Anda di halaman 1dari 114

NORTHWESTERN UNIVERSITY

FDTD Computational Electromagnetics Modeling of Microcavity


Lasers and Resonant Optical Structures
A DISSERTATION
SUBMITTED TO THE GRADUATE SCHOOL
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
for the degree
DOCTOR OF PHILOSOPHY
Field of Electrical Engineering

By

Susan Carol Hagness


EVANSTON, ILLINOIS
June 1998

c Copyright by Susan C. Hagness 1998


All Rights Reserved

ii

ABSTRACT

FDTD Computational Electromagnetics Modeling of Microcavity


Lasers and Resonant Optical Structures
Susan C. Hagness
Recent advances in materials technology and fabrication techniques have made
it feasible to consider optical resonators having physical dimensions of the order of
the optical wavelength. Microcavity resonators have electromagnetic wave transport phenomena as a critical operating factor. A direct result of the progress in
nanofabrication techniques is the increased need for accurate models of the electromagnetic eld behavior of these novel structures. Numerical simulations provide a
framework for quick low-cost feasibility studies and allow for design optimization
before devices are fabricated. Furthermore, accurate computations can provide
a detailed understanding of the complex physical phenomena inherent in optical
microcavities.
This dissertation addresses the numerical modeling of microcavity lasers and
resonators directly from the time-dependent Maxwell's equations. The primary
goals of this research are to develop time-domain numerical algorithms for modeling
electromagnetic wave interactions with linear and nonlinear optical gain media and
to characterize and optimize the design of novel microcavity resonators. For this
research, the nite-di erence time-domain (FDTD) method serves as the basis for
the computational studies and algorithm development. The devices of interest here
iii

present novel geometries and material composition varying on sub-micron scales


as well as nonlinear dynamics, all of which are amenable to FDTD analysis.
Towards these goals, an algorithm for modeling frequency-dependent optical
gain media with saturation e ects has been developed and rigorously validated.
This algorithm is used to model vertical-cavity surface-emitting lasers. A comparison between uniform-gain and periodic-gain con gurations within the cavity
illustrates the advantage of the periodic-gain structure for providing lower lasing
thresholds and higher e ciencies. FDTD simulations are also used to characterize
key design parameters of waveguide-coupled microcavity ring and disk resonators.
These compact high-Q resonators o er a wide spacing between the longitudinal
mode resonances and may serve as useful components in future high-density photonic integrated circuits associated with optical communications, computing, and
signal processing.

iv

Contents
1 Introduction

2 Overview of Microcavity Resonators

2.1 Microcavity Resonators as Passive Components . . . . . . . . . . . 9


2.2 Microcavity Resonators as Active Components: Microlasers . . . . . 10

3 Methods in Computational Electromagnetics

12

3.1 Maxwell's Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 12


3.2 Methods Based on Simpli ed Wave Equations . . . . . . . . . . . . 14
3.3 Full-Wave Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4 FDTD Modeling of Optical Waveguides

18

4.1 Waveguide Termination . . . . . . . . . . . . . . . . . . . . . . . . . 18


4.2 Waveguide Source Condition . . . . . . . . . . . . . . . . . . . . . . 19
4.3 Validation: Calculation of the Propagation Constant . . . . . . . . 22

5 Wave Propagation in Optical Media

26

5.1 Advances in Modeling Linear and Nonlinear Frequency-Dependent


Optical Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.2 FDTD Algorithm for Gain Media . . . . . . . . . . . . . . . . . . . 30
5.3 Validations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

6 FDTD Modeling of Vertical Microcavity Lasers

46

6.1 Passive studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49


6.2 Active studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
vii

7 FDTD Modeling of Ring and Disk Resonators


7.1
7.2
7.3
7.4
7.5
7.6
7.7

FDTD Modeling Considerations . . . . . . . . . . . . . . . . . .


Ring Resonators Coupled to Straight Waveguides . . . . . . . .
Ring Resonator Coupled to Curved Waveguides . . . . . . . . .
Elongated Ring Designs . . . . . . . . . . . . . . . . . . . . . .
Microcavity Ring Resonances . . . . . . . . . . . . . . . . . . .
Microcavity Disk Resonances . . . . . . . . . . . . . . . . . . .
Suppression of Higher-Order Radial Whispering-Gallery Modes .

8 Conclusions and Future Research

57
.
.
.
.
.
.
.

.
.
.
.
.
.
.

57
61
65
67
72
79
85

92

8.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
8.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

References

94

Appendix: Additional Doctoral Research

103

Vita

124

viii

List of Figures
1
2
3

4
5
6

Scanning electron microscope image of a small portion of a chip of


GaAs/AlAs VCSELs with diameters ranging from 1 to 5 m. . . . .
Scanning electron microscope image of GaAs/AlAs microcavity
VCSELs with aspect ratios of 20:1. . . . . . . . . . . . . . . . . . .
Scanning electron microscope image of an InGaAsP/InGaAs suspended photonic wire microcavity. Holes etched in the waveguide
from 1-D photonic bandgap structures. . . . . . . . . . . . . . . . .
Scanning electron microscope image of 10.5- m-diameter AlGaAs/
GaAs disk and ring resonators coupled to 0.5- m-wide waveguides.
Cross-section of a dielectric slab waveguide with electric and magnetic eld components shown for a TE waveguide mode. . . . . . .
Excitation spectrum (dotted line) superimposed on the waveguide
dispersion relations (solid lines) for the rst three TE modes of a
symmetric strongly guiding dielectric slab waveguide. . . . . . . . .
Comparison of FDTD-computed and theoretical results for the (a)
longitudinal propagation constant and (b) group velocity vg of the
fundamental TE mode propagating in the strongly guiding dielectric
slab planar waveguide described by Figures 5 and 6. . . . . . . . . .
Comparison of FDTD results and theory for the ampli cation and
phase factors of a pulse propagating a distance of one dielectric
wavelength in a linear gain medium with o = ;5000 S/m. . . . . .
Unsaturated gain and loss spectra for a Fabry-Perot microcavity
laser. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ix

5
6

7
8
20

21

25

39
41

10 Histogram of the rst 10,000 outputs of a pseudorandom number


generator for a Gaussian random number with mean = 0:0 and
variance = x2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11 (a) FDTD-computed time evolution of the electric eld outside the
Fabry-Perot laser cavity. (b) Expanded timescale of the steady-state
region showing a single-mode oscillation at !o. . . . . . . . . . . . .
12 FDTD-computed L-I curve for the Fabry-Perot laser, showing the
dependence of the output intensity on the peak of the gain curve. .
13 Schematic of microcavity lasers having vertical distributed Bragg reectors and uniform (right) or periodic (left) gain structures within
the cavity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14 FDTD-computed re ectivity and phase of the re ection coe cient
for the (a) bottom and (b) top mirrors of the vertical microcavity. .
15 FDTD-computed re ectivity spectrum for the vertical microcavity. .
16 Superposition of the unsaturated gain spectrum of GaAs for three
values of o , the loss characteristic of the mirrors (from Fig. 14),
and the cold-cavity resonances (from Fig. 15). . . . . . . . . . . . .
17 Spatial variation of the refractive index (top) and optical gain (middle and bottom) within the VCSELs with uniform and periodic gain
con gurations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18 Normalized steady-state electric eld envelope within the VCSEL. .
19 FDTD-computed output intensity vs. gain for VCSELs comprised
of either a uniform-gain structure (UGS) or a periodic-gain structure
(PGS). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
x

42

43
44

47
50
51

52

54
54

55

20 Schematic of a microcavity ring or disk resonator coupled to two


straight waveguides. . . . . . . . . . . . . . . . . . . . . . . . . . .
21 FDTD-computed coupling coe cient, , as a function of frequency
and gap size, g, for the geometry of Figure 20, (5.0- m-diameter
ring resonator case). The electric eld is vertically polarized, i.e.,
perpendicular to the plane of Figure 20. . . . . . . . . . . . . . . .
22 Comparison of the FDTD-computed coupling coe cient for the
V-pol and H-pol cases of Figure 20 (ring) with d = 5:0 m and
g = 0:232 m. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23 V-pol FDTD-computed coupling coe cient, , as a function of frequency and gap size, g, for the geometry of Figure 20, (10.0- mdiameter ring resonator case). . . . . . . . . . . . . . . . . . . . . .
24 Schematic diagram of a microcavity ring resonator coupled to a
curved waveguide. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25 FDTD-computed coupling coe cient as a function of frequency
and waveguide bend radius, , for a 5.0- m-diameter ring ( g =
0:205 m, V-pol case). . . . . . . . . . . . . . . . . . . . . . . . . .
26 Schematic of a microcavity ring resonator (elliptical con guration)
coupled to straight waveguides. . . . . . . . . . . . . . . . . . . . .
27 FDTD-computed coupling coe cient (V-pol case) as a function
of frequency and semimajor axis, a, for an elliptical ring resonator
coupled to straight waveguides (b = 2:5 m, g = 0:3 m). . . . . . .

xi

58

62

63

64
66

67
68

69

28 FDTD-computed coupling coe cient as a function of wavelength


and the gap width, g, for a 5.0- m-diameter circular ring resonator
(dotted lines) and an elliptical ring resonator with semimajor and
semiminor axis lengths of a = 6:0 m and b = 2:5 m, respectively
(solid lines). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29 Schematic of a microcavity ring resonator (racetrack con guration)
coupled to straight waveguides. . . . . . . . . . . . . . . . . . . . .
30 FDTD-computed coupling coe cient (V-pol case) as a function
of frequency and straight-section length, L, for a racetrack ring resonator coupled to straight waveguides (d = 5:0 m, g = 0:3 m). . .
31 FDTD-computed coupling coe cient as a function of wavelength
and the gap width, g, for a 5.0- m-diameter circular ring resonator
(dotted lines) and a racetrack ring resonator with 5.0- m-diameter
semi-circular sections and 1.5- m-long straight sections (solid lines).
32 Visualization of the FDTD-computed initial coupling and circulation of the exciting 20-fs optical carrier pulse around the 5.0- mdiameter microcavity ring resonator. . . . . . . . . . . . . . . . . .
33 FDTD-computed transmittance spectrum of the 5.0- m-diameter
microcavity ring resonator. . . . . . . . . . . . . . . . . . . . . . . .
34 Visualizations of the FDTD-computed sinusoidal steady-state electric eld in the 5.0- m-diameter ring having the properties summarized in Figures 32, 33, and Table 2. . . . . . . . . . . . . . . . . . .

xii

70
71

72

73

74
75

78

35 Visualization of the FDTD-computed initial coupling and circulation of the exciting 20-fs optical carrier pulse around the 5.0- mdiameter microcavity disk resonator. . . . . . . . . . . . . . . . . .
36 FDTD-computed transmittance spectrum of the 5.0- m-diameter
microcavity disk resonator. . . . . . . . . . . . . . . . . . . . . . . .
37 Visualizations of the FDTD-computed sinusoidal steady-state electric eld in the 5.0- m-diameter disk resonator coupled to straight
0.3- m-wide waveguides for single-frequency excitations at port A
of WG1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
38 Progression of the FDTD-computed transmittance of a 5.0- mdiameter microcavity. . . . . . . . . . . . . . . . . . . . . . . . . . .
39 FDTD-computed transmittance of a 5.0- m-diameter microcavity
disk resonator coupled to straight waveguides of the following widths:
(a) 0.20 m (b) 0.30 m (c) 0.35 m and (d) 0.38 m. . . . . . .

xiii

80
81

84
87

90

List of Tables
1
2
3

Refractive indices used in the FDTD models of the VCSELs of


Fig. 13. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Resonance data from Figure 33 for the 5.0- m-diameter microcavity
ring resonator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
Resonance data from Figure 36 for the 5.0- m-diameter microcavity
disk resonator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

xiv

1 Introduction
Several decades ago, the concept of \integrated optics" was introduced 1]. In a
photonic integrated circuit (PIC), active and passive optical devices are interconnected by optical waveguides, in a manner similar to the wire interconnections between transistors in electronic integrated circuits. While there has been signi cant
development towards the realization of PICs, the miniaturization and integration
of optical components and devices has lagged considerably in comparison to the
achievements of large-scale integration in microelectronics.
Recent advances in materials technology and fabrication techniques have made
it feasible to consider optical resonators having physical dimensions of the order
of the optical wavelength (\microcavities"). Microcavity resonators may serve as
versatile passive and active components in photonic integrated circuits. Interest
in optical microcavities has been stimulated by the desire to achieve higher performance and higher-density integration for future devices associated with optical
communications, computing, and signal processing. For photonic devices to ever
succeed in complementing or replacing their microelectronic counterparts in largescale integrated circuits, novel optical microstructures must be explored.
Microcavity resonators have electromagnetic wave transport phenomena as a
critical operating factor. A direct result of the progress in nanofabrication techniques is the increased need for accurate models of the electromagnetic eld behavior of these novel optical devices. Numerical simulations provide a framework for
quick low-cost feasibility studies and allow for design optimization before devices
are fabricated. Furthermore, accurate computations can provide a detailed understanding of the complex physical phenomena inherent in optical microcavities.
1

2
This dissertation addresses the numerical modeling of microcavity lasers and
resonators directly from the time-dependent Maxwell's equations. The primary
goals of this research are to investigate electromagnetic wave interactions with linear and nonlinear optical gain media, to characterize and optimize the design of
novel microcavity resonators, and to develop time-domain numerical algorithms
to rigorously describe the physics of interest. Presently, the most commonly used
numerical methods in optics are limited in their applicability due to the approximations involved in reducing Maxwell's equations to simpli ed wave equations.
In order to implement reliable electrodynamics models of the micron-scale optical devices of interest here, a time-domain full-wave Maxwell's equations solver
is required. For this research, the nite-di erence time-domain (FDTD) method
has served as the basis for the computational studies and algorithm development.
Since its introduction in 1966 2], the FDTD method has been re ned and extended to e ciently treat a wide range of problems involving the interaction of
electromagnetic waves with complicated structures and materials. The devices of
interest here present novel geometries and material composition varying on submicron scales as well as nonlinear dynamics, all of which are amenable to FDTD
analysis.
Chapter 2 provides an overview of microcavity resonators and their advantages for both passive and active applications in optical communications and computing. Chapter 3 reviews a variety of numerical methods available for solving
Maxwell's equations and highlights the capabilities of the nite-di erence timedomain (FDTD) method. In Chapter 4, the basic issues related to modeling optical waveguides are discussed. Chapter 5 provides a discussion of the relevant

3
physics of optical media, and reviews the advances in FDTD algorithms for the
modeling of linear and nonlinear interactions of light with optical materials. An
algorithm for modeling frequency-dependent optical gain media with saturation effects is presented and validated. In Chapter 6, the FDTD algorithm for gain media
is used to model vertical-cavity surface-emitting lasers (VCSELs). A comparison
is made between uniform-gain and periodic-gain con gurations within the VCSEL.
Chapter 7 focuses on microcavity ring and disk resonators that are coupled to adjacent input/output waveguides via the evanescent eld. Key design parameters
are characterized using FDTD simulations. A summary of results and a discussion
of future research is presented in Chapter 8.

2 Overview of Microcavity Resonators


Microcavity structures can be categorized into two groups based on the physical
mechanism by which the cavity con nes the optical mode 3]:
Group 1: Con nement by multiple re ections, as in vertical-cavity
surface-emitting lasers (VCSELs) and photonic crystals and
Group 2: Con nement by total internal re ection, as in whisperinggallery-mode disk and single-mode ring cavities.
Since the rst demonstration in 1979 4], VCSELs have received much attention primarily due to interest in the development of high-density laser arrays in
optical communications and optical interconnect applications. In contrast to a
conventional edge-emitting diode laser, the light is emitted from the VCSEL perpendicular to the wafer substrate. In simplest terms, a VCSEL consists of a semiconductor laser diode sandwiched vertically between two highly re ective mirrors.
The mirrors usually consist of either dielectric or semiconductor distributed Bragg
re ectors (DBRs). In this context, a DBR is a multi-layer stack of alternating
materials where the thickness of each layer is =4. In the center of the cavity,
between the mirrors, is a either a bulk gain region or one or more quantum wells.
Using molecular-beam epitaxy, a wafer of VCSELs is grown layer by layer in
the vertical direction. This creates a planar one-dimensional cavity without lateral
guiding. A three-dimensional microcavity structure can be fabricated by means of
deep etching using an ion-assisted process 5, 6]. By 1989, more than one million
pillar-like microlasers of this type were fabricated on a single GaAs semiconductor
chip 7], as shown in Figure 1. Figure 2 shows a scanning electron microscope
4

5
(SEM) image of ultra-small (0.4- m-wide, 8- m-high) microcavity VCSELs 8].

Figure 1: Scanning electron microscope image of a small portion of a chip of


GaAs/AlAs VCSELs with diameters ranging from 1 to 5 m 7].
Photonic crystals 9] are arti cial structures that have a periodic variation of
the refractive index in one, two, or three dimensions. Analogous to the energy
gap in pure semiconductor crystals in which electrons are forbidden, these photonic bandgap (PBG) structures have a frequency stopband over which there is no
transmission of electromagnetic waves 10]. Similar to a donor or acceptor state in
a doped semiconductor, a small defect introduced into the photonic crystal creates
a resonant mode at a frequency that lies inside the bandgap 11]. The defect in
the periodic array behaves as a microcavity resonator.
A novel class of coplanar microcavities is based on a strongly guiding semiconductor waveguide that con nes the light in the transverse dimensions and a

Figure 2: Scanning electron microscope image of GaAs/AlAs microcavity VCSELs


with aspect ratios of 20:1 8].
one-dimensional PBG structure that con nes the light in the longitudinal dimension. Figure 3 is a SEM image of one such structure, an \air-bridge" 12] or
\photonic-wire" 13] microcavity consisting of a series of etched air holes along a
semiconductor waveguide suspended in air. The microcavity of Figure 3 is comprised of a 0.45 0.79 m InGaAsP/InGaAs waveguide with a defect introduced
into a linear array of 40 etched holes.
Ring and disk resonators based on high-index-contrast, strongly guiding waveguides can be designed with diameters as small as 1 to 2 m with negligible bending
loss 14]. Microdisk 15] and microcylinder 16] lasers make use of high-re ectivity
whispering-gallery modes 17]. Microdisk lasers have been fabricated from a wide
variety of material systems the common design feature is that the disk is supported

Figure 3: Scanning electron microscope image of an InGaAsP/InGaAs suspended


photonic wire microcavity. Holes etched in the waveguide from 1-D photonic
bandgap structures 13].
on a smaller-diameter semiconductor pedestal (for example, see 18, 19, 20, 21]).
The light is emitted in the plane of the disk as well as vertically due to the presence
of the pedestal 22]. Several methods have been proposed for the directional coupling of light output horizontally from these lasers, including the introduction of an
asymmetry in the shape of the microdisk resonator 23] and the use of a double-disk
structure 24]. Several groups have demonstrated strongly-guiding semiconductor
ring lasers 25, 26, 27, 28] with diameters in the range of 10-500 m and large
waveguide widths of 4-20 m. To obtain light output, these ring resonators are
coupled to output guides via Y junctions.

8
Most recently, nanofabrication techniques have been used to create semiconductor microcavity ring and disk resonators that are evanescently coupled to submicron-width waveguides across sub-micron-width air gaps. One such example is
a 10- m-diameter photonic-wire ring laser coupled to a U-shaped waveguide 29].
Figure 4 shows SEM images of AlGaAs/GaAs microcavity ring and disk resonators
coupled to straight waveguides 30]. The air gap between the adjacent waveguides
and the ring or disk is only 100 nm. Further discussion of these types of resonators
is presented in Chapter 7.

Figure 4: Scanning electron microscope image of 10.5- m-diameter AlGaAs/GaAs


disk and ring resonators coupled to 0.5- m-wide waveguides 30].

2.1 Microcavity Resonators as Passive Components

Passive microcavity resonators are useful components for wavelength ltering, routing, switching, modulation, and multiplexing/demultiplexing applications. The
quality of a resonator for optical communications is measured by several properties including the following:
1. the free-spectral range (the spacing between adjacent resonances)
2. the quality factor (a measure of the e ectiveness of the resonator in storing
energy)
3. the nesse (the ratio of the FSR to the width of the resonance) and
4. the extinction ratio (the ratio of the transmission at resonance to the o resonance transmission).
The ideal resonator for a wavelength-division multiplexing (WDM) system has a
wide free spectral range (FSR), a high quality factor (Q), and a large nesse to accommodate many channels, high on-resonance transmission to minimize insertion
loss, and a large extinction ratio to minimize crosstalk. In particular, a ring or
disk resonator having an FSR larger than 30 nm is desirable for accommodating
the WDM channels within the erbium-ampli er communications window.
The Fabry-Perot etalon is the simplest form of an optical resonator cavity.
Using Fabry-Perot resonators, tuning ranges of 3 nm have been reported 31]. A
variety of circular resonators have also been explored for applications in optical
communications. Ring resonators based on waveguide structures with weak lateral
con nement have very low propagation loss 32, 33] however, the small refractive

10
index contrast of the weakly guiding structures leads to high radiative bending
losses at ring/disk diameters below 1 mm. Therefore, the FSR, which is inversely
proportional to the diameter, has been limited to 26 GHz (0.15 nm at 1.3 m) 34]
for a single ring and 100 GHz (0.8 nm at 1.55 m) 35] for a double ring resonator.
Various techniques have been reported for increasing the FSR without decreasing
the diameter, including a triple-coupler ring resonator that o ers twice the FSR
of the double ring resonator with the same radius of curvature 36].
To achieve a FSR on the order of tens of nanometers, the size of the resonator
cavity must be on the order of just a few micrometers. Microcavity resonators,
such as those shown in Figures 3 and 4, o er FSRs that are more than an order-ofmagnitude wider than those previously achieved with larger weakly-guiding ring
resonators. With high-quality etching, the scattering losses can be kept low enough
to achieve simultaneously a high nesse. Additionally, these resonators have the
potential for high-density integration for example, microcavity resonators can be
integrated with low-threshold photonic-wire microcavity lasers 29] to form photonic integrated circuits. This ease in integration represents a major advantage
over Fabry-Perot type resonators that require cleaved end facets. Potential applications of microcavity resonators include micron-size channel-dropping lters 37],
frequency-domain switches, and intensity/phase modulators.

2.2 Microcavity Resonators as Active Components: Microlasers


The conventional semiconductor diode laser has revolutionized the storage and
communication of information. Even so, the miniaturization and integration of
the conventional device has lagged behind its microelectronic counterparts. The

11
dimensions of the optical cavity of a typical diode laser greatly exceed the photon
wavelength. Furthermore, the several-hundred-microns-long cavity is formed from
a waveguide with cleaved facets, which limits its large-scale integration onto chips.
Distributed feedback and distributed Bragg re ector lasers have been studied extensively for monolithic integration because the resonators can be made without
cleaved facets.
More recently, signi cant advances in fabrication techniques have led to the
miniaturization of semiconductor lasers to the point where the cavity length is
comparable to the emitted radiation wavelength. The microcavity laser not only
allows for high-density integration, but also has the potential for realizing the
ultimate performance of a semiconductor laser through the control of spontaneous
emission. The spontaneous emission coupling e ciency sp, de ned as the rate of
spontaneous emission into the lasing mode divided by the total rate of spontaneous
emission 38], approaches unity in an ideal microcavity laser. In contrast, sp
0:001% in a conventional semiconductor diode laser. An increase in sp reduces
the lasing threshold and, via a corresponding increase in the stimulated decay rate,
increases the modulation bandwidth 39]. Using the classical e ect of reducing
the number of modes interacting with the active material 40], sp can be greatly
increased by decreasing the cavity length to the point where the longitudinal cavity
mode spacing is larger than the gain bandwidth of the active material.

3 Methods in Computational Electromagnetics


A variety of numerical methods are currently in use for modeling the propagation
of electromagnetic waves in optical media. The methods available o er a wide
range of modeling capabilities their accuracy depends on the type of approximations used to simplify the governing Maxwell's equations, and how valid those
approximations are for the particular device being modeled. This chapter presents
a survey of numerical methods for solving Maxwell's equations. The underlying
conclusion is that the nite-di erence time-domain (FDTD) technique o ers the
most appropriate and accurate level of analysis for micron-scale photonic and optoelectronic devices and systems.

3.1 Maxwell's Equations


Classical electromagnetic theory is based on Maxwell's equations. The timedependent equations in di erential form are (in MKS units):
r E~ = ; @@tB~
(1)
(2)
r H~ = J~ + @@tD~
r D~ =
(3)

B~ = 0

(4)

where E~ and H~ are the electric and magnetic eld vectors, respectively, J~ is the
current density vector, and is the charge density. The electric and magnetic ux
densities, D~ and B~ , arise in response to the interaction between the elds and the
medium. The constitutive equations

J~ = E~
12

(5)

13

D~ = o E~ + P~

(6)

B~ = oH~ + M~

(7)

relate the electromagnetic elds to the macroscopic parameters of the medium.


Here, o and o are the permittivity and permeability of free space and is the
electric conductivity. The polarization P~ and magnetization M~ are the dipole
moments per unit volume of the medium. It should be noted that the time domain
equations (5), (6), and (7) assume that the medium response is instantaneous, that
is, that the material parameters are independent of frequency. This is not always
a valid assumption, as discussed in Chapter 5.
Semi-classical theory treats the electromagnetic eld as classical radiation that
interacts with a quantized atomic system 41]. The electromagnetic eld induces
an electric dipole moment in the material. The polarization vector P~ is obtained
from the quantum theory of the atomic system that provides the gain mechanism.
P~ is in turn used in the classical Maxwell's equations to understand the e ect of
the gain medium on the eld.
The physics of the semiconductor material can be described by the phenomenological polarization of injected electron-hole pairs. The transfer of energy from the
polarization of the electron-hole pairs to the electric eld, and the reverse process,
are the phenomena of stimulated emission and absorption, respectively. This semiclassical approach is su cient to describe the gain mechanism of a laser medium.
However, the noise properties due to spontaneous emission require the fully quantized theory and are consequently lost.

3.2 Methods Based on Simpli ed Wave Equations

14

Two popular methods based on simpli ed wave equations are the coupled mode
theory (CMT) 42, 43] and the beam propagation method (BPM) 44]. CMT is
an envelope-function approach that has been successfully applied to simple optical ber and waveguide geometries where the coupling can be represented as a
distributed perturbation polarization source for example, the approximations of
the standard CMT are valid for the distributed feedback laser 45] where the grating is weak. BPM, a frequency-domain approach, is one of the most widely-used
methods to simulate guided-wave propagation in integrated photonic devices, such
as Y -junction semiconductor rib waveguides 46]. In its original formulation, the
BPM solves the paraxial wave equation under the scalar approximation, providing a spectral propagation algorithm for an arbitrary incident beam propagating
in a medium of slowly-varying refractive index. The standard BPM has several
fundamental limitations 47] which restrict its applicability to problems in which
the following assumptions are valid: paraxial propagation, small contrast in the
dielectric structure, negligible wave polarization, and most importantly, negligible
re ected waves. In recent years, considerable e ort has been made toward extending the CMT and BPM beyond their limitations. Nevertheless, optical wave
radiation, re ection, and di raction e ects relevant to the design of wavelengthscale structures are di cult or impossible to obtain with these methods.

3.3 Full-Wave Methods


A full-wave Maxwell's equations solver is required for modeling complex geometrical features and inhomogeneous materials. Until recently, frequency-domain

15
matrix-based methods, including the method of moments (MoM) 48] and nite
element methods (FEM) 49], had been the most widely-used for obtaining detailed
full-wave, full-vector models of electromagnetic wave interactions. Although these
methods have mostly been used to study scattering and radiation in traditional
electromagnetics problems, a few recent applications to the study of integrated
optics have been reported 50, 51]. The boundary-integral formulation of MoM requires the di cult derivation of a geometry-speci c Green's function and the computationally di cult solution of a dense system of linear equations. The partial
di erential equation formulation of FEM leads to sparse matrices and a relatively
simple mesh generation for the structure to be modeled.
Full-wave, full-vector methods in the time domain include the nite-di erence
time-domain (FDTD) method 2, 52] and related nite-volume time-domain algorithms. The FDTD method is an explicit grid-based technique for the direct
solution of the fundamental Maxwell's curl equations. The original Yee algorithm
is a computationally e cient approach for modeling sinusoidal or impulsive electromagnetic wave interactions with arbitrary three-dimensional structures. The
derivation of the basic FDTD equations has been documented extensively elsewhere 53, 54], so it will not be repeated here. The standard algorithm, based on
centered nite di erences for the space and time derivatives, is second-order accurate in space and time. The leapfrog integration scheme marches the discretized
electric and magnetic elds forward in time. In comparison to FEM and MoM,
FDTD involves no computationally burdensome matrix inversions, thereby permitting orders-of-magnitude larger problems to be modeled. FDTD permits transient
electromagnetic excitations and the computation of impulse responses whereas

16
FEM and MoM frequency-domain solutions are limited to sinusoidal steady-state
conditions. Furthermore, the time-domain nature permits the modeling of nonlinear dynamics.
Today, the FDTD method is one of the most popular numerical methods for
several classes of electromagnetics problems 55]. Its many advantages have made
it the method of choice in modeling and designing complicated RF and microwave
circuits with dimensions comparable to the radiation wavelength. In order to
maintain high accuracy and numerical stability at optical frequencies, the grid
cell size and the time step are on the order of tens of nanometers and tens of
attoseconds (10;18 s), respectively. Thus, until recently, most optical devices had
distance and/or time scales that made them inappropriate for FDTD analysis.
However, for the emerging class of micrometer- and nanometer-scale integrated
optical devices, FDTD modeling has the potential to play a useful and practical
design role.
In contrast to the more commonly used numerical methods in traditional integrated optics, FDTD makes no assumptions about the direction of wave propagation or the time-varying nature of the elds. In particular, there are no slowlyvarying envelope approximations, no plane-wave assumptions, and no modal expansions. Since FDTD rigorously enforces boundary conditions at material interfaces, it is capable of modeling the geometrical complexities and material inhomogeneities of realistic micron-scale photonic and optoelectronic devices, including side-wall roughness, curved guiding structures with small bend radii, tapered
waveguides, deep gratings, and novel cavity shapes (rings, disks, cylindrical posts,
defects in photonic crystals, etc.). Consequently, the optical wave phenomena in-

17
herent in photonic structures that are comparable in size to the wavelength are
fully accounted for. Given the advanced state of computing power of workstations and supercomputers, a complete FDTD analysis of the electrodynamics of
microcavity lasers and resonators is indeed computationally feasible.

4 FDTD Modeling of Optical Waveguides


Waveguides proposed for micron-scale integrated optical devices have a submicronwide core layer and a large refractive index contrast between the core and cladding
layers. The strong optical mode con nement that results from such a design leads
to signi cant waveguide dispersion. The relatively strong dependence of the waveguide propagation constant upon frequency impacts several key design parameters
for microcavity resonators composed of such waveguides, including the Q and the
FSR. Accurate FDTD modeling of highly dispersive waveguides requires careful
consideration of two important issues:
1. how to terminate waveguides that extend beyond the grid boundaries and
2. how to source speci c waveguide modes for the case of a pulsed excitation.
This chapter discusses considerations for modeling highly dispersive optical waveguides in the context of FDTD Maxwell's equations solutions.

4.1 Waveguide Termination


The accuracy of the FDTD method for optical waveguiding devices was rst
demonstrated for the directional coupler 56]. Since then, Berenger's \perfectly
matched layer" (PML) 57] absorbing boundary condition (ABC) has been introduced. The PML is a nonphysical absorber adjacent to the outer grid boundary
that permits outgoing waves of arbitrary frequency and direction of propagation
to rapidly decay while maintaining the same velocity and eld impedance of the
medium on the interior side of the grid boundary. The PML ABC was shown to
provide e ective re ection coe cients as low as 1/3000th those of standard ABC's,
18

19
essentially providing a means to terminate 2-D and 3-D FDTD grids without reection 57, 58]. The improved performance of the PML ABC relative to any
earlier technique has provided the increased dynamic range required for numerical simulations such as the analysis of antire ection coatings 59] or resistively
loaded antennas 60]. Of particular interest here is the use of the PML to terminate waveguides extending beyond the FDTD grid boundaries. Reuter et al. 61]
demonstrated that multimodal and dispersive waveguiding structures extending
beyond the edges of the conventional FDTD grid can be terminated accurately by
the PML, with re ection coe cients less than -75 dB, for a wide range of group
velocities.

4.2 Waveguide Source Condition


Consider a two-dimensional problem where the z-directed electric eld is normal to
the x-y plane of the grid. Any structure being modeled is assumed to be of in nite
extent in the z-direction. Maxwell's equations for the electric and magnetic elds,
Ez , Hx, and Hy in a linear nondispersive nonmagnetic material are given by

@Hx = ; 1 @Ez
(8)
@t
o @y
@Hy = 1 @Ez
(9)
@t
o @x
"
#
@Ez = 1 @Hy ; @Hx ; E :
(10)
z
@t
@x
@y
This is a transverse electric (TE) optical problem where there is no electric eld
in the direction of propagation. Figure 5 shows the cross-section of a three-layer
dielectric slab waveguide and the eld components associated with the TE wave.

20
n1
Hy
Ez

Hx

n2

direction
of propagation

n3

Figure 5: Cross-section of a dielectric slab waveguide with electric and magnetic


eld components shown for a TE waveguide mode.
Sourcing the numerical analog of a propagating TE waveguide mode is accomplished by specifying an electric eld distribution along a one-dimensional transverse cross section of the waveguide. For the case of sinusoidal excitation, the
transverse eld distribution of the desired guided mode is obtained by numerically
solving a determinantal equation for the transverse propagation constant in the
core region and decay constants in the cladding regions. For example, given the
refractive indices of the three layers of the waveguide (n1 n2 n3) and the width of
the core (w), the dispersion relations for the various waveguide modes can be computed, as shown by Kogelnik and Ramaswamy 62]. With this data, the normalized
longitudinal propagation constant corresponding to the desired frequency of excitation and mode number is selected. The e ective refractive index is calculated
from the normalized propagation constant then, given the e ective index, the longitudinal propagation constant is calculated. Finally, the transverse propagation
constant and decay constants are calculated directly from .

21
The case of a pulsed source, such as a Gaussian pulse modulating a carrier wave,
poses more of a challenge. Even if the frequency spectrum of the pulse is narrow
enough and the carrier frequency is low enough to allow only the fundamental
mode to propagate, there still exists a spread of values along the fundamental
mode dispersion curve corresponding to the spread of frequencies of interest, as
illustrated in Figure 6 for the slab waveguide of Figure 5 with n1 = 1:0, n2 = 3:2,
n3 = 1:0, and w = 0:3 m.

normalized propagation constant

1.0
m=0
0.8

0.6
m=1
0.4

0.2
m=2
0.0
0.0

1.0

2.0
14
frequency (10 Hz)

3.0

4.0

Figure 6: Excitation spectrum (dotted line) superimposed on the waveguide dispersion relations (solid lines) for the rst three TE modes of a symmetric strongly
guiding dielectric slab waveguide. The refractive indices of the cladding and 0.3m-wide core are 1.0 and 3.2, respectively.
In this case, the transverse mode pro le for the pulse can be approximated
using the value of at the carrier frequency. The fact that transverse distribution

22
of the source is not exact for all frequencies in the pulse spectrum means that a
small amount of energy will be shed from the waveguide core early on in the FDTD
simulation. The accuracy of simulations in which the waveguide is immediately
surrounded by the PML ABC is not be a ected by the shed energy. However,
Chapter 7 presents examples in which the shed energy from the waveguide would
interact with other structures in the grid, thereby introducing unwanted noise into
the computation.
An e ective solution to this problem is to use \bootstrapping", that is, to
perform a preliminary FDTD simulation of a su ciently long version of the waveguide. During this preliminary simulation, the energy shed from the waveguide is
absorbed by the PML. The time history of the transverse electric eld distribution
is stored as the pulse passes an observation plane at the far end of the grid where
the exact mode has been established. The stored data are used as the waveguide
source in subsequent simulations involving that waveguide. The level of numerical
noise introduced into the grid from the bootstrapped source falls well below the
noise level arising from re ections at the PML-grid boundaries. Therefore, for all
practical purposes, this method serves as a perfect pulsed source condition.

4.3 Validation: Calculation of the Propagation Constant


The accuracy of the FDTD method for modeling propagating waveguide modes
is demonstrated through the computation of and the group velocity, vg . The
longitudinal propagation constant is formally de ned as
= +j

(11)

23
where is the eld attenuation constant and is the phase constant. In situations
where there is no attenuation ( = 0), is often called the propagation constant.
The group velocity is de ned as
vg = d!
(12)
d :
The strongly guiding planar waveguide in this study consists of a 0.3- m-wide
core with a high index of refraction (n = 3:2), such as that of a semiconductor
material, surrounded by cladding of a low refractive index (n = 1) on both sides,
such as that of air. The PML ABC is used to terminate the waveguide in the FDTD
grid therefore, the waveguide is successfully simulated as extending beyond the
edges of the FDTD grid. The PML is also used to terminate the cladding layers
on either side of the waveguide. Since the only structure in this simulation is the
straight waveguide, the bootstrapped source is not required. The inexact mode
pro le for the pulsed source of the waveguide is used any energy shed from the
waveguide is directly absorbed by the PML. The waveguide must simply be made
long enough to ensure that the mode has settled down before any data is recorded.
A single Gaussian pulse with a 20-fs temporal width between the 1=e points,
modulating a carrier frequency of 0 = 2:0 1014 Hz, is launched in the waveguide
at one end of the grid using the fundamental (m = 0) mode pro le computed for
the carrier frequency. Figure 6 shows the source spectrum superimposed on the
normalized waveguide dispersion curves for the rst three TE modes of waveguide.
At 0 , this waveguide can support only the m = 0 mode, which has even symmetry
about the longitudinal axis of the waveguide, and the m = 1 mode, which has odd
symmetry. Because the numerical excitation has precisely even symmetry, only
the m = 0 mode is generated by the impulsive source. At the end of the grid,

24
two xed observation points located along the longitudinal axis of the core record
electric- eld data as the pulse passes. By taking the ratio of the discrete Fourier
transforms (DFTs) of the two time histories, is computed over the full bandwidth
of the pulse.
In Figure 7a, the FDTD-computed propagation constant as a function of frequency is compared with the exact denormalized dispersion curve. The group
velocity is computed from the numerical data for using a second-order-accurate
central di erence approximation to the derivative the results for vg are shown in
Figure 7b. The error in at 0 is less than two parts per thousand at a grid resolution of dx = 13:6 nm the results at this resolution have graphically converged to
the exact curve. The error in vg at the same resolution is less than four parts per
thousand. At twice the grid resolution (dx = 6:8 nm), the numerical results for vg
converge to the exact curve the error drops to less than one part per thousand.

25
2.0e+07
1

propagation constant (m )

(a)
1.5e+07

1.0e+07

theory
FDTD

5.0e+06

0.0e+00
1.0

1.5

2.0
2.5
14
frequency (10 Hz)

3.0

1.0e+08

group velocity (m/s)

(b)

theory
FDTD

9.5e+07

9.0e+07

8.5e+07

8.0e+07
1.0

1.5

2.0
2.5
14
frequency (10 Hz)

3.0

Figure 7: Comparison of FDTD-computed and theoretical results for the (a) longitudinal propagation constant and (b) group velocity vg of the fundamental
TE mode propagating in the strongly guiding dielectric slab planar waveguide
described by Figures 5 and 6.

5 Wave Propagation in Optical Media


In general, the macroscopic parameters describing a medium of propagation are
dependent on frequency, due to the variety of resonances (rotational, vibrational,
ionic, electronic) that are available to couple to the electromagnetic eld. If the
excitation pulse width, t, is much longer than the nite response time, , of the
medium, such as in the limit of single-frequency sinusoidal excitation, then the
medium can be modeled as having an instantaneous response time { that is, the
medium parameters can be treated as frequency-independent constants. On the
other hand, if t
, such as for an ultra-short pulse excitation, then the medium
response lags signi cantly behind the pulse therefore, when the medium response
is nite, the frequency-dependence of the medium parameters must be included.
At relatively low light intensities, the optical properties of dielectrics and semiconductors do not depend on the illumination intensity. In this case, the polarization depends linearly upon the strength of the electric eld and can be written
as
P~ (!) = o (1)(!)E~(!)
(13)
where (1) is the linear susceptibility. The response of a material to very intense
light may become nonlinear. In many cases when the excitation frequencies are
far from a medium resonance, the intensity-dependent polarization can be written
as a power series expansion as follows:

P~ (!) =

(1)

(!)E~(!) +

(2)

(!)E~ 2 (!) +

(3)

(!)E~ 3(!) + : : :

(14)

where (2) and (3) are the second-order and third-order nonlinear susceptibilities.
However, in the case of resonantly excited transitions of the material system, such
26

27
a power series expansion does not converge. A saturable absorber is one example where perturbative techniques fail to provide an adequate description of the
response. Instead, the nonlinear relationship between the absorption coe cient
and the intensity I of the applied electromagnetic eld can be written as
(!) =
where

1 + I=Is

(15)

is the small-signal absorption coe cient and Is is the saturation intensity.

5.1 Advances in Modeling Linear and Nonlinear FrequencyDependent Optical Media


The dispersive nature of materials was not a concern in early FDTD simulations,
largely because the applications were in the realm of traditional electromagnetics
such as microwave plane-wave scattering at single frequencies from complicated objects 63]. However, for disciplines outside of this subset of electromagnetics, such
as the interaction of light with optical materials, the material dispersion must be
accounted for accurately. There were at least two motivations for the development
of an FDTD formulation for dispersive materials. The realization of the power
of FDTD over BPM for modeling complicated geometries of integrated optical
components prompted the application of FDTD (using single-frequency sinusoidal
excitation) to guided-wave optical structures 56]. Also, one of the primary advantages of the FDTD method { the ability to obtain wideband frequency spectra
in a single simulation using a short pulse excitation and a discrete Fourier transformation (DFT) { was negated by the fact that many materials of interest could
not be characterized by frequency-independent parameters and therefore could not
be accurately modeled for pulsed excitations by the FDTD method in its original

28
formulation for nondispersive materials.
The rst FDTD formulations to account for material dispersion, presented independently by Luebbers et al. 64] in 1990 and Bui et al. 65] in 1991, demonstrated
a recursive method for e ciently evaluating the electric ux density as a convolution of the electric eld and the susceptibility function. This was accomplished
by discretizing the integral as a running sum, and assuming that the susceptibility function could be expressed as a decaying exponential in time. This recursive
convolution (RC) scheme, originally applied to modeling rst-order Debye media
(single pole), was later extended to Drude media 66] including anisotropic magnetized plasmas 67], and N th-order Lorentz media (2N poles) 68]. The original
RC approach was only rst-order accurate in x. More recently, the accuracy
has been improved using piecewise linear integration (PLRC) 69] and trapezoidal
integration (TRC) 70].
An alternate method, termed the auxiliary di erential equation (ADE) scheme,
was developed independently by Kashiwa and Fukai 71] in 1990 and Joseph et al.
72] in 1991. The ADE method involves the discretization of a di erential equation obtained from the relevant constitutive relation and is second-order accurate
in x. The stability and phase error properties of the ADE schemes were analyzed by Petropoulos 73]. In addition to the initial applications of modeling Debye
and Lorentz media, the ADE approach has been applied to the modeling of wave
propagation in the dispersive ionosphere 74] and in a plasma 75, 76]. The initial disadvantage of the ADE approach was that it required more backstorage of
variables than the RC scheme. However, recent improvements have signi cantly
reduced the storage requirements 77, 78, 79].

29
ADE formulations have also enabled the modeling of nonlinear dispersive media 80, 81]. This modeling capability has been applied to a variety of second- and
third-order nonlinear phenomena, including temporal and spatial soliton propagation 80, 82, 83, 84], self-focusing of optical beams 81], scattering from linearnonlinear interfaces 85], pulse propagation through nonlinear corrugated waveguides 86], pulse-selective behavior in nonlinear Fabry-Perot cavities 87], and
second-harmonic generation in nonlinear waveguides 84].
While the RC and ADE approaches have received the most attention, several
other schemes for implementing models of electronic dispersion have been proposed
these schemes fall into the class of methods based on digital signal processing
techniques. Linear material dispersion has been included in the FDTD method
using a discrete feedback lter 88]. A formulation based on Z transforms 89] for
modeling linear dispersion can also be used to treat nonlinear optical phenomena,
as was demonstrated for the one-dimensional propagation of solitons 90].
FDTD algorithms for modeling optical gain media on a macroscopic level have
also been developed. Using their approach for modeling linear dispersion, Hawkins
and Kallman 91] modeled tilted waveguide traveling wave ampli ers using the
FDTD algorithm. Two di erent approaches for modeling the linear gain medium
of the ampli er were taken. The rst involved the standard Yee formulation with
a constant negative conductivity parameter providing frequency-independent linear gain the second involved the extended FDTD algorithm for modeling linear
dispersion, with parameters selected to provide gain rather than loss. Part of the
research for this dissertation involved in the development of an ADE formulation
to account for materials exhibiting frequency-dependent gain including the nonlin-

30
ear e ect of gain saturation. The details of this work are highlighted in the next
section.
Finally, there have been a few recent advances in modeling the interaction of
light with optical materials on a microscopic level. Ziolkowski et al. 92] have developed an iterative predictor-corrector FDTD method of solving the semi-classical
Maxwell-Bloch equations for a two-level atom. This approach has enabled the
modeling of ultra-fast pulse interactions and self-induced transparency. An alternative formulation for incorporating the atomic rate equations into the FDTD
model eliminates the need for an iterative scheme 93].

5.2 FDTD Algorithm for Gain Media


This section presents the FDTD formulation described above that permits the modeling of frequency-dependent optical gain media, including the e ect of gain saturation. For simplicity, the algorithm is described in the context of a one-dimensional
problem with electric and magnetic eld components, Ez and Hy , propagating
along the x direction through a nonmagnetic, isotropic medium. The approach
described here can be easily extended to arbitrary two- and three-dimensional
problems.
First, the original FDTD algorithm is reviewed for the purpose of comparison to
the extended algorithm. In the original formulation, the eld-material interactions
are described by frequency-independent constants therefore Jz = Ez and Dz =
o Ez + Pz = o (1 + )Ez = Ez , where the conductivity, , the susceptibility,
, and the permittivity, , are assumed constant over the entire spectrum. Then

31
Maxwell's curl equations (1) and (2) in one dimension are

@Hy = 1 @Ez
(16)
@t
o @x
@Hy :
z
Ez + @E
=
(17)
@t
@x
Using centered nite di erences for the space and time derivatives, the curl equations can be expressed as second-order accurate nite di erence equations:
1
1
Hyn+ 2 (i + 12 ) = Hyn; 2 (i + 12 ) +

t E n(i + 1) ; E n(i)]
z
x z

(18)

t
1 ; 2(i)(i) t n
1
1
(i) x
E (i) +
Hyn+ 2 (i + 12 ) ; Hyn+ 2 (i ; 21 )]: (19)
(i) t z
(i) t
1 + 2 (i)
1 + 2 (i)
The vector eld component Vzn(i) denotes sampling at space point x = i x and
time point t = n t. To obtain the solution for the eld components, the two-step
recursive process given by (18) and (19) is iterated to the desired nal observation
time.
In the formulation for frequency-dependent eld-material interactions, the timedomain constitutive relations can no longer be expressed in the simple form used
above therefore, (17) must be replaced by

Ezn+1(i) =

@Hy :
z
=
Jz + @D
@t
@x

(20)

Here, the constitutive relations are simplest in the frequency-domain form of


Jz (!) = (!)Ez(!) and Dz(!) = oEz (!)+ Pz(!) = o 1+ (!)]Ez(!) = (!)Ez (!).
One way to incorporate gain into the FDTD algorithm is through the polarization term of Maxwell's equations. Several numerical schemes for including the
linear e ects of material dispersion in FDTD have been developed these were discussed in Section 5.1. The common feature among these schemes is the use of

32
the frequency-dependent constitutive relation for the electric ux density. Ideally,
any one of these schemes could produce the linear e ect of frequency-dependent
gain through an appropriate choice of parameters in the expression for electric
susceptibility or permittivity. However, at least two of these schemes di er in their
stability properties for modeling linear gain. Using the ADE method 72, 80], initial attempts to simulate linear gain were unsuccessful the algorithm was unstable
for a choice of susceptibility parameters giving gain. Using a similar choice of parameters with a di erent scheme (a variant 94] of the RC method 64]), Hawkins
and Kallman 91] reported stable results for linear gain.
Alternatively, gain can be incorporated as a frequency-dependent conductivity through the constitutive relation for the current density. The idea of using a
negative conductance for gain is an extension of the decades-old concept in microwave engineering and electronics. For example, the tunnel diode has a long
history of use as a microwave negative-resistance ampli er. Initial attempts to
treat the frequency-dependent conductivity were based on the original RC scheme
of Luebbers et al. 64]. While the resulting algorithm for dispersive gain media
appeared to be stable, its accuracy could not be validated. The accuracy of the
RC scheme applied to dispersive lossy media is known to be only rst-order in x
53]. Neither the PLRC nor the TRC methods were investigated.
The method described below, as reported in 95], incorporates gain through a
frequency-dependent conductivity and is based on the ADE scheme. This work
was developed independently from the work of Hawkins and Kallman 91] and
covers linear as well as nonlinear gain. The algorithm assumes that the optical
gain medium is homogeneously broadened wherein the atoms of the gain medium

33
are indistinguishable and have the same atomic transition frequency, !o. The
relaxation processes are included in a phenomenological manner through the time
constant T2 (or decay rate 1=T2). This dipole relaxation time accounts for the fact
that any phase coherence introduced into the system of atoms by the action of an
electric eld will be lost in a time interval T2 once the electric eld is turned o 41].
Therefore the small-signal linear gain coe cient is governed by a single Lorentzian
pro le having a width determined by T2. The gain coe cient should also include
the large-signal nonlinear e ect of saturation, which is due to the decrease of the
population inversion with eld intensity. Accordingly, the frequency-dependent
conductivity is expressed as
(!) = JE z((!!))
z

"

= 1 I 1 + j (!o =;2 ! )T + 1 + j (!o=+2 ! )T


1 + Is
o 2
o 2
"
#
+ j!T2)
= 1 I (1 + !2To2(1
2 2
1 + Is
o 2 ) + 2j!T2 ; ! T2

(21)

where the Fourier transform F (!) of the time-domain function F (t) is de ned
as follows: (all equations in this section are derived consistently assuming this
de nition)
Z1
F (!)
F (t)e;j!tdt:
(22)
;1

The form for (!) is chosen to ensure that the corresponding time-domain conductivity is real and causal:
(t) =

;t=T2 u(t)

T2 cos(!ot)e

(23)

where u(t) is the unit step function. Here o is proportional to the peak value of
the gain set by the pumping level and the resulting population inversion, and Is
is the saturation intensity. The damped resonant frequency, that is, the frequency

34

at which the response is maximized, is given by !d = !o2 ; 1=T22. The peak of


the gain curve is (!d) = o=2 and the full-width-at-half-maximum bandwidth
is ! ' 2=T2. Since (!) accounts for any gain (or loss) in the medium, the
susceptibility can been reduced to a frequency-independent real constant, such
that Dz = Ez where = o r .
To see how this gives gain, consider the linear case when the intensity, I , is
small compared to the saturation intensity, Is. The expression for (!) simpli es
and can be separated into real and imaginary parts,
(!) =
=

R (! ) + j I (! )

1 + (!o2 + !2)T22] + j o!T2 ;1 + (!o2 ; !2)T22 ] : (24)


1 + (!o2 ; !2)T22]2 + 4!2T22
1 + (!o2 ; !2)T22]2 + 4!2T22
o

The homogeneous wave equation for a monochromatic wave in this medium, obtained by combining (16) and (20) and assuming the form Ez (x t) = Ez (x !)ej!t,
is given by
!
@ 2 Ez + !2 ~ (!) ; j R (!) E = 0:
(25)
@x2 c2 r
!o z
The e ective relative permittivity, de ned for ! 6= 0 as ~r (!) r ; I (!)=! o,
clearly depends on frequency. The constant r = n2 is the relative permittivity at
in nite frequency. Furthermore by assuming the form Ez (x !) = Eoe;jkcx,
!

2
kc = !c2 ~r (!) ; j R!(!) :
o
2

(26)

The complex wave number, kc, may be separated into real and imaginary parts,
jkc = + j , where is the eld attenuation/ampli cation constant and is
the phase constant. The exponential propagation factor, e;jkcx, then becomes
e;j xe; x . The term e; x in this expression acts as an ampli er when is negative.

35
The expressions for and , however, are somewhat complicated they can be
simpli ed by making the reasonable assumption that R (!)=! o ~r (!) 1, that is,
the material is low-gain (or low-loss). In this case,
(!) ' 2c Rn(~!(!) )
(27)
o
(!) ' n~(!) !
(28)
c
q
where n~ (!) = ~r (!). From (27), we conclude that if R (!) is negative, then is
negative, providing gain. According to (24), this is the case when the parameter
o is chosen to be negative. Physically, this corresponds to requiring a population
inversion. A similar derivation for the case where the gain is implemented via
a frequency-dependent susceptibility would have indicated similar conditions. In
that case, for (!) = R (!) + j I (!), the system would act as an ampli er for a
positive I (!). In either case, the expressions for and are of the same form as
those obtained by means of semi-classical density matrix theory 41].
As an initial step, the following algorithm assumes a constant saturation intensity. The spatially dependent intensity, I = 12 cn o Eo, is treated as a feedback
parameter in time. Taking the inverse Fourier transform of (21) provides the following auxiliary di erential equation that can by solved simultaneously with (20):
2
@Ez :
z
2 @ Jz
(1 + !o2T22)Jz + 2T2 @J
+
T
=
s
(29)
o Ez + s o T2
2
2
@t
@t
@t
1
Here s 1+I=I
is the saturation coe cient that contains feedback information
s
of the latest peak electric eld. The goal of implementing a central di erencing
procedure of (20) and (29) at time step n + 21 is achieved by de ning an auxiliary
variable, Fz , and rewriting (29) as two rst-order di erential equations:
z
Fz @J
(30)
@t

36

@Ez :
z
(1 + !o2T22)Jz + 2T2 Fz + T22 @F
=
s
(31)
o Ez + s o T2
@t
@t
The system of coupled di erential equations (20), (30), and (31) are di erenced
using semi-implicit representations for Fz , Jz , and Ez at time step n + 12 . Solving
the resulting system of nite di erence expressions for the three unknowns Fzn+1(i),
Jzn+1(i), and Ezn+1(i) yields the following explicit expressions:
1
1
Fzn+1(i) = A1 Hyn+ 2 (i + 12 ) ; Hyn+ 2 (i ; 21 )] + A2 Ezn(i) + A3Jzn(i) + A4 Fzn(i) (32)

Jzn+1(i) = Jzn(i) + 2t Fzn+1(i) + Fzn(i)]


(33)
1
1
Ezn+1(i) = Ezn(i) ; 2 t Jzn+1(i) + Jzn(i)] + tx Hyn+ 2 (i + 12 ) ; Hyn+ 2 (i ; 21 )] (34)
where
(35)
A1 = 4 t s(i) ox( t + 2T2) A2 = 8 s(i) o t
2 2
4
t
2
(1
+
!
T2 ) + s(i) o( t + 2T2)]
o
A3 = ;

2
2 2
A4 = ; 8 T2 ( t ; T2) ; ( t) 2 (1 + !o T2 ) + s(i) o( t + 2T2)]

(36)
(37)

= 8 T2( t + T2) + ( t)2 2 (1 + !o2T22) + s(i) o ( t + 2T2 )]


(38)
s(i) = 1I (i) I (i) = 21 cn o Ezpeak (i)]2 :
(39)
1 + Is
For o = 0 and T2 = 0, the update equations given by (32), (33), and (34) reduce
to the update equation (19) for frequency-independent lossless ( = 0) media, as
they should.
For a linear medium, there is no feedback the saturation coe cient reduces
to s = 1, because the intensity is negligible compared to the saturation intensity.
For a nonlinear medium, the saturation coe cient is updated as follows. If the
electric eld at time step n is greater than the electric eld at time step n ; 1

37
at the same location in space, then the saturation coe cient is updated using
Ezpeak (i) = Ezn(i). On the other hand, if the electric eld at time step n has
decreased from its previous value, then the saturation coe cient is not updated
hence s(i) remains based on the latest peak electric eld. In this manner, intensity
feedback in the time domain retains as much as possible its frequency domain
meaning. Note that since the feedback is performed independently at each grid
location, this method simulates a spatially inhomogeneously broadened medium
in which spatial hole burning may occur.
Equations (18), (32), (33), and (34) compose the complete FDTD time-stepping
algorithm for the single Lorentzian optical gain medium using the auxiliary di erential equation approach. The computational model is now a four-step recursive
process that retains the fully explicit nature of the original frequency-independent
FDTD algorithm and requires storage of elds only one step back in time. More
complicated gain spectra can be approximated by using a linear combination of
Lorentzians.

5.3 Validations
The accuracy of the method presented in the previous section is demonstrated with
two validation studies. In the rst study, the accuracy of the FDTD algorithm
for linear frequency-dependent gain is tested. In the second study, the FDTD
algorithm for frequency-dependent gain and gain saturation is applied to a onedimensional Fabry-Perot etalon, the simplest geometry for a laser oscillator.

38

Linear frequency-dependent gain


The accuracy of the algorithm presented in Section 5.2 is rst demonstrated for a
linear (small-signal) gain medium. As reported in 95], the electric eld propagation
factor, de ned by e;jkc(x2;x1 ) = Ez (x2 !)=Ez (x1 !), is computed for a gain medium
with o = 0:89 m, T2 = 0:07 ps (gain spectrum bandwidth = 5 THz, n = 3:59,
and s = 1. A single Gaussian pulse with a 5-fs temporal width between the 1=e
points (a bandwidth greater than 5 THz) and a carrier frequency of !o propagates
through the medium, data are taken every time step at two xed observation
points separated by a distance of l = x2 ; x1 = o=n. By taking the ratio of the
discrete Fourier transforms of the pulses at the two locations, the complex-valued
propagation factor from the FDTD model is calculated over the full bandwidth
of the pulse. In Figure 8, the ampli cation factor, e; l , and the phase, l, of the
propagation factor (minus the amount n!l=c) are compared with the exact pro les.
If strictly l is plotted as a function of frequency, the only clearly observable
behavior is the linear dependence of n!l=c the ne details of the contribution
from I (w) are hidden. The reason is evident from (28), which can be written as
s

(!) ' n! 1 ; I (!) :


c
!or
Since I (!)=! o r is small,
expansion, yielding

(40)

can be simpli ed further using a truncated series

(!) ' n! ; 1 I (!) :


(41)
c 2 c on
From (41), it is clear that l, the phase of the propagation factor, is dominated by
the n!l=c behavior. Plotting l ; n!l=c allows the subtle dispersive properties of
the gain medium to be observed. At a very ne grid resolution of x = o=400n =

39
amplification factor
1.03

theory
FDTD

1.02

1.01
1.00
1.8

1.9

2.0
2.1
2.2
15
(10 rad/s)

2.3

2.4

phase (in degrees)


1.0

theory
FDTD

0.5

0.0
0.5
1.0
1.8

1.9

2.0
2.1
2.2
15
(10 rad/s)

2.3

2.4

Figure 8: Comparison of FDTD results and theory for the ampli cation and phase
factors of a pulse propagating a distance of one dielectric wavelength in a linear
gain medium with o = ;5000 S/m.

40
0:62 nm ( t = 0:002068 fs), the deviation from the exact values over the complete
frequency range is less than 2 parts per 10,000 in the ampli cation factor and less
than 0:006 in the phase factor. This indicates the validity and potential for high
accuracy of the FDTD model over extremely large instantaneous bandwidths.

Gain saturation
Next, the accuracy of the algorithm presented in Section 5.2 is demonstrated for
a nonlinear (large-signal) gain medium 95]. The structure used for this study
consists of a one-dimensional Fabry-Perot cavity lled entirely with a GaAs gain
medium ( o = 0:89 m 96], T2 = 0:07 ps 97], n = 3:59 97], and Is = 65:2
kW/cm2 ) of length l = 12:4 m. The mirrors at each end of the cavity are
formed by planar GaAs/air interfaces. Therefore, the re ectivity, R, of each end
facet is independent of frequency. The mth resonance for this simple cavity is
given by m = 2nl=m. Figure 9 shows the unsaturated gain curve for an abovethreshold choice of o . Three longitudinal modes lie in the region between 880
and 900 nm where the unsaturated gain exceeds the round-trip loss. Since this
is a homogeneously broadened system, the longitudinal mode with the highest
unsaturated gain is expected to clamp the gain curve at the loss line, yielding singlemode operation. In this example, the frequency of the lasing mode is designed to
be !m = !o, that is, the peak of the gain curve. (Technically, the peak of the gain
curve occurs at !d, as discussed in Section 5.2 however, for T2 = 0:07 ps, !d is
equivalent to !o to four decimal points.) For a cavity with no internal loss, the
material gain required at threshold is given by
th

= 21l ln R:

(42)

41

gain or loss (per m)

0.20

0.15

mirror losses
gain (0=7000 S/m)
cavity modes

0.10

0.05

0.00
840

850

860

870

880
890
900
wavelength (nm)

910

920

930

Figure 9: Unsaturated gain and loss spectra for a Fabry-Perot microcavity laser.
The longitudinal modes of the cavity are shown as vertical lines.
Using (27) to determine the threshold value of o yields a theoretical value of
oth = ;1760 S/m.
Within a semi-classical framework, spontaneous emission can be included in
Maxwell's equations as a noise current 98]. A pseudorandom number generator
for a Gaussian random number sequence 99] with zero mean and a variance of
1:0 10;6 (a standard deviation of x = 0:001) is used to implement the noise current inside the laser cavity. The approximately Gaussian random number sequence
is obtained by summing uniformly distributed random numbers therefore, the generation of one Gaussian random number requires the generation of twelve uniformly
random numbers (twelve subroutine calls). A histogram of the rst 10,000 outputs
of the pseudorandom number generator is shown in Figure 10. Each bar gives the

42
relative frequency of the random number falling within the range of the bar { for
example, 0:000 < X 0:001, 0:001 < X 0:002, and so on. The noise source is

relative frequency of X

0.40

0.30

0.20

0.10

0.00
50.004
0 X
X 4X 30.002
X 2X X 0.000

2X 3X0.004
4X 5X
0.002

random number X

Figure 10: Histogram of the rst 10,000 outputs of a pseudorandom number generator for a Gaussian random number with mean = 0:0 and variance = x2 .
implemented in the FDTD algorithm as a virtual current source:

Ezn+1(isource ) = Ezn(isource ) + X

(43)

where X is the output of the pseudorandom number generator. Since the generation of multiple Gaussian random numbers at each time step is computationally
burdensome, the source is con ned to only one grid cell within the cavity. This
simpli cation, which reduces the simulation run time by almost forty-fold in this
example, has no observable e ect on the results.

43

electric field (kV/m)

The FDTD-computed time evolution of the electric eld outside the laser cavity
is shown in Figure 11. In Figure 11a, the transient response is shown after a delay,

1000
500
0
500
1000
0.0

electric field (kV/m)

(a)

1000

1.0

2.0

3.0
4.0
time (ps)

5.0

6.0

7.0

(b)

500
0
500
1000
6.50

6.51

6.52

6.53

time (ps)

Figure 11: (a) FDTD-computed time evolution of the electric eld outside the
Fabry-Perot laser cavity. (b) Expanded timescale of the steady-state region showing a single-mode oscillation at !o.
the electric eld oscillations build up rapidly from the noise background, then level
o to a steady-state amplitude as the gain saturates. Figure 11b shows an expanded
timescale of the sinusoidal steady-state region of Figure 11a, which illustrates that
the FDTD model correctly predicts a lasing frequency of !o. This simulation,

44

performed for o = ;7000 S/m, was repeated for smaller values of o . In each
simulation, the output intensity was computed from the steady-state data. The
results, plotted as a light-current (L-I) curve in Figure 12, show that the output
intensity varies linearly with the above-threshold gain level. This is expected for
a homogeneously broadened system. Furthermore, the FDTD simulations provide
an accurate estimate of the gain threshold. Extrapolating the data in Figure 12
yields an estimate of oth ' ;1780 S/m, which corresponds closely to the value of
;1760 S/m calculated above.

FDTD computed laser output

Output Intensity (kW/cm )

100
80
60
40
20
0

2000

4000
o (S/m)

6000

8000

Figure 12: FDTD-computed L-I curve for the Fabry-Perot laser, showing the dependence of the output intensity on the peak of the gain curve.
The FDTD model does not treat the initial carrier dynamics that determine
the turn-on delay instead, the model assumes that the population inversion already exists when the simulation begins. Therefore, the delay in the build-up and

45
saturation of the oscillations observed in Figure 11 depends solely on the noise
power (the variance), which is chosen arbitrarily. Another parameter that is chosen somewhat arbitrarily is the saturation intensity Is. Decreasing Is while keeping
the noise level constant simply decreases the steady-state output intensity the delay remains constant. Most importantly, varying the noise level or the saturation
intensity has no e ect on the lasing wavelength or gain threshold obtained from
the simulations.
This simulation of a simple Fabry-Perot laser demonstrates the validity of this
method for providing accurate predictions of the gain threshold and the lasing
wavelength. A formal stability analysis of the FDTD algorithm for frequencydependent gain with saturation is di cult because of the nonlinearity however,
modeling experience to date, which includes time-stepping the algorithm in onedimension for more than 1 million time steps, shows that the optical e ect of gain
saturation requires no more a restrictive time step than is needed to ensure the
stability of the original FDTD algorithm.

6 FDTD Modeling of Vertical Microcavity Lasers


Vertical-cavity surface-emitting lasers (VCSELs) based on simple Fabry-Perot cavities having high-re ectivity dielectric quarter-wave stacks for mirrors have attracted much attention over the last decade because of their desirable lasing characteristics: low power consumption, high output power, and single-mode operation.
The low lasing threshold of VCSELs can be further improved by using a periodicgain con guration within the microcavity. By placing thin gain segments along
the electric eld standing wave maxima, the longitudinal con nement factor, ;,
is maximized, thereby reducing the material gain threshold. Corzine et al. 96]
developed an approximate analytical method for optimizing the design of the periodic gain structure. They demonstrated that close to a factor of 2 reduction in
the material gain threshold is possible.
In this chapter, the FDTD algorithm described in Section 5.2 is used to model
VCSELs with uniform-gain and periodic-gain structures (UGS and PGS) 95]. The
microcavity geometries, shown schematically in Figure 13, are modeled as onedimensional structures since the lateral dimensions are relatively large (> 5 m)
96]. Each is assumed to be grown on a GaAs substrate with mirrors comprised
of distributed Bragg re ectors (DBRs) consisting of Al0:2Ga0:8As/AlAs alternating
quarter-wave layers. The top mirror has nine pairs while the bottom mirror has
12.5 pairs. The DBRs are designed to have a Bragg wavelength of B = 0:87 m
and a peak re ectivity of 92.8%. The refractive index of AlxGa1;xAs is given
experimentally by
n = 3:590 ; 0:710x + 0:091x2
(44)
46

47
for light at a wavelength of 0:9 m at 297 K 97]. Table 1 lists the materials
and the corresponding refractive indices. The optical path length of the cavity is
l = 27 B=2, which corresponds to a physical length of L = l=n.
Periodic Gain Structure (PGS)

/2

Uniform Gain Structure (UGS)

ACTIVE
PASSIVE
DBR

Figure 13: Schematic of microcavity lasers having vertical distributed Bragg reectors and uniform (right) or periodic (left) gain structures within the cavity.
For the uniform-gain case, the cavity consists of a solid, active GaAs region
( o = 0:89 m, T2 = 0:07 ps, n = 3:59, Is = 65:2 kW/cm2 ) of length d surrounded
on each side by a passive GaAs region (n = 3:59). To make a direct comparison
between the UGS and PGS, GaAs is used for both the active and passive regions
of the cavity in order to avoid the e ect of refractive index variations within the
cavity. The ll factor, de ned as d=L, is chosen to be 0:5. For the periodic-gain

48
material refractive index
GaAs
3.59
Al0:2Ga0:8As
3.45164
AlAs
2.971
air
1.0
Table 1: Refractive indices used in the FDTD models of the VCSELs of Fig. 13.
case, the cavity consists of thin active GaAs segments, each with a thickness t,
separated by passive GaAs segments. The optical path length between the gain
segments is equal to one-half the lasing wavelength, laser=2. The total thickness
of the gain segments is equal to d. These design parameters permit comparison of
the FDTD results with the analysis presented by Corzine et al. 96].
Following 96], for a cavity with no internal loss, the material gain required at
threshold is given by
1 1 ln R
(45)
th =
; 2L
where R is the geometric mean mirror re ectivity and ; is the longitudinal connement factor. For the UGS, ; is simply the ll factor:
;UGS = d :
L

(46)

Assuming a uniform standing wave pattern and ideal mirror re ectivities, the longitudinal con nement factor for the PGS can be approximated as
8

<
sin ( =t 2 )] =
;PGS = d :1 +
L
( =t 2 )

(47)

where = laser=n. When t = =2, Equation (47) shows that ; for the PGS
reduces to that of the UGS. When t ! 0, that is, when the gain segments are

49
extremely thin, ; is maximized at twice that of the UGS. Therefore, according to
Equation (45), the gain threshold can be reduced by as much as 50%. Equation
(47) can be expressed in terms of the ll factor as follows:
(

sin ( Ld )]
;PGS = Ld 1 +
( Ld )

(48)

since L = M =2 and d = Mt, where M is the number of thin gain segments in


the PGS. For the design studied here, the gain threshold for the PGS according
to (45), (46), (47), and (48) is approximately 39% less than the gain threshold for
the UGS.

6.1 Passive studies


First, the FDTD method without gain is used to determine the passive characteristics of the cold-cavity VCSEL. Here, the UGS and PGS are identical because
the refractive indices of the passive and active regions are the same. For each
of the mirrors, the frequency-dependent re ection coe cient is computed with a
single FDTD run, using a short-pulse excitation and DFTs of the incident and reected time-domain data. The FDTD grid for the simulation of the bottom DBR
consists of a half space of GaAs (for the cavity side of the mirror), 12.5 pairs of
AlGaAs/AlAs quarter-wave layers, and a half space of GaAs (for the substrate
side of the mirror). In Figure 14a, the FDTD results for the bottom DBR are
compared with the analytical spectra of the re ectivity (the squared amplitude
of the re ection coe cient) and the phase of the re ection coe cient, given by
the matrix method 100]. At a grid resolution of x = B=124nH ' 2:0 nm, the
deviation from the matrix method at the re ectivity peak is less than 3 parts per
10,000. For the top mirror, the FDTD grid consists of a half space of GaAs (for the

50

reflectivity

1.0
0.5
0.0
0.0

phase (in degrees)

FDTD
theory (matrix method)

(a)

0.5

1.0
/B

1.5

2.0

0.5

1.0
/B

1.5

2.0

360
180
0
0.0

reflectivity

1.0
0.5
0.0
0.0

phase (in degrees)

FDTD
theory (matrix method)

(b)

0.5

1.0
/B

1.5

2.0

0.5

1.0
/B

1.5

2.0

360
180
0
0.0

Figure 14: FDTD-computed re ectivity and phase of the re ection coe cient for
the (a) bottom and (b) top mirrors of the vertical microcavity.

51
cavity side of the mirror), 9 pairs of AlGaAs/AlAs quarter-wave layers, and a half
space of air. Similar accuracy is obtained for the top DBR, shown in Figure 14b.
The cold-cavity resonant modes are determined by computing the re ectivity
spectrum of the entire microcavity. In Figure 15, ve resonances (821 nm, 843
nm, 870 nm, 897 nm, and 923 nm) are seen to fall within the broad stopband.
Knowledge of the spacing between these resonant frequencies, here about 27 nm,
permits determination of the e ective cavity length. This is longer than the physical length of the cavity because of eld penetration into the mirrors. In the case
considered here, the e ective cavity length is approximately 34 half Bragg wavelengths, compared to the length between mirrors of 27 half Bragg wavelengths.
1.0

reflectivity

0.8
0.6
0.4
0.2
0.0
0.75

0.80

0.85
0.90
0.95
wavelength (m)

1.00

Figure 15: FDTD-computed re ectivity spectrum for the vertical microcavity.

52

6.2 Active studies

Following the methodology presented in the second validation study of Section 5.3,
the FDTD algorithm for gain is used to determine the lasing wavelength and gain
threshold for the VCSELs of Figure 13. Figure 16 graphs the unsaturated GaAs
gain curve for three di erent values of o (dotted lines), the mirror loss curve calculated from Figure 14 (solid line), and the cold-cavity resonances determined from
Figure 15 (thin vertical dotted lines) 101]. The GaAs gain spectrum is centered at

gain or loss (per m)

0.08
mirror losses
0=2000 S/m
0=3000 S/m
0=4000 S/m
cavity modes

0.06

0.04

0.02

0.00

840

850

860

870

880

890

900

910

920

930

wavelength (nm)

Figure 16: Superposition of the unsaturated gain spectrum of GaAs for three
values of o , the loss characteristic of the mirrors (from Fig. 14), and the coldcavity resonances (from Fig. 15).
890 nm while the mirror re ectivity spectrum is centered at the Bragg wavelength
of 870 nm. The gain bandwidth is narrower than the re ectivity bandwidth of
the mirrors. At the 870-nm cavity resonance, the mirror loss exceeds the optical

53
gain for each value of o shown, even though the mirror loss is at a minimum here.
However, at the 897-nm cavity resonance, the optical gain exceeds the mirror loss
for values of o more negative than approximately -3000 S/m. Therefore, since the
897-nm longitudinal mode experiences the greatest net gain, lasing should occur
at this wavelength.
Figure 17 (top) shows the spatial variation of the index of refraction for both
the UGS and PGS microcavities. Here, the index varies periodically in the mirror
regions between 2.97 (AlAs layer) and 3.45 (AlGaAs layer) and is independent of
the speci c gain structure. Figure 17 (middle) plots o as a function of position
within the microcavity to show the region of uniform gain in the UGS that is
centered within the otherwise passive cavity-mirror structure. Figure 17 (bottom)
illustrates the periodic-gain con guration where o alternates between 0 (passive
segment) and -10,000 S/m (active segment).
Figure 18 plots the FDTD-computed envelope of the electric eld within the
uniform-gain VCSEL at the sinusoidal steady state. Here, the decaying elds on
either side of the cavity illustrate the penetration of the mode into the passive
mirrors. This simulation yields a lasing wavelength of just under 900 nm, as
expected from the discussion of Figure 16. The standing wave maxima within
the cavity are spaced periodically by an optical path length of laser=2. These
maxima determine the locations of the thin gain segments within the periodicgain VCSEL, as shown in Figure 17 (bottom). The spatial distribution of the
steady-state electric eld that develops in the PGS is similar to that for the UGS.
For each lasing structure, the simulation is performed for several di erent values
of o and the output intensity is recorded, as shown in Figure 19.

index of refraction

54
4.0
3.0
(UGS or PGS)

2.0
1.0
0.0
0.0

1.0

2.0

3.0
4.0
length (m)

5.0

6.0

0 (kS/m)

15
UGS

10
5
0
0.0

1.0

2.0

3.0
4.0
length (m)

5.0

6.0

0 (kS/m)

15
PGS

10
5
0
0.0

1.0

2.0

3.0
4.0
length (m)

5.0

3.0
4.0
length (m)

5.0

6.0

normalized electric field

Figure 17: Spatial variation of the refractive index (top) and optical gain (middle
and bottom) within the VCSELs with uniform and periodic gain con gurations.
1.0
0.8
0.6
0.4
0.2
0.0
0.0

1.0

2.0

6.0

Figure 18: Normalized steady-state electric eld envelope within the VCSEL.

55
4.0
PGS
UGS

Output Intensity

3.0

2.0

1.0

0.0

10
0 (kS/m)

15

20

Figure 19: FDTD-computed output intensity vs. gain for VCSELs comprised of
either a uniform-gain structure (UGS) or a periodic-gain structure (PGS).
A comparison of the FDTD-computed UGS and PGS data sets reveals the
following advantages of the PGS:
1. a substantial reduction in the gain threshold for lasing and
2. a higher output intensity for a given gain level.
Extrapolation of the data of Figure 19 towards zero output intensity yields a lasing
threshold of oth ' ;3260 S/m for the case of the UGS, about as expected from the
discussion of Figure 16. For the case of the PGS, oth ' ;1790 S/m. Therefore, the
FDTD simulations indicate that the lasing threshold for the PGS is approximately
45% less than the threshold for the UGS. This compares to a 39% reduction in the
gain threshold predicted by (48), (45), and (46). The FDTD calculation is a more

56
accurate measure of the threshold reduction since it accounts for the fact that the
mirrors are not perfect re ectors.

7 FDTD Modeling of Ring and Disk Resonators


The waveguide-coupled microcavity ring and disk resonators that were introduced
in Chapter 2 are further explored in this chapter. In the numerical models discussed
here, the standard FDTD method is used to solve Maxwell's equations in two
dimensions for the TE set of eld components Ez , Hx, and Hy .

7.1 FDTD Modeling Considerations


Figure 20 shows a typical geometry of a microcavity ring or disk resonator. Adjacent waveguides WG1 and WG2 serve as evanescent wave input and output
couplers. Using the bootstrapping technique discussed in Section 4.2, a fundamental mode is sourced at the left end of WG1. Here, the excitation is a 20-fs
Gaussian pulse modulating a 200-THz carrier. The eld components in the pulsed
waveguide mode are recorded as it propagates past cross-sections P1 and P2 in
WG1, and in the microring or microdisk at P3.
Power spectra at these cross-sections are calculated as follows. First, DFTs of
the time histories of the electric and magnetic elds are computed at each grid
point along each cross-section. The extent of each cross-section is wide enough
to detect essentially all of the power in the waveguide mode. Next, using the
complex-valued Fourier transform data, the time-average longitudinal Poynting
power density at each cross-section grid point is calculated as a function of frequency by taking 21 Real(E H ). Then the total longitudinal power ux passing
through the waveguide cross-section at each frequency is calculated by summing
the Poynting power densities at the given frequency at all grid points along the
cross-section. In this manner, a single computer run provides a complete spectral
57

58
characterization of the coupling e ciency (the ux through P3 normalized by the
incident ux through P1) or the transmittance (the normalized ux through P2).
C
on

WG2

P3

P2

P1
g

on ,
off

off

WG1
A

Figure 20: Schematic of a microcavity ring or disk resonator coupled to two straight
waveguides. An o -resonance signal entering port A remains in WG1 and exits at
port B. An on-resonance signal is switched to WG2 and exits at port C.
Where benchmark data are available, the accuracy level of this modeling approach is found to be excellent for the assumed standard grid resolution of 13.6
nm. For example, power balances in the propagating mode computed in this manner are found to hold within parts per ten thousand for cases where the optical
radiation is negligible.
In the vertical direction, each waveguiding structure considered here consists
of a 0.45- m-thick GaAs guiding layer with AlGaAs cladding layers above and

59
below. Strong lateral con nement is achieved in the semiconductor waveguide by
etching down through the lower cladding layer. The structure is weakly guiding
in the vertical direction to allow for the input coupling of light from an external
ber or planar-emitting device. The vertical waveguide thickness and material
composition is accounted for by computing the e ective refractive index ne for
the fundamental mode at = 1:5 m, and then using ne as the bulk material
index of the core in the two-dimensional simulations. For vertical core thicknesses
of 0:3 m to 0:5 m, ne is between 3.2 and 3.3. In the cases considered here, the
waveguide core (n = 3:2) is bordered on each side by air (n = 1).
As discussed in Section 2.1, a universal measure of the e ectiveness of an optical
resonator is the quality factor, Q, which is de ned as
Q = !U
= ; !U :
(49)
P
dU=dt
Here, U is the eld energy stored by the resonator and P = ;dU=dt is the power
dissipated by the resonator. Mechanisms which can a ect the Q of microcavity ring
resonators are intrinsic material absorption, radiation loss caused by waveguide
bending and surface roughness scattering, and coupling from the cavity to the
adjacent waveguides.
The ion-assisted etching process results in a surface roughness of the sidewalls
of the waveguides that is experimentally observed to be on the order of 10 to 20 nm
30]. This fabrication artifact causes two problems due to its tendency to scatter
light propagating within the waveguiding structure: 1) radiation to the far eld,
and hence, a partial loss of the stored energy and a reduction of the resonator Q
102] and 2) backre ection into the counterpropagating mode 103]. The 13.6-nm
grid-cell size is approximately 1/40th of the optical wavelength within the wave-

60
guide core material. This gridding is ne enough to directly model the 10-20-nm
waveguide surface roughness by implementing a standard, staircased approximation of the curving waveguide in the manner of the basic FDTD algorithm. (In
fact, several variations of the staircase t have been constructed to examine the
optical e ects of randomness in the fabrication process). Thus, the FDTD model
naturally includes the scattering losses. This grid-cell size is also ne enough to
permit highly detailed models of electromagnetic coupling across the air gaps between the coupling waveguides and the resonators, even down to the minimum
distance that can be reliably fabricated with current technology (on the order of
100 nm).
Radiation losses to the far eld can also be caused by waveguide bends. However, because of the strong mode con nement, the bending losses are negligible
in resonators with diameters greater than 1-2 m 14]. For the passive devices
considered here, the intrinsic absorption loss is negligible because the wavelength
of interest corresponds to an energy that is well below the bandgap energy.
The importance of accurate modeling of the highly dispersive semiconductor
waveguides is illustrated with the following equation for the FSR between the mth
and (m + 1)th longitudinal modes:
(
)
c
m
+
1
m
(50)
FSR =
de ( m+1 ) ne ( m+1 ) ; de ( m ) ne ( m )
where ne is the e ective index of the waveguide mode and de is the e ective
diameter of the ring or disk. The quantity de corresponds to the circumferential
path traveled by the peak of the guided mode that is, de is the distance between
the two peaks of the transverse mode pro le found along the ring/disk diameter.
For semiconductor microcavity ring and disk resonators, ne ( m ) 6= ne ( m+1 )

61
because the waveguide dispersion is strong and the longitudinal mode spacing is
so wide. Also, de is signi cantly smaller than the diameter d and is a function of
frequency. Therefore, (50) cannot be reduced to FSR = c=d ne , as is commonly
done for larger weakly guiding resonators. Similarly, the Q is a ected by the
relatively large dispersion 104].

7.2 Ring Resonators Coupled to Straight Waveguides


In Figure 20, the coupling between the waveguides and the microcavity ring or
disk resonator is controlled by two parameters that a ect the coupling interaction
length: g, the size of the coupler air gap and d, the outer diameter of the ring or
disk. Accordingly, the coupling e ciency of this geometry can be investigated by
varying g and d 105]. In the present simulations, the width of WG1, WG2, and
the ring waveguide is 0.3 m. The straight waveguides support only one symmetric
and one antisymmetric mode at = 1:5 m. The coupling coe cient, , de ned
as the percentage of power coupled into the ring from WG1, is computed using the
procedure outlined in Section 7.1.
Consider a 5.0- m-diameter ring coupled to a straight single-mode input waveguide. Figure 21 shows the FDTD-computed coupling coe cient for the verticalpolarization (V-pol) case where the electric eld is perpendicular to the plane of
the ring for g varying from 0.19 to 0.26 m. Two well-known coupling phenomena
are illustrated in this gure 105]:
1. For a xed frequency, decreases as the physical gap size increases.
2. For a xed physical gap size,
e ective gap size increases.

decreases at higher frequencies where the

62
6.0
g=0.191 m
g=0.205 m
g=0.218 m
g=0.232 m
g=0.245 m
g=0.259 m

coupling coefficient, (%)

5.0

4.0

3.0

2.0

1.0

0.0
175

200
frequency (THz)

225

Figure 21: FDTD-computed coupling coe cient, , as a function of frequency


and gap size, g, for the geometry of Figure 20, (5.0- m-diameter ring resonator
case). The electric eld is vertically polarized, i.e., perpendicular to the plane of
Figure 20.
As seen in Figure 21, at = 1:5 m ( = 200 THz), ranges between 0.7% and
2.8% for a change of only 68 nm in g. This compares favorably to the desired range
for , typically 0.5% to 3%, needed to achieve good transmission characteristics
and large extinction ratios. (Convenient rules of thumb are that should be
greater than the round-trip cavity loss, and should be symmetric for both WG1
and WG2 relative to the ring or disk.) While it is feasible to fabricate the small
gaps required to achieve the desired range of , the FDTD simulations indicate

63
that is very sensitive to g. Therefore, since it is di cult to fabricate identical
gaps to very high precision, the coupling from the ring to WG1 and WG2 may be
slightly asymmetric.
Figure 22 compares the V-pol coupling coe cient curve of Figure 21 obtained
for g = 0:232 m with analogous FDTD results for the same gap size computed
for the horizontal-polarization (H-pol) case corresponding to the TM set of eld
components, Hz , Ex, and Ey , where the electric eld is in the plane of the ring
105]. The coupling coe cient for the H-pol case at 1.5 m is 1.9% compared with
6.0
Vpol
Hpol

coupling coefficient, (%)

5.0

4.0

3.0

2.0

1.0

0.0
175

200
frequency (THz)

225

Figure 22: Comparison of the FDTD-computed coupling coe cient for the V-pol
and H-pol cases of Figure 20 (ring) with d = 5:0 m and g = 0:232 m.

64
1.2% for the V-pol case. These results show that the H-pol coupling exceeds the
V-pol coupling for most of the spectrum of interest. In order to avoid high lateral
coupling of light to the media surrounding the waveguides and the ring or disk,
relatively wide boundary trenches must be etched, as seen in the SEM images of
Figure 4. While the remainder of this chapter focuses on the V-pol case, it is noted
that with a su ciently wide trench, the H-pol case is qualitatively the same.
The V-pol FDTD results for a 10- m-diameter ring resonator are shown in
Figure 23. This ring has about twice the of the 5.0- m-diameter ring for a given
12.0
g=0.191 m
g=0.205 m
g=0.218 m
g=0.232 m
g=0.245 m
g=0.259 m

coupling coefficient, (%)

10.0

8.0

6.0

4.0

2.0

0.0
175

200
frequency (THz)

225

Figure 23: V-pol FDTD-computed coupling coe cient, , as a function of frequency and gap size, g, for the geometry of Figure 20, (10.0- m-diameter ring
resonator case).

65
gap size and frequency, for example, ranging from 1.3% to 5.5% at = 1:5 m.
The interaction length for the larger ring is still short enough to avoid backcoupling
from the ring to the waveguide.

7.3 Ring Resonator Coupled to Curved Waveguides


In addition to varying the gap size and ring diameter, the waveguide-to-ring coupling can be manipulated by adding curvature to the input and output waveguides.
This section presents the case wherein the adjacent waveguide remains straight
from the source plane to the point of smallest separation from the ring, and then
is provided with a controlled double curvature to allow variability of the interaction
length 105].
Figure 24 illustrates the geometry used for the curved-waveguide models. At
the point of smallest separation from the 5.0- m-diameter ring, the gap width is
0.205 m. The curved part of the input waveguide has a bend radius of . For a
given value of , the curved waveguide section bends toward the ring for 1/8th of
the ring's circumference and then bends away from the ring in exactly the same
manner. The end of the waveguide is extended so that it can be terminated into
the PML normal to the grid boundary.
Figure 25 graphs the FDTD-computed for this structure. The top curve
( = 1) shows the previous case of the straight waveguide. In successive curves,
the bend radius decreases slightly. The minimum possible value for is r + g,
where r is the radius of the ring. In this extreme, the gap between the ring and
waveguide is constant over a 45 angle. The FDTD results illustrate the following
phenomena:

66

r
45 o

g
bend radius =

Figure 24: Schematic diagram of a microcavity ring resonator coupled to a curved


waveguide. The bend radius of the waveguide's outer curves is .
1. The coupling drops as the input waveguide bends more toward the ring, suggesting an increasing phase mismatch between the signals in the waveguide
and in the ring.
2. For the smaller values of , is less sensitive to frequency.
Therefore, curving the adjacent waveguide toward the ring increases the interaction
length, but introduces an undesirable path length di erence due to the di ering
bend radii of the ring and adjacent waveguide. In order to achieve an increase
in the coupling through the use of a curved adjacent waveguide, the length of
curvature of the waveguide should be no longer than a very small fraction of the
circumference of the ring.

67
6.0
= infinity
=10.2 m
= 5.2 m
= 4.0 m
= 3.3 m
= 3.0 m

coupling coefficient, (%)

5.0

4.0

3.0

2.0

1.0

0.0
175

200
frequency (THz)

225

Figure 25: FDTD-computed coupling coe cient as a function of frequency and


waveguide bend radius, , for a 5.0- m-diameter ring ( g = 0:205 m, V-pol case).

7.4 Elongated Ring Designs


As discussed in Section 7.2, the strong optical con nement of the waveguides requires very close proximity of the waveguide to the ring in order to achieve the
desired level of coupling. An important step towards practical fabrication and
commercialization of these devices is the exploration of designs that might alleviate the need for such narrow gaps. In particular, a wider air gap may allow
reliable and repeatable fabrication of these resonators with reduced sensitivity to
fabrication variations.

68
In order to achieve the desired coupling level with a wider gap, the interaction
length over which the coupling takes place must be increased. In the previous
section, it was shown that the coupling actually drops when the adjacent waveguide
curves along the ring for more than just a small fraction of the circumference of
the ring. Here, changing the shape of the ring instead of the adjacent waveguide is
investigated. For example, elongating the ring to form an elliptical con guration
or a \racetrack" con guration leads to an increased interaction length without
introducing a phase mismatch 106].

b
a

Figure 26: Schematic of a microcavity ring resonator (elliptical con guration)


coupled to straight waveguides.
Consider the elliptical ring resonator of Figure 26 con gured with a gap width
of 0.3 m and a semiminor axis length of b = 2:5 m. Figure 27 shows the FDTDcomputed coupling coe cient as a function of frequency and the semimajor axis

69
length, a. The solid line (a = 2:5 m) represents the coupling for the 5.0- m6.0
a=2.5 m
a=3.0 m
a=4.0 m
a=5.0 m
a=6.0 m

coupling coefficient, (%)

5.0

4.0

3.0

2.0

1.0

0.0
175

200
frequency (THz)

225

Figure 27: FDTD-computed coupling coe cient (V-pol case) as a function of


frequency and semimajor axis, a, for an elliptical ring resonator coupled to straight
waveguides (b = 2:5 m, g = 0:3 m).
diameter circular ring resonator previously studied. For the circular ring, the
coupling coe cient at = 1:5 m is approximately 0.3% at this larger gap width.
By elongating the ring to form an ellipse with a major axis of 2a = 12 m,
increases to 1.5% at = 1:5 m. Figure 28 compares the FDTD-computed
for the elliptical ring resonator (a = 6:0 m, b = 2:5 m) and the previous
5.0- m-diameter circular ring resonator as a function of the gap width, g. The
dotted curves represent at the wavelengths 1.5 m, 1.55 m, and 1.6 m for the

70
circular ring, and the solid curves represent at the same three wavelengths for
the elliptical ring.
5.0
=1.60 m
=1.55 m
=1.50 m

coupling coefficient, (%)

4.0

elliptical ring
3.0

circular ring

2.0

1.0

0.0

0.20

0.22

0.24
0.26
0.28
gap width (m)

0.30

0.32

Figure 28: FDTD-computed coupling coe cient as a function of wavelength


and the gap width, g, for a 5.0- m-diameter circular ring resonator (dotted lines)
and an elliptical ring resonator with semimajor and semiminor axis lengths of
a = 6:0 m and b = 2:5 m, respectively (solid lines).
Next, consider the racetrack ring resonator of Figure 29 con gured with a gap
width of 0.3 m. In this study, the diameter of the semi-circular sections is xed at
d = 5:0 m. In Figure 30, the FDTD-computed coupling coe cient is plotted as a
function of frequency and straight-section length, L. Here, ranges between 0.3%
and 2.0% for values of L ranging from 0.0 (the case of the circular ring resonator
studied previously) to 2.1 m at = 1:5 m. Thus, a larger coupling coe cient is
achieved with the racetrack con guration than with the elliptical con guration of

71

L
d

Figure 29: Schematic of a microcavity ring resonator (racetrack con guration)


coupled to straight waveguides.
the same perimeter. Figure 31 compares the FDTD-computed for the racetrack
ring (L = 1:5 m, d = 5:0 m) and the previous 5.0- m-diameter circular ring
resonator as a function of the gap width, g.
As illustrated in Figures 28 and 31, a speci c coupling level at a given wavelength is achieved at a larger gap width with the elongated resonator in comparison
to the circular ring resonator. These results illustrate the potential advantage of
elongated ring structures in allowing wider air gaps while providing the desired
coupling levels.

72
6.0
L=0.0 m
L=0.3 m
L=0.6 m
L=0.9 m
L=1.2 m
L=1.5 m
L=1.8 m
L=2.1 m

coupling coefficient, (%)

5.0

4.0

3.0

2.0

1.0

0.0
175

200
frequency (THz)

225

Figure 30: FDTD-computed coupling coe cient (V-pol case) as a function of


frequency and straight-section length, L, for a racetrack ring resonator coupled to
straight waveguides (d = 5:0 m, g = 0:3 m).

7.5 Microcavity Ring Resonances


The focus of this section and the next is on the resonance behavior of microcavity
rings and disks 105]. The resonant nature of these structures is illustrated in
Figure 20. If the signal that enters port A is on-resonance with the ring or disk,
then that signal couples into the cavity from WG1, couples out from the cavity into
WG2, and exits the device at port C. Alternatively, a signal that is o -resonance
with the cavity remains in WG1 and exits at port B. The output at port B is often

73
5.0
=1.60 m
=1.55 m
=1.50 m

coupling coefficient, (%)

4.0

3.0

circular ring

racetrack ring

2.0

1.0

0.0

0.20

0.22

0.24
0.26
0.28
gap width (m)

0.30

0.32

Figure 31: FDTD-computed coupling coe cient as a function of wavelength


and the gap width, g, for a 5.0- m-diameter circular ring resonator (dotted lines)
and a racetrack ring resonator with 5.0- m-diameter semi-circular sections and
1.5- m-long straight sections (solid lines).
called the channel-dropping transmittance the output at port C is correspondingly
called the channel-passing transmittance. The rings and disks support both a
clockwise and a counterclockwise propagating mode. By sending a signal into
port A, only the counterclockwise propagating mode is directly excited. Backward
re ections in the coupling regions and backscattering due to sidewall roughness
can excite the counterpropagating mode.
Referring to Figure 20, consider a 5.0- m-diameter ring resonator with a gap
of 0:232 m between the ring and each adjacent straight waveguide. All straight
and ring waveguides are 0.3 m wide. Exciting WG1 in the same manner as in the

74
coupling studies, the simulation is continued in time as the pulse initially couples
into the ring and repeatedly travels around it. Figure 32 is a visualization of
snapshots in time of the FDTD-computed electric eld (V-pol) as the pulse rst
couples into the ring and completes one round trip. For best display dynamic
range, the gray scale in each snapshot is normalized to the maximum electric- eld
in that snapshot.

Figure 32: Visualization of the FDTD-computed initial coupling and circulation


of the exciting 20-fs optical carrier pulse around the 5.0- m-diameter microcavity
ring resonator.
The coupling of energy out to the two adjacent waveguides at each pass results

75
in a series of output pulses of diminishing amplitude at the output ports of WG1
and WG2. To determine the spectral properties of the transmittance at port B,
the FDTD simulation is run until the energy in the resonator and the amplitude of
the pulse train have su ciently died down. This permits convergence of the DFT
that is performed concurrently with the FDTD time-stepping.
Figure 33 graphs the calculated transmittance spectrum for this microcavity
ring. The Q of the mth resonance is obtained directly from the power spectrum by
forming the ratio of the center frequency, m , to the width of the resonance, m ,
at the half-power points. The resonant frequencies and Qs calculated from the data
in Figure 33 are listed in Table 2. The FSR values, also listed in Table 2, decrease
as the wavelength decreases. This behavior is a consequence of the relatively large
waveguide dispersion, as seen in (50). The nesse, calculated from the FSR values
and the widths of the resonances, falls in the range between 115 and 160.
1.0

normalized power

0.8

0.6

0.4

m=25

m=26

m=27

m=28

m=29

0.2

0.0
185

190

195
200
frequency (THz)

205

210

Figure 33: FDTD-computed transmittance spectrum of the 5.0- m-diameter microcavity ring resonator.

76

m
25
26
27
28
29

(THz) m (nm) Q FSR (nm)


185.85 1613.10 3700
50.66
191.88 1562.44 5000
47.56
197.90 1514.88 4000
44.73
203.92 1470.15 4000
42.17
209.94 1427.98 4000

Table 2: Resonance data from Figure 33 for the 5.0- m-diameter microcavity ring
resonator.

Impact of Sidewall Roughness


Based on only the coupling, which decreases at higher frequencies as seen in Figure 21, the Q should increase with frequency. Such behavior is observed for the
rst two resonances listed in Table 2 (m = 25 26), but then the apparent Q drops
for the higher-frequency resonances. This drop is due to stronger scattering at
higher frequencies where imperfections in the etched sidewalls (approximated in
the FDTD model using staircasing) have more of an e ect. In fact, on an expanded
frequency scale, it is clear that each of the m = 28 and m = 29 resonances actually
represents two subresonances. The sidewall roughness causes backscattering into
a counterpropagating mode which splits the resonance peak 103]. In the case of
the higher-frequency resonances of Figure 33, the subresonances are not split far
enough to be individually discernible, so the apparent Q of each overall resonance
is lowered.
The drop in the apparent Q is veri ed by simulating a microcavity ring res-

77
onator that has rougher sidewalls by a factor of two. This involves modifying the
staircased approximation of the sidewalls to mimic a twice-coarser grid in the vicinity of the walls, but actually retaining the ne 13.6-nm grid-cell size everywhere.
In this case, all ve resonances are split as a result of higher scattering losses across
the 25-THz spectrum. Further, the two highest-frequency resonances are split far
enough apart to individually resolve each subresonance. This study reveals that
the Q increases with increasing mode number for each subresonance, as expected
based on the strong dependence of coupling on frequency.

Rejection Ratio
The on/o ratio, or rejection ratio, is de ned here as the ratio of power transmitted
at a resonance frequency to the power not transmitted at that frequency. Consider
the 30-nm communications window of erbium ampli ers that is centered around
1.55 m. Focusing on this window, that is, around the m = 26 resonance (191.9
THz) in Figure 33, the rejection ratio is approximately 72:1. By slightly decreasing
the gap size, the rejection ratio can be increased to more than 500:1. In general,
this ratio can be optimized over any desired band spanning a couple of THz by
adjusting the gap size to change the coupling with respect to the scattering losses
in that wavelength range.

Extinction Ratio
Figure 34 illustrates the potential for high extinction ratios. First, the sinusoidal
steady-state electric eld is computed for an excitation at a nonresonant frequency
of 193.4 THz (1550 nm), shown in Figure 34a. In this case, nearly 100% of the
signal remains in WG1. Second, the sinusoidal steady-state electric eld is com-

78
puted for an excitation at the m = 26 resonant frequency of 191.9 THz. As shown
in Figure 34b, nearly 100% of the signal in this case is switched to WG2.

(a)
(b)
Figure 34: Visualizations of the FDTD-computed sinusoidal steady-state electric
eld in the 5.0- m-diameter ring having the properties summarized in Figures 32,
33, and Table 2. Here, a single-frequency excitation is applied at port A of WG1:
(a) o -resonance signal at 193.4 THz (b) on-resonance signal at 191.9 THz (the
m = 26 resonance).

E ective Diameter
Since ne is a known function of frequency for the straight waveguides, an estimate
of de can be obtained using mc= m ne for the ring resonator. For the resonances
in Figure 33, de increases with frequency from 4.717 to 4.720 m. If the circumferential path of the mode were along the center of the waveguide, de would be 4.7
m (halfway between the outer-rim diameter of 5.0 m and the inner-rim diameter
of 4.4 m). The estimate of de is slightly larger, agreeing with the expectation

79
that the peak of the transverse mode pro le shifts toward the outer edge of the
curved waveguide.

Overall Assessment
The 5.0- m-diameter microcavity ring resonator demonstrates narrow channel
wavelength selectivity (high nesse) and constant transmittance over FSR values as wide as 6.026 THz (50.66 nm). This structure exhibits excellent rejection
and extinction ratios. Its Q is limited primarily by the degree of sidewall roughness
which is a function of the fabrication process and its quality control.

7.6 Microcavity Disk Resonances


Referring again to Figure 20, consider a 5.0- m-diameter disk resonator with a gap
of 0:232 m between the disk and each adjacent straight 0.3- m-wide waveguide.
The disk geometry is identical to that of the 5.0- m-diameter ring studied above
except that there is no inner rim. Exciting WG1 in the same manner as in the
coupling and ring-resonance studies, the simulation is continued in time as the
pulse initially couples into the disk and repeatedly travels around it. Figure 35 is
a visualization of snapshots in time of the FDTD-computed electric eld (V-pol)
as the pulse rst couples into the disk and completes one round trip. Again, for
best display dynamic range, the gray scale in each snapshot is normalized to the
maximum electric- eld in that snapshot.
Figure 35 can be compared directly with Figure 32, the corresponding visualization for the ring resonator studied earlier. These gures illustrate that there
is a di erence in the group velocities between the ring and the disk. In the nal
frame of Figure 32, the pulse coupled into the ring is just short of completing one

80

Figure 35: Visualization of the FDTD-computed initial coupling and circulation


of the exciting 20-fs optical carrier pulse around the 5.0- m-diameter microcavity
disk resonator.
round trip after 240 fs. The pulse also maintains its single-mode transverse pro le.
However, the same pulse has a greater group velocity in the disk and therefore
propagates farther around the disk after 240 fs. Further, the pulse in the disk is
also broadened and is starting to trace out the whispering-gallery modes.
Figure 36 graphs the calculated transmittance spectrum for the microcavity
disk. The three sets of resonances in this gure correspond to the rst-order,
second-order, and third-order radial whispering-gallery modes of the disk. As the

81
radial order (the number of electric eld intensity maxima in the radial direction)
increases, the transmission characteristics worsen because the coupling between
the disk and the adjacent waveguides is not optimal for the higher-order radial
modes.
1.0
m=19

m=21

m=20

m=22

normalized power

0.8

0.6
m=24

m=23

m=25

0.4

0.2

m=29
m=27

0.0
185

m=30

m=28

190

195
200
frequency (THz)

205

210

Figure 36: FDTD-computed transmittance spectrum of the 5.0- m-diameter microcavity disk resonator.
Table 3 lists the resonant frequencies, Qs, and FSR values calculated from the
data in Figure 36. In this table, q indicates the radial order of the mode and \NA"
indicates that the resonance is too weak to obtain accurate data. In comparison
with the resonance data for the ring, the rst-order radial mode resonances for the
disk exhibit several interesting properties:
1. The locations of the resonances are at lower frequencies because the e ective
index of the fundamental whispering-gallery modes in the solid disk is very
di erent from that of the narrow-waveguide ring.
2. The FSR values are even larger that those for the ring because the lack of

82

q m
1 27
1 28
1 29
1 30
2 23
2 24
2 25
3 19
3 20
3 21
3 22

(THz) m (nm) Q FSR (nm)


189.20 1584.51 7600
51.32
195.54 1533.19 7500
47.98
201.85 1485.21 7000
45.03
208.16 1440.18 9100
191.29 1567.24 8700
52.69
197.94 1514.55 9900
49.12
204.58 1465.43 11000
187.83 1596.10 9900
57.09
194.80 1539.01 9700
52.96
201.74 1486.05 8800
(NA)
208.65 (NA) (NA)

Table 3: Resonance data from Figure 36 for the 5.0- m-diameter microcavity disk
resonator.
an inner rim allows the e ective diameter to be smaller for the fundamental
whispering-gallery modes of the disk than for the ring. This phenomenon
is also evident in comparing the FSR of the rst-order modes of the disk
with the FSR of the higher-order modes. As the order of the radial mode
increases, the peak of the mode moves toward the center of the disk. Therefore, de decreases even further for the higher-order modes. Just as with the
microcavity ring, the FSR decreases as the wavelength decreases within a
given set of microdisk resonances.

83
3. The linewidths are narrower than for the ring. The higher fundamental-mode
Qs of the disk resonator result from the reduced scattering loss arising from
the lack of an inner rim. In an ideal disk, the low-order modes have the
highest Q. However, since the lowest-order mode is con ned most closely to
the perimeter of the disk, it is most sensitive to sidewall roughness. Therefore,
as seen in Table 3, the higher-order mode resonances have slightly higher Qs
because of lower scattering loss.
To observe the whispering-gallery modes of the microdisk resonator, singlefrequency FDTD simulations are performed at the resonance frequencies pinpointed by Figure 36. Figure 37a shows the sinusoidal steady-state electric eld
pattern for an excitation at the nonresonant frequency =193.4 THz ( = 1550.0
nm). From the data graphed in Figure 36 and the data visualized in Figure 37a,
it can be shown that 99.98% of the power in the incident signal at this frequency
remains in WG1.
Figure 37b visualizes the steady-state electric eld pattern for an excitation at
=189.2 THz ( = 1584.5 nm). According to Figure 36, this frequency corresponds
to the m = 27, q = 1 resonance. The eld pattern within the microdisk illustrates
a rst-order radial whispering-gallery mode. This visualization also illustrates the
e ect of resonant eld enhancement inside the microcavity. Since the gray scale
here is normalized to the peak resonant elds, the elds in WG1 and WG2 appear
to be very weak. However, they are at the same level as the elds in WG1 in
Figure 37a. For this on-resonance case, 99.79% of the signal in WG1 is switched
to WG2.

84

(a)

(b)

(c)
(d)
Figure 37: Visualizations of the FDTD-computed sinusoidal steady-state electric
eld in the 5.0- m-diameter disk resonator coupled to straight 0.3- m-wide waveguides for single-frequency excitations at port A of WG1: (a) o -resonance signal
(193.4 THz) (b) on-resonance signal, rst-order radial mode (189.2 THz) (c)
second-order radial mode resonance (191.3 THz) and (d) third-order radial mode
resonance (187.8 THz).

85
In the same manner, Figure 37c visualizes the steady-state electric eld pattern
for an excitation at =191.3 THz ( = 1567.2 nm), the frequency of the m = 23,
q = 2 resonance in Figure 36. This graphic illustrates two electric eld oscillations
in the radial direction, and shows the e ect of the modal peak shifting away from
the disk edge.
Finally, Figure 37d shows the steady-state electric eld pattern for an excitation
at =187.8 THz ( = 1596.1 nm), the frequency of the m = 19, q = 3 resonance in
Figure 36. This graphic illustrates the third-order radial whispering-gallery mode
and further shows the progressive e ect of the modal peak shifting away from the
edge of the disk.
The visualizations of Figure 37 show how each resonant whispering-gallery
mode is con ned to an annulus around the perimeter of the disk wherein the
width of the annulus increases as the radial order increases. A comparison between Figures 37a and 37b indicates the potential for high extinction ratios (and
low crosstalk between channels) in these devices if the higher-order modes of the
microdisk can be suppressed.

7.7 Suppression of Higher-Order Radial Whispering-Gallery


Modes
As demonstrated above, microcavity disk resonators supporting whispering-gallery
modes have less sidewall scattering loss and higher Q in comparison to single-mode
microcavity rings. However, microdisks have multiple sets of resonances due to the
presence of fundamental and higher-order radial modes. Suppression of the higherorder modes is necessary to use the disks as single-mode laser sources or as WDM
devices having low crosstalk across a wide spectrum.

86
This section explores two methods for suppressing the higher-order radial modes
of microcavity disk resonators 105, 107]: (1) etching out the center of the disk and
(2) choosing the width of the adjacent waveguide to minimize coupling between
the waveguide mode and the higher-order modes of the disk. In each case, the
resonances are manipulated by taking advantage of the spatial characteristics of
the whispering- gallery modes of the disk.

Etching Out the Center of the Disk


The center of the disk can be etched out to create larger losses for the higherorder modes that propagate in that region. Here, it is very instructive to consider
a series of intermediate disk geometries that span the extremes of the solid disk
and the narrow-waveguide ring. Only the cavity geometry is perturbed the air
gap between the 0.3- m-wide waveguides and the microcavity remains constant
(g = 0:232 m).
Consider starting with the solid 5.0- m-diameter microcavity disk resonator.
For convenience, Figure 38a repeats the FDTD-computed transmittance of this
solid disk that was previously shown in Figure 36.
Now, assume that the center of the disk is etched out to form a wide ring
with an inner rim diameter di of 3.0 m, thereby yielding a waveguide width w of
1.0 m. Figure 38b shows the FDTD-computed transmittance for this structure.
The rst-order and second-order whispering-gallery-mode resonances remain unchanged from that of the solid disk. The e ect of creating an inner ring boundary is
to shift the third-order resonances toward higher frequencies and to lower their Q.
This begins to suppress the third-order radial modes that have elds concentrated
toward the center of the disk.

87
1.0
m=19

m=21

m=20

m=22

normalized power

0.8

0.6
(a)
m=24

m=23

m=25

0.4

0.2

m=29
m=27

0.0
185

m=30

m=28

190

195
200
frequency (THz)

205

210

1.0
m=22

m=21

m=20

m=19

normalized power

0.8

0.6
(b)
m=23

0.4

0.2

0.0
185

190

m=30

m=29

m=28

m=27

m=25

m=24

195
200
frequency (THz)

205

210

1.0

normalized power

0.8

0.6
(c)
0.4

0.2

0.0
185

m=24

m=23

m=22

m=27

190

m=28

m=25

m=29

195
200
frequency (THz)

m=30

205

210

Figure 38: Progression of the FDTD-computed transmittance of a 5.0- m-diameter


microcavity.

88
1.0

normalized power

m=22

m=21

0.8

m=23

m=20

0.6
(d)
0.4
m=30

m=29

m=28

m=27

0.2

0.0
185

190

195
200
frequency (THz)

205

210

1.0

normalized power

0.8

0.6
(e)
0.4

m=25

m=26

m=27

m=28

m=29

0.2

0.0
185

190

195
200
frequency (THz)

205

210

Figure 38: (continued) The resonator is coupled to straight 0.3- m-wide waveguides. The inner-rim diameter di is increased in four steps: (a) solid disk
(di = 3:5 m) (b) di = 3:0 m, ring waveguide width w = 1:0 m (c) di = 3:5 m,
w = 0:75 m (d) di = 4:0 m, w = 0:5 m (e) di = 4:4 m, w = 0:3 m.
In the next step, di is increased to 3.5 m so that w = 0:75 m. Figure 38c
shows the computed transmittance for this case. Here, the third-order-mode resonances are completely suppressed and each second-order resonance is shifted upward in frequency. Only the rst-order resonances remain una ected.
Proceeding further, di is increased to 4.0 m so that w = 0:5 m. In Fig-

89
ure 38d, the computed transmittance for this case shows substantial suppression
and upshifting of the second-order resonances. Further, the fundamental-mode
resonances are now slightly upshifted and their Qs slightly lowered.
Finally, di is increased to 4.4 m and w = 0:3 m. This yields the geometry
of the single-mode, narrow-waveguide ring studied in Section 7.5. Figure 38e repeats for convenience the computed transmittance for this ring previously shown
in Figure 33.
In summary, as the etched-out inner diameter of the disk increases, each set
of resonances monotonically shifts toward higher frequencies. This can be seen by
tracking the location of speci c resonances through the sequence of Figures 38a-e.
A smooth transition occurs between the multi-whispering-gallery-mode resonances
of the solid 5.0- m-diameter disk and the single-mode resonances of the 5.0- mdiameter ring formed by 0.3- m-wide waveguide.

Choosing the Width of the Adjacent Waveguide


Properly choosing the width of the adjacent waveguide minimizes coupling between
the waveguide mode and the higher-order whispering-gallery modes of the disk.
Consider the FDTD-computed transmittance spectra for the 5.0- m-diameter microcavity disk resonator coupled to two straight adjacent waveguides with an air
gap width of g = 0:2 m. Figure 39a shows transmittance results when the
adjacent waveguides are 0.2 m wide. Here, there are four sets of resonances corresponding to the rst four radial whispering-gallery modes of the disk. (Each
resonance is labeled with its radial number.)
In Figure 39b, the widths of the adjacent waveguides are increased to 0.3 m.
For this case, the fourth-order radial mode resonances disappear and the third-

90
0.8

1.0

(a)

normalized power

normalized power

1.0

q=4

0.6
0.4

q=3
q=2

0.2

q=1

0.0
1.55

1.56

1.57
1.58
wavelength (m)

1.59

(c)

q=2

normalized power

normalized power

q=3

0.6
0.4

q=2
0.2

q=1
1.56

1.57
1.58
wavelength (m)

1.59

1.60

1.0

0.6
0.4

q=1

0.2
0.0
1.55

(b)

0.0
1.55

1.60

1.0
0.8

0.8

1.56

1.57
1.58
wavelength (m)

1.59

1.60

0.8

q=2

(d)

0.6
0.4

q=1

0.2
0.0
1.55

1.56

1.57
1.58
wavelength (m)

1.59

1.60

Figure 39: FDTD-computed transmittance of a 5.0- m-diameter microcavity disk


resonator coupled to straight waveguides of the following widths: (a) 0.20 m (b)
0.30 m (c) 0.35 m and (d) 0.38 m.
order radial mode resonances are weak.
In Figure 39c, the widths of the adjacent waveguides are further increased to
0.35 m. Now, the third-order resonances are eliminated, and the second-ordermode resonances are weak.
Finally, in Figure 39d, the widths of the adjacent waveguides are increased
to 0.38 m. Here, the second-order resonances are strongly suppressed, verging
on extinction. As the waveguide width approaches this optimum value, the Qs
of the fundamental modes increase. However, without changing the gap size, the
minimum on-resonance transmission also increases. Once the optimum waveguide

91
width is determined from the standpoint of suppressing unwanted higher-order
modes, the gap size can be adjusted to give the best fundamental-mode transmission characteristics (namely, low on-resonance transmission and high Q).

8 Conclusions and Future Research


8.1 Summary

This dissertation has addressed FDTD computational modeling and design of


micron- and nanometer-scale integrated optics devices, speci cally microcavity
lasers and resonators, for high-density photonic integrated circuits. The primary
goals of this research were to develop and re ne FDTD algorithms for modeling the
electrodynamics of optical materials and structures, and to apply these techniques
to the emerging class of microcavity devices.
During the course of this research, FDTD formulations based on the ADE
approach were developed for modeling linear and nonlinear absorbing and gain
media 72, 80, 95]. Rigorous validations of these algorithms have demonstrated the
potential for high accuracy of these models over large bandwidths. The technique
for modeling a saturable homogeneously broadened gain medium was applied to
analyze the operation of VCSELs 95]. The comparison between uniform and
periodic gain con gurations within the microcavity illustrated the advantage of the
periodic-gain structure for providing lower lasing thresholds and higher e ciencies.
The FDTD approach proved to be much more accurate in quantifying the lasing
wavelength and gain threshold than previously used analytical approaches. Finally,
this dissertation presented detailed FDTD modeling results for waveguide-coupled
microcavity ring and disk resonators, and discussed the key design parameters and
tradeo s that were identi ed as a result of these modeling studies 105, 106, 107].

92

8.2 Future Work

93

The advent of advanced nanofabrication techniques has led to the realization of


optical resonators having physical dimensions of the order of the optical wavelength. These microcavities have been proposed for a rich variety of applications
in high-density photonic integrated circuits. The results reported in this dissertation indicate the utility of FDTD Maxwell's equations solvers for serving as useful
and practical design tools for this emerging class of integrated optical devices and
circuits. In fact, the usage of FDTD in micrometer- and nanometer-scale integrated
optics should eventually be similar in scope to current FDTD applications in the
engineering design of linear and nonlinear microwave circuits. The long-term goal
is to help bring about fundamental advances in the area of novel integrated optical
devices and circuits through the use of large-scale computational simulations.
Areas of future research include the following:
Further investigation of microcavity ring and disk resonators to determine
optimum designs that are less sensitive to fabrication variations.
Studies of all-optical switching e ects in high-Q microcavity ring and disk
resonators composed of nonlinear media.
Exploration of the class of microresonators and other nanometer-scale devices
based on photonic bandgap structures, including waveguide structures, power
splitters, and lasers.
Extension of the explicit FDTD-ADE model incorporating atomic rate equations 93] to model microcavity lasers and ultra-fast optical phenomena in
two- and three-dimensions.

References

1] S. E. Miller, \Integrated optics: an introduction," Bell Syst. Tech. J., vol.


48, pp. 2059{2069, Sept. 1969.
2] K. S. Yee, \Numerical solution of initial boundary value problems involving
Maxwell's equations in isotropic media," IEEE Trans. Antennas Propagat.,
vol. 14, no. 3, pp. 302{307, 1966.
3] T. Baba, \Photonic crystals and microdisk cavities based on GaInAsP-InP
systems," IEEE J. Selected Topics in Quantum Electronics, vol. 3, no. 3, pp.
808{830, June 1997.
4] H. Soda, K. Iga, C. Kitahara, and Y. Suematsu, \GaInAsP/InP surface
emitting injection lasers," Jpn. J. Appl. Phys., vol. 18, pp. 2329{2330, Dec.
1979.
5] K. Iga, F. Koyama, and S. Kinoshita, \Surface emitting semiconductor
lasers," IEEE J. Quantum Electron., vol. 24, pp. 1845{1855, 1988.
6] J. L. Jewell, A. Scherer, S. L. McCall, Y. H. Lee, S. Walker, J. P. Harbison,
and L. T. Florez, \Low-threshold electrically pumped vertical-cavity surfaceemitting microlasers," Electon. Lett., vol. 25, no. 17, pp. 1123{1124, Aug.
1989.
7] Y. H. Lee, J. L. Jewell, A. Scherer, S. L. McCall, J. P. Harbison, and L. T.
Florez, \Room-temperature continuous-wave vertical-cavity single-quantumwell microlaser diodes," Electon. Lett., vol. 25, no. 20, pp. 1377{1378, Sept.
1989.
8] J. L. Jewell, A. Scherer, B. Van der Gaag, L. M. Schiavone, J. P. Harbison,
and L. T. Florez, \Microcavity VCSELs," in Quantum Electronics and Laser
Science (QELS) Conference, Washington, DC, 1993, pp. 52{53.
9] E. Yablonovitch, \Photonic band-gap structures," J. Opt. Soc. Am. B, vol.
10, no. 2, pp. 283{295, Feb. 1993.
10] E. Yablonovitch, \Inhibited spontaneous emission in solid-state physics and
electronics," Phys. Rev. Lett., vol. 58, no. 20, pp. 2059{2062, May 1987.
11] E. Yablonovitch, T. J. Gmitter, R. D. Meade, A. M. Rappe, K. D. Brommer,
and J. D. Joannopoulos, \Donor and acceptor modes in photonic band
structure," Phys. Rev. Lett., vol. 67, pp. 3380{3383, 1991.
12] P. R. Villeneuve, S. Fan, J. D. Joannopoulos, K.-Y. Lim, G. S. Petrich, L. A.
Kolodziejski, and R. Reif, \Air-bridge microcavities," Appl. Phys. Lett., vol.
67, no. 2, pp. 167{169, July 1995.

94

95
13] J. P. Zhang, D. Y. Chu, S. L. Wu, W. G. Bi, R. C. Tiberio, R. M. Joseph,
A. Ta ove, C. W. Tu, and S. T. Ho, \Nanofabrication of 1-D photonic
bandgap structures along a photonic wire," IEEE Photon. Technol. Lett.,
vol. 8, no. 4, pp. 491{493, Apr. 1996.
14] E. A. J. Marcatili, \Bends in optical dielectric guides," Bell Syst. Tech. J.,
vol. 48, no. 5, pp. 2103{2132, Sept. 1969.
15] S. L. McCall, A. F. J. Levi, R. E. Slusher, S. J. Pearton, and R. A. Logan,
\Whispering-gallery mode microdisk lasers," Appl. Phys. Lett., vol. 60, no.
3, pp. 289{291, Jan. 1992.
16] A. F. J. Levi, R. E. Slusher, S. L. McCall, S. J. Pearton, and W. S. Hobson,
\Room-temperature lasing action in In0:51 Ga0:49P/In0:2 Ga0:8As microcylinder laser diodes," Appl. Phys. Lett., vol. 62, no. 17, pp. 2021{2023, Apr.
1993.
17] Lord Rayleigh, \The problem of the whispering gallery," Scienti c Papers
(Cambridge University, Cambridge, England), vol. 5, pp. 617{620, 1912.
18] M. Hovinen, J. Ding, A. V. Nurmikko, D. C. Grillo, J. Han, L. He, and
R. L. Gunshor, \Blue-green laser emission from ZnSe quantum well microresonators," Appl. Phys. Lett., vol. 63, no. 23, pp. 3128{3130, Dec. 1993.
19] U. Mohideen, W. S. Hobson, S. J. Pearton, F. Ren, and R. E. Slusher,
\GaAs/AlGaAs microdisk lasers," Appl. Phys. Lett., vol. 64, no. 15, pp.
1911{1913, Apr. 1994.
20] D. Y. Chu, S. T. Ho, X. Z. Wang, B. W. Wessels, W. G. Bi, C. W. Tu, R. P.
Espindola, and S. L. Wu, \Observation of enhanced photoluminescence in
erbium-doped semiconductor microdisk resonator," Appl. Phys. Lett., vol.
66, no. 21, pp. 2843{2845, May 1995.
21] B. Corbett, J. Justice, L. Considine, S. Walsh, and W. M. Kelly, \Lowthreshold lasing in novel microdisk geometries," IEEE Photon. Technol.
Lett., vol. 8, no. 7, pp. 855{857, July 1996.
22] A. F. J. Levi, R. E. Slusher, S. L. McCall, T. Tanbun-Ek, D. L. Coblentz,
and S. J. Pearton, \Room temperature operation of microdisc lasers with
submilliamp threshold current," Electon. Lett., vol. 28, no. 11, pp. 1010{
1011, May 1992.
23] A. F. J. Levi, R. E. Slusher, S. L. McCall, J. L. Glass, S. J. Pearton, and
R. A. Logan, \Directional light coupling from microdisk lasers," Appl. Phys.
Lett., vol. 62, no. 6, pp. 561{563, Feb. 1993.
24] D. Y. Chu, M. K. Chin, W. G. Bi, H. Q. Hou, C. W. Tu, and S. T. Ho,
\Double-disk structure for output coupling in microdisk lasers," Appl. Phys.
Lett., vol. 65, no. 25, pp. 3167{3169, Dec. 1994.

96
25] A. F. Jezierski and P. J. R. Layborn, \Integrated semiconductor ring lasers,"
IEE Proc. J, vol. 135, no. 1, pp. 17{24, Feb. 1988.
26] T. Krauss, P. J. R. Laybourn, and J. Roberts, \CW operation of semiconductor ring lasers," Electon. Lett., vol. 26, pp. 2095{2097, Dec. 1990.
27] J. P. Hohimer, D. C. Craft, G. R. Hadley, G. A. Vawter, and M. E. Warren,
\Single-frequency continuous-wave operation of ring resonator diode lasers,"
Appl. Phys. Lett., vol. 59, no. 26, pp. 3360{3362, Dec. 1991.
28] H. Han, M. E. Favaro, D. V. Forbes, and J. J. Coleman, \Inx Ga1;xAsAly Ga1;y As-GaAs strained-layer quantum-well heterostructure circular ring
lasers," IEEE Photon. Technol. Lett., vol. 4, no. 8, pp. 817{819, Aug. 1992.
29] J. P. Zhang, D. Y. Chu, S. L. Wu, W. G. Bi, R. C. Tiberio, C. W. Tu,
and S. T. Ho, \Directional light output from photonic-wire microcavity
semiconductor lasers," IEEE Photon. Technol. Lett., vol. 8, no. 8, pp. 968{
970, Aug. 1996.
30] D. Ra zadeh, J. P. Zhang, S. C. Hagness, A. Ta ove, K. A. Stair, R. Tiberio,
and S. T. Ho, \Waveguide-coupled AlGaAs/GaAs microcavity ring and disk
resonators with high nesse and 21.6-nm free spectral range," Opt. Lett.,
vol. 22, no. 16, pp. 1244{1246, 1997.
31] A. Dentai, J. Stone, E. C. Burrows, C. A. Burrus, L. W. Stulz, and M. Zirngibl, \Electrically-tunable semiconductor Fabry-Perot lter," IEEE Photon.
Technol. Lett., vol. 6, no. 5, pp. 629{631, May 1994.
32] T. Kominato, Y. Ohmori, N. Takato, H. Okazaki, and M. Yasu, \Ring resonators composed of GeO2 -doped silica waveguide," J. Lightwave Technol.,
vol. 10, no. 12, pp. 1781{1787, Dec. 1992.
33] R. Adar, M. R. Serbin, and V. Mizrahi, \Less than 1 dB per meter propagation loss of silica waveguides measured using a ring resonator," J. Lightwave
Technol., vol. 12, no. 8, pp. 1369{1372, Aug. 1994.
34] P. Heimala, P. Katila, J. Aarnio, and A. Heinamaki, \Thermally tunable
integrated optical ring resonator with poly-Si thermistor," J. Lightwave
Technol., vol. 14, no. 10, pp. 2260{2267, Oct. 1996.
35] S. Suzuki, K. Oda, and Y. Hibino, \Integrated-optic double-ring resonators
with a wide free spectral range of 100 GHz," J. Lightwave Technol., vol. 13,
no. 8, pp. 1766{1771, Aug. 1995.
36] G. Barbarossa, A. M. Matteo, and M. N. Armenise, \Theoretical analysis
of triple-coupler ring-based optical guided-wave resonator," J. Lightwave
Technol., vol. 13, no. 2, pp. 148{157, Feb. 1995.
37] B. E. Little, S. T. Chu, and H. A. Haus, \Micro-ring resonator channel
dropping lters," in Proc. IEEE Lasers and Electro-Optics Soc. Annual
Meeting, San Francisco, CA, Oct. 1995, vol. 2, pp. 233{234.

97
38] Y. Suematsu and K. Furuya, \Theoretical spontaneous emission factor of
injection laser," Trans. IECE Japan, vol. 60, no. 9, pp. 467{472, 1977.
39] Y. Yamamoto, S. Machida, K. Igeta, and G. Bjork, \Controlled spontaneous
emission in microcavity semiconductor lasers," in Coherence, Ampli cation,
and Quantum E ects in Semiconductor Lasers, Y. Yamamoto, Ed. John
Wiley and Sons, New York, 1991.
40] G. Bjork and Y. Yamamoto, \Analysis of semiconductor microcavity lasers
using rate equations," IEEE J. Quantum Electron., vol. 27, no. 11, pp.
2386{2396, Nov. 1991.
41] D. Marcuse, Principles of Quantum Electronics, Academic Press, New York,
1980.
42] A. Yariv, \Coupled mode theory for guided wave optics," IEEE J. Quantum
Electron., vol. 9, pp. 919{933, 1973.
43] D. Marcuse, Theory of Dielectric Optical Waveguides, 2nd ed., Academic
Press, Boston, 1991.
44] M. D. Feit and Jr. J. A. Fleck, \Computation of mode properties in optical
ber waveguides by a propagating beam method," Appl. Opt., vol. 19, no.
7, pp. 1164{1164, Apr. 1980.
45] H. Kogelnik and C. V. Shank, \Coupled wave theory of distributed feedback
lasers," J. Appl. Phys., vol. 43, pp. 2327{2335, 1972.
46] D. Yevick and B. Hermansson, \New formulation of the beam propagation
method: application of rib waveguides," IEEE J. Quantum Electron., vol.
25, pp. 221{229, Feb. 1989.
47] J. Van Roey, J. van der Donk, and P. E. Lagasse, \Beam-propagation
method: analysis and assessment," J. Opt. Soc. Am., vol. 71, no. 7, pp.
803{810, July 1981.
48] K. R. Umashankar, \Numerical analysis of electromagnetic wave scattering
and interaction based on frequency-domain integral equation and method of
moments techniques," Wave Motion, vol. 10, pp. 493{525, 1988.
49] M. N. O. Sadiku, \A simple introduction to nite element analysis of electromagnetic problems," IEEE Trans. Educ., vol. 32, no. 2, pp. 85{93, 1989.
50] W. Yang and A. Gopinath, \A boundary integral method for propagation
problems in integrated optical structures," IEEE Photonics Technol. Lett.,
vol. 7, no. 7, pp. 777{779, 1995.
51] L. Bersiner, U. Hempelmann, and E. Strake, \Numerical analysis of passive
integrated-optical polarization splitters: comparison of nite-element and
beam-propagation method results," J. Opt. Soc. Am. B, vol. 8, no. 2, pp.
422{433, Feb. 1991.

98
52] A. Ta ove and M. E. Brodwin, \Numerical solution of steady-state electromagnetic scattering problems using the time-dependent Maxwell's equations," IEEE Trans. Microwave Theory Tech., vol. MTT-23, no. 8, pp. 623{
630, 1975.
53] A. Ta ove, Computational Electrodynamics: The Finite-Di erence TimeDomain Method, Artech House, Boston, MA, 1995.
54] K. S. Kunz and R. J. Luebbers, The Finite Di erence Time Domain Method
for Electromagnetics, CRC Press, Boca Raton, FL, 1993.
55] K. L. Shlager and J. B. Schneider, \A selective survey of the nite-di erence
time-domain literature," IEEE Antennas Propagat. Magazine, vol. 37, no. 4,
pp. 39{56, 1995.
56] S.-T. Chu and S. K. Chaudhuri, \A nite-di erence time-domain method
for the design and analysis of guided-wave optical structures," J. Lightwave
Technol., vol. 7, no. 12, pp. 2033{2038, 1989.
57] J.-P. Berenger, \A perfectly matched layer for the absorption of electromagnetic waves," J. Comput. Phys., vol. 114, no. 1, pp. 185{200, 1994.
58] D. S. Katz, E. T. Thiele, and A. Ta ove, \Validation and extension to three
dimensions of the Berenger PML absorbing boundary condition for FD-TD
meshes," IEEE Microwave Guided Wave Lett., vol. 4, no. 8, pp. 268{270,
1994.
59] J. Yamauchi, M. Mita, S. Aoki, and H. Nakano, \Analysis of antire ection coatings using the FD-TD method with the PML absorbing boundary
condition," IEEE Photonics Technol. Lett., vol. 8, no. 2, pp. 239{241, 1996.
60] S. C. Hagness, A. Ta ove, and J. E. Bridges, \Wideband ultralow reverberation antenna for biological sensing," Electron. Lett., vol. 33, no. 19, pp.
1594{1595, 1997.
61] C. E. Reuter, R. M. Joseph, E. T. Thiele, D. S. Katz, and A. Ta ove, \Ultrawideband absorbing boundary condition for termination of waveguiding
structures in FD-TD simulations," IEEE Microwave Guided Wave Lett., vol.
4, no. 10, pp. 344{346, 1994.
62] H. Kogelnik and V. Ramaswamy, \Scaling rules for thin- lm optical waveguides," Appl. Opt., vol. 13, no. 8, pp. 1857{1862, Aug. 1974.
63] A. Ta ove, \Review of the formulation and applications of the nitedi erence time-domain method for numerical modeling of electromagnetic
wave interactions with arbitrary structures," Wave Motion, vol. 10, no. 6,
pp. 547{582, 1988.

99
64] R. J. Luebbers, F. Hunsberger, K. S. Kunz, R. B. Standler, and M. Schneider, \A frequency-dependent nite-di erence time-domain formulation for
dispersive materials," IEEE Trans. Electromagn. Compat., vol. 32, no. 3, pp.
222{227, 1990.
65] M. D. Bui, S. S. Stuchly, and G. I. Costache, \Propagation of transients in
dispersive dielectric media," IEEE Trans. Microwave Theory Tech., vol. 39,
no. 7, pp. 1165{1172, 1991.
66] R. J. Luebbers, F. Hunsberger, and K. S. Kunz, \A frequencydependent nite-di erence time-domain formulation for transient propagation in plasma," IEEE Trans. Antennas Propagat., vol. 39, no. 1, pp. 29{34,
1991.
67] F. Hunsberger, R. J. Luebbers, and K. S. Kunz, \Finite-di erence timedomain analysis of gyrotropic media. I: Magnetized plasma," IEEE Trans.
Antennas Propagat., vol. 40, no. 12, pp. 1489{1495, 1992.
68] R. J. Luebbers and F. Hunsberger, \FDTD for N th-order dispersive media,"
IEEE Trans. Antennas Propagat., vol. 40, no. 11, pp. 1297{1301, 1992.
69] D. F. Kelley and R. J. Luebbers, \Piecewise linear recursive convolution for
dispersive media using FDTD," IEEE Trans. Antennas Propagat., vol. 44,
no. 6, pp. 792{797, 1996.
70] R. Siushansian and J. LoVetri, \A comparison of numerical techniques for
modeling electromagnetic dispersive media," IEEE Microwave Guided Wave
Lett., vol. 5, no. 12, pp. 426{428, 1995.
71] T. Kashiwa and I. Fukai, \A treatment by the FD-TD method of the dispersive characteristics associated with electronic polarization," Microwave Opt.
Technol. Lett., vol. 3, no. 6, pp. 203{205, 1990.
72] R. M. Joseph, S. C. Hagness, and A. Ta ove, \Direct time integration of
Maxwell's equations in linear dispersive media with absorption for scattering
and propagation of femtosecond electromagnetic pulses," Optics Lett., vol.
16, no. 18, pp. 1412{1414, 1991.
73] P. G. Petropoulos, \Stability and phase error analysis of FD-TD in dispersive
dielectrics," IEEE Trans. Antennas Propagat., vol. 42, no. 1, pp. 62{69, 1994.
74] L. J. Nickisch and P. M. Franke, \Finite-di erence time-domain solution of
Maxwell's equations for the dispersive ionosphere," IEEE Antennas Propagat. Magazine, vol. 34, no. 5, pp. 33{39, 1992.
75] J. L. Young, \A full nite di erence time domain implementation for radio
wave propagation in a plasma," Radio Sci., vol. 29, no. 6, pp. 1513{1522,
1994.

100
76] S. C. Hagness, unpublished work (codes were developed for researchers at
the Battelle Paci c Northwest Laboratories for the modeling of optical wave
propagation in metals).
77] J. L. Young, \Propagation in linear dispersive media: Finite di erence timedomain methodologies," IEEE Trans. Antennas Propagat., vol. 43, no. 4,
pp. 422{426, 1995.
78] M. Okoniewski, M. Mrozowski, and M. A. Stuchly, \Simple treatment of
multi-term dispersion in FDTD," IEEE Microwave Guided Wave Lett., vol.
7, no. 5, pp. 121{123, 1997.
79] T. O. Korner and W. Fichtner, \Auxiliary di erential equation: E cient
implementation in the nite-di erence time-domain method," Optics Letters,
vol. 22, no. 21, pp. 1586{1588, 1997.
80] P. M. Goorjian, A. Ta ove, R. M. Joseph, and S. C. Hagness, \Computational modeling of femtosecond optical solitons from Maxwell's equations,"
IEEE J. Quantum Electron., vol. 28, no. 10, pp. 2416{2422, 1992.
81] R. W. Ziolkowski and J. B. Judkins, \Full-wave vector Maxwell equation
modeling of the self-focusing of ultrashort optical pulses in a nonlinear Kerr
medium exhibiting a nite response time," J. Opt. Soc. Am., B Opt. Phys.,
vol. 10, no. 2, pp. 186{198, 1993.
82] R. M. Joseph, P. M. Goorjian, and A. Ta ove, \Direct time integration of
Maxwell's equations in two-dimensional dielectric waveguides for propagation
and scattering of femtosecond electromagnetic solitons," Optics Lett., vol.
18, no. 7, pp. 491{493, 1993.
83] R. M. Joseph and A. Ta ove, \Spatial soliton de ection mechanism indicated
by FD-TD Maxwell's equations modeling," IEEE Photonics Technol. Lett.,
vol. 6, no. 10, pp. 1251{1254, 1994.
84] R. M. Joseph and A. Ta ove, \FDTD Maxwell's equations models for nonlinear electrodynamics and optics," IEEE Trans. Antennas Propagat., vol.
45, no. 3, pp. 364{374, 1997.
85] R. W. Ziolkowski and J. B. Judkins, \Applications of the nonlinear nite
di erence time domain (NL-FDTD) method to pulse propagation in nonlinear media: Self-focusing and linear-nonlinear interfaces," Radio Sci., vol. 28,
no. 5, pp. 901{911, 1993.
86] R. W. Ziolkowski and J. B. Judkins, \Nonlinear nite-di erence time-domain
modeling of linear and nonlinear corrugated waveguides," J. Opt. Soc. Am.,
B Opt. Phys., vol. 11, no. 9, pp. 1565{1575, 1994.
87] S. A. Basinger and D. J. Brady, \Finite-di erence time-domain modeling of
dispersive nonlinear Fabry-Perot cavities," J. Opt. Soc. Am., B Opt. Phys.,
vol. 11, no. 8, pp. 1504{1511, 1994.

101
88] C. Hulse and A. Knoesen, \Dispersive models for the nite-di erence timedomain method: Design, analysis, and implementation," J. Opt. Soc. Am.,
A Optics Image Sci. Vision, vol. 11, no. 6, pp. 1802{1811, 1994.
89] D. M. Sullivan, \Frequency-dependent FDTD methods using Z transforms,"
IEEE Trans. Antennas Propagat., vol. 40, no. 10, pp. 1223{1230, 1992.
90] D. M. Sullivan, \Nonlinear FDTD formulations using Z transforms," IEEE
Trans. Microwave Theory Tech., vol. 43, no. 3, pp. 676{682, 1995.
91] R. J. Hawkins and J. S. Kallman, \Lasing in tilted-waveguide semiconductor
laser ampli ers," Opt. Quantum Electron., vol. 26, pp. S207{S217, 1994.
92] R. W. Ziolkowski, J. M. Arnold, and D. M. Gogny, \Ultrafast pulse interactions with two-level atoms," Phys. Rev. A, vol. 52, no. 4, pp. 3082{3094,
1995.
93] A. S. Nagra and R. A. York, \FDTD analysis of wave propagation in nonlinear absorbing and gain material," IEEE Trans. Antennas Propagat., vol.
46, no. 3, pp. 334{340, Mar. 1998.
94] R. J. Hawkins and J. S. Kallman, \Linear electronic dispersion and nitedi erence time-domain calculations: A simple approach," J. Lightwave Technol., vol. 11, no. 11, pp. 1872{1874, 1993.
95] S. C. Hagness, R. M. Joseph, and A. Ta ove, \Subpicosecond electrodynamics of distributed Bragg re ector microlasers: Results from nite di erence
time domain simulations," Radio Sci., vol. 31, no. 4, pp. 931{941, 1996.
96] S. W. Corzine, R. S. Geels, J. W. Scott, R.-H. Yan, and L. A. Coldren, \Design of Fabry-Perot surface-emitting lasers with a periodic gain structure,"
IEEE J. Quantum Electron., vol. 25, no. 6, pp. 1513{1524, June 1989.
97] Y. Suematsu and A. R. Adams, Handbook of Semiconductor Lasers and
Photonic Integrated Circuits, Chapman and Hall, London, 1994.
98] G. P. Agrawal and N. K. Dutta, Semiconductor Lasers, 2nd ed., Van Nostrand Reinhold, New York, 1993.
99] S. D. Stearns and R. David, Signal Processing Algorithms Using Fortran and
C, PTR Prentice Hall, Englewood Cli s, NJ, 1993.
100] M. Born and E. Wolf, Principles of Optics, Pergamon Press, Oxford, 1980.
101] S. C. Hagness, S. T. Ho, and A. Ta ove, \Finite-di erence time-domain
(FDTD) computational electrodynamic simulations of microlaser cavities in
one and two spatial dimensions," in Computational Electromagnetics and Its
Applications, M. D. Salas, R. A. Nicolaides, and T. G. Campbell, Eds., pp.
229{251. Kluwer Academic Publishers, Netherlands, 1997.

102
102] B. E. Little and S. T. Chu, \Estimating surface-roughness loss and output
coupling in microdisk resonators," Opt. Lett., vol. 21, no. 17, pp. 1390{1392,
Sept. 1996.
103] B. E. Little, J.-P. Laine, and S. T. Chu, \Surface-roughness-induced contradirectional coupling in ring and disk resonators," Opt. Lett., vol. 22, no.
1, pp. 4{6, Jan. 1997.
104] R. E. Slusher, A. F. J. Levi, U. Mohideen, S. L. McCall, S. J. Pearton, and
R. A. Logan, \Threshold characteristics of semiconductor microdisk lasers,"
Appl. Phys. Lett., vol. 63, no. 10, pp. 1310{1312, Sept. 1993.
105] S. C. Hagness, D. Ra zadeh, S. T. Ho, and A. Ta ove, \FDTD microcavity simulations: Design and experimental realization of waveguide-coupled
single-mode ring and whispering-gallery-mode disk resonators," J. Lightwave
Technol., vol. 15, no. 11, pp. 2154{2165, 1997.
106] S. C. Hagness, D. Ra zadeh, S. T. Ho, and A. Ta ove, \FDTD analysis
and comparison of circular and elongated ring designs for waveguide-coupled
microcavity ring resonators," in Integrated Photonics Research Technical
Digest, 1998, Victoria, B.C., Canada, Mar. 1998, vol. 4, pp. 227{229.
107] S. C. Hagness, D. Ra zadeh, S. T. Ho, and A. Ta ove, \Suppression of
higher-order radial whispering gallery modes in waveguide-coupled microcavity disk resonators," in Proc. IEEE Lasers and Electro-Optics Soc. Annual
Meeting, San Francisco, CA, Nov. 1997, vol. 2, pp. 160{161.

Anda mungkin juga menyukai