Introduction To Slab Dielectric Waveguides
Uploaded by
qfascxvIntroduction To Slab Dielectric Waveguides
Uploaded by
qfascxvNotes on Integrated Optics
Introduction to
Slab Dielectric Waveguides
Prof. Elias N. Glytsis
November 19, 2020
School of Electrical & Computer Engineering
National Technical University of Athens
c 2020 Prof. Elias N. Glytsis
This page was intentionally left blank......
Contents
1 Introduction 1
2 Ray Approach for Guided Modes 5
2.1 Alternate Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3 Electromagnetic Approach for Guided Modes 9
3.1 TE Guided Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 TM Guided Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4 Substrate Modes 18
5 Radiation Modes 20
6 Unphysical Modes (or Nonphysical Modes) 22
7 Evanescent Modes 23
8 Normalized Slab-Waveguide Variables 26
9 Cutoff Conditions 28
10 Power Considerations 32
10.1 TE Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
10.2 TM Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
11 Multi-Layered Slab Waveguides 36
11.1 TE Guided Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
11.2 TM Guided Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
11.3 TE and TM Guided Modes using Chen’s Approach . . . . . . . . . . . . . . . . 42
12 Finite-Difference Frequency-Domain Method 44
12.1 FDFD Method Based on Yee’s Cell . . . . . . . . . . . . . . . . . . . . . . . . . 47
12.1.1 Average-Scheme Modification for Arbitrary Boundary Location . . . . . 55
13 Graded-Index Slab Waveguides 61
14 Lossy, Leaky, and Active Planar Multilayer Waveguides 72
14.1 Basic Theory of the Principle Argument Method (APM) . . . . . . . . . . . . . 72
14.2 Basic Theory of the Abd-ellal, Delves, and Reid (ADR) Method . . . . . . . . . 74
14.3 Basic Theory of the Derivative-Free Zero-Extraction by Phase-based Enclosure
Method (DFZEPE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
14.4 Application of APM, ADR, and DFZEPE methods to Planar Multilayer Waveg-
uides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
14.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
14.5.1 Lossless Dielectric Waveguide . . . . . . . . . . . . . . . . . . . . . . . . 86
3
14.5.2 Low-Loss Dielectric Waveguide . . . . . . . . . . . . . . . . . . . . . . . 88
14.5.3 Active Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
14.5.4 Leaky Modes of an ARROW Waveguide . . . . . . . . . . . . . . . . . . 91
References 97
4
This page was intentionally left blank......
INTRODUCTION TO
SLAB DIELECTRIC WAVEGUIDES
1. Introduction
Planar waveguides are critical elements in present day optical devices such as modulators, semi-
conductor lasers, couplers, wavelength filters, polarizers, and optical interconnects. Passive
waveguides, electro-optic components, transmitters, receivers, active waveguides, and driving
electronics can be integrated into an optical/electronics chip using planar technology, similar to
microelectronics. Optical waveguides are the fundamental elements that interconnect the vari-
ous devices of an photonic integrated circuit, just as a metallic strips make interconnections in
an electronic integrated circuit. Although the operation of waveguide devices is well researched
and understood, their particular performance relies on many parameters such as geometry,
freespace wavelength, initial field distribution, material characteristics, as well as electro-optic,
acousto-optic, or other driving conditions. The device parameters must be optimized before
fabricating a well operating device. With large-scale optoelectronic circuits, accurate modeling
is predominant because of the numerous resources required to fabricate a chip. Optical waveg-
uide design relies on simulating the propagation of light signals, determining the waveguide
modes, evaluating the mode coupling, and assessing the loss and gain of the structure. Un-
like electrical current that flows through a metal strip (wire) according to Ohm’s law, optical
waves travel in the waveguide in characteristic optical modes. A mode can be defined as a
spatial distribution of optical energy in one or more dimensions that remains constant in space
perpendicularly to the direction of propagation. A mode is an eigenfunction of the Maxwell’s
equations for the waveguiding structure.
For the successful design and optimization of waveguiide-based devices, robust and reliable
numerical methods are needed in order to determine accurately the propagation characteris-
tics of their modes. For example, the determination of guided-mode characteristics in optical
waveguides comprised of lossless and/or lossy materials, results in the solution of a transcen-
dental complex equation [1]. In addition, the design of semiconductor lasers/amplifiers require
the utilization of both lossy and active (for gain) materials [2]. Recently, there has been a
lot of interest in non-Hermitian optical structures where both gain and loss are present [3, 4].
1
Furthermore, in waveguiding structures, the determination of leaky-mode characteristics that
can compactly model the radiation field could be important. A practical example of this is
found in the antiresonant reflecting optical waveguides (ARROW) [5, 6]. ARROW structures
have received significant attention due to their low transmission loss, high polarization sensi-
tivity, as well as compatibility with optical fibers for telecommunications applications. The
determination of complex, in general, eigenvalues facilitates the evaluation of the correspond-
ing field distributions, power density, radiation losses, modal gain (for active waveguides) and
possibly other parameters of interest. For this reason it has been a great interest over the
last few decades to find reliable and robust techniques for calculating the zeros of an analytic
function in the complex plane. Common general numerical analysis methods that are used are
the Newton’s method (where the function’s derivative is used) and Müller’s method (where the
function’s derivative is not needed) [7–9]. However, these methods require a good estimate of
the root of the complex function which in most practical problems is not available. Further-
more, usually not even the number of zeros is known. A successful numerical method should be
able to extract all the zeros of the complex function in a domain of interest without any prior
knowledge of either their number or their approximate location.
One of the first numerical methods for the solution of complex transcendental equations is
based on the Cauchy integration method introduced by Delves and Lyness [10]. This technique
finds the number of zeros and approximates the roots of the function in a specified domain by
evaluating a polynomial with the same roots. The polynomial roots could be used as estimates
for further root improvements via iterative methods. The first applications of the approach
to multilayer planar waveguides using the transfer matrix approach [11–13] was presented by
Smith et al. [14, 15] and Anemogiannis and Glytsis (known as the argument principal method
- APM) [16]. Since then similar methods appear in the literature. Chen et al. [17] used
APM for multilayer lossy anisotropic waveguides, Kwon and Shin applied APM to isolated
roots [18, 19], and Michalski and Mustafa also utilized APM in conjunction to either transfer-
or scatter-matrix formulation and automatic differentiation [20]. These methods are based on
the numerical evaluation of the Cauchy integrals of the waveguide dispersion equation and
require the derivative of the dispersion equation that can be calculated either numerically [21]
or analytically. Some alternative methods that have been proposed in the literature are the
reflection-pole method [22, 23], the wavevector density method [23], a variational method [24],
2
a method based on smooth transition between bound modes of a closed waveguide and leaky
modes of an open waveguide [25], an iterative ”downhill” method [26], and a two-dimensional
minimization [27], to mention a few. Of course the fully numerical methods based on finite
elements [28, 29] or finite differences [30, 31] should also be mentioned. These latter methods
can also be applied to three-dimensional waveguides (channel waveguides) since they represent
completely numerical solution of the Maxwell’s equations. However, they require specialized
boundary layers (such as Perfectly Matched Layers [32]) around the structure in order to treat
open dielectric waveguides.
One of the drawbacks of the APM method is the use of the derivative of the function which
makes it rather cumbersome especially when the derivative cannot be calculated analytically.
Therefore, there has been significant effort to focus on globally reliable methods that find the
zeros of a complex function without the use of its derivative. Anemogiannis et al. [33] first
proposed the ADR method which is based on an algorithm provided by Abd-ellal, Delves, and
Reid [34] (the ADR acronym was derived from their last names initials). The ADR method
uses integrals (representing function moments) on a contour enclosing a specified domain that
do not involved the derivative of the function. The ADR method requires though a double
number of integrals as compared to APM method in order to determine the coefficients of the
approximating polynomial. In addition, it has some difficulty in estimating the actual number
of zeros inside the complex-plane domain under investigation. Ying and Katz [35] proposed a
simple method to determine the exact number of zeros of an analytic function based on the its
winding number in a bounded domain. Based on Ying and Katz method Kravanja and Van
Barel [36] and Gillan et al. [37] introduced a reliable derivative-free method for determining
the roots of an analytic function in a bounded complex domain. Their method is based on
calculating similar integrals as in ADR and using orthogonal polynomial theory to estimate
the zeros of the function by solving a general eigenvalue problem. Refinement of the zeros
is achieved using the Halley/Aitken method [37]. Recently Semwal and Rastogi applied the
Gillan’s method to multilayer planar waveguides [38].
Another simple method proposed recently [39], is based on Ying and Katz [35] winding num-
ber evaluation and uses zero enclosing by evaluating only the winding number of the function
in successively shrinking rectangles until every root (function’s zero) has been surrounded by a
pre-specified dimensions corresponding rectangle in the complex plane. Then the center of each
3
rectangle can serve as an estimate of the zero of the function in the complex plane. The winding
number around each shrinking rectangle is calculated by using only the phase of the function
and without any integral evaluation or derivative knowledge. Then, the Müller’s algorithm [9]
with deflation [8] is used for the iterative refinement of the root. Root bracketing by using the
winding number was also proposed by Ying and Katz [40], Dellnitz et al. [41] and Kwon and
Shin [18] but in conjuction to the Cauchy’s integrals. Furthermore, very recently Kowalczyk and
Marynowski [42] have used the winding number approach on a self-adaptive triangular-based
mesh to find the roots of an analytic function for radiation and propagation electromagnetic
problems. This approach is purely based on the successive enclosing by shrinking rectangles
of each zero (complex propagation constant) of the dispersion function of multilayer planar
waveguides and its based exclusively on the phase only of the function without using any inte-
grals or approximating polynomials. For brevity this method will be referred as Derivative-Free
Zero-Extraction by Phase-based Enclosure (DFZEPE). The DFZEPE approach is much simpler
than the adaptive Delaunay triangulation method proposed by Kowalczyk [42, 43].
In this chapter, the fundamental theory of optical modes in a waveguiding structures is
developed in order to acquire a general understanding of the nature of light propagation in an
optical waveguides. In the first part of the chapter the simple single-film planar waveguide will
be analyzed using both a ray approach as well as a fully electromagnetic approach. Then the
classification of the modes of a planar optical waveguide into TE and TM will be introduced.
Guided modes, substrate modes, radiation modes, evanescent modes, and unphysical modes
will be discussed. Then, the normalized variable will be introduced along with the cut-off
conditions. Power considerations will also be discussed. Later in the chapter, the multilayer
planar waveguides are presented in conjunction to the transfer-matrix technique. The usage of
the Finite-Differences Frequency-Domain (FDFD) method for the study of planar multilayer
waveguides is also explained with example cases. Then the graded-ndex waveguide analysis is
shown. Finally, s short review of the most useful methods for extracting complex in general
propagation constants for lossy and active waveguides will presented along with the definition of
the leaky modes which are not formal eignesolutions of the Maxwell’s equations. Some excellent
books exist on the topic on optical waveguides including [1, 13, 44–50] that may include parts
of the topics discussed in this chapter.
4
2. Ray Approach for Guided Modes
The basic geometry of a dielectric slab waveguide is shown in Fig. 1. The light is confined along
the x-axis and propagates along the z-axis in the coordinate system shown. The waveguide
is assumed to be uniform along the y-axis and therefore this comprises a two-dimensional
electromagnetic problem. The basic principle behind the concept of light guiding in dielectric
waveguide is the phenomenon of total internal reflection. If light can be launched into the
film layer of the waveguide it can remain in the film, provided that the angle of incidence θ
between the film-cover and the film-substrate regions is greater than both of the corresponding
critical angles. The refractive indices of the three regions are nc , nf , and ns for cover, film,
and substrate regions, respectively while the film thickness is h and the free-space wavelength
of the light that can travel guided within this structure is λ0 . Of course it is necessary that
nf > max{nc , ns } in order to guarantee that total internal reflection can occur in both film-
substrate and film-cover boundaries. The angle θ, known also as the zig-zag angle, for a specific
mode should satisfy the condition max{θcr,f s, θcr,f c} < θ < π/2, where θcr,f s = sin−1 (ns /nf ) and
θcr,f c = sin−1 (nc /nf ), are the critical angles for the film-substrate and film-cover boundaries,
respectively. Without loss of generality, in the following discussion, it is assumed that ns ≥ nc
(which is usually the most practical case). In the special case that ns = nc the waveguide is
characterized as a symmetric slab waveguide.
Figure 1: The geometric configuration of a three-region slab dielectric waveguide.
It is well known that there is an induced phase shift upon total internal reflection at both
5
boundaries. Therefore there should be a self-consistency condition that must be satisfied in
order for the plane waves shown in the ray picture of Fig. 1 to comprise a valid solution for a
guided mode.
In order to quantify the self-consistency condition Fig. 2 will be used [47]. In this figure
two possible rays are shown along with two wavefronts the one specified by AB and the other
by DC. The wavefronts AB and DC are selected to be at an infinitesimal distance away from
the film-cover and film-substrate boundaries, respectively. This implies that the ray from point
B to point D has not suffered any total internal reflections. In contrast, the ray from point
A to point C has suffered two total internal reflections at the two boundaries. Following the
phases accumulated by points A and B of the initial wavefront as the rays move towards to
their positions C and D of the second wavefront shown, it is necessary to require that the
accumulated phase from A → C and from B → D should differ at most by 2πν where ν is an
integer. This condition will guarantee that the second wavefront will remain a valid wavefront
of the guided mode. Mathematically, the previous conditions can be written as follows
−k0 nf (AC) + 2φpf s + 2φpf c − {−k0 nf (BD)} = 2πν, ν = 0, ±1, ±2, · · · (1)
| {z } | {z }
phase from A → C phase from B → D
where k0 = 2π/λ0 , and φpf s and φpf c (p = TE or TM) are the phase shifts that occur upon total
internal reflection at the film-substrate and film-cover boundaries respectively. These phase
shifts are functions of the angle θ and are given by the equations (as it was presented in the
Review of Electromagnetic Principles notes)
q
n2f sin2 θ − n2w
TE
φf w (θ) = tan−1 , for TE Polarization, (2)
nf cos θ
q
n2 n2f sin2 θ − n2w
TM f
φf w (θ) = tan−1 2
, for TM Polarization, (3)
nw nf cos θ
where w = c or s.
The distances (AC) and (BD) can be easily determined from the geometry shown in Fig.
2. Specifically, (AC) = h/ cos θ and (BD) = (AD) sin θ = [(OA) − (OD)] sin θ = [h tan θ −
h/ tan θ] sin θ = h[sin2 θ − cos2 θ]/ cos θ. Replacing the previous expressions in Eq. (1) and
6
0 2 3
! " # $ %
+ , - . /
& ' ( & ) % * ) $
Figure 2: The slab waveguide geometric configuration showing two rays and and two selective wavefronts
(AB and DC). For self-consistency it is necessary that the accumulated phase from A → C and from B → D
should differ by 2πν where ν is an integer. The points A and C are assumed to be infinitesimaly before/after
the film-cover/film-substrate boundaries respectively.
changing the signs assigning only positive (or zero) values for ν the following equation is derived
2k0 nf h cos θ − 2φpf s (θ) − 2φpf c (θ) = 2πν, ν = 0, 1, 2, · · · (4)
where p = TE or TM . Of course solution for θ of the previous equation has meaning for a guided
mode only when max{θcr,f s, θcr,f c} < θ < π/2. This equation can be solved only numerically
(using for example the bisection method). For every value of ν solutions for θνp (p = TE or TM
) can determined. There can be none, one, or multiple solutions depending on the parameters
of the waveguide and the free-space wavelength. A graphical representation of the solution is
shown in Fig. 3 for the example case where λ0 = 1.0µm, h = 1.2µm, nc = 1.0, nf = 2.2, and
ns = 1.5. The solutions for the zig-zag angles θνp can be visualized as the intersection of two
curves representing the functions f1 (θ) = 2k0 nf h cos θ and f2(θ) = 2φpf s (θ) + 2φpf c (θ) + 2πν (p =
TE or TM). In this particular example there are 4 TE modes (TE0 , TE1 , TE2 , and TE3 ) and
4 TM modes (TM0 , TM1 , TM2 , and TM3) that can be supported. It can be observed that the
TE TM TE TM
solutions θν and θν satisfy the inequality θν > θν (for any ν) due to the larger phase shift
TE TM
[φf w (θ) < φf w (θ), w = c or s] upon total internal reflection for the TM polarization.
In general, the solution for the zig-zag angle θ = θνp (where p = TE or TM ) is characteristic
of the guided mode. The associated to θνp electromagnetic field represents the guided mode TEν
or TMν and has a characteristic profile. The complete electromagnetic field will be determined
7
0
= 1 m, n c = 1, n f = 2.2, n s = 1.5, h = 1.2 m
40 40
30 30
Function f 1( )
Function f 2( )
20 20
10 10
TE
TM
0 0
0 15 30 45 60 75 90
Angle ,(deg)
Figure 3: The graphical representation of the solution of the dispersion equation 2k0 hnf cos θ − 2φpfc −
2φpfs = 2νπ (p = TE or TM , and ν = 0, 1, · · · ). The solution for the zig-zag angle θ is shown as the
intersection of two simple curves, the f1 (θ) = 2k0 hnf cos θ and f2 (θ) = 2φpfc (θ) + 2φpfs (θ) + 2νπ. For the
example case shown λ0 = 1.0µm, h = 1.2µm, nc = 1.0, nf = 2.2, and ns = 1.5. The solutions for the zig-
TE TE TE TE
zag angles are θ0 = 80.5307◦, θ1 = 70.8514◦, θ2 = 60.7247◦, and θ3 = 49.9121◦ for the TE modes,
TM TM TM TM
and θ0 = 79.6444◦, θ1 = 69.0254◦, and θ2 = 57.8963◦, and θ3 = 46.5355◦ for the TM modes. The
corresponding critical angles are θcr,fc = 27.0357◦ and θcr,fs = 42.9859◦.
in the next section when the full electromagnetic approach will be presented. From the angle
θνp , the effective refractive index of the mode can be defined as Nνp = nf sin θνp . Furthermore,
the effective propagation constant is related to the effective index as βνp = k0 Nνp = k0 nf sin θνp
(p = TE or TM ).
2.1 Alternate Approach
A similar approach based on a self-consistency condition is presented in this section and is
based on the geometry shown in Fig. 1. In this ray diagram of a guided mode in the waveguide
a plane wave is associated with each ray. Specifically (neglecting polarization) the three plane
8
waves, S0 , S1, and S2 , can be described by the following expressions
S0 = S0 (x, z) = E0 e−jk0 nf (x cos θ+z sin θ) , (5)
S1 = S1 (x, z) = E1 e−jk0 nf (−x cos θ+z sin θ) , (6)
S2 = S2 (x, z) = E2 e−jk0 nf (x cos θ+z sin θ) . (7)
The plane wave S1 is produced by a reflection of S0 at the film-cover boundary. Similarly, plane
wave S2 is produced by reflection of S1 at the film-substrate boundary. These arguments can
be quantified as follows
p
S1(x = h, z) = rf c S0 (x = h, z) ⇒ E1 e+jk0 nf h cos θ = ej2φf c E0e−jk0 nf h cos θ , (8)
p
j2φf s
S2 (x = 0, z) = rf s S1 (x = h, z) ⇒ E2 = e E1 , (9)
where rf c and rf s are the reflection coefficients upon total internal reflection at the film-cover
and film-substrate interfaces, respectively, and p = TE or TM . From the last two equations it
can be deduced that
p p
E2 = E0 ej(−2k0 nf h cos θ+2φf s +2φf c ) . (10)
The last equation requires that in order for S2 to be equivalent to S0 their amplitudes must
differ by at most by 2πν, i.e.
p p
E2 = E0 ej2πν = E0 ej(−2k0 nf h cos θ+2φf s +2φf c ) , (11)
and from the last expression Eq. (4) is derived again.
3. Electromagnetic Approach for Guided Modes
The electromagnetic fields representing a mode for the slab waveguide problem (using the
coordinate axes system shown in Fig. 1) can be written in the form
~ = [Ex (x)x̂ + Ey (x)ŷ + Ez (x)ẑ] exp(−jβz),
E (12)
~ = [Hx (x)x̂ + Hy (x)ŷ + Hz (x)ẑ] exp(−jβz),
H (13)
where Ew and Hw (w = x, y, z) represent the electric and magnetic field components and β is
a propagation constant. Using Maxwell equations in their differential time-harmonic form, for
a homogeneous, linear, lossless, and isotropic material, of permittivity and permeability µ0
9
(non-magnetic) with the above field expressions results in the following sets of equations (ω is
the radial frequency)
0 −jωµ0 0 0
E
Ey β2 y
−jω + j 0 0 0
d
Hz ωµ0 Hz
= , (14)
dx Hy
0 0 0 jω Hy
Ez β2 Ez
0 0 jωµ0 − j 0
ω
β
− 0
" # " #
Hx E
ωµ0 y
= . (15)
Ex β Hy
0
ω
From the above equations it is straightforward to distinguish two set of field components,
the {Ey , Hx , Hz } and {Hy , Ex , Ez }, which are independent from each other due to the zeros
appearing in the matrices of Eqs. (14) and (15). These two sets can be used to define the TE-
modes solutions for {Ey , Hx , Hz }, and the TM-modes solutions for {Hy , Ex , Ez }. It is mentioned
that these two triplets of field components are independent only in the isotropic case. If the
material becomes anisotropic the zero elements are replaced by non-zero ones and all six field
components become coupled. Then a waveguide mode, usually called hybrid mode, can be
characterized by all six field components as {Ex , Ey , Ez , Hx , Hy , Hz }.
In a homogeneous, isotropic, lossless, non-magnetic, and linear material the solutions to the
Helmholtz’s equation are in the form of plane waves. For example, if solutions of the form of
Eqs. (12), (13) are sought the Helmholtz’s equation becomes
~
d2 U ~ = 0,
+ (k02n2 − β 2)U (16)
dx 2
~ for electric of magnetic field, respectively, and n = √ε = /0 is the
p
~ = E
where U ~ or H
refractive index of the material (ε is the relative permittivity of the material). The solutions of
the above equation have the form
~ =U
U ~ + exp(−j~k+ · ~r) + U
~ − exp(−j~k− · ~r), (17)
~ − are the vector amplitudes of the two plane wave solutions and ~k+ and ~k− are
~ + and U
where U
their corresponding wave-vectors. The “k+ /k− ” terms represent the waves propagating towards
10
the positive/negative x direction. By applying the above solutions to each of the three separate
regions of the slab waveguide of Fig. 1, one can write the following equations
~ c1 e−j(kcx x+kcz z) + U
U ~ c2 e−j(−kcx x+kcz z) , x > h,
~ =
U ~ f 1e−j(kf x x+kf z z) + U
U ~ f 2e−j(−kf x x+kf z z) , 0 < x < h, (18)
~ −j(ksx x+ksz z) ~ −j(−ksx x+ksz z)
Us1 e + Us2 e , x < 0,
where the subscripts c, f, and s denote fields as well as wavevector components in cover, film,
and substrate, respectively. From the phase matching condition along the boundaries (x = 0
for film-substrate and x = h for film-cover) the following relation is necessary
kcz = kf z = ksz = β. (19)
At the same time the wave-vector components should satisfy the plane wave dispersion equation
2
of the form kwx + β 2 = k02 n2w (for w = c, f, and s). In order to warrant total internal reflection
at the film-cover and film-substrate boundaries β must satisfy the inequality k0 nc ≤ k0 ns <
β < k0 nf . Then the x components of the wavevectors become
q
2
kcx = k02n2c 2
− β < 0 ⇒ kcx = ±j β 2 − k02n2c = ±jγc , (20)
q
2 2 2 2
kf x = k0 nf − β > 0 ⇒ kf x = ± k02 n2f − β 2, (21)
q
2 2 2 2
ksx = k0 ns − β < 0 ⇒ ksx = ±j β 2 − k02n2s = ±jγs , (22)
where the signs of the imaginary wave-vector components must be selected in such a way in
order to warrant exponentially decaying solutions in the cover and the substrate regions. Thus,
from the two possible solutions in the cover and the substrate regions only the exponentially
decaying one are retained. Therefore the field solutions can be summarized as
~ c e−γc (x−h) e−jβz ,
U x > h,
h i
~
U= ~ f 1 e−jkf x x + U
U ~ f 2e+jkf x x e−jβz , 0 < x < h, (23)
~ γs x −jβz
Us e e , x < 0,
The unknown constants of the above equations (field amplitudes and β) should be determined by
suitable application of the boundary conditions. The boundary conditions are the continuity of
11
the tangential electric field and tangential magnetic field components at the film-cover and film-
substrate boundaries. In order to proceed with the boundary conditions and the determination
of the unknowns the two distinguished families of modes (TE and TM ) will be treated separately
since they are decoupled (in isotropic regions).
3.1 TE Guided Modes
As it was explained in the previous section the TE mode solutions include the field triplet
{Ey , Hx , Hz }. Then the electric field described in Eq. (23) can written in the form
Ec e−γc (x−h) e−jβz , x ≥ h,
E~ = ŷ Ef 1e−jkf x x + Ef 2 e+jkf x x e−jβz , 0 ≤ x ≤ h, (24)
Es eγsx e−jβz , x ≤ 0,
The magnetic field components can be determined from Maxwell’s equations, Hx = (1/jωµ0 )(dEy /dz)
and Hz = −(1/jωµ0 )(dEy /dx), and are given by
Ec e−γc (x−h) e−jβz , x ≥ h,
β
Hx = − Ef 1e−jkf x x + Ef 2e+jkf x x e−jβz , 0 ≤ x ≤ h, (25)
ωµ0
Es eγs x e−jβz , x ≤ 0,
and by
−jγc Ec e−γc (x−h) e−jβz , x ≥ h,
1
Hz = kf x Ef 1e−jkf x x − kf x Ef 2e+jkf x x e−jβz , 0 ≤ x ≤ h, (26)
ωµ0
jγs Es eγs x e−jβz , x ≤ 0.
Using the continuity of the tangential electric and magnetic field components across the
film-cover [Ey (x = h+ ) = Ey (x = h− ) and Hz (x = h+ ) = Hz (x = h− )] and film-substrate
boundaries [Ey (x = 0+ ) = Ey (x = 0− ) and Hz (x = 0+ ) = Hz (x = 0− )] the following system of
equations is formed
−1 e−jkf x h e+jkf x h 0 Ec 0
jγc kf xe−jkf x h −kf x e+jkf x h 0 Ef 1 0
= . (27)
0 1 1 −1 E 0
f2
0 kf x −kf x −jγs Es 0
| {z }
ÃT E (β 2)
12
In order to have nontrivial solutions of the previous equation it is necessary that the determinant
of ÃT E be set to zero. After some manipulations the following dispersion equation is derived
γs γc
+
kf x kf x
det{ÃT E (β 2)} = 0 =⇒ tan(kf x h) = γs γc . (28)
1−
kf x kf x
A graphical representation of Eq. (28) is shown in Fig. 4. There can be none, one, or multiple
solutions depending on the waveguide parameters and the free-space wavelength. For this
particular example (the waveguide parameters are given in the figure caption) there 4 TE and
4 TM solutions. This example case is the same with the one used for the graphical representation
of Eq. (4).
n c = 1, n f = 2.2, n s = 1.5, h = 1.2 m, 0
=1 m
10
4
Functions f 1 (N) or f 2 (N)
-2
-4
-6 f1
f2 for TE
-8
f2 for TM
-10
1.5 1.6 1.7 1.8 1.9 2 2.1 2.2
Effective Index, N
Figure 4: A graphical representation of the solution of Eq. (28) for TE modes. The functions f1 and f2 are
2
given by f1 (N ) = tan(kfx h) and f2 (N ) = [(γs /kfx) + (γc /kfx )]/[1 − (γc γs /kfx )], where kfx = k0 (n2f − N 2 )1/2 ,
γc = k0 (N 2 − n2c )1/2 , γs = k0 (N 2 − n2s )1/2 , and N is the effective index. The case parameters are nc = 1,
nf = 2.2, ns = 1.5, h = 1.2µm, and λ0 = 1.0µm. The points along the effective index axis where f1 is
discontinuous and infinite are given by N (m) = [n2f − ((2m + 1)/4)2 (λ0 /h)2 ]1/2 (m = 0, 1, · · · ). Similarly
f2 becomes infinite and discontinuous at N = [(n4f − n2s n2c )/(2n2f − n2c − n2s )]1/2 . In the case of TM modes
the solution of Eq. (34) is also shown. In the latter case f1 remains the same but f2 becomes f2 (N ) =
[(n2f /n2s )(γs /kfx ) + (n2f /n2c )(γc /kfx )]/[1 − (n4f /n2c n2s )(γc γs /kfx
2
)]. In the TM case f2 becomes discontinuous at
N being the solution of (1 − a )N + (2nf a − ns − nc )N + (n2c n2s − n4f a2 ) = 0, that lies in the interval
2 4 2 2 2 2 2
ns < N < nf (where a = n2s n2c /n4f ).
13
It is straightforward to show that the last equation is equivalent to Eq. (4) (for p = TE). It
is worth mentioning that the dispersion equation is actually a function of β 2 (or N 2 ) and not
of just β (or N). This implies that if βν is a solution of Eq. (28) then −βν is also a solution
that corresponds to the same mode propagating backwards (along the −z axis). If βν satisfies
Eq. (28) then the matrix ÃT E becomes singular and correspondingly the boundary conditions
contained in Eq. (27) become dependent. This fact provides the flexibility of selecting a free
parameter and then solve for the electric field amplitudes Ec , Ef 1, Ef 2, and Es as functions of
this parameter. It is a simple task to perform this procedure and find that Ec = E0 cos(kf x h −
TE TE TE TE
φf s ), Ef 1 = (E0/2) exp(+jφf s ), Ef 2 = (E0 /2) exp(−jφf s ), and Es = E0 cos φf s where E0
TE
is the free parameter and φf s = tan−1 (γs /kf x ). The amplitude E0 is the free parameter and
can also be determined if information is known about the power that the mode carries. It is
mentioned that all amplitudes are calculated for β = βν and therefore are characteristic of the
T Eν mode. Then the final form of the electric field of the T Eν mode is given by
TE −γc (x−h)
cos(kf x h − φ fs ) e
(x ≥ h),
E~ ν = ŷEyν (x)e−jβν z = ŷE0 TE
cos(kf xx − φf s ) e−jβν z (0 ≤ x ≤ h), (29)
TE
cos φf s e γs x (x ≤ 0),
where for simplicity the subscript ν, which is characteristic of the T Eν mode, has been omitted
TE
from E0 , kf x , γc , γs and φf s . The term Eyν (x) denotes the characteristic profile of the TEν
mode. Example TEν mode electric field patterns are shown in Fig. 5 where the effective index
solutions are also included. It can be observed that the number of zero-crossings of the electric
field profile for TEν mode is ν.
3.2 TM Guided Modes
The TM mode solutions include the field triplet {Hy , Ex , Ez }. Then the magnetic field repre-
sented in Eqs. (23) can written in the form
Hc e−γc (x−h) e−jβz , x ≥ h,
~ = ŷ
H Hf 1 e−jkf x x + Hf 2 e+jkf x x e−jβz , 0 ≤ x ≤ h, (30)
Hs eγs x e−jβz , x ≤ 0,
The electric field components of the TM guided mode can be easily determined from Maxwell’s
equations, i.e. Ex = −(1/jω)(dHy /dz) and Ez = (1/jω)(dHy /dx), and are given by
14
TE 0 -mode (N 0 = 2.170022) TE 1 -mode (N 1 = 2.078276)
1 1
0.8
Normalized E y
0.5
Normalized E
0.6
0
0.4
-0.5
0.2
0 -1
-4 -2 0 2 4 -4 -2 0 2 4
Distance x ( m) Distance x ( m)
TE 2 -mode (N 2 = 1.919016) TE 3 -mode (N 3 = 1.683127)
1 1
Normalized E y
Normalized E y
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-4 -2 0 2 4 -4 -2 0 2 4
Distance x ( m) Distance x ( m)
Figure 5: The normalized electric field profiles, Eyν (x), for TEν modes (ν = 0, 1, 2,and 3). The slab
waveguide parameters are nc = 1, nf = 2.2, ns = 1.5, h = 1.2µm, and λ0 = 1.0µm. The vertical lines indicate
the film-substrate and film-cover boundaries. The effective indices are also shown on the top of each plot.
1
H e−γc (x−h) e−jβz ,
2 c
x > h,
nc
β 1
Ex = 2
Hf 1 e−jkf x x + Hf 2 e+jkf x x e−jβz , 0 < x < h, (31)
ω0
nf
1
2 Hs eγsx e−jβz ,
x < 0,
ns
and by γc
+j H e−γc (x−h) e−jβz ,
2 c
x ≥ h,
n c
" #
1 kf x kfx
Ez = Hf 1e−jkf x x − 2 Hf 2 e+jkf x x e−jβz , 0 ≤ x ≤ h, (32)
ω0
n2
f n f
γ s
−j 2 Hs eγsx e−jβz ,
x ≤ 0.
ns
Using the continuity of the tangential electric and magnetic field components across the
15
film-cover [Hy (x = h+ ) = Hy (x = h− ) and Ez (x = h+ ) = Ez (x = h− )] and film-substrate
boundaries [Hy (x = 0+ ) = Hy (x = 0− ) and Ez (x = 0+ ) = Ez (x = 0− )] the following system of
equations is formed
−1 e−jkf x h e+jkf x h 0
γc kf x kf x +jkf x h Hc 0
−j 2 − 2 e−jkf x h e 0
nc nf n2f
Hf 1 0
= . (33)
0 1 1 −1 Hf 2
0
kf x kf x γs Hs 0
0 − 2 2
j 2
nf nf ns
| {z }
ÃT M (β 2)
In order to have nontrivial solutions of the previous equation it is necessary that the determinant
of ÃT M be set to zero. After some manipulations, similar to the TE polarization case, the
following dispersion equation is derived
n2f γs n2f γc
+
2 n2s kf x n2c kf x
det{ÃT M (β )} = 0 =⇒ tan(kf x h) = . (34)
n4f γc γs
1− 2 2 2
ns nc kf x
It is straightforward to show again that the last equation is equivalent to Eq. (4) (for p = T M).
A graphical representation of Eq. (34) is also shown in Fig. 4. Similarly to the TE polarization
case a solution ±βν of Eq. (34) corresponds to the ±z-axis propagating mode (assuming that
k0 ns < βν < k0 nf ). If βν satisfies Eq. (34) then the matrix ÃT M becomes singular and
correspondingly the boundary conditions contained in Eq. (33) become dependent. This provide
the flexibility of selecting a free parameter and then solve for the magnetic field amplitudes Hc ,
Hf 1 , Hf 2 , and Hs as functions of this parameter. It is a simple task to perform this procedure
TM TM TM
and find that Hc = H0 cos(kf x h− φf s ), Hf 1 = (H0 /2) exp(+jφf s ), Hf 2 = (H0 /2) exp(−jφf s ),
TM TM
and Hs = H0 cos φf s where H0 is the free parameter and φf s = tan−1 [n2f γs /(n2s kf x )]. The
amplitude H0 is the free parameter and can also be determined if information about the power
that the mode carries is known. It is mentioned that all amplitudes are calculated for β = βν
and therefore are characteristic of the TMν mode. Then the final form of the magnetic field of
16
the TMν mode is given by
TM
cos(kf x h − φf s ) e−γc (x−h)
(x ≥ h),
~ ν = ŷHyν (x)e−jβν z
H = ŷH0
TM
cos(kf x x − φf s ) e−jβν z (0 ≤ x ≤ h), (35)
TM
cos φf s eγs x (x ≤ 0),
where for simplicity the subscript ν, which is characteristic of the TMν mode, has been omitted
TM
from H0 , kf x , γc , γs and φf s . The term Hyν (x) denotes the profile of the TMν mode. Example
TMν mode electric field patterns are shown in Fig. 6 where the effective index solutions are
also shown. Again it is observed that the number of zero-crossings of the magnetic field profile
for TMν mode is ν.
TM 0 -mode (N 0 = 2.164164) TM 1 -mode (N 1 = 2.054226)
1 1
0.8
Normalized H y
Normalized H y
0.5
0.6
0
0.4
-0.5
0.2
0 -1
-4 -2 0 2 4 -4 -2 0 2 4
Distance x ( m) Distance x ( m)
TM 2 -mode (N 2 = 1.863593) TM 3 -mode (N 3 = 1.596761)
1 1
Normalized H y
Normalized H y
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-4 -2 0 2 4 -4 -2 0 2 4
Distance x ( m) Distance x ( m)
Figure 6: The normalized magnetic field profiles, Hyν (x), for TMν modes (ν = 0, 1, 2, and 3). The slab
waveguide parameters are nc = 1, nf = 2.2, ns = 1.5, h = 1.2µm, and λ0 = 1.0µm. The vertical lines indicate
the film-substrate and film-cover boundaries. The effective indices are also shown on the top of each plot.
17
4. Substrate Modes
In the previous sections solutions for guided modes were determined. This was based on the
assumption that k0 ns < β < k0 nf (or equivalently ns < N < nf ) in order to guarantee
that total internal reflection occurs across the film-cover and film-substrate boundaries. When
k0 nc < β < k0 ns (or nc < N < ns ) then the electromagnetic field in the substrate region is
propagating instead of evanescent. The modes that have this property are called substrate
modes and radiate power into the substrate as they propagate. For guided modes the solutions
for β (or N) were discretized. In the case of the substrate modes the solutions for β form a
continuum. Thus, any β in the interval k0 nc < β < k0 ns can be a solution for a substrate mode.
It is straightforward to determine the electromagnetic field that correspond to a substrate mode.
The electric or magnetic field of a TE or TM substrate mode can be written as
Uc e−γc (x−h) e−jβz , x ≥ h,
~
U = U(x, z)ŷ = ŷ Uf 1 e−jkf x x + Uf 2e+jkf x x e−jβz , 0 ≤ x ≤ h, (36)
Us1 e−jksx x + Us2 e+jksx x e−jβz , x ≤ 0,
where U = Ey for TE substrate modes and U = Hy for TM substrate modes, and ksx =
(k02 n2s − β 2)1/2 > 0 for the range of β that is valid for substrate modes. Using the continuity of
the tangential electric and magnetic field components 4 equations can be specified. However, the
number of unknowns is six, i.e. Uc , Uf 1, Uf 2, Us1 , Us2 , and β. Therefore, there is some flexibility
in satisfying the boundary conditions. For example, any β in the interval k0 nc < β < k0 ns can
satisfy the boundary conditions. After some manipulations of the resulting boundary conditions
it can be shown that the electric field of a TE substrate mode is given by
E~β = ŷEyβ (x)e−jβz = (37)
TE
cos φf c e−γc (x−h)
(x ≥ h),
TE
ŷE0 cos[kf x (x − h) + φf c ] e−jβν z (0 ≤ x ≤ h),
TE kf x TE
cos(kf x h − φf c ) cos(ksx x) + sin(kf x h − φf c ) sin(ksx x)
(x ≤ 0),
ksx
18
TE
where E0 is a free parameter, Eyβ (x) is the substrate TE mode profile, and φf c = tan−1 (γc /kf x ).
Similarly the magnetic field for a TM substrate mode is given by
H~β = ŷHyβ (x)e−jβz = (38)
TM
cos φf c e−γc (x−h)
(x ≥ h),
TM
cos[kf x(x − h) + φf c ]
ŷH0 e−jβν z (0 ≤ x ≤ h),
TM kf x/n2f TM
cos(kf x h − φf c ) cos(ksx x) + sin(kf x h − φ f c ) sin(ksx x)
(x ≤ 0),
ksx /n2s
TM
where H0 is a free parameter, Hyβ (x) is the substrate TM mode profile, and φf c = tan−1 (γc n2f /kf x n2c ).
Sample normalized electric/magnetic field profiles for TE/TM substrate modes are shown in
Fig. 7.
TE Substrate Mode, (N eff = 1.4500) TM Substrate Mode, (Neff = 1.4500)
1 1
0.5 0.5
Normalized Ey
Normalized Hy
0 0
-0.5 -0.5
-1 -1
-6 -4 -2 0 2 4 -6 -4 -2 0 2 4
Distance x (µm) Distance x (µm)
TE Substrate Mode, (N eff = 1.3000) TM Substrate Mode, (Neff = 1.3000)
1 1
0.5 0.5
Normalized Ey
Normalized Hy
0 0
-0.5 -0.5
-1 -1
-6 -4 -2 0 2 4 -6 -4 -2 0 2 4
Distance x (µm) Distance x (µm)
Figure 7: The normalized electric/magnetic Eyβ (x)/Hyβ (x) field profiles for TE/TM substrate modes. The
slab waveguide parameters are nc = 1, nf = 2.2, ns = 1.5, h = 1.2µm, and λ0 = 1.0µm. The vertical lines
indicate the film-substrate and film-cover boundaries. The effective indices (selected arbitrarily in the interval
nc = 1 < N < 1.5 = ns ) are also shown on the top of each plot.
19
5. Radiation Modes
When 0 < β < k0 nc (or equivalently 0 < N < nc ) then the field in all three regions (cover,
film, and substrate) is propagating. The modes that have this property are called r adiation
modes and radiate power into both the cover and the substrate regions as they propagate. As
in the case of the substrate modes, the solutions for β form a continuum and for the radiation
modes. Thus, any β in the interval 0 < β < k0 nc can be a solution for a radiation mode. It
is straightforward to determine the electromagnetic field of a radiation mode. The electric or
magnetic field of a TE or TM radiation mode can be written as
Uc1 e−jkcx (x−h) + Uc2 e+jkcx (x−h) e−jβz , x ≥ h,
~ = U(x, z)ŷ = ŷ
U Uf 1 e−jkf x x + Uf 2e+jkf x x e−jβz , 0 ≤ x ≤ h, (39)
Us1 e−jksx x + Us2 e+jksx x e−jβz , x ≤ 0,
where U = Ey for TE radiation modes and U = Hy for TM radiation modes, and kcx =
(k02 n2c − β 2)1/2 > 0 for the range of β that is valid for radiation modes. Using the continuity of
the tangential electric and magnetic field components 4 equations can be specified. However,
the number of unknowns is seven, i.e. Uc1 , Uc2 , Uf 1 , Uf 2 , Us1 , Us2 , and β. Therefore, there is
even more flexibility in satisfying the boundary conditions. For example, any β in the interval
0 < β < k0 nc can satisfy the boundary conditions. After some manipulations of the resulting
boundary conditions it can be shown that the electric field of a TE radiation mode is given by
E~β = ŷEyβ (x)e−jβz = (40)
"
1 kf x
1+ cos[kcx (x − h) + kf xh − φ] +
(x ≥ h),
2 kcx
#
kf x
1− cos[kcx (x − h) − kf x h + φ]
kcx
ŷE0 e−jβz
cos(kf x x − φ)
(0 ≤ x ≤ h),
" #
1 kf x kf x
1+ cos[ksx x − φ] + 1 − cos[ksx x + φ]
(x ≤ 0),
2 ksx ksx
where E0 and φ are two free parameters. Similarly the magnetic field for a TM radiation mode
20
is given by
H~β = ŷHyβ (x)e−jβz = (41)
"
1 kf x /n2f
1+ cos[kcx (x − h) + kf xh − φ] +
(x ≥ h),
2 kcx /n2c
#
kf x /n2f
1− cos[kcx (x − h) − kf x h + φ]
kcx /n2c
ŷH0 e−jβz
cos(kf x x − φ)
(0 ≤ x ≤ h),
" #
2 2
1 kf x /nf kf x /nf
1+ cos[ksx x − φ] + 1 − cos[ksx x + φ]
(x ≤ 0),
2 ksx /ns2 ksx /ns2
where H0 and φ are two free parameters. Sample normalized electric/magnetic field profiles for
TE/TM radiation modes are shown in Fig. 8.
TE Radiation Mode, , Neff = 0.95, φ = 135 ° TM Radiation Mode, , Neff = 0.95, φ = 135 °
1 1
0.5 0.5
Normalized Ey
Normalized Hy
0 0
-0.5 -0.5
-1 -1
-6 -4 -2 0 2 4 6 8 -6 -4 -2 0 2 4 6 8
Distance x (µm) Distance x (µm)
TE Radiation Mode, , Neff = 0.95, φ = 30 ° TM Radiation Mode, , Neff = 0.95, φ = 30 °
1 1
0.5 0.5
Normalized Ey
Normalized Hy
0 0
-0.5 -0.5
-1 -1
-6 -4 -2 0 2 4 6 8 -6 -4 -2 0 2 4 6 8
Distance x (µm) Distance x (µm)
Figure 8: The normalized electric/magnetic field profiles, Eyβ (x)/Hyβ (x), for TE/TM radiation modes. The
slab waveguide parameters are nc = 1, nf = 2.2, ns = 1.5, h = 1.2µm, and λ0 = 1.0µm. The vertical lines
indicate the film-substrate and film-cover boundaries. The effective indices (arbitrarily chosen in the interval
0 < N < 1 = nc ) are also shown on the top of each plot along with the choice of the free parameter φ.
21
6. Unphysical Modes (or Nonphysical Modes)
When k0 nf < β < ∞ (or equivalently nf < N < ∞) then the field in all three regions (cover,
film, and substrate) is comprised of evanescent terms, usually including both exponentially
increasing and exponentially decreasing terms. The modes that have this property are called
unphysical (or nonphysical) modes since there are valid solutions of the Maxwell’s equations
but they cannot be excited since they have infinite power. For this reason they are called
unphysical modes and are not usually referred to most of the textbooks. The analysis of these
modes is similar with the case of radiation modes with the exception that all field components
are comprised of both increasing and decreasing exponential terms. As in the case of the
substrate/radiation modes, their solutions for β form a continuum. Thus, any β in the interval
k0 nf < β < ∞ can be a solution for an unphysical mode. It is straightforward to determine
the electromagnetic field of an unphysical mode. The electric or magnetic field of a TE or TM
unphysical mode can be written as
Uc1 e−γc (x−h) + Uc2 e+γc (x−h) e−jβz , x ≥ h,
~ = U(x, z)ŷ = ŷ [Uf 1e−γf x + Uf 2e+γf x ] e−jβz ,
U 0 ≤ x ≤ h, (42)
[Us1 e−γs x + Us2 e+γs x ] e−jβz , x ≤ 0,
where U = Ey for TE unphysical modes and U = Hy for TM unphysical modes, and γf =
(β 2 − k02n2f )1/2 > 0 for the range of β that is valid for unphysical modes. If one tries to find
a solution without the increasing exponentials in the cover and the substrate regions (i.e.,
for Uc2 = Us1 = 0), one ends with a dispersion equation of the form (for TE polarization)
tanh(γf h) = −[(γs /γf ) + (γc /γf )]/[1 + (γc γs /γf2 )] that does not have a real solution for β in the
range of the unphysical modes (a similar equation, that does not have a real solution, exists
for the TM polarization too). Therefore, it is required to retain all the terms appearing in the
above equation similarly to the radiation modes case. Using the continuity of the tangential
electric and magnetic field components 4 equations can be specified. However, the number of
unknowns is seven, i.e. Uc1 , Uc2 , Uf 1, Uf 2, Us1 , Us2 , and β (similar to the radiation modes case).
Therefore, there is a lot of flexibility in satisfying the boundary conditions. For example, any β
in the interval k0 nf < β < ∞ can satisfy the boundary conditions. After some manipulations
of the resulting boundary conditions it can be shown that the electric field of a TE unphysical
22
mode is given by
E~β = ŷEyβ (x)e−jβz = (43)
"
1 γf
1+ cosh[γc (x − h) + γf h − φ] +
(x ≥ h),
2 γc
#
γf
1− cosh[γc (x − h) − γf h + φ]
γc
ŷE0 e−jβz
cosh(γf x − φ)
(0 ≤ x ≤ h),
" #
1 γf γf
1+ cosh[γs x − φ] + 1 − cosh[γs x + φ]
(x ≤ 0),
2 γs γs
where E0 and φ are two free parameters. Similarly the magnetic field for a TM unphysical
mode is given by
H~β = ŷHyβ (x)e−jβz = (44)
"
1 γf /n2f
1+ cosh[γc (x − h) + γf h − φ] +
(x ≥ h),
2 γc /n2c
#
γf /n2f
1− cosh[γc (x − h) − γf h + φ]
γc /n2c
ŷH0 e−jβz
cosh(γf x − φ)
(0 ≤ x ≤ h),
" #
1 γf /n2f γf /n2f
1+ cosh[γs x − φ] + 1 − cosh[γs x + φ]
(x ≤ 0),
2 γs /n2s γs /n2s
where H0 and φ are two free parameters. Sample normalized electric/magnetic field profiles for
TE/TM radiation modes are shown in Fig. 9.
7. Evanescent Modes
In some cases there might be a need to use waveguide modes that have a purely imaginary
propagation constant β̃, such that β̃ = ±jβ (where β > 0). In this case the waveguide field
~ = U(x)
could be written as U ~ ~
exp(−j β̃z) = U(x) exp(−βz), where for forward evanescent modes
the β̃ = −jβ was selected in order for the evanescent mode to decay along the propagation
direction z. Strictly speaking, the evanescent modes are not propagating since they decay along
the propagation direction. However, they are proper modal solutions of the Helmholtz wave
equation as it is applied to the slab waveguide problem. The acceptable imaginary propagation
23
TE Unphysical Mode, , N eff = 2.25, φ = 0 TM Unphysical Mode, , Neff = 2.25, φ = 0
1 1
0.8 0.8
Normalized Ey
Normalized Hy
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-1 -0.5 0 0.5 1 1.5 -1 -0.5 0 0.5 1 1.5
Distance x (µm) Distance x (µm)
TE Unphysical Mode, , N eff = 2.25, φ s = 1.7783 TM Unphysical Mode, , Neff = 2.25, φ s = 1.7783
1 1
0.8 0.8
Normalized Ey
Normalized Hy
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.5 0 0.5 1 1.5 -0.5 0 0.5 1 1.5
Distance x (µm) Distance x (µm)
Figure 9: The normalized electric/magnetic field profiles, Eyβ (x)/Hyβ (x), for TE/TM unphysical modes.
The slab waveguide parameters are nc = 1, nf = 2.2, ns = 1.5, h = 1.2µm, and λ0 = 1.0µm. The vertical lines
indicate the film-substrate and film-cover boundaries. The effective indices are also shown on the top of each
plot along with the choice of the free parameter φ. For symmetry of the fields in the film layer the φ parameter
was chosen as φ = φs = γf h/2 in the bottom two plots.
constants are of the form β̃ = −jβ (for the forward evanescent modes) and β̃ = +jβ (for the
backward evanescent modes), i.e. they all lie along the imaginary β̃ axis. These evanescent
modes form a continuum. For example any value of β > 0 gives a propagation constant β̃ = −jβ
which is an acceptable solution to Maxwell equations in the slab waveguide geometry.
It is straightforward to determine the electromagnetic field of an evanescent mode. The
electric or magnetic field of a TE or TM forward propagating evanescent mode can be written
as 0 (x−h) 0 (x−h) −βz
−jkcx +jkcx
U c1 e + U c2 e e , x ≥ h,
h i
~
U = U(x, z)ŷ = ŷ Uf 1 e −jkf0 x x
+ Uf 2e +jkf0 x x
e−βz , 0 ≤ x ≤ h, (45)
0 0
Us1 e−jksx x + Us2 e+jksx x e−βz , x ≤ 0,
where U = Ey for TE forward evanescent modes and U = Hy for TM forward evanescent
24
0
modes, and kcx = (k02n2c + β 2)1/2 > 0, kf0 x = (k02n2f + β 2)1/2 > 0, and ksx
0
= (k02 n2s + β 2)1/2 > 0,
for the range of β that is valid for forward evanescent modes (i.e. the negative imaginary axis of
the propagation constant region). Using the continuity of the tangential electric and magnetic
field components 4 equations can be specified. However, the number of unknowns is seven, i.e.
Uc1 , Uc2 , Uf 1 , Uf 2, Us1 , Us2 , and β. Therefore, there is even more flexibility in satisfying the
boundary conditions. For example, any β in the interval 0 < β < k0 nc can satisfy the boundary
conditions. After some manipulations of the resulting boundary conditions it can be shown
that the electric field of a TE evanescent mode is given by
E~β = ŷEyβ (x)e−βz = (46)
"
1 kf0 x 0
1+ 0 cos[kcx (x − h) + kf0 x h − φ] +
(x ≥ h),
2 kcx
#
kf0 x 0
1− 0 cos[kcx (x − h) − kf0 x h + φ]
kcx
ŷE0 e−βz
cos(kf0 x x − φ)
(0 ≤ x ≤ h),
" #
0 0
1 kf x kf x
1+ 0 0
cos[ksx x − φ] + 1 − 0 0
cos[ksx x + φ]
(x ≤ 0),
2 ksx ksx
where E0 and φ are two free parameters (similar to the radiation modes). In an analogous
manner, the magnetic field for a TM forward evanescent mode is given by
H~β = ŷHyβ (x)e−βz = (47)
"
1 kf0 x /n2f 0
1+ 0 2
cos[kcx (x − h) + kf0 xh − φ] +
(x ≥ h),
2 kcx /nc
#
kf0 x /n2f 0
1− 0 2
cos[kcx (x − h) − kf0 x h + φ]
kcx /nc
ŷH0 e−βz
cos(kf0 x x − φ)
(0 ≤ x ≤ h),
" #
0 2 0 2
1 k /n k /n
1+ 0
fx f
cos[k 0
x − φ] + 1 −
f x f
cos[k 0
x + φ]
(x ≤ 0),
2 sx 0 2 sx
2 k /n
sx s k /n sx s
where H0 and φ are again the two free parameters. Sample normalized electric/magnetic field
profiles for TE/TM forward evanescent modes are shown in Fig. 10.
The various modes of the lossless slab waveguide problem (the example is for the single-
film layer slab waveguide but it can be easily generalized to multilayer-film slab waveguides) are
shown in the complex propagation constant (β̃ = βr +jβi) diagram in Fig. 11. The only discrete
25
TE Evanescent Mode, , N = - j0.5, φ = 30 ° TM Evanescent Mode, , N = - j0.5, φ = 30 °
eff eff
1 1
0.5 0.5
Normalized E y
y
Normalized H
0 0
-0.5 -0.5
-1 -1
-2 -1 0 1 2 3 4 -2 -1 0 1 2 3 4
Distance x (µm) Distance x (µm)
° °
TE Evanescent Mode, , N = - j0.5, φ = 135 TM Evanescent Mode, , N = - j0.5, φ = 135
eff eff
1 1
0.5 0.5
y
y
Normalized E
Normalized H
0 0
-0.5 -0.5
-1 -1
-2 -1 0 1 2 3 4 -2 -1 0 1 2 3 4
Distance x (µm) Distance x (µm)
Figure 10: The normalized electric/magnetic field profiles, Eyβ (x)/Hyβ (x), for TE/TM forward evanescent
modes. The slab waveguide parameters are nc = 1, nf = 2.2, ns = 1.5, h = 1.2µm, and λ0 = 1.0µm. The
vertical lines indicate the film-substrate and film-cover boundaries. The effective indices are also shown on the
top of each plot along with the choice of the free parameter φ.
spectrum is of the guided modes as it can be seen. The substrate, radiation, unphysical, and
evanescent modes from a continuum. In the same diagram the leaky modes are also shown. The
leaky modes is an approximation of the radiation field by a series of discrete modes of complex
propagation constant. These leaky modes will be discussed at a later section of this chapter.
In the case of loss or gain in the film region, the guided modes will have nonzero component
of their complex propagation constant, βi < 0 for loss and βi > 0 for gain (for the forward
propagating modes and the reverse of this for the backward propagation modes).
8. Normalized Slab-Waveguide Variables
The dispersion Eq. (4) can also be written in terms of normalized variables. Such a rep-
resentation is useful for any three-region slab waveguide. The normalized parameters are the
26
7 8 9 : ;
b U ^ c T U S V
o h p q g h f i j k l m e n
d e f g h f i j k l m e n
r s U X a \ ^ a X t
y U V ] U t ] R X Q R S T U S V
N F O J E F D G
` R V a \
` R V a \ u v ] V a V
H I F J K L C G I M
` R V a \
Q R S T U S V
Q R S T U S V
W X Y Z [ \ ] ^ U _ ` R V a \
w v x \ t S U t a
P
` R V a \
= > = @ = A
6 6 6 5 5 5
= @
= A = >
? ? ?
7 < 9 :
;
b U ^ c T U S V
w v x \ t S U t a
b U ^ c T U S V
` R V a \
P
W X Y Z [ \ ] ^ U _ ` R V a \
b U ^ c T U S V
u v ] V a V
B C D E F D G
Q R S T U S V
` R V a \
H I F J K L C G I M
r s U X a \ ^ a X t
` R V a \
z R X t ] X v v { w Y a ^ t S v {
| } ~
Figure 11: A diagram of the various modes of a lossless slab waveguide. The guided modes (forward or
backward) from a discrete spectrum while the radiation, substrate, and evanescent modes (forward or backward
form a continuum spectrum. The unphysical mode solutions have also a continuum spectrum (forward or
backward). The leaky modes are not exactly formal modes but they represent some form of normalization of
the radiation spectrum (forward or backward).
normalized frequency, V , the normalized effective propagation constant (or normalized effective
index), b, and are defined as follows
2π q 2
V = h nf − n2s , (48)
λ0
N 2 − n2s n2f sin2 θ − n2s
b = = , (49)
n2f − n2s n2f − n2s
n2s − n2c
aT E = , (50)
n2f − n2s
n4f n2s − n2c n4f
aT M = = 4 aT E , (51)
n4c n2f − n2s nc
27
where the aT E and aT M are defined as the asymmetry factors for TE and TM modes respectively.
Using these normalized variables in Eq. (4) it is straightforward to derive the following dispersion
equation for the case of TE polarization
(r ) (r )
√ b b + a
V 1 − b − tan−1 − tan−1 TE
= νπ, ν = 0, 1, · · · . (52)
1−b 1−b
A similar normalized dispersion equation can not be done for the TM modes. It can be shown
that by defining the normalized variables, bT M = (n2f /qn2s )[(N 2 − n2s )/(n2f − n2s )] and q =
(N 2 /n2f ) + (N 2 /n2s ) − 1, the resulting normalized dispersion equation for the TM modes is given
by
(s ) (s )
√ nf p bT M bT M + aT M (1 − bT M d)
V q 1 − bT M − tan−1 − tan −1
= νπ, (53)
ns 1 − bT M 1 − bT M
where ν = 0, 1, · · · and d = [1 − (ns /nf )2][1 − (nc /nf )2 ]. It is obvious that the above normalized
equation is not completely normalized since the ratios ns /nf , nc /nf , and ns , nf , are necessary
for the evaluation of bT M . However, if nf − ns ' 0 then the normalized dispersion equation for
the TM modes becomes similar to the one for the TE modes if the corresponding to the TM
modes asymmetry factor, aT M , is used.
Solving Eq. (52) for various values of the normalized frequency, V , of the asymmetry factor,
a (aT E or aT M ), and of ν, normalized diagrams such the one shown in Fig. 12 can be generated.
These are exact for TE modes and only approximately correct for TM modes as it was explained
above. Measuring from the diagram the value(s) of bν , for given V and a, the effective index of
the T Eν mode(s) (and approximately for T Mν mode(s) with a = aT M ) can be determined as
Nν = [n2s + bν (n2f − n2s )]1/2.
9. Cutoff Conditions
When the effective index of a guided mode becomes smaller than the substrate refractive index
(Nν < ns ) then the mode starts radiating into the substrate and becomes a substrate mode as
it was discussed in the previous sections. The condition Nν = ns (or equivalently βν = k0 ns )
represents a critical condition below which the guided field is cut-off and from guided mode
becomes a substrate (or in general radiation) mode. Equivalently, this occurs when the zig-zag
angle θ becomes equal to the critical angle at the film-substrate interface, θcr,f s (it is reminded
28
1.0
0.9
Normalized Propagation Constant, b 0.8
0.7
0.6
0.5
0.4
0.3
0.2 a = 0
1
0.1 10
10 6
0.0
0 2 4 6 8 10 12 14 16
Normalized Frequency, V
Figure 12: The normalized propagation constant b versus normalized frequency V for four selected asymmetry
factors (a = 0, 1, 10, and 106 ) for the first 5 modes (ν = 0, 1, 2, 3, and 4). A small segment for ν = 5 and a = 0
is shown on the right bottom corner of the diagram.
that nc < ns < nf ). Then, Eq. (4) becomes
q √
k0 h n2f − n2s − tan−1 ( aw ) = νπ, w = T E, T M. (54)
The last equation is a cutoff condition and can be solved for one parameter retaining the others
constant. For example, if it solved with respect to the waveguide thickness, h, the cutoff
thicknesses can be determined as follows
√
TE νπ + tan−1 ( aT E )
hcut,ν = q , (55)
k0 n2f − n2s
√
TM νπ + tan−1 ( aT M )
hcut,ν = q , (56)
2 2
k0 nf − ns
where aT E and aT M are the asymmetry factors for TE and TM polarization respectively and
are defined in Eqs. (50) and (51). It is interesting to notice that for symmetric waveguides,
aT E = aT M = 0 the cutoff thicknesses for the lowest order mode (ν = 0) are zero. I.e., for
symmetric waveguides, any film thickness, independently of how small it is, can support the
29
lowest order modes TE0 and TM0 . In Fig. 13 the effective index of various modes is shown as a
function of the film layer thickness. From these plots the cutoff thicknesses are the points from
which each of the βν curves starts.
nc = 1, nf = 1.56, n s = 1.45, =1 m
0
1.56
TE
TM
1.54
Effective Index, N
1.52
1.5
1.48
1.46
1.44
0 2 4 6 8 10
Film Thickness, h ( m)
n = 1, n = 2.2, n = 1.5, =1 m
c f s 0
2.3
2.2
2.1
Effective Index, N
1.9
1.8
1.7
TE
1.6
TM
1.5
0 1 2 3 4 5
Film Thickness, h ( m)
Figure 13: The effective index variation as a function of the waveguide film thickness. The slab waveguide
parameters are shown on the top of each plot. It is mentioned that the effective indices of up to the tenth TE
or TM are included.
If Eq. (54) is solved with respect to the free-space wavelength or frequency the following
30
cutoff wavelengths and frequencies can be defined for each mode and polarization
q
TE
2πh n2f − n2s
λ0,cut,ν = √ , (57)
νπ + tan−1 ( aT E )
q
TM
2πh n2f − n2s
λ0,cut,ν = √ , (58)
νπ + tan−1 ( aT M )
(59)
√
TE νπ + tan−1 ( aT E )
ωcut,ν = c q , (60)
h n2f − n2s
√
TM νπ + tan−1 ( aT M )
ωcut,ν = c q . (61)
h n2f − n2s
It is evident again that for symmetric waveguides the cutoff wavelength is infinite and the
corresponding cutoff frequencies are zero for the zero order modes. An example ω − β diagram
is shown in Fig. 14 where the cutoff frequencies can be shown for the first five TE and TM
modes. The waveguide parameters are included in the figure caption.
nc = 1, nf = 2.2, n s = 1.5, h = 1 m
10 15
3 (c/nc ) (c/ns )
(rad/sec)
2.5
(c/nf)
2
Radial Frequency,
1.5
0.5 TE-modes
TM-modes
0
0 0.5 1 1.5 2
Propagation Constant, (m -1 ) 10 7
Figure 14: A sample ω − β diagram for a three layer slab waveguide with nc = 1.0, nf = 2.2, ns = 1.5, and
h = 1µm for both TE and TM polarizations.
31
10. Power Considerations
In order to determine the time-average power that each mode can carry along the propagation
direction z, it is necessary to evaluate the z-component of the time-average Poynting vector
and integrate it over the xy-plane. Then, the power carried by the ν-th mode can be express
as Z Z
+∞ +∞
1 ~ ×H
~ ∗ } · ẑdxdy,
P = Re{E (62)
−∞ −∞ 2
where the integration is over the entire xy plane. However, in slab waveguides, there is no
y-dependence, and therefore the integration over the y-axis should give infinite power! As a
result in slab waveguides the power per unit y length is of interest and the previous equation
is written as
Z +∞
1 n~ ~ ∗
o
P = Re E × H · ẑdx =
−∞ 2
Z 0 Z h Z +∞
1 ~ ~ ∗ 1 ~ ~ ∗ 1 ~ ×H
~ ∗ } · ẑdx,(63)
= Re{E × H } · ẑdx + Re{E × H } · ẑdx + Re{E
2 2 2
| −∞ {z } |0 {z } |h {z }
Ps Pf Pc
where it is implied that P is expressed in Watts/per unit y length and the integral has been
separated in three parts Ps , Pf , and Pc each representing the power per unit y length in each
of the three waveguide regions, substrate, film, and cover, respectively. The x-component of
the average Poynting vector should be zero since no power is propagating in the x-direction for
guided modes.
10.1 TE Modes
Using the field Eqs. (29) and (25) in the case of TE modes the Poynting vector can be written
as (1/2)Re{−Ey Hx∗ } and the resulting power components become
r TE
cos2 (φf s )
TE
Nν 0 2
Ps = |E0 | , (64)
4 µ0 γs
TE r
" TE TE
#
Nν 0 sin(2k f x h − 2φ fs ) + sin(2φ fs )
Pf = |E0 |2 h + , (65)
4 µ0 2kf x
r TE
Nν
TE
0 cos2 (kf xh − φf s )
Pc = |E0 |2 , (66)
4 µ0 γc
32
TE r TE r
Nν 0 2 1 1 Nν 0 TE
P = |E0| h + + = |E0 |2hef f,ν , (67)
4 µ0 γs γc 4 µ0
TE
where hef f,ν is the effective thickness of the TEν mode and P = Ps + Pf + Pc . It is reminded
TE
that φf s = tan−1 (γs /kf x ). It is interesting to specify the percentage of power being carried by
the TEν mode in each of the three regions. Specifically, from the above equations the following
ratios can be defined
TE
Ps 1 cos2 (φf s )
= TE , (68)
P hef f,ν γs
" TE TE
#
Pf 1 sin(2kf x h − 2φf s ) + sin(2φf s )
= TE h+ , (69)
P hef f,ν 2kf x
TE
Pc 1 cos2 (kf x h − φf s )
= TE . (70)
P hef f,ν γc
TE
For example, the dependence of hef f,ν , and of Ps /P , Pf /P , and Pc /P of the film thickness h is
shown in Figs. 15 and 16 for the first 10 modes of a slab waveguide.
nc = 1, nf = 1.56, n s = 1.45, 0
=1 m
14
TE
12 TM
Effective Thickness, h eff
10
0
0 2 4 6 8 10
Film Thickness, h ( m)
Figure 15: A sample heff,ν versus film thickness diagram for a three layer slab waveguide with nc = 1.0,
nf = 1.56, ns = 1.45, and h = 1µm for both TE and TM polarizations. The effective thicknesses of the first 10
TE and TM modes are shown.
It is straightforward to show that the average Poynting vector along the x-direction is
Px = (1/2)Re{Ey Hz∗ }. Specifically, using Eqs. (29) and (26) it is easy to show that Px = 0 as
it is expected for guided modes.
33
TE, nc = 1, nf = 1.56, n s = 1.45, =1 m
0
1
Normalized Power, P w/P, (w=c,f,s)
0.8
0.6 P c /P
P f /P
P s /P
0.4
0.2
0
0 2 4 6 8 10
Film Thickness, h ( m)
TM, n c = 1, nf = 1.56, n s = 1.45, =1 m
0
1
Normalized Power, P w/P, (w=c,f,s)
0.8
0.6 P c /P
P f /P
0.4 P s /P
0.2
0
0 2 4 6 8 10
Film Thickness, h ( m)
Figure 16: The power percentages in the cover, film and substrate regions as a function of the waveguide
film thickness. The slab waveguide parameters are shown on the top of each plot. It is mentioned that the first
ten TE and TM modes are included.
10.2 TM Modes
Using the field Eqs. (35) and (31) in the case of TM modes the Poynting vector can be written
as (1/2)Re{Ex Hy∗} and the resulting power components become
TM r TM
Nν µ0 2
cos2(φf s )
Ps = |H0 | , (71)
4 0 n2s γs
TM r
" TM TM
#
Nν µ0 1 sin(2kf x h − 2φ f s ) + sin(2φ f s )
Pf = |H0 |2 2 h + , (72)
4 0 nf 2kf x
r TM
Nν µ0
TM
2
cos2(kf x h − φf s )
Pc = |H0 | , (73)
4 0 n2c γc
TM r TM r
Nν µ0 2 1 1 1 Nν µ0 1 TM
P = |H0 | 2 h + + = |H0 |2 2 hef f,ν , (74)
4 0 nf qs γs qc γc 4 0 nf
34
TM
where hef f,ν is the effective thickness of the TMν mode and P = Ps + Pf + Pc , and qc =
TM TM TM TM
(Nν )2 /n2c + (Nν )2 /n2f − 1, qs = (Nν )2/n2s + (Nν )2 /n2f − 1. It is reminded that in this
TM
case φf s = tan−1 [(γs /n2s )/(kf x /n2f )]. It is interesting to specify the percentage of power being
carried by the TMν mode in each of the three regions. Specifically, from the above equations
the following ratios can be defined
TM
Ps n2f cos2(φf s )
= TM , (75)
P hef f,ν n2s γs
" TM TM
#
Pf n2f 1 sin(2kf x h − 2φf s ) + sin(2φf s )
= TM h+ , (76)
P hef f,ν n2f 2kf x
TM
Pc n2f cos2(kf x h − φf s )
= TM . (77)
P hef f,ν n2c γc
TM
For example, the dependence of hef f,ν , and of Ps /P , Pf /P , and Pc /P of the film thickness
h is shown in Figs. 15 and 16 for the first 10 modes of a slab waveguide. Similar to the TE
modes, it is straightforward to show that the average Poynting vector along the x-direction is
Px = (1/2)Re{−Ez Hy∗ }. Specifically, using Eqs. (35) and (32) it is easy to show that Px = 0 as
it is expected for guided modes.
35
11. Multi-Layered Slab Waveguides
Multi-layered slab waveguides are of practical interest since they represent realizable semicon-
ductor waveguides that are used in semiconductor lasers. Furthermore, even dielectric waveg-
uides of more than a single film region are used. An example is the case that a buffer layer
may be needed to be inserted between the film region and metallic electrodes that are used
in electro-optic applications. Also, many times from a fabrication point of view, the resulting
waveguides end up with a refractive index profile that is of graded (smooth) variation instead of
step (abrupt) variation. Examples of such waveguides are the ones fabricated via ion-exchange,
ion-bombardment, or metal-in-diffusion techniques. In latter cases the graded-index waveguides
can be approximated by an arbitrary number of layers of constant index thus approximating
the smoothly varying refractive index. This technique can also be used to characterized fabri-
cated waveguides. For example, if the mode propagation constants can be measured (which is
feasible through a prism-coupling method) the parameters of the fabricated waveguide can be
determined by an inverse problem to the one that determines the propagation constants. For
all these reasons, the extension of the three region slab waveguide to the general multi-layered
case is presented here.
A general multi-layered slab waveguide is shown in Fig. 17. The waveguide is comprised of
N film regions (of refractive index ni and thickness hi ), in addition to the cover and substrate
regions, of refractive indices, nc and ns , respectively. The x = 0 is now selected at the cover-film
layer #1 boundary and its direction is towards the substrate. This direction is opposite of the
one used in Fig. 1. This is chosen since it matches the coordinate system that was utilized in
the computer program that was pre-written to these notes in order to implement the procedure
for the multi-layered slab waveguide. All the basic equations are the same with the only care
needed in the selection of the signs of the exponential terms of the evanescent waves in cover
and substrate regions.
The approach to determine the modes (TE or TM) in the case of a multi-layered slab
waveguide is conceptually identical with the one that was followed in Secs. 3.1 and 3.2. For
example, if the TE modes should be determined, then the unknown electric field amplitude
coefficients are Ec , E11, E12, · · · , Ei1 , Ei2 , · · · , EN 1 , EN 2, Es , as well as the propagation
constant βν . Therefore the number of unknowns is 2(N + 1) excluding βν . The number of
36
¥
¤ ¦ §
¥
¤ ¦ ¨
¥
¤ ¦ ©
¡ ¢ £ ¤ £
Figure 17: A general multilayered slab waveguide. The waveguide is comprised of N layers, with refractive
indices, ni , and layer thicknesses, hi (i = 1, 2, · · · , N ). The cover (x < 0) has a refractive index of nc , whereas
PN
the substrate (x > 1 hi ) has a refractive index of ns .
equations that can be derived from the boundary conditions are 2(N + 1). The boundary
conditions result in a matrix equation similar to Eq. (27) but of dimension 2(N + 1) × 2(N + 1).
The determinant of the coefficient matrix is the dispersion equation and solutions of this in the
interval k0 ns < βν < k0 maxi {ni } determine the propagation constants βν . However, in this
section the Transfer Matrix technique as described in Ref. [11–13] will be presented.
11.1 TE Guided Modes
As in the case of the three-layer slab waveguide, the TE modes are characterized by the elec-
tromagnetic fields {Ey , Hx , Hz }. Using the first two of Eqs. (14) and the first of Eqs. (15) the
following equations can be written for a region of refractive index n
0 −jωµ0
d Ey 2 Ey
= β , (78)
dx Hz −jω + j 0 Hz
ωµ0
β
Hx = − Ey . (79)
ωµ0
37
The solution of Ey in each region can be summarized as follows
Ec eγc x e−jβz , x ≤ 0,
~ = ŷ
E Ei1 e−jkxi (x−Di ) + Ei2 e+jkxi (x−Di ) e−jβz , Di ≤ x ≤ Di+1 , (80)
Es e−γs (x−DN +1 ) e−jβz , x ≥ DN +1 ,
where the distances Di are given by
i−1
X
Di = h` , with D1 = 0, i = 1, 2, · · · , N, (81)
`=1
and γc = (β 2 − k02n2c )1/2, γs = (β 2 − k02 n2s )1/2, kxi = (k02n2i − β 2)1/2 (i = 1, 2, · · · , N), and
Ec , Es , and Ei1, Ei2 the unknown amplitude constants. The tangential magnetic field can be
determined from Eqs. (78) and is given by
γc Ec eγc x e−jβz , x ≤ 0,
1
Hz = −jkxi Ei1e−jkxi (x−Di ) + jkxi Ei2 e+jkxi (x−Di ) e−jβz , Di ≤ x ≤ Di+1 , (82)
−jωµ0
−γs Es e−γs (x−DN +1 ) e−jβz , x ≥ DN +1 .
The next step is to use the boundary conditions at x = 0, D2 , · · · , DN +1 for continuity of Ey
and Hz . The resulting equations are written in the following form
1 1 E11 1
= Ec , at x = 0, (83)
−jkx1 jkx1 E12 γc
At the boundary between the i−th and the (i + 1)−th layer the following equations can be
derived
e−jkxi hi e+jkxi hi Ei1 1 1 Ei+1,1
= , at x = Di+1 ,
−jkxi e−jkxi hi jkxi e+jkxi hi Ei2 −jkx,i+1 jkx,i+1 Ei+1,2
(84)
PN
whereas at the last boundary at x = DN +1 = `=1 h` the following equations are derived
e−jkxN hN e+jkxN hN EN 1 1
= Es , at x = DN +1 . (85)
−jkxN e−jkxN hN jkxN e+jkxN hN EN 2 −γs
Now Eq. (83) can be written as
E11 1 jkx1 −1 1
= Ec , (86)
E12 2jkx1 jkx1 1 γc
| {z }
M̃1,0
38
defining the transfer matrix M̃1,0 between regions 0 (cover) and 1 (first film layer). Some special
care should be taken in the case that kx1 = 0 [in this case the electric field solution in region
1 is linear, i.e. Ey1 (x) = E1,2 (x − D1 ) + E1,1] and M̃1,0 becomes the unity matrix. Boundary
condition expressed in Eq. (84) can be written as
−jk h kxi +jk h kxi
e xi i 1 + e xi i 1 −
Ei+1,1 1 kx,i+1 kx,i+1 Ei1
= , (87)
Ei+1,2 2 −jkxi hi kxi +jkxi hi kxi Ei2
e 1− e 1+
kx,i+1 kx,i+1
| {z }
M̃i+1,i
thus defining the transfer matrix M̃i+1,i between i−th and (i + 1)-th regions. As a special case,
when kxi = 0 [the electric field solution in region i is linear, i.e. Eyi (x) = Ei,2 (x − Di ) + Ei,1]
and the matrix M̃i+1,i becomes
1
1 hi −
1 jkx,i+1
M̃i+1,i = 1 . (88)
2 1 hi +
jkx,i+1
Another special case is when kx,i+1 = 0 [the electric field solution in region i + 1 is linear, i.e.
Ey,i+1 (x) = Ei+1,2 (x − Di+1 ) + Ei+1,1 ] and the matrix M̃i+1,i becomes
1 e−jkxi hi e+jkxi hi
M̃i+1,i = . (89)
2 −jkxi e−jkxi hi +jkxi e+jkxi hi
Finally, Eq. (85) can be written as
1 e−jkxN hN e+jkxN hN EN 1
Es = , (90)
−γs −jkxN e−jkxN hN jkxN e+jkxN hN EN 2
| {z }
M̃N +1,N
thus, defining the last transfer matrix M̃N +1,N between regions N and substrate. In the case
that kxN = 0 [the electric field solution in region N is linear, i.e. Ey,N (x) = EN,2 (x−DN )+EN,1]
and the matrix M̃N +1,N becomes
1 1 hi
M̃N +1,N = . (91)
2 0 1
Using the previous equations all amplitudes Ei1 and Ei2 can be eliminated as follows
1 1 m11 m12 1
Es = M̃N +1,N M̃N,N −1 · · · M̃2,1 M̃1,0 Ec = Ec . (92)
−γs | {z } γc m21 m22 γc
M̃
39
QN
In the previous equation M̃ = `=0 M̃N +1−`,N −` is the total transfer matrix between the cover
and the substrate regions. Equation (92) results in
n o
m11 + m12 γc −1 Ec
= 0 =⇒ det ÃT E (β 2) = 0, (93)
m21 + m22 γc γs Es
| {z }
ÃT E (β 2 )
and equivalently
n o
det ÃT E (β 2) = 0 ⇒ γs m11 + γc m22 + γc γs m12 + m21 = 0. (94)
Solutions, βν , of the last equation for k0 ns < βν < k0 maxi {ni } specify the TEν modes. Using
Eqs. (86)-(90) all the unknown amplitude coefficients can be determined using a free parameter
(for example Es or Ec ). For an example multi-layered waveguide (for N = 3) the corresponding
TE modes electric field profiles are shown in Fig. 18.
TE 0 -mode (N TE
0
= 1.52154577) TE 1 -mode (N TE
1
= 1.50837665)
1 1
Normalized Electric Field, E y
Normalized Electric Field, E y
0.8
0.5
0.6
0.4
-0.5
0.2
0 -1
-6 -4 -2 0 2 4 6 8 -6 -4 -2 0 2 4 6 8
Distance x ( m) Distance x ( m)
Figure 18: The TEν (ν = 0, 1) modes electric field profiles of a sample multi-layered waveguide are shown.
The waveguide is comprised of three film layers and its paramaters are nc = 1.45, n1 = 1.56, n2 = 1.45,
n3 = 1.56, ns = 1.45, λ0 = 1.0µm, h1 = 0.75µm, h2 = 0.5µm, and h3 = 0.75µm. The effective indices
NνT E = βν /k0 are shown on each plot. The dashed lines specify the boundaries between differing regions.
11.2 TM Guided Modes
As in the case of the three-layer slab waveguide, the TM modes are characterized by the
electromagnetic fields {Hy , Ex , Ez }. Using the last two of Eqs. (14) and the second of Eqs. (15)
the following equations can be written for a region of refractive index n
d Hy
0 jω
H
y
= β2 , (95)
dx Ez jωµ0 − j 0 Ez
ω
40
β
Ex = Hy . (96)
ω
The solution of Hy in each region can be summarized as follows
Hc eγcx e−jβz , x ≤ 0,
~ = ŷ
H Hi1 e−jkxi (x−Di ) + Hi2 e+jkxi (x−Di ) e−jβz , Di ≤ x ≤ Di+1 , (97)
Hs e−γs(x−DN +1 ) e−jβz , x ≥ DN +1 ,
where the distances Di are given by Eq. (81), and Hc , Hs , and Hi1 , Hi2 the unknown amplitude
constants. The tangential magnetic field can be determined from Eqs. (95) and is given by
γ
c
H eγc x e−jβz ,
2 c
x ≤ 0,
nc
1 kxi −jkxi (x−Di ) kxi +jkxi (x−Di )
Ez = −j 2 Hi1 e + j 2 Hi2 e e−jβz , Di ≤ x ≤ Di+1 , (98)
jω0
ni ni
γs −γ (x−DN +1 ) −jβz
− 2 Hs e s
e , x ≥ DN +1 .
ns
At the boundary between the i−th and the (i + 1)−th layer the boundary conditions can be
written as
e−jkxi hi e+jkxi hi 1 1
Hi1 Hi+1,1
kxi −jkxi hi kxi = kx,i+1 kx,i+1 , at x = Di+1 .
j 2e −j 2 e+jkxi hi Hi2 j 2 −j 2 Hi+1,2
ni ni ni+1 ni+1
(99)
The last equation is the analogous of Eq. (84) with the difference that kxi → −kxi /n2i and sim-
ilarly for kx,i+1 whereas those terms appear as amplitude terms. The same replacement is valid
for the equations analogous to Eqs. (83), (85) where γc → −γc /n2c and γs → −γs /n2s whereas
they appear as amplitude terms. Consequently it is not necessary to repeat the procedure for
the TM modes. For example the dispersion equation can be written as
n o γs γc γc γs
det ÃT M (β 2) = 0 ⇒ − 2 m11 − 2 m22 + 2 2 m12 + m21 = 0. (100)
ns nc nc ns
The matrix elements mij (i, j = 1, 2) can be determined in the same manner as for the
TE modes with the replacements made previously for the terms that appear as amplitude
terms (not inside exponentials or sines and cosines). For an example multi-layered waveguide
(for N = 3) the corresponding TM modes magnetic field profiles are shown in Fig. 19.
41
TM 0 -mode (N TM
0
= 1.51817637) TM 1 -mode (N TM
1
= 1.50411134)
1 1
Normalized Magnetic Field, H y
Normalized Magnetic Field, H y
0.8
0.5
0.6
0.4
-0.5
0.2
0 -1
-6 -4 -2 0 2 4 6 8 -6 -4 -2 0 2 4 6 8
Distance x ( m) Distance x ( m)
Figure 19: The TMν (ν = 0, 1) modes magnetic field profiles of a sample multi-layered waveguide are
shown. The waveguide is comprised of three film layers and its paramaters are nc = 1.45, n1 = 1.56, n2 = 1.45,
n3 = 1.56, ns = 1.45, λ0 = 1.0µm, h1 = 0.75µm, h2 = 0.5µm, and h3 = 0.75µm. The effective indices
NνT E = βν /k0 are shown on each plot. The dashed lines specify the boundaries between differing regions.
11.3 TE and TM Guided Modes using Chen’s Approach
In this section the transfer matrix technique as used by Chen et al. [17] for the study of
multilayer planar waveguides is shortly summarized. For the TE modes the field components
for a region of refractive index n can be related by Eq. (78). The solution of Ey in each region
can be summarized as follows
Ec eγc x e−jβz , x ≤ 0,
n o
~
E = ŷ Ei1 cos [kxi (x − Di )] + Ei2 sin [kxi (x − Di )] e−jβz , Di ≤ x ≤ Di+1 , (101)
Es e−γs (x−DN +1 )e−jβz , x ≥ DN +1 ,
Pi−1
where the distances Di are given as before by Di = `=1 h` , with D1 = 0, i = 1, 2, · · · , N, and
γc = (β 2 − k02 n2c )1/2, γs = (β 2 − k02n2s )1/2, kxi = (k02 n2i − β 2)1/2 (i = 1, 2, · · · , N), and Ec , Es ,
and Ei1 , Ei2 the unknown amplitude constants. The tangential magnetic field Hz can easily be
determined from Eqs. (78) and (101). Using the continuity of the tangential field components
across the layer boundaries it is straightforward to relate the tangential fields (Eyi , dEyi /dx) =
(Ui , Vi ) where i = 0, 1, · · · , N with (U0 , V0 ) the tangential fields at the cover/layer-1 interface
(Ui , Vi ) the tanglential fields at the layer-i/layer-(i + 1) interface, and (UN , VN ) the tangential
fields at the layer-N/substrate interface. It can be shown [13, 17] that
U0 UN m11 m21 UN
= M̃1 M̃2 · · · M̃N −1 M̃N = . (102)
V0 | {z } VN m21 m22 VN
M̃
42
In the previous equation M̃ is the total transfer matrix between the cover and the substrate
regions where
sin(kxi hi )
cos(kxi hi ) −
M̃i = kxi if kxi 6= 0, (103)
kxi sin(kxi hi ) cos(kxi hi )
1 −hi
M̃i = , if kxi = 0, (104)
0 1
where i = 1, 2, · · · , N. For nontrivial solutions of Eq. (102) the following dispersion equation
can be easily derived
FT E (β) = γs m22 + γc m11 − γc γs m12 − m21 = 0. (105)
In the case of the TM modes the tangential fields are [Hyi , (1/n2i )dHyi /dx] = (Ui , Vi ) with
i = 0, 1, 2, · · · , N similar to the TE case. With these tangential fields, it is straightforward to
show that the dispersion equation can be derived by a simple replacement of kxi → +kxi /n2i
(only in amplitude terms of the previous equations and not inside phase terms), γs → +γs /n2s ,
and γc → +γc /n2c . For completeness in the case of kxi = 0 (linear field solution inside layer-i)
the element −hi of matrix M̃i appearing in Eq. (104) is replaced by −n2i hi . Then the resulting
dispersion equation is
γs γc γc γs
FT M (β) = 2
m11 + 2 m22 − 2 2 m12 − m21 = 0. (106)
ns nc nc ns
It is worth mentioning that both functions FT E and FT M are actually functions of β 2 which
implies that if +β satisfies Eq. (105) or (106) so is −β which corresponds to the backward in
z propagating mode.
43
12. Finite-Difference Frequency-Domain Method
In section the finite-difference frequency-domain (FDFD) method will be applied to the analysis
of general isotropic slab waveguides. The wave Helmholtz’s equations can be written in the
following vector form
" #
~
∇ε
2~ ~ ~ + k 2 εE
~ = 0,
∇E + ∇ ·E 0 (107)
ε
" #
~
∇ε
2~ ~ ~ ~ = 0,
∇ H + × ∇ × H + k02 εH (108)
ε
where ε is the relative permittivity of the region. In the case of TE modes solutions the electric
~ = ŷEy (x) exp(−jβz) and Eq. (107) can be simplified in the form
field is of the form E
d2 Ey 2 2 2
+ k0 n (x) − β Ey = 0, (109)
dx2
~ term is along the x direction whereas E
since the ∇ε ~ is along the y direction, and n2 (x) =
~ =
ε(x). Similarly, in the case of TM modes solutions, the magnetic field is of the form H
ŷHy (x) exp(−jβz) and Eq. (108) can be simplified in the form
d2 Hy 1 dn2 (x) dHy 2 2
2
− 2
+ k0 n (x) − β 2 Hy = 0, or (110)
dx n (x) dx dx
2 d 1 dHy
n (x) 2
+ k02 n2 (x) − β 2 Hy = 0, (111)
dx n (x) dx
where for the TM case the corresponding equation is more complicated than the TE one.
In the finite-difference frequency-domain (FDFD) method the derivatives are approximated
using central differences. For example, if U = Ey or Hy , U can be expanded into a Taylor series
around a point x. These leads to the following equations
dU ∆x d2 U (∆x)2 d3 U (∆x)3 d4 U (∆x)4
U(x + ∆x) ' U(x) + + 2 + 3 + 4 , (112)
dx 1! dx 2! dx 3! dx 4!
dU ∆x d2 U (∆x)2 d3 U (∆x)3 d4 U (∆x)4
U(x − ∆x) ' U(x) − + 2 − 3 + 4 , (113)
dx 1! dx 2! dx 3! dx 4!
and adding the last two equations the following equation can be derived
d2 U
U(x + ∆x) + U(x − ∆x) = 2U(x) + (∆x)2 + O[(∆x)4], (114)
dx2
where the O[(∆x)4] is the term that depends on the forth power of ∆x and can be neglected if
∆x is small (i.e. |∆x| 1). Now the discretization along the x axis is considered. For example
44
the continuous x axis is replaced by the points xi = j∆x where (i = 0, 1, · · · , M, M + 1). Then
if x = xi , x + ∆x = xi+1 , and x − ∆x = xi−1 , U(xi ) = Ui , U(xi+1 ) = Ui+1 , and U(xi−1 ) = Ui−1 ,
and n2 (xi) = n2i . The geometry is summarized in Fig. 20.
Ê Ë Ì Í Î Ï Ð Í Ñ Ï Ò Ó Ô Õ Ì Î Ð Í Ö Ó Ë Ë
Æ Ç
² ³
¼ ½ ¾ ¿ À
ª «
° ±
× Ø
Ù Ú Û Ü Ý Þ
¬
¶ ·
® ¯
¸ ¹
Á Â Ã Á Ä À Å Ä ¿ ´ µ
È É
ß à á â ã ä å â æ ä ç è é ê á ã å â ë è à à
Figure 20: The one dimensional (along the x axis) computational window for the application of the finite-
difference frequency-domain (FDFD) in general slab waveguides. It is assumed a multi-layered slab waveguide
without thus affecting the generality of the method. The top and bottom lines comprise either electric or
magnetic walls (for TE and TM modes respectively) where the electric or magnetic fields are forced to zero.
The distances dc and ds adjust the total width of the computational window. The grid along the x axis is
comprised of points xi , i = 1, 2, · · · , M .
Then from the last equation the second derivative can be approximated (central differences)
as follows
d2 U Ui−1 − 2Ui + Ui+1
= . (115)
dx2 (∆x)2
Using the approximation for the second derivative in Eq. (109) it can be transformed in the
following discretized form
Ui−1 − 2Ui + Ui+1
+ k02 n2i Ui = β 2Ui , i = 0, 1, · · · , M + 1. (116)
(∆x)2
45
It is straightforward to make the transformation X = k0 x and re-write the previous equation
in the form
Ui−1 − 2Ui + Ui+1
+ n2i Ui = N 2 Ui , i = 0, 1, · · · , M + 1. (117)
(∆X)2
where ∆X = k0 ∆x and N = β/k0 is the effective index. Applying the last equation for all points
xi and forcing the field to be zero at the upper and lower electric walls (i.e, U0 = UM +1 = 0 it
is straightforward to form the following system of equations
a1 b 0 ... 0 0 U1 U1
b a2 b 0 ... 0 U2 U
.. .2
.. .
0 b a3 b ... . . .
. = N2 .
. .. .. . . .. .. , (118)
.. . . . . . .
. ..
0 ... . . . b aM −1 b UM −1 UM −1
0 ... ... 0 b aM UM UM
where the coefficients ai = n2i − [2/(∆X)2 ] for i = 1, 2, · · · , M and b = 1/(∆X)2 . The last
equation is an eigenvalue/eigenvector equation of the form Av = λv. The eigenvalues of interest
are the ones for which ns < N < max{n(x)} and the corresponding eigenvectors are the
ones that specify the profile of the mode. The above matrix is a tridiagonal matrix which is
characteristic of the one-dimensional finite-difference implementation.
In order to apply the same approach for the case of TM polarization, Eq. (111), the
more complicated first term has to be expressed in discretized form. Following the procedure
explained in Ref. [51] it can be shown that
2 d 1 dU 1 2n2i 1 2n2i
n (x) = (Ui+1 − Ui ) − (Ui − Ui−1 ). (119)
dx n2 (x) dx (∆X)2 n2i+1 + n2i (∆X)2 n2i + n2i−1
Inserting the last approximation into Eq. (111) the following discretized wave equation is ob-
tained
1 2n2i 1 2n2i 2n2i
Ui−1 + n2i − + 2 Ui
(∆X)2 n2i + n2i−1 (∆X)2 n2i+1 + n2i ni + n2i−1
1 2n2i
+ Ui+1 = N 2 Ui , i = 1, 2, · · · , M. (120)
(∆X)2 n2i+1 + n2i
Writing the set of equations for each i (i = 1, 2, · · · , M) the following eigenvalue/eigenvector
46
problem can be formed
a 1 c1 0 ... 0 0 U1 U1
b2 a 2 c2 0 ... 0 U2 U2
.. .. ..
0 b3 a3 c3 ... . . 2 .
. .. .. .. .. .. .. =N .. , (121)
.. . . . . . . .
0 ... . . . bM −1 aM −1 cM −1 UM −1 UM −1
0 ... ... 0 bM aM UM UM
where now the matrix elements are given by
2 1 2n2i 2n2i
ai = ni − + 2 , (i = 1, 2, · · · , M), (122)
(∆X)2 n2i+1 + n2i ni + n2i−1
1 2n2i
bi = , (i = 1, 2, · · · , M) (123)
(∆X)2 n2i + n2i−1
1 2n2i
ci = , (i = 1, 2, · · · , M). (124)
(∆X)2 n2i+1 + n2i
An example case is presented next. A simple three-region slab waveguide is selected from
semiconductor materials. The parameters are nc = 1.0, nf = 3.4, ns = 3.1, h = 1.0µm, and
the freespace wavelength is λ0 = 1.3µm. The distances from the electric walls are dc = 0.5µm
and ds = 1.0µm. The grid spacing ∆x is selected as fraction of the freespace wavelength, i.e.,
∆x = λ0 /L where L an integer. The waveguide with these parameters supports two TE and
two TM modes. The results are summarized in Table 1. It is mentioned that the important
parameter for convergence is the number of grid points per minimum wavelength as this is
measured inside the material. Therefore, for the example case here the minimum wavelength is
inside the film region λmin = λ0 /3.4. Consequently, the smallest ∆x from the table corresponds
to λ0 /200 = λmin /58.8, i.e. in the denser grid results shown there are at least 58 grid points
per minimum wavelength. The resulting profiles of the supported modes that can be extracted
as the corresponding eigenvectors from Eqs. (118), (121), are shown in Fig. 21.
12.1 FDFD Method Based on Yee’s Cell
In this section the Yee’s cell [52] is going to be utilized for the determination of the effective
propagation constants (effective indices) and the corresponding field profiles for both TE and
TM modes for multilayered isotropic slab waveguides. The approach described by Rumpf [53]
will be applied appropriately modified in this case. It is assumed that all the materials are
non-magnetic (µr ' 1).
47
Table 1: Effective Indices Calculated by FDFD
TE Modes TM Modes
∆x N0T E N1T E N0T M N1T M
λ0 /20 3.3585913 3.2362825 3.3518412 3.2129633
λ0 /40 3.3579455 3.2333453 3.3515178 3.2109972
λ0 /60 3.3578155 3.2327610 3.3514548 3.2106177
λ0 /80 3.3577736 3.2325729 3.3514346 3.2104963
λ0 /100 3.3577538 3.2324843 3.3514251 3.2104392
λ0 /120 3.3577424 3.2324333 3.3514196 3.2104064
λ0 /150 3.3577336 3.2323938 3.3514154 3.2103809
λ0 /200 3.3577268 3.2323632 3.3514215 3.2103612
Exact 3.3577180 3.2323308 3.3514080 3.2103532
TE modes: n c = 1, nf = 3.4, n s = 3.1, λ 0 = 1.3µm, h = 1µm TM modes: n c = 1, nf = 3.4, n s = 3.1, λ 0 = 1.3µm, h = 1µm
1 1
Normalized Magnetic Field, Hy
Normalized Electric Field, E y
0.5 0.5
0 0
-0.5 -0.5
cover (nc) film (nf ) substrate (n s) cover (nc) film (nf ) substrate (n s)
-1 -1
-0.5 0 0.5 1 1.5 2 -0.5 0 0.5 1 1.5 2
Distance, x (µm) Distance, x (µm)
Figure 21: Example electric and magnetic field profiles determined by the FDFD method. This is a case
of a slab waveguide of nc = 1.0, nf = 3.4, ns = 3.1, h = 1.0µm, and λ0 = 1.3µm. The computational window
distances dc = 0.5µm and ds = 1.0µm were used. In these plots the x-axis matches the one used in Fig. 1. The
value of ∆x = λ0 /200 is used in these plots. The values of the effective indices are given in Table 1.
Starting from the TE modes determination, it is reminded that the non-zero field compo-
nents are (Ey , Hx , Hz ) where each field component is a function of (x, z) of the form Ey (z, x) =
Ey (x) exp(−jβz) = Ey (x) exp(−jk0 Nz) (as an example), where Ey (x) is the mode profile (for
the electric field - similar notation was used all over this chapter) and β, N are the effective
propagation constant and the effective refractive index of the TE mode that is sought, and
48
k0 is the freespace wavenumber. Using the normalization of the magnetic field components of
the form Hw0 = −jZ0 Hw (according to Rumpf [53]) where Z0 is the free space impedance, the
Maxwell’s equations can be written in the form
∂Ey ∂Ey
−= k0 Hx0 =⇒ − 0 = Hx0 , (125)
∂z ∂z
∂Ey ∂E y
+ = k0 Hz0 =⇒ + 0 = Hz0 , (126)
∂x ∂x
∂H 0 ∂Hx0 ∂H 0 ∂Hx0
− z+ = k0 r (x)Ey =⇒ − 0z + 0
= r (x0)Ey , (127)
∂x ∂z ∂x ∂z
where x0 = k0 x and z 0 = k0 z are normalized coordinates.
At this point the space domain discretization is done according to Yee’s scheme as shown
in Fig. 22. The fields are discretized in the coordinate system shown as Ey (zi , xj ) = Eyi,j taking
into account the staggering between various field components.
Eyi+1,j − Eyi,j i+1/2,j j
=⇒ jNEyj = Hx0 e−jN ∆z /2,
0
− 0
= Hx0 (128)
∆z
Eyi,j+1 − Eyi,j 0 i,j+1/2
Eyj+1 − Eyj j+1/2
= H z =⇒ = Hz0 , (129)
∆x0 ∆x0
H 0 i,j+1/2 − Hz0 i,j−1/2 Hx0 i+1/2,j − Hx0 i−1/2,j
− z + = jr Eyi,j =⇒
∆x0 ∆z 0
H 0 j+1/2 − Hz0 j−1/2 j
− z − jNHx0 = jr Eyj . (130)
∆x0
When ∆z 0 → 0 the 2D Yee’s grid collapses to the compact 1D Yee’s cell as it is shown also in the
right column (b figures) of Fig. 22. Therefore, the positions of Eyj and Hx0 j coincide (these field
components become collocated) while the Hz0 j+1/2 are shifted by ∆x0/2 (forming a staggered
1D grid). Now let’s define the vectors Ẽy = [Ey1, Ey2 , · · · , EyM ]T , H̃x0 = [Hx0 1 , Hx0 2 , · · · , Hx0 M ]T ,
and H̃z0 = [Hz0 1+1/2, Hz0 2+1/2, · · · , Hz0 M +1/2]T . Now let’s define the difference operators according
to Rumpf [53] as follows
−1 1 0 ... 0 0 1 0 0 ... 0 0
0 −1 1 0 ... 0 −1 1 0 0 ... 0
. ..
. . . ..
0 0 −1 1 0 −1 1 0 ... .
D̃xE0 = . D̃xH0 = .
.. .. . . .. . , .. .. . . . . .. . (131)
.. . . . . .. .. . . . . .
0 ... ... 0 −1 1 0 . . . . . . −1 1 0
0 ... ... 0 0 −1 0 ... ... 0 −1 1
49
î
ð
ñ ò
ï
ø ù ú û ý þ ü
ð
õ ÷
ø ù ú û ý ÿ ü
í
î ñ ò
õ ö
õ ö
ø ù ú û ü
ì
ì
ó ô
ï
ø ù ý ÿ ú û ü
ø ù þ ú û ü ø ù ÿ ú û ü ø ù ý þ ú û ü
õ ÷
ø ù ú û ÿ ü
í
!
ø ù ú û þ ü
8
&
' (
%
&
: ;
? @ A
. / 0 1 3 4 2
& <
=
+ -
? @ B
. / 0 1 3 5 2 9
7
8
#
< >
&
$ ' (
+ ,
+ ,
. / 0 1 2
: ;
"
"
) *
%
. / 3 5 0 1 2
. / 6 4 0 1 2 . / 6 5 0 1 2 . / 3 4 0 1 2
+ -
. / 0 1 6 5 2
? C B
# 7
< =
P W J M J N N L
U V V
? C A
. / 0 1 6 4 2
D E F G H I J J K L M J N N D O F H M P Q R E S T I J J K L M J N N
A
Figure 22: A 2D Yee’s cell and its 1D compact counterpart as it applies to the geometry of the slab waveguide
for both TE (top) and TM (bottom) modes. The point (zi , xj ) is represented as (i, j).
50
Using the previous definitions, equations (128–130) can be written in the following matrix form:
jN Ẽy = H̃x0 , (132)
1 E
0
D̃x0 Ẽy = H̃z0 , (133)
∆x
1 H 0
− D̃ 0 H̃ − jN H̃x0 = ˜r Ẽy . (134)
∆x0 x z
where ˜r is a diagonal matrix that contains the relative permittivity, i.e. ˜jr = r (xj ). Eliminat-
ing the magnetic field components from the last equations it is straightforward to obtain the
following eigenvalue/eigenvector equation
1
˜r + 0 2
D̃x0 D̃x0 Ẽy = N 2 Ẽy ,
H E
(135)
(∆x )
where N, and Ẽy are the eigenvalues and the corresponding eigenvectors (electric field profiles)
of the TE modes, respectively.
Observing Eq. (130) there might be a dilemma of what relative permittivity to use especially
in the case that the node is at an abrupt permittivity discontinuity. The only thing that is
certain is that Ey electric field is continuous at the permittivity discontinuity boundary. Based
on this, the following Taylor expansions can be made:
∂Dy ∂Dy
Dy (x + ∆x) = Dy (x) + ∆x = r (x)Ey (x) + ∆x ,
∂x ∂x
∂Dy ∂Dy
Dy (x − ∆x) = Dy (x) − ∆x = r (x)Ey (x) − ∆x .
∂x ∂x
Adding up the last two equations the following could be obtained
1
r (x)Ey (x) = [Dy (x + ∆x) + Dy (x − ∆x)] =
2
1
= [(x + ∆x)Ey (x + ∆x) + r (x − ∆x)Ey (x − ∆x)] =
2
1
= [(x + ∆x) + r (x − ∆x)] Ey (x),
2
where the continuity of Ey was utilized, i.e. Ey (x + ∆x) = Ey (x − ∆x) = Ey (x) for ∆x being
small. The previous averaging should be inserted into the eigenvalue/eigenvector equation (135)
in the form of changing the relative permittivity of the boundary nodes as
1 j−1
jr = r + j+1
r . (136)
2
As an example case, the same single film layer semiconductor waveguide that was presented
in the previous section is used . This three-region slab waveguide has the parameters nc = 1.0,
51
nf = 3.4, ns = 3.1, h = 1.0µm, and the freespace wavelength is λ0 = 1.3µm. The distances
from the electric walls are dc = 0.5µm and ds = 1.0µm. The results for the effective indices of
the two supported TE modes are shown in Table 2 and are compared to the previous solutions
of the Helmholtz equation with the collocated fields. It is observed that with any of the three
methods the results are very close. The bottom half of the table shows the error which is defined
as Error(%) = 100[Nexact − N(∆x)]/Nexact . It is observed that the Yee’s cell with averaging
at the boundary nodes gives slightly better results for the practical case of λ0 /80 ' λmin /20
where λmin = λ0 /nf .
In the case of TM modes the non-zero field components are (Hy , Ex , Ez ) where again each
field component is of the form Hy (z, x) = Hy (x) exp(−jβz) = Hy (x) exp(−jk0 Nz) (as an
example), where Hy (x) is the mode profile (for the magnetic field) and β, N are the effective
propagation constant and the effective refractive index of the TM mode that is sought, and k0
is the freespace wavenumber. Using again the normalization of the magnetic field components
of the form Hw0 = −jZ0 Hw the Maxwell’s equations can be written in the form
∂Hy0 ∂Hy0
− = k0 r Ex =⇒ − 0 = r Ex , (137)
∂z ∂z
∂Hy0 ∂Hy0
= k0 r Ez =⇒ = r Ez , (138)
∂x ∂x0
∂Ex ∂Ez ∂Ex ∂Ez
− = k0 Hy0 =⇒ − = Hy0 , (139)
∂z ∂x ∂z 0 ∂x0
where x0 = k0 x and z 0 = k0 z are normalized coordinates.
At this point the space domain discretization is done according to Yee’s scheme as shown in
Fig. 22. The fields are discretized in the coordinate system shown as Hy0 (zi, xj ) = Hy0 i,j taking
into account the staggering between various field components.
Hy0 i,j − Hy0 i−1,j j 0
− 0
= i−1/2,j
r Ex i−1/2,j =⇒ jNHy0 = jrxxEx j e+jN ∆z /2, (140)
∆z
0 i,j 0 i,j−1
Hy − Hy i,j−1/2 i,j−1/2
Hy0 j − Hy0 j−1 j−1/2
= r E z =⇒ = rzz Ez j−1/2 , (141)
∆x0 ∆x0
Ex i+1/2,j − Ex i−1/2,j Ez i,j+1/2 − Ez i,j−1/2 i,j
− + = Hy0 =⇒
∆z 0 ∆x0
Ez j+1/2 − Ez j−1/2 j
−jNEx j − 0
= Hy0 , (142)
∆x
j−1/2
where jrxx = r (xj ) and rzz = r (xj−1/2). When ∆z 0 → 0 the 2D Yee’s grid collapses to the
compact 1D Yee’s cell as it is shown also in the right column (b figures) of Fig. 22. Therefore,
52
Table 2: Effective Indices (of TE Modes) Calculated by FDFD either using Helmholtz Equation
(collocated fields) or using Yee’s Cell (staggered fields)
TE Modes
Helmholtz Yee’s Cell Yee’s Cell
No Averaging Averaging
∆x N0T E N1T E N0T E N1T E N0T E N1T E
λ0 /20 3.3585913 3.2362825 3.3585914 3.2362911 3.3570617 3.2310872
λ0 /40 3.3579455 3.2333453 3.3579314 3.2332928 3.3575479 3.2319896
λ0 /60 3.3578155 3.2327610 3.3578157 3.2327737 3.3576393 3.2321748
λ0 /80 3.3577736 3.2325729 3.3577720 3.2325782 3.3576744 3.2322471
λ0 /100 3.3577538 3.2324843 3.3577531 3.2324938 3.3576897 3.2322788
λ0 /120 3.3577424 3.2324333 3.3577426 3.2324471 3.3576981 3.2322964
λ0 /150 3.3577336 3.2323938 3.3577338 3.2324078 3.3577053 3.2323113
λ0 /200 3.3577268 3.2323632 3.3577268 3.2323769 3.3577109 3.2323231
λ0 /500 3.3577193 3.2323297 3.3577195 3.2323443 3.3577169 3.2323357
λ0 /750 3.3577185 3.2323262 3.3577187 3.2323409 3.3577176 3.2323371
λ0 /1000 3.3577183 3.2323250 3.3577184 3.2323397 3.3577178 3.2323376
Exact 3.3577180 3.2323308 3.3577180 3.2323308 3.3577180 3.2323308
TE Modes
Helmholtz Yee’s Cell Yee’s Cell
No Averaging Averaging
∆x N0T E N1T E N0T E N1T E N0T E N1T E
Error (%) Error (%) Error (%) Error (%) Error (%) Error (%)
λ0 /20 −0.0260 −0.1223 −0.0260 −0.1225 0.0195 0.0385
λ0 /40 −0.0068 −0.0314 −0.0064 −0.0298 0.0051 0.0106
λ0 /60 −0.0029 −0.0133 −0.0029 −0.0137 0.0023 0.0048
λ0 /80 −0.0017 −0.0075 −0.0016 −0.0077 0.0013 0.0026
λ0 /100 −0.0011 −0.0047 −0.0010 −0.0050 0.0008 0.0016
λ0 /120 −0.0007 −0.0032 −0.0007 −0.0036 0.0006 0.0011
λ0 /150 −0.0005 −0.0019 −0.0005 −0.0024 0.0004 0.0006
λ0 /200 −0.0003 −0.0010 −0.0003 −0.0014 0.0002 0.0002
λ0 /500 0.0000 0.0000 0.0000 −0.0004 0.0000 −0.0002
λ0 /750 0.0000 0.0001 0.0000 −0.0003 0.0000 −0.0002
λ0 /1000 0.0000 0.0002 0.0000 −0.0003 0.0000 −0.0002
53
the positions of Hy0 j and Ex j coincide (these field components become collocated) while the
Ez j−1/2 are shifted by −∆x0/2 (forming a staggered 1D grid). Now let’s define the vectors H̃y0 =
[Hy0 1, Hy0 2 , · · · , Hy0 M ]T , Ẽx = [Ex1 , Ex 2, · · · , Ex M ]T , and Ẽz = [Ez 1−1/2, Ez 2−1/2, · · · , Ez M −1/2]T .
Using the previous definitions, equations (140–142) can be written in the following matrix form:
jN H̃y0 = ˜rxxẼx , (143)
1 H 0
D̃ 0 H̃ = ˜rzz Ẽz , (144)
∆x0 x y
1 E
−jN Ẽx − D̃ 0 Ẽz = H̃y0 . (145)
∆x0 x
where ˜rxx and ˜rzz are diagonal matrices that contain the relative permittivity, i.e. ˜jrxx = r (xj )
and ˜jrzz = r (xj+1/2). Eliminating the electric field components from the last equations it is
straightforward to obtain the following eigenvalue/eigenvector equation,
1
˜rxx + ˜rxx D̃x0 ˜rzz D̃x0 H̃y = N 2 H̃y ,
E −1 H
(146)
(∆x0)2
where N, and H̃y are the eigenvalues and the corresponding eigenvectors (magnetic field profiles)
of the TM modes, respectively.
Observing Eq. (140) there might be a dilemma of what relative permittivity to use especially
in the case that the node is at an abrupt permittivity discontinuity (similarly to the TE mode
case). The only thing that is certain in this case, is that Dx electric displacement field is
continuous across the permittivity boundary (since there are no free charges). Based on this,
the following Taylor expansions can be made:
Dx (x + ∆x) ∂Ex
= Ex (x + ∆x) = Ex (x) + ∆x ,
r (x + ∆x) ∂x
Dx (x − ∆x) ∂Ex
= Ex (x − ∆x) = Ex (x) − ∆x .
r (x − ∆x) ∂x
The last two equations can be manipulated as following in order to obtain the most appropriate
expression for the relative permittivity that will be used in Eq. (140).
1 Dx (x + ∆x) Dx (x − ∆x)
+ = Ex (x),
2 r (x + ∆x) r (x − ∆x)
1 1 1
Dx (x) + = Ex (x),
2 r (x + ∆x) r (x − ∆x)
Dx (x)
= Ex (x),
r,eq (x)
Dx (x) = r,eq (x)Ex (x),
−1
1 1 1
where r,eq (x) = + .
2 r (x + ∆x) r (x − ∆x)
54
where used of the boundary condition Dx (x + ∆x) = Dx (x − ∆x) = Dx (x) when ∆x is small.
The above averaging (harmonic average) should be inserted into the eigenvalue/eigenvector
equation (146) in the form of changing the relative permittivity of the boundary nodes of ˜r as
−1
1 1 1
jrxx = j−1
+ j+1 . (147)
2 rxx rxx
Using the same example as before the above procedure is implemented for the determination
of the TM modes. The results are tabulated in Table 3. Again it is observed that the Yee’s
cell with the harmonic averaging gives slightly better results than the solution of Helmholtz’s
equation using the Kawano scheme described in the previous subsection. The Yee’s cell without
the averaging performs consistently worse than the other two schemes.
12.1.1. Average-Scheme Modification for Arbitrary Boundary Location
The previously presented average scheme for the FDFD method for slab waveguides is valid
only for the special case where the permittivity discontinuity plane boundary passes through an
electric field node (Ey node for TE modes) [Eq. (136)] or a magnetic field node (Hy node for TM
modes) [Eq. (147)]. In this section let’s consider the case that the permittivity discontinuity
boundary crosses a Yee’s cell at an arbitrary location. This is shown schematically for the case
of TE and TM modes in Figs. 23a and 23b, respectively.
55
Table 3: Effective Indices (of TM Modes) Calculated by FDFD either using Helmholtz Equation
(collocated fields) or using Yee’s Cell (staggered fields)
TM Modes
Helmholtz Yee’s Cell Yee’s Cell
Kawano [51] No Averaging Averaging
∆x N0T M N1T M N0T M N1T M N0T M N1T M
λ0 /20 3.3518412 3.2129633 3.3496125 3.2041094 3.3512255 3.2110965
λ0 /40 3.3515178 3.2109972 3.3504428 3.2067131 3.3513622 3.2105083
λ0 /60 3.3514548 3.2106177 3.3507380 3.2077721 3.3513869 3.2104143
λ0 /80 3.3514346 3.2104963 3.3509033 3.2083858 3.3513963 3.2103793
λ0 /100 3.3514251 3.2104392 3.3509982 3.2087442 3.3514004 3.2103641
λ0 /120 3.3514196 3.2104064 3.3510631 3.2089916 3.3514026 3.2103558
λ0 /150 3.3514154 3.2103809 3.3511305 3.2092505 3.3514045 3.2103487
λ0 /200 3.3514215 3.2103612 3.3511995 3.2095179 3.3514060 3.2103430
λ0 /500 3.3514086 3.2103396 3.3513237 3.2100031 3.3514076 3.2103368
λ0 /750 3.3514082 3.2103373 3.3513516 3.2101129 3.3514078 3.2103361
λ0 /1000 3.3514081 3.2103364 3.3513656 3.2101682 3.3514078 3.2103358
Exact 3.3514080 3.2103532 3.3514080 3.2103532 3.3514080 3.2103532
TM Modes
Helmholtz Yee’s Cell Yee’s Cell
Kawano [51] No Averaging Averaging
∆x N0T M N1T M N0T M N1T M N0T M N1T M
Error (%) Error (%) Error (%) Error (%) Error (%) Error (%)
λ0 /20 −0.0129 −0.0813 0.0536 0.1945 0.0054 −0.0232
λ0 /40 −0.0033 −0.0201 0.0297 0.1134 0.0014 −0.0048
λ0 /60 −0.0014 −0.0082 0.0200 0.0804 0.0006 −0.0019
λ0 /80 −0.0008 −0.0045 0.0153 0.0613 0.0003 −0.0008
λ0 /100 −0.0005 −0.0027 0.0124 0.0501 0.0002 −0.0003
λ0 /120 −0.0003 −0.0017 0.0103 0.0424 0.0002 −0.0001
λ0 /150 −0.0002 −0.0009 0.0083 0.0343 0.0001 0.0001
λ0 /200 −0.0004 −0.0002 0.0063 0.0260 0.0001 0.0003
λ0 /500 0.0000 0.0004 0.0025 0.0109 0.0000 0.0005
λ0 /750 0.0000 0.0005 0.0017 0.0075 0.0000 0.0005
λ0 /1000 0.0000 0.0005 0.0013 0.0058 0.0000 0.0005
56
Z «
b c d e f g ³ ´ µ ¶ · ¸
à á
u v w x y z { Æ Ç È É Ê Ë Ì
æ ç
| } ~ Í Î Ï Ð Ñ Ò Ó
\ ] ®
n o ¿ À
Y ª
^ _ ` a ¯ ° ± ²
l m ½ ¾
j k » ¼
p q r s t Á Â Ã Ä Å
X ©
X ©
[ ¬
h i | } ~ ¹ º
Í Î Ñ Ò Ï Ð Ó
| } ~ | } ~ | } ~ Í Î Ô Õ Ï Ð Ó Í Î Ô Ò Ï Ð Ó Í Î Ñ Õ Ï Ð Ó
¦ § ¨ ö ÷ ø
Y ª
| } ~ Í Î Ï Ð Ô Ò Ó
â ã
¡ ¢ £ ¤ ¥ è é ê ë ì í í ì î ì í ï ð ñ ò ó ô õ ê ï
Þ ß
| } ~ Í Î Ï Ð Ô Õ Ó
ä å
Ö × × Ø Ù Ú Û Ú Ü Ü Ý
(a) (b)
Figure 23: (a) A 2D Yee’s cell as it applies to the geometry of the slab waveguide for for TE modes. The
point (zi , xj ) is represented as (i, j). The location of the boundary is shown with dashed-dotted line. It is
assumed that the grid size along the z direction tends to zero (∆z → 0). For the sake of simplicity the magnetic
fields and the grid lengths are shown as unprimed. (b) Same as (a) but for the case of TM modes.
In order to determine the averaging scheme in this case the procedure developed by Hwang
and Cangellaris [54] will be utilized. For the TE modes, let’s apply the Ampere’s law (in time-
harmonic form) in its integral form, for the area S specified by the cell centered at the Eyi,j and
the contour C defined by the perimeter of the cell (see Fig. 23a).
ZZ I
jω ~ ~
D · dS = H ~ =⇒
~ · d`
S C
Z xj +∆x/2 Z zi +∆z/2 I
jω0 r (x)Ey (z, x)dxdz = [Hx dx + Hz dz] =⇒
xj −∆x/2 zi −∆z/2 C
Z xj +∆x/2
jω0 r (x)Eyj (x)dx[e−jk0 N zi ∆z] = − Hzj+1/2 − Hzj−1/2 [e−jk0 N zi ∆z] −
xj −∆x/2
Hxj [−jk0 N∆ze−jk0 N zi ]∆x =⇒
Z xj +∆x/2
jω0 r (x)Eyj (x)dx = − Hzj+1/2 − Hzj−1/2 − Hxj [−jk0N]∆x =⇒
xj −∆x/2
jω0 [F 1 + (1 − F )2] Eyj ∆x = − Hzj+1/2 − Hzj−1/2 − Hxj [−jk0N]∆x =⇒
Hz0 j+1/2 − Hz0 j−1/2 j
[F 1 + (1 − F )2] Eyj = jr,ef f Eyj = − 0
− jNHx0 , (148)
∆x
where the last equation is the same with Eq. (130) with ejr replaced by jr,ef f . In the above
derivation of the RHS integral with respect to x variable the continuity of the Ey (tangential
57
to the boundary) component was used. Therefore, in this arbitrary location of the permittivity
discontinuity boundary the corresponding averaging is expressed by the following equation
[which is the generalized form Eq. (136)],
jr,ef f = [F 1 + (1 − F )2] = jr = F rj−1 + (1 − F )j+1
r . (149)
The above equation reduces to the Eq. (136) in the case that the boundary passes from the Eyi,j
node since in this case F = 1/2. Since the materials are non-magnetic no averaging is needed
for the other equations since the relative permeability is everywhere unity.
Now let’s examine the case of the TM modes. Again the Ampere’s law (in time-harmonic
i−1/2,j
form) will be used. Assume the cell that is centered at the point of Ex and is perpendicular
to the xz-plane and apply again the Ampere’s law as it was done previously. The following can
be easily shown (where in this case S the area of the cell of size ∆z × Ly with Ly a unit length
in the y direction and C its perimeter)
ZZ I
jω ~ ~
D · dS = ~ · d`
H ~ =⇒
S C
Z Z zi +∆z/2 I
jω r (x)Ex(x, z)dydz = Hy dy =⇒
Ly zi −∆z/2 C
Z zi +∆z/2
jω0 r (x)Ex (x)e−jk0 N z dzLy = Hyi−1,j − Hyi,j Ly = Hyj e−jk0 N zi jk0 N∆zLy =⇒
zi −∆z/2
zi +∆z/2
e−jk0 N z
jω0 r (x)Ex(x) = Hyj e−jk0 N zi jk0 N∆z =⇒
−jk0 N zi −∆z/2
jω0 r (x)Ex(x) = jk0 NHyj =⇒
j
r (x)Ex(x) = jNHy0 . (150)
The last equation is similar to Eq. (140). However, due to the existence of the permittivity
boundary there is some ambiguity in the selection of the term r (x)Ex (x). For example the
surface S in Ampere’s law can be any surface that ends up on the contour C. Therefore x
could be selected anywhere in the cell. For this reason it is reasonable to assume that since Dx
is continuous at the permittivity boundary to take an average value of Ex (x) and an effective
relative permittivity term. The weighted average of the the electric field at the center of this
58
cell is
F 1−F
Exj = F Ex(x−
b ) + (1 − F )Ex(x+
b ) = Dxj +
r1 r2
Dxj
= where
jr,ef f
−1
F 1−F
r,ef f = + = jrxx, (151)
r1 r2
and xb is the position of the permittivity discontinuity boundary while Dxj is the approximately
constant electric flux density within the cell used. The same result can be obtained if the cell
centered at Hyi,j is used and the Faraday’s law is used with S the area of the cell (∆x × ∆z) and
contour C the perimeter of the cell. The above defined effective permittivity is the generalization
of the average permittivity defined in Eq. (147). Of course these two expressions coincide in
the case of the permittivity boundary passing through the Hyi,j node (F = 1/2).
In the TM modes case there is a need to apply the Ampere’s law to another cell to generate
i,j−1/2
the equivalent of Eq. (141). In this case the cell centered at Ez is utilized with its area
being perpendicular to xz plane and its dimensions of ∆x × Ly .
ZZ I
jω D ~
~ · dS = H ~ =⇒
~ · d`
S C
Z Z xj I
jω r (x)Ez (x, z)dydx = Hy dy =⇒
Ly xj −∆x C
Z xj
jω0 r (x)Ez (x)e−jk0 N z dxLy = Hyi,j − Hyi,j−1 Ly = Hyj − Hyj−1 e−jk0 N zi Ly =⇒
xj −∆x
Z xj
jω0 r (x)Ez (x)dx = Hyj − Hyj−1 =⇒
xj −∆x
jω0 [F1r1 + (1 − F1)r2] Ezj−1/2 ∆x = Hyj − Hyj−1 =⇒
Hy0 j − Hy0 j−1
[F1r1 + (1 − F1)r2 ] Ezj−1/2 = rzz
j−1/2 j−1/2
Ez = , where
∆x0
j−1/2
r,ef f = F1r1 + (1 − F1)r2 = rzz , (152)
where the term F1 = F + 1/2 (if F ≤ 1/2). In the case that F > 1/2 then the cell which is
i,j+1/2
centered on Ez is affected. Summarizing the new effective permittivity term to be used in
Eq. (141) is given by
j−1/2
F1r1 + (1 − F1)r2 = rzz , F1 = F + 1/2, for F ≤ 1/2,
r,ef f = (153)
j+1/2
F2r1 + (1 − F2)r2 = rzz , F2 = F − 1/2, for F > 1/2.
59
In the case that the permittivity boundary passes through the Hyi,j node (F = 1/2) then
r,ef f = r1 as it was used in the previous section. I.e., in this case the relative permittivity of
Eq. (141) does not require any averaging.
When a single film layer slab waveguide is analyzed, there is always possible to select ∆x
in such a manner as to force either electric field nodes (for TE modes) or magnetic field nodes
(for TM modes) to be located exactly on the two permittivity discontinuities boundaries. This
simplifies significantly the averaging scheme which is as the one presented in the previous
section. However, when there are more film layers, in general it is not possible to have field
nodes located at all permittivity boundaries. As a comment, it can be stated, that a similar
procedure could be used in the case that magnetic materials are used. In the latter case similar
appropriate average relative permeabilities can be defined.
60
13. Graded-Index Slab Waveguides
As it was mentioned earlier, waveguide fabrication techniques such as metal-in-diffusion, ion
bombardment, or ion migration result in waveguides with graded variation of the refractive
index instead of the step-index profile that was studied up to this point. An example graded-
index waveguide is shown in Fig. 24. A more general refractive index profile of a graded-index
waveguide is shown in Fig. 25. Starting from the left-hand side, there is the cover region of
refractive index nc , then there is the substrate top region that is affected by the fabrication
process, of refractive index ns (x), and finally, deep into the substrate (i.e. for large enough values
of x), is the unaffected from the fabrication process part of the substrate with refractive index
ns . After, analyzing multi-layered slab waveguides any resulting graded index profile could
potentially be approximated by a stair-step approximation and could be analyzed according to
the theory developed in Sec. 11. However, in this section a short summary of an approximate
technique will be presented, even if its utility at the present day computer era is questionable.
The method has the initials WKB (from the initials of the researchers who first proposed it,
Wentzel, Kramers, and Brillouin) and was originated in the area of quantum mechanics.
ü ý ù þ
ü ý ù þ
Figure 24: An example graded-index slab waveguide with an exponential-like refractive index profile. The
dimension along the y-axis is much larger than the dimension alon the x-axis so the light confinement is along
the x direction and its propagation along the z direction.
Before presenting the basics of WKB method it is useful to start from the Maxwell’s equa-
tions once more but now assuming that = (x) for the graded-index waveguide. Assuming
that the materials are linear, nonmagnetic, and isotropic, the Helmholtz’s equation is written
61
Figure 25: An example refractive index profile of a graded-index waveguide. For a mode of effective index
0
Neff the turning points xA and xB are shown. For a mode of effective index Neff < n(x = x0A ) the turning
0 0
points are xA (which is at the cover/graded-index-layer boundary) and xB could vary within the graded-index
region.
as
~ ∇
∇ ~ ·E
~ − ∇2 E
~ = k02 n2 (x)E.
~ (154)
In order to simplify the solution it is now assumed that TE polarized solutions are sought.
~ = ŷEy (x) exp(−jβz). The y polarized electric field is perpedicular to the gradient
Therefore, E
~ 2·E
of the relative permittivity variation, i.e. ∇n ~ = 0. Using the Gauss’s law ∇
~ ·D
~ =0⇒
~ 2·E
∇n ~ + n2 ∇
~ ·E
~ =0⇒∇
~ ·E
~ = 0. Consequently the divergence of the electric field is zero,
and the previous Helmholtz’s equation simplifies to the form
d2 Ey 2 2 2
+ k0 n (x) − β Ey (x) = 0. (155)
dx2
The last equation is a second-order ordinary differential equation with varying coefficients.
Usually, there are only very specific relative permittivity, n2 (x), profiles for which analytical
solutions are feasible. For example analytical solutions can be found in the case of parabolic,
linear, and exponential profile [44]. In general, though there are no analytical solutions of this
equation. As it was mentioned earlier, one approximate approach is the WKB method. In this
62
method a trial solution for Ey is selected of the form ψ = ψ0 exp[jk0 S(x)]. Expanding S(x) in
the form
1 1
S(x) = S0 (x) + S1 (x) + 2 S2(x) + · · · , (156)
k0 k0
results in a trial solution of the form
1
ψ(x) = ψ0 exp jk0 S0 (x) + jS1 (x) + j S2(x) + · · · , (157)
k0
where all higher order terms can be neglected since k0 is a large number for optical frequencies.
Using the previous trial solution in Helmholtz’s equation, and retaining terms of the same order
in k0 , the S0 and S1 terms are given by
Z
1 2 2 1/2
S0 (x) = k0 n (x) − β 2 dx, (158)
k0
j dS0
S1 (x) = ln , (159)
2 dx
and the resulting trial solution ψ(x) has the form
Z
ψ0
√ exp ±j Qdx , if [k02n2 (x) − β 2] = Q2 > 0,
Q
Ey (x) = ψ(x) = Z (160)
ψ0
√ exp ± P dx , if [k02n2 (x) − β 2] = −P 2 < 0.
P
The above solutions for the example case shown in Fig. 25 take the form [44]
Z xA
ψ0 2 2 2 0 1/2 0
exp − [β − k0 n (x )] dx , x < xA ,
2[β 2 − k02 n2 (x)]1/4 x
Z x
ψ0 π
2 2 0 2 1/2 0
Ey (x) = ψ(x) = cos [k0 n (x ) − β ] dx − , xA < x < xB ,
[k02n2 (x) − β 2]1/4 xA 4
Z x
ψ
0 2 2 2 0 1/2 0
exp − [β − k0 n (x )] dx , x > xB ,
2[β 2 − k02 n2 (x)]1/4 xB
(161)
where the solution between the turning points is propagating and elsewhere is evanescent. The
above solution diverges when x = xA or x = xB , i.e. at the turning points. Therefore, the
above solutions are valid somewhat away from the turning points. Near the turning points
special connections formulas are necessary (see Ref. [1]) in order to describe the electric field.
However, further details of the WKB will not be pursued here. The interested reader can refer
to Refs. [1] and [44]. Instead, a simple approach developed in Ref. [45] will be presented. By
63
ÿ
Figure 26: A picture diagram inside a graded-index waveguide for a guided mode. The turning points at
xA and xB are shown. In addition, between x and x + dx a wavevector ~k(x) is shown where |~k(x)| = k0 n(x) as
an approximation.
doing so a dispersion equation for the determination of the propagation constant β can be easily
developed.
Using the ray diagram shown in Fig. 26 it can be assumed that locally, between x and x + dx
the ray “sees” a constant refractive index of n(x). Consequently, the plane wave solution can
be used for the ray between x and x + dx. This means that the local wavevector ~k(x) satisfies
locally the dispersion for a plane wave. Therefore,
kx2 (x) + β 2 = k02 n2 (x) = |~k(x)|2. (162)
Defining again β = k0 N, where N = Nef f is the effective index of the resulting guided mode, and
using the angle θ(x) as defined in Fig. 26, it can be easily shown that θ(x) = cos−1 [N/n(x)].
Of course θ(x) takes real values when n(x) > N = Nef f as it can be seen in Fig. 26 when
xA < x < xB . However, when x < xA or x > xB , θ(x) takes purely imaginary values. At the
turning points θ(xA ) = θ(xB ) = 0, thus the ray becomes parallel to the z axis at the turning
points. For a given value of the effective index, the turning points can be evaluated by inverse
solution of the equation n(x) = N ⇒ xw = n−1 (N) (w = A or B) and n−1 (N) specifies the
inverse function. Using the analogous of Eq. (4) for this case, the following dispersion condition
64
can be written
Z xB
−2 kx (x)dx + φA + φB = 2νπ, ν = 0, ±1, ±2, · · · , (163)
xA
where φA and φB are the phase shifts occurring at the turning points xA and xB , respectively.
These phase shifts can be determined by the use of Fresnel reflection coefficient in the limiting
case between xA − ∆x < x < xA + ∆x (and similarly for xB ) where ∆x (∆x > 0) is a small
variation around xA (or xB ). Using the TE polarization Fresnel coefficient for the ray going
from xA + ∆x to xA − ∆x (as seen in Fig. 26) the following can be written
kx (xA − ∆x)
1−
kx (xA + ∆x) − kx (xA − ∆x) kx (xA + ∆x)
r= = , (164)
kx (xA + ∆x) + kx (xA − ∆x) kx (xA − ∆x)
1+
kx (xA + ∆x)
and using Eq. (162)
[n2(xA − ∆x) − N 2 ]1/2
1−
[n2 (xA + ∆x) − N 2 ]1/2
r= . (165)
[n2 (xA − ∆x) − N 2 ]1/2
1+ 2
[n (xA + ∆x) − N 2 ]1/2
At this point, it is necessary to use a Taylor series expansion for the square of the refractive
index and retain up to first-order terms. This results in the following approximations:
dn dn
n2 (xA − ∆x) ' n2 (xA ) − ∆x = N 2 − ∆x, (166)
dx dx
dn dn
n2 (xA + ∆x) ' n2 (xA ) + ∆x = N 2 + ∆x. (167)
dx dx
Using the previous approximations the reflection coefficient becomes
2 1/2 1/2
[n (xA − ∆x) − N 2 ] −(dn/dx)∆x
1− 1−
[n2 (xA + ∆x) − N 2 ] +(dn/dx)∆x
r = 2 1/2
' 1/2 =
[n (xA − ∆x) − N 2 ] −(dn/dx)∆x
1+ 1+
[n2 (xA + ∆x) − N 2 ] +(dn/dx)∆x
1−j π
= = 1 exp −j = 1 exp(jφA ). (168)
1+j 2
Therefore, the resulting phase shift φA = −π/2. Similarly, the phase shift at turnig point xB is
also φB = −π/2. Replacing the two phase factors in Eq. (163) the dispersion equation for the
determination of the effective index for a graded index slab waveguide becomes
Z xB =n−1 (N ) p
2 0 2 0 1
k0 (n (x ) − N ) dx = ν + π, ν = 0, 1, 2, · · · . (169)
xA =n−1 (N ) 2
65
In the latter dispersion equation the unknown effective index N is inside the integrand as well
as in the integration limits. Of course the above equation can be solved only numerically. The
solution provides the effective index of the mode while the field profile can be approximated by
Eqs. (161) (relatively away though from the turning points).
Example 1: The first example case is comprised of a graded index slab waveguide with a
relative permittivity profile of the form
2 (x − x0 )2
n (x) = εs + ∆ε exp − , (170)
w02
where it is assumed that the variation is inside the substrate. For εs = 4.80, ∆ε = 0.045,
x0 = 8µm, w0 = 2µm, and λ0 = 0.6328µm, the relative permittivity variation is shown in
Fig. 27 along with a stair-step approximation by 50 layers (in the interval between x0 − 4w0 <
x < x0 + 4w0 ). The effective indices were determined with three methods: (a) using the multi-
layered waveguide analysis with the stair-step approximation of the profile (comprised of 50 film
layers), (b) using the finite-difference frequency-domain (FDFD) method (presented in Sec. 12),
and (c) the WKB method, i.e. the numerical solution of Eq. (169).
Number of Discrete Layers = 50
2.204
2.202
2.2
Graded Index Profile
2.198
2.196
2.194
2.192
2.19
0 2 4 6 8 10 12 14 16
Distance x (µm)
Figure 27: An example graded index waveguide profile of the form n2 (x) = εs + ∆ε exp[−(x − x0 )2 /w02 ]. The
parameters are εs = 4.80, ∆ε = 0.045, x0 = 8µm, w0 = 2µm, and λ0 = 0.6328µm. A stair-step approximation
of the graded index profile is also shown for 50 layers of equal thickness of h = 8w0 /50.
The effective indices are shown tabulated in Table 4. It can be seen that the effective indices
agree very well for all methods. The multi-layered waveguide analysis determines the solution
66
Table 4: Effective Indices of Example # 1 Graded Index Waveguide
Multilayered Waveguide Finite Finite
Approximation Difference Difference WKB
T Eν Number of Layers = 50 ∆x = 0.05µm ∆x = 0.025µm Approximation
Nν Nν Nν Nν
T E0 2.198894532 2.198926188 2.198925969 2.198818399
T E1 2.195047282 2.194992344 2.194991579 2.194884443
T E2 2.192239107 2.192152744 2.192151661 2.192051926
T E3 2.190896246 Not Found Not Found Not Found
T E3 which both the FDFD and the WKB methods fail to determine. For the FDFD analysis this
is due to the relatively small computational window that is used. Since T E3 is near cutoff the
evanescent tails extend well beyond the area shown in Fig. 27 and larger computational window
could be needed for the determination of this mode. The WKB approximation fails to determine
this mode since it is near cutoff and the corresponding approximations may not hold or it could
be just a numerical issue. Furthermore, the T E3 mode found by the multilayered approximation
of the graded-index profile could be a product of this discretization. The normalized electric field
profiles of the calculated T Eν modes are shown in Fig. 28 when the multi-layered slab waveguide
theory is used in conjunction with the stair-step approximation of the graded refractive index
profile. Similar graphs are shown in Fig. 29 when the FDFD method is used.
Example 2: In this example the graded relative permittivity profile has the following form
2
x
εs + ∆ε exp − 2 , if x > 0,
2 w0
n (x) = (171)
2
nc , if x < 0.
For εs = 4.80, ∆ε = 0.045, w0 = 4µm, and λ0 = 0.6328µm, and nc = 1.0, the relative
permittivity variation is shown in Fig. 30 along with a stair-step approximation by 40 layers (in
the interval between 0 < x < 5w0 ) and it is shown in Fig. 30. Again, the effective indices are
determined with three methods: (a) the multi-layered waveguide analysis with the stair-step
approximation of the profile (40 film layers), (b) using the finite-difference frequency-domain
(FDFD) method (presented in Sec. 12), and (c) the WKB method. However, for the WKB
method, some special care has to be taken since the turning point xA falls always at the cover-
67
TE TE
TE0 (N0 = 2.198894) TE1 (N1 = 2.195047)
1 1
0.8
0.5
Normalized Ey
Normalized E
0.6
0.4
−0.5
0.2
0 −1
−5 0 5 10 15 20 25 −5 0 5 10 15 20 25
Distance x (µm) Distance x (µm)
TE TE
TE2 (N2 = 2.192239) TE3 (N3 = 2.190896)
1 1
0.8
0.6
0.5
0.4
Normalized Ey
Normalized Ey
0.2
0
0
−0.2
−0.5
−0.4
−0.6
−0.8 −1
−5 0 5 10 15 20 25 −5 0 5 10 15 20 25
Distance x (µm) Distance x (µm)
Figure 28: Normalized electric field profiles of T E0 , T E1 , T E2 , and T E3 modes for the example # 1 graded-
index slab waveguide. The profile was calculated using the multi-layered slab waveguide theory. The vertical
dashed lines show the cover-first film layer boundary (left dashed line at x = 0) and the boundary between
the last film layer and substrate region (right dashed line). The boundaries between the film layers due to the
stair-step approximation are not shown for simplicity of the figure.
substrate boundary and this means that the phase term φA should be given by the standard
term expressing the phase shift upon total internal reflection between the graded index layer
near the edge and the cover index layer. Therefore Eq. (169) has to be modified as
Z xB =n−1 (N ) p ( 1/2)
2 2 −
N − n (x = 0 ) π
k0 (n2 (x0) − N 2 ) dx0 −tan−1 2 + 2
− = νπ, ν = 0, 1, 2, · · · .
xA =0 n (x = 0 ) − N 4
(172)
or equivalently
Z ( 1/2)
xB =n−1 (N ) p
N 2 − n2c π
k0 (n2 (x0) − N 2 ) dx0 −tan−1 − = νπ, ν = 0, 1, 2, · · · .
xA =0 εs + ∆ε − N 2 4
(173)
The effective indices are shown tabulated in Table 5. It can be seen that the effective
indices agree very well for all methods. The normalized electric field profiles of the calculated
68
FDFD Results
1
TE
2
0.8
TE1
0.6 TE
0
Normalized Modal Field
0.4
0.2
−0.2
−0.4
−0.6
−0.8
−1
0 2 4 6 8 10 12 14 16
x−axis (µm)
Figure 29: Normalized electric field profiles of T E0 , T E1 , and T E2 modes for the example # 1 graded-index
slab waveguide. The profile was calculated using the FDFD method with ∆x = 0.025µm.
Number of Discrete Layers = 40 Number of Discrete Layers = 40
2.5
2.2
Graded Index Profile
Graded Index Profile
2 2.195
2.19
1.5
2.185
2.18
1
−5 0 5 10 15 20 −4 −2 0 2 4 6 8 10
Distance x (µm) Distance x (µm)
Figure 30: Another example of a graded index waveguide profile of the form n2 (x) = εs + ∆ε exp[−x2 /w02 ]
for x > 0, n2 (x) = n2c for x < 0 and . The parameters are εs = 4.80, ∆ε = 0.045, nc = 1.0, w0 = 4µm,
and λ0 = 0.6328µm. A stair-step approximation of the graded index profile is also shown for 40 layers of
equal thickness of h = 5w0 /40. The right plot is a zoomed version of the left in order to show the stair-step
approximation.
T Eν modes are shown in Fig. 31 when the multi-layered slab waveguide theory is used in
conjunction with the stair-step approximation of the graded refractive index profile. Similar
graphs are shown in Fig. 32 when the FDFD method is used.
69
Table 5: Effective Indices of Example # 2 Graded Index Waveguide
Multilayered Waveguide Finite Finite
Approximation Difference Difference WKB
T Eν Number of Layers = 40 ∆x = 0.05µm ∆x = 0.025µm Approximation
Nν Nν Nν Nν
T E0 2.197868744 2.197905044 2.197877837 2.197825988
T E1 2.194253274 2.194235177 2.194204855 2.194152293
T E2 2.191721704 2.191680005 2.191658955 2.191615558
TE TE
TE0 (N0 = 2.197869) TE1 (N1 = 2.194253)
0.8
1
0.6
0.4
0.8
0.2
Normalized Ey
y
Normalized E
0.6
0
−0.2
0.4
−0.4
0.2
−0.6
−0.8
0
−1
−5 0 5 10 15 20 −5 0 5 10 15 20
Distance x (µm) Distance x (µm)
TE
TE2 (N2 = 2.191722)
1
0.8
0.6
0.4
Normalized Ey
0.2
−0.2
−0.4
−0.6
−0.8
−5 0 5 10 15 20
Distance x (µm)
Figure 31: Normalized electric field profiles of T E0 , T E1 , and T E2 modes for the example # 2 graded-index
slab waveguide. The profile was calculated using the multi-layered slab waveguide theory (using 40 layers in the
stair-step approximation). The vertical dashed lines show the cover-first film layer boundary (left dashed line at
x = 0) and the boundary between the last film layer and substrate region (right dashed line). The boundaries
between the film layers due to the stair-step approximation are not shown for simplicity of the figure.
70
FDFD Results
1
TE
2
0.8 TE
1
0.6 TE0
Normalized Modal Field
0.4
0.2
−0.2
−0.4
−0.6
−0.8
−5 0 5 10 15 20
x−axis (µm)
Figure 32: Normalized electric field profiles of T E0 , T E1 , and T E2 modes for the example # 2 graded-index
slab waveguide. The profile was calculated using the FDFD method with ∆x = 0.025µm.
71
14. Lossy, Leaky, and Active Planar Multilayer Waveg-
uides
In this section numerical methods for determining the complex propagation constants and their
corresponding modal characteristics for planar multilayer waveguides are presented. In the
previous lossless waveguides results presented, the real propagation constants were determined
using the bisection method [8]. However, the calculation of the complex propagation constants
that they can respresent lossy, active, or leaky modes, is a lot more complicated. Initially, three
similar globally converging, and robust numerical methods for determining zeros of analytic
functions will be summarized. Then these methods are applied to the case of the planar multi-
layer waveguides. The three numerical mathematical methods are (a) the Argument Principal
Method (APM) [10], (b) the Abd-ellal, Delves, and Reid (ADR) method [34, 36, 37], and the
winding-number based on the analytic function’s phase method [35]. In all cases the methods
find good estimates of the function’s zeros which are subsequently refined using the Müller’s
iterative method [9] with deflation [8].
14.1 Basic Theory of the Principle Argument Method (APM)
The application of the argument principle method for the determination of the modes of planar
multilayer waveguides was first published by Smith et al. [14,15] and Anemogiannis and Glytsis
[16]. The argument principle method (APM) [10] relies on the generalized argument principle
theorem from complex analysis. This theorem relates the number of zeros of an analytic function
to a contour integral, and it has the form [21]
I 0 Nz
1 k f (z)
X
σk = z dz = ζik , (174)
j2π C f(z) i=1
where f(z) is an analytic function inside and on the simple closed contour C (Fig. 33a) and
on which no zeros are located. In addition, ζik are the i = 1, · · · , Nz zeros of f(z) inside C
raised to the k-th power. The quantity σk is their summation and f 0 (z) denotes the derivative
of f(z) with respect to z. For the case that f 0 (z) is not analytically given, it can be calculated
numerically via the Cauchy’s theorem [21], according to which the first derivative of a function
f(z) at the point z = z0, f 0 (z0) is given by
I
0 1 f(z)
f (z0) = dz , (175)
j2π D (z − z0)2
72
where D is any simple path which encloses the point z0 and f(z) is analytic inside and on D.
The path integral of Eq. (174) evaluated for k = 0 gives the number of zeros of f(z) enclosed
by contour C. Also for k = 1, · · · , Nz , the same integral produces the summations σ1, · · · , σn
which can be used to evaluate the coefficients of a polynomial S(z) of degree n, which has the
same roots, ζ1 , · · · , ζNz , as the function f(z) inside C. The approximation polynomial, S(z),
can be written as
Nz
X
S(z) = c` z ` , (176)
`=0
with cNz = 1. The coefficients ck are given via the Newton recursive formulas [55, 56]
(Nz − `) c` + σ1c`+1 + σ2c`+2 + · · · + σNz −` cNz = 0 for ` = Nz − 1, · · · , 0 . (177)
The polynomial, S(z), can be solved by well known techniques [55], which in general are more
efficient and accurate for low–degree polynomials. Therefore the maximum number of zeros
enclosed by C must be low, i.e., Nz ≤ 4, as proposed in [10, 21, 57]. Following the above
procedure, the initial difficult problem of finding the zeros of an arbitrary function, f(z), [or
the poles of f −1 (z)], is transformed to the problem of finding the zeros of the polynomial, S(z),
for which a variety of reliable numerical methods exist. By suitably dividing the region where
the zeros, ζi , are located and using Eq. (174) with k = 0, the degree of S(z) can be set such
that the polynomial root-solver can be more effective.
In practice, the zeros of the polynomial S(z), ζi0 , do not coincide with the zeros, ζi , of the
initial function f(z), due to errors introduced by the numerical evaluation of integrals of Eqs.
(174) and (175). For this reason, a further refinement must be done on the roots of S(z),
applying a zero-finding numerical method [9, 55] with initial conditions ζi0 .
As an example case the function f(z) = e3z + 2z cos z − 1, from Ref. [37], is considered. The
derivative of the function is f 0 (z) = 3e3z + 2 cos z − 2z sin z. The contour ABCD is specified in
Ref. [37] by the A, B, C, and D (see Fig. 33b caption). In order to determine the zeros of f(z)
inside the domain enclosed by ABCD the APM method is applied. Using Eq. (174) for k = 0
the number of zeros is determined to be σ0 = Nz = 6. Then the terms σk (k = 1, 2, · · · Nz )
are calculated from Eq. (174) and are shown in Table 6. The zero estimates and the refined
by Müller’s method zeros are shown in Table 7. It is observed that the zero estimates and the
refined zeros are practically identical which means that the integrations were very accurately
performed. Furthemore, it can be observed that the σk coefficients increase in magnitude as k
73
increases. Therefore, as it was mentioned previously, it is safer to apply the method in regions
that contain Nz ≤ 4 roots. In the case that Nz > 4 the initial contour can be subdivided into
smaller contours that each one contains smaller than 4 roots.
Table 6: APM σk ’s of function f (z) = e3z + 2z cos z − 1 from Ref. [37].
k σk Coefficients
0 6.0000000000000 − j0.000000000000000
1 2.0467702626400 − j0.000000000000000
2 − 14.1574340854587 − j0.000000000000001
3 − 84.7999601511681 − j0.000000000000007
4 − 30.4653454987963 + j0.000000000000006
5 696.0895474735294 − j0.000000000000028
6 2527.5240087190520 + j0.000000000000123
Table 7: APM zeros estimates and refined zeros of function f (z) = e3z + 2z cos z − 1 from Ref. [37].
APM Zeros Estimates Zeros after Müller’s Refinement
−1.844233953262218 − j0.000000000000002 −1.844233953262213 + j0.000000000000000
0.000000000000019 + j0.000000000000007 0.000000000000000 − j0.000000000000000
0.530894930292922 − j1.331791876751122 0.530894930292931 − j1.331791876751121
0.530894930292921 + j1.331791876751117 0.530894930292931 + j1.331791876751121
1.414607177658184 − j3.047722062627170 1.414607177658184 − j3.047722062627173
1.414607177658189 + j3.047722062627174 1.414607177658184 + j3.047722062627173
14.2 Basic Theory of the Abd-ellal, Delves, and Reid (ADR) Method
The algorithm summarized here extracts the zeros of a complex function, f(z), without using
knowledge of the function’s derivative. The algorithm used in [10], is rather complicated since
it uses the multivalued log[f(z)] function. Another algorithm which was first presented by
Abd-ellal, Delves, and Reid, (ADR method) [34], can be applied to meromorphic functions,
i.e., complex functions that are analytic in the region enclosed by a closed contour C, except at
a finite number of points (poles) inside C. The authors in [34] describe four different algorithms
which extract both poles and zeros of a meromorphic function in a complex region, and apply
their algorithms inside the unit circle on well behaved (polynomial) functions.
74
Here, the Abd-ellal et al. [34] results are adopted in order to be used for the eigenvalue
dispersion equations of optical, microwave, and semiconductor structures. In the case that the
function f(z) has Nz zeros inside the complex contour C, a polynomial can be constructed
having the same zeros as f(z). The polynomial has the form
PNz (z) = c0 + c1 z + c2 z 2 + · · · + cNz −1 z Nz −1 + z Nz , (178)
and its coefficients, ci , (i = 0, · · · , Nz − 1), are obtained after solving the system of linear
equations [34].
N
X z −1
cj Gr+j + Gr+Nz = 0, for r = 0, 1, 2, · · · , (Nz − 1). (179)
j=0
The Gk quantities are given as
I
1 zk
Gk = dz , for k = 0, 1, 2, · · · , (2Nz − 1). (180)
j2π C f(z)
The number of zeros Nz inside the domain enclosed by contour C (see Fig. 33a) is critical
in the above process and can be determined according to the argument principle theorem [21]
(including possible multiplicities) [35] by
I 0
1 f (z) 1 1
Nz = dz = ∆C {ln [f(z)]} = ∆C {arg [f(z)]} , (181)
j2π f(z) j2π 2π
where f 0 (z) is the first derivative of f(z) and the contour C is traveled counterclockwise.
Therefore, knowing Nz from Eq. (185), the coefficients Gk (k = 0, 1, 2, · · · , 2Nz − 1) can be
calculated from Eq. (180), and then the polynomial coefficients cj (j = 0, 1, 2, · · · , Nz −1 can be
evaluated from the solution of Eq. (179). Finally, the roots of the polynomial of Eq. (178) can
be used as estimates in the Müller’s iterative procedure for the final refinement of the function’s
zeros.
An alternative method for finding the zeros estimates has been proposed by Kravanja et
al. [36] and Gillan et al. [37]. Their approach calculates the Gk coefficients as they are defined
75
in Eq. (180). Then the following matrices can be defined:
G1 G2 G3 . . . GNz
..
G G3 G4 . . . .
H < = .2
.. .. . . .. (182)
.. . . . .
GNz . . . . . . . . . G2Nz
G0 G1 G2 . . . GNz −1
..
G1 G2 G3 . . . .
H = .. .. .. . . .. (183)
. . . . .
GNz −1 . . . . . . . . . G2Nz −2
The zeros of the f(z) function can be calculated by the following generalized eigenvalue problem,
H < − ζ 0H V = 0, (184)
where ζ 0 correspond to the ζ10 , · · · , ζN0 z are the estimates of the zeros of the function, and V are
the corresponding wavefunctions of the generalized eigenvalue problem (that are not utilized
and therefore are computed).
As an example case, again the function f(z) = e3z + 2z cos z − 1 from Ref. [37], is considered.
The contour ABCD is specified in Ref. [37] by the A, B, C, and D (see Fig. 33b caption).
In order to determine the number of zeros of f(z) inside the domain enclosed by ABCD the
ADR method is applied. Using Eq. (185) the number of zeros is determined to be Nz = 6.
Then the coefficients Gk (k = 0, 1, · · · , 2 × 6 − 1) are given in Table 8. The corresponding zero
estimates, ζi0 as solutions of the generalized eigenvalue problem [Eq. (184)] along with the final
refined by Müller’s method are shown in Table 9. It is observed that the estimates are identical
to the refined roots and they agree perfectly with the ones specified in Ref. [37]. Also it can
be observed that the Gk coefficients increase in magnitude as k increases. Therefore, it is safer
to apply the method in regions that contain Nz ≤ 4 roots. In the case that Nz > 4 the initial
contour can be subdivided into smaller contours that each one contains smaller than 4 roots as
it was proposed for APM in Sec. 14.1.
14.3 Basic Theory of the Derivative-Free Zero-Extraction by Phase-
based Enclosure Method (DFZEPE)
The idea of the zero extraction by phase-based enclosure is based on the Ying and Katz [35]
winding number evaluation of an analytic function. Assume a function f(z) (where z = x + jy)
76
Table 8: ADR Gk ’s of function f (z) = e3z + 2z cos z − 1 from Ref. [37].
k Gk Coefficients
0 − 0.07519739 − j0.00000000
1 0.27503184 + j0.00000000
2 − 0.90593139 − j0.00000000
3 2.07560724 + j0.00000000
4 − 2.01447904 + j0.00000000
5 2.44932797 + j0.00000000
6 − 22.63060397 + j0.00000000
7 14.84101715 + j0.00000000
8 99.50440329 + j0.00000000
9 456.78080464 − j1.832 × 10−13
10 − 483.81132715 − j9.047 × 10−14
11 −5297.84498932 − j2.099 × 10−12
Table 9: ADR zeros estimates and refined zeros of function f (z) = e3z + 2z cos z − 1 from Ref. [37].
ADR Zeros Estimates Zeros after Müller’s Refinement
−1.844233953262213 − j0.000000000000000 −1.844233953262213 + j0.000000000000000
−0.000000000000017 − j0.000000000000032 0.000000000000000 − j0.000000000000000
0.530894930292934 − j1.331791876751112 0.530894930292930 − j1.331791876751121
0.530894930292934 + j1.331791876751058 0.530894930292931 + j1.331791876751121
1.414607177658188 − j3.047722062627168 1.414607177658184 − j3.047722062627173
1.414607177658189 + j3.047722062627171 1.414607177658184 + j3.047722062627173
77
which is analytic in a bounded domain enclosed by the rectangular contour C (defined by
ABCD) shown in Fig. 33a. According to the argument principle theorem [21] the number of
zeros (including possible multiplicities), Nz , of function f(z) within the domain in the interior
of contour C is given as [35]
I 0
1 f (z) 1 1
Nz = dz = ∆C {ln [f(z)]} = ∆C {arg [f(z)]} , (185)
j2π f(z) j2π 2π
where f 0 (z) is the first derivative of f(z) and the contour C is traveled counterclockwise. The
range of the argument function “arg” is in the interval (−π, +π]. For a straight section defined
by [zi, zi+1 ] (along the contour C) that satisfies |∆[zi ,zi+1 ] arg[f(z)]| ≤ π the phase accumulation
z
is Φt |zi+1
i = arg[f(zi+1 )/f(zi )]. Therefore, the total length of the contour C can be decomposed
into linear segments satisfying the previous condition, by 4M points (M points per side of the
rectangle) such as C = ∪4M
i=1 [zi, zi+1 ] (i = 1, 2, · · · , 4M with z1 = z4M +1 ). Then, Nz can be
expressed in the form [35]
4M
1 X f(zi+1 ) 1
Nz = arg = Φt , (186)
2π i=1 f(zi ) 2π
where Φt is the total accumulated phase along the closed contour ABCD.
The last equation, Eq. (186), is the heart of the Derivative-Free Zero-Extraction by Phase-
based Enclosure (DFZEPE) method. Assume that the size of the sub-rectangle that approxi-
mates each zero of the function in the complex plane has pre-selected dimensions wx × wy as
shown in Fig. 33a where wx Lx = x2 − x1 and wy Ly = y2 − y1. The wx and wy specify the
zeros-enclosing sub-rectangles maximum dimensions and are directly related to the accuracy
of the estimation of the function’s f(z) zeros. For well separated zeros wx and wy could be of
the order of 10−1 Lx and 10−1 Ly , respectively. For closely located zeros wx and wy could be
of the order of 10−6 Lx and 10−6 Ly or even smaller. However, it is not necessary to isolate all
roots since the Müller’s algorithm which employs deflation can detect them inside the wx × wy
enclosing rectangle in most cases. In general though for close-spaced roots, isolation of the
roots is more reliable. Of course in the isolated roots the deflation procedure is unnecessary.
The DFZEPE algorithm is summarized in the following steps:
• Define the number of points M per segment of each rectangle or sub-rectangle for the use
of Eq. (186). This number could vary from 50-200 for functions f(z) with well-separated
78
4 wx = wy = 1
"
w x = w y = 0.5
3 w x = w y = 0.25
! - - 0
% ) )
2
'
&
y-coordinate
.
0
)
1
1
$ %
$ &
& )
0
#
/
.
% )
% )
-1
1
* +
.
/
-2
& )
& )
-3
,
( (
' % (
0
& ) )
% )
* ,
-2 0 2 4
x-coordinate
(a) (b)
Figure 33: (a) A rectangular bounded domain in the complex plane (z = x + jy). The rectangular
contour ABCD encloses the domain where the function f (z) is analytic. The “×” correspond to
zeros (roots) of f (z) in this domain. The DFZEPE procedure encloses each root of the function by
successively shrinking sub-rectangles. The sub-rectangles of dimensions ≤ wx, wy correspond to the
final sub-rectangles that enclose each of the zeros. (b) An example case for f (z) = e3z + 2z cos z − 1
(from Ref. [37]) where the final sub-rectangles are shown for various selections of wx and wy for
visualization of the DFZEPE process. The “×” correspond to the positions of the zeros as specified in
Ref. [37]. The realistic values of wx and wy that where used for this example were wx = wy = 0.0025
but in the scale of the figure these small sub-rectangles will show like dots. The A, B, C, and D points
correspond to x1 = −2.2, x2 = 2.5, y1 = −3.5, and y2 = 4.5 as given in Ref. [37].
zeros up to 500-1000 (or even more if root isolation is sought) for functions with very
closely located zeros and/or fast phase variation. This can be done in an adaptive manner.
For example, one could start with M = M0 and then repeat the process for M = 2M0
and M = 3M0 . If the results are identical (number and location of roots) then M0 is
sufficient for all phase accumulation computations. However, it’ should be pointed out
that the selection of M is strongly dependent on the selection of wx and wy . Actually,
the latter are more critical for the DFZEPE method since large wx and wy results in
crude two-dimensional bracketing of the roots and this could cause failure of the Müller’s
refinement algorithm (employing deflation which is needed if the roots are not isolated).
• Subdivide the initial rectangle ABCD into four equal sub-rectangles using the horizontal
divider segment H1m G1m = Lx and the vertical divider segment E1m F1m = Ly . Now each
79
of the four equal sub-rectangles has dimensions 0.5Lx × 0.5Ly .
• Using Eq. (186) check the number of zeros within each of the newly-formed four sub-
rectangles. Check the value of the function along the sub-rectangles boundaries. In case
that |f(z)| ≤ εF (where εF is related to the computer precision and to the function behav-
ior within the complex domain of interest) move divider H1m G1m slightly to the bottom
of the midpoint in y-direction and/or divider E1m F1m slightly to the left to the midpoint
in the x direction (for each sub-rectangle). In this manner, mitigation of the boundary
proximity or even crossing to a zero is achieved. In the case of dividers movement from
the midpoints, recalculate the number of zeros of each sub-rectangle.
• In case that the number of zeros of a sub-rectangle is zero disregard it.
• If the number of zeros of a sub-rectangle is greater or equal to one continue subdividing
it while width > wx or height > wy where width ' 0.5` Lx and height ' 0.5` Ly denote
the dimensions of the sub-rectangle after ` subdivisions. For example in Fig. 33a two
subdivisions are shown to illustrate the process (see the dividers E2m F2m , H2m G2m , and
E3m F3m which represent the second-level subdivisions (` = 2). Observe that the second
subdivision process occurred only in the sub-rectangles that include zeros inside. Store the
four corners of each sub-rectangle and its Nz if Nz ≥ 1 at the end of this loop and repeat
the process for all stored sub-rectangles while the conditions width > wx or height > wy
hold.
• Use the centers of the converged sub-rectangles that enclose the zeros as the estimates of
the zeros of function f(z).
• Use these zero estimates in the Müller iterative algorithm with deflation [8] to obtain re-
finement of the function’s f(z) zeros. Store the final refined zeros. Since the Müller’s
method requires three initial starting points, if ze is the zero estimate provided by
DFZEPE’s enclosure process, the three initial points are ze and ze ± min{wx , wy }/2.
• End of the algorithm.
As an example case the function f(z) = e3z + 2z cos z − 1 from Ref. [37] is considered.
The contour ABCD is specified in Ref. [37] by the A, B, C, and D (see Fig. 33b caption).
80
1
0.5
(z)/
0
Phase,
-0.5
-1
0 500 1000 1500 2000 2500 3000 3500
8
AB BC CD DA
(z)/2
6
t
Total Phase,
0
0 500 1000 1500 2000 2500 3000 3500
Sample Points of ABCD, (z)
Figure 34: (a) The phase variation around ABCD, Φ(z) = arg[f (z)], for P the function f (z) =
e3z +2z cos z −1 (from Ref. [37]) and (b) the total phase accumulation Φt(z) = Ii=1 arg[f (zi+1 )/f (zi )]
where I is an index such zI = z and Φt(z1 ) = 0. The AB, BC, CD, and DA show the phase
accumulation within each segment of the contour C. Obviously, the total accumulated phase is 6 × 2π
that reveals the existence of 6 zeros (Nz = 6) within the domain enclosed by C.
In order to determine the number of zeros of f(z) inside the domain enclosed by ABCD
the DFZEPE method is applied. The Φ(z) = arg[f(z)] and the accumulated phase Φt (z) =
PI
i=1 arg[f(zi+1 )/f(zi )] where I is an index such zI = z are shown in Fig. 34 for demonstration
purposes. It is obvious that the function f(z) has six zeros in the specified domain. It is also
interesting to mention that the largest phase accumulation occurs on the BC segment of the
contour. For illustration purposes the convergence of the subdivision process is shown Fig.
33b for three different selections of wx = wy = 1, 0.5, and 0.25. It can be observed how the
converged sub-rectangles enclose tigher the zeros of the function as wx and wy become smaller.
In the actual numerical test wx = wy = 0.0025, M = 250, and εF = 10−11 were selected and
the zero estimates along with the Müller’s refinement are shown in Table 10. The agreement
with the results provided by Gillan et el. [37] is perfect for the precision shown in the table.
It is worth mentioning that the number of points per contour segment, M, could be selected
81
Table 10: DFZEPE zeros estimates and refined zeros of function f (z) = e3z + 2z cos z − 1 from
Ref. [37].
DFZEPE Zeros Estimates Zeros after Müller’s Refinement
−1.844165039062500 + j0.000976562500000 −1.844233953262213 + j0.000000000000000
0.000317382812500 + j0.000260416666667 0.000000000000000 − j0.000000000000000
0.531323242187500 − j1.331054687500000 0.530894930292931 − j1.331791876751121
0.531323242187500 + j1.331054687500000 0.530894930292931 + j1.331791876751121
1.415112304687500 − j3.047851562500000 1.414607177658184 − j3.047722062627173
1.415112304687500 + j3.047851562500000 1.414607177658184 + j3.047722062627173
smaller and the wx and wy can be selected larger if it is not required that the mode indices to
be separated by different enclosing sub-rectangles. In such a case the final sub-rectangles could
enclose more than a single mode but the Müller’s algorithm with deflation can successfully
distinguish all the zeros since their number is known. This was tested for modes separated by
∆Nef f,r ≤ 10−6 with wx = wy = 10−3 with success in the case of two coupled waveguides that
was reported in Ref. [38]. However, it is always safer to isolate by different enclosing rectangles
each mode (since for the case of multilayer waveguides there are no multiple zeros). Cases
where the final sub-rectangles could enclose more than a single zero could exist in problems
where there are zeros with higher than one multiplicities.
14.4 Application of APM, ADR, and DFZEPE methods to Planar
Multilayer Waveguides
A general planar multilayer waveguide is shown in Fig. 35a. The waveguide is uniform in the y
direction, while the light confinement is in the x direction, and the light propagation is in the z
direction. It is comprised of the cover region (x < 0) of refractive index nc , the substrate region
P
(x > N i=1 hi ) of refractive index ns , and N film layers of height hi , and refractive index ni
(where i = 1, 2, · · · , N). The refractive indices could be of the form n = nR + jnI where nR > 0
and nI = 0 for lossless layers, nR > 0 and nI < 0 for lossy layers, and nR > 0 and nI > 0
or active layers. Of course this selection of signs is consistent with exp[+j(ωt − kz)] (ω is the
radial frequency and k the corresponding wavenumber) field dependences for modes traveling
in the positive z direction. The freespace wavelength is λ0 and the freespace wavenumber is
k0 = 2π/λ0 . The transfer matrix technique [11–13] is used in order to find the dispersion
82
equation for the guided modes of this planar waveguide. The resulting dispersion equation is
[Eqs. (105) and (106)]
γs γc γc γs
F (β) = F (k0Nef f ) = m22 + m11 − m12 − m21 = 0, (187)
as ac ac as
where ac = as = 1 for TE polarization, and ac = n2c , as = n2s for TM polarization and mij
(i, j = 1, 2), γs , γc are defined in Sec. 11. The electric and magnetic field phasors are of the form
A(x) exp(−jβz). The propagation constant β can be in general complex (for lossy or active
waveguides) and can be written in the form β = k0 Nef f = k0 (Nef f,r + jNef f,i ) where Nef f,r is
the real part and Nef f,i is the imaginary part of the effective index of a mode, respectively. For
lossy waveguides and for propagation in the positive z direction Nef f,r > 0 and Nef f,i < 0 while
for active waveguides Nef f,r > 0 and Nef f,i > 0. Of course, for lossless waveguides, Nef f,i = 0
for all guided modes. At the branch-cut points Nef f = nc and Nef f = ns it is important to
remain in the same Riemann surface when Nef f (and correspondingly β) are varying in the
complex plane. This can be accomplished by selecting the sign of γc and γs in such a way in
order to satisfy the conditions Re{γc } > 0 and Re{γs } > 0 (where Re{∗} denotes real part)
that guarantee confinement within the guiding layers.
P Q R R S T
7 8
? @ A B C
d e f g h i j f k g
I 3 4
4
H J B C K W ^
7
O O
a b a
b p
[ o
l l l
n
W X Y Z
_ W ]
P Q R R S r
I 3 5
H J B C L
7 :
n n
U V
n
U c
3 6
I
; <
[ \
H J B C M
N q N O O
` `
l l l l k m m h i j f k g
D E F D G C H G B 7 9
>
(a) (b)
Figure 35: (a) The geometry of a general planar multilayer waveguide. The z-axis is the propagation
axis and the x axis is the confinement axis. The i-th layer has a refractive index ni and a height of hi .
The number of layers is N . The substrate and the cover refractive indices are ns and nc , respectively.
(b) Zero search in the complex domain for guided (lossless, lossy, or active) and leaky waveguide
modes.
The case of leaky modes requires a little more attention. The leaky modes are a com-
pact representation of the radiation field of the waveguide [58, 59]. Therefore, even for lossless
83
waveguides the leaky modes have complex propagation constants (with Nef f,i < 0) that repre-
sent the radiation of the electromagnetic field into the substrate or into the cover or into both
regions. In order to remain in the same Riemann surface when Nef f (and β) are varying in
the complex plane, the sign of γc and γs must be selected in such a way in order to satisfy the
conditions [16, 20, 33]: (a) Re{γc } > 0 and =m{γs } > 0 (where =m{∗} denotes imaginary part)
for confinement in the cover region and radiation into the substrate region; (b) =m{γc } > 0
and =m{γs } > 0 for radiation into both the cover and the substrate regions. These conditions
are such if ns > nc . In the case of nc > ns the roles of γs and γc as well as of substrate and
cover regions in the previous conditions are interchanged.
The modes of the planar multilayer waveguide can be determined by applying either the
APM (Sec. 14.1), the ADR (Sec. 14.2), or the DFZEPE procedure (Sec. 14.3) to the function
of Eq. (187). The contours Ag Bg Cg Dg (for guided modes) and A` B` C` D` for leaky modes
are shown in Fig. 35b. The right part of A` B` C` D` domain (E` B` C` F` ) contains leaky modes
radiating into the substrate region (assuming ns > nc ) but confined in the cover region, while
the left part (A` E` F` D` ) contains leaky modes radiating into both the substrate and the cover
regions. The guided modes of lossy waveguides will lay in the bottom part (Nef f,i < 0) of the
complex plane while the guided modes of active waveguides could lay either in the top part
(Nef f,i > 0) (if they experience gain) or in the bottom part (if they experience loss). Similarly,
the leaky modes will lay on the bottom half (Nef f,i < 0) but for the range 0 < Nef f,r <
max{nc , ns } = ns (since it is assumed that ns > nc ). All these are true for forward mode
propagation (z > 0). The converged sub-rectangles (with exaggerated size) that enclose the
location of the mode effective indices are also shown in Fig. 35b which will be the result of
the DFZEPE subdivision process. The APM and ADR methods do not require any systematic
zero enclosure process since their main result is the coefficients of an approximating polynomial
or the eigenvalues of a generalized eigenvaule problem that involves orthogonal approximating
polynomials, respectively. Then the estimated by the centers of the final sub-rectangles mode
effective indices are inserted into the Müller’s refinement process (with deflation) for the final
determination of the mode complex effective indices. Similarly, the APM and ADR methods use
their approximating zeros as estimates for the Müller’s refinement process. It is true that the
approximating zeros of the APM and ADR methods are usually closer than the approximations
of the DFZEPE method but after the refinement process the final roots (modal propagation
84
constants) of all methods are identical. After the effective index of each mode is determined it
is straightforward to calculate all modal field components using the transfer matrix technique
presented in Sec. 11.
Furthermore, the APM method requires the derivative of the function. The derivative could
be calculated numerically using Eq. (175) or can be determined analytically in the case of
multilayer planar waveguides [14, 17]. Defining as before β = k0 Nef f = k0 (Nef f,r + jNef f,i) the
function f(β) = F (Nef f ) in Eq. (187). Then the derivative with repsect to Nef f of the function
is given by
dF 1 dγs γs dm22 1 dγc γc dm11
= m22 + + m11 +
dNef f as dNef f as dNef f ac dNef f ac dNef f
γs dγc γc dγs γc γs dm12 dm21
− m12 − m12 − − . (188)
asac dNef f as ac dNef f ac as dNef f dNef f
In the above equation the required derivatives are given by
dγc Nef f
= k02 , (189)
dNef f γc
dγs Nef f
= k02 , (190)
dNef f γs
N N
dM̃ X dM̃i Y
= M̃j
(191)
dNef f dNef f
i=1
j=1
j6=i
hi sin(kxi hi ) aihi cos(kxi hi ) ai sin(kxi hi )
dM̃i 2
− 3
= k02 Nef f kxi kxi kxi
, (192)
dNef f sin(kxi hi ) hi hi sin(kxi hi )
− − cos(kxi hi )
ai kxi ai kxi
where ai = 1 for TE modes and ai = n2i for TM modes. In the special case that a kxi = 0 the
corresponding M̃i matrix is treated as constant in the previous equations and its derivative is
not needed.
14.5 Results
Only waveguide structures that have been published in the literature will be tested to demon-
strate the effectiveness of the APM, ADR, and DFZEPE methods. All these methods produced
identical results. However, in this section the results were obtained using the DFZEPE method
which is the simplest and does not require either the derivative of the function or any complex
integrations. Specifically, the following cases will tested: (a) A case of a lossless multilayer
85
waveguide [11, 16, 17, 38] where both guided and leaky modes will be determined. (b) A case of
a slightly-lossy multilayer waveguide [11, 16, 17, 38] where the guided modes and leaky modes
will be determined. (c) A case of an active waveguide [2, 38] from doped semiconductor ma-
terials where guided modes and sample leaky modes will be determined, and (d) A case of an
ARROW waveguide [17, 38] where only leaky modes exist. The results of the DFZEPE method
for the effective modal indices in all cases are identical to the published results within the nu-
merical accuracy that is provided in the literature. Therefore, a deviation from the published
results in terms of a percentage difference is meaningless. All the results were calculated on an
Intel(R) Core(TM) i7-7700, 3.60 GHz processor. The computation times varied from 4-20 sec
for most cases. The most time-demanding calculations were for the ARROW waveguide were
the computation times were between 160-400 sec (results of Table 14).
14.5.1. Lossless Dielectric Waveguide
This multilayer lossless waveguide was presented initially in Ref. [11]. It is comprised of 4 film
layers (N = 4) and its parameters are given in the caption of Table 11. To apply the DFZEPE
method for the guided modes the following parameters were selected: x1 = 1.501 >∼ ns , x2 =
1.659 <
∼ nmax = maxi {ni }, y1 = −0.25, y2 = 0.20, wx = wy = 0.005, M = 250, and εF = 10
−11
(this parameter checks the proximity of zero to contour segments and it is discussed in Sec.
14.3). The effective indices of the guided TE and TM modes are shown in Table 11. Observe
that the imaginary part of the effective index of each guided mode is zero as it is expected since
the waveguide is lossless.
In the same table are also shown the effective indices of TE and TM leaky modes that are
radiating into the substrate but are confined in the cover. For these leaky modes the parameters
of the DFZEPE method are x1 = 1.001 >
∼ nc , x2 = 1.499 <
∼ ns , y1 = −0.25, y2 = 0.20, wx =
wy = 0.005, M = 250, and εF = 10−11 . There are five TE and five TM leaky modes that
radiate into the substrate only. If x1 = 0.001 >
∼ 0 and y1 = −5.0 the DFZEPE method detects
additional leaky modes that radiate into both the substrate and the cover regions. For latter
case the DFZEPE method identified 25 additional TE and 26 additional TM leaky modes
that radiate in both cover and substrate with effective indices 0 < Nef f,r < nc = 1.0 and
−5 < Nef f,i < 0. Since these leaky modes have much higher losses (compare to the ones that
radiate only into the substrate) are not of much interest. However, for completeness all TE and
86
Table 11: Effective indices of guided and leaky modes of a multilayer lossless waveguide from Ref. [11]
λ0 = 0.6328 µm, nc = 1.0, ns = 1.45, n1 = 1.66, n2 = 1.53, n3 = 1.60, n4 = 1.66, h1 = h2 = h3 = h4 =
0.5 µm.
Mode Nef f = Nef f,r + jNef f,i Mode Nef f = Nef f,r + jNef f,i
Guided Modes
TE0 1.62272868 + j0 TM0 1.62003132 + j0
TE1 1.60527569 + j0 TM1 1.59478848 + j0
TE2 1.55713615 + j0 TM2 1.55498069 + j0
TE3 1.50358711 + j0 TM3 1.50181780 + j0
Leaky Modes - Substrate Radiating
TE4 1.46185664 − j0.00715587 TM4 1.45153498 − j0.01192359
TE5 1.38248922 − j0.01816588 TM5 1.37066437 − j0.03014206
TE6 1.28136443 − j0.03587739 TM6 1.27373706 − j0.05679177
TE7 1.14231446 − j0.05287607 TM7 1.15731285 − j0.08757849
TE8 1.00303702 − j0.07077094 TM8 1.03695026 − j0.10307808
Leaky Modes - Substrate/Cover Radiating
TE9 0.80402477 − j0.15549191 TM9 0.96341519 − j0.165250319488834
TE10 0.49261437 − j0.33590355 TM10 0.76239325 − j0.222733601859893
TE11 0.29877905 − j0.69942867 TM11 0.46058337 − j0.370232916081077
TE12 0.25212085 − j1.00504264 TM12 0.24771086 − j0.718279099245514
TE13 0.25050946 − j4.95922057 TM13 0.18839165 − j1.013610354108087
TE14 0.24991236 − j4.79387973 TM14 0.14364341 − j1.274982619497127
TE15 0.24586267 − j4.62848156 TM15 0.12685382 − j1.524936733594305
TE16 0.24161573 − j4.45910352 TM16 0.11859313 − j1.949587977521766
TE17 0.24026651 − j4.28845411 TM17 0.11837505 − j1.753757137213375
TE18 0.23986335 − j4.12047459 TM18 0.10254993 − j2.145238444255885
TE19 0.23543581 − j3.95230034 TM19 0.10232727 − j2.720059502187552
TE20 0.23077652 − j3.77917269 TM20 0.09894013 − j2.349610858911572
TE21 0.22948257 − j3.43177029 TM21 0.09740231 − j2.546721039915246
TE22 0.22944624 − j3.60410128 TM22 0.09539809 − j3.435254140965898
TE23 0.22470439 − j3.25890969 TM23 0.09169923 − j4.123746135758958
TE24 0.22207063 − j1.26968361 TM24 0.09054857 − j2.896630870912667
TE25 0.21976462 − j2.71626746 TM25 0.08946336 − j4.796978313580165
TE26 0.21968734 − j3.07943141 TM26 0.08924637 − j3.085873731780236
TE27 0.21863277 − j2.89664559 TM27 0.08914771 − j3.271127601096534
TE28 0.21504559 − j1.94511532 TM28 0.08509431 − j3.603072946747228
TE29 0.21495125 − j2.53439786 TM29 0.08488961 − j3.964394978515887
TE30 0.21238769 − j1.74120004 TM30 0.08452455 − j3.785071780722443
TE31 0.21178594 − j1.51632839 TM31 0.08236169 − j4.640441344235671
TE32 0.21039529 − j2.14368451 TM32 0.08209538 − j4.286677523338038
TE33 0.21009037 − j2.34244177 TM33 0.08181951 − j4.464558548926473
TM34 0.08025401 − j4.956855738939457
87
Chilwell waveguide leaky mode distribution, N z,TE = 30, N z,TM = 31
1
-1
Neff,i -2
-3
-4
TE-modes
TM-modes
-5
-6
0 0.5 1 1.5
Neff,r
Figure 36: The leaky mode distribution for TE and TM modes for the Chilwell [11] waveguide
of Table 11. There are 29 TE leaky modes and 30 TM modes. The parameters for the DFZEPE
method where x1 = 0.001, x2 = 1.499, y1 = −0.25 and y2 = 0.20.
TM leaky modes (radiating into both cover and substrate) are shown in Table 11. A subset
of these leaky modes (with Nef f,i > −3.70) has been published in Ref. [25]. If y1 < −5 many
more leaky modes are detected but since they all have high radiation losses may not be a good
representation of the radiation field [58, 59] and therefore are not of much interest. A graphical
representation of all the leaky modes of Table 11 are shown in Fig. 36. Furthermore, the
electric field profile of the first leaky (substrate-radiating) TE (TE4) mode, and the magnetic
field profile of the first leaky (substrate-radiating) TM (TM4) mode, are shown in Fig. 37. The
radiation of these modes into the substrate is apparent. The increasing amplitude of the fields
in the substrate region denotes the radiation of the modes and is a characteristic of the leaky
mode approximation (implying that power from previous z positions of the mode is arriving at
higher points into the substrate).
14.5.2. Low-Loss Dielectric Waveguide
This is the same multilayer waveguide that was presented in the previous section (Sec. 14.5.1)
with the addition of some small loss in two of its film layers [11]. The parameters of the structure
are given in the caption of Table 12. In order to apply the DFZEPE method for the guided
88
TE 4 -mode (N 4 = 1.46185664-0.00715587065i)
0.8
y
Normalized Electric Field, E
0.6
0.4
0.2
-0.2
-0.4
-0.6
-0.8
-6 -4 -2 0 2 4 6 8
Distance x ( m)
TM 4 -mode (N 4 = 1.45153498-0.0119235986i)
0.8
y
Normalized Magnetic Field, H
0.6
0.4
0.2
-0.2
-0.4
-0.6
-0.8
-6 -4 -2 0 2 4 6 8
Distance x ( m)
Figure 37: The electric field profile of the first leaky (substrate-radiating) TE (TE4 ) mode, and the
magnetic field profile of the first leaky (substrate-radiating) TM (TM4 ) mode, for the Chilwell [11]
waveguide of Table 11. Observe the radiation into the substrate region as well as the increasing
amplitude of the radiation field.
89
modes the same parameters with the ones used in Sec. 14.5.1 were used. The effective indices
of the guided TE and TM modes are shown also in Table 12. Observe that now all modes
have negative imaginary part of their effective index due to the lossy layers of the waveguide.
Table 12: Effective indices of guided and leaky modes of a multilayer small-loss waveguide from
Ref. [11] λ0 = 0.6328 µm, nc = 1.0, ns = 1.45, n1 = 1.66 − j1.66 × 10−4 , n2 = 1.53 − j1.53 × 10−4 ,
n3 = 1.60, n4 = 1.66, h1 = h2 = h3 = h4 = 0.5 µm.
Mode Nef f = Nef f,r + jNef f,i × 10+4 Mode Nef f = Nef f,r + jNef f,i × 10+4
Guided Modes
TE0 1.62272868 − j0.00673727 TM0 1.62003131 − j0.00892759
TE1 1.60527569 − j1.66244285 TM1 1.59478847 − j1.65565266
TE2 1.55713612 − j0.20880097 TM2 1.55498066 − j0.23704828
TE3 1.50358696 − j0.55032495 TM3 1.50181764 − j0.42530043
Leaky Modes - Substrate Radiating
TE4 1.46185448 − j0.00726710 TM4 1.45153751 − j0.01202887
TE5 1.38249997 − j0.01827662 TM5 1.37068384 − j0.03024261
TE6 1.28137151 − j0.03596266 TM6 1.27375077 − j0.05687731
TE7 1.14233026 − j0.05299360 TM7 1.15732794 − j0.08766890
TE8 1.00303470 − j0.07087449 TM8 1.03694118 − j0.10316486
Even though for this lossy structure leaky modes were not reported in the literature, they
are included here for completeness. Specifically, the leaky modes that radiate only into the
substrate are shown in Table 12. It can be observed that the real parts of the leaky mode
effective indices are very close to the ones appearing for the lossless waveguide in Table 11.
However, their imaginary parts of their effective indices are in magnitude slightly higher since
they represent both the loss due to absorption in the lossy layers as well as the loss due to
radiation into the substrate.
14.5.3. Active Waveguide
Semiconductor materials with loss and gain are utilized for optical amplifiers or lasers. In this
example case the structure proposed by Visser et al. [2] based on InP and InGaAsP system is
analyzed using the DFZEPE method. The same case is also reported in Ref. [38]. It is comprised
of 4 film layers with either loss or gain. The structure’s parameters are defined in the caption of
Table 13. To apply the DFZEPE method for the guided modes the following parameters were
90
selected: x1 = 3.17 >
∼ ns , x2 = 3.59 <
∼ maxi [Ree{ni }], y1 = −0.25, y2 = 0.20, wx = wy = 0.005,
M = 500, and εF = 10−11 . The results are summarized in Table 13 for both TE and TM guided
modes. Observe that TE0 and TM1 modes have positive imaginary part of their effective index
which imply gain due to the active layer. These modes are mainly concentrated in the active
layer (corresponding to InGaAsP, n3 = 3.60 + j0.002, and h3 = 0.15 µm). On the other hand,
T M0 mode has a negative imaginary part which imply loss (this is actually a plasmon mode
mainly concentrated at the interface between the thin metal layer (corresponds to a thin gold
layer with n1 = 0.18 − j10.2 and thickness of h1 = 0.04 µm) and the second film layer (which
corresponds to InP with n2 = 3.16 − j0.0001 and h2 = 1.0 µm). These modes normalized
electric and magnetic fields profiles are shown in Fig. 38.
Table 13: Effective indices of guided modes and first leaky mode of a multilayer active waveguide from
Ref. [2,38] λ0 = 1.30 µm, nc = 1.0, ns = 3.16, n1 = 0.18−j10.2, n2 = 3.16−j0.0001, n3 = 3.6+j0.002,
n4 = 3.16 − j0.0001, h1 = 0.04 µm, h2 = 1.0 µm, h3 = 0.15 µm, h4 = 3 µm.
Mode Nef f = Nef f,r + jNef f,i × 10+4 Mode Nef f = Nef f,r + jNef f,i × 10+4
Guided Modes
TE0 3.28088001 + j 9.13918191 TM0 3.33449848 − j 75.18872326
TM1 3.24809848 + j 5.46307013
First Leaky Mode
T E1 3.13650356 − j376.20259075 TM2 3.13622674 − j377.75840427
For completeness a search for leaky modes into the substrate was performed. Using x1 =
1.001 >
∼ nc , x2 = 3.159 <
∼ ns , while retaining the rest of the parameters of DFZEPE the same
as in the case of guided modes 12 TE leaky modes and 11 TM leaky modes were determined
using the specified values of y1 and y2. All these leaky modes radiate into the substrate and
they all have negative imaginary parts of their effective indices revealing radiation loss. For
demonstration purposes, only the first leaky mode for each polarization is shown in Table 13.
14.5.4. Leaky Modes of an ARROW Waveguide
Antiresonant reflecting optical waveguides (ARROW) utilize antiresonant (multi-interference)
reflection as the guiding mechanism instead of total internal reflection [5, 6]. The structure
under investigation was also used by Chen et al. [17] and Semwal and Rastogi [38]. It is
91
TE 0 -mode (N 0 = 3.28088001+0.000913918191i) TM 0 -mode (N 0 = 3.33449848-0.00751887233i)
1 1
y
y
0.8
Normalized Magnetic Field, H
Normalized Electric Field, E
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0 -0.2
-5 0 5 10 -5 0 5 10
Distance x ( m) Distance x ( m)
TM 1 -mode (N 1 = 3.24809848+0.000546307013i)
1
y
0.8
Normalized Magnetic Field, H
0.6
0.4
0.2
-0.2
-5 0 5 10
Distance x ( m)
Figure 38: The active and lossy modes TE0 , TM0 , and TM1 , for the Semwal [38] active waveguide
of Table 13. Observe that the active modes are confined in the active layer (TE0 and TM1 ), while the
lossy mode (plasmon mode) TM0 is confined in the metal-semiconductor interface. The thin metal
layer of n1 = 0.18 − j10.2 and h1 = 0.04 µm is represented by the thicker dashed line on the left part
of the figures.
92
comprised of N = 9 film layers with interchanging high and low refractive indices on top of a
high index substrate (silicon). The structure parameters are defined in the caption of Table 14.
Only leaky modes are supported by this structure. The DFZEPE method parameters were:
x1 = 1.4501, x2 = 1.499 < −6
∼ maxi [Re{ni }], y1 = −0.25, y2 = 0.20, wx = 2.5×10 , wy = 5×10
−4
M = 500 − 1000, and εF = 10−4 . The value of εF was much larger than previous cases since
the dispersion function reaches values of the order of 1044 in the domain of interest. Therefore,
εF represents a drop of the function of almost 48 orders. Furthermore, the large number of M
is necessary due to the large phase variation that is observed in the vertical segments of the
ABCD contour. For demonstration, the phase variation of this ARROW waveguide is shown
in Fig. 39.
Table 14: Effective indices of leaky modes of a multilayer ARROW waveguide from Ref. [38] λ0 =
0.6328 µm, nc = 1.0, ns = 3.50, n1 = 1.46, n2 = 1.50, n3 = 1.46, n4 = 1.50, n5 = 1.46, n6 = 1.50,
n7 = 1.46, n8 = 1.50, n9 = 1.46, h1 = 2.00 µm, h2 = 0.448 µm, h3 = 4.00 µm, h4 = 0.448 µm,
h5 = 2.00 µm, h6 = 0.448 µm, h7 = 4.00 µm, h8 = 0.448 µm h9 = 2.00 µm.
Mode Nef f = Nef f,r + jNef f,i × 10+4 Mode Nef f = Nef f,r + jNef f,i × 10+4
TE0 1.473925808 − j0.000000801 TM0 1.473275805 − j0.000005809
TE1 1.473697976 − j0.000017405 TM1 1.473027205 − j0.032900856
TE2 1.473696644 − j0.005452261 TM2 1.473026854 − j0.000035036
TE3 1.473459693 − j0.000001142 TM3 1.472767027 − j0.000008508
TE4 1.457920191 − j0.007106241 TM4 1.457925423 − j0.045880488
TE5 1.457791244 − j0.009053396 TM5 1.457782773 − j0.057163274
TE6 1.453780369 − j0.114698816 TM6 1.453795448 − j0.645756672
TE7 1.453045406 − j0.420121480 TM7 1.452928429 − j2.555862981
TE8 1.451864807 − j0.693651857 TM8 1.451781628 − j4.567101184
TE9 1.450269491 − j0.732515868 TM9 1.450247659 − j4.357488809
The results are summarized in Table 14 for both TE and TM guided modes. The leaky
modes of the ARROW waveguides are numerically harder to find due to the close proximity of
the zeros. In addition for the accurate computation of the accumulated phase Φt around each of
the sub-rectangles a much higher number of points per segment M was used. The accumulated
phase increases fast in the BC part of the rectangle contours while decreases fast in the DC
parts of the contours (see Fig. 39). This means that it is critical to count all of the phase
jumps in order to accurately determine the accumulated phase. This is the main reason for the
93
1
0.5
(z)/
0
Phase,
-0.5
-1
0 2 4 6 8 10
104
40
AB BC CD DA
(z)/2
30
t
Total Phase,
20
10
0
0 2 4 6 8 10
Sample Points of ABCD, (z) 104
Figure 39: (a) The phase variation around ABCD, Φ(z) = arg[f (z)], for the function of the ARROW
waveguide and x1 = 1.4501, x2 = 1.4999, y1 = 0.20, y2 = −0.25, M = 20000 and (b) the total phase
P
accumulation Φt (z) = Ii=1 arg[f (zi+1 )/f (zi)] where I is an index such zI = z and Φt (z1 ) = 0. The
AB, BC, CD, and DA show the phase accumulation within each segment of the contour C. Obviously,
the total accumulated phase is 10 × 2π that reveals the existence of 10 leaky modes (Nz = 10) within
the domain enclosed by C. Observe the very fast variation along the vertical segments of the contour.
94
TE 1 -mode (N 1 = 1.47369798-1.74048619e-09i) TE 2 -mode (N 2 = 1.47369664-5.45226138e-07i)
1 1
y
0.8 0.8
Normalized Electric Field, E
Normalized Electric Field, E
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-5 0 5 10 15 20 25 -5 0 5 10 15 20 25
Distance x ( m) Distance x ( m)
TM 1 -mode (N 1 = 1.4730272-3.29008566e-06i) TM 2 -mode (N 2 = 1.47302685-3.50360353e-09i)
1 1
y
y
0.8 0.8
Normalized Magnetic Field, H
Normalized Magnetic Field, H
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-5 0 5 10 15 20 25 -5 0 5 10 15 20 25
Distance x ( m) Distance x ( m)
Figure 40: The close in effective index leaky modes TE1 and TE2 , as well TM1 , and TM2 , for the
ARROW waveguide [38] of Table 14. Observe that the leaky modes that closer to the substrate radiate
more and have larger(in magnitude) Nef f,i.
10 -3 ARROW leaky mode distribution, N z,TE
= 53, N z,TM = 54
2
-2
Neff,i
-4
-6
TE-modes
TM-modes
-8
-10
1 1.1 1.2 1.3 1.4 1.5
Neff,r
Figure 41: The leaky (substrate-radiating) mode distribution for TE and TM modes for the
ARROW waveguide of Table 14. There are 53 TE leaky modes and 54 TM modes (the first ten of
TE and TM are shown in the table). There are no leaky modes in the 1.5 − 3.5 Nef f,r interval. The
parameters for the DFZEPE method where x1 = 1.001, x2 = 3.499, y1 = −0.25 and y2 = 0.20.
95
large number of points needed. From Table 14, for both polarizations, there are two clusters of
modes. One that is near 1.45 (6 leaky modes) and one around 1.47 (4 leaky modes). Because
the accumulated phase around a contour can accurately determine the number of modes the
process of finding closely spaced modes can be performed in parts of the domain by changing
appropriately the DFZEPE parameters. It is mentioned that only the 1.45 leaky mode cluster
is published in Ref. [17] and only a subset of them (3 out the 6 modes from 1.45 cluster and
1 out of 4 modes from 1.47 cluster) in Ref. [38]. However, the proposed DFZEPE method
can reliably determine all the leaky modes of the ARROW waveguide. For example, the total
number of substrate-radiating leaky modes in the interval of Nef f,r from 1 − 3.5 is 53 for TE
and 54 for TM polarization, respectively. It is interesting to note that there are no leaky modes
found in the interval Nef f,r ∈ (1.5 − 3.5). Only the first ten are given with their effective indices
in Table 14. However, the distribution of all the substrate-radiating leaky modes is shown in
Fig. 41. For illustration purposes the normalized electric and magnetic fields profiles of the
leaky modes TE1 and TE2, as well as TM1, and TM2, are shown in Fig. 40, respectively. It is
interesting to note that the modes that are closer to the substrate experience higher radiation
losses as it is naturally expected due to their proximity to the high index silicon substrate.
96
References
[1] A. W. Snyder and J. D. Love, Optical Waveguide Theory. London: Chapman & Hall, 1983.
[2] T. D. Visser, H. Blok, and D. Lenstra, “Modal analysis of a planar waveguide with gain
and losses,” IEEE J. Quantum Electron., vol. 31, pp. 1803–1810, Oct. 1995.
[3] M. Principe, G. Castaldi, M. Consales, A. Cusano, and V. Galdi, “Supersymmetry-inspired
non-Hermitian optical couplers,” Sci. Rep. UK, vol. 5, pp. 8568:1–7, Feb. 2015.
[4] E. G. Turitsyna, I. V. Shadrivov, and Y. S. Kivshar, “Guided modes in non-Hermitian
optical waveguides,” Phys. Rev. A, vol. 96, p. 033824, Sept. 2017.
[5] M. A. Duguay, Y. Kokubun, T. L. Koch, and L. Pfeiffer, “Antiresonant reflecting optical
waveguides in SiO2-Si multilayer structures,” Appl. Phys. Lett., vol. 49, pp. 13–15, July
1986.
[6] T. Baba and Y. Kokubun, “Dispersion and radiation loss characteristics of antiresonant re-
flecting optical waveguides - numerical results and analytical expressions,” IEEE J. Quan-
tum Electron., vol. 28, pp. 1689–1700, July 1992.
[7] J. F. Traub, Iterative Methods for the Solution of Equations. Englewood Cliffs, NJ:
Prentice-Hall, 2005.
[8] S. D. Conte and C. de Boor, Elementary Numerical Analysis: An Algorithmic Approach,
ch. The solution of nonlinear equations, pp. 72–127. International Series in Pure and
Applied Mathematics, 3rd Ed., New York: McGraw-Hill Book Company, 1980.
[9] D. E. Müller, “A method for solving algebraic equations using an automatic computer,”
Mathematical Tables and Other Aids to Computation, vol. 10, no. 56, pp. 208–215, 1956.
[10] L. M. Delves and J. N. Lyness, “A numerical method for locating the zeros of an analytic
function,” Math. Comput., vol. 21, pp. 543–560, 1967.
[11] J. Chilwell and I. Hodhkinson, “Thin-films field-transfer matrix theory of planar multilayer
waveguides and reflection from prism-loaded waveguide,” J. Opt. Soc. Amer. A, vol. 1,
pp. 742–753, July 1984.
97
[12] L. M. Walpita, “Solutions for planar optical waveguide equations by selecting zero elements
in a characteristic matrix,” J. Opt. Soc. Amer. A, vol. 2, pp. 595–602, Apr. 1985.
[13] T. Tamir, Guided-Wave Optoelectronics. Springer series in Electronics and Photonics 26,
New York: Springer-Verlag, 2 ed., 1988.
[14] R. E. Smith, S. N. Haude-Walter, and G. W. Forbes, “Numerical determination of planar
waveguide modes using the analyticity of the dispersion equation,” Opt. Lett., vol. 16,
pp. 1316–1318, Sept. 1991.
[15] R. E. Smith, S. N. Haude-Walter, and G. W. Forbes, “Mode determination for planar
waveguide using the four-sheeted dispersion relation,” IEEE J. Quantum Electron., vol. 28,
pp. 1520–1526, June 1992.
[16] E. Anemogiannis and E. N. Glytsis, “Multilayer waveguides: Efficient numerical analysis
of general structures,” J. Lightwave Technol., vol. 10, pp. 1344–1351, Oct. 1992.
[17] C. Chen, P. Berini, D. Feng, S. Tanev, and V. P. Tzolov, “Efficient and accurate numerical
analysis of multilayer planar optical waveguides in lossy anisotropic media,” Opt. Express,
vol. 7, pp. 260–272, Oct. 2000.
[18] M.-S. Kwon and S.-Y. Shin, “Simple and fast numerical analysis of multilayer waveguide
modes,” Opt. Commun., vol. 233, no. 1, pp. 119–126, 2004.
[19] M.-S. Kwon, “A numerically stable analysis method for complex multilayer waveguides
based on modified transfer-matrix equations,” J. Lightwave Technol., vol. 27, pp. 4407–
4414, Oct. 2009.
[20] K. A. Michalski and M. M. Mustafa, “Numerically stable and reliable computation of
electromagnetic modes in multilayered waveguides using the Cauchy integration method
with automatic differentiation,” IEEE Trans. Microwave Theory Tech., vol. 66, pp. 3981–
3992, Sept. 2018.
[21] J. D. Brown and R. V. Churchill, Complex Variables and Applications, ch. 4 and 7. New
York: McGraw-Hill Education, 9th ed., 2014.
98
[22] A. Ghatak, K. Thyagarajan, and M. Shenoy, “Numerical analysis of planar optical waveg-
uides using matrix approach,” J. Lightwave Technol., vol. 5, pp. 660–667, May 1987.
[23] E. Anemogiannis, E. N. Glytsis, and T. K. Gaylord, “Determination of guided and leaky
modes in lossless and lossy planar multilayer optical waveguides: Reflection pole method
and wavevector density method,” J. Lightwave Technol., vol. 17, pp. 929–941, May 1999.
[24] K. Mehrany, S. Khorasani, and B. Rashidian, “Variational approach for extraction of
eigenmodes in layered waveguides,” J. Opt. Soc. Amer. B, vol. 19, pp. 1978–1981, Sept.
2002.
[25] J. Petracek and K. Singh, “Determination of leaky modes in planar multilayer waveguides,”
IEEE Phot. Technol. Lett., vol. 14, pp. 810–812, June 2002.
[26] K. H. Schlereth and M. Tacke, “The complex propagation constant of multilayer waveg-
uides: an algorithm for a personal computer,” IEEE J. Quantum Electron., vol. 26, pp. 627–
630, Apr. 1990.
[27] G. P. Zouros, “CCOMP: An efficient algorithm for complex roots computation of deter-
minantal equations,” Comp. Phys. Comm., vol. 222, pp. 339–350, 2018.
[28] M. Koshiba, H. Kumagami, and M. Suzuki, “Finite-element solution of planar arbitrarily
anisotropic diffused optical waveguides,” J. Lightwave Technol., vol. 3, pp. 773–778, Aug.
1985.
[29] M. Koshiba and H. Kumagami, “Theoretical study off silicon-clad planar diffused optical
waveguides,” IEE Proc., Pt. J, vol. 134, pp. 333–338, Dec. 1987.
[30] E. Schweig and W. B. Bridges, “Computer analysis of dielectric waveguides: A finite-
difference method,” IEEE Trans. Microwave Theory Tech., vol. 32, pp. 531–541, May
1984.
[31] C. M. Arft and A. Knoesen, “An efficient finite-difference frequency-domain method in-
cluding thin layers,” Microw. Opt. Techn. Let., vol. 43, pp. 40–44, Oct. 2004.
[32] S. D. Gedney, “An anisotropic perfectly matched layer absorbing media for the truncation
of FDTD lattices,” IEEE Trans. Antennas & Prop., vol. 44, pp. 1630–1639, Dec. 1996.
99
[33] E. Anemogiannis, E. N. Glytsis, and T. K. Gaylord, “Efficient solution of eigenvalue equa-
tions of optical waveguiding structures,” J. Lightwave Technol., vol. 12, pp. 2080–2084,
Dec. 1994.
[34] L. F. Abd-ellal, L. M. Delves, and J. K. Reid, “A numerical method for locating zeros and
poles of a meromorphic function,” in Numerical Methods for Nonlinear Algebraic Equations
(P. Rabinowitz, ed.), ch. 3, pp. 47–59, London: Gordon and Breach, Science Publishers
Ltd. 1970.
[35] X. Ying and I. N. Katz, “A reliable argument principle algorithm to find the number of
zeros of an analytic function in a bounded domain,” Numer. Math., vol. 53, pp. 143–163,
Jan. 1988.
[36] P. Kravanja and M. van Barel, “A derivative-free algorithm for computing zeros of analytic
functions,” Computing, vol. 63, pp. 69–91, 1999.
[37] C. J. Gillan, A. Schuchinsky, and I. Spence, “Computing zeros of analytic functions in
the complex plane without using derivatives,” Comp. Phys. Comm., vol. 175, pp. 304–313,
Aug. 2006.
[38] G. Semwal and V. Rastogi, “Derivative free method for computing modes of multilayer
planar waveguides,” Opt. Quant. Electron., vol. 46, pp. 423–438, Mar. 2014.
[39] E. N. Glytsis and E. Anemogiannis, “A simple derivative-free method for zero-extraction
by phase-based enclosure for determination of complex propagation constants in planar
multilayer waveguides,” Appl. Opt., pp. xxx–xxx, Dec. 2018.
[40] X. Ying and I. N. Katz, “A simple reliable solver for all the roots of a nonlinear function
in a given domain,” Computing, vol. 41, pp. 317–333, Dec. 1989.
[41] M. Dellnitz, O. Schutze, and Q. Zheng, “Locating all the zeros of an analytic function in
one complex variable,” J. Comput. Appl. Math., vol. 138, no. 2, pp. 325–333, 2002.
[42] P. Kowalczyk and W. Marynowski, “Efficient complex root tracing algorithm for propa-
gation and radiation problems,” IEEE Trans. Antennas & Prop., vol. 65, pp. 2540–2546,
May 2017.
100
[43] P. Kowalczyk, “Complex root finding algorithm based on Delaunay triangulation,” vol. 41,
pp. 19:1–19:13, June 2015.
[44] M. J. Adams, An introduction to Optical Waveguides. Chichester: John Wiley & Sons,
Inc., 1981.
[45] D. L. Lee, Electromagnetic Principles of Integrated Optics. New York: John Wiley & Sons,
Inc., 1986.
[46] R. G. Hunsperger, Integrated Optics: Theory and Technology. Optical Sciences, New York:
Springer-Verlag, 2009.
[47] D. Marcuse, Theory of Dielectric Optical Waveguides. San Diego: Academic Press Inc.,
1991.
[48] C. Vassallo, Optical Waveguide Concepts. Optical Wave Sciences and Technology, New
York: Elsevier Science Publishing Company, Inc., 1991.
[49] A. B. Buckman, Guide-Wave Photonics. New York: Sauders College Publishing, 1992.
[50] K. Okamoto, Fundamentals of Optical Waveguides. Burlington, MA: Academic Press as
Inprint of Elsevier Inc., 2 ed., 2006.
[51] K. Kawano and T. Kitoh, Introduction to Optical Waveguide Analysis. New York: John
Wiley & Sons, 2001.
[52] K. S. Yee, “Numerical solution of initial boundary value problems involving Maxwell’s
equations in isotropic media,” IEEE Trans. Antennas & Prop., vol. 14, pp. 302–307, May
1966.
[53] R. C. Rumpf, “Simple implementation of arbitrary shaped total-field/scattered-field re-
gions in finite-difference frequency-domain,” Progress in Electromagnetics Research B,
vol. 36, pp. 221–248, 2012.
[54] K.-P. Hwang and A. C. Cangellaris, “Effective permittivities for second-order accurate
FDTD equations at dielectric interfaces,” IEEE Microw. Wirel. Compon. Lett., vol. 11,
pp. 158–160, Apr. 2001.
101
[55] A. S. Householder, The Numerical Treatment of a Single Nonlinear Equation. International
series in pure and applied mathematics, McGraw-Hill, 1970.
[56] L. C. Botten, M. S. Craig, and R. C. McPhedran, “Complex zeros of analytic functions,”
Comp. Phys. Comm., vol. 29, no. 3, pp. 245–259, 1983.
[57] L. C. Botten, M. S. Craig, and R. C. McPhedran, “Highly conducting lamellar diffraction
gratings,” Opt. Acta, vol. 28, no. 8, pp. 1103–1106, 1981.
[58] S.-L. Lee, Y. Chung, L. A. Coldren, and N. Dagli, “On leaky mode approximations for
modal expansion in multilayer open waveguides,” IEEE J. Quantum Electron., vol. 31,
pp. 1790–1802, Oct. 1995.
[59] R. E. Smith and S. N. Haude-Walter, “Leaky guiding in nontransparent waveguides,”
J. Opt. Soc. Amer. A, vol. 12, pp. 715–724, Apr. 1995.
102
You might also like
- Photodiodes - From Fundamentals To ApplicationsNo ratings yetPhotodiodes - From Fundamentals To Applications290 pages
- Gupta Et Al. 1996 - Microstrip Lines and Slotlines. 2nd Ed.100% (9)Gupta Et Al. 1996 - Microstrip Lines and Slotlines. 2nd Ed.547 pages
- Three Phase Power Transformer Modeling Using FEM For Accurate Prediction of Core and Winding Loss.No ratings yetThree Phase Power Transformer Modeling Using FEM For Accurate Prediction of Core and Winding Loss.6 pages
- A Numerically Stable Analysis Method For Complex Multilayer Waveguides Based On Modified Transfer-Matrix EquationsNo ratings yetA Numerically Stable Analysis Method For Complex Multilayer Waveguides Based On Modified Transfer-Matrix Equations8 pages
- DesignBuilder Simulation Training - HSD PDF100% (2)DesignBuilder Simulation Training - HSD PDF20 pages
- Ece Vi Antennas and Propagation (10ec64) Notes100% (1)Ece Vi Antennas and Propagation (10ec64) Notes112 pages
- Dielectric Slab Waveguide: 1 Total Internal ReflectionNo ratings yetDielectric Slab Waveguide: 1 Total Internal Reflection3 pages
- 352 39135 EC341 2015 2 2 1 Final EC341 2014 2015 Fall 1100% (2)352 39135 EC341 2015 2 2 1 Final EC341 2014 2015 Fall 12 pages
- Simplified Signal Flow Graph Analysis (Kuhn HP 1963)No ratings yetSimplified Signal Flow Graph Analysis (Kuhn HP 1963)8 pages
- L1M2 - TL and Impedance Matching Techniques PDFNo ratings yetL1M2 - TL and Impedance Matching Techniques PDF41 pages
- 11-THREE PORT DEVICES E H MAGIC TEE-02-Aug-2019Material - I - 02-Aug-2019 - Prabhu-Tee - E - H - Magic - Tee - S - MatrixNo ratings yet11-THREE PORT DEVICES E H MAGIC TEE-02-Aug-2019Material - I - 02-Aug-2019 - Prabhu-Tee - E - H - Magic - Tee - S - Matrix77 pages
- Analysis of Vectorial Mode Fields in Optical Waveguides by A New Finite Difference MethodNo ratings yetAnalysis of Vectorial Mode Fields in Optical Waveguides by A New Finite Difference Method8 pages
- Ec342 - Electromagnetic Waves and Wave GuidesNo ratings yetEc342 - Electromagnetic Waves and Wave Guides3 pages
- RF Circuit Design Lecture Notes - Google Search50% (2)RF Circuit Design Lecture Notes - Google Search1 page
- Solution To Microwave Engineering Pozar Chapter 14 Example 8 With MATLABNo ratings yetSolution To Microwave Engineering Pozar Chapter 14 Example 8 With MATLAB2 pages
- Dental Material MCQ DR Akram Altamimi - WatermarkNo ratings yetDental Material MCQ DR Akram Altamimi - Watermark30 pages
- Errata Digital Communications: John G. Proakis and Masoud Salehi Fifth Edition, Mcgraw-Hill, 2008No ratings yetErrata Digital Communications: John G. Proakis and Masoud Salehi Fifth Edition, Mcgraw-Hill, 20083 pages
- Chapter 3 Optical Components in Silicon PhotonicsNo ratings yetChapter 3 Optical Components in Silicon Photonics179 pages
- High-Quality Optical Ring Resonator-Based Biosensor For Cancer DetectionNo ratings yetHigh-Quality Optical Ring Resonator-Based Biosensor For Cancer Detection9 pages
- Toughening Mechanisms of Nanoparticle-Modified Epoxy PolymersNo ratings yetToughening Mechanisms of Nanoparticle-Modified Epoxy Polymers12 pages
- Lecture 23 Oblique Incidence and ReflectionNo ratings yetLecture 23 Oblique Incidence and Reflection21 pages
- Characterizing Losses in Microstrip Transmission Lines - ThesisNo ratings yetCharacterizing Losses in Microstrip Transmission Lines - Thesis67 pages
- Synopsis On Mathematical Model of Microwave Transmission100% (1)Synopsis On Mathematical Model of Microwave Transmission3 pages
- Guided Wave Study of Asymmetric Planner Slab Dielectric Optical WaveguideNo ratings yetGuided Wave Study of Asymmetric Planner Slab Dielectric Optical Waveguide3 pages
- Ch2 Characteristics of Semiconductor MaterialsNo ratings yetCh2 Characteristics of Semiconductor Materials32 pages
- Fundamentals of Microwave Communications (Microwave Planning Guide) 61-68No ratings yetFundamentals of Microwave Communications (Microwave Planning Guide) 61-688 pages
- System Level Simulation of Acoustic DevicesNo ratings yetSystem Level Simulation of Acoustic Devices8 pages
- Bahl - RF and Microwave Coupled-Line CircuitsNo ratings yetBahl - RF and Microwave Coupled-Line Circuits24 pages
- Heat Transfer IES GATE IAS 20 Years Question and Answers by S K Mondal85% (13)Heat Transfer IES GATE IAS 20 Years Question and Answers by S K Mondal97 pages
- Rectangular To Circular Waveguide TransitionsNo ratings yetRectangular To Circular Waveguide Transitions2 pages
- ICARA. Induced-Current Analysis of Reflector AntennasNo ratings yetICARA. Induced-Current Analysis of Reflector Antennas9 pages
- 2 - Atomic Structure (Engg) Sol - FinalNo ratings yet2 - Atomic Structure (Engg) Sol - Final16 pages
- Thermodynamics of Magnetic Refrigeration: Andrej Kitanovski, Peter W. EgolfNo ratings yetThermodynamics of Magnetic Refrigeration: Andrej Kitanovski, Peter W. Egolf19 pages
- Influence of Thermal Annealing Temperatures On Powder Mould Effectiveness To Avoid Deformations in ABS and PLA 3DPrinted PartsPolymersNo ratings yetInfluence of Thermal Annealing Temperatures On Powder Mould Effectiveness To Avoid Deformations in ABS and PLA 3DPrinted PartsPolymers19 pages
- Surface & Coatings Technology: Emerson Escobar Nunez, Seung Min Yeo, Kyriaki Polychronopoulou, Andreas A. PolycarpouNo ratings yetSurface & Coatings Technology: Emerson Escobar Nunez, Seung Min Yeo, Kyriaki Polychronopoulou, Andreas A. Polycarpou12 pages
- Geometry and Meshing Module: Niversity of Oronto Aculty of Pplied Cience and NgineeringNo ratings yetGeometry and Meshing Module: Niversity of Oronto Aculty of Pplied Cience and Ngineering10 pages
- Potential Applications of Metallic GlassesNo ratings yetPotential Applications of Metallic Glasses8 pages
