and spectrum cleanup in add-drop filters. IEEE Phot. Tech. Lett. 11,
1423–1425, 1999.
166. S. T. Chu, B. E. Little, W. Pan, T. Kaneko, and Y. Kukubun. Second-order filter response from
parallel coupled glass microring resonators. IEEE Phot. Tech. Lett. 11, 1426–1428, 1999.
167. K. Djordjev, S. J. Choi, S. J. Choi, and P. D. Dapkus. Microdisk tunable resonant filters and
switches. IEEE Phot. Tech. Lett. 14, 828–830, 2002.
168. O. Schwelb and I. Frigyes. Vernier operation of series coupled optical microring resonator
filters. Microwave Opt. Technol. Lett. 39, 258–261, 2003.
169. J.-L. Gheorma and R. M. Osgood. Fundamental limitations of optical resonator based high-
speed EO modulators. IEEE Photon. Tech. Lett. 14, 795–797, 2002.
170. G. Griffel. Vernier effect in asymmetrical ring resonator arrays. IEEE Phot. Technol. Lett. 12,
1642–1644, 2000.
171. P. Urquhart. Compound optical-fiber-based resonators. J. Opt. Soc. Am. A 5, 803–812, 1988.
172. K. Oda, N. Takato, and H. Toba. Wide-FSR waveguide double-ring resonator for optical FDM
transmission system. J. Lightwave Technol. 9, 728–736, 1991.
173. M. Mohageg, A. Savchenkov, D. Strekalov, A. Matsko, V. Ilchenko, and L. Maleki.
Reconfigurable optical filter. Electron. Lett. 41, 91–92, 2005.
174. C. K. Madsen and J. H. Zhao. A general planar waveguide autoregressive optical filter.
J. Lightwave Technol. 14, 437–447, 1996.
175. A. B. Matsko, A. A. Savchenkov, and L. Maleki. Vertically coupled whispering-gallery-mode
resonator waveguide. Opt. Lett. 30, 3066–3068, 2005.
176. L. Maleki, A. B. Matsko, A. A. Savchenkov, and D. Strekalov. Slow light in vertically coupled
whispering gallery mode resonators. Proc. SPIE, 6130, 61300R, 2006.
177. A. Yariv, Y. Xu, R. K. Lee, and A. Scherer. Coupled–resonator optical waveguide: a proposal and
analysis. Opt. Lett. 24, 711–713, 1999.
178. J. E. Heebner, P. Chak, S. Pereira, J. E. Sipe, and R. W. Boyd. Distributed and localized feedback
in microresonator sequences for linear and nonlinear optics. J. Opt. Soc. Am. B 21, 1818–1832,
2004.
179. M. Sumetsky. Vertically-stacked multi-ring resonator. Opt. Express 13, 6354–6375, 2005.
180. X. S. Yao and L. Maleki. Optoelectronic microwave oscillator. J. Opt. Soc. Am. B 13, 1725–1735,
1996.
181. D. Strekalov, D. Aveline, N. Yu, R. Thompson, A. B. Matsko, and L. Maleki. Stabilizing an
optoelectronic microwave oscillator with photonic filters. J. Lightwave Technol. 21, 3052–3061,
2003.
182. A. B. Matsko, L. Maleki, A. A. Savchenkov, and V. S. Ilchenko. Whispering gallery mode based
optoelectronic microwave oscillator. J. Mod. Opt. 50, 2523–2542, 2003.
183. A. B. Matsko, A. A. Savchenkov, N. Yu, and L. Maleki. Whispering-gallery-mode resonators as
frequency references. I. Fundamental limitations. J. Opt. Soc. Am. B 24, 1324–1335, 2007.
184. A. A. Savchenkov, A. B. Matsko, V. S. Ilchenko, N. Yu, and L. Maleki. Whispering-gallery-
mode resonators as frequency references. II. Stabilization. J. Opt. Soc. Am. B 24, 2988–2997,
2007.
185. M. L. Gorodetsky and I. S. Grudinin. Fundamental thermal fluctuations in microspheres. J. Opt.
Soc. Am. B 21, 697–705, 2004.
186. F. L. Walls and J. R. Vig. Fundamental limits on the frequency stabilities of crystal oscillators.
IEEE Tran. Ferroelectrics and Freq. Control 42, 576–589, 1995.
187. V. B. Braginsky, M. L. Gorodetsky, and S. P. Vyatchanin. Thermodynamical fluctuations and
photo-thermal shot noise in gravitational wave antennae. Phys. Lett. A 264, 1–10, 1999.
188. C. Audoin. Proceedings of the International School of Physics Enrico Fermi, Course LXVIII, A. F.
Milone and P. Giacomo (Eds). Amsterdam, North-Holland, p. 169, 1980.
189. A. Feldman, D. Horowitz, R. M. Waxler, and M. J. Dodge. Optical materials characterization.
National Bureau of Standards Technical Note 993. U.S. Government Printing Office, Washington
DC, 1979.
190. V. B. Braginsky, V. P. Mitrofanov, and V. I. Panov. Systems with small dissipation. The University
of Chicago Press, Chicago, IL, 1985.
191. K. Numata, A. Kemery, and J. Camp. Thermal-noise limit in the frequency stabilization of lasers
with rigid cavities. Phys. Rev. Lett. 93, 250602, 2004.
Contents
4.1 Introduction......................................................................................................................... 212
4.2 Polygonal-Shaped Microdisk Resonators with Directional Coupling........................ 214
4.2.1 Overview of Polygonal-Shaped Microresonators.............................................. 214
4.2.2 N-Bounce Orbits in Polygonal-Shaped Microresonators.................................. 216
4.2.3 Modes in Square-Shaped Microdisk Resonators............................................... 217
4.2.4 Directional Coupling via Polygonal-Shaped Microdisk Flat Sidewalls.........222
4.2.5 Sharp Corner Radiative Loss and Corner Rounding........................................225
4.2.6 Experimental Demonstrations.............................................................................. 226
4.3 Spiral-Shaped Microdisk Resonators with Nonevanescent Coupling........................ 227
4.3.1 Overview of Spiral-Shaped Microresonators..................................................... 227
4.3.2 Numerical Simulations.......................................................................................... 230
4.3.3 Experimental Demonstrations.............................................................................. 233
4.3.4 Tilted Notch-Coupled Waveguide Design for Mode Matching....................... 236
4.4 Silicon Electro-Optic Modulators Using Microdisk Resonators.................................. 238
4.4.1 Overview of Silicon Electro-Optic Modulators.................................................. 238
4.4.2 Principle of Microresonator-Based Modulators................................................. 239
4.4.3 Microdisk Resonator-Based Modulators............................................................. 240
4.4.3.1 Experimental Demonstrations............................................................... 241
4.4.3.2 Toward GHz-Speed Microdisk Resonator-Based Modulators.......... 241
4.5 Coherent Interference of Optical Resonances................................................................. 244
4.5.1 Overview of Coherent Interference in Photonic Resonators............................ 244
4.5.2 Reconfigurable Microring Resonator-Based Add-Drop Filters Using
Fano Resonances..................................................................................................... 245
4.5.3 Coherent Interference between a Resonance Pathway and a
Feedback Pathway.................................................................................................. 248
4.5.3.1 Coherent Feedback-Coupled Filters...................................................... 248
4.5.3.2 Coherent Feedback-Coupled Modulators and Logic Devices........... 253
4.6 Summary and Outlook...................................................................................................... 256
Acknowledgments....................................................................................................................... 257
References...................................................................................................................................... 257
211
© 2009 by Taylor & Francis Group, LLC
4.1 Introduction
Optical microresonators [1–5] that partially confine light by total internal reflection (TIR)
at the microresonator sidewalls are elegant micro-photonic structures. Such TIR-based
resonators have demonstrated high-Q resonances and can often be realized by means
of conventional fabrication techniques. Although early work on optical microresonators
dates back more than six decades [6–8], many high-impact microresonator technologies
from university and industry laboratories only came out in the 1990s. Recent progress in
microresonator technology is in part fueled by the availability of advanced fabrication
techniques for patterning fine structures on various material systems. Notably, we see
microresonator-based lasers [9–16] in different material systems and in different cavity
shapes. We also see microresonator-based filters [17–20], switches [21,22], and modulators
[19,23,24] for next-generation wavelength-division multiplexing (WDM) optical communi-
cations. Moreover, refractive-index-based biochemical sensors using microresonators are
also attracting considerable research interest [25–28].
Among all material systems, silicon, the most common semiconductor material for
computer chips, has only recently attracted a surge of rekindled interest for photonics
applications. Traditionally, silicon is not regarded as a photonic material other than for
the manufacture of photodetectors and solar cells. It is well-understood that silicon is an
indirect-bandgap material that does not efficiently emit light. Furthermore, bulk silicon is a
centrosymmetric crystal that does not exhibit linear electro-optic effect. Nonetheless, there
are also major merits that render silicon a good material platform for fabricating photonic
devices. On the material front, silicon is a robust material that is widely available and can
be purified to a high level. The conductivity of silicon can also be controlled by doping
impurities. Moreover, the natural oxide of silicon, silicon dioxide (SiO2), is a very stable
and excellent insulator, offering a very high etching selectivity with respect to silicon. On
the semiconductor optics front, silicon has a relatively wide bandgap of ~1.1 eV and thus
is transparent to the telecommunication wavelengths (~1.3 − ~1.7 µm). The high material
refractive index of ~3.45 also enables tight optical confinement. Hence, in hindsight it is not
surprising that silicon emerges as a material platform for fabricating highly compact low-
loss integrated waveguide devices for applications in telecommunication wavelengths.
The prospect of integrating microelectronics and photonics on the same silicon platform
using the complementary metal-oxide-semiconductor (CMOS) fabrication process is tech-
nologically appealing. Indeed, submicrometer-scale silicon wire waveguides have shown
a propagation loss of only ~1 − 2.5 dB/cm [29,30]. The waveguide propagation loss mainly
originates from sidewall roughness-induced scattering. Over recent years, considerable
research work [31] has been devoted to minimize sidewall roughness by using CMOS-
compatible deep-UV photolithography [29] or e-beam lithography [30] and optimized dry
etching processes (e.g. double thermal oxidation [30]). Furthermore, these wire waveguides
not only enable low-loss optical interconnection on a chip but also evanescent coupling to a
microstructure that is positioned in close proximity. In this respect, silicon photonics offers
a useful technology platform for fabricating and testing microresonator-based passive and
active devices, which would otherwise be more difficult or costly to fabricate using other
materials.
Over the past few years, we have witnessed unprecedented growth in research and
development activities on exploiting silicon for photonics [32]. Excellent reviews [33–36]
and textbooks [37,38] on silicon photonics have been written. One can perhaps argue that
the surge of silicon photonics research is driven by the increasing need of ultra-short-reach
Table 4.1
Summary of Our Experimental Demonstrated Polygonal-Shaped Microdisk Resonator-Based
Add-Drop Filters
Cavity Maximum FSR Extinction
Year Author Materials Shape Modes Q Finesse (nm) Ratio (dB)
2004 Ma et al. [71] SiN-on-SiO2 Hexagon Single 1,300 7 7.7 20
2004 Li et al. [74] SiN-on-SiO2 Octagon Single 1,150 5 7.1 3.6
2005 Li et al. [75] SiN-on-SiO2 Octagon Single 6,400 30 7.4 14
SOI Octagon Double 10,000 30 4.7 15
2006 Li et al. [76] SOI Hexagon Double 7,600 20 4.2 20
SOI Square Multi 2,600 8 4.8 25
input/output coupling with straight waveguides through a relatively wide gap, yet with-
out the complications of using racetrack-shaped microring resonators. However, based on
coupled-mode theory analysis, Manolatou et al. [66] also pointed out that resonances in
square-shaped microresonators are standing-wave modes. This means that for a square-
shaped microresonator-based add-drop filter with four symmetrically coupled waveguide
ports, there is only a maximum of 25% transmission at the drop-port.
In order to apply waveguide-coupled polygonal-shaped microdisk cavities for device
applications such as add-drop filters which impose traveling-wave resonances, our research
group have addressed key issues regarding their (i) standing-wave or traveling-wave modal
characteristics, (ii) waveguide-to-microresonator coupling characteristics, and (iii) cavity
loss mechanisms.
On addressing (i), Fong et al. [69] numerically examined the standing-wave modal
structures and transmissions of waveguide-coupled square-shaped microresonators
(see Section 4.2.3), and proposed that by properly cutting the square cavity corners the
traveling-wave components determined by the input-coupling direction can become dom-
inant [70]. Ma et al. [71] experimentally demonstrated that traveling-wave resonances in
waveguide-coupled hexagonal-shaped microdisk resonators can be preferentially coupled
by optimizing the waveguide width for phase matching the waveguide mode with the
cavity resonances (see Section 4.2.2). While, Li et al. [74] experimentally demonstrated
traveling-wave resonances in waveguide-coupled octagonal-shaped microdisk cavity-
based add-drop filters (see Section 4.2.6).
On addressing (ii), Li et al. [76] numerically simulated the effect of the cavity flat sidewall
length on the waveguide evanescent coupling, and revealed that the coupling is directional
(see Section 4.2.4). The coupled wavevector propagation direction essentially depends on
the waveguide width and the coupling sidewall length. The directional evanescent cou-
pling enables preferential coupling with only a few modes (and ideally singlemode) in albeit
highly multimode large-sized polygonal-shaped microcavities. As a proof-of-concept, Li
et al. [76] experimentally demonstrated only a few modes in waveguide-coupled 50-µm
square, hexagonal, and octagonal-shaped microdisk cavities on SOI substrates.
On addressing (iii), Zheng et al. [73] and Li et al. [75] also experimentally demonstrated that
the cavity loss due to the polygonal-shaped microcavity corners can be mitigated by round-
ing the sharp corners with a tailored radius of curvature (see Sections 4.2.5 and 4.2.6).
Meanwhile, Huang’s group analytically investigated resonance modes in a discrete
square-shaped microresonator. Huang’s group systematically classified the modes in
square-shaped microdisk resonators according to the symmetry point group C4v [78]. Based
on the so-called mode match method [79], they showed that for modes with integer mode
numbers in the x- and y-axes differing by two (see Section 4.2.3), the coupling between a
standing-wave mode and its 90°-rotated mode results in mode splitting. One of the split
modes displays WG-like standing-wave modal structure with zero-amplitude field at
the cavity corners, and thereby favors high-Q resonance. The other split mode exhibits
field extrema at the cavity corners and only supports low-Q resonance. Based on Fong et
al. [70] Huang’s group [80] further revealed that the cut-cornered square microresonator
traveling-wave modes can be due to mode-coupling between two modes that are related
by their spatial symmetry properties.
Other researchers also recognized the potentials of exploiting polygonal-shaped micro-
cavities as microlaser cavities. In contrast to the conventional circular-shaped microdisk
lasers that isotropically emit light along the cavity rim, polygonal-shaped microcavities
have the advantage of attaining multiple directional emissions from the flat sidewalls
[15]. Various polygonal-shaped microdisk or micropillar lasers have been demonstrated
including hexagonal-shaped micropillar lasers in dye-doped zeolites [81], square-shaped
microdisk lasers in dye-doped polymers [15], square-shaped micropillar lasers in
polymer-coated silica [82], and triangular-shaped microdisk lasers in InP/GaInAsP
systems [83].
(d) (e) (f )
θin θin
θin
Figure 4.1
Ray-optics picture of resonance orbits in (a, d) square-shaped microcavities, (b, e) hexagonal-shaped microcavi-
ties, and (c, f) octagonal-shaped microcavities. (a–c) N-bounce closed round-trip orbits. The dotted arrows in (b)
and (c) represent three- and four-bounce modes. (d–f) N-bounce open orbits that are wavefront-matched upon
each round trip. The dashed lines represent the matched wavefronts. θin: incident angle.
(a) (d) 80
TM 1.41/1.90/1.41
TM 1.41/2.78/1.41
(b)
θin 60 ky kwg
фin
50 kx
(c)
n1 n2 n1
θin w
40
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Waveguide width, w (µm)
Figure 4.2
(a–c) Schematics of waveguide-coupled (a) square-, (b) hexagonal-, and (c) octagonal-shaped microdisk resona-
tors. The cavity N-bounce ray orbits of incident angle θin are wavefront-matched with the waveguide propagation
modes of different waveguide widths w. (d) Calculated waveguide mode angle φin as a function of waveguide
width w, assuming symmetric slab structures with effective refractive index contrast n1/n2/n1 relevant to
our discussion. We choose n1/n2/n1 of 1.41/1.90/1.41 for the TM mode (circle), 1.41/2.78/1.41 for the TM mode
(triangle), and 1.62/2.78/1.62 for the TE mode (square). kwg: waveguide wavevector, ky: waveguide wavevector
longitudinal component, kx: waveguide wavevector transverse component.
[84]. Besides, odd M corresponds to opposite mode-field distribution parities in the x- and
y-directions, whereas even M corresponds to the same mode-field distribution parity in the
x- and y-directions.
We express the two-dimensional standing-wave scalar field pattern of mode (mx, my) in a
square-shaped microcavity as [69],
where ω is the mode angular frequency, and A is the field amplitude. We assume the x − y
coordinates origin at the lower left-hand corner of the square-shaped microcavity (see
Figure 4.3a). Figure 4.3a and b illustrates the calculated mode-field patterns of (mx, my) = (7, 9),
and (9, 7) using Equation 4.1.
We note that modes (mx, my) and (my, mx) in a square-shaped microcavity are spectrally
degenerate yet spatially non-degenerate. The spatially non-degenerate modes are related
by 90° rotation, as expected from a four-fold rotational symmetry. We also see from ray-
optics that the two spatially nondegenerate modes have incident angles θin and 90°-θin
on the same cavity sidewall. Thus, when the square-shaped microcavity is waveguide-
coupled, it is possible that only one of the spatially non-degenerate modes is wavefront-
matched with the waveguide mode.
In the case of spatially nondegenerate (mx, my) and (my, mx) modes coherently coupled,
the steady-state mode-field pattern is a linear superposition of the spectrally degenerate
modes as follows [69]:
Emx,my (x, y) eiωt = A eiωt sin(mx πx/a) sin(my πy/a) + B ei(ωt - δ) sin(my πx/a) sin(mx πy/a), (4.2)
o
x
Figure 4.3
Calculated standing-wave mode-field patterns of (a) (mx, my) = (7, 9), (b) (mx, my) = (9, 7), (c) superposition of modes
(7, 9) and (9, 7) with relative phase δ = 0, and (d) superposition of modes (7, 9) and (9, 7) with relative phase δ = π.
where A and B are the normalized field amplitudes of the spectrally degenerate modes,
and δ is the relative phase. Figure 4.3c and d illustrates the calculated mode-field patterns
of the supposition of modes (7, 9) and (9, 7) with t = 0, A = B, δ = 0 and π using Equation 4.2.
The in-phase supposition with field maxima at the cavity corners may induce an excessive
radiative loss, and thus suggest a low-Q resonance. Whereas, the π-out-of-phase super-
position with zero-amplitude field along the square-shaped microcavity diagonals may
favor a low radiative loss at the cavity corners, and thereby suggest a high-Q WG-like reso-
nance. We remark that our heuristic modeling is consistent with the theoretical analysis
by Huang’s group [78,79].
Here, we numerically simulate the transmissions and mode-field distributions of
silicon waveguide-coupled square-shaped microdisk resonators, using a commercially
available finite-difference time-domain (FDTD) simulation tool [85]. We adopt an effec-
tive refractive index approach to approximately account for the vertical dimension in
a t wo-dimensional (2-D) calculation. Figure 4.4 illustrates the silicon wire waveguide
cross-section and the effective 2-D structure for the FDTD calculations. We consider a
submicrometer singlemode silicon rib waveguide comprising a 0.34-µm thick silicon
waveguide core, a 0.05-µm thick silicon slab layer, with an air upper-cladding and a silica
under-cladding. We note that the thin slab layer structure adjacent to the waveguide side-
walls is instrumental for dopant implantations in electro-optic active devices (see Section
4.4.3). We simulate the transmission spectra for the TM polarization, to be consistent with
the scalar-field approximation adopted in the k-space modeling. Both the waveguide core
and the microdisk assume the same effective refractive index, neff = 2.78, while the slab
region assumes neff = 1.41.
Figure 4.5a shows the simulated throughput-, drop- and add-port transmission spectra of
the waveguide-coupled square-shaped microcavity-based add-drop filter of size a = 5 µm,
waveguide width w = 0.3 µm, and gap separation g = 0.2 µm for the TM polarization. The
transmission spectra are multimode. We identify the mode numbers using Fourier analysis
of the simulated mode-field patterns. We label the resonance by the dominant mode num-
bers (mx, my), indexed with the estimated relative phase δ (assuming 0 or π) based on the
calculated temporal evolution of the mode-field pattern (e.g. (11, 16)0, (12, 16)π). We see that
the drop-port and the add-port exhibit similar transmissions, with a maximum transmis-
sion of ~20%, as expected from standing-wave modes and in the presence of cavity radiative
loss. Figure 4.5b shows the schematic of an open-orbit ray propagating in a waveguide-
coupled square-shaped microcavity-based add-drop filter. The clockwise (CW) traveling-
wave orbit (solid arrows) walk-offs and backward-couples to the counterclockwise (CCW)
Throughput
Add
a
nair = 1.0
w
Silicon 0.34 µm nair = 1.0 neff = 1.41
0.05 µm nsi = 3.5 TM g
nsi = 3.5 neff = 2.78
Buried oxide
nox = 1.45 nox = 1.45
Silicon substrate
Input
Drop
Figure 4.4
Cross-section of the silicon rib waveguide and the effective 2-D structure for the TM-polarization. neff = 2.78 for
the 0.34-µm thick silicon waveguide core and microdisk, and neff = 1.41 for the 0.05-µm thick silicon slab region,
assuming buried oxide and air cladding.
(a) (b)
1.0
(13,15)π (13,13)
Throughput
Add
0.8
Normalized intensity
(14,14)
0.6 (13,14)π
(12,14)π
(10,17)0 (12,15)π
θin
0.4 (12,16)π (11,16) 0 (10,16)0
(11,17)π (11,15)π
0.2
Drop
Input
0.0
1400 1450 1500 1550 1600
Wavelength (nm)
Figure 4.5
(a) 2-D FDTD-simulated TM-polarized transmission spectra of the waveguide-coupled square-shaped
m icrocavity-based add-drop filter of a = 5 µm, w = 0.3 µm, and g = 0.2 µm. Solid line: throughput. Dashed line:
drop-port. Dotted line: add-port. The multimode resonances are labeled based on their simulated mode-field
patterns. (b) Schematic of a standing-wave orbit in a waveguide-coupled square-shaped microcavity-based
add-drop filter. Solid arrows: clockwise circulation; dotted arrows: counterclockwise circulation.
traveling-waves (dotted arrows) with the same incident angle, forming a standing-wave
mode which yields undesirable transmissions in the add-port. Furthermore, our numeri-
cal simulations reveal that modes of odd and even M exhibit parity at different transmis-
sions and mode-field patterns [69]. We find that the odd M modes display lower Q’s than
the even M modes. Figure 4.6a and b shows for instance the simulated mode-field pattern
and the Fourier transform analysis for odd M mode (11, 16)0. Figure 4.6c shows the cal-
culated mode-field pattern of mode (11, 16)0 in a discrete microcavity using Equation 4.2,
with A = 0.77 for mode (11, 16) and B = 0.23 for mode (16, 11) following the Fourier analysis.
In both the simulated and the calculated odd M mode-field patterns, we find field extrema
at the cavity corners, suggesting high radiative loss at the cavity corners and thus a low-Q
resonance. Whereas, Figure 4.6d and e shows the simulated mode-field pattern and the
Fourier transform analysis of even M mode (12, 16)π. Figure 4.6f shows the calculated mode-
field pattern of mode (12, 16)π, with A = 0.62 for mode (12, 16) and B = 0.38 for mode (16, 12).
Here, we see zero-amplitude fields at the cavity corners, and zero-amplitude field patterns
(a) (b) 25
(c)
20 (11,16)
15
ky(π/a)
10
(16,11)
5
0
0 5 10 15 20 25
kx(π/a)
(d) (e) 25
(f )
20 (12,16)
15
ky(π/a)
10
(16,12)
5
0
0 5 10 15 20 25
kx(π/a)
(g) (h) 25
(11,17) (i)
20
15
ky(π/a)
10
(17,11)
5
0
0 5 10 15 20 25
kx(π/a)
(j) (k) 25
(10,16) (l)
20
(12,14)
15
ky(π/a)
10
(14,12)
5
0
0 5 10 15 20 25
kx(π/a)
Figure 4.6
(a, d, g, j) FDTD-simulated mode-field patterns of a waveguide-coupled square-shaped microdisk resonator for
modes (11,16)0, (12,16)π, (11,17)π, and (12,14)π. (b, e, h, k) Fourier analysis of the simulated cavity mode-field pat-
terns. (c, f, i, l) Calculated mode-field patterns of a discrete square-shaped microcavity with the amplitude ratios
of the superposed modes following (b, e, h, k).
Figure 4.6g and h shows for instance the simulated mode-field pattern and the Fourier
transform analysis of even M mode (11, 17)π (|mx − my| = 6). Figure 4.6i shows the calculated
mode-field pattern of mode (11, 17)π, with A = 0.69 for mode (11, 17) and B = 0.31 for mode
(17, 11). We see more cross-like features than mode (12, 16)π with |mx − my| = 4 and mode
(7, 9)π with |mx − my|= 2 (see Figure 4.3d). Moreover, for the same even M, we note the
resonance Q scales inversely as |mx − my|. The maximum Q corresponds to |mx − my| = 2,
which gives a WG-like mode-field pattern [79] with a single cross-like feature at the cavity
center. For instance, Figure 4.6j and k shows the simulated mode-field pattern and the
Fourier transform analysis of mode (12, 14)π, which displays high-Q. Yet, we note that
the mode-field pattern is distorted due to the interference from mode (10, 16). Figure 4.6l
shows the calculated WG-like mode-field pattern of mode (12, 14)π, taking into account the
interference from mode (10, 16).
ky(π/a)
θin a
фin θin 6500
kx 20
5500
10
θin
0 4500
0 10 20 30 40 50
kx(π/a)
w g
(b) (d) (f )
50
2000
R 40
1800
30
ky(π/a)
1600
20
1400
10 ∆θ 1200
0
0 10 20 30 40 50
kx(π/a)
Figure 4.7
(a, b) Schematics of a waveguide-coupled microstructure with (a) a flat sidewall of length a, and (b) a curved
sidewall of radius R. The coupled k-vectors along the flat sidewall are directional. The coupled k-vectors along
the curved sidewall exhibit more angular components. φin: waveguide mode angle relative to the sidewall nor-
mal, θin: k-vector angle relative to the microdisk sidewall normal. (c, d) FDTD-simulated TE-polarized H-field
patterns for a silicon waveguide-coupled microstructure with (c) a flat sidewall of length a = 10 µm, and (d) a
curved sidewall of radius 5 µm. The dashed-line square box indicates the area of size a under Fourier analy-
sis. We adopt an effective refractive index contrast of 2.78–1.62, assuming a 0.2-µm thick silicon waveguide
core and microdisk and a 0.05-µm thick silicon slab region in SOI for the TE mode. (e, f) Fourier analysis of
the simulated field patterns in the microstructures. The contours represent the k-vectors spanning within
half-maximum of the angular distributions. The dashed arrows indicate the coupled k-vectors center angles
θin. The dashed lines indicate the k-space angular distribution ∆θ. (Reprinted from Li, C., Zhou, L., Zheng, S.,
and Poon, A. W., IEEE J. Select. Topics Quantum Electron., 12, 1438–1449, 2006. With permission.)
waveguide mode angles φin’s (see Figure 4.2d). Whereas, θin/∆θ linearly rises with a, and
displays no saturation up to θin/∆θ exceeding 30 for a≈50 µm, and is largely independent of
the waveguide widths adopted here.
Furthermore, we examine the waveguide-microstructure field amplitude coupling coef-
ficient κ under various configurations. We obtain κ from our simulations as the amplitude
ratio of the total field along the entire upper and right boundaries of the dashed-line square
(Figure 4.7c and d) to the input-coupled field amplitude at the waveguide. Figure 4.9 shows
the extracted κ with flat sidewalls of various a’s of different waveguide widths. We note
that κ rises with a, yet drops with w. Thus, we see that with fixed w, we can obtain higher
coupling coefficients for polygonal-shaped microdisks of a longer sidewall length.
Here, we assume the same waveguide coupling structure as above and numerically sim-
ulate a silicon waveguide-coupled octagonal-shaped microdisk resonator. In order to see
the effect of the microcavity flat and curved coupling sidewall, we also numerically model
a silicon waveguide-coupled circular-shaped microdisk resonator of the same cavity size.
Figure 4.10a and b shows the FDTD-simulated TE-polarized throughput- and drop-port
transmission spectra of the waveguide-coupled octagonal-shaped and circular-shaped
microdisk resonator-based filters. The octagonal-shaped microdisk has a sidewall-to-sidewall
(a) (b)
52 35
w = 0.34 µm 30
50
Central angle θin (°)
25
48 w = 0.3 µm 20
θin/∆θ
46 15
10 w = 0.26 µm
44 w = 0.26 µm w = 0.3 µm
5 w = 0.34 µm
42 0
0 10 20 30 40 50 0 10 20 30 40 50
Sidewall length a (µm) Sidewall length a (µm)
Figure 4.8
(a) FDTD-simulated coupled wavevectors center angle θin in the waveguide-coupled microstructures with
various flat sidewall lengths a, for waveguide width w = 0.34 µm, 0.3 µm, and 0.26 µm. θin saturates at angles
determined by the waveguide mode (shown as horizontal dotted lines). (b) FDTD-simulated coupled wavevec-
tors directionality measured as θin /∆θ in the waveguide-coupled microstructures with various flat sidewall
lengths a. The dashed line is visual aid. (Reprinted from Li, C., Zhou, L., Zheng, S., and Poon, A. W., IEEE J. Select.
Topics Quantum Electron., 12, 1438–1449, 2006. With permission.)
1.0
w = 0.26 µm
0.8
w = 0.3 µm
0.6
w = 0.34 µm
κ
0.4
w = 0.5 µm
0.2
w = 0.6 µm
0.0
5 10 15 20
Sidewall length a (µm)
Figure 4.9
Simulated coupling coefficient κ for waveguide coupling to planar microstructures as a function of sidewall
length with various waveguide widths. The dashed lines are visual aid. (Reproduced with permission from IEEE,
Li, C., Zhou, L., Zheng, S., and Poon, A. W., IEEE J. Select. Topics Quantum Electron., 12, 1438–1449, 2006. © IEEE.
With permission.)
distance L = 18 µm and a side-coupling length a≈7.5 µm. The circular-shaped microdisk has
a radius of R = 9 µm. Based on the waveguide-coupled microstructure analysis (Figure 4.8),
we choose waveguide width w = 0.3 µm with θin≈48.0° and ∆θ exceeding 9.0°. Our simulation
reveals only two traveling-wave modes (denoted as A and B) for the o ctagonal-shaped micro-
disk, whereas we discern five traveling-wave modes in the circular-shaped microdisk. This
exemplifies that a waveguide-coupled polygonal-shaped microdisk resonator with a long flat
sidewall is favorable for preferential coupling with a few and ideally single modes, whereas
the waveguide-coupled circular-shaped microdisk resonator tends to couple with a large
number of modes. Figure 4.10c shows the simulated steady-state H-field pattern at resonance
(a) (b)
Throughput Throughput
0 20 40
0
–10 20
Norm. Int. (dB)
60 2
ky (π/a)
0 40
1
20
θin ~ 45°
–0.8 0 0
0 20 40 60 80
kx (π/a)
Figure 4.10
(a) FDTD-simulated TE-polarized throughput- and drop-port transmission spectra of a waveguide-coupled
octagonal-shaped microdisk resonator-based filter with coupling length a≈7.5 µm, L = 18 µm, w = 0.3 µm, and
g = 0.3 µm. (b) FDTD-simulated TE-polarized throughput- and drop-port transmission spectra of a waveguide-
coupled circular-shaped microdisk resonator-based filter with R = 9 µm, w = 0.3 µm, g = 0.3 µm. (c) Simulated
steady-state H-field pattern of resonance A. (d) Fourier analysis of the mode-field pattern. The arrow indicates
the center angle θin≈45° and distribution ∆θ≈9.5° of the coupled k-vectors.
(a) (b)
L L
Diffraction
a a
Figure 4.11
Schematics of waveguide-coupled octagonal-shaped microresonator-based filters with (a) sharp cavity corners
that impose surface wave diffraction loss, and (b) rounded cavity corners that alleviate the surface wave dif-
fraction loss.
4.2.6 Experimental Demonstrations
Here, we review representative measurements of waveguide-coupled octagonal-shaped
microdisk resonators in a SiN-on-SiO2 substrate [75]. Figure 4.12a through f shows the
top-view scanning electron micrographs (SEMs) of the fabricated 50-µm microdisk reso-
nator-based filters with R/L = 0 − 0.5 (in steps of 0.1) [75]. The waveguide width is 0.6 µm.
Figure 4.12g through l shows the measured TM-polarized throughput-port and drop-
port transmission spectra. We observed that the sharp-cornered and round-cornered
octagonal-shaped microdisk resonators are nearly singlemode. In contrast, the circular-
shaped microdisk resonator (R/L = 0.5) exhibits multimode. With R/L = 0.4, we obtained
an optimum extinction ratio of ~14.5 dB, a Q~6,400, and an on-off ratio of ~13.5 dB. The
measured FSR for R/L = 0.4 is ~7.35 nm, which suggests a WG-like mode grazing along
the round-cornered octagonal-shaped microdisk circumference. Whereas, the measured
FSR for R/L = 0 and 0.1 is ~7.45 nm, which is consistent with relatively short eight-bounce
round-trip lengths [74].
Figure 4.13a shows the rising trend of the measured Q with R/L ratio for w = 0.6 µm, sug-
gesting that the round-cornered microresonators favor high-Q resonances. Figure 4.13b
shows the measured Q with various w for R/L = 0.3, revealing that the resonance Q is
optimized at w = 0.8 µm. We attribute the optimized Q to the nearly k-vector matching
between the microresonator mode and the waveguide mode. Based on the waveguide
dispersion analysis shown in Figure 4.2d, we see that for an effective refractive index
contrast n1 /n2 /n1 = 1.41/1.90/1.41 (for our fabricated device with a 1.1-μm thick SiN
device and a 0.15-μm thick SiN slab), the waveguide mode angle φin≈70° at w = 0.8 µm
is favorable for coupling to the octagonal-shaped microcavity eight-bounce or WG-like
orbits [76].
However, we found a number of anomalies in our measurements that are worth detailed
modeling. Specifically, we observed an exceptional case at R/L = 0.2 (Figure 4.12i), which
suggests weak coupling. Figure 4.13c shows the measured extinction ratio with R/L ratio
(and a) for w = 0.6 µm and w = 0.5 µm. For both w’s, the extinction ratio drops to a mini-
mum at R/L = 0.2 (a = 12.43 µm). We attribute the appearance of weak coupling to possible
destructive interference among multiple light paths that were coherently coupled along
this particular side-coupling length. Interestingly, we also observed that with R/L = 0.25
and w = 0.5 µm, the measured extinction ratio in the throughput-port transmission spec-
trum displays a maximum of ~24 dB. We found a corresponding narrowband dip near the
center of the drop-port peak transmission, as shown in Figure 4.13d. The drop-port sharp
resonance dips resemble resonance line shapes expected in coupled microresonators [88].
a –20 g
0
20 –40
b –20
h
0.1
–40
15
c –20 i
0.2
–40
Norm. Int. (dB)
10
–20
d j
0.3
–40
e –20
5 k FSR≈7.35 nm
0.4 Q≈6,400
–40
f –20
1
0 0.5
–40
Figure 4.12
(a–f) Scanning electron micrographs of the fabricated filters with R/L ratios from 0 (sharp-cornered) to 0.5
(circular). (g–l) Measured TM-polarized throughput-port and drop-port transmission spectra. Throughput
(solid line), drop (dashed line). (Reprinted from Optical Society of America. Li C., and Poon, A. W., Opt. Lett. 30,
546–548, 2005. With permission from the Optical Society of America.)
(a) 15 (b) 8
12
Measured Q (×103)
Measured Q (×103)
6
9
6
4
3
0 2
0.0 0.1 0.2 0.3 0.4 0.5 0.5 0.6 0.7 0.8 0.9
R/L Ratio w (µm)
Sidewall length a (µm)
(c) (d)
20 15 10 5 0
–20
20
Extinction ratio (dB)
10 –40
w = 0.6 µm –50
w = 0.5 µm
0
0.0 0.1 0.2 0.3 0.4 0.5 –60
1520 1525 1530 1535 1540
R/L Ratio
Wavelength (nm)
Figure 4.13
(a) Measured Q at various R/L ratios with w = 0.6 µm. (b) Measured Q for R/L = 0.3 at various waveguide width
w. (c) Measured extinction ratio at various R/L ratios for w = 0.6 µm (solid line) and w = 0.5 µm (dashed line). (d)
Measured TM-polarized throughput-port and drop-port transmission spectra with R/L = 0.25 and w = 0.5 µm.
(Reprinted from Optical Society of America. Li C., and Poon, A. W., Opt. Lett. 30, 546–548, 2005. With permission
form the Optical Society of America.)
r(фr)=r0(1-εфr/2π)
(a) (b) (c)
фr
r0 CW CCW
r0ε
Figure 4.14
(a) Schematic of the spiral-shaped microdisk resonator. (b) Schematic CW traveling orbit with partial transmis-
sion at the spiral notch for each round trip. (c) Schematic CCW traveling orbit that allows input-coupling via the
spiral notch yet the orbit bypasses the notch for each round trip.
coupling with the microresonator modes. Chern et al. [16] and various research groups
[90–94] have experimentally demonstrated unidirectional lasing emission with refractive
output-coupling from the spiral notch. Meanwhile, theoretical calculations of lasing modes
revealed distinct modal structures in a spiral-shaped microcavity [95,96]. Furthermore,
Chang’s group [97,98] also demonstrated novel optical devices with high-Q resonances
using refractive coupling via the spiral notch between a spiral-shaped microdisk laser and
a semicircular-shaped microdisk amplifier. Thus, it is known that refractive coupling via
the spiral notch preserves high-Q resonances.
Another uniqueness of spiral-shaped microresonators is that the cavity shape entirely
lacks rotational symmetry. It is thus conceivable that the rotational asymmetry breaks the
spatial degeneracy between CW and CCW traveling-wave modes, as depicted in Figure 4.14b
and c. We see that CW traveling-wave component spatially overlaps with the spiral notch
which enables partial transmission via the notch every round trip. Whereas, CCW traveling-
wave component tends to bypass the spiral notch yet can be refractively input-coupled at the
spiral notch.
Previously [99], in order to experimentally study CW and CCW traveling-wave modes
in spiral-shaped microresonators, we devised passive waveguide-coupled spiral-shaped
microdisk resonator-based filters, as shown in Figure 4.15. We seamlessly integrated a
waveguide to the spiral notch in order to facilitate direct coupling to the spiral-shaped
microdisk. Another waveguide was side-coupled to the microdisk, such that CW and CCW
transmissions can be separately characterized by input-coupling light into the microdisk
from each of the three waveguide ports. We remark that the waveguide notch-coupling is
fundamentally different from butt-coupling a waveguide to a circular-shaped microdisk.
In the spiral-shaped microdisk case, the cavity ray orbit has varying incident angle at each
reflection (due to the linear spiral shape) and that potentially facilitates the cavity mode
to be mode-matched with the notch-coupled waveguide mode. Whereas, in the circular-
shaped microdisk case, the WG mode has a fixed incident angle at each reflection, and
thus it is difficult to mode-match the butt-coupled waveguide with the WG mode without
severely spoiling the phase-matching condition or the resonance Q. It is worth mentioning
that Fujii et al. [93] also reported a spiral-shaped microdisk laser with a waveguide butt-
coupled at the spiral notch.
The filter functionalities depend on to which waveguide port the light is input-coupled.
Figure 4.15a shows the notch-filter functionality, which is based on CCW traveling-wave
modes that are evanescently input-coupled from the side waveguide yet tend to by-pass the
notch every round trip without direct output-coupling. Figure 4.15b shows the drop-filter
CCW CW CCW
Figure 4.15
Schematics of the waveguide-coupled spiral-shaped microdisk resonator-based filters: (a) notch filter with side-
coupled CCW traveling-wave modes, (b) drop filter with side-coupled CW traveling-wave modes, and (c) add-only
filter with notch-coupled CCW traveling-wave modes. (Reprinted from Optical Society of America. Lee, J. Y., Luo,
X., and Poon, A. W., Opt. Express 15, 14650–14666, 2007. With permission from The Optical Society of America.)
(a) 0
–5
–15
Q~500 FSR~21.6 nm
B
–10
–20
Figure 4.16
(a) FDTD-simulated TE-polarized throughput-port transmission spectra of notch filter (dashed line) and drop
filter (solid line). (b) FDTD-simulated TE-polarized drop-filter drop-port transmission spectrum (solid line) and
add-only filter throughput-port transmission spectrum (dotted line). We adopt ro = 10 µm, ε = 0.04, and effective
refractive index contrast of 1.9–1.4.
(iii)
(iii)
(i)
Figure 4.17
Loss mechanisms in a waveguide-coupled spiral-shaped microdisk resonator for CCW and CW traveling
waves. (i) Direct output-coupling via the notch-waveguide. (ii) Diffraction and scattering at the notch junction.
(iii) Distributed cavity loss along the cavity sidewall. (iv) Material absorption.
Figure 4.17 illustrates the various cavity loss mechanisms for CCW and CW waves. The
total cavity loss comprises (i) direct output-coupling via the notch-waveguide, (ii) dif-
fraction and scattering at the notch junction, (iii) distributed cavity loss along the cavity
sidewall (including curved sidewall diffraction, refraction, and roughness-induced scat-
tering), and (iv) material absorption. Given CW waves partially output-couple via the
notch-waveguide, the losses from (ii) and (iii) should therefore be lower. Whereas, it is
(a) CCW (notch filter) (b) CW (drop filter) (c) CCW (add-only filter)
20 1.2 20 1.2 20 1.2
Normalized field
Normalized field
Normalized field
Resonance A
amplitude
amplitude
amplitude
y (µm)
10 10 10
Normalized field
Normalized field
Resonance B
amplitude
amplitude
amplitude
y (µm)
10 10 10
Figure 4.18
FDTD-simulated TE-polarized steady-state resonance mode H-field patterns for (a), (d) notch filter, (b, e) drop
filter, and (c, f) add-only filter. (a–c) resonance A, (d–f) resonance B.
possible that CCW waves see relatively high losses from (ii) and/or (iii) in order to bal-
ance the relatively low-loss from (i). A large distributed cavity loss for CCW waves can
be originated from a less confined traveling-wave orbit, also suggesting CCW and CW
waves display different mode-field distributions. We remark that scattering at the notch
junction and along the cavity sidewall may also result in cross-coupling between CCW
and CW waves [16].
Figure 4.18a through f shows the FDTD-simulated TE-polarized mode-field (H-field)
patterns at resonances A and B (see Figure 4.16) for notch filter (a and d), drop filter
(b and e), and add-only filter (c and f). In both resonance wavelengths, our simulations
reveal WG-like traveling-wave mode-field patterns that are rotational asymmetric. The
simulated mode-field patterns are consistent with the theoretically calculated WG-like
resonances in a spiral-shaped microdisk without waveguide coupling [95,96], suggesting
that the perturbation from the notch-coupled waveguide to the spiral-shaped cavity
modes is not significant. Both resonances display qualitatively similar field patterns, with
weaker field amplitudes in notch filter than in drop filter or add-only filter. This shows
that the side-coupled CCW mode and the side-coupled CW mode exhibit asymmetrical
modal distributions despite their reciprocal transmissions. We attribute the asymmetri-
cal modal distributions between the side-coupled CCW and CW modes to the rotational
asymmetric cavity shape. This is also based on our earlier numerical work which sug-
gested that the side-coupled CCW and CW traveling waves have essentially identical
k-vector distributions at the input-coupling sidewall [99], and thus the input-coupling
conditions do not differentiate the steady-state CCW and CW mode-field patterns.
Besides, the side-coupled CCW mode and the notch-coupled CCW mode display non-
identical modal distributions, despite their same sense of circulation. We attribute the
larger field amplitude in the notch-coupled CCW mode than the side-coupled CCW mode
to a possible larger input-coupling at the notch. Whereas, we note that the side-coupled CW
mode and the notch-coupled CCW mode, while both enable spatial overlap with the notch,
exhibit field-patterns that are very similar to but not identical. Notably, for resonance B,
the notch-coupled CCW mode field-amplitude is about twice as high as that of the side-
coupled CW mode field-amplitude. This suggests that the notch input-coupling can be
more efficient than the conventional evanescent input-coupling given particular resonance
wavelengths.
4.3.3 Experimental Demonstrations
Here, we review our experimental demonstration of reciprocal transmissions and asym-
metrical modal distributions in waveguide-coupled spiral-shaped microdisk resonators.
We fabricated the resonators on a SiN-on-SiO2 substrate using standard silicon microelec-
tronics processes [99]. Spiral-shaped microdisks with fixed radius (r0 = 25 µm) yet various
deformations (ε = 0.016, 0.032, 0.064, 0.08, and 0.16) were fabricated in the same chip. Figure
4.19a shows the SEM of the spiral-shaped microdisk filter with r0 = 25 µm and ε = 0.016. The
ε value gives a notch size of 0.4 µm for singlemode waveguide notch-coupling. Figure 4.19b
through d shows the zoomed SEMs of the notch coupling regions with ε = 0.016, 0.032, and
0.064. The wider notch waveguide supports multimode. We note that the side-coupled
waveguide width can be separately optimized for singlemode propagation.
We measured the transmission spectra of the notch/drop/add-only filter configurations
by separately input-coupling light to the three waveguide ports [99]. Figure 4.20 schemati-
cally shows the experimental setup. The inset shows the optical micrograph of the filter
configurations. For each input-coupling configuration, we also collected the out-of-plane
light scattering near the notch coupling region and the side-coupling region by linearly
scanning a vertical lensed-fiber at a height of a few microns above the cavity top sur-
face. The light scattering spectra and spatial distributions indicated the cavity mode-field
distributions.
Figure 4.21a shows the measured TE-polarized throughput-port transmission spectra
of notch and drop filters [99] for the spiral-shaped microdisk with ε = 0.016. The evanes-
cently input-coupled CCW and CW traveling-wave transmissions are reciprocal related
with essentially identical resonances and line shapes. The FSR of ~7.2 nm suggests a
wavefront-matched round-trip path length that is consistent with the spiral-shaped
(a) (b)
r0ε~0.4 µm
(c)
r0~25 µm r0ε~0.8 µm
(d)
r0ε~1.6 µm
w~0.4 µm
Figure 4.19
(a) SEM of the fabricated spiral-shaped microdisk resonator-based filter in SiN. (b–d) Zoom-in SEMs of the
notch junctions with ε = 0.016, 0.032, and 0.064.
Out-of-plane
scattering
PM SM lensed-fiber
PM SM
Input lensed-fiber Output
Output Input
Chip
Input Throughput
CW input
CCW input
(drop filter)
(notch filter)
Input
Throughput CCW
CW
µm
50
Drop
CCW input
(add-only filter)
Add
Figure 4.20
Schematic of the experimental setup for transmission and out-of-plane scattering measurements of the three
filter configurations. Bottom inset: optical micrograph of the spiral-shaped microdisk filter with arrows denot-
ing the three filter operations. PM: polarization maintaining, SM: singlemode. (Reprinted from Optical Society
of America. Lee, J. Y., Luo, X., and Poon, A. W., Opt. Express 15, 14650–14666, 2007. With permissoin from The
Optical Society of America.)
(a) (b)
–10
–20
Norm. Int. (dB)
Q~8,000 –30
–20 FSR~7.2 nm Q~12,000
Q~9,000
–40
1550 1555 1560 1550 1555 1560
(c) Wavelength (nm) Wavelength (nm)
(d)
–15
Norm. Int.y (dB)
1540 1545 1550 1555 1560 1540 1545 1550 1555 1560
Wavelength (nm) Wavelength (nm)
(e) (f )
–15
–15
Norm. Int. (dB)
Norm. Int. (dB)
–20
–20
–25
–25
–30
Q~700
–30
1550 1555 1560 1565 1570 1550 1555 1560 1565 1570
Wavelength (nm) Wavelength (nm)
Figure 4.21
(a, c, e) Measured TE-polarized throughput-port transmission spectra of notch filter (dashed line) and drop
filter (solid line). (b, d, f) Measured TE-polarized drop-port transmission spectrum of drop filter (solid line) and
throughput-port transmission spectrum of add-only filter (dotted line). The spiral-shaped microdisk has r0 = 25
µm. (a, b) ε = 0.016, (c, d) ε = 0.032, and (e, f) ε = 0.064. (Figure 4.21a and b are reproduced from Optical Society
of America. Lee, J. Y., Luo, X., and Poon, A. W., Opt. Express 15, 14650–14666, 2007. With permission from The
Optical Society of America.)
at the notch junction. Yet, the light scattering intensity profiles vary in detail among the
three filter configurations, suggesting asymmetric modal distributions that depend on the
input-coupling port. Each filter configuration displays light scattering intensity profiles
that are consistent between the two resonance wavelengths.
Figure 4.22d depicts the measured light scattering spectra at the side-coupling region.
The estimated Q values from the light scattering spectrum of the drop filter configura-
tion are again consistent with those estimated from the transmission spectra. Figure 4.22e
and f shows the light scattering profiles near the side-coupling region spanning ~13 µm
in the x-direction at resonance wavelengths of 1552.06 nm and 1553.02 nm. At each reso-
nance wavelength, the scattering intensity profiles again vary among the three filter con-
figurations, suggesting cavity modal structures that depend on the input-port. Moreover,
the light scattering profile distributions in the vicinity of the side-coupling region are
significantly different from those collected in the vicinity of the notch-coupling region,
indicating a nonsymmetric modal distribution for each filter configuration.
CCW
Drop Add-only (add-only filter)
filter filter
Figure 4.22
(a) Measured out-of-plane light scattering spectra near the notch junction of ε = 0.016 filter. The estimated Q val-
ues are labeled. (b–c) Light scattering intensity profiles over a ~11-µm span near the notch junction in the x-di-
rection at resonances of (b) 1552.06 nm, and (c) 1553.02 nm. (d) Measured out-of-plane light scattering spectra
near the side-coupling region. (e–f) Light scattering intensity profiles over a ~13-µm span near the side-coupling
region in the x-direction at resonances of (e) 1552.06 nm, and (f) 1553.02 nm. The SEMs depict the scanning fiber
field of view, positions, and the scanning directions. (Reprinted from Optical Society of America. Lee, J. Y., Luo,
X., and Poon, A. W., Opt. Express 15, 14650–14666, 2007. With permission from the Optical Society of America.)
(a)
–10
Figure 4.23
(a) Schematic of the waveguide-coupled spiral-shaped microdisk resonator-based filter with an angle-tilted
notch-coupled waveguide. Arrows depict the light propagation. Dashed line: matched wavefront. Inset: Design
parameters of the spiral notch-to-waveguide junction. θt: tilting angle; d: width of the tilted waveguide. (b) SEM
of our fabricated device in SiN, with r0≈25 µm, ε≈0.016, w≈0.5 µm, and g≈0.3 µm. Left inset: zoom-in SEM of the
side-coupling region. Bottom insets: zoom-in SEMs of the notch-coupling region with different tilting angles.
(c–f) Measured TE-polarized throughput-port transmission spectra with different angle-tilted notch-coupled
waveguides for ε = 0.016. Both Q and extinction ratio vary with the tilt angle.
transmission spectra of drop filter. The measured FSR of ~6.8 nm is consistent with the
microcavity circumference. The observed inhomogeneously broadened resonance Q’s and
extinction ratios significantly vary with θt. The measured transmission spectra thus sug-
gest that the notch coupling efficiency and/or the cavity loss can be tailored by varying the
notch-waveguide tilting angle. We note that the detailed dependence on θt also depends
on the spiral shape parameter ε.
where ∆ne (∆nh) is the change in refractive index resulting from a change in free electron
(hole) carrier concentration (cm-3), ∆Ne (∆Nh), ∆αe (∆αh) is the change in absorption coeffi-
cient resulting from a change in free electron (hole) carrier concentration (cm-3), ∆Ne (∆Nh).
There are three common types of silicon electro-optic modulators depending on the
microelectronic devices employed to induce carrier dispersion effect: (i) carrier-injection
type using forward-biased p-i-n diodes [23,24,56], (ii) carrier-accumulation type using
metal-oxide-semiconductor (MOS) capacitors [41,10], and (iii) carrier-depletion type using
reverse-biased p-n diodes [55,102 − 105]. Figure 4.24a through c illustrates the three types
of silicon electro-optic modulators.
n+-poly
––––––––
Si + + + ++ + + + – – – –
p+ + p+ p+ – – n+
+ + –– – n p-Si + p-Si
+ + – + ++
Buried oxide Buried oxide Buried oxide
Figure 4.24
Schematics of three common mechanisms of silicon electro-optic modulators using carrier dispersion effect.
(a) Carrier-injection-based using a forward-biased p-i-n diode. (b) Carrier-accumulation-based using a MOS
capacitor. (c) Carrier-depletion-based using a reverse-biased p-n diode. The dashed arrows represent the carri-
ers transport directions.
(a) (b)
Free carries
“on”
“off ”
“on”
Intensity
λ0 Optical response
“off ”
λoff
TO
EO λon
λon λoff
Wavelength Time
Figure 4.25
(a) Schematic resonance modulation upon a square-wave electrical signal with forward and reverse biases. Solid
line: resonance during off-cycle of the electrical signal, dotted line: resonance during on-cycle. TO: Thermo-
optic effect, EO: Electro-optic effect. (b) Schematic free-carrier concentration modulation and the resulting
modulated optical signals at wavelengths λoff and λon.
As the driving signal period becomes comparable to the carriers’ transit times, the car-
rier injection into and extraction from the p-i-n diode intrinsic region cannot totally fol-
low the driving signal. This results in congestion of carriers in the intrinsic region, and
consequently narrows the dynamic range of the resonance wavelength shifts. The reso-
nance wavelengths of the diode off- and on-cycles thus approach each other, resulting in
an electrically-limited modulation depth.
The modulator bandwidth can therefore be electrically limited by the carrier dynam-
ics or optically limited by the photon cavity lifetime τph [56], depending on which sets the
lower limit. For example, assuming a fast 10-ps carrier transit time in a carrier-injection or
carrier-depletion-based modulator, and a relatively low-Q of 103, the modulator bandwidth
is electrically limited as it only takes τph≈Q/ω~1 ps to build up or decay the optical modes
but ~10 ps to build up or decay the carrier concentrations. Whereas, for a relatively high-Q
of 105 (τph<100 ps), the modulator is optically limited as it now takes much longer time to
build up or decay the optical mode than to build up or decay the carrier concentrations.
We note that numerical modeling work suggested that silicon microring modulators using
an integrated p-i-n-i-p diode structure can reach 40 Gb/s [107].
It is also important to attain a large extinction ratio for microresonator-based modulators by
approaching the critical coupling condition [108]. In the case of a single-waveguide-coupled
microresonator configuration (notch filter), the critical coupling imposes τ = Ac, where τ is
the bus waveguide field amplitude transmission coefficient, and Ac is the fraction of the
electric field amplitude that remains upon a cavity round-trip (Ac = 1 means lossless). For
instance, for modulator with an electrically limited bandwidth using a cavity of Q≈103,
Αc~0.9 and thus τ~0.9 in order for an extinction ratio larger than 20 dB. This suggests a
relatively large waveguide coupling coefficient κ~0.4 (assuming lossless waveguide-
microresonator coupling τ2 + κ2 = 1), imposing either a narrow coupling gap or a long inter-
action length for the conventional microdisk and microring resonators.
near the microdisk rim, we can selectively inject carriers into the optical modes using
an embedded p-i-n diode, which has the lossy heavily-doped regions only in the micro-
disk center and outside the microdisk rim. In addition, the heavily-doped region in the
microdisk also serves to suppress high-order WG modes that extend deep into the cavity,
enabling the microdisk modulator to be ideally singlemode [109].
Figure 4.26a shows the schematic of our silicon electro-optic modulator comprising a
10-µm diameter microdisk resonator selectively embedded with a lateral p-i-n diode on
an SOI substrate. The microdisk rim region is left undoped, and thus enables low-loss
propagation for the low-order WG modes. It is also critical that the doped region in the
slab layer surrounds almost the entire microdisk [23], thus enabling a uniform carrier
injection into the microdisk rim and a fast sweeping out of the injected carriers before car-
rier recombination. The inset shows the schematic cross-section of the lateral p-i-n diode
in our microdisk modulator. Although the p-i-n diode intrinsic region width of 2.25 µm is
sufficiently wide to house low-order WG modes, it is too wide for attaining a GHz-speed
response.
Figure 4.26b shows the optical micrograph of our fabricated 10-µm diameter microdisk
modulator. The dashed lines highlight the bus waveguide and the microdisk resonator
underneath the oxide upper-cladding layer. Aluminum wires are connected to the p-i-n
diode doped regions. We detailed the fabrication processes elswhere [24].
242
(a) (b)
Output 0.35 µm
Output
1×1020 cm–3
e
Microdisk 2.25 µm
av
n+ Doped
W
p+
0.21 µm
Trench n+ Si Trench
Input
Si 0.21 µm p+ 30 nm
SiO2 2 µm SiO2 3 µm 2 µm Waveguide
Si substrate
Si substrate AI
5 µm
Input
Figure 4.26
(a) Schematic of the silicon microdisk resonator-based modulator using a selectively embedded p-i-n diode. The p-i-n diode (and a 2-µm wide trench) surrounds almost
the entire microdisk except in the vicinity of the coupled waveguide. Inset: cross-sectional view of the lateral p-i-n diode. The n + -doped region radius is 3.25 µm.
The p + -doped region width is 3 µm. The p-i-n diode intrinsic region width is 2.25 µm. (b) Optical micrograph of our fabricated p-i-n diode integrated microdisk
modulator. (Reprinted from Optical Society America. Zhou, L., and Poon, A. W., Opt. Express 14, 6851–6857, 2006. With permission form The Optical Society of
America.)
2/23/09 12:08:09 PM
(a) (c)
–20
0
Normalized intensity (dB)
–25
Voltage (V)
–24 Mode B Input
0
Pulse 5 ns
Normalized intensity (dB)
–3
–6
–28 6 Mode A
Optical signal (a.u.)
–32 ~9 dB 2
10 Mode B
0V
–36 5
~0.07 nm 0.85V (24 µA)
0
1547.0 1547.2 1547.4 1547.6 0 4 8 12
Wavelength (nm) Time (ns)
Figure 4.27
(a) Measured TE-polarized throughput-port transmission spectrum of the microdisk resonator-based modu-
lator without biasing the embedded p-i-n diode. ER: extinction ratio. (b) DC tuning of mode B. Resonance
wavelength blueshifts by ~0.07 nm upon a 0.85 V forward bias across the p-i-n diode. (c) Measured modulation
bandwidths for mode A (open circles) and mode B (solid squares). The driving electrical signal is 0.9 V forward
bias and -6.0 V reverse bias. (d) Optical transient responses to a single 5-ns input electrical pulse for modes A
and B. Mode A: τrise~ 417 ps and τfall~1.1 ns; mode B: τrise~850 ps and τfall~2.1 ns. (Reprinted from Optical Society
America. Zhou, L., and Poon, A. W., Opt. Express 14, 6851–6857, 2006. With permission from The Optical Society
of America.)
i. On the electrical front, our semiconductor device simulations (using MEDICI) sug-
gest that with the p-i-n diode intrinsic region width of 1.5 µm and the microdisk
etched height of 0.18 µm, we can attain a rise time of ~0.33 ns and a fall time of
~0.37 ns (i.e. GHz modulation bandwidth) using a forward bias of 1.25 V and a
reverse-bias of - 4 V. A sub-microampere injection current is sufficient to induce
a refractive index change of ~10-3, which enables tuning one resonance line width
(0.36 nm) of a Q~4,500 mode.
ii. On the optical front, our 3D-FDTD simulations using a commercially available
FDTD simulation tool [110] suggest that we can adopt a 5-μm diameter microdisk
with the p-i-n diode as mentioned in (i) in order to confine the first-order WG mode
within ~0.75 μm distance from the microdisk sidewall. Figure 4.28a and b shows that
(a) (c)
0.6 0.6
0.4 0.4
z (µm)
z (µm)
n+ p+ n+ p+
0.2 0.2
0 0
0.5 1.0 1.5 2.0 2.5 3.0 0.5 1.0 1.5 2.0 2.5 3.0
(b) (d)
3.5 0.35
3 0.3
Normalized E-field
Normalized E-field
2.5 0.25
2 0.2
1.5 0.15
1 0.1
0.5 0.05
0 0
0.5 1.0 1.5 2.0 2.5 3.0 0.5 1.0 1.5 2.0 2.5 3.0
x (µm) x (µm)
Figure 4.28
3-D FDTD-simulated TE-polarized (a), (b) first- and (c), (d) second-order WG mode-field distributions of a 5-µm
diameter silicon microdisk resonator with the selectively embedded highly doped regions. (a, c) Cross-sectional
view. (b, d) Radial distribution in the mid-height of the microdisk (z = 0.3 µm). The cavity-field amplitude is
normalized to the input E-field amplitude.
the WG mode can be properly confined without significant spatial overlaps with or
attenuation by the highly doped regions inside and outside the microdisk. In con-
trast, Figure 4.28c and d shows that the mode field amplitude of the second-order
WG-mode is about an order of magnitude smaller than the first-order mode.
(a) Transmission
Asymmetric fano
lineshape
Interference
exp(iф)
Coherent background
(b)
exp(iф)
Selectively ER
tuning
EIT
L
(c)
EIT
exp(iф)
Interference
Figure 4.29
Schematics of three microresonator-based devices utilizing coherent interference of optical resonances. (a) Fano
interference: a resonance pathway interferes with a coherent background pathway. (b) Coherent feedback: a
resonance pathway interferes with a coherent feedback. (c) EIT: a resonance pathway interferes with another
resonance pathway of the same or detuned wavelength. ER: extinction ratio.
Directional SiO2
κ
coupler Out2
n+ Si p+
τ
SiO2
n+
Si substrate
Input p+ Out1
Phase
shifter
(b)
λ2 Microring
resonator
Add MZI Port 2 λ1, λ3
Throughput
Input Drop
∆Φ = 0
λ1, λ2, λ3 Port 1 λ2
Phase shifter
λ1 λ2 λ3
(c)
λ2
Port 2 λ2
Add
Drop
Input Throughput
∆Φ = π
λ1, λ2, λ3 Port 1 λ1, λ3
λ1 λ2 λ3
Figure 4.30
(a) Schematic of the MZI-coupled racetrack-shaped microring resonator-based reconfigurable add-drop filter.
Inset: schematic cross-sectional view of the integrated lateral p-i-n diode in the phase shifter. (b, c) Schematics
of the Fano resonance-based reconfigurable add-drop filter transmissions upon two relative phase conditions:
(b) without phase shift (∆Φ = 0), and (c) with π phase shift (∆Φ = π) of the nonresonance arm. (Reprinted from
Optical Society of America. Zhou, L., and Poon, A. W., Opt. Lett. 32, 781–783, 2007. With permission from The
Optical Society of America.)
Figure 4.30b and c illustrates the device principle. Without an induced phase shift at the
coherent background pathway (∆Φ = 0), off-resonance wavelengths λ1 and λ3 are output-
coupled at port-2 acting as a throughput-port, while resonance wavelength λ2 is routed to
port-1 acting as a drop-port. With a ∆Φ = π phase shift imposed on the coherent background
pathway, the output-port transmissions are swapped with off-resonance wavelengths λ1
and λ3 routed to port-1 and resonance transmission of λ2 at port-2. As an add-function, in
the case that resonance wavelength λ2 is launched from the add-port, then λ2 is output-
coupled at port-2 with ∆Φ = 0 and routed to port-1 with ∆Φ = π.
We model the MZI-microresonator output-ports transmission spectra using scattering-
matrix approach, following Zhou and Poon [124]. We represent the output-ports electric
fields, Eo1 (port-1) and Eo2 (port-2), and the input-port and add-port electric fields, Ein and
Ead, in matrix form as follows,
τ - Ac e - iϕ c - iβL
Eo1 cos θ2 - i sin θ2 - iϕ c
e 0 cos θ1 - i sin θ1 Ead (4.5)
=
E - i sin θ 1 - τA e
o2 cos θ
c - i sin θ cos θ1 Ein
2
2
0 te - i(βL- ∆Φ))
1
where θ1 and θ2 determine the directional couplers power coupling ratios, ϕc is the cavity
round-trip phase, β is the propagation constant in the wire waveguide, L is the MZI arm
length, and ∆Φ = (∆Φo + ∆Φc) is the phase difference between the resonance arm and the
coherent background arm, comprising an initial phase mismatch ∆Φo and a free-carrier
induced phase shift ∆Φc = −∆neffk0L. We assume the waveguide effective refractive index
is neff ( = β/k0, where k0 is the free-space propagation constant) and its free-carrier induced
dispersion and absorption are ∆neff and ∆αeff. The fraction of the electric field transmitted
through the coherent background arm is given as t = exp (−∆αeffL/2).
Assuming ideal 3-dB couplers (θ1 = θ2 = π/4) and the light is launched from the input-port
(Ead = 0), we express the input-normalized transmission intensities as follows [124],
2 2
Io1 Eo1 1 τ - Ac e - iϕ c - iβL
= = - e + te - i(βL- ∆Φ ) (4.6)
I in Ein 4 1 - τAc e - iϕ c
2 2
Io 2 Eo 2 1 τ - Ac e - iϕ c - iβL
= = e + te - i(βL- ∆Φ ) (4.7)
I in Ein 4 1 - τAc e - iϕ c
Equations 4.6 and 4.7 show that Io1 and Io2 are interchanged every π phase change of ∆Φ.
Specifically, in the case that ∆Φ equals even (odd) numbers of π, Io1 displays a symmet-
ric resonance transmission peak (transmission dip) and Io2 exhibits a symmetric reso-
nance transmission dip (transmission peak), suggesting drop/throughput functionalities.
Likewise, we can deduce the add functionality assuming the light is launched from the
add-port (Ein = 0).
Figure 4.31a through c shows the top-view and zoom-in view optical micrographs of our
Fano resonance-based device. The MZI arm length is 1.2 mm in order to accumulate 2π
phase tuning. Two long aluminum pads are connected to the laterally embedded 1.2-mm
long p-i-n diode. However, we caution that the long metal pads do not favor high-speed
switching.
Our experiment revealed that the Fano resonance-based add-drop filter can be electri-
cally reconfigured by a sub-0.1 V change in the driving voltage. Figure 4.32a through e
(a)
50 µm
Pad
Output
Input 2
1
(b) (c)
5 µm 0.5 µm
Microresonator p-i-n diode
Trench
25 µm
Waveguide
10 µm
Aluminum
Waveguide
Figure 4.31
(a) Optical micrograph of the fabricated device. The MZI arm length is 1.2 mm. (b) Optical microscope zoom-in
view of the racetrack-shaped microring coupled straight waveguide. The microring curved waveguide radius is
25 µm. The lateral interaction length is 10 µm. (c) Optical microscope zoom-in view of the p-i-n diode embedded
phase shifter. (Reprinted from Optical Society of America. Zhou, L., and Poon, A. W., Opt. Lett. 32, 781–783, 2007.
With permission from The Optical Society of America.)
(a) Vd = 0 V (f ) ∆Φc = 0
1 0.4
Port-1
0 0.0
2 0.8
(b) (g) ∆Φc = 0.6π
1 Vd = 0.85 V 0.4
0 0.0
Normalized intensity
2 Vd = 0.89 V 0.8
Intensity (a.u.)
0 0.0
2 0.8
Vd = 0.92 V
1 (d) (i) ∆Φc = 1.6π 0.4
0 0.0
2 Vd = 0.96 V 0.8
(e) (j) ∆Φc = 2.2π
1 0.4
0 0.0
1562 1563 1564 1565 1562 1563 1564 1565
Wavelength (nm) Wavelength (nm)
Figure 4.32
(a–e) Measured TE-polarized output-ports transmission spectra of the racetrack-shaped microring resonator-
coupled MZI with bias voltages Vd = (a) 0 V, (b) 0.85 V, (c) 0.89 V, (d) 0.92 V, and (e) 0.96 V. (f–j) Scattering-matrix-
modeled output-ports transmission spectra with the free-carrier induced phase shift ∆Φc = (f) 0, (g) 0.6π, (h) 1.1π,
(i) 1.6π, and (j) 2.2π. Solid lines: port-2, dotted lines: port-1. (Reprinted from Optical Society of America. Zhou, L.,
and Poon, A. W., Opt. Lett. 32, 781–783, 2007. With permission from The Optical Society of America.)
Input p+
Vd
Ei n+ Feedback
E1
E2 n+ Si p+
Eo SiO2
Si substrate
Output
Figure 4.33
Schematic of a silicon waveguide cross-coupled microring resonator-based reconfigurable filter. Inset: cross-
sectional schematic of the lateral p-i-n diode embedded in the external feedback-waveguide section. Vd: driving
voltage. (Reprinted from Optical Society of America. Zhou, L., and Poon, A. W., Opt. Express, 15, 9194–9204, 2007.
With permission from The Optical Society of America.)
on the coherent cross-coupled feedback phase and amplitude. Thus, by means of electri-
cally tuning the feedback phase and amplitude using carrier dispersion effect, we can
selectively enhance or suppress resonances in the waveguide transmission.
Using scattering-matrix approach [130], we express the steady-state relationship between
the input- and output-coupled fields, Ei and Eo, as,
- iϕ
Eo τe ke - i(ϕ + π/2 ) bγe - iφ 0 τe - iϕ ke - i(ϕ + π/2 ) Ei
=
E2 ke - i(ϕ + π/2 ) (4.8)
τe - iϕ 0 ae - iθ ke - i(ϕ + π/2 ) τe - iϕ E1
We also express the steady-state relationship between the microring internal fields E1 and
E2 as
E1 = ae − iθ E2 (4.9)
where E1 and E2 are the fields inside the microring just prior to the input-coupler and
just after the output-coupler, κ is the coupling coefficient of the symmetric input/output-
couplers (τ2 + κ2 = 1 for lossless coupling), ϕ is the couplers transmission phase change (ϕ = 0
for circular microring), a is the half-circular microring transmission factor, θ is the half-
circular microring phase change, b is the feedback-waveguide amplitude transmission fac-
tor, γ is the amplitude transmission factor under free-carrier absorption (γ = 1 means no
carrier injection), and φ = φo + δφ is the feedback-waveguide phase change including a pas-
sive path-length phase change φo and a carrier-injection induced phase shift δφ. We note
that θ is related with the half-circular microring length La as θ = neffLa(2π/λ). Whereas, φo is
related with the feedback-waveguide length Lb as φο = neffLb(2π/λ). We remark that δφ and
γ are related by carrier dispersion effect.
From Equations 4.8 and 4.9, we obtain the filter transmission as,
2 2
Iout Eo τ 2bγe - i( φ+ 2 ϕ ) + k 2 ae - i(θ+ 2 ϕ + π ) + a 2bγe - i( 2 θ+ 4ϕ + φ+ π )
= = (4.10)
I in Ei 1 - [ τ 2 a 2 e - i( 2 θ+ 2 ϕ ) + k 2 abγe - i(θ+ 2 ϕ + φ+ π ) ]
ϕ ϕ
θ θ ф
Figure 4.34
Schematics of two approximations of the round-trip phase change for the waveguide cross-coupled microring
resonator. (a) Microring approximation. (b) Feedback-loop approximation. Arrows represent the light propaga-
tion directions.
Intensity
λm+1 FSR λm λm–1 λm+1 FSR λm λm–1
λm+1
∆ф1
γ1
∆фm+1 ∆фm+1
∆фm 2nπ ∆фm+1–2nπ
∆ф
∆ф
∆фm–1 ∆фm+1–4nπ
2nπ
Intensity
Intensity
γ1
Wavelength Wavelength
Figure 4.35
(a) Schematics of the resonance-dependent coherent feedback interference. Resonances λm + 1 and λm see dif-
ferent feedback phase angles ∆φm + 1 and ∆φm, resulting in different line shapes. (b, c) Schematics of the micror-
ing resonance spectrum, the wavelength-dependent feedback phase ∆φ, and the coherent feedback-interfered
transmission spectrum. (b) ∆φ does not satisfy the phase-matched feedback condition. (c) ∆φ nearly satisfies the
phase-matched feedback condition. (Reprinted from Optical Society of America. Zhou, L., and Poon, A. W., Opt.
Express, 15, 9194–9204, 2007. With permission from The Optical Society of America.)
Figure 4.35b illustrates the microring resonance spectrum, the wavelength-dependent feed-
back phase ∆φ, and the coherent feedback-interfered transmission spectrum. In general, ∆φ
values at resonance wavelengths λm + 1, λm, and λm-1 do not differ by integer number of 2π.
In order to obtain nearly uniform resonance extinction ratios and line shapes, we can
design Lb such that every resonance wavelength sees nearly the same feedback phase and
feedback amplitude. We refer to this as phase-matched feedback condition [130]:
where Lres is the resonance loop length. For devices close to the microring approximation,
we see that Lres≈ 2(La + Lc), where Lc is the coupler length. In the case that the phase-matched
feedback condition is nearly satisfied by properly choosing Lb , ∆φ values at resonance
wavelengths differ by nearly integer number of 2π. Thus, our modeling suggests that the
resonance line shapes are nearly uniform across multiple FSRs, as shown in Figure 4.35c.
However, we note that for devices close to the feedback-loop ring approximation, Lres
exceeds Lb, and thus there is no Lb’s satisfying Equation 4.12.
Figure 4.36a and b shows the optical micrographs of two typical feedback-coupled devices
with different Lb design parameters. In both devices the microring is weakly coupled with
the feedback-waveguide. Device I comprises a circular-shaped microring of radius 25 µm,
an arbitrary chosen Lb = 180 µm, and an integrated p-i-n diode spanning 130 µm. Device II
comprises a circular-shaped microring of radius 25 µm, Lb = 230 µm (with phase-matched
feedback using ν = 1), and an integrated p-i-n diode spanning 180 µm.
Figure 4.36c through e shows the measured and modeled TE-polarized singlemode
transmission spectra of device I under forward biases of Vd = 0 V, 0.9 V, and 1.0 V. The
FSR of 3.8 nm is consistent with the microring round-trip path length, suggesting that
the device is in the microring approximation regime. While the resonance wavelengths
are almost fixed upon the three Vd’s, the resonance line shapes exhibit modulations over
multiple FSRs. The extinction ratio values vary significantly (most pronounced for reso-
nance C) yet with only slight variations in the Q values.
(a) (b)
10 µm 10 µm
(I) (II)
p-i-n
diode
Microring Microring
25 25
µm µm
Output
Input
2 2
1
1 C (f ) Vd = 1.2 V δф = –π, γ = 0.81 ER > 20dB
B γ=1 0
(c) A δф = 0 2
0
Intensity (a.u.)
Intensity (a.u.)
2 0.9 V 1
(g) 1.4 V δф = –1.4π, γ = 0.73
0
1 2 (h) δф = –2π, γ = 0.46
γ = 0.98 2.0 V
(d) δф = –0.13π 1
0
2 1.0 V 0
2 (i) 3.0 V δф = –2.4π, γ = 0.33 0.5 nm
1 1
3.8 nm γ = 0.96
(e) δф = –0.27π
0 0
1548 1554 1560 1566 1560 1565 1570 1575 1580
Wavelength (nm) Wavelength (nm)
Figure 4.36
(a, b) Optical micrographs of two typical devices (I) and (II) with different feedback-coupled waveguide lengths
on a SOI substrate. (c–e) Measured (solid lines) and modeled (dashed lines) TE-polarized transmission spectra of
device (I) with bias voltages Vd of (c) 0 V, (d) 0.9 V, and (e) 1.0 V. (f–i) Measured (solid lines) and modeled (dashed
lines) TE-polarized transmission spectra of device (II) under various bias voltages Vd of (f) 1.2 V, (g) 1.4 V,
(h) 2.0 V, and (i) 3.0 V. ER: extinction ratio. (Reprinted from Optical Society of America. Zhou, L., and Poon, A.
W., Opt. Express, 15, 9194–9204, 2007. With permission from The Optical Society of America.)
The modeled transmission spectra (using Equation 4.10) assume a = 0.96 and b = 0.95.
We choose κ = 0.34 for weak coupling. We determine the wavelength-dependent feedback
phase φo by beam-propagation method (BPM) simulations and thereby numerically incor-
porating the waveguide dispersion. We fit the measured transmission spectra by choosing
δφ = 0 and γ = 1 for Vd = 0 V, and γ remains ~1 for small Vd’s. We assume δφ and γ are inde-
pendent on wavelength within the spectral range of interest.
Whereas, device II displays nearly uniform electrically reconfigurable resonance line
shapes over multiple FSRs as expected from the phase-matched feedback condition, and
a FSR that is within the weakly-coupled microring approximation regime. Figure 4.36f
through i shows the measured and modeled TE-polarized transmission spectra of device
II upon Vd = 1.2 V, 1.4 V, 2.0 V, and 3.0 V. We see uniformly distributed resonance dips with
extinction ratio >20 dB (Vd = 1.2 V), modulated all-pass transmission (Vd = 1.4 V), asymmet-
ric Fano-like line shapes (Vd = 2.0 V), and resonance peaks within a broad resonance dip as
in EIT-like transmission [88,115] (Vd = 3.0 V).
(a) (b)
p-i-n diode
(A)
Pad
Feedback
A waveguide
20 µm
B (c)
Input Output
Waveguide
p-i-n diode
(B)
50 µm Microring
resonator 20 µm
Figure 4.37
(a) Optical micrograph of the fabricated coherent feedback-coupled modulator/switch with dual electrical
inputs (two pairs of metal pads) on a silicon chip. (b, c) Zoom-in view optical micrographs of the two embedded
p-i-n diodes and their metal contacts along the feedback waveguide (node A) and the microring arcs (node B).
(Reprinted from Optical Society of America. Li, C., Zhou, L., and Poon, A. W., Opt. Express 15, 5069–5076, 2007.
With permission from The Optical Society of America.)
Figure 4.37b and c shows the zoom-in view optical micrographs of the diode-integrated
feedback waveguide and microring. One p-i-n diode of ~135 µm in length (node A) is later-
ally integrated across the feedback waveguide, and two other p-i-n diodes of ~59 µm in
length each (node B) with the same cross-section as node A are laterally integrated across
the microring arcs.
–20
1
Voltage (V)
(f )
–30 (a) VA = 0 V, VB = 0 V
Normalized intensity (dB)
0 0 1 0 1 1 1 0 1
–20
–30 (b) –1
VA= 1.2 V
–20
1.0 (g)
–30 (c)
Intensity (a.u.)
Figure 4.38
Measured TE-polarized transmission spectra of the coherent feedback-coupled microring resonator upon vari-
ous DC-bias voltages across nodes A and B. (a) VA = VB = 0 V, (b) VA = 1.2 V, VB = 0 V, (c) VA = 1.2 V, VB = 1.0 V, (d) VA = 0 V,
VB = 1.1 V, and (e) VA = 1.0 V, VB = 1.1 V. (f) 200-Mbps NRZ electrical input data stream with ± 1-V applied across
node A or B. (g) Measured optical waveform upon node B modulation at a probe wavelength in the vicinity of
the resonance upon VA = 1.2 V. (h) Measured optical waveform upon node A modulation at a probe wavelength
in the vicinity of the resonance upon VB = 1.1 V. Reprinted from Optical Society of America. Li, C., Zhou, L., and
Poon, A. W., Opt. Express 15, 5069–5076, 2007. With permission from The Optical Society of America.)
In order to examine the modulator response with a high data rate, we apply a 200-Mbps
nonreturn-to-zero (NRZ) electrical driving signal of bit sequence “01011101” with ± 1 V,
as shown in Figure 4.38f. Figure 4.38g shows the modulated optical waveform at a probe
wavelength of 1550.2 nm with the driving signal applied across node B upon a DC bias
VA = 1.2 V. The modulation depth is consistent with the intensity modulation of ~11 dB upon
DC biasing (see Figure 4.38b and c). Likewise, Figure 4.38h shows the modulated optical
waveform at a probe wavelength of 1549.1 nm with the driving signal applied across node A
upon a DC-bias VB = 1.1 V. The modulation depth is also consistent with the intensity modu-
lation of ~8.3 dB upon DC biasing (see Figure 4.38d and e). We note that the modulator 3-dB
bandwidth is only ~0.9 GHz, measured from modulating either node A or B. We attribute
the bandwidth limitations to both the unoptimized device design (including the use of long
metal pads) and the relatively low drive voltages.
4.5.3.2.2 Demonstration of OR-logic Functionality
Here, we discuss possible logic devices using the same feedback-coupled structure with
dual electrical inputs. Figure 4.39a through d shows the measured TE-polarized transmis-
sion spectra for four electrical signal input states (VA, VB) of (0 V, 0 V), (1.0 V, 0 V), (0 V, 1.0 V),
and (1.0 V, 1.0 V). We choose resonance “*” that displays an extinction ratio of ~12 dB when
the device is unbiased. The resonance is partially suppressed (and blue-shifted) when only
node A is switched on, and totally suppressed when only node B is switched on. When
both nodes A and B are switched on, the resonance reappears with a spectral blueshift of
~0.25 nm. Thus, at a probe wavelength in the vicinity of resonance “*”, the optical trans-
mission varies between a high-level (Figure 4.39b through d) and a low-level (Figure 4.39a)
upon different electrical signal input states.
Hence, in the case that two electrical data are simultaneously applied across nodes A
and B, the electrical states can effectively impose logic functionality onto the optical
transmission near a microring resonance. Figure 4.39e through g presents our measured
electro-optic logic switching using the cross-coupled microring. We applied two different
200-Mbps NRZ electrical data with ± 1 V across nodes A and B. The measured optical
–30 (a) 1
(e)
–35 12dB
VA = 0 V, VB = 0 V 0 1100
Voltage (V)
–40 *
Normalized intensity (dB)
–30 (b) –1
–35 1
(f )
–40 VA = 1.0 V, VB = 0 V
0 1 001
–30 (c)
–35 –1
Opt. Int. (a.u.)
Figure 4.39
Measured TE-polarized transmission spectra of the coherent feedback-coupled microring resonator-based
device with dual electrical inputs for four electrical signal input states (VA, VB) of (a) (0 V, 0 V), (b) (1.0 V, 0 V),
(c) (0 V, 1.0 V), and (d) (1.0 V, 1.0 V). λprobe: probe wavelength. (e, f) 200-Mbps NRZ electrical input data applied
across nodes A and B. (g) Measured optical waveform at a probe wavelength near resonance “*”. (Reprinted
from Optical Society of America. Li, C., Zhou, L., and Poon, A. W., Opt. Express 15, 5069–5076, 2007. With permis-
sion from The Optical Society of America.)
Hence, along with the carrier dispersion based resonance wavelength tuning techniques
reviewed here, we see the potential of realizing nonblocking N × N microring resonator-
coupled crossing-based switch nodes, for scalable photonic interconnect networks-on-chip
in multi-core computing systems. It is conceivable that such photonic networks-on-chip
technology offers an optical solution for broadband high-throughput short-distance data
communications with low latency and low-power consumption. Nonetheless, many criti-
cal issues are still debatable such as whether to put light sources on- or off-chip, and how
logic devices can be embedded for intelligent microresonator switching, while keeping
low-power consumption for the entire microsystem. Besides, in order to overcome the
inherent limitations of photonics in networks-on-chip applications, namely buffering and
signal processing, we believe the integration of microelectronics and photonics is inevi-
table and will bring up many new major design/fabrication challenges and re-thinking
[40] for computer architectures, optoelectronics integrated systems, and silicon photonics.
Acknowledgments
Andrew W. Poon would like to dedicate this Chapter to his thesis advisor, Professor Richard
K. Chang of Yale University, for his inspiration and foresight on optical microresonators.
Many of the concepts reviewed here including Fano resonances, square-shaped microre-
sonators, and spiral-shaped microresonators are based on early work done and ideas dis-
cussed in Chang’s lab while Andrew W. Poon was still a graduate student. Andrew W.
Poon would also like to thank all his former and present students at the Photonic Device
Laboratory. Their collective effort over the past 6 years has certainly put PDL at HKUST
on the world map of silicon photonics and optical microresonators research. In particular,
Andrew W. Poon would also like to acknowledge former students N. Ma, C. Y. Fong, H.
T. Lee, K. K. Tsia, and S. Zheng for their early contributions. The research on microreso-
nators and silicon photonics have been substantially supported by the Research Grants
Council and the University Grants Committee of The Hong Kong Special Administrative
Region, China, under Project HIA01/02.EG05, Project 6166/02E, Project 6254/04E, Project
618505, and Project 618506. Andrew W. Poon would also like to acknowledge support by
the Institute of Integrated Micro Systems of HKUST, under Project I2MS01/02.EG07. X. Luo
would like to acknowledge support from NANO Fellowship of HKUST.
References
1. R. K. Chang and A. J. Campillo, Eds., Optical Processes in Microcavities. Singapore: World
Scientific, 1996.
2. K. J. Vahala, Ed. Optical Microcavities. Singapore: World Scientific, 2004.
3. R. K. Chang and Y. Pan. Linear and non-linear spectroscopy of microparticles: basic principles,
new techniques and promising applications. Faraday Discussions 137, 9–36, 2008.
4. A. B. Matsko and V. S. Ilchenko. Optical resonators with whispering-gallery modes—part I:
basics. IEEE J. Select. Topics Quantum Electron. 12, 3–14, 2006.
5. V. S. Ilchenko and A. B. Matsko. Optical resonators with whispering-gallery modes—part II:
applications. IEEE J. Select. Topics Quantum Electron. 12, 15–32, 2006.
30. F. Xia, L. Sekaric, and Y. Vlasov. Ultracompact optical buffers on a silicon chip. Nature Photonics 1,
65–71, 2007.
31. D. Van Thourhout, P. Dumon, W. Bogaerts, G. Roelken, D. Taillaert, G. Priem, and R. Baets.
Recent progress in SOI nanophotonic waveguides. In Proceedings of the 31st European Conference
on Optical Communication (ECOC ’05), Glasgow, Scotland, September 25–29, 2005, 241–244.
32. See for example Proceedings of the 1st–4th IEEE International Conferences on Group IV Photonics,
2004–2007.
33. M. Lipson. Guiding, modulating, and emitting light on silicon - challenges and opportunities.
J. Lightwave Technol. 23, 4222–4238, 2005.
34. B. Jalali and S. Fathpour. Silicon photonics. J. Lightwave Technol. 24, 4600–4765, 2006.
35. C. Gunn, CMOS photonics technology platform. Proceedings of SPIE Photon. West, San Jose, CA,
January 25, 2006, Invited Paper 6125-01.
36. R. Soref. The past, present, and future of silicon photonics. IEEE J. Select. Topics Quantum
Electron. 12, 1678–1687, 2006.
37. G.T. Reed and A.P. Knights. Silicon Photonics: An Introduction. John Wiley, Chichester, UK, 2004.
38. L. Pavesi and D. J. Lockwood, Eds. Silicon Photonics. New York: Springer, 2004.
39. B. A. Small, B. G. Lee, K. Bergman, Q. Xu, and M. Lipson. Multiple-wavelength integrated pho-
tonic networks based on microring resonator devices. J. Opt. Netw. 6, 112–120, 2007.
40. A. Shacham, K. Bergman, and L. P. Carloni. On the design of a photonic network-on-chip.
Proceedings of the First International Symposium on Networks-on-Chip, Nocs’07, 53–64, 2007.
41. A. Liu, R. Jones, L. Liao, D. Samara-Rubio, D. Rubin, O. Cohen, R. Nicolaescu, and M. Paniccia.
A high-speed silicon optical modulator based on a metal-oxide-semiconductor capacitor. Nature
427, 615–618, 2004.
42. H. Rong, A. Liu, R. Jones, O. Cohen, D. Hak, R. Nicolaescu, A. Fang, and M. Paniccia. An all-
silicon Raman laser. Nature 433, 292–294, 2005.
43. J. Niehusmann, A. Vörckel, P. H. Bolivar, T. Wahlbrink, W. Henschel, and H. Kurz. Ultrahigh-
quality-factor silicon-on-insulator microring resonator. Opt. Lett. 29, 2861–2863, 2004.
44. I. Kiyat, A. Aydinli, and N. Dagli. High-Q silicon-on-insulator optical rib waveguide racetrack
resonators. Opt. Express 13, 1900–1905, 2005.
45. M. A. Popović, T. Barwicz, M. R. Watts, P. T. Rakich, L. Socci, E. P. Ippen, F. X. Kärtner, and H. I.
Smith. Multistage high-order microring-resonator add-drop filters. Opt. Lett. 31, 2571–2573, 2006.
46. F. Xia, L. Sekaric, M. O’Boyle, and Y. Vlasov. Coupled resonator optical waveguides based on
silicon-on-insulator photonic wires. Appl. Phy. Lett. 89, 041122, 2006.
47. P. Koonath, T. Indukuri, and B. Jalali. Monolithic 3-D silicon photonics. J. Lightwave Technol. 24,
1796–1804, 2006.
48. M. Soltani, S. Yenagnarayanan, and A. Adibi. Ultra-high Q planar silicon microdisk resonators
for chip-scale silicon photonics. Opt. Express 15, 4694–4704, 2007.
49. M. Borselli, T. J. Johnson, and O. Painter. Accurate measurement of scattering and absorption
loss in microphotonic devices. Opt. Lett. 32, 2954–2956, 2007.
50. S. Blair and K. Zheng. Intensity-tunable group delay using stimulated Raman scattering in
silicon slow-light waveguides. Opt. Express 14, 1064–1069, 2006.
51. M. S. Nawrocka, T. Liu, X. Wang, and R. R. Panepucci. Tunable silicon microring resonator with
wide free spectral range. Appl. Phys. Lett. 89, 071110, 2006.
52. I. Kiyat, A. Aydinli, and N. Dagli. Low-power thermooptical tuning of SOI resonator switch.
IEEE Photon. Technol. Lett. 18, 364–366, 2006.
53. G. N. Nielson, D. Seneviratne, F. Lopez-Royo, P. T. Rakich, Y. Avrahami, M. R. Watts, H. A.
Haus, H. L. Tuller, and G. Barbastathis. Integrated wavelength-selective optical MEMS switch-
ing using ring resonator filters. IEEE Photon. Technol. Lett. 17, 1190–1192, 2005.
54. M. M. Lee and M. C. Wu. Variable bandwidth of dynamic add-drop filters based on coupling-
controlled microdisk resonators. Opt. Lett. 31, 2444–2446, 2006.
55. A. Huang, C. Gunn, G. Li, Y. Liang, S. Mirsaidi, A. Narasimha, and T. Pinguet. A 10Gb/s
photonic modulator and WDM MUX/DEMUX integrated with electronics in 0.13 μm SOI
CMOS. 2006 IEEE International Solid-State Circuits Conference, San Francisco, CA, 2006.
56. Q. Xu, S. Manipatruni, B. Schmidt, J. Shakya, and M. Lipson. 12.5 Gbit/s carrier-injection-based
microring silicon modulators. Opt. Express 15, 430–436, 2007.
57. S. F. Preble, Q. Xu, B. S. Schmidt, and M. Lipson. Ultrafast all-optical modulation on a silicon
chip. Opt. Lett. 30, 2891–2893, 2005.
58. V. R. Almeida, C. A. Barrios, R. R. Panepucci, and M. Lipson. All-optical control of light on a
silicon chip. Nature 431, 1081–1084, 2004.
59. H. Rong, S. Xu, Y. Kuo, V. Sih, O. Cohen, O. Raday, and M. Paniccia. Low-threshold continuous-
wave Raman silicon laser. Nature Photonics 1, 232–237, 2007.
60. A. W. Fang, R. Jones, H. Park, O. Cohen, O. Raday, M. J. Paniccia, and J. E. Bowers. Integrated
AlGaInAs-silicon evanescent race track laser and photodetector. Opt. Express 15, 2315–2322,
2007.
61. J. Van Campenhout, P. Rojo-Romeo, P. Regreny, C. Seassal, D. Van Thourhout, S. Verstuyft,
L. Di Cioccio, J.-M. Fedeli, C. Lagahe, and R. Baets. Electrically pumped InP-based microdisk
lasers integrated with a nanophotonic silicon-on-insulator waveguide circuit. Opt. Express 15,
6744–6749, 2007.
62. Q. Xu, V. R. Almeida, and M. Lipson. Micrometer-scale all-optical wavelength converter on
silicon. Opt. Lett. 30, 2733–2735, 2005.
63. S. F. Preble, Q. Xu, and M. Lipson. Changing the colour of light in a silicon resonator. Nature
Photonics 1, 293–296, 2007.
64. A. Vörckel, M. Mönster, W. Henschel, P. H. Bolivar, and H. Kurz. Asymmetrically coupled
silicon-on-insulator microring resonators for compact add–drop multiplexers. IEEE Photon.
Technol. Lett. 15, 921–923, 2003.
65. A. Melloni, F. Morichetti, and M. Martinelli. Polarization conversion in ring resonator phase
shifters. Opt. Lett. 29, 2785–2787, 2004.
66. C. Manolatou, M. J. Khan, S. Fan, P. R. Villeneuve, H. A. Haus, and J. D. Joannopoulos. Coupling
of modes analysis of resonant channel add–drop filters. IEEE J. Quantum Electron. 35, 1322–1331,
1999.
67. A. W. Poon, F. Courvoisier, and R. K. Chang. Multimode resonances in square-shaped optical
microcavities. Opt. Lett. 26, 632–634, 2001.
68. Y. Pan and R. K. Chang. Highly efficient prism coupling to whispering gallery modes of a
square µ cavity. Appl. Phys. Lett. 82, 487–489, 2003.
69. C. Y. Fong and A. W. Poon. Mode field patterns and preferential mode coupling in planar
waveguide-coupled square microcavities. Opt. Express 11, 2897–2904, 2003.
70. C. Y. Fong and A. W. Poon. Planar corner-cut square microcavities: ray optics and FDTD analy-
sis. Opt. Express 12, 4864–4874, 2004.
71. N. Ma, C. Li, and A. W. Poon. Laterally coupled hexagonal micropillar resonator add-drop
filters in silicon nitride. IEEE Photon. Technol. Lett. 16, 2487–2489, 2004.
72. N. Ma, F. K. Tung, S. Lui, and A. W. Poon. Hexagonal micro-pillar cavities: multimode reso-
nances and open-loop resonance linewidth broadening. Proceedings of SPIE 4986, 153–160,
2003.
73. S. Zheng, N. Ma, and A. W. Poon. Experimental demonstration of waveguide-coupled hex-
agonal micropillar resonators with round-corners in silicon nitride. Proceedings of Conference on
Lasers Electro-Optics. IEEE and Optical Society of America, 443–445, 2005.
74. C. Li, N. Ma, and A. W. Poon. Waveguide-coupled octagonal microdisk channel add-drop
filters. Opt. Lett. 29, 471–473, 2004.
75. C. Li and A. W. Poon. Experimental demonstration of waveguide-coupled round-cornered
octagonal microresonators in silicon nitride. Opt. Lett. 30, 546–548, 2005.
76. C. Li, L. Zhou, S. Zheng, and A. W. Poon, Silicon polygonal microdisk resonators. IEEE J. Select.
Topics Quantum Electron. 12, 1438–1449, 2006.
77. C. Li, L. Zhou, and A. W. Poon. Waveguide-coupled square micropillar resonator-based devices:
channel filters and electro-optic switches with embedded p-i-n diodes. IEEE Proceedings of the
8th International Conference on Transparent Optical Networks, 96–99, 2006.
78. W. H. Guo, Y. Z. Huang, Q. Y. Lu, and L. J. Yu. Whispering-gallery-like modes in square resona-
tors. IEEE, J. Quantum Electron. 39, 1106–1110, 2003.
79. W. H. Guo, Y. Z. Huang, Q. Y. Lu, and L. J. Yu. Modes in square resonators. IEEE, J. Quantum
Electron. 39, 1563–1566, 2003.
80. Q. Chen, Y. D. Yang, and Y. Z. Huang. Finite-difference time-domain analysis of deformed
square cavity filters with a traveling-wave-like filtering response by mode coupling. Opt. Lett.
32, 967–969, 2007.
81. U. Vietze, O. Krauß, and F. Laeri, G. Ihlein and F. Schüth, B. Limburg and M. Abraham. Zeolite-
dye microlasers. Phys. Rev. Lett. 81, 4628–4631, 1998.
82. H. J. Moon, K. An, and J. H. Lee. Single spatial mode selection in a layered square microcavity
laser. Appl. Phys. Lett. 82, 2963–2965, 2003.
83. Q. Chen, Y. H. Hu, Y. Z. Huang, Y. Du, and Z. C. Fan. Equilateral-triangle-resonator injection
lasers with directional emission. IEEE J. Quantum Electron. 43, 440–444, 2007.
84. A. W. Poon. Optical resonances of two-dimensional microcavities with circular and non-circular
shapes. PhD thesis, Yale University, 2001.
85. FullWAVE, Rsoft Inc. Research Software, http://www.rsoftinc.com.
86. M. Lohmeyer. Mode expansion modeling of rectangular integrated optical microresonators.
Opt. Quantum Electron. 34, 541–557, 2002.
87. M. Hammer. Resonant coupling of dielectric optical waveguides via rectangular microcavities:
the coupled guided mode perspective. Opt. Commun. 214, 155–170, 2002.
88. Q. Xu, S. Sandhu, M. L. Povinelli, J. Shakya, S. Fan, and M. Lipson. Experimental realization of
an on-chip all-optical analogue to electromagnetically induced transparency. Phy. Rev. Lett. 96,
123901, 2006.
89. J. U. Nöckel and A. D. Stone. Ray and wave chaos in asymmetric resonant optical cavities.
Nature 385, 45–47, 1997.
90. M. Kneissl, M. Teepe, N. Miyashita, N. M. Johnson, G. D. Chern, and R. K. Chang. Current-
injection spiral-shaped microcavity disk laser diodes with unidirectional emission. Appl. Phys.
Lett. 84, 2485–2487, 2004.
91. T. Ben-Messaoud and J. Zyss. Unidirectional laser emission from polymer–based spiral micro-
disks. Appl. Phys. Lett. 86, 241110, 2005.
92. A. Fujii, T. Nishimura, Y. Yoshida, K. Yoshino, and M. Ozaki. Unidirectional laser emission from
spiral microcavity utilizing conducting polymer. Jap. J. Appl. Phys. 44, L1091–L1093, 2005.
93. A. Fujii, T. Takashima, N. Tsujimoto, T. Nakao, Y. Yoshida, and M. Ozaki. Fabrication
and unidirectional laser emission properties of asymmetric microdisks based on poly
(p-phenylenevinylene) derivative. Jap. J. Appl. Phys. 45, L833–L836, 2006.
94. R. Audet, M. A. Belkin, J. A. Fan, B. G. Lee, K. Lin, F. Capasso, E. E. Narimanov, D. Bour,
S. Corzine, J. Zhu, and G. Höflerf. Single-mode laser action in quantum cascade lasers with
spiral-shaped chaotic resonators. Appl. Phys. Lett. 91, 131106, 2007.
95. T. Y. Kwon, S. Y. Lee, M. S. Kurdoglyan, S. Rim, C. M. Kim, and Y. J. Park. Lasing modes in a
spiral shaped dielectric microcavity. Opt. Lett. 31, 1250–1252, 2006.
96. S. Y. Lee, S. Rim, J. W. Ryu, T. Y. Kwon, M. Choi, and C. M. Kim. Quasiscarred resonances in a
spiral shaped microcavity. Phys. Rev. Lett. 93, 164102, 2004.
97. R. K. Chang, G. E. Fernandes, and M. Kneissl. The quest for uni-directionality with WGMs in
μ-Lasers: coupled oscillators and amplifiers. Proceedings of the 8th International Conference on
Transparent Optical Networks, 1, 47–51, 2006.
98. G. D. Chern, G. E. Fernandes, R. K. Chang, Q. Song, L. Xu, M. Kneissl, and N. M. Johnson.
High-Q-preserving coupling between a spiral and a semicircle µ-cavity. Opt. Lett. 32, 1093–
1095, 2007.
99. J. Y. Lee, X. Luo, and A. W. Poon. Reciprocal transmissions and asymmetric modal distributions
in waveguide-coupled spiral-shaped microdisk resonators. Opt. Express 15, 14650–14666, 2007.
100. M. Born and E. Wolf. Principles of Optics, 7th Ed. Cambridge: Cambridge University Press, 1999,
724–726.
101. L. Liao, D. Samara-Rubio, M. Morse, A. Liu, D. Hodge, D. Rubin, U. Keil, and T. Franck. High
speed silicon Mach-Zehnder modulator. Opt. Express 13, 3129–3135, 2005.
102. F. Y. Gardes, G. T. Reed, and N. G. Emerson. A sub-micron depletion-type photonic modulator
in silicon on insulator. Opt. Express 22, 8845–8854, 2005.
103. A. Liu, L. Liao, D. Rubin, H. Nguyen, B. Ciftcioglu, Y. Chetrit, N. Izhaky, and M. Paniccia. High-
speed optical modulation based on carrier depletion in a silicon waveguide. Opt. Express 15,
660–668, 2007.
104. D. Marris-Morini, L. Vivien, S. Maine, E. Cassan, S. Laval, P. Lyan, and J-M. Fédéli. Low loss
optical modulator in a silicon waveguide based on a carrier depletion horizontal structure.
Proceedings of the IEEE 4th International Conference on Group IV Photonics, 195–197, 2007.
105. L. Liao, A. Liu, D. Rubin, J. Basak, Y. Chetrit, H. Nguyen, R. Cohen, N. Izhaky, and M. Paniccia.
40 Gbit/s silicon optical modulator for high-speed applications. Electron. Lett. 43, 20072253,
2007.
106. R. A. Soref and B. R. Bennett. Electrooptical effects in silicon. IEEE J. Quantum Electron. QE-23,
123–129, 1987.
107. S. Manipatruni, Q. Xu, and M. Lipson. PINIP based high-speed high-extinction ratio micron-
size silicon electro-optic modulator. Opt. Express 15, 13035–13042, 2007.
108. A. Yariv. Universal relations for coupling of optical power between microresonators and dielec-
tric waveguides. Electron. Lett. 36, 321–322, 2000.
109. L. Zhou and A. W. Poon. Silicon-on-insulator tunable waveguide-coupled microdisk resonators
with selectively integrated p-i-n diodes. Proceedings of Conference on Lasers Electro-Optics. IEEE
and Optical Society of America, 116–118, 2005.
110. OmniSim, Photon Design Ltd., http://www.photond.com.
111. J. E. Heebner and R. W. Boyd. “Slow” and “fast” light in resonator-coupled waveguides. J. Mod.
Opt. 49, 2629–2636, 2002.
112. J. K. S. Poon, L. Zhu, G. A. DeRose, and A. Yariv. Polymer microring coupled-resonator optical
waveguides. J. Lightwave Technol. 24, 1843–1849, 2006.
113. B. E. Little, S. T. Chu, J. V. Hryniewicz, and P. P. Absil. Filter synthesis for periodically coupled
microring resonators. Opt. Lett. 25, 344–346, 2000.
114. U. Fano. Effect of configuration interaction on intensities and phase shifts. Phys. Rev. 124, 1866–
1878, 1961.
115. D. D. Smith, H. Chang, K. A. Fuller, A. T. Rosenberger, and R. W. Boyd. Coupled-resonator-
induced transparency. Phys. Rev. A 69, 063804, 2004.
116. L. Maleki, A. B. Matsko, A. A. Savchenkov, and V. S. Ilchenko. Tunable delay line with interact-
ing whispering-gallery-mode resonators. Opt. Lett. 29, 626–628, 2004.
117. A. B. Matsko, A. A. Savchenkov, D. Strekalov, V. S. Ilchenko, and L. Maleki. Interference effects
in lossy resonator chains. J. Mod. Opt. 51, 2515–2522, 2004.
118. S. Fan. Sharp asymmetric line shapes in side-coupled waveguide-cavity systems. Appl. Phys.
Lett. 80, 908–1000, 2002.
119. A. Chiba, H. Fujiwara, J. Hotta, S. Takeuchi, and K. Sasaki. Fano resonance in a multimode
tapered fiber coupled with a microspherical cavity. Appl. Phy. Lett. 86, 261106, 2005.
120. H. T. Lee and A. W. Poon. Fano resonances in prism-coupled square micropillars. Opt. Lett. 29.
5–7, 2004.
121. H. T. Lee, L. Zhou and A. W. Poon. Fano resonances in prism-coupled multimode square micro-
pillar resonators. Opt. Lett. 30. 1527–1529, 2005.
122. Y. Lu, J. Yao, X. Li, and P. Wang. Tunable asymmetrical Fano resonance and bistability in a
microcavity-resonator-coupled Mach-Zehnder interferometer. Opt. Lett. 30, 3069–3071, 2005.
123. N. K. Hon and A. W. Poon. Silica polygonal micropillar resonators: Fano lineshapes tuning by
using a Mach-Zehnder interferometer. Proceedings of SPIE 6101, 176–182, 2006.
124. L. Zhou and A. W. Poon. Fano resonance-based electrically reconfigurable add-drop filters in
silicon microring resonator-coupled Mach-Zehnder interferometers. Opt. Lett. 32, 781–783, 2007.
125. G. Lenz and C. K. Madsen. General optical all-pass filter structures for dispersion control in
WDM systems. J. Lightwave Technol. 17, 1248–1254, 1999.
126. W. Green, R. Lee, G. DeRose, A. Scherer, and A. Yariv. Hybrid InGaAsP-InP Mach-Zehnder
racetrack resonator for thermooptic switching and coupling control. Opt. Express 13, 1651–1659,
2005.
127. G. T. Paloczi, Y. Huang, A. Yariv, and S. Mookherjea. Polymeric Mach-Zehnder interferometer
using serially coupled microring resonators. Opt. Express 11, 2666–2671, 2003.
128. S. Mookherjea, Mode cycling in microring optical resonators. Opt. Lett. 30, 2751–2753, 2005.
129. M. R. Watts, T. Barwicz, M. Popovic, P. T. Rakich, L. Socci, E. P. Ippen, H. I. Smith, and
F. Kaertner. Microring-resonator filter with doubled free-spectral-range by two-point coupling.
Proceedings of Conference on Lasers and Electro-Optics 1, 273–275, 2005.
130. L. Zhou and A. W. Poon, Electrically reconfigurable silicon microring resonator-based filter
with waveguide-coupled feedback. Opt. Express 15, 9194–9204, 2007.
131. C. Li, L. Zhou, and A. W. Poon. Silicon microring carrier-injection-based modulators/switches
with tunable extinction ratios and OR-logic switching by using waveguide cross-coupling. Opt.
Express 15, 5069–5076, 2007.
132. S. Zheng, H. Chen, and A. W. Poon. Microring-resonator cross-connect filters in silicon nitride:
rib waveguide dimensions dependence. IEEE J. Select. Topics Quantum Electron. 12, 1380–1387,
2006.
133. H. Chen and A. W. Poon. Low-loss multimode-interference-based crossings for silicon wire
waveguides. IEEE Photon. Technol. Lett. 18, 2260–2262, 2006.
134. F. Xu and A. W. Poon. Microring resonator-coupled multimode-interference-based crossings
in 2x2 cross-grid array filters. Proceedings of the IEEE 4th International Conference on Group IV
Photonics, 16–18, 2007.
William H. Steier
University of Southern California
Hidehisa Tazawa
University of Southern California
Contents
5.1 Introduction......................................................................................................................... 266
5.2 Ring Resonator Basics and EO Modulators and Switches............................................ 267
5.2.1 Ring Resonator Basics............................................................................................ 267
5.2.2 Electro-Optic Ring Modulators............................................................................ 270
5.2.3 Bandwidth of Ring Resonant Modulators.......................................................... 271
5.2.3.1 Lumped Circuit Electrode...................................................................... 272
5.2.3.2 Traveling Wave Electrode....................................................................... 272
5.2.4 Electro-Optic Polymer Traveling-Wave Ring Modulator.................................. 277
5.2.4.1 Fabrication................................................................................................. 277
5.2.4.2 Optical and Electro-Optical Properties................................................ 277
5.2.4.3 Traveling Wave Electrode Properties.................................................... 278
5.2.4.4 Modulation at the First FSR Spacing of 28 GHz.................................. 280
5.2.4.5 Modulation at Multiples of the FSR....................................................... 283
5.3 Optical Signal Processing Using Ring Resonator.......................................................... 286
5.3.1 Theory...................................................................................................................... 286
5.3.1.1 Fundamentals of OSP.............................................................................. 286
5.3.1.2 Representations of OSP........................................................................... 288
5.3.1.3 Operations of OSP.................................................................................... 289
5.3.2 Structure................................................................................................................... 290
5.3.3 Analysis.................................................................................................................... 291
5.3.3.1 Configurable Couplers............................................................................ 292
5.3.3.2 Racetrack................................................................................................... 292
5.3.3.3 MZ Section................................................................................................ 293
265
© 2009 by Taylor & Francis Group, LLC
5.1 Introduction
Optical waveguide ring resonators, shown in Figure 5.1, are the simplest form of an inte-
grated optical resonant structure. They were first proposed by Marcatili1 and in the past
several years photonic ring resonators have been investigated and demonstrated in several
material systems including SiO2,2 Si,3,4 III-V semiconductors,5 and polymers.6 They have
been demonstrated as passive wavelength filters,7,8 slow wave devices,9 and complex multi-
pole filters.2 If the rings are made of an electro-optic material the resonant wavelength
can be voltage or current tuned and the devices become optical switches or modulators.
Tuning based on charge injection, depletion layer modulation, and the Pockel’s effect10
is in the literature. The modulation bandwidth or the switching speed is limited by the
bandwidth of the resonator or the inherent speed limitation of the electro-optic effect. The
highest speed devices are based on the Pockel’s effect.
In first section of this chapter we review the basics of micro-resonators, Pockel’s
effect modulators, and the bandwidth limitations. In the second section, our work on
a1 a2
κ, τ
b1 b2
Figure 5.1
The generic geometry of ring resonator (one bus waveguide).
operating polymer ring resonant modulators at modulation frequencies over 100 GHz will
be reviewed. This very high speed operation is based on the modulation frequency match-
ing the free spectral range (FSR) of the resonator. The modulation frequency matches the
FSR but the band width is still limited by the Q of the resonator. This high frequency
band-limited modulation has been demonstrated in LiNbO3 disc resonators and in poly-
mer ring waveguide resonators.
In the third section, the great potential of using tunable ring resonators for very high
speed optical signal processing will be reviewed. One of the most promising technologies
for the general optical signal processing is the optical delay line circuits.11–17 The ring reso-
nator is a crucial building block for the optical delay line circuits enabling the feedback
control of the optical signal. Together with other optical building blocks such as optical
coupler, we have proposed this general optical structure as optical signal processor (OSP).18
In addition to generating arbitrary and complex functionality by employing optical delay
line structures, use of the electro-optic polymer OSP permits operation at high speed.
a2 τ ik a1 2 2
b = ik τ b , τ + k = 1 (5.1)
2 1
where b, r, and L are propagation constant, intensity attenuation coefficient, and perimeter
of a ring, respectively. Thus a and q represent the loss factor and phase shift after one
round trip. From Equations 5.1 and 5.2, one obtains
τ - a e - iθ
a2 = a1 . (5.3)
1 - τa e - iθ
The same result can be obtained by the summation of multiple round trips
τ - a e - iθ
= a1
1 - τa e - iθ
This expression is useful in analyzing the properties of EO modulation later. The trans-
mission of a ring resonator is
T (θ) =
a2
2
= 1-
(1 - a 2 )(1 - τ2 ) .
(5.5)
a1 ( 1 - aτ )2 + 4aτ sin 2 (θ/2 )
Optical resonances occur at the points q = 2mp, where m is some integer. When the inter-
nal losses in ring, a, are equal to the coupling losses, t, the transmitted power vanishes.
This condition is called critical coupling. For a < t, the resonator is said to be undercou-
pling and for a > t, the resonator is said to be overcoupled.
The critically coupled ring resonator can be used as a notch filter. In case of low loss (a ≈ 1)
and over-coupling (a > t), the relatively flat intensity response and the rapid phase change
at resonances can be used as a phase filter or all-pass filter. Figure 5.2 shows the optical
1
0.85 (over)
0.95 (under)
Intensity
0.5
0
2mπ 2(m+1)π
Phase θ
3π
0.85 (over)
2π
Phase shift
0.95 (under)
−π
2mπ 2(m+1)π
Phase θ
Figure 5.2
The intensity transfer function and phase shift of a ring resonator (α = 0.9, τ = 0.95 and 0.85).
intensity transfer function T(θ) and the phase shift of slightly over and under coupled ring
resonators.
The full phase width at half maximum (FWHM), Dq, and the finesse of a transfer func-
tion are obtained from Equation 5.5
1- aτ 2(1 - aτ)
∆θ = 4 sin -1 ≈
2 + 2a 2 τ 2 aτ
(5.6)
2 π π aτ
F= ≈ .
∆θ 1 - aτ
The separation of each resonance peaks is the free spectral range, FSR, which is given
by
λ2
FSR λ = (5.7)
ng L
where ng is the waveguide effective index of refraction. Using the relation, F = FSR/Df, the
full frequency and wavelength width at half maximum (FWHM) are
c λ2
∆f = , ∆λ = . (5.8)
Fng L Fng L
f λ Fng L
Q= = = . (5.9)
∆f ∆λ λ
To derive the expression of intrinsic or unloaded QU, setting τ = 1 and α ≈ 1,
The final portion of Equation 5.10 is useful when r is given in dB/cm and l in μm. Thus,
QU is dominated only by the attenuation coefficient, which will be a combination of mate-
rial absorption, bending loss, and scattering loss. The intrinsic finesse is also obtained as
2π
FU ≈ . (5.11)
ρL
1
T = 1- 2 (5.12)
F
1 + π θ2o
πn3g rΓV0
θ = θ0 + ∆βL, ∆β = , (5.13)
λg
where q0 is the bias phase, L is the perimeter of the ring, ng is the effective index of the ring
waveguide, l is the free space optical wavelength, r is the EO coefficient, g is the electrode
gap, and G is the electrical-optical overlap integral. If the wavelength of the incident light is
set to the high slope point on the transmission, the output intensity will be strongly modu-
lated with a small modulating voltage. This situation is illustrated in Figure 5.3. The high
modulation depth results from the sharp slope of the transmission spectrum and higher F
provides higher sensitivity.
From Equation 5.12, the slope of the transmission, dT/dθ0, is maximum at
π
θ0 = (5.14)
3F
At this point T = 0.25 and
9F 9 F πn03rΓL
∆T = ∆θ0 = ∆V (5.15)
8 3π 8 3 π λg
Equation 5.15 gives the change in transmission for a change in voltage when the trans-
mission is optically biased to T = 0.25, the maximum slope point.
By comparing |∆T/∆V|max of a ring modulator with that of a Mach-Zehnder (MZ) modula-
tor, an equivalent Vπeq, can be defined as
Intensity
∆T
Voltage
∆V
Figure 5.3
Conceptual drawing of the modulation by a ring modulator.
-1 -1
π ∆T = π ∆T ∆θ
Vπeq = (5.16)
2 ∆V max 2 ∆θ ∆V max
( ∆∆Vθ ) ,
-1
VπMZ = π (5.17)
VπMZ 9F
eq = 2 × ∆T∆θ = (5.18)
Vπ 4 3π
if the device parameters are same. For example, for a finesse of 10 and |dT/dq|max = 2.07 at
q0 = 0.058p. An MZ modulator must have an electrode length of 4.14 × L to obtain the same
modulation sensitivity as the ring modulator with a finesse of 10. The resonator with F = 10
increased the EO interaction length by a factor of 4.14.
∞
Eout (t) = τ - k 2 ∑ τ n-1a n exp [ - i(nθ + δ n sin(ω mt - nφ))] Ein (t) (5.19)
n= 1
where f = wm/FSR. It is assumed that the group index and effective index of a waveguide
are the same. The dispersions of propagation constants, coupling coefficients, and loss are
neglected.
Equation 5.20 takes into account the transit-time. Now the optical power of the transmit-
ted signal can be given by substituting Equation 5.20 into Equation 5.19 and calculating the
first component of the Fourier expansion of the output intensity
T
2
Iω m =
T ∫E
0
out (t) 2e iω mt dt (5.21)
where T = 2p/wm. The electrical signal power in a photo detector is proportional to the
square of the optical signal power and this must be taken into account in calculating the
bandwidth of the electrical signal power. Figure 5.4 shows the small signal frequency
responses in dBe of the modulated output from lumped ring modulators, optically biased
to the maximum slope with a finesse of 10. The response is normalized by low modula-
tion frequency response. The modulating frequency is normalized by the FSR. The modu-
lation is clearly limited by optical resonator bandwidth. The 3 dBe bandwidth BWopt is
0.0845 × FSR. Notice that this bandwidth is somewhat larger than the half of the resonator
line-width (0.05 × FSR for F = 10).
V ( z, t ) = V0 sin ω m t - ( nm
c
z ) (5.22)
–10
–15
–20
0 0.05 0.1 0.15 0.2 0.25
Modulation frequency/FSR
Figure 5.4
The small signal frequency responses (in dBe) of the modulated output from a ring modulator with a lumped
electrode, optically biased to the maximum slope with a finesse of 10. The responses are normalized by low
modulation frequency response.
Optical
input
z
Optical
input
0 MMW
Microwave in
input
Figure 5.5
Traveling wave ring modulator.
where nm is the microwave effective index. The voltage seen at position z along the ring
resonator waveguide by photons that enter the z = 0 at t = t is given by
V ( z, t ) = V0 sin ω m t - ( ∆n
c
z ) (5.23)
where Dn = nm − ng and ng is the optical index. The modulation index dn is given by
( )
n- 1 L
= ∆βL
sin ( ψ / 2 ) sin ( nφ/2 )
( ψ / 2 ) sin (φ/2 )
sin ω mt - -
2 (
ψ n+1
2
φ ) (5.24)
25
20 F=30
F=10
10
–5
0 0.5 1 1.5 2 2.5 3
Modulation frequency (FSR)
Figure 5.6
Modulation frequency responses of velocity matched traveling wave ring modulators with the finesse of 10 and
30. The responses are normalized by the response by an MZ modulator with same electrode length.
where tr = n0L/c is the optical round trip time and y = wmDnL/c is the velocity matching fac-
tor. The optical power of the transmitted signal can be obtained by substituting Equation
5.24 into Equation 5.19, and then using Equation 5.21.
The small signal frequency responses in dBe of the modulated output from velocity
matched (Dn = 0) ring modulators, optically biased to the maximum slope, with a finesse
of 10 and 30 are shown in Figure 5.6. The responses are normalized by the small sig-
nal response of a MZ modulator biased at quadrature with same electrode length. The
modulating frequency is normalized by the FSR. The velocity matched traveling wave
ring modulators provide high modulation efficiency at frequencies around all multiples of
FSR. The 3 dBe bandwidth for F = 10 is 0.174 and for F = 30 is 0.0516. Notice that this band-
width is somewhat larger than the resonator line-width (0.1 for F = 10 and 0.033 for F = 30).
Comparing Figures 5.4 and 5.6 for the F = 10 and baseband resonance curves, the 3 dBe
bandwidth of the traveling wave electrode, 0.087, is slightly wider than that for the lumped
electrode, 0.0845. This means there is a small transit-time effect on the bandwidth of the
lumped electrode.
In addition to the velocity mismatch, another bandwidth limitation in traveling wave
modulators is the loss of the microwave transmission line. For a given electrode dimen-
sion, the high frequency microwave loss is determined by the skin depth and one expects
a loss in dB/cm of a = a0 f 1/2 where a0 depends on electrode conductivity and geometry.
Assuming no velocity mismatch, the effect of loss on the modulation index dn is10
n- 1 L
= ∆βL
1 - e - a mL sin ( nφ/2 )
a mL sin ( φ/2 ) (
sin ω mt -
n+1
2
φ ) (5.25)
where am = a / 8.7 converts from power loss in dB/cm to an exponential amplitude loss
coefficient.
Table 5.1
Numerical Example Values of an EO Polymer Device
Index of polymer no 1.6
EO coefficient r 50 pm/V
Overlap integral G 1
Wavelength of light l 1.31 mm
Electrode gap g 10 mm
Tuning sensitivity df/dV 1.47 GHz/V
V MZ
π L 6.4 Vcm
Vπeq L (Finesse 10) 1.58 Vcm
0
Ring
Response (dB in electrical)
–5
–10
–15
MZ
–20
0 15 30 45 60 75 90
Modulation frequency (GHz)
Figure 5.7
Effect of velocity mismatch (Dn = 0.2, no microwave loss) in modulation frequency response of the ring modu-
lator with the FSR of 30 GHz (L = 6.25 mm) and F = 10 compared to the equivalent broadband MZ modulator
(L = 4.04 × 6.25 mm). The responses are normalized by the low-frequency response.
To evaluate the effect of velocity mismatch and microwave loss in traveling wave ring
modulators, the small signal frequency responses of the output with a MZ modulator with
equal low-frequency Vp is compared. As a numerical example, the parameters in Table 5.1
are used. Assuming a FSR of 30 GHz, a finesse of 10, and critical coupling, the perimeter
of the ring resonator is set to be 6.25 mm and the Vπeq is 2.53 V. The electrode length of the
equivalent MZ modulator is 2.53 cm.
Figure 5.7 shows the effect of velocity mismatch in the modulation frequency response of
the ring modulator and the equivalent MZ modulator for Dn = 0.2. The mismatch in LiNbO3
is much larger (Dn~2) but can be reduced by design while the mismatch is negligible in EO
polymer modulators. Dn = 0.2 was chosen as a compromise which clearly shows the effect.
Figure 5.8 shows the effect of microwave loss in the modulation frequency response of
the ring modulator and the MZ modulator with equal low-frequency Vp. The microwave
power loss coefficient of the microstrip line is 0.7 dB/cm GHz which is typical for modu-
lators. No velocity mismatch is assumed. In the both cases, the ring modulator shows
higher modulation efficiency at frequencies around multiples of FSR than the equivalent
MZ modulator because of the shorter electrode length.
0
Ring
–10
–15
–20
0 15 30 45 60 75 90
Modulation frequency (GHz)
Figure 5.8
Effect of microwave loss (a = 0.7 dB/cm GHz , Dn = 0) in modulation frequency response of the ring modula-
tor with the FSR of 30 GHz (L = 6.25 mm) and F = 10 compared to the equivalent broadband MZ modulator
(L = 4.04 × 6.25 mm). The responses are normalized by the low-frequency response.
Transmission
Modulation
sidebands Modulation
sidebands
Frequency Frequency
Optical frequency
(c) Modulation Modulation
sideband sideband
Transmission
Frequency
Free spectral range
Figure 5.9
This sideband picture of the modulation provides another explanation for the peaks in the response of the ring
resonant modulator at a modulation frequency which matches the FRS.
The high response of the ring resonant modulator at the FSR modulation frequency can
be understood by considering the position of the optical sidebands due to the modulation
as shown in Figure 5.9. In part A the modulation frequency is low enough to fit the side-
bands in the first resonant response of the ring. In part B, the modulation frequency has
increased and the modulation sidebands fall outside the ring response and the modulator
has poor efficiency. In part C the modulation frequency has increased to near the FSR and
the sidebands again fall under an optical response on the ring.
In summary, ring resonator based traveling wave modulators show high modula-
tion efficiency at frequencies around all multiples of the FSR and have better tolerance
for the velocity mismatch and microwave loss of the electrode than traveling wave MZ
modulators, thus making them useful in microwave photonics applications. An exam-
ple other than communication applications is metrology systems.23 Efficient generation
of modulation sidebands improves the precision of measurements. A ring modulator is
mathematically identical to a Fabry–Perot modulator.24 In the traveling wave Fabry–Perot
modulator,25 however, the microwave can interact only with one direction of propagated
light. In traveling wave ring modulators, the microwave can interact with light in the full
length of a round trip due to unidirectional propagation. Hence, they have a great poten-
tial not only for modulators but also for other applications such as comb generation and
pulse generation previously demonstrated by Fabry–Perot modulators.25,26
r = 1 mm
RF out RF in
Au AJL8/APC
UFC170 4 µm
1 µm
2 µm
UV15 2 µm 2 µm 2 µm 5 µm
Si
Au
Figure 5.10
The top view optical microscope image and schematic cross section of the fabricated modulator.
respectively. Since the material loss of AJL8/APC is approximately 2 dB/cm, the excess
2.8 dB/cm is the scattering loss from the sidewall roughness due to fabrication. From this
data, the parameters in the theoretical transfer function (Equation 5.5) are a = 0.696 and
t = 0.783. The EO tuning sensitivity, measured by applying a 100 Hz triangular signal,
(Figure 5.13) was found to be 1 GHz/V, which corresponds to an effective EO coefficient
r = 33 pm/V of the core layer. The VπIM of the modulator at a low modulation frequency are
estimated to be 7.5 V.
Patterned Cr/Au
Si
Spin cladding
UV15
Lithography
RIE
Spin core
AJL8/APC
Spin cladding
UFC170
Corona poling
Cr/Au deposition
lithography
Au plating
Au
Figure 5.11
Fabrication of traveling wave ring resonator.
–3
–9
–12
–15
–10 0 10 20 30 40
Frequency detuning from resonance (GHz)
Figure 5.12
The transmission spectrum of the TM mode of the ring modulator.
c1
Figure 5.13
The modulated TM light from the ring modulator (Vpp = 20 V).
transmission line. At the other end of the line, a RF terminator is required to minimize
the reflection. A 50 W chip resistor (State of the Arts S0202AF) is put on the side of the pad,
and then the contact on resistor is connected to the termination pad on the device by Au
ribbon. The ground plane is also bonded to common ground by Au ribbons. The situation
is illustrated in Figure 5.15. Figure 5.16 shows the reflection S-parameter, S11, from 1 to 40
GHz, as measured by a network analyzer. S11 is less than −10 dB from 0 to 30 GHz, which
indicates the RF impedance match is reasonable.
394 µm
Top view
500 µm 517 µm
250 µm
Ground Contact pad
Cross section
Substrate
Polymer
Figure 5.14
The detailed dimensions of the feeding pad.
Bottom electrode
Microstrip
Device
Au Ribbon
Ground
Contact
Chip resistor
Figure 5.15
Schematic drawing of the microstrip termination.
–20
–30
0 10 20 30 40
Frequency (GHz)
Figure 5.16
S11 (reflection) of the traveling wave electrode.
–25
22 GHz
–30 28 GHz
34 GHz
–35
Optical power (dBm)
–40
–45
–50
–55
–60
–0.3 –0.2 –0.1 0 0.1 0.2 0.3
Wavelength shift (nm)
Figure 5.17
Optical spectra of the modulated light at 22 GHz, 28 GHz, and 34 GHz.
the same bias point in each modulation frequency, the input light was tuned to minimum
transmission which is at resonance of the ring modulator. The observed sidebands were
therefore due to phase modulation and the carrier is suppressed. The data shows that the
sideband power peaks at the modulation frequency of 28 GHz, which is the FSR of the
ring, and the sideband power decreases as the RF frequency is tuned away from the FSR.
Figure 5.18 shows the modulation frequency dependence of optical sideband powers.
The side band power shows the minimum at the modulation frequency of 14 GHz, which
is off-resonance of the resonator, and the maximum at 28 GHz. The 3 dBe bandwidth of
the detected signal power is the same as the 3 dB bandwidth of the sideband power. From
Figure 5.18, the signal 3 dBe bandwidth is ~7 GHz. This modulator would therefore find
application in analog optical links with the sub-carrier frequency of around 28 GHz with
a useable bandwidth of ~7 GHz.
–30
–35
–40
–45
–50
0 10 20 30 40
Modulation frequency (GHz)
Figure 5.18
The modulation frequency dependence of optical sideband powers.
–10
+0.014 nm
Resonance
–20 –0.014 nm
Optical power (dBm)
–30
–40
–50
–60
1309.4 1309.5 1309.6 1309.7 1309.8 1309.9 1310
Wavelength (nm)
Figure 5.19
The bias dependence of 28 GHz modulation spectra.
The bias dependence of 28 GHz modulation spectra is shown in Figure 5.19. When the
input light is tuned to a resonance, the carrier is suppressed by 15 dB by the optical reso-
nance and the 28 GHz modulation sidebands are observed. At this point there is only a
small intensity modulation at 28 GHz. When the input light is tuned at near a half trans-
mission point, ± 2.5 GHz (± 0.014 nm) from a resonance, the carrier suppression is approxi-
mately 3 dB and the 28 GHz modulation sidebands are also observed. At this point there
is significant intensity modulation at 28 GHz.
Power
meter
Isolator
BWO
Photodetector
3dB Coupler
Figure 5.20
Millimeter-wave ring modulator test setup. The optical coupler after the ring modulator allowed continuous
power monitoring at the photodetector to ensure proper wavelength bias for pure phase modulation. To obtain
the optical modulation response, the modulated sideband power to optical carrier ratio was measured on the
optical spectrum analyzer (OSA) and divided by the monitored input modulation power from the backward
wave oscillator (BWO) millimeter-wave source, as measured at the millimeter-wave 3 dB coupler.
wavelength of 1.31 mm. After passing through a polarization controller, the light was butt-
coupled to the device. The tunable laser was tuned to a minimum of the ring modulator’s
transfer function, ensuring that pure optical phase modulation with suppressed carrier
occurs. The modulated optical output was divided into two paths by a 3 dB optical coupler.
One arm of the optical output was connected to an optical power meter to ensure phase
modulation of the ring modulator throughout the experiment by holding the transmitted
optical power to a minimum and performing fine tuning of the wavelength of the tunable
laser as needed. The other arm was connected to an optical spectrum analyzer (OSA), the
Ando AQ6319, which has a spectral resolution of 0.01 nm, yielding a measurement uncer-
tainty of about 1 GHz.
A backward wave oscillator (BWO), micro-now model 705C with model 728 BWO head,
was used as a millimeter wave source, producing up to 17 dBm in millimeter-wave power
within the W-band (75–110 GHz). The millimeter-wave setup consisted of entirely W-band
waveguide components to minimize resonances before the device. An isolator was placed
between the BWO and the waveguide to reduce back-reflected waves and a 3 dB coupler
was used after the isolator. One of the coupler’s outputs was used to drive the ring modu-
lator and the other output was coupled to an Anritsu ML83A power meter with W-band
MP81B head. The power transfer function from both outputs of the coupler was fully char-
acterized across the W-band measuring a power difference between both outputs of less
than 2 dB. In addition, this frequency scan showed a smooth spectral profile of the BWO
output, indicating that resonances within the BWO and the external components were
negligible. The millimeter-wave signal was coupled to the device via a co-planar (G-S-G)
W-band probe (Picoprobe). The 10 µm EO polymer stack was scratched on either side of
the microstrip to expose the ground plane. Direct contact was then established between
the signal tips and ground tips of the probe with the microstrip and exposed ground plane
areas, respectively. The microstrip line was terminated by an on-chip commercial 50 W
resistor that has been rated to 50 GHz.
We observed the modulation sideband power on the optical spectrum analyzer and
divided by the millimeter-wave power input to the probe, as measured at the coupler
before the probe. The largest power input into the probe was 8 dBm. The experimentally
obtained optical modulation response data across the W-band is plotted in Figure 5.21.
A simulated optical modulation response for phase modulation is also plotted in the same
–10
–15
Figure 5.21
Experimental optical modulation response plotted along with simulated optical modulation response of a ring
modulator and an MZ modulator. Both velocity mismatch (Dn = 0.1) and microwave loss (0.8 dB/cm GHz ) are
considered in the simulation. Both simulated devices have the same effective low-frequency and are biased at
the nulls of their transfer function to produce pure phase modulation. The agreement between the experimen-
tal data and simulation validates the proposed analysis.
figure. Pure phase modulation occurred since the ring modulator was wavelength biased
at the null of its transfer function. For comparison purposes, the simulation results plot-
ted in Figure 5.21 give the optical modulation response for both the ring resonator and an
equivalent MZ modulator when both are biased for phase modulation. The experimental
data was normalized to the simulation results by matching the responses at the 84 GHz
resonance. The simulation included a velocity mismatch of Dn = 0.1 and microwave loss of
the electrodes of 0.8 dB/cm GHz . There is a good agreement between the analysis/simu-
lation above and the experimental results. The third and fourth modulation resonances
at 84 GHz and 111 GHz are evident in the experimental data, rising approximately 10 dB
above the experimental off-resonance response. Figure 5.22a and b show the optical carrier
(center) and modulation sidebands that were seen on the OSA when obtaining the modu-
lation response data at the third and fourth order resonances. If the FSR is 28 GHZ, the
fourth resonance should occur at 112 GHz. The difference could be error in the microwave
frequency measurement or that the FSR is slightly less than 28 GHz.
In addition, the modulation response in the D-band (110–180 GHz) was also obtained.
The microwave module, including the BWO source (Model 731 BWO head), the coupler
and the waveguide, were replaced by components in the D-band. The D-band coupler was
also characterized by the same procedure as above. The probe accepts a W-band wave-
guide and therefore a D-band to W-band waveguide transition was used in this setup.
The probe used to apply the millimeter wavelength driving signal is rated to 120 GHz and
has higher millimeter-wave loss and potential oscillations in its transmission spectrum
above 120 GHz. This makes full characterization of the ring modulator throughout the
D-band difficult. Nonetheless, the device showed substantial modulation at the fifth and
sixth order resonances of 139 GHz and 165 GHz, while remaining mostly ineffective at
off-resonance frequencies. Figure 5.23a and b are the experimentally obtained OSA traces
showing the optical carrier and modulated sidebands at 139 GHz and 165 GHz, illustrating
the observed modulation response at resonance.
–60 –60
–70 –70
–80 –80
–90 –90
1309.7 1310.2 1310.7 1311.2 1311.7 1309.7 1310.2 1310.7 1311.2 1311.7
Wavelength (nm) Wavelength (nm)
Figure 5.22
(a) and (b). Phase modulation sidebands observed on the OSA at 84 GHz (3 × FSR) and 111 GHz (4 × FSR).
(a) (b)
–30 –40
–40 –50
fm = 165 GHz
fm = 139 GHz
Optical power (dBm)
–50 –60
PSD (dBm)
–60 –70
–70 –80
–80 –90
–90 –100
1309.6 1310.1 1310.6 1311.1 1311.6 1311 1311.5 1312 1312.5
Wavelength (nm) Wavelength (nm)
Figure 5.23
(a) and (b). The measured output optical spectrum of the ring modulator upon modulation by the BWO at the
resonant millimeter-wave frequencies of 139 GHz (5 × FSR), and 165 GHz (6 × FRS).
Figure 5.24
Example block diagrams of OSP circuits. The input is split into multiple waveguides. The individual optical sig-
nal experiences equally different time delays represented by Z -1, and amplitude and phase changes represented
by the ak, bk, ck, and dk coefficients. At the output port, they are combined again to generate an output signal.
Example block diagrams of OSP circuits are shown in Figure 5.24. The input optical sig-
nal is launched at the left side and split into multiple waveguides. The individual optical
signal experiences equally different time delays represented by t and Z -1 in the diagrams,
and amplitude and phase changes represented by the ak, bk, ck, and dk coefficients. At the
output port located at the right side of the diagram, they are combined again to generate
an output signal.
Two typical configuration structures are distinguished; transversal form28 and lattice
form13,14 as seen in Figure 5.24a and b, respectively. The transversal form is a parallel struc-
ture since the signal is processed in a parallel way while the lattice form is a serial struc-
ture where signals are processed in series. If we define a waveguide branch arm, an arm
can be classified into two types; forward feeding arm and backward feeding arm. Since
the response of forward feeding arms is finite in time, it is called finite impulse response
(FIR). In a similar way, the response of backward feeding arms is called infinite impulse
response (IIR) since it is infinite. The number of the forward feeding arms and of back-
ward feeding arms represent the order of FIR and IIR respectively. The origin of this ter-
minology is from the digital filter theory in electronics.29 According to this theory, it is
well known that any arbitrary response can be obtained when FIR and IIR are combined
together. The order of FIR and IIR is associated with how arbitrary an OSP can describe
the response.
Several physical values are important for characterizing an OSP. They are the unit delay
time, t, and the number of arms that split or combine lights (N and M in Figure 5.24a).
The unit delay time of an OSP is analogous to the sampling time in discrete-time signal
processing. It represents a time resolution for the OSP to process a signal. Due to this time
resolution, the frequency response of an OSP is periodic and the period is proportional to
1/t. This period is called the free spectral range (FSR).
The number of optical splitting and combining arms determines the spectral resolution
of the OSP within a FSR. As the number of arms increases, a sharper frequency response
can be obtained and hence a more arbitrary response is possible. In that sense, it is analo-
gous to the number of eigen functions in the frequency domain because the overall fre-
quency response of an OSP is a response combination of the individual arms.
y(t) - ∑
n= 1
bn y(t - nτ) = ∑ a x(t - mτ)
m= 0
m (5.26)
where the coefficients, am and bn, stand for the amplitude and phase changes of the m’th
forward feeding arm and the n’th backward feeding arm respectively. The coefficients are
complex values in general because the signals, (x)t and (x)t, stand for coherent optical fields,
and are thus considered to have both amplitude and phase. The amplitude and the phase
of a coefficient corresponds to the amplitude and phase change of an individual signal.
The second method for describing this system is the transfer function representation. In
general, the coefficients, am and bn, and the delay time, t in Equation 5.26 can be functions
of time as well if we change (or modulate) those parameters in time. If their varying rate is
comparable to that of the optical signal such as in electro-optic polymer waveguides, their
time-dependent effect must be considered. We sometimes apply time-varying signals to
the time-delay lines to effectively change t in our experiments. However, we assume that
the time-varying rate is small enough or the delay control mode (see Section 5.3.1.3 for its
definition) is only considered in the OSP operation. In this case, the OSP becomes a linear
system and is similar to digital filter in many ways. Indeed, the OSP has the same trans-
mission characteristics and features as those of a corresponding digital filter. Therefore,
the conventional Z transform analysis, which is frequently used with digital filters, can be
applied to analysis of the OSP. In Z transformation, the transfer function is represented as
the rational functions of Z.
If we calculate the Fourier transformation of Equation 5.26 with respect to the eigen
functions of ejpwt where p is any integer and w is the optical angular frequency and replace
the e-jwt with Z -1, we obtain the transfer function, H(Z), in Z domain.
∑ aZ
M
-m
m
H (Z) = m= 0 (5.27)
1- ∑ b Z
N
-n
n
n= 1
∏
M
(1 - ρi Z -1 )
H (Z) = A i=1
(5.28)
∏
N
(1 - γ j Z -1 )
j=1
Intensity response
0.2
0.6
0
–0.2
Unit 0.4
–0.4
circle
–0.6
Pole 0.2
–0.8
Zero
–1
0
–1 –0.8–0.6–0.4 –0.2 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 250 300 350
Normalized frequency
Figure 5.25
Pole/zero diagram examples. (a) Pole/zero diagram. Suppose that an OSP is configured such that it has poles
(star dots) and zeros (circle dots) located in the Z space (b) Corresponding transfer function in frequency domain
(linear scale). Around zeros, the magnitude of the transfer function is small while it becomes large near poles.
where ri and gj are the ith zero and jth pole and A is the amplitude factor. Figure 5.25a and b
shows an example of pole/zero diagram and its corresponding transfer function, respec-
tively. As the optical frequency increases, the variable Z in Equation 5.28 follows the unit
circle counter-clockwise in the Z space. Around the zeros, the magnitude of the transfer
function becomes small while it increases near the poles.
Once we find the transfer function, H(z), the functionality of the OSP is exactly defined.
This can be done by defining the coefficients, am and bn. Therefore, operation of an OSP
depends on how flexible and stable these coefficients can be tuned.
as will be shown in Sections 5.3.4 and 5.3.5, we sometimes apply time-varying signals to
the delay lines or power splitting control elements to effectively change the delay time or
the coefficients, respectively.
We investigate the OSP applications using the delay control mode rather than the fre-
quency control mode or the coefficients control mode when we discuss its applications in
Section 5.3.5. If the frequency control mode is used in an OSP with the electro-optic poly-
mer waveguides, the OSP can tune such optical filters much faster. However, the filter per-
formance will be limited if electro-optic polymers are used in place of silica waveguides
due to its intrinsic optical loss. Furthermore, operating our polymer OSP in the coefficients
control mode is not particularly interesting, either. The OSP we investigate is classified
into the lattice form structure. In the lattice form structure, the ck and dk coefficients in
Figure 5.24b are complex functions of actual electrical biases, which makes it difficult to
find the relationship between its output function and the electrical biases.13,14 In addition,
we have not found any useful and unique applications using this mode partly because the
optical loss limits the performance of the OSP building blocks.
One can find that the frequency control mode is based on the same principle as the delay
control mode. In Equation 5.27, the delay time t and the optical frequency w are multiplied
together. Therefore, either affects the transfer function in the same way. We can change the
delay time using the electro-optic effect with a fixed input optical frequency (wavelength)
or the optical frequency can be changed using a fixed delay time. Either approach will
generate the same output response. This is the basic concept for the arbitrary waveform gen-
erator, we investigate in Section 5.3.5.1 and it becomes more clear with the driving formula
of Equation 5.34.
5.3.2 Structure
Since multiple power splitting is hard to obtain and not effective in the optical waveguide,
the lattice form structure is considered for implementation of our OSP. The lattice form is
a series structure with a certain type of a unit block. The unit block which we have chosen
consists of a symmetric MZ interferometer and a racetrack waveguide as shown in Figure
5.26. “Symmetric” means that the lengths of the two waveguide arms of the MZ interfer-
ometer are the same. A racetrack structure is used so that the straight waveguide section
has an extended coupling region for coupling inside and outside of the ring.
This unit block, originally proposed by Jinguji,14 has two input ports and two output
ports. Any input port can be used for operation while the two output ports are related to
Фr
tr
t1 t2
Фm
Figure 5.26
Unit block of OSP. It is a two-port input and two-port output system consisting of a symmetric MZ interferom-
eter and a racetrack structure. Four electrodes control the locations of a zero and a pole.
each other by a conjugate power relation. A conjugate power relation means that the total
sum of two output powers is the same as the input power if the system is lossless.
This structure generates one zero and one pole simultaneously. The degree of freedom
to locate a zero or a pole in the Z space is two because a zero or a pole is a complex value,
and hence has both amplitude and phase. The locations of the zero and the pole in the Z
space are controlled by the four different electrodes. Note that t1 and t2 are functionally
redundant.
A multiple-block OSP consists of cascaded multiple unit blocks as shown in Figure 5.27.
With N cascades of unit blocks, our OSP can generate N zeros and pole pairs. Therefore, the
degree of freedom of the N cascaded OSP is N × 4, which are controlled by N × 4 electrodes
on top of the waveguides. In the strict sense, the definition of the unit block in Figure 5.26
is wrong if the multiple-block is considered. It should be one of the divided sections in
Figure 5.27 except the first block, which is just a configurable coupler. To avoid confusion,
we name the structure in Figure 5.26 as a “one-block OSP” and one of the divided sections
in Figure 5.27 as a “unit block”.
The unit block contains two configurable couplers and two phase shifters. They are the
“splitting coupler”, the “racetrack coupler“, the “MZ phase shifter“, and the “racetrack
phase shifter”, labeled as t1, tr, fm and fr, respectively, shown in Figure 5.26. They are used
for configuring one zero and one pole generated by a unit block and controlled by micros-
trip electrodes located on top of the waveguides via the electro-optic effect.
In order to have a useful phase shift at the racetrack phase shifter, the perimeter of the
racetrack must be large enough. On the other hand, the round trip loss of the racetrack
will be too large if the perimeter is large. Therefore, we design the radius of the racetrack
to be 1.2 mm. The interaction lengths of the racetrack coupler and the splitting coupler are
designed as 2 mm and 6 mm, respectively. The detail design parameters of the individual
building blocks or components are discussed in the following sections.
5.3.3 Analysis
As discussed earlier, the one-block OSP generates a single zero and a single pole simulta-
neously. The degrees of freedom to locate a zero or a pole in the Z space is two since a zero
or a pole is a complex value, and thus has both amplitude and phase. Their locations in
the Z space are controlled by the four different electrodes shown by the red shaded bars in
Figure 5.26. With N cascades of the unit block, an OSP can generate N number of zero and
pole pairs as shown Figure 5.27. The poles of the whole OSP system are the same as those
of the individual unit blocks. On the other hand, the zeros of the whole system are not the
Figure 5.27
Multiple structure of OSP. It consists of N cascaded unit blocks.
same as those of the unit blocks since both the output power and the conjugated power
of a unit block are cascaded to the next unit block by coupling each other. Because of this
problem, it is not trivial to identify the zeros of a multiply cascaded structure. Jinguji et al.
demonstrated the synthesis method for analyzing these structures.14 However, their tech-
nique cannot be applied generally to a lossy system since their technique assumed that the
unit structure is lossless.
However, dealing with just a one-block OSP is relatively easy and straight forward. By
using the scattering matrices, we can derive the scattering parameters, which are useful to
understand the operation of the one-block OSP. First, before deriving the scattering matri-
ces, we define the scattering matrices of the individual components. And then, we cascade
their matrices to find the scattering matrices of the entire one-block OSP.
where t and k (|t|2 + |k|2 = 1) are the transmission and coupling ratio of each coupler and a
represents the optical losses inside the coupler. Note that the matrix should be an unitary
matrix if a lossless case is assumed (a = 1).31 In a real situation, all these numbers are com-
plex. If we change or tune the coupling ratio, the phases as well as the amplitudes of all val-
ues change. Due to the electro-optic effect, they are all functions of the applied voltage.
5.3.3.2 Racetrack
The schematic diagram for a racetrack structure is shown in Figure 5.28a. In order to find
the transfer function for the racetrack, we use the scattering matrix of the racetrack cou-
pler, which is Equation 5.29. In the steady state condition, the input electric field, Ein and the
output electric field, Eout, satisfy Equation 5.30.
(a) (b) Фr
Фr
tr A
E΄ B
t1 t2
Ein Eout Фm
tr
A B S1 S2 S3
Figure 5.28
(a) Racetrack structure with appropriate symbols for mathematical analysis. (b) Unit block of OSP with appro-
priate symbols for mathematical analysis. The individual scattering matrix transfer function consists of two dif-
ferent terms. One, labeled as “Path A”, is associated with the light beam which propagates through the racetrack
and the other, labeled as “Path B’, is with the light beam which propagates through the other MZ arm.
where E′ is a steady state solution at the starting point of the racetrack as indicated in
Figure 5.28a, tc and tr are the transition time in the coupler (from A to B counter-clockwise)
and in the racetrack (from B to A counter-clockwise), and ar and aRT1 is the optical loss fac-
tor from A to B and from B to A, respectively. The optical loss factors represent the electric
field attenuation on a linear scale and become unity in lossless waveguides. The param-
eters tr and kr denote the transmission and coupling coefficients for the racetrack coupler.
The optical loss factor is a measure of electric field intensity attenuation on a linear scale.
We solve Equation 5.30 to find out the transfer function of the racetrack, HRT, which is the
complex ratio of Eout with respect to Ein.
where t( = tc + tr) is the total round trip time, and aRT( = aRT1ar )is the total round trip loss.
5.3.3.3 MZ Section
In the symmetric MZ section, two light beams propagate independently. Hence, the scat-
tering matrix for the MZ section, S2 becomes,
H RT 0
S2 = (5.32)
0 a me jφm
where am and fm are the optical loss factor and the phase shift in the MZ phase shifter.
S11 S21
S = S1 S2 S3 = (5.33)
S12 S22
where S1 and S3 are the scattering matrices for the two splitting couplers and S2 is the MZ
section as shown in Figure 5.28b. The calculation is straight forward and the scattering
parameters can be summarized as,
∗ j τω j φ r
t -a e e jφ m
Slm = a lm r RT +Rlm e
1- tr a RT e jτω e jφ r (5.34)
Path B
Path A
where,
a 11 = a 1a 2a r (+1)t1t2
a 12 = a 1a 2a r (- j)t1c2∗
a 21 = a 1a 2a r (- j)c1t2
a 22 = a 1a 2a r (-1)c1c2∗
As seen in the under brackets in Equation 5.34, the individual scattering matrix transfer
function consists of two different terms. One, labeled as “Path A’, is associated with the
light beam which propagates through the racetrack and the other, labeled as “Path B’, is
for the light beam which propagates through the other MZ arm. The Path A and B terms
correspond to two paths indicated in Figure 5.28b. Therefore, the Path A term contains the
transfer function of the racetrack while the Path B term is independent of the optical delay
line formed by the racetrack.
The two terms are summed at the output port depending on the coupling ratio of the two
splitting couplers. The amplitude of Rlm stands for the normalized intensity of Path B with
respect to the maximum intensity of Path A. Note that the amplitude and the phase of Path
B beam of light are controlled by the two splitting couplers and the fm electrode, respectively
and that the transfer function and the resonant wavelength of the racetrack (Path A light
beam) are controlled by the tr and the fr electrodes, respectively. The mathematical represen-
tation in Equation 5.34 is useful for an intuitive and physical understanding of the one-block
OSP structure and it is used when we verify the OSP experimentally in Section 5.3.4.6.
Another useful and more mathematical way to represent the scattering matrices is using
the concept of poles and zeros. Further simplification of Equation 5.34 results in:
γ = a RT tr (5.37)
1 + tr Rlm
ρlm = γ
| tr |2 +tr Rlm
All transfer functions have the same pole, g, while rlm is the zero obtained from the Slm
scattering matrix element. As we discuss before in Section 5.3.1.2, using poles and zeros is
convenient to understand the output characteristics of the OSP.
Note that the phase of g defined in Equation 5.36 is not configurable. The definition of
the pole should include g as well as either ejør or ejωτ in Equation 5.36 because the two terms
contribute to the pole phase. However, we define our pole as shown in Equation 5.36 since
it is convenient to understand two similar operation modes; frequency control mode or
delay control mode. In this way, the similarity between two operation modes of the OSP,
as discussed in Section 5.3.1.3, becomes clear. In the frequency control mode, the ejτω term
in Equation 5.36 is varying and represents the unit delay Z-1 in the Z space. In this case,
the actual pole becomes gejør, thus, the pole phase is controlled by the φr electrode. From a
practical point of view, the frequency response of the OSP is periodic with respect to the
FSR and the response shifts in the frequency domain with respect to the φr value. On the
other hand, in the delay control mode, the delay is controlled by ejør in Equation 5.36, which
represents the unit delay Z-1 in the Z space. In this case, the actual pole becomes gejωτ, thus,
the pole phase is controlled by the input optical frequency (ω) With a similar analogy from
a practical point of view, the amplitude and phase response of the OSP is periodic with
respect to φr and the response shifts in the φr (or voltage) domain with respect to the input
optical frequency. Note also that the phase of the zero also depends on both parameters
in the same manner. By changing any of two parameters, the phases of both pole and zero
are changing in the same amount. However, the zero has additional configurable degrees
of freedom such as t1, t2 and φm as seen in Equation 5.36.
all these components are working correctly. Therefore, it is important to characterize and
verify the individual components before integrating them. In order to verify the compo-
nents, we also fabricate individual components. In this section, we discuss the experimen-
tal results for individual components” verification.
5.3.4.1 Racetrack
First, we characterize the racetrack, which is one of the crucial components in an OSP.
Figure 5.29 shows the spectral response of a racetrack and its experimental measurement
setup. The racetrack measured has the same design as the OSP, where the bending radius
of the racetrack is 1.2 mm and the interaction length of the racetrack coupler is 2 mm. An
Ando AQ4321D is used for the tunable laser source at 1.5 µm with TM mode polarization
control. As seen in Figure 5.29b, the measurement is done in a 0.6 nm wavelength span and
the FSR, extinction ratio, finesse and Q-factor (loaded) are measured 0.12 nm (15.5 GHZ),18
dB, 3.36 and 4.34 × 104, respectively. Using these values, the effective group refractive index,
αRT value, tr value, total round trip optical loss and the optical loss inside the ring are calcu-
lated as 1.66, 0.608, 0.535, 4.4 dB and 3.85 dB/cm, respectively. Figure 5.29b also shows the
simulated spectral response using these values. Since the propagation loss of the straight
waveguide is measured around 2 dB/cm, the excess loss inside the racetrack is expected
to be from scattering loss due to the roughness at the interface between core and cladding
materials. A similar propagation loss inside the racetrack has been obtained using 1 mm
bending radius and similar electro-optic polymer material.32
We drive voltages of ± 20 V on the tr electrode to verify the operation of the racetrack
coupler as seen in Figure 5.30a. Figure 5.30b shows the spectral response while varying
the driving voltages. The spectral response when no voltage is applied is shown as well
for purpose of a comparison. As seen in Figure 5.30b, the response (extinction ratio) is
changed depending on the voltage. By fitting the measured response, we find that tr
becomes 0.535 ± 0.25 with an applied voltage to the racetrack coupler of ± 20 V. We also find
that the local minima shift with applied voltage.
−5
−10
Tunable laser Photo
detector
−15
2.4 mm
Polarization 2 mm −20
Polarizer 1550 1550.1 1550.2 1550.3 1550.4 1550.5 1550.6
control
Wavelength (nm)
Figure 5.29
Experimental setup and spectral response of a racetrack which has the same design as the OSP used. The free
spectral range (FSR), extinction ratio, finesse and Q-factor are measured 0.12 nm (15.5 GHz), 18 dB, 3.36 and
4.34 × 104, respectively.
Figure 5.30
Experimental setup and spectral response of a racetrack when ± 20 V is applied to the racetrack coupler. By fit-
ting the measured response, we find that tr becomes 0.535 ± 0.25 with an applied voltage to the racetrack coupler
of ± 20 V.
1
−20
Photo 0.8
Tunable laser
detector 0.6
0.4
Polarization 0.2
Driving Polarizer
control
voltage 0
−10 −5 0 5 10
Time (ms)
Figure 5.31
Intensity response of a racetrack when ± 20 V peak-to-peak triangular signal is applied to the racetrack phase
shifter. The measurement shows that the half-wave voltage (Vp) of the fr phase shifter is 18.3 V. The correspond-
ing r33 coefficient is also calculated as 23 pm/V.
For verification of the racetrack phase shifter, we fix the input optical wavelength at
1.5 µm and apply ± 20 V peak-to-peak triangular signal to the φr electrode. Its experimental
setup and response are shown in Figure 5.31a and b, respectively. The measurement shows
that the half-wave voltage (vπ) of the racetrack phase shifter is 18.3 V. The corresponding r33
coefficient is also calculated as 23 pm/V.
Normalized intensity
1 1
−20 −20
0.8 0.8
0.6 0.6
0.4 0.4
0.2 S11 Intensity 0.2 S11 Intensity
S12 Intensity S12 Intensity
0 0
−10 −5 0 5 10 −10 −5 0 5 10
Time (ms) Time (ms)
0 0.75
0.6 −20 0.7
0.65
0.5 0.6
0.55
0.4 0.5
S11 Intensity 0.45
S12 Intensity
0.3 0.4
−10 −5 0 5 10 −20 −15 −10 −5 0 5 10 15 20
Time (ms) Driving voltage (V)
Figure 5.32
Experimental measurements of S11 and S12 scattering parameters for the different racetrack couplers.
Figure 5.32 shows the measured S11 and S12 responses of the racetrack couplers. From the
measured response, we calculate the transmission coefficients for the couplers, which are
also shown in Figure 5.32d. Comparing to the computer simulations, the measurements
show that the measured transmission coefficients are smaller than the simulated values.
Furthermore, the different polarity of voltages leads to different output response even
though output responses should be even functions with respect to applied voltages since
the waveguides are symmetric. This implies that two waveguides inside the couplers are
mismatched already due to imperfect fabrication, which has also been found in conven-
tional electro-optic MZ devices.33
Figure 5.33 shows measured S11 and S12 responses of the splitting couplers. Since the
interaction length is designed large enough (6 mm), both outputs reach their maximum
and minimum intensities during the application of the triangular voltage waveform on the
electrodes. The turn on/off voltage of the splitting coupler is 30 V and its extinction ratio is
approximately 10 dB, implying that its transmission coefficient, t1 or t2, can be configurable
between 0.4 and 0.9 with an applied voltage of 30 V.
Drv Voltage 20
Normalized intensity
1
−20
0.8
0.6
0.4
0.2
S11
S12
0
0 20 40 60 80 100
Time (ms)
Figure 5.33
Experimental measurements of S11 and S12 scattering parameters for the splitting coupler.
0.75 10
Normalized intensity
5 mm 0.25 −10
Driving Polarizer
Tunable laser 0 −20
voltage −10 −5 0 5 10
Time (ms)
Figure 5.34
Intensity response of a racetrack when a ± 20V peak-to-peak triangular signal is applied to the fm electrode. The
measurement shows that the half-wave voltage (Vp) of the MZ phase shifter is 33 V. The corresponding r33 coef-
ficient is calculated as 20 pm/V.
signal to the φm electrode. Its experimental setup and response are shown in Figure 5.34a
and b, respectively. The measurement shows that the half-wave voltage (Vπ) of the MZ
phase shifter is 33 V. The corresponding r33 coefficient is calculated as 20 pm/V.
• The FSR of the racetrack is 0.12 nm (15.5 GHZ) and the αRT value is constant 0.608.
The transmission coefficient, tr, varies between 0.535–0.25 and 0.535 + 0.25 with
± 20 V applied voltage to the tr electrode.
Race-Track
Фr FSR 0.12 nm
αRT 0.608
tr 0.535 ± 0.25
tr with ± 0.20 V
Фr 0 to 2π
t1 t2 with ± 30 V
Coupler Фm
Phase shifter
Фm0 to 2π with ± 33 V
0.4 < t1 < 0.9 with 30 V
same as c1, t2 , c2
Figure 5.35
Summary of one-block OSP. Based on the measurement of individual components, configurable parameters are
summarized with their configurable amount and driving voltages.
• The phase, φr, can be fully configurable (0–2p) with ± 30 V applied voltage to the φr
electrode.
• The phase, φm, of the MZ phase shifter can be fully configurable (0–2p) with ± 33 V
applied voltage to the φm electrode.
• The splitting coupler, t1 and t2, can be configurable between 0.4 and 0.9 with 30 V.
Z Space
0.48 0.17
Pole
Figure 5.36
Pole locations in Z space which can be implemented by one-block OSP. The amplitude of poles is bounded from
0.17 to 0.48 while the pole phase is configured from 0 to 2p as indicated in the shaded region.
However |Rlm| is bounded due to the bounded transmission coefficients of the splitting
couplers. As in Equation 5.34, |Rlm| is the electric field amplitude ratio between Path A and
Path B, which is configured by the splitting couplers. From the bounded transmission coef-
ficients of the splitting couplers, we find the maximum and minimum values of |Rlm|.
On the other hand, the phase of Rlm depends on φm. Since φm can be configured completely
from 0 to 2π by ± 33 V as shown in Figure 5.35, the phase of Rlm is also configured com-
pletely from 0 to 2π. Therefore, the zero can be located in the entire Z space except near
two points, which are the pole and the zero of the racetrack (a RT /tr∗ ) . The shaded area in
Figure 5.37a shows conceptually possible zero locations when the pole and the zero of the
racetrack are given as shown. We use computer simulations to find the possible locations
of the zeros that can be implemented by the fabricated one-block OSP. Figure 5.37b, c, and
d shows their results in the Z space when the pole has the minimum (0.17), middle (0.325),
and maximum (0.48) possible amplitude, respectively. For three plots, a dot closer to the
origin and the other dot are indicating the pole and the zero of the racetrack, respectively.
The zero cannot be located inside the two circles, which indicate boundaries set by the
bounded |Rlm|. As the pole locates near to the unit circle, the zero can be located throughout
the entire Z space.
5.3.4.6 DC Operation
The multiple-block OSP is complicated and it is difficult to understand its operation intui-
tively. However, the one-block OSP is somewhat simple and it is possible to understand its
operation with physical insights. The output of a one-block OSP consists of two different
light paths as seen in Figure 5.28b. Path A is through the racetrack structure while Path B
is through the other straight MZ branch. The amplitude and phase responses of Path A
150 1 30
0.5
180 0
210 330
Pole
240 300
Pole/|t_r|^2
270
90 90
2 2
120 60 120 60
1.5 1.5
150 1 30 150 1 30
0.5 0.5
180 0 180 0
Figure 5.37
The shaded area in (a) shows conceptually possible zero locations when the pole and the zero of the racetrack
(pole/|tr|2) are given. (b), (c), and (d) show simulation results of the possible zero locations in the Z space when
the pole has the minimum (0.17), middle (0.325), and maximum (0.48) amplitude, respectively. For three plots, a
dot closer to origin and the other dot are indicating the pole and the zero of the racetrack, respectively. The zero
cannot be located inside the two circles, which indicate boundaries set by the bounded |Rlm|.
are same as those of ring resonators. On the other hand, the amplitude response of Path
B is only a function of splitting couplers (t1 or t2) and its phase response is independent of
φr. Figure 5.38 summarizes an intuitive understanding of the roles of bias change on the
operation of the one-block OSP. The splitting couplers (t1 or t2) change mostly the extinc-
tion ratio of the intensity response of a one-block OSP. The change of φm changes the shape
of the intensity response of a one-block OSP. We also find that the racetrack coupler (tr)
changes mostly the sharpness of the amplitude and phase response of a one-block OSP.
Figure 5.39 shows various amplitude responses of a one-block OSP with different bias
values and its experimental setup. We apply a ± 20 V peak-to-peak triangular voltage signal
t1 or t3 Ф1 t2
Extinction
Shapes Sharpness
ratio
Figure 5.38
Summary of biases’ role.
Modulation Scope
Polarization
control
Polarizer
CW laser
t1 Bias Фm Bias tr Bias
Figure 5.39
Various intensity responses of one-block OSP and its experimental setup when different biases are applied. The
splitting couplers (t1 or t2) mostly change their extinction ratios while the MZ phase shifter (fm) mostly changes
their shapes.
to the φr electrode and different bias voltages to the t1, φm, and tr electrodes. A continuous
laser source at 1.55 μm is used with a polarization control to launch TM mode light into
the input of the OSP and its S11 output port is measured with an oscilloscope, which is
connected to a photodetector. Using the X–Y display feature of the oscilloscope, we see the
amplitude response of the OSP in Figure 5.39b.
The first plot in Figure 5.39b shows the amplitude response when no biases are applied
to the OSP. It is one of the typical response shapes of the one-block OSP. The second plot
in Figure 5.39b shows the amplitude response when a different voltage is applied to the t1
electrode. Due to the change of the t1 value, the second plot shows an extinction ratio of the
response changes as expected. The third and fourth plots in Figure 5.39b shows another
amplitude response when different voltages are applied to the φm electrode. Due to the
change of the φm value, the plots show changes of the response shapes as we expected.
5.3.5 Applications
Due to its intrinsic arbitrariness, the OSP can be used for many applications. Applications
presented here are for high speed situations, for which a conventional PLC cannot be used.
As for the potential high speed analog applications, we investigate arbitrary waveform gen-
erators and true-time delay elements using our one-block OSP. For discrete-time applications,
optical pulse code generator and optical discrete-time signal processor will be introduced.
Scope
CW laser
Polarization Polarizer
t1 Bias Фm Bias tr Bias
control
0.9 −10
0.8 0.8
0.7
0.6 0.6
0.5
0.4 0.4
0.3
0.2 0.2
0.1
0 0
−20 −15 −10 −5 0 5 10 15 20 0.49 0.5 0.51 0.52 0.53
Applied voltage (V) Time (ms)
Figure 5.40
Rectangular signal generation using a one-block OSP and its experimental setup. (a) A ± 10 V peak-to-peak sinu-
soidal voltage input with a proper offset bias is applied to the fr electrode and other biases are properly adjusted
by voltage supplies. (b) We obtain a proper transfer function for the rectangular signal generation. (c) We utilize
the sharp transition region in the transfer function to generate a rectangular output signal.
with polarization control and the TM mode output of S22 is measured by the photodetector
and the oscilloscope. First, we apply a ± 20 V peak-to-peak sinusoidal voltage input to the
φr electrode. When the other biases are properly adjusted by voltage supplies as shown in
Figure 5.40a, we obtain the OSPs transfer function as a function of the applied voltage as
shown in Figure 5.40b. We then apply a ± 10 V peak-to-peak sinusoidal voltage input with
a proper offset bias to the φr electrode while the other biases are properly adjusted. We
utilize the sharp transition region in the transfer function in Figure 5.40b to generate the
sharper rectangular output signal. As shown in Figure 5.40c, a rectangular voltage signal
is obtained. As the number of unit blocks increases, a more rectangular shape is possible.
Furthermore, this waveform can be quickly changed to another desired shape with differ-
ent sets of parameters due to the fast electro-optic effect. In Section 5.3.5.2, we use a similar
transfer function from Figure 5.40b to generate another linear signal.
d n |Slm |2
Pn = (5.39)
dφ nr
where Pn is the nth high-order harmonic term and n is an integer larger than 0. Instead
of using the analytical Taylor expansion technique, we utilize a numerical method to
calculate the high-order nonlinear harmonic terms with various poles, zeros and biases.
As a result, we find the most linear region of the response (in terms of smallest high-
er-order terms43) when the pole, the zero, and the bias are 0.27∠ -1.037 rad, 1∠ 0 and
2.2019 rad, respectively. In this case, the third, fourth, and fifth harmonic terms van-
ish while the first and second harmonic terms are calculated as 0.287/rad1, 0.001/rad2,
respectively.
The simulated normalized transfer function using these values are shown in Figure
5.41a, where the linear bias is indicated with a red thick line. The normalized transfer
function of a conventional MZ modulator is also shown for the purpose of comparison.
The corresponding pole/zero diagram is also shown in Figure 5.41b. The calculation shows
0.8 0.4
Bias
0.2
0.6 0
–0.2
0.4 –0.4
–0.6
0.2 –0.8
–1
0
0 100 200 300 –1 –0.8–0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1
φr
Pole at 0.27 < 59.42°
Zero at 1
Figure 5.41
(a) Simulated linearized transfer function using a one-block OSP. It is also compared with that of a conventional
MZ modulator. (b) Pole/zero and bias location for the linearized modulation. The zero, the pole, and the bias are
located in 1, 0.27 < -1.037 rad, and 2.2019 rad, respectively.
that the most linearized modulation occurs away from the pole location, implying that a
small resonance is required for linearization. The relatively small correction of the nonlin-
ear response from phase change is sufficient. Therefore, the optical loss issue is somewhat
mitigated in the linearized modulator. In order to demonstrate a linearized modulator,
we use the same experimental setup as shown in Figure 5.40 except that we utilize the
linear slope of the transfer function and we apply a triangular signal rather than a sinu-
soidal signal since a triangular signal allows easier verification of linearity. As shown in
Figure 5.42c, the output response linearly follows the input signal.
However, more rigorous methods should be applied to check the linearity of a modula-
tor such as two-tone test measurement method.44
Scope
CW laser
Polarization Polarizer
control t1 Bias Фm Bias tr Bias
Normalized intensity
0.8 −10
0.7 0.8
0.6
0.5 0.6
0.4
0.3 0.4
0.2
0.2
0.1
0 0
−20 −15 −10 −5 0 5 10 15 20 0 0.2 0.4 0.6 0.8 1
Applied voltage (V) Time (ms)
Figure 5.42
Linear signal generation using a one-block OSP and its experimental setup. (a) A ± 12 V peak-to-peak triangular
voltage input with a proper offset bias is applied to the fr electrode and the other biases are properly adjusted
by voltage supplies. (b) We obtain a proper transfer function for linear signal generation. (c) We utilize the linear
transition region in the transfer function to generate a linear output signal.
FWHM bandwidth of 0.2 GHz) is launched into the one-block OSP and then the voltage
applied to the electrode on top of the ring waveguides is set to 0 and Vp. Figure 5.43 shows
the intensities of the output pulses with different delay times of 123 ps and 33 ps, respec-
tively. Hence, 90 ps difference between maximum and minimum time delay is obtained.
The insertion loss is approximately 10 dB and the shape and the insertion loss of the output
pulses are almost the same regardless of the delay time.
Figure 5.44a shows the experimental setup for the TTD measurement using the one-
block OSP. In order to generate an input optical pulse for the OSP, we use a mode-lock
pulse laser (Clark-MXR Inc.). The optical pulse from the mode-lock laser is set to generate
0.15
φr = 0
φr = Vπ
0.05
0
−3000 −2000 −1000 0 1000 2000 3000
Time (ps)
Figure 5.43
Simulated intensities of the output optical pulses when a Gaussian optical pulse with 1.06 ns pulse width
(FWHM) is launched into the one-block OSP. The difference between maximum and minimum time delay of
90 ps is obtained by applying a different voltage to the fr electrode.
60 B C
B 40
RF Optical
Amp modulator 20
C PD &
Bias control 0
EDFA Scope 0 10 20 30 40 50
Time (ns)
Figure 5.44
TTD measurement experimental setup. (a) Experimental setup. (b) Photograph of laboratory setup. (c) Input
optical pulses corresponding to the points labeled in (a).
a short electrical pulse, which is shown and labeled as A in Figure 5.44a and b. We then
amplify the electrical pulse with an RF amplifier, which has a bandwidth from 1 MHz to
1 GHz. The amplified pulse is shown and labeled as B in Figure 5.44a and b after 28 dB
attenuation because the maximum peak voltage of the amplified pulse is about 2 V, which
is too high for detection with our sampling scope. We apply the amplified electrical pulse
signal as the driving input of an optical LiNbO3 amplitude modulator (Lucent) and obtain
a few nanosecond optical pulses at the modulator output, which is used for the input of the
OSP after optical amplification from an erbium-doped fiber amplifier (EDFA). The input
pulse for the OSP is also shown and labeled as C in Figure 5.44a and b.
By using multiple voltage supplies, a function generator and oscilloscope, we find a bias
condition which satisfies the APF condition. Once the APF condition is met, the intensity
response seen on the oscilloscope is constant as a function of voltage at the fr electrode.
Next, the output optical pulses from the OSP are measured by a sampling oscilloscope,
3
1
2.5
0.8
Normalized intensity
2
Volt (mV)
0.6
1.5
0.4
1
0.5 0.2
0 0
−3000 −2000 −1000 0 1000 2000 3000 −1000 −800 −600 −400 −200 0
Time (ps) Time (ps)
Figure 5.45
Output pulses from the TTD measurement setup with different voltages. (a) The output pulses detected at the
sampling oscilloscope and (b) their normalized and zoomed plots. The outputs have similar output shape and
insertion losses, and they are time-delayed each other. Maximum delayed amount measured is 135 ps.
triggered by the mode-lock laser. Figure 5.45a shows the output signals measured by the
sampling oscilloscope with three different voltages applied to the fr electrode. As seen
in Figure 5.45a, outputs have similar shapes and are time-delayed with respect to each
other. Figure 5.45b also shows a close-up of the normalized output signals. The maximum
delayed amount measured is 135 ps corresponding to a free space delay of 4 cm, which is
close to our simulation result.
Since Z -1 stands for one unit delay as we discussed before, this OSP generates pulse codes
with coefficients of a0, a1, a2, a3 and so forth. These coefficients or sequence numbers are
complex in general, having amplitudes and phases, which are interpreted as the amplitudes
and phases of the pulses. As in the case of analog applications, the individual parameters
0 0 0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Figure 5.46
Pulse codes are generated using the OSP. For example, four different pulse codes are shown using a two-block
OSP in this figure with four different pole phases and same zeros as in the pole/zero diagrams (stars for poles
and dots for zeros) from (a) to (d). Intensities of the generated pulses are normalized to the input pulse intensity.
In (a), the parameters are set to generate zeros and poles at (-1, 1) and (-0.5, -0.5) and the generated intensities
are (0.37, 0.37, 0.02, 0.09, 0.07, 0.03, 0.02, . . .).
of the OSP determine these coefficients (pulse sequence) all together and the complexity of
the generated pulse sequence depends on the number of OSP blocks.
Figure 5.46 shows a simulation result of the generated pulse codes from a two-block OSP.
In this simulation, four different pulse codes are generated with four different pole phases
while maintaining the zeros as in the pole/zero diagrams from Figure 5.46b through e.
Since a two-block OSP is assumed, two zeros and two poles are configurable. The indi-
vidual subfigures show the pole/zero diagrams denoting their locations in the complex
Z domain and also show the intensity of the generated pulses sequentially, which are
normalized to the input pulse intensity. In Figure 5.46b, the parameters are set to generate
zeros and poles at (-1,1) and (-0.5, -0.5) and the generated intensities are (0.37, 0.37, 0.02,
0.09, 0.07, 0.03, 0.02, …). In Figure 5.46c, the poles are changed to (-0.5j, -0.5j) while main-
taining the zeros and the generated intensities are (0.38, 0, 0.59, 0, 0.04,0, …) (likewise in
Figures 5.40 and 5.46d and e).
Since pulse code generators can be operated at GHz rates, it will be useful in generating
security codes of optical pulses and in general optical communications.
If x(n) and y(n) are used to denote an input and output discrete-time signal respectively,
the OSP behaves like an operator to transform x(n) into y(n). Consider a system where the
input and output satisfy an Nth-order linear constant coefficient differential equation of
the form:
N M
y(n) - ∑
k =1
bk y(n - k ) = ∑ a x(n - k)
k =0
k (5.41)
0.2
0.5
0.1
0 0
–0.1
–0.5
Impulse –0.2
0 1 2 3 4 5 6 7 8 response 0 2 4 6 8
Input pulse sequence, x(n) 0.4 function Output pulse sequence, y(n)
0.2
0
0 1 2 3 4 5 6 7 8
Figure 5.47
Equation 5.42 is computed in real time using the one-block OSP for the input sequence of x(n), which was gener-
ated in Figure 5.46b. The input sequence is (0.61, -0.61, -0.15, 0.31, -0.27, 0.19, -0.14, . . .) and the output sequence is
(0.15, 0.08, -0.04, -0.10, -0.07, 0, -0.04, 0.01, -0.01, . . . ). Pole/Zero diagram for the configured OSP and its impulse
response function are also shown.
This is similar to the equation introduced in Section 5.3.1.2. Here we replace the round trip
time with a symbolic sequence. Therefore, if an OSP is configured properly to have these
am and bn coefficients, the OSP can compute this differential equation to generate a real
time solution for output y(n) with given input x(n). For example, consider a system which
has a differential equation of:
1 1 1
y(n) - y(n - 1) = x(n) + x(n - 1) (5.42)
2 4 4
Since this system has one zero (one delayed) and one pole system (one recursive), only
one unit block is needed. By using the Z-transform, we also find out that this system has a
zero of -1 and a pole of 1/2. Figure 5.47 shows the computed sequence of y(n) for the input
sequence of x(n), which was generated in Figure 5.46b. For the input sequence of (0.61, -0.61,
-0.15, 0.31, -0.27, 0.19, -0.14, … ), the one-block OSP computes Equation 5.42 to generate an
output code as (0.15, 0.08, -0.04, -0.10, -0.07, 0, -0.04, 0.01, -0.01, … ) in real time. Note that
pulse codes are plotted in units of electric fields instead of the intensity since electric field
is the actual signal for the OSP. Because the electric field has phase in general, the pulse
codes are all complex numbers. In Figure 5.47, we set the parameters in such a way that the
pulses are all real numbers for simplicity.
The ability to compute and process optical signals at very high speeds will be truly use-
ful for optical packet communication networks and therefore are a very exciting prospect
for communication providers.
References
1. E. Marcatili. Bends in optical dielectric guides. The Bell System Technical Journal, 48, 2103–2132,
1969.
2. B. E. Little, S. T. Chu, W. Pan, and Y. Kokubun. An eight-channel add-drop filter using vertically
coupled microring resonators over a cross grid. IEEE Photonics Technology Letters, 11, 691–693,
1999.
3. B. E. Little, J. S. Foresi, G. Steinmeyer, E. R. Thoen, S. T. Chu, H. A. Haus, E. P. Ippen, L.
Kimerling, and W. Greene. Ultra-compact si-sio2 microring resonator optical channel dropping
filters. IEEE Photonics Technology Letters, 10, 529–551, 1998.
4. Q. Xu, B. Schmidt, S. Pradhan, and M. Lipson. Micrometre-scale silicon electro-optic modula-
tor. Nature, 435, 325–327, 2005.
5. K. D. Djordjev, S. J. Choi, S. J. Choi, and P. D. Dapkus. Active semiconductor microdisk device.
IEEE Journal of Lightwave Technology, 20(1), 105–113, 2002.
6. D. Rafizadeh, J. P. Zhang, S. C. Hagness, A. Taflove, K. A. Stair, S. Ho, and R. C. Tiberio.
Waveguide-coupled algaas/gaas microcavity ring and disk resonators with high finesse and
21.6-nm free spectral range. Optics Letters, 22(16), 1244–1226, 1997.
7. P. Rabiei, W. H. Steier, C. Zhang, and L. R. Dalton. Polymer micro-ring filters and modulators.
IEEE Journal of Lightwave Technology, 20(11), 1968–1975, 2002.
8. B. E. Little, S. T. Chu, H. A. Haus, J. Foresi, and J.-P. Laine. Microring resonator channel drop-
ping filters. IEEE Journal of Lightwave Technology, 15, 998–1005, 1997.
9. S. O. Transmission, group delay, and dispersion in single-ring optical resonators and add/drop
filters-a tutorial overview. IEEE Journal of Lightwave Technology, 22, 1380–1394, 2004.
10. R. C. Alferness. Waveguide electooptic modulators. IEEE Transactions on Microwave Theory and
Techniques, MTT-30, 1121–1137, 1982.
11. C. K. Madsen and G. Lenz. Optical all-pass filters for phase response design with applications
for dispersion compensation. IEEE Photonics Technology Letters, 10(7), 994–996, 1998.
12. C. K. Madsen. General iir optical lter design for wdm applications using all-pass filters. IEEE
Journal of Lightwave Technology, 18(6), 860–868, 2000.
13. K. Jinguji and M. Kawachi. Synthesis of coherent two-port lattice-form optical delay-line circuit.
IEEE Journal of Lightwave Technology, 13(1), 73–82, 1995.
14. K. Jinguji. Synthesis of coherent two-port optical delay-line circuit with ring waveguides. IEEE
Journal of Lightwave Technology, 14(8), 1882–1898, 1996.
15. N. Takato, T. Kominato, A. Sugita, K. Jinguji, H. Toba, and M. Kawachi. Silica-based integrated
optic mach-zehnder multi/ demultiplexer family with channel spacing of 0.01–250 nm. IEEE
Journal of Selected Areas in Communications, 8(6), 1120–1127, 1990.
16. G. Lenz, B. J. Eggleton, C. K. Madsen, and R. E. Slusher. Optical delay lines based on optical
filters. IEEE Journal of Quantum Electronics, 37(4), 525–532, 2001.
17. K. P. Jackson, S. A. Newton, B. Moslehi, M. Tur, C. C. Cutler, J. W. Goodman, and H. J. Shaw.
Optical fiber delay-line signal processing. IEEE Transactions on Microwave Theory and Techniques,
MTT-33(3), 193–210, 1985.
18. B.-J. Seo, S. Kim, B. Bortnik, H. Fetterman, D. Jin, and R. Dinu. Optical signal processor using
electro-optic polymer waveguides. IEEE Journal of Lightwave Technology, submitted.
19. V. S. Ilchenko, A. A. Savchenkov, A. B. Matsko, and L. Maleki. Sub-micro watt photonic micro-
wave receiver. IEEE Photonics Technology Letters, 14, 1602–1604, 2002.
20. D. A. Cohen, M. Hossein-Zadeh, and A. F. J. Levi. High-q microphotonic electro-optic modula-
tor. Solid-State Electronics, 45, 1577–1589, 2001.
21. V. S. Ilchenko, A. A. Savchenkov, A. B. Matsko, and L. Maleki. Whispering-gallery-mode
electro-optic modulator and photonic microwave receiver. Journal of the Optical Society of America
B, 20, 322–342, 2003.
22. M. Hossein-Zadeh and A. Levi. A new electrode design for microdisk electro-optic rf modula-
tor. The Conference on Lasers and Electro-Optics, proceedings are published by IEEE, 863–865, 2003.
23. O. P. Lay, S. Dubovitsky, R. D. Peters, J. P. Burger, S.-W. Ahn, W. H. Steier, H. R. Fetterman,
and Y. Chang. Mstar: a submicrometer absolute metrology system. Optics Letters, 28, 890–892,
2003.
24. I.-L. Gheorma and R. M. Osgood Jr. Fundamental limitations of optical resonator based on
high-speed eo modulators. IEEE Photonics Technology Letters, 14, 795–797, 2002.
25. T. Saitoh, S. Mattori, S. Kinugawa, K. Miyagi, A. Taniguchi, M. Kourogi, and M. Ohtsu.
Modulation characteristic of waveguide-type optical frequency comb generator. IEEE Journal of
Lightwave Technology, 16, 824–832, 1998.
26. M. Kato, K. Fujiura, and T. Kurihara. Generation of super-stable 40 ghz pulses from fabry-perot
resonator integration with electro-optic phase modulator. Electronics Letters, 40, 299–301, 2004.
27. T. Sadagopan, S. J. Choi, P. D. Dapkus, and A. E. Bond. Optical modulators based on deple-
tion width translation in semiconductor microdisk resonators. IEEE Photonics Technology Letters,
17(3), 567–569, 2005.
28. K. Sasayama, M. Okuno, and K. Habara. Coherent optical transversal filter using silica-based
waveguides for high-speed signal processing. IEEE Journal of Lightwave Technology, 9(10), 1225–
1230, 1991.
29. A. V. Oppenheim and R. W. Schafer. Discrete-Time Signal Processing. Prentice Hall, New Jersey,
1989.
30. C. J. Kaalund and G.-D. Peng. Pole-zero diagram approach to the design of ring resonator-based
filters for photonic applications. IEEE Journal of Lightwave Technology, 22(6), 1548–1559, 2004.
31. Yariv. Critical coupling and its control in optical waveguide-ring resonator systems. IEEE
Photonics Technology Letters, 14(4), 483–485, 2002.
32. H. Tazawa, Y.-H. Kuo, I. Dunayevskiy, J. Luo, A. K.-Y. Jen, H. R. Fetterman, and W. H. Steier.
Ring resonator-based electrooptic polymer traveling-wave modulator. IEEE Journal of Lightwave
Technology, 24(9), 3514–3519, 2006.
33. K. Geary, S.-K. Kim, B.-J. Seo, and H. R. Fetterman. Mach-zehnder modulator arm length mis-
match measurement technique. IEEE Journal of Lightwave Technology, 23, 1273–1277, 2005.
34. M. R. Fetterman and H. R. Fetterman. Optical device design with arbitrary output intensity as
a function of input voltage. IEEE Photonics Technology Letters, 17(1), 97–99, 2005.
35. I. Y. Poberezhskiy, B. Bortnik, S.-K. Kim, and H. R. Fetterman. Electro-optic polymer frequency
shifter activated by input optical pulses. Optics Letters, 28(17), 1570–1572, 2003.
36. T. R. Halemane and S. K. Korotky. Distortion characteristics of optical directional coupler mod-
ulators. IEEE Transactions on Microwave Theory and Techniques, 38(5), 669–673, 1990.
37. L. M. Johnson and H. V. Roussell. Reduction of intermodulation distortion in interferometric
optical modulators. Optics Letters, 13(10), 928–930, 1998.
38. D. J. M. Sabido, M. Tabara, T. K. Fong, C.-L. Lu, and L. G. Kazovsky. Improving the dynamic
range of a coherent am analog optical link using a cascaded linearized modulator. IEEE Photonics
Technology Letters, 7(7), 813–815, 1995.
39. J. L. Brooks, G. S. Maurer, and R. A. Becker. Implementation and evaluation of a dual parallel
linearization system for am-scm video transmission. IEEE Journal of Lightwave Technology, 11(1),
34–41, 1993.
40. R. B. Childs and V. A. O’Byrne. Multichannel am video transmission using a high-power nd:yag
laser and linearized external modulator. IEEE Journal on Selected Areas in Communications, 8,
1369–1376, 1990.
41. Y.-C. Hung and H. R. Fetterman. Polymer-based directional coupler modulator with high
linearity, IEEE Photonics Technology Letters, 17, no. 12, pp. 2565–2567, Dec 2005.
42. X. Xie, J. Khurgin, J. Kang, and F.-S. Chow. Linearized mach–zehnder intensity modulator. IEEE
Photonics Technology Letters, 15(4), 531–533, 2003.
43. J. Yang, F. Wang, X. Jiang, H. Qu, M. Wang, and Y. Wang. Influence of loss on linearity of
microring-assisted mach–zehnder modulator, Optics Express, 12, no. 18, pp. 4178–4188, Sep 2004.
44. P.-L. Liu, B. J. Li, and Y. S. Trisno. In search of a linear electrooptic amplitude modulator. IEEE
Photonics Technology Letters, 3(2), 144–146, 1991.
45. H. Zmuda, R. A. Soref, P. Payson, S. Johns, and E. N. Toughlian. Photonic beamformer for phased
array antennas using a fiber grating prism. IEEE Photonics Technology Letters, 9(2), 241–243,
1997.
46. J. L. Corral, J. Marti, S. Regidor, J. M. Fuster, R. Laming, and M. J. Cole. Continuously variable
true time-delay optical feeder for phased-array antenna employing chirped fiber gratings. IEEE
Transactions on Microwave Theory and Techniques, 45(8), 1531–1536, 1997.
47. T. Kawanishi, S. Oikawa, K. Higuma, and M. Izutsu. Electrically tunable delay line using an
optical single-side-band modulator. IEEE Photonics Technology Letters, 14(10), 1454–1456, 2002.
48. N. A. Riza, M. A. Arain, and S. A. Khan. Hybrid analog–digital variable fiber-optic delay line.
IEEE Journal of Lightwave Technology, 22(2), 619–624, 2004.
49. B.-J. Seo and H. R. Fetterman. True time delay element in lossy environment using eo wave-
guides. IEEE Photonics Technology Letters, 18(1), 10–12, 2006.
Melanie Lebental
Ecole Normale Superieure of Cachan
Université Paris-Sud
Eugene Bogomolny
Université Paris-Sud
Joseph Zyss
Ecole Normale Superieure of Cachan
Contents
6.1 Introduction......................................................................................................................... 318
6.2 Introduction to Flat Organic Micro-Cavities.................................................................. 318
6.2.1 Context..................................................................................................................... 318
6.2.2 Basic Elements......................................................................................................... 320
6.3 Polymer-Based Technology and Process......................................................................... 324
6.3.1 Materials................................................................................................................... 325
6.3.2 Etching Methods..................................................................................................... 326
6.4 Optical Tests........................................................................................................................ 327
6.4.1 Background.............................................................................................................. 327
6.4.2 Various Set-Up......................................................................................................... 328
6.4.3 Spectra...................................................................................................................... 329
6.5 Theoretical Approaches..................................................................................................... 333
6.5.1 General Methodology............................................................................................334
6.5.2 Spectrum.................................................................................................................. 335
6.5.3 Directions of Emission...........................................................................................340
6.5.4 Light Patterns Inside the Cavities........................................................................344
6.5.4.1 Benefit of “Scarring”................................................................................344
6.5.4.2 Perturbation Approach...........................................................................346
6.6 Conclusion...........................................................................................................................348
Acknowledgments....................................................................................................................... 349
Appendix A: Lyapounov Coefficient for Unstable Periodic Orbits................................... 349
References...................................................................................................................................... 350
317
© 2009 by Taylor & Francis Group, LLC
6.1 Introduction
This chapter is mainly devoted to the basic concepts that underlie the dynamic features
of light confined in quasi two-dimensional micro-cavities. Moreover, in accordance with
the general orientation of this book, there is special emphasis on practical applications,
or the onset of such in this relatively new and emerging field at the border of science and
technology. The implementation of physical phenomena in organic materials is justified
by the appealing assets of this broad class of materials, both in terms of their functional
versatility, that lends them to a molecular engineering approach and the availability and
maturity of a relatively easy enabling technology. Nevertheless, the foundations as well as
major applications of organic materials outreach any specific material technology, similar
possibilities in glass and semiconductor materials bear specific pros and cons. In the first
section, we will review the main motivation which have lead us to select organic materials
and the candidate technology in the context of solid state micro-cavity lasers. Their struc-
tural and functional flexibility, compliance to etching, abiding to relatively easy and
unconstrained design rules and basic technological fabrication steps will be considered
in the second part. In the final sections, different experimental lay-outs and methods to
extract relevant information will be considered and subsequently challenged at different
levels for specific generic cavity shapes, from nonchaotic and integrable to chaotic.
diodes (OLEDs) with applications ranging from display to biomedicine [10]. The success of
this approach derives from a sequence of optimization steps starting from the tailoring of
upstream material properties all the way to the fabrication of confining devices allowing
for the concentration and steering of light in structures of adequate shape.
This area of investigation has been active for more than three decades, experiencing peri-
odic revivals each time some significant progress has emerged in physical concepts, mate-
rial engineering methods or enabling technologies. In that respect, silica has always been a
starting material of choice, hard to compete with in view of its excellent optical quality, but
depending on rather demanding and rigid processing conditions at higher flame tempera-
tures which have limited its application in integrated optics. Nevertheless, optical micro-
spheres have been investigated for their record high figures of merit and ability to sustain
long lived whispering gallery modes (WGM) with high confinement features along micro-
tore volumes close to the equator [11,12]. Such configurations, while offering the advan-
tage of high-Q factors, are inherently deprived of the planarity that is demanded for easy
coupling and connecting with usual optical integrated inputs and outputs. Moreover, the
rather rigid structure of silica makes it difficult to vary at will its composition and the outer
shape of micro-spheres, which calls for other types of materials and technologies, provid-
ing more flexibility both in terms of internal composition and boundary definition.
It was therefore natural to consider the possibilities of polymers, and more generally
of organic materials, to further investigate and exploit the properties of WGMs in a more
accessible and flexible two-dimensional format, i.e. a thickness comparable to the wave-
length (~1 µm) and planar dimensions about one or two orders of magnitude higher. Indeed
earlier technological processes have evidenced a quasi-unlimited freedom of design by
tailoring the outer boundary of a limited film portion by adequate masking techniques
(see Section 6.3.2).
Such planar cavity structures can also be achieved with inorganic semiconductors how-
ever with less ease of fabrication and a significantly higher refractive index (typically from
2.5 to 3.5 in contrast with 1.5 for usual organic materials). As the refractive index gov-
erns the total internal reflection angle, the higher its value, the more confining the cav-
ity will be. Referring to a ray picture in the limits of geometric optics (e.g. a cavity size
much larger than the wavelength, typically hundred times more), the number of confined
modes abiding to the total internal reflection condition will then be significantly smaller
for lower index materials. This leads to “cleaner” spectral features for the out-coupled
emission, making it amenable to simpler physical interpretation. Moreover, this phenom-
enon implies direct practical applications, for instance if light must be emitted in a planar
geometry along a preferential direction. It is therefore of primary interest to distort the
circular boundary of a micro-disk onto a more adequate shape so as to favor such a desired
emission pattern. But this distortion is sometimes not sufficient to achieve the directional
emission since a large amount of modes can coexist in the cavity. Then the low refractive
index brings an additional way to select the required modes [13,14].
As specified in the title, we are working on micro-lasers and not on passive resonators.
However the laws governing an active cavity in contrast to a passive one are still under-
studied, the materials being either organic or inorganic. In fact the standard principles
of laser can be hardly extended to these open resonators due to their small size and their
relatively low quality factor which usually implies strong spatial hole-burning and no
orthogonality between resonances [15]. The successful analysis with fundamental equa-
tions coupling field, polarization and inversion density shows that the active modes are not
so different from that of passive cavities [16,17]. Only passive properties are thus involved*
while, from an experimental point of view, the use of dye-doped polymers reveal a double
asset. First it opens the way to well-behaved micro-lasers [18] with potential applications
as light emitters or sensors; and secondly the laser effect appears as an ideal probe located
in the resonator itself, thereby allowing to select (or not) the modes to be excited by an effi-
cient and relatively straightforward optical pumping scheme.
Our investigations aim at answering a number of fundamental questions, some of which
are driven by practical considerations. Among the central issues that we wish to address
are the behaviors of a so-called chaotic cavity (for example stadium or cut disk) whereby
periodic (closed) orbits are an exception somehow buried within a sea of dominant chaotic
trajectories where rays experience erratic bouncing [19]. Indeed at the onset of the research
reported here, little was known about the spectral and angular emission properties of such
cavities, where the erratic behavior of rays could be expected to lead to featureless spectral
and angular properties. We will show that, in general, this is not the case. Moreover, these
results can be accounted for by calling on the more elaborate wave chaos tool-box [19] and
the design of the cavity shape can then be optimized for applications where specific prop-
erties of the spectrum or the directional emission are involved.
In summary, we will proceed to show that polymer-based micro-cavities are appealing
structures in the investigation of a variety of both technologically and physically mean-
ingful issues due to their external boundary shapes. We will confront experimental results
based on the emissive properties of a series of cavities with generic shapes so as to test vari-
ous theoretical modeling approaches ranging from fully numerical (solution of Helmholtz
equations) to more analytical methods. A most interesting outcome of this work, is the
definition of a general methodology by which the nature of the underlying orbits can be
inferred from the Fourier transform of the emission spectrum, which is applicable to a
broad variety of shapes (chaotic or not) and class of materials.
* For specific properties where the influence of lasing has been proven, comments will specify how to take it
into account.
† We restrict our study to passive cavities, the influence of lasing being then deduced from general passive
Boundary
h
Inner
neff Equation 6.4
x y
Outer
Buffer layer neff = 1
Figure 6.1
(a) Schematic view of a cylindrical cavity with h thickness. (b) Schematic view of the system after reduction to
a two-dimensional problem.
this frame, we assume that the electromagnetic fields, far away from the cavity, behave like
outgoing waves with an eikr propagative phase factor, without any ingoing waves and the
related e-ikr phase factor. The difficulty of dealing with this problem in general comes from
its three-dimensional and vectorial nature. To our knowledge, the only case where a quasi
analytical solution could be found is that of an infinite cylinder with a circular basis [22].
To overcome this difficulty, the electromagnetic fields are commonly separated into two
different polarizations: the electric (resp. magnetic) field lies in the (x,y) plane, while the
magnetic (resp. electric) field has a nonzero component in the z direction, referred to as TE
(resp. TM) polarization‡. This separation is based on a Fourier transform and assumes that
the Hz and Ez components are uncoupled. This is therefore exact for a cavity infinite either
in the z or in the xy directions. But this assumption fails due to the boundaries where these
components are coupled by Maxwell equations (see e.g. Dubertrand et al. [23]). Far from
the boundaries, in the cavity bulk, the separation into TE and TM polarizations seems
usually a good approximation. This condition is more or less verified for our micro-lasers,
since the h thickness is less than 1 μm for a typical size of 100 μm in the xy plane. It will
thus be used in this chapter as in most papers dedicated to this subject (see e.g. [24,25]).
The problem then reduces to the scalar Helmholtz equation for the Ez or Bz component
(called Ψ hereafter), the other one being directly inferred from Maxwell equations:
( ∆ x + k 2 n2 ( x))Ψ ( x) = 0 (6.1)
where k = ω/c is the vacuum wave-number and n( x) the local refractive index. The Ψ wave-
function depends on the three space variables and, due to the assumption of infinite layers
in the xy plane, it can be expressed as Ψ( x , y , z) = φ( z)ψ ( x , y ). This separation of the xy and
z variables is called effective refractive index approximation and leads to
( ∆ xy + k 2 neff
2
)ψ ( x , y ) = 0 (6.2)
‡ This definition is consistent in this chapter. In the literature, these terms are sometimes interchanged.
l=0
n1 1.50
nk l=1
n q p
neff
l=2
n2 z 1.46
x,y l=3
l=4
1.42
0 1 2 3 4
h/λ
Figure 6.2
(a) Cross-section of the layers and notations for refractive indexes and propagation wavenumbers. The propaga-
tive layer is a polymer (PMMA) doped with a laser dye (DCM) (n = 1.54) between confinement layers assumed
to be infinite in the vertical direction: the air (n1 = 1) and the buffer layer (another polymer (SOG) (n2 = 1.42) or
silica (n2 = 1.45)). (b) Effective refractive index versus the thickness over wavelength variable calculated from
Equation 6.4. The refractive indices are assumed to be constant: 1 for air, 1.42 for the buffer layer, and 1.54 for the
cavity layer (horizontal lines). The TE polarization is plotted with solid lines and TM polarization with dotted
lines. Integer l (see Equation 6.4) increases from left to right starting from l = 0.
inferred from φ(z). In fact, from Equation 6.1, it follows that φ( z) ∝ e iqz in the cavity layer,
and so k 2 neff
2
= k 2 n2 - q 2. The q value is determined from the phase coherence condition
after one round trip (length 2h) and one reflection on both interfaces (cavity layer/air and
cavity layer/buffer layer):
where r1 (resp. r2) is the reflection coefficient at the interface between air (resp. the buffer
layer) and the cavity layer. It reduces to a pure dephasing due to the assumption of an
evanescent wave in the vertical direction outside the cavity layer (total internal reflection
at both interfaces). From Equation 6.3 the expression of q follows and then the effective
refractive index (here, for instance, for TM polarization):
neff
2
- n12 neff
2
- n22
kh n2 - neff
2
= arctan 2 2
+ arc
c tan 2 2
+ lπ l ∈ N (6.4)
n - neff n - neff
where n1 and n2 stand for the refractive indexes of the adjacent layers. The refractive indexes
are inferred from ellipsometric measurements and the plots of Figure 6.2b are directly cal-
culated from Equation 6.4. This graph shows that under a certain thickness cut-off (about
0.3 μm) no mode propagates in the cavity, and that between this inferior cut-off and the
next threshold (typically h~λ) there exists only one vertical excitation for each polarization.
Our cavities are designed according to this latter configuration, in general close to the
upper boundary (h/λ~1) to maximize the volume and then the gain. This refractive index
approximation is not specific to organic material and is commonly used also for inorganic
devices.
At this latter step, the problem reduces to the scalar and two-dimensional Helmholtz
equation (Equation 6.2) and the boundary shape is defined from the cylinder section (see
Figure 6.1b). Then the neff effective refractive index outside the cavity is neff = 1 and is calcu-
lated from Equation 6.4 inside the cavity. The boundary shape has a prominent influence
on spectra and resonance patterns while the effective refractive index approximation fails
close to the boundary. This paradoxical approach could be only decided with a full three-
dimensional treatment. For in-plane scale much larger than the wave-length (typically 100
times larger for our cavities), effective refractive index approximation has proved to be
relevant. For consistence, the boundary conditions are inferred from Maxwell equations
of the three-dimensional system (see Dubertrand et al. [23]):
ψin = ψout (6.5)
∂ ψ in ∂ ψ out
= for TM polarization (6.6)
∂ν ∂ν
1 ∂ ψ in ∂ ψ out
= for TE polarization (6.7)
n2 ∂ ν ∂ν
where the label in (resp. out) stands for inside (resp. outside) and ν is the coordinate normal
to the one-dimensional boundary.
This problem is significantly simpler than the original one, since it reduces to a two-
dimensional one. It is however still hardly solvable due to two separate difficulties: the
diffraction at the corners and the boundary shape.
Actually the diffraction coefficient of a plane wave incident on a dielectric corner is not
yet known, but it was analytically calculated for a metallic one by Sommerfeld at the end of
the nineteenth century [26]. As a consequence, even a system as simple as a square shaped
dielectric cavity currently remains an unsolved challenge. This problem is directly con-
nected to more general ones such as the propagation of electromagnetic waves in urban
environments, where some solutions derived from micro-cavity studies could also prove
useful.
To explain the additional difficulty due to the boundary shape, we resort to geometrical
optics, i.e. when the wavelength is much smaller than the typical size, and therefore rays
follow the trajectories of point-like particles. In fact the dynamics of rays depends strongly
on the boundary shape. It can be integrable for instance inside a square or a disk, chaotic
inside a stadium or a cut disk (see Figure 6.4), mixed, pseudo-integrable, etc.§ [19]. The
route from a geometrical optics system to the equivalent wave optics one (usually referred
to as quantization) is straightforward for an integrable system because, by definition, there
exists as many constants of motion as degrees of freedom.¶ For instance in a circle, the
constants of motion are the energy and the angular momentum (or the incident angle). But
for a chaotic system, where, by definition, there are less constants of motion than degrees
of freedom, the standard quantization method fails and other ways must be explored. This
is the broad domain of research called wave chaos.** Earlier works were based mainly on
metallized micro-cavities in the millimeter wave regime, whereby electromagnetic modes
are fully confined within the metallic boundaries, as a result of strict cancelation of the
field outside of the cavity and therefore referred to as “closed system” as in the following.
§ This is true for closed cavities such as metallic ones, from which light cannot escape, while for dielectric ones,
refraction selects some rays and modifies the dynamics.
¶ Moreover the constants of motion must commute with each other [27].
** In general it is called “quantum chaos” in reference to the original studies on quantum systems. But the con-
cepts and results easily extend to any physics domain where waves are concerned.
In our case, a dielectric interface does not impose cancelation of the field out of the cav-
ity and allows indeed, not only for refractive out-coupling but also for evanescent wave
generation, even in the case where total internal reflection conditions are fully satisfied.
Connection of the inside and outside parts of the cavities provides new physical proper-
ties that cannot occur in strictly confining cavities and that is further enhanced by the
use of low refractive index materials such as polymers. As the topic of this book is “prac-
tical applications”, we will not emphasize this aspect and restrict our chapter to poten-
tially applicable results derived from these studies, in particular in Section 6.5. However, a
wealth of literature has been devoted to this field [19,20,27].
rganic-inorganic blending where the polymer host provides an ideal scaffold for a diver-
o
sity of guests and properties (such as would allow for simultaneous lasing and electro-op-
tic properties) which cannot be readily achieved in purely crystalline inorganic materials,
opens the way to multi-functional structures.
Finally, keeping in mind the perspective of life science applications and the need for a
new generation of bio-chips with sensing and diagnostic potential, there is clearly a need
for a patterned and eventually active organic or polymer intermediate functional layer,
capable of connecting a biological recognition layer (e.g. targeted to specific cells, tissues or
macromolecular assemblies of proteins) with an inorganic substrate (e.g. a silicon chip to
drive or retrieve and process input or output electronic signals, such as based on an array
of field effect transistors). A physically functional polymer layer bears more potential than
a purely inorganic substrate to allow for additional biological functions, for example by
addition of a biopolymer over-layer, or by featuring intrinsic biopolymer character, so as
to provide adequate and sensitive linkage between the biological and physical parts of a
biosensor [30].
In the context of this chapter, we illustrate the implementation of such advantages in
a more modest, however highly rewarding context, that of patterning flexibility, achiev-
ing with simple means over the same substrate, a quasi-unlimited variety of micro-cavity
contours. The advantage of ease and speed of fabrication, by way of easily accessible tech-
nological tools, has been used to explore a wealth of cavity shapes over a broad range
of dimensions, which would hardly have been possible under the more rigid conditions
imposed by inorganic semiconductor fabrication. We have been mainly using, as will be
detailed in the following sections, relatively basic equipment for micro-photonics device
fabrication processes, such as micro-photo-lithography, reactive ion etching or metal evap-
oration, excluding the use of ultra-vacuum deposition techniques. When higher quality
factors than those achieved by such well established and “elementary” techniques are
needed, direct patterning of polymers by electron or focused ion beam is feasible and pro-
vides the required level of resolution, down to the nano-scale.
6.3.1 Materials
One of the main advantages of organic compounds is the broad diversity of materials with
very specific properties. Various arrangements lead to efficient organic micro-lasers, but
each one must fulfill the following requirements.
• The polymer matrix, which makes up the cavity, must be as transparent as possible
in the pumping and emission ranges of the active material to avoid additional
losses. More specifically, the choice depends on the etching technique (e.g. UV
sensitivity) and on the active material (e.g. solvent compatibility and glass transi-
tion temperature). In our case, polymethyl methacrylate (PMMA) is well-adapted
and some studies even report that it makes the photo-degradation of the laser dye
reversible [31,32]. It may be responsible for the very long sample life-time, up to
several years of storage at regular room conditions.
• The active material can be organic or inorganic [29]. Our micro-lasers are based on
[2-[2-[4-(dimethylamino)phenyl]ethenyl]-6-methyl-4H-pyran-4-ylidene]-propane-
dinitrile (DCM) but other laser dyes can also be used: nile-red, rhodamine [33], etc.
The typical doping amounts to a few percents in weight and cannot be increased
by much due to the dye emission quenching.
• A buffer layer is usually needed to ensure the vertical confinement since the refrac-
tive index of the gain layer (n ~ 1.5) is in general lower than that of the wafer (n = 3.9
for silicon at 600 nm). Two microns of silica (n = 1.45) or Spin-On Glass (n = 1.42)
fulfill this purpose perfectly well.
• The wafer can be either opaque (silicon) or transparent (glass) depending on the
application. Micro-lasers on flexible wafers have also been realized [34].
These requirements allow for a broad choice of emission wavelengths and optical proper-
ties (e.g. addition of nonlinear materials [3]).
6.3.2 Etching Methods
Thin film deposition by spin-coating is one of the great advantages of organic mate-
rials as opposed to their inorganic counterparts which generally require much more
elaborate high vacuum techniques. In fact this method to make a layer is cheap, fast,
easy, reproducible and allows for a fine control of the thickness (typical uniformity ±10
nm). Once the necessary layers have been laid, standard photolithography and reactive
ion etching (RIE) steps can be performed as described in Figure 6.3. The use of a com-
mercial Cr/quartz mask ensures a good etching quality (roughness scale less than 500
nm) and a broad diversity of shapes (see Figure 6.4). The simplicity of the device (the
single PMMA-DCM layer) reduces the processing time: in 2 hours, several hundred of
micro-cavities of different sizes and shapes can be simultaneously made on a standard
wafer.
Usually we use the photolithography and RIE process, but different etching processes
are available, each one with various grading in etching and ease-of-use:
Photo−resist A cavity
Mask
PMMA−DCM
Photolithography RIE
Silica
Silicon
Figure 6.3
Cross-sectional view for the “photolithography+RIE process” steps, following the spin-coating of the PMMA-
DCM and photoresist layers.
Figure 6.4
Top view photographs of micro-cavities by an optical microscope. From left to right: stadium (a rectangle between
two half-circles), cut disk, mushroom, and cardioid. Their typical size is of the order of 100 µm with a thickness
(direction perpendicular to the sheet) less than 1 µm.
Figure 6.5
Scanning electron microscope images of PMMA cavities etched with an electron beam (Courtesy of C. Ulysse,
LPN, France). The specific scale is shown in the bottom right corner of each picture.
The fabrication of standard organic micro-lasers only requires a relatively cheap technol-
ogy, such as available in standard clean room conditions. Furthermore, the use of organic
materials gives access to a broad variety of optical properties.
†† Each wafer contains a few hundred cavities. The cavity to be pumped is selected with a microscope.
L1
M Pump
laser
L2
a cavity
Detector
Rotating mount
Figure 6.6
Schematic view of a standard experimental set-up. M stands for mirror and L for lens.
spectrometer) via a lens (L2). Concerning the pumping, the size of the beam cross-sectional
area can be adjusted with a lens (L1) and its shape eventually controlled to select specific
lasing modes of the cavity [38]. In the following, we will consider uniform pumping only,
in a so-called “flooded” geometry.
This set-up is similar to optical pumping experiments with inorganic micro-lasers. The
only difference is the constraint on the pulse duration of the pump laser. In fact laser dyes
show dark states, referred to as triplet states, with typical transition times from the excited
state of about 100 ns. The pump pulse duration must thus be less than a few nano-seconds
[39,40] (30 ps in our case).
The laser threshold depends on the cavity shape and the organic compounds. For a
typical stadium made of 5% DCM doped PMMA, it amounts to about 100 pJ/µm2 [18] and
can be decreased using a better etching process and an integrable shape such as square or
disk because of the better confinement of the modes therein. Besides, as the available gain
decreases with the size of the cavity, the minimal size to get a laser effect is imposed by
the damage threshold of the layer.
One of the advantages (or disadvantages, depending on the point of view) of laser dyes is
their homogeneous gain broadening. We will take advantage of it in Section 6.4. Concerning
the lasing dynamics, it induces a competition between modes, probably controlled by the
fluorescence statistics, and then implies variability from pulse to pulse: peaks stay at the
same position, but their relative intensities change. This variation disappears after averag-
ing over a few dozen pulses. In the following, we will be considering spectra after integra-
tion over 30 pump pulses.
• Spectra and direction of emission: these properties can be studied at the same time
with the standard set-up described in Figure 6.6. In this configuration, the detector
is a spectrometer coupled to a cooled CCD camera with a rotating mount to turn the
cavity in front of it. The intensity emitted in each direction is then deduced from
integration over all the pixels of the spectrum. The resolution depends mainly on
the angular aperture of the detection system. In Figure 6.6, it is connected to the
size of the lens and the distance to the sample. In our experiments, it is about ±5°
and can be easily decreased.
• Lateral far-field imaging. The side of the cavity can be imaged, changing the detector
system of Figure 6.6 into a microscope objective and a cooled CCD camera. Figure 6.7
shows two examples of stadium images collected for different positions of the cavity
(a) (b)
Figure 6.7
Images of stadium sides collected in the far-field (a) in the direction of the small axis, (b) in the direction of
maximal emission. The area of highest intensity are indicated in white. The maximal intensity in (b) is about 15
times higher than in (a).
with respect to the detector. This set-up provides additional information about the
distribution of light around the cavity [41]. But emission mechanisms at the bound-
ary can not be directly inferred because of the far-field collection: near-field imaging
is then required.
• Top far-field imaging. At low pump intensity, it is possible to look trough a micro-
scope and observe directly the cavity from the top when it is emitting. The use of a
CCD camera provides quantitative information about vertical emission [42,43]. But
this set-up is not adapted to the study of low loss resonances, since these modes
are confined in the plane of the cavity and thus present an evanescent amplitude
profile in the vertical direction.
• Near-field imaging is the most suitable approach to test the spatial pattern of low
loss modes. A few reports are devoted to this approach in the case of quasi two-
dimensional cavities [44–46]. But these studies have been further developed in
micro-spheres (see e.g. [47]) and in photonic-crystal micro-cavities (see e.g. [48]).
• To track a wave-packet, two different methods have been demonstrated. The first
one [49,50] is based on the near-field optical set-up described elsewhere [44,45]. The
time resolution is obtained by detecting the femto-second pulse injected inside the
resonator and using interferences with a referent beam. The second method [51]
takes advantage of the nonlinear properties of the laser dye. The femto-second
pulse injected inside the cavity acts as a pump beam while a delayed probe beam
lights the cavity uniformly. The wave-packet is then tracked via the two-photon
triggered fluorescence. In both cases, the cavity is used as a passive resonator.
6.4.3 Spectra
The spectrum of an organic micro-laser (or a standard laser) is made of two main elements:
its envelope and the peaks. The former one results from the gain property of the mate-
rial. For instance, Figure 6.8a shows the fluorescence and amplified spontaneous emission
(ASE) spectra of a PMMA-DCM layer before etching. The width of the fluorescence spec-
trum is of the order 50–70 nm, which is typical for a laser dye, while the ASE spectrum is
narrower and shifted to the red, as a result of competition between gain and absorption
losses [52]. Besides, the peak properties are imposed by the resonator itself, both for their
width and position, at least in a pumping range close to the laser threshold‡‡. And their
amplitudes depend on the envelope as shown in Figure 6.8b.
‡‡ The peaks could become narrower due to lasing and be shifted by frequency pulling but we have not observed
so far such a phenomenon in our experiments. To our knowledge, there does not exist any report on mode-
locking or, more generally, interaction between modes in polymer micro-lasers, but these nonlinear effects
could occur.
0.8
Normalized intensity
Normalized intensity
Fluorescence
ASE
0.6 0.5
0.4
0.2
0.0
Figure 6.8
(a) Spectra collected with the set-up of Figure 6.6. Continuous line: fluorescence spectrum of a PMMA-DCM
layer on a glass substrate excited with a continuous pumping. The refractive index of the glass is higher than
the refractive index of the active layer, preventing any propagation in the latter one. Dotted line: ASE spectrum
of a PMMA-DCM layer on SOG on Si layer excited with a pulse pumping. (b) Laser spectra of the same cavity
exposed to a pulse pump energy of 8 a.u. (in continuous) and 20 a.u. (in dotted).
(a) (b)
4000 1.0
0.8
Fourier transform (a.u.)
3000
Intensity (counts)
0.6
2000
0.4
1000
0.2
0 0.0
605 610 615 620 625 0.0 0.5 1.0 1.5 2.0×106
Wavelength (nm) Optical length (nm)
Figure 6.9
(a) Spectrum of a Fabry–Perot micro-laser of width w = 150µm. (b) Normalized Fourier transform of the spec-
trum (a) expressed as intensity versus wavenumber. The arrow indicates the “first peak”.
The positions of the peaks hold a wealth of information about the light pattern inside the
cavity. In fact they are connected to geometric features of periodic orbits. For instance, let
us consider a Fabry–Perot resonator§§ of w width (distance between mirrors). The experi-
mental spectrum in Figure 6.9a is made of almost regularly spaced peaks. The correspond-
ing wavenumbers must verify the phase coherence condition along the single periodic
orbit of L = 2w length
r2 eiLkn eff = 1
where r stands for the reflection coefficient and neff for the effective refractive index (see
Section 6.3). The solutions of this equation are complex numbers: the imaginary part
§§ The corresponding polymer micro-cavity is a very long stripe [53].
800 Experiments
Linear fit
Slope 1.64 × 2w
700 Slope 1.50 × 2w
500
400
300
Figure 6.10
Optical length versus Fabry–Perot width w. The experiments (circle points) are linearly fitted by the solid line.
The dashed line corresponds to the theoretical prediction for a refractive index n = 1.64 without any adjusted
parameter. For comparison, the dark dotted line is similar to the dashed line, but for a refractive index n = 1.50,
i.e. the effective refractive index in the absence of group velocity corrections.
corresponds to the width of the resonance and the real part (called km) gives the position of
a peak in the spectrum and verifies
Some groups use the resonance widths, for instance towards sensor applications [54],
while we focus here on their positions through Equation 6.8 as a way to identify the lasing
periodic orbits via their length L. From Equation 6.8, the peaks are regularly spaced by a
wave-number interval expressed as
2π
δ k = km+1 - km = (6.9)
Lneff
This relation suggests that the periodic orbit length can be directly inferred from an experi-
mental spectrum, by way of measuring the distance between adjacent peaks¶¶. Nevertheless
the use of Equation 6.9 for organic micro-lasers leads to a systematic deviation from the
expected length, as illustrated in Figure 6.10. This graph shows the optical length versus
the width w of Fabry–Perot resonators. The circle points correspond to the quantity 2π/δk
inferred from the experimental spectra*** and are linearly fitted by the solid line. For com-
parison the straight line 2wn has been plotted with a dotted line with n = 1.50 standing for
the effective refractive index, showing some disagreement with the error bars†††.
¶¶ The effective refractive index neff is inferred independently from the cavity thickness and bulk refractive
indexes (see Section 6.3).
*** In fact this quantity can be directly measured via the Fourier transform of the experimental spectrum when
expressed as intensity versus wavenumber. The optical length corresponds to the position of the first peak
which is indicated by an arrow in Figure 6.9b.
††† In our experiments, the measure of the optical length is limited by the precision of the discrete Fourier trans-
form. So the error bars correspond to the width of the first peak (see Figure 6.9b).
This discrepancy comes from the dispersion of the refractive index, whereas Equation
6.9 is correct if the refractive index does not depend on the wavenumber. Obviously dis-
persion can not be discarded and a more complete expression is derived from Equation 6.8
after differentiating with respect to k, namely:
∂neff
δ km [neff ( km ) + km ( km )] L = 2 π. (6.10)
∂k
Thus a more comprehensive expression of the refractive index is given by the well-known
group velocity formula:
∂neff
n full = neff + km (6.11)
∂k
The term neff corresponds to the phase velocity and km (∂neff /∂k ) to the group velocity.
The latter contribution accounts for about 10% of the former one, which explains the dis-
crepancy in Figure 6.10. The group velocity correction is made of two different contribu-
tions resp. accounting for bulk and effective dispersions as summarized in Figure 6.11.
The bulk refractive index of the active layer is inferred from ellipsometric measurements,
evidencing a nonnegligible slope due to the proximity of the absorption band. The disper-
sion originating from the effective refractive index must also be taken into account, as the
thickness is of the same order of magnitude as the wavelength. Both contributions can be
calculated without any micro-laser spectrum. Thus the group velocity term is inferred
independently. For our structures, the full-index value accounts for nfull = 1.64 ± 0.01 and
does not depend significantly on thickness and polarization, allowing its use as such in
the following. Taking into account other error sources (spectrometer calibration, discreti-
zation, etc.), the global uncertainty on the geometrical length of the periodic orbit is esti-
mated to be less than 3%. The excellent agreement between this theoretical prediction and
experiments as shown in Figure 6.10 confirms our approach.
Having demonstrated the efficiency of our orbit length measurement method, we can
proceed using it to study different resonators with more complex shapes. For instance, let us
(a) (b)
1.54
1.57
1.56
Effectif index
1.50
Bulk index
1.55
1.46
1.54
1.53 1.42
Figure 6.11
(a) Refractive index versus the frequency for a 5% DCM doped PMMA layer, inferred from ellipsometric mea-
surements. (b) Effective refractive index versus the parameter “thickness over wavelength” for the TE funda-
mental mode calculated for constant bulk refractive indexes shown by the horizontal lines. For both graphs, the
vertical lines indicate the experimental range: from 600 nm to 630 nm for a 0.6 µm thickness.
consider the square and try to find out its dominant periodic orbit. Such an orbit is known
to minimize the refractive losses, so Fabry–Perot‡‡‡ orbits should be excluded whereas the
so-called diamond, such as drawn in Figure 6.12a, is to be expected. Figure 6.12b shows
the comparison between the length inferred from experiments and the expected diamond
orbit length. Such good agreement confirms our expectation that the dominant periodic
orbit is indeed the diamond. Based on such encouraging results, we propose in Section
6.5.2 a general method to predict the dominant periodic orbit in more complex situations
of increasing complexity.
(a) (b)
700
600
Optical length (µm)
500
a
400
300
Figure 6.12
(a) Two representations accounting for the diamond-like periodic orbit. They have the same fundamental fea-
tures: length and incident angles. (b) Optical length versus square side width a. The experiments (circle points)
are linearly fitted by the solid line. The dashed line corresponds to the theoretical prediction (diamond periodic
orbit) without any adjusted parameter, namely 2 2 an full .
‡‡‡ In general, we call Fabry–Perot a periodic orbit connecting two opposite plane sides.
to easier technological fabrication conditions allowing to test many cavity shapes, such
less stringent confinement conditions entailing stronger dissipation is a central corner-
stone of the use of polymer in the context of this chapter.
The aim of this section is to provide practical tools to predict the main features of the
electromagnetic field both inside and outside micro-cavities. After a paragraph dealing
with the application range and numerical simulations, three topics will be more specifi-
cally explored: spectral features, directions of emission and intra-cavity light pattern.
These will encompass most of the necessary elements to optimize a device or account for
an already existing one, for instance in order to combine ultra-low loss and directional
light output in a same resonator [55].
6.5.1 General Methodology
As noted in Section 6.2, we focus hereafter on passive resonators. Actually the influence
of the additional nonlinearity due to lasing seems to be very limited [16,17] and can be
accounted for from simple arguments to be developed at the end of Section 6.5.2. Therefore
tools proposed in the following sections can be directly used for passive devices and read-
ily extended to active ones.
As demonstrated in Section 6.2, the underlying physics of micro-cavities is based on solu-
tions of the passive two-dimensional stationary Helmholtz equation (see Equation 6.2):
( ∆ + ni2,o k 2 )ψ i,o = 0
where the label i (resp. o) stands for inside (resp. outside), and ψ corresponds either to the
vertical component of the electric or the magnetic field, depending on the polarization.
In this modeling, the effective refractive index (so far noted as ni,o) takes into account the
vertical direction (i.e. perpendicular to the substrate). The values ni = 1.5 and n0 =1 are used
in the following as being typical for organic materials, but the methods developed in this
section can be readily extended to higher values such as ni ~ 3 for inorganic materials.
However a low refractive index is sometimes a great advantage because refraction leads to
a strong selection and the specificities of nonescaping light are then emphasized.
Our purpose here is to provide “user-friendly” recipes to account for the main features
of the devices, with predictions from geometrical optics used whenever possible. Such
treatment makes sense considering the much larger typical size of cavities (a ~ 100 µm) as
compared to the wavelength (λ ~ 0.6 µm). Full wave solutions are used only if no other way
is known so as to account for the resonance patterns as shown at the end of Section 6.5.
To reinforce the relevance and validity of our approach, it is compared with numerical
simulations based on ray propagation or Maxwell equations, for which the precision is
consistent with our experimental studies and has been tested apart [56,57].
The ray simulations consist in statistics over a large number of rays (about 106) with ran-
dom starting points. Each ray propagates inside the cavity and is reflected or refracted at
the boundary depending on its incident angle.
More elaborate solutions of Maxwell equations become necessary if wave properties
are involved. Different methods have been proposed, that can be roughly separated into
two classes: time-dependent and stationary. The former one, finite-difference time-do-
main (FDTD), is broadly developed and available as a routine package but requires a time
consuming discretization procedure both inside and outside the cavity. However it has
already been performed close to the experimental range (see e.g. [33]). In the following
section, we use a stationary algorithm, known as boundary element method [56,57], which
provides spectra and resonance patterns from a discretized mesh localized only on the
cavity boundary.
6.5.2 Spectrum
The signature of a resonator is its spectrum. It is related to the scattering matrix (see e.g.
[20]), which is useful for numerical simulations and fundamental approaches. But in the
context of practical applications, we will consider the spectrum as a sequence of peaks
such as in Figure 6.9a to be further accounted for and predicted. Their relative intensities
depend on the gain media and pumping conditions. We will therefore focus on the more
robust quantities of position and periodicity.
In the case of a Fabry–Perot cavity (width w), the light is organized along the single
straight line periodic orbit of length L = 2w. The spectrum reveals this structure from the
phase coherence condition (see Section 6.4)
r2 eilkn = 1
for a medium without any refractive index dispersion. In general the peaks of an arbitrary
resonator belong to different combs, more or less in relation with the different periodic
orbits. For a Fabry–Perot resonator, there is a single periodic orbit and then a single comb.
At this stage, two questions can be raised:
A rigorous and quantitative answer to issue (1) is given by the trace formula. This approach
connects the density of states (i.e. the spectrum) to a sum over periodic orbits in the limit of
geometrical optics. To be more precise, for a cavity without loss the spectrum is a sequence
of peaks located at km that can be individually represented as Dirac functions:
d( k ) = ∑ δ(k - k )
m
m (6.12)
This expression underlies the wave point of view. In the geometrical optics approximation,
it has been analytically proved§§§ to be equivalent to
d( k ) ∝ d ( k ) + d( k ) = d ( k ) + ∑C
p
p cos( kLp + ϕ p ) (6.13)
§§§ he demonstration has been carried out for integrable and chaotic shapes [27,59]. The terms “integrable”
T
and “chaotic” refer to the formally equivalent dynamical system where a particle stands for the light and is
reflected at the boundary.
Here d ( k ) represents a mean of the density of states [58,59], which does not contain orbit
contributions leading us to concentrate on the more relevant d( k ) oscillating part. The latter
is expressed as a sum over the periodic orbits labeled by p, Lp corresponding to the length
of the p periodic orbit and Cp being a factor depending only on the geometrical properties
of the p periodic orbit. The additional phase ϕp will be of no use in the following and can
be omitted. The transition from Equations 6.12 to 6.13 is exact in the limit of geometrical
optics, that means when ka → ∞ where a is the typical size of the system.
The trace formula then provides a positive answer to issue (1): the spectrum of any reso-
nator is made of a sum over periodic orbits. But with some reservations. First the transition
from Equations 6.12 to 6.13 has been demonstrated for a broad diversity of systems, but not
for all of them. Secondly, this demonstration applies for closed systems (i.e. systems where
light can not escape the cavity) and the extension to resonators has only been conjectured
and not proved [20]. Thus the expression of the Cp coefficients in Equation 6.13 is not pre-
cisely known in the case of dielectric cavities, requiring further investigations.
With the answer to issue (1) being positive, in general, question (2) becomes relevant to
predict the spectrum of an arbitrary shape. For instance it has been experimentally dem-
onstrated in Section 6.4 that the dominating periodic orbit in a square is the diamond. But
the diamond apart, there exists a lot of different periodic orbits in a square. Their weigh-
ing depends on the attached Cp coefficients, which appear directly in the Fourier trans-
form of the spectrum (Equation 6.13). This point clarifies the data processing proposed
in Section 6.4 for experimental spectra, while a similar process can also be applied with
spectra inferred from numerical simulations [60,61].
The analytical expression of the Cp coefficients is well-known for closed systems. For
dielectric resonators, we have shown in some particular cases [21,53] that they seem identi-
cal, however additionally weighted by refractive losses:
d( k ) = ∑r C
p
p p cos( knLp + ϕ p ) (6.14)
For instance, the rp coefficient for the Fabry–Perot periodic orbit corresponds to
r p = r2
where
n-1
r=
n+1
leading to rp = 0.04 for n = 1.5. On the other hand, the diamond periodic orbit in the square
is confined by total internal reflection for n > 2 1.41, then rp = 1. So we notice that the
refractive losses have a significant influence. Thus if there exist periodic orbits confined by
total internal reflection, they should dominate the spectrum.
In a general way, it is necessary to evaluate the Cp coefficients. For instance in a circular
cavities, all N-polygons with N ≥ 4 (i.e. square, pentagon, hexagon, etc.) are confined by
total internal reflection. But their Cp coefficients are different and we have demonstrated
that the square periodic orbit is dominating [53].
The Cp coefficients are known for two kinds of periodic orbits, namely those isolated and in
family. These notions are illustrated in Figure 6.13 on the example of the pentagonal cavity.
One of the most obvious periodic orbits is the inscribed pentagon plotted in Figure 6.13a.
Hereafter it is referred to as a simple pentagon periodic orbit so as to avoid possible con-
fusion. It belongs to the class of isolated orbits. Actually if one of the bounces is slightly
moved, it leads to a very different trajectory. This is illustrated in Figure 6.13b and c. On the
other hand, the double pentagon periodic orbit plotted in Figure 6.13d, is in family since it
can be continuously deformed without losing its fundamental features: length and incident
angles. For instance, the arrows in Figure 6.13d indicate how to deform the double pentagon
all the way to the five-pointed star periodic orbit of Figure 6.13e, which is a boundary case
of the double pentagon periodic orbit.
In a circular cavity, the polygons are in family. For illustration, Figure 6.14a shows two
representations of the square periodic orbit family. When all these representations are
plotted, we notice that a portion of the cavity is not crossed by any ray. Its boundary, called
(a) (b) (c) (d) (e)
Figure 6.13
Periodic orbits in a pentagonal cavity. (a) Simple pentagon periodic orbit. (b) In continuous line, slight displace-
ment of the simple pentagon periodic orbit. (c) Dotted line: simple pentagon periodic orbit. Continuous line:
double pentagon periodic orbit. (d) Arrows indicate possible deformations of the double pentagon periodic
orbits. (e) Five-pointed star periodic orbit obtained from continuous deformation of the double pentagon peri-
odic orbit.
(a) (b)
1.0
Cp
0.5
0.0
5.0 5.5 6.0 6.5
Geometrical length
Figure 6.14
Circular cavity. (a) Two representations of the square periodic orbit family. The caustic is indicated in dotted
line. (b) Cp coefficients versus the length of the periodic orbits for the disk from Equation 6.15. From left to right,
the vertical solid lines correspond to square, pentagon, hexagon, etc., and the vertical dotted line indicates the
position of the perimeter. The Cp coefficient for the square periodic orbit has been normalized to unity. Triangle
and diameter are not plotted since they are not confined by total internal reflection for n = 1.5.
caustic, is indicated in dotted in Figure 6.14a. In a general way, for a periodic orbit in fam-
ily, the Cp coefficients are
Ap
Cp = (6.15)
Lp
where Ap is the area covered by the family. For a circular cavity of radius R and the fam-
ily of a polygon with N sides, AN corresponds to the area between the boundary and the
caustic and is written as AN = πR 2 sin2(π/N). As the length of the orbit is LN = 2Nr sin(π/N),
the Cp coefficients are
( )
3/ 2
1 π
CN ∝ R sin
N N
These coefficients are plotted in Figure 6.14b. The coefficient corresponding to the square
periodic orbit is higher and is thus expected to dominate the spectrum, which has been
experimentally demonstrated [53]. As the Cp coefficients of the other periodic orbits are not
so different, they also contribute to the spectrum but with a smaller weight.
If both kinds of periodic orbits coexist, i.e. isolated and in family, the trace formula pre-
dicts that in family periodic orbits dominate the spectrum. For instance, in the case of the
pentagonal cavity where the simple pentagon periodic orbit is isolated and the double
pentagon is in family, it has been demonstrated both experimentally and via numerical
simulations, that the dominant periodic orbit is indeed the double pentagon [53].
There exists a last configuration where it is possible to predict the dominating periodic
orbits: if all the periodic orbits are unstable, and thus isolated. Figure 6.15b explains how the
stability of a periodic orbit can be characterized. The continuous curve corresponds to the
periodic
orbit. For a typical starting point slightly shifted by an infinitesimal displacement
| dr |, one considers
the end point of the resulting trajectory after one round trip dr ′. If | dr ′|
is smaller than | dr |, the orbit is stable, whereas it is unstable if it is larger. The λp Lyapounov
coefficient quantifies this instability. Appendix A specifies how it can be calculated in a prac-
tical way for optical cavities. In the case of an unstable periodic orbit, the Cp coefficient is
1
Cp = λ L (6.16)
sinh p2 p
(a) (b)
|dr΄| |dr|
Figure 6.15
(a) Periodic orbits in a stadium-shaped cavity: rectangle in solid line and elongated hexagon in dotted line.
(b) Diagram to explain how to quantify the stability of a periodic orbit.
This means that the dominating periodic orbit minimizes the λpLp product, so as to make
it both shorter and less unstable (small Lp and λp).
As an example, let us consider a stadium-shaped cavity (see Figure 6.15a). It has been
proved to be fully chaotic [62], leading to unstability for all the periodic orbits¶¶¶. The a
priori dominating periodic orbit is confined by total internal reflection, so the compet-
ing orbits are shaped as elongated polygons**** as plotted in Figure 6.15a. They exhibit
comparable lengths, corresponding approximately to the perimeter. But their Lyapounov
coefficients are significantly different. It is shown in Appendix A, that they do increase
with the number of bounces on circular boundaries. Thus the dominating periodic orbit
is expected to have the minimum number of bounces and so to be the rectangle. A more
refined approach is proposed in [61].
To summarize, in the limit of geometrical optics, resonator spectra are made of peak
combs in relation with the periodic orbits. The competing periodic orbits are weighted by
a coefficient rpCp, taking into account their refractive (rp) and geometrical (Cp) features. At
this time, it is possible to predict the dominating periodic orbits in three configurations
• The periodic orbits are living in family with their weighing given by formula
(Equation 6.15).
• The periodic orbits are unstable (and so isolated) with their weighing given by
formula (Equation 6.16).
• Isolated and in family periodic orbits coexist. Then the in family confined periodic
orbits dominate.
To conclude this section devoted to spectra, we must come back to the influence of the
detection set-up and the laser effect. Actually, predictions from trace formula stand for
what happens inside a passive resonator, and thus favor well-confined periodic orbits. But
in practice, we detect what is emitted and so the hierarchy can be seen as different from
that inferred from trace formula predictions. For instance the diamond is the dominant
periodic orbit in a square-shaped micro-cavity, as appears in Figure 6.12b. These experi-
mental spectra have been collected in a direction perpendicular to the square sides for
a low pump energy. If the pump energy is increased high enough, then the Fabry–Perot
periodic orbit can start to lase. As its outside coupling is very efficient, it dominates the
spectrum at the expense of the diamond periodic orbit, which is confined by total internal
reflection [21]. Moreover the energy injection scheme can deeply modify the hierarchy (e.g.
to pump a ring from an adjacent waveguide). Depending on the application, this selec-
tion can turn out to be detrimental or advantageous. It can be prevented with a uniformly
pumped active medium to ensure a homogeneous distribution of light or, on the other
hand, enhanced by local pumping.
In this section, a concrete method based on periodic orbits has been presented to predict
the spectrum of an arbitrary resonator. This approach is exact for passive cavities and can
be easily extended depending on the pumping and the emission process.
¶¶¶ ctually this definition also allows for a countable set of periodic orbits with marginal stability living in
A
family. In a stadium-shaped cavity, there is one such periodic orbit: the Fabry–Perot orbit which connects the
straight lines. But in practice, it is neglected because of its important refractive losses.
**** The elongated polygons are confined by total internal reflection only in the strong limit of geometrical
optics, i.e. |kr| 1. For |kr| 1, these orbits have refracted losses due to the curvature of the boundary.
(a) (b) 90
135 45
y
Ω
x 180 0
225 315
270
Figure 6.16
(a) Proposed diffraction process at a polygon corner. (b) Normalized far-field patterns for micro-squares
etched by different process. The triangle hourglass correspond to a square shaped micro-laser etched by a
“photolithography+RIE” process and then selected for the quality of its corners, while the circled points refer
to a cavity etched by e-beam.
(a) 90 (b) 90
135 45 135 45
180 0 180 0
270 270
Figure 6.17
Experimental far-field patterns in polar coordinates for chaotic cavities. (a) Stadium-shaped cavities with l/r = 0.5
(continuous line) and l/r = 1 (dashed line). See notations in Figure 6.18b. (b) Cut-disk with d = 0.9. The parameter
d is defined as the radius proportion which has not been removed from the circle (see [21]).
(a) (b)
1.0
l/r = 0.3
0.5
Normalized intensity
0.0
1.0
χ0
l/r = 1
0.5
0.0 θ
1.0 r
l/r = 2 l
0.5
0.0
0 20 40 60 80
Emission angle θ (degrees)
Figure 6.18
(a) Far-field pattern of individual well-confined wave-functions for micro-stadiums with different l/r shape
ratios. From top to bottom: (l/r=0.3, Re(kr)=39.31); (l/r=1, Re(kr)=13.88); (l/r=2, Re(kr)=23.29). The θ angle is defined
according to the main axis of the stadium as in (b). (b) Outline of the lens model and some notations.
Experiments and ray simulations, which agree with each other, show a great depen-
dence on the direction of emission with the l/r shape parameter (see Figures 6.17a and 6.19).
This feature is very useful as well for practical applications as for testing our model. This
model, which is referred to as “lens model” in the following, is based on three assump-
tions, illustrated on the example of the stadium:
• The confined light is localized close to the boundary of the cavity to limit the
refraction losses, like the so-called WGM. This assumption is confirmed by wave
and ray simulations. Thus, due to the stadium geometry, the emission occurs at a
half-circle, originating from the other half-circle.
• A kind of ergodicity appears on the half-circles, which means that the source points
(dark spot on the right of Figure 6.18b) are more or less uniformly distributed on the
60
Ray simulations
Wave simulations
50 Lens model
Experiments
40
θ (degrees)
30
20
10
0
0.5 1.0 1.5 2.0 2.5 3.0 3.5
l/r
Figure 6.19
Direction of maximal emission versus the l/r shape for a stadium-shaped cavity. Comparison between experi-
ments, ray simulations, wave simulations, and predictions of the lens model. The wave simulation points cor-
respond to an averaging over a representative set of high-Q wave-functions, while their error bars represent
the standard deviation.
Then within this model, we just need to calculate the average over the θ angles for a start-
ing point covering the right half-circle. Due to the equivalent averaging over S or χ , the
0
direction of emission is defined by
χ
2
1
< θ> =
χ 2 - χ1 ∫ θ(χ ) dχ
χ1
0 0
(6.17)
It seems that the choice of an integration density such as dχ0, dsin(χ0) or dϕ(χ0) is not rel-
evant compared to the experimental precision. The limiting angles, χ1 and χ2, are directly
connected to refraction or geometrical constraints. In the stadium example, χ1 is the critical
angle, and χ2 the boundary angle, χ0 corresponding to the incident ray at the connection
between the circular and straight part.
This <θ> value can be analytically calculated versus the l/r stadium shape and is plotted
as the inferior dashed line in Figure 6.19. The other dashed lines correspond to equivalent
quantities but considering reflections on the straight boundaries. The detailed formulae
are developed elsewhere [13,21] and their predictions agree with ray simulations (second
branch not represented in Figure 6.19) and wave simulations. Probably due to the limited
resolution, the second peak has not yet been shown experimentally.
The good agreement between this lens model, experiments and simulations, confirms in
some way the above assumptions: a kind of ergodicity and focusing due to circular bound-
aries. Even without calculating Equation 6.17, these tools allow to qualitatively predict and
interpret the directions of emission of a wide range of chaotic polymer§§§§ cavities.
• The wave-functions propagate along the periodic orbit and imitate its periodicity:
Ψ( x , y ) = ψ (k y y ) e ik x x
§§§§ he extension to high refractive index (n~3−4) is far from obvious because of the increasing number of rays
T
confined in the cavity, which leads to a more complicated dynamics [14].
(a) (b)
y
x
Figure 6.20
(a) To clarify the plotting of ray trajectories, the rays are not reflected when bouncing at the boundary, but propa-
gate along a straight line, while a symmetrical cavity is being plotted across the boundary. The cavity boundary
is indicated in continuous line and the ray in dotted line. (b) The channel of the double pentagon periodic orbit
covers the hatched area. Their lateral boundaries are plotted in continuous horizontal line and one representa-
tive of this orbit in dotted line.
with
where δ takes into account the dephasing at each reflection on the boundary, and
Lp is the length of the periodic orbit. In the limit of geometrical optics, the trans-
versal excitation is negligible compared to the longitudinal one, i.e. κy κx, then
Equation 6.18 is similar to the basic elements of Section 6.2.
• The transversal profile, ψ(kyy), is defined by geometrical constraints. For the pen-
tagonal cavity, the corners force the high-Q wave-functions to vanish in their vicin-
ity so as to minimize their diffraction losses. In the case of closed cavities [68], it
leads to a vanishing along all lateral boundaries of the channel, as indicated with
the horizontal continuous lines in Figure 6.20b. As the diffraction coefficient on
a dielectric corner is still unknown, no similar demonstration can be invoked for
dielectric cavities, but it seems reasonable to propose a similar assumption, and so
ψ(kyy)=sin(kyy)
with
kylp = qπ q∈N∗
At this step, the wave-functions can be folded back to the original cavity. Once symmetry
class and polarization have been taken into account, the agreement with wave numerical
simulations is very satisfactory [53]. Beyond its quantitative efficiency, the main interest of
this method lies in predictions of resonance structures. In fact, it appears in Figure 6.20
that the channel never cross the cavity center as the rays passing through the corners
delineate the limit of the double pentagon periodic orbit. Thus it draws a pentagonal caus-
tic which is clearly seen in the wave-function plotted in Figure 6.21a. This scheme can be
generalized to other shapes such as the hexagonal cavity presented in Figure 6.21b.
The efficiency of this approach encourages an extension to a broad diversity of shapes,
based on the clearly dominating periodic orbit which often underlies the high-Q modes.
However the prediction of resonance structures requires transversal boundary condi-
tions. In the case of the polygonal cavities reported above, they can be deduced from the
(a) (b)
Figure 6.21
Numerical simulations: modulus squared of wave-functions for the TM polarization and a particular sym-
metry class. In this grey scale, black area corresponds to minimal intensity zones. (a) Pentagonal cavity, ak =
123.0−0.02i. (b) Hexagonal cavity, ak = 90.99−0.12i. a is the side length.
corresponding metallic cavities where a robust theory has been developed [68]. For other
cavity shapes, such as chaotic, strong evidence inclines to possible extensions and studies
are still in progress.
r = R + λƒ(θ)
δa k2 1
where δa is the area modified from the circular cavity, represented by dashed regions
in Figure 6.22. From this criterion, the wavelength must be larger than the deformation
scale.
Different expansion methods are well-documented, mainly in the context of quantum
physics. Hereafter, we focus on the Green function method since it avoids definition prob-
lems in the modified area [23]. It is directly based on the two-dimensional Helmholtz
equation
( ∆ + k 2 n2 ( x )) ψ ( x ) = 0
Figure 6.22
Example of a cavity shape which can be studied by the perturbation method.
with position dependent “potential” n2 ( x). For perturbed cavity
n2 ( x ) = n02 ( x ) + δ n2 ( x )
where n02 ( x ) is the “potential” for the pure circular cavity
2 n2 when |x| < R
n (x ) =
0
1 when | x| > R
and δn2 ( x ) is the perturbation to the regular potential. Then the Helmholtz equation can
be rewritten in the form
( ∆ x + k 2n02 (x )) ψ(x ) = - k 2δ n2 (x )ψ(x )
and its formal solution is given by the following integral equation
∫
ψ ( x ) = - k 2 G( x , y )δ n2 ( y )ψ ( y )dy (6.19)
Here G( x, y ) is the Green function for the dielectric circular cavity
and describes the field
produced at point x by a delta-function source situated at point y. The explicit expressions
of this function are presented in Appendix C of [23]. “Expression” is written in the plural
since the plane—and then the study—is divided into three circular regions, r<R1, R1<r<R2,
and r>R 2 defined by
R1 = min θ (R , R + λf (θ))
(6.20)
R2 = max θ (R , R + λf (θ)).
Their boundaries
are indicated by dashed circles in Figure 6.22, the deformation “poten-
tial” δn2 ( x ) being nonzero only in the second region R1<r<R 2. Due to singular character
of the Green function, wave functions inside each region are represented by different
expressions, leading to different perturbation expansions. But continuity of the perturba-
tion wave-functions are maintained thanks to the Green function.
The main steps of the calculations are detailed in [23], where they are performed up to
second order λ. In [23], this method is applied to the example of the cut disk and leads to
very good agreement with numerical simulations both for the wave-functions¶¶¶¶ and the
spectrum, while this system presents a double difficulty due to its chaoticity and dielec-
tric corners. In a more general way, this approach seems very promising for cavity shapes
where wave numerical simulations reveal a low efficiency such as cusp or concave bound-
aries. All the more so as there are no restrictions due to the refractive index, and it can also
be directly applied to inorganic cavities.
To summarize this section, we have proposed two different approaches to predict and
explain the resonance structures. The former one is an ad hoc “scar” model based on the
dominant periodic orbit and therefore useful in the geometrical optics limit. The other
approach is a perturbation expansion from the circular cavity and leads to analytical for-
mulae valid for wavelengths of the same order of magnitude or larger than the small per-
turbation parameter.
6.6 Conclusion
The engineering and physical properties of polymer based micro-cavities have to be con-
sidered in the dual scope of light–matter interactions, where physical and chemical view-
points are inter-twinned and are considered on a par. A supposedly “purely” physical
approach, is traditionally more attached to photons per se, in the legitimate search of gen-
eral physical laws governing their behavior, the validity of which should extend above and
beyond any specific class of materials. However, the legitimate search for generality does
not imply ignorance of the material point of view, as photonic effects obviously require
material interactions to manifest themselves. In our opinion, this quest has been amply
illustrated above in the case of polymer based micro-lasers.
Polymers provide, both in terms of their specific technological assets as well as physi-
cal properties, a unique enabling arena allowing progress towards more general (e.g.
ultimately material independent) mathematical physics rules governing the confinement
and out-coupling of light in dissipative billiard-like micro-cavities. As far as technologi-
cal assets are concerned, it is the versatility of polymer technologies allowing to fabri-
cate and screen a wealth of micro-cavity shapes (in fact much beyond those that could
be discussed within the limited space of this chapter). This versatility has been taken
advantage of here, so as to experimentally probe and theoretically analyze previously
unexplored and hardly accessible configurations with the potential to challenge existing
models.
As far as the physical properties of polymers are concerned, it is their low index of refrac-
tion and the ensuing dissipative nature of micro-cavities (as opposed to higher index of
inorganic semiconductor based cavities), that has been specifically targeted in this study.
This leads to the dominance of easily identified “scar” orbits which provide a simple and
¶¶¶¶ I t must be stressed that this approach leads to the wave-functions inside and outside the cavity, and thus
predicts the far-field as well.
illustrative geometric support to the field distribution. Such a situation is in contrast with
that of micro-cavities based on higher index materials where the more messy electromag-
netic speckle-like wave-functions do not lend themselves to the extraction of clearly distinc-
tive geometric optics orbits. As shown in Section 6.5, this approach has lead to conjecture
the extension of the trace formula to the case of chaotic dissipative micro-cavities, thus
building-up on the work by Gutzwiller et al. [53,61] and extending its validity range, based
on the interpretation of our polymer based micro-cavity experiments.
Far from an embarrassment or a degrading factor as opposed to the universality of phys-
ics, respectful consideration of materials in their diversity and specificity entails a benefit. It
would be a shame to ignore the opportunities of exploiting different material properties and
functions, if only as a useful help and handle towards reaching an increased level of “imma-
terial” generality. But material science and, more generally, chemistry deserve better than
being considered as a transient and utilitarian technological playground towards physics.
Indeed, the application and engineering potential of micro-cavity lasers will critically and
specifically depend on and benefit from specific material choices, making progress in this
important and promising area of research critically dependant on joint efforts in physics,
both theoretical and experimental, as well as material science and technology
Acknowledgments
The authors are grateful to their collaborators at Laboratoire de Photonique Quantique et
Moléculaire, N. Djellali, R. Hierle, J.-S. Lauret, and J. Lautru, at Laboratoire de Physique
Théorique et Modèles Statistiques, R. Dubertrand and C. Schmit, and at Laboratoire de
Photonique et de Nanostructures, C. Ulysse and Y. Chen.
| Λ | = e λL
where L is the length of the periodic orbit. The monodromy matrix is calculated by using
the basic matrix ∏(L) (propagation over a distance L) and R(χ) (reflection with incident
angle χ and curvature radius ρ).
-1 0
1 L
Π(L) = R(χ) = 2
0 1 -1
ρ cos(χ)
For instance, in a stadium-shaped cavity, the monodromy matrix of the rectangle periodic
orbit (Figure 6.15a) can be written [Π(L1)R(π/4)Π(L2)R(π/4)]2 with L1 = 2l/r + 2 and L2 = 2 .,
assuming r = 1. For a periodic orbit in a shape of an elongated polygon with m bounce on
each half-circle, the trace of the monodromy matrix can be written as
()
2
16 m l 16 m2 l
trM = 2 + +
sin ( 2πm ) r sin 2 ( 2πm ) r
References
1. H. S. Nalwa and S. Miyata editors. Nonlinear optics of organic molecules and polymers. CRC
Press, 1997.
2. B. Valeur. Molecular fluorescence. Principle and applications. Wiley-VCH, Weinheim, 2002.
3. J. Zyss. Molecular nonlinear optics: materials, physics, and devices. Academic Press, Boston,
MA, 1994.
4. H. C. Song, M. C. Oh, S. W. Ahn, W. H. Steier, H. R. Fetterman, C. Zhang. Flexible low volt-
age electro-optic polymer modulators. Applied Physics Letters, 82, 4432, 2003.
5. Y. Huang, G. T. Paloczi, J. K. S. Poon, and A. Yariv. Demonstration of flexible free standing
all polymer integrated optical ring resonator devices. Advance Materials, 16, 44, 2004.
6. R. Levenson and J. Zyss. Polymer based optoelectronics: from molecular nonlinear optics
to device technology. In Material for optoelectronics, M. Quillec editor. Kluwer, Dordrecht,
1996.
7. P. Labbe, A. Donval, R. Hierle, E. Toussaere, and J. Zyss. Electro-optic polymer based device
technology for optical telecommunication. In Molecular photonics: material, physics, and
devices, special issue (in English) of the Compte-rendu de l’Académie des Sciences, 3, 2002.
8. S. Kim, W. Yuan, K. Geary, Y.-C. Hung, H. R. Fetterman, D.-G. Lee, and C. Zhang. Electro-
optic phase modulator using metal-defined polymer optical waveguide. Applied Physics
Letters, 87, 011107, 2005.
9. R. Song and W. Steier. Overlap integral factor enhancement using buried electrode structure
in polymer Mach-Zehnder modulator. Applied Physics Letters, 92, 031103, 2008.
10. A. Moliton. Optoelectronics of molecules and polymers. Springer, 2006.
11. V. Sandoghar, F. Treussart, J. Hare, V. Lefèvre-Seguin, J.-M. Raimond, and S. Haroche. Very
low threshold whispering-gallery-mode microsphere laser. Physical Review A, 54, R1777,
1996.
12. D. W. Vernooy, V. S. Ilchenko, H. Mabuchi, E. W. Streed, H. J. Kimble. High-Q measurements
of fused-silica microspheres in the near infrared. Optics Letters, 23, 247, 1998.
13. M. Lebental, J. S. Lauret, J. Zyss, C. Schmit, and E. Bogomolny. Directional emission of
stadium-shaped microlasers. Physical Review A, 75, 033806, 2007.
14. S. Shinohara, T. Fukushima, and T. Harayama. Light emission patterns from stadium-
shaped semiconductor microcavity lasers. arxiv:0706.0106, 2007.
15. H. Türeci, D. Stone, and L. Ge. Theory of the spatial structure of nonlinear lasing modes.
Physical Review A, 76, 013813, 2007.
16. T. Harayama, T. Fukushima, P. Davis, P. O Vaccaro, T. Miyasaka, T. Nishimura, and T. Aida.
Lasing on scar modes in fully chaotic microcavities. Physical Review E, 67, 015207(R), 2003.
17. S.-Y. Lee, M. S. Kurdoglyan, S. Rim, and C.-M. Kim. Resonance patterns in a stadium-shaped
microcavity. Physical Review A, 70, 023809, 2004.
18. M. Lebental, J.-S. Lauret, R. Hierle, and J. Zyss. Highly directional emission of stadium-
shaped polymer microlasers. Applied Physics Letters, 88, 031108, 2006.
19. H.-J. Stöckmann. Quantum chaos, an introduction. Cambridge University Press, Cambridge,
1999.
20. P. Cvitanovic´, R. Artuso, R. Mainieri, G. Tanner, and G. Vattay. Chaos: classical and quantum.
chaosBook.org, Niels Bohr Institute, Copenhagen, Cambridge, 2005.
21. M. Lebental. Quantum chaos and organic micro-lasers. PhD dissertation, http://tel.
archives-ouvertes.fr/2007.
22. H. G. L. Schwefel, A. D. Stone, and H. E. Tureci. Polarization properties and dispersion rela-
tions for spiral resonances of a dielectric rod. Journal of the Optical Society of America B, vol.
22, 2295, 2005.
23. R. Dubertrand, E. Bogomolny, N. Djellali, M. Lebental, and C. Schmit. Circular dielectric
cavity and its deformations. Physical Review A, 77, 013805, 2008.
24. P. K. Tien. Light waves in thin films and integrated optics, 10, 2395, 1971.
25. C. Vassallo. Optical waveguide concepts. Elsevier, Amsterdam-Oxford-New York-Tokyo,
1991.
26. Sommerfeld. Optics. Lectures on theoretical physics volume IV. Academic Press, 1964.
27. M. C. Gutzwiller. Chaos in classical and quantum mechanics. Springer, 1991.
28. B. Gates, Q. Xu, M. Stewart, D. Ryan, C. Willson, and G. Whitesides. New approaches to
nanofabrication: Molding, printing, and other techniques. Chemical Review, 105, 1171, 2005.
29. V. Klimov, A. Mikhailovsky, S. Xu, A. Malko, J. Hollingsworth, C. Leatherdale, H.-J. Eisler,
and M. Bawendi., Optical gain and stimulated emission in nanocrystal quantum dots.,
Science, 290, 314, 2000.
30. C.-Y. Chao, W. Fung, and L. J. Guo. Polymer microring resonators for biochemical sensing
applications. IEEE Journal of Selected Topics in Quantum Electronics, 12, 134, 2006.
31. B. F. Howell and M. G. Kuzyk. Amplified spontaneous emission and recoverable photodeg-
radation in polymer doped with dispersed orange 11. Journal of the Optical Society of America,
19, 1790, 2002.
32. B. F. Howell and M. G. Kuzyk. Lasing action and photodegradation of disperse orange 11
dye in liquid solution. Applied Physics Letters, 85, 1901, 2004.
33. W. Fang and H. Cao. Wave interference effect on polymer microstadium laser. Applied
Physics Letters, 91, 041108, 2007.
34. Kallinger, M. Hilmer, A. Haugeneder, M. Perner, W. Spirkl, U. Lemmer, J. Feldmann,
U. Scherf, K. Müllen, A. Gombert, and V. Wittwer. A flexible conjugated polymer laser.
Advanced Materials, 10, 920, 1998.
35. Y. Huang, G. Paloczi, J. Scheuer, and A. Yariv. Soft lithography replication of polymeric
microring optical resonators. Optics Express, 11, 2452, 2003.
36. M. Armani, A. Srinivasan, and K. J. Vahala. Soft lithographic fabrication of high Q polymer
microcavity arrays. Nano Letters, 7, 1823, 2007.
37. Y. Chen. Nanoembossing techniques. In Encyclopedia of Nanoscience and Nanotechnology, H.S.
Nalwa, editor. American Scientific Publisher, 2004.
38. T. Ben Messaoud and J. Zyss. Unidirectional laser emission from polymer-based spiral
microdisks. Applied Physics Letters, 86, 241110, 2005.
39. O. G. Peterson, J. P. Webb, W. C. McColgin, and J. H. Eberly. Organic dye laser threshold.
Journal of Applied Physics, 42, 1917, 1971.
40. P. P. Sorokin, J. R. Lankard, V. L. Moruzzi, and E. C. Hammond. Flashlamp-pumped organ-
ic-dye lasers. Journal of Chemical Physics, 48, 4726, 1968.
41. H. G. L. Schwefel, N. B. Rex, H. E. Tureci, R. K. Chang, A. D. Stone, T. Ben Messaoud, and
J. Zyss. Dramatic shape sensitivity of directional emission patterns from similarly deformed
cylindrical polymer lasers. Journal of the Optical Society of America, 21, 923, 2004.
42. W. Fang, H. Cao, and G. S. Solomon. Control of lasing in fully chaotic open microcavities by
tailoring the shape factor. Applied Physics Letters, 90, 081108, 2007.
43. J. Gao, P. Heider, C. J. Chen, X. Yang, C. A. Husko, C. W. Wong. Observations of inte-
rior whispering gallery modes in asymmetric optical resonators with rational caustics.
arxiv:0707.2552, 2007.
Misha Sumetsky
OFS Laboratories
Contents
7.1 Introduction......................................................................................................................... 355
7.2 Microfiber Photonics.......................................................................................................... 356
7.3 Microfiber Loop Resonator (MLR)................................................................................... 359
7.3.1 Theory of an MLR . ................................................................................................ 359
7.3.1.1 Transmission Amplitude........................................................................ 360
7.3.1.2 Q-Factor, Extinction Ratio, and Finesse................................................ 361
7.3.1.3 Models of Directional Coupling ........................................................... 362
7.3.2 Experimental Demonstration and Applications of MLR.................................. 363
7.3.2.1 MLR Fabricated by Macro-Manipulation............................................. 363
7.3.2.2 Knot MLR Fabricated by Micro-Manipulation.................................... 368
7.4 Microfiber Coil Resonator (MCR)..................................................................................... 370
7.4.1 Theory of an MCR.................................................................................................. 371
7.4.1.1 Higher-Order MLR.................................................................................. 372
7.4.1.2 Uniform MCR........................................................................................... 372
7.4.1.3 MCR Transmission Line.......................................................................... 373
7.4.2 Experimental Demonstration and Application of MCR................................... 375
7.4.2.1 MCR in Air................................................................................................ 375
7.4.2.2 MCR in Low-Index Polymer................................................................... 377
7.4.2.3 MCR Microfluidic Sensor........................................................................ 378
7.5 Conclusion........................................................................................................................... 380
References...................................................................................................................................... 381
7.1 Introduction
This Chapter reviews recent progress in theory, fabrication, and application of optical
microfiber (MF) resonators, i.e., a microfiber loop resonator (MLR) and a microfiber coil
resonator (MCR). Generally, MF devices and circuits can be created by manipulation
(bending, looping, coiling, twisting, crossing, etc.) of uniform and tapered MFs. An opti-
cal MF is usually fabricated from a standard telecom optical fiber by drawing and has a
diameter of ~1 μm. The interest in photonic devices fabricated of MFs is caused by small
losses and potential ability of micro-assemblage in 3D. Eventually, these properties could
make possible the creation of MF devices, which are more compact and perform bet-
ter than those fabricated lithographically. Furthermore, some MF-based devices possess
355
© 2009 by Taylor & Francis Group, LLC
functionalities, which are not possible or much harder to achieve by other means. MLR
and MCR resonators together with MF waveguide are the basic elements of MF pho-
tonics, which is introduced in Section 7.2. Section 7.3 reviews the theory, experimental
demonstration, and application of an MLR. Section 7.4 reviews the theory, experimental
demonstration, and application of an MCR.
ring resonators coupled to a bus waveguide. From this Figure, the following advantages of
an MCR device become clear: (a) smaller dimensions due to 3D assemblage; (b) simple low-
loss input and output connections; (c) creation of a single continuous fiber; (d) possibility
of fabrication with feedback; and (e) smaller transmission loss than that of a planar wave-
guide, whose surface roughness is introduced by etching. Possible functional elements,
which can be created by wrapping an MF around a cylindrical surface, are a delay line, a
sequence of uniform and chirped MLR, and a uniform MCR illustrated in Figure 7.2a, b,
c and d, respectively. Eventually, the MF coils can be assembled into more complex func-
tional MF circuits as shown in Figure 7.3.
In order to create a photonic circuit composed of MFs, one has to learn how to manipulate
them. The simplest elements of an MF circuit are straight segments, bends, couples, and
loops. These elements can be composed in a plane and are similar to the corresponding
elements of lithographically-fabricated photonic circuits. Using MFs, one can create more
compact photonic circuits compared to the conventional planar circuits by assembling
them in 3D, e.g., by wrapping on an optical rod. Figure 7.4a shows an example of the MF
macro-manipulation, which was used in earlier studies [14,21,22] for creating of an MF loop
Figure 7.1
Example of shrinking a planar ring resonator circuit into an MCR. (a) Planar circuit consisting of ring resona-
tors coupled to a bus waveguide; (b) MCR with the same functionality. (From Sumetsky, M., IEEE J. Lightwave
Technol., 26, 21–27, 2008.)
Figure 7.2
Simplest MF devices, which can be fabricated by coiling an MF around optical rod: (a) a delay line; (b) and (c) a
uniform and a chirped sequence of loop resonators; (d) a coil resonator.
in free space illustrated in Figure 7.5a and described in Section 7.3. Macro-manipulation of
an MF in free space does not allow the creation of MF bends with small radii of curvature.
However, this problem can be solved with optical rods [23–26] as described in Section 7.4.
Figure 7.4b illustrates the micro-manipulation of MFs using the tips of a scanning tunnel-
ing microscope. With this technique, fabrication of short MF segments, an MF coupler, and
an MF knot illustrated in Figure 7.5b were demonstrated [10,27–31].
Figure 7.3
Illustration of a MF photonic circuit.
Biconical Microfibers
fiber taper
3D translation ex substrate
stage Low refractive ind
Rotation
stage
Figure 7.4
(a) MF macro-manipulation with translation and rotation stages; (b) MF micro-manipulation with two STM tips,
which hold and bend an MF.
(a) (b)
Figure 7.5
(a) an MF loop resonator; and (b) an MF knot resonator.
Coupling region
s
A– A+
A–
01
02 01
A+
02
1 2
s=0
Cross-section of equal
propagation constants
Figure 7.6
Illustration of an MLR. The coupling region is outlined. The numbers 1 and 2 correspond to the coupled micro-
fiber segments 1 and 2. (From Sumetsky, M., Dulashko, Y., Fini, J. M., Hale, A., and DiGiovanni, D. J., IEEE J.
Lightwave Technol., 24, 242–250, 2006.)
holds for an MLR shown in Figure 7.5a and may be wrong for a knot MLR shown in
Figure 7.5b where the twisting effects are significant. In the latter case, the results pre-
sented in this subsection should be modified by taking into account coupling between
polarization states [33].
s
d A1
0
∫
k( s)exp i ∆β( s)ds
s1
= i (7.1)
ds A2 s
s1
∫
k( s)exp - i ∆β( s)ds
0
∆β( s) = β1 ( s) - β 2 ( s)
where κ(s) is the inter-fiber coupling coefficient. From this equation, the transmission
amplitude of the MLR, T, can be found in the form, which is similar to that of the ring
resonator [34]:
e - a/2 e iΨ - sin( K )
T= ,
1 - sin( K )e - a/2 e iΨ
Ψ = Re(Θ), a = 2 Im(Θ) << Ψ , (7.2)
S
1
Θ=
2 ∫
(β1 ( s) + β 2 ( s))ds + ϕ , ϕ << Ψ.
0
Here, the parameter K and phase ϕ characterizes the coupling between the adjacent MF
segments, parameter α characterizes the roundtrip attenuation, and S is the circumference
of the MLR. The MLR is nonreflecting device and, for lossless propagation, |T| = 1. The
losses in the MF loop are taken into account by the complex-valued propagation constant
β, and, consequently, by a complex valued phase Θ. If losses are uniformly distributed
along the MF length, we have Im(∆β) = 0 and the imaginary component of propagation con-
stant does not enter the coupling wave equation, Equation 7.2. If the coupling parameter
K = 0 then the loop is decoupled and, from Equation 7.2, the MLR transmission behavior is
similar to that of a straight MF. In the opposite case of strong coupling, the resonances in
transmission amplitude occur only if the coupling parameter K is close to the values
π
Km = (2 m + 1), m = 1, 2 ,..., (7.3)
2
which correspond to the full transmission of electromagnetic field from one of the adjacent
MF segments to another. Then the resonances in transmission amplitude correspond to
the condition:
Ψ = Ψmn = π(m + 2n) (7.4)
where n is an integer.
If the propagation losses are ignored, then the propagation constant is real and T = 1.
In this case, the MLR performs as an all pass filter [35] and the resonances appear in the
group delay, td, only. The group delay is determined through the derivative of the phase of
T with respect to radiation wavelength λ:
n f λ 2 d ln(T )
td = , (7.5)
2 πc dλ
where c is the speed of light and nf is the effective refractive index of the MF.
For relatively low losses, a<<1, the resonances are well separated. In this case, in the
neighborhood of the resonance wavelength λ0 calculated from Equations 7.3 and 7.4, the
expression for the transmission amplitude is simplified:
( γ a - γ c ) + i(λ - λ 0 )
T= ,
( γ a + γ c ) - i(λ - λ 0 )
(7.6)
1 ∂Ψ -1 1 ∂Ψ -1
γa= a, γ c = ( K - K m )2
2 ∂λ λ = λ 0 2 ∂λ λ = λ 0
Here the parameters γc and γa are the coupling and attenuation parameters, respectively.
From Equation 7.6, the transmission power of an MLR is:
( γ a - γ c )2 + ( λ - λ 0 )2
P= (7.7)
( γ a + γ c )2 + ( λ - λ 0 )2
λ0
Q= . (7.8)
2( γ a + γ c )
The Q-factor defined by Equation 7.8 is often called the loaded Q-factor. Sometimes it is
also useful to introduce the intrinsic Q-factor,
λ0
Qint = . (7.9)
2γ a
This Q-factor takes into account only internal losses of the MLR. From Equation 7.7, the
extinction ratio of the resonance is:
γ + γ a 2
R = 10 ⋅ log c (7.10)
γ c - γ a
The finesse of a resonator is the ratio of its free spectral range (FSR) to the resonance
FWHM: F = FSR/∆λ. From Equation 7.2
2π
F= . (7.11)
( K - K m )2 + a
Equations 7.8 through 7.11 show that the MLR parameters contributing to its Q-factor,
extinction ratio, and finesse can be divided into geometric and material parameters. The
quantity γc ~ (K - Km)2 is a geometrical parameter. It characterizes the deviation from the con-
dition of full transmission of electromagnetic field from one of the adjacent MF segments
to another. This parameter is determined by the shape of the coupling MF segments and
their mutual position. The quantity γa ~ a is a material parameter because it characterizes
losses in the MF due to microscopic elastic and inelastic scattering of light. Currently, the
propagation losses achieved for the silica microfiber with diameter d ~ 1 µm is ~10 - 3 mm - 1
[13]. From Equation 7.9 the corresponding estimate of the intrinsic Q-factor is Qint~107.
K = K m + ( K 0 - K m )2 + ε 2 ,
s2
s2 s2 (7.12)
ε=
∫
s1
∫ ∫
∆β( s)sin 2 ds′k( s′) ds, K 0 = k( s)ds.
s s1
Equation 7.12 allows to estimate the reduction of the MLR Q-factor due to microfiber non-
uniformity. From this equation, the minimum possible deviation of the coupling parame-
ter, K, from its resonance value Km is ε. Assuming κ(s2–s1) ~ 1, ε ~ ∆β/κ ~ ∆dβ/κd, ∂β/∂λ ~ β/λ,
(where d is the MF diameter and ∆d is the characteristic diameter variation), and neglect-
ing losses, γ = 0, we have from Equation 7.9 [22]:
k 2 d 2S
Qint ~ . (7.13)
( ∆d)2 β
Setting the coupling length s2–s1 ~ κ - 1 ~ 10 µm, the MF diameter d ~ 1 µm, the loop length
S ~ 103 µm, the propagation constant β ~ 5 µm, and the relative diameter variation ∆d/d ~ 10 –3
over 10 µm, we have Qint~106. From this estimate, the MF uniformity is critical for achiev-
ing the high Q-factor. In certain situations, Equation 7.13 can significantly underestimate
the Q-factor. For example, if the coupling region is symmetric with respect to the plane
s = 0 then κ(s) is a symmetric and ∆β(s) is an antisymmetric function. Then, at resonance,
K0 = (π/2)(2m + 1), Equation 7.12 gives ε = 0. This means vanishing of γc in the considered
first-order approximation and possibility of much greater Q-factors.
Assume now that the diameters of adjacent MF segments and, hence, their propagation
constants are equal at point s = 0 (see Figure 7.6), i.e., ∆β(0) = 0, and diverge away from this
point. In this case, coupling can be studied using the Landau-Zener approximation devel-
oped in quantum mechanics [36]. If the coupling region is symmetric with respect to the
plane s = 0 then ∆β(0) = 0 and dκ/ds s = 0 = 0. In the immediate neighborhood of the point s = 0
we can set ∆β(s) = vs and κ(s) = κ(0). Then the coupling wave equation, Equation 7.1 can be
reduced to the parabolic cylinder differential equation. As a result, the coupling parameter
is determined by the following Landau-Zener formula [36,37]:
πk 2 (0)
K = arcsin 1 - exp - . (7.14)
ν
From this equation, if the propagation constant is an adiabatically slow varying function,
so that v << πκ2(0), then K is exponentially close to the resonant coupling parameter K = π/2.
In the opposite case, v >> πκ2(0), the coupling coefficient K is close to zero and the adjacent
MF segments are decoupled. Equation 7.14 is important for modeling of the MLR. It defines
the rate of MF tapering at which one can expect the proximity of the coupling parameter
to its resonance value.
0.5 mm
Figure 7.7
Optical microscope image of an MLR. (From Sumetsky, M., Dulashko, Y., Fini, J. M., Hale, A., and DiGiovanni,
D. J., IEEE J. Lightwave Technol., 24, 242–250, 2006.)
and output ends aligned parallel to each other. The parallel alignment of the adjacent MF
segments was possible due to surface attraction forces (Van der Waals and electrostatic),
which kept the ends together, apparently overcoming the elastic forces that would tend to
straighten out the microfiber. In Sumetsky et al. [22], the characteristic diameter of a MF
used for fabrication of a MLR was ~0.9 µm. For this MLR, the MF diameter variation in the
coupling region was slower than for the MLR of Ref. [21], which resulted in smaller losses,
higher coupling efficiency, and, eventually, in higher Q-factor of the MLR.
P = P1 + P2 (7.15)
where each of the transmission powers, P1 and P2 corresponds to one of the polarization
states, TM or TE. The transmission powers Pj in Equation 7.15 are defined by Equation 7.7.
The experimental spectrum was fitted with Equations 7.6 and 7.7 and averaged over the
interval of instrumental averaging equal to 0.01 nm. For the case of Figure 7.8c, an example
(a)
0 0 0
–2 –2 –2
–4 –4 –4
–6 –6 –6
–8 –8 –8
–2
–4
–6
–8
–10
0 0 0
–1 –1 –1
1522 1523 1544 1545 1546 1568 1569 1570
0
–1
1520 1530 1540 1550 1560 1570
Wavelength (nm)
(c) 0 0 0
Transmission power (dB)
–1 –1 –1
–2 –2 –2
1520.5 1521.0 1545.5 1546.0 1569.0 1569.5
0
–1
–2
1520 1530 1540 1550 1560 1570
Wavelength (nm)
Figure 7.8
Transmission spectra of MCR tuned to loaded Q-factor of (a) 22,000, (b) 98,000, and (c) 120,000. (From Sumetsky, M.,
Dulashko, Y., Fini, J. M., Hale, A., and DiGiovanni, D. J., IEEE J. Lightwave Technol., 24, 242–250, 2006.)
Experiment
Theory
Theory, smoothed over 0.01 nm
1
Transmission power, linear (a.u.)
0.9
0.8
0.7
Figure 7.9
Fitting the experimental data of Figure 7.8c near a resonance (solid) with the theoretical calculation (dashed)
averaged over the interval of instrumental averaging equal to 0.01 nm (dotted). (From Sumetsky, M., Dulashko, Y.,
Fini, J. M., Hale, A., and DiGiovanni, D. J., IEEE J. Lightwave Technol., 24, 242–250, 2006.)
of fitting the theoretical model to the experimental resonance peak is shown in Figure 7.9.
The best fit is achieved for the following parameters: α1 = 0.14, α2 = 0.19, sin(K1) = 0.981, and
sin(K 2) = 0.985. From the image of the MLR shown in Figure 7.7, the width of the coupling
region is ~10–100 µm; therefore the coupling coefficient κ~10 –1 – 10 –2 µm–1. It can be shown
that for the propagation constant β ~ 5 µm–1 and the characteristic length of its variation
Sc ~ 103 µm, the coupling parameter defined by Equation 7.14 can be away from the reso-
nance value as well as exponentially close to it. For the known MF diameter variation
(measured, e.g. with an SEM or with the more accurate method [16]), the accurate values
for the parameters entering Equation 7.14 can be calculated numerically.
Beam
chopper
MLR
Laser
beam
Figure 7.10
Periodic on/off heating of a MLR using a CO2 laser beam and a beam chopper. (From Sumetsky, M., Dulashko, Y.,
Fini, J. M., Hale, A., and DiGiovanni, D. J., IEEE J. Lightwave Technol., 24, 242–250, 2006.)
(a) 1550.4 nm
1550.12 nm
0
–1
–2
–3
1550.0 1550.5 1551.0 1551.5 1552.0
Transmission power (dB)
Wavelength (nm)
Figure 7.11
(a) Transmission spectrum of MLR used as a temperature sensor. Transmission power of MLR periodically
heated by CO2 laser as a function of time at wavelength 1550.4 nm (b) and 1550.12 nm (c). (From Sumetsky, M.,
Dulashko, Y., Fini, J. M., Hale, A., and DiGiovanni, D. J., IEEE J. Lightwave Technol., 24, 242–250, 2006.)
shown in Figure 7.11c. From the transient oscillations in Figure 7.11c, the magnitude of tem-
perature variation was estimated as 0.4 K, and the thermal equilibration time constant as
3 ms, in agreement with the theoretical prediction. For a regular optical fiber, which has a
diameter 100 times greater, this constant is, proportionally, 0.3 s. The measured tempera-
ture range in this experiment was ~1 K. The range can be increased by orders of magnitude
with decreasing Q and/or choosing a spectral region having smaller steepness. The tem-
perature resolution of the demonstrated MLR thermometer is determined by resolution of
the used optical spectrum analyzer and can be as small as ~0.1 mK.
(a)
50 µm
(b)
Coupler Taper
Taper
Knot
Figure 7.12
(a) SEM image of the twisted overlap region of a knot MLR with diameter of 290 µm. The MLR is fabricated from a
2.66-µm diameter MF. (b) Schematic diagram of a microfiber knot with two ends connected to standard fibers for
light launching and collection through fiber tapers. (From Jiang, X., Tong, L., Vienne, G., Guo, X., Tsao, A., Yang, Q.,
and Yang, D., Appl. Phys. Lett., 88, 223501, 2006.)
regular MLR in Ref. [22] (see Section 7.3.2.1.1). In addition, the transmission spectrum of a
regular MLR shown in Figure 7.8 is noticeably smoother and more uniform than that of a
knot MLR shown in Figure 7.13. Presumably, this is the result of twisting and polarization
mode coupling in the knot MLR. Nevertheless, a knot MLR is a much more robust device
than a regular MLR. This allows use of a knot MRL in contact with liquid and solid sur-
rounding medium. In particular, the authors of Ref. [28] demonstrate a knot MLR in water
with a Q-factor of 31,000. This indicates on opportunity of using a knot MLR as a microflu-
idic chemical and biological sensor. In addition, a knot MLR can be safely imbedded into
a UV curable polymer. After curing, the device is secured in a solid polymer matrix and
exhibits the resonant spectrum [31]. An optical microscope image of a knot MLR embed-
ded in fluoroacrylate from Ref. [31] is shown in Figure 7.14.
(a) (b) 0
0
–3
Transmission (dB)
Transmission (dB)
–2 –6
–9
0
–4
–12
–5
–15
–10
–6 1573.2 1573.6
–18
1650 1652 1654 1656 1658 1660 1568 1570 1572 1574 1576 1578 1580
Wavelength (nm) Wavelength (nm)
Figure 7.13
Transmission spectra of (a) a 396-µm diameter knot MLR assembled using a 2.66-µm diameter MF and (b) a
850-µm diameter knot MLR assembled using a 1.73-µm diameter MF. The inset shows a single resonance peak.
(From Jiang, X., Tong, L., Vienne, G., Guo, X., Tsao, A., Yang, Q., and Yang, D., Appl. Phys. Lett., 88, 223501, 2006.)
1 mm
Figure 7.14
Optical microscope image of a knot MLR embedded in fluoroacrylate. The MF diameter is 5 µm. (From Vienne, G.,
Li, Y., and Tong, L., IEICE Trans. Electron. E90, 415–421, 2007.)
500
400
Intensity (a.u.)
300
C
200
B
100
A
0
564 568 572 576 580 584
Wavelength (nm)
Figure 7.15
A Laser emission from a 350-μm diameter knot MLR fabricated from a 3.9-μm MF pumped by a pulsed
frequency-double Nd:YAG laser. Typical lasing groups centered around wavelengths 567 nm (Group A), 570 nm
(Group B) and 580 nm (Group C) are observed. (From Jiang, X., Song, Q., Xu, L., Fu, J., and Tong, L., Appl. Phys.
Lett., 90, 233501, 2007.)
mentioned in the beginning to this Chapter: smaller dimensions, simple low-loss input
and output connections, possibility of fabrication with feedback, smaller transmission loss
than that of a planar waveguide.
A simple case of an MCR is a higher-order MLR, determined as a coiled MF, which con-
sists of several independent MLRs as, e.g., shown in Figure 7.2b. The design of higher-order
MLRs is similar to that of all-pass ring resonators [35]. Alternatively, an MCR may be repre-
sented by a uniformly wrapped self-coupling coil shown in Figure 7.2d studied in Ref. [50].
Figure 7.2c illustrates an MCR wrapped on a nonuniform core studied in Refs. [51,52].
Currently, fabrication of MCRs is in its initial stage. The first experimental demonstra-
tion of an MCR containing more than one resonant loop was performed in Ref. [23] (see
also Ref. [24]). Independently, a MCR, which exhibited single loop resonances, were dem-
onstrated in Refs. [25,26].
A1 0 χ12 ( s) 0 ... 0 0 0 A1
A2 χ 21 ( s) 0 χ 23 ( s) ... 0 0 0 A2
... 0 χ 32 ( s) 0 ... 0 0 0 ...
d
Am = i ... ... ... ... ... ... ... Am (7.15)
ds
... 0 0 0 ... 0 χ M -1, M -22 ( s) 0 ...
A 0 0 0 ... χ M - 2 , M -1 ( s) 0 χ M -1, M ( s) AM -1
M -1
A 0 0 0 ... 0 χ M , M -1 ( s) 0 A
M M
s
s1
∫
χ pq ( s) = k pq ( s)exp i (β p ( s) - β q ( s))ds .
Here, S is the length of a single turn and coefficients Am(s) are defined at the turns s1 < s < s1 + S
and satisfy the continuity conditions:
s1 + S
Am+1 ( s1 ) = Am ( s1 + S)exp[i
∫ β (s)ds],
s1
m m = 1, 2..., M - 1. (7.16)
Using Equations 7.15 and 7.16 it is straightforward to calculate the resonance spectrum of
MCRs with interturn coupling changing along the length of the coiled MF. In Refs. [51,52]
these equations were generalized to the case of nonuniform MCR with changing radius of
turns. The transmission amplitude of an MCR can be calculated with Equations 7.15 and 7.16
numerically. Here we consider two situations when these equations allow analytical solu-
tion and full analysis. In the first case, an MCR is composed of M independent MLR. In the
second case the MCR is composed of a uniformly coiled MF with M turns.
M M
T= ∏m= 1
Tm , td = ∑t
m= 1
dm . (7.17)
Therefore, the design of the higher-order MLR generating the predetermined group delay
dependence is similar to the design of the all-pass ring resonators (see Ref. [35] and refer-
ences therein).
dA1
= k A2
ds
dAm
= k( Am-1 + Am+1 ), m = 1, 2 ,..., M - 1 (7.18)
ds
dAM
= k AM - 1
ds
where κ is the constant coupling coefficient between adjacent turns. Solution of Equations 7.18
and 7.16 is [50]:
(a) (b)
Figure 7.16
(a) Higher-order MLR and (b) uniform MCR.
Am ( s) = ∑a
n= 1
n1 Amn ( s), amn = amn -1
,
(
Amn ( s) = exp 2 iks cos
πn
M+1
sin) (
πmn
M+1 )
, m, n = 1, 2 ,..., M
∑
M
T = ( A1 (0))-1 an,1 AMn (S), (7.20)
n= 1
For the lossless MCR considered in this Section, |T| = 1. Equation 7.19 shows that the opti-
cal properties of MCR depend on three dimensionless parameters: the number of turns,
M, the dimensionless propagation constant, B = βS, and the dimensionless coupling param-
eter, K = κS. The parameters B and K, which define the eigenmodes of the MCR can be
determined from the condition td = ∝.
The transmission spectrum of a uniform MCR with two turns (N = 2) is similar to the
transmission spectra of the loop and ring resonators [11,34]. In this case, the eigenmodes
exist under the conditions of the full interturn light transition, κS = (π/2)(2m–1), and the
constructive interference of electromagnetic field at the adjacent turns, βS = (π/2)(4n–2m + 1),
where m and n are integers. If these conditions are met, the light in the input and output
waveguides of the MCR is completely trapped by the MCR. During a roundtrip, the light
transfers from one turn to another 2m–1 times and fully returns to the original turn with
the same phase.
For the uniform MCR with M > 2, more than two turns are coupling simultaneously.
This leads to a complex interplay of propagation along the MF and propagation through
interturn coupling. Under certain conditions, the MCR with M > 2 possesses eigenmodes.
Figure 7.17 shows the surface relief of the time delay td(B,K) for the MCRs with differ-
ent numbers of turns, M. Because td(B + 2π,K) = td(B,K), only one full period of td(B,K),
2πn < B < 2π(n + 1), where n is an integer, is shown. It is seen that the shape of the MCR spec-
trum qualitatively depends on the value of K. The top row of plots in Figure 7.17 shows that
the spectrum behavior is periodic as a function of coupling parameter, K, only for M = 2
and M = 3. For M ≥ 4, the dependence on coupling parameter, K, is no longer periodic.
Notice the interesting evolution of the group delay spectrum with growth of M, which is
shown in the lower rows of plots in Figure 7.17. All features of the spectrum including the
eigenvalues, which appear for certain number of turns M in the lower row, do not disap-
pear for larger M. These features move along the straight axial lines towards the point of
spectral collapse, (Bc,Kc) = (π(2n – 1/2),1/2), where n is an integer. The features shrink pro-
portionally to their distance from the collapse point and fully collapse at (Bc,Kc).
15
10
Coupling parameter, K = κS
Fractal collapse
2.0
1.5
1.0
0.5
K=0.5
2πn
2π(n+1)
2πn
2π(n+1)
2πn
2π(n+1)
2πn
2π(n+1)
2πn
2π(n+1)
2πn
2π(n+1)
2πn
2π(n+1)
2πn
2π(n+1)
Dimensionless propagation constant, B = βS
Figure 7.17
Surface plots of the time delay in the plane (B,K) for the number of turns M equal to 2 through 7, 10, and 20.
Brighter points correspond to larger time delay. For M = 2, 3, 4, and 5, the points corresponding to MCR eigen-
modes are marked by black dots. Upper row of plots show the time delay for 0 < K < 20, while the lower plots
show the time delay for 0 < K < 2.5 with higher resolution. In the lower plots, the circles mark similar features
(V-shaped and W-shaped). For M → ∞, all similar features tend to the collapse point. The ordinate of collapse
points, K = 0.5, is marked by a dotted line. (From Sumetsky, M., Opt. Express, 13, 4331–4340, 2005.)
Then Equation 7.18 is solved in the form Am( ± k ) ( s) = exp[2 ik cos(ξS)s ± iξSm], where the inte-
ger m is a number of a turn and ξ is an effective propagation constant. The continuity con-
dition, Equation 7.16, applied to this solution yields the dispersion relation [50]:
c
ω(ξ ) = ( ξ - 2k cos(ξS)) (7.21)
nf
where ω = cβ/nf is the frequency of electromagnetic field, c is the speed of light, and nf
is the effective refractive index of the MF. It can be found from Equation 7.21 that the
MCR transmission line does not have stop bands. Three qualitatively different situations
occur depending on the value of coupling parameter, K. If K < 1/2, the dispersion relation
is monotonic as shown in Figure 7.18a1. In this case the dispersion relation is qualitatively
similar to that of a SCISSOR shown in Figure 7.18c [34]. The crossover situation occurs
at K = 1/2. The corresponding dispersion relation is shown in Figure 7.18a2. In this case,
function ω(ξ) has inflection points at ξ n = π(2 n - 21 )/S, where n is an integer. In the vicinity
of these points, ω(ξ) ≈ ω(ξn) + α(ξ - ξn)3, and the group velocity is vanishing simultaneously
with the inverse group velocity dispersion. Similar phenomenon has been predicted for
photonic crystal waveguides [53]. A pulse in the neighborhood of these points experiences
dramatic distortion. Finally, for the case of strong coupling illustrated in Figure 7.6a3,
π 2m 2
π 2m 7 2π(m–1)
2
π 2m 9 a3 K = 1.5 2π(m+1)
2
π 2m 7
2
π 2m 3
2
1 2πm
π 2m 2
π 2m 5
2
π 2m 7 2π(m–1)
2
π 2m 5 π 2m 1 π 2m 3 π 2m 7
2 2 2 2
Dimensionless propagation constant, B
2π(m–2) 2π(m–1) 2πm 2π(m+1) 2π(m+2)
Dimensionless propagation constant, B
Figure 7.18
Comparison of characteristic dispersion relations for an MCR, a CROW, and a SCISSOR. (a) Dispersion relations
for an MCR with the coupling parameter K = 0.25 (a1), K = 0.5, (a2), and K = 1.5 (a3); (b) dispersion relation of a
CROW; (c) dispersion relation of a SCISSOR. (From Sumetsky, M., Opt. Express, 13, 4331–4340, 2005.)
when K > 1/2, the dispersion relation has the minima and maxima points. At these points,
the group velocity is equal to zero. In the neighborhood of these points, propagation of
light is similar to propagation near the band edges of photonics crystals and CROWs [34],
which is illustrated in Figure 7.18b. However, in contrast to the CROW, the MCR is an all-
pass and no bandgap structure.
fiber. A 345 µm diameter silica fiber was used as an optical rod. The taper was connected
to a broadband light source and to an optical spectrum analyzer, which controlled the MF
transmission loss. The transmission spectrum of the free taper is shown in Figure 7.20,
curve 1. Before wrapping, the silica rod was oriented perpendicular to the MF and put in
contact with it. At this step, the transmission spectrum of the MF exhibited whispering
gallery modes resonances generated in the rod, shown in Figure 7.20, curve 2. Then, the
MF was wrapped on a silica rod using the rotation and 3D translation stages. In the process
of wrapping performed by a rotation stage, the angle between the MF and the rod was
Optical
spectrum
Broadband
analyzer
light source
otatio
R
345 µm diameter
optical fiber
Axis of
rotation
Fiber taper
3D
Translation
Figure 7.19
Illustration of the setup for fabrication of an MCR in air. (From Sumetsky, M., IEEE J. Lightwave Technol., 26, 21–27,
2008.)
0 1
–5 2
–10
3
Transmission power (dB)
–15
–20
4
–25
–30
–35
1 – free MF
–40 2 – MF touches the rod
3 – Coiled MF
–45
4 – Critical coupling
–50
1534 1536 1538 1540 1542 1544 1546 1548 1550 1552
Wavelength (nm)
Figure 7.20
Transmission spectrum of an MCR fabricated in air. (1) A free MF; (2) a straight MF in contact with the optical
rod; (3) an MCR tuned to the highest Q; (4) an MCR tuned to the highest extinction ratio. (From Sumetsky, M.,
IEEE J. Lightwave Technol., 26, 21–27, 2008.)
controlled by a 3D translation stage. This angle was gradually increased so that it achieved
~900 when the uniform MF part touched the rod (see Figure 7.19). This happened after ~2
turns were made. The angle between the MF and the rod was tuned in order to form a
MF closed loop and to arrive at the transmission spectrum resonances with the greatest
Q-factor and extinction ratio shown in Figure 7.20, curves 3 and 4. In order to reduce the
insertion losses outside the loop section, it was necessary to wrap an MF adiabatically. In
particular, the first contact point between the MF and the rod should be at a thicker section
of the fiber waist. In addition, the MF should be slowly (adiabatically) reduced in thickness
in continuous contact with the rod down to the radius of the MF loop. In the described
experiment, these conditions were not fully achieved. Consequently, significant insertion
loss was observed. In Figure 7.20, curve 3 is the transmission spectrum of the MF after it
was wrapped on the rod and the resonant loop was aligned to maximize the resonance
Q-factor of the device. The Q-factor ~20000 has been achieved. The insertion loss of the fab-
ricated MCR was ~9 dB, which was primarily generated outside the resonant loop. The loss
was caused by discontinuities in the contact between the rod and the MF, contamination of
the silica surface, and nonadiabaticity of wrapping. Curve 4 in Figure 7.20 is the transmis-
sion spectrum of the same MF after it was aligned in order to arrive at the condition of criti-
cal coupling (γa = γc in Equation 7.7), which maximized the extinction ratio of transmission
power oscillations. The achieved extinction ratio was greater than 20 dB.
Fabrication of an MCR on a central rod with the refractive index lower than that of an
MF may be more preferable because coupling between the MF mode and the rod modes
can be better suppressed. In Ref. [25] an MCR was created by wrapping a silica MF on a
Teflon AF rod with refractive index 1.29. The maximum achieved Q-factor of MCR was
~10,000, i.e. still less than that achieved for a silica rod. Potentially, the obtained Q-factor
can be increased by improving the smoothness and uniformity of the low-index rod.
Thus, to date, fabrication of MCRs in air encountered several problems resulting in con-
siderable propagation loss. The loss is introduced by contamination of the MF and rod
surfaces and by discontinuities of the MF-rod contact line. It is believed that further devel-
opment of MCR fabrication methods in air will allow to improve its Q-factor and reduce
the insertion loss.
Broadband Optical
light spectrum
source analyzer
Fiber taper
Fiber spring
Microfiber coil
3D Translation
stage
Figure 7.21
Illustration of a setup for the fabrication of an MCR in index-matching liquid. (From Sumetsky, M., IEEE J.
Lightwave Technol., 26, 21–27, 2008.)
propagates along the MF. The MF was wrapped onto the rod by rotating the shaft and the
Γ-shaped leg. Fabrication of the MCR was performed in several steps. After the MF was
immersed into the liquid polymer, the transmitted power decreased from ~ -0.5 dB in free
space to ~ -1 dB. (curves 1 and 2 in Figure 7.22). This was caused by scattering of light at
the meniscus surface. Next, the MCR was created by wrapping the MF around the rod. In
the process of wrapping of the first turn, no noticeable transmission losses were observed.
Curves 3 and 4 in Figure 7.22 show the transmission spectrum of the taper at rotation
angles 180° and 360°, respectively. At rotation angle of 360° and greater, the relative posi-
tion of MF ends were manipulated to create a closed loop and to initiate self-coupling. The
self-coupling showed itself for the first time at ~ 450° when oscillations in transmission
spectrum appeared (curve 5 in Figure 7.22). At larger angles, a more pronounced resonant
spectrum was observed. Curve 6 in Figure 7.22 shows the resonant spectrum exhibiting
the Q-factor of 61000, observed at 720°, i.e., for the two-turn MCR. From curve 6, which was
fitted using Equation 7.7, the total loss of the resonant loop was estimated to be ~ 0.06 dB
[24]. The resonance structure of this spectrum corresponded to a double loop MCR. This
conclusion was based on the fact, that the group of four peaks, which appears periodically
in curve 6, corresponds to two resonances of different polarization of the first and second
loop of the MCR.
0 Free
Immersed
180°
360°
–0.5 1
–1 3
Transmission power (dB)
0 450°
720°
–1
5
6
–2
–3
1530 1531 1532 1533 1534 1535
Wavelength (nm)
Figure 7.22
Transmission spectrum of the MF in the process of MCR fabrication: (1) in air; (2) in liquid polymer; (3) at 180° of
rotation; (4) at 360° of rotation; (5) at 450° of rotation; (6) at 720° of rotation. (From Sumetsky, M., IEEE J. Lightwave
Technol., 26, 21–27, 2008.)
demonstrated the MCR refractometric sensor that was created from an embedded MCR
by removing its supporting rod. This device is an advanced version of an optical liquid
ring resonator sensor, composed of an ultra-thin microcapillary coupled to an MF imbed-
ded into a cured polymer [49]. The sensor was fabricated as follows. First, a 50-mm length
and 2.5 µm diameter MF was fabricated. Then, the MF was wrapped on a 1 mm diam-
eter polymethylmethacrylate (PMMA) rod. The PMMA refractive index was in the range
of 1.49–1.51. The whole structure was repeatedly coated by the low-index Teflon solution
601S1-100-6 (DuPont, USA), which was dried in air. Next, the dried embedded MCL was
left soaking into acetone to dissolve the support rod. Finally, the microfluidic MCR sensor
with a ~1 mm diameter microchannel was fabricated. The sensor consisted of a MCR with
five turns and had a channel inside as illustrated in Figure 7.23. The MCR was very close
to the channel. The sensitivity of this device was measured by inserting the sensor in a
beaker containing mixtures of isopropyl and methanol with ratios 60%, 61.5%, 63%, 64.3%,
65.5%, 66.7%, and 67.7%. The refractive indexes of pure isopropyl and methanol at 1.5 µm
are 1.364 and 1.317, respectively. Figure 7.24 shows the spectra recorded at 1530 nm. The
resonator peak shifted towards longer wavelengths with growth of the refractive index.
The calculated sensitivity of this MCR sensor was ~40 nm/RIU, which is comparable to the
sensitivity ~60 nm/RIU obtained for the embedded ring resonator sensor in Ref. [49].
Teflon Teflon
Analyte
Figure 7.23
The cross section of an MCR microfluidic sensor. (From Xu, F., and Brambilla, G., Appl. Phys. Lett., 92, 101126,
2008.)
–2
Output (dBm)
–4
1
2
–6 3
4
5
–8 6
7
Figure 7.24
Output spectrum of the MCR microfluidic sensor in mixtures of isopropyl and methanol. The isopropyl
fraction is (1) 60%; (2) 61.5%; (3) 63%; (4) 64.3%; (5) 65.5%; (6) 66.7%, and (7) 67.7%, respectively. (From Xu, F.,
and Brambilla, G., Appl. Phys. Lett., 92, 101126, 2008.)
7.5 Conclusion
In this Chapter, a novel platform for the fabrication of photonic circuits based on MF
assembly in air and in an optical medium was reviewed. The MLR and MCR fabrica-
tion technique developed to date is a significant step toward controllable manufacture of
MF devices, e.g., delay lines, filters, dispersion compensators, miniature lasers, and sen-
sors. The described approaches of MLR and MCR manufacturing are very simple and far
from perfection. Nevertheless, the initial results obtained in theory, fabrication, and first
applications of MLRs and MCRs are promising. It is feasible that further development of
the theory of MF-based devices and progress in their fabrication will lead to the creation
of robust, low-loss, and highly compact MF photonic circuits with numerous potential
applications.
References
1. N. S. Kapany. High-resolution fibre optics using sub-micron multiple fibres. Nature, 184, 881–
883, 1959.
2. N. S. Kapany and J. J. Burke. Fiber optics. IX. Waveguide effects. J. Opt. Soc. Am., 51, 1067–1078,
1961.
3. C. Caspar and E.-J. Bachus. Fibre-optics micro-ring-resonator with 2 mm diameter. Electron.
Lett., 25, 1506–1508, 1989.
4. H. S. Mackenzie and F. P. Payne. Evanescent field amplification in a tapered single-mode fibre.
Electron. Lett., 26, 130–132, 1990.
5. G. J. Pendock, H. S. MacKenzie, and F. P. Payne. Dye lasers using tapered optical fibers. Appl.
Opt., 32, 5236–5242, 1993.
6. W. M. Henry and F. P. Payne. Solid-state tapered optical fibre devices. J. Opt. Quantum Electron.,
27, 185–191, 1995.
7. J. C. Knight, G. Cheung, F. Jacques, and T. A. Birks. Phase-matched excitation of whispering-
gallery mode resonances by a fiber taper. Opt. Lett., 22, 1129–1131, 1997.
8. J. Bures and R. Ghosh. Power density of the evanescent field in the vicinity of a tapered fiber. J.
Opt. Soc. Am. A, 16, 1992–1996, 1999.
9. T. A. Birks, W. J. Wadsworth, and P. St. J. Russell. Supercontinuum generation in tapered fibers.
Opt. Lett., 25, 1415–1417, 2000.
10. L. Tong, R. R. Gattass, J. B. Ashcom, S. He, J. Lou, M. Shen, I. Maxwell, and E. Mazur.
Subwavelength-diameter silica wires for low-loss optical wave guiding. Nature, 426, 816–819,
2003.
11. M. Sumetsky. Optical fiber microcoil resonator. Opt. Express, 12, 2303–2316, 2004.
12. G. Brambilla, V. Finazzi, and D. J. Richardson. Ultra-low-loss optical fiber nanotapers. Opt.
Express, 12, 2258–2263, 2004.
13. S. G. Leon-Saval, T. A. Birks, W. J. Wadsworth and P. St. J. Russell. Guidance properties of low-
contrast photonic bandgap fibres. Opt. Express, 12, 2864–2869, 2004.
14. M. Sumetsky, Y. Dulashko, and A. Hale. Fabrication and study of bent and coiled free silica
nanowires: Self-coupling microloop optical interferometer. Opt. Express, 12, 3521–3531, 2004.
15. G. Brambilla, F. Koizumi, X. Feng, and D. J. Richardson. Compound-glass optical nanowires.
Electron. Lett. 41, 400–402, 2005.
16. M. Sumetsky, Y. Dulashko, J. M. Fini, A. Hale, and J. W. Nicholson. Probing optical microfiber
nonuniformities at nanoscale. Opt. Lett., 31, 2393–2395, 2006.
17. M. Sumetsky. How thin can a microfiber be and still guide light? Opt. Lett., 31, 870–872, 2006.
18. M. Sumetsky. How thin can a microfiber be and still guide light? Errata, Opt. Lett., 31, 3577–
3578, 2006.
19. M. Sumetsky. Optics of tunneling from adiabatic nanotapers. Opt. Lett., 31, 3420–3422, 2006.
20. M. Sumetsky, Y. Dulashko, P. Domachuk, and B. J. Eggleton. Thinnest optical waveguide:
experimental test. Opt. Lett., 32, 754–756, 2007.
21. M. Sumetsky, Y. Dulashko, J. M. Fini, and A. Hale. Optical microfiber loop resonator, Appl. Phys.
Lett., 86, 161108, 2005.
22. M. Sumetsky, Y. Dulashko, J. M. Fini, A. Hale, and D. J. DiGiovanni. The microfiber loop resona-
tor: theory, experiment, and application. IEEE J. Lightwave Technol., 24, 242–250, 2006.
23. M. Sumetsky, Y. Dulashko, and M. Fishteyn. Demonstration of a multi-turn microfiber coil
resonator. In Proceedings of Optical Fiber Communication Conference, Anaheim, CA, 2007.
Postdeadline papers, Paper PDP46.
24. M. Sumetsky. Basic elements for microfiber photonics: micro/nanofibers and microfiber coil
resonators. IEEE J. Lightwave Technol., 26, 21–27, 2008.
25. X. Fei and G. Brambilla. Manufacture of 3-D microfiber coil resonators. IEEE Photon. Technol.
Lett., 19, 1481–1483, 2007.
26. F. Xu and G. Brambilla. Embedding optical microfiber coil resonators in Teflon. Opt. Lett., 32,
2164–2166, 2007.
27. L. M. Tong, J. Y. Lou, R. R. Gattass, S. L He, X. W. Chen, L. Liu, and E. Mazur. Assembly of silica
nanowires on silica aerogels for microphotonic devices. Nano Lett., 5, 259–262, 2005.
28. X. Jiang, L. Tong, G. Vienne, X. Guo, A. Tsao, Q. Yang, and D. Yang. Demonstration of optical
microfiber knot resonators. Appl. Phys. Lett., 88, 223501, 2006.
29. X. Jiang, Q. Yang, G. Vienne, Y. Li, L. Tong, J. Zhang, L. Hu. Demonstration of microfiber knot
laser. Appl. Phys. Lett., 89, 143513, 2006.
30. X. Jiang, Q. Song, L. Xu, J. Fu, and L. Tong. Microfiber knot dye laser based on the evanescent-
wave-coupled gain. Appl. Phys. Lett., 90, 233501, 2007.
31. G. Vienne, Y. Li and L. Tong. Microfiber resonator in polymer matrix, IEICE Trans. Electron., E90,
415–421, 2007.
32. L. F. Stokes, M. Chodorow, and H. J. Shaw. All-single-mode fiber resonator. Opt. Lett., 7, 288–
230, 1982.
33. W. Snyder and J. D. Love. Optical Waveguide Theory. Chapman & Hall, London, 1983.
34. J. Heebner, T. Ibrahim, and R. Grover. Optical Microresonators: Theory, Fabrication, and Applications.
Springer-Verlag, New York, 2007.
35. G. Lenz, B. J. Eggleton, C. K. Madsen, and R. E. Slusher. Optical delay lines based on optical
filters. IEEE J. Quant. Electron., 37, 525–532, 2001.
36. L. D. Landau and E. M. Lifshitz. Quantum Mechanics. Pergamon Press, Oxford, 1977.
37. R. B. Smith. Analytic solution for linearly tapered directional couplers. J. Opt. Soc. Am., 66,
882–892, 1976.
38. C. K. Madsen, S. Chandrasekhar, E. J. Laskowski, M. A. Cappuzzo, J. Bailey, E. Chen, L. T.
Gomez, A. Griffin, R. Long, M. Rasras, A. Wong-Foy, L. W. Stulz, J. Weld, and Y. Low. An inte-
grated tunable chromatic dispersion compensator for 40 Gb/s NRZ and CSRZ. In Optical Fiber
Communication Conference, Anaheim, CA, 2002. Postdeadline papers, Paper FD9.
39. G. Bourdon, G. Alibert, A. Beguin, B. Bellman, and E. Guiot. Ultralow loss ring resonators using
3.5% index-contrast Ge-doped silica waveguides. IEEE Photon. Technol. Lett., 15, 709–711, 2003.
40. B. E. Little, S. T. Chu, P. P. Absil, J. V. Hryniewicz, F. G. Johnson, F. Seiferth, D. Gill, V. Van,
O. King, and M. Trakalo. Very high-order microring resonator filters for WDM applications.
IEEE Photon. Technol. Lett., 16, 2263–2265, 2004.
41. J. Niehusmann, A. Vörckel, P. H. Bolivar, T. Wahlbrink, W. Henschel, and H. Kurz. Ultrahigh-
quality-factor silicon-on-insulator microring resonator. Opt. Lett., 29, 2861–2863 , 2004.
42. K. J. Vahala. Optical microcavities. Nature, 424, 839–846, 2003.
43. B. E. Little, T. Chu, H. A. Haus. Second-order filtering and sensing with partially coupled tra-
veling waves in a single resonator. Opt. Lett., 23, 1570–1572, 1998.
44. E. Krioukov, D. J. W. Klunder, A. Driessen, J. Greve, and C. Otto. Integrated optical microcavi-
ties for enhanced evanescent-wave spectroscopy. Opt. Lett., 27, 1504–1506, 2002.
45. S. Ashkenazi, C.-Y. Chao, L. J. Guo, and M. O’Donnell. Ultrasound detection using polymer
microring optical resonator. Appl. Phys. Lett., 85, 5418–5420, 2004.
46. A. B. Matsko, V. S. Ilchenko. Optical resonators with whispering-gallery modes - Part I: Basics.
IEEE J. Sel. Top Quant. Electron., 12, 3–14, 2006.
47. V. S. Ilchenko, A. B. Matsko. Optical resonators with whispering-gallery modes - Part II:
Applications. IEEE J. Sel. Top Quant. Electron., 12, 15–32, 2006.
48. M. White, H. Oveys, and X. Fan. Liquid core optical ring resonator sensors. Opt. Lett., 31, 1319–
1321, 2006.
49. M. Sumetsky, R. S. Windeler, Y. Dulashko, and X. Fan. Optical liquid ring resonator sensor. Opt.
Express, 15, 14376–14381, 2007.
50. M. Sumetsky. Uniform coil optical resonator and waveguide: transmission spectrum, eigen-
modes, and dispersion relation. Opt. Express, 13, 4331–4340, 2005.
51. F. Xu, P. Horak, and G. Brambilla. Conical and biconical ultra-high-Q optical-fiber nanowire
microcoil resonator. Appl. Opt., 46, 570–573, 2007.
52. F. Xu, P. Horak, and G. Brambilla. Optimized design of microcoil resonators. J. Lightwave
Technol., 25, 1561–1567, 2007.
53. Figotin and I. Vitebskiy. Electromagnetic unidirectionality in magnetic photonic crystals. Phys.
Rev. B, 67, 165210, 2003.
54. F. Xu, P. Horak, and G. Brambilla. Optical microfiber coil resonator refractometric sensor. Opt.
Express, 15, 7888–7893, 2007.
55. F. Xu and G. Brambilla. Demonstration of a refractometric sensor based on optical microfiber
coil resonator. Appl. Phys. Lett., 92, 101126, 2008.
Contents
8.1 Introduction......................................................................................................................... 386
8.1.1 Background.............................................................................................................. 386
8.1.2 Optical Ring Resonator Biosensors...................................................................... 386
8.1.3 Opto-Fluidic Ring Resonator (OFRR) Biosensors.............................................. 388
8.2 Theoretical Analysis........................................................................................................... 389
8.2.1 Model........................................................................................................................ 389
8.2.2 Bulk Refractive Index Sensitivity (BRIS)............................................................. 390
8.2.2.1 Wall Thickness Dependence.................................................................. 391
8.2.2.2 Mode Number Dependence................................................................... 391
8.2.2.3 OFRR Size Dependence.......................................................................... 393
8.2.3 Noise Analysis......................................................................................................... 393
8.2.3.1 Thermally Induced Noise....................................................................... 393
8.2.3.2 Amplitude Noise...................................................................................... 394
8.2.3.3 Pressure Induced Noise.......................................................................... 395
8.2.3 Relation between BRIS and the Sensitivity to Molecule Binding.................... 396
8.2.4 Detection Limit....................................................................................................... 397
8.3 Experimental Investigations............................................................................................. 398
8.3.1 OFRR Fabrication.................................................................................................... 398
8.3.2 Experimental Setup................................................................................................ 399
8.3.3 Q-Factor of the OFRR.............................................................................................400
8.3.4 BRIS Characterization............................................................................................ 401
8.3.5 Characterization of Thermally-Induced Noise................................................... 401
8.3.6 Surface Activation................................................................................................... 402
8.3.7 Bio/chemical Molecule Detection......................................................................... 403
8.3.7.1 Protein Detection..................................................................................... 403
8.3.7.2 DNA Detection.........................................................................................405
8.3.7.3 Virus Detection........................................................................................ 407
8.3.7.4 Bacterium and Whole Cell Detection.................................................... 407
8.3.7.5 Pesticide Detection................................................................................... 409
385
© 2009 by Taylor & Francis Group, LLC
8.1 Introduction
8.1.1 Background
Optical label-free bio/chemical sensors are highly desirable for molecular analysis in many
applications such as disease detection, genomics, proteomics, environmental protection,
homeland security, and drug discovery. They enable detection of bio/chemical molecules
without the laborious fluorophore attachment processes that may interfere with the mol-
ecule’s bio/chemical functions, and allow for kinetics measurement in real time to provide
important insight into molecule–molecule interaction. A well-known example is the surface
plasmon resonance based bioanalytical instrument [1,2], as commercialized by Biacore, Inc.
[3], which combines advantageous label-free sensing and high detection sensitivity.
Optical waveguides can also be utilized as a label-free bio/chemical sensor [4–7]. A typi-
cal label-free waveguide bio/chemical sensor is illustrated in Figure 8.1A. The surface of
the waveguide is first immobilized with a layer of bio/chemical recognition molecules
such as antibodies. The guided light traveling along the waveguide has an evanescent
field extending into the surrounding medium (e.g., water) for approximately 100 nm and
is capable of interacting with the bio/chemical molecules near the waveguide surface. A
small change in refractive index (RI) near the waveguide surface when target analyte is
captured causes modifications in the optical sensing transduction signal such as phase or
intensity change at the output.
Waveguide based bio/chemical sensors have been in existence for over 20 years and
found applications in many fields [4]. However, this type of biosensor has a fundamental
limitation. In the waveguide bio/chemical sensor, the sensing signal accumulates along the
waveguide while interacting with the analyte. A longer light–analyte interaction length
results in a lower (i.e., better) detection limit. As a consequence, a waveguide as long as
a few centimeters is needed to detect trace amount of bio/chemical molecules, which has
significantly impaired the sensor multiplexing capability and tremendously increased the
sample consumption [4–7].
A B
Biorecognition WGM
molecule Analyte
Figure 8.1
(A) An optical biosensor based on a straight waveguide. (B) An optical ring resonator in which the light circu-
lates along the resonator surface.
repeatedly with the analytes on its surface through the WGM evanescent field, as illus-
trated in Figure 8.1B. In contrast to a straight waveguide based bio/chemical sensor, the
effective light–analyte interaction length of a ring resonator is no longer determined by the
sensor’s physical size, but rather by the number of revolutions of the light supported by
the resonator, which is characterized by the resonator quality factor, or the Q-factor. The
effective length Leff is related to the Q-factor by [51,54]:
Qλ
Leff = , (8.1)
2 πn
where λ is wavelength and n is the RI of the ring resonator. For example, for a ring
resonator with a Q-factor of 106, n = 1.45, and λ = 1550 nm, Leff can be as long as 17 cm.
Consequently, despite the small physical size of a resonator (10–200 µm in diameter), a
ring resonator can deliver sensing performance superior to a waveguide while using
orders of magnitude less surface area and sample volume. Furthermore, due to the small
size of ring resonators, high density sensor integration becomes possible. Optical ring
resonators have previously been implemented in the form of dielectric microspheres,
microcylinders, and microtoroids [8–17,19–28,46–50,55], as well as planar ring resonator
on a substrate [29–45].
Similar to all other optical label-free bio/chemical sensors, the ring resonator utilizes the
RI change to perform detection. The WGM spectral position, i.e., resonant wavelength, λ,
is related to the RI through the resonant condition [51]:
2 πRneff
λ= , (8.2)
m
where R is the ring outer radius, neff is the effective RI experienced by the WGM, and m is
an integer that describes the WGM angular momentum. neff changes when the RI near the
ring resonator surface is modified due to the capture of target molecules on the surface,
which in turn leads to a shift in the WGM spectral position, as illustrated in Figure 8.2.
Thus, by directly or indirectly monitoring the WGM spectral shift, it is possible to obtain
both quantitative and kinetic information about the binding of molecules near the surface.
This label-free sensing mechanism allows for the detection of bio/chemical molecules in
their natural form without the laborious fluorescent labeling process, as well as the detec-
tion of nonfluorescent chemical molecules.
A B
Binding
λ1 λ2 Time
Figure 8.2
(A) The WGM resonance shifts in response to the attachment of the target molecules to the sensor surface.
(B) Sensorgram is obtained by recording the WGM spectral position over time.
A B C D E
Low RI
Fluidic oil
cell
Figure 8.3
(A) Solid microsphere in a liquid well [9,19]. (B) Liquid droplets [51,58]. (C) Liquid droplet formed on an ultrahy-
drophobic surface. (D) Liquid droplet immersed in oil [65]. (E) Planar ring resonators [31,33].
A Analyte B
WGM
OFRR
Wav
egu ide
Figure 8.4
(A) Conceptual illustration of the OFRR. Analyte flows in the capillary and is captured by bio/chemical recog-
nition molecules immobilized on the OFRR interior surface. The WGM has an evanescent field in the core and
interacts with analyte in the core. A waveguide in contact with the OFRR is used to couple the light into (or out
of) the OFRR. (B) Conceptual illustration of an OFRR array.
AJ m ( km ,l n1r ) (r ≤ r1 )
( 1)
Em ,l (r ) = BJ m ( km ,l n2 r ) + CH m ( km ,l n2 r ) (r1 ≤ r ≤ r2 ), (8.3)
DH m(1) ( km ,l n3r ) (r ≤ r2 )
Core
Surrounding
medium
RI
r1 r2 r
Figure 8.5
Three-layer model.
where Jm and H m(1) are the mth Bessel function and the mth Hankel function of the first kind,
respectively. The RI of the core, wall, and the surrounding medium is described by n1, n2,
and n3. The terms r1 and r2 represent the inner and outer radius of the OFRR, respectively,
and km,l is the amplitude of the wave vector in vacuum for the lth order radial WGM. The
WGM has two polarization states with the electric field being along or perpendicular to
the OFRR longitudinal direction. Using this three-layer model, the WGM spectral position,
the radial distribution of the light, and radiation-limited Q-factor (in case that the OFRR is
immersed in a medium other than air) can be obtained as a function of wall thickness, the
OFRR size, and angular momentum of light, etc.
λ0
S= η, (8.4)
neff
where λ0 and neff are the wavelength in vacuum and the effective RI of the WGM, respec-
tively. η is the fraction of light energy in the core. When the WGM is mostly confined within
the wall
Wall
Core
Wall
1.0
Core
1.0
0.8 0.8
Intensity (A.U.)
Intensity (A.U.)
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
46 47 48 49 50 51 46 47 48 49 50 51
Radius (µm) Radius (µm)
Figure 8.6
Radial distribution of WGMs (m = 271, l = 2. Wall thickness = 3 µm) (A) and 2.0 µm (B) OD = 100 µm. Refractive
index: n1 = 1.33, n2 = 1.45, n3 = 1.0. λ~1550 nm. Fraction of light in the core: η = 3% (A) and 14.6% (B). Polarization
is along the OFRR longitudinal direction.
λ0
S= η,
nwall
Wall
Core
1.0 1.0
0.8 0.8
Intensity (A.U.)
Intensity (A.U.)
0.6 0.6
0.4 0.4
0.2 0.2
Core
Wall
0.0 0.0
46 47 48 49 50 51 45 46 47 48 49 50 51
Radius (µm) Radius (µm)
Figure 8.7
Radial distribution of WGMs. (A) m = 271, l = 2, λ = 1554.198 nm. (B) m = 256, l = 3, λ = l = 1554.527 nm. OD = 100
µm. Wall thickness = 3 µm. Refractive index: n1 = 1.33, n2 = 1.45, n3 = 1.0. Fraction of light in the core: η = 3% (A)
and 14.9% (B). Polarization is along the OFRR longitudinal direction.
Inner Outer
surface surface
1.0
0.8
Intensity (A.U.)
0.6
0.4
0.2
0.0
36 38 40 42 44 46 48 50 52
Radius (µm)
Figure 8.8
Radial distribution of a high order WGM. (A) m = 204, l = 13, λ = 1557.064 nm. OD = 100 µm. Wall thickness = 10 µm.
n1 = 1.33, n2 = 1.45, n3 = 1.0. η = 56.4%. Polarization is along the OFRR longitudinal direction.
providing a means to distinguish the target analytes captured by the recognition mol-
ecules from those in solution. Nevertheless, these high order modes can be employed for
refractometric measurement of solution and biosensing in free-solution that do not rely on
the binding to a substrate [89].
where α = 1/r ∂r/∂T is the OFRR thermal expansion coefficient and κwall(core) = ∂n/∂T is the
thermo-optic coefficient for the OFRR wall (core). The first term on the right side of
Equation 8.5 is the noise due to thermal expansion, and the second and the third term rep-
resent the respective noise induced by temperature induced refractive index change in the
wall and core. For a thick-walled OFRR, the amount of the light in the core is negligible.
Therefore, the thermal noise is mainly determined by the first two terms on the right side
of Equation 8.5, which leads to a temperature dependent WGM shift of approximately
8 pm/K (or 1 GHz/K) at 1550 nm for silica glass, as shown in Figure 8.10. When the OFRR
wall becomes thinner, the evanescent field of the WGM in the core, and hence the third term
on the right side of Equation 8.5, can not be ignored. Since the thermo-optic coefficient for the
core, which is typically water (or other organic solutions), is negative and large in magnitude
in comparison with that for the wall [93,94], the thermally induced noise is suppressed, as
demonstrated in Figure 8.10 with the 3.5 µm thick wall. Figure 8.10 further shows that at cer-
tain wall thickness, the water effect will completely counterbalance the thermal noise induced
by the wall, leading to elimination of the thermal noise to the first order. This phenomenon
provides a means for us to optimize the OFRR wall thickness to minimize the thermally
100
1 1E-3
0.1 1E-4
Figure 8.9
OFRR sensitivity and fraction of light in core as a function of the OFRR size for the second order WGM. Wall
thickness is kept constant at 4 µm. n1 = 1.33, n2 = 1.45, n3 = 1.0. λ~1550 nm. Dashed line shows the fraction of light in
the core in the limit of a straight waveguide of 4 µm wide. Polarization is along the OFRR longitudinal direction.
8
WGM spectral shift (pm)
6
7.5 µm wall
4
3.5 µm wall
2.6 µm wall
0
Figure 8.10
The WGM spectral response to temperature change. The OFRR is made of fused silica and filled with water.
OD (outer diameter) = 100 µm; αwall = 5 × 10 -7 K-1; κwall = 6.4 × 10 -6 K-1; κcore = -1 × 10 -4 K-1. Refractive index: n1
= 1.33; n2 = 1.45; n3 = 1. Wavelength: 1550 nm. Wall thickness is labeled on the figure. Zero thermal sensitivity
is achieved when the OFRR wall thickness is 2.6 µm. Polarization is along the OFRR longitudinal direction.
OFRR wall thickness is 2.6 µm. Polarization is along the OFRR longitudinal direction. (Reused from Fan, X.,
White, Ian M., Zhu, H.,. Suter, Jonathan D., and Oveys, H. Proc. SPIE, 6452, 64520M 1–20, 2007. © SPIE. With
permission.)
induced noise. Note that similar phenomenon was also theoretically studied by coating the
microspheres with a dielectric layer with a negative thermal-optic coefficient [95].
A B 1000
20 dB
1.2 100
40 dB
10
3σ noise (pm)
60 dB
1.0
Intensity (A.U.)
1 60 dB
0.1 80 dB
0.8
0.01
40 dB
0.6 1E-3
1E-4
0.4
–0.8 –0.4 0.0 0.4 0.8 1000 10000 100000 1000000 1E7 1E8
Wavelength (pm) Q-factor
Figure 8.11
(A) WGMs with Gaussian noise added for SNR of 40 dB and 60 dB (SNR = peak power of signal divided by vari-
ance of noise distribution). Curves are shifted vertically for clarity. (B) Results of Monte Carlo simulations over
a range of Q-factors and amplitude noise variances. Vertical axis represents the accuracy in WGM spectral peak
determination. WGM wavelength is 1550 nm. (Reused from White, Ian M. and Fan, X., Optics Express, 16, 2008.
© Optical Society of America. With permission.)
simulation plotted in Figure 8.11B shows the three standard deviations (3σ) value of the
determined maximum for signal-to-noise-ratio (SNR) ranging from 20 to 80 dB, and for
Q-factors between 108 and 103. The simulation results reveal that having a high Q-factor is
advantageous in reducing the spectral noise of the sensor, as the mode with narrower line
width filters the spectral noise more effectively. The simulation results show an empirical
linear dependence of the standard deviation on the linewidth and an exponential relation-
ship with the SNR [96]:
∆λ
s≈ , (8.6)
4.5(SNR 0.25 )
where σ is the standard deviation of the resulting spectral variation and ∆λ is the full-
width-half-max of the mode amplitude and is related to the Q-factor by Q = λ/∆λ. SNR is
in linear units in this expression. For Q = 106 and SNR = 106, 3σ = 0.03 pm at 1550 nm wave-
length, or 1/50 of the WGM linewidth. When SNR is increased to 108, 3σ becomes 0.01 pm,
or 1/150 of the WGM linewidth.
∆λ ∆r r ∆P
= = ⋅ , (8.7)
λ r t E
where E is the OFRR Young’s modulus. For fused silica used in OFRRs, E = 7x1010 Pa. For an
OFRR with r = 100 µm and t = 2 µm, Equation 8.7 becomes:
t
∆P
Figure 8.12
OFRR radius changes under differential pressure (∆P) between the core and the surrounding medium.
∆λ
≈ 10-9 ∆P(Pa). (8.8)
λ
In order to keep the pressure induced noise small in comparison with the spectral reso-
lution set by the WGM spectral linewidth (or the Q-factor), restrictions on the external
pressure that pushes the liquid sample through the OFRR and on the flow rate may
need to be imposed, especially when mechanical pumps such as peristaltic pumps are
used. For example, if the Q-factor is 106 and if we assume that the spectral resolution is
one hundredth of the WGM spectral linewidth, then the variation in differential pres-
sure should be kept below 10 Pa, which can easily be maintained during an experiment.
In addition, the pressure noise introduced by a mechanical pump has a well-defined
frequency (e.g., a few Hz for a peristaltic pump). Therefore, it can be removed rela-
tively easily during the data processing stage. Furthermore, when electro-kinetic pump
such as electro-osmosis method is used, the pressure induced noise can be completely
removed.
2 2
∆λ 2 π nwall - ncore nwall
= s pa ex 2 2 S, (8.9)
λ ε 0λ ncore
where σp is the surface density of the captured biomolecules and α ex is the excess polar-
izability of the molecule. ε0 is vacuum dielectric constant. Equation 8.9 shows that the
contribution of each molecule to the WGM spectral shift is weighted by the OFRR BRIS.
In addition, since the BRIS measurement is nondestructive and is easy to accomplish
(see Section 8.3.4. for experimental results on BRIS measurement), Equation 8.9 provides
a method to characterize the OFRR performance in bio/chemical molecule sensing
using BRIS. Note that Equation 8.9 is valid regardless of the WGM mode number and
polarization; it is also broadly applicable to other types of ring resonators based on
microspheres and planar ring resonators where exterior surface may be used for bio/
chemical sensing [70].
where δ is the system spectral resolution in units of pm. Figure 8.13 plots the bulk RI detec-
tion limit for various solvent absorption.
The detection limit for molecule density of the OFRR surface can be deduced using
Equations 8.9 and 8.10 [70]:
2
ncore ε 0λ
DLmolecule = DLbulk . (8.11)
nwall nwall - ncore πa ex
2 2 2
As discussed in Refs. [70,73,96,97], DLbulk is on the order of 10-7 RIU, leading to a detec-
tion limit for biomolecules (protein and DNA) on the order of 0.1–1 pg/mm2, which
is highly competitive as compared to the state-of-the-art label-free optical sensors
[1,3,9,12,30,31,33,34,98–104].
10–3
10–4
Detection limit (RIU)
10–5
η = 0.01
10–8
η = 0.05
η = 0.1
10–9
10–2 10–1 100 101 102 103 104 105
–1
α (m )
Figure 8.13
Calculated DL as a function of solvent absorption coefficient for η = 0.01, 0.05, and 0.1. For solid lines, only the
mode amplitude noise is considered. For the dashed lines, temperature-induced spectral fluctuation (0.01 pm)
is also considered. λ = 1550, Q 0 = 108, SNR = 60 dB. (Reused from Ian M. White and Xudong Fan, Optics Express
16, 1020–1028, 2008. © Optical Society of America. With permission.)
where R0, t0, and Vf are the original outer radius, wall thickness, and feed-in speed,
r espectively. R1, t1, and Vp are the final outer radius, wall thickness, and pulling speed,
respectively. If we assume that there is no surface tension induced capillary collapse
involved, the capillary aspect ratio, i.e., OD/t should remain constant. Therefore, by con-
trolling the ratio between the feed-in and pulling speeds, we are able to control the final
size of the OFRR and its wall thickness.
CO2 Laser
lasers micrometer
Feed Pull
Plastic enclosure
Figure 8.14
OFRR pulling station.
A Preform OFRR C
5 µm
Heating zone
B
Bulge-shaped OFRR
SEM image
5.0kV 15.5mm×1.10k SE(M) 7/2/07 50.0 µm
Figure 8.15
(A) OFRR made from a preform. (B) SEM image. (C) Bulge-shaped OFRR.
In practice, a certain degree of collapsing may occur due to glass surface tension, as
reported in the literature on capillary pulling [105–108], leading to a thicker wall, i.e., a lower
aspect ratio than the initial value. The collapsing effect can be mitigated by using a higher
pulling speed and/or lower heating temperature. In our experiment, we start with a com-
mercial fused silica capillary (Polymicro TSP530660, OD = 616 µm and wall thickness: t = 40
µm). After systematic investigation in pulling parameters, and quality check of the final size
and the wall thickness using an optical microscope, it is found that the original aspect ratio
can well be maintained after pulling, as shown in Figure 8.15B. In fact, over 95% of the origi-
nal aspect ratio was reported in a previous study [105]. In addition, OFRRs of various shapes
can also be fabricated by modulating the CO2 laser or pulling/feed speed during fabri-
cation (Figure 8.15C). The OFRR prepared in this manner have an OD of approximately
50–100 μm and wall thickness of 3–6 μm (see Figure 8.15B).
In order to achieve an even thinner wall, low concentration of hydrofluoric acid (HF)
is passed through the OFRR to slightly etch the glass. However, HF etching may result
in surface roughness and causes the Q-factor to degrade. Direct fabrication of the thin-
walled OFRRs is of great interest to preserve high Q-factors and to reduce fabrication time.
Recently, capillaries with sub-micron wall thickness walls have been achieved when the
preform is pressurized and tens of meters long thin-walled OFRR have also been fabri-
cated using a commercial draw tower [81,83].
A Detector # 1 B C
0.08 1.2
Detector #1
0.04
Detector #2 Optical
fiber taper
0.8
Detector #2
OFRR
D = 75 µm
0.00
–0.4 0.0 0.4
Sample WGM spectral position (pm)
OFRR
Taper
Tunable D = 4 µm
diode laser
Figure 8.16
(A) Experimental setup. (B) A picture of the LCORR in touch with an optical fiber taper. (C) Transmission signal
from Detector #1 and scattering signal from Detector #2 indicate the WGM spectral position.
A B
OFRR
Copper plate
Fiber taper
Tubing
OFRR Fiber
taper
Tunable diode laser
TEC
Heat sink
Copper
TEC
plate
Heat sink
Thermistor
Tubing
Figure 8.17
Top view (A) and side view (B) of the experimental setup with TEC.
light coupled into the OFRR is scattered off the OFRR surface and can be detected as a
spectral peak with a detector placed above the OFRR (Detector #2). Either of these mea-
sured signals can be used to indicate the WGM spectral position. While the first approach
is easier to implement, the second scheme is more suitable for an OFRR array.
As schematically shown in Figure 8.17, the OFRR is then fixed onto a copper plate (1 cm2)
that, in turn, sits on top of a thermo-electric cooler (TEC) unit (Marlow Industries, Inc.)
connected to a temperature controller (ILX Lightwave LDT-5910B). A thermister is embed-
ded in the copper plate 2 mm away from the coupling region with thermally conductive
epoxy (Arctic Silver). A custom fitted acrylic cap is placed on top of the copper plate to pro-
tect the system from air current disruption. Fluidic delivery is handled by a syringe pump
(Harvard Apparatus). The reagent is sucked through Tygon tubing from a 1-mL polyethyl-
ene source well, which can be conveniently filled and emptied via manual pipette during
the experiment. All experiments are performed at room temperature.
Q -1 = Qrad
-1 -1
+ Qwall -1
+ Qsca -1
+ Qsol , (8.13)
where Qrad, Qwall, and Qsca are the Q-factor determined by the radiation loss, loss in the wall
medium, and loss resulting from the surface scattering. Qsol = 2πn2 /ηλαsol, where αsol is the
absorption coefficient of the solvent. For low order WGMs, Qrad>>1011. Fused silica absorp-
tion and surface roughness induced losses have been investigated in detail with micro-
spheres, which show a Q-factor as high as 8x109 [54,109]. Since the OFRR uses the same
material and is fabricated under the similar heating condition (before HF etching) as for
a microsphere, it should have a Qwall and Qsca well above 109. For biosensing, the solvent is
basically water, which has low absorption at visible spectrum [110], therefore, the solvent
absorption induced Q-factor, Qsol should be in excess of 108 at visible spectrum. At a longer
wavelength (e.g., 1550 nm), Qsol suffers severely from the high absorption of water. Heavy
water, which has a much lower absorption, can be used to replace water for biosensing [28].
Although high Q-factor can be obtained upon fabrication, the Q-factor is degraded due to
HF etching, which results in a Q-factor on the order of 106. The best Q-factor experimentally
achieved is 107 at 690 nm [77]. Recent experimental results in OFRR laser (no HF etching
involved) studies suggest that the Q-factor of the OFRR can be on the order of 109 [111].
20
15
WGM spectral shift (pm)
1%
0.75%
16 10
0.25%
WGM spectral shift (pm)
5 0.1%
12
0
0 50 100 150 200 250 300
8
Time (s)
0
0 2 4 6
RI change (×10–4 RIU)
Figure 8.18
WGM spectral shift versus RI change in the core. Sensitivity is 22 nm/RIU. The inset is the sensorgram showing
various concentrations of ethanol passing through the OFRR.
A 200 B
Slope=7.6 10
100 0
Slope=5.4
R2=0.9955
–5
–10
0 5 10 15 20 25 0 20 40 60 80
∆T (K) Time (s)
Figure 8.19
(A) WGM shift in response to temperature change for thick-walled OFRR (square, wall thickness = 8 µm) and
thin-walled OFRR (triangle, wall thickness = 4 µm). Solid lines are simulation curves based on Equation 8.3.
(B) OFRR noise with thermo-electric cooler. σ = 0.0048 pm. Wavelength = 1550 nm. Wavelength scanning step
size = 0.004 pm. (Reused from Xudong Fan, Ian M. White, Hongying Zhu, Jonathan D. Suter, and Hesam Oveys,
Proc. SPIE, 6452, 64520M, 1–20, 2007. © SPIE. With permission.)
A Sample B
Sensing C
50
OFRR channel Sensing channel
Sensing 45
OFRR
channel 40
WGM shift (pm)
Laser 35 Reference
30
25
20
Reference 15
Reference Sensing channel-reference
channel channel 10
5
2 3 4 5
Time (min)
Figure 8.20
(A) Schematic of an OFRR with a reference channel; (B) A picture of such a setup; (C) Common-mode noise reduc-
tion by subtracting the reference from the actual sensing signal. (Reused from Xudong Fan, Ian M. White, Hongying
Zhu, Jonathan D. Suter, and Hesam Oveys, Proc. SPIE, 6452, 64520M 1–20, 2007. © SPIE. With permission.)
A Photo-patterned
biorecognition molecules
OFRR
wall Photo-mask
Figure 8.21
(A) Molecule patterning through photo-assisted molecular grafting. (B) Results were achieved with a single-
window photomask. The molecules are labeled with fluorescent dye molecules to guide the eye. (Reused from
Xudong Fan, Ian M. White, Hongying Zhu, Jonathan D. Suter, and Hesam Oveys, Proc. SPIE, 6452, 64520M 1–20,
2007. © SPIE. With permission.)
A 120 B 140
100 120
100
40 40 0.1
20 0.01
20 0.01 0.1 1 10 100 1000
0 BSA concentration (nM)
0
0 100 200 300 400 500 0 200 400 600 800 1000 1200 1400 1600
Time (s) BSA concentration (nM)
Figure 8.22
(A) Sensorgram for BSA binding. (B) WGM spectral shift vs.BSA concentration for the OFRR with BRIS of
31 nm/RIU. The LCORR starts to saturate at 200 nM and the saturation WGM shift is 129 pm. Half saturation
concentration: 8 nM. Inset: log–log scale. Solid line is the linear fit in log–log scale. (Reused from Hongying Zhu,
Ian M. White, Jonathan D. Suter, Paul S. Dale, and Xudong Fan, Optics Express, 15, 9139–9146, 2007. © Optical
Society of America. With permission.)
200
WGM spectral shift (pm)
150
100
50
0
0 5 10 15 20 25 30 35 40
BRIS (nm/RIU)
Figure 8.23
Saturation WGM spectral shift vs. OFRR BRIS. The solid curve is the theoretical calculation based on
Equation 8.9. n2 = 1.45, n3 = 1.333. σp = 2.9 × 1012 cm -2 and αex = (4πε0) × (3.85 × 10 -21) cm3. (Reused from Hongying
Zhu, Ian M. White, Jonathan D. Suter, Paul S. Dale, and Xudong Fan, Optics Express, 15, 9139–9146,
2007. © Optical Society of America. With permission.)
BSA molecules are removed from the OFRR. The remaining 20% permanently remains on
the OFRR surface due to the lack of available cleavage sites [21,70].
As one of practical applications, we use the OFRR for detection of breast cancer bio-
marker. Breast cancer is the most frequently diagnosed malignancy in women, ranking
second among cancer death in women, especially in Western countries [115]. As it has
enormous impact on women’s health, a lot of efforts might have been put into breast can-
cer research than any other malignancy. Early diagnosis and assess treatment prognosis
are very important for malignancy treatment. Tissue based assay test of estrogen receptor
(ER) and progesterone receptor (PR) are the most routine method for predicting hormone
100
60
40
20
0
0 400 800 1200 1600
Time (s)
Figure 8.24
BSA molecules bind to the OFRR surface, followed by the cleavage by trypsin. BSA concentration: 0.1 mg/mL.
Trypsin concentration: 0.1 mg/mL. Arrow indicates the time when trypsin is added. (Reused from Hongying
Zhu, Ian M. White, Jonathan D. Suter, Paul S. Dale, and Xudong Fan, Optics Express, 15, 9139–9146, 2007. © Optical
Society of America. With permission.)
dependence of breast cancer and assess prognosis. However there are very few mature
techniques for detecting the sensitive and specific circulating breast cancer biomarker.
Herein, we attempt to apply the OFRR for real-time, label-free, rapid detection of CA15-3,
one of the most widely used circulating cancer biomarker that is recommended for the
monitoring of the course and response to therapy of breast cancer [116,117]. During the
experiment, the OFRR surface is first immobilized with Anti-CA15-3 antibodies through
the glutaraldehyde. The unoccupied sites on the sensing surface are then blocked by casein
blocking buffer. After this, the OFRR is ready for CA15-3 measurement. Figure 8.25 plots
the OFRR’s response to various concentration of CA15-3, which covers the CA15-3 cut-off
point (between 20 and 40 Units/mL according to clinical studies [118]) that indicates the
presence of the circulating tumor cells in blood. The detection time for each concentration
takes only a few minutes. These results demonstrate the potential of the OFRR as a diag-
nostic instrument for rapidly detecting the cancer biomarker. For further development of
OFRR based device for cancer biomarker detection, real serum samples obtained from
patients will be used to test in the future.
1
10 100 1000
CA15-3 concentration (unit/mL)
Figure 8.25
Detection of CA 15-3 breast cancer marker. Dashed line represents 40 Units/mL.
A B
30
2.5 µM 25-mer
WGM spectral shift (pm)
10 µM 10
20
10
0 0
25 50 100 25 5 2 1 0
Strand length (bases) # of Mismatched bases
Figure 8.26
(A) WGM shift for various concentrations and strand length variables. (B) WGM shift for various strand mismatches.
Concentration: 10 µM. (Reused from Jonathan D. Suter, Ian M. White, Hongying Zhu, Huidong Shi, Charles W.
Caldwell, and Xudong Fan, Biosensors and Bioelectronics, 23, 1003–1009, 2008. © Elsevier. With permission.)
mismatched bases between probe and target DNA strands increases, the overall prob-
ability of binding decreases, resulting in a lower overall spectral shift. Figure 8.26B
shows about a 10% drop in signal for a 1-base mismatch, but results from the literature
suggest that this result could be improved to 90% depending on the experimental con-
ditions used [12].
Surface coverage is easily analyzed from the spectral response based on Equation 8.9
[70]. We estimate that the OFRR is capable of detecting a DNA surface coverage as small as
6.8 × 109 molecules/cm2 with 100-base targets [79]. This represents approximately 0.012% of
the theoretical maximum surface loading. Fractional coverage, however, is highly depen-
dent upon the size of the adsorbed molecule, as larger molecules interfere sterically with
each other, generally leading to reduced densities [79].
40
30
10
(4)
0 50 100 150
Time (s)
Figure 8.27
Sensorgrams for various 25-mer DNA target. (1) 100% complementary; (2) 2-base mismatch; (3) 5-base mismatch;
(4) completely mismatch. (Reused from Jonathan D. Suter, Ian M. White, Hongying Zhu, Huidong Shi, Charles W.
Caldwell, and Xudong Fan, Biosensors and Bioelectronics, 23, 1003–1009, 2008. © Elsevier. With permission.)
Figure 8.28 shows the net WGM shift in response to a series of target DNA concentrations
between 0.5 and 500 nM using a 6.8 nm/RIU OFRR, which is modeled using a Michaelis-
Menton type curve. The inset graph shows detection of 10 pM target using a more sensitive
OFRR. This demonstrates the ability of the sensor to detect extremely small concentrations
and proves that greater performance is possible if higher sensitivities are used.
10
72
Figure 8.28
Concentration dependent WGM shift for 25-mer strands. Inset shows the WGM shift from 10 pM tar-
get DNA using a 37 nm/RIU OFRR. (Reused from Jonathan D. Suter, Ian M. White, Hongying Zhu,
Huidong Shi, Charles W. Caldwell, and Xudong Fan, Biosensors and Bioelectronics, 23, 1003–1009, 2008.
© Elsevier. With permission.)
100
WGM spectral shift (pm)
10
1 2
10 103 104 105 106 107 108 109 1010 1011 1012
M13 concentration (pfu/mL)
Figure 8.29
WGM spectral shift measured for various concentrations of M13 phage in PBS buffer. Concentration varies from
2.3 × 103 to 2.3 × 1011 pfu/mL. (Reused from Hongying Zhu, Ian M. White, Jonathan D. Suter, Mohammed Zourob,
and Xudong Fan, Analyst, 132, 356–360, 2008. © The Royal Society of Chemistry. With permission.)
typically a few micrometers in size. Optical label-free detection of bacterium has been
achieved using SPR [120–122], waveguide [123,124], and photonic crystal sensors [102].
Recently, microspheres and planar ring resonators were also employed [45,125]. The detec-
tion limit ranges from 105 to 107 cfu/mL (cfu: colony-forming unit, it simply represents the
number of bacteria or cells). While the performance of the OFRR is similar to that of other
label-free optical sensor (the detection limit is on the order of 105 cfu/mL), it benefits from
3
108 cfu/mL E.Coli
2
Figure 8.30
Sensorgram for 108 cfu/mL E. coli running through the OFRR.
integrated microfluidics and the circular shape of the capillary for on-line rapid detection
of bacteria. Figure 8.30 shows an example of typical OFRR response to bacteria.
A 40 B
40
30
30
20
20
10
10 R2 = 0.9530
0
1 8 15 22 29 36 43
0
–10 4.E-11 4.E-10 4.E-09 4.E-08 4.E-07 4.E-06
Time (sec) Concentration (M)
Figure 8.31
(A) Sensorgram for pesticide (parathion-methyl) binding and subsequent removal from the OFRR interior sur-
face. (B) WGM shift for various concentrations of pesticide.
+ –
V A
Electrode
Fiber taper
Analyte
OFRR
Photodetector
Figure 8.32
Schematic of electrophoresis experimental setup.
A B 70
120
WGM spectral shift (pm)
80 60
Applied voltage (V)
40
50
0
200 40
–40
0
–80 30
–200
0 100 200 300 400 500 600 0 100 200 300 400
Time (s) Time (s)
Figure 8.33
(A) The WGM spectral shift shows that analyte solution is driven into and out of the OFRR when the voltage
supply switches polarity. (B) Electrophoretic separation between sucrose and ribose.
1/2
λ n2 λ2
t= (2 N + 1) 1 - 12 + 2 2 ( N = 0, 1, 2 ,…), (8.13)
4n2 n2 4n2 h
where λ is the laser wavelength in vacuum, n1 and n2 are the refractive index of the core
and high index layer, respectively, and h is the thickness of the core. Under this antireso-
nance condition, the ARROW prevents the light in the core from leaking into the substrate,
and in the meantime presents a sufficient evanescent field above the core for the coupling
between the ARROW and the OFRR. In an ARROW, only TE-polarized light will propagate
whereas the TM-mode is very lossy and can not be transmitted through the ARROW.
Similar to the coupling between the ARROW and a microsphere ring resonator that has
previously been achieved [129–131], the coupling between the ARROW and the OFRR can
be accomplished by placing the OFRR in touch with the ARROW channels, as shown in
A B OFRR
OFRR WGM
profile Sample
Fluid sample
WGM Guided
optical
field WGM
Arrow Core h
High index layer
Arrow
Figure 8.34
(A) OFRR-ARROW system. (B) OFRR cross section on top of an ARROW structure. Light polarization is along
the OFRR longitudinal direction. (Reused from Ian M. White, Hesam Oveys, Xudong Fan, Terry L. Smith, and
Junying Zhang, Applied Physics Letters, 89, 191106, 2006. © American Institute of Physics. With permission.)
A
Output fiber B
ARROW 1
C
ARROW 2
60
Coupling (%)
Intensity (A.U.)
ARROW 1
ARROW 2
ARROW 3
ARROW 4
40
ARROW 3
20
OFRR
0
5 µm 2.0 2.5 3.0 3.5 4.0 4.5 5.0
ARROW height (µm)
Silica ARROW 4
n = 1.45
h
SiO2, n = 2.0, 180 nm
SiO2, n = 1.45, 2 µm
Si water
–20 –10 0 10 20
Input fiber Relative wavelength (pm)
Figure 8.35
(A) Experimental setup for simultaneous excitation of the WGMs along an OFRR with an ARROW. The inset in
the picture shows the detail of the ARROW structure designed to for 980 nm. (B) WGM spectra at the ARROW
output. (C) Coupling depth for various ARROW heights (h). (Reused from Ian M. White, Hesam Oveys, Xudong
Fan, Terry L. Smith, and Junying Zhang, Applied Physics Letters, 89, 191106, 2006. © American Institute of Physics.
With permission.)
Figure 8.35A and B. Furthermore, Figure 8.35C shows that the coupling can be adjusted
by changing the ARROW core height [132,133]. Our results will enable the development of
integrated high throughput sensor arrays.
B
OFRR
WGM Sample
profile
A
Guided
optical
SiO2 core field WGM
Au cladding
SiO2 lower cladding layer
Si substrate Si substrate
Figure 8.36
(A) Cross-section of waveguide chip structure. (B) Ring resonator in contact with waveguide for WGM excita-
tion. (Reused from Ian M. White, Jonathan D. Suter, Hesam Oveys, Xudong Fan, Terry L. Smith, Junying Zhang,
Barry J. Koch, and Michael A. Haase, Optics Express, 15, 646–651, 2007. © Optical Society of America. With
permission.)
Acknowledgments
The authors acknowledge the financial support from the 3M Non-Tenured Faculty Award,
the University of Missouri Research Board Award (05-013), the University of Missouri
A B
1.0 1.0 3.0 µm WG
0.9 690 nm
0.9
0.8
Intensity (A.U.)
Intensity (A.U.)
0.7 0.8
0.6 0.7
0.5
1550 nm 0.6 2.5 µm WG
0.4
0.5 2.0 µm WG
0.3
–4 –2 0 2 4 –4 –3 –2 –1 0 1 2 3 4
Relative wavelength (pm) Relative wavelength (pm)
Figure 8.37
(A) WGMs in an OFRR of 140 µm in diameter for 690 nm and 1550 nm that are excited by a gold-clad waveguide
of 2.5 µm in height. (B) Lorentzian fits of recorded WGMs in OFRR excited by 1550 nm light coupled from gold-
clad waveguides of heights of 2.0, 2.5, and 3.0 µm. (Reused from Ian M. White, Jonathan D. Suter, Hesam Oveys,
Xudong Fan, Terry L. Smith, Junying Zhang, Barry J. Koch, and Michael A. Haase, Optics Express, 15, 646–651,
2007. © Optical Society of America. With permission.)
A B
Low RI polymer
1.0
Intensity (A.U.)
0.8
0.6
Q = 6×106
0.4
–40 –20 0 20 40
WGM spectral position (pm)
Figure 8.38
(A) OFRR-microfiber system is embedded in a low index polymer. (B) High Q-factors can still be obtained for
the system in (A).
References
1. J. Homola, S. S. Yee, and G. Gauglitz. Surface plasmon resonance sensors: review. Sensors and
Actuators B 54, 3–15, 1999.
2. C. L. Baird, and D. G. Myszka. Current and emerging commercial optical biosensors. Journal of
Molecular Recognition, 14, 261–268, 2001.
71. X. Fan, I. M. White, H. Zhu, J. D. Suter, and H. Oveys. Overview of novel integrated optical ring
resonator bio/chemical sensors (invited paper). SPIE Laser Resonators and Beam Control, 6452,
64520M, 2007.
72. I. M. White, J. D. Suter, H. Oveys, X. Fan, T. L. Smith, J. Zhang, B. J. Koch, and M. A. Haase.
Universal coupling between metal-clad waveguides and optical ring resonators. Optics Express,
15, 646–651, 2007.
73. H. Zhu, I. M. White, J. D. Suter, M. Zourob, and X. Fan. An integrated refractive index
optical ring resonator detector for capillary electrophoresis. Analytical Chemistry, 79, 930–
937, 2007.
74. J. D. Suter, I. M. White, H. Zhu, and X. Fan. Thermal characterization of liquid core optical ring
resonator sensors. Applied Optics, 46, 389–396, 2007.
75. I. M. White, H. Oveys, and X. Fan. Liquid core optical ring resonator sensors. Optics Letters, 31,
1319–1321, 2006.
76. I. M. White, H. Oveys, X. Fan, T. L. Smith, and J. Zhang. Integrated multiplexed biosensors
based on liquid core optical ring resonators and antiresonant reflecting optical waveguides.
Applied Physics Letters, 89, 191106, 2006.
77. S. I. Shopova, H. Zhu, X. Fan, and P. Zhang. Optofluidic ring resonator based dye laser. Applied
Physics Letters, 90, 221101, 2007.
78. H. Zhu, I. M. White, J. D. Suter, M. Zourob, and X. Fan. Opto-fluidic micro-ring resonator for
sensitive label-free detection. Analyst, 132, 356–360, 2008.
79. J. D. Suter, I. M. White, H. Zhu, H. Shi, C. W. Caldwell, and X. Fan. Label-free quantitative
DNA detection using the liquid core optical ring resonator. Biosensors and Bioelectronics, 23,
1003–1009, 2008.
80. S. I. Shopova, J. M. Cupps, P. Zhang, E. P. Henderson, S. Lacey, and X. Fan. Opto-fluidic
ring resonator lasers based on highly efficient resonant energy transfer. Optics Express, 15,
12735–12742, 2007.
81. M. Sumetsky, R. S. Windeler, Y. Dulashko, and X. Fan. Optical liquid ring resonator sensor.
Optics Express, 15, 14376–14381, 2007.
82. T. Ling and L. J. Guo. A unique resonance mode observed in a prism-coupled micro-tube reso-
nator sensor with superior index sensitivity. Optics Express, 15, 17424–17432, 2007.
83. V. Zamora, A. Díez, M. V. Andrés, and B. Gimeno. Refractometric sensor based on whispering
gallery modes of thin capillaries. Optics Express, 15, 12011–12016, 2007.
84. J. P. Landers. Handbook of Capillary Electrophoresis. Boca Raton, FL.
85. M. Cai, and K. Vahala. Highly efficient optical power transfer to whispering-gallery modes by
use of a symmetrical dual-coupling configuration. Optics Letters, 25, 260–262, 2000.
86. I. M. White, J. Gohring, and X. Fan. SERS-based detection in an optofluidic ring resonator plat-
form. Optics Express, 15, 17433–17442, 2007.
87. N. A. Mortensen, S. Xiao, and J. Pedersen. Liquid-infiltrated photonic crystals: enhanced light–
matter interactions for lab-on-a-chip applications. Microfluidics and Nanofluidics, 3, 1613–4982,
2007.
88. H.-J. Moon, and K. An. Interferential coupling effect on the whispering-gallery mode lasing in
a double-layered microcylinder. Applied Physics Letters, 80, 3250–3252, 2002.
89. D. J. Bornhop, J. C. Latham, A. Kussrow, D. A. Markov, R. D. Jones, and H. S. Sørensen. Free-
solution, label-free molecular interactions studied by back-scattering interferometry. Science,
317, 1732–1736, 2007.
90. M. K. Chin, and S. T. Ho. Design and modeling of waveguide-coupled single-mode microring
resonators. Journal of Lightwave Technology, 16, 1433–1446, 1998.
91. B. R. Johnson. Theory of morphology-dependent resonances: shape resonances and width for-
mulas. Journal of the Optical Society of America A., 10, 343–352, 1993.
92. I. M. White, J. Gohring, Y. Sun, G. Yang, S. Lacey, and X. Fan. Versatile waveguide-coupled
optofluidic devices based on liquid core optical ring resonators. Applied Physics Letters, 91,
241104, 2007.
93. D. R. Linde. The CRC Handbook of Chemistry and Physics. Boca Raton, FL, 2005.
94. J. P. Longtin, and C.-H. Fan. Precision laser-based concentration and refractive index measure-
ment of liquid. Microscale Thermophysical Engineering, 2, 261–272, 1998.
95. M. Han, and A. Wang. Temperature compensation of optical microresonators using a surface
layer with negative thermo-optic coefficient. Optics Letters, 32, 1800–1802, 2007.
96. I. M. White, and X. Fan. On the performance quantification of resonant refractive index sensors.
Optics Express, 16(2), 1020–1028, 2008.
97. I. M. White, H. Zhu, J. D. Suter, X. Fan, and M. Zourob. Label-free detection with the liquid
core optical ring resonator sensing platform. In A. Rasooly and K. E. Herold, Biosensors and
Biodetection: Methods and Protocols, Volume 1: Optical-Based Detectors (Methods in Molecular
Biology) Humana Press, New York.
98. B. Cunningham, P. Li, B. Lin, and J. Pepper. Colorimetric resonant reflection as a direct bio-
chemical assay technique. Sensors and Actuators B, 81, 316–328, 2002.
99. B. Cunningham, J. Qiu, P. Li, and B. Lin. Enhancing the surface sensitivity of colorimetric reso-
nant optical biosensors. Sensors and Actuators B, 87, 365–370, 2002.
100. B. T. Cunningham, P. Li, S. Schulz, B. Lin, C. Baird, J. Gerstenmaier, C. Genick, F. Wang, E. Fine, and
L. Laing. Label-free assays on the BIND system. Journal of Biomolecular Screening, 9, 481–490, 2004.
101. L. L. Chan, B. T. Cunningham, P. Y. Li, and D. Puff. A self-referencing method for microplate
label-free photonic-crystal biosensors. Sensors Journal, IEEE, 6, 1551, 2006.
102. B. Lin, P. Li, and B. T. Cunningham. A label-free biosensor-based cell attachment assay for char-
acterization of cell surface molecules. Sensors and Actuators B: Chemical, 114, 559–564, 2006.
103. M. M. Varma, D. D. Nolte, H. D. Inerowicz, and F. E. Regnier. Spinning-disk self-referencing
interferometry of antigen–antibody recognition. Optics Letters, 29, 950–952, 2004.
104. M. M. Varma, H. D. Inerowicz, F. E. Regnier, and D. D. Nolte. High-speed label-free detection
by spinning-disk micro-interferometry. Biosensors and Bioelectronics, 19, 1371–1376, 2004.
105. A. D. Fitt, K. Furusawa, T. M. Monro, and C. P. Please. Modeling the fabrication of hollow
fibers: Capillary drawing. Journal of Lightwave Technology, 19, 1924–1931, 2001.
106. S. C. Xue, R. I. Tanner, G. W. Barton, R. Lwin, M. C. J. Large, and L. Poladian. Fabrication of
microstructured optical fibers-part I: problem formulation and numerical modeling of transient
draw process. Journal of Lightwave Technology, 23, 2245–2254, 2005.
107. S. C. Xue, R. I. Tanner, G. W. Barton, R. Lwin, M. C. J. Large, and L. Poladian. Fabrication
of microstructured optical fibers—Part II: numerical modeling of steady-state draw process.
Journal of Lightwave Technology, 23, 2255–2266, 2005.
108. S. C. Xue, M. C. J. Large, G. W. Barton, R. I. Tanner, L. Poladian, and R. Lwin. Role of material
properties and drawing conditions in the fabrication of microstructured optical fibers. Journal
of Lightwave Technology, 24, 853–860, 2006.
109. D. W. Vernooy, V. S. Ilchenko, H. Mabuchi, E. W. Streed, and H. J. Kimble. High-Q measure-
ments of fused-silica microspheres in the near infrared. Optics Letters, 23, 247–249, 1998.
110. L. Kou, D. Labrie, and P. Chylek. Refractive indices of water and ice in the 0.65- to 2.5-Mum
spectral range. Applied Optics, 32, 3531–3540, 1993.
111. S. Lacey, I. M. White, Y. Sun, S. I. Shopova, J. M. Cupps, P. Zhang, and X. Fan. Versatile opto-
fluidic ring resonator lasers with ultra-low threshold. Optics Express, 15, 15523–15530, 2007.
112. F. S. Ligler, M. Breimer, J. P. Golden, D. A. Nivens, J. P. Dodson, T. M. Green, D. P. Haders, and
O. A. Sadik. Integrating waveguide biosensor. Analytical Chemistry, 74, 713–719, 2002.
113. M. Y. Balakirev, S. Porte, M. Vernaz-Gris, M. Berger, J.-P. Arie, B. Fouque, and F. Chatelain.
Photochemical patterning of biological molecules inside a glass capillary. Analytical Chemistry,
77, 5474–5479, 2005.
114. M. A. Holden, S.-Y. Jung, and P. S. Cremer. Patterning enzymes inside microfluidic channels via
photoattachment chemistry. Analytical Chemistry, 76, 1838–1843, 2004.
115. Cancer Facts and Figures. American Cancer Society, Atlanta, Georgia Publisher, American Cancer
Society 2007.
116. F. Safi, I. Kohler, E. Röttinger, and H. Beger. The value of the tumor marker CA 15-3 in diagnos-
ing and monitoring breast cancer. A comparative study with carcinoembryonic antigen. Cancer,
68, 574–582, 1991.
117. M. J. Duffy. CA 15-3 and related mucins as circulating markers in breast cancer. Annals of Clinical
Biochemistry, 36, 579–586, 1999.
118. D. Aguiar-Bujanda. False elevation of serum CA 15-3 levels in patients under follow-up for
breast cancer. The Breast Journal, 10, 375–376, 2004.
119. M. Cargill, D. Altshuler, J. Ireland, P. Sklar, K. Ardlie, N. Patil, C. R. Lane, E. P. Lim,
N. Kalyanaraman, J. Nemesh, L. Ziaugra, L. Friedland, A. Rolfe, J. Warrington, R. Lipshutz,
G. Q. Daley, and E. S. Lander. Characterization of single-nucleotide polymorphisms in coding
regions of human genes. Nature Genetics, 22, 231–238, 1999.
120. B. Barlen, S. D. Mazumdar, O. Lezrich, P. Kämpfer, and M. Keusgen. Detection of Salmonella by
surface plasmon resonance. Sensors, 7, 1427–1446, 2007.
121. V. Nanduri, A. K. Bhunia, S.-I. Tu, G. C. Paoli, and J. D. Brewster. SPR biosensor for the detection
of L. monocytogenes using phage-displayed antibody. Biosensors and Bioelectronics, 23, 248–252,
2007.
122. A. D. Taylor, J. Ladda, Q. Yu, S. Chena, J. Homola, and S. Jiang. Quantitative and simultaneous
detection of four foodborne bacterial pathogens with a multi-channel SPR sensor. Biosensors and
Bioelectronics, 22, 752–758, 2006.
123. M. Zourob, S. Mohr, B. J. T. Brown, P. R. Fielden, M. B. McDonnell, and N. J. Goddard. An
integrated metal clad leaky waveguide sensor for detection of bacteria. Analytical Chemistry, 77,
232–242, 2005.
124. R. Horváth, H. C. Pedersen, N. Skivesen, D. Selmeczi, and N. B. Larsen. Optical waveguide
sensor for on-line monitoring of bacteria. Optics Letters, 28, 1233–1235, 2003.
125. H.-C. Ren, F. Vollmer, S. Arnold, and A. Libchaber. High-Q microsphere biosensor - analysis for
adsorption of rodlike bacteria. Optics Express, 15, 17410–17423, 2007.
126. G. Yang, I. M. White, and X. Fan. An opto-fluidic ring resonator biosensor for the detection of
organophosphorus pesticides, under review. Sensors and Actuators B, 133, 105–112, 2008.
127. M. A. Duguay, Y. Kokubun, T. L. Koch, and L. Pfeiffer. Antiresonant reflecting optical wave-
guides in SiO2-Si multilayer structures. Applied Physics Letters, 49, 13–15, 1986.
128. T. Baba, Y. Kokubun, T. Sakaki, and K. Iga. Loss reduction of an ARROW waveguide in shorter
wavelength and its stack configuration. Journal of Lightwave Technology, 6, 1440–1445, 1988.
129. B. E. Little, J.-P. Laine, D. R. Lim, H. A. Haus, L. C. Kimerling, and S. T. Chu. Pedestal antireso-
nant reflecting waveguides for robust coupling to microsphere resonators and for microphoto-
nic circuits. Optics Letters, 25, 73–75, 2000.
130. J.-P. Laine, B. E. Little, D. R. Lim, H. C. Tapalian, L. C. Kimerling, and H. A. Haus. Microsphere
resonator mode characterization by pedestal anti-resonant reflecting waveguide. IEEE Photonics
Technology Letters, 12, 1004–1006, 2000.
131. J. P. Laine, B. E. Little, D. R. Lim, H. C. Tapalian, L. C. Kimerling, and H. A. Haus. Planar
integrated wavelength-drop device based on pedestal antiresonant reflecting waveguides and
high-Q silica microspheres. Optics Letters, 25, 1636–1638, 2000.
132. I. M. White, H. Oveys, X. Fan, T. L. Smith, and J. Zhang. Integrated multiplexed biosensors
based on liquid core optical ring resonators and antiresonant reflecting optical waveguides.
Applied Physics Letters, 89, 191106, 2006.
133. I. M. White, H. Oveys, X. Fan, T. L. Smith, and J. Zhang. Demonstration of a liquid core opti-
cal ring resonator sensor coupled with an ARROW waveguide array. In SPIE Integrated Optics:
Devices, Materials, and Technologies XI, Y. Sidorin, and C. A. Waechter, eds., 2006, 6475–6403.
134. M. Sumetsky. Basic Elements for Microfiber Photonics: Micro/Nanofibers and Microfiber Coil
Resonators. Journal of Lightwave Technology, 26, 21–27, 2008.
135. On-Column Micro Gas Chromatography Detection with Capillary-Based Optical Ring
Resonators Analytical Chemistry 80, 2232–2238, 2008.
Rick C. J. Hsu
Broadcom Corporation
Gary Betts
Photonic Systems Inc.
Contents
9.1 Introduction.........................................................................................................................422
9.2 The ADNERF Concept.......................................................................................................423
9.3 Immunity to High EM Fields............................................................................................425
9.4 Thermal Consideration...................................................................................................... 426
9.5 Temperature Dependence of Dielectric Antenna........................................................... 428
9.6 Receiver Sensitivity............................................................................................................ 428
9.7 Choice of EO Material........................................................................................................430
9.8 Receiver Dynamic Range................................................................................................... 431
9.9 RF Gain in the Optical Front-End..................................................................................... 432
9.10 Heterogeneous Dielectric Antenna for Wideband Operation.....................................433
9.11 Microwave Ceramics for Dielectric Antenna.................................................................434
9.12 EO Resonator Design: Whispering Gallery Versus Fabry–Perot . ..............................434
9.13 Optical Power Limit in a Resonant Field Sensor............................................................ 436
9.14 Maximum Power in Disk and Ring Resonators............................................................. 436
9.15 Fabry–Perot LiNbO3 Resonant Modulator...................................................................... 437
9.16 Other Applications of the ADNERF Technology........................................................... 437
9.17 Competing Technologies................................................................................................... 437
9.18 Summary..............................................................................................................................440
Acknowledgment......................................................................................................................... 441
Appendix A: Theory of Resonant EO Field Sensors............................................................ 441
A.1 Equivalent Eπ of Resonant Modulators................................................................ 441
421
© 2009 by Taylor & Francis Group, LLC
9.1 Introduction
The trend toward circuits with smaller sizes and voltages renders modern electronics
highly susceptible to damage from high power electromagnetic sources. Radio frequency
(RF) communication systems are particularly vulnerable because the antenna provides a
direct port of entry for electromagnetic radiation. The threats to civil society posed by such
weapons are viewed as a grim but real possibility in the post September 11 world [1–3].
This chapter describes a new type of RF receiver front-end that offers protection against
electromagnetic weapons. The receiver features a complete absence of electronic circuitry
and metal interconnects, the traditional “soft spots” of a conventional RF receiver. It
exploits a dielectric resonator antenna (DRA) to capture and deliver the RF signal onto an
electro-optic field sensor. The dielectric approach has an added benefit: it reduces physi-
cal size of the front-end—an important benefit in mobile applications. Also, the resonant
enhancement of the RF field in the dielectric antenna enhances the receiver sensitivity.
It has been known that electronic equipment will stop working when brought into close
proximity of TV or radio towers. Separately, it was found in the early 1960s that a high
altitude nuclear explosion can destroy electronics in satellites, ships and homes hundreds
of miles away. These observations served as motivation for development of microwave
based directed energy weapons, a recent example of which is U.S. Navy’s NIRF (neutral-
izing improvised explosive devices using RF). Bordering the worlds of science and science
fiction, electromagnetic weapons have fascinated the public and have been the subject of
many movies, including A View to a Kill (1985) and Ocean’s Eleven (2001)—to name a few.
In the real world, there is great concern that these sources may become widely available
and accessible to unconventional adversaries through the use of flux compression genera-
tors (FCG)—a low cost method for creating intense bursts of electromagnetic energy [4].
These generators produce a power surge capable of destroying or damaging sensitive cir-
cuitry in electronic systems. They can shut down telecommunications networks without
leaving behind a trace of the attacker [2].
High power electromagnetic weapons include narrow band sources known as high power
microwaves (HPM) and broadband (pulse) sources known as electromagnetic pulse (EMP)
devices. Ironically, the danger caused by HPM and EMP sources is exacerbated by the impact
of the celebrated Moore’s law, which describes the electronic industry’s relentless pursuit of
miniaturization. The scaling has resulted in a continuing decrease in the maximum voltage
that such circuits can tolerate, rendering modern electronics highly susceptible to damage
from high power electromagnetic sources. For example the breakdown voltage of the 1.5-nm
gate oxide (used in state-of-the-art integrated circuits) during a 100 ns pulse has been mea-
sured to be less than 5 V [5]. This value should be compared to the capabilities of HPM
weapons, which can deliver transient potentials of several kilovolts inside a circuit.
Figure 9.1
Example of damage suffered by metallic wiring and interconnects caused by high electric fields.
The breakdown of dielectrics is only one of the failure modes of electronic circuits.
Catastrophic damage also results from melting and arcing of metallic interconnects by the
large current surge produced by an HPM source [2]. The close spacing of interconnects in
modern integrated circuit exacerbates this problem. Integrated circuit components (transis-
tors, diodes, inductors, capacitors) and metallic interconnects represent the “soft spots” of
integrated circuits. An example of damage to metallic components is shown in Figure 9.1.
In an RF front-end, the receiver is more vulnerable and more difficult to protect than
the transmitter because of the sensitive electronics (e.g., the low noise amplifier) it con-
tains. Presently, there are few practical options for protecting high frequency receivers
from high power electromagnetic radiation (a review of conventional protection technol-
ogies appears later in this chapter). The All-Dielectric Non-Electronic Radio Front-End
Technology (ADNERF) is aimed at addressing this need.
RF radiation
DRA
Optical carrier
EO
resonato
r
Laser
Electronic PD
circuitry
RF out
Figure 9.2
Concept of the photonic-assisted all-dielectric RF front-end technology. An EO powered dielectric antenna cap-
tures the free-space RF signal. The optical link provides complete electrical isolation between the air interface
and the electronic circuitry, which is located only after the photodetector (PD) [8,9].
Dielectric antenna
Electro-optic
field sensor
Figure 9.3
Proof of concept demonstration ADNERF receiver.
EO resonator 0.9
0.8
0.6
0.5
0.4
0.3
0.2
0.1
Figure 9.4
Simulated received RF electric field inside the dielectric antenna and EO resonator for the structure shown in
Figure 9.5. Note the high intensity locations occur in pairs that are 180 degrees out of phase. We will exploit this
property of dielectric antenna to achieve differential modulation and detection [8].
RF
EO
Crystal O/E
RF
Optical
Dielectric I/O ports
antenna
Figure 9.5
The all-dielectric RF front-end. A resonant ceramic antenna captures the RF signal and concentrates it onto
an EO microresonator, resulting in RF modulation of an optical carrier. A photodetector suitably located at a
remote or shielded enclosure delivers the RF output to an LNA. The use of optics creates complete charge isola-
tion between the LNA and air. The doubly resonant structure (RF and optical) offers extremely high sensitivity,
and can amplify the RF before the LNA [7,8].
–20
With DRA
RF power after photodetection (dBm)
Without DRA
–40
–60
–80
–100
–120
6.544
Frequency (GHz)
Figure 9.6
Demodulated microwave spectrum after photodetection with (solid line) and without (dashed line) the dielec-
tric resonator antenna. Horn antenna feed power is 10 dBm and resolution BW=10 KHz [7–9].
wavelength divided by the square root of the antenna dielectric constant, which leads to a
value of about 5 mm (10 GHz signal and a dielectric constant of 36). Even if the DRA ampli-
fies the field from a 100 MW/m2 pulse by a factor of 100, the electric field inside the DRA
(and the LiNbO3) will be around 105 V/cm, well below the breakdown field of the materi-
als. The field occurring in the front-end transistor would be 103–104 times larger, which
would certainly destroy the transistor.
equation relating power dissipation, applied electric field properties and dielectric proper-
ties of the material can be derived from Maxwell’s equation [10],
P = 2πƒε 0 ε″E2 (9.1)
Where P is the power dissipation (W m-3), f is the electric field frequency, ε0 is the permit-
tivity of free space (8.85 ⋅ 10-12 F m-1), ε″ is the dielectric loss factor of the material (ε″ = ε′
tanδ) and E is the average electric field strength (V m-1). Equation 9.1 assumes that the elec-
tric field is uniform throughout the material. The sample dimensions, therefore, should be
sufficiently small so that the effect of penetration depth, which is inversely proportional
to the dielectric loss factor, is negligible. The dissipation of microwave power within the
material results in an increase in the temperature of the material. The heating rate is given
by the equation,
dT P (9.2)
=
dt ρCp
Where dT/dt is the heating rate (ºC s -1), P is the power dissipation (W m-3), ρ is the density
of the material (Kg m-3) and Cp is the specific heat capacity of the material (J g-1 °C-1).
For the DRA of our experiment, power dissipation of Equation 9.1 should be calculated
based on the peak electric field inside the resonator to give an idea of the maximum tem-
perature rise occurring inside it. Moreover, it should be noted that the ratio of the peak
electric field inside the resonator to that of the incident wave is defined as field enhancement
factor. Therefore Equation 9.1 is modified in this fashion:
P = 2πƒε0ε″ ⋅|EDRA|2 =2πƒε0ε″ ⋅|E0|2 ⋅ β2 (9.3)
Where β is the field enhancement factor—the ratio of the field inside the DRA to that
outside of it. On the other hand, incident wave electric field magnitude is related
to its irradiance through I = 1/2 ⋅ cε0|E 0|2 giving rise to the new equation for power
dissipation:
4 πfε ′′ ⋅ I ⋅ β 2
P= (9.4)
c
dT 4 πfε ′′ ⋅ I ⋅ β 2
= (9.5)
dt ρ ⋅ c ⋅ cp
Now, for a short pulse (∆t) of HPM, maximum temperature rise inside the resonator (∆T)
can be written as:
4 πfε ′′ ⋅ β 2 ⋅ ( I .∆t )
∆T = (9.6)
ρ ⋅ c ⋅ cp
But I . ∆t is called single pulse fluence (with the dimension of J/m2) and is a measure of the
total energy (per unit area) that is incident on the antenna.
© 2009 by Taylor & Francis Group, LLC
Table 9.1
Material Parameters for LiNbO3 and DRA
cp
ε′ tanδ ρ (kg·m-3) (J·kg-1·K-1) β
DRA 100 4.5×10 –4 4380 600 20
LiNbO3 30 1.5×10–3 4650 630 67
Given that the single pulse fluence in air is limited by the air’s dielectric breakdown
threshold, it is possible to estimate the highest power density and energy fluence that can
be beamed on a target a given distance away by an HPM source of given size, frequency
and pulse width.
Taking the dielectric breakdown threshold into account, the theoretical maximum of
single pulse fluence at 300 m from the source is around 103 mJ/cm2 for a 1 µs pulse [11].
The fluence of existing EMP sources are believed to be still several orders of magnitude
below this breakdown threshold. Assuming a source of one order of magnitude below this
threshold and using Table 9.1 for the material parameters, Equation 9.6 leads us to a peak
temperature rise of about 2ºC inside the DRA and about 13ºC inside the LiNbO3. Both of
which are tolerable by the materials.
ance of free space. We define β = ErfDRA/Erfinc as the ratio of the field inside the DRA to that
outside of it. This parameter is the antenna’s field enhancement factor and depends on its
dielectric constant and geometry.
Prfinc
DRA
Emin
εLN 1 εLN
Emin E
εDRA β εDRA min
Figure 9.7
ADNERF receiver considered in the sensitivity calculations.
The field in the DRA is related to the field experienced by the EO modulator
(resonator) by
ErfDRA ε
= m
Erfm ε DRA
where εm and εDRA are the dielectric constants of the EO material and DRA. The depth of
modulation, m, of the light in the resonant modulator can be written as
πErfm Pi
m=
Eπeff Po
where
4 3 π 2λ
Eπeff =
9 Fn3r33L
and F = EO resonator finesse, and L = 2πr. We assume the resonant modulator is biased at
25% transmission point (see Appendix A).
If we let Po be the optical power at the detector, then the RF power from the detector,
PDrf, is
PDrf =
( RPorf )2 Rd
8
where R = receiver responsivity, and Rd = detector impedance. Here, Porf = mPo is the peak-
to-peak RF modulation of the optical carrier.
The CNR, over the RF bandwidth, B, is:
( mRP0 )2 8 (RP0 )2
CNR = 2 2 s 2RIN = (RIN ) B s 2shot = 2 q(RP0 )B
s RIN +s shot +s 2
thermal 2
Figure 9.8
Minimum detectable power into antenna aperture for a resonant LiNbO3 modulator receiver. The notation p-p
indicates the modulator has been push–pull poled. The table shows the parameters used in the calculations.
Where the terms in the denominator describe the RIN, shot, and thermal noise powers. For
an optimum receiver, the received power is high enough to overcome the thermal noise,
so we can ignore this term.
Setting CNR to one, and relating m to the incident RF power, the minimum detectable
power per unit bandwidth, or equivalently the receiver sensitivity, is calculated to be:
2 2 2
1 2 qR R2 ε 2 3 πλ Ae
P min
= + ( RIN ) m 3
2 ε DRA 9n r33L R 2 ⋅ η
rf
β ⋅ F Po
It is readily seen that high finesse, F, high β, and high optical power, Po, all result in better
sensitivity. In particular, the sensitivity improves with the square of β and F. The sensitiv-
ity improves linearly with Po in the shot noise limited case (first term in square brackets)
and becomes independent of Po when RIN noise dominates (second term).
Figure 9.8 is a plot of minimum detectable power as a function of average optical power.
A LiNbO3 resonator is considered and the parameters used in the calculations are shown
in the accompanying table.
polymer ring resonators have been reported. For a review of LiNbO3 microdisk resona-
tors and their use as EO modulators see Ref. [12]. Polymer EO materials have been dem-
onstrated as ring waveguide resonant modulators operating at up to six times the FSR
(162 GHz).
A high n3r33, and a low εr are advantageous. In addition high optical power is advanta-
geous to drive the sensitivity to the RIN limit. Table 9.2 shows the trade-off between power
handling capacity and sensitivity.
Table 9.2
Material Composition
EO Material n3r33 /εr (pm/V) Imax (W/µm2)
LiNbO3 11 0.3
Polymer 5.0 × 103 5 × 10-3
FSR
Output optical power
Frequency
Optical frequency
Sideband Sideband
Figure 9.9
Transmission spectrum of an EO resonator (top) and the spectrum of double-side-band modulated optical car-
rier (bottom). The RF modulation frequency must be equal to the free spectral range (FSR) of the resonator. The
modulator is “biased” by tuning the optical frequency to an optimum point on the transmission spectrum.
–40
–120
2H
–160
Noise floor
–200
–160 –120 –80 –40 0 40
Input power (dBm)
Figure 9.10
Dynamic range (DR) calculation of the resonant modulator biased for maximum linearity. Lines show the out-
put signal power along with 2nd harmonic (2H) and odd-order intermodulation (IM) distortion. The receiver
achieves a third-order spur free dynamic range (SFDR) of 127.7 dB/Hz4/5.
loop is used to lock the tunable laser at a desired laser frequency to the EO resonator for
maximum slope of the transmission function. The mathematical model describing these
features of resonant EO modulators is described in Appendix A. The results, as they relate
to the dynamic range goal of this program, are shown in Figure 9.10. The plot shows the
dynamic range (DR) for a LiNbO3 resonant modulator biased for maximum linearity. The
lines show the output power along with 2nd harmonic (2H) and odd-order intermodula-
tion (IM) distortion. The receiver achieves a third-order spur free dynamic range (SFDR)
of -127.7 dB/Hz4/5.
Є3<Є2<Є1
Є1 Є2 Є3
Figure 9.11
Concentric heterogenous dielectric antenna structure. Such structures offer a larger sensitivity-bandwidth
product than a uniform structure.
11.25 mm
11.25 mm
ε = 10 1.5 mm
ε = 38 ε = 38
9 mm 8 mm
ε = 10 1.5 mm
Resonant frequency = 7.43 GHz Resonant frequency = 7.54 GHz
Loss tangent = 0.0005 Loss tangent = 0.0005
BW = 5.5 MHz BW = 6.7 MHz
EF = 23 EF = 26.6
BW · EF = 130 BW · EF = 180
Figure 9.12
Stacked dielectric antenna structure and results of preliminary (nonoptimized) simulations. The structure on
the right offers larger sensitivity-bandwidth combination than a uniform structure and offers a path toward
receiver optimization. EF: field enhancement factor.
Table 9.3
Materials with Large Dielectric Constant
Loss Temperature
Dielectric Tangent Coefficient coefficient
Material Constant (1/Q) of Dielectric Constant Comments
D-100 (TiO2) 100 <0.0010 -575 E-6 Limited or no
compositional variability.
MCT-100 through MCT-140 100-140 <0.0015 -1070 E-6 Composition variable.
Higher permittivities lead
to more negative
temperature coefficients.
Ba-Ln-Ti-O Tetragonal 90-140 <0.001 Can be tuned by Extremely anisotropic
Tungsten Bronze composition to near 0 material for both
(Ln=La,Nd,Sm and Gd) for permittivity of 90. permittivity and
temperature coefficient.
Ba-Ln-Ti-Nb(Ta)-O 120-200 <0.001 Temperature coefficient Extremely anisotropic
Tetragonal Tungsten Bronze depends on material for both
composition. permittivity and
temperature coefficient.
z axis
R
d
WGM propagation
Figure 9.13
The geometry of a microdisk resonator which supports whispering gallery mode (WGM) propagation.
For the EO sensors based on optical resonance, one immediate path to increasing the
EO sensitivity is to push–pull pole the EO material. Because the RF frequency matches
the FSR of the resonator, the resonator can only be exposed to the RF field over one half of
the optical path. Poling the LiNbO3 or the polymer in opposite directions over each half
of the optical path allows the full resonator to be exposed to the RF field which decreases
the Eπ by a factor of 2 and increases the sensitivity by 6 dB.
The proof of concept ADNERF demonstration made use of LiNbO3 microdisk resona-
tors, such as the one shown in Figure 9.13. This approach combines LiNbO3, a proven mate-
rial for EO modulation, with high finesse achievable in microdisk structures. While this
approach can achieve high sensitivity and dynamic range, it suffers from the limited band-
width and limited optical power that can be used with high finesse disk resonators. To
overcome these limitations, alternate approaches to EO field sensors can be considered.
One such approach to high finesse combined with high optical power is the Fabry–Perot
resonator. The mode size in the Fabry–Perot can be more than an order of magnitude
larger than the ~3 µm2 mode size in the microdisk resonators. The small mode volume in
the microdisk is a manifestation of the side wall’s vertical curvature (Figure 9.13). For the
same finesse and LiNbO3 thickness (~1 mm), this device will have a much larger mode
area and hence can handle orders of magnitude more optical power. It is unlikely that
these devices will have the ultra-high finesse possible (>1000) in the microdisk, however a
lower finesse implies a larger bandwidth. Therefore, the choice between Fabry–Perot and
microdisk depends on the bandwidth and the required optical power at the detector that
are needed to achieve the system performance.
I4 λ oQL F
= =
2
I1 2 π aRneff aπ
Where QL = loaded Q; a = exp(-απR) = E field transmission for one round trip in the resona-
tor; α = power absorption coefficient; R = ring radius; neff = effective index of refraction; and
F = finesse. Critical coupling and λ tuned to resonance is assumed. When λ is tuned to
T = 0.5 (for maximum linearity), I4 is reduced by 0.5.
The critical issue is the optical power, Po, which is incident on the detector. If this power
is limited by the maximum circulating power within the resonator, Pmax, then the maxi-
mum power on the detector, Po(max) is
I1
0.25 Pmax
I4 Pmax aπ
Po ( max ) = =
0.75 3F
approximately a factor of ten smaller than in the modulators (~30 µm2). The resulting opti-
cal intensity will therefore be 10 times higher in the microdisk. So damage may occur. At
this point, no conclusive data exists for the maximum intensity in LiNbO3. Any observable
damage may be related to the epoxy used in the fiber connections and not the crystal itself.
Also, excessive stimulated Raman scattering, or photorefractive effect may limit the power
before crystal damage occurs.
resonators can be fabricated with a finesse of 100 or greater with the mode cross-section
significantly larger than in the disk and consequently these resonant modulators should
be able to handle significantly higher optical power. For example, a Fabry–Perot resona-
tor with a mode area of 100 µm2 would be able to handle ~10 W of circulating optical
power. For F = 100 the output power will be 240 mW; more than enough for a single
detector. Since the modulating RF field generated by the dielectric antenna is larger that
100 µm 2 there is no penalty for the larger mode size. Since the propagation loss in LiNbO3
is low, the finesse, F, is determined by the mirror reflection, R. For F = 100, a reflectivity
R = 0.97 will be needed (F = π R/(1 - R)). This is a reasonable reflectivity for dielectric
coatings.
up to 5.5 KV ESD protection [5]. While the literature on this topic is vast, the basic tech-
nique is rather simple and is described in Figure 9.14. ESD is primarily a low frequency
phenomenon and most of its power is out of band of the LNA. Providing a short circuit
to ground at d.c. and very low frequencies, a shunt inductor protects the input node of
the LNA. The desired RF signal, being at higher frequency, avoids the inductor and is a.c.
coupled by the coupling capacitor to into the LNA. To protect against residual ESD that
gets passed the inductor, diodes are used to clamp the LNA input, preventing its voltage
from exceeding the gate breakdown voltage. In practice, diodes with both polarities must
be used although for simplicity we only show one polarity in Figure 9.14.
These schemes have several limitations. First, the capacitive loading caused by the pres-
ence of extra components at the input will compromise the noise figure. Because of capaci-
tive loading, this technique does not scale well into very high frequencies, namely beyond
10 GHz. To get a high-level ESD protection a large size diode is needed to conduct the
ESD current, which results in a large capacitive load at the input/output (I/O) pins. Using
several diodes in series can reduce the capacitance; however, the cascaded voltage drops
across the diode string will exceed the breakdown voltage of the device in the core circuit
[13]. At or near resonance frequency, the shunt circuit’s equivalent capacitance increases
and its impedance decreases, significantly degrading circuit performance. To avoid this,
the resonant frequency must be much higher than the operating frequency. For example,
a resonant frequency greater than 20 GHz had to be used for a 10 GHz circuit [13]. This
requires small diode capacitances resulting in low levels of protection.
Although these techniques can and are being further optimized and may be applied to
protection against HPM/EMP, their fundamental limitations are inescapable. They work
best when the threat is out of band. Performance is compromised by the extra circuitry
added to the signal path with the penalty becoming more severe for large and in-band
threats. Also they don’t scale well to very high frequency (due to capacitive loading).
Ironically, they rely on integrated circuit components with similar vulnerabilities to those
they are supposed to protect! Electronic protection techniques will be unable to protect
against future threats which, undoubtedly, will be designed to exploit the fundamental
limitations of the technologies.
Plasma limiters are another class of solutions (Figure 9.15). Here, a hollow metallic wave-
guide filled with an inert gas undergoes breakdown when an EMP pulse arrives. The
resulting conductive plasma reflects back the incoming EMP pulse. These devices have
Coupling
Input
L
Inductor Diode
Ground
Figure 9.14
ESD protection circuitry. The shunt inductor directs low frequency out-of-band portion of ESD to ground. The
coupling capacitor a.c. couples the incoming RF signal to the LNA. Diodes clamp the LNA input preventing it
exceed gate breakdown voltage. Such solutions work best for out of band threats. The additional components
tied to the LNA input compromise its noise Figure and frequency response.
Figure 9.15
A plasma limiter consists of a gas filled hollow metallic waveguide. When an intense EM field arrives, the gas
undergoes discharge. The conductive gas reflects the pulse. Protection is only triggered when pulse power
reaches the kW level. Also, there is a delay associated with discharge build up. At lower levels and shorter
times, no protection exists, rendering this technology unsuitable for protecting very sensitive electronics (www.
accurate-automation.com).
high power handling capability and hence are superior to ESD technologies. However,
they do not provide continuous protection, in fact, there is no protection before the plasma
is activated. State-of-the-art devices known to us (www.accurate-automation.com) have a
rise time of about ~1 ns and a trigger level ~5 kW. Until these levels are reached, electro-
magnetic energy leaks in and can destroy the LNA. Therefore, they are not adequate for
protecting very sensitive electronics, and do not protect against very short pulse threats.
Their bulky size notwithstanding, gas discharges are inherently unstable. For that reason,
some means of “priming” the gas is necessary in order to ensure stable activation on each
and every pulse. Older designs employ a small electrode, called a “keep-alive”, for this
purpose, which requires application of a very large d.c. voltage. In addition to the incon-
venience of requiring a power supply, the keep-alive also limits useful tube life and is a
source of a small amount of excess noise [14]. In order to overcome these problems, radio-
active priming was developed. The radioactive source provides a source of electrons for a
more stable activation of plasma, and has been the standard approach in modern devices.
These issues notwithstanding, the leakage of gas during long periods in the field, and the
inability to monitor the gas pressure, pose reliability concerns. In other words, the gas
discharge may not activate when needed.
The ferrite limiter is another type of limiter that requires activation [14]. It employs fer-
rite material which is mounted along waveguide walls of a metallic waveguide. The ferrite
is magnetically biased with permanent magnets. The electrons in the ferrite will precess
around the magnetic field lines. When the input power reaches a critical threshold, the RF
energy will couple into the precession motion causing the ferrite to absorb power as the
RF passes down the waveguide. The absorbed energy is converted to heat. Its advantage is
that it has fast (but not zero) activation and recovery time. Its main disadvantages are that
it exhibits relatively high insertion loss and its performance is very sensitive to changes in
ambient temperature. Also, its leakage is too high for a sensitive receiver to sustain. Thus,
it is normally followed by a diode limiter [14].
Another limiter technology is the multipactor; a vacuum device which employs second-
ary emission materials to absorb RF power. When the input RF power reaches a critical
threshold, multipaction, enhanced by the presence of secondary emission material, takes
place. This causes power to be absorbed as the RF energy passes down the waveguide.
The absorbed energy is converted to heat. It requires a number of d.c. supplies to operate
certain of its key components. In addition, it must be liquid cooled, and it has a limited
lifetime. It has large leakage and thus must be followed by a diode limiter to clean up the
leakage.
To protect sensitive RF front-ends against HPM/EMP treats with arbitrary spectrum,
power levels, and temporal profiles, an entirely new technology is clearly needed. The tech-
nology must eliminate metallic interconnects, metallic based devices such as inductors and
capacitors, and transistors. It must provide complete charge isolation between the air inter-
face and any electronic circuitry that eventually appear in the signal path. In addition, a
reliable protection is one that is always on, i.e. it does not need to be activated when certain
conditions are reached. Neither plasma/ferrite limiters nor diode/inductor circuits repre-
sent a true and complete solution. They do not provide leakage-free, and activation-less
protection. On the other had, the ADNERF technology does. The success of the venerable
optoisolator in protecting digital circuitry is a testament to the potential of this solution.
9.18 Summary
The trend towards reduced geometry and voltage has rendered modern electronics highly
susceptible to damage from EMP attack. The failure mechanisms in integrated circuits are
two-fold: breakdown of dielectrics used in transistor gates and capacitors, and melting
and arcing in metallic electrodes and interconnects.
ADNERF is a fundamentally new RF receiver technology that is inherently immune to
electromagnetic induced damage. It intends to eliminate electronic devices, metallic wir-
ing and interconnects from the RF front-end—the elements that constitute the “soft spots”
in a conventional receiver. The new receiver can match conventional receivers in terms of
bandwidth, sensitivity and dynamic range. It is compact and offers the potential for chip
scale integration.
The approach is to use a dielectric antenna coupled to an EO crystal to create an all-di-
electric RF front-end. The dielectric antenna captures the incoming RF and concentrates it
onto the EO crystal. The RF is modulated onto an optical carrier which delivers it to a pho-
todetector followed by an LNA. The optical link so created provides complete charge iso-
lation between the air interface and the LNA. It also makes it possible to remote the LNA
and subsequent circuitry away from the front-end. The doubly resonant design (RF reso-
nant in the dielectric antenna and optical resonant in the EO microdisk) offers extremely
high receiver sensitivity. The front-end features a complete absence of any electronic and
metallic components, and can amplify the RF even before it reaches the LNA.
There are other approaches to protect against the electromagnetic pulse threat such as
the diode limiter, the plasma limiter, or the ferrite limiter. None of these competing tech-
nologies represents a complete and reliable solution. Protection does not exist at all times;
it is only activated when certain conditions are reached. None provides leakage-free and
activation-less (continuous) protection. They do not eliminate metallic components from
the front-end. The ADNERF technology offers all these features.
The ability of optics to protect electronics against electrical transients is already proven,
as evidenced by the popularity of the venerable Optoisolator, the de facto standard in pro-
tecting digital logic circuitry against high voltage transients. ADNERF is a modern optoiso-
lator; one that has no electronic and metallic components, is analog, and has the sensitivity
and dynamic range required by communication and radar systems.
Acknowledgment
This work was supported by DARPA, and by U.S. Army through the DETEC program.
The LiNbO3 microdisk resonator used in the prototype device was fabricated by Dr. Lute
Maleki’s group at JPL.
dT π
=
dV 2Vπ
In a similar way, the effective Vπeff of a resonant modulator is based on the slope at the
optical bias point selected. For the MZ modulator, Vπ can be expressed as
( )
-1
π dT
Vπ =
2 dV
( )
-1
π dT dθ
Vπeff =
2 dθ dV
where
πn3r33VL
θ = θo + ,
λg
g is the electrode gap, L is the circumference of the ring, and θ0 is the optical bias point.
The transmission of a ring when critically coupled can be written as
1
T (θ) = 1 -
( ) sin θ/2
2
2F
1+ 2
π
θo π
sin =
2 2 3F
4 3 πλg
Vπeff = .
9 Fn3r33L
In the ADNERF application, we are concerned about detecting a microwave E field and
it is more appropriate to consider Eπeff where
4 3 πλ
Eπeff = .
9 Fn3r33L
In ADNERF, it is assumed that the microwave frequency matches the FSR of the resona-
tor and that only half the ring is exposed to the microwave E field. Therefore L becomes
L/π. Alternately the EO effect in the ring in reversed by poling in each half of the ring and
the entire ring is exposed to the E field. In this case L becomes 2L/π.
The advantage provided by the optical resonance can be seen by comparing it to an
equivalent MZ modulator with the same Eπeff. For the same Eπeff, the interaction length of
the MZ must be increased by the factor of 9 F/( 4 3 ). The resonant modulator with a FSR
of 10 GHz will have a circumference of ~1.5 cm. For F = 100, an MZ modulator with same
Eπeff would require an interaction length of 62 cm! The advantage of the resonant optical
modulator can be clearly seen, particularly for the case of high finesse resonators.
1 1
T= + δ - c3δ 3 + c 4δ 4 - c6δ 6 ...
4 2
πEm
δ=
Eπeff
At this bias point there is no second order distortion and Eπeff is minimized. For the
Fabry–Perot resonator all of the above results apply except the output is biased to 75%
transmission.
1 1 5 2 5 4 10 5
T= + δ- δ + δ - δ ...
2 2 22 11 11
where
πEm
δ= .
Eπeff
πλ
Eπeff = .
Fn3r33L
Hence at this bias point, the third order intermode is reduced but Eπeff increases by 1.1 dB
and the d.c. optical power level increases by 3 dB. The 3 dB increase in optical power at the
detector increases CNR when shot noise is limiting. In this case also, L becomes L/π or L
becomes 2L/π as explained above.
For the Fabry–Perot resonator all of the above results apply and the output is biased to
50% transmission.
In sub-octave bandwidth systems, the even order terms are not of importance and if
the optical bias point is set at ∆ƒFWHM/2 to eliminate the third-degree term, the remain-
ing distortion is fifth-degree. By assuming reasonable link parameters (PL = 100 mW, laser
(RIN) = –165 dB/Hz, and detector responsivity = 0.7 A/W) a third-order SFDR of 127.7
dB/Hz4/5 can be expected with micro-resonant modulators as shown in Figure 9.A.1. In
Figure 9.A.1 the horizontal axis is the RF power in the resonant modulator and does not
include the Erf enhancement of the DRA and the F of the resonator.
–40
–120
2H
–160
Noise floor
–200
–160 –120 –80 –40 0 40
Input power (dBm)
Figure 9.A.1
SFDR calculation of the resonant modulator biased for maximum linearity. SFDR=127.7 dBHz4/5.
The bandwidth of the linearity is an issue. The linearity bandwidth will be limited
because the sidebands of the modulation will fall at different points on the resonator trans-
mission curve and therefore will be frequency filtered by the resonator response curve.
1 εM E
Erf = Emod β = DRA
β ε DRA Erf
Erf2
Prf = ADRA ADRA = Effective area of the DRA
2ηo
Here, β is the field enhancement factor of the DRA, ε is the dielectric constant and ηo the
free space impedance. Combining these equations,
2
2 2 ηo Prf ε DRA
Emod = β
ADRA εM
Given a modulation depth m, the transmission of the EO resonant modulator biased for
maximum sensitivity is,
1 1 πErfm P
T= + m m = eff i
4 2 Eπ Po
where Eπeff is the effective half wave voltage of the EO modulator. Now, let i = detector cur-
rent = (m/2)Po R , where R is the detector responsivity, and Po is the optical power at the
detector. This leads to an RF power at the detector,
1 2
Pdet = i RD RD = detector resistance
2
2 2
RD 2 π ηo Prf ε DRA
= ( PR
i ) eff β
4 Eπ ADRA ε M
ε DRA 38 A
ηo = 377 , ADRA = 10-4 m2 , β = 25, = , RD = 50Ω , R = 0.8 , Pdet = 2.8 × 102
εM 35 W
ε DRA 38 A
ADRA = 10-4 m2 , β = 25, = , RD = 50Ω , R = 0.8 , Pdet = 2.8 × 102 Po2 Prf
εM 35 W
If Po = 0.1 W, then Pdet = 11.2Prf. In other words, the system has an RF power gain of 11.2.
If the resonator is push–pull poled, this decreases Eπeff by ½ and increases the gain by an
additional factor of 4.
References
1. Special Issue on High-Power Electromagnetics (HPEM) and Intentional Electromagnetic
Interference (IEMI), IEEE Trans. EMC, 46, 2004.
2. M. Abrams, Dawn of the e-bomb. IEEE Spectrum, 40, 24–30, 2003.
3. Report of the Commission to Assess the Threat to the United States from Electromagnetic
Pulse (EMP) Attack, Volume 1: Executive Report 2004, http://www.globalsecurity.org/wmd/
library/congress/2004_r/04-07-22emp.pdf
4. C. M. Fowler, R. S. Caird, and W. B. Garn. An introduction to explosive magnetic flux compression
generators. Los Alamos National Laboratory, Report LA-5890-MS, 1975.
5. D. Linten, et al. An integrated 5 GHz low-noise amplifier with 5.5 kV HBM ESD protection in
90 nm RF CMOS. 2005 Symposium on VLSI Circuits, 2005. Digest of Technical Papers. pp. 86–89,
Kyoto, Japan, June 2005.
6. R. Dettmer, Dielectric antenna make waves. IEE Review, 49, 28–31, 2003.
7. R. C. J. Hsu, A. Ayazi, B. Houshmand, and B. Jalali. All-dielectric wireless receiver. International
Microwave Symposium, Honolulu, HI, 221–224, June 2007.
8. R. C. J. Hsu, A. Ayazi, B. Houshmand, and B. Jalali. All-dielectric photonic-assisted radio front-
end technology. Nature Photonics, Vol. 1, 535–538, 2007.
Kerry J. Vahala
California Institute of Technology
Contents
10.1 Introduction....................................................................................................................... 447
10.2 Theoretical Framework of Dynamic Back-Action........................................................ 450
10.2.1 Coupled Mode Equations.................................................................................. 450
10.2.2 Modifications due to Dynamic Back-action: Method of
Retardation Expansion...................................................................................... 452
10.2.3 Sideband Formalism.......................................................................................... 457
10.3 Opto-mechanical Coupling and Displacement Measurements................................. 462
10.3.1 Mechanical Modes of Optical Microcavities.................................................. 462
10.3.2 Measuring the Opto-mechanical Response...................................................464
10.3.3 Displacement Sensitivity................................................................................... 465
10.4 Blue Detuning: Mechanical Gain and Parametric Oscillation Instability............... 466
10.4.1 Threshold and Mode Selection Mechanisms................................................. 466
10.4.2 Threshold Dependence on Optical Q and Mechanical Q............................. 467
10.4.3 Oscillation Linewidth........................................................................................ 471
10.5 Red Detuning: Radiation Pressure Cooling................................................................. 471
10.5.1 Experimental Setup............................................................................................ 471
10.5.2 Experimental Observation of Cooling............................................................. 472
10.5.3 Quantum Limits of Radiation Pressure Back-action Cooling..................... 476
10.5.4 Physical Interpretation of the Quantum limits of Back-action Cooling...... 477
10.6 Summary and Outlook.................................................................................................... 478
Acknowledgments.......................................................................................................................480
References......................................................................................................................................480
10.1 Introduction
Recent years have witnessed a series of developments at the intersection of two, previously
distinct subjects. Optical (micro) cavities [2] and micro (nano) mechanical resonators [3],
each a subject in their own right with a rich scientific and technological history, have, in a
* This chapter was originally published in Optics Express [1]. [Kippenberg, T. J., and Vahala, K., Optics Express,
15, 17172, 2007.]
447
© 2009 by Taylor & Francis Group, LLC
theory itself are in play here [40]. On the practical side, there are opportunities to harness
these forces for new metrology tools [35] and even for new functions on a semiconductor
chip (e.g., oscillators [21,22], optical mixers [41], and tuneable optical filters and switches
[42,43]. It seems clear that a new field of cavity opto-mechanics has emerged, and will soon
evolve into cavity quantum opto-mechanics (cavity QOM) whose goal is the observation and
exploration of quantum phenomena of mechanical systems [44] as well as quantum phe-
nomena involving both photons and mechanical systems.
The realization of dynamical, opto-mechanical coupling in which radiation forces medi-
ate the interaction, is a natural outcome of underlying improvements in the technologies of
optical (micro) cavities and mechanical micro (nano-) resonators. Reduction of loss (increas-
ing optical and mechanical Q) and reductions in form factor (modal volume) have enabled
a regime of operation in which optical forces are dominant [21,22,26,28,30,31,34,35,43]. This
coupling also requires coexistence of both high-Q optical and high-Q mechanical modes.
Such coexistence has been achieved in the geometries illustrated in Figure 10.1. It also
seems likely that other optical microcavity geometries such as photonic crystals [45] can
exhibit the dynamic back-action effect provided that structures are modified to support
high-Q mechanical modes.
To understand how the coupling of optical and mechanical degrees of freedom occurs
in any of the depicted geometries, one need only consider the schematic in the upper
panel of Figure 10.1. Here, a Fabry–Perot optical cavity is assumed to feature a mirror that
also functions as a mass-on-a-spring (i.e. is harmonically suspended). Such a configura-
tion can indeed be encountered in gravitational wave laser interferometers (such as LIGO)
and is also, in fact, a direct representation of the “cantilever mirror” embodiment in the
lower panel within Figure 10.1. In addition it is functionally equivalent to the case of a
(a)
(b)
Figure 10.1
(a) A cavity opto-mechanical system consisting of a Fabry–Perot cavity with a harmonically bound end mirror.
(b) Different physical realizations of cavity opto-mechanical experiments employing cantilevers, micromirrors,
microcavities, nano-membranes and macroscopic mirror modes. Red and green arrows represent the optical
trajectory and mechanical motion (From Optics Express. With permission.).
microtoroid embodiment (also shown in the lower panel), where the toroid itself provides
both the optical modes as well as mechanical breathing modes (see Figure 10.1 and discus-
sion below). Returning to the upper panel, incident optical power that is resonant with a
cavity mode creates a large circulating power within the cavity. This circulating power
exerts a force upon the “mass-spring” mirror, thereby causing it to move. Reciprocally, the
mirror motion results in a new optical round trip condition, which modifies the detuning
of the cavity resonance with respect to the incident field. This will cause the system to
simply establish a new, static equilibrium condition. The nonlinear nature of the coupling
in such a case can manifest itself as a hysteretic behavior and was observed over two
decades ago in the work by Walther et al. [18]. However, if the mechanical and optical Qs
are sufficiently high (as is further detailed in what follows, such that the mechanical oscil-
lation period is comparable or exceeds the cavity photon lifetime) a new set of dynamical
phenomena can emerge, related to mechanical amplification and cooling.
In this Chapter, we will give the first unified treatment of this subject. Although micro-
toroid optical microcavities will be used as an illustrative platform, the treatment and phe-
nomena are universal and pertain to any cavity opto-mechanical systems supporting high
Q optical and mechanical modes. In what follows we begin with a theoretical framework
through which dynamic back-action can be understood. The observation of micromechani-
cal oscillation will then be considered by reviewing both old and new experimental results
that illustrate the basic phenomena [21–23]. Although this mechanism has been referred
to as the parametric instability, we show that it is more properly defined in the context of
mechanical amplification and regenerative oscillation. For this reason, we introduce and
define a mechanical gain, its spectrum, and, correspondingly, a threshold for oscillation.
Mechanical cooling is introduced as the reverse mechanism to amplification. We will then
review the experimental observation of cooling by dynamic back-action [28,46] and also
the quantum limits of cooling using back action [47,48].
Finally, we will attempt to discuss some of the possible future directions for this new
field of research.
da 1 1 1
= - i∆( x)a - + a+i s (10.1)
dt 2 τ 0 2 τ ex τ ex
Aside from the x-dependence, the first equation governs the dynamics of the optical
field according to the formalism of Haus [50], i.e., |a|2 is the stored cavity energy, whereas
|s|2 denotes the launched input power upon the cavity system. Moreover the optical field
decays with a rate 1/2 τ = (1/2 τ 0 ) + (1/2 τ ex ) and ∆(x) = ω − ω0(x) accounts for the detuning of
the pump laser frequency ω with respect to the cavity resonance w0 (x) (which, as shown
below, depends on the mechanical coordinate, x). The power coupling rate into outgoing
modes is described by the rate 1/τex, whereas the intrinsic cavity loss rate is given by 1/t0.
In the subsequent discussion, the photon decay rate is also used κ ≡ 1/τ.
The second equation describes the mechanical coordinate (x) accounting for the movable
cavity boundary (i.e. mirror), which is assumed to be harmonically bound and undergoing
harmonic oscillation at frequency Ωm with a power dissipation rate Γ m = (Ω m/Qm ). Moreover,
meff is the effective mass of the mirror mode and will be discussed in a later section. This
mass describes in large part the strength of the coupling between optical and mechanical
mode. For an excellent treatment of its determination and derivation the reader is referred
to Ref. [51]. The radiation pressure forcing function is given by FRP (t) = (ζ/c)(| a |2/Trt ) , where
the dimensionless parameter z takes on the value 2πn for a whispering gallery mode
microcavity (consisting of a dielectric material with refractive index n), and z = 2 for a
Fabry–Perot cavity; and where Trt is the cavity round trip time (note that the intracavity cir-
culating power is given by | a |2/Trt ). Moreover the term FL(t) denotes the random Langevin
force, and obeys 〈FL(t)FL(t′)〉 = ΓmkBTRmeffδ(t − t′), where kB is the Boltzmann constant and TR is
the temperature of the reservoir. The Langevin force ensures that the fluctuation dissipa-
tion theorem is satisfied, such that the total steady state energy in the (classical) mechani-
cal mode Em (in the absence of laser radiation) is given by Em = ∫Ω2meff | χ(Ω)FL(Ω) | 2 dΩ = kBTR,
-1
where the mechanical susceptibility χ(Ω) = meff /(iΩΓ m + Ω m2 - Ω 2 ) has been introduced. Of
special interest in the first equation is the optical detuning ∆(x) which provides coupling
to the second equation through the relation:
ω0
∆( x ) = ∆ + x (10.3)
R
This relation assumes that, under circumstances in which the mass-spring is at
rest (x = 0), the optical pump is detuned by ∆ relative to the optical mode resonance. Two
cases of interest, both illustrated in Figure 10.1, will emerge: blue detuned (∆ > 0) and red-
detuned (∆ < 0) operation of the pump-wave relative to the cavity resonance. It is important
to note that quadratic coupling can also be realized, e.g., Ref. [31], however, this case will
not be considered here.
Before discussing the physics associated with the optical delay (which gives rise to
dynamical back-action), we briefly divert to a static phenomena that is associated with
the steady state solutions of the above coupled equations: the mirror bistability [18] and
multi-stability [52]. Indeed, as noted earlier, the radiation-pressure-induced bi-stability
was observed over two decades ago [18] using a harmonically-suspended mirror. In brief,
considering purely the static solutions for the displacement (i.e., x ) of the above set of
equations, it becomes directly evident that the equilibrium position of the mechanical
oscillator will depend upon the intra-cavity power. Since the latter is again coupled to
the mechanical displacement, this leads to a cubic equation for the mirror position x as a
function of applied power:
τ2 Trt cmeff 2 2 ω
2
| s |2 = Ω m x 4τ ∆ + 0 x + 1 (10.4)
τ ex 4ζ R
For sufficiently high power, this leads to bi-stable behavior (namely for a given detuning
and input power the mechanical position can take on several possible values).
8 ∆τ 2 ω
0 0
Pcav = Pcav - Pcav 0x (10.5)
(
4τ 2 ∆ 2 + 1 )
2
R
64 ∆τ 3 ω dx
+ τPcav
0
0 (10.6)
(
4τ 2 ∆ 2 + 1 )2
R dt
2 8 n2 ω 0 2 ∆τ
∆Ω m = F C⋅ P
k Ω m Ω mmeff c 2 2 2
(
4τ ∆ + 1 )
(10.7)
16n3ω 0R 8 ∆τ
P
Γ = -F 3
C⋅ (10.8)
k Ω m
(
Ω mmeff c 3 4τ 2 ∆ 2 + 1 )
2
Consequently, the modified (effective) damping rate of the mechanical oscillator is
given by:
In both Equations 10.7 and 10.8, we have stressed the fact that these expressions are valid
only in the weak retardation regime in which k >> Ωm. The sign of Γ (and the correspond-
ing direction of power flow) depends upon the relative detuning of the optical pump with
respect to the cavity resonant frequency. In particular, a red-detuned pump (∆ < 0) results
in a sign such that optical forces augment intrinsic mechanical damping, while a blue-
detuned pump (∆ > 0) reverses the sign so that damping is offset (negative damping or
amplification). It is important to note that the cooling rate in the weak-retardation regime
depends strongly (∝F 3) on the optical finesse, which has been experimentally verified as
discussed in Section 10.4.2. Note also that maximum cooling or amplification rate for given
power occurs when the laser is detuned to the maximum slope of the Lorenzian cavity;
these two cases are illustrated in Figure 10.1. These modifications have been first derived
by Braginsky [11] more than 3 decades ago and are termed dynamic back-action. Specifically,
an optical probe used to ascertain the position of a mirror within an optical resonator, will
have the side-effect of altering the dynamical properties of the mirror (viewed as a mass-
spring system).
The direction of power-flow is also determined by the sign of the pump frequency
detuning relative to cavity resonant frequency. Damping (red tuning of the pump) is
accompanied by power flow from the mechanical mode to the optical mode. This flow
results in cooling of the mechanical mode. Amplification (blue tuning of pump) is, not
surprisingly, accompanied by net power flow from the optical mode to the mechanical
mode. This case has also been referred to as heating, however, it is more appropriately
referred to as amplification since the power flow in this direction performs work on the
mechanical mode. The nature of power flow between the mechanical and optical compo-
nents of the system will be explored here in several ways, however, one form of analysis
makes contact with the thermodynamic analogy of cycles in a Clapeyron or Watt dia-
gram (i.e., a pressure–volume diagram). In the present case—assuming the mechanical
oscillation period to be comparable to or longer than the cavity lifetime—such a diagram
can be constructed to analyze power flow resulting from cycles of the coherent radiation
gas interacting with a movable piston-mirror [54] (see Figure 10.2). In particular, a plot of
radiation pressure exerted on the piston-mirror versus changes in optical mode volume
provides a coordinate space in which it is possible to understand the origin and sign of
work done during one oscillation cycle of the piston mirror. Considering the oscillatory
motion of the piston-mirror at some eigenfrequency Ωm, then because pressure (pro-
portional to circulating optical power) and displaced volume (proportional to x) have a
quadrature relationship (through the dynamic back-action term involving the velocity
dx/dt), the contour for a PV cycle will encompass a nonzero area, giving the net-work
performed during one cycle of mechanical oscillation. Note that the area reduces as the
photon lifetime shortens (i.e., as retardation is weakened). Also, the sense in which the
PV cycle is traversed is opposite for the two cases of red and blue detuning of the pump
(i.e., the area and hence work-done changes sign). For blue detuning, the radiation gas
does net work on the piston while the reverse is true for red detuning. The fact that
cooling is possible for the case of a red detuned pump is the result of both: the sign of
work in this case being such that the piston does positive work on the photon gas, and,
equally important, that the photon gas (if in a coherent state) provides only quantum,
back-action on the piston. This makes the photon gas effectively very cold.
The cooling and amplification processes can also be understood in terms of the cre-
ation of stokes and anti-Stokes sidebands [21]. Oscillatory motion of the cavity mirror
will create sidebands on the probe wave as the circulating optical power is Doppler-
shifted by the mirror’s motion (or equivalently the expansion and contraction of the
Photon pressure
Displacement (volume)
Figure 10.2
Work done during one cycle of mechanical oscillation can be understood using a PV diagram for the radiation
pressure applied to a piston-mirror versus the mode volume displaced during the cycle. In this diagram the cycle
follows a contour that circumscribes an area in PV space and hence work is performed during the cycle. The sense
in which the contour is traversed (clockwise or counterclockwise) depends upon whether the pump is blue or red
detuned with respect to the optical mode. Positive work (amplification) or negative work (cooling) are performed
by the photon gas on the piston mirror in the corresponding cases (From Optics Express. With permission.).
whispering gallery in the case of a toroid or disk resonator). As shown in Figure 10.3,
these motional sidebands will have asymmetric amplitudes owing to the fact that the
pump wave frequency is detuned from the cavity resonance. The two cases of interest
(red and blue detuning) are illustrated in the Figure 10.2 and produce opposite asymme-
try. Intuitively, the sideband that is closer to the cavity resonance has its amplitude reso-
nantly enhanced. This will be rigorously derived in the next Section. This asymmetry
indicates a net deficit (blue detuned) or surplus (red detuned) of power in the transmit-
ted pump wave. The sideband asymmetry also produces strong amplitude modulation
that can be measured as a photocurrent upon detection of the transmitted power from
the resonator.
Amplification and damping (i.e., cooling) have been verified experimentally. In particu-
lar, a mechanical mode subjected to dynamic backaction through a properly detuned opti-
cal pump wave will exhibit a thermal noise spectrum whose line-shape is modified by the
presence of added optical damping or gain (negative damping). In either case, the damp-
ing rate will depend in a linear fashion on the coupled optical pump power for fixed pump
detuning. Before proceeding with the further development of the theory, it is useful to con-
sider the measurement of this damping dependence in an actual system (further details
regarding the realization of amplification and cooling will be the subject of Sections 10.4
and 10.5). Measurement of this behavior is possible by probing the motion of the cavity
mirror using the transmitted optical pump wave. If the pump wave is detuned relative to
the cavity resonance, it is power-modulated by the resonator since the mechanical motions
vary in time the cavity resonance. (Equivalently, the asymmetric motional sidebands on
the transmitted pump are detected.) The detected amplitude modulation contains infor-
mation about the underlying mechanical motion. If, for example, the mirror is undergoing
regenerative oscillations (see Section 10.4), then spectral analysis of the photocurrent will
reveal the oscillation frequency (and even the effective temperature) and measurement
Power
Power
κ κ
∆ ∆
ω0 ω0 Frequency
Frequency
(b)
} Ωm
} Ωm
} Ωm
m
Power
Power
} Ω
Frequency Frequency
Figure 10.3
The two manifestations of dynamic back-action: blue-detuned and red-detuned pump wave (green) with respect
to optical mode line-shape (blue) provide mechanical amplification and cooling, respectively. Also shown in the
lower panels are motional sidebands (Stokes and anti-Stokes fields) generated by mirror vibration and subse-
quent Doppler-shifts of the circulating pump field. The amplitudes of these motional sidebands are asymmetric
owing to cavity enhancement of the Doppler scattering process (From Optics Express. With permission.).
of the modulation power can be used to ascertain the mechanical oscillation amplitude.
In cases where the mirror is excited only by thermal energy, the spectrally-broad ther-
mal excitation (upon measurement as a photocurrent spectral density) provides a way to
observe the oscillator line-shape and thereby determine its linewidth and effective damp-
ing rate. If the optical pump wave is weak, this damping rate will reflect the intrinsic
loss (intrinsic mechanical Qm = Ωm/Γm) of the mechanical mode. On the other hand, as the
probe power is increased, it will modify the damping rate causing the line to narrow (blue
detuned) or broaden (red detuned) in accordance with the above model.
Figure 10.4 presents such line-shape data taken using a microtoroid resonator. The power
spectral density of the photocurrent is measured for a mechanical mode at an eigenfre-
quency of 40.6 MHz; three spectra are shown, corresponding to room temperature intrin-
sic motion (i.e., negligible pumping), mechanical amplification and cooling. In addition
to measurements of amplification and damping as a function of pump power (for fixed
detuning), the dependence of these quantities on pump detuning (with pump power fixed)
can also be measured [28]. Furthermore, pulling of the mechanical eigenfrequency (caused
by the radiation pressure modification to mechanical stiffness) can also be studied [53]. A
summary of such data measured using a microtoroid in the regime where k W is pro-
vided in Figure 10.5. Both the case of red- (cooling) and blue- (amplification) pump detun-
ing are shown. Furthermore, it can be seen that pump power was sufficient to drive the
mechanical system into regenerative oscillation over a portion of the blue detuning region
(section of plot where linewidth is nearly zero). For comparison, the theoretical prediction
is shown as the solid curve in the plots. Concerning radiation-pressure-induced stiffness,
it should be noted that for red-detuning, the frequency is pulled to lower frequencies (stiff-
ness is reduced) while for blue-detuning the stiffness increases and the mechanical eigen-
frequency shifts to higher values. This is in dramatic contrast to similar changes that will
0
40.58 40.59 40.6
Frequency (MHz)
Figure 10.4
Dynamics in the weak retardation regime. Experimental displacement spectral density functions for a mechan-
ical mode with eigenfrequency 40.6 MHz measured using three, distinct pump powers for both blue and red
pump detuning. The mode is thermally excited (green data) and its linewidth can be seen to narrow under blue
pump detuning on account of the presence of mechanical gain (not sufficient in the present measurement to
excite full, regenerative oscillations); and to broaden under red pump detuning on account of radiation pressure
damping (From Optics Express. With permission.).
Parametric
100 oscillation
regime
50
57.86
Frequency (MHz)
57.84
57.82
57.8
−2 −1 0 1 2
Normalized detuning τ(ω−ω0)
Figure 10.5
Upper panel shows mechanical linewidth (δm = Γeff/2π) and shift in mechanical frequency (lower panel) mea-
sured versus pump wave detuning in the regime where Ωm<< k. For negative (positive) detuning cooling (ampli-
fication) occurs. The region between the dashed lines denotes the onset of the parametric oscillation, where gain
compensates mechanical loss. Solid curves are theoretical predictions based on the sideband model (see Section
10.2.2). (From Schliesser, A., Del’Haye, P., Nooshi, N., Vahala, K. J., and Kippenberg, T. J., Physical Review Letters,
97, 243905, 2006) (From Optics Express. With permission.).
be discussed in the next Section. While in Figure 10.5 the absolute shift in the mechanical
eigenfrequency is small compared to Wm, it is interesting to note that it is possible for this
shift to be large. Specifically, statically unstable behavior is possible if the total spring
constant reaches a negative value. This has indeed been observed experimentally in gram
scale mirrors coupled to strong intra-cavity fields by the LIGO group at MIT [30].
While the above approach provides a convenient way to understand the origin of gain
and damping and their relationship to nonadiabatic response, a more general understand-
ing of the dynamical behavior requires an extension of the formalism. Cases where the
mechanical frequency, itself, varies rapidly on the time scale of the cavity lifetime cannot
be described correctly using the above model. From the viewpoint of the sideband picture
mentioned above, the modified formalism must include the regime in which the sideband
spectral separation from the pump can be comparable or larger than the cavity linewidth.
a≡ ∑ε a
n= 0
n
n
Here, the zeroth order perturbation amplitude (a0) describes the cavity build-up factor in
the absence of the opto-mechanical coupling
i 1
τ ex
a0 (t) = s (10.10)
i∆ + 1
2τ
where the detuning is given by D. Making the assumption that the mechanical system
is undergoing harmonic motion x(t) = x ∙ cos(Ωmt), it is possible to solve for the first-order
perturbation term:
e iΩmt e - iΩmt
a1 (t) = ia0 (t)2 τω 0 +
2 i( ∆ + Ω m )τ + 1 2 i( ∆ - Ω m )τ + 1
Inspection shows that this 1st order term consists of two, independent fields, a frequency
upshifted anti-Stokes sideband (ωAS = ω + Ωm), and a frequency down shifted (ωS = ω − Ωm)
Stokes sideband produced by the harmonic mechanical motion. These fields account
for Doppler shifting of the circulating pump field caused by the motion of the mirror or
dielectric cavity. The parameter ε is sufficiently small to neglect the higher-order terms
in the perturbation. The radiation pressure force acting on the mechanical oscillator in
the toroidal (or other whispering gallery mode) cavity is FRP (t) = (2 πn/c)(| a |2/Trt ) (and
for a Fabry–Perot cavity FRP (t) = (2/c)(| a |2/Trt )). Here, Trt is the cavity round trip time and
|a|2 and is to 1st order by a 2 ≅ a0 2 + 2 εℜ( a0∗a1 ) . The radiation pressure force can now be
expressed in terms of in-phase and quadrature components (with respect to the harmonic
displacement, assumed above):
FRP(t) = cos(Ωmt)FI + sin(Ωmt)FQ
16 πn ω 0τ 3 /τ ex 2( ∆ + Ω m )τ 2( ∆ - Ω m )τ
FI = + ε 4( ∆ + Ω )2 τ 2 + 1 + 4( ∆ - Ω )2 τ 2 + 1 P (10.11)
cTrt 4 ∆ 2 τ 2 + 1 m m
8 πn ω 0τ 2 /τ ex 2τ 2τ
FQ = - ε - P (10.12)
cTrt 4 ∆ 2 τ 2 + 1 4( ∆ + Ω m )2 τ 2 + 1 4( ∆ - Ω m )2 τ 2 + 1
The in-phase part of this force is responsible for changes in rigidity whereas the quadra-
ture part is responsible for changes in the damping factor, which leads to cooling (or ampli-
fication). From the force, the net power (Pm) transferred from the mechanical mode to (or
from) the optical mode can be calculated via the relation Pm = FQ ⋅ sin(Ωmt) ⋅ x yielding:
8 πn ω 0τ 3 /τ ex Ω m 2τ 2τ
Pm = x 2 (10.13)
cTrt 4τ 2 ∆ 2 + 1 2 R 4( ∆ + Ω m )2 τ 2 + 1 - 4( ∆ - Ω m )2 τ 2 + 1 P
2 8 n2 ω 0 1 1
Γ = -F 2 C ⋅ - P (10.14)
Ω mmeff c 4( ∆ - Ω m ) τ + 1 4( ∆ + Ω m ) τ + 1
2 2 2 2
Here the finesse F has been introduced (as before) as well as the previously introduced
dimensionless coupling factor C = (τ / τ ex ) /( 4 ∆ 2 τ 2 + 1) which takes on a value between 0 and 1.
Analysis of the above formula allows derivation of the optimum detuning, for which the
cooling or amplification rate is maximum. Note that in the limit of k >> Ωm one recovers the
result of the previous section (Equation 10.8), as
1 1 1 8 ∆τ 2
lim - =
Ω m →0 2 Ω m 4( ∆ - Ω m ) τ + 1
2 2
4( ∆ + Ω m )2 τ 2 + 1 4∆ 2 τ 2 + 1 ( )
2 (10.15)
Hence as noted before, in the weak retardation regime, the maximum cooling (or amplifi-
cation) rate for a given power occurs when the laser is detuned to the maximum slope of the
cavity Lorenzian, i.e., ∆ = k/2, in close analogy to Doppler cooling [7] in atomic physics. On
the other hand, in the resolved sideband regime optimum detuning occurs when the laser is
tuned either to the lower or upper “motional” sideband of the cavity, i.e., ∆ = ± Ωm. This behavior
is also shown in Figure 10.6. The above rate modifies the dynamics of the mechanical oscilla-
tor. In the case of amplification (blue detuned pump) it offsets the intrinsic loss of the oscilla-
tor and overall mechanical damping is reduced. Ultimately, a “threshold condition” in which
mechanical loss is completely offset by gain occurs at a particular pump power. Beyond this
power level, regenerative mechanical oscillation occurs. This will be studied in Section 10.4.
For red detuning, the oscillator experiences enhanced damping. Beyond the power
flow analysis provided above, a simple classical analysis can also be used to understand
how such damping can result in cooling. Specifically, in the absence of the laser, the mean
energy (following from the equation for x) obeys Em /dt = - Γ m Em + kBTR Γ m, implying
that the mean energy is given by the reservoir temperature 〈Em〉 = kBTR. When considering
the modifications to this equation resulting from back-action damping, an additional loss
term appears in the equation for the average energy:
d
Em = -(Γ m + Γ ) Em + kBTR Γ m (10.16)
dt
1.5 Amplification
1
Normalized detuning ∆/∆opt
0.5
0
Cooling
−0.5
−1
Figure 10.6
The mechanical gain and cooling rate as a function of detuning and inverse cavity decay rate (normalized with
respect to the fixed mechanical frequency). Also shown is the optimum amplification and cooling rate for fixed
frequency (dotted lines). In the simulation, pump power and cavity dimension are fixed parameters (From
Optics Express. With permission.).
Note that within this classical model, the cooling introduced by the laser is what has
been described by some authors [29] as “cold damping”: the laser introduces a damping
without introducing a modified Langevin force (in contrast to the case of intrinsic damp-
ing). This is a key feature and allows the enhanced damping to reduce the mechanical
oscillator temperature, yielding as a final effective temperature for the mechanical mode
under consideration:
Γm
Teff ≅ TR (10.17)
Γm + Γ
It deserves notice that the above formula predicts that arbitrarily low temperatures can
be attained. As discussed later, the laser damping also introduces a small noise term, due
to the quantum nature of light, which adds a further Langevin force to previous equations.
This will be considered in the last section and is shown to provide the ultimate cooling
limit of this technique. Moreover, it is noted that the above formula is only valid [47,48] as
long as Γ << k and as (Γ m/Γ m + Γ ) > (1/Qm ).
For completeness, the in-phase component of the radiation pressure force is also investi-
gated. This component of the force causes a change in the mechanical oscillator’s rigidity,
and its adiabatic contribution is the well-known optical spring effect described earlier.
Specifically, the change in mechanical resonance frequency (from its intrinsic value) is
given by [28]:
2 8 n2 ω 0 ∆ - Ωm ∆ + Ωm
∆Ω m = F 2 Cτ ⋅ + P (10.18)
Ω mmeff c 4( ∆ - Ω m ) τ + 1 4( ∆ + Ω m ) τ + 1
2 2 2 2
Note that in the regime where the mechanical frequency is comparable to, or exceeds
the cavity decay rate it’s behavior is quite different from that described earlier for the
conventional adiabatic case; and was only recently observed experimentally [28]. As noted
earlier, in the adiabatic regime, the mechanical frequency is always downshifted by a red-
detuned laser (i.e. a reduced rigidity). However, when Ωm > 1/2k, an interesting phenomena
can occur. Specifically, when the pump laser detuning is relatively small, the mechanical
frequency shift is opposite in sign to the conventional, adiabatic spring effect. The same
behavior can occur in the case of amplification (∆ > 0). Furthermore, a pump detuning exists
where the radiation-pressure induced mechanical frequency shift is zero, even while the
damping/amplification rate is nonzero. The latter has an important meaning as it implies
that the entire radiation pressure force is viscous for red detuning, contributing only to
cooling (or to amplification for blue detuning). Figure 10.6 shows the attained cooling rate
(as a contour plot) for fixed power and mechanical oscillator frequency as a function of
normalized optical cavity decay rate and the normalized laser detuning (normalized with
respect to ∆ opt = Ω m2 + k 2 /4 . Evidently, the highest cooling/amplification rates are achieved
in this resolved sideband regime, provided the pump laser is detuned to + Ωm or − Ωm (i.e.
corresponding to the cavities lower and upper motional sideband).
An important feature of cooling and amplification provided by dynamic back-action
is the high level of mechanical spectral selectivity that is possible. Since, the cooling/
amplification rates depend upon asymmetry in the motional sidebands, the optical lin-
ewidth and pump laser detuning can be used to select a particular mechanical mode to
receive the maximum cooling or amplification. In effect, the damping/amplification rates
shown above have a spectral shape (and spectral maximum) that can be controlled in an
experimental setting as given by horizontal cuts of Figure 10.6. This feature is important
since it can provide a method to control oscillation frequency in cases of regenerative oscil-
lation on the blue detuning of the pump. Moreover, it restricts the cooling power to only
one or a relatively small number of mechanical modes. In cases of cooling, this means that
the overall mechanical structure can remain at room temperature while a target mechani-
cal mode is refrigerated.
The full-effect of the radiation-pressure force (both the in-phase component, giving rise
to a mechanical frequency shift, and the quadrature-phase component which can give
rise to mechanical amplification/cooling) were studied in a “detuning series”, as intro-
duced in the last Section. The predictions made in the Ωm > 0.5k regime have been verified
experimentally as shown in Figure 10.7. Specifically, the predicted change in the rigidity
of the oscillator was experimentally observed as shown in Figure 10.7 and is in excellent
agreement with the theoretical model (solid red line). Keeping the same sample but using a
different optical resonance with a linewidth of 113 MHz (57.8 MHz mechanical resonance),
the transition to a pure increasing and decreasing mechanical frequency shift in the cool-
ing and amplification regimes was observed as shown in the preceding Section, again
confirming the validity of our theoretical model based on the motional sidebands.
60
Cooling Amplification
50
Linewidth δm (kHz)
40
30
20
10
57.83
Frequency (MHz)
57.825
57.82
−2 −1 0 1 2
Normalized detuning τ(ω−ω0)
Figure 10.7
Dynamics in the regime where Wm~k as reported in Ref. [28]. Upper panel shows the induced damping/amplifi-
cation rate (δm = Γeff/2π) as a function of normalized detuning of the laser at constant power. The points represent
actual experiments on toroidal microcavities, and the solid line denotes a fit using the sideband theoretical
model (Equations 10.14 and 10.18). Lower panel shows the mechanical frequency shift as a function of normal-
ized detuning. Arrow denotes the point where the radiation pressure force is entirely viscous causing negligible
in phase, but a maximum quadrature component. The region between the dotted lines denotes the onset of the
parametric instability (as discussed in Section 10.3). (From Schliesser, A., Del’Haye, P., Nooshi, N., Vahala, K. J.,
and Kippenberg, T. J., Physical Review Letters, 97, 243905, 2006). (From Optics Express. With permission.).
50 (µm)
30 (µm)
n=3
r(t)
V(n,1) = V(1,0)
100
Frequency (MHz)
50
60
40
40 30
20
20
10
0 0
15 20 25 30 35 40 30 35 40
Cavity length (L in µm) Radius (µm)
Figure 10.8
Upper panel: SEM images and mechanical modes of several types of whispering gallery mode microcavities:
toroid microcavities [59] microdisks [58] and microspheres [57]. Also shown are the stress and strain field in cross
section of the fundamental radial breathing modes, which include radial dilatation of the cavity boundary. Lower
panel: the dispersion diagram for the lowest lying, rotationally symmetric mechanical modes for a toroid (as a
function of its undercut) and for a microsphere (as a function of radius) (From Optics Express. With permission.).
have been observed at 50 MHz and room temperature, comparing favorably with the best
nano-mechanical resonators [16] to date at low temperatures.
While the micromechanical modes coexist within the same physical structure as the
optical WGM, it is important to note that there is high level of spatial separation between
the modes of these physical systems. In fact, the optical whispering gallery confines opti-
cal energy to the extreme periphery of the device, while the mechanical mode impacts
the entire structure. It is therefore possible to affect changes in the mechanical Q and
eigenfrequency spectrum by introduction of micromechanical probes without affecting
in any way the optical properties of the resonator. This feature can provide additional
ways to investigate the physics of these devices. For example, a micromechanical probe,
when scanned across the surface of a microtoroid, is found to modify the eigenfrequency
of a mechanical mode in proportion to the amplitude-squared of the mode function.
By measuring the mechanical eigenfrequency during such a scan (using the optical probe
technique described above), the underlying mechanical mode can be “imaged”. Figure 10.9
provides images taken of both a fundamental and first-excited mode in a microtoroid.
1 f = 15 MHz
Relative Freq. Shift
0.8
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70
1
f = 6.6 MHz
Relative Freq. Shift
0.8
0.6
0.4
0.2
0
0 10 20 30 40 50 60
Position in (µm)
Figure 10.9
Scanning probe microscopy of the two lowest lying micromechanical resonances of a toroid microcavity. Lower
graph: The normalized mechanical frequency shift for the first mode as a function of position. Upper graph: The
normalized frequency shift for the second mechanical mode as a function of scanned distance across the toroid.
Superimposed is the scaled amplitude (solid line) and the amplitude squared (dotted line) of the mechanical
oscillator modes obtained by finite element simulation of the exact geometry parameters (as inferred by SEM)
(From Optics Express. With permission.).
IR FSFC PC FC
External cavity Power
meter
diode laser isolator
965 nm
AUX N2
Locking
electronics
Spectrum
analyzer
Oscilloscope SC
DET
Figure 10.10
Experimental setup for the observation of cavity cooling or oscillation of a mechanical oscillator. All relevant data
from the electronic spectrum analyzer and the oscilloscope are transferred to a computer controlling the experiment.
More details in the text. IR: iris, FSFC: free-space to fibre coupler, PC: fibre polarization controller, FC: fibre coupler,
AUX: auxiliary input, SC: sealed chamber, DET: fast photoreceiver (From Optics Express. With permission.).
λ
δx min ≅ (10.19)
8 πF ηP / ω
For numbers typical of the toroidal microcavity work (F≈4000, η≈1 μW λ = 1064 nm) this
implies a displacement sensitivity of δxmin ≅ 5 ⋅ 10-19 m/ Hz . Note that this is a remark-
ably low level, which has been experimentally achieved at the LKB [35] and also for moni-
toring of the toroid cavity modes [46,64]. Interestingly, it is, and in principle, even sufficient
to detect the zero point motion δxmin ≅ 5·10− 16 m within a 1-kHz resolution bandwidth. In
practice, however, such a value can only be achieved in cases where true quantum-limited-
readout is present, necessitating lasers which operate with quantum limited amplitude
and phase noise, i.e. do not have excess classical noise in either of the two quadratures (the
latter is the case of Nd:YAG lasers for frequencies above ca. 1 MHz). For the experimental
measurements described [28] herein the actual displacement sensitivity did not achieve
this level owing to the fact that the diode lasers exhibited excess phase noise, and lim-
ited the sensitivity to a value of about δxmin ≅ 5 ⋅ 10-18 m/ Hz . Recent work, however, has
also obtained higher sensitivity by employing low noise lasers [46]. A typical calibrated
5
n=3
Displacement SD (M/Hz0.5 × 10−17) 4.5
4
n=3
3.5
3
n=1
2.5
n=2
2 n=1
1.5
1
0.5
n=1
0
5 10 15 20 25 30 35 40 45 50 55 60
Frequency (MHz)
Figure 10.11
Calibrated [46] displacement spectral density as measured by the setup shown in Figure 10.10. The peaks
denote different mechanical eigenmodes of the toroidal microcavity. The probe power is sufficiently weak
such that the mechanical modes amplitude is dominated by Brownian motion at room temperature and
backaction effects are negligible. Cross-sectional representations of the n = 1,2,3 modes and their correspond-
ing spectral peaks are also given as inferred by finite element simulations. (From Optics Express. With
permission.).
and broadband displacement spectrum that can be attained with diode lasers is shown
in Figure 10.11. It reveals several mechanical modes, which can be accurately assigned via
finite element modeling.
-1
Ω2 meff c 2 1 1
Pthresh = m ⋅ - (10.20)
Qm ω 0 F 8n C 4( ∆ - Ω m ) τ + 1 4( ∆ + Ω m ) τ + 1
2 2 2 2 2 2
-1
Ωm meff c 3 8 ∆τ
Pthresh = 2
(10.21)
k Ω m Qm ω 0 F 16n RC ( 4 ∆ τ + 1)
3 3 2 2
It is worth noting that even though cooling does not exhibit a similar threshold condi-
tion, the above condition, when expressed for the case of red-detuning of the pump wave,
gives the condition in which the radiation pressure cooling rate equals the heating rate of
the mechanical mode. As such, a factor of two in temperature reduction (cooling) already
requires that parametric oscillation is observable for the corresponding blue detuning.
An important feature of the mechanical gain is its dependence with respect to the
mechanical eigenfrequency. Mechanical modes whose eigenfrequencies fall near the peak
of this curve have the lowest threshold pump power for oscillation. The general shape
of this curve can be inferred from Figure 10.6, which shows a contour plot of both the
gain and the cooling rate versus the normalized mechanical eigenfrequency and the nor-
malized detuning. As noted earlier, horizontal slices of this plot give the gain (or cool-
ing rate) spectral shapes. Experimental control of the spectral peak of the gain (cooling
rate) is obviously important since it determines which mechanical modes oscillate or
receive maximum cooling. The contour of maximum gain appears as a dashed contour
in the plot (likewise there is a corresponding contour for maximum cooling rate). The
unresolved sideband case (case of weak retardation) in the figure provides a convenient
physical limit in which to illustrate one form of spectral control. In this case, as noted
before, the contour of maximum gain (cooling) occurs when the pump wave is detuned
to the half-max position of the optical lineshape function (cf. the vertical axis value of the
contour in unresolved sideband regime). The maximum mechanical gain increases along
this contour as the parameter Wm/k increases towards 0.5 and then diminishes beyond
this value (for pump-wave detuning fixed). (Beyond this value, even somewhat before it
is reached, the pump detuning must be adjusted continuously in concert with increases
in Wm/k to remain on the contour of maximum gain. This is a result of transition into the
sideband resolved regime as described earlier). This behavior can be understood in the
context of the motional sideband (Stokes/anti-Stokes waves) description provided earlier
(see Figure 10.2). Specifically, the case of a pump wave detuned to the half-max point is
diagrammed in Figure 10.12 for three values of the parameter Wm/k (<, = ,> 0.5). The corre-
sponding sideband configuration in each of these cases is also illustrated for comparison.
Since mechanical gain is largest when the sideband asymmetry is maximum, the interme-
diate case of Wm/k = 0.5 will exhibit maximum gain in the scenario depicted in Figure 10.12.
Through adjustments of the optical mode linewidth (as can be done by controlling wave-
guide loading of the microresonator) an optimum Wm/k can be set experimentally. This
method has, in fact, been used to provide targeted oscillation of mechanical modes (even
into the microwave regime) through control of optical waveguide loading. It is impor-
tant to note however that the above considerations assume a constant normalized detun-
ing. If the detuning is allowed to vary as well, maximum cooling or amplification rates
always occurs in the resolved sideband regime, when the detuning equates to ∆ = ± Ωm
(cf. Figures 10.6 and 10.12). The inherent advantages of this regime are that it enables a
higher level of asymmetry in the sidebands. As described in the next section, sideband
asymmetry takes on even greater significance in the context of ground state cooling.
Power
Power
κ κ κ
Figure 10.12
Back-action tuning for mode selection for a fixed laser detuning corresponding to ∆ = k/2. The target mode that
receives maximum gain (or optimal cooling for ∆ = –k/2) can be controlled by setting the cavity linewidth to
produce maximum sideband asymmetry for that particular mechanical mode. In this schematic, three mechan-
ical modes (having frequencies Ωm,i i = 1,2,3 interact with an optical pump interact with the optical cavity mode,
however, in the present scenario only the intermediate mode experiences maximum gain (or cooling) since its
sideband asymmetry is maximal (since Ωm,2/k = 0.5). It is important to note however, that if the laser detuning
is allowed to vary as well, the highest frequency mode would experience the largest gain if ∆ = Ωm,2 was chosen
(From Optics Express. With permission.).
20
Radial displacement (pm)
15
10
0 20 40 60 80
Optical power (µW)
Figure 10.13
Regenerative oscillation amplitude plotted versus pump power. The threshold knee is clearly visible. In this
case a threshold of 20 µW is observed. (From Kippenberg, T. J., Rokhsari, H., Carmon, T., Scherer, A., and Vahala,
K. J., Physical Review Letters, 95, 033901, 2005) (From Optics Express. With permission.).
The measurement of these inherent dependences provides further evidence of the underlying
nature of the interaction. Measurement of the threshold is straightforward and involves moni-
toring the photocurrent of the detected transmission either in the time domain or on a spec-
trum analyzer (as described earlier). The amplitude of oscillations at a particular mechanical
eigenfrequency will exhibit a “threshold knee” when plotted versus the coupled optical pump
power. This knee is an easily measurable feature and one example is provided in Figure 10.13.
The first observation of radiation pressure parametric oscillation instability, as predicted
by Braginsky, was made in toroidal microcavities [21–23]. The setup employed a tapered
optical fiber coupling such that optical loading could be controlled during the experi-
ment. This enabled control of the optical linewidth as described above to effect control of
the specific mechanical mode designated for oscillation. When exciting the cavity using a
blue-detuned laser pump, an oscillatory output of the cavity could be observed, indicative
of the excitation of mechanical modes. The oscillation is readily observable in microtoroids
with threshold values in the microwatt range. Indeed, by using typical parameters in the
above threshold equation (F≈105, meff ≈10−11 kg, Qm≈104) a threshold in the range of a few
microwatts is predicted (which is even below the threshold for Raman [65] and paramet-
ric oscillations [66]). The fact that this process can be observed in a reproducible manner
allows probing of fundamental metrics of these phenomena including the above-noted
mechanical and optical Q-dependences.
Measurement of the dependence of the parametric oscillation threshold on the mechanical Q
is shown in Figure 10.14. In this experiment, first reported in Ref. [21], the mechanical Q-factor
was varied while the optical Q factor was left unchanged. In a variation on the mechani-
cal probing technique described above to provide image scans of the mechanical mode, a
microprobe, in the form of a sharp silica fiber tip, was brought into contact, at a fixed posi-
tion, with the interior of the microcavity. This caused dissipative coupling of the mechanical
mode, decreasing its value from an initial, room temperature and ambient pressure Q-value
of 5000 to below 50. As noted above, while the mechanical probe modifies the Q (and weakly,
the eigenfrequency) of the mechanical mode, it has no effect on the optical performance of
the whispering gallery. Therefore, by using this method, the dependence of threshold on
mechanical Q can be probed in a nearly ideal way. Gradual change could be induced by varia-
tion of the tip pressure. Figure 10.14 shows the result of this measurement for the n = 1 flexural
mode. As is evident, there is excellent agreement with the theoretical prediction.
Next, we examine the measured oscillation threshold dependence on the optical Q
factor. To illustrate behavior occurring in the sideband resolved and unresolved regimes
800
Probe
700
600
400
300
200
100
0
0 0.002 0.004 0.006 0.008 0.01 0.012
Inverse mechanical Q
Figure 10.14
Main figure: The observed threshold for the parametric oscillation (of an n = 1 mode) as a function of inverse
mechanical quality factor. In the experiment, variation of Q factor was achieved by placing a fiber tip in
mechanical contact with the silica membrane, which thereby allowed reduction of the mechanical Q (cf. inset).
The mechanical mode was a 6 MHz flexural mode. (From Optics Express. With permission.).
two mechanical modes were measured: a fundamental (n = 1) at 4.4 MHz and a third-or-
der mode (n = 3) at 49 MHz. The optical Q factor was adjusted by exciting different radial
and transverse optical modes. For lower optical Q, wherein the mechanical oscillation
frequency falls within the cavity bandwidth (i.e. the adiabatic regime), a rapid dependence
1/Q3 is observed, which is in agreement with the threshold equation (i.e. Pthresh∝1/F 3 cf.
Equation 10.21). However, the scaling of threshold changes once a transition from the unre-
solved (weak retardation) to the resolved sideband limit occurs. Indeed, for Ωm k, the
threshold dependence on optical Q (and finesse) weakens and eventually approaches an
asymptotic value for Ωm >> k. The deviation from the cubic dependence is indeed observed
experimentally as shown in Figure 10.15. The solid line in the figure is a prediction based
on the threshold Equation 10.20 with effective mass as an adjustable parameter (where
optimum detuning is assumed, and optimum coupling for each optical Q value). The
inset of Figure 10.15 shows the threshold behavior for the n = 3 mechanical mode which
for which Ωm k is satisfied for the entire range of Q values. As expected, a much weaker
dependence on optical Q-factor is found as predicted theoretically. Direct comparison with
the n = 1 mode data shows that oscillation on the n = 1 mode is preferred for lower optical
Qs. Indeed, preference to the n = 3 mode was possible by increased waveguide loading of
the microcavity in agreement with theory. The solid curve in the inset gives the single-
parameter fit to the n = 3 data yielding, meff = 5 × 10− 11 kg which is a factor of 660 lower than
the mass of the n = 1 mode.
105
104
n = 3 mode
Threshold (µW)
103
104
Oscillation threshold (µW)
102
103
101 6
10 107 108
Optical quality factor
102
n = 1 mode
101
106 107 108 109
Optical quality factor (Q0)
Figure 10.15
Main panel shows the measured mechanical oscillation threshold (in microwatts) plotted versus the optical Q
factor for the fundamental flexural mode (n = 1,Ωm/2π = 4.4 MHz, meff ≈ 3.3 × 10 − 8 kg, Qm ≈ 3500 ). The solid line is
a one-parameter theoretical fit obtained from the minimum threshold equation by first performing a minimi-
zation with respect to coupling (C) and pump wavelength detuning (∆), and then fitting by adjustment of the
effective mass. Inset: The measured threshold for the 3rd order mode (n = 3, Ωm/2p = 49 MHz, meff ≈ 5 × 10 − 11 kg,
Qm ≈ 2800) plotted versus optical Q. The solid line gives again the theoretical prediction. The n = 1 data from the
main panel is superimposed for comparison. (From Kippenberg, T. J., Rokhsari, H., Carmon, T., Scherer, A., and
Vahala, K. J., Physical Review Letters, 95, 033901, 2005) (From Optics Express. With permission.).
2
kBTR Ω m
∆Ω m ≅
P Qm
Here, P is the power dissipated in the mechanical oscillator, i.e. P = Γ mmeff Ω m2 x 2 .
Consequently the equation predicts that the linewidth and mechanical amplitude satisfy
∆Ωm ∝ 1/〈x2〉. Indeed, experimental work has confirmed this scaling in toroidal microcavi-
ties [67,68]. The measured, inverse-quadratic dependence (as first reported in Refs. [53,67]
is presented in Figure 10.16. Fundamental linewidths that are sub Hertz have been mea-
sured, however, with improvements in mechanical Q-factor, these values are expected to
improve, owing to the inverse-quadratic dependence appearing in the above formula.
10
δv= 53 x–2.04
Mechanical oscillation linewidth (Hz)
T = 390 K
1
δv= 38 x–1.98
T = 300 K
2 4 6 8 10
Mechanical oscillation amplitude (pm)
Figure 10.16
Linewidth measurements from Refs. [67,53] of the opto-mechanical oscillator for different amplitudes of oscilla-
tion plotted in picometers. The measurement is done at room temperature (dots) and at temperature 90°C above
room temperature (stars). The solid lines and the corresponding equations are the best fits to the log–log data.
Solid line denotes theoretically expected behavior (From Optics Express. With permission.).
detuning for ∆ = ± k/2 (since the mechanical amplification or cooling rate is much smaller
than the intrinsic mechanical damping rate). Conversion to displacement spectral density
then requires knowledge of the effective mass [51] of the mechanical mode which could be
independently determined in two ways. First, from finite-element simulation of the actual
cavity geometry parameters, as inferred from scanning electron microscopy images. As
detailed in Ref. [21] this is accomplished via the relation meff = (Em/Ω 2m (δR)2 ), where dR is
the mechanical energy causing a displacement in the radial direction. Second, the effec-
tive mass was determined by experimental measurements, by recording the threshold for
parametric oscillation for blue detuning and subsequently inverting the threshold equation
for the effective mass. Note that in the described experiments, both techniques agree very
well, from which the effective mass of the radial breathing mode of Figure 10.17 is inferred
to be meff = 1.5 × 10−11kg. Correspondingly, the rms motion caused by Brownian motion of the
radial breathing mode at room temperature is on the order of 〈x2〉1/2≈5 × 10−14m.
Figure 10.18 shows the displacement spectral density for the radial breathing mode
under conditions of constant red detuning (∆ = − k/2) for varying input power levels P. By
extrapolating the resonant frequency and linewidth to zero power (cf. inset) the intrinsic
resonance frequency and an intrinsic mechanical Q factor of 2890 were extracted. Note
that much higher mechanical Q (>50,000) are possible in an evacuated chamber [60] and by
optimizing the microcavity shape to reduce dissipative clamping losses. In the reported
experiment the optical linewidth was 50 MHz, equivalent to an optical Q factor of 4.4 × 106.
When varying the pump power, a clear reduction of the noise spectra is observed as shown
250
Linewidth (δm in kHz)
2.5 Teff = 74 K
200
150 Teff = 25 K
2 100
Teff = 34 K
Displacement (10−17 M/Hz1/2)
50 Γm/2π = 20 kHz
Teff = 11 K
0
0 1 2
1.5
Pump power (mW) ∆ · τ = 0.5
0.5
Figure 10.17
Main figure shows the normalized, measured noise spectra around the mechanical breathing mode frequency
for ∆ ⋅ τ ≈ − 0.5 and varying power (0.25, 0.75, 1.25, and 1.75 mW). The effective temperatures were inferred using
mechanical damping, with the lowest attained temperature being 11 K. (b) Inset shows increase in the linewidth
(damping δm = Γeff/2π) of the 57.1–MHz mode as a function of launched power, exhibiting the expected linear
behavior as theoretically predicted. (From Schliesser, A., Del’Haye, P., Nooshi, N., Vahala, K. J., and Kippenberg,
T. J., Physical Review Letters, 97, 243905, 2006) (From Optics Express. With permission.).
−30
Amplitude repsonse
−30 −40
−50
Amplitude response (dBV)
−35
−60
−70
−40 50 55 60 65
Frequency (MHz)
−45
−50
−55
102 104 106 108
Frequency (Hz)
Figure 10.18
The frequency response from 0 to 200 MHz of a toroidal opto-mechanical system. The plateau occurring above
1MHz is ascribed to the (instantaneous) Kerr nonlinearity of silica (dotted line). The high-frequency cutoff is due
to both detector and cavity bandwidth. The response poles at low frequency are thermal in nature. Inset: Data in
the vicinity of mechanical oscillator response which shows the interference of the Kerr nonlinearity and the radi-
ation pressure-driven micromechanical resonator (which, on resonance, is π/2 out-of-phase with the modulating
pump and the instantaneous Kerr nonlinearity). From the fits (solid lines) it can be inferred that the radiation
pressure response a factor of 260 larger than the Kerr response and a factor of ×100 larger than the thermo-
mechanical contribution (Adapted from Schliesser, A., Del’Haye, P., Nooshi, N., Vahala, K. J., and Kippenberg, T.
J., Physical Review Letters, 97, 243905, 2006) (From Optics Express. With permission.).
in the main panel. The shape of the spectra changes in two dramatic ways. First the lin-
ewidth of the mechanical spectra increase, owing to the fact that the light field provides
a viscous force, thereby increasing the damping rate. Note that this rate varies linearly
with applied power (cf. inset). Moreover, and importantly, the areas of the mechanical
curves also reduce, which are a direct measure of the mechanical breathing modes (RBM)
temperature. Indeed the effective temperature of the RBM is given by kBTeff = ∫Ω2meff Sx(Ω)
dΩ and correspondingly the peak of the displacement spectral density is reduced in a qua-
dratic fashion with applied laser power. While measurement of the calibrated noise spec-
tra is the most accurate way to determine the effective temperature, the simplified analysis
presented earlier (and neglecting any other heating mechanism) yields a temperature (cf.
Equation 10.17) given by the ratio of damping rates (with and without the pump laser):
Γm
Teff ≅ TR (10.22)
Γm + Γ
As noted before this formula only proves valid when (Γ m/Γ m + Γ ) < 1/Qm and for cool-
ing rates satisfying Γ << k. Note that the maximum temperature reduction factor is bound
by the cavity decay rate ~G/k. Thus, the highest temperature reduction factor which can
be attained is given by ~k/Gm. Furthermore, additional modifications of this expression
are necessary when entering the limit of temperatures, which correspond to only a few
quanta, and will be considered in the next section.
For the highest pump power (2 mW and 970 nm), the effective temperature was reduced
from 300 K to 11 K. This experiment, reported in Ref. [28], along with two earlier publica-
tions [25,26] represent the first demonstration of radiation-pressure back-action cooling.
It is noted for completeness that the experiment described in Ref. [26] also attributed an
appreciable cooling effect due to thermal effects in the mirror coatings. Indeed, of the
physical mechanisms that can create opto-mechanical coupling temperature is another
possibility. Temperature variations introduced by absorption, for example, create a well-
known, trivial coupling, by way of thermal expansion (and absorption of photons). Indeed
both mechanical amplification [72] and cooling [69] have been demonstrated using this
mechanism. In the present, microtoroid studies, thermal effects are negligible. This is
known, first, because of the parametric-instability studies of the previous section, where
both the observed threshold dependence on optical Q as well as the magnitude of the
threshold power, itself, are in excellent agreement with the theory of radiation-pressure-
induced coupling. Second, the recent observation of microwave-rate parametric oscilla-
tions [62] confirms the broadband nature of the underlying mechanism, greatly exceeding
bandwidths possible by thermal coupling mechanisms.
To quantitatively assess the contributions of thermal and radiation-pressure induced
effects in the present context, pump-probe-type response measurements were performed
using a second laser coupled to the cavity and operating in the wavelength region around
1550 nm. This laser serves as pump, providing a sinusoidally modulated input power
P(t) = P0(1 + ε sin(Ωt)), which in turn causes the optical resonances to periodically shift via
both thermal effects and radiation-pressure-induced mechanical displacement, but also
via the Kerr-nonlinearity of fused silica. These shifts are then “read out” with the 965
nm probe laser tuned to the wing of an optical resonance. Modulating the power of the
pump laser and demodulating the detected probe power with the same frequency, it is
possible to measure the microcavity response caused by all three nonlinearities (thermal,
Kerr, and radiation pressure). Note that due to their different spectral response, the three
nonlinearities can be readily differentiated when performing a frequency sweep. While
electronic modulation and demodulation are conveniently accomplished with an elec-
tronic network analyzer, we use a fibre-coupled interferometric LiNbO3 amplitude modu-
lator to generate the modulated pump light. Data were taken on a logarithmic frequency
scale between 50 Hz and 200 MHz, and, subsequently, on a linear frequency scale in the
interval 50–65 MHz. From finite-element simulation, this resonance can be identified as
the radial breathing mode. The result of this measurement, as first reported in Ref. [28], is
shown in Figure 10.18. Several features are evident from the graph and are now discussed
in detail.
First, the plateau in the response, for frequencies beyond 1 MHz and up to the cavity (and
detector) cut-off at 200 MHz, is attributed to the intensity-dependent refractive index (Kerr
effect) of fused silica [73,74]. Importantly, since both thermal and mechanical responses
exceed the Kerr nonlinearity in some frequency domains, the Kerr effect (plateau) pro-
vides a precisely known reference for all observed nonlinearities. Second, the response up
to a frequency of about 1 MHz can be well fitted assuming the sum of two single-poled
functions with cut-off frequencies of 1.6 kHz and 119 kHz. It has been attributed to the
response related to convective and conductive heat exchange of the cavity mode with its
environment [28,57]. Concerning this thermal-related response function, it is important to
note that a temperature change to the silica microcavity causes resonant frequency shifts
via both a change in the refractive index and a displacement due to thermo-mechanical
expansion. However, temperature-induced index changes dominate over thermal expan-
sion by a factor of at least 15 for glass. Therefore, the thermo-mechanical contribution to the
A+
nf = (10.23)
A - A+
-
In the unresolved k >> Wm sideband case, (which has been referred to as “weak binding”
in atomic laser cooling [79]) this prevents ground state cooling as:
k
nf ≈ 1 (10.24)
4Ω m
k2
nf ≈ 1 (10.25)
16Ω m2
If also the contribution from reservoir heating is included, the final occupancy that can
be achieved then takes the form
A+ Γm
nf = + nR
A - A+ Γ m + Γ
-
Here the nR average denotes the average occupancy of the harmonic oscillator prior to
applying radiation pressure cooling. As before, this limit is valid over the range where the
cooling satisfies (Γ m + Γ/Γ m ) < nR and G < k. Note that from the above expression it becomes
clear that pre-cooling of the mechanical oscillators is advantageous and is indeed pres-
ently undertaken by several groups [31]. Resolved sideband cooling has also recently been
demonstrated experimentally [46].
Ωm << κ Ωm >> κ
ΩmΩm Ωm Ωm
}
}
Power
Power
κ κ
Frequency Frequency
~= κ/4Ω >> 1
nmin = ~= κ2/16Ω2 << 1
nmin =
m m
Figure 10.19
Quantum limits of radiation pressure cooling. Left panel shows the regime of unresolved sidebands whereas
the right panel shows the resolved sideband case. Note that only in the resolved sideband regime average final
occupancies below unity can be attained (i.e. ground state cooling). (Adapted from Schliesser, A., Riviere, R.,
Anetsberger, G., Arcizet, O., and Kippenberg, T. J., Nature Physics, doi:10.1038/nphys939, 2008) (From Optics
Express. With permission.).
in Section 10.2 the mechanical motion produces blue-shifted sidebands which remove
mechanical quanta from the mechanical oscillator. These generated photons decay during
the cavity lifetime 1/k into the outside world (i.e. into the tapered optical fiber waveguide,
or through the finite reflection of the mirror in the case of a Fabry–Perot). The finite decay
time of the cavity however entails that the energy that is carried away by a blue-shifted
photon has a Heisenberg limited uncertainty of ∆E = /∆t ≈ k . Consequently, the mean
energy of the mechanical oscillator E = Ωm(n + 1/2) cannot be lower than this limit, imply-
ing a final Doppler temperature of TD ≈ κ/kB in entire analogy to laser cooling in atomic
physics [7]. This Doppler temperature entails that the final occupancy is much larger than
unity in the case where the mechanical oscillator frequency is smaller than the cavity lin-
ewidth (since k >> Wm, see Figure 10.19). In contrast, ground state cooling is possible when
then energy quantum of the mechanical oscillator ( Ωm) is made large in comparison with
the energy scale set by the Doppler temperature. This required the “strong binding condi-
tion” to be satisfied, i.e. Wm >> k (Figure 10.19).
A second consideration that can be used to successfully estimate the quantum limits of
back-action cooling is to consider the mechanical mirror as performing a measurement on
the intra-cavity photons. As in the case of photo-detector shot noise, the random arrival
time of photons onto the mirror will entail a fluctuation of the radiation pressure force that
will heat the mechanical oscillator. If N is the average number of photons in the cavity, the
fluctuation of photon numbers (governed by Poissonian statistics) is given by N , entail-
2 1/ 2
ing a radiation pressure force fluctuation of ∆FRP = N (2k/Trt ) . Approximating this
fluctuation as a white noise spectrum over the cavity bandwidth, the corresponding force
spectral density of the radiation pressure quantum noise is given by δFRP 2 2
(Ω) = ∆FRP /k.
This effectively white spectrum provides a driving force for the mechanical oscillator,
and can be used to determine the final temperature via the relation kBTF = ∫Ω2meff χeff(Ω)δ
FRP(Ω) 2dΩ = Γ − 1meffδFRP(Ωm) 2 (where χeff(Ω) is the modified mechanical oscillators sus-
ceptibility). Two limits can be derived. First, assuming that the cavity linewidth exceeds
the mechanical oscillator frequency, the noise spectrum is approximately given by
2
δFRP (Ω) ≈ ( P/ω )(1/2)τ(k/Trt )2 = ( Pω / 2 L2 )τ 2 which yields, i.e. kBTf ≈ κ/4 (i.e. n f ≈ k/4Ω m
recovering the result derived in Refs. [47,48]). Here we have assumed a Fabry–Perot cavity of
length L. On the other hand considering the resolved sideband regime (i.e. assuming laser
and D = − Wm) the spectral density of the radiation pressure quantum fluctuations at the fre-
quency of the mechanical oscillator are given δFRP 2
(Ω m ) ≈ ( P ω/2 L2 Ω m ) yielding the result
kBT f = (Ω m/2)(1 + k 2/8Ω 2m ) or n f ≈ k 2/16Ω 2m. Thus, the final occupancy in the unresolved and
resolved sideband case can be understood as arising from the quantum fluctuation of the
intra-cavity field and are in agreement with the results of a rigorous calculation.
[18] to chaotic regimes [61], some of which have already been observed in an experimental
setting and allow one to create switchable tunable optical filters [43]. Furthermore, para-
metric amplification of mechanical modes constitutes an entirely new way of creating the
analogue of a “photonic quartz oscillator” which is driven purely by the radiation pressure
of light and whose linewidth is limited by thermal noise. With continued improvements in
mechanical Q that are already underway to address the requirements of cooling-related-
research, there is the potential for realization of a new class of ultra-stable, narrow lin-
ewidth RF oscillators.
On the cooling side, there is a rich history of theoretical proposals pertaining to entan-
gling mechanical oscillators with a light field using radiation pressure [13,36,37]. Most of
these will require achieving temperatures at which the mechanical system is close-to, or
at, the quantum ground state. With the rapid progress towards realization of ground-state
cooling, it now seems likely that many of these ideas and proposals can be tested over
the next decade. Similarly, macroscopic mechanical modes can provide a new medium in
which to explore quantum information phenomena, as has been true in the rich scientific
arena of cavity QED [80]. Indeed, one can view cavity QOM as the logical extension of
cavity QED into the macroscopic realm.
Another interesting goal is to reach a regime where cavity QOM phenomena become
observable. For example, a regime where the back-action from the quantum noise of the
radiation pressure dominates the thermal noise [9,10] would enable experimental realiza-
tions of proposals such as the quantum nondemolition measurement of photon number
or pondermotive squeezing [13]. Moreover, recent work from Yale [31] has demonstrated
cavity OM systems which realize a quadratic coupling to the mechanical coordinate,
thereby lending themselves to perform QND measurements of the mechanical eigenstate.
As originally noted, both cooling [24] (“tranquilizer”) and parametric instability [81] were
first conceived theoretically in the context of gravitational-wave detection by Braginsky.
No doubt, the better understanding of these phenomena gained by their demonstration in
the microscale will benefit this important field. Along these lines, since the initial observa-
tion of radiation pressure parametric instability in microtoroids [21–23] in 2005, the MIT
gravitational-wave group has reported observation of parametric instability [82]. Even
more recently they have also reported the cooling of gram scale mirror modes [30]. These
results bode well for further progress in this field.
From a practical point-of-view, the ability to achieve cooling and oscillation of microme-
chanical modes on a semiconductor chip bodes well for realization of new technologies
that could leverage these new tools. Specifically, miniaturization and integration of these
functions with electronics and other optical functions is already possible because of this
microfabricated, chip-based platform. Also significant is that radiation-pressure cooling
through dynamic back action, as already noted, is a highly targeted form of cooling in
which a selected mechanical mode(s) can be precisely defined to receive the benefit of
cooling, while other modes remain at elevated or even at room temperature. Indeed, the
first demonstrations of back-action cooling were to temperatures in the range of 10 K, but
featured mechanical structures that were otherwise uncooled and at room temperature.
As a result, this novel form of cooling offers ultra-low temperature performance with rela-
tively low power requirements (milli-Watts) and without the need for cryogenics, vacuum
handling, or any of the other necessities of conventional refrigeration. This feature, above
all others, would seem to offer the greatest advantages in terms of new technologies.
Finally, the range of phenomena that have been described here extend over the entire elec-
tromagnetic spectrum. Indeed, the concept of dynamic back action was conceived-of first
in the microwave realm [11]. It is also important to note that there are electro-mechanical
systems that provide an analogous form of back-action cooling [16,17]. These systems can
potentially provide a means to achieve the quantum ground-state.
In summary, cavity (quantum) opto-mechanics represents many of the concepts of atomic
and molecular physics, however embodied in an entirely different macroscale system. It is
currently experiencing rapid experimental and theoretical success in various laboratories
worldwide and offers entire new inroads for basic science and potentially new technolo-
gies. It seems clear that this field is entering an exciting period of experimental science.
Acknowledgments
KJV acknowledges the Caltech Lee Center and DARPA for supporting this work. TJK
acknowledges support via a Max Planck Independent Junior Research Group, a Marie
Curie Excellence Grant (MEXT-CT-2006-042842) and the Nanosystems Initiative Munich
(NIM). The authors kindly thank Albert Schliesser, Olivier Arcizet, Jens Dobrindt , Stefan
Weis, Emanuel Gavartin and Mani Hossein-Zadeh and Jiang Li for contributions to this
review.
References
1. T. J. Kippenberg and K. Vahala, Optics Express, 15, 17172, 2007.
2. K. J. Vahala. Nature, 424, 839, 2003.
3. H. G. Craighead. Science, 290, 1532, 2000.
4. T. W. Hänsch and A. L. Schawlow. Optics Communications, 13, 68, 1975.
5. D. J. Wineland, R. E. Drullinger, and F. L. Walls. Physical Review Letters, 40, 1639, 1978.
6. S. Chu, L. Hollberg, J. E. Bjorkholm, A. Cable, and A. Ashkin. Physical Review Letters, 55, 48,
1985.
7. S. Stenholm. Reviews of Modern Physics, 58, 699, 1986.
8. C. M. Caves. Physical Review D, 23, 1693, 1981.
9. K. Jacobs, I. Tittonen, H. M. Wiseman, and S. Schiller. Physical Review A, 60, 538, 1999.
10. I. Tittonen, G. Breitenbach, T. Kalkbrenner, T. Muller, R. Conradt, S. Schiller, E. Steinsland,
N. Blanc, and N. F. de Rooij. Physical Review A, 59, 1038, 1999.
11. V. B. Braginsky. Measurement of Weak Forces in Physics Experiments. University of Chicago Press,
Chicago, IL, 1977.
12. V. B. Braginsky and F. Khalili. Quantum Measurement, Cambridge University Press, Cambridge,
England, 1992.
13. S. Mancini and P. Tombesi. Physical Review A, 49, 4055, 1994.
14. S. Bose, K. Jacobs, and P. L. Knight. Physical Review A, 56, 4175, 1997.
15. M. D. LaHaye, O. Buu, B. Camarota, and K. C. Schwab. Science, 304, 74, 2004.
16. Naik, O. Buu, M. D. LaHaye, A. D. Armour, A. A. Clerk, M. P. Blencowe, and K. C. Schwab.
Nature, 443, 193, 2006.
17. K. Brown, J. Britton, R. Epstein, J. Chiaverini, D. Leibfried, and D. Wineland. Physical Review
Letters, 99, 137205, 2007.
18. Dorsel, J. D. McCullen, P. Meystre, E. Vignes, and H. Walther. Physical Review Letters, 51, 1550, 1983.
19. S. Sheard, M. B. Gray, C. M. Mow-Lowry, D. E. McClelland, and S. E. Whitcomb. Physical Review
A, 69, 2004.
20. V. B. Braginsky, S. E. Strigin, and S. P. Vyatchanin. Physics Letters A, 287, 331, 2001.
21. T. J. Kippenberg, H. Rokhsari, T. Carmon, A. Scherer, and K. J. Vahala. Physical Review Letters,
95, 033901, 2005.
22. H. Rokhsari, T. J. Kippenberg, T. Carmon, and K. J. Vahala. Optics Express, 13, 5293, 2005.
23. T. Carmon, H. Rokhsari, L. Yang, T. J. Kippenberg, and K. J. Vahala. Physical Review Letters, 94,
2005.
24. V. B. Braginsky and S. P. Vyatchanin. Physics Letters A, 293, 228, 2002.
25. O. Arcizet, P. F. Cohadon, T. Briant, M. Pinard, and A. Heidmann. Nature, 444, 71, 2006.
26. S. Gigan, H. R. Bohm, M. Paternostro, F. Blaser, G. Langer, J. B. Hertzberg, K. C. Schwab, D.
Bauerle, M. Aspelmeyer, and A. Zeilinger. Nature, 444, 67, 2006.
27. M. Poggio, C. L. Degen, H. J. Mamin, and D. Rugar. Physical Review Letters, 99, 2007.
28. Schliesser, P. Del’Haye, N. Nooshi, K. J. Vahala, and T. J. Kippenberg. Physical Review Letters, 97,
243905, 2006.
29. P. F. Cohadon, A. Heidmann, and M. Pinard. Physical Review Letters, 83, 3174, 1999.
30. T. Corbitt, Y. B. Chen, E. Innerhofer, H. Muller-Ebhardt, D. Ottaway, H. Rehbein, D. Sigg, S.
Whitcomb, C. Wipf, and N. Mavalvala. Physical Review Letters, 98, 150802, 2007.
31. J. D. Thompson, B. M. Zwickl, A. M. Yarich, F. Marquardt, S. M. Girvin, and J. Harris.
arXiv:0707.1724, 2007.
32. S. Mancini, D. Vitali, and P. Tombesi. Physical Review Letters, 80, 688, 1998.
33. S. Vandermeer. Reviews of Modern Physics, 57, 689, 1985.
34. D. Kleckner and D. Bouwmeester. Nature, 444, 75, 2006.
35. O. Arcizet, P. F. Cohadon, T. Briant, M. Pinard, A. Heidmann, J. M. Mackowski, C. Michel, L.
Pinard, O. Francais, and L. Rousseau. Physical Review Letters, 97, 133601, 2006b.
36. V. Giovannetti, S. Mancini, and P. Tombesi. Europhysics Letters, 54, 559, 2001.
37. S. Mancini, V. Giovannetti, D. Vitali, and P. Tombesi. Physical Review Letters, 88, 120401, 2002.
38. J. M. Courty, A. Heidmann, and M. Pinard. Physical Review Letters, 90, 2003.
39. O. Arcizet, T. Briant, A. Heidmann, and M. Pinard. Physical Review A, 73, 2006.
40. W. Marshall, C. Simon, R. Penrose, and D. Bouwmeester. Physical Review Letters, 91, 2003.
41. M. Hossein-Zadeh and K. J. Vahala. Volume 20, No. 4, February 2008, Page(s): 234–236.
42. M. L. Povinelli, J. M. Johnson, M. Loncar, M. Ibanescu, E. J. Smythe, F. Capasso, and J. D.
Joannopoulos. Optics Express, 13, 8286, 2005.
43. M. Eichenfeld, C. Michael, R. Perahia, and O. Painter. Nature Photonics, 1, 416, 2007.
44. K. C. Schwab and M. L. Roukes. Physics Today, 58, 36, 2005.
45. Y. Akahane, T. Asano, B. S. Song, and S. Noda. Nature, 425, 944, 2003.
46. A. Schliesser, R. Riviere, G. Anetsberger, O. Arcizet, and T. J. Kippenberg. Nature Physics,
doi:10.1038/nphys939, 2008.
47. F. Marquardt, J. P. Chen, A. A. Clerk, and S. M. Girvin. Physical Review Letters, 99, 093902, 2007.
48. I. Wilson-Rae, N. Nooshi, W. Zwerger, and T. J. Kippenberg. Physical Review Letters, 99, 093902,
2007.
49. K. Law. Physical Review A, 51, 2537, 1995.
50. H. A. Haus. Electromagnetic Fields and Energy. Prentice Hall, Englewood Cliff, NJ, 1989.
51. M. Pinard, Y. Hadjar, and A. Heidmann. European Physical Journal D, 7, 107, 1999.
52. F. Marquardt, J. Harris, and S. Girvin. Physical Review Letters, 96, 2006.
53. M. Hossein-Zadeh and K. J. Vahala. Optics Letters, 32, 1611, 2007.
54. K. Karrai. Nature, 444, 41, 2006.
55. V. Vuletic and S. Chu. Physical Review Letters, 84, 3787, 2000.
56. P. Maunz, T. Puppe, I. Schuster, N. Syassen, P. W. H. Pinkse, and G. Rempe. Nature, 428, 50,
2004.
57. V. B. Braginsky, M. L. Gorodetsky, and V. S. Ilchenko. Physics Letters A, 137, 393, 1989.
58. T. J. Kippenberg, S. M. Spillane, D. K. Armani, and K. J. Vahala. Applied Physics Letters, 83, 797,
2003.
59. K. Armani, T. J. Kippenberg, S. M. Spillane, and K. J. Vahala. Nature, 421, 925, 2003.
Contents
11.1 Introduction to Optical Frequency Combs.....................................................................483
11.2 Frequency Comb Generation from a Monolithic Silica Microresonator..................... 485
11.2.1 Physics of the Comb Generation Process............................................................ 485
11.2.2 Verification of the Comb Spectrum Equidistance.............................................. 488
11.2.2.1 Multiheterodyne Spectroscopy.............................................................. 488
11.2.2.2 Proving the Equidistance of the Mode Spacing at the mHz Level..... 489
11.2.3 Dispersion in Toroidal Microresonators.............................................................. 491
11.3 Stabilization of the Comb.................................................................................................. 493
11.3.1 Principle................................................................................................................... 494
11.3.2 Implementation....................................................................................................... 495
11.3.3 Characterization of the Locking Mechanism..................................................... 496
11.3.4 Actuation Properties.............................................................................................. 498
11.4 Generation of a Stabilized Microwave Repetition Rate Frequency Comb................. 499
11.4.1 Monolithic Frequency Comb Generators with Microwave
Repetition Rate........................................................................................................500
11.4.2 Stabilization and Characterization of a Microwave Frequency Comb........... 501
11.5 Conclusion........................................................................................................................... 503
Acknowledgments.......................................................................................................................504
References......................................................................................................................................504
483
© 2009 by Taylor & Francis Group, LLC
In time domain, fceo describes the phase slippage of the pulse envelope with respect to
the carrier. Thus, any comb line of the resulting output spectrum can be expressed by
f m = fceo + mfrep, where m is an integer. Measurement and control of frep and fceo allows
to phase-coherently link optical frequencies across the entire spectrum spanned by the
comb [8,9]. In this case, every comb tooth is uniquely determined by the experimentally
controlled quantities fceo and frep. Therefore, any arbitrary optical frequency within the
spectrum of the comb may be synthesized, or, vice versa, any given optical frequency
within this spectrum may be phase-coherently compared to the frequencies fceo and frep in
the RF domain, which can be referenced to microwave time standard [10,11]. This unique
ability has made the frequency comb a revolutionary tool in metrology and laser spec-
troscopy, and, serving as a clockwork mechanism, has enabled the development and
practical use of time standards in the optical domain [2,3,12,13]. Beyond these advances
in metrology and precision measurements, frequency combs have given new impetus
to applications such as broadband laser-based gas sensing [14–16] or cavity ring down
spectroscopy [17].
In an earlier work [18], Kourogi generated optical frequency combs by inserting a phase
modulator inside a cavity. Injection of a continuous wave laser resonant with one of the
cavity mode lead to the generation of multiple sidebands when the modulator driving fre-
quency is matched to a multiple of the cavity’s modespacing, in this case, the bandwidth of
the comb was limited by the dispersion of the cavity and phase modulator. Combs gener-
ated in this way could span several terahertz. This work was refined in later research at the
Joint Institute for Laboratory Astrophysics (JILA) [19,20].
Prior to this work, the ability to link an optical frequency down to the RF domain
in a phase-coherent way was pursued in experiments at the Physikalisch-Technische
Bundesanstalt (PTB) in Braunschweig, Germany, following an approach taken at the
National Bureau of Standards (NBS, now National Institute of Standard and Technology,
NIST) in Boulder, CO. Nonlinear optical processes, in conjunction with a plethora of aux-
iliary oscillators, were used to phase-coherently link a Cs atomic clock reference to the
optical domain [21]. For the task of bridging a large frequency interval, researchers at the
Max Planck Institute of Quantum Optics (MPQ) in Garching, Germany, had used another
approach: in their so-called optical interval divider chain, the interval to be measured was
successively divided using several identical compact semiconductor lasers, until a direct
microwave signal could be measured [22].
A revolutionary and extremely powerful tool—that made frequency chains obsolete—
was realized with the demonstration that the optical frequency combs from a femtosecond
laser can be fully stabilized and phase-coherently compared to the RF standards. While
mode-locked lasers naturally generate an optical frequency comb owing to the periodic
nature of the optical emission, a key ability of a frequency comb is to stabilize the optical
spectrum in terms of the absolute frequency of each comb tooth. While locking of the mode
spacing of a mode-locked laser can be readily implemented, full stabilization requires mea-
suring and locking the carrier envelope frequency (fceo). A powerful technique to measure
the carrier envelope frequency and thereby creating an optical frequency synthesizer is
the use of an f − 2f (or 2f − 3f) interferometer, which requires in the former case however an
optical spectrum that spans a full octave. With the advent of photonic crystal fibers, that
exhibited a zero dispersion wavelength in the 800 nm range, it became possible to broaden
the output spectrum of a Ti:Sapphire laser system to a full octave and thereby measure (and
stabilize) for the first time the carrier envelope offset frequency [10,11,13]. This marked the
beginning of a new era of frequency metrology, and led to new advances in diverse fields
such as optical frequency metrology and precision measurements and in applications such
as broadband laser-based gas sensing [17] and molecular fingerprinting [14–16]. Moreover
frequency combs are an integral part of a new era of atomic clocks, based on optical tran-
sitions (exhibiting very narrow transition linewidth and high frequency) such as those
afforded by a single trapped 133Hg + ion [23], or Sr atoms confined in an optical lattice [24].
* This process has also been referred to as hyperparametric frequency conversion [25].
νpump + νpump νI + νS
m–1
m
Silica Tapered fiber
microtoroid evanescent
m+1
coupling
m
νS νpump νI ν
20 λp = 1,550 nm
∆λ = 7 nm νpump + νI νII + νS
Optical power (dBm)
10
m+2
m+1
0
–10
–20
m–1
m
–30
Figure 11.1
Optical frequency comb generation in a microresonator. Panel (a) shows the measured comb spectrum via the
output of the tapered optical fiber. The individual comb modes are spaced by the free spectral range of the cav-
ity. The change in the baseline observed close to the pump wavelength is due to the erbium doped fiber ampli-
fier used to enhance the launched pump power. Panel (b): Physical principle of the comb generation process.
Degenerate and nondegenerate four-wave mixing allow conversion of a CW laser into an optical frequency
comb. (Adapted from Del’Haye, P., Schliesser, A., Arcizet, O., Wilken, T., Holzwarth, R., and Kippenberg, T. J.,
Nature, 450, 1214, 2007.)
on an FWM process mediated by the intensity dependent refractive index of silica (the Kerr
nonlinearity, n2 = 2.3 × 10 − 16 cm2/W) that initiates the process and results in the creation of
coherent signal and idler photons from two pump photons: νpump + νpump → νI + νS (see Figure
11.1b). The conservation of energy ensures that the generated photon pairs are symmetric in
frequency with respect to the pump. This mechanism is resonantly enhanced by the cavity
if idler, signal and pump frequencies all coincide with optical modes of the microresona-
tor. Parametric oscillations from the pump into signal and idler sidebands had first been
observed in optical microcavities in 2004, in both silica microtoroidal resonators [30] and
crystalline CaF2 resonators [31] with very low thresholds (tens of μW), as a consequence
of the cavity enhancement. CaF2 resonators are typically larger and consequently have a
smaller free spectral range in the order of 10 GHz. The larger mode volume in this kind of
resonators is compensated by a higher optical quality factors of more than 1010 [32].
It is important to note that this process can cascade via nondegenerate FWM among
pump and first-order sideband, to produce higher order sidebands (for example νpump +
νI → νII + νS) as can be seen in Figure 11.1. This cascaded mechanism ensures that the fre-
quency difference of pump and first-order sidebands νpump − νI = νS − νpump is exactly
transferred to all higher-order inter-sideband spacings. Thus, provided that the cavity
exhibits a sufficiently equidistant mode spacing, successive FWM to higher orders intrin-
sically leads to the generation of phase-coherent sidebands with equal spacing, that is, an
optical frequency comb (termed Kerr comb in the remaining discussion). Indeed, the process
Pump
Cavity FSR
Figure 11.2
The role of dispersion in broadband comb generation. Due to the dispersion in the cavity, its free spectral range
(FSR) varies with the optical frequency, so that the cold cavity resonances (blue Lorentzians) are not spaced
equidistantly in frequency space. The generated optical frequency comb (red lines), in contrast, is perfectly
equidistant. Therefore the generated sidebands walk off from the cavity resonances with increasing sideband
order, reducing the cavity enhancement of the four-wave mixing process. As a consequence, uncompensated
cavity dispersion can eventually limit the comb bandwidth.
can cascade over a wide frequency span, determined by the condition that the frequency
of higher order converted photons must be sufficiently close to a microcavity optical reso-
nance so that the process can still be resonantly enhanced by the optical cavity [31] (cf.
Figure 11.2). Therefore, the comb generation process is naturally limited by the fact that
dispersion renders the cavity modes intrinsically nonequidistant.
Dispersion in microcavities arises from two contributions resulting from material and
resonator waveguide dispersion. In the case of silica microcavities these two contributions
have opposite signs allowing partial compensation of dispersion in the 1550-nm window.
Indeed, measurements have verified that the optical modes exhibit only around 20 MHz
deviation from equidistance for the cold cavity over a 100-nm wide span. This value is
comparable to the frequency bandwidth of the individual optical resonances, of the order
of 10 MHz for the considered sample when overcoupled [33], which explains the broad-
band frequency comb generation. Note that self- and cross-phase modulation mechanisms
also help to compensate the residual dispersion [30].
It is important to note that the parametric process itself could in principle produce couples
of signal/idler sidebands that are only pair-wise equidistant but not mutually equidistant
as required for a comb. It is thus required to verify the equidistance of the spectrum. In
the next section, a measurement is described which has allowed to prove the equidistance
between the different lines at the mHz level, demonstrating the generation of a frequency
comb from a continuously pumped monolithic high-Q microcavity [29].
Finally, in addition to energy conservation, parametric oscillations also require angu-
lar momentum conservation to be satisfied for a rotational invariant geometry. In a
microcavity of the whispering gallery type, the optical eigenmodes are also eigenstates
of the angular momentum and have discrete propagation constants given by βm = m/R
as a consequence of the periodic boundary condition (ei2πβmR = 1), where m is the angular
mode number. Consequently the angular phase matching condition is intrinsically sat-
isfied when the signal and idler photons correspond to adjacent cavity modes, since in
this case 2βpump = βI + βS. Analogous reasoning also applies to the case of nondegenerate
FWM (e.g. βpump + βI = βII + βS). It is important to note that in contrast to the second order
nonlinearity, where the phase matching condition has to be ensured for frequency dif-
ferences in the optical range, the FWM only requires phase matching condition over the
frequency range occupied by the optical modes involved in the parametric frequency
conversion process.
FFT
PD Microtoroid
QWP
QWP
HWP
HWP
HWP
ESA
Kerr comb
Grating PBS PBS Mode spacing 1 THz
Beat detection unit
(b)
300
Signal (µV)
0
1 2 3 4 5 6 7 8 9
Frequency (MHz)
Figure 11.3
Multi-heterodyne beat note detection. (a) Experimental setup employed. ECDL, external cavity diode laser;
FLFC, femtosecond laser frequency comb; PBS, polarizing beam splitter; HWP, half-wave retarder plate; QWP,
quarter-wave retarder plate; PD, photodiode; FFT, fast-Fourier transform analyzer; ESA, electronic spectrum
analyzer. (b) Radio frequency spectrum of nine simultaneously oscillating microcavity modes beating against
reference comb modes. (Data from Del’Haye, P., Schliesser, A., Arcizet, O., Wilken, T., Holzwarth, R., and
Kippenberg, T. J., Nature, 450, 1214, 2007.)
possible to make the spacing Df of the resulting beats as small as ~1 MHz, so that all beat
frequencies f0 + k∆f of interest fall in the window between DC and fr/2.
The results for such a measurement on nine microcavity comb lines incident on a sin-
gle BDU, covering a span of 50 nm or 6.5 THz, are shown in Figure 11.3c. No deviation
from equidistance can be found at the level of 5 kHz, limited by the acquisition time.
This proves the equidistance for all nine lines at a level of 10 −9 (or 10 −11 when referenced
to the optical carrier). While higher resolutions and signal-to-noise ratios (SNR) would in
principle be available by extending the measurement time, the mutual fluctuations of the
two free-running comb generators however prevent such improvements in practice.
11.2.2.2 Proving the Equidistance of the Mode Spacing at the mHz Level
Verification of the equidistance with higher resolution necessitates reducing the mutual
fluctuations of the microcavity and reference comb. This can be accomplished by locking
the EDCL generating the microcavity comb to the reference comb using an offset lock.
To this end, a first BDU is used to detect the beat between the ECDL and the closest
reference comb mode. The phase of the resulting RF-beat is compared to a stable
RF-reference at the “offset-frequency” f0, obtained from a synthesizer referenced to the
MPQ in-house hydrogen maser. By feeding back on the frequency of the pumping laser,
the two RF signals are held in phase. On time scales longer than the feedback’s response
time the frequency of the pump laser is thus rigidly connected to the closest reference
frequency comb mode, with a frequency offset of f0. The stability of the measurement
is further improved by locking the repetition rate of the reference comb to around
100 MHz. These improvements enable measurements of much longer durations, up to
several hundreds of seconds, without degradation of the beat signals due to mutual fluc-
tuations. Measurement of radio frequencies on such a long timescale can be accomplished
using radio-frequency counters. In order to achieve the SNR required for reliable count-
ing, it is advantageous to detect each beat in a dedicated BDU. This allows an individual
optimization of the polarization and power ratio of the input beams as well as a selection
of narrow spectral regions reducing the amount of background light incident on the pho-
toreceiver. The achieved SNR exceeding 30 dB in a 500-kHz bandwidth proved sufficient,
in spite of still lacking about 26 dB to the shot-noise limit [36] of SNR opt ηPrcm /hν ⋅ RBW ,
detection efficiency h~0.5, power of the reference comb mode beating with the microcav-
ity Kerr comb mode Prcm 50 nW, optical frequency ν 200 THz and detection band-
width RBW = 500 kHz.
To verify the equidistance of the microcavity Kerr comb, at least three modes have to be
measured simultaneously. One mode of the microcavity comb is the pump laser itself; its
offset to the closest reference comb mode is locked to f0. It is therefore sufficient to measure
two more beats of the microcavity comb against the reference comb. Figure 11.4 shows the
setup employed in the most accurate measurement of microcavity combs performed so far
[29]. The two other counted beats f1 and f2 correspond to the microcavity comb lines which
are located at five and seven FSRs from the pump laser. A measure for the deviation from a
perfectly equidistant mode spacing is the difference of the average mode spacing between
the three measured lines ε = (ƒ2 − ƒ1)/(7 − 5)−(ƒ1 − ƒ0)/5. For reproducible results, all counters
were referenced to the same hydrogen maser and triggered by a common trigger signal to
ensure simultaneous measurements.
A slightly modified RF setup could even make one counter obsolete, by directly count-
ing a frequency ratio instead of absolute counting of two radio frequencies. This was
achieved by mixing the outputs of both BDU2 and BDU3 with the reference frequency f0
in order to directly obtain f1 − f0 and f 2 − f0 at the outputs of the mixers. Counting the ratio
R = (f 2 − f0)/(f 1 − f0) allows direct calculation of the deviation ε according to ε = (R − 7/5)·
(f1 − f0)/2, where f1 − f0 provides only a scale factor and needs not to be known very accu-
rately. This method yielded slightly better results, which is attributed to latencies in the
counters degrading temporal overlap of the counting intervals in spite of a common trig-
ger signal.
Figure 11.5 illustrates the results and statistics of a series of measurements in the two-
counter-configuration. The results obtained for ε were found to be normally distributed
around a mean of (− 0.91 ± 5.5) mHz with a standard deviation of 322 mHz. Together with
many more measurements in the ratio-counting configuration, accumulating to a total
measurement time of several hours, the mean deviation from equidistance could be
bracketed to (− 0.8 ± 1.4) mHz, confirming the uniformity of the comb spacing to a level
of 7.3 × 10−18 relative to the optical carrier, and 5.2 × 10−16 relative to the 2.1 THz span of the
measured microcavity comb line [29].
(a)
Kerr comb
Pump
laser
∆ν ≈ 1THz Pump Sideband N Sideband M
BDU 1 BDU 2 BDU 3
Reference comb
f r ≈ 100 MHz
(b)
20 Pump
M · ∆ν ν2
N = 5, M = 7
ν0
Optical power (dBm)
10 N · ∆ν
ν1
0
–10
–20
–30
1540 1550 1560 1570
Wavelength (nm)
Figure 11.4
High-resolution measurement of the microcavity comb structure. (a) Simplified schematic of the setup. Three
independent BDUs are used. With the signal from BDU 1, the pump laser is locked to one comb-line of the refer-
ence comb with an offset frequency f0. BDU 2 and BDU 3 count the beats of the 5th and 7th sideband of the micro-
cavity comb spectrum shown in (b) against the reference comb modes. (Adapted from Del’Haye, P., Schliesser,
A., Arcizet, O., Wilken, T., Holzwarth, R., and Kippenberg, T. J., Nature, 450, 1214, 2007.)
( )
1/ 3
c m+1/ 2
νm = m + 1/ 2 + η1 + … , (11.1)
2 πnR 2
(a)
3493 counts Mean –0.91±5.46 mHz
200 1 s gate time
Counts
Standard deviation 322 mHz
100
0
–1.5 –1.0 –0.5 0.0 0.5 1.0 1.5
Deviation from equidistant modespacing, ε (Hz)
(b)
10–15
Allan deviation, σA
10–16
σA = 1.3×10–15·τ–0.53
1 10 100 1,000
Gate time, τ (s)
Figure 11.5
Verification of equidistant mode spacing. (a) The statistics of the measurement results for the deviation from
equidistance ε in a 1-s gate time. The results are normally distributed. (b) Allan deviation as a function of gate
time, displaying an approximately inverse square-root dependence. (Adapted from Del Haye, P., Schliesser, A.,
Arcizet, O., Wilken, T., Holzwarth, R., and Kippenberg, T. J., Nature, 450, 1214, 2007.)
where c is vacuum speed of light, n the refractive index, R the cavity radius and −η1 the first
zero of the Airy function (η1 ≈ 2.34). As a consequence, the variation of the free spectral
range ∆νFSR = (νm + 1− νm)−(νm − νm−1)≈ ∂2νm/∂m2 is given by
( )
-5 / 3
c η m+1/ 2 c
∆νFSR = - ⋅ 1 ≈ - 0.41 m-5// 3 < 0 (11.2)
2 πnR 18 2 2 πnR
Evidently, the free spectral range reduces with increasing frequency corresponding to a
negative group velocity dispersion (GVD). Low frequency modes exhibit a shorter round
trip time than high frequency modes (i.e. the dispersion is normal). This behavior can be
understood by noting that higher frequency modes spatially move progressively closer to
the cavity boundary, making the classical optical trajectory larger.
A second contribution is due to the material dispersion of the fused silica material con-
stituting the resonator. Considering the fact that the refractive index n is actually a func-
tion of frequency (and therefore mode number m), n ≡ n(m), the GVD of fused silica alone
would lead to a FSR variation of
∂2 c c 2λ 2
∆νFSR ≈ 2 ⋅ m ≈ 2 3 2 ⋅ GVD , (11.3)
∂m 2 πn(m)R 4 π n R
where GVD = − (λ/c)(∂2n/∂λ2) is the group velocity dispersion of fused silica. This mate-
rial parameter is well-known to change its sign in the 1300-nm wavelength region from
60
Resonator (R=40 µm)
40 Resonator + Material (R=40 µm)
Resonator + Material (R=80 µm)
Variation of FSR (MHz)
20
–20
–40
–60
600 800 1000 1200 1400 1600 1800 2000 2200
Wavelength (nm)
Figure 11.6
Variation of the free spectral range of two whispering-gallery microsphere resonators of radii 40 µm and
80 µm including the effect of silica dispersion via the Sellmeier equation. The red curve shows only the
geometric dispersion.
Idler
Frequency (kHz) 200
0
Difference
–200 Signal
Figure 11.7
Temporal variation of the signal and idler sidebands and their frequency difference measured on two different
Beat Detection Units (5 s gate time), while the pump frequency is locked on the fiber reference comb. The two
comb lines remain equidistant, but drift in time.
comb spectrum can be controlled in terms of its two degrees of freedom (mode spacing
and frequency offset). In the following sections it is shown that monolithic microresona-
tor based frequency combs fulfill both of these requirements, allowing full control of the
optical mode spectrum and generation of mode spacings in the microwave frequency
range. We explain how to gain control on the frequency comb spectrum generated and
describe the principle of the stabilization, its implementation and discuss the locking
properties.
11.3.1 Principle
An optical comb spectrum can be entirely described by two parameters, namely the
absolute frequency of one arbitrary mode and its mode spacing. In the case of a micro-
cavity comb the pump frequency νpump is directly part of the generated spectrum, since
the equidistance was checked when using both the pump beam or the corresponding
comb line (i.e. before and after the microcavity). Therefore the generated spectrum can
be described as: νm = νpump + m ƒrep m∈Z, consequently, full stabilization of the optical
f requency comb requires both control of the pump frequency and the comb’s mode
spacing. The latter can be achieved by identifying the pump power and frequency as
two actuators.
In standard femtosecond comb generators, the control of the comb’s repetition rate can
be accomplished by tuning the cavity length, in general via a piezoelectric element sup-
porting one cavity mirror. This control of the cavity length can obviously not be imple-
mented in the present case since there is no movable part in the monolithic microcavity. It
is however possible to control the cavity optical path length by exploiting the temperature
dependent refractive index of glass: changing the intracavity intensity modifies the cavity
temperature through photon absorption, which in turn affects the refractive index. In the
following section it is described how this locking technique is implemented relying on
the thermorefractive effect. Note that the different additional nonlinear mechanisms that
define the actual mode spacing of the Kerr comb (e.g. self/cross phase modulation [30])
also contribute to its intensity dependence.
11.3.2 Implementation
The setup used in this stabilization experiment (cf. Figure 11.8) is similar to the one
described in the previous section, except that the second beat note is now used to derive
an error signal, fed back in order to control the pump intensity. The latter is achieved with
a tunable gain erbium doped fiber amplifier (EDFA).
In such a configuration, the signal used to control the repetition rate is not directly pro-
portional to the comb mode spacing since it is also affected by the pump frequency drifts.
However since the latter is independently locked on an optical reference (one mode of the
fully-stabilized fiber comb), one can perfectly use the second beat for testing the viability
of the locking principle. In the following, it is shown that it is possible to directly measure
the comb repetition rate and use it for locking.
To create an error signal suitable for locking the mode spacing, the beat generated on
BDU2 (see Figure 11.8) is compared to a reference signal with a phase detector whose
output serves as an error signal. It is then amplified, filtered and injected on the cur-
rent input of the 980-nm diode laser pumping the EDFA, without affecting the 1550-nm
pump frequency, as a direct feedback on the pump laser current would have induced.
Once locked, one can tune the mode spacing by simply modifying the frequency of the
reference signal.
Figure 11.8b represents the time dependence of the beat between the Kerr comb’s 7th side-
band with the respective reference comb line, counted with a 1-s gate time and expressed
in equivalent mode spacing, by dividing the measured fluctuations by a factor of 7. During
the first 65 min, the lock is disabled and one thus simply measures the free drift of the
mode spacing. It exhibits fluctuations on a 5-s timescale (see inset) which is attributed to
(a) (b)
Optical
Laser
amplifier
Equivalent mode spacing drift (kHz)
Beat 1 Spacing
control 40
Frequency TOF
control Beat 2
40
Spacing (kHz)
Figure 11.8
(a) Experimental setup used to stabilize the Kerr comb. The mode spacing is inferred from the beat of the 7th
sideband with the reference comb and is controlled via an intensity regulation implemented with the optical
amplifier TOF: tunable optical filter. (b) The fluctuations of the second beat note frequency, counting the beat
between the 7th Kerr sideband and an adjacent reference comb mode, can be efficiently reduced by the lock,
demonstrating the capability to gain control on the comb spectrum. They are expressed in equivalent mode
spacing fluctuations by dividing by a factor 7, the pump frequency remaining locked to one line of the refer-
ence comb. Out-of-loop measurements on the 5th sideband confirmed the stabilization effect. (From Del’Haye,
P., Arcizet, O., Schliesser, A., Holzwarth, R., and Kippenberg, T.J., Full Stabilization of a microresonator-based
optical frequency comb. Phys. Rev. Lett., 101, 053903, 2008.)
a change in the pump intensity and coupling strength and longer time drifts due to envi-
ronment temperature evolution. After 65 min, the lock is activated, and one can observe a
net suppression of the mode spacing fluctuations, without affecting the pump frequency
lock. This behavior can also be observed on an independent out-of-loop signal originating
from the beat of the 5th sideband of the Kerr comb with the reference comb. This shows the
potential of the technique described to gain control over the entire the Kerr comb. Note
that during the experiments, the microcavity remains locked to the pump frequency, in
particular because of the so called thermal “self-locking mechanism” [45] which will be
explained later in the chapter.
Optical
(a) (c) Laser
amplifier
0
variation (MHz)
Mode spacing
Modulation
–.2
Correction
–.4
Spacing
78 79 80 Error
control
Pump power (mW) Beat 2
VCO
(b)
0 (d)
Resonance shift
–30
Power (dBm)
(GHz)
–1
–90
70 74 78 82 10 1000
Pump power (mW) Frequency (Hz)
Figure 11.9
(a) Static mode spacing dependence on the pump power. (b) Frequency of the thermally induced bistable turn-
ing point for different injected intensities, when scanning the laser frequency sufficiently slowly for letting it
thermalize. (c) Setup used for characterizing the lock properties: an external perturbation can be applied in
order to measure how the system reacts. (d) Frequency response of the correction signal to an external per-
turbation: the actuation bandwidth is of the order of 10 kHz. Similar characterizations have been performed
on microwave repetition rate comb generators (From Del’Haye, P., Arcizet, O., Schliesser, A., Holzwarth, R.,
and Kippenberg, T.J., Full Stabilization of a microresonator-based optical frequency comb. Phys. Rev. Lett., 101,
053903, 2008.)
experimentally measured that a 1 GHz change in the pump frequency linearly affects the
mode spacing by only 1 MHz.
Next, the origin of the intensity dependence of the mode spacing was investigated. From
the microtoroid’s high quality factors and small mode volumes, several nonlinear mecha-
nisms can contribute to this effect by changing its geometric and refractive properties. It
has been shown that the cavity static and low frequency response (below 10 kHz) to an
intensity change is governed by the absorption-induced heating and concomitant refrac-
tive index change and thermal expansion, which are dominant compared to the Kerr or
radiation pressure contributions [46]. The corresponding variation of the microcavity’s
repetition rate due to a temperature change δT can be written: δ ƒrep/ƒrep = (αl + αn) δT. The
comparison of αn = 1.0 × 10−5 K−1, the thermorefractive coefficient of silica, to αl = 5.5 × 10−7 K−1
its thermal expansion coefficient, allows to conclude that the main contribution arises from
the temperature dependence of the refractive index. As a consequence, a 0.1 K change in
the toroid temperature induces a 400 kHz change of the repetition rate.
The same nonlinear mechanism is also responsible for the optical bistability mentioned
above, observed when scanning the microcavity resonance with the laser. The frequency
of the bistability turning point depends on the injected power, its dependence on the pump
power is represented in Figure 11.9b. It exhibits a characteristic linear behavior with a
slope of Γ = − 93 MHz/mW. This measurement was taken by scanning the laser frequency
sufficiently slowly such that thermal equilibrium of the toroidal cavity was reached for
each detuning value. It is possible to estimate the toroid thermalization time with the char-
acteristic diffusion time of a thermal wave from the center of the optical mode to the toroid
† Note however that the specific geometry of the toroid induces a more complex thermal response than this
simple estimation.
0 1
20 dB
0 2
0 3
Transmitted power (dBm)
0 5
0 6
0 7
0 8
0 Pump 10
1561 nm
Figure 11.10
Spectra obtained with the same microcavity, for different input powers ranging from 400 µW to 70 mW. For
increasing pump power it is possible to observe higher order comb repetition rate, corresponding to multiples
(up to a factor of 10) of the cavity FSR.
Ωc = D/δ2 ≈ 2π × 16 kHz, and is also comparable to the efficiency range of the locking electron-
ics that we used. Thus the above measurements have revealed that the locking technique
has the potential to reach high frequencies of actuation (up to 10 kHz), which is on par with
the actuation range of standard comb generators based on piezo-mounted mirrors.
5 mm
100 µm
Fiber transmission (arb)
0
–1000 –500 0 500 1000
Laser frequency (GHz)
Figure 11.11
Upper left: Photograph of a chip supporting several millimeter sized optical resonators; Upper right: Optical
microscope view of a 750-µm diameter toroid and the tapered fiber. Lower panel: Transmission spectrum of
660-µm diameter toroidal microcavity coupled to a tapered optical fiber, exhibiting a mode spacing of 86 GHz.
than 10 times lower than what has been observed in smaller toroidal microcavities (and
>100 times lower than in crystalline microresonators), it was sufficient to observe paramet-
ric oscillations at a threshold value of ca. 10 mW.
Using a 750-μm diameter sample it was possible to generate a frequency comb with a
86 GHz repetition rate, by pumping the toroid at 1550 nm with a 150-mW continuous-wave
laser. The typical optical spectrum measured after the cavity is represented in Figure 11.12a.
The comb spans over 60 nm which is lower than what was demonstrated with smaller cavi-
ties [29]. This is a consequence of the increased dimensions of the toroid, which imply a
larger minimum threshold for parametric oscillations [30]:
(1 + K )3 π 2 neff
2
RAeff
min
Pthres = Q0-2 , (11.4)
K 2 λ 0n2
The expression presents a linear dependence in the effective mode volume 2πRAeff
(K = Q 0/Qex is the coupling parameter, Q 0 and Qex the intrinsic and external (related to
the coupling) optical quality factor, and n 2 the intensity dependent refractive index).
For the 750-μm cavities, the diameter is responsible for a ten-fold increase in thresh-
old, and the effective mode area is also expected to increase from the higher inner
toroid diameter. Those effects directly increase the threshold and similarly affect the
strength of the induced nonlinearities. Note that there is no fundamental limit which
prevents higher Q factors. Indeed, Q > 109 have been measured in millimeter sized
microspheres [48].
(a) (b) 0
Pump laser 85.65762
10
Relative beat strength (dB)
GHz
100 kHz
Optical power (dBm)
0 83 GHz
–10
–20
–30
–80
–40
1540 1560 1580 1600 85.6574 85.6576 85.6578
Wavelength λ (nm) Frequency (GHz)
Figure 11.12
(a) Typical optical spectrum measured after the millitoroid. The comb exhibits a 83 GHz comb spacing whose
intensity beat can be directly measured using a fast photodiode. (b) Typical locked microwave beat (measured
by downconverting the signal to 30 MHz using a harmonic mixer). The width of the coherent spike is deter-
mined by the resolution bandwidth of the measurement (10 Hz). (From Del’Haye, P., Arcizet, O., Schliesser, A.,
Holzwarth, R., and Kippenberg, T. J., Full Stabilization of a microresonator-based optical frequency comb. Phys.
Rev. Lett., 101, 053903, 2008.
Allan deviation
photodiode –40
Optical 10–10 –60
Laser
amplifier 1550 nm Stabil –80
ized m
Kerr comb ode s
pacin85.6574 85.6576 85.6578
TOF g Frequency (GHz)
Figure 11.13
(a) Setup used for characterizing and stabilizing the microwave frequency comb. The use of a fast photodiode
combined with a harmonic mixer allows to directly measured the 86 GHz beat corresponding to the mode
spacing of the comb generated from our millimetric sized cavity. TOF: tunable optical filter. (b) Measured Allan
deviation of the comb repetition rate as a function of gate time for the free running comb and once locked. (From
Del’Haye, P., Arcizet, O., Schliesser, A., Holzwarth, R., and Kippenberg, T. J., Full Stabilization of a microresonator-
based optical frequency comb. Phys. Rev. Lett., 101, 053903, 2008.)
whose 6th harmonics falls in the vicinity of the 86 GHz signal to be measured in order to
down-convert the signal to countable frequencies, of the order of 30 MHz.
In order to use the beat signal for locking the comb, it was phase compared with a reference
signal and the error signal provided was similarly used as feedback to the optical amplifier
pump current. The setup is described in Figure 11.13. The technique allowed direct locking
of the repetition rate of the comb without relying on the fiber reference comb. The typical
locked 86 GHz signal is represented in Figure 11.12b, revealing the coherent spike and the ca.
10 kHz wide noise reduction accomplished by the feedback loop. This noise pedestal reflects
the cavity optical path fluctuations above the frequency range where the lock is compen-
sating. Many mechanisms contribute to this noise [42,49], in particular the laser’s classical
frequency noise, which is not distinguishable in this measurement, the laser’s shot noise, the
thermally and optically induced noise processes that affect the microcavity’s optical path
length through its refractive index, expansion and higher frequency mechanical vibrations.
Note that even if the opto-mechanical properties of the millitoroid have been intentionally
deteriorated, by maintaining a very small undercut and leaving it in air for providing vis-
cous damping that decreased its mechanical Q factor, the optomechanical instability [50,51]
i.e. radiation pressure driven regenerative mechanical oscillations, is however very easily
observed when the laser detuning is significant, and in general prevents from generating a
stable optical comb (for a laser strongly blue detuned). In addition, to limit this optomechan-
ical instability, the fiber was in general intentionally placed in contact to the microcavity for
suppressing these radiation-pressure-induced mechanical oscillations.
The long term stability of the microwave beat was investigated using an RF counter, the
Allan deviation measurements for the free running beat and the locked signal are shown
in Figure 11.13b. The influence of the lock is observable, dramatically increasing the long
term stability of the beat and yielding a repetition rate stability at the level of 10 mHz in
1 s, corresponding to an Allan deviation of 10 −12 for the in-loop measurement. This number
reflects the large stability achieved on the optical path length of the toroid. The out-of-
loop measurements implemented on the first idler sideband yielded rms fluctuations ∆ν1
below 1 Hz for a 1 hour measurement with a 1-s gate time [44]. The measured fluctuation
Oscilloscope 0
–10 0 10
Time delay (ps)
Figure 11.14
(a) Autocorrelator based on a periodically poled LiNbO3 non linear crystal. (b) Second order crosscorrelation
trace obtained with a 750 µm diameter microcavity.
expressed in an equivalent fluctuation of cavity optical path length, gives a relative sta-
bility of ∆Lopt = 2πR∆ν1/ν1 ≈ 10 am in 1 s. The demonstrated capacity to actively control
the resonator effective optical length, combined with their intrinsic stability, could allow
microcavities to become a future promising tool for time transfer. [43].
11.5 Conclusion
In summary, this chapter has introduced a radically different frequency comb generator
approach, unique in several ways. First, the approach offers unprecedented form factor,
allowing a monolithic approach to optical frequency comb generation. Due to the high
Q-factor in conjunction with tapered optical fiber coupling, the approach is moreover
highly efficient, with conversion efficiencies frequently observed above 80%. Second, the
approach does not rely on the use of atomic or molecular resonances, and therefore can
achieve broadband gain in a frequency window only limited by material absorption or dis-
persion. Owing to the large transparency window of glass, the frequency window extends
from the UV to 2 μm. Third, the use of monolithic microresonators allows to significantly
augment the repetition rate. While the highest repetition rates of conventional femtosec-
ond lasers so far are approaching 10 GHz, the equivalent cavity length of 3 cm is difficult
to reduce further. Moreover, high repetition rate combs have so far only been achieved in
Ti:Sa laser systems, whereas for 1.5 μm fiber based frequency comb repetition rates have
been generally much lower. The present approach can operate at previously unattainable
repetition rates exceeding 10 GHz, and with a high power per individual comb mode,
which is useful in applications where the access to individual comb modes is required,
such as optical waveform synthesis, high capacity telecommunications, spectrometer cali-
bration [52] or direct comb spectroscopy. Finally, several new prospects exist that bode
well for future work and for the future of this technology.
Since this work, the generation of frequency combs in this manner has also been
reported in CaF2 resonators [49]. This work allowed reducing the repetition rate to
10 GHz and moreover showed the promise of this technology to serve as low phase noise
microwave oscillator. These developments show that microresonator-based frequency
comb generation bear promise for the future, possibly providing a route to all chip scale
atomic clocks.
Acknowledgments
We thank T. W. Hänsch for discussions and suggestions. T. J. K. acknowledges support
via an Independent Max Planck Junior Research Group. This work was funded as part of
a Marie Curie Excellence Grant (RG-UHQ) and the DFG-funded Nanosystems Initiative
Munich (NIM). We thank J. Kotthaus for access to clean room facilities for sample
fabrication.
References
1. T. W. Hänsch. Passion for precision (Nobel lecture). ChemPhysChem, 7: 1170–1187, 2006.
2. S. T. Cundiff and J. Ye. Colloquium: Femtosecond optical frequency combs. Rev. Mod. Phys., 75:
325–342, 2003.
3. T. Udem, R. Holzwarth, and T. W. Hänsch. Optical frequency metrology. Nature, 416: 233–237,
2002.
4. R. J. Jones, K. D. Moll, M. J. Thorpe, and J. Ye. Phase-coherent frequency combs in the vacuum
ultraviolet via high-harmonic generation inside a femtosecond enhancement cavity. Phys. Rev.
Lett., 94: 193201, 2005.
5. C. Gohle, T. Udem, M. Herrmann, J. Rauschenberger, R. Holzwarth, H. A. Schuessler, F.
Krausz, and T. W. Hänsch. A frequency comb in the extreme ultraviolet. Nature, 436: 234–237,
2005.
6. U. Keller. Recent developments in compact ultrafast lasers. Nature, 424:831–838, 2003.
7. J. N. Eckstein, A. I. Ferguson, and T. W. Hänsch. High-resolution two-photon spectroscopy with
picosecond light pulses. Phys. Rev. Lett., 40: 847–850, 1978.
8. Th. Udem, J. Reichert, R. Holzwarth, and T. W. Hänsch. Accurate measurement of large optical
frequency differences with a mode-locked laser. Optics Lett., 24: 881–883, 1999.
9. Th. Udem, J. Reichert, R. Holzwarth, and T. W. Hänsch. Absolute optical frequency mea-
surement of the cesium D1 line with a mode-locked laser. Phys. Rev. Lett., 82: 3568–3571,
1999.
10. D. J. Jones, S. A. Diddams, J. K. Ranka, A. Stentz, R. S. Windeler, J. L. Hall, and S. T. Cundiff.
Carrier-envelope phase control of femtosecond mode-locked lasers and direct optical frequency
synthesis. Science, 288: 635–639, 2000.
11. R. Holzwarth, T. Udem, T. W. Hänsch, J. C. Knight, W. J. Wadsworth, and P. S. J. Russell. Optical
frequency synthesizer for precision spectroscopy. Phys. Rev. Lett., 85: 2264–2267, 2000.
12. J. Reichert, M. Niering, R. Holzwarth, M. Weitz, Th. Udem, and T. W. Hänsch. Phase coherent
vacuum-ultraviolet to radio frequency comparison with a mode-locked laser. Phys. Rev. Lett.,
84: 3232–3235, 2000.
13. S. A. Diddams, D. J. Jones, J. Ye, S. T. Cundiff, J. L. Hall, J. K. Ranka, R. S. Windeler, R. Holzwarth,
T. Udem, and T. W. Hänsch. Direct link between microwave and optical frequencies with a
300 THz femtosecond laser comb. Phys. Rev. Lett., 84: 5102–5105, 2000.
37. B. Min, L. Yang, and K. J. Vahala. Perturbative analytic theory of an ultrahigh-Q toroidal micro-
cavity. Physical Rev., A, 76: 013823, 2007.
38. S. Schiller. Asymptotic-expansion of morphological resonance frequencies in Mie scattering.
Appl. Optics, 32(12): 2181–2185, 1993.
39. Tobias Kippenberg. Nonlinear Optics in Ultra-high-Q Whispering-Gallery Optical Microcavities.
PhD thesis, California Institute of Technology, Pasadena, 2004.
40. I. H. Agha, Y. Okawachi, M. A. Foster, J. E. Sharping, and Gaeta A. L. Four-wave-mixing para-
metric oscillations in dispersion-compensated high-Q silica microspheres. Physical Rev., A, 76:
043837, 2007.
41. A. B. Matsko, A. A Savchenkov, N. Yu, and L. Maleki. Whispering-gallery-mode resonators as
frequency references. I. Fundamental limitations. J. Opt. Soc. Am. B, 24: 1324–1335, 2007.
42. A. A. Savchenkov, A. B. Matsko, V. S. Ilchenko, N. Yu, and L. Maleki. Whispering-gallery-mode
resonators as frequency references. II. Stabilization. J. Opt. Soc. Am. B, 12: 2988–2997, 2007.
43. P. Del’Haye, O. Arcizet, A. Schliesser, R. Holzwarth, and T. J. Kippenberg. Full stabilization of
a microresonator-based optical frequency comb. Phys. Rev. Lett., 101, 053903, 2008.
44. T. Carmon, L. Yang, and K. J. Vahala. Dynamical thermal behavior and thermal self-stability of
microcavities. Optics Express, 12: 4742–4750, 2004.
45. A. Schliesser, P. Del’Haye, N. Nooshi, K. J. Vahala, and T. J. Kippenberg. Radiation pressure
cooling of a micromechanical oscillator using dynamical backaction. Phys. Rev. Lett., 97: 243905,
2006.
46. A. Bartels, C. W. Oates, L. Hollberg, and S. A. Diddams. Stabilization of femtosecond laser fre-
quency combs with subhertz residual linewidths. Optics Lett., 29: 1081, 2004.
47. D.W. Vernooy, V.S. Ilchenko, H. Mabuchi, E.W. Streed, and H.J. Kimble. High-Q measurements
of fused-silica microspheres in the near infrared. Optics Lett., 23: 247–249, 1998.
48. A. A. Savchenkov, A. B. Matsko, V. S. Ilchenko, I. Solomatine, D. Seidel, and Lute Maleki.
Tunable optical frequency comb with a crystalline whispering gallery mode resonator. Phys.
Rev. Lett. 101: 093902, 2008.
49. T. J. Kippenberg, H. Rokhsari, T. Carmon, A. Scherer, and K. J. Vahala. Analysis of radiation-
pressure induced mechanical oscillation of an optical microcavity. Phys. Rev. Lett., 95: 033901,
2005.
50. O. Arcizet, P. F. Cohadon, T. Briant, M. Pinard, and A. Heidmann. Radiation-pressure cooling
and optomechanical instability of a micromirror. Nature, 444: 71–74, 2006.
51. M.T. Murphy, T. Udem, R. Holzwarth, A. Sizmann, L. Pasquini, C. Araujo-Hauck, H. Dekker,
S. Odorico, M. Fischer, T. W. Hänsch, and A. Manescau. High-precision wavelength calibration
with laser frequency combs. Mon. Not. Roy. Astron. Soc., 380: 839–847, 2007.
Jacob Khurgin
Johns Hopkins University
Contents
12.1 Introduction....................................................................................................................... 507
12.2 Single and Coupled Microresonators as Optical Delay Elements.............................508
12.3 Single and Coupled Microresonators as Optical Switches......................................... 513
12.4 Loss and GDD Limitations in CRS Delay Lines........................................................... 518
12.5 Mitigation of GDD............................................................................................................ 524
12.6 Conclusions........................................................................................................................ 526
References...................................................................................................................................... 527
12.1 Introduction
Recent years have seen spectacular progress in development of linear and nonlinear opti-
cal devices based on microresonators. As one can observe from reading other chapters in
this volume, high Q microresonators made from various materials have been successfully
fabricated and used for such diverse applications as delay lines [1], microwave photonics
[2,3], sensors [4,5] and others. Especially impressive have been recent advances in active
microresonators made from electro-optic or nonlinear materials. Active microresonators
are expected to be employed as miniature optical modulators [6], switches [7], tunable
filters [8] and delay lines [9], that are amenable to large scale integration culminating one
day in development of photonic integrated circuits.
The main advantage of optical methods of information transmission is, of course, their
large capacities that can reach tens of terabits per second (TBps) in a single optical fiber.
Processing of information at such speeds electronically is difficult to fathom not only with
today’s semiconductor-based electronics, but even in the foreseeable future. This discrep-
ancy between optical transmission and electronic processing results in traffic jams on the
information superhighway occurring wherever the information carried by photons needs
to be switched and routed electronically. Therefore, optical methods of processing informa-
tion, at least at some rudimentary level, are desperately needed to relieve this congestion.
So far, unfortunately, optical devices performance has lagged behind that of their electronic
counterparts. This disadvantage is rooted in the most basic features distinguishing photons
from electrons: photons move with high speed, carry no charge, and are not subject to the
Pauli principle, which makes all interactions involving photons relatively weak compared
507
© 2009 by Taylor & Francis Group, LLC
to electronic interactions. Thus photons must propagate over long distances, measured in
millimeters, to carry out the same tasks that can be accomplished electronically over a few
micrometers. Clearly, until effective means of confining photons into tighter spaces are
found and perfected, photonic integration is simply not going to happen. Microresonator
(MR) technology is ideally suited for effective photon confinement is small volumes and
that is why development of microresonators is pursued with such dynamism.
As with any promising technology, MR technology is not without limitations. As the
term “microresonator” itself suggests, the ability of photon confinement rests upon the
existence of a strong resonance in its characteristics, with all the desirable properties con-
fined to a relatively narrow band near this resonance. This “gain-bandwidth product”
limitation affects the performance of the MR-based devices. Understanding of bandwidth
limitations of MRs is important because it allows one to decide on applicability of MR
technology for a given application with subsequent optimization of the MR design for a
particular task. In this chapter we consider the bandwidth limitations of MRs in potential
applications as optical delays and optical switches. We compare single MRs with multiple
coupled ones and show that, while multiple MRs are the best choice for delay lines, in the
optical switches the choice should depend on both bit rate and the strength of index modu-
lation. Then we study delay lines based on multiple coupled MRs and show that their per-
formance at large bandwidths is equally affected by two factors—dispersion of loss and
dispersion of group delay. Finally we describe some ideas for the dispersion mitigation.
Eout (ω ) ρ - e jωτ
= ≡ e jΦ ( ω ) (12.1)
Ein (ω ) 1 - ρe jωτ
where ρ2 = 1-κ2 and τ = Ln/c is a ring round trip time. As the name “all-pass” implies, the
amplitude transmission is unity for all frequencies, and only the phase is affected as
k 2 sin ( ω - ω 0 ) τ
tan Φ(ω ) = (12.2)
(1 + ρ2 )cos (ω - ω 0 )τ - 2ρ
where ω0 = 2πmτ−1 is a resonance frequency. The group delay of the APF is
∂ Φ(ω )
Td (ω ) = = Td( 0) + Td( 2 ) ( ω - ω 0 )2 + … (12.3)
∂ω
(a) (b)
out
jκ
jκ jκ
in ρ out
in ρ through
Figure 12.1
(a) An all-pass filter (APF) based on a single ring microresonator. (b) An add-drop filter (ADF) based on a single
ring microresonator.
(a) (b)
3 12
2 10
1 8
Phase (rad)
Delay (ps)
0 6
–1 4
–2 2
–3 0
–150 –100 –50 0 50 100 150 –150 –100 –50 0 50 100 150
Frequency (GHz) Frequency (GHz)
Figure 12.2
(a) Phase characteristics of the APF of Figure 12.1a. (b) Group delay in the APF of Figure 12.1a
1+ρ
Td( 0) = τ (12.4)
1-ρ
( 1 + ρ)ρ ρ
3 = - [Td ]
( 0) 3
Td( 2 ) = - τ 3 (12.5)
(1 - ρ) ( 1 + ρ )2
In Figure 12.2a and b the phase and group delay characteristics of an APF with L = 36 µm, n = 2.1
L = 36 µm, n = 2.1 (τ = 250 fs) and κ = 0.3 [1] are shown. As one can see, large group delays (exceeding τ
by a factor of 40) can be achieved in a relatively narrow bandwidth near ν0.
0.8
Transmission
0.6
∆νpass
0.4
0.2
0
–150 –100 –50 0 50 100 150
Frequency (GHz)
Figure 12.3
Transmission characteristics of the ADF of Figure 12.1b.
Eout (ω ) k2
= (12.6)
Ein (ω ) 1 - ρ2 e jωτ
has a Fabry–Perot like spectrum of intensity transmission shown in Figure 12.3 and can be
characterized by a full width at half maximum (FWHM)
The delay time of the signal, i.e. the time spent inside the resonator is
Td =
∂Φ
∂ω (
= τ k -2 -
1
2
) (12.8)
and we obtain a simple relationship between delay and width of the pass-band
Td∆ωpass = 2 - κ2 (12.9)
Irrespective of the application or MR, the longer the light spends inside the MR, the
stronger is the effect one can impose on the light. Therefore, Equation 12.9 would lead to
a gain-bandwidth product for any application. In this chapter we shall limit ourselves to
digital signals; thus we shall consider bit rate limitations, but the implications for band-
width limitations of analog signals are obvious.
We consider an on-off-keyed signal with a Gaussian profile whose FWHM is equal to
one half of the bit interval B. The temporal profile of such a signal is P(t) = exp(-16ln(2)B2t2),
while the power spectrum of signal is also Gaussian with FWHM
∆ωsig = 8ln(2)B (12.10)
Since one needs to maintain ∆ωsig ≤ ∆ωpass then, according to Equation 12.9, the delay-bit
rate product, or the number of stored bits in a single resonator is
2 - k2
N = Td B ≤ ≈ 0.36 (12.11)
8 ln(2)
Although the APF is transparent at any frequency, one can still talk about the useful
bandwidth of the APF defining is as the bandwidth at which the group delay changes by
a factor of two
2
Td ∆ω pass = (1 + ρ) (12.12)
ρ
For any resonator with a reasonably high Q, ρ approaches unity and a relation almost
identical to Equation 12.11 can be obtained, indicating that one cannot possibly store even
one bit of information in a single resonator without too much distortion. For this reason
one should consider combining the resonators. Just as with single resonators two differ-
ent arrangements can be made—the all pass arrangement, or the side-coupled integrated
spaced sequence of resonators (SCISSOR) [12,13] shown in Figure 12.4a. The dispersion
relation of SCISSORS is the same as Equation 12.3, but the relation between the group
delay and GDD is changed from Equations 12.4 and 12.5 and can be written as
1+ρ τ
Td(,0S) = N r τ = Nr (12.13)
1-ρ k1
and
Td(,2S) = - N r τ 3
(1 + ρ)ρ = - N -2 T (0) 3 1 - k 12 ≈ - 1 N -2 T (0) 3
r d ,S r d ,S (12.14)
( 1 - ρ )3 4 4
1-ρ
k1 = (12.15)
1+ρ
Out
In
In Out
Figure 12.4
Coupled resonator structures (CRS). (a) SCISSOR. (b) CROW.
(a) (b)
ω
4
∆ωpass
3
π 1
Φ 0
–100 –50 0 50 100
Frequency (GHz)
Figure 12.5
(a) Dispersion and (b) group delay (per resonator) of a CROW.
The dispersion curve is shown in Figure 12.5a and consists of the series of pass bands
around resonant frequencies ωm = 2πm/τ separated by the wide gaps. The width of the pass
band is
∂Φ 1 ∂3Φ
Φ = Φ0 + (ω - ω 0 ) + ( ω - ω 0 )3 + ... =
∂ω 6 ∂ω 3
(12.17)
1
= Φ 0 + Td(,0C) ( ω - ω 0 ) + Td(,2C) ( ω - ω 0 )3
3
∂Φ τ
Td(,0C) = = Nr (12.18)
∂ω 2k
( ) (1 - k ) ≈ 21 N
3
1 ∂3Φ 1 τ -2 3
Td(,2C) = = Nr 2
r Td(,0C) (12.19)
2 ∂ω 3 2 2k
The group delay per one ring in a CROW with the same τ = 250 fs [17] is plotted in
Figure 12.5b. The relation (Equation 12.19) is almost identical (except the factor of 2) in
magnitude but opposite in sign to the SCISSOR relation (Equation 12.14). We shall return
[16] to the difference in sign later on, but for the time being we shall concentrate on the
salient feature of the coupled resonators, the fact that the pass band in either CROW or
SCISSOR is independent of the number of resonators. The delay time Td(0) can be increased
without affecting the width of pass band, and the delay-bit rate product, and the delay-
bandwidth product can be increased compared to a single resonator. One should remem-
ber, however, that the GDD (Equation 12.14 or 12.19) will still limit this product.
There can be different ways to gauge the GDD limitation—the one we adapt here is
rather simple. We want the differential group delay within the FWHM (Equation 12.10) of
the signal spectrum to be less than one half of the bit interval B −1 to avoid excessive inter
symbol interference (ISI). Thus the GDD limitation is
3
2 [ 4 ln(2)]2 Td(,2C)B3 ≈ B3 [ 4 ln(2)]2 N r-2 Td(,0C) ≤ 1 (12.20)
independent of the bit rate and the delay per ring [17]. Thus, to store N bits of information
in CROW without incurring too much distortion that can cause intersymbol interference,
one should use at least Nr,C ≈ 2.8N3/2, while for SCISSOR it is at least Nr,S ≈ 2.2N3/2. The dif-
ference is insignificant so it makes sense to use a “rule of thumb” relation
Nr ≈ 3N3/2 (12.22)
Using this relation one can immediately get the required delay per resonator for a given
bit rate as
1
Td( 0) = N/BN r ≈ (12.23)
3 N 1/2 B
From this relation, with a given resonator dimensions, one can determine the coupling
coefficients using Equation 12.4 or 12.18 for SCISSOR and CROW, respectively. Thus for
the linear task of storage it is always preferable to use multiple MR structures rather than
single MR’s. The analysis performed so far has not included the influence of loss incurred
in the MR’s. We shall return to the issue of loss in the delay line later on, but next we turn
our attention to optical switching and how it can be improved with MR’s.
L0 ∆φ = π
∆n
Figure 12.6
Optical switch based on MZI.
contain material whose refractive index can be varied by current injection, as in Si devices,
by a low frequency electric field (electro-optic effect), or by an optical field (nonlinear opti-
cal effect). With regard to the all-optical switches, we shall consider here only the case
when the switching (pump) optical field is not resonant with the MR, while the signal is
resonant. The double-resonant case, which is more difficult to implement in practice, is
treated in Refs. [18,19]. No matter what mechanism causes the refractive index change ∆n,
the effect can be characterized by the maximum attainable index change ∆nmax and by the
speed characterized by the cut-off frequency Fmax. From a more fundamental point of view
the index change is limited by the material damage resulting from excessive electric or
optical fields or from an extremely high current injection level. From more practical con-
siderations the index change can be limited by voltage, current or optical power levels at
the disposal of the designer. Speed limitations can be intrinsic (recombination time in cur-
rent injection schemes) or extrinsic (parasitic capacitances and losses at high frequencies
in electro-optic devices). The cut-off frequencies range from the GHz range in Si current
injection devices [7,20] to tens of THz in nonlinear optical devices based on the Kerr effect
[21]. But as a rule, the faster the speed of the effect, the weaker it is, because the slow index
change mechanisms usually rely upon the accumulation of carriers (or temperature rise)
while the faster mechanisms have to depend on the instantaneous values of the input. As
a result, the product of ∆nmax Fmax (by using analogy from electronics, one can call it “gain
bandwidth product”) does not vary much. For instance in a Si-based scheme with carrier
injection, the refractive index change [7] ∆nmax ~10−3 and Fmax ~1GHz, while in the LiNbO3
the bandwidth can be as high as Fmax ~100GHz, but the index change is much smaller, ∆nmax
~10−5 [22]. In a typical ultra-fast nonlinear optical switch the index changes also do not
exceed 10−5 [21] but the bandwidth is measured in THz.
To attain switching in the MZI of Figure 12.6a, an optical path length change equal to
one half wavelength must be induced in one of the arms and this gives us a switching
condition that the required product of index change and length be
(∆nL)MZl = λ0/2 (12.24)
If one considers other nonresonant switching schemes, for instance the ones relying
upon directional couplers or Bragg gratings, they all have constraints on the length-index
change product that are very similar to Equation 12.24.
Since the index change is limited, one can introduce the “minimum switching distance”
in the MZI interferometer as
L0 = λ0∆nmax/2 (12.25)
Now the “gain” of the resonant scheme is its ability to reduce ∆nL, thus reducing the size
and power consumption of the device. Let us see how it can be done and what the bit rate
limitations are.
When the effective index of refraction is changed via either an electro-optic or a nonlin-
ear optical process
n = n + ∆n (12.26)
∆n
∆τ = τ (12.27)
n
∆n
∆ω 0 = -ω 0 (12.28)
n
as shown in Figure 12.7a. Consider now a single resonator of Figure 12.1b. If the pass band
width ∆ωpass of Equation 12.7 is equal to the signal bandwidth ∆ωsig, then shifting the res-
onance frequency by ∆ω0 = ∆ωpass = ∆ωsig will cause switching in this resonance-shifting
scheme, and combining Equation 12.28 with Equation 12.10 one obtains the switching
condition.
c c 8 ln(2) λ 0 λ
( ∆nL )r = ∆nτ = 8 ln(2)Bτ = Bτ ≈ 1 . 7 Bτ 0 (12.29)
n ω0 π 2 2
(a)
1
0.8 (b) L
Transmission
Δφ = π
0.6 Δωpass Δn
Signal
0.4 Δωsig
0.2
0
ω0 – Δω0 ω0 ω (arb. units)
Figure 12.7
(a) Optical switch based on resonance shift in a single MR. (b) Optical switch based on coupled multiple MR’s
inside an MZI slow light switch.
It would appear that one can increase the gain using smaller and smaller radii, but,
since maximum index change is limited, one can also define the maximum switching bit
rate as
Bmax, r =
ω 0 ∆n
8 ln(2) n ( ) max
= 1.1ν0 ( ∆nn ) max
(12.32)
The resonant gain at maximum bit rate in the resonance shifting single resonator scheme
is then
-1
Bcut ,r ∆n L
Gr (Bmax, r ) = ≈ 0.65 ν0τ max ≈ 0 (12.33)
Bmax ,r n L
Since L0 even for the largest value of ∆nmax ~10−3 is in the mm range, and the rings can
have circumference of less than 100 µm, the maximum bit rate is always at least an order of
magnitude smaller than the cut-off bit rate. In Figure 12.8 we have plotted as dashed lines
the resonant gain as a function of bandwidth for two different schemes—one is a narrow-
band Si current injection scheme with ∆nmax ~10−3 (L0 ~0.7 mm @1550 nm) [7] and Fmax ~10
GHz and the other is a GaAs photonic switch based on the optical Kerr effect [21] with ∆nmax
~3 × 10−5 (L0 ~2.5 cm @1550 nm) and Fmax > 10 THz. We assumed that the ring resonators
have about 36 µm circumference or τ ~ 250 fs. For the Si scheme of Figure 12.8a we obtain
Bmax,r = 110 GBps which is in fact much faster than Fmax; hence the speed of Si is intrinsically
limited to about 10 GBps (this is shown by thick line). But at this fairly low bit rate one can
achieve a tremendous gain of about 300 compared to the MZI. In the GaAs photonic switch
(Figure 12.8b), on the other hand, Bmax,r ~3 GHz which is orders of magnitude less than
its intrinsic potential of almost unlimited bandwidth. Therefore one should consider the
(a) (b)
105 105
104 104
Resonant gain
Resonant gain
102 102 Nr = 10
Nr = 10 Nr = 100
101 101
Nr = 100 Nr = 1000
Fmax Nr=1000
100 100
108 109 1010 1011 1012 108 109 1010 1011 1012
Bit rate (Bps) Bit rate (Bps)
Figure 12.8
Resonant gain (decrease in ∆nL relative to MZI) for two different optical switches (a) Si switch with current
injection. (b) GaAs all optical switch.
means to improve the speed of these devices using coupled MR’s. It is obvious that combin-
ing MR’s in CROW or SCISSORS and still using the resonance shifting scheme will simply
use more rings to achieve the same shift of ω0 and thus will only make things worse. For
this reason we shall consider a different scheme that takes advantage of longer propagation
times in the coupled MR’s while still relying on MZI geometry to achieve switching. We
refer to this scheme shown in Figure 12.7b as the “slow light scheme” [18,19].
With the phase characteristic of CROW or SCISSOR described by Equation 12.17 the
phase change caused by the change of index is
∂Φ ( 0) ∆n
∆Φ = - ∆ω 0 = TdC ω0 =π (12.34)
∂ω n
must be still enforced to avoid excessive ISI. Then we quickly obtain the required index
change
∆n B
= N r-2/3 [ 4 ln(2)]2/3 (12.36)
n 2 ν0
c cB λ
( ∆nL )Nr = N r ∆nτ r = N r1/3 [ 4 ln(2)]2/3 τ r ≈ 2 N r1/3Bτ r 0 (12.37)
n 2 ν0 2
and the gain is actually reduced by a factor of Nr1/3 compared to a single resonator reso-
nance shifting scheme (Equation 12.30).
( Nr )
Bcut ≈ 0.5τ r-1N r-1/3 ≈ Bcut ,r N r-1/3 (12.39)
But the main consequence is the fact that the maximum bit rate increases according to
Equation 12.36.
( Nr )
Bmax ≈ N r2/3 ν0 ( ∆nn ) max
≈ N r2/3Bmax,r (12.40)
which allows us to fully use the potential of fast-switching materials. We have plotted
the gain versus bit rate in the slow light scheme as solid lines in Figure 12.8 for different
numbers of resonators. As one can see from Figure 12.8a for the Si-based scheme where the
index change is large and intrinsically slow, using coupled resonators offers no advantage
relative to a simple resonance-shifting scheme with a single resonator. The increase in the
number of resonators causes a drop in the resonant gain and thus raises required switch-
ing power. This increased power demand comes with no switching speed benefit because
the switching speed is determined by the intrinsic speed of the device. But for the rela-
tively weak and fast scheme, such as the GaAs nonlinear all-optical switch of Figure 12.8b,
the coupled MR scheme (in either SCISSOR or CROW configuration) greatly enhances the
speed of the device from GBps for a single MR to hundreds of GBps, and still reduces the
power requirements compared to a simple MZI by a factor of up to 100 or more. Therefore,
the decision of whether to use a single (or a few) MR scheme with resonance shift or a mul-
tiple coupled MR slow light scheme should be based on the relation between the intrinsic
limitations of the material and required operational speed.
∆Td = Td(ω) − Td(ω0) (12.42)
Now the signs of ∆Td are different for CROW and SCISSOR designs and the loss implica-
tions are different. For CROW
(2)
∆TdC ≈ TdC (ω - ω 0 )2 > 0 (12.43)
indicating that the high frequencies in the signal spectrum are attenuated and the signal
broadens in time causing ISI. Substituting Equation 12.43 into Equation 12.41 one obtains
the expression for the output signal power spectrum
4 ln 2(ω - ω 0 )2 16 ln(2)acB2 ( 2 )
Pout (ω ) ~ exp - 1 + TdC (12.44)
( ∆ω sig ) 2
n
-1/2
16 ln(2)acB2 ( 2 )
∆ω sig,out = ∆ω sig 1 + TdC (12.45)
n
and the temporal FWHM is wider by the same factor. Limiting the temporal broadening to
a factor of 21/2 we obtain a loss-related bit-rate limitation
16 ln(2)acn -1B2TdC
(2)
≤1 (12.46)
Now, this condition must be added to the GDD condition (Equation 12.20)
Consider the Gaussian pulse with FWHM of 50 ps (Figure 12.9a) propagating through
the CROW delay line with Td(,2C) = 3 × 10 4 ps 3 and loss acn -1 = 0.01 ps -1. In Figure 12.9b we
show the impact of GDD only on the shape of output pulse. The pulse is somewhat broad-
ened and there are few decaying oscillations following the pulse which come from the
higher frequency components that had longer delay times. In Figure 12.9c we also include
the loss into the consideration, and the impact of the loss is twofold. The signal broadens as
a result of spectral narrowing, and at the same time the oscillations following the pulse get
dampened, because high frequency components get attenuated more. Overall one com-
bines two conditions (Equations 12.46 and 12.47) into a single one by insisting that
2
Bloss
2 [ 4 ln(2)]2 Td(,2C) B3 1 + ≤1 (12.48)
B2
where we have introduced the “loss bit rate” as Bloss = ac/2 n ln(2) the bit rate at which the
signal will experience 3 dB loss per 1/4 of the bit interval.
We can re-normalize it to the loss per ring a R = acτ/n and obtain
0.8
0.4 b
0.2
a
0
–200 –100 0 100 200
Time (ps)
Figure 12.9
Evolution of the Gaussian signal through the CROW delay line. (a) Input pulse. (b) Output pulse disregarding
loss. (c) Output pulse with loss.
Clearly at bit rates less than Bloss the loss distortion dominates while at higher rates GVD
becomes the overriding factor. Using Equations 12.18 and 12.19 we express the GDD as
Td(,2C) =
1 -2 ( 0) 3
2
1
2
N2 τ 2
N r [Td ,C ] ( 1 - k 2 ) = N 3B-3 N r-2 1 - r2 B
N 2
( ) (12.50)
2
Bloss
1+
[ 4 ln(2)]2 N 3 N r-2 B2 ≤ 1 (12.51)
2
B B2
1 + (N ) 1 + loss
Bcut B2
(N )
Bcut = ( 2 ln 2 τ )-1 N -1/2 = a -R1Bloss (12.52)
which can be understood as the bit rate at which the CRS offers only a 2 length advan-
tage over a simple straight waveguide. Thus in lieu of Equation 12.22 we obtain a more
refined important relation for the required number of resonators which includes bit rate
dependence.
2
4
1 + Bloss /B2
N r (B, N ) ≈ 2.8 N 3/2 , (12.53)
1 + [ B/Bcut ]
(N ) 2 2
1 + Bloss/B2
The dependence (Equation 12.53) is plotted in Figure 12.10 for four different values
of N using an example of coupled ring resonators from resonators having about 36 µm
104
Bloss
103 N = 250
Required number
of resonators Nr
N = 50
102 S
N = 10
101 (N)
Bmin
N=2
(N)
Bcut
100
109 1010 1011 1012 1013
Bit rate (Bps)
Figure 12.10
Number of resonators required to store N bits in a CROW delay line as a function of bit rate.
circumference or τ ~ 250 fs and a 0.04 dB (αR = 0.01 that yields Bloss~25 GBps) [1,26]. This
value indicates that with the present day technology both GDD and loss distortion play
roughly equal roles for 10–40 GBps signals, while at higher rates GDD is expected to domi-
nate. Clearly the plot can be split into three distinct regions
In the first region, the loss-induced distortion dominates and the required number of
resonators is inversely proportional to the square root of the bit rate. In the second region
the required number of resonators is determined by dispersion and remarkably does not
depend on the bit rate, Finally, at high bit rates the coupling coefficient κ becomes so close
to unity that the CROW delay line becomes essentially equivalent to a simple meandering
waveguide consisting of Nr half-circles and having a total length of cn -1NB-1 .
It should be noted that the condition for the existence of the second, GVD-limited, region
(N )
is, Bloss << B << Bcut or αR << N1/2; hence for the present structure the region should exist for
delay lines with a few hundred bits of storage capacity, albeit at very high bit rates. As one
can see, the “flat” region is clearly present in Figure 12.10 for N = 2 bits and N = 10 bits, but
it is far less prominent for N = 50 bits and barely visible at N = 250 bits.
We can now determine he slow down factor S = κ −1 in different regions. Note that the
slow down factor is the distance between the actual curve Nr(N,B) and the dashed line 2N/
Bτ for the simple meandering waveguide. Thus we obtain
As one can see from Figure 12.10 it is feasible to obtain the slow down factor on the order
of few tens for 10 GBps bit rate and N = 10 bits.
It is now important to return to the issue of the influence of ring loss on the dispersion
characteristics by imposing the condition that round trip loss αR should be much smaller
that coupling coefficient κ = S −1. Since κ decreases with bit rate we need to consider only the
low bit range of B<<Bloss where one immediately obtains from Equation 12.55.
By itself, spectral broadening should not be a problem because it may mean that the pulse
gets compressed in time domain; however, once the dispersion is taken into account, the
pulse shape deteriorates dramatically because the high frequency components propagate
much faster than the low frequency ones. Consider what happens with the same 50 ps long
Gaussian pulse of Figure 12.11a as it propagates through the SCISSOR with Td(,2C) = -3 × 10 4 ps 3
and the same loss acn -1 = 0.01 ps -1. In Figure 12.11b, we show the impact of GDD only on the
shape of the output pulse. The pulse is somewhat broadened and there are few decaying
“precursor” oscillations ahead of the pulse comprised by the higher frequency compo-
nents of the pulse that have shorter delay times. In Figure 12.11c, we also include the loss
and see that, while the main pulse does get narrower, the precursor pulses get amplified
0.8
Optical power (arb. units)
c
0.6
0.4 b
0.2
a
0
–200 –100 0 100 200
Time (ps)
Figure 12.11
Evolution of the Gaussian signal through the SCISSOR delay line. (a) Input pulse. (b) Output pulse disregarding
loss. (c) Output pulse with loss.
thus greatly distorting the signal. Therefore, to avoid excessive ISI the group delay over the
expanded bandwidth ∆ωsig,out should be less than one half of the bit interval, i.e
2 [ 4 ln(2)]2 Td(,2S)B3
≤ 1 (12.58)
1 - 16 ln(2)acn -1B2Td(,2S)
From this equation and using Equations 12.13 and 12.14 we obtain
2
1 + Bloss/B2
N r (B, N ) ≈ 2.2 N 3/2 (12.59)
1 + [ B/Bcut ] (1 + Bl2oss/B2 )
(N ) 2
where Bloss is still defined as Equation 12.49 while the cut-off bit rate for SCISSOR is 21/2
larger than the CROW cut-off rate
= ( 2 ln 2 τ ) N -1/2
(N ) -1
Bcut (12.60)
The dependence (Equation 12.59) is plotted in Figure 12.12 for four different values of
N using the same rings as in CROW example. Once again the plot can be split into three
distinct regions
The main difference from CROW is in the B -1 dependence of the required number of
resonators on the bit rate for bit rates below Bloss. Therefore, the slow down factor S for
SCISSOR is constant for B < Bloss and equal to roughly
105
Bloss
104
Required number
of resonators Nr
N = 250
103
S N = 50
102
N = 10
(N)
101 Bmin N=2
B(N)
cut
100
109 1010 1011 1012 1013
Bit rate (Bps)
Figure 12.12
Number of resonators required to store N bits in a SCISSOR delay line as a function of bit rate.
This is a very practical limitation on the slow down effect in SCISSOR, indicating that
with about 1% loss per ring one can expect that, independent of bandwidth, 100 bits can be
stored with a slow down factor of 10. We shall conclude our comparison of loss and group
delay dispersion in SCISSOR and CROW with the statement that for low loss case SCISSOR
offers an advantage while for the higher loss the benefits of CROW become more pro-
nounced. In the end, though the choice may have to be hinged on the technology because
with ring resonators both CROW and SCISSOR can be implemented with equal ease, while
photonic crystals CROW can be fabricated much easier than SCISSOR.
( )
3
1 τ 3 1 τ
Td( 2 ) = Td(,2C) + Td(,2S) = NC - N S = 0; (12.63)
2 2k 4 k1
(a) Out
Drop
In SCISSOR CROW
(b)
CROW
3
Delay per ring (ps)
2.5
1.5 SCISSOR
1
–100 –50 0 50 100
Frequency (GHz)
Figure 12.13
(a) Mixed SCISSOR–CROW delay line. (b) Group delay (per resonator) in the mixed delay line (solid line) and
its components (dashed lines).
where κ1 was defined in Equation 12.15. For Nc = Ns one readily obtains κ1 = 22/3κ. In
Figure 12.13b the results of dispersion compensation are shown. To keep things in pro-
spective we have normalized the group delay to the total number of rings. One can see
that the bandwidth gets enlarged, but, unfortunately, the dispersion compensation is not
complete, since the next order GDD terms Td( 4) have the same sign in CROW and SCISSOR
and one is left with the GDD of the form
where
( ) ( )
5
N C τ 5 N S τ N r τ 5 1 -4 ( 0) 5
Td( 4) = + ≈ = N r (Td ) (12.65)
24 2k 8 k1 12 2k 12
We then evoke once again the GDD limitation in the absence of losses and obtain in lieu
of Equation 12.20:
N r-4
2 [ 4 ln(2)]4 Td( 4)B5 ≈ B5 [ 4 ln(2)]2
6
[ NB-1 ] ≤ 1
5
(12.66)
From which we obtain the expression for the number of resonators required to store N
bits
Nr ≈ 2N5/4 (12.67)
This is indeed a substantial improvement over the Nr ~ 3N3/2 dependence in the absence
of compensation. For instance, without compensation, storing 50 bits would require 1000
MR’s, but with compensation 260 MR’s would suffice. The cut-off bit rate expression with
compensation becomes
(N )
Bcut ~ τ -1N -1/4 (12.68)
δω 0 δτ k1
= = ≈ 10-4 (12.69)
ω0 τ 3ω 0 τ
(b) 3.5
3
(a)
2.5
In Out 1.5
0.5 2δν
0
–200 –100 0 100 200
Frequency (GHz)
Figure 12.14
(a) Balanced SCISSOR delay line. (b) Group delay (per resonator) in the balanced SCISSOR line (solid line) and
its components (dashed lines).
12.6 Conclusions
In this chapter, we have analyzed limitations imposed on single and multiple coupled MR’s
(CRS) by group delay dispersion combined with loss. We have considered applications of
MR’s as delay lines and as optical switches. In case of delay lines we have shown that in
order to avoid distortion and ISI one has to use CRS with a large number of MR’s in CROW
or SCISSOR arrangements and that the required number of resonators Nr increases super
linearly with the number of stored (delayed) bits N as Nr ~ N3/2. We have also demonstrated
(N )
that there exists a cut-off bit rate Bcut ~ N -1/2 beyond which CRS offers no advantage over
a simple waveguide as an optical delay element. In addition we have explained that the
group delay dispersion distorts the signal shape both directly and also via loss-induced
changes in the spectrum. With the current state of MR technology both direct and loss-
related bit rate limitations are of equal importance. We have also reviewed the means of
mitigating the GDD.
For optical switching applications we have considered two alternative schemes. One is a
single-resonator optical switch in which the resonance frequency gets shifted and causes
the change in transmission. The other is the slow light scheme where the phase shift gets
accumulated as the light passes through CRS. We have demonstrated that for each scheme
there exists a maximum bit rate beyond which the MR offers no advantage over the stan-
dard MZI switch. We have shown that for relatively low speed switches (up to 10 GBs) the
single resonator scheme works as good if not better that the one with CRS, but for the high
bit rates the slow light scheme has a distinct advantage.
With the results obtained in this chapter one should be able to decide which MR scheme
can offer more benefits for each specific application and material.
References
1. F. Xia, L. Sekaric, and Y. Vlasov. Ultracompact optical buffers on a silicon chip. Nature Photonics,
1, 65–71, 2006.
2. H. Tazawa, Y-H. Kuo, I. Dunayevskiy, J. Luo, Al K.-Y. Jen, H. R. Fetterman, and W. H. Steier. Ring
resonator-based electrooptic polymer traveling-wave modulator. IEEE J. Lightwave Technol., 24,
3514–3519, 2006.
3. D. A. Cohen, M. Hossein-Zadeh, and A. F. J. Levi. High-Q microphotonicelectro-optic modula-
tor. Solid-State Electron, 45(9), 1577–1589, 2001.
4. B. Bhola, H-C. Song, H. Tazawa, W. H. Steier. Polymer micro-resonator strain sensors. Photon.
Technol. Lett., 17, 867–869, 2005.
5. A. Savchenkov, A. B. Matsko, M. Mohageg, and L. Maleki. Ringdown spectroscopy of stimulated
Raman scattering in a whispering gallery mode resonator. Opt. Lett., 32, 497–499, 2007.
6. B. Schmidt, Q. Xu, J. Shakya, S. Manipatruni, and M. Lipson. Compact electro-optic modulator
on silicon-on-insulator substrates using cavities with ultra-small modal volumes. Opt. Express,
15, 3140–3148, 2007.
7. M. Lipson. Compact electro-optic modulators on a silicon chip. IEEE J. Sel. Top. Quant., 12(6),
1520–1526, 2006.
8. C. K. Madsen and G. Lenz. Optical all-pass filters for phase response design with applications
for dispersion compensation. IEEE Photon Tech. Lett., 10, 994–996, 1998.
9. J, K. Poon, L. Zhu, G. A. DeRose, and A. Yariv. Transmission and group delay of microring
coupled-resonator optical waveguides. Opt. Lett., 31, 456–458, 2006.
10. S. Xiao, M. H. Khan, H. Shen, and M. Qi. Multiple-channel silicon micro-resonator based filters
for WDM applications. Opt. Express, 15, 7489–7498, 2007.
11. M. S. Nawrocka, T. Liu, X. Wang, and R. R. Panepucci. Tunable silicon microring resonator with
wide free spectral range. Appl. Phys. Lett., 89, 071110 2006.
12. J. E. Heebner and R. W. Boyd. ‘Slow’ and ‘fast’ light in resonator-coupled waveguides. J. Mod.
Opt., 49, 2629–2636, 2002.
13. J. E. Heebner, R. W. Boyd, and Q-H Park. SCISSOR solitons and other novel propagation effects
in microresonator-modified waveguides. J. Opt. Soc. Am. B, 19, 722–731, 2002.
14. Yariv, Y. Xu, R. K. Lee, A. Scherer. Coupled-resonator optical waveguide: a proposal and analy-
sis. Opt. Lett., 24, 711–713, 1999.
15. Melloni, F. Morichetti and M. Martnelli. Linear and nonlinear pulse propagation in coupled
resonator slow-wave optical structures. Opt. Quant. Electron., 35, 365–378, 2003.
16. J. B. Khurgin. Expanding the bandwidth of slow light photonic devices based on coupled reso-
nators. Opt. Lett., 30, 513–515, 2005.
17. J. B. Khurgin. Optical buffers based on slow light in EIT media and coupled resonator
structures—comparative analysis. J. Opt. Soc. Am. B, 22, 1062–1074, 2005.
18. M. Soljacic, S. G. Johnson, S. Fan, M. Inanescu, E. Ippen, and J. D. Joannopulos. Photonic-
crystal slow-light enhancement of nonlinear phase sensitivity. J. Opt. Soc. Am. B, 19, 2052–2059,
2002.
19. J. B. Khurgin. Performance of nonlinear photonic crystal devices at high bit rates. Opt. Lett., 30,
643–645, 2005.
20. Q. Xu, S. Manipatruni, B. Schmidt, J. Shakya, and M. Lipson. 12.5 Gbit/s carrier-injection-based
silicon micro-ring silicon modulators. Opt. Express, 15(2), 430–436, 2007.
21. P. Millar, J. S. Aitchison, J. U. Kang, G. I. Stegeman, A. Villeneuve, G. T. Kennedy, and W. Sibbett.
Nonlinear waveguide arrays in AlGaAs. J. Opt. Soc. Am. B, 14, 3224–3231, 1997.
22. D. A. Nikogosyan. Properties of Optical and Laser Related Materials. A Handbook. Wiley, NY,
1997.
23. J. B. Khurgin. Power dissipation in slow light devices: a comparative analysis. Opt. Lett., 32,
163–165, 2007.
Shayan Mookherjea
University of California, San Diego
Contents
13.1 Introduction....................................................................................................................... 529
13.2 Waveguiding Principles and the Dispersion Relationship......................................... 531
13.2.1 Waveguide Mode................................................................................................ 531
13.2.2 Dispersion Relationship.................................................................................... 531
13.2.3 Tail of the Dispersion Relationship.................................................................. 532
13.3 Localization in the Presence of Disorder......................................................................534
13.4 Nonlinear Localization.................................................................................................... 539
13.4.1 Quadratic Dispersion at the Band Edge.......................................................... 539
13.4.2 The Nonlinear Evolution Equation..................................................................540
13.4.3 Time-Invariant Evolution.................................................................................. 541
13.4.4 The “Super-Resonant” Mode............................................................................542
13.4.5 Nonlinear Anderson Localization...................................................................544
13.5 Cascaded Versus Nested Coupled-Resonator Structures...........................................545
13.5.1 Slow Light in Fabry–Perot and Gires–Tournois Resonators.........................545
13.5.2 Slow Light in the Coupled Fabry–Perot Structure......................................... 547
13.5.3 Advantages and Disadvantages of the Nested Architecture.......................548
13.6 Summary............................................................................................................................ 550
Acknowledgments....................................................................................................................... 551
References...................................................................................................................................... 551
13.1 Introduction
There are various aspects of localization which are important in condensed matter physics,
including the metal-insulator (Anderson) transition, localization of charges in amorphous
and disordered semiconductors, tunneling and breakdown in insulators, oscillations of
the conductivity in inversion layers and field-effect transistors. In optics, light can be local-
ized in k-space, as shown by coherent backscattering in aqueous solutions of polystyrene
beads [1]. The width of the backscattering cone and the angle of its cusp are inversely
proportional to l, the mean free path of photons. This called “weak localization” and is a
precursor to strong, i.e., Anderson localization.
Localization of light has been demonstrated relatively recently, in three and two dimen-
sions [2,3]. In such cases, recurrent multiple scattering of light in random media leads to
529
© 2009 by Taylor & Francis Group, LLC
strong localization if the Ioffe-Regel criterion is satisfied, kl ≤ 1 where k = 2π/λ and l is the
elastic mean free path of photons. Since the product kl diverges in both the short wave-
length (kl ∼ 1/λ) and long wavelength (kl ∼ λ3) limits, localization can take place only in a
limited wavelength range [4].
Localization in a truly-1D or quasi-1D structure (in the latter case, the number of trans-
verse modes N = k2 A/4π >> 1 where A is the cross-sectional area of the waveguide) is poten-
tially simpler to achieve because of the restrictions on the propagation constant kz imposed
by the geometry, which makes it easier to realize sufficiently strong multiple scattering.
In a long quasi-1D structure, energy injected at any point of the input is equally likely
to emerge at any point of the output, and modes are completely mixed by the medium.
A diffusion-theory based approach to the study of photon transport may be adopted.
However, in a truly-1D structure, the diffusion time scale is absent, and there are only the
ballistic propagation regime and the localized regime [5].
An optical slow-wave structure consists of a chain of resonant cavities (such as a defect
cavity resonator in a photonic crystal slab, or a Fabry–Perot resonator) or other forms of
resonators such as microrings, microdisks etc. [6,7]. In this context, localization is defined
as an exponential maximum of the intensity in real (coordinate) space along the length
of the waveguide. Confinement in the transverse plane is provided by the index contrast
between the core and the cladding of the waveguide cross-section.
We will discuss here two distinct approaches to potentially localizing light in an optical
slow-wave structure: (a) taking advantage of Anderson localization at the band edge in a
disordered structure [8,9], or (b) utilizing the Kerr optical nonlinearity in conjunction with
the unique dispersion relationship of the slow-wave structure to create slowly-moving or
stopped soliton-like pulses [10,11]. (Very recently, researchers have also been investigat-
ing combinations of the two approaches—control over Anderson localization utilizing the
optical Kerr effect in nonlinear coupled-waveguide structures [12].)
A typical realization of the optical slow-wave structure is the coupled-resonator optical
waveguide (CROW), which is formed by placing optical resonators in a linear (or two/
three dimensional) array, so as to guide light from one end of this chain to the other [13–15].
Energy transfer occurs because of nearest-neighbor interaction: the spatial overlap between
the fields of adjacent unit cells. Although each unit cell could be custom-designed, which
is done in the synthesis of optical filters with specified transfer functions [16], it is usually
assumed that a slow-wave waveguide consists of nominally identical unit cells.
Since there is translational periodicity along the length of the waveguide, the modes of
the structure can be described by Bloch functions [17]. In the presence of disorder, coher-
ent back-scattering is restricted only to certain Bragg-selected channels, making the pos-
sibility of localization more likely [4]. We have fabricated such structures using a high
index contrast material system (silicon for the waveguide core, and silicon dioxide or air
for the waveguide cladding regions) which enhances the perturbation-induced coupling
of modes—the coupling coefficient is (approximately) proportional to the difference in
relative permittivity of the core and the cladding.
The dispersion relationship of an infinitely-long slow-wave waveguide can be found
using tight-binding theory, as described below. For a finite-length structure, a matrix
theory has been developed [18]. Many potential applications of slow-wave structures,
such as slow light, enhanced optical nonlinearities etc., involve the use of pulses rather
than continuous-wave (CW) light. Given an explicit relationship (albeit a nonlinear one)
between ω and k, one can derive expressions describing pulse propagation [19]. We have
derived such expressions in both the linear dispersion approximations [20] while taking
into account the higher orders of dispersion [21]. The latter result is unusual in optical
waveguide theory, as it is one of the rare cases when one can write down a nonperturba-
tive propagation equation for any input envelope to all orders of dispersion, and therefore,
pulse distortion can be exactly quantified.
e - iω k tE wg (r ) = e - iω k t ∑e
n
inR k⋅e z
E res (r - nRe z ), (13.1)
where n is an index over the constituent resonators. Often, each constituent resonator has
two (or more) mutually orthogonal eigenmodes, and then, each set of modes leads to a
distinct waveguiding band [22]. Equation 13.1 has the Bloch form, and in drawing the dis-
persion relationship, we restrict the range of k to the first Brillouin zone, |k|R<π.
2
∇ 2E wg (r ) + ε wg ω2k E wg = 0, (13.2)
c
and use the fact that Eres obeys exactly the same equation but substituting εres in place of
εwg, and Ω in place of ωk. After some algebra [13], the dispersion relationship for a CROW
waveguide mode is of the familiar tight-binding “cosine” form:
where the various parameters in the above equation all have physical meaning. Ω is the
eigenfrequency of the individual resonators, and ∆α and k are overlap integrals involv-
ing the individual resonator modes and the spatial variation of the dielectric constant,
defined as
∫ d r[ε
2
∆a = 3
wg (r - Re z ) - ε res (r - Re z )] E res(r ) ,
(13.4)
k=
∫ d 3r[ε res (r - Re z ) - ε wg (r - Re z )]E res (r ) ⋅ E res (r - Re z ),
where, as before, εres is the dielectric coefficient of an individual resonator (in isolation),
and εwg is the dielectric coefficient of the waveguide. ∆α represents the shift of the eigen-
frequency of a single resonator when it is coupled to its neighbors, and k is the coupling
coefficient between two adjacent resonators. Consequently, the position and slope of the
dispersion relationship are controlled by the geometric design of the waveguide through
the overlap integrals defined in Equation 13.4.
The slowing factor for the waveguide describes how slow light is (at band-center) com-
pared to a conventional waveguide,
c/neff 1
S≡ ≈ (13.5)
max.v g 2 | k | m
where m = neffR/λ is the mode number expressing how many wavelengths λ/neff fit into a
single period R. It is a constant in a particular design, and reflects the characteristic length
scale of the structural elements, e.g., m ≈ 1−10 for coupled Fabry–Perot resonators, whereas
m ≈ 50−100 for coupled microrings.
Based on Equation 13.3, the waveguiding bandwidth (defined as one-half of the end-to-
end width) is
2|k|mc c
∆ω = = , (13.6)
neff R Sneff R
which shows that the slowing factor S and waveguide bandwidth ∆ω are inversely propor-
tional to each other, through their dependence on |κ|. Therefore, there is a fundamental
tradeoff between how slow a pulse can propagate in a coupled-resonator optical slow-
wave structure, and its transmission bandwidth. A representative figure-of-merit, defined
as the product of the slowing factor and bandwidth, is inversely proportional to the length
of a unit cell, R.
Values of |κ| for typical coupled-resonator optical slow-wave structures are listed in
Table 13.1. For coupled microresonators, a representative range for k is 0.05–0.25, with
larger values resulting in a larger transmission bandwidth, and smaller values resulting
in a larger slowing factor.
Table 13.1
Typical Values of Microresonator Coupling Coefficients, Which are Used to Construct Coupled-
Resonator Optical Slow-Wave Structures
Microrings:
(a) ω−ω0
1.010
∆ω
1.005
(b)
1.000 Density of states
2.0 2.0
0.995
K π
R
0.990
0.99 1
ω−ω0
1.0 1.0
∆ω
Norm. frequency
0 0
−1.0 −1.0
Dispersion
−2.0 −2.0
0 0.2 0.4 0.6 0.8 1 0 1 2 3
K ρ̂(ω)
Norm. wavenumber π/R
Figure 13.1
Dispersion and density of states for a weakly disordered slow-wave structure. (a) Dispersion relationship calcu-
lated ab initio for a slow-wave structure with 1% disorder in the nearest-neighbor unit cell coupling coefficients.
For such weak disorder, the dispersion follows the tight-binding (cosine) function almost exactly, except at
the band edge, shown by the inset. The band edge tail shows that vg ≡ dw/dk is non-zero. (b) In the presence
of disorder, the density of states is calculated over the same range of frequencies, showing that the density of
states, ρˆ (ω ) , does not diverge to infinity at the band edges. (From Mookherjea, S., Park, S. J., Yang, S. S. H., and
Bandaru, P. R. Nature Photonics, 2(2), 90–93, 2008. With permission.)
To understand why group velocity does not go to zero, it is necessary to consider the
connection between group velocity and the density of states. In a weakly disordered slow-
wave structure, the group velocity vg is inversely proportional to the density of states, and
is given by [23]:
vg =
1 1
ˆρ(ω ) 2 π/R(1+
1 dφ
NR dk
, ) (13.7)
where ρ̂ is the optical density of states (normalized to unit integral over the waveguide
band), and φ is the disorder-induced (scattering) phase shift. In a perfectly ordered struc-
ture, ρˆ (ω edge ) → ∞ at the band edges [24] and thus vg → 0. Whereas, as shown in Figure 13.1b,
in a disordered structure, ρˆ (ω edge ) only reaches a certain maximum value that depends on
the statistics of variation of the elements of M—most importantly, the mean and standard
deviation of the distribution of coupling coefficients, k and δκ, respectively [23].
In the presence of disorder, represented most conveniently by the standard deviation in
the inter-unit-cell coupling coefficients δκ normalized to its mean k, we have shown using
ab initio simulations that [23]:
v g at band center
v g at band edge
= 0.667 log 10
k
δk( )
+ 0.313. (13.8)
Consequently, Sbe, which is defined as the band edge slowing factor taking into account
the effects of disorder, is
Sbe ≡ c = λ 1 . (13.9)
vg at band edge R (δk 2⋅k )1/ 3
The band edge slowing factor is a measure of how much a slow-wave structure, or CROW,
is free from disorder. Equation 13.9 shows that Sbe → ∞ (since vg at band edge → 0) if δκ → 0.
But, for a typical structure, if κ = 10−2, δκ = 5% of k, and R = 10λ, then Sbe = 74, a much more
modest slowing factor. Experimental observations also indicate Sbe ≈ 10−100 [25,26].
A 1-cm long slow-wave structure which achieves maximum S = 100 may be replaced sim-
ply by a spool of fiber of length 1 cm × S × neff/nfiber ≈ 2 m, which is considerably easier to
implement in practice, and can be packaged into a fairly small volume. An alternate chip-
scale method to achieve comparable slowing factors is described in Section 13.5. Therefore,
using a coupled-resonator slow-wave structure only to slow down the propagation of light
is unlikely to yield a commercially viable technology.
However, we may gain new insights from the study of fundamental physical phenom-
ena related to electron transport within the context of optical slow-wave structures, such
as Anderson localization. On-chip tunability of delay and localization at speeds in excess
of achievable by thermal tuning of fiber Bragg gratings may yet offer a considerable advan-
tage over the state-of-the-art.
localized modes decay exponentially with distance, the average of log T (λ) over many
realizations of disorder decays linearly with the sample thickness, <logT> = −L/(ξ + κe−1)
where ξ is a length scale called the localization length and κe−1 is an extinction coefficient
(units of inverse length) that takes into account the total loss, including scattering losses
and absorption. A slightly different expression, <logT> = −L(1/ξ + κe) has also been proposed
[28]. Although the averaging is sometimes carried out over wavelength (assuming ergodic-
ity) for the sake of experimental convenience—since the averaging over many realizations
of disorder would require fabricating and measuring many samples—wavelength averag-
ing is not suitable for dispersive structures such as the coupled-resonator waveguide.
Another way of examining localization is in the time domain: the Lorentzian line shape
of a localized mode generates an asymmetric time response with an exponentially decay-
ing tail [27]. Again, the highly dispersive nature of a coupled-resonator waveguide near
the band edge makes this approach difficult for a slow-wave structure.
A third way of investigating localization, which has been used recently in several exper-
iments [8,9,12] is to input light into the structure using a tunable CW laser and directly
image the optical fields in a surface-normal configuration. Care should be taken not to
perturb the fields strongly, e.g., by introduction of a metal-coated fiber tip in close vicin-
ity of the waveguide. Here, localization is defined by the exponential maximum of the
intensity in real (coordinate) space along the length of the waveguide. Confinement in the
transverse plane is provided by the index contrast between the core and the cladding of
the waveguide cross-section.
As we will discuss in this section, how disorder in the slow-wave structure can lead to
localization of optical fields at frequencies near the band edges. This is shown in Figure
13.2. When there is no disorder, the spatial distribution of the field is
(a) (b)
16
Mode number
1
0.9995 1 0 0.25 0.50 0.75 1
(ω − ω0)/∆ω Distance along structure
Figure 13.2
Spectral distribution of modes with disorder-induced localization. (a), Eigen-frequencies from a single Monte-
Carlo calculation are labeled from the band edge inwards (1 = farthest detuning from band-center), showing that
disorder creates states beyond the band edge of the ideal structure. The corresponding density of states is shown
in Figure 13.1. (b) The corresponding field distributions along the (normalized) length of the slow-wave structure,
showing that field distributions for frequencies near the band edge, at the bottom of the figure, are localized, and
those inside the band, towards the top of the figure, are extended. (From Mookherjea, S., Park, S. J., Yang, S. S. H., and
Bandaru, P. R. Nature Photonics, 2(2), 90–93, 2008. With permission.)
〈 z|Ψ k 〉 ≡∑e n= 1
- inkR
E single (r - nRzˆ ) (13.10)
i.e., a Bloch sum over the single-resonator fields E single(r). ψk satisfies the generalized eigen-
value equation
∇ × ∇ × |ψk 〉 = εwgEk|ψk〉 (13.11)
where Ek ≡ (ωk/c)2. Field distributions for modes near the band edge are shown in Figure
13.3. In the disorder-free case, the eigenfrequencies do not exceed ω0 + ∆ω, unlike in the
disordered case shown in Figure 13.2.
The band edge modes do not couple to the input and output edges of the structure even
in the ideal disorder-free case, as can be seen from the left and right edges of Figure 13.3b.
But they can be excited by light already propagating in the waveguide at a nearby wave-
length if the refractive index of the waveguide is changed, e.g., by the electro-optic or
thermo-optic effect. Notice a change of less than 0.1% of the bandwidth of the dispersion
relationship is sufficient; the dispersion bandwidth is itself on the order of a few nanome-
ters in a typical CROW. In practice, transmission at the band edge may not fall by more
than 10–20 dB in a typical structure, in part to the presence of substrate-guided or leaky
modes, in addition to those considered in Equation 13.10.
We write the disorder as εwg(r) → εwg(r) + δεwg(r), and the new field distributions as |ψk〉,
which satisfy
∇ × ∇|ψk〉 − δεwgEk/ψk〉 = εwgEk|ψk 〉. (13.12)
(a) (b)
16
Mode number
1
0.999 1 0 0.25 0.50 0.75 1
(ω − ω0)/∆ω Distance along structure
Figure 13.3
Spectral distribution of modes in the absence of disorder. (a) Eigen-frequencies from a single Monte-Carlo cal-
culation are labeled from the band edge inwards (1 = farthest detuning from band-center). In the absence of
disorder, the eigenfrequencies approach but do not exceed the band edge ω0 + ∆ω (compare with Figure 7.2a).
(b) The corresponding field distributions along the (normalized) length of the slow-wave structure, which are
similar to the amplitudes of the supermodes of waveguide arrays [43].
which may be verified by multiplying the left and right hand sides by (εwg−H).
In terms of Green’s function for the disordered structure, G(z,z′; Ek), the solution may be
written as
G(r,r′; Ek) is not known exactly, of course, but may be found quasi-analytically from vari-
ous well-known theories such as the average t-matrix approximation (ATA), or the coher-
ent potential approximation (CPA) [24]. However, these methods do not work well in the
region of our interest—the band edge, where the density of states is dominated by contri-
butions from resonance or localized eigenstates, as the numerical results suggest.
To obtain a general idea of the behavior of Equation 13.14, one may approximate G(z,z′; Ek)
by G 0(z,z′; Ek) for z′ ≠ z, where G 0 is Green’s function for the unperturbed structure [24],
and for z′ = z, by −πρ(z,E), in terms of the local density of states, calculated numerically for
the perturbed structure. Thus writing the disordered fields directly in terms of the per-
turbed density-of-states shows that those fields ψk(r) which are most perturbed are those
for which ρ(Ek) is the most significantly affected—i.e., the band edge states.
Although it has been known theoretically that all the eigenmodes are localized for any
degree of disorder (however small) if the structure is infinitely long [29], light may not be
localized if the physical length of the slow-wave structure does not exceed the localization
length. In a structure of finite length, localized modes are first observed near the band
edges, as the modes at band-center are localized on a length-scale larger than the length
of the system. To observe localization effects, fairly long slow-wave structures consisting
of a large number of coupled resonators are needed, as is necessary anyway for practical
applications of slow-wave structures as optical interconnects, which also require fairly
long device lengths [30]. An estimate for how many resonators should be in the chain in
order to observe localization can be obtained by the following argument: for a given value
of N, the (dimensionless) energy separation δE between the two eigenmodes closest to the
band edge is
πN π( N - 1)
δE = 2k cos - cos ≈ 3k π 2/N 2 (13.15)
N +1 N +1
for large N, where N is the number of resonators in the chain, and k is the average coupling
coefficient. Localization occurs when this energy separation is approximately equal to the
energy perturbation due to disorder, which is ≈ δκ/√N, after taking into account the effects
of exchange narrowing [31]. The equality implies that N = (3π2κ/δκ)2 ⁄3 ≈ 9(κ/δκ)2/3 which
evaluates to 93 unit cells for 3% disorder, and 66 unit cells for 5% disorder. (Higher values
of disorder may not obey the ballistic propagation model in this waveguide geometry.)
Three recent experiments on localization in 1D waveguide structures have fabricated
waveguides consisting of (in two cases) 100 unit cells, in which each unit cell consisted of
a lithographically defined resonator [8] or a sequence of side-coupled waveguides [12], and
(in one case) about 250 unit cells of a single-line-defect photonic crystal waveguide [9].
One of these experiments [8] has demonstrated localization in a slow-wave structure
(see Figure 13.4) consisting of a sequence of cuboidal resonators (the volume of a resonator
(a)
In Out
(b) (c)
Transition
1 µm
Acc. V Spot Magn Det WD 5 µm
10.00 kV 2.0 5945x JSP Acc. V Spot Magn Det WD 1 µm
SE 30.8
10.00 kV 2.0 16543x SE 30.8 JSP
Figure 13.4
Slow-wave coupled-resonator waveguides. (a) The fabricated structure on an SOI chip consists of a single
mode waveguide loaded periodically with resonators, with two transition sections to better match impedances
between the input/output sections and the slow-wave section. The top and side claddings are air, and the bot-
tom cladding is silicon dioxide. (b) A magnified view of the transition section, and (c) a magnified view of the
slow-wave section, with the periodicity R being 2.75 µm. (From Mookherjea, S., Park, S. J., Yang, S. S. H., and
Bandaru, P. R. Nature Photonics, 2(2), 90–93, 2008. With permission.)
(a) (b)
1574.0 (c)
1.2
ξ~ω–2
Intensity profile
1574.5
Wave length (nm)
1.0
Log10 (ξ)
1575.0
1575.5 0.8
Figure 13.5
Experimental measurements of localization. (a) Measured transmission spectrum for the band edge wavelengths
of a coupled-resonator structure as shown in Figure 13.4b, Measured spatial profiles of a representative extended
field distribution at 1576.16 nm, and localized field distributions at 1575.64 nm, 1575.60 nm, 1575.10 nm, and 1575.00
nm (from bottom to top). (c) Log-log plot of the localization length ξ̂ (normalized units) versus frequency ω̂ (normal-
ized units), in agreement with the theoretically-calculated shape. (From Mookherjea, S., Park, S. J., Yang, S. S. H., and
Bandaru, P. R. Nature Photonics, 2(2), 90–93, 2008. With permission.)
detuning from the frequency of maximum localization, in agreement with the theoreti-
cally expected behavior [28,29]. Based on simulations, we predicted that for a finite-length
structure consisting of a chain of 100 coupled resonators, the minimum localization length
should be approximately 5.6 unit cells. By way of comparison, from Figure 13.5c, the small-
est localization length measured was 6 unit cells (17 µm). Note that unlike the theoretical
calculations which assume an infinitely-long structure, “perfect” localization (log ξˆ → 0) is
not observed in a finite-length structure.
At the present time, given the fabrication difficulties in making long sequences of cou-
pled resonators, all the three experiments carried out recently [8,9,12] have explored some-
what different aspects of localization. Future studies will undoubtedly converge to a more
unified view of localization in slow-wave structures, and possible applications in creating
high-Q states, or trapping and releasing light dynamically.
1 i
ε( z , t) =
2π ∫
dK ε 0 ( K )exp aK 2t - iKz ,
2
(13.16)
1/2
t2
w(t) = w0 1 + 2 , (13.17)
TD
π /R
dK - i∆ω t cos[( k0 + K )R ]
ε(r , t) = e - iω 0t
∫
- π /R
e
2π
c k0 + K (t) E wg (r ) . (13.18)
Note that since nonlinear phenomena such as the Kerr effect change the relative weights
of the eigenmodes as the waveform evolves with time, we have appropriately introduced a
time-dependency in the superposition coefficients ck appearing in Equation 13.18.
The exact nature in which these coefficients change with time is given by substituting
the above expression into the nonlinear wave equation, written out with an explicit non-
linear polarization term describing the Kerr effect,
∂2 ∂2
∇ × ∇ × ε(r , t) - µε wg (r ) 2 ε(r , t) = µ PNL (r , t)
∂t ∂t 2 (13.19)
where
3
PNL (r , t) = ε 0χ( 3) E(r , t) 2E(r , t). (13.20)
4
in the instantaneous response approximation.
In simplifying the terms, we use the normalization
M
∫
M dr ε wg (r ) ∑E
m= 1
res (r - mRe z ) 2 = 1 (13.21)
∫ ∑E
4
γ = 2 n0n2 ε 0 ω 0 dr res (r - mRe z ) , (13.22)
m
using the relationship 3χ(3)/8 = n0n2 [36], and we have ignored the dispersion (varia-
tion in ω) of γ. (Conversion to real world units follows conveniently from the equation
n2 = n02 + 2n0n2|E|2 where n0 is the linear refractive index.)
If we assume that ck0 + K(t) varies slowly over time intervals ∼ O(2π/ω0), as is usually the
case, then we obtain
π/R
dc k0 + K (t) dK1 dK 2
dt
= iγ
∫ ∫ 2π
- π/R
2π
exp[- ikΩt{- cos[( k0 + K1 )R] + cos[( k0 + K 2 )R]
where K1 + K = K 2 + K3 and γ is the nonlinearity coefficient in the CROW geometry, described
above.
Equation 13.23 can be written in a different basis set. The an’s defined as
π /R
dK ∆a
an (t) =
∫
- π /R
2π
c k0 + K (t)exp [ in( k0 + K )R ]exp i
2
Ω - kΩ cos[( k0 + K )R] t
(13.24)
are the coefficients that appear in the expansion of the field in terms of individual resona-
tor modes, rather than the waveguide modes. The resultant differential equation is
dan ∆a k
i + Ω an - Ω ( an+1 + an-1 ) + γ | an |2an = 0. (13.25)
dt 2 2
Equation 13.25 is the nonlinear generalization of the linear evolution equation for an(t)
discussed later (Equation 13.45), along with a minor redefinition of the center frequency,
represented by the second term, from one centered at an optical frequency to base-band
(see Equation 13.24).
ck0 + K(t) = Ak0 + K(t)exp[iφk0 + K(t)] (13.26)
We will look for solutions that retain their shape, i.e., dA/dt = 0. Substituting Equation
13.26 into Equation 13.23 and separating the real and imaginary parts, we obtain a pair of
equations,
dAk0 + K dK1 dK 2
dt
= -γ
∫∫ 2π 2π
n Φ,
Ak0 + K1 Ak0 + K2 Ak0 + K3 sin (13.27)
d φ k0 + K γ dK1 dK 2
dt
=
Ak0 + K ∫∫ 2π 2π
Ak0 + K1 Ak0 + K2 Ak0 + K3 cos Φ , (13.28)
where Φ is defined as
Based on Equation 13.27, the A’s will be independent of t if sin Φ ≡ 0 for all t. This implies
that cosΦ = 1, and based on Equation 13.28, we take φk0 + K to be a linear function of t,
φk0 + K(t) = a + bt + κΩtcos[(k0 + K)R], (13.30)
where a and b are constants independent of t and k. We drop the constant a which repre-
sents a fixed phase that can be absorbed into the initial conditions. Substituting this form
for φk0 + K(t) into Equation 13.28, we get
π/R
γ dK1 dK 2
b + k Ω cos[( k0 + K )R] =
Ak0 + K ∫ ∫ 2π
- π/R
2π
Ak0 + K1 Ak0 + K2 Ak0 + K3 . (13.31)
π/R
( KR)2 γ dK1 dK 2
b + kΩ = kΩ
2
+
Ak0 + K ∫ ∫ 2π
- π/R
2π
Ak0 + K1 Ak0 + K2 Ak0 + K3 (13.32)
We assume that the A’s are defined to be zero outside the regions of integration - π/R and
π/R so that the limits of integration can be taken as - ∞ to ∞. Equation 13.32 may then be
solved [10],
where K is a spectral width parameter whose relevance will become clear in the following
discussion. Substituting Equation 13.33 into Equation 13.32, we get
2
( KR)2 2 γ πKR
b + kΩ = kΩ + 2 Ak( 00 )+ K ( KR ) 2
+ (13.34)
2 (2 πR)2 2
kΩ
Ak( 00 )+ K = -(2 πR)2 . (13.35)
4γ
Since the left-hand side represents a real and positive number, we require that k as
defined in Equation 13.4 be a negative number (as is physically expected from the mean-
ing of εwg and εres). This is equivalent to anomalous dispersion in optical fibers and similar
waveguides.
Using Equations 13.30 and 13.35 in Equation 13.26, we write the final expression for
ck0 + K(t),
where
kΩ
c k0 + K (0) ≡2 πR - sech( K/K ), | KR | ≤ π. (13.37)
4γ
In light of Equation 13.37, the integral on the second line of Equation 13.38 is not express-
ible in a simpler form. However, if KR <∼ 1, the hyperbolic secant function decays rapidly,
and the limits of integration may be changed to (-∞,∞). The integral then can be evaluated
easily—the Fourier transform of a hyperbolic secant is itself a hyperbolic secant function.
We derive the approximation,
∑ sech
- ik Ωt 1+ π 2 ( KR )2 / 8 kΩ πK
E(r , t) ≈ e - iω 0t e - πKR nR E res (r - nRzˆ ) (13.39)
4γ 2
n
The modulus of the amplitude |ε(z,t = 0)| normalized to its maximum value (in this
approximation) is plotted in Figure 13.6. Values of the hyperbolic secant function in
× 105
6
4
|E| [(V/cm)]
0
−300 −200 −100 0 100 200 300
Distance (µm)
Figure 13.6
Calculation of a super-resonant field distribution, assuming that the individual resonator eigenmodes are
Gaussian in shape with width 1 µm, with inter-resonator spacing R = 10 µm, background refractive index
n0 = 2.5, nonlinear index n2 = 1.5 × 10 - 13cm2/W and K =1/(5R). The input power is 1 Watt, and the cross-sectional
dimensions are taken as 0.5 µm × 0.5 µm. The inter-resonator coupling coefficient is assumed to be of magnitude
|κ| = 0.01 . The dotted line is an envelope—a hyperbolic secant—connecting the excitation coefficients multiply-
ing the individual resonator eigenmodes.
Equation 13.39 at nR (which has the dimensions of length) are the weights of the individual
resonator eigenmodes. In this approximation, the envelope of these weights is a hyperbolic
secant function whose width is inversely proportional to K.
As we had expected from physical arguments, the envelope of ε(r,t) is a stationary state
that is independent of time: its spatial distribution at t = 0 is maintained for all subse-
quent t. This is consistent with the observation that although the group velocity disper-
sion coefficient is nonzero, the group velocity itself is zero. We call the stationary state a
super-resonant field since it is formed in a waveguide that itself comprises the coupling
of individual (stationary) resonator modes. At the present time, the super-resonant mode
of a coupled-resonator optical slow-wave structure has not been experimentally observed;
however, it has been verified by simulations [37].
∂U n
-i = β nU n + C (U n+1 + U n-1 ) + γ | U n |2 U n (13.40)
∂z
where βn is the propagation constant of the nth waveguide, C is the nearest-neighbor cou-
pling coefficient (in units of inverse distance), and γ represents the effect of the Kerr non-
linear index.
Disorder is introduced by randomly changing the widths of each waveguide (a nominal
value of the width is 4 µm), so that the parameters βn become random in the range β0 ± ∆.
The measure of disorder is set by the ratio ∆/c = 1. The input intensity was below the self-
focusing limit so as to operate in the weakly nonlinear regime.
Recall from our earlier discussion that the two band edges of Equation 13.3 are char-
acterized by k = 0 and k = ± π/R, which represent a relative phase shift kR = 0 and kR = ± π
between adjacent unit cells. It was observed by Lahini et al. that a positive Kerr nonlin-
earity tends to further localize those “flat phased” modes in which the phase-difference
between nearest-neighbors is close to 0, and tends to delocalize those “staggered” modes
in which the phase-difference between nearest-neighbors is close to π.
It was further observed, by injecting light of different intensities into a single waveguide
and monitoring the evolution along z, that localization in a coupled 1D structure emerges
out of the ballistic propagation regime without a signature of diffusive behavior [5], as
would be observed in higher dimensions [2,3].
1 + r22
T FP = T (13.41)
1 - r22
(a) R
In r2 Out
(b) Dielectric
In –r2
Air
Partial
Out reflector
r1 100% reflector
L
Figure 13.7
Cascaded and nested slow-wave architectures. (a) A cascaded resonator slow-wave architecture, consisting of
identical Fabry–Perot etalons (thickness d) in a linear array with inter-resonator spacing R. (b) A nested resona-
tor slow-wave architecture, consisting of a Fabry–Perot etalon inside a Gires–Tournois resonator of length L. The
amplitude reflection coefficients at various air-dielectric interfaces are marked.
s d(2 φ GT )
τ GT = (13.42)
1 + (s 2 - 1)sin 2 φ GT dω
Combining these three facts (a–c), we can readily conclude that when a maximum-delay
Fabry–Perot etalon is placed inside a Gires–Tournois resonator, as shown in Figure 1.7b,
the only effect is to modify φGT which appears in Equation 13.42,
ω
φ GT → φ FP+GT = (L - d) + φ FP . (13.43)
c
Therefore, the group delay (upon reflection) of the nested structure shown in Figure
13.7b is
s d(2 φ FP+GT )
τ FP+ GT = 2 2
1 + (s - 1)sin φ FP+GT dω
s 2(L - d) (13.44)
= 2 2 + 2τ FP .
1 + (s - 1)sin φ FP+GT c
≡ n
According to Equation 13.43, we can control φFP + GT, for example, by changing L by translat-
ing either of the mirrors comprising the Gires–Tournois etalon. In doing so, we can tune
the pre-factor η in Equation 13.44 continuously between a minimum value ≈ 1/σ (antireso-
nance) to a maximum value ≈ σ (resonance). In other words, the group delay of the nested
resonator structure (in reflection) can be σ times larger than twice the maximum group
delay of the Fabry–Perot resonator alone (in transmission). The minimum group delay of
the nested structure can also be made a factor of σ smaller, which may be useful in some
applications.
dan k
i + Ω an + Ω ( an- 1 + an+ 1 ) = 0 (13.45)
at 2π
where Ω is the resonant frequency of a single resonator and κ is the (dimensionless) nearest-
neighbor coupling coefficient. Assuming periodic boundary conditions,* we guess a solu-
tion of the form
1
an (t) = e i(ω t- nkmR) (13.46)
N
where km = m2π/(NR). Substituting Equation 13.46 into Equation 13.45, we obtain that
ωm k
= 1 + cos( kmR), m = 0, 1,…, N - 1 (13.47)
Ω π
are the N normalized eigenfrequencies of the chain. In the limit of large N, Equation 13.47
defines ω as a continuous function of k, which, in the first Brillouin zone, takes values from
–π/R to π/R. The group velocity is defined as νg(ω) ≡ dω/dk from Equation 13.47.
* In assuming periodic boundary conditions, we rely on our physical intuition that the “bulk” optical proper-
ties within the waveguide are unaffected by the exact nature of the boundaries. In chains consisting of only
a few unit cells, this need not be true [39]. Chains of resonators with gain may also be sensitive to boundary
conditions [40].
A pulse propagates with minimum distortion if it is centered at mid-band, i.e., with car-
rier frequency ω = Ω. The time taken by such a pulse to propagate across M unit cells of the
slow-wave structure is
MRneff
τ(CROW
M)
= . (13.48)
kc
1- | r2 |2
| k | ∈ ,1 , (13.49)
1+ | r2 |2
which correspond to the maximum and minimum group delays of the slow-wave struc-
ture, respectively. Using the minimum value of |k| (for the largest possible delay), setting
R = L, and thereafter equating Equation 13.48 and 13.44, we obtain
M=s
2d
L ( τ
τ )
2 + | k | GT ≈ 2 s , (13.50)
since typically, κ<<1, τGT ≈ τ, and 2d ≈ L., As expected, M is proportional to σ, and there-
fore, the higher the quality factor of the Gires–Tournois resonator, the longer the length of
the corresponding unfolded chain that achieves the same delay.
At the present time, the state-of-the-art in cascaded identical microrings is 12 sequen-
tially coupled resonators [26]. (Longer chains of microrings have recently been reported
[25] but their transmission spectra are complicated due to mis-matching of resonances in
long chains of not-exactly-identical resonators.) On the other hand, achieving σ = 6 is fairly
straightforward in a Gires–Tournois resonator, corresponding to an intensity reflection cof-
ficient |r1|2 = 51% ≈ 3 dB at the front mirror. Therefore, for small delays, the obtainable delay
from the nested structure would easily compare with the best cascaded micro-resonator
structures demonstrated, which require considerably more effort to fabricate.
1. The nested structure avoids some of the disorder-induced limitations of slow light
in the unfolded coupled-resonator structure, whose performance greatly suffers
if the inter-resonator spacings are not exactly identical [23]. Conceptually unfold-
ing the nested resonator architecture into a chain of coupled resonators, it is clear
that no matter what is the fine-structure of the unit-cell, the very same unit cell is
repeated (approximately 2σ times, as we have shown in Equation 13.50). Thereby,
The main disadvantage of the nested architecture is that its enhanced group delay is
obtained at the expense of reducing the bandwidth over which the delay peaks. Let us
define 2δω (nested) as the separation between those frequencies at which τFP + GT falls to 50%
of its peak value. Assuming 2δω (nested) is narrow enough so that the Fabry–Perot etalon
remains near its resonance, we can use Equation 13.44 to find
c)
(
L-d
(2δω ( nested )
)×s + τ = 2.
FP (13.51)
bandwidth
half-delay
Using the minimum value of |k| from Equation 13.49, we can re-write Equation 13.41 as
τFP = neff d/(|κ|c) The bandwidth of the coupled-resonator chain is obtained from Equation
13.47 as
2|k|c
2δω ( chain ) = (13.52)
neff L
(although usually only one-half of this bandwidth has approximately linear dispersion.)
Therefore, we find
2δω ( nested ) 1 L
= ≈ if | k | 1. (13.53)
2δω ( chain ) | k | L - d d sd
s +
neff L L
Although this equation seems to suggest that the ratio of the bandwidths could approach
unity if d = L/σ, such a configuration (thin Fabry–Perot etalon in a long Gires–Tournois
resonator) results in the bandwidth of the nested structure being dominated by the Gires–
Tournois resonator, rather than the Fabry–Perot etalon, and considerably reduces the value
of M in Equation 13.50.
However, the bandwidth limitations can be circumvented by spectrally separating the
input wavelengths using a diffraction grating, and using an array of nested-resonator
devices to tune the delay in each narrow passband. Although we end up consuming the
same amount of dielectric material and physical space as the slow-wave chain to regain
the bandwidth, there is no longer any need to maintain phase-lock between the individual
elements, as there would be in an ideal cascaded chain.
The significant attraction of the CROW (cascaded, rather than nested, resonator) archi-
tecture lies then in the new physical phenomena that can be observed in such structures,
such as Anderson localization or Kerr-effect induced (solitonic) localization, which require
a long unfolded optical path length to manifest themselves.
13.6 Summary
Recent research in slow-light has shown that it is difficult to achieve large slowing factors
in practical devices. Each approach has its own limitations: the coupled-resonator opti-
cal waveguide (CROW) architecture is sensitive to fabrication errors which result in mis-
matching of resonances along the chain. This causes dephasing of the transmitted wave,
and variations in the group delay, particularly at the band edges of the dispersion relation-
ship, where light is expected to propagate slowly.
Disordered 1D structures are naturally prone to Anderson localization. Nano-lithographic
tools, particularly the use of electron-beam lithography, allows the fabrication of slow-wave
waveguides in the Mie scattering regime, where the dimension of a unit cell is comparable
to the wavelength of light. Such structures can now be fabricated using materials of high
index contrast, such as silicon and oxide or air, and with surface roughness on the order of
a few nanometers. Most of the slow-wave structures demonstrated can also be fabricated
using UV or deep-UV lithographic tools. Importantly, improvements in optical coupling
techniques to nano-photonic waveguides of sub-micron dimensions, as well as the devel-
opment of experimental methods to measure optical fields in situ have made experimental
studies of optical localization in these waveguides possible.
Localization of light in a slow-wave structure can be investigated in both linear and non-
linear regimes. The former is closely connected with the usual theory of Anderson local-
ization in disordered periodic structures, which has been an area of considerable interest
in the last 20 years. Nonlinear localization can occur either in the perfectly-ordered case,
or from the interplay of disorder and nonlinear self-focusing. The latter is of more practical
interest, but has so far been demonstrated only in the quasi-1D case, where the structure is
not single-moded in the transverse waveguide dimension. Single-mode waveguides are of
more practical interest for applications.
Future studies of localization in slow-wave structures will investigate the role of diffu-
sive processes (which are predicted to be absent in this geometry, but play a signature role
in quasi-1D, 2D and 3D Anderson localization), and measure the time constants associ-
ated with localization. Control over localization, whether by optical, thermal, or electronic
means, is also important. Since these waveguides are viewed as possibly playing the role
of intra-chip interconnects in future micro-processor technologies, it is necessary to mea-
sure their dispersion and polarization properties, and at the same time, minimize both the
absorption and coupling losses by proper design of the input and output transitions.
In summary, research on the coupled-resonator optical slow-wave structure is rapidly
moving forward on both the theoretical and experimental fronts, and involves both basic
physics and fundamental engineering topics. The waveguides display novel linear dis-
persion properties, interesting pulse propagation characteristics, and distinct nonlin-
ear behavior. Using modern-day lithographic technology and commonly-used material
systems which are compatible with the other developing areas of research in on-chip
photonics, we can now fabricate and demonstrate slow-wave structures of considerable
length, and gradually make the transition between laboratory demonstrations and practi-
cal applications.
Acknowledgments
The author is grateful to his students Jung S. Park, Michael L. Cooper, Marco A. Escobar,
and Andrew Oh for assistance, and also to P. R. Bandaru, S. H. Yang, U. Levy, J. B. Khurgin,
V. N. Astratov, J. E. Ford, G. C. Papen, and J. E. Sipe for useful discussions. This work
was funded in part with support from the National Science Foundation (L. Goldberg, R.
Hui and E. G. Johnson). The San Diego Supercomputing Center provided computational
resources.
References
1. P.-E. Wolf and G. Maret. Weak localization and coherent backscattering of photons in disor-
dered media. Phys. Rev. Lett., 55(24): 2696–2699, 1985.
2. D. S. Wiersma, P. Bartolini, A. Lagendijk, and R. Righini. Localization of light in a disordered
medium. Nature, 390: 671–673, 1997.
3. T. Schwartz, G. Bartal, S. Fishman, and M. Segev. Transport and Anderson localization in disor-
dered two-dimensional photonic lattices. Nature, 446: 52–55, 2007.
4. S. John. Strong localization of photons in certain disordered dielectric superlattices. Phys. Rev.
Lett., 58: 2486–2489, 1987.
5. F. M. Izrailev, T. Kottos, A. Politi, and G. P. Tsironis. Evolution of wave packets in quasi-one-
dimensional and one-dimensional random media: Diffusion versus localization. Phys. Rev. E,
55(5): 4951–4963, 1997.
6. A. Melloni, F. Morichetti, and M. Martinelli. Optical slow wave structures. Optics and Photonics
News, 14: 44–48, 2003.
7. M. L. Povinelli, S. G. Johnson, and J. D. Joannopoulos. Slow-light, band-edge waveguides for
tunable time delays. Opt. Express, 13(18): 7145–7159, 2005.
8. S. Mookherjea, J. S. Park, S.-H. Yang, and P. R. Bandaru. Localization in silicon nanophotonic
slow-light waveguides. Nature Photonics, 2: 90–93, 2008.
9. J. Topolancik, B Ilic, and F. Vollmer. Experimental observation of strong photon localization in
disordered photonic crystal waveguides. Phys. Rev. Lett., 99: 253901, 2007.
10. S. Mookherjea and A. Yariv. Kerr-stabilized super-resonant modes in coupled-resonator optical
waveguides. Phys. Rev. E, 66: 046610, 2002.
11. D. N. Christodoulides and N. K. Efremidis. Discrete temporal solitons along a chain of nonlin-
ear coupled microcavities embedded in photonic crystals. Opt. Lett., 27(8): 568–570, 2002.
12. Y. Lahini, A. Avidan, F. Pozzi, M. Sorel, R. Morandotti, D. N. Christodoulides, and Y. Silberberg.
Anderson localization and nonlinearity in one-dimensional disordered photonic lattices. Phys.
Rev. Lett., 100: 013906, 2008.
13. A. Yariv, Y. Xu, R. K. Lee, and A. Scherer. Coupled-resonator optical waveguide: a proposal and
analysis. Opt. Lett., 24(11): 711–713, 1999.
39. Y.-H. Ye, J. Ding, D.-Y. Jeong, I. C. Khoo, and Q. M. Zhang. Finite-size effects on one-dimensional
coupled-resonator optical waveguides. Phys. Rev. E, 69: 056604, 2004.
40. J. K. S. Poon and A. Yariv. Active coupled-resonator optical waveguides—part I: Gain enhance-
ment and noise. J. Opt. Soc. Am. B, 24(9): 2378–2388, 2007.
41. G. T. Paloczi, Y. Huang, A. Yariv, and S. Mookherjea. Polymeric Mach-Zehnder interferometer
using serially coupled microring resonators. Optics Express, 11: 2666–2671, 2003.
42. M. F. Yanik and S. Fan. Stopping light all optically. Phys. Rev. Lett., 92: 083901, 2004.
43. A.Yariv. Optical Electronics in Modern Communications. 5th Edition. Oxford, New York, 1997.