chapter 1 photonic crystals: properties and applications
TRANSCRIPT
Chapter 1
Photonic crystals: properties and
applications
This chapter is intended as a brief overview of the history, concepts, characteristics
and applications of photonic crystals (PhCs). A single chapter is insufficient to re-
view a field that continues to grow almost exponentially in the literature almost 20
years after it began, so attention is given to aspects most relevant to the work pre-
sented in the remainder of this thesis. Sections 1.1–1.4 introduce the concepts of one-,
two-, and three-dimensional photonic crystals and bandgaps and the motivation that
led to their development. The theory of band structures and Bloch modes of uniform
two-dimensional photonic crystals and photonic crystal slabs is then considered in Sec-
tions 1.5 and 1.6. Section 1.7 is concerned with the properties and applications of defects
in PhCs, and specifically those designed for resonant behaviour and waveguiding. Inte-
gration of these two functions into photonic devices is discussed in the context of two
basic operations: coupling and add/drop filtering. Finally, Section 1.8 focuses on a dif-
ferent class of applications, namely those that exploit the unusual dispersion properties
of the Bloch modes that exist in uniform photonic crystals.
1.1 Introduction
The optical properties of periodic structures can be observed throughout the natural
world, from the changing colours of an opal held up to the light to the patterns on a
butterfly’s wings. Nature has been exploiting photonic crystals for millions of years [1],
but humans have only recently started to realise their potential. One-dimensional peri-
odic structures in the form of thin film stacks have been studied for many years [2], but
three-dimensional photonic crystal were first proposed by Yablonovitch [3] and John [4]
in 1987. Yablonovitch proposed that three-dimensional periodic dielectric structures
could exhibit an electromagnetic bandgap - a range of frequencies at which light cannot
propagate through the structure in any direction. He also predicted that unwanted
spontaneous emission within a semiconductor can be prevented by structuring the ma-
5
6 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
terial so that the frequencies of these emissions fall within a photonic bandgap; since
no propagating states exist at that frequency, emission is effectively forbidden. John [4]
showed that many of the properties of PhCs survive even when the periodic lattice
becomes disordered. In such structures, if the index contrast is sufficiently large, strong
light localisation can still occur, in analogy to the electronic bandgaps of amorphous
semiconductors. Perhaps more relevant to much of the PhC research that has followed,
and to the topic of this thesis, was Yablonovitch’s [3] interpretation of the cavity modes
that can be introduced into a periodic structure by creating a defect or “phase slip”.
While resonant cavities in distributed feedback lasers had had already been demon-
strated using this approach [5], Yablonovitch showed that modes could be localised in
three-dimensions and explained the effect in terms of defect states in the photonic band-
gap. From this observation, and the initial proposals for limiting spontaneous emission,
the concept of controlling light with periodic structures has developed rapidly into a
topic of worldwide research.
Bandgaps in periodic materials were already well understood from solid-state physics,
where the presence of electronic bandgaps in semiconductors has revolutionised electron-
ics. Many of the concepts from solid-state research have been carried over to photonic
crystals including the notation and nomenclature, and perhaps this is what has allowed
the field to make such rapid progress in less than twenty years.
1.2 One-dimensional photonic crystals
Although the term photonic crystal (PhC) is relatively recent, simple one-dimensional
(1D) PhCs in the form of periodic dielectric stacks have been used for considerably
longer [2]. Their wavelength-selective reflection properties see them used in a wide
range of applications including high-efficiency mirrors, Fabry-Perot cavities, optical
filters and distributed feedback lasers. As illustrated in Fig. 1.1, the simplest PhC is
an alternating stack of two different dielectric materials. When light is incident on
such a stack, each interface reflects some of the field. If the thickness of each layer is
chosen appropriately, the reflected fields can combine in phase, resulting in constructive
interference and strong reflectance, also known as Bragg reflection. In contrast to
two- and three-dimensional PhCs, 1D Bragg reflection occurs regardless of the index
contrast, although a large number of periods is required to achieve a high reflectance
if the contrast is small. Since the absorption in dielectric optical materials is very
low, mirrors made from dielectric stacks are extremely efficient, and can be designed
to reflect almost 100% of the incident light within a small range of frequencies. The
main limitation of these dielectric mirrors is that they only operate for a limited range
of angles close to normal incidence.
Another, more recent application of 1D PhCs is the fibre Bragg grating (FBG), in
which the refractive index of the fibre core is varied periodically along its axis, typically
approximating a sinusoidal profile. This case is somewhat more complex because the
1.3. TWO-DIMENSIONAL PHOTONIC CRYSTALS 7
x
y
zd
Figure 1.1: Schematic of a one-dimensional photonic crystal consisting of a periodic
stack of dielectric layers with period d.
refractive index varies continuously, rather than discretely, as in the previous example,
but the properties are essentially the same. The main difference is that the refractive
index contrast in the FBG is so small (∆n ≤ 0.5%) that the operational bandwidth
is very narrow and thousands of periods are typically required to obtain the desired
reflectance properties. FBGs are now an integral part of fibre optic systems, being used
in dispersion compensation, filters, and a wide range of other applications [6].
1.3 Two-dimensional photonic crystals
Both two-dimensional (2D) and three-dimensional (3D) PhCs can be thought of as
generalisations to the 1D case where a full 2D or 3D bandgap appears only if the 1D
Bragg reflection condition is satisfied simultaneously for all propagation directions in
which the structure is periodic. For most 2D periodic lattices this occurs providing the
index contrast is sufficiently large, but for 3D structures only certain lattice geometries
display the necessary properties, and only then for large enough index contrasts (Ref. [7],
Chapter 6). The 3D case is discussed in more detail in Section 1.4.
Instead of a stack of uniform dielectric layers, 2D PhCs typically consist of an array
of dielectric cylinders in a homogeneous dielectric background material, as illustrated
in Fig. 1.2, although there are many other possible geometries. If the refractive index
contrast between the cylinders and the background is sufficiently large, 2D bandgaps can
occur for propagation in the plane of periodicity — perpendicular to the rods. Light at
a frequency within the bandgap experiences Bragg reflection in all directions due to the
periodic array of cylinders. However, as in the 1D case where light could still propagate
in two-dimensions, in a 2D PhC propagation can still occur in the non-periodic direction,
parallel to the cylinders. Thus, an alternative means of confinement is required in the
third dimension to avoid excessive losses due to diffraction and scattering. This issue
is discussed further in Section 1.6.
As in semiconductor devices, much of the interest in photonic crystals arises not
8 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
(e)
(c) (d)
(a) (b)
Figure 1.2: (a) Schematic of a 2D hole-type PhC slab consisting of low refractive
index cylinders in a high-index slab. (b) Scanning electron microscope (SEM) image
of a fabricated hole-type PhC in a silicon slab (from Ref. [8]). (c) Schematic of a 2D
rod-type PhC consisting of high refractive index cylinders in a low-index background.
(d) SEM image of a fabricated rod-type PhC formed from GaAs rods on a low-index
aluminium oxide layer (from Ref. [9]).
from the presence of a bandgap alone, but rather from the ability to create localised
defect states within the bandgap by introducing a structural defect into an otherwise
regular lattice. For example, the removal of a single cylinder from a 2D PhC creates
a point-like defect or resonant cavity, and the removal of a line of cylinders can create
a waveguide that supports propagating modes. Many potential applications based on
this concept have been proposed and demonstrated, a number of which are discussed
in more detail in Section 1.7.
A second class of 2D PhC applications exploits the unique properties of the propa-
gating modes that exist outside the bandgaps in defect-free PhCs. The discrete transla-
tional symmetry of PhCs imposes strict phase conditions on the field distributions that
they support. As a result, only a discrete number of modes are supported for any given
frequency and light propagating in these Bloch modes can have very different proper-
ties to light in a homogeneous medium. We review some of the potential applications
related to this in Section 1.8.
1.4. THREE-DIMENSIONAL PHOTONIC CRYSTALS 9
1.4 Three-dimensional photonic crystals
Three-dimensional PhCs have proved to be the most challenging PhC structures to
fabricate. Whereas 2D PhC research has gained significant benefit from well-established
1D PhC thin-film and semiconductor processing technology such as plasma deposition
and electron-beam lithography, fabrication of 3D PhCs has required the development
of entirely new techniques. For this reason, it was more than three years after the
initial proposal for 3D band gap materials [3, 4] before a structure was calculated to
exhibit a bandgap for all directions and all polarizations [10]. The design consisted
of dielectric spheres positioned at the vertices of a diamond lattice. This followed
experimental reports the previous year in which a partial bandgap in a face-centred-
cubic (FCC) lattice of spheres was mistakenly identified as a complete bandgap [11].
This latter result highlighted the requirement for rigorous theoretical and computational
tools capable of dealing with high-index contrast dielectrics.
Since these early studies, a wide range of 3D PhC geometries exhibiting complete
bandgaps have been demonstrated both in theory and experiment. As an example, a
3D “woodpile” PhC is shown in Fig. 1.3. Due to the challenges involved in fabricating
high-quality structures with features on the scale of optical wavelengths, early photonic
crystal experiments were performed at microwave and mid-infrared frequencies [12–
14]. With the improvement of fabrication and materials processing methods, smaller
structures have become feasible, and in 1999 the first 3D PhC with a bandgap at
telecommunications frequencies was reported [15]. Since then, various lattice geometries
have been reported for operation at similar frequencies [16–18].
Waveguiding and the introduction of intentional defects in 3D PhCs has not pro-
gressed as rapidly as in 2D PhCs, largely due to the fabrication difficulties and the more
complex geometry required to achieve 3D bandgaps. Theoretical studies have demon-
strated the potential for novel photonic circuit designs [19, 20], but to date only a few
experimental results have been reported [21, 22]. Although much of the recent interest
in PhCs has focused on telecommunication related applications, the original concept
of controlling spontaneous emission has not been forgotten. Recent experiments have
demonstrated both inhibition and enhancement of spontaneous emission from quantum
dots embedded in both in 2D [23] and 3D [21, 24] PhCs. The presence of a photonic
bandgap at black-body radiation frequencies has also been shown to modify the thermal
emission properties of heated tungsten 3D PhCs [25,26].
3D PhCs formed in low-index contrast materials such as silica or polymer are also
potentially useful for applications such as those described in Section 1.8 where a com-
plete bandgap is not required. Superprism effects have been calculated in 3D polymer
PhCs [27], and tunable bandgap effects have also been demonstrated using both non-
linear [28] and liquid crystal tuning [29,30].
10 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
(a) (b)
Figure 1.3: An example of a 3D PhC woodpile structure known to exhibit a full photonic
bandgap. (a) Schematic of an ideal woodpile PhC. (b) SEM image of a real 3D woodpile
structure fabricated in silicon (from Ref. [14]).
1.5 Band structure and Bloch modes of 2D pho-
tonic crystals
The remainder of this thesis is concerned with 2D PhCs and their application to optical
processing and photonic integrated circuits. In this section we review the properties of
the 2D band structure and associated Bloch modes of uniform PhC lattices. A truly
2D PhC is invariant in the direction parallel to the cylinder axis, and thus has cylin-
ders of infinite length. For a rigorous analysis of such a structure, the out-of-plane
wavevector component must be included in a full 3D calculation, but this can be im-
practical for large structures given the computational demands of 3D calculations. In
a 2D calculation only in-plane propagation is considered, but there are some simple
modifications that can be made to correct partially for out-of-plane wavevector com-
ponents. Although real PhC structures have finite-length cylinders and usually rely
on a slab waveguide geometry to prevent out-of-plane losses, the underlying physics is
the same, and significant insight can be obtained by considering the ideal case. Unless
otherwise stated, the structures studied throughout this thesis are treated as 2D and
the light is taken to be propagating in the plane of the PhC. The implications of this
approximation are discussed in general terms in Section 1.6.4, and more specifically in
the chapters relating to each of the devices.
1.5.1 Band structure calculation
Figures 1.4(a) and (d) show two typical 2D PhC geometries consisting of a square
and a triangular array of cylinders of radius r and refractive index ncyl embedded in a
background dielectric of refractive index nb. The choice of index contrast (ncyl > nb or
ncyl < nb) has a number of important consequences, not only from the point-of-view of
fabrication, but also in terms of fundamental properties. To distinguish between these
1.5. BAND STRUCTURE AND BLOCH MODES OF 2D PHOTONIC CRYSTALS11
two geometries, we will refer to the former as rod-type and the latter as hole-type PhCs.
Examples of both are shown in Fig. 1.2.
Many of the techniques for solving quantum mechanical problems in solid state
physics can be used for electromagnetic fields by casting Maxwell’s equations in the
form of an eigenvalue equation in terms of either the electric (E) or magnetic (H) field
∇2E = −(
ωc
)2εE, (1.1)
∇×(
1ε∇×
)H =
(ωc
)2H , (1.2)
where ε = ε(x, y, z) is the dielectric constant, which may depend also on frequency, and
c is the speed of light. A time dependence of exp(−iωt) has been included implicitly
and we assume that ε is everywhere real and positive, and that µ = µ0 corresponding to
a lossless, non-magnetic material. Whereas plane wave based numerical methods (see
Section 1.9) typically solve Eq. (1.2), the Bloch mode matrix method described in Chap-
ter 2 involves the solution of Eq. (1.1). We discuss here the band structure calculations
in terms of Eq. (1.1), although the conclusions are equally valid for Eq. (1.2). In Eq. (1.1)
the eigenvalue term (ω/c)2ε is analogous to the energy eigenvalue in Schrodinger’s equa-
tion describing the propagation of electrons in a potential.
For a 2D PhC such as those illustrated in Fig. 1.2, the periodic modulation of the
dielectric constant can be written expressed as ε(r) = ε(r+lpq). Here, lpq = pe1+qe2 is
a general lattice vector defined in terms of the basis vectors e1 and e2, and r = r(x, y)
is a vector in the plane perpendicular to the cylinders. The propagation of a wave in
such a periodic medium is governed by the Floquet-Bloch theorem which states that the
solutions to Eq. (1.1) correspond to plane wave fields modulated by a periodic function.
These Bloch modes have the form
Emk(r) = eik0.rumk(r), (1.3)
where umk(r) has the periodicity of the PhC, i.e. umk(r) = umk(r + lpq). Thus, for a
given Bloch vector k0 = (kx, ky), the eigenvalue equation (1.1) can be solved to yield
a collection of Bloch mode eigenfunctions Emk and corresponding eigenvalues ωm(k0).
When plotted over reciprocal space, the ωm(k0) form discrete bands, where each band
is a continuous function of k0, indexed by m = 1, 2, 3 . . . with increasing frequency
as seen in Fig. 1.5. A 2D bandgap occurs at frequencies where no solutions exist
for all possible (real) k0-vectors. Note that an equivalent calculation for the modes
of a uniform dielectric of refractive index n yields a conical band surface defined by
(ω/c)2 = (k0/n)2, where k0 is the free-space wavevector.
The eigensolutions are also periodic functions of k0 on the reciprocal lattice defined
by the basis vectors b1 and b2, where ei.bj = 2πδij, as shown in Figs. 1.4(b) and
(e). Thus, to characterise the band structure completely, it is necessary only to find
solutions for k0 vectors within the first Brillouin zone, defined as the region of reciprocal
space centred on k0 = 0 in which any two wavevectors are separated by less than a
reciprocal lattice vector. The first Brillouin zones of a square and a triangular lattice
are illustrated in Figs. 1.4(c) and (f) respectively.
12 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
For lattices with rotational and/or mirror symmetry in addition to translational
symmetry, the complete band structure can be obtained by applying these symmetry
operations to the solutions in an even smaller region of reciprocal space - the irreducible
Brillouin zone, indicated by the white triangles in Figs. 1.4(c) and (f). The origin of
the Brillouin zone is labelled as the Γ-point in both cases, while the other two vertices
are labelled M and X for the square lattice and M and K for the triangular lattice.
For many applications it is sufficient to calculate only the modes corresponding to
k0 vectors lying on the edge of the irreducible Brillouin zone, since the local minima
and maxima of the band structure tend to lie on the symmetry axes and at the high-
symmetry points of the Brillouin zone. Thus, a typical band diagram for a 2D PhC
shows only the band structure along the edges of the irreducible Brillouin zone between
each pair of labelled symmetry points, as can be seen in Figs. 1.5(c) and (d). From
such a diagram, the position and width of all bandgaps can be determined, along with
many other characteristic features of the band structure. In Section 1.5.3 we consider
the two most common lattice types used for 2D PhCs – square and triangular – and
give examples of the band structure for these lattices.
1.5.2 TE and TM modes
In the analysis described in Section 1.5.1, no assumptions are made about the polarisa-
tion of the electromagnetic fields. However, in the case of a 2D PhC, where propagation
is restricted to a plane perpendicular to the cylinders, the solutions decouple into two
distinct polarisation states - transverse electric (TE) and transverse magnetic (TM).
For TE (TM) modes, the electric (magnetic) field lies in the xy-plane, while the mag-
netic (electric) field is aligned with the z-axis, along the cylinders. In general the band
structures of the TE and TM states are quite different due to the boundary conditions
imposed at the dielectric interfaces. Therefore most 2D PhCs are designed for operation
in a single polarisation state. The choice depends largely on the geometry of the PhC
and the desired properties; in rod-type geometries, the largest bandgaps tend to occur
for TM polarisation, so for bandgap applications this is usually the best choice; the
reverse is the case in hole-type PhCs where TE polarised states tend to have the largest
bandgaps (Ref. [7], Chapter 5).
1.5.3 Band structures of square and triangular lattices
The first 2D PhC structures found to exhibit photonic bandgaps were square and trian-
gular lattices of cylinders [31,32]. These geometries are still the most common, and the
ones considered in this thesis. Systematic studies of both rod- and hole-type geometries
have been undertaken to identify the optimal choice of cylinder radius, index contrast
and lattice type in order to maximise the frequency range of the bandgap for either
one or both polarisations [33, 34]. In all cases, there is a minimum refractive index
contrast below which gaps do not exist for either polarisation, and not all geometries
1.5. BAND STRUCTURE AND BLOCH MODES OF 2D PHOTONIC CRYSTALS13
e1
e2
d
d e1
e2
G
M
X
G
KM
b2
b1
b1
2
d
p
b2
Real lattice Reciprocal lattice Brillouin zone
x
y
kx
ky
Square
Triangular
(f)(e)(d)
(b) (c)(a)
Figure 1.4: (a) and (d): Diagrams of square and triangular PhC lattices in real space
generated by the basis vectors e2 and e2. (b) and (e): The corresponding reciprocal
lattices and basis vectors b1 and b2. The boundaries of first Brillouin zones are indicated
by the dotted lines. (c) and (f): Representation of the first Brillouin zones and the
irreducible region (white triangles) of the two lattices.
have overlapping TE and TM bandgaps, regardless of the index contrast.
Figure 1.4 shows a square and a triangular lattice represented in both real and
reciprocal space along with the basis vectors used to generate any lattice vector. The
first Brillouin zone of each lattice type is shown in parts (b), (c), (e) and (f) of the figure.
In (c) and (f), the irreducible Brillouin zones are defined by the lines of symmetry joining
the high symmetry points, which are labelled according to convention.
As an example, we consider here the band structures of two typical PhC geometries:
the first is a rod-type PhC consisting of a square lattice of silicon cylinders of index
ncyl = 3.4 and radius r = 0.25 d embedded in silica with index nb = 1.46, and the
second is a hole-type PhC formed by a triangular lattice of air holes (ncyl = 1) of
radius r = 0.32 d in silicon (nb = 3.4). When calculating the properties of PhCs, it
is standard practice to express all spatial dimensions relative to the lattice period d
as this allows structures to be scaled for operation at any desired frequency, although
material dispersion must be taken into account for large changes of scale. Thus, the
cylinder radius is typically written as r/d and the wavelength as λ/d. The dimensionless
frequency d/λ is equivalent to the parameter ωd/2πc used in some texts.
14 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
Figure 1.5(a) shows the first four TM polarised bands (see Section 1.5.2) of the
square lattice PhC plotted as a function of frequency over first Brillouin zone. Since
the eigenvalues ωm(k0) are continuous functions of k0 within each band, they form
band surfaces. Observe that there is a bandgap between the first and second bands.
Figure 1.5(b) is a contour plot of the second band showing the equifrequency, or isofre-
quency contours over the Brillouin zone. Visualising the bands in this manner can be
informative for some applications, but as mentioned in Section 1.5.1, it is often sufficient
to calculate the bands along the perimeter of the irreducible Brillouin zone, indicated
by the dashed line. Figure 1.5(c) shows such a band diagram for the same PhC, for
both the TE and TM polarised modes. The TM bandgap is clearly seen over the nor-
malised frequency range of 0.249 < d/λ < 0.298, but there is no equivalent gap for TE
polarisation. Figure 1.5(d) shows the band diagram for the triangular lattice PhC with
the parameters given earlier in this section. In this case, there is a large bandgap for
TE polarisation between 0.216 < d/λ < 0.298 but none for TM.
�����
��������������������������� �����
� ππ� π π ��
�������� �����
Γ
�
�
"!�# "!�#%$ "!�#'& "!�#)( "!�#)* "!+& "!+&,$
-/.λ
01 λ
243
���5���������� ����������������6������
��������� �7����������8�����������
9;:=< 9;>?<
24@ 24324@
Γ A B Γ Γ ΓB C
�� �
01 λ
01 λ
Figure 1.5: (a) TM band surfaces for a 2D square rod-type photonic crystal with the
parameters defined in the text. (b) Equifrequency contour plot for the second band in
(a). (c) Band diagram for the same structure showing TE and TM modes on the edge
of the irreducible Brillouin zone. (d) Band diagram for the 2D triangular hole-type
photonic crystal with the parameters given in the text.
For bandgap applications, the position and size of the gap are usually the critical
1.6. PHOTONIC CRYSTAL SLABS 15
parameters, while the details of the band structure above and below the gap are much
less important. Recently, however, the unusual propagation properties of the Bloch
modes have been attracting renewed attention and a number of novel applications have
been proposed. These applications require the band surfaces to be engineered to have
very specific properties, so a full calculation of the bands over the Brillouin zone is
required. We discuss some of these in-band applications in Section 1.8.
1.6 Photonic crystal slabs
We have so far discussed only the properties of purely 2D PhC geometries which are
invariant along the z-axis, and thus have cylinders of infinite length. This approximation
is valid if the cylinders are many wavelengths long, but for practical purposes this is
not always desirable or possible to achieve. Even in geometries that can be considered
as purely 2D, diffraction losses are high since light is free to propagate out of the xy-
plane even when there is a bandgap for in-plane propagation. An exception to this
are photonic bandgap fibres in which light propagates almost parallel to the cylinders
and is confined to a central core by the periodic cladding [35]. But for most other
applications, a method is required to confine the light in the out-of-plane direction
within a 2D PhC of finite height. One way to achieve this is to cap the slab above
and below with another PhC structure that has a bandgap for propagation in the z-
direction. Suggestions have been made for 1D [36, 37], 2D [38] and 3D [39] capping
layers, but to date these approaches have not been demonstrated experimentally.
A more common technique is to confine the light in the slab using total internal
reflection (TIR) at the slab/cladding interface as in a planar dielectric waveguide. This
approach was proposed in 1994 [40], and demonstrated experimentally in 1996 [41], but
was not studied rigorously until 1999 [42] for uniform PhC slabs and later for PhC
waveguides [38, 43, 44]. To achieve TIR, the refractive index of the cladding must be
lower than that of the slab, or in the case of rod-type PhCs, the high-index rods. A
bulk dielectric such as air or silica is typically used for the low index cladding material,
as illustrated in the examples of Fig. 1.2. A more detailed discussion of slab design in
both rod- and hole-type PhCs is provided in Sections 1.6.2 and 1.6.3.
1.6.1 The lightline
As with any index-guiding structure, for light to be confined within the slab, it must
satisfy a total internal reflection condition to ensure that light in the core does not
couple to the radiation modes of the cladding. The dispersion relationship for a mode
in a homogeneous dielectric cladding of refractive index ncl is given by(nclω
c
)2
= k20 + k2
z , (1.4)
16 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
where k0 = (kx, ky) is the wavevector component in the plane of the slab. Thus,
if k20 > (nclω/c)
2, then kz is imaginary and the field in the cladding is evanescent,
corresponding to TIR. If k20 < (nclω/c)
2 however, kz is real and light can propagate
in the cladding. When plotted in k − ω-space, the relation k0 = (nclω/c)2 defines the
light cone, as illustrated by the white cone in Fig. 1.6. In a conventional band diagram
this appears as a lightline, such as those in Fig. 1.7. Above the lightline a continuum
of propagating modes exists in the cladding, whereas no propagation is allowed below.
Therefore, modes of the PhC that lie inside the cone (above the lightline) can couple to
the radiation modes and leak into the cladding but modes that lie below the lightline
decay exponentially in the cladding and are guided within the slab.
G
M K
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
d/l
kx
ky
43p
-43p
00
-23p
23p
Figure 1.6: Band surface plot showing the first four bands of the hole-type PhC con-
sidered in Section 1.5.3, but with a finite slab of height 0.4 d surrounded by air, as in
Fig. 1.5(d). The white surface centred at Γ represents the light cone. Modes outside the
cone are guided in the slab, while those inside the cone can radiate into the cladding.
The band properties of a PhC slab are in many ways similar to those of the 2D PhCs
discussed in Section 1.5.3, but a number of additional effects must also be considered
including the slab thickness, refractive index contrast and symmetry properties of the
slab and the cladding. One of the most important differences is the effect of out-of-plane
propagation on the mode structure. In the 2D analysis, the TE and TM polarised modes
are decoupled and can be treated separately, but this is not the case for the modes of
a PhC slab. However, if the slab and cladding are symmetric about the z = 0 plane,
1.6. PHOTONIC CRYSTAL SLABS 17
these modes can be classified according to the symmetry of the Hz field component
with respect to the reflection plane. In this case, the TE and TM modes of a purely
2D structure map onto even and odd slab modes respectively and bandgaps can exist
for one or other of these symmetries. These are not true bandgaps because light can
always propagate above the lightline in cladding modes. However they correspond to
gaps in the spectrum of guided modes, and thus share many of the properties of 2D
photonic bandgaps.
1.6.2 Hole-type photonic crystal slabs
The fabrication and experimental demonstration of PhC-based devices has progressed
much more rapidly for hole-type PhCs than for rod-type structures, despite both ge-
ometries receiving almost equal attention in theoretical studies. This difference is in
part due to the fabrication challenges involved in forming circular rods with smooth,
vertical sides, but is also related to the issue of vertical confinement. In a hole-type
PhC, the high-index material surrounding the holes can also serve as a high-index slab
for TIR guidance, but in rod-type PhCs, where the high-index regions are separated
from one another, alternative methods are required to achieve TIR while maintaining
structural integrity. This issue is discussed further in Section 1.6.3.
There are two cladding designs that are used most for hole-type PhC slabs. The first
of these consists of a high refractive index PhC slab on top of a lower refractive index
substrate with an air cladding above. Such structures are relatively easy to fabricate
using existing thin film deposition techniques to lay down the multiple layers [41, 45],
but the resulting asymmetric slabs do not allow for the simple splitting of the odd and
even modes. More recently, methods have been developed to fabricate thin membranes
of PhC material suspended in air, producing symmetric air-clad PhC slabs, also known
as membrane-type or air-bridge PhCs. High-quality low-loss PhCs have been fabricated
in both Si and GaAs slabs using this method [8,46].
Figure 1.7(a) shows the band diagram calculated for the same photonic crystal as
in Fig. 1.5(d), but now for a finite slab of height 0.4 d with air above and below. The
corresponding band surfaces are shown in Fig. 1.6. Only the modes below the lightline
are plotted, while the region above the lightline is shaded light grey and contains a
continuum of radiation modes. The even mode bandgap in the range 0.292 ≤ d/λ ≤0.388 is shaded dark grey and corresponds to the TE bandgap in the 2D calculation,
although it has shifted to higher frequencies.
1.6.3 Rod-type photonic crystal slabs
Rod-type PhC slab geometries are inherently different from hole-type structures because
the regions of high refractive index are not connected, and thus the rods must be
attached to a substrate material. Index guidance at the bottom of the rods is achieved
by choosing a substrate of lower refractive index, or even etching the rods into the
18 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
Γ M K Γ
odd
even
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
d/l
Γ X M Γ
odd
even
0.5
0.4
0.3
0.2
0.1
0.0
d/l
(a) (b)
Figure 1.7: (a) Band diagram of a hole-type PhC slab with the same parameters as in
Fig. 1.5(d) and a slab height of 0.4 d. The cladding above and below the slab is air. (b)
Band diagram of a rod-type PhC with the same parameters as in Figs. 1.5(a)–(c), but
for cylinders of height 1.5 d surrounded above and below by a silica cladding. Only the
guided modes below the cladding lightline are plotted. The regions shaded light grey
lie above the lightline and contain a continuum of cladding modes.
substrate so that the index of the rods is lower at the bottom than at the top. If
the rods are left surrounded by air, then index confinement at the top of the rods
is provided by the rod-air interface [47]. The main drawback of this geometry is the
strong asymmetry due to the different cladding materials above and below the rods.
One solution is to fill the region between and above the rods with a low-index material
that matches the substrate index, thus creating an up-down symmetric array of finite
rods analogous to the hole-type PhC membrane structure [48, 49]. A similar design
was proposed by Martinez et al., [50] in which two different materials were used – one
to fill the spaces between the rods, and another with a slightly lower index for the
cladding above and below. However it is not clear whether such a small index contrast
contributes significantly to the guidance, given the already strong contrast between the
rods and the cladding. In these last two approaches, it is important that a sufficiently
large index contrast is maintained between the rods and the surrounding material to
ensure that the in-plane bandgap is not destroyed.
Figure 1.7(b) shows the band diagram calculated for the same photonic crystal
as in Figs. 1.5(a)–(c), but now for finite rods of height 1.5 d with the SiO2 back-
ground extending above and below. The bandgap for odd-symmetry modes in the
range 0.293 ≤ d/λ ≤ 0.314 corresponds to the TM bandgap in the 2D calculation, again
shifted to higher frequencies.
1.6. PHOTONIC CRYSTAL SLABS 19
1.6.4 2D vs 3D modelling of PhC slabs
Rigorous theoretical analysis of PhC slabs requires full 3D modelling tools, but these
involve intensive numerical calculations that can take days or weeks to run on a desk-
top computer, or may be impossible due to memory requirements. Although parallel
computing methods and access to supercomputers may be an option, significant results
and insight can be obtained using 2D methods.
When operating below the lightline, the main difference between 2D and 3D calcula-
tions is a frequency shift in the band structure. This occurs because the band structure
is largely determined by the wavevector component k0 that lies in the plane of the PhC,
rather than the total k. Since k0 = k − kz, out-of-plane propagation has the effect of
reducing k0 for a given frequency. Thus, band features associated with a particular
k0 are shifted to higher frequencies, as observed in Sections 1.6.2 and 1.6.3. In many
cases this effect can be corrected to some extent by replacing the refractive index of
the background material in the 2D calculation with an effective refractive index. For
hole-type PhCs, an appropriate effective index is found by first calculating the guided
modes of the slab without any holes. The effective indices of the slab modes are pro-
portional to their in-plane wavevector. Since most PhC slabs are designed to support
only a single slab mode, calculating the 2D band structure with the effective index of
the fundamental slab mode approximates the effect of out-of-plane propagation [51].
For relatively low slab-cladding index contrasts in both symmetric and asymmetric
hole-type PhC slabs, the effective index method generally agrees well with the full 3D
band structure and waveguide mode calculations [51]. Experimental results have also
shown good agreement with 2D calculations [52, 53]. The approximation deteriorates
for high index contrasts, such as in membrane-type slabs, as the tighter slab mode
confinement introduces significant waveguide dispersion and other effects not easily
accounted for in 2D calculations. The method is also unsuitable for studying strongly
resonant behaviour in cavities as the operation of these structures can be severely
degraded by out-of-plane losses. In these cases it is essential to understand and control
the radiation properties [54] and hence 3D calculations are generally required. Another
case where 3D methods are needed is the interaction region between modes of different
symmetry [55].
Applying the effective index approximation to rod-type PhCs is somewhat less
straightforward since the material between the rods is often the same as the cladding.
In this case, or for hole-type PhCs with large holes, it may be appropriate to calculate
first an average refractive index of the PhC, taking into account the refractive indices
and the relative fill-fractions of the cylinders and background material (Ref. [56], Chap-
ter 10). The average index value could then be used as the slab index when calculating
the guided slab modes. Although applying an average index to a PhC is strictly valid
only when the wavelength is significantly larger than the lattice period, this approach
may at least allow the effective index method to be applied to rod-type PhCs as an
intermediate approximation between 2D and full 3D calculations. However there are
20 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
no reports in the literature of this method being tested.
In Chapters 3 and 4 we present several novel PhC waveguide-based devices, all of
which are studied using the purely 2D techniques described in Chapter 2. Although the
devices have not been explicitly designed for implementation in a PhC slab geometry,
their operation relies on generic PhC properties such as waveguiding, mode coupling
and omnidirectional reflection. The exception to this is Chapter 5, where 3D numerical
calculations are used to verify the efficient coupling characteristics of rod-type PhCs.
Rather than presenting devices optimised for fabrication, this thesis is intended to
highlight some of the unique properties of 2D photonic crystals and demonstrate novel
ways to utilise these properties.
1.7 Defect states in PhCs: cavities and waveguides
Guiding and trapping light using waveguides and resonant cavities are two fundamental
optical functions that enable a range of all-optical devices to be created. Waveguides
not only perform the tasks of their electrical analogues, wires, by transferring light from
one part of a circuit to another, but are used in many other devices such as couplers,
junctions and interferometers. Resonant cavities have many potential applications that
make use of the sharp spectral response and very strong field intensities that occur at
resonance. There are various methods for achieving efficient waveguiding and many
others for producing high quality optical cavities, but very few single technologies allow
both to be engineered in a single integrated structure. Two dimensional photonic
crystals can provide just such a combination due to their versatile geometry that allows
simultaneous fine tuning of a number of parameters. 3D PhCs could potentially provide
even greater control over light, but they are considerably more challenging to fabricate.
In addition, the complex lattice structures required to achieve full 3D bandgaps can
make it very difficult to create high quality defects.
The possibility of producing cavities and waveguides in PhCs by introducing defects
was recognised not long after the first 2D bandgap structures were proposed. In 1994,
Meade et al. suggested that both cavities and waveguides could be created in 2D
PhCs by simply changing or removing a single cylinder or a whole line of cylinders,
respectively [40]. A defect created in this way is surrounded by the uniform PhC that
acts as an omnidirectional mirror for light at frequencies in the bandgap, thus trapping
light in the defect region. In a cavity defect, this can result in sharp resonant behaviour,
whereas a linear defect allows light to propagate along the defect while being confined
in the transverse direction by perfectly reflecting PhC walls. In a PhC slab, this is only
true for light confined vertically by TIR, but the behaviour is essentially the same.
1.7. DEFECT STATES IN PHCS: CAVITIES AND WAVEGUIDES 21
1.7.1 Cavities
Resonant cavities provide both sharp spectral responses and large field enhancement
within the cavity when the resonance condition is satisfied. The former can be used for
narrow bandwidth filters and wavelength selective couplers, both of which are required
in wavelength-division-multiplexing (WDM) optical systems for operating on individual
frequency channels. High field intensities due to light being trapped in a small cavity
can enhance light-matter interaction, making them ideal for photonics applications such
as lasers and nonlinear optics. They are also useful in sensing applications and for more
fundamental research into cavity quantum electrodynamics and control of spontaneous
emission. A review of many of these applications is given in Ref. [57].
An ideal optical cavity would trap light indefinitely and have a single resonant
frequency, but real cavities are limited by losses, both due to radiation and absorption.
A measure of how close a cavity approximates an ideal resonator is given by the quality
factor Q, which is proportional to the lifetime of light in the cavity,
Q = ω0stored energy
power loss per cycle= −ω0
U
dU/dt, (1.5)
where ω0 is the resonant frequency, and U is the energy stored in the cavity (Ref. [58],
chapter 8). Therefore, an ideal cavity would have an infinite Q. A more practical
definition for the purposes of measuring Q in experiment or simulation is given by the
relative width of the resonance,
Q =ω0
∆ω, (1.6)
where ∆ω is the full-width at half-maximum (FWHM) of the resonant power spectrum.
Field intensities are also determined by the volume of the cavity V , so to achieve the
optimum field enhancement, the ratio Q/V should be maximised. Silica microspheres
hold the record for the highest Q values, with experimental demonstrations of Q > 109
[59], but the geometry is not easily integrated with other photonic devices and the
modal volumes are relatively large. Toroid microcavities are perhaps more promising
as they have been demonstrated with Q > 108 in an on-chip silicon-on-insulator (SOI)
geometry [60]. However, despite being fabricated on-chip, a tapered optical fibre was
still required for coupling to the toroid. While this method allows fine-tuning, it may
not be suitable for dense integration as the condition for coupling between the fibre and
the resonator is likely to be delicate and hence sensitive to perturbations. In contrast,
PhC cavities can be coupled into directly from nearby waveguides in the same structure,
requiring no additional alignment or coupling control.
Whereas a cavity in a 3D PhC could theoretically provide perfect confinement,
cavities in PhC slabs rely on TIR for confinement in the vertical direction, and so
their performance is largely dependent on out-of-plane losses. For this reason, it is
informative to separate the contributions due to in-plane effects Q‖ and out-of-plane
22 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
effects Q⊥. These are related to the total Q of the cavity by
1
Q=
1
Q‖+
1
Q⊥. (1.7)
From this equation it can be seen that the cavity Q is limited by the smaller of Q‖ and
Q⊥. Q‖ depends on a number of factors including the number of PhC layers surrounding
the cavity and the scattering losses due to imperfections in the lattice. Q⊥ depends on
the out-of-plane radiation, most of which occurs when light is reflected at the interfaces
surrounding the cavity [61].
Recall from Section 1.5.1 that Bloch modes of a 2D PhC extend throughout the
crystal lattice and can thus be characterised by a single frequency eigenvalue ωm(k0)
for every in-plane wavevector k0. When light is confined to a small cavity, it is no
longer represented by a single wavevector, but the modal field instead contains a super-
position of spatial frequencies in the same way that a finite beam can be expressed as a
superposition of plane waves. Any components with k20 < (nclω/c)
2 lie inside the light
cone, and can couple to radiation modes leading to losses that reduce Q⊥ and hence
result in significant reductions in the total Q of the cavity.
The first PhC cavities to be fabricated exhibited Q values of only a few hundred
[62,63]. These low values were due not only to out-of-plane losses inherent in the cavity
design, but also to scattering losses introduced by structural variations and surface
roughness. In 2001, Vuckovic et al. reported a systematic theoretical study of various
cavity designs to establish methods for reducing radiation losses [64]. Cavities with
Q > 104 were calculated by varying the radius and shape of the cylinders in the area
surrounding the defect. Structures based on some of these designs were also fabricated
and measured to have Q = 2800 [65].
Many of these early PhC cavities were designed for lasing and included quantum
wells in the slab structure that produce photoluminescence when optically pumped
from outside the slab but the first experimental report of direct coupling between a
PhC cavity and a waveguide was in 2001 [66], where a cavity was created in a section
of PhC between the ends of two waveguides. When light was coupled into the end of
one waveguide, a sharp resonant transmission of Q = 816 was observed
Although radiation losses were considered in the various cavity designs discussed
above, the direct relationship between Q and the wavevector components of the mode
was not explored. The wavevector distribution of a mode at the resonant frequency is
given by the 2D spatial Fourier transform of the field. In this k-space representation,
the light cone forms a circle of radius (nclω/c). Any component of the Fourier transform
that lies within the circle contributes to radiation losses and hence reduces Q⊥ and the
total Q of the cavity. Theoretical studies have shown that by careful adjustment of the
size and shape of the cavity, it is possible to redistribute almost all of the mode energy
outside the light cone [67, 68]. The first experimental demonstration of this design
approach produced Q values of 1.3 × 104 [69]. An alternative approach to reducing
radiation losses was proposed by Johnson et al. whereby the cavity is modified to
1.7. DEFECT STATES IN PHCS: CAVITIES AND WAVEGUIDES 23
achieve cancellation of specific radiated field components in the far-field [70].
There are many possible ways to optimise the geometry of a cavity, and it is not
immediately obvious which one should be taken. In Ref [69], the defect was formed by
reducing the size of many holes in a large defect region, with the holes near the centre
of the region being changed more than the ones closer to the outside. A number of
other ‘graded lattice’ designs have been proposed with predicted Q values of more than
105 [71, 72]. The feature of all these designs is that there are no abrupt changes in the
lattice structure at the edges of the defect. Using a simple 1D model, Akahane et al.
[54] showed that this feature is the key to high-Q cavity design; the modal field should
vary smoothly across the edges of the defect. By shifting the position of two holes, a
Q of 4.5× 104 was measured in a cavity formed by the removal of only three holes in a
triangular lattice. The small cavity size resulted inQ/V = 1.2×105/λ3, at least an order
of magnitude higher than any other optical resonator at the time ∗. Further refinements
in the cavity design have more than doubled these values [74]. The same technique
was also used to design a different cavity geometry with an experimentally measured
Q = 6 × 105 [75]. In the last three examples, light was coupled into the cavity from
a PhC waveguide separated by a number of lattice periods - a simple, yet convincing
demonstration of the possibilities for integrated optical circuitry. Bistable switching for
powers as low as 40µW was demonstrated recently using a PhC cavity coupled between
two waveguides that exhibited a linear transmission resonance of Q = 3.3× 104 [76].
Although the study of high-Q cavities is not a major theme of this thesis, resonance
effects are considered in detail and some of the design rules used to reduce radiation
losses in cavities could be applied to the resonant coupling devices presented here.
1.7.2 Waveguides
The concept of creating a linear defect in a PhC slab for waveguiding was already
introduced at the start of Section 1.7, but we now consider the properties in more
detail. In many ways the design of efficient waveguides is considerably easier than the
design of resonant cavities because the confinement of light by the PhC is required
only in one direction within the slab. There are two conditions for propagating modes
to exist in any waveguide, whether metal, dielectric or PhC: first, the walls must be
reflective, and second a phase condition must be satisfied. The latter requirement
results in a discrete spectrum of guided modes. Hollow metal waveguides are used for
guiding microwaves as metals have relatively low absorption at those frequencies, and
they reflect light for all incident angles. Dielectric waveguides, on the other hand, rely
on total internal reflection, so they only guide modes that satisfy the TIR condition
on all the walls. PhC slab waveguides use the in-plane photonic bandgap for lateral
mode confinement and TIR for vertical confinement. Since the PhC walls reflect light
for all incident angles, PhC waveguides have many similarities to metal waveguides and
∗Higher values have since been reported in toroid microcavities [73].
24 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
some of the concepts of microwave circuit design such as impedance matching have been
applied to PhCs [77,78].
There is justifiably some debate as to whether photonic crystal waveguides provide
any benefit over high-index contrast dielectric waveguides, such as those based on the
well-established SOI technology [79]. For simply transferring light from one part of an
optical circuit to another, SOI waveguides not only tend to have lower losses, but can
also be positioned anywhere, rather than having to conform to a PhC lattice. PhC
waveguides, on the other hand, can be directly integrated with other PhC structures
such as cavities to produce functional devices. In addition, light guided by a photonic
bandgap can exhibit very different behaviour to TIR guided light, including very strong
dispersion and strong reflections that we exploit in Chapters 3 and 4.
1.7.3 Linear waveguide modes
A linear defect is typically created in a PhC by removing or changing nearest-neighbour
cylinders along one of the symmetry directions of the lattice. In a square lattice this
usually corresponds to the Γ−X direction, while in a triangular lattice the waveguide
lies along Γ−K, as illustrated in Fig. 1.8(b) and (c). Since the waveguide only breaks
the PhC symmetry in one direction, the structure remains periodic along the waveguide
axis. Thus the modes are characterised by a wavevector k directed along the guide and
a corresponding frequency. As for any 1D periodic lattice, the waveguide dispersion
curves can be represented on a band-diagram showing the mode frequency as a function
of the Bloch vector k0 over the 1D Brillouin zone defined by −π/d ≤ k0 ≤ π/d, where
d is the periodicity in the waveguide direction. For the common waveguide directions
described above for the square and triangular lattices, the period along the waveguide
is just the lattice period d = d, but for waveguides in other directions, generally d 6= d.
The bands of the PhC surrounding the waveguide can be represented on the same
plot by projecting the 2D band diagram onto the k−axis of the waveguide. In a square
lattice for example, the Γ−X band structure is plotted as before, but the modes lying
in the rest of the Brillouin zone are plotted as a function of their kx component only.
Such a combined waveguide dispersion and projected PhC band diagram clearly shows
where the waveguide modes lie in relation to the bandgap and the Bloch modes of the
PhC. For 3D calculations, the lightline can be projected onto the waveguide axis in the
same way. More details on the construction of these diagrams are given by Johnson et
al. [38]. As an example, Fig. 1.8 shows the dispersion curves and modal field profiles
for a waveguide formed by leaving out a single line of holes in the Γ − K direction
of the hole-type PhC considered in Section 1.5.3. For TE polarisation there are two
modes: one with even symmetry, extending down from the top of the bandgap to the
cutoff frequency d/λ = 0.221, and one with odd symmetry that lies in the frequency
range 0.245 ≤ d/λ ≤ 0.256. The properties of these modes are discussed in more detail
in Section 1.7.4.
Waveguide modes lying below the lightline are lossless in an ideal PhC slab. In prac-
1.7. DEFECT STATES IN PHCS: CAVITIES AND WAVEGUIDES 25
0.2
0.22
0.24
0.26
0.28
0.3
0.32
0.180 0.50.25
k d( )2 /p
d/l
PhC bands
PhC bands(a)
(b)
(c)
Figure 1.8: (a) Dispersion curves for a waveguide formed in the triangular PhC lattice of
Section 1.5.3 by removing a line of cylinders along the Γ−K direction. Two waveguide
modes exist in the bandgap: an even symmetry mode (solid curve) and an odd symmetry
mode (dashed curve). Hz field distributions of the odd and even modes are shown in
(b) and (c) respectively for k = π/d, clearly illustrating the difference in symmetry.
tice however, propagation losses are significant. The first experimental demonstration
of PhC waveguides was performed using millimetre waves in a lattice of alumina rods
with a period of 1.27 mm, in which propagation losses of 0.3 dB/cm were measured [80].
The lattice in this case was essentially 2D, since the rods were considerably longer than
the wavelength, and there was no guidance in the vertical direction. Waveguiding at
telecommunications wavelengths in PhC slabs was first observed in 1999 but transmis-
sion spectra and loss measurements were not reported until 2001 [81, 82]. Much effort
has gone into reducing loss since then with recent reports of 15 dB/cm propagation loss
in SiO2 clad structures [83] and < 5 dB/cm in air-clad Si slabs [84,85]. These losses are
within an order of magnitude of high-index-contrast SOI waveguides, for which losses
of 0.8 dB/cm have been reported [86].
Apart from the initial millimetre-wave experiments, the great majority of wave-
guiding demonstrations in PhC slabs have involved hole-type PhCs. Although a few
experimental demonstrations of waveguiding in rod-type PhCs have been reported, the
propagation losses are at least an order of magnitude greater the lowest-loss waveguides
in hole-type structures [47,48].
Below the lightline, the main contribution to waveguide loss is thought to be struc-
tural disorder. The length scales of this disorder can range from several tens of nm,
in the case of hole position and radius, down to only a few nm, in the case of surface
roughness. Positional and size disorder has been shown to reduce the size of the band-
gap in bulk PhC, and simultaneously decrease the transmission bandwidth and total
transmission of PhC waveguides [87, 88]. Calculations have also shown that the first
26 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
row of cylinders on the waveguide edges is the most important for minimizing in-plane
scattering losses; as long as these cylinders are in the correct places, disorder in the
rest of the lattice has a minimal effect on the waveguide transmission properties [89].
Surface roughness has only recently been included rigorously in scattering loss calcu-
lations for PhC waveguides using a Green function theory. Very good agreement with
experimental loss measurements was demonstrated, suggesting that the dominant loss
mechanisms were identified [84].
The prospect of low-loss bends in PhC waveguides has perhaps attracted more
attention than any other aspect of PhC applications [90]. High-efficiency bends have
been demonstrated in many different photonic crystal geometries [48, 80, 82, 91]. Since
the PhC waveguide walls reflect light for all incident angles, sharp bends do not suffer
from the same bend losses that occur in dielectric waveguides [92], but this also means
that without careful design, they can reflect a considerable amount of light as well. In
contrast, if a dielectric waveguide is bent too sharply, the TIR condition is violated
and light radiates into the cladding. Although this is major drawback of low-index-
contrast dielectric waveguides originally proposed for integrated optics applications,
the stronger index confinement in SOI waveguides allows low-loss curved waveguide
bends of less than 5µm in radius [93]. Alternative methods have also been proposed to
achieve even sharper bends [94].
1.7.4 Dispersion
One aspect where PhC waveguides differ significantly from TIR-based waveguides is
in their dispersion properties. Large and variable dispersive effects are a characteris-
tic feature of light propagation in all periodic media, originating from the wavelength
dependence of coherent scattering in these materials. Dispersion control is particularly
important when transmitting short pulses through waveguides because of the corre-
spondingly broad frequency spectrum. If the dispersion properties change significantly
over the spectral range of the pulse it can lead to distortion or, in the worst case de-
struction of the signal. This effect can also be used to great advantage, however if the
properties can be controlled. By tuning the dispersion appropriately, dispersive effects
of other components in the system can be compensated for, or alternatively pulses can
be slowed down, compressed or stretched. Nonlinear interaction lengths can also be
significantly reduced by slowing the light down.
The dispersion properties of a waveguide mode are characterised in part by the
group velocity, vg = dω/dk0 and the group velocity dispersion GVD = dvg/dk0; the
slope and the curvature of the mode dispersion curve, respectively. Figure 1.8(a) shows
a typical pair of dispersion curves for a waveguide formed in a triangular hole-type
lattice. Consider first the even mode represented by the solid curve. The curve is
almost straight over most of its frequency range, indicating a constant group velocity
and a near-zero GVD. As the mode approaches cutoff however, it flattens out and
vg → 0 at the edge of the Brillouin zone, as expected for the modes of most periodic
1.7. DEFECT STATES IN PHCS: CAVITIES AND WAVEGUIDES 27
structures. The sharp change in the slope just before cutoff also results in a large
GVD at these frequencies, a property first measured experimentally with an indirect
interferometric technique [95]. In contrast, the odd mode, represented by the dashed
curve in Fig. 1.8(a) has zero vg in the middle of the Brillouin zone, as well as at the
edges. Group velocities as low as 0.02c, where c is the speed of light in a vacuum, have
been demonstrated experimentally via group delay measurements [83]. Even smaller
values of vg < 0.001c have been inferred from direct time-resolved observations of pulse
propagation in PhC waveguides, although these have not been confirmed rigorously
using other measurement techniques [96].
One of the advantages of the geometrical complexity of PhC waveguides is that
there are a vast number of possibilities for tailoring the dispersion properties to suit the
desired application. Some of these include changing the waveguide width [97], changing
the radius of some of the cylinders [98, 99], or even placing additional cylinders along
the waveguide to create a coupled cavity waveguide (CCW) [100, 101]. Other, more
complicated coupled resonator designs have been proposed for stopping light altogether,
raising the possibility of optical data storage or buffering [102].
1.7.5 Integrated waveguide and cavity devices
Waveguides and cavities form the building blocks for a vast range of optical components
and devices. In this section we consider two classes of device that have been successfully
demonstrated in photonic crystals by integrating waveguides and resonant cavities.
Waveguide couplers
Directional couplers perform various functions in conventional optical circuits, includ-
ing power splitting and combining and wavelength-selective coupling. Switching and
wavelength tuning can also be achieved by incorporating nonlinear or other tunable
elements into the structure (See, for example, Ref. [103], Chapters 17, 18).
In its most basic form a coupler consists of two parallel identical waveguides running
together over some distance L. When placed closely together, the modes in each guide
interact to produce supermodes — modes of the coupled waveguide system. If each
waveguide in isolation supports only a single mode, then the coupled system generally
splits these into two supermodes, one at a lower frequency and one at a higher frequency
than the original. For coupled dielectric waveguides, the fundamental, lowest frequency
mode is always symmetric (even) with respect to the symmetry axis of the waveguide
pair, while the second mode is antisymmetric (odd), with a nodal line running along
the axis. Two identical coupled PhC waveguides can also support a symmetric and an
antisymmetric mode, but the lowest frequency mode is not always the symmetric one.
This issue is studied and explained in more detail in Chapter 3. Although this property
may introduce some additional design considerations for PhC waveguide couplers, the
coupling mechanism is unchanged as it depends only on the magnitude of the mode
28 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
splitting, and not on the order of the two modes. The mode splitting is characterised by
the frequency-dependent parameter ∆β = |β+−β−|/2, where the β± are the propagation
constants (or wavevectors) of the even and odd supermodes respectively.
If monochromatic light of unit power is injected into one of the waveguides and
allowed to propagate, it couples into the other waveguide. After a propagation distance
L, the power remaining in the input guide is given by cos2(∆βL), while the power
transferred to the second guide is sin2(∆βL). The coupling length is the distance over
which a π phase difference accumulates between the two modes
Lc =π
2∆β=
π
|β+ − β−|, (1.8)
corresponding to the propagation distance required for all of the light in one guide to
be transferred into the other. Couplers of length Lc/2 transfer 50% of the light into the
other guide and are widely used as 3 dB power splitters and combiners. The shortest
coupling lengths achievable in dielectric waveguides are several tens of microns, and
even this requires novel coupler designs [93, 104]. Coupled PhC waveguides however,
exhibit very strong mode splitting, and coupling lengths of only a few wavelengths have
been predicted [105,106]. Various PhC directional coupler designs have been proposed
and demonstrated experimentally [107, 108]. Modified coupler designs incorporating a
short waveguide cavity between the two guides have also been suggested to reduce the
coupling length further and improve coupling efficiency [109].
The sensitivity of PhC waveguide dispersion properties to small variations in ge-
ometry or refractive index makes them ideal for coupler-based switching applications.
Optical switches based on both electro-optic and nonlinear refractive index modula-
tion have been proposed as methods of tuning the coupling strength to allow or block
coupling between the guides [110, 111]. A basic add/drop filter can also be formed by
arranging multiple couplers along a single waveguide, each designed to couple out a
single wavelength channel of a WDM signal [112]. One limitation of using a standard
directional coupler response for wavelength switching or filtering is that the transmis-
sion spectrum is almost sinusoidal. To achieve a transmission band narrow enough to
filter a single WDM or DWDM channel, the coupler must be many coupling lengths
long, and even then there are transmission peaks close by on either side of the desired
frequency. Although it has been suggested that this property could be used for a chan-
nel interleaving device, it is by no means ideal for add/drop functions. Some alternative
designs for improved add/drop performance will now be considered.
Channel add/drop filters
The ability to add or remove one or more wavelength channels from a multiplexed signal
is an essential function in optical networks, and as such, all-optical add/drop filters are
one of the components required for future all-optical processing. Almost all proposed
PhC add/drop filters utilise a resonant cavity coupled to one or more waveguides to
improve the wavelength selectivity of the device. Light propagating along a waveguide
1.7. DEFECT STATES IN PHCS: CAVITIES AND WAVEGUIDES 29
will couple to a nearby cavity if the frequency matches, or is close to the resonant
frequency of the cavity. Once in the cavity, the light can be dropped into a nearby
waveguide, or even radiated out of the plane for monitoring purposes or coupling into
an optical fibre. Schematics of these approaches are shown in Fig. 1.9.
forwarddrop port
backwarddrop port
transmissioninput port
l1
l1, 2 3l l, l2,l3
resonantcavityresonantcavity
drop port
l1, 2 3l l,
l1
l2,l3
resonantcavity
transmissioninput port
transmissioninput port
l1 radiativedrop port(out-of-plane)
l2,l3
resonantcavity
(a)
(b)
(c)
l1, 2 3l l,
Figure 1.9: Schematics showing three different add/drop coupler geometries. (a) and
(b) Two alternative methods for in-plane cavity coupled waveguides. (c) Out-of-plane
channel drop filter.
A geometry for in-plane channel dropping from one waveguide to a parallel wave-
guide as shown in Fig. 1.9(a) was first analysed by Fan et al., [113] who showed that two
resonant states with different symmetries are required for complete power transfer, and
furthermore, that these states must be degenerate. These conditions are necessary to
ensure light is not coupled into the backward drop port or reflected back into the input
waveguide. Various PhC based designs displaying these properties have been proposed
and demonstrated experimentally either using two distinct cavities or a single cavity
supporting two modes [83, 114, 115]. The use of multiple coupled cavities can also be
used to improve the spectral shape of the coupling response. A single cavity resonance
typically exhibits a Lorentzian response, whereas multiple coupled cavities can be de-
30 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
signed to have a flat-topped response with steeper sides, thereby reducing overlap with
nearby channels while improving the transmission within the desired channel [74,116].
The filter design shown in Fig. 1.9(b) works in a similar fashion to that in (a), although
it is not clear from the numerical results whether mode symmetry plays the same role
in determining the efficiency [83,117]. One clear advantage, however is the ability to in-
clude additional drop ports along the same input waveguide, thereby allowing multiple
channels to be dropped.
An alternative to coupling light into another waveguide is to radiate it out of the
plane, as illustrated in Fig. 1.9(c). The main drawback of this technique is that the
maximum theoretical coupling efficiency is only 50%, with the remainder coupling back
into the input waveguide in both the forward and backward directions [118, 119]. Un-
like the cavities required for designs (a) and (b), where radiation losses have to be
minimised, the cavities in this case must be intentionally designed to radiate, while still
exhibiting a sharp resonance. Experimental demonstration of a multiple wavelength
filter incorporating seven different cavities was recently reported with drop efficiencies
of up to 40% [120]. Thermo-optic tuning of a single cavity with a blue laser has also
been used to achieve dynamic control of the coupling frequency, with shifts of nine
times the resonance width demonstrated [121].
1.8 In-band applications
In this section we consider the possible applications of uniform PhCs operating at
frequencies where propagating states exist. As the band diagrams in Fig. 1.5 illustrate,
the Bloch modes can have complicated dispersion profiles, and these lead to a range
of unusual propagation effects, some of which are discussed in this section. Here the
behaviour is illustrated by considering the transmission of light from a homogeneous
dielectric into a PhC, or vice-versa, but interfaces between two PhCs or between a
waveguide and a PhC have qualitatively similar properties.
1.8.1 Coupling to Bloch modes
The coupling of plane waves from a dielectric to the Bloch modes of a PhC can be
understood in terms of momentum conservation or as a generalised version of Snell’s
law. Figure 1.10(b) illustrates a graphical construction of Snell’s law for an interface
between two dielectrics labelled 1 and 2, as shown in part (a) of the figure. A plane wave
of frequency ω propagating in medium 1 is characterised in k-space by a wavevector of
magnitude k1 = n1ω/c pointing in the direction of propagation. The set of all possible
wavevectors at this frequency can thus be represented by a circle of radius k1, indicated
by the solid green circle in the diagram. Similarly, the set of plane waves in medium 2
lie on a circle of radius k2 = n2ω/c, indicated by the dashed circle. These circles are
equifrequency contours (EFCs) of the conical band surfaces of media 1 and 2. If one
1.8. IN-BAND APPLICATIONS 31
or both of the media is anisotropic, the EFCs become elliptical, rather than circular,
since n depends on the propagation direction [122].
A refraction problem can be constructed for a given incident angle θi by first drawing
the incident wavevector into the previous diagram, indicated by the solid green arrow.
Momentum conservation requires the wavevector component parallel to the interface,
k‖ = k1 sin(θi) to be the same in both media, as represented by the dashed-dotted
line perpendicular to the interface. The wavevector of the refracted wave is given by
the intersection of this line with the EFC of medium 2, indicated by the dashed blue
arrow. Since the light is being transmitted through the interface, it is necessary to
choose the intersection corresponding to energy flowing away from the interface. The
energy flow is given by the group velocity vg = ∇k0ω(k0), which is a vector normal to
the EFC in the direction of increasing frequency. For an isotropic dielectric, the group
and phase velocities are parallel, and hence vg ‖ kpc. Snell’s law can be found using
simple geometry to show n1 sin(θi) = n2 sin(θ2).
A similar method is followed for an interface between a dielectric and PhC, but there
are three significant differences. First, it is clear from Fig. 1.5(b) that the EFCs of PhC
Bloch modes are not circular, so the refraction properties are in general anisotropic.
Second, the phase and group velocities are not necessarily in the same direction. Third,
the momentum conservation rule must be generalised to take account of the periodicity
in the direction along the interface, dint. Hence, the construction lines perpendicular
to the interface representing conservation of k‖ must be repeated for k‖ + m(2π/dint),
where 2π/dint is a reciprocal lattice vector parallel to the interface andm = 0,±1,±2, . . .
[123,124]. In the example shown in Fig. 1.10(d), only the first Brillouin zone is drawn,
but in general the surrounding zones must also be considered.
For a dielectric, the circular EFCs have a radius proportional to the frequency, so kpc
and vg are both directed radially outwards in the same direction. In a PhC, however, the
situation can be very different, as illustrated by the example in Fig. 1.10(d) which shows
an EFC for a band with a maximum at the Γ-point. In this case, the EFCs are almost
hexagonal and their size decreases with increasing frequency so that vg is directed
inwards from the contour. To ensure that energy flows away from the interface the
upper intersection point is chosen, resulting in the kpc and vg indicated. θpc is given by
the direction of vg. Coupling to PhCs is not always as simple as the example given here.
Depending on the shape of the band surfaces, it is sometimes possible for light to couple
into more than one part of a band, or into more than one band simultaneously, although
these situations are not desirable for most practical applications. Similarly, coupling
into higher-order Brillouin zones can sometimes occur for non-normal incidence, or
when the interface is not aligned to a symmetry axis of the crystal [125]
Although this construction method provides a simple and intuitive way of under-
standing where the light can go, like Snell’s law, it does not provide information about
transmission and reflection amplitudes at the interface. For dielectric interfaces, these
are characterised by the Fresnel coefficients [122]. In Chapter 2 we introduce a gen-
eralised form of Fresnel coefficient that characterises transmission between dielectric
32 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
k1
k2
k//
qi
q2
kx
ky
n2
n1
qi
q2
x
y
qi
n1
qpc
kpc
k1
vg
kx
qpc
qi
(b)(a)
(c) (d)
Figure 1.10: Diagrams of refraction at an interface between (a) two homogeneous
isotropic dielectrics and (c) a dielectric and a PhC. (a), (c) Ray diagram of incident,
reflected and refracted light for incidence from above. (b), (d): Construction diagrams
for refraction in k-space showing the incident wavevectors k1, refracted wavevectors k2
and kpc, and the group velocity vectors of the refracted light vg. Solid green circles in
(b) and (d) are the equifrequency contours of the medium 1 at the incident frequency.
Dashed blue curves are the EFCs of the dielectric medium 2 and the PhC.
and PhC or between two different PhCs. The issue of improving transmission from a
dielectric to a uniform PhC is considered in detail in Chapter 5.
1.8.2 p and q parameters
From Section 1.8.1 it is clear that θpc has a strong dependence on the local shape of
the EFC where it intersects the k‖ conservation line in the construction diagram. Thus,
changing the incident angle or the frequency can result in large variations of θpc. These
two effects are characterised by the parameters
p ≡ ∂θpc
∂θi
∣∣∣∣ω
, q ≡ ∂θpc
∂ω
∣∣∣∣θi
, (1.9)
1.8. IN-BAND APPLICATIONS 33
which were first introduced by Baba et al. in the context of PhC superprisms [126].
In Sections 1.8.4 and 1.8.5 we consider two examples of unusual propagation behaviour
in uniform PhCs that require careful control of p and q. First, however, we consider
a more general property exhibited when light is coupled into a PhC, namely negative
refraction.
1.8.3 Negative refraction
Theoretical studies predict that materials with both a negative permittivity ε and per-
meability µ should exhibit a range of unusual properties including a negative refractive
index and anti-parallel group and phase velocities [127]. Furthermore, a material with
n = −1 has been shown to act as a perfect lens, capable of sub-wavelength resolu-
tion [128]. In contrast, the resolution limit of conventional optics in the far-field is
equal to the wavelength, regardless of the size and precision of the lens. Although
no naturally-occurring materials are known to exhibit negative refractive indices, com-
posite “meta-materials” with these properties have been proposed and demonstrated
experimentally for microwave frequencies [129].
Photonic crystals have been known for some time to exhibit unusual refraction
properties, including ultra-refraction, where a beam incident from air is refracted away
from the normal, implying a refractive index of n < 1 [130, 131]. Negative refraction
in PhCs, where the refracted and incident beams lie on the same side of the normal
was first observed experimentally in a 3D PhC [132] and since then has been studied in
detail in both theory and experiment [123,133–136]. Negative refraction is exhibited in
PhCs wherever the local curvature of the EFC is such that vg is directed away from the
normal on the same side as the incident beam. Moreover, if the EFCs are approximately
circular, and vg is directed radially inwards, then the PhC displays negative refraction
for all incident angles. For this situation, the concept of an effective refractive index
based on the radius of the EFC has been proposed [123,133].
Although PhCs are not true negative refractive index materials, they can still exhibit
some similar novel properties due to their negative refractive behaviour. A number
of theoretical studies have predicted that a finite slab of PhC that exhibits negative
refraction for all incident angles can focus an image to sub-wavelength resolution [137,
138]. This effect has been demonstrated experimentally with microwaves [139,140] but
has not yet been reported at telecommunications wavelengths.
1.8.4 Auto-collimation
Auto-collimation, or self-guiding, in photonic crystals allows a narrow beam to prop-
agate without any significant broadening or change in the beam profile. This effect
was first observed experimentally in a 3D PhC [141], where it was shown to occur at
inflection points on EFCs. A subsequent theoretical analysis provided a framework for
the design of auto-collimating 2D PhCs, and also demonstrated that PhCs designed for
34 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
collimation properties do not need to exhibit bandgaps [142]. This is an advantage of
many in-band applications: since they rely on the shape of the bands, rather than a
bandgap, many of the properties occur for lower refractive index contrasts, and hence a
larger range of materials become available for fabrication. Auto-collimation of an inci-
dent beam was first demonstrated experimentally in a PhC slab for light incident from
a waveguide. Relatively low-loss collimated beams were observed propagating through
the crystal for a range of angles [143].
Collimation occurs when an incident beam is coupled to a Bloch mode in a region
where the EFC is locally straight. Quantitatively, this can be expressed in terms of the
p-parameter (1.9), which indicates the sensitivity of θpc to variations in θi at a fixed
frequency. An incident Gaussian beam is not represented by a single wavevector, but
can be considered as a superposition of plane waves with a small angular spread centred
on the beam direction. Thus, an incident Gaussian beam in the construction diagram
of Fig 1.10 becomes a narrow cone of wavevectors that couple into a region of the EFC,
rather than a single point. Figure 1.11(a) illustrates the result of coupling to a region of
the EFC with strong curvature (large p); the plane wave components are refracted over
a range of angles ∆θpc and hence the beam broadens. In extreme cases the beam can
even be split into multiple beams [125]. Therefore, collimation requires a locally flat
region of the EFC where p ≈ 0, as shown in Fig. 1.11(b) [126]. A recent analytic study
of the p and q parameters has shown that every PhC band has regions where p = 0, but
in some cases this occurs over only a small range of incident angles [144]. If an EFC
has long straight sides, rather than just local points of inflection, then collimation is
possible for a range of incident angles. Square lattice PhCs exhibiting collimation for
incident angles in the range −54◦ ≤ θi ≤ 54◦ have been calculated [142].
Since the collimated beams do not require a waveguide or even a bandgap for lateral
confinement auto-collimation has been proposed as a novel method to achieve compact,
reconfigurable photonic circuits. Multiple beams can, in principle, pass through each
other without any resulting cross-talk [145, 146] and various circuit components such
as beam splitters [147, 148] and mirrors have been demonstrated experimentally [146].
Circuits based on auto-collimation would also be more tolerant to lateral coupling mis-
alignment than PhC waveguides which are typically only one or two lattice periods
wide. In Chapter 5 we consider the issue of coupling through dielectric–PhC inter-
faces, and present examples of high-efficiency coupling into a PhC designed to exhibit
auto-collimation for a range of incident angles.
1.8.5 Superprism effects
Superprism effects, where the angle of a transmitted beam is depends strongly on the
incident angle or frequency, have been proposed as a method of achieving ultra-compact
WDM multiplexing and demultiplexing in photonic circuits. Superprism behaviour
was first reported in 3D PhCs in a series of papers by Kosaka et al., where large beam
deflections were demonstrated for changes in both the incident angle and frequency [132,
1.8. IN-BAND APPLICATIONS 35
k1
vg
kx
qi
Dqi
Dqpc
vg
kx
qi
k1
Dqi
Dqpc~0
ky
(a) (b)
ky
Figure 1.11: Refraction construction diagrams similar to Fig. 1.10(d), but for an incident
beam of finite width, and hence a spread of incident angles. The incident wavevector
k1 and group velocity vg of the refracted beam are shown. (a) Example when |p| � 1
showing the spreading of the incident beam by the sharply curved EFC. (b) Example
when the EFC is straight, and hence |p| ≈ 0, resulting in collimation of the refracted
beam.
149]. Wavelength separations of 1 nm were shown to produce a 3◦ deflection - around
500 times that obtained in conventional prisms. Further, the authors suggested that
a PhC only 190µm long would be sufficient to separate a WDM signal into individual
channels [149].
Although studies of PhC prisms were reported earlier than the work described above,
the parameters were chosen so that the EFCs were round. The prisms therefore dis-
played refraction properties similar to an isotropic medium, albeit with larger dispersion
and a corresponding moderate improvement in the wavelength resolution [150]. The
difference between this early study and those that followed was the use of non-circular
EFCs to enhance the sensitivity significantly for a small range of incident angles and
frequencies. The large deflections observed in Ref [149] were attributed to a sharply
curved section of the EFC close to the region being accessed by the incident beam.
Small changes in the input conditions moved the outgoing wavevector along the curve,
resulting in rapid variation in the group velocity direction, and hence the angle of the
transmitted beam. Similar results have been demonstrated in 2D PhC slabs, with
angular resolutions of up to 1◦/nm at telecommunications frequencies [151,152].
In order for a superprism to resolve two closely spaced wavelengths, the angular
separation must be converted into a spatial separation sufficient to resolve the two
beams. The required separation therefore depends on the beam width; wider beams
36 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
require greater separation. Unfortunately the sharp curves in the EFC that produce
strong angular resolution also cause spreading of narrow beams, as was discussed in
Section 1.8.4, so a balance is required. To quantify these two competing effects, Baba
et al. [126] introduced the collimation and angular resolution parameters, p and q re-
spectively (1.9) and proposed that a realistic measure of resolution should be the ratio
r = q/p, not simply the value of q as suggested in earlier superprism discussions [149].
This alternative measure led to the conclusion that PhCs of more than 1 cm2 would be
required to resolve the channels of a DWDM network, a similar size to conventional
silica-based arrayed waveguide grating filters.
One reason for such long PhCs is the requirement that beams be separated spatially
before leaving the PhC. Such a condition is only necessary if the input and output
interfaces are parallel since in this case the separated wavelengths all leave the PhC in
the same direction. This can be seen by considering again the diagram in Fig 1.10(d).
The construction lines are unchanged for coupling out of the crystal, and thus the
transmitted beams are composed of the same wavevectors as the incident beam. A
more compact superprism is possible if the transmitted beams maintain their angular
separation after leaving the PhC. One way to achieve this is to angle the second interface
relative to the first, thereby changing the k‖ conservation condition for the output
coupling. This approach has been studied in two different geometries in Refs [153]
and [154].
1.9 Computational methods
In Section 1.5 the basic theory of band structures was discussed, and the eigenvalue
equations (1.1) and (1.2) that determine the Bloch modes were introduced. Although
the solutions and their significance were considered, we have not yet described how the
equation should be solved. Furthermore, aside from identifying the bandgaps, the band
structure of an infinite periodic lattice provides little information about transmission
and reflection or many other aspects of PhC operation. No single formulation provides
the computational tools necessary for an efficient and rigorous study of all photonic
crystal devices. Hence, many different theoretical methods have been developed, some
of which are discussed in this section.
The most appropriate numerical method for a given problem depends largely on the
structure being studied and the data that is required. Some methods provide versatility
at the expense of efficiency and/or accuracy, while others are highly efficient but can
only deal with a limited range of geometries or specific problems. Another consider-
ation is whether a method solves the problem in the time or frequency domain. In
time domain methods, the solution is calculated as a function of time, and can thus be
used to track the propagation of a pulse or observe other transient behaviour. This is
especially important when studying many nonlinear effects for which time dependent
material properties must be included explicitly. A frequency spectrum can be calcu-
1.9. COMPUTATIONAL METHODS 37
lated by launching a short pulse and taking a Fourier transform of the time response,
however long simulations are required to obtain the steady-state response of high-Qstructures. Frequency domain methods on the other hand, calculate the solution as a
function of frequency, which makes them more appropriate for calculating steady-state
frequency responses and band structures, and for studying systems with strongly dis-
persive materials. There is also considerable overlap between the application of these
two methods; while the frequency spectra can be obtained by taking a Fourier trans-
form of a time response, it is also possible to calculate the time response of a device if
the full frequency response is known.
Finite-Difference-Time-Domain (FDTD) methods are the most general and widely-
used tool for PhC calculations as they are flexible and can handle almost any geom-
etry [155]. The method involves the solution of Maxwell’s equations via numerical
integration on a discretised grid in both time and space. Since the grid must be fine
enough to resolve the smallest feature in the problem, dealing with extended PhC struc-
tures can be very demanding of computational time and memory. FDTD can be used
to simulate finite or periodic structures through the appropriate choice of absorbing or
periodic boundary conditions.
Another numerical propagation technique, this time in the frequency domain, is the
beam propagation method (BPM) which is more commonly used for studying optical
fibres and dielectric waveguides [156,157], but has also been applied to PhCs [158]. This
approach involves the launch of an initial field distribution u(x, y, z = 0) which is then
propagated numerically in the z direction. The method was initially implemented for
forward propagation only and is most efficient for structures that vary slowly along the
propagation direction. Adaptations of the original method have allowed bi-directional
propagation and improved the versatility, but BPM can not deal with the same scope
of problems as FDTD methods.
Also in the frequency domain, plane wave methods are the most appropriate for
band structure calculations and other problems that can be modelled as periodic sys-
tems [159]. In order to solve the eigenvalue form of Maxwell’s equation (1.2), the
refractive index profile and the electric and magnetic fields are first decomposed into
their respective Fourier components. The expansions are truncated to a finite number of
plane wave orders, and substituted into Eq. (1.2), resulting in a matrix eigenvalue equa-
tion that is solved numerically. Plane wave methods are not limited to band structure
calculations of simple lattices. For instance, waveguides and cavities can be simulated
by creating a periodic lattice of supercells, each one containing an array of cylinders
with the desired defect introduced. If the supercell is sufficiently large that defects in
nearby cells do not interact significantly, then the solutions to the eigenvalue problem
will approximate those for a single isolated defect. Waveguide dispersion curves are
commonly calculated in this way. Supercells are also used when calculating the 3D
band structure of a 2D PhC slab; the slab and surrounding cladding are replicated
periodically in the vertical direction with enough cladding between each slab to ensure
negligible coupling. Although this method is useful for calculating modes bound within
38 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
the slab, cladding modes can couple to nearby supercells, and are therefore not an
accurate approximation of the modes of an isolated slab. As with all supercell treat-
ments, the plane wave method is not able to calculate radiation losses, since every cell
is surrounded by others, and any light leaking out of one supercell passes into the ones
nearby and vice-versa.
A range of alternative frequency domain methods have also been developed, with
each one using a different set of orthogonal basis functions to represent the fields and
the index profile of the PhC. Although the plane wave method is versatile and can be
used to represent many different structures, it is not efficient for calculating complex or
tightly bound defect modes as this requires many thousands of plane waves. One way
to improve this is to expand the fields using functions that more closely represent the
modal profiles, and therefore require fewer terms in the expansion. One approach that
is most suitable for studying localised modes is based on Wannier functions which are
derived from the Bloch modes of the uniform PhC and therefore exhibit many of the
symmetry properties of the underlying lattice. The calculation of Wannier functions
from the Bloch modes is non-trivial, but when optimised, they provide a highly-efficient
basis set for PhC waveguide and defect calculations [160].
For PhCs consisting of circular cylinders, one of the most efficient modal techniques
is the multipole method [161–163]. In this case, the basis functions are cylindrical
harmonics centred on each of the cylinders, thereby exploiting the circular geometry
and allowing the boundary conditions to be expressed analytically. For this reason,
the multipole method is accurate and efficient when studying structures with very high
index contrasts or absorbing materials. Multipole methods can be applied to either
finite clusters of cylinders or periodic systems, with the former allowing radiation losses
and edge effects to be investigated.
Transfer matrix methods refer to a large class of techniques in which the structure
to be modelled is considered as a stack of individual layers, where transmission and
reflection from each layer can be characterised by scattering matrices. Transmission
through the whole stack is calculated by multiplying the scattering matrices together in
a recursive manner. Thus, if there are multiple identical layers, the scattering matrices
need only be calculated once, and can then be reused every time the layer appears
in the stack. Transfer matrix methods are a common technique for modelling thin-
film stacks, where the layer-by-layer approach is an obvious choice [164], but they can
also be applied to PhCs, which can often be treated as a periodic stack of diffraction
gratings [165, 166]. Calculating the band structure of a uniform PhC thus requires
only the knowledge of the scattering properties of a single layer. Various numerical
methods can be used to calculate the scattering matrices, including differential Fourier
methods [167,168] and finite-element methods [165].
The computational method chosen for most of the numerical calculations in this the-
sis is a recently developed 2D Bloch mode scattering matrix method [169, 170], which
is highly efficient for studying 2D PhC structures consisting of distinct segments. The
formulation is based on transfer matrix techniques where the fields are expanded in
1.10. DISCUSSION AND THESIS OUTLINE 39
terms of the natural Bloch mode basis of the PhCs. As we discuss in Chapter 2, the
significant advantage and novelty of this approach is the derivation of Bloch mode scat-
tering matrices which play the role of generalised Fresnel coefficients to describe the
reflection and scattering of light at a PhC interface. This allows many complicated
PhC structures to be treated using methods developed for thin-film optics, and also
provides a systematic approach for deriving semianalytic models of many structures.
The numerical implementation of the Bloch mode method used here also takes advan-
tage of the multipole method for grating calculations, and as a result gains some of
the additional benefits of this highly efficient and accurate technique for dealing with
circular cylinders.
1.10 Discussion and thesis outline
We have reviewed in this chapter some of the unique properties of photonic crystals and
the opportunities they present for the design of efficient, compact devices for a range of
photonics applications. As the most recent experimental results demonstrate, photonic
crystal research is entering a new phase. Fabrication techniques can now provide the
quality and control of structural detail necessary for optimisation and fine-tuning of
device designs, and as a consequence, modelling and simulation of new designs will
be more relevant than ever. Given the inherently complex nature of photonic crystal
structures, there is little doubt that they will remain more difficult to fabricate than
alternative integrated photonics technologies such as high-index SOI based components.
Therefore, to compete seriously with more conventional technologies, PhC devices must
either provide functions that are not easy to achieve using other methods, or perform
the same functions in a cheaper, more efficient and compact fashion that outweighs the
additional fabrication effort. Therefore, the ultimate goal is to demonstrate compact,
integrated photonic devices that satisfy both of these criteria. In the following chapters
we present results and techniques that may contribute towards the future realisation of
this goal.
Chapter 2 provides an overview of the Bloch mode scattering matrix method which
is the main computational tool that we use in later chapters. This method is not only
an efficient and accurate 2D numerical tool, but can also be used to derive semianalytic
treatments for many structures. We describe first the general formulation of the method,
then discuss several issues regarding the application of the method to specific structures
considered in this thesis.
In Chapter 3 we introduce a new class of device in which directional coupling and
resonant reflection effects are combined. The characteristics of these devices vary con-
siderably with the geometry, which allows us to demonstrate a high-Q resonant filter,
a wide bandwidth, high-transmission Y-junction and a coupled cavity delay line. The
devices are analysed using accurate 2D numerical results obtained from the Bloch mode
method, and semianalytic models that provide insight into the underlying physics.
40 CHAPTER 1. PHOTONIC CRYSTALS: PROPERTIES AND APPLICATIONS
The concepts of mode coupling and resonances are extended in Chapter 4, where
we study the effects of mode recirculation in Mach-Zehnder interferometers (MZIs)
and show that a perfectly reflecting cladding such as a photonic crystal can result in
manifestly different transmission characteristics compared with conventional, dielectric
waveguide-based MZIs. We derive a semianalytic model for this class of device and
demonstrate that rather than degrading the performance, recirculation can provide an
enhanced response. Numerical results are presented for both rod- and hole-type PhC
structures to verify these effects.
Chapter 5 concerns an entirely different type of resonance and mode coupling be-
haviour that we have identified in uniform rod-type PhCs. We show that scattering
resonances in the high-index rods provide a mechanism for light to be coupled effi-
ciently into the Bloch modes of the PhC for a wide range of angles. These results
may lead to more efficient designs for in-band applications such as superprisms and
auto-collimation devices. We demonstrate one such example by optimising a structure
to exhibit both auto collimation and very low reflectance for a wide range of incident
angles, and confirm the 2D results with rigorous 3D simulations.
The thesis concludes with a brief outlook on the future of photonic crystal devices
and a discussion of out-of-plane losses and other issues that must be considered in order
for the devices we have proposed to be demonstrated experimentally.
Chapter 2
Bloch mode scattering matrix
method
This chapter provides an overview of the Bloch mode scattering matrix method – the
main numerical tool used in this thesis. It is not the intention to provide a full derivation
of the formulation, but rather to present the details of the method that relate directly
to the results that follow in Chapters 3–5. The majority of this chapter is based on the
work of Botten et al. in Refs. [169] and [171], with the exception of Sections 2.4 and 2.5
in which we discuss a number of issues relating to the application of the Bloch mode
method to the specific structures of relevance to this thesis.
2.1 Introduction
While the calculation of 2D PhC band structures has become routine and can be per-
formed with a minimum of computational expense, studying propagation in more com-
plicated PhC structures remains a challenge. The main computational method used
throughout this thesis is the recently developed Bloch mode scattering matrix method
(BMM), which is not only an efficient 2D numerical tool but also provides a systematic
and intuitive approach for deriving semianalytic models of many structures. We use
both of these features here to gain insight into the underlying physics of resonances and
mode coupling behaviour in photonic crystal devices.
The computational efficiency of the method is a result of two key techniques that
are combined in the BMM. The first of these is the transfer matrix method, which has
been used for many years in the field of thin-film optics [2], and more recently has been
applied to PhC problems [165,166,172]. The second technique is the use of the natural,
Bloch mode basis to expand the fields.
To apply transfer matrix methods to a complex structure, it is necessary to break
the structure down into basic units – layers that can be characterised relatively easily
in terms of their transmission and reflection properties. In a conventional thin-film
stack the basic layer is typically a homogeneous dielectric material. The interfaces
41
42 CHAPTER 2. BLOCH MODE SCATTERING MATRIX METHOD
between layers are characterised by Fresnel transmission and reflection coefficients while
propagation within a layer is described by plane waves. When applied to a periodic
geometry such as a slab of PhC consisting of periodic gratings as shown in Fig. 2.1,
the fields are expressed in terms of the Bloch modes of the slab, and we can derive
generalised Fresnel coefficients (in this case matrices) that describe transmission and
reflection of these modes at the interfaces of the slab. The derivation of these Fresnel
matrices is the main conceptual advance of the method we describe here, as it allows
many relatively complicated PhC structures to be treated using the framework of thin-
film optics.
In order to represent accurately the fields in a photonic crystal device using a Bloch
mode expansion, it is necessary to include all of the modes, both propagating and
evanescent. Recall from Section 1.5.1 that the standard technique for solving Eq. (1.1)
is to calculate the eigenvalues (ω/c)2ε as a function of the real Bloch vector k0, which
leads to the band structure. The approach described here involves an alternative eigen-
value equation which we solve at a fixed frequency to yield the Bloch vectors. If the
frequency lies in a bandgap, then all of the Bloch wavevector solutions are complex,
corresponding to evanescent Bloch states that can be excited at an interface but which
decay exponentially within the PhC. For frequencies where propagating modes exist,
there are a finite number of solutions for which k0 is real, corresponding to these modes,
and an infinite number of solutions for which k0 is complex, again corresponding to
evanescent states.
Bloch modes can be expanded as a superposition of plane wave orders, where both
propagating and evanescent plane waves are required to represent the modes accurately.
Hence, propagating Bloch modes may have evanescent plane wave components and
evanescent Bloch modes may have propagating plane wave components. While the
evanescent Bloch states carry no energy through an infinite structure, they can couple
energy through thin PhC barriers via evanescent coupling, much like frustrated total
internal reflection in conventional optics. The evanescent states are important when
matching fields at an interface between two different PhCs or between a homogeneous
medium and a PhC. In practice, the field expansions are truncated to include only a
finite number of modes where the truncation order depends on the specific geometry and
the required accuracy. For many structures, the essential physics is described accurately
by the propagating modes alone, in which case the evanescent Bloch modes can be
dropped from the treatment, resulting in simple expressions involving only a few modes.
We provide numerical examples of this asymptotic approximation in Section 3.3.2.
A typical structure that we consider is illustrated in Fig. 2.1, consisting of multiple
segments, M1, M2, . . . , MN , each of which is formed by a stack of identical layers with
a transverse period of Dx. Media M1 and MN are semi-infinite in the vertical direction
and form the input and output domains of the problem. While the example shows all
segments containing photonic crystal material, they may also contain a homogeneous
medium such as a dielectric or free-space. Each individual layer is treated as a diffraction
grating with a period Dx corresponding to the size of the supercell. A uniform 2D PhC
2.1. INTRODUCTION 43
M1
M3
M2
M4
layer
stack of L2 layers
(n = 0)
(n = L2)
Dx
x
yy = y1
y = y2
Figure 2.1: Schematic of a photonic crystal structure formed by joining segments M1−MN , each consisting of identical grating layers (here N = 4). All layers must have the
same supercell period Dx, but the positions and properties of the cylinders within a
layer can all be different. Segments M2 −MN−1 contain Lm layers, segments M1 and
MN are semi-infinite. Layer interfaces within each segment are numbered n = 0 to
n = Lm.
structure is formed if the grating consists of a periodic array of cylinders with period d,
and Dx is an integer multiple of d. Functional elements such as waveguides are created
in the supercell by removing or modifying cylinders from the grating as shown. When
operated in the bandgap of the bulk photonic crystal, the supercell must be sufficiently
large to effectively isolate the elements in one cell from those in neighbouring cells. For
many structures we find that a separation of at least 10d is sufficient for coupling effects
to be neglected, but more may be required when modelling strongly resonant structures
such as those studied in Chapter 3. We discuss convergence with supercell size in more
detail in Section 2.4.
In Section 2.2 we describe the steps required to obtain the Bloch modes for a typical
segment in Fig. 2.1 from the scattering properties of a single grating layer. We then
proceed to couple segments together in Section 2.3, considering initially a single interface
between two semi-infinite segments which allows the derivation of generalised Bloch
mode Fresnel matrices. Calculation of the complete multi-segment structure follows
in direct analogy to the modelling of multi-layer thin-film stacks via recursion. In
Sections 2.4 and 2.5 we discuss several practical issues relating to the application of the
Bloch mode method to modelling PhC devices.
44 CHAPTER 2. BLOCH MODE SCATTERING MATRIX METHOD
2.2 From a grating to a PhC
In this section we outline the derivation of the Bloch modes of a typical PhC segment
in Fig. 2.1 formed from a stack of identical periodic grating layers. In the most general
case, each layer is offset from the previous one by δx in the horizontal direction and δyin vertical direction, as shown in Fig. 2.2. The choice of these offsets determines the
symmetry properties of the lattice as we discuss in Section 2.2.2.
2.2.1 Grating characterisation
Since each layer in a segment is a periodic grating with period Dx, both the layers and
the segments that they form are coupled together by plane wave diffraction orders s,
the directions θs of which are governed by the grating equation
k sin θs = αs = α0 +2πs
Dx
, s = 0,±1,±2, . . . (2.1)
where α0 = k0x is the component of the Bloch vector along the direction of periodicity
of the gratings, and k = 2π/λ is the wavenumber in the background medium. The
integers s in Eq. (2.1) run over a finite set of propagating orders, and an infinite set
of evanescent orders. Between each grating layer the background medium is taken
to be homogeneous, and so the fields can be represented in the plane wave basis by
an expansion over all plane wave orders. The response of the grating to an incident
field is characterised by the plane wave reflection and transmission scattering matrices
R, T and R′, T
′, with the two pairs corresponding to incidence from above and below,
respectively. For example, the element Rpq specifies the reflected amplitude in plane-
wave order p due to unit amplitude incidence from above in plane-wave order q. The
tilde notation is used here to differentiate between the plane-wave scattering matrices
for a single grating and those of a PhC segment, which we introduce in Section 2.3.
The grating scattering matrices can be calculated using a variety of techniques,
including integral equation methods, differential Fourier methods and finite-element
methods [165, 167, 168, 173–175]. The BMM implementation described here uses the
multipole method [176], which is one of the most efficient numerical techniques for
dealing with circular cylinders. The main disadvantage of the multipole method is that
individual grating layers cannot interpenetrate. Although this limits somewhat the
geometries that can be studied, in situations where it is valid, the multipole method
provides many significant advantages over other techniques. For example, the energy
conservation and reciprocity relations are embedded in the multipole formulation and
these features are inherited by the scattering matrices and the Bloch modes. This
issue has some important consequences regarding the verification of the method and
convergence of solutions, as we discuss further in Section 2.4.
2.2. FROM A GRATING TO A PHC 45
e2f1
+
f2
-
f1
-
f2
+
y
x
dx
dy
Figure 2.2: Geometry of a single grating layer in a periodic stack defined by the lattice
vector e2 = (δx, δy). The plane wave field coefficients above the grating are given by f−1and f+
1 , corresponding to upward and downward propagating components respectively.
Similarly, the fields below the grating are given by f−2 and f+2 .
2.2.2 Basic PhC lattices
Here we consider briefly the steps required to form a segment of square or triangular
PhC from a stack of identical grating layers with a supercell period Dx, containing
multiple cylinders. The cylinders in each supercell are typically arranged with a local
period d, and Dx is chosen to be an integral multiple of d so that dislocations are not
formed at the edges of the supercell.
The diagrams in Fig. 2.3 illustrate how a PhC segment is formed from a stack of
gratings where each successive grating is shifted by δx along the x-axis and δy along
the y-axis. Stacking gratings of local period d with δx = 0 and δy = d for example,
results in a square lattice oriented with the interface parallel to the Γ − X direction
as shown in Fig. 2.3(a). A straight waveguide is created simply by removing one or
more cylinders from each layer. As mentioned in Section 2.2.1, in order to use the
multipole method for calculating the grating scattering matrices, the layers must not
interpenetrate. Therefore we are restricted to cylinder radii r < δy/2. This is not a
problem for the square array in Fig. 2.3(a) since the gratings do not begin to penetrate
until the cylinders themselves touch at r = d/2, but as we see in the next examples,
the interpenetration can limit the study of other lattice types and orientations. We
note again that this is a limitation of the multipole grating formulation, and not a
fundamental limitation of the Bloch mode method.
The square lattice in (a) can also be oriented with the interface parallel to the Γ−Mdirection by starting with a grating layer of period d′ =
√2d, and stacking them with
δx = d′/2 =√
2d/2 and δy = d′/2 =√
2d/2 as shown in Fig. 2.3(b). In this case,
interpenetration of layers restricts the cylinder radius to r <√
2d/4. Observe that
the waveguide formed by removing a single cylinder from each layer is identical to the
waveguide in Fig. 2.3(a).
Figures. 2.3(c) and (d) show triangular lattices of period d oriented respectively with
interfaces parallel to the Γ−K and the Γ−M directions. The example in (c) is formed
46 CHAPTER 2. BLOCH MODE SCATTERING MATRIX METHOD
G
M
X
dG M
Xd
d’
G
KM
d
d
(a) (b)
(c) (d)
G
K
M
d’
d
dx
dy
Figure 2.3: Diagram showing four PhC lattices with period d formed by stacking in-
dividual grating layers of period d′ with a horizontal offset δx and vertical offset δy as
given in the text. The waveguides are formed by removing one cylinder from each layer.
(a) Square lattice oriented with the interface parallel to the Γ−X direction. (b) Square
lattice with interface parallel to Γ−M . (c) Triangular lattice with interface parallel to
Γ−K. (d) Triangular lattice with interface parallel to Γ−M .
from gratings of period d with δx = d/2 and δy =√
3d/2 and the example in (d) has
a grating period of d′ =√
3d, stacked with δx = d′/2 =√
3d/2 and δy =√
3d′/6 = d/2.
Hence the maximum cylinder radii to avoid interpenetration in these two cases are
respectively r =√
3d/4 and r = d/4.
The horizontal offset δx is included in the formulation by applying the translation
operator Q = diag[exp(iαsδx/2)] to the original grating scattering matrices [177]. This
translation operator adjusts the phase of each scattered plane wave order to account for
the relative shift between the upper and lower interfaces. For example, the plane-wave
transmission matrix T becomes QTQ, and the reflection matrix R becomes QRQ−1.
A similar translation operator can also be defined for the vertical offset between each
layer. In the following treatment, it is assumed that the plane-wave scattering matrices
have already been transformed in this way.
2.2.3 Bloch mode calculation
On the interface between two grating layers in a given segment, the plane wave field ex-
pansions are represented by vectors of plane wave amplitudes f− and f+, corresponding
respectively to the downward and upward propagating plane wave orders of Eq. (2.1).
Both propagating and evanescent orders must be included to represent the fields accu-
2.2. FROM A GRATING TO A PHC 47
rately. The fields f1 and f2 on either side of the grating in Fig. 2.2 are related via a
transfer matrix T which can be derived from the plane wave scattering matrices such
that
f2 = T f1, where fj =
[f−jf+j
], (2.2)
and
T =
[T− R
′T′−1
R R′T′−1
−T′−1
R T′−1
]. (2.3)
The periodic stack of gratings must also satisfy a 1D Bloch condition, f2 = µf1, where
µ = exp(−ik0.e2), k0 is the Bloch vector and e2 is the basis vector of the lattice
defined in Fig. 2.2. Combining the Bloch condition with Eq. (2.2) leads to the eigenvalue
equation
T f = µf . (2.4)
In practice, the form of the transfer matrix given in Eq. (2.3) results in a numerically
unstable eigenvalue equation, so Eq. (2.4) must be expressed in an alternative, more
robust form before the eigensolutions can be calculated [178, 179]. We do not discuss
the details here.
2.2.4 Bloch mode partitioning
The solutions to Eq. (2.4) yield the Bloch modes, expressed as a superposition of plane
waves, and their associated eigenvalues, µ. In practice, the plane wave field expansions
must be truncated to a finite number of orders, running from s = −Npw to s = Npw,
resulting in transfer matrices of even order with dimension 2(2Npw + 1) × 2(2Npw +
1). This allows the Bloch modes to be partitioned into equal numbers of downward
and upward modes. The details of the partitioning are given in Ref. [169], but for
the purposes of this discussion it is sufficient to note that the non-propagating states
with |µ| 6= 1 are classified according to the direction in which they decay, while the
directionality of the propagating modes (|µ| = 1) is inferred from the energy flux normal
to the grating. Hence the non-propagating modes with |µ| < 1 are classified as forward
(downward) states, while those with |µ| > 1 decay in the opposite direction, and are
thus classified as backward (upward) states. With the modes partitioned in this way,
the transfer matrix can be written in its diagonalised form
T = FLF−1, (2.5)
which includes the whole family of eigenvalue equations T F = FL from Eq. (2.4).
Here, F is a matrix whose columns are the eigenvectors of T , and L is a diagonal ma-
trix containing the corresponding eigenvalues µ. When partitioned using the approach
described above, F and L have the forms
F =
[F− F ′
−F+ F ′
+
], L =
[Λ 0
0 Λ′
]. (2.6)
48 CHAPTER 2. BLOCH MODE SCATTERING MATRIX METHOD
The columns of the matrices F∓, F ′∓ are formed from the downward and upward plane-
wave components of the eigenvectors for the downward and upward Bloch modes respec-
tively, and Λ and Λ′ are diagonal matrices containing the corresponding eigenvalues µj.
The columns of F are scaled so that physical properties such as reciprocity and energy
conservation, and modal orthogonality are represented in a physically normalised form.
The matrices F∓ and F ′∓ provide a direct mapping between the Bloch mode basis
and the plane wave basis. We can thus treat propagation within a segment in terms of
the Bloch modes of that segment, but transform the fields into the common plane-wave
basis to match the fields at interfaces between different segments. This is the topic of
the next section.
2.3 From a PhC segment to a PhC device
Propagation of light through a complex PhC structure such as that illustrated in Fig. 2.1
can be characterised by two basic processes. The first is propagation within a segment
of identical gratings and the second is reflection and transmission at a single interface
between two different segments. In the BMM, the first of these is described in terms of
the Bloch modes of that segment, while the second is characterised in terms of Bloch
mode transmission and reflection matrices that play the role of generalised Fresnel
coefficients. We note that the interface could instead be characterised by plane-wave
scattering matrices, but as we demonstrate in the remainder of this chapter, the Bloch
mode scattering matrices allow photonic crystal structures to be treated using the
formalism developed for thin-film optics.
2.3.1 Propagation in a single PhC medium
Consider first the propagation of light in a single finite segment of PhC consisting of
L grating layers as illustrated in Fig. 2.4. Light incident on the segment from above,
whether from free-space or another PhC, excites downward propagating Bloch modes
at the interface with amplitudes given by the vector c−. Similarly, light incident from
below excites upward propagating Bloch modes, c+. Hence, at interface n inside the
segment, the downward and upward Bloch mode amplitudes are given by c−(n) = Λnc−and c+(n) = Λ′ −(L−n)c+, where the Λ matrices introduced in Section 2.2.4 describe
propagation through a single layer, and hence Λn describes propagation through n
layers. Similarly, Λ′ is describes propagation in the backward direction. The plane-wave
expansion coefficients f(n) for fields at interface n are obtained via a simple change of
basis using F∓ and F ′∓ which gives
f(n) =
[f−(n)
f+(n)
]=
[F−F+
]c−(n) +
[F ′−
F ′+
]c+(n). (2.7)
We use this plane wave expansion in Sections 2.3.2 and 2.3.3 in order to match the
fields between two different segments with different Bloch mode bases.
2.3. FROM A PHC SEGMENT TO A PHC DEVICE 49
n = 0
n = L
n
L - n
c-c-(n) = Λnc-
c+(n) = ΛL-nc+c+
Figure 2.4: Downward and upward Bloch mode amplitudes c−(n) and c+(n) defined
at interface n in a segment of length L in terms of Bloch modes sourced respectively
from the interfaces above (c−) and below (c+) the segment. Each grating layer in the
segment is identical.
Although the full complement of propagating and evanescent Bloch modes may be
required to match the fields at input and output interfaces, the evanescent modes excited
at the top interface decay inside the segment as µn, since |µ| < 1 (see Section 2.2.4), and
similarly for evanescent fields excited from below. Thus, in a long segment (typically
L > 5) that supports both propagating and evanescent modes, the propagating modes
dominate and in many cases the evanescent states can be dropped from the calculation.
Examples of this approximation are given in Section 3.3.2 and [180], and a rigorous
asymptotic analysis is described in Ref. [169]. Note that while this simplification reduces
the number of Bloch modes in the analysis, a sufficiently large set of both propagating
and evanescent plane wave orders is still required to represent each of the propagating
Bloch modes accurately.
2.3.2 Interfacing two semi-infinite segments
In this section we consider the reflection and transmission of Bloch modes at an interface
between two semi-infinite media, M1 and M2, as shown in Fig. 2.5, and derive the PhC
analogues of Fresnel reflection and transmission coefficients. These generalised Fresnel
coefficients are Bloch mode scattering matrices with the domain and range defined by
the Bloch modes of the input and output media.
The interface shown in Fig. 2.5 is a fictitious line between the bottom grating layer
of M1 and the top grating layer of M2. In M1, the downward and upward mode Bloch
mode amplitudes at the interface are given by the vectors c−1 and c+1 respectively, and
similarly in M2. Since the interface lies in the homogeneous medium between layers
50 CHAPTER 2. BLOCH MODE SCATTERING MATRIX METHOD
c2
+c2
-
c1
+c1
-
M1
M2
Figure 2.5: Diagram of a single interface between two semi-infinite media, M1 and M2.
The Bloch mode amplitudes at the interface are given by the vectors c∓1 in M1 and c∓2in M2.
there is no change in material properties and thus the fields and their upward and
downward components must be continuous across it. As the Bloch mode expansions
in two different media cannot be directly compared, the fields on either side of the
interface are first transformed into the common plane-wave basis. The field continuity
relationship therefore becomes[F−
1
F +1
]c−1 +
[F−′
1
F +′
1
]c+
1 =
[F−
2
F +2
]c−2 +
[F−′
2
F +′
2
]c+
2 , (2.8)
where the F∓j are the partitioned eigenvector matrices introduced in Section 2.2.4,
defined in medium j.
The generalised Fresnel (Bloch mode) transmission and reflection matrices are de-
fined by the relations
c+1
def= R12c
−1 + T21c
+2 (2.9)
c−2def= T12c
−1 + R21c
+2 , (2.10)
which express the Bloch modes propagating away from the interface in each region in
terms of the incident Bloch modes. The Bloch mode scattering matrices are set in sans
serif typeface to distinguish them from the plane-wave scattering matrices, which are
presented in standard Roman typeface throughout. Solving Eq. (2.8) and expressing the
outgoing fields in terms of the incoming fields then leads to the following expressions
for the Bloch mode reflection and transmission matrices:
R12 = (F +′
1 )−1(I −R∞
2 R∞′
1
)−1
(R∞2 −R∞
1 ) F−1 , (2.11a)
T12 = (F−′
2 )−1(I −R∞′
1 R∞2
)−1 (I −R∞′
1 R∞1
)F−
1 , (2.11b)
where
R∞j = F +
j (F−j )−1, R∞′
j = F−′
j (F +′
j )−1. (2.11c)
2.3. FROM A PHC SEGMENT TO A PHC DEVICE 51
Similar expressions may be obtained for R21 and T21. In the most general treatment, ad-
ditional Bloch mode matrices are required for each propagation direction. For instance
T12 describes forward (downward) propagation from M1 to M2 in while T′12 describes
backward (upward) propagation from M1 to M2. To simplify the notation, we assume
throughout this thesis that the segments are numbered in order from top to bottom as
in Fig. 2.1, and thus the propagation direction is implicit in the Bloch mode matrix
subscripts.
In Eq. (2.11c), R∞1 denotes the plane-wave reflection matrix that characterises re-
flection of a plane-wave field incident from above on a semi-infinite slab of M1. Cor-
respondingly, R∞′
1 is the reflection matrix for plane-wave incidence from below. An
arbitrary lateral shift can be introduced at the interface by applying a translation oper-
ator Q of the same form as in Section 2.2.1 to the plane-wave scattering matrices of one
or both segments. This operation is used in Section 3.3.1 to study transmission through
a waveguide dislocation. In Chapter 5 we study the coupling of light from free-space
into the Bloch modes of a uniform PhC. For this application, R∞ proves particularly
useful as it allows the coupling properties to be separated completely from the effects
of finite-thickness PhCs. The Bloch modes are normalised so that the reflectance of
a propagating plane wave order p into reflected order q is simply given by the square
modulus of the corresponding element of R∞, |R∞qp|2.
2.3.3 Propagation through three segments
We now combine the results of the previous two sections to consider propagation from
M1 to M3 in the three segment structure illustrated in Fig. 2.6, where segments M1 and
M3 are semi-infinite and segment M2 consists of L grating layers. This analysis reveals
clearly the similarity of the Bloch mode method to the standard approach in thin-film
optics.
In the most general case, each interface in Fig. 2.6 is described by four Fresnel
matrices, Rij, Tij, Rji, and Tji, which can be calculated from Eqs. (2.11). Here, the
subscript ij corresponds to propagation from segment Mi to Mj through a single in-
terface. Transmission from M1 to M3 is characterised by an incident field composed
of downward propagating Bloch modes with amplitudes c−1 , a reflected field of up-
ward propagating modes c+1 = R13c
−1 , and a transmitted field of downward propagating
modes, c+3 = T13c
−1 in M3.
In segment M2, the fields have both downward and upward Bloch mode components,
c−2 and c+2 , originating from the top and bottom interfaces respectively, as shown in
Fig. 2.6. The field matching equations at the two interfaces are then
c+1 = R12c
−1 + T12Λ
Lc+2 , c−2 = T12c
−1 + R21Λ
Lc+2 , (2.12a)
c+2 = R23Λ
Lc−2 , c−3 = T23ΛLc−2 , (2.12b)
where Eqs. (2.12a) and (2.12b) correspond to field matching at the top and bottom
interfaces respectively. Note that Λ−1 has been substituted here for Λ′, a result due
52 CHAPTER 2. BLOCH MODE SCATTERING MATRIX METHOD
c2
+
c2
-
c1
+c1
-c1
+c1
-M1
c3
+c3
-
M2
M3
L
LL +c2
LL -c2
Figure 2.6: Diagram of a three-segment structure consisting of semi-infinite segments
M1 and M3 above and below segment M2, which is formed from L grating layers.
to the forward/backward pairing of the Bloch mode eigenvalues that holds for all PhC
lattices, as shown in Ref. [169]. Expressions for T13 and R13 are obtained by solving
these four equations, which gives
R13 = T21ΛLR23Λ
L(I − R21Λ
LR23ΛL)−1
T12 + R12, (2.13a)
T13 = T23ΛL
(I − R21Λ
LR23ΛL)−1
T12. (2.13b)
Given that the BMM is cast in a form that resembles thin-film optics, it is not
surprising that Eqs. (2.13) are generalisations of the Airy formulae for a Fabry-Perot
interferometer (see for example Ref. [122], Chapter 7). The term (I − R21ΛLR23Λ
L)−1
in both expressions encapsulates the multiple reflections within segment M2, while the
remaining terms correspond to reflection and transmission at the interfaces when the
light enters and leaves M2.
2.3.4 Propagation through N segments
With the results of Sections 2.3.2 and 2.3.3, we are now in a position to analyse a
generic N -segment structure of the type shown in Fig. 2.7, consisting of semi-infinite
media M1 and MN , and finite segments M2–MN−1 of lengths L2–LN−1. This is achieved
via a straightforward process of recursion, starting from the interface between segments
MN and MN−1 and working upwards to the complete structure which is characterised
by the Bloch mode scattering matrices R1,N and T1,N , defined according to
c+1 = R1,Nc
−1 , c−N = T1,Nc−1 . (2.14)
2.3. FROM A PHC SEGMENT TO A PHC DEVICE 53
Mn
MN-1
M2
M1 (semi-infinite)
MN (semi-infinite)
�� ����
���
���
��� � ��� ��� ��
� � ���������� ��
� � �������������� ��� �� ���� � ���� ����� � �
Mn-1
Mn+1
Figure 2.7: Recursive coupling of grating stacks. An incident field from above, c−1 , in
semi-infinite stack M1, is reflected back into M1 (c+1 ) and transmitted through the N
stacks to semi-infinite stack MN (c−N).
We begin by considering the fields at an interface between segments Mn and Mn+1
as shown in Fig. 2.7. Following the notation used in Eq. (2.15a), we define Bloch mode
scattering matrices that describe the reflection and transmission properties of the stack
comprising segments Mn+1, Mn+2 . . .MN , such that
c+n = Rn,Nc−n , c−N = Tn,Nc−n . (2.15a)
Turning our attention now to the interface at the top of Mn, we consider propagation
from Mn−1 through Mn and eventually into MN . Using Eq. (2.15a), we can treat the
stack of segments below Mn as a single interface separating Mn and MN described
by Fresnel matrices Rn,N and Tn,N which reduces the problem to the three segment
structure considered in Section 2.3.3, where the three segments are Mn+1, Mn and MN .
Substituting the Bloch mode scattering matrices for the two interfaces into Eqs. (2.13),
we derive the recurrence relations
Rn−1,N = Rn−1,n + Tn,n−1ΛLnn Rn,NΛLn
n
·(I− Rn,n−1Λ
Lnn Rn,NΛLn
n
)−1Tn−1,N
(2.15b)
Tn−1,N = Tn,NΛLnn
(I− Rn,n−1Λ
Lnn Rn,NΛLn
n
)−1Tn−1,N . (2.15c)
Thus, R1,N and T1,N can be derived for any number of segments by repeated application
of Eqs. (2.15b) and (2.15c), starting from the last interface and working upwards in
Fig. 2.7.
54 CHAPTER 2. BLOCH MODE SCATTERING MATRIX METHOD
2.4 Numerical aspects
For the numerical implementation of the BMM we use a combination of Mathematica
and FORTRAN code. Mathematica is used as a front-end and for all the matrix ma-
nipulation and solution of eigenvalue problems, while the more numerically intensive
grating scattering calculations are implemented in FORTRAN, with the two linked by
MathLink. While the code has been undergoing development during the course of this
thesis, and is by no means optimised, the majority of the results presented here were
calculated on a laptop computer with a 1.8 GHz Pentium 4 processor and 512 MB of
memory.
The numerical accuracy of the Bloch mode method depends on four main param-
eters: the truncation orders of the multipole, plane wave and Bloch mode expansions
and the size of the supercell. Since energy conservation and reciprocity are both pre-
served analytically in the formulation as a result of using the multipole method (see
Section 2.2.1), these physical properties cannot be used to verify the accuracy of a
calculation. Hence, the main test for accuracy is numerical convergence with respect
to the truncation orders and the supercell dimension. In the current numerical im-
plementation, an equal number of plane wave and Bloch mode orders are retained in
the full calculation, so it is necessary to check convergence of only three parameters,
rather than four. In practice it is often convenient to set “safe” values that guarantee
convergence for a wide range of structures. Although this means sacrificing some nu-
merical efficiency, it reduces the need for convergence tests every time a new structure
is calculated. It is still important, however, to identify how the geometry of a struc-
ture influences the accuracy of the calculation, and the optimal truncation orders and
supercell size.
As a general rule, the multipole truncation order Nm, required for a given accu-
racy depends most strongly on the relative cylinder separation within the grating; more
multipole orders are needed as the relative cylinder radius, r/d increases. The BMM
calculations presented throughout thesis have all been performed with a multipole trun-
cation order of Nm = 7 ie. 2Nm + 1 = 15 multipole terms are used in the field
expansions centred on each cylinder within the supercell. This is sufficient to ensure
convergence for the largest cylinders considered, which have a radius of r = 0.35d.
The number of plane-waves required for convergence depends on the size of the
supercell since the relative spacing of orders is given by the grating equation (2.1). We
have already noted that Dx must be sufficiently large to ensure that there is negligible
coupling between functional elements such as waveguides or cavities in neighbouring
supercells. Table 2.1 gives an example of the convergence behaviour as a function of Dx
and the plane-wave truncation order, Npw. The value listed in the table is the modal
propagation constant β = arg(µ) (units of d) calculated for a single waveguide in a
square lattice, rod-type PhC with cylinders of radius r = 0.3d and refractive index
n = 3 in air (n = 1). The data is calculated for a normalised wavelength of λ = 3.15d.
Each column corresponds to a different sized supercell, and the β values are listed as a
2.5. MODELLING COMPLEX STRUCTURES 55
Dx/d
Npw 11 13 15 17 19 21 23
26 1.115906 1.115406 1.115293 1.115246 1.115148 1.115033 1.115011
28 1.115907 1.115406 1.115297 1.115261 1.115240 1.115104 1.115024
30 1.115907 1.115407 1.115297 1.115271 1.115245 1.115213 1.115074
32 1.115907 1.115407 1.115297 1.115273 1.115260 1.115240 1.115169
34 1.115907 1.115407 1.115297 1.115273 1.115266 1.115250 1.115238
36 1.115907 1.115407 1.115298 1.115273 1.115268 1.115261 1.115242
38 1.115907 1.115407 1.115298 1.115273 1.115268 1.115265 1.115254
40 1.115907 1.115407 1.115298 1.115274 1.115268 1.115267 1.115262
42 1.115907 1.115407 1.115298 1.115274 1.115268 1.115267 1.115266
44 1.115907 1.115407 1.115298 1.115274 1.115268 1.115267 1.115266
46 1.115907 1.115407 1.115298 1.115274 1.115268 1.115267 1.115267
Table 2.1: Convergence of the modal propagation constant β = arg(µ) (units of d),
as a function of the plane-wave truncation order Npw and supercell period Dx for the
waveguide described in the text.
function of Npw. Note that the plane-wave expansion is over all orders s in the range
−Npw ≤ s ≤ Npw, so the total number of plane waves is 2Npw + 1.
Observe that within each column of Table 2.1 β converges to 7 figures accuracy, and
these converged values approach a stable solution as Dx increases. All of the BMM
results given in Chapters 3 and 4 are calculated with Dx = 21d and Npw = 44. For
these parameters, β is stable to 6 figures in Table 2.1. While most calculations do
not require such a high level of accuracy, it can be necessary when studying strongly
resonant structures with extremely narrow spectral features. We note, finally, that we
have confirmed the accuracy of the BMM by comparison with a recently developed
Wannier function method [160] and found agreement of results to better than 1 part in
1000.
2.5 Modelling complex structures
In this section we consider two examples that highlight several important aspects relat-
ing to the application of the Bloch mode matrix method to PhC structures consisting of
multiple segments. While the Bloch mode scattering matrices allow simple expressions
such as Eqs. (2.13) to be written down, the calculation of the grating scattering matri-
ces is still a significant numerical task. Efficient modelling approaches can reduce the
computational demand significantly. Any structure that is modelled using the BMM
must be treated as a stack of individual segments formed from gratings. As we demon-
strate in the first example there may be more than one way to do this so it is important
to consider the geometry of the structure carefully before starting a calculation. For
56 CHAPTER 2. BLOCH MODE SCATTERING MATRIX METHOD
M1
M2
(a)
M1
M2
M3
(b)
Figure 2.8: Diagram showing two alternative ways to model the waveguide dislocation
shown in (a) using the Bloch mode method. (a) A single interface formed from two
identical waveguide segments displaced laterally. (b) The same dislocation as in (a)
rotated by 45◦ to form a three-segment structure.
complicated devices it is often beneficial to break the structure down into basic func-
tional components which can be characterised individually before being combined into
the complete structure. We consider such an example in Section 2.5.2.
2.5.1 Lattice orientation
The first example we consider is taken from Section 3.3.1, where we study transmission
through a dislocation in an otherwise straight waveguide extending infinitely above and
below the dislocation. The structure illustrated in Fig. 2.8(a) is an example of such a
dislocation formed in a square lattice of period d by shifting the lower section of the
PhC by 2d′ = 2√
2d along the Γ − M direction, where d′ is the local period of the
grating. If the structure is rotated anti-clockwise by 45◦, we obtain the structure shown
in Fig. 2.8(b), which is identical to the folded directional coupler geometry considered in
Section 3.4. The shading of the cylinders indicates the two different ways to model the
dislocation using the BMM. The structure in (a) is treated as a single interface between
two identical semi-infinite sections of PhC with a lateral displacement, whereas the
structure in (b) is treated as a three-segment problem consisting of identical top and
bottom segments M1 and M3 and a short segment of double waveguide M2.
Consider first the structure in Fig. 2.8(a). Since there is only a single interface, the
transmission and reflection characteristics are given by the Fresnel matrices defined in
Eqs. (2.11) which are functions of the plane-wave scattering matrices R∞1 and R∞
2 , of
the semi-infinite segments. Since M1 and M2 are identical, R∞1 and R∞
2 are related by a
simple translation operation as described in Section 2.3.2 (see also Section 3.3.1). Thus,
to calculate the transmission properties of the dislocation at one wavelength requires
the scattering matrices for a single grating, after which the remaining computation
involves only matrix manipulation.
When the dislocation is oriented as in Fig. 2.8(b), it becomes a three-segment struc-
ture of the type treated in Section 2.3.3, where the Bloch mode transmission and re-
2.5. MODELLING COMPLEX STRUCTURES 57
M1
M2
M3
M4
M5
M6
M7
L2
L3
L4
L5
L6
Figure 2.9: Diagram of a PhC Mach-Zehnder interferometer of the type studied in
Chapter 4. The dashed lines indicated the six interfaces that must be considered in the
BMM analysis of the device. The cylinders that form each segment are indicated by
the shading.
flection properties are given by Eqs. (2.13). Modelling this structure therefore requires
the characterisation of two interfaces, plus an additional grating calculation since the
gratings that form segment M2 are different to those in M1 and M3. As the calculation
of the grating scattering matrices is the most numerically intensive part of method, it
is clearly preferable to treat the dislocation as a single interface as in Fig. 2.8(a). We
have performed both calculations and confirmed that they are numerically consistent.
There are, of course other advantages to the single-interface treatment, not least being
the option of non-integer displacements. Quantitatively, the three-segment calculation
takes approximately 70% longer than the single interface. While this is a matter of
only a few seconds for the simple structure considered here, it illustrates the impor-
tance of lattice orientation in the BMM, and may be important in larger, more complex
structures.
2.5.2 Composite devices
One of the many advantages of the Bloch mode method is that it provides a systematic
approach for combining basic functional components into more complex PhC devices.
We have already seen how segments consisting of identical gratings can be coupled to-
gether using Bloch mode scattering matrices and simple recursion equations (2.15), but
this approach can be extended to the coupling of basic components such as junctions,
PhC mirrors and waveguides. Here we consider one such example: the Mach-Zehnder
interferometer that we study in detail in Chapter 4.
The diagram in Fig. 2.9 shows a symmetric Mach-Zehnder interferometer formed in
a triangular lattice PhC. At a basic level, a symmetric Mach-Zehnder consists of two
beam splitters or junctions joined by two identical arms as shown. The details of the
58 CHAPTER 2. BLOCH MODE SCATTERING MATRIX METHOD
design shown here are discussed in Chapter 4 but are unnecessary for the purposes of
this section. As indicated by the dashed lines, the device can be modelled as a seven-
segment structure, requiring a minimum of four distinct types of grating; gratings in
M1 and M7 are missing one cylinder; gratings in M2 and M6 are missing two cylinders;
gratings in M3 and M5 are missing three cylinders; and the gratings forming M4 have
two missing cylinders separated by three cylinders. Observe that segments M6 and M2
are included to ensure the structure remains symmetric with respect to the input and
output waveguides. This is necessary as result of the lateral shift introduced between
each grating to create the triangular lattice. An equivalent device formed in a square
lattice can be modelled using only five segments and three different types of grating, as
can be seen in Fig. 4.6(a).
Propagation between M1 and M7 can be calculated using the recurrence relations
(2.15), which start by characterising propagation from M5 to M7 then work backwards
to the M1–M2 intersection. When designing a Mach-Zehnder interferometer, however,
it may be necessary to characterise the properties of the junctions independently of the
complete structure. Consider first the input junction consisting of segments M1–M4.
The transmission and reflection properties of the junction are given by the Bloch mode
scattering matrices R14 and T14, where T14 describes the transmission amplitudes of the
Bloch modes from the single input waveguide into the two arms of the interferometer
in M4. These matrices, and their counterparts corresponding to propagation from M4
into M1, are calculated using Eqs. (2.15) with N = 4. Similarly, the output junction is
characterised by the matrices R47 and T47.
Once the complete set of scattering matrices has been calculated for the junctions, it
is straightforward to calculate the properties of the complete structure using Eq. (2.3.3),
where the scattering matrices of the junctions replace the scattering matrices of the
interfaces, such that
R17 = T41ΛL44 R47Λ
L44
(I− R41Λ
L44 R47Λ
L44
)−1T14 + R14, (2.16a)
T17 = T47ΛL44
(I− R41Λ
L44 R47Λ
L44
)−1T14. (2.16b)
This approach is simply a re-ordering of the recursion process described in Section 2.3.4,
and therefore does not provide any computational advantage for a single calculation.
However, the advantages of this component-based approach are obtained when the
Mach-Zehnder interferometer is modified but the junctions are left unchanged. This
occurs in Chapter 4, for instance, when varying the length of the arms or introducing
a phase delay in one arm. In this case the junction scattering matrices need only be
calculated once for each wavelength, and can then be recalled whenever the junction is
used, saving considerable computational expense. Even further efficiencies are gained
if the device is described accurately by only the propagating Bloch modes, as then the
complete junction is characterised by just nine numbers – the elements of the scattering
matrices corresponding to the single propagating mode in M1 and the two modes in
M4. Another example where the component-based approach is particularly useful is
2.6. DISCUSSION 59
in the analysis of the Fabry-Perot interferometer considered in Section 3.3.2 where we
show that a PhC partial mirror is described accurately by a single pair of (complex)
scalar Fresnel coefficients.
As we have demonstrated here, the modelling of even relatively simple devices can
be simplified further by considering first, the functional components that make up
the device. Ultimately, this approach could lead to the development of a component
“library” containing the Bloch mode scattering matrices for each structure. This would
provide a rapid and efficient tool for developing more complex PhC circuits with many
coupled components.
2.6 Discussion
In this chapter we have described the theoretical framework of the Bloch mode scatter-
ing matrix method, and discussed a number of issues relating to the practical application
of the method to studying photonic crystal devices. We have shown that the structure
of the formulation closely resembles that of thin-film optics, with familiar scalar quan-
tities such as Fresnel coefficients being generalised to appropriate matrix forms. While
the method is a highly efficient numerical tool for studying many PhC devices, there
are geometrical limitations imposed by the transfer matrix approach, since all struc-
tures must be described in terms of discrete PhC segments. As we demonstrate in the
remainder of this thesis, however, a wide range of PhC-based devices can be analysed
in this way.
A geometry that we have not discussed in detail is coupling between a homogeneous
dielectric medium such as air, and a PhC. Here, the same supercell must be used in
the dielectric and in the PhC so the Bloch modes outside the PhC are the diffracted
plane wave orders given by the grating equation (2.2). In Chapter 5 we consider the
transmission of an incident plane wave into a uniform PhC, but the method is not
limited to a single plane wave order. If the supercell period Dx is large, and hence the
plane wave orders are closely spaced, it is possible to simulate an incident beam of finite
width by taking an appropriate superposition of plane wave orders.
Perhaps the most important advantage of the Bloch mode method compared to
many other numerical techniques is the physical insight that it provides as a result of
using the natural Bloch mode basis. This allows many seemingly complicated scattering
problems to be expressed in terms of familiar, well understood structures that can be
analysed from first-principles. While we have not demonstrated this aspect here, the
following chapters provide numerous examples.
Chapter 3
Coupled waveguide devices
In this chapter we introduce a new class of photonic crystal device in which directional
coupling and resonant reflection effects are combined. We demonstrate, using both 2D
numerical calculations and simple semianalytic models, that these devices exhibit not
only high-Q reflection resonances but also high transmission over wide-bandwidths. We
present three novel PhC devices that exploit these features – a notch-rejection filter, a
Y-junction and a coupled waveguide delay line.
Much of the work presented in this chapter has been published or is due to appear.
In particular, Sections 3.3.1, 3.3.2, and 3.4–3.6 are based on Refs. [170, 171, 181, 182],
with additional discussions and analysis provided where necessary.
3.1 Introduction
Waveguide coupling and resonant cavities were discussed separately in Sections 1.7.1
and 1.7.5, and in Section 1.7.5 we reviewed several results where a single waveguide
was used to couple light to and from a resonant cavity. In contrast, here we combine
waveguide coupling and resonance effects in a single structure to obtain a range of
unusual transmission characteristics. We present rigorous 2D numerical calculations
using the Bloch mode method, but also show that these structures can be modelled
accurately using simple modal methods and semianalytic techniques. These approaches
provide physical insight into the behaviour of the devices allow rapid and efficient
parameter optimization. The physical interpretation involves two different forms of
waveguide coupling: coupling between parallel waveguides and end-coupling between
two waveguide segments. These two effects are discussed separately in Sections 3.2 and
3.3.
In Section 3.2 we justify numerically the application of conventional tight-binding
approximations [164] to PhC waveguides and present a detailed discussion of coupled
PhC waveguide dispersion properties. Two examples of end-coupling are considered in
Section 3.3: waveguide dislocations and a waveguide-based Fabry-Perot interferometer.
These examples not only provide insight into the coupling of Bloch modes through
61
62 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
waveguide discontinuities, but also demonstrate the efficiency of the Bloch mode method
when studying such structures.
The second half of the chapter consists of Sections 3.4–3.6 where we present three
closely related structures that derive their properties from a combination of waveguide
coupling and resonant reflection at waveguide interfaces. The first of these devices is the
“folded directional coupler” (FDC) - a simple, high-Q notch-rejection filter, the second
is a wide-bandwidth symmetrical Y-junction and the third is a delay line formed by
coupling together multiple FDCs. The properties of all three structures are described
accurately by the semianalytic model derived in Section 3.4.1.
3.1.1 Photonic crystal and numerical parameters
Unless otherwise stated, the numerical examples provided throughout this chapter are
based on two generic photonic crystal lattices. PC1 is a rod-type PhC consisting of a
square lattice of rods of radius r = 0.3d and index ncyl = 3.0 in a background of air (nb =
1). This structure has a TM bandgap in the wavelength range 2.986 < λ/d < 3.774,
corresponding to 0.265 < d/λ < 0.337. The results presented for PC1 are therefore
for TM polarised light. PC2 is a hole-type PhC consisting of a triangular lattice of air
holes (ncyl = 1) in a dielectric background of index nb = 3.37, corresponding to GaAs.
The hole radius is r = 0.3d, which gives a TE bandgap in the range 3.335 < λ/d <
4.629 (0.216 < d/λ < 0.299). Hence the results for PC2 are for TE polarised light.
We consider in this chapter a range of PhC structures and devices, and provide
numerical examples of each to demonstrate the typical characteristics. Although ex-
amples are provided for either PC1 or PC2 in each case, the results are general to
both rod- and hole-type PhCs. The semianalytic analysis of these structures devel-
ops physical insight into device behaviour, thereby allowing simple design rules to be
identified. Within the assumptions of these models, the design rules are not specific
to the underlying PhC lattice and can thus be applied to a range of PhC geometries.
However, this does not imply that all PhCs are equally suitable for a particular device;
the dispersion and modal properties can vary widely, and it is these parameters that
have a significant effect on device characteristics.
The full Bloch mode calculations presented in this chapter were performed with a
supercell period of Dx = 21d to ensure that coupling between supercells is negligible.
Eighty-nine plane wave orders ([−44,+44]) were used to represent the fields between
each layer of cylinders, and the scattering matrices were calculated with 17 multipole
orders ([−8,+8]). For many of the calculations presented here, a supercell of 11d and
45 plane wave orders provides satisfactory accuracy. However, the larger supercell is
required for studying the strong resonant behaviour in Section 3.4, since a very high
spectral resolution is required to resolve the resonant features.
3.2. COUPLED WAVEGUIDE MODES 63
3.2 Coupled waveguide modes
The derivations of the semianalytic methods in Sections 3.4 and Chapter 4 rely on
conventional weak-coupling approximations to describe the modes of parallel coupled
waveguides. Although this is a standard technique for studying conventional dielectric
waveguides, its application to PhC waveguides, which can exhibit strong coupling, has
not been studied systematically. In this section we consider PhC waveguide coupling
in detail and demonstrate numerically that weak-coupling theories predict accurately
the modal fields of coupled PhC waveguides, even when the mode interaction is sig-
nificant. The results also highlight important differences between conventional coupled
waveguides and PhC waveguides, such as mode ordering and symmetry properties.
Recall from Section 1.7 that the supermodes of a two-waveguide system are classified
as either even (+) or odd (−) according to their symmetry with respect to the mirror
plane between the guides. When infinitely far apart, the waveguides are isolated from
each other and so the supermodes are degenerate, being characterised by the propaga-
tion constant β of the single waveguides. As the guides are brought together, the modes
in each waveguide begin to interact and the supermodes split into two non-degenerate
modes |ψ±〉 with propagation constants β±. This splitting is analogous to the splitting
of atomic energy levels when two atoms approach each other.
Coupling between more than two identical waveguides is essentially the same as
the two-waveguide case, however the modes of the individual guides are split into N
supermodes, whereN is the number of waveguides. Approximate modal solutions can be
derived for weakly coupled conventional waveguides using perturbation methods and a
tight-binding approximation that considers only coupling between adjacent waveguides.
We briefly outline here the results derived in Chapter 11 of Ref. [164], using bra-ket
notation to represent the modes. The supermodes in this case are written as linear
combinations of the individual waveguide modes,
|Ψs〉 = 1AN
∑Nn=1Cn|ψn〉,
where Cn = sin(
nsπN+1
), and AN =
(∑Nn=1 |Cn|2
)1/2
, (3.1)
for s = 1, 2, 3, . . . , N and |ψn〉 is the unperturbed mode of the nth waveguide. Here
AN is a normalization factor to ensure 〈Ψ|Ψ〉 = 1, and hence each supermode carries
unit power. In this approximation the propagation constants of the supermodes can be
determined by overlap integrals of the modes in neighbouring waveguides, although we
do not use this result in our analysis.
The application of this simple coupled-mode-theory (CMT) to photonic crystal
waveguides has not been studied rigorously, although a similar approach reported by
Kuchinsky et al. showed good agreement with 2D numerical results in the case of two
coupled waveguides [183]. In this chapter and in Chapter 4 we apply CMT to several
PhC waveguide-based devices and derive semianalytic models that display excellent
agreement with full 2D numerical calculations. The devices consist of both two and
64 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
three coupled waveguide systems, the properties of which we consider in more detail in
Sections 3.2.1 and 3.2.2.
3.2.1 Coupling of two waveguides
We consider first the simplest coupled PhC waveguide example: two identical wave-
guides in PC1, each formed by removing a single line of cylinders in the Γ−X direction,
as shown in Figs. 3.1(e) and (f). The electric field of a single waveguide mode, calcu-
lated using the Bloch mode method, is plotted in Figs. 3.1(a) and (d) for a normalised
frequency of d/λ = 0.32. The dark regions in (d) correspond to negative phase and the
white regions to positive phase. The profile in (a) is calculated along the white line
shown between the cylinders in (d). Note that the field extends several periods into the
PhC walls of the waveguide.
In the tight-binding approximation the odd and even supermodes are given by
Eq. (3.1) which yields
|ψ±〉 = (|ψL〉 ± |ψR〉)/√
2 . (3.2)
Here |ψL〉 and |ψR〉 are the unperturbed modes of the left and right waveguides, respec-
tively, which are simply obtained by applying a lateral translation to the waveguide
mode in Fig. 3.1(a). Figures. 3.1(b) and (c) show the electric fields of |ψ±〉 calculated
using Eq. (3.2) (dashed curves) and the full Bloch mode method (solid curves) for two
PhC waveguides separated by a single line of cylinders as shown in Figs. 3.1(e) and (f).
While the tight-binding approximation is most accurate between the layers of cylin-
ders (red curves), reasonable agreement is also obtained for the fields in the cylinders
(blue curves). Such good agreement is surprising given that the waveguides are only
separated by a single line of cylinders and thus interact strongly with each other. The
tight-binding treatment also provides a reasonable approximation for coupled hole-type
PhC waveguides, although we do not present numerical results here. We note however,
that the approximation was applied successfully to design the hole-type PhC Y-junction
used in the Mach-Zehnder interferometer in Section 4.4.
Figure 3.2(a) shows the projected band structure and dispersion curve for the single
waveguide shown in the inset. The PhC bands are shaded dark grey and the bandgap
is shaded light grey. In contrast to the waveguide modes of the hole-type PhC shown
in Fig. 1.8, there is at most one guided mode for all frequencies in the bandgap. This is
a property common to most single line defect waveguides in rod-type PhCs, and one of
their advantages over hole-type PhCs. Observe also that the dispersion slope of the rod-
type PhC waveguide is positive, while the even mode of the hole-type PhC waveguide
has a negative dispersion slope. Figures 3.2(b)–(f) show the supermode dispersion
curves for coupled waveguides with separations of Nc = 5 to Nc = 1 respectively,
where Nc is the number of cylinder layers between the waveguides, as shown in the
inset diagrams. The even and odd modes are indicated by dotted and dashed curves
respectively. For comparison, the single waveguide dispersion curve from (a) is also
plotted in each of (b)–(f), clearly showing that the mode splitting is almost symmetric
3.2. COUPLED WAVEGUIDE MODES 65
(a) (c)(b)
(d) (e) (f)
Figure 3.1: Electric fields of the modes in a single waveguide (a,d) and two waveguides
separated by a row of cylinders (b,c,e,f). The solid curves in (a)–(c) are calculated along
the corresponding red and blue lines in (d)–(f) using the full Bloch mode method. The
dashed curves in (b) and (c) are the approximate result given by Eq. (3.2). The light
and dark regions in (d)–(f) correspond respectively to positive and negative phases.
about the single waveguide mode, and hence the average of the two modes β = (β+ +
β−)/2 ≈ β. This feature is important for the semianalytic analysis in Section 3.4. The
strong mode coupling for Nc = 1 and Nc = 2 results in large mode splitting ∆β = |β+−β−|/2, a feature that has attracted much attention recently. Recall from Section 1.7.5
that large mode splitting results in short coupling lengths (1.8) and therefore provides
the possibility of compact coupler-based devices [105–107,110–112].
The dispersion curves in Fig. 3.2 also illustrate an unusual mode ordering property
of coupled PhC waveguides which occurs when the dominant Bloch factor of the bulk
crystal, µ, is negative [184]. When Nc = 1, 3, 5, the lowest frequency (fundamental)
mode has odd symmetry, whereas for Nc = 2, 4, the fundamental mode is even. In
contrast, the fundamental mode of a symmetric coupler formed in conventional wave-
guides is always even and the second mode is odd. An interesting consequence of this
is the ability to design waveguide structures that support only an odd mode. Consider,
for instance, the dispersion curves for the Nc = 1 case in Fig. 3.2(f). The even mode,
indicated by the dotted curve, is cut off at a frequency of d/λ = 0.305, below which
66 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
0 0.1 0.2 0.3 0.4 0.50.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
0 0.1 0.2 0.3 0.4 0.50.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
0 0.1 0.2 0.3 0.4 0.50.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
0 0.1 0.2 0.3 0.4 0.50.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
0 0.1 0.2 0.3 0.4 0.50.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
0 0.1 0.2 0.3 0.4 0.50.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
b pd/2
b pd/2
b pd/2
b pd/2
b pd/2
b pd/2
d/l
d/l
d/l d/l
d/l
d/l
N =5c
N =2c
N =4c N =3c
N =1c
(a) (b)
(c) (d)
(e) (f)
Figure 3.2: Projected band structures and dispersion curves for the various waveguide
structures shown in the insets. Dark grey regions represent PhC bands, while the light
grey region shows the bandgap. (a) Single waveguide (b)-(f): Two coupled waveguides
separated by Nc = 5 − 1 rows of cylinders respectively. The dotted curve corresponds
to the even supermode, and the dashed curve corresponds to the odd supermode. The
solid curve is the single waveguide dispersion plotted in (a).
only the odd supermode can propagate. In an ideal symmetric structure, an even mode
incident from another waveguide would be perfectly reflected by the waveguide pair.
Similar concepts of mode symmetry filtering play an important role in the recirculating
Mach-Zehnder interferometer studied in Chapter 4.
3.2.2 Coupling of three waveguides
We now consider the effect of adding a third waveguide to the system in Section 3.2.1
to form a waveguide structure that supports three supermodes, |ψ1,2,3〉 respectively. In
this case the correct linear superpositions of waveguide modes is not as obvious as in
the two waveguide example. These are calculated by substituting N = 3 into Eq. (3.1)
3.2. COUPLED WAVEGUIDE MODES 67
(a) (b) (c)
(d) (e) (f)
Figure 3.3: Contour and line plots of the supermode electric fields in a three waveguide
system with Nc = 1(a–c) and Nc = 2 (d–f). The line plots above each figure show the
electric field profile along the white lines in the contour plots. The solid curves show
the rigorous numerical field calculation and the dashed curves are the coupled mode
approximation (3.1).
which gives
|ψ1〉 = (|ψL〉+√
2 |ψC〉+ |ψR〉)/2,|ψ2〉 = (|ψL〉 − |ψR〉)/
√2, (3.3)
|ψ3〉 = −(|ψL〉 −√
2 |ψC〉+ |ψR〉)/2,
where |ψL〉, |ψC〉 and |ψR〉 are the unperturbed modes of the left, centre and right
waveguides respectively. Observe that |ψ1〉 and |ψ3〉 are even modes, whereas |ψ2〉 is
odd. For conventional waveguides, |ψ1〉 is the fundamental mode, followed by |ψ2〉 and
|ψ3〉 but this is not necessarily the case for PhC waveguides. The contour plots in
Fig. 3.3 show the mode fields for Nc = 1 (a–c) and Nc = 2 (d–f) at d/λ = 0.32. The
modes are listed in order of decreasing β, where β = 1.473, 1.171, 0.377 for modes (a)–
(c) and β = 1.320, 1.164, 0.907 for (d)–(f). As for double waveguides, the mode order
68 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
0 0.1 0.2 0.3 0.4 0.50.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
b pd/2
d/l
(a)
0 0.1 0.2 0.3 0.4 0.50.22
0.24
0.26
0.28
0.3
0.32
0.34
0.36
b pd/2
d/l
(b)
(b)(a)
Figure 3.4: Dispersion curves for three coupled PhC waveguides with (a) Nc = 1 and
(b) Nc = 2. The dotted and dashed-dotted represent the two even modes |ψ1〉 and |ψ3〉respectively, and the dashed curve corresponds to the odd mode |ψ2〉. The solid line in
both plots is the dispersion of a single waveguide mode.
changes depending on whether Nc is even or odd, but here it is the two even modes,
|ψ1〉 and |ψ3〉, that swap positions. This is clearly seen in the line plots above each of
the figures which show the electric field along the white lines in each of the contour
plots (solid curves) and the approximate result given by Eqs. (3.3) (dashed lines). Once
again the agreement between the rigorous numerical calculation and the coupled mode
approximation is excellent, with the two curves being indistinguishable for Nc = 2.
The supermode dispersion curves are shown in Fig. 3.4 for Nc = 1 and Nc = 2,
where the dotted and dashed-dotted curves correspond to the two even modes |ψ1〉 and
|ψ3〉 respectively, and the dashed curves corresponds to the odd mode |ψ2〉. The single
waveguide dispersion is also indicated by the solid curve. Since the three supermodes
are split almost symmetrically relative to the single waveguide mode, the dispersion
curve of |ψ2〉 is very close to that of the unperturbed waveguide. Note that in contrast
to double waveguides, it is not possible to design a triple waveguide structure that only
supports an odd mode since |ψ2〉 always lies between the two even modes.
3.3 Waveguide coupling at interfaces
The second type of waveguide coupling that plays an important role in the context
of this thesis is mode coupling through an interface between the ends of two PhC
waveguides. In this case, the coupling properties are often a combination of direct
transmission and resonant coupling effects. The Bloch mode scattering matrix method
is an ideal numerical tool for studying this form of coupling as it allows interface effects
to be treated separately from propagation along sections of straight waveguide. In
particular, the generalised Fresnel matrices introduced in Section 2.3.2 describe the
coupling from one Bloch mode basis to another. Here we present two examples that
provide physical insight into Bloch mode coupling behaviour, and at the same time
demonstrate the advantages of the Bloch mode method for studying such problems. In
3.3. WAVEGUIDE COUPLING AT INTERFACES 69
Section 3.3.1 we consider a lateral dislocation in a waveguide that can be treated as a
single interface between two identical PhC stacks and therefore requires the calculation
of only one set of scattering matrices. Coupling through multiple interfaces in the form
of a Fabry-Perot cavity is considered in Section 3.3.2 where we show that the cavity
resonance is described accurately by considering only the propagating Bloch modes.
3.3.1 Waveguide dislocations
Recall from Section 1.7.2 that an ideal, straight PhC waveguide can guide light without
loss or reflection. In practical situations, however, non-uniformities in the crystal or
waveguide structure introduce losses due to reflection and scattering. In this section
we consider the effect of a linear dislocation through the crystal and waveguide. Two
structures are considered: a dislocation perpendicular to the guide direction (along the
Γ − X axis of the crystal), as in Fig. 3.5(a), and a dislocation at a 45◦ angle to the
waveguide (along the Γ−M crystal axis) (Fig. 3.5(b)). The bulk crystal in Fig. 3.5(b) is
the same square lattice of period d as in (a), but rotated by 45◦. Thus, when calculating
the scattering matrices for the rotated crystal, the lattice period within each grating
layer is d′ =√
2d.
The dislocation is modelled as an interface between two identical semi-infinite crys-
tals as discussed in Section 2.5.1. To introduce a lateral shift of sx parallel to the
grating layers as shown in Fig. 3.5, we apply a translation operator Qs = diag[eiαpsx/2
]to the appropriate R∞ matrices and define R∞
sx = Q−1s R∞Qs and R∞′
sx = QsR∞′
Q−1s as
described in Section 2.3.2. Here, the switching of Qs and Q−1s simply changes the sign
of sx depending on whether the dislocation is viewed from above or below. The Fresnel
matrices are then calculated using Eqs. (2.11) with R∞sx replacing R∞. Figure 3.6 shows
the transmission through the dislocated waveguides in Fig. 3.5 as a function of displace-
ment sx. The wavelength is fixed at λ = 3.25d. Since the period of the supercell used
in these calculations is Dx = 21d, the transmission curves have the same periodicity.
Most of the features in the spectra of Fig. 3.6 are easily understood. Both cases
exhibit 100% transmission for sx = 0, since this corresponds to an infinite guide with no
dislocation. However, the two curves show very different behaviour when a dislocation is
introduced. In the case of the perpendicular dislocation (Fig. 3.5(a)), the transmission
curve is symmetric about sx = 0, as expected since the guide is symmetric about the
axis transverse to the interface. The transmission peaks correspond to local maxima
in the overlap of the fields in the two waveguide sections. These peaks occur close to
integer values of sx/d since at these points the cylinders are aligned at the interface
between the two sections so the only dislocation occurs at the waveguide ends.
For the 45◦ dislocation, the transmission curve is strongly asymmetric in sx, which
can be understood from the diagram in Fig. 3.5(b). When sx > 0 (as in the diagram),
the structure resembles two overlapping waveguides at 45◦ separated by an arrangement
of cylinders. In this case, the overlap region allows strong coupling between the two
waveguides, and hence strong resonant transmission peaks. The first two of these
70 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
D =21dx
D =21d’x
d
d’= 2 d
sx
s ’x
(a)
(b)
Figure 3.5: (a) Straight waveguide with dislocation along the Γ−X crystal axis. The
circles represent the cross-section of the dielectric cylinders forming the 2D photonic
crystal. The cylinder shading indicates the two distinct waveguide sections considered
in the calculations. (b) Identical waveguide, but with dislocation along the Γ − M
crystal axis. Both diagrams show the supercell used in the calculations, and the dash-
dot line indicates the line of dislocation. The waveguides extend infinitely in the vertical
direction.
peaks occur at sx ≈ 1d′ and sx ≈ 2d′, for which the transmission exceeds 99.5%. For
integer values of sx/d′, the structure resembles the folded directional coupler discussed
in Section 3.4. If the guide is displaced in the sx < 0 direction, however, the two guides
are effectively being pulled apart in both directions, and so the transmission is seen
to decrease rapidly below 1% and stay low as |sx| is increased. Such an asymmetry
in transmission with respect to displacement could potentially be used in a sensitive
directional motion detector [185].
3.3.2 Fabry-Perot cavity
Resonant devices are an essential component in modern optics, being used in filters,
switches, couplers and many other other applications. We reviewed in Section 1.7 some
of the properties and applications of resonant PhC cavities, most of which were formed
by removing or modifying one or several cylinders in an otherwise uniform PhC. Here
we consider a more conventional resonant cavity design – a Fabry-Perot cavity – and
show that it can be modelled accurately using semianalytic methods.
The Fabry-Perot (FP) resonator is one of the most basic resonant devices, consist-
3.3. WAVEGUIDE COUPLING AT INTERFACES 71
10 5 0 5 10Displacement, sx (units of d)
0.2
0.4
0.6
0.8
1
T
Figure 3.6: Transmission of dislocated straight guides, for dislocations parallel to the
Γ−X (solid curve) and Γ−M (dashed curve) crystal axes at λ/d = 3.25.
ing of two high-reflectivity mirrors on either end of a cavity. Away from resonance,
the transmission of the Fabry-Perot device is typically very small, depending on the
reflectivity of the individual mirrors. On resonance, however, the small transmitted
amplitude through the first mirror interferes constructively inside the cavity, and high
transmission is possible. If the mirrors are identical (balanced), the resonant transmis-
sion can reach 100% but this is reduced if the reflectivities are unequal. The finesse,
F , of a balanced FP cavity is defined as the ratio of the fringe separation (free spectral
range) to the full width at half maximum (FWHM) of the transmission peak. It is a
function of the mirror reflectivity only, given by
F = π√R/(1−R), (3.4)
where R = |r|2 is the reflectance of each mirror (Ref. [122], Chapter 7). In a physical
sense, the finesse is a measure of how many times light circulates through the cavity be-
fore it has all leaked away. Thus, for high finesse cavities, the mirrors must be strongly
reflecting, requiring R > 0.9985 for F > 2000. The resonance width is characterised by
the “quality factor” Q, defined in Section 1.7.1 as the ratio of the resonant frequency
(ω0) to the FWHM (∆ω) of the transmission intensity peak (1.6). In practical PhC
devices, however, radiation losses can reduce Q significantly. As we discussed in Sec-
tion 1.7.1, cavity losses can be minimised with careful cavity design, for instance by
modifying the cavity walls to reduce coupling to radiation modes [54]. Since the issue of
losses is relevant to all of the structures considered in this thesis, we provide a general
discussion of the matter in Chapter 6.
72 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
L
d
mirror depth
air holes
Figure 3.7: Diagram of the Fabry-Perot resonator formed in a single-mode waveguide
in PC2. The mirrors are formed by “plugs” of cylinders. For the example here, all
cylinders are identical.
Mirror properties
Fabry-Perot devices in photonic crystals have been studied previously in various con-
texts [112,186–188]. Here we apply the Bloch mode method to the simple Fabry-Perot
cavity shown in Fig. 3.7, formed by placing two “plugs” of Nplug cylinders in a single-
mode waveguide formed in the hole-type PhC, PC2. Although their general behaviour
is similar to that of conventional Fabry-Perot resonators, PhC-based Fabry-Perot de-
vices have one significant difference that arises because the mirrors in a PhC FP are
distributed reflectors, in which the light penetrates several layers into the crystal struc-
ture. This introduces additional spectral features such as resonances in the mirror
reflectivity due to reflections within the mirror itself. The reflectivity of the mirrors can
be increased by adding more cylinders, or by changing the properties of the cylinders
forming the mirror, such as the radius [186]. Figure 3.8(a) shows the reflectivity of a
single mirror as Nplug is increased from 1 to 6 cylinders. As expected, the reflectivity
increases with the number of cylinders, however the effect is not monotonic at all wave-
lengths due to resonant reflections from either end of the plug, as seen in the curves
corresponding to Nplug = 3− 6.
The properties of an individual mirror can be calculated with the Bloch mode
method as a 3-section waveguide structure, where the two end sections are identical
3.3. WAVEGUIDE COUPLING AT INTERFACES 73
10-7
10-5
10-3
10-1
123
456
Mirror thickness (cylinders)
T
3.4 3.5 3.6 3.7 3.8
d/l
10-11
10-8
10-5
10-2
10-14
T
(a)
(b)
Figure 3.8: (a) Transmittance (log scale) as a function of normalised wavelength,
through a single mirror (plug) formed in a single-mode waveguide in PC2 for mir-
rors with Nplug = 1 − 6. (b) Transmittance of a Fabry-Perot cavity of length L = 5d
for mirrors with Nplug = 1− 6.
single waveguides, and the central section is composed of complete layers of cylinders
(possibly with changes made to the central one). The transmission and reflection of the
mirror are therefore described by Eqs. (2.13). The scattering matrices are calculated
for each of the interfaces, with any symmetries being taken into account to save com-
putation time. In this case the mirror is symmetric with identical waveguides on both
sides, so T21 = T23 and R21 = R23, which reduces significantly the required scattering
matrix calculations. Many calculations can benefit from an understanding of the sym-
metry properties of the structure as demonstrated by the examples in this chapter and
Chapter 4.
Unless the central blocking cylinder is made very small, the plug section does not
support any propagating modes, and hence the reflection of the propagating modes at
the 1−2 interface is 100%. However, the mirror is of finite thickness so energy is coupled
into the cavity by evanescent modes within the plug, giving a non-zero transmission of
74 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
the propagating Bloch mode from section 1 to section 3.
Fabry-Perot properties
As shown in Fig. 3.7, the FP cavity is formed by placing two such mirrors in a single-
mode waveguide, to form a cavity of length L. The transmission through the complete
structure can be calculated in two ways. The first approach is to apply Eq. (2.15) with
N = 5, thereby treating it as a 5 layer structure consisting of the input and output
waveguides, the two mirrors and the cavity. The second, and more efficient, method
is to treat each mirror as a single interface characterised by the scattering matrices
Rplug = R and Tplug = T which are calculated from Eqs. (2.13) as described above. This
allows the whole FP structure to again be treated as a 3-layer problem using Eqs. (2.13),
where the matrices T12 = T21 = T23 = Tplug describe the transmission of Bloch modes
through the mirror, and R12 = R21 = R23 = Rplug describe reflection from the mirror.
Thus, once the mirror properties have been calculated, the length of the cavity can be
varied without the need to repeat any scattering matrix calculations.
Figure 3.8(b) shows the transmission spectra for a cavity of length L = 5d with
increasing mirror thickness. As expected from Eq. (3.4), the peak width decreases
(Q increases) as the reflectivity of the mirrors is increased. At frequencies far from
the reflection resonances shown in Fig. 3.8(a), the reflectivity of the mirrors increases
exponentially with Nplug so Q also increases rapidly with the size of the mirror. The
transmission spectra become more complicated in the vicinity of a mirror reflection
resonance, since the reflectance changes significantly across the width of the cavity
resonance. As a consequence of this, Eq. (3.4), which is calculated for a fixed R loses
some accuracy.
Table 3.1 shows the quality factor of the Fabry-Perot resonance near d/λ = 0.369
as a function of the number of cylinders in the mirror for Nplug = 1 − 4. Since these
calculations are purely 2D, out-of-plane losses would have to be controlled to achieve
such large Q values in realistic slab geometries. We discuss this issue in Chapter 6.
Experimental measurements ofQ = 7300 have been reported for a similar FP cavity [83]
and Q = 1.2 × 104 was measured for a structure with multi-mode rather than single-
mode waveguides [188]. Although these values are several orders of magnitude lower
than the highest-Q PhC cavities discussed in Section 1.7.1 [75], the in-line geometry of
the FP structure provides a simple design for an integrated notch-pass filter.
Propagating mode analysis
The results presented so far include both propagating and evanescent Bloch mode or-
ders. It is not always necessary, however, to include the evanescent orders and in these
cases, many of the complicated matrix expressions can be replaced with semianalytic
scalar forms [180]. For the FP cavity described above, the mirror sections have no prop-
agating modes and therefore rely solely on evanescent modes to carry energy through
the reflecting layers into the cavity. Between the mirrors, however, most of the energy
3.3. WAVEGUIDE COUPLING AT INTERFACES 75
Nplug λ0(d) (full) λ0(d) (prop) Q (full) Q (prop) F [= π√R/(1−R)]
1 3.68692 3.68692 78 78 6.8
2 3.69090 3.69091 1200 1200 97
3 3.690642 3.690656 6.6× 104 6.4× 104 5.4× 103
4 3.6905974 3.6906108 1.3× 106 1.1× 106 1.1× 105
Table 3.1: Resonant wavelength (units of d), Q and F for a Fabry-Perot cavity of
length L = 5d in PC2 for mirrors of thickness Nplug. Resonant wavelength and Qvalues are given for both the full numerical calculation (full) and the propagating mode
approximation (prop).
is carried by the single propagating Bloch mode, suggesting that a propagating mode
analysis might be appropriate. We describe here the application of this approach to the
FP structure, although we note that a rigorous asymptotic analysis for a general class
of three layer structures was reported by White et al. [170].
The most significant advantage in using Bloch mode methods derives from their
capacity to model efficiently, long propagation spans in a fixed medium. Since this
is characterised by the matrix ΛL in Eqs. (2.13), it follows that for sufficiently long
spans (typically L > 5), the evanescent modes can be disregarded since they decay
exponentially with L and therefore the calculations are dominated by the propagating
states. To apply this to the Fabry-Perot resonator in Fig. 3.7, we first calculate the
transmission and reflection matrices through a single mirror using the full Bloch mode
calculation. Since these describe coupling between single-mode waveguides on either
side of the mirror, they reduce to complex scalar transmission and reflection coeffi-
cients when only the single propagating mode is considered in each segment. Similarly,
the matrix Λ in Eqs. (2.13) reduces to the eigenvalue µ corresponding to the single
propagating mode of the waveguide. Substituting these scalar values into Eq. (2.13b)
results in a simple scalar equation that gives the transmission through the complete FP
structure as a function of the cavity length, L. Since the properties of the mirror are
wavelength-dependent, the scattering matrices must be calculated for each wavelength
of interest. In practice, once the matrices have been calculated in full, only the elements
corresponding to the propagating mode(s) need to be saved.
The effect of neglecting the evanescent modes within the cavity can be seen in
Table 3.1, where the resonant wavelength and Q for the FP cavity of length 5d are
compared for the full evanescent field calculation and the reduced, propagating mode
calculation. For all but the highest Q cavities, the agreement is excellent. If the cavity
length is reduced, it might be expected that the evanescent fields would have a more
significant effect on the results. However, even for cavity lengths of 2 lattice periods, the
difference between the propagating mode result and the full result is relatively small.
For mirrors that are 3 layers thick, the difference in resonant frequency calculated with
the two methods is less than 0.05%, and the Q values agree to 2 figures. Table 3.1 also
shows the finesse calculated using the standard FP equation (3.4) with the reflectance
76 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
values R = 1− T obtained from the transmission data plotted in Fig. 3.8(a).
The two examples in this section demonstrate clearly the advantages of the Bloch
mode matrix method for studying PhC waveguide structures consisting of a few different
waveguide sections. We have shown that the calculational demands can be reduced
by taking into account the symmetry properties of the structures being considered,
and even greater reductions can be made by considering only the propagating modes.
When these two simplifications are combined with the coupled mode theory results
of Section 3.2, they provide a powerful set of tools for developing simple semianalytic
models to describe relatively complex photonic crystal structures. In the remainder of
this chapter and in Chapter 4, we study several different PhC devices that exhibit the
two types of coupling discussed in Sections 3.2 and 3.3 in addition to Fabry-Perot-like
resonances. Elegant semianalytic models are derived for each device, and the results
are compared to the rigorous 2D numerical results obtained with the full Bloch mode
method.
3.4 Folded directional coupler
In this section we discuss the folded directional coupler (FDC): a novel notch-rejection
filter that exploits the mode coupling properties of a directional coupler and the sharp
resonances of a Fabry-Perot resonator. This same concept is extended in Sections 3.5
and 3.6 to design an efficient, wide-bandwidth junction and a novel coupled resonator
waveguide that could be used as a delay line.
Figure 3.9(a) illustrates the basic geometry of the FDC which consists of a single-
mode input waveguide (region M1), a single-mode output waveguide (M3), and an
overlap region (M2), of length L, where the modes of the two guides couple. The
similarity to a directional coupler is obvious, but in addition to coupling behaviour, re-
flections from the closed waveguide ends result in resonant transmission characteristics.
We demonstrate that an ideal FDC structure that is less than 3µm long can be tuned
to have a notch-rejection width of less than 10 GHz, corresponding to Q > 104, and
with minor structural modifications, this can be increased to Q > 6× 105.
The concept of combining Fabry-Perot effects in a directional coupler geometry was
studied previously by Safaai-Jazi et al. [189] as a way to improve mode discrimination
and longitudinal mode spacing in resonant cavities. Although the FDC shares similari-
ties with the earlier device, there are also distinct differences. In Ref. [189] the resonant
cavity is formed by placing two partially reflecting mirrors across a pair of coupled
waveguides to form a four-port device. In contrast, the FDC is a two-port device since
one of the waveguides at each end of the cavity is blocked by a fully reflective photonic
crystal mirror, while the other waveguide is left open. This difference in the reflection
mechanism allows the FDC to exhibit complementary transmission behaviour to FP
devices. FP structures are normally highly reflective, only transmitting when a res-
onance condition is satisfied (notch-pass filter), whereas the FDC can be designed to
3.4. FOLDED DIRECTIONAL COUPLER 77
L
(a) (b)
M2
M3
M1
�������
����� �
������ �
����� � �
Figure 3.9: (a) Geometry of the FDC consisting of three sections M1, M2 and M3,
of identical cylinders. The shading indicates the three waveguide different sections.
(b) Schematics showing all of the propagating modes in the structure. The input and
output guides each support a single mode, while region M2 supports an even and an
odd supermode, |ψ±〉, indicated by the dashed and dotted curves respectively.
transmit at most frequencies and only reflect on resonance (notch-rejection filter).
In general, FDC structures exhibit a complicated and varied range of transmission
spectra that depend strongly on the structural parameters. Several examples are shown
in Fig. 3.10, where the transmission is plotted as a function of frequency for different
cavity lengths (a) and guide separations (b). These curves were calculated using the
full numerical formulation of the Bloch Mode method. Observe that there are several
common features in the spectra in Fig. 3.10 such as resonances where the transmission
vanishes, and regions where the transmission is flat and close to 100%. The solid
curves in both parts of Fig. 3.10 correspond to the same FDC with a cavity length of
L = 5d and guide separation of Nc = 1. This structure exhibits a sharp resonance at
d/λ = 0.331 where the transmittance drops from almost 100% to < 0.1% for a very
narrow range of frequencies. In Section 3.4.3 we use the semianalytic model derived
in Section 3.4.1 to identify the origin of this resonance, and find an expression for the
resonance width in terms of the waveguide dispersion parameters.
In Sections 3.4.1 and 3.4.2 we describe two alternative but equivalent derivations of
a simple semianalytic model for the FDC structure illustrated in Fig. 3.9(a), both of
which are based on mode coupling and symmetry arguments. In the first method, we
begin from the coupled mode approximation for two identical coupled waveguides (3.2)
and follow the propagation of an input mode through the FDC. The second derivation
is based on the Bloch mode scattering matrix formulation, which is first reduced to
include only the propagating modes, as in Section 3.3.2, then simplified further by
replacing the numerically calculated scattering matrices by idealised expressions. The
78 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
0.29 0.3 0.31 0.32 0.33
d/l
0
0.2
0.4
0.6
0.8
1
T
0.29 0.3 0.31 0.32 0.33
d/l
0
0.2
0.4
0.6
0.8
1
T
(a) (b)
Figure 3.10: FDC transmission spectra for (a) FDCs with Nc = 1 and cavity lengths
L = 4d (dashed), L = 5d (solid) and L = 6d (dashed-dotted) and (b) FDCs of length
L = 5d and guide separations of Nc = 1 (solid), Nc = 2 (dashed) and Nc = 3 (dashed-
dotted).
resulting model requires numerical input of only the waveguide dispersion properties
and the structure length to describe accurately the transmission properties of the FDC.
It also provides physical insight into many of the spectral features observed in Fig. 3.10.
We compare the results to full numerical calculations using the Bloch mode method.
3.4.1 Semianalytic model: coupling analysis
Consider first the modes in each section of the FDC shown in Fig. 3.9(b). The input
and output waveguides each support a single mode, |ψin〉 and |ψout〉, respectively, while
the coupling region supports even and odd supermodes |ψ±〉 with propagation constants
β±. As in Sections 1.7.5 and 3.2.1, we define ∆β = |β+ − β−|/2 and β = (β+ + β−)/2,
so that the coupling length Lc = π/(2∆β).
Recall from Section 3.2.1 that the supermodes of two identical coupled waveguides
can be approximated by linear superpositions of the individual waveguide modes using
Eq. (3.2). The comparison with rigorous numerical results in Fig. 3.1 clearly demon-
strates that this approximation remains accurate even for PhC waveguides separated
by a single line of cylinders and can therefore be applied to the modes in region M2
of the FDC. Since the input and output waveguides continue into region M2, the cou-
pled mode result provides a simple relationship between the modes in each of the three
regions. From Eq. (3.2) we can write
|ψin/out〉 = (|ψ+〉 ± |ψ−〉)/√
2, |ψ±〉 = (|ψin〉 ± |ψout〉)/√
2 . (3.5)
We now follow the propagation of a mode as it passes through the FDC. Light
entering from the input waveguide in mode |ψin〉 is coupled into the supermodes of
region M2 in the ratio given by Eq. (3.5), where we ignore back reflection at this first
3.4. FOLDED DIRECTIONAL COUPLER 79
interface since the input waveguide is continuous. The even and odd supermodes then
propagate a distance L through region M2, advancing in phase by factors exp(iβ+L)
and exp(iβ−L) respectively so that the field at the output interface is given by
Ψ(L)(1) =(eiβ+L|ψ+〉+ eiβ−L|ψ−〉
)/√
2 ,
=(eiβ+L + eiβ−L
)|ψin〉/2 +
(eiβ+L − eiβ−L
)|ψout〉/2 ,
= eiβL (cos(∆βL)|ψin〉+ i sin(∆βL)|ψout〉) , (3.6)
where we have used the definitions of β and ∆β to write β± = β±∆β. The superscript
(1) on the left hand side of equation (3.6) indicates that this is the first pass of the
light through M2. Notice that the modes of the two waveguides have a relative phase
difference of π/2 due to the mode coupling, while they share a common phase term
exp(iβL) which depends on the average propagation constant.
At the M2–M3 interface, the right-hand waveguide continues into M3 without in-
terruption, whereas the left-hand waveguide ends. We therefore assume that the field
component in mode |ψout〉 is transmitted with amplitude τ1 = i exp(iβL) sin(∆βL),
while the remaining field in mode |ψin〉 is reflected by the photonic crystal back into
the left-hand waveguide. For the purposes of this model, we neglect any phase change
that may occur when |ψin〉 is reflected, and so the reflected mode has an amplitude
of exp(iβL) cos(∆βL). Note that if the FDC was formed from conventional dielec-
tric waveguides, most of the light in the closed waveguide would be radiated into the
cladding and lost. The reflected field propagates back through region M2 to the first
interface, where it is again expressed in terms of the individual waveguide modes as
Ψ(0)(1) = e2iβL(cos2(∆βL)|ψin〉+ i cos(∆βL) sin(∆βL)|ψout〉
). (3.7)
Here, as at the output interface, the field in the open guide is transmitted into the input
waveguide with amplitude ρ1 = exp(2iβL) cos2(∆βL) while the field in the right-hand
waveguide is reflected back into region M2. We again see that both terms in Eq. (3.7)
have gained an additional factor of exp(iβL) and the individual waveguide modes are
again π/2 out of phase.
The total transmission and reflection amplitudes of the FDC, denoted τ and ρ
respectively, are found by repeating this calculation for each pass through M2 and
adding coherently the transmitted and reflected components, τn and ρn respectively,
80 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
which yields
τ = ieiβL sin(∆βL) + ie3iβL cos2(∆βL) sin(∆βL)
− ie5iβL cos2(∆βL) sin3(∆βL) + ie7iβL cos2(∆βL) sin5(∆βL)− . . .
= i exp(iβL) sin(∆βL)1 + exp(2iβL)
1 + sin2(∆βL) exp(2iβL)
(3.8a)
ρ = e2iβL cos2(∆βL)− e4iβL cos2(∆βL) sin2(∆βL)
+ ie6iβL cos2(∆βL) sin4(∆βL)− e8iβL cos2(∆βL) sin6(∆βL)− . . .
=cos2(∆βL) exp(2iβL)
1 + sin2(∆βL) exp(2iβ)L.
(3.8b)
Hence the reflectance is
R = |ρ|2 =cos4(∆βL)
cos4(∆βL) + 4 sin2(∆βL) cos2(βL), (3.9)
and the transmittance is simply T = 1−R. Note that the summation for τ in Eq. (3.8a)
is a geometric series from the second term onwards, whereas the summation for ρ in
Eq. (3.8b) is geometric series from the first term, as for a standard FP interferometer
[122]. The relative phase of each term is determined by the value of exp(iβL), which is
associated with a single pass through M2. We return to this property when considering
reflection resonances in Section 3.4.3.
Here we have derived a semianalytic model to describe the transmission properties
of the FDC that requires numerical input of only the modal dispersion properties in
region M2. These are calculated easily using the Bloch mode method or another PhC
calculation tool, and substituted into Eq. (3.9) to provide a very efficient formulation
to study FDC behaviour. The properties predicted by the model are discussed in
Section 3.4.3, but first we outline an alternative derivation of Eqs. (3.8) based on Bloch
mode scattering matrices.
3.4.2 Semianalytic model: Bloch mode matrix method
The FDC shown in Fig. 3.9(a) can be analysed using the Bloch mode method as a
three layer structure, and is therefore described by Eqs. (2.13) in terms of the Fresnel
matrices at each interface. Propagation from M1 to M3 is characterised by six Fresnel
matrices, R12, R21, R23, T12, T21 and T23. We showed in Section 3.3.2 that it is not
always necessary to retain the evanescent Bloch modes in the calculation if the interfaces
in the structure are sufficiently far apart. In this limit, accurate results can often be
obtained by including only the propagating modes in each section. This approach can
be applied to the FDC, however in contrast to the Fabry-Perot example, there are two
propagating modes in the central section, rather than one.
3.4. FOLDED DIRECTIONAL COUPLER 81
The truncated matrix dimensions are determined by the number of propagating
modes on either side of the relevant interface. For instance, R12 describes modes incident
and reflected in section M1, which only supports a single mode, hence R12 is a 1 ×1 matrix, and for the same reason, R21 and R23 are 2 × 2 matrices. T12 describes
transmission of the single mode in M1 into the two modes of M2, and hence it is a
column vector with two rows (2× 1). Similarly, T21 and T23 are 1× 2 row vectors. The
matrix Λ, which represents propagation through region M2, reduces to a 2×2 diagonal
matrix of eigenvalues µi = exp(iβjd), where j = 1, 2. When these reduced matrices are
substituted into Eq. (2.13b), the resulting transmission coefficient is a complex scalar
quantity.
While the transmission properties of the FDC are described accurately by such a
propagating mode treatment, the computational requirements are only modestly re-
duced since the full plane wave basis is still required to represent the Bloch modes.
We can however, determine approximate analytic expressions for the Fresnel matrices
using the same assumptions as in Section 3.4.1. At the first interface, |ψin〉 is perfectly
transmitted into |ψ±〉 with amplitudes given by Eq. (3.5) which gives
R12 = 0 , T12 =
[1/√
2
1/√
2
]. (3.10)
At the M2–M3 interface, the field in the right-hand guide is transmitted freely into the
output guide, so we can again use Eq. (3.5) to write
T23 = [ 1/√
2 −1/√
2 ]. (3.11)
The reflection matrix at the second interface is a 2 × 2 matrix, where element (i, j)
is the reflection coefficient for the reflection of mode j into mode i (travelling in the
opposite direction), and modes 1 and 2 are the even and odd supermodes, respectively.
Once again, assuming that a mode in the right guide is completely transmitted while a
mode in the left guide is completely reflected back into the left guide, we can write
R23 .
[1/√
2 1/√
2
1/√
2 −1/√
2
]=
[1/√
2 0
1/√
2 0
]⇒ R23 =
[1/2 1/2
1/2 1/2
]. (3.12)
Similar arguments are used to find the final two scattering matrices,
R21 =
[1/2 −1/2
−1/2 1/2
]and T21 = [ 1/
√2 1/
√2 ] . (3.13)
Equations (3.8) are derived by substituting these ideal Fresnel matrices into Eqs. (2.13)
and making the appropriate substitutions for β and ∆β. Although the result is iden-
tical to that derived using the coupled mode analysis in Section 3.4.1, the technique
described in this section provides a systematic approach that is more easily applied to
complicated structures. In addition, it is possible to improve the accuracy of the model
by replacing the approximate matrix entries derived above by more accurate estimates.
82 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
For instance, it is straightforward to include a small reflection at the first interface by
setting R12 6= 0 and making the appropriate modifications to T12 to maintain energy
conservation.
3.4.3 Transmission properties
In this section we investigate the transmission characteristics of the FDC and compare
the semianalytic model derived in Section 3.4.1 to full numerical calculations using the
Bloch mode method. The model is used to identify parameters for which the FDC
operates as a high-Q notch-rejection filter with a resonance width determined largely
by the mode splitting parameter ∆β. We also discuss methods of tuning the FDC
geometry to increase Q and reduced the device length.
The transmission spectra in Fig. 3.10 exhibit a number of high reflection and high
transmission features that we now identify using the model derived in Section 3.4.1.
Each of these features is driven by one of two physical effects; the features associated
with particular values of ∆βL result from coupling between the waveguides inM2, while
those features associated with particular βL values are caused by mode recirculation
in M2. We consider first the high transmission features. From Eq. (3.9) it can be seen
that all incident light is transmitted when cos(∆βL) = 0 which corresponds to L = Lc.
The physical explanation for this is clear from Fig. 3.9: light incident in mode |ψin〉is completely coupled into |ψout〉 at the end of M2 where it is transmitted into the
output waveguide with minimal reflection. This single-pass behaviour can be seen in
the field intensity plot of Fig. 3.11(a), calculated for a FDC with L = 4d and Nc = 1
at d/λ = 0.319. For these parameters, T > 99.9%. Provided that cos(βLc) 6= 0, the
reflection zero of Eq. (3.9) has a quartic dependence on cos(∆βL), resulting in the flat-
topped, high-transmission parts of the spectra in Fig. 3.10(a). The Nc = 2 and Nc = 3
curves in Fig. 3.10(b) do not exhibit this behaviour since the coupling is much weaker
and hence L < Lc.
If the cavity length is doubled to L = 2Lc, then all of the light is coupled back
into mode |ψin〉 by the end of M2, where it is reflected back through M2 and out
of the input guide. From Eq. (3.9) it is seen that L = 2Lc gives sin(∆βL) = 0, and
hence R = 1. Figure 3.11(b) shows the field intensity in a FDC with Nc = 1 and
L = 8d, calculated at d/λ = 0.317, for which T < 1%. The small frequency difference
is attributed to an additional phase shift introduced when the mode is reflected – recall
that we ignored any phase change on reflection by setting the reflection coefficient equal
to 1 in Section 3.4.1.
For the purpose of designing a notch-rejection filter, we are interested in the second
condition for which R = 1 in Eq. (3.9), namely when cos(βL) = 0, or exp(iβL) = ± i.Unlike the transmission and reflection features that we have just discussed, this reflec-
tion resonance is a result of mode recirculation through the double guides in M2. To
see this, consider again the series for ρ and τ in Eqs. (3.8a) and (3.8b) which corre-
spond to the transmitted and reflected field amplitudes on each pass through section
3.4. FOLDED DIRECTIONAL COUPLER 83
0 0.2 0.4 0.6 0.8 1
Field intensity (a.u.)
(a) (b)
Figure 3.11: (a) Field intensity in a FDC of length L = 4d at d/λ = 0.319, for which
the reflected power is less than 0.1%. (b) Field intensity for a FDC of length L = 8d at
d/λ = 0.317d, for which R > 99.9%. The geometry of the two devices is identical aside
from the length.
M2. When the resonance condition is satisfied the field transmitted on the first pass,
τ1 is 180◦ out-of-phase relative to transmitted fields on each subsequent pass. On the
other hand, the reflection terms ρn are all in-phase and hence they combine construc-
tively to produce the resonant reflection. This resonance is the cause of all of the local
transmission minima seen in Fig. 3.10 since L < 2Lc in each example and hence the
first R = 1 condition identified above is not satisfied. Observe that the resonance width
varies considerably for the different structures, but appears to be narrowest when the
transmission on either side of the resonance is close to 1. This is not a coincidence, as
we now show.
Suppose that the resonance occurs at a frequency ω = ω0, and denote the values of
∆β and β at resonance by ∆β0 and β0. For frequencies ω = ω0 + δω close to ω0, ∆β
and β are expanded to first order as
∆β = ∆β0 + δω∆β1 , β = β0 + δωβ1 , (3.14)
where
∆β1 =
(d∆β
dω
)ω=ω0
and β1 =
(dβ
dω
)ω=ω0
.
Since we are interested in the resonance related to the zeros of cos(βL), we set β0 = nπ/(2L)
for n = 1, 3, 5, . . . and substitute Eqs. (3.14) into Eq. (3.8b) to obtain
ρ =−1
1− 2iδω/∆ω, R =
1
1 + (2δω/∆ω)2, (3.15)
84 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
where R = 1/2 at ω = ω0 ±∆ω/2 and
∆ω =cos2(∆β0L)
β1L. (3.16)
Note that Eqs. (3.15) describe a classic Lorentzian reflection resonance, as expected for
a simple resonant cavity [113,190].
Our earlier observation that the resonance is narrowest when the surrounding trans-
mission is high is verified by Eq. (3.16), since bothR→ 0 and ∆ω → 0 as cos(∆β0L) → 0.
This is a desirable feature for notch-rejection filters which ideally transmit all frequen-
cies ω 6= ω0. We note also that the β1 term in Eq. (3.16) is a function of the group
velocities of |ψ+〉 and |ψ−〉. Since β is well approximated by the dispersion of a single
waveguide for frequencies well above cutoff (see Section 3.2.1), we have β1 ≈ 1/vg. Thus
Q is approximately inversely proportional to the group velocity of the waveguides at the
resonant frequency. A similar dependence is displayed by other coupled-cavity designs,
and this has been used to improve their resonant characteristics [83, 191]. Equations
(3.9),(3.15) and (3.16) also show that the resonant frequency for a FDC of length L is
determined largely by β, whereas the resonance width depends strongly on ∆β. Since
∆β is more sensitive than β to changes in the coupling conditions, it is therefore possible
to tune the resonance width by modifying the structure of region M2, for instance by
changing the waveguide separation or modifying the cylinders between the two guides.
An example of the latter method is given in Section 3.4.4, where we demonstrate tuning
of Q by almost four orders of magnitude.
The results of Eqs. (3.9) and (3.15) (dashed curves) are compared to the full nu-
merical calculations (solid curves) in Figs. 3.12(a) and (b) for a FDC of length L = 5
and Nc = 1. The dashed curve in (a) is given by Eq. (3.9) with L = 6.067d, which was
chosen to give the best agreement with the numerical result. Although L was defined
in Fig. 3.9 as the number of PhC layers forming region M2, this is not necessarily
the most accurate value to use in Eq. (3.9) since the waveguide ends act as distributed
reflectors and hence the phase change on reflection can be relatively large. This effect
was neglected in Section 3.4.1, but can be partly corrected for by adjusting L. In many
cases a fixed L value gives reasonable agreement for a small range of frequencies, as can
be seen in Figs. 3.12(a) and (b), but this does not take into account the wavelength
dependence of the phase change. Attempts to include a wavelength-dependent length
correction by considering the phase of the reflected mode have proved no more accurate
than the simple result, suggesting that there are additional physical effects that are not
accounted for in our model. In most cases, the length adjustment necessary for the best
agreement with numerical results is approximately one lattice period or less.
The transmission spectrum in Fig. 3.12(b) is a high-resolution plot of the sharp
reflection resonance observed at d/λ = 0.33107 in (a) where the resonant reflection
exceeds 99.95%. For operation at λ = 1.55µm, corresponding to d = 513 nm, the
length of M2 is only L = 5d = 2.6µm or 1.7 wavelengths, and the resonance width is
approximately 0.11 nm. From the resonance we calculateQ = 1.3×104 for the numerical
3.4. FOLDED DIRECTIONAL COUPLER 85
0.31 0.315 0.32 0.325 0.330
0.2
0.4
0.6
0.8
1
d/l
T
0.2
0.4
0.6
0.8
1
0
T
d/l
(a)
(b)
(c)
105
100
102
103
Field intensity (a.u)
104
101
0.33102 0.33106 0.33110 0.33114
Figure 3.12: (a) Transmission spectrum as a function of normalised frequency for an
un-optimised FDC of length L = 5d and Nc = 1. The solid curve in (a) and (b) is
the full numerical result and the dashed curve is given by Eq. (3.9) with L = 6.066.
(b) High resolution plot of the resonance in (a) centred at d/λ = 0.33107. (c) Field
intensity in the FDC on resonance. Note that the colour scaling is logarithmic and
spans five orders of magnitude.
result (solid curve) and Q = 2.4 × 104 for the approximate result (dashed curve). As
expected, there is a strong field enhancement in regionM2 when the resonance condition
is satisfied, as shown in Fig. 3.12(c), where the field intensity on resonance is plotted
on a log scale. The peak intensity is more than 4 orders of magnitude greater than the
input intensity, which is consistent with the Q of the resonance. Despite the factor of
two difference in the peak width, the simple model predicts the broader characteristics
of the FDC with reasonable accuracy, and as we show in Section 3.4.4, it can be used
as an efficient aid to device optimisation.
3.4.4 High-Q tuning
In this section we demonstrate fine-tuning of the FDC response by changing the radius
of the cylinders that separate the two guides in region M2. The resonance width is
varied by almost four orders of magnitude to obtain an optimised FDC design of length
L = 4d that exhibitsQ > 6×105. The change inQ is predicted accurately by Eq. (3.16).
We begin with the FDC illustrated in Fig. 3.11 (a), which has a length of L = 4d.
86 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
0.312 0.314 0.316 0.318 0.32
0.2
0.4
0.6
0.8
1
00.31
d/l
T
0.26 0.28 0.3 0.32 0.3410
103
104
105
106
102
Q
rc
Bloch mode method (L=4d)Model (L = 4.83d)Model (L = 4.90d)
r =0.25dc
r =0.26d
r =0.27d
r =0.28d
r =0.29d
r =0.30d
r =0.31d
r =0.32d
r =0.33d
r =0.34d
r =0.35d
c
c
c
c
c
c
c
c
c
c(a)
(b)
Figure 3.13: (a) Transmission spectra for a FDC of length L = 4d as the cylinder radius
between the two waveguides (rc) is changed between 0.25d and 0.35d. The black curve
corresponds to the original structure with rc = 0.34d. (b) Variation of Q with rc. The
black points correspond to the curves in (a), while the red and blue points are given by
Eq. (3.16) for two different value of L.
The transmission spectrum for the unmodified structure is indicated by the dashed line
in Fig. 3.10(a) and exhibits a resonant reflection close to d/λ = 0.315 with Q = 1300.
The same curve is shown in Fig. 3.13(a) for a narrower range of frequencies, where
the black curve corresponds to the original FDC with a cylinder radius of r = 0.3d.
The remaining curves in Fig. 3.13(a) show the transmission spectrum as the radius of
the four cylinders between the guides is varied between rc = 0.25d and rc = 0.35d.
Observe that the resonant frequency decreases with increasing radius, while the peak
width initially narrows to a minimum at rc = 0.27d, then broadens again.
Figure 3.13(b) is a plot of Q as a function of rc, where the black circles are calculated
directly from the transmission spectra in (a). Observe that Q varies by almost four
orders of magnitude, and reaches a maximum of Q = 6× 105 when rc = 0.27d. In this
3.4. FOLDED DIRECTIONAL COUPLER 87
105
100
102
103
Field intensity (a.u)10
410
610
1
Figure 3.14: Resonant field intensity in the optimised FDC with L = 4d and Nc = 1.
The four central cylinders have a radius of rc = 0.27d: 0.03d smaller than the other
cylinders. Note that the colour scaling is logarithmic and spans six orders of magnitude.
purely 2D geometry, Q can be made arbitrarily large in principle by adjusting rc to
match the ∆ω = 0 condition given by Eq. (3.16). The red and blue points plotted in
Fig. 3.13 are given by Eq. (3.16) for L = 4.83d and L = 4.90d respectively, where ∆β0
and β1 have been calculated for each value of rc. The choice of L = 4.83d gives the best
agreement between Eq. (3.9) and the rc = 0.3d curve in Fig. 3.13, whereas L = 4.9d
provides the best agreement with the numerically calculated Q values.
The resonant field intensity inside the optimised FDC is shown in Fig. 3.14. Observe
that the peak intensity inside region M2 is more than five orders of magnitude larger
than the incident field, clearly illustrating the high-Q properties of the FDC.
The results in this section provide a convincing demonstration of how the semian-
alytic model can be used to great advantage when optimizing design parameters. In
this example, the results are summarised by the plot in Fig. 3.13(b) which indicates the
variation of Q as a function of rc. Each black point in (b) corresponds to a single curve
in (a), each of which must be calculated with enough points to resolve the resonance.
The curves in (a) each consist of approximately 250 points, and many more would be
required if the frequency grid was uniform. One data point takes approximately 17 s to
88 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
calculate using the Bloch mode method on a Pentium 4 (2.8GHz) desktop computer,
which gives a total computation time of 13 hours to generate the 11 curves in Fig. 3.13.
In contrast, the blue and red data points in Fig. 3.13(b) are given by Eq. (3.16), for
which we only need to calculate the coupled waveguide dispersion curves for each value
of rc. Since β and ∆β are slowly varying functions of frequency only 11 data points
were used to calculate the dispersion curves and then a polynomial function was fitted
to allow interpolation and differentiation. Each point takes approximately 7 s to cal-
culate, so the total calculation time for the 11 different rc values was approximately
810 s – fifty-five times faster than the full calculation. Once the dispersion data was
calculated, the red and blue data points in Fig. 3.13(b) were generated in a few seconds
from Eq. (3.16).
Although the accuracy of the model depends on the choice of L, this can be esti-
mated using a number of methods. For instance, the red points in Fig. 3.13(b) were
calculated using the L value that gave the best match between Eq. (3.9) and the original
spectrum. Even if this meant one full transmission spectrum was required, the reduc-
tion in computation time is still significant. Approximate methods do not replace full
numerical calculations, which would still be required for the final optimization process,
however they can be used to restrict the computational demand to a minimum.
3.4.5 Comparison with side-coupled cavity
Although it is not immediately obvious from the geometry, the FDC is closely related
to other waveguide-coupled cavities such as those in Refs. [115, 120, 121, 192], in which
a short line-defect is coupled by evanescent fields to a straight PhC waveguide as il-
lustrated in Fig. 3.15(a). A number of semianalytic theories have been developed for
describing the coupling in this type of structure based on tight-binding approximations
and scattering-theory [113,190,193,194]. Here we show that the simple mode-coupling
analysis developed in Section 3.4.1 also provides an accurate approximation for simple
side-coupled cavity geometries. A feature of these designs is that the in-plane Q, in-
creases exponentially with the distance between the cavity and the waveguide [192]. In
contrast, it is clear from Fig. 3.10 that the FDC resonances start to disappear as the
waveguide separation increases. To understand this difference we consider the simple
side-coupled cavity (SCC) design illustrated in Fig. 3.15(a), which resembles the FDC
in that it consists of single input and output waveguides coupled into a section of double
waveguide. This structure can be modelled using the same semianalytic method applied
to the FDC in Section 3.4.1. The only difference in the analysis is that the light in the
right-hand waveguide is reflected at the second interface of the SCC, whereas in the
FDC the light in the left-hand waveguide was reflected. The equation for the reflected
power of the SCC is thus found to be
R =sin4(∆βL)
sin4(∆βL) + 4 cos2(∆βL) sin2(βL), (3.17)
3.4. FOLDED DIRECTIONAL COUPLER 89
2 3 4 5 6 710
2
103
104
105
106
Nc
Q
L
Nc
(a) (b)
Figure 3.15: (a) Diagram of the side-coupled cavity geometry considered in Sec-
tion 3.4.5. The example shown here is for Nc = 6. (b) Plot of Q as a function of
Nc for a cavity with L = 2, as in (a). Black points are calculated using the Bloch mode
method and the red points are given by Eq. (3.18) for L = 2.882.
which is identical in form to Eq. (3.9) but with the cos(∆βL) and sin(∆βL) terms
swapped and sin(βL) in place of cos(βL). In analogy to the FDC, the resonance of
interest occurs in the SCC for sin(βL) = 0 and has a width approximated by
∆ω =sin2(∆β0L)
β1L, (3.18)
which is the equivalent of Eq. (3.16).
Let us now compare Eqs. (3.16) and (3.18) to identify the differences between the
FDC and the SCC. For both geometries, the position of the resonance is determined by
βL whereas the resonance width depends on ∆βL and β1L. Recall from Section 3.4.3,
that the maximum Q of the FDC occurs when cos(∆βL) ≈ 0, and thus we were able
to tune the Q in Section 3.4.4 by modifying the coupling conditions, and hence the
value of ∆β in the cavity region. Similarly, to maximise the Q of the SCC, we want
sin(∆βL) = 0 in Eq. (3.18). This can be achieved in one of two ways. The first is to
design the cavity such that L = 2mLc, where m is an integer, and hence ∆βL = mπ,
but this requires a cavity twice the length of the equivalent FDC. The second option is
to let ∆β → 0, which can be done by increasing the waveguide separation. This latter
approach for tuning Q is the most common technique used in the literature (see for
example Ref. [192]).
While we have only considered a basic SCC geometry, it is interesting to compare
the relative sizes of a FDC and a SCC with the same Q. In Section 3.4.4 we designed a
FDC with L = 4d and Q = 6× 105. This resonance corresponded to βL = 3π/2 – the
second zero of cos(βL). A shorter FDC would not have been possible in this geometry,
since we require L = Lc in order to maximise Q, and hence the shortest length is
determined by the maximum ∆β that can be achieved. The SCC does not have the
same length limitation since Q can always be increased by decreasing ∆β. Hence, the
90 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
first practical resonance occurs when βL = π. This condition is satisfied for d/λ ≈ 0.316
for a cavity of L = 2d as illustrated in Fig. 3.15(a). Figure 3.15(b) is a plot of Q as
a function of Nc for Nc = 1 − 7 where the black points were calculated using the full
Bloch mode method and the red points are given by Eq. (3.18) with L = 2.882d. This
value of L was chosen so that the resonant frequency predicted by Eq. (3.17) matched
the numerical results for Nc = 4. As in the FDC example, the semianalytic model
provides an accurate estimate of the resonant properties of the cavity.
Despite the exponential increase of Q with Nc, the SCC still requires Nc > 6 for
Q > 105, which corresponds to a distance of more than 7d between the cavity and
the waveguide. This is approximately twice the size of the FDC design considered
in Section 3.4.4. Although a rigorous comparison of relative sizes would include the
area of PhC required around each design to confine the light in the plane, the simple
comparison here suggests that the FDC compares favourably with alternative high-Qfilter designs.
3.4.6 Discussion
In this section we have demonstrated a novel filter design that exhibits high Q reso-
nances in a compact geometry. Since it is essentially an in-line device it can be easily
coupled to, and the resonance properties can be optimised by modifying the geometry
of the coupling region. As with any resonant cavity, however, the control of out-of-
plane losses is critical if such high-Q values are to be achieved in practice. Although we
have not considered losses in our analysis, it is expected that low-loss designs could be
optimised in much the same way as the cavities discussed in Section 1.7.1. We return
to this issue in Chapter6.
The comparison between the FDC and a simple side-coupled cavity filter in Sec-
tion 3.4.5 shows that the two devices have similar resonance characteristics when oper-
ated as narrow-band filters, but we have not compared other functions such as add-drop
behaviour or nonlinear switching characteristics. A more detailed study of these prop-
erties would be required in order to establish the relative merits of the two geometries.
Although we do not consider the structure here, we note that the modification pro-
posed by Fan [195] to achieve asymmetric line shapes in side-coupled cavities can also
be applied in the case of the FDC.
In addition to the high-Q resonances studied in this section, the transmission spectra
in Fig. 3.10(a) also exhibit regions of flat, almost perfect transmission. This character-
istic behaviour is exploited in Section 3.5 to design an efficient wide-bandwidth sym-
metric Y-junction that is incorporated into a PhC-based Mach-Zehnder interferometer
in Chapter 4. We also study the transmission properties and band structure of a chain
of identical FDCs in Section 3.6.
3.5. COUPLED Y-JUNCTION 91
3.5 Coupled Y-junction
Efficient, wide-bandwidth Y-junctions or beam splitters are required for compact inte-
gration of multiple optical devices on photonic chips. As a basic component of many
integrated optical devices, these junctions must be designed with a minimum back
reflection and maximum transmission over the operating bandwidth. A number of
studies have been made into the optimal method of splitting a single guided mode into
two modes, and the approaches taken can be broadly classified into two groups. The
first design approach has been to optimise simple Y-junction designs, while the sec-
ond approach has been to design novel alternative structures exploiting the properties
of photonic crystals. Good transmission has been demonstrated in Y-junctions both
theoretically [77,196–198], and experimentally [198,199] using optimisation techniques
such as placing one, or several, “tuning” cylinders near the junction of three wave-
guides. High transmission junctions, with calculated transmissions up to 99% have
been designed, however in most cases, the transmission bandwidth decreases rapidly
as the maximum transmission approaches 100% [77]. Typical bandwidths reported in
the literature for hole-type PhC junctions are on the order of 10 − 40 nm for 95%
transmission.
An alternative to the Y-junction is essentially a directional coupler with L = Lc/2
so that light incident in one of the guides is split evenly between the two guides. Such
a structure was proposed by Martinez et al. [200] as a means of splitting an input
beam for a Mach-Zehnder interferometer application. The optimal coupling length is
wavelength dependent, but may be robust enough to give a satisfactory bandwidth.
One of the big advantages of a coupler-based beam splitter is that the back reflection
can be negligible. If the coupler is not perfectly matched to the wavelength, the splitting
between the output guides changes, but the total transmission of the splitter remains
high. This is an important issue in compact devices where back reflections can cause
unwanted interference effects. A modified version of the directional coupler junction
was demonstrated by Sugimoto et al. [109] in which a cavity was placed between the two
waveguides to create a similar geometry to the coupled cavity add-drop filter illustrated
in Fig. 1.9(a). Couplers of this type have been successfully integrated into PhC Mach-
Zehnder interferometers [201].
Here, we present a novel coupled Y-junction design which operates similarly to
a directional coupler junction, but with the symmetry properties of a Y-junction that
ensure the transmission into each output arm is identical even when the coupling length
is not perfectly matched to the wavelength. In addition, recirculation inside the junction
results in high transmission over a wide bandwidth. We show that the junction is
closely related to the folded directional coupler, and a modal analysis similar to that in
Section 3.4 yields identical expressions for the transmission and reflection. While similar
structures have been studied in dielectric waveguides [202, 203], the PhC-based device
gains a significant advantage from the perfectly reflecting cladding which provides mode
recirculation that increases the transmission bandwidth.
92 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
L
(a) (b)
�������
������
���� �����
����� � �
Figure 3.16: (a) Geometry of the coupled Y-junction. (b) Schematics of the symmetric
propagating modes in the three waveguide sections.
The general geometry of the coupled Y-junction is illustrated in Fig. 3.16(a). A
single input guide enters a coupling region of length L consisting of three identical
waveguides. At the output interface, the central waveguide ends and the two remaining
waveguides continue as outputs. Light entering the triple guide region in the central
waveguide is coupled into the two guides on either side. If the length of the triple guide
region is such that all of the light has been transferred into the outer guides at the end
of the triple guide region, then almost all light propagates into the two output guides.
If the output guides are sufficiently far apart, the coupling between them is negligible.
In the junction shown in Fig. 3.16 the output guides are separated by five layers of
cylinders so the coupling length is several hundred lattice periods. Hence, for compact
devices the guides can be treated as isolated from each other.
A semianalytic model for the coupled Y-junction is derived in the same way as
the FDC model in Sec. 3.4.1, and we show here the two structures are described by
an identical set of equations. We must first consider the propagating modes within
each of the three regions. The input guide supports a single even mode, the output
guides support an odd and an even mode, and the central triple guide section supports
three modes - two even and one odd - as discussed in Section 3.2.2. However, since
the structure is symmetric and the input waveguide only supports a single even mode,
light can only couple to the even modes in each section. Hence, the single input mode
couples into two modes in the central region, which in turn couple into a single output
mode, in much the same way as in the FDC. Schematics of the even modes in each
region of the coupled Y-junction are shown in Fig. 3.16(b).
Recall from Section 3.2.2 that the supermodes of the triple guide region can be
approximated by linear superpositions of the individual waveguide modes. Hence we
3.5. COUPLED Y-JUNCTION 93
can use Eq. (3.3) to write
|ψ1〉 = (|ψL〉+√
2 |ψin〉+ |ψR〉)/2 = (|ψin〉+ |ψout〉)/√
2,
|ψ3〉 = −(|ψL〉 −√
2 |ψin〉+ |ψR〉)/2 = (|ψin〉 − |ψout〉)/√
2,
(3.19)
where |ψL〉, |ψin〉 and |ψR〉 are the modes of the left, centre (input) and right waveguides
respectively, and |ψout〉 = (|ψL〉+ |ψR〉)/√
2 is the even mode of the output waveguides.
As in the case of the FDC, the propagation constants of |ψ1〉 and |ψ3〉 are expressed as
β1 = β + ∆β and β3 = β −∆β, where β = (β1 + β3)/2 and ∆β = |β1 − β3|/2.
Equations (3.19) reveal the direct mapping between the coupled Y-junction and
the FDC; |ψ1〉 and |ψ3〉 are equivalent to |ψ+〉 and |ψ−〉 in Eq. (3.5) while the input
and output modes of the two structures are also equivalent. Hence, the transmission
properties of the coupled Y-junction are described by Eqs. (3.8) and (3.9) with β and
∆β calculated for the even modes of the triple guide section.
Although the coupled Y-junction also exhibits the sharp resonances of the FDC, it
is the flat, high-transmission regions of the spectrum that we wish to exploit for the
purposes of this device. In Section 3.4.3 we identified this feature with the condition
cos(∆βL) = 0, which corresponds to L = Lc. When this condition is satisfied, all of the
incident light is coupled from the central input waveguide into the output waveguides
on the first pass through the triple guide section, and so the transmission is high.
Since the reflectance given by Eq. (3.9) is quartic in cos(∆βL), the high transmission
is typically flat-topped and broad. In order to maximise the transmission bandwidth,
it is desirable to have cos(βLc) ≈ 1 so that the reflection resonances are far from the
maximum transmission frequency, although reasonable bandwidths are possible without
this condition, as demonstrated in Section 4.4.1.
Figures 3.17(a) and (b) show the transmission through two different coupled Y-
junctions, with lengths L = 5d and L = 7d respectively. The junction is formed in
the PC1 lattice, with the output guides separated by 5 rows of cylinders, and each of
the guides in the central region separated by 2 rows of cylinders. The lattice period
is d = 491 nm in both examples, corresponding to a cylinder radius of 173 nm. For
lengths of L = 5d and L = 7d, the Y-junction has 95% transmission bandwidths of
approximately 92 nm and 83 nm, respectively at λ = 1.55µm, and 99% transmission
bandwidths of about 80 nm, and 76 nm. The dashed curves in Figs. 3.17(a) and (b) are
the semianalytic result (3.9) calculated with L = 5.75d and L = 7.80d respectively for
the two devices. The solid curves show the transmission calculated using the full Bloch
mode method. As in the examples of Section 3.4, the difference between the length
defined in Fig. 3.16 and the length parameter in the model results from phase changes
upon reflection, and the ambiguity in defining a reflection plane at either end of the triple
guide region. The semianalytic model was used to design the two junctions, and it can
be seen that the transmission bandwidth obtained from the full Bloch mode calculation
exceeds the bandwidth predicted by the model. In Section 4.4.1 we present two other
coupled Y-junction designs - one in a rod-type PhC and the other in a triangular
94 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
0.2
0.4
0.6
0.8
1
T
1.48 1.5 1.52 1.54 1.56 1.58 1.6
0.2
0.4
0.6
0.8
1
T
92nm
83nm
λ (µm)
(a)
(b)
L = 5d
L = 7d0
Figure 3.17: Transmission curves for the coupled Y-junction with (a) L = 5d and (b)
L = 7d. Wavelengths have been scaled so that the high transmission bands occur near
λ = 1.55µm. The 95% transmission bandwidths are 92 nm and 83 nm respectively. The
solid curves are calculated using the full Bloch mode method, and the dashed curves
are the semianalytic result given by Eq. (3.9) for L = 5.75d and L = 7.80d.
lattice hole-type PhC. The hole-type PhC junction has a 95% transmission bandwidth
of 39 nm, which compares favourably with alternative junction designs reported in the
literature [77,197,198]. It is expected that the bandwidths could be further increased by
tuning the structure of the triple guide region as demonstrated in Section 3.4.4 for the
FDC. We note that since this work was reported in Ref. [204], similar junction designs
have been proposed by other authors [205, 206], however the structures we present
here exhibit superior bandwidth and transmission properties. There have also been
several experimental demonstrations of efficient junction designs obtained via numerical
optimisation of the structure [199,207].
When designing the coupled Y-junction for maximum bandwidth, the splitting of
the modes in the triple guide region must be considered. Since the successful operation
of the device requires two propagating even modes to exist in this region, we must
operate in the frequency range above the cut-off of the second even mode, and below
3.6. SERPENTINE WAVEGUIDE 95
the high-frequency edge of the bandgap. If the three guides are too close together,
the mode splitting becomes very large, and the frequency range for operation becomes
limited. This is clearly seen in the triple waveguide dispersion curves for Nc = 1
in Fig. 3.4. On the other hand, if the guides are too far apart, the coupling length
becomes prohibitively large. Thus, the choice of guide separation and its effect on
mode splitting must be considered. In the examples chosen above, the second even mode
cutoff occurs at λ ≈ 3.32d, giving a maximum bandwidth of ∆λ = 0.33d corresponding
to approximately 160 nm at λ = 1.55µm. The transmission bandwidth can also be
increased by operating at frequencies where the dispersion curves of all three modes
are approximately parallel, such that ∆β and hence Lc are very weak functions of
wavelength [205].
The coupled Y-junction design provides a novel solution to the problem of achieving
high-transmission, wide-bandwidth beamsplitting, by combining the efficient coupling
properties of a directional coupler with the symmetrical splitting of more conventional
Y-junction designs. Such properties make this junction ideal for use in PhC-based
Mach-Zehnder interferometers where high performance junctions are required. We
demonstrate this application in Chapter 4. While the 2D results presented here are
promising, further work is required to give an accurate comparison with alternative
Y-junction designs, not only in terms of transmission and bandwidth, but also regard-
ing fabrication tolerance. One issue of particular interest is the sensitivity to structural
variations that break the symmetry properties of the junction. We have only considered
the even modes in our analysis, however any breaking of the symmetry allows light to
also couple into the odd modes. Although this is likely to result in unequal splitting
between the two output waveguides, it is not clear whether the back reflection would
increase significantly.
3.6 Serpentine waveguide
In this section we consider the serpentine waveguide – a period coupled-cavity waveguide
formed by cascading multiple FDC structures as shown in Fig. 3.18. Periodic coupled
resonators such as coupled-resonator optical waveguides (CROWs) and coupled ring
resonators have been studied previously for a range of applications, including add-drop
filtering, slow-light propagation, and nonlinear switching [101, 102, 208–213]. The ser-
pentine waveguide is of particular interest as a slow-light structure due to the existence
of very low group velocity bands as we demonstrate here. In Section 3.6.1 we derive a
dispersion relation for the serpentine waveguide in terms of the transmission and reflec-
tion properties of a single FDC. The dispersion properties of the periodic structure can
thus be evaluated using either the full Bloch mode method or the semianalytic expres-
sions in Eq. (3.8). We calculate the band structure and transmission properties of two
different serpentine waveguides in Section 3.6.2 and discuss their general properties.
96 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
L1
L2
L1
L /22
Dy
y3
y2
y1
L /22
(a) (b)
Figure 3.18: Two equivalent serpentine waveguide geometries (assuming no tunnelling
through the guide ends). Both are characterised by the period Dy and the double and
single guide lengths L1 and L2.
3.6.1 Derivation of the dispersion relation
The unit cell of the serpentine waveguide is a pair of coupled FDC devices joined input-
to-output by a single waveguide of length L2 to create a periodic structure with a period
Dy = 2(L2 + L1) as shown in Fig. 3.18(a). Each of the individual components of this
structure is essentially the same as those of the isolated FDC device. In the sections of
length L1 there are two parallel waveguides that support a pair of propagating modes
which allow light to couple from one guide to the other. The single guide sections of
length L2 act as the input/output guides to the next/previous coupling region. If L2
is sufficiently long that evanescent tunnelling between the blocked waveguide ends is
negligible, the only connection between the coupling regions is via the one propagating
mode of the single guide. In this limit, the structures shown in Figs. 3.18(a) and (b) are
equivalent and hence a single FDC, rather than a complete unit cell, is the minimum
segment of the serpentine required to determine the band structure. As we show in
Section 3.6.3, this property holds for the structure in Fig. 3.18(a) even when the single
mode approximation is invalid.
We now proceed to analyse the structure using a single mode approximation which is
3.6. SERPENTINE WAVEGUIDE 97
valid when the L2 is sufficiently long to ensure coupling between the double waveguide
segments in Fig. 3.18 occurs only via the propagating mode in the single waveguide
section. Consider first the transmission and reflection characteristics of a single period
of the serpentine waveguide, bounded by the interfaces y1 and y3 shown in Fig. 3.18.
Using the notation of Chapter 2 we can express the fields at interface yj in terms of
their upward and downward components b+j and b−j respectively. In the propagating
mode analysis here, the b±j are complex amplitudes of the single propagating waveguide
mode, which are related via a transfer matrix according to[b−3b+3
]= Ts
[b−1b+1
]. (3.20)
Denoting the complex transmission and reflection amplitudes of a single period by
τs and ρs respectively, we can write down the transfer matrix
Ts =
[τs − ρ2
s/τs ρs/τs−ρs/τs 1/τs
]. (3.21)
Since the basic unit of the serpentine is the FDC bounded by y1 and y2 in Fig. 3.18,
we can rewrite Eq. (3.20) as
Ts = T 2f , where Tf =
[τf − ρ2
f/τf ρf/τf−ρf/τf 1/τf
](3.22)
is the transfer matrix for a FDC with input and output guides of length L2/2. Here,
τf and ρf are the transmission and reflection coefficients of a FDC with an additional
phase term exp(iβL2) to include propagation between the FDC and interfaces y1 and
y2. β is the propagation constant of the single input and output mode. Note that a
more general relationship between Ts and Tf , valid for all L2, is derived in Section 3.6.3.
The dispersion relation for the periodic serpentine structure is obtained by solving
the eigenvalue problem Tsb = µb, where the eigenvalues are µ = exp(iqDy) and q is
the Bloch factor in the longitudinal direction of the serpentine. Since Ts = T 2f , the
eigenvalue equation can be written as Tfb = ±µ1/2b, which has solutions that satisfy
µ+ 1
∓µ1/2=τ 2f − ρ2
f + 1
τf. (3.23)
This expression can be further simplified using the energy conservation properties
|ρf |2 + |τf |2 = 1 and arg(τf )− arg(ρf ) = ±π/2, to yield the dispersion relation
∓ cos(qDy/2) = <(1/τf ). (3.24)
The right hand side of Eq. (3.24) can be calculated using either the full Bloch mode
method or the semianalytic expression in Eq. (3.8), with τf = τ exp(iβL2) to include the
additional phase due to propagation through the single waveguide sections. The latter
98 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
method results in a semianalytic form of the right hand side, which can provide a better
understanding of the serpentine waveguide properties. Substituting the semianalytic
expression for τf into Eq. (3.24) yields
cos(qDy/2) = sin(∆βL) sin(βL1 + βL2) +cos(∆βL1) sin(2βL1 + βL2)
2 sin(∆βL1) cos(βL1), (3.25)
which can be evaluated after numerical input of the waveguide dispersion properties
and the lengths L1 and L2. In the next section we consider the properties predicted by
this dispersion relation and compare the results to full numerical calculations.
3.6.2 Serpentine waveguide properties
We consider first the different solutions to Eq. (3.25), and identify band features that
are directly related to the properties of a single FDC, and others that result from the
periodicity. The serpentine waveguide exhibits transmission bands for frequencies where
the right hand side of Eq. (3.25), <(1/τf ), has magnitude less than unity. Correspond-
ingly, band gaps, which lie between the transmission bands, occur when |<(1/τf )| > 1.
There are three distinct types of band gap, each originating through a distinct physical
effect. Following from Eq. (3.25), or from the analysis of a single FDC in Section 3.4.3,
there are two conditions for which τf = 0. The first occurs when sin(∆βL1) = 0, which
corresponds to L1 being an integer multiple of the coupling length Lc. The second
condition occurs when cos(βL) = 0, corresponding to the reflection resonance identified
in Section 3.4.3 as being suitable for narrow-band filtering. When either of these con-
ditions is satisfied, the right hand side of Eq. (3.25) diverges, and a band gap appears
in the serpentine band diagram. Similar resonator band gaps are displayed by coupled
ring-resonator structures [211] and other periodic coupled resonator systems [193]. We
label the two types of resonator band gaps identified here as RG1 and RG2 respectively.
The third class of band gap that appears in the serpentine band diagram is a Bragg
gap [211] that occurs as a result of the overall periodicity of the serpentine structure,
rather than a single feature of the FDC.
Figure 3.19(a) shows the band diagram of a serpentine waveguide with L1 = L2 = 5d,
formed in the square rod-type PhC, PC1. The FDCs making up this device are identical
to the L = 5d structure shown in Fig. 3.12(c), which has the transmission properties
plotted in Fig. 3.12(a) and (b). The very narrow RG2-type band gap near d/λ = 0.33
corresponds to the sharp reflection resonance of the FDC. In Fig. 3.19(b), the intensity
transmission is shown for the single FDC (half a period), a single whole period and two
whole periods of the serpentine waveguide. The band diagram was calculated using
τf obtained from the full Bloch mode method and the transmission spectra were also
calculated using the full numerical approach. Counting from the top of Fig. 3.19(a),
we see that top two resonator gaps are both of type RG2, and hence they correspond
to strong features in all three of the transmission curves. In contrast, the Bragg gaps
do not correspond to any obvious features in the FDC transmission spectrum, and
3.6. SERPENTINE WAVEGUIDE 99
0.29
0.3
0.31
0.32
0.33d/
λ
���������
1
Transmission
Bragg gap
Resonator gap type RG2
Resonator gap type RG2
Bragg gap
Bragg gap
ππ/2π/4 3π/40� � ��� ��� ���
(a) (b)
Figure 3.19: (a) Band diagram for a serpentine waveguide with L1 = L2 = 5d, where q is
the Bloch coefficient along the waveguide and Dy = L1 +L2. (b) Transmission through
a FDC with the same parameters (dashed), one period of the serpentine waveguide
(dotted) and two periods (solid). Note that for frequencies below d/λ, the double guide
cavity only supports a single mode (see Fig. 3.2(f)), and thus the analytic result of
Eq. (3.25) does not apply.
only begin to appear in the multiple FDC serpentine structure. Given the resonance
conditions for RG1 and RG2, and since β > ∆β, it is apparent that resonator gaps of
type RG1 occur less frequently than RG2-type gaps.
If the structure is made slightly longer, a RG1-type gap appears, as shown in
Fig. 3.20(a), which shows the band structure for a serpentine waveguide with L1 =
L2 = 7d. The solid curves in (a) are the full numerical result given by Eq. (3.24) cal-
culated using the Bloch mode method, and the dashed curves are calculated using the
semianalytic expression (3.25), with L1 = 7.5d and L2 = 6.7d, which were chosen to
give the best agreement to the numerical result. Figure 3.20(b) shows the transmis-
sion spectrum through a single FDC with L = 7d (dashed curve), and two and three
periods of the serpentine waveguide, indicated by the dotted and solid curves respec-
tively. Observe that for these parameters, the resonator gaps are relatively strong even
for such a small number of periods, whereas the Bragg gaps only appear as very weak
perturbations in the transmission spectrum for two periods.
Another feature associated with both types of resonator band gap is the positioning
of the bands above and below the gaps. In a one-dimensional Kronig-Penney model, the
right hand side (RHS) of the dispersion relation is continuous, and hence only direct
gaps exist since the RHS cannot change sign without passing through zero (see, for
example Ref. [164], chapter 6). In the case of the resonator gaps however, the RHS
of the dispersion relation (3.25) diverges when either of the resonance conditions is
satisfied. When this happens, the RHS can change sign without passing back through
zero, resulting in an indirect band gap. Indirect band gaps can be seen in Figs. 3.19(a)
and Fig. 3.20(a) at each of the resonance gaps. Although indirect band gaps occur
100 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
0.31
0.315
0.32
0.325
0.33d/
λ
Resonator gap type RG2
Bragg gap
Resonator gap type RG1
Bragg gap
Transmission0 0.2 0.4 0.6 0.8 1��������� ππ/2π/4 3π/40
(a) (b)
Figure 3.20: (a) Band diagram for a serpentine waveguide with L1 = L2 = 7d. The
solid curve is calculated with the full numerical simulation while the dashed curve is
calculated using Eq. (3.25). (b) Transmission through a FDC with L = 7d (dashed), 2
periods of the serpentine guide (dotted) and 3 periods of the serpentine guide.
commonly in two-dimensional photonic crystals, such features have only recently been
observed in other coupled resonator waveguides consisting of a linear periodic array of
resonant elements [211].
The serpentine waveguide also exhibits a number of other interesting band structures
that can be created with appropriate choices of the lengths L1 and L2 and waveguide
separation, Nc. For example, the sharp resonance at d/λ = 0.331 in Fig. 3.19(a) creates
very flat bands which correspond to low group velocities, and hence large group delays.
We showed in Section 3.4.4 that the width of a RG2-type resonance is very sensitive to
structural changes in the double waveguide section, so a similar tuning method could be
applied to adjust the group velocity response. An alternative mechanism for achieving
low group velocity bands is discussed below.
Another unusual band feature occurs when two or more consecutive band gaps
are resonator gaps, as in Fig. 3.20 where the band gaps at centre frequencies d/λ =
0.3135 and d/λ = 0.3183 are resonator gaps of type RG1 and RG2 respectively. Since
each of these gaps is indirect, the three bands above, below and between the gaps are
almost parallel to one other and all have a positive slope. With optimisation of the
geometry it may be possible to design a band structure with several equally-spaced
bands, each exhibiting a similar group-velocity profile. The width of the central band
is also very sensitive to the relative positions of the two resonances. Since the RG1 gap
depends on ∆β and the RG2 gap depends on β, it is possible to change their relative
positions by changing L1 or tuning the geometry of the double guide region. If the
two gaps are brought close together, the band between them becomes flat, and as the
two resonances pass through each other, the slope of the band, and hence the group
velocity changes sign. The group velocity could thus be controlled by tuning the relative
resonance positions. Similar properties have been identified in other coupled-resonator
3.6. SERPENTINE WAVEGUIDE 101
y1
y2
y3
y1
y2
� R1�
~
T1�
~
FDC1
FDC2� R
2�
~
T2�
~
y2
y3
y1
y2
�
�
y2
y3
R2’ �
~
T1’ �
~
T2’ �
~
R1’ �
~
Forward Backward
(a) (b)
L1
L /22
L2
L1
L /22
Figure 3.21: (a) Schematic of the serpentine waveguide structure shown in Fig. 3.18(a).
(b) The FDC structures that are combined to form the structure in (a). Each FDC is
characterised by plane wave scattering matrices for a field δ incident from above (T, R)
and below (T′, R
′).
waveguides, and have been proposed as a basis for all-optical trapping of light [102,214].
Further work is required to determine whether a serpentine structure can be designed
to exhibit the same properties.
3.6.3 Symmetry properties of the serpentine waveguide
In Section 3.6.1 we derived the dispersion properties of the serpentine waveguide in
the limit where the single waveguide regions in Figs. 3.18(a) and (b) were sufficiently
long to neglect evanescent coupling between the ends of the blocked waveguides. In
this limit, the Bloch mode transfer matrix of the full serpentine period is equal to the
square of the transfer matrix of a single FDC (3.22). Here we use symmetry relations
to derive a more general form of Eq. (3.22) for the structure in Fig. 3.18(a) which is
valid for all L2.
Consider first the serpentine structure shown in Fig. 3.21 which consists of two
FDCs (FDC1 and FDC2) joined by a single waveguide. Since FDC2 is identical to
FDC1 after reflection about the dashed-dotted line in Fig. 3.21(a), the plane wave
102 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
scattering matrices of the two structures are related according to
R1 = SR2S T1 = ST2S
R′1 = SR
′2S T
′1 = ST
′2S, (3.26a)
where S is the reversing permutation matrix
S =
0 · · · 1
· · · 1 ·· · · · ·· 1 · · ·1 · · · 0
.
Here, the plane wave scattering matrices T1, R1, T2, R2 correspond to transmission
and reflection of a field δ incident in the forward (downward) direction and T′1, R
′1, T
′2, R
′2
correspond to incidence in the backward (upward) direction as shown in Fig. 3.21(b).
Furthermore, it can be seen from Fig. 3.21(a) that forward propagation from y1 to y2
is equivalent to backward propagation from y3 to y2 such that
T1 = T′2 T2 = T
′1
R1 = R′2 R2 = R
′1. (3.26b)
Following the notation of Chapter 2, the plane wave fields at either end of FDC1
and FDC2 are related by the transfer matrices
T pwf1 =
T1 − R′1 T
′ −1
1 R1 R′1 T
′ −1
1
−T′ −1
1 R1 T′ −1
1
, (3.27a)
T pwf2 =
T2 − R′2 T
′ −1
2 R2 R′2 T
′ −1
2
−T′ −1
2 R2 T′ −1
2
. (3.27b)
Hence, given the relations (3.26a) and (3.26b), it follows that
T pwf2 = S T pw
f1 S, where S =
[S
S
]. (3.28)
We now convert the above transfer matrix formulation from the plane wave basis
to the waveguide mode basis (Bloch mode basis). At interface yi in Fig. 3.22, the field
can be represented as an expansion of plane waves with amplitudes f±i or Bloch modes
with amplitudes c±i . At y1 and y2 these expansions are related according to[f−1f+1
]= F1
[c−1c+
1
] [f−2f+2
]= F2
[c−2c+
2
], (3.29a)
3.6. SERPENTINE WAVEGUIDE 103
f1
±
f2
±
f3
±
c1
±
c2
±
c3
±
y1
y2
y3
�f�
BM
�f�
BM
�f�
pw
�f�
pw
�s
BM
Figure 3.22: Schematic of the unit cell of the serpentine waveguide formed from two
FDC structures. The forward and backward propagating fields at interfaces y1 − y3
are represented in the basis of plane waves f±i or Bloch modes c±i . The fields at each
interface are related by plane wave and Bloch mode transfer matrices, T pw and T BM
respectively.
where the F i provide the change of basis from plane waves to Bloch modes as defined
Section 2.2.4.
By definition, the plane wave transfer matrix T pwf1 relates[
f−2f+2
]= T pw
f1
[f−1f+1
], (3.29b)
which can be converted to Bloch mode form using Eqs. (3.29a) such that[c−2c+
2
]= F−1
2 T pwf1 F1
[c−1c+
1
]= T BM
f1
[c−1c+
1
], (3.29c)
where
T BMf1 = F−1
2 T pwf1 F1. (3.29d)
Similarly, the equivalent Bloch mode transfer matrix for FDC2 is
T BMf2 = F−1
1 T pwf2 F2, (3.29e)
which can be expressed in terms of T BMf1 using Eqs. (3.28) and (3.29d) as
T BMf2 = F−1
1 ST pwf1 SF2 = F−1
1 SF2T BMf1 F−1
1 SF2. (3.29f)
Therefore the Bloch mode transfer matrix for the full period of the serpentine waveguide
in Fig. 3.22 is
T BMs = T BM
f2 T BMf1 = QT BM
f1 QT BMf1 =
(QT BM
f1
)2, (3.30)
104 CHAPTER 3. COUPLED WAVEGUIDE DEVICES
where
Q = F−11 SF2. (3.31)
Recall from Section 3.6.1 that the dispersion relation for the serpentine waveguide
is obtained by solving for the eigenvalues of T BMs . In the monomodal approximation
we observed that T BMs = T BM
f1
2and hence the eigenvalues µ were found by solving the
eigenvalue equation T BMf1 b = ±µ1/2b. Equation (3.30) shows that the general solution
can also be found in terms of a single FDC, where the relevant eigenvalue equation is(QT BM
f1
)B = µ1/2B. (3.32)
Here B is a matrix of eigenvectors expanded in the full Bloch mode basis containing
both propagating and evanescent modes. Although we do not derive the result here, Qis a diagonal matrix of elements ±1, and so (Q)2 = I. In the limit where L2 is large,
T BMf1 is dominated by the single propagating Bloch mode in the waveguide, and hence
Q has little effect on the solutions of Eq. (3.32), which then reduces to the single mode
result derived in Section 3.6.1.
3.7 Conclusion
We have covered in this chapter a range of topics associated with directional coupling
and resonant reflection in photonic crystals, and have presented several devices that
exhibit unusual transmission characteristics as a result of combining the two effects.
These devices are described accurately by semianalytic models which allow rapid and
efficient optimization and provide physical insight into the properties observed in full
numerical simulations.
The semianalytic formulations and the physical interpretation of the numerical re-
sults are based on the two coupling effects considered in Sections 3.2 and 3.2. Coupling
between parallel waveguides was considered in Section 3.2 where we demonstrated that
weak-coupling approximations typically used for dielectric waveguides can also be ap-
plied to photonic crystal waveguides. The second type of coupling occurs when two
different waveguide sections are interfaced end-to-end, as in the two examples presented
in Section 3.3. In this case, the coupling properties are influenced not only by direct
propagation from one waveguide to the other, but also by resonant coupling through
the photonic crystal barrier, should one exist. The devices presented in Sections 3.4 —
3.6 derive their properties from a combination of these two types of coupling.
Chapter 4
Recirculating Mach-Zehnder
interferometer
In this chapter we use many of the techniques and results from Chapter 3 to investigate
the characteristics of photonic crystal Mach-Zehnder interferometers (MZIs). We show
that the MZIs formed from waveguides in a perfectly reflecting cladding can display
manifestly different transmission characteristics from conventional MZIs due to mode
recirculation and resonant reflection. In devices of this type, recirculation effects can
result in a significantly sharper switching response than in conventional interferometers.
A simple and accurate analytic model is derived to describe this behaviour and we
propose specific PhC structures in both rod-type and hole-type geometries that display
these properties.
This chapter is based largely on the work published in Ref. [204], with some addi-
tional details included relating to model in Section 4.2 and the design and simulation of
the photonic crystal based Mach-Zehnder interferometer in Section 4.4. The discussion
in Section 4.6 is based on work to be published in Ref. [182].
4.1 Introduction
Mach-Zehnder interferometers (MZIs) are used extensively in optical systems as filtering
and switching devices and in phase measurement and detection applications. Their
many uses and functions have led to the fabrication of MZI devices in bulk-optics [103],
fibres [215], planar dielectric waveguides [216], rib waveguides [217], and more recently
2D PhCs [200,201,218–220]. Although the morphology of all MZIs is the same regardless
of the type of waveguide in which they are formed, the operational characteristics can
be strongly influenced by the waveguide and cladding properties.
Three generic MZI designs are illustrated in Figure 4.1. The example in (a) is a
bulk-optics MZI, while those in (b) and (c) are waveguide-based designs that can be
formed in either dielectric or PhC waveguides. In each case, an incident beam is split
into two beams at the first junction, with each one taking a different path through
105
106 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
(a) (c)
in
out 1
out 2
in out
(b)
in 1
in 2
out 1
out 2
Figure 4.1: Examples of three generic symmetric MZI designs with different types of
junction. (a) Bulk-optics MZI using 50/50 beam-splitters and mirrors. (b) Directional
coupler-based MZI with two possible input and output waveguides. (c) Y-junction MZI
with single input and output waveguides.
the interferometer before being recombined at the second junction. In a symmetric,
or balanced, MZI, the two paths are identical as in the diagram, whereas the paths
of an unbalanced MZI are different so as to introduce a relative phase delay. The
devices illustrated in Fig. 4.1 (a) and (b) each have two output ports where the power
coupled into each depends on the relative phase delay between the two arms of the
interferometer. For an ideal device, the combined output power is equal to the input
power, and the light only passes through the interferometer once. In contrast, the MZI
in Fig. 4.1(c) only has a single output waveguide. In this case, any light that is not
coupled into the output waveguide must either be radiated out of the device or reflected
back into the interferometer. If the junction is formed in a dielectric waveguide relying
on TIR for guidance, most of the non-transmitted light is radiated into the cladding, as
happened with the terminated waveguides discussed in Chapter 3. For a PhC waveguide
junction however, the cladding is perfectly reflecting so radiation cannot occur and thus
any light that is not transmitted must be reflected. The resulting recirculating MZI
exhibits strong resonance effects that are not usually observed in conventional dielectric
MZIs or coupler-based MZIs of the type illustrated in Figs. 4.1(a) and (b). The clear
observation of these effects relies critically on the use of efficient, symmetric beam
splitters that exhibit very low reflectance (< 2%).
In Section 4.2 we present a simple modal analysis to explain the transmission char-
acteristics of recirculating MZIs and derive an elegant semianalytic expression that
accurately describes their behaviour. The properties of these devices are strikingly
different from conventional MZIs, and our analysis shows that they are typical of a gen-
eral class of MZI with perfectly reflecting cladding and single-mode input and output
waveguides. One feature of these devices is an almost square transmission response
with phase and thus a significantly sharper switching response than conventional MZIs.
Such behaviour is a highly desirable characteristic for both linear and nonlinear switch-
ing applications. As in Chapter 3, the analysis and numerical results are purely 2D,
but the results hold generally if radiation losses are low. This issue is discussed in more
4.2. MODAL ANALYSIS 107
detail in Chapter 6.
Two specific examples of PhC waveguide-based recirculating MZIs are proposed in
Section 4.4. Both examples use the coupled Y-junction design presented in Section 3.5
which provides the necessary high-transmission and mode rejection properties to observe
the recirculation effects. In this case, the perfectly reflecting cladding is provided by
the photonic bandgap of the PhC waveguide walls. While it has been proposed that the
strong reflections and associated interference observed in many PhC based devices may
limit their practical use in optical systems [221], we show here that these effects can
also be used to great advantage. Similar recirculation characteristics can be obtained
in a bulk-optics geometry by including two additional mirrors in the system illustrated
in Fig. 4.1(a). We analyse such a system in Section 4.6 and show that the transmission
properties are analogous to those of the PhC MZIs.
4.2 Modal analysis
Consider first the MZI illustrated in Fig. 4.2(a), consisting of single input and output
waveguides, and two junctions Y1 and Y2 joined by waveguide arms A1 and A2 of length
L. In this analysis we assume that the Y-junctions are ideal and split light equally
between A1 and A2. If an element is placed in A2 to introduce a phase difference ϕ
between the two arms, the intensity of light transmitted into the output depends on
ϕ. Active devices such as switches and filters can be made by varying ϕ to modulate
the output [218, 220, 222], or in passive devices, the measured output can be used to
determine ϕ [215]. In a symmetric, or balanced, MZI ϕ = 0, so light entering from
the input guide recombines in-phase at Y2 and is transmitted into the output guide - a
balanced MZI transmits equally well at all wavelengths.
ϕ
input outputwaveguide junctionsY1 Y2
A1
A2
(a) (b)
Figure 4.2: (a) Schematic of a simple MZI with single input and output waveguides
joined by arms A1 and A2 via Y-junctions Y1 and Y2. (b) Schematic of the propagating
modes in each of the waveguide sections in (a). The green and red curves show the
even and odd superpositions of the individual waveguide modes.
The transmission properties of the MZI in Fig. 4.2(a) can be derived by considering
the propagating modes in each of the waveguide sections, in a procedure similar to that
108 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
used to analyse the coupled waveguide devices in Chapter 3. An alternative derivation
based on ideal Bloch mode scattering matrices is outlined in Section 4.5. All of the
waveguides in Fig. 4.2 are taken to be identical, with each supporting a single mode
of even symmetry with respect to the waveguide axis. Thus, the input and output
guides support one mode each, denoted by |ψin〉 and |ψout〉 respectively, and the two-
guide region supports one mode in each of A1 and A2, denoted |ψ1〉 and |ψ2〉, as shown
in Fig. 4.2(b). Alternatively, the modes in A1 and A2 are well approximated as a
combination of two supermodes of even (+) and odd (−) symmetry with respect to the
axis of the interferometer, indicated by the green and red curves in Fig. 4.2(b). The
supermodes are approximately related to the modes in the individual guides by
|ψ±〉 =1√2
(|ψ1〉 ± |ψ2〉) , |ψ1,2〉 =1√2
(|ψ+〉 ± |ψ−〉) . (4.1)
In an ideal structure, A1 and A2 are separated sufficiently that coupling between the
two guides is negligible, in which case the propagation constants of |ψ±〉 are equal to
the propagation constant of an isolated waveguide, β+ = β− = β respectively. This ap-
proximation is not always accurate however, in which case an alternative approximation
can be made to the semianalytic model, as discussed in Section 4.4.3.
Since mode |ψin〉 is symmetric with respect to A1 and A2, it can only couple to the
supermode with the same symmetry, |ψ+〉, so for an ideal Y1 junction, mode |ψin〉 is
fully transmitted into |ψ+〉. When this mode propagates through A1 and A2, the effect
of ϕ is to introduce an asymmetric field component, equivalent to transferring some of
the energy into |ψ−〉. As at Y1, the |ψ+〉 component of the field entering Y2 can couple
into the output guide, but the |ψ−〉 component cannot and must be either radiated or
reflected.
To this point, the modal description holds for MZIs in general. The essential differ-
ence lies in what happens to |ψ−〉 at Y2. In dielectric waveguides, most of the light in
|ψ−〉 is radiated into the cladding and lost, and only a very small amount is reflected
back into the arms of the interferometer [223]. However, if light is unable to radiate into
the cladding, any light in |ψ−〉 must be reflected at Y2, back into the interferometer.
This odd mode reflection results in the unique resonant behaviour of recirculating MZIs
with single-mode input and output waveguides.
We now follow the propagation of a mode Ψ with unit energy 〈Ψ|Ψ〉 = 1 entering an
unbalanced MZI from the input guide and consider the effect of odd mode reflection at
Y2. After passing through Y1, the resulting symmetric field distribution can be expressed
in terms of |ψ1〉 and |ψ2〉 using Eq. (4.1) to give
|Ψ(0)〉(1) = |ψ+〉 =1√2
(|ψ1〉+ |ψ2〉) .
Here the superscript (1) indicates that we are considering the fields on the first pass
through the interferometer. The modes propagate through each arm essentially inde-
pendently of each other, advancing in phase as they propagate. At Y2, the combined
4.2. MODAL ANALYSIS 109
field in A1 and A2 is
|Ψ(L)〉(1) =(eiβL|ψ1〉+ ei(βL+ϕ)|ψ2〉
)/√
2 = eiχ (cos(ϕ/2)|ψ+〉 − i sin(ϕ/2)|ψ−〉) ,(4.2)
where χ = βL + ϕ/2 is the average phase of the two arms and β is the propagation
constant of the single propagating mode of an isolated waveguide. At Y2, the even field
component is transmitted into the output guide, while the odd mode is reflected back
into A1 and A2. Note that in a conventional MZI the odd mode is radiated into the
cladding and the transmission is simply given by the even mode term in Eq. (4.2),
T = |eiχ cos(ϕ/2)|2 = cos2(ϕ/2) =1
1 + tan2(ϕ/2). (4.3)
In the PhC structure, however, the reflected odd field component, −i exp(iχ) sin(ϕ/2)|ψ−〉,propagates back through the two arms, accumulating a further phase ϕ so that the field
arriving back at Y1 is
|Ψ(0)〉(2) = −ieiχ sin(ϕ/2)(eiβL|ψ1〉 − ei(βL+ϕ)|ψ2〉
)/√
2
= −ie2iχ sin(ϕ/2) (cos(ϕ/2)|ψ−〉 − i sin(ϕ/2)|ψ+〉) . (4.4)
Once again the odd field component is reflected back into the interferometer, while the
even component is transmitted through the junction, in this case back into the input
waveguide.
The total transmission is given by the superposition of the fields transmitted through
Y2 each time the odd mode is reflected. Denoting the contribution after n passes through
the interferometer by tn, the transmitted field is
t =∞∑
n=1
tn (4.5)
= eiχ cos(ϕ/2)− e3iχ sin2(ϕ/2) cos(ϕ/2)− e5iχ sin2(ϕ/2) cos3(ϕ/2)− . . . (4.6)
= eiχ cos(ϕ/2)− e3iχ sin2(ϕ/2) cos(ϕ/2)
1− e2iχ cos2(ϕ/2), (4.7)
where the second term in (4.7) is the sum of the geometric series∑∞
n=2 tn. Hence, the
transmitted intensity is given by
T = |t|2 =4 sin2(χ) cos2(ϕ/2)/ sin4(ϕ/2)
1 + 4 sin2(χ) cos2(ϕ/2)/ sin4(ϕ/2). (4.8)
Note that the form of Eq. (4.7) closely resembles the Airy-formula [122] for reflection
from a Fabry-Perot cavity – a feature we return to in Section 4.3. The reflected in-
tensity (R = 1 − T ) can be derived in the same way by summing the field amplitudes
reflected back into the input waveguide. Observe that Eq. (4.8) is a function of both
ϕ and βL = β(λ)L, in contrast to the conventional MZI transmission (4.3), which is a
110 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
function of ϕ only. Thus, the waveguide dispersion properties are especially important
in recirculating MZIs. Apart from the different functional forms of Eqs. (4.3) and (4.8),
the additional dependence on wavelength and structure length leads to significantly dif-
ferent transmission properties. The explicit dependence on structure length is a direct
result of the mode recirculation inside the device.
4.3 Transmission characteristics
In this section we study the transmission characteristics of recirculating MZIs described
by Eqs. (4.7) and (4.8). For generality the normalised units βL are used, although we
note that when modelling a realistic device, the waveguide dispersion β = β(λ) must be
taken into account. Hence for a specific MZI design, the dispersion properties provide
a direct mapping between βL and wavelength. The effects of waveguide dispersion are
discussed in Section 4.4, where the transmission properties of two PhC-based MZIs are
presented in terms of wavelength.
Figures 4.3(a) and (b) are contour plots of transmittance as a function of βL and ϕ
for a conventional and a recirculating MZI respectively, where the red areas correspond
to high transmittance and the blue areas correspond to low transmittance. The trans-
mittance of the conventional MZI is given by Eq. (4.3) so there is no variation with βL,
but only with ϕ. For the recirculating MZI however, there is a strong dependence on
both parameters but it can be seen that T = 1 when ϕ = 0 and T = 0 when ϕ = π,
as for the conventional MZI. In (b) the slightly tilted, low transmittance ‘valleys’ cor-
respond to strong resonances due to the multiple reflections of the odd mode. At the
centre of the valleys, the transmittance drops to zero and the valleys become rapidly
narrower as ϕ approaches 0 or 2π. The slight tilt of the valleys from vertical is due to
the dependence of χ on both βL and ϕ in Eq. (4.8).
We consider first the properties of Eq. (4.8) as ϕ is varied at a fixed βL, correspond-
ing to traversing a vertical line across Fig. 4.3(b). Figures 4.4(a)–(e) show the trans-
mittance of a recirculating MZI as a function of ϕ for βL = 0, π/8, π/4, 3π/8 and π/2
respectively, indicated by the solid curves. The corresponding transmittance of a con-
ventional MZI (4.3) is indicated in each plot by the dashed curves. Observe from
Fig. 4.3(b) that for most values of βL, a vertical line of varying ϕ intersects one end
of a low transmission valley, resulting in a response with a strong resonant reflection
as seen in Fig. 4.4(b). The width of the resonance decreases to zero as ϕ → 0 or
ϕ→ 2π, at which point T = 1. Of particular interest is the response of the MZI when
βL = π/2 , 3π/2 , 5π/2 . . ., where the diagonal valleys in Fig. 4.3(b) intersect the low
transmittance band at ϕ = π. The solid curve in Fig. 4.4(e) shows the response as
a function of ϕ when βL = π/2. Comparing this with the dashed curve of the con-
ventional MZI, it is seen that the recirculating MZI response exhibits a significantly
sharper transition between ‘on’ and ‘off’ states. Substituting the condition identified
4.3. TRANSMISSION CHARACTERISTICS 111
0
p
2p
p
2
32p
j
0 2p 4p 6p
bL
0 0.2 0.4 0.6 0.8 1
0
p
2p
p
2
32p
j
0 2p 4p 6p
bL
(a) (b)Transmission
Figure 4.3: Contour plots of transmittance as a function of the dimensionless parameter
βL and the phase difference ϕ for (a) a conventional MZI described by Eq. (4.3) and
(b) a recirculating MZI described by Eq. (4.8)
above for βL into Eq. (4.8), we find the transmittance to be
T =4 cot4(ϕ/2)
1 + 4 cot4(ϕ/2)=
1
1 + tan4(ϕ/2)/4. (4.9)
Comparing Eqs. (4.3) and (4.9), we see that the recirculating MZI response is quar-
tic in tan(ϕ/2), compared with the quadratic response of a conventional MZI. The
change in ϕ required to switch the transmittance from 0.9 to 0.1 in the recirculating
MZI is approximately 0.32π, whereas a conventional MZI would require a change of
0.59π. Although cos4(ϕ/2) responses can be achieved with a pair of conventional MZIs,
and higher order behaviour occurs for a chain of such devices [224], we find that at
least six identical conventional MZIs would be required to obtain a superior switching
performance to that of a single recirculating MZI. Note also, that the transmittance
given by Eq. (4.9) has a fourth-order variation near both T = 0 and T = 1, whereas a
chain of MZIs always exhibits a quadratic variation in the region around T = 1. The
steep, almost square response of the recirculating MZI is highly desirable for switching
applications, and could be used to improve performance in both linear and nonlinear
devices.
If βL is varied while ϕ is fixed, as in Figs. 4.4(f)–(j), the transmission properties are
very different from those shown in Figs. 4.4(a)–(e). For a given ϕ, the transmittance
T = T (βL) given by Eq. (4.8) resembles the expression for the reflectance of a Fabry-
Perot etalon [122] with finesse
F =π| cos(ϕ/2)|
1− | cos(ϕ/2)|. (4.10)
112 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
bL=0
bL=p/2
bL=3p/8
bL=p/4
bL=p/8
j=0
j=p/4
j=p/2
j=3p/4
j=p
0
1
0.2
0.4
0.6
0.8
T
0
1
0.2
0.4
0.6
0.8
T
0
1
0.2
0.4
0.6
0.8
T
0
1
0.2
0.4
0.6
0.8
T
0
1
0.2
0.4
0.6
0.8
T
0 p 2pp
23p
2
0 p 2p 3p 6p5p4p
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)
(j)
j bL
Figure 4.4: Solid curves: (a)–(e): Power transmitted by an ideal recirculating MZI as a
function of ϕ for βL = 0, π/8, π/4, 3π/8, π/2 respectively. (f)–(j): Power transmitted by
an ideal recirculating MZI as a function of βL for ϕ = 0, π/4, π/2, 3π/4, π respectively.
Dashed curves: Power transmitted by a conventional MZI for the same parameters.
Thus, the transmission behaviour of the recirculating MZI is essentially the reverse of
a Fabry-Perot etalon, exhibiting resonant reflection, rather than resonant transmission.
This is seen by considering horizontal lines of constant ϕ in Fig. 4.3(b), and the examples
in Figs. 4.4 (f)–(j). As ϕ approaches 0 or 2π, the resonances become narrower, and the
finesse increases according to Eq. (4.10). As ϕ approaches π, both the fringe visibility
and the finesse decrease. These reversed Fabry-Perot transmission characteristics could
be used in the design of a tunable notch rejection filter where the finesse and resonance
position could be adjusted by varying β and ϕ.
In practice both β and ϕ are functions of wavelength, and so it is difficult to design a
device to exhibit transmission spectra identical to those shown in Figs. 4.3(f)–(j). Any
realistic measurement of transmission spectra is likely to correspond to the transmission
along a curved line in Fig. 4.3(b), with the exact shape of such a curve being determined
4.4. PHOTONIC CRYSTAL RECIRCULATING MZI STRUCTURES 113
by the method used to induce ϕ. This aspect is discussed further in Section 4.4.3.
4.4 Photonic crystal recirculating MZI structures
In Section 4.2 we derived a simple model for the transmission of an ideal MZI with
a perfectly reflecting cladding and single-mode input and output guides, and showed
that recirculation and resonant properties occur due to the reflection of modes with
odd symmetry at the waveguide junctions. In this section we present both rod- and
hole-type PhC MZI designs that display these characteristics and compare full numer-
ical calculations with the semianalytic results of Eq. (4.8). For these comparisons, the
waveguide dispersion properties are required as numerical input into Eq. (4.8), however
this is a relatively simple calculation that does not require additional knowledge of the
MZI structure.
The rod-type PhC considered throughout this section consists of a square lattice
of rods with index ncyl = 3.4 and radius r = 0.18d in air. This structure has a TM
bandgap in the normalised frequency range 0.302 ≤ d/λ ≤ 0.445. We present results
for a structure with lattice period d = 570 nm, chosen so that the MZI operates close
to λ = 1.55µm, within the bandgap spanning 1.28µm ≤ λ ≤ 1.88µm. A waveguide
formed by removing a single line of cylinders supports a single propagating mode of
even symmetry in the wavelength range 1.28µm ≤ λ ≤ 1.83µm.
For the hole-type PhC we use the same parameters as in Section 1.5.3, namely a
triangular lattice of air holes with radius r = 0.32d in a silicon background of nb = 3.4.
A lattice period of 350 nm is chosen so that the TE bandgap in the range 0.216 <
d/λ < 0.298 corresponds to a wavelength range 1.17µm ≤ λ ≤ 1.62µm. Since a W1
waveguide can support two modes in the bandgap (see Fig. 1.8), we operate only in the
low-frequency region of the bandgap where a single mode is supported for 1.435µm ≤λ ≤ 1.580µm. This mode has a large group velocity dispersion close to cutoff at the
low frequency end of this range which distorts the transmission profile of the MZI
considerably, as seen in Section 4.4.3.
4.4.1 Design
The most important components in a PhC-based recirculating MZI are the Y-junctions
or beam splitters. Recall from Section 4.2 that the Y-junctions in the model were
taken to be ideal, transmitting modes of even symmetry and reflecting modes of odd
symmetry. In practice, the even mode is not perfectly transmitted, and the additional
reflected field components can result in extra Fabry-Perot effects [218] that make the
odd mode resonances difficult to identify. From our numerical studies, we find that
the odd mode effects are most clearly observed when the even mode reflectance is less
than 2%. Thus, to observe the properties predicted by Eq. (4.8), the junction design is
critical. In Section 3.5 we proposed a PhC coupled Y-junction design that exhibits very
high transmission over a wide bandwidth due to a combination of directional coupling
114 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
and mode recirculation behaviour. This type of junction is ideal for use in recirculating
PhC MZI designs as demonstrated by the examples in this section. We note that Wang
et al. proposed a similar coupled junction design for a surface plasmon MZI based
on metal gap waveguides, although results were only reported for a balanced MZI so
recirculation effects were not observed [225].
1.5 1.55 1.60.95
0.96
0.97
0.98
0.99
1
T
l m( m)
1.65
88nm 67nm
(a) (b)
(c) (d)
36nm
1.54 1.55 1.56 1.57
l m( m)
0.95
0.96
0.97
0.98
0.99
1
T
L =5dy L =11dy
Figure 4.5: (a) and (b): Y-junction designs used for the rod- and hole-type PhC MZIs
illustrated in Figs. 4.6(a) and (b) respectively. (c) Solid curve: transmission spectrum of
the Y-junction shown in (a). Dashed curve: transmission of the PhC MZI in Fig. 4.6(a)
with ϕ = 0 and LMZ = 15d. (d) Solid curve: transmission spectrum of the Y-junction
shown in (b). Dashed curve: Transmission of the PhC MZI in Fig. 4.6(b) with ϕ = 0
and LMZ = 32d. Note that the transmission axis starts at 95% in both plots.
Figures 4.5 (a) and (b) show two coupled Y-junctions designed for wide bandwidth
transmission in the rod- and hole- type PhCs defined at the start of Section 4.4. The
transmission spectra for these junctions are plotted in (c) and (d) of the same figure
(solid curves). The spectrum in (c) has two transmission bands of widths 88 nm and
67 nm respectively where T > 98%, but between these two regions is a sharp resonant
reflection, the cause of which is discussed in Section 3.5. Transmission spectra are cal-
culated over the combined wavelength range of the two bands and thus the resonant
reflection feature can be seen in some of the results for the rod-type PhC MZI. Since
the resonance is sharp, however, the MZI properties are only affected in a very nar-
row wavelength region. The hole-type PhC junction shown in Fig. 4.5(b) has a 98%
transmission bandwidth of 36 nm, as shown by the solid curve in Fig. 4.5(d). When
4.4. PHOTONIC CRYSTAL RECIRCULATING MZI STRUCTURES 115
the junctions are operated in reverse as beam combiners, the even mode is transmitted
into the single output waveguide with the same efficiencies shown in Figs. 4.5(c) and
(d), while the symmetrical design of the junctions provides the necessary odd mode
reflection properties.
The PhC junctions illustrated in Figs. 4.5(a) and (b) are formed into MZIs by joining
two identical junctions together with PhC waveguides as shown in Fig. 4.6. The total
length of the device, including the junctions of length LY is given by LMZ as shown. In
Section 4.4.2 we demonstrate that a phase difference can be introduced between the two
arms by modifying the radius of several cylinders on the edge of one waveguide, thereby
changing the optical path length relative to the other arm. The number of cylinders
modified on each edge is given by the parameter Lϕ, defined in Fig. 4.6, where the
modified cylinders are shaded a different colour for illustration purposes. Recall that
we discussed in Section 2.5.2 some of the issues involved with modelling a balanced
PhC MZI using the Bloch mode method. To introduce the modified cylinders in one
arm we must include a new segment in the structure, and thus an additional grating
calculation is required. To ensure that the modified segment does not interfere with
the junction operation, at least two layers of symmetric double waveguide are retained
where the arms enter the junctions. Thus, the MZI in Fig. 4.6(a) must be calculated as
a seven segment structure, while the one in Fig. 4.6(b) is calculated as a nine segment
structure by adding an extra segment in the middle of M4 in Fig. 2.9. The Bloch mode
method results presented in this chapter we calculated with a grating supercell period
of Dx = 21d and a plane-wave truncation order of Npw = 44, where Dx and Npw are
defined in Chapter 2.
Recall from Section 4.2 that when ϕ = 0 an MZI is symmetric (balanced) and
therefore transmits equally well at all wavelengths. In a practical PhC based device,
this is only expected for wavelengths within the transmission band of the junction.
The dashed curves in Figs. 4.5(c) and (d) show the transmission spectra for balanced
MZIs of length LMZ = 15d (rod-type PhC) and LMZ = 32d (hole-type PhC) respectively.
Note that although the transmission bandwidth is reduced slightly relative to the single
junction in both cases, the transmission remains high across the band with only a few
shallow ripples due to weak Fabry-Perot resonances of the even mode.
4.4.2 Tuning ϕ
Most MZI applications require ϕ to be tuned for the purposes of modulating or switching
the output signal. This has been demonstrated in PhC based MZIs using thermo-
optic [218] and nonlinear [220] effects to induce a small refractive index change in one
arm. Liquid crystal tuning of the refractive index has also been proposed [222]. For
the purposes of demonstrating the response of recirculating MZIs, here we generate ϕ
by changing the radii of Lϕ/d pairs of cylinders on either side of one waveguide from
r to r as indicated in Fig. 4.6. This approach is chosen for numerical convenience as
it is easier to modify the properties of individual cylinders rather than the background
116 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
Lj
LY
LMZ
Lj
LY
LMZ
(a)
(b)
Figure 4.6: Diagrams showing the basic geometry of the PhC MZIs studied in this
chapter. (a) Rod-type PhC MZI formed from the junction shown in Fig. 4.5(a). (b)
Hole-type PhC MZI formed from the junction shown in Fig. 4.5(b). The cylinders
highlighted in (a) and (b) are modified to generate ϕ as described in Section 4.4.2.
refractive index in the Bloch-mode scattering matrix method.
The modal propagation constant in the modified waveguide section differs from
the original waveguide by ∆β � β, so a mode propagating through the modified
waveguide arm accumulates a total phase difference of approximately ϕ = ∆βLϕ relative
to the other arm. Thus, the maximum ϕ that can be introduced depends on Lϕ and
the change in radius ∆r = r − r. However ∆r must be sufficiently small to avoid
significant reflection caused by the change in waveguide properties and hence many
cylinders typically need to be modified to achieve even a relatively small phase change.
This is shown in Fig. 4.7(a) where ϕ is plotted as a function of ∆r for Lϕ = 5d
and Lϕ = 10d for the rod-type PhC waveguides used in Fig. 4.1(a) at λ = 1.553µm.
Hole-type PhC waveguides exhibit similar behaviour when tuned in the same way.
Observe that ϕ is doubled when Lϕ is increased from 5d to 10d as expected. The total
reflected power from the modified waveguide section is also plotted for the two lengths
in Fig. 4.7(b). The reflectance does not increase linearly with Lϕ because the reflections
originate only at the two ends of the modified section. Since any additional reflection
4.4. PHOTONIC CRYSTAL RECIRCULATING MZI STRUCTURES 117
effects could interfere with the recirculation properties of the MZI, we limit ∆r such
that the R . 1%.
In Section 4.4.3 we use the Bloch mode method to calculate the transmission spectra
of the PhC MZIs shown in Fig. 4.6 for different values of Lϕ. Since ∆β and hence ϕ are
both functions of wavelength, this calculation is not directly equivalent to the results
plotted in Fig. 4.4(f)–(j), where ϕ was held constant. In order to compare Eq. (4.8) with
the full numerical results, the wavelength dependence variation of ϕ must be included.
The dashed and dash-dot curves plotted in Fig. 4.8 show ϕ as a function of wavelength
for the different Lϕ used in the numerical calculations of Section 4.4.3.
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0-0.02 0.02 0.04
R0
0-0.02 0.02 0.04
j
p
2
p
2 (a) (b)
L dj=5
L dj=10
L dj=5
L dj=10
Dr/d Dr/d
Figure 4.7: (a) Phase difference induced by changing the radii of Lϕ pairs of cylinders
by ∆r in one arm of the rod-type PhC MZI in Fig. 4.6(a). (b) Reflectance of the
modified waveguide section for the same parameters. Note that both ϕ = 0 and R = 0
when r = r = 0.18d (∆r = 0). The calculations are for λ = 1.553µm, corresponding to
λ/d = 2.689.
4.4.3 Numerical results
Before proceeding to full numerical calculations, we consider first the transmission prop-
erties predicted by Eq. (4.8) for the two MZIs illustrated in Fig. 4.6. Figures 4.8(a) and
(b) are contour plots of transmittance as a function of wavelength and ϕ calculated
using Eq. (4.8) for rod- and hole-type PhC MZIs of length L = 36d and L = 40d respec-
tively. The waveguide dispersion β = β(λ) was calculated using the Bloch mode method
and used as numerical input. These plots are essentially the same as Fig. 4.3(b), where
the horizontal axis has been rescaled to take into account the length and dispersion
parameters of the devices in Fig. 4.6.
There are some clear differences between the contour plots in Figs. 4.8(a) and (b).
First, the low transmittance valleys run in the opposite diagonal direction because of
the different dispersion slopes of the waveguide modes in the two PhC structures. Sec-
ond, in (b) the tilt and the separation of the resonances changes significantly at longer
wavelengths, whereas in (a) the valleys remain almost parallel and equally spaced. This
is also a result of the different dispersion slopes. The waveguide modes of the MZI in
118 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
1.5 1.55 1.6
2p
p
0
32p
p
2
l m( m)
j
1.52 1.54 1.56 1.58
2p
p
0
32p
p
2
l m( m)
j
0 0.2 0.4 0.6 0.8 1
Transmission
Figure 4.8: (a) Contour plot of transmittance as a function of λ and ϕ calculated using
Eq. (4.8) for the PhC waveguides in Fig. 4.6(a) with LMZ = 36d. The dashed and
dashed-dotted curves show the phase change induced by ∆r = 0.02d for Lϕ = 8d and
Lϕ = 16d respectively as described in Section 4.4.3. (b) As in (a) but for the waveguides
in Fig. 4.6(b) with LMZ = 40d. The dashed and dashed-dotted curves correspond to
the phase change for ∆r = −0.02d and Lϕ = 4d and Lϕ = 8d respectively.
Fig. 4.6(a) are well away from cutoff, so the dispersion slope is relatively constant,
resulting in the regular features of Fig. 4.8(a). In contrast, the waveguide modes of
the MZI in Fig. 4.6(b) are closer to cutoff and the resulting strong dispersion distorts
the transmission profile. The spacing of the resonances decreases with increasing wave-
length, while the resonances become narrower, maintaining a constant finesse. Despite
this distortion, the square response predicted by Eq. (4.9) still occurs when βL satisfies
the conditions identified in Section 4.3.
Rod-type PhC MZI
Figures 4.9(a) and (b) show the transmission spectra of two MZIs of the type shown
in Fig. 4.6(a) with Lϕ = 8d and Lϕ = 16d respectively for r = 0.2d (∆r = 0.02d) and
LMZ = 36d. The solid curves are calculated using the full Bloch mode numerical calcu-
lation, and the dashed and dashed-dotted curves are the result of Eq. (4.8), calculated
along the corresponding curves in Fig. 4.8(a) to account for the wavelength depen-
dence of ϕ. Both spectra exhibit the reversed Fabry-Perot-like behaviour predicted by
Eq. (4.8) with the finesse of the reflection resonances decreasing with increasing phase
according to Eq. (4.10). Note also that the sharp reflection feature exhibited by the
Y-junction in Fig. 4.5(a) can just be seen in Fig. 4.9(a) at λ ≈ 1.57µm, very close to
a resonance of the MZI. The two resonances coincide exactly in Fig. 4.9(b) and cannot
4.4. PHOTONIC CRYSTAL RECIRCULATING MZI STRUCTURES 119
0.2
0.4
0.6
0.8
1
T
0.2
0.4
0.6
0.8
1
T
l m( m)
1.54 1.55 1.56 1.57
0.2
0.4
0.6
0.8
1
0
1.54 1.55 1.56 1.57
0.2
0.4
0.6
0.8
1
0
1.5 1.54 1.58 1.6200
TT
l m( m)
l m( m)l m( m)
1.5 1.54 1.58 1.62
(a) (b)
(c) (d)
Figure 4.9: (a) and (b): Transmission spectra of the rod-type PhC MZI (Fig. 4.6(a))
calculated using the Bloch mode matrix method with LMZ = 36d and ∆r = 0.02d for
(a) Lϕ = 8d and (b) Lϕ = 16d (solid curve). Dashed and dashed-dotted curves are
given by Eq. (4.8) calculated along the corresponding curves in Fig. 4.8(a). (c) and
(d): Transmission spectra of the hole-type PhC MZI (Fig. 4.6(b)) calculated using the
Bloch mode matrix method with LMZ = 40d and ∆r = −0.02d for (a) Lϕ = 4d and
(b) Lϕ = 8d (solid curve). Dashed and dashed-dotted curves are given by Eq. (4.8)
calculated along the corresponding curves in Fig. 4.8(b).
be distinguished.
The parameters L and Lϕ used in Eq. (4.8) are not necessarily identical to those
defining the structure in Fig. 4.3(d). Since PhCs behave as distributed reflectors, defin-
ing a structure length is somewhat ambiguous. We find that the best agreement in
resonance position occurs when we take L in Eq. (4.8) to be the total length of the
structure including the junctions, LMZ, as shown in Fig. 4.6, with a slight correction
for the phase change introduced by reflection from the PhC. For the comparisons in
Figs. 4.9(a) and (b) we found the best agreement for L = 36.75d. Under these condi-
tions, the agreement between the rigorous numerical result and the semianalytic model
is excellent. In particular, the model predicts the resonance widths to a high accu-
racy, and with the corrected length parameter the resonance positions are also closely
matched.
120 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
In Section 4.3 we showed that recirculating MZIs exhibit a considerably sharper
response to ϕ than conventional, single-pass MZIs, with that response described by
Eq. (4.9) when βL = π/2 , 3π/2 , 5π/2 . . . To demonstrate this property in a full nu-
merical calculation ϕ must be varied while keeping the wavelength fixed. Recall from
Section 4.4.2 that the maximum ϕ that can be induced by varying the cylinder radii is
limited by Lϕ and the maximum ∆r that can be used without introducing additional
reflections into the system. From Fig. 4.7(b) it can be seen that the reflectance remains
below approximately 1% in the range −0.025 ≤ ∆r/d ≤ 0.035. This results in a phase
variation of −0.36π . ϕ . 0.36π for an MZI with Lϕ = 10d, as shown by the dashed
curve in Fig. 4.7(a).
0
0.2
0.4
0.6
0.8
1
T
p
p/2
- /2p
-p
0 j
Dr-0.02 -0.01 0 0.01 0.02 0.03
Figure 4.10: Solid (blue) curve: Transmission as a function of ∆r for a rod-type PhC
MZI (Fig. 4.6(a)) with LMZ = 36d and Lϕ = 22d at λ = 1.553µm. Dashed-dotted
curve: phase difference induced by the radius modification ∆r. Dashed (red) curve:
Quartic response given by Eq. (4.9).
The range of ϕ is maximised in the rod-type PhC MZI of length LMZ = 36d by
choosing Lϕ = 22d so that there are only two lattice periods of undisturbed waveguide
separating the modified waveguide section and junctions at either end. The dashed-
dotted curve in Fig. 4.10 shows ϕ as a function of ∆r for λ = 1.553µm and the solid
curve shows the MZI transmittance calculated using the Bloch mode method for the
same wavelength. The dashed curve is calculated using Eq. (4.9) with ϕ = ϕ(∆r) given
by the dashed-dotted curve. Again there is very good agreement between the rigorous
numerical calculation and the semianalytic model derived in Section 4.2. Given the
restrictions placed on ϕ by the choice of tuning method, a longer structure would be
required to obtain a full 2π phase change, but shorter devices may be possible with
alternative tuning methods.
4.4. PHOTONIC CRYSTAL RECIRCULATING MZI STRUCTURES 121
Hole-type PhC MZI
Figures 4.9(c) and (d) show the transmittance as a function of wavelength for two
MZIs of the type shown in Fig. 4.6(b) with Lϕ = 4d and Lϕ = 8d respectively for
r = 0.30d (∆r = −0.02d) and LMZ = 40d. The solid curves are the full numerical result
and the dashed and dashed-dotted curves are given by Eq. (4.8), calculated along the
dashed (Lϕ = 4d) and dashed-dotted (Lϕ = 8d) curves in Fig. 4.8(b), corresponding to
the wavelength variation of ϕ = ∆βLϕ. As in the rod-type PhC example, the parameter
L used in Eq. (4.8) does not match LMZ exactly. For the comparisons in Figs. 4.9(c)
and (d) the best agreement was found for L = 39.26d.
Once again we see that the semianalytic model predicts the finesse of the resonances
with good accuracy. The resonance positions line up closely at the shorter wavelength
end of the transmission region, however the agreement is poorer at longer wavelengths,
as a result of the strong modal dispersion close to cutoff. We have found that the agree-
ment is improved at longer wavelengths if β is replaced by the propagation constant of
the odd supermode β− in Eq. (4.8). This modification is reasonable since the resonances
inside the interferometer are driven by multiple reflections of the odd mode. For the
rod-type PhC examples in Figs. 4.9(a) and (b), the modes are well away from cutoff
and β and β− are almost equal, so the comparison is not improved significantly by this
modification. For consistency however, the semianalytic curves plotted in Figs. 4.9(a)
and (b) are also calculated using β− in place of β.
5 10 15 20 25 30
0.2
0.4
0.6
0.8
1
00
-p
- p2
0
T
L /dj
j
-p/2
- p3 /2
Figure 4.11: Blue data points: transmission as a function of Lϕ for a hole-type PhC
MZI (Fig. 4.6(b)) with LMZ = 72d and r = 0.266 d (∆r = −0.054) at λ = 1.5392µm.
The dotted line between the points is an aid to the eye. Dashed-dotted curve: phase
difference induced by ∆r. Dashed (red) curve: quartic response given by Eq. (4.9).
In order to demonstrate the response of the hole-type MZI as ϕ is varied, we now
consider a similar hole-type PhC MZI to the previous example, but with a length of
122 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
LMZ = 72 d. This longer structure allows us to change ϕ through a complete 2π period.
For this example we fix the radius of the modified cylinders to r = 0.266 d, corresponding
to ∆r = −0.054 d, then vary ϕ by changing the number of modified cylinders, Lϕ.
Figure 4.11 shows the transmitted power as a function of Lϕ for 0d ≤ Lϕ/2 ≤ 32d. The
blue data points were calculated using the full Bloch mode method, and the dashed red
curve is the quartic response given by Eq. (4.9), where the dependence of ϕ on Lϕ is given
by the dashed-dotted curve and the right-hand axis. Observe that there is excellent
agreement between the approximate and the numerical calculation for Lϕ < 15d, but
for Lϕ > 15d the numerical results appear to be shifted relative to the model. We
attribute this to the change in ∆β that results from the radius change in the MZI arm.
In the derivation of the semianalytic model in Section 4.2, we assume that β changes
by ϕ/2, but that ∆β remains small and approximately zero. This is valid when ϕ is
small, but becomes less accurate as ϕ → π. Although this may make it difficult to
obtain the predicted quartic response over a full 2π change in ϕ, the steep switching
behaviour between T = 0 and T = 1 is still displayed in the example shown in Fig. 4.11.
Alternative tuning methods may also make it possible to reduce this effect.
4.5 Semianalytic Bloch mode matrix derivation
In Section 4.2, the transmission equations (4.7) and (4.8) were derived from a simple
modal analysis and compared in Section 4.4.3 with full numerical results calculated
using the Bloch mode method. Identical expressions can be obtained via the Bloch
mode matrix method by considering only the propagating modes in each region, and
filling the Bloch mode scattering matrices with idealised values corresponding to the
assumptions of Section 4.2. This alternative approach is outlined here.
The MZI of Fig. 4.2 can be modelled as a three layer structure as shown in Fig. 4.12.
Two semi-infinite regions form the input and output waveguides on either end of the
central section which consists of the two arms of the interferometer. Transmission
through the three-layer structure is given by the matrix equation
t = T23P (I− R21PR23P)−1 T21, (4.11)
where the operator P describes Bloch mode propagation between interfaces 1 and 2.
We can write this as
P =
[1/√
2 1/√
2
1/√
2 −1/√
2
]−1 [ei(χ−ϕ/2) 0
0 ei(χ+ϕ/2)
] [1/√
2 1/√
2
1/√
2 −1/√
2
](4.12a)
=
[eiχ cos(ϕ/2) −ieiχ sin(ϕ/2)
−ieiχ sin(ϕ/2) eiχ cos(ϕ/2)
], (4.12b)
where the diagonal matrix in Eq. (4.12a) describes propagation through the two arms of
the interferometer, while the other two matrices provide a change of basis between the
odd/even supermodes and the modes of the individual arms. The diagonal elements
4.5. SEMIANALYTIC BLOCH MODE MATRIX DERIVATION 123
T12
R12
T23
R23
T21
R21
P
Interface 1 Interface 2
1 2 3
Figure 4.12: Diagram showing the three layer structure considered in Section 4.5 to
derive a semianalytic Bloch mode scattering matrix model of a recirculating MZI. In-
terface 1 lies between the input waveguide and the two arms and is characterised by
the scattering matrices R12, T12, R21 and T21. Interface 2 lies between the two arms
and the output waveguide and is characterised by the scattering matrices R23 and T23.
of the matrix in Eq. (4.12b) can also be obtained directly from Eqs. (4.2) and (4.4) by
extracting the odd and even mode amplitudes before and after a single pass through
the interferometer.
Analytic approximations to the scattering matrices in Eq. (4.11) are obtained by
considering the transmission and reflection of modes at the interfaces between the input
and output waveguides and interferometer arms. In Section 4.2 we assumed that a mode
in the input guide is perfectly transmitted into the even mode inside the interferometer.
In the Bloch mode matrix notation, this corresponds to setting R12 = 0 and
T12 =
[1
0
],
where the first and second elements of T12 represent transmission into the even and
odd supermodes respectively. Light incident on either interface from inside the inter-
ferometer is filtered by the junction according to the odd and even mode components;
the even mode is perfectly transmitted, while the odd mode is reflected. Hence, the
remaining Bloch mode scattering matrices in Eq. (4.11) can be written as
R21 = R23 =
[0 0
0 1
]T21 = T23 = [1 0] .
Equations (4.7) and (4.8) can be derived by substituting the analytic matrix expressions
above into Eq. (4.11).
Although this method of deriving the semianalytic result may seem identical to the
modal method in Section 4.2, it provides an intermediate step between an ideal model
124 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
and the a full numerical treatment. For instance, if the full Bloch mode matrices are
truncated to include only the propagating modes, the complex reflection and trans-
mission coefficients are retained but the matrices have the dimensions of the idealised
version above and can thus be manipulated very efficiently. Alternatively, the effects of
non-ideal transmission and reflection can be included in an intuitive way by modifying
the analytic matrix expressions appropriately.
4.6 Recirculation in coupler-based MZIs
Recall from Section 4.2 that mode recirculation occurred because the input and output
waveguides of the MZI only supported a single mode of even symmetry, and thus any
odd field component created inside the interferometer is reflected by both junctions.
In Section 4.3 we showed that rather than degrading the performance of the device,
recirculation enhances the response for some wavelengths, and so it may be a desirable
feature to generate in other MZI geometries. While mode recirculation is likely to occur
in PhC-based MZIs of the type Fig. 4.1(c), MZIs of the type illustrated in Figs. 4.1(a)
and (b) do not exhibit the same behaviour because the beam splitter and coupler based
junctions can transmit all incident fields. Mode recirculation can be achieved in a
such MZIs, however, by blocking one each of the input and output ports and reflecting
the light back into the interferometer. The resulting transmission characteristics are
identical to those derived in Section 4.2. Here we use a conventional MZI analysis with
the addition of reflections from mirrorsm1 andm2 to model the bulk-optics recirculating
MZI of the type illustrated in Fig. 4.1(a). A similar approach could also be applied to
the coupler-based MZI in (b).
in
out 1
50/50 beamsplitters
m1
in
out
50/50 beamsplitters
h
h
out 2(a) (b)
m2
L1
L2
L1
L2
B1
B2
B1
B2
Figure 4.13: (a) Conventional bulk-optics MZI using beam splitters. (b) Modified bulk-
optics MZI designed to exhibit recirculation effects. The two additional mirrors m1 and
m2 reflect light back into the interferometer.
Figures 4.13(a) and (b) respectively show the conventional bulk optics MZI from
4.7. DISCUSSION 125
Fig. 4.1(a) and a modified version designed to exhibit mode recirculation. Consider first
the design illustrated in (a). Light entering via the input port is split at the first beam
splitter B1 and the two beams travel different paths before being recombined at the
second beam splitter B2. Suppose that the two arms of the interferometer have lengths
L1 and L2 = L1 + δL respectively, as shown in Fig. 4.13. A beam propagating along
path 1 therefore advances in phase by Φ1 = k0L1, while a beam propagating along path
2 advances by Φ2 = k0(L1 + δL) = Φ1 + ϕ, where k0 = 2π/λ. The light transmitted
into output 1 is a superposition of two beams; the first is transmitted through B1 then
propagates a distance L1 and is reflected at B2; the second beam is reflected at B1 then
propagates a distance L2 and is transmitted through B2. The superposition of these
two beams and their accumulated phases is given by
t1 = τeiΦ1ρ+ ρeiΦ2τ = 2ρτeiΦ1 cos(ϕ/2), (4.13)
where we have assumed that ρ = 1/√
2 exp(iφr) and τ = 1/√
2 exp(iφt) are the re-
flection and transmission coefficients of the beam splitters B1 and B2. Therefore, the
transmitted intensity for the single pass MZI is T1 = cos2(ϕ/2), identical to Eq. (4.3).
Similarly, the field transmitted into output 2 is given by
t2 = τeiΦ1τ + ρeiΦ2ρ = ie2iφreΦ1 sin(ϕ/2), (4.14)
where we have taken the fields reflected by the beam splitters to be π/2 out of phase
relative to the transmitted fields i.e. φt = φr + π/2. Comparing Eqs. (4.13) and (4.14)
with Eq. (4.2) it can be seen that the field transmitted at output 1 is equivalent to the
even mode component in Eq. (4.2), while the field transmitted at output 2 is equivalent
to the odd mode component.
Now consider the effect of placing two additional mirrors m1 and m2 in the positions
indicated in Fig. 4.13(b). Mirror m1 reflects the field t2 back into the interferometer
with an additional phase of exp(2ik0h) due to propagation between B2 and m1. The
reflected beam splits and propagates through the interferometer back to B1 where some
is transmitted into the input port, while the rest is reflected at m2. This selective re-
flection at either end of the interferometer is analogous to the odd-mode recirculation
between the junctions of the PhC MZIs in Section 4.4, and results in identical recir-
culation properties. By tracking the recirculating beam back and forth through the
interferometer, the total transmitted power is found to be given by Eq. (4.8), where
χ = βL+ ϕ/2 is replaced by χ = k0(L1 + 2h) + φr + ϕ/2.
4.7 Discussion
In this chapter we have shown that MZIs formed from waveguides with a perfectly
reflecting cladding display significantly different transmission characteristics to conven-
tional MZIs. In particular, reflections due to modal symmetry mismatches result in
mode recirculation and resonant transmission behaviour. Photonic crystals provide one
126 CHAPTER 4. RECIRCULATING MACH-ZEHNDER INTERFEROMETER
means of obtaining a reflecting cladding with the required properties and we have pro-
posed two PhC MZI designs that exhibit the predicted mode recirculation behaviour.
The semianalytic model derived in Section 4.2 gives insight into the physics of MZIs
and provides a powerful design tool for predicting the properties of relatively com-
plicated devices knowing only the dispersion properties of a single waveguide. When
combined with rigorous numerical techniques such as the Bloch mode matrix method,
new designs can be optimised rapidly over a broad parameter space. An example of this
was given in Sections 4.3 and 4.4.3 where we first identified using Eq. (4.8), the condi-
tions under which a recirculating MZI exhibits a quartic phase dependence (4.9), then
demonstrated the predicted behaviour in a PhC MZI (Fig. 4.10). The model provided
an accurate estimate of the optimum wavelength and structural parameters, thereby
saving a considerable amount of computational time compared to an all-numerical pa-
rameter search.
Although several PhC MZI designs have been fabricated, the recirculation properties
that we predict here have not been identified experimentally for several reasons. First,
the devices studied in Refs. [200, 201, 220] all use directional coupler-based junctions
that transmit both odd and even symmetry modes, and are thus inherently single-pass
MZIs. Second, although the designs in Refs [218] and [219] both have single input
and output waveguides, and would thus be expected to exhibit recirculation properties,
the junctions do not have the necessary high transmission properties identified in Sec-
tion 4.4.1 to observe clearly the odd-mode resonance effects. The coupled Y-junction
proposed in Chapter 3 exhibits high transmission over a wide bandwidth and is thus
an ideal junction for designing efficient MZIs.
A theoretical and experimental investigation of an MZI with a realistic tuning mech-
anism, rather than the model adopted for convenience in Section 4.4.2 would be of
interest. Among other things, this would provide a better estimate of the length re-
quirements for tuning the phase. Three-dimensional modelling in a slab design would
also be valuable.
Chapter 5
Efficient wide-angle transmission
into rod-type photonic crystals
In this chapter we demonstrate highly efficient coupling into uniform rod-type photonic
crystals with unmodified interfaces. Reflected powers of less than 1% are calculated for
two-dimensional structures at incident angles up to 80◦. These properties are closely
related to forward scattering resonances of the high index cylinders, and as a con-
sequence, are not displayed by rod-type photonic crystals. We use these results to
optimise a realistic PhC slab structure to exhibit both high efficiency coupling and
auto-collimation and present numerical results for two- and three-dimensional finite
difference time domain calculations. The results could also be applied to the design of
efficient superprisms and other in-band dispersion based PhC devices.
This chapter contains a substantial amount of unpublished work in Sections 5.1–5.4
which is currently being prepared for publication. The results presented in Section 5.5
are based on Ref. [226] which is due to appear in Applied Physics Letters in September
2005.
5.1 Introduction
We now turn our attention away from photonic bandgap devices and consider what is
perhaps a more fundamental photonic crystal problem, that of coupling light from a
homogeneous dielectric material such as free-space into the propagating Bloch modes of
a uniform PhC. As we saw in Section 1.8, the unusual dispersion properties of PhC bands
lead to a range of interesting effects such as superprism behaviour and auto-collimation.
To use these effects in practical devices it is essential that light can be coupled efficiently
into and out of the PhC structure with minimal back-reflection. This is necessary not
only to keep the insertion losses low, but also to prevent light from being scattered
into other parts of a device where it may cause interference and cross-talk. The latter
issue will especially important in compact integrated optical circuits consisting of many
components on a single chip.
127
128 CHAPTER 5. EFFICIENT WIDE-ANGLE TRANSMISSION
While one of the main attractions of PhC devices is their compact size, this also
presents many practical challenges when interfacing these PhC components to con-
ventional optical systems. Several efficient methods have been developed for coupling
to waveguide-based devices [8, 227–230], but the challenge remains to find a practical
solution for coupling light into the extended Bloch modes of a uniform PhC lattice for in-
band applications. The main approach reported in the literature for air-hole type PhCs
has been to modify one or several of the interface layers to produce an anti-reflection
coating on the front and back faces of the structure. One such method is analogous to
the application of thin-film anti-reflecting coatings to optical elements, where resonant
effects are used to enhance the transmission properties over a narrow frequency band-
width, although the bandwidth can be improved with multi-layered gratings [231]. A
second approach, proposed by Baba et al. [232] is to change the size and/or shape of the
air holes to gradually modify the field profile as it approaches the PhC slab, effectively
creating an apodised interface. A third possibility suggested by Witzens et al. [233] is
to place a waveguide above a PhC slab and use evanescent coupling to transfer light
from the waveguide into counter-propagating Bloch modes of the PhC.
Here we study the transmission properties of rod-type, rather than hole-type, PhCs
and demonstrate that it is possible to couple more than 99% of incident light into such
structures for a wide range of incident angles without making any modifications to the
interface. We start in Section 5.2 by considering the basic 2D geometry illustrated in
Fig. 5.1, in which a plane wave is incident on a semi-infinite PhC from a homogeneous,
semi-infinite region above. The incident medium is taken to continue into the PhC
as the background material. Both square and triangular rod-type PhCs oriented as
in Figs. 5.1(a) and (b) respectively, are found to exhibit almost perfect wide-angle
transmission features for a wide range of structural parameters.
In Sections 5.3 and 5.4 we show that this behaviour is closely related to the scattering
properties of the individual cylinders which in turn influence the transmission properties
of the gratings that form the PhC lattice. We identify the conditions under which
high efficiency coupling can occur and show in Section 5.4 that these provide a simple
explanation as to why efficient coupling is achieved more easily in rod-type PhCs than
in hole-type PhCs. In Section 5.5 we optimise a purely 2D rod-type PhC to exhibit
simultaneously very low reflectance and auto-collimation. The successful combination
of these two properties is demonstrated for an incident Gaussian beam using Finite
Difference Time Domain (FDTD) simulations, first for a purely 2D geometry, and finally
in a realistic rod-type PhC slab consisting of finite length silicon rods in a background
of silica.
5.2 Plane wave coupling to semi-infinite PhCs
In this section we consider the two geometries shown in Figs. 5.1(a) and (b), where a
plane wave is incident at an angle θi onto a semi-infinite 2D PhC consisting of dielectric
5.2. PLANE WAVE COUPLING TO SEMI-INFINITE PHCS 129
0 20 40 60 800.1
0.2
0.3
0.4
0.5
0.6
0.7
d/l
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 20 40 60 80
d/l
Transmission
0 0.2 0.4 0.6 0.8 1
G X
Mqi
(a)
KM
G
qi
(b)
(c) (d)
qi (degrees) qi (degrees)
Figure 5.1: Geometry and transmittance contour plots for a plane wave incident on a
semi-infinite PhC of rods (ncyl = 3) in air (n = 1). (a) Square lattice oriented with
Γ − X direction normal to the interface. (b) Triangular lattice with Γ −M direction
normal to the interface. (c) and (d) Transmittance contour plots for varying θi and d/λ
for the PhCs in (a) and (b) respectively. Below the dashed white line there is only a
single propagating plane wave grating order.
rods of refractive index ncyl, and radius r embedded in a dielectric background nb. To
calculate the transmission into the PhC, we use the Bloch mode method to calculate the
matrix R∞, defined in Section 2.3.2 as the plane wave reflection scattering matrix for a
plane wave incident from free-space onto a semi-infinite PhC. Since we are interested in
the reflected power, and no energy is carried by the evanescent states, R∞ is truncated
to contain only those elements corresponding to reflection into propagating plane wave
orders. Thus, when only a single reflected order is supported, R∞ can be treated as a
scalar and the total reflected power is R = |R∞|2. When multiple reflected orders are
present, the total reflectance is calculated as a sum over the reflectance into each order.
For the purely 2D structures that we consider first, the transmittance is then simply
calculated as T = 1−R.
The Bloch mode method is ideal for these calculations, as it is very efficient and can
130 CHAPTER 5. EFFICIENT WIDE-ANGLE TRANSMISSION
20
40
60
80
00.25 0.3 0.35 0.4
qm
ax(d
egre
es)
r/d
T>99%T>95%T>90%
20 40 60 80
0.992
0.994
0.996
0.998
1
00.99
qi (degrees)
T
(a) (b)
Figure 5.2: (a) Transmittance as a function of incident angle for the square (solid
curve) and triangular (dashed curve) lattices in Fig. 5.1. Note the vertical axis starts
at T = 99%. (b) Maximum acceptance angle θmax plotted as a function of normalised
radius for a square lattice of rods with ncyl = 3.0 in air. The three curves correspond
to T > 90%, 95%, and 99% in the range 0 ≤ θi ≤ θmax.
deal with semi-infinite structures without the need for absorbing boundary conditions
or other techniques required by many other simulation methods to avoid reflection from
the rear surface of finite structures. In addition, the method implicitly calculates the
scattering properties of a single grating, from which the PhC lattices are formed. As we
show in Section 5.4, studying the properties of a single grating provides an intermediate
case between a single cylinder and the properties of a full 2D lattice.
The contour plots in Figs. 5.1(c) and (d) show the transmittance as a function
of incident angle and normalised frequency for a TM-polarised plane wave incident
on the semi-infinite PhCs shown in (a) and (b) respectively. The PhCs both have
period d, and consist of rods with ncyl = 3.0 and r = 0.34d embedded in free-space
(nb = 1). Recall from Section 1.8.1 that when coupling into a PhC, the incident
wavevector component parallel to the interface (kx here) must be conserved and so the
incident field can only couple to specific regions of each PhC band. We can therefore
understand the relatively complex structure of the transmission plots in Figs. 5.1(c)
and (d) as a projection of the band diagram onto the kx component of the incident
wavevector, where kx = 2π/λ sin(θi). Hence, the dark blue regions of zero transmission
correspond to full and partial bandgaps in the projected band structure. Note that the
transmission into the lowest frequency band does not exceed 90%, but almost perfect
transmission is possible into some of the higher order bands.
One of the most striking features in Fig. 5.1 is the dark red, high transmission
‘finger’ extending almost horizontally across both transmission plots at approximately
d/λ = 0.34, corresponding to transmission into the second band in both cases. At this
frequency, indicated by the dotted green lines in Figs. 5.1(c) and (d), the transmittance
into the square lattice PhC exceeds 99% for incident angles 0◦ ≤ θi ≤ 80◦. This can
5.2. PLANE WAVE COUPLING TO SEMI-INFINITE PHCS 131
be seen in Fig. 5.2(a), which shows the transmittance as a function of incident angle at
d/λ = 0.34 for the two lattices. The triangular lattice exhibits T > 99% for all angles
in the range 0◦ ≤ θi ≤ 37◦. Observe that a second region of almost perfect transmission
occurs at frequencies close to d/λ = 0.6, for which the transmission exceeds 99% for
θi ≤ 25◦ in both cases. Such efficient coupling over a wide range of angles is far
superior to the published results for air hole PhCs. In the remainder of this section and
Sections 5.3–5.4 we identify the physical origin of this behaviour and use the results to
explain why hole-type PhCs do not have similar properties.
The high transmission features exhibited by the PhCs in Fig. 5.1 are not unique
to the structural parameters chosen for the examples. In fact, highly efficient, wide-
angle coupling appears to be a common feature of rod-type PhCs, occurring for incident
fields of both TE and TM polarisations and for hexagonal and square lattices with a
range of refractive indices and radii. As an example of the robust nature of the high
transmission, Fig. 5.2(b) shows the maximum acceptance angle θmax for transmission
into the second PhC band, plotted as a function of cylinder radius for a square lattice
of rods with ncyl = 3.0 in air. The three curves correspond to T > 90%, 95%, and 99%
for all incident angles 0 ≤ θi ≤ θmax.
2.5 3.0 3.5 4.0
0.24
0.22
0.2
0.18
0.16
0.14
0.12
0.1
0.08
r/l
normalised lattice period (d/r)
T
0
0.2
0.4
0.6
0.8
1
Figure 5.3: Plane wave transmission for normal incidence into a square lattice of rods
with fixed radius r and ncyl = 3.0 plotted as a function of normalised lattice period d/r
and normalised frequency r/λ. The dashed black curves are the T = 98% contours. The
green dotted lines at r/λ = 0.115 and r/λ = 0.204 indicate the maximum transmission
frequencies for r = 0.34d (d/r = 2.94) as shown in Figs. 5.1(c) and (d).
Another unusual property of the high transmission features is weak dependence on
the lattice geometry. While the frequency at which they occur is a strong function of
the cylinder radius and refractive index, it is a much weaker function of the cylinder
132 CHAPTER 5. EFFICIENT WIDE-ANGLE TRANSMISSION
position. Evidence of this is seen in Figs. 5.1(c) and (d) where the high transmission
bands occur at d/λ = 0.34 and d/λ = 0.6 for both the square and the triangular
lattice. Perhaps more surprising is the observation that the maximum transmission
frequencies change very little when the cylinder separation (lattice period) is varied but
the cylinders remain unchanged. This is demonstrated in Fig. 5.3, which is a contour
plot of the plane wave transmittance for normal incidence into a square lattice similar
to that of Fig. 5.1(a), but with a fixed cylinder radius and varying lattice period d. The
horizontal axis is expressed as the normalised lattice period d/r and the vertical axis is
the frequency r/λ, also normalised to the cylinder radius. In these units, the structure
calculated in Fig. 5.1(c) with r = 0.34d corresponds to d/r = 2.94. Observe that the
two high transmission bands occur at approximately r/λ = 0.115 and r/λ = 0.205 and
do not change significantly as the lattice period is varied by more than 60%. Other
features in the same transmission plot are clearly influenced more strongly by a change
in lattice period. This very weak dependence on structure suggests that the efficient
coupling in rod-type PhCs is driven by a property of the individual rods, rather than
by a property of the periodic lattice. We investigate this further in Sections 5.3 and
5.4.
5.3 Single cylinder scattering
In this section we show that the high transmission features of rod-type PhCs are closely
related to the scattering characteristics of the individual rods. We identify two scat-
tering parameters, namely the scattering cross-section and the asymmetry parameter,
that provide an intuitive physical interpretation of the results in Section 5.2.
Although photonic crystals rely intrinsically on coherent scattering for many of their
unique properties, the individual scattering elements contribute to the detailed charac-
teristics of the structure. In Refs. [234,235] it was shown that there is a correspondence
between the position of bandgaps in three-dimensional (3D) PhCs formed from dielec-
tric spheres, and the Mie resonances of the spheres. Similar results have been shown for
2D PhCs consisting of dielectric cylinders [236] but to our knowledge, the relationship
between scattering resonances and PhC transmittance has not been directly studied.
Here we show that there is not only a correspondence between cylinder resonances and
band position, but also a direct relationship between forward scattering and the trans-
mittance of plane waves incident on a PhC. We note that although the single cylinder
treatment presented in this section is strictly 2D, the results are also applicable to
realistic PhC slab structures, as demonstrated in Section 5.5.
The scattering characteristics of an infinite dielectric cylinder are well understood,
and can be calculated for non-conical incidence following the treatment of van de
Hulst [237]. Results for TM polarization are presented here, but we note that sim-
ilar scattering behaviour also occurs for TE polarization. Consider first a TM polarised
wave in a background medium of index (nb = 1), incident on a dielectric cylinder of
5.3. SINGLE CYLINDER SCATTERING 133
r
q
r
Figure 5.4: Single cylinder scattering geometry. An incident plane wave from above is
scattered in directions θ, where θ = 0 corresponds to forward scattering.
radius r and index ncyl in the plane perpendicular to cylinder axis as shown in Fig. 5.4.
The total field outside the cylinder is expressed as the sum of the incident and scattered
components, which allows weak scattering effects to be separated from the dominant
incident field. At large distances ρ from the cylinder, the scattered field can be written
as [237]
Ez ∝√
2
πkρexp [−ikρ+ iωt− i3π/4]T (θ), (5.1)
where k = 2π/λ,
T (θ) = b0 + 2∞∑
n=1
bn cos(nθ), (5.2)
and1
bn= 1 + i
ncylJ′n(ncylkr)Yn(kr)− Jn(ncylkr)Y
′n(kr)
ncylJ ′n(ncylkr)Jn(kr)− Jn(ncylkr)J ′n(kr).(5.3)
Note that these equations can be applied to an arbitrary index contrast after appropriate
scaling of ncyl and k.
Two scattering parameters relevant to this discussion are the scattering cross-section
per unit length of the cylinder,
cscat =2
πk
∫ 2π
0
|T (θ)|2dθ =4
k
∞∑n=−∞
|bn|2, (5.4)
and the asymmetry parameter, g = 〈cos(θ)〉, the average cosine of the scattering angle.
The scattering cross-section is a measure of the total energy scattered by the cylinder
in all directions, while the asymmetry parameter gives an indication of whether the
scattered energy is in the forward or backward direction. Here, |g| < 1 and g > 0
(g < 0) corresponds to dominant forward (backward) scattering. In the long wavelength
134 CHAPTER 5. EFFICIENT WIDE-ANGLE TRANSMISSION
limit, as k → 0, the bn coefficients (5.3) scale approximately as k2(n+1), and hence the
field expansion (5.1) is dominated by the b0 term, and cscat → 0.
0.1 0.2 0.3 0.4 0.5 0.6
0.2
0.4
0.6
0.8
1
0 0.70
d/l
T
0.05 0.1 0.15 0.2
(c) (g)(f)(e)(d)
r/l
csca
t,g
(a.u
)
0
(a)
(b)
Square
Triangular
cscat
g
0
Figure 5.5: (a) Normal incidence transmission spectrum into the square (solid curve)
and the triangular (dashed curve) PhCs in Fig. 5.2.(b) Single-cylinder scattering cross-
section (dashed curve) and asymmetry parameter g (dashed-dotted curve) plotted as a
function of r/λ for the same frequency range as (a). (c)–(g) Polar plots the scattering
pattern |T (θ)|2/k for r/λ = 0.034, 0.085, 0.116, 0.163 and 0.204. The light is incident
from above.
Figure 5.5(a) shows the normal incidence transmittance spectrum for the square
and triangular lattices considered in Fig. 5.1. Note that the transmission peaks at
d/λ = 0.34 and d/λ = 0.6 almost coincide for the two structures, even though the band
gap frequencies shift significantly. The curves plotted in Fig. 5.5(b) show the scattering
cross-section and asymmetry parameter as a function of r/λ for light scattered from
a single cylinder of index ncyl = 3.0 in air. Figures 5.5 (c)-(g) are polar plots of the
scattering profile given by |T (θ)|2/k, where T (θ) is defined in Eq. (5.2). The profiles
are plotted for frequencies in the first band, the first gap, the second band, the second
gap and the third band respectively. Observe from Eq. (5.4) that the area enclosed by
these scattering diagrams is proportional to the scattering cross-section.
Figure 5.5 shows a clear correspondence between the position of the bands (indicated
by non-zero transmittance) and peaks in cscat, as has been observed by others [235,236].
5.4. GRATING PROPERTIES 135
Observe also that the transmittance into the second and third bands is better than
99%, whereas transmittance into the first band does not exceed 90% – a feature we
also identified in Section 5.2. The asymmetry parameter curve and the polar plots
provide an explanation for this difference. In the second and third bands, where the
transmittance is highest, the peak in cscat coincides approximately with a peak in g,
indicating a forward scattering resonance. In the first band g remains positive but
approaches zero, despite there being a small, broad peak in cscat. Thus, although
there is still significant scattering from each cylinder at low frequencies, the scattering
direction is almost isotropic, with Eqs. (5.1), (5.2), and (5.4) being dominated by the
b0 term as described above. This is clearly seen in the scattering patterns plotted
in Figs. 5.5(c) and (d). As the frequency is increased, higher-order scattering terms
become significant, thus introducing dipole-like and higher order scattering behaviour
which can result in strong directional scattering. For example, the cscat peaks near
d/λ = 0.36 and d/λ = 0.60 correspond to maxima in b1 and b2 respectively, with both
resonances producing strong forward scattering, indicated by the g curve of Fig. 5.5(b)
and the scattering patterns in Figs. 5.5(e) and (g).
The observation that forward scattering resonances enhance the transmission into a
PhC is a physically intuitive explanation for the results presented in Sec. 5.2, however
we have not discussed the possibility of coupling between closely spaced cylinders. In
Section 5.4 we consider the properties of a grating formed from a single line of cylinders,
which provides an intermediate case between the single cylinder results of Section 5.3
and the full 2D PhC results in Section 5.2.
5.4 Grating properties
Recall from Chapter 2 that we consider a PhC to be a stack of identical cylinder grating
layers that are coupled together by diffracted plane wave orders given by the grating
equation (2.2). In this chapter we consider only gratings with period d consisting of
identical cylinders and so we would expect the scattering properties of a grating to
approach those of the individual cylinders as d/λ → ∞. For finite lattice periods we
would still expect the forward scattering resonances of the single cylinders identified in
Section 5.3 to have strong influence on the scattering properties of a grating formed
from those cylinders.
We consider first the transmission properties of the grating that is used to form the
PhCs shown in Fig. 5.1. The square and triangular lattices are both formed from the
same grating by stacking each layer as described in Section 2.2.2. Figure 5.6(a) shows
the transmittance as a function of θi and d/λ for the grating of period d formed from rods
with ncyl = 3.0 and r = 0.34d. The results are calculated directly from the plane-wave
scattering matrix T, where we sum the total transmittance over all propagating grating
orders. Observe that the two high-transmission features of the grating at d/λ = 0.34
and d/λ = 0.6 resemble closely those in Figs. 5.1 (c) and (d). The agreement is even
136 CHAPTER 5. EFFICIENT WIDE-ANGLE TRANSMISSION
more apparent in Fig. 5.6(b) where we compare the normal incidence reflectance of the
grating and the square PhC with the same parameters. The strong correlation between
low grating reflectance and efficient coupling into the PhC further reinforces the single
cylinder scattering explanation. It also explains why the transmission peaks occur at
the same frequencies in the square and triangular lattices.
0 20 40 60 80
0.1
0.2
0.3
0.4
0.5
0.6
0.7
qi
d/l
0 0.2 0.4 0.60.001
0.01
0.1
1
d/l
R
(a) (b)
T
0
0.2
0.4
0.6
0.8
1
Figure 5.6: (a) Plane wave transmittance through a grating of period d consisting of rods
with ncyl = 3.0 and r = 0.34d, plotted as a function of incident angle and normalised
frequency. (b) Plane wave reflectance spectra (log scale) for normal incidence on the
grating (dashed red curve) and a semi-infinite square PhC lattice formed from the
grating (solid blue curve).
The challenge of establishing a rigorous proof of the relationship between the forward
scattering resonances of a cylinder and high grating transmission is a topic of ongoing
research, and is not solved in this thesis. The multipole method may be particularly
helpful in this regard as it includes implicitly the scattering properties of the individual
cylinders in the grating scattering formulation [176] thereby providing a mathematical
connection between the two structures. An alternative approach may be to consider
first the interaction of the scattered fields from two neighbouring cylinders, then use a
tight-binding approximation similar to that of Lidorikis et al. [236] to characterise the
scattering from a grating. We show here, however that the single-cylinder interpreta-
tion provides a simple physical explanation as to why rod-type PhCs exhibit superior
coupling properties compared to hole-type PhCs in agreement with our earlier results.
This argument relies on key observations: the first is that strong forward scattering is
required from each cylinder, and the second is that there must only be a single propagat-
ing diffracted grating order. The second condition can be understood in the following
way.
When a plane-wave is incident on a grating with period d, the number of propagating
grating orders depends on both the ratio of the incident wavelength to the period and
the incident angle. The condition for only a single reflected and single transmitted
grating order to exist can be derived from the grating equation (2.1). If the background
5.5. EFFICIENT COUPLING FOR AUTO-COLLIMATION 137
material has a refractive index nb, then the single order condition can be expressed as
d
λ<
1
nb(1 + sin(θi)), (5.5)
where λ is the free-space wavelength. Consider now the properties of the grating at a
forward scattering resonance of the cylinders. If Eq. (5.5) is satisfied, the light scattered
forward by each cylinder can only couple into the zeroth (specular) grating order, and
thus there is very little overlap between the scattered fields resulting in high transmission
through the grating. If multiple grating orders are present, they provide a means for
the fields scattered by each cylinder to interfere with each other via coupling to higher
diffraction orders. When this occurs, phase differences between the zeroth and higher
orders become an additional complicating factor. Moreover, it is considerably easier to
satisfy the conditions for zero reflectance into a single grating order than into multiple
reflected orders simultaneously. The relationship between single cylinder scattering and
grating transmission thus becomes much more complex and the simple arguments that
we applied in Section 5.3 are no longer valid. This interpretation is consistent with our
observations that the zero reflectance features in Fig. 5.6(b) disappear at frequencies
where multiple grating orders exist.
The results and observations of this section and Section 5.3 provide a simple expla-
nation as to why rod-type PhCs typically exhibit highly efficient coupling over a wide
range of incident angles, whereas air-hole PhCs may require interface modifications in
order to achieve satisfactory results [231–233]. In Section 5.3 we showed that the reso-
nance condition for strong forward scattering is not satisfied at low frequencies where
light is scattered almost isotropically. The same applies to scattering from cylinders
with a low refractive index contrast relative to the background. Efficient coupling is
therefore more likely to occur in the second PhC band or higher, while additional grating
orders appear if the frequency is too high. For rod-type PhCs, where the background
refractive index is lower than that of the rods, these two conditions are relatively easy to
satisfy, as shown in Figs. 5.1(c) and (d) where the dashed white lines indicate indicate
the upper frequency limit given by Eq. (5.5). In air hole PC structures it is much more
difficult to satisfy both conditions simultaneously since non-isotropic scattering from
air holes only occurs at relatively high frequencies but at the same time the high index
background material reduces the frequency at which multiple plane wave orders appear.
For example, if the background material is silicon (nb = 3.4) the first non-zero grating
order appears for large angle of incidence at frequencies above d/λ = 0.15, which is well
below the second band for most air-hole PhCs.
5.5 Efficient coupling for auto-collimation
So far in this chapter we have demonstrated very low reflection coupling to rod-type
PhCs over a wide range of incident angles, however we have not considered how the light
138 CHAPTER 5. EFFICIENT WIDE-ANGLE TRANSMISSION
propagates once inside the structure. In order to exploit the efficient coupling for in-
band applications such as superprism effects and auto-collimation, we must optimise the
PhC structure so that the light is coupled into bands which have the desired dispersion
properties for these effects. In this section we present a realistic rod-type PhC slab
design that exhibits auto-collimation and less than 1.5% reflection loss at each interface
for angles of incidence up to θi = 25◦. We first describe in Section 5.5.1 the optimisation
process using 2D calculations, then in Sections 5.5.2 and 5.5.3 we present both 2D and
3D results that demonstrate the successful combination of the two effects. We also show
that this simple PhC structure can be used as a collimating beam combiner.
n = 1.460slab
n = 1.458clad
n = 1.458clad
Si rodsn = 3.4cyl
h =3d
x
zy
qi
Figure 5.7: Geometry of the rod-type PhC slab consisting of a triangular lattice of
silicon rods (ncyl = 3.4) in a slab of silica (nslab = 1.460) of height h = 3d where d is
the lattice period. The structure is clad above and below by a cladding with refractive
index nclad = 1.458.
Figure 5.7 shows the PhC slab geometry considered here, consisting of a triangular
lattice of silicon rods (ncyl = 3.4) embedded in a slab of silica with index nslab = 1.46.
The lattice is oriented such that the interface is parallel to the Γ −M direction. The
slab and the cylinders have a finite height of h = 3d, and they are capped above
and below with a cladding of index nclad = 1.458. As we demonstrate in this section
a cylinder radius of r = 0.35d results in both low reflectance and auto-collimation
properties. This geometry resembles that proposed by Martinez et al. [50], although
the slab height is increased to improve the mode overlap with the incident Gaussian
beam in the 3D simulations. The design and modelling of this structure is performed
using a combination of 2D and 3D Finite Difference Time Domain (FDTD) simulations,
in addition to Bloch mode method calculations and the use of commercial plane-wave
band structure software.
5.5. EFFICIENT COUPLING FOR AUTO-COLLIMATION 139
5.5.1 Design
We describe here the process of optimising the PhC in Fig. 5.7 for auto-collimation
and transmittance at incident angles of θi = 22.5◦, as this allows the PhC to be used
as a simple beam combiner for two incident beams with an angular separation of 45◦.
Although designed for a specific incident angle, the properties of the optimised structure
are relatively robust. The reflectance remains below 1.5% for all incident angles θi <
25◦, and auto-collimation is observed throughout this range.
We consider first the transmission of a plane wave incident at θi = 22.5◦ on the PhC
in Fig. 5.7, which we approximate as a 2D PhC lattice of rods with index ncyl = 3.4 in
a background of index nb = 1.458. As we demonstrate in Sections 5.5.2 and 5.5.3, the
2D results provide a good approximation to the 3D slab structure, thus providing an
efficient tool for optimizing parameters before launching computationally intensive 3D
simulations. The contour plot in Fig. 5.8(a) shows the transmittance into the second
band of the PhC as a function of the cylinder radius, where the contours correspond
to transmittances of T = 90%, 95%, 98% and 99%. These results were obtained using
the Bloch mode method as described in Section 5.2. Observe that for all cylinder radii
in the range 0.31 ≤ r ≤ 0.4 there are frequencies for which the transmittance exceeds
99%. This region defines the ideal part of parameter space for obtaining efficient auto-
collimation.
r/d
d/l
0.3 0.32 0.34 0.36 0.38 0.4
0.24
0.26
0.28
0.3
0.32
0.34
kx
ky
p/2
p
- /2p
-p
p/2 p- /2p-p
0
0
T>99%
(a) (b)
Figure 5.8: (a) Transmittance contour plot for a plane wave incident at θi = 22.5◦ onto
a 2D triangular PhC lattice of silicon rods embedded in silica. The contour levels are
at T = 90%, 95%, 98% and 99%. The green points represent the optimum frequency
for collimation at θi = 22.5◦. (b) Background: contour plot of 1/p calculated over the
Brillouin zone for the second band of the PhC in (a) with r = 0.35d. White regions
correspond to maximum 1/p (strong collimation). Blue curves: equi-incident angle
contours for θi = ±22.5◦. Red curve: d/λ = 0.29 equifrequency contour.
140 CHAPTER 5. EFFICIENT WIDE-ANGLE TRANSMISSION
Recall from Section 1.8.4 that auto-collimation occurs when an incident beam cou-
ples into a Bloch mode where there is a point of inflection in the equifrequency curve
[141]. At this point, the small range of incident k-vectors present in the beam all
propagate in approximately the same direction within the crystal, and thus the beam is
collimated. If the equifrequency curve is relatively flat, then two beams can be launched
into the crystal at different angles, but once inside, they are collimated and propagate
through the crystal in approximately the same direction. Quantitatively, collimation is
measured by the parameter p defined in Eq. (1.9) [126]. For good collimation, we re-
quire small p values so that the beam direction inside the crystal θc, varies slowly with
θi. The green circles plotted on top of the transmission contours in Fig. 5.8 (a) show
the optimum collimation frequency as a function of the cylinder radius, for θi = 22.5◦.
The diagram in Fig. 5.8(b) illustrates how the optimum collimation frequency is
calculated for each cylinder radius using a similar approach to Baba et al. [238]. We
first calculate the 2D band surface for the second band using the commercial software
BandSOLVE from which we evaluate the p parameter over the Brillouin zone as shown
in Fig. 5.8(b), where the grey contours in the background represent the value of 1/p,
calculated for a cylinder radius of r = 0.35d. The narrow white regions of the 1/p plot
correspond to large 1/p values, and hence indicate the regions where collimation occurs
(the horizontal white line on the ky = 0 axis indicates the interface). The blue curves
in Fig. 5.8(b) are the equi-incident angle contours [126] that indicate the regions of the
Brillouin zone that can be accessed by varying the frequency while θi = 22.5◦. The
green circle indicates the region we wish to couple into with the incident beam, defined
by the intersection of the blue equi-incident angle contour and the white collimation
region. Thus, the optimum frequency for collimation at θi = 22.5◦ is given by the
frequency of the second band at this intersection point. In the example shown here, the
optimum collimation frequency for r = 0.35d is approximately d/λ = 0.290, indicated
by the red equifrequency curve in Fig. 5.8(b). To plot the green line in Fig. 5.8(a)
we repeat the process described above for a range of cylinder radii. The results show
that while high transmittance and auto-collimation are independent properties, they
are not incompatible. Figure 5.8(a) shows that both properties occur simultaneously
for cylinder radii in the range 0.335d ≤ r ≤ 0.382. For the results presented in the
remainder of this chapter we consider rods with r = 0.35d, for which the optimum
collimation occurs at approximately d/λ = 0.29, as shown in Fig. 5.8(b).
5.5.2 Results: 2D vs 3D coupling properties
Before presenting the collimation results, we demonstrate first that the efficient coupling
conditions identified with the 2D calculations in Section 5.5.1 provide a good approxi-
mation to the slab structure in Fig. 5.7, which requires full 3D numerical simulations.
The commercial FDTD software FullWave is used for the 3D calculations with perfectly
matched layer boundary conditions on all sides of the computational domain to simulate
a semi-infinite cladding above and below the slab. We consider an incident Gaussian
5.5. EFFICIENT COUPLING FOR AUTO-COLLIMATION 141
beam with an elliptical cross-section with full widths at half maximum (FWHM) of
wxy = 5d and wz = 2.5d, launched with TM polarization. While there is some loss at
the interface of the PhC due to a mismatch in the beam profile, the reflected power
remains very low, which is essential to avoid interference effects in compact devices.
Figure 5.9 shows the 3D reflectance spectra for Gaussian beams with θi = 0◦ (solid
curve) and θi = 22.5◦ (long dashed curve) incident on 16 layers of the PhC in Fig. 5.7.
Also shown are the equivalent reflectance spectra for plane waves incident on a 2D
semi-infinite PhC, calculated using the Bloch mode matrix method. On all curves the
minimum reflected power occurs for scaled frequencies between 0.29 < d/λ < 0.295,
with bandwidths of more than 15% of the central frequency in which the reflectance is
below 10%. The 3D results show a minimum reflectance of 0.2% and 1.4% for θi = 0◦
and θi = 22.5◦ respectively, compared to < 0.1% in the 2D plane-wave results, a
variation attributed to the angular spread of the Gaussian beam in the 3D calculations.
Note also that the Fabry-Perot fringes in the FDTD results are much stronger for
the normally incident beam since the finite beam width weakens multiple reflection
interference for oblique incidence.
0.24 0.26 0.28 0.3 0.32 0.34 0.36 0.38 0.40
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
d/l
Re
fle
cte
d p
ow
er
2D R
2D R∞
∞
i
i
, = 0
, = 22.5
3D FDTD, 16 layers, = 0
3D FDTD, 16 layers, = 22.5
q
q
q
q
i
i
o
o
o
o
Figure 5.9: 2D plane wave reflectance spectra (R∞: Bloch mode method) for a semi-
infinite PC and 3D FDTD reflectance spectra for a Gaussian beam of width wxy = 5d
and height wz = 2.5d incident on 16 layers of cylinders. Results are shown for θi = 0◦
and θi = 22.5◦.
The agreement between the 2D and 3D calculations in Fig. 5.9 is surprisingly good
given that the 2D calculation is for plane waves incident on a semi-infinite 2D PhC,
142 CHAPTER 5. EFFICIENT WIDE-ANGLE TRANSMISSION
while the 3D results are calculated for a relatively narrow Gaussian beam incident on a
PhC slab consisting of only 16 grating layers. This means that the 2D calculations can
provide an efficient tool for initial optimization calculations over a broad parameter
space as demonstrated in Sec. 5.5.1. The results in Section 5.5 show that the 2D
optimisation of the collimation parameter p is also a good approximation to the 3D
results.
5.5.3 Results: collimation
We consider now the second optimised feature of the PhC – auto-collimation. Ob-
serve that the approximately hexagonal equifrequency curve in Fig. 5.8(b) has slightly
concave, rather than flat sides. This is a consequence of optimising the structure for col-
limation at non-normal incidence. We have found however, that reasonable collimation
is achieved for all incident angles θi ≤ 25◦.
(a) (b)
PC
Figure 5.10: (a) Electric field from a 2D FDTD simulation for an incident Gaussian
beam launched from below at a frequency of d/λ at θi = 22.5◦ incident on 30 layers of
PhC. The dashed lines indicate the upper and lower interfaces of the lattice. The cylin-
ders are omitted so that the field inside the PhC can be seen more clearly. (b) Electric
field for the same Gaussian beam propagating in the uniform background material with
no PhC present.
Figure 5.10(a) shows the electric field for an incident Gaussian beam at θi = 22.5◦
launched from below with a FWHM of 5.7d onto a PhC consisting of 30 grating layers.
The cylinders have not been drawn in the figure so that the fields inside the PhC can
be more easily seen, but the dashed lines indicate the upper and lower boundaries of
the structure. The results were calculated using the commercial 2D FDTD software
FullWAVE for a frequency of d/λ = 0.29. For comparison, Fig. 5.10(b) shows the fields
in the background medium when the PhC is removed. The collimation effects of the
PhC can be clearly seen; the beam is deflected when it enters the PhC, and propagates
through the structure at an angle θc ≈ 0. At the rear interface the beam leaves the
5.5. EFFICIENT COUPLING FOR AUTO-COLLIMATION 143
PhC and continues to propagate in the direction of the original beam. The beam on
either side of the PhC does not broaden significantly, and remains narrower than the
beam propagating in the dielectric background. As expected from the 2D calculations
in Fig. 5.9, the reflectance is less than 1%, and no reflected fields can be observed
Fig. 5.10(a).
45o
45o
(a) (b)
Figure 5.11: (a) Electric field calculated using the 2D FDTD method for a PhC beam
combiner. Two Gaussian beams of width 5.7d are incident on the front face of a 40-layer
PhC with θi = ±22.5◦. (b) Electric field for the same parameters as (a) but without
the PhC.
Since the beam direction inside the crystal is approximately normal to the interface
of the PhC, two beams incident at θi = ±22.5◦ will propagate in approximately the
same direction through the crystal. We use this property to demonstrate a simple PhC
beam combiner exhibiting low reflection, wide angle coupling and collimation that could
be used to interact multiple beams for studying nonlinear effects in PCs. In Fig. 5.11(a)
we present 2D results for the case where two Gaussian beams with the same parameters
as in Fig. 5.10 are focused onto the front face of a 40-layer PC. Interference of the beams
results in the three-lobed pattern in front of the interface in Fig. 5.11(a). Inside the PC,
the beams are collimated and propagate in approximately the same direction, emerging
144 CHAPTER 5. EFFICIENT WIDE-ANGLE TRANSMISSION
through the rear interface where they continue to propagate in their original directions.
For comparison, Fig. 5.11(b) shows the propagation of the beams in the background
material without the PhC. As we demonstrate in the 3D examples in Fig. 5.12, the field
profile at the input interface is effectively transferred through the crystal to the rear
interface with only minor distortion. This behaviour is maintained regardless of the
relative phase difference between the beams. PhC designs of this type could potentially
be used for transferring more complex field modulations such as images from one side of
the crystal to the other. In contrast, if the two beams were coupled into a single finite
waveguide, the conversion between waveguide and free space modes would destroy the
original field profile without careful waveguide design.
In this final example we demonstrate auto-collimation behaviour using full 3D nu-
merical simulations for the PhC slab illustrated in Fig. 5.7 with the parameters opti-
mised for the 2D PhC in Section 5.5.1. Figure 5.9 shows that the minimum reflectance
for the 3D structure occurs for a frequency of approximately d/λ = 0.295, slightly
higher than for the purely 2D lattice. Hence the results that follow are calculated at
this frequency, for which the reflectance is approximately 1.4%. Figure 5.12(a) shows
the fields inside a PhC of 16 layers calculated using 3D FDTD software for two incident
Gaussian beams as in Fig. 5.11. The beams have a width of wxy = 5d and height of
wz = 2.5d and are launched in phase with other.
These 3D simulations are computationally intensive and are limited by the memory
requirements for large structures. We are thus limited to relatively small computational
domains that do not allow us to consider long propagation regions before and after the
PhC. Instead, the overlapping Gaussian beams are launched just before the front face
of the PhC to reproduce the input field of the two converging beams in Fig. 5.11. With
this modification, we have been able to simulate propagation through 16 layers of PhC,
but this is close to the limit of current computational resources.
As in the 2D results in Fig. 5.11(a), the fields at the front and rear interfaces of
the PhC in Fig. 5.12(a) have a very similar profile. This can be seen more clearly
in Fig. 5.12(b) and (c) where the field intensities at the front and rear interfaces are
plotted for different phase differences between the two input beams. The dashed curve
in Fig. 5.12(b) is a Gaussian profile with a width of 5d cos(22.5◦), corresponding to the
envelope of the two input beams, while the dashed curve in Fig. 5.12(c) is a Gaussian
of width 5.8d cos(22.5◦), indicating that the beams are not perfectly collimated. Apart
from this slight broadening, the field profiles at the back interface closely resemble those
at the front interface for each of the relative phase differences. We have confirmed that
this occurs regardless of the lateral alignment of the beams with the cylinders of the
interface.
While we did not explicitly consider the out-of-plane losses in the optimisation, the
operating frequency of d/λ = 0.29 lies close to the light line, and so the light is relatively
well confined once it enters the PhC region where the high index rods provide a large
index contrast in the vertical direction. The beam profiles in Figs. 5.12 however indicate
a decrease in the peak intensity. This is due in part to the spreading of the beam as it
5.5. EFFICIENT COUPLING FOR AUTO-COLLIMATION 145
propagates, but also to a mismatch between the incident Gaussian beam profile and the
vertical width of the guided mode, which results in additional scattering losses. The
good agreement between the 2D and 3D reflectance spectra in Fig. 5.9 suggests that
improved mode matching in the vertical plane would increase the transmission into the
PhC without a significant increase in the reflected power.
-6 -4 -2 0 2 4 6x (d)
1
0
0.2
0.4
0.6
0.8
Inte
nsity (
a.u
.)
j
j p
j p
j p
= 0
= /2
=
= 3 /2
(b)
-6 -4 -2 0 2 4 6
1
0
0.2
0.4
0.6
0.8
Inte
nsity (
a.u
.)
x (d)
(c)
(a)
Figure 5.12: (a) 3D FDTD simulation showing the |Ey| field component in the PC
for two in-phase Gaussian beams launched with θi = ±22.5◦. (b) Field intensity (nor-
malised) at the input interface for beams with phase differences ϕ = 0, π/2, π, and 3π/2.
The dashed curve is a Gaussian profile of width wxy = 5d cos(22.5◦). (c) Intensity at
the output interface for the phase differences in (b). The dashed curve is a Gaussian
profile of width 5.8d cos(22.5◦).
146 CHAPTER 5. EFFICIENT WIDE-ANGLE TRANSMISSION
5.6 Discussion
In this chapter we have studied the coupling of light from a homogeneous dielectric into
the Bloch modes of uniform rod-type PhCs and have shown that very efficient coupling
is possible over a wide range of angles without modifying the interface of the crystal.
Numerical studies reveal that this behaviour is a generic property of many rod-type
PhCs, and thus they present an attractive class of structure for studying in-band PhC
applications such as auto-collimation and superprism effects. In contrast, coupling to
air-hole type PhCs is typically much less efficient and requires interface modifications or
novel coupling techniques such as those proposed in Refs. [231–233]. While a rigorous
proof of the single cylinder scattering interpretation is still being developed, there is
sufficient evidence to connect the efficient coupling properties of the PhCs with forward
scattering resonances of the cylinders. We have shown that this interpretation provides
a simple and intuitive physical explanation for the coupling properties of rod-type PhCs
and the difference between these and hole-type PhCs.
The recognition that forward scattering resonances play a large rope in determining
the transmission properties of a PhC could also lead to improved designs for hole-type
PhCs in which the hole shape is optimised to produce strong forward scattering. If
this was to work in practice, the modifications would require careful design so as not
to interfere with other desirable properties. While the additional fabrication challenges
may be comparable to the interface modifications that have been proposed, a theoretical
study of this approach may be worthwhile.
Analogies can be made between the transmission properties observed here and loss
mechanism in a new class of microstructured optical fibre consisting of high refrac-
tive index cylindrical inclusions in a low index background [239]. These fibres rely on
“anti-resonant reflection” and exhibit high loss features in the transmission spectrum
that correspond to forward scattering resonances of the individual cylinders [240]. We
note also that there are similarities between our results and recent studies of perfect
transmission through periodically structured metallic films [241–243]. In particular,
Garcıa et al. identified Mie resonances as the cause of enhanced transmission through
periodic arrays of silica spheres embedded in a metallic film [242]. In another recent
paper, Mandatori et al. demonstrated theoretically, wide-angle transmission into 1D
multilayer structures consisting of birefringent materials [244]. High transmissions were
obtained for incident angles up to 80◦ by choosing the birefringent properties of each
layer to balance the angular transmission dependence due to increased path length and
variation of Fresnel coefficients. While it is not immediately clear whether the wide-
angle PhC transmission investigated in this chapter can be interpreted in a similar way,
a future comparison of the two structures is warranted.
Since rod-type PhCs exhibit efficient coupling over a wide parameter space, this
provides significant flexibility to optimise a structure for additional characteristics as
demonstrated in Section 5.5 for auto-collimation. These results have been used to
design an efficient PC self-collimating beam combiner that could be used for studying
5.6. DISCUSSION 147
nonlinear interactions in PhCs. Many other in-band applications could benefit from
these design techniques and results. For example PhC superprisms require light to be
coupled efficiently through both interfaces while maintaining a good beam shape [126]
and there are also proposals for optical chips based on self collimated beams [146,245],
for which this work would clearly be of great benefit.
The ability to combine auto-collimation and efficient coupling without complex in-
terface modification is surprising if considered from the viewpoint of conventional optics.
At a dielectric interface, strong beam deflections require a large index contrast but this
is accompanied by high reflectance. In contrast, here we have demonstrated deflections
of more than 20◦ with a reflectance of less than 1%. This is only possible as a result of
the unique dispersion properties of photonic crystals.
In conclusion, we have found that coupling into regular rod-type PhCs can be signif-
icantly more efficient over a wide range of incident angles than for PCs with air holes,
and this can be achieved without modifications to the interface. We have demonstrated
that 2D simulations of this behaviour can be used to optimise the parameters for real-
istic PC slab geometries, and that these properties are retained by the 3D structures.
Chapter 6
Discussion and conclusions
In this thesis we have considered a number of photonic crystal structures and devices
and identified the basic physical effects that determine their behaviour. While we have
focussed on two aspects of these unique structure – resonances and mode coupling
properties – the theoretical approach taken here is much more widely applicable. In
particular, the combined use of simple physical models and accurate numerical methods
provides a powerful and efficient array of tools for studying many structures. As we
demonstrated, this is aided greatly by the Bloch mode scattering matrix method, which
is not only an efficient numerical method but also provides framework for deriving
semi-analytic models. Perhaps one of the greatest advantages to be gained from this
approach is the physical insight, and as a consequence, intuition which can lead to
simple design rules. These will become ever more important as photonic crystal devices
enter commercial applications.
In Chapters 3 and 4 we presented several PhC waveguide based devices designed for
operation in a photonic bandgap. The three main structures we considered are the FDC,
the coupled Y-junction and the recirculating Mach-Zehnder interferometer. While these
devices have been demonstrated only in a purely 2D geometry, the underlying concepts
are general and rely predominantly on modal dispersion properties and characteristics
common to most PhCs rather than specific details of the waveguides. The geometries
that we used for the numerical demonstration of these concepts were designed without
considering the out-of-plane losses that are present in realistic PhC slab structures.
As such, the specific design parameters are intended as examples only, and do not
necessarily correspond to practical devices. The logical next step, therefore, is to use
the insights gained from this research to develop complete designs for experimental
demonstration.
As we have mentioned several times throughout this thesis, one of the important
challenges in designing efficient resonant devices is controlling out-of-plane losses. Al-
though the Y-junction and the Mach-Zehnder interferometer presented here are not
high-Q resonant structures like the FDC, the unique properties of all three structures
are a result of multiple reflections between the ends of blocked waveguides, and so the
149
150 CHAPTER 6. DISCUSSION AND CONCLUSIONS
design considerations are common. We discuss here three distinct issues that must be
addressed in order to achieve realistic device designs. They are:
1. Control of waveguide dispersion properties for modes guided below the lightline.
2. Out-of-plane scattering losses at the reflective interfaces.
3. Tolerance to symmetry breaking and other fabrication variations.
These issues are common to many PhC slab-based devices, and thus there are many
examples in the literature where they have been successfully addressed.
Let us consider first the waveguide dispersion properties. As the semi-analytic mod-
els derived in Sections 3.4.1 and 4.2 show, the performance of these devices depends
strongly on the dispersion. While dispersion tuning is relatively straightforward in the
2D analysis where radiation losses are ignored, the available parameter space is reduced
considerably in slab structures by the requirement to operate below the lightline. How-
ever, techniques such as changing the waveguide width [97], or tuning the radius of the
cylinders close to waveguide [98] provide significant control over the dispersion. We
expect that tuning mechanisms such as these can be used do optimise the waveguide
design to permit our devices to operate below the lightline.
The second issue – reducing scattering losses at the reflective interfaces – is some-
what more of a challenge. While waveguide modes that lie below the cladding lightline
are theoretically lossless in a straight waveguide, considerable losses may still occur
when additional features such as junctions, bends or mirrors are introduced. Since the
devices we have proposed all require light to be reflected by the PhC at the end of a
blocked waveguide, the control of losses at these mirrors is essential. While there has
been considerable effort to reduce the radiation losses of high-Q PhC cavities via vari-
ous structural tuning methods as discussed in Section 1.7.1, [54,61,68,70], the radiation
losses that occur at a single reflective interface have received less attention. The physi-
cal mechanisms of this loss have been studied in some detail in Refs. [61,246,247], and
a number of approaches have been proposed to minimise the effects. Although the re-
sults were reported for waveguides with a 1D Bragg mirror, the physical interpretation
applies equally to 2D PhC mirrors in a slab geometry.
In Ref. [247], Bragg reflection from a PhC mirror is described as a three-step process;
first the forward propagating mode in the waveguide must couple to the fundamental
forward (non-propagating) Bloch mode of the PhC; second, light is coupled from the
forward travelling Bloch mode into its counterpropagating pair; and finally the back-
ward travelling Bloch mode is coupled into the backward propagating waveguide mode.
While the second process is lossless provided the Bloch modes are guided by the slab,
radiation losses may occur during the first and last stages of the process due to possible
mode mismatches between the modes of the PhC and the modes of the waveguide. Sev-
eral adiabatic tapering techniques have been proposed to optimise the mode-matching
in 1D periodic structures [246, 247] for high reflectance and low radiation losses. A
151
similar approach could be applied to the blocked waveguides in the FDC and the other
devices to minimise the radiation loss at these interfaces.
An alternative highly efficient waveguide mirror was demonstrated recently by Song
et al. [248], formed by coupling two slightly different waveguides. Almost perfect re-
flection occurs at frequencies where the input waveguide supports a propagating mode
but the output waveguide supports no propagating modes (ie: in the frequency interval
between the waveguide mode cutoff and the bottom of the bandgap). These results
are consistent with a mode-matching interpretation since the very similar geometry of
the two waveguides is likely to allow efficient coupling between the propagating and
nonpropagating modes. While this design is suitable for a single waveguide mirror,
some modification would be required to apply to the FDC or the Y-junction, where
waveguides on both sides of the interface are required to support propagating modes.
Reducing the scattering losses upon reflection is likely to result in efficient Y-junction
and Mach-Zehnder designs, but in order to achieve the predicted high-Q resonances in
the FDC, it may be necessary to consider the complete cavity design. While reducing
the out-of-plane scattering losses at each reflective interface is critical to achieve a high-
Q response, a second mechanism due to “radiation-loss recycling” has also been shown
to play a significant role in these structures [61]. Since this effect depends on the details
of the cavity as well as the interfaces, more complicated optimisation approaches may
be required. In Section 3.4.5 we show that the FDC is closely related to other common
PhC cavity designs, so the same optimisation techniques developed for those structures
could be applied [54,67,68,70,74].
The third and final consideration for developing practical PhC device designs is
their tolerance to fabrication errors, although recent demonstrations of ultrahigh-Qcavities [74] and low-loss waveguides [83–85] indicate that these errors can be tightly
controlled. While we cannot predict the sensitivity of the devices proposed here without
a numerical study, we can make some general observations. One aspect common to
all of our structures, is their dependence on modal symmetry, a property we used to
develop the semi-analytic models in Sections 3.4.1 and 4.2. Although it is unlikely
that small structural variations which break the symmetry will destroy the properties
of these devices, they may be more sensitive to this type of fabrication error. The
Y-junction in particular requires further study in this regard, since an asymmetric
structure would allow light to couple into the odd supermode of the triple waveguide
section – a possibility forbidden in the symmetric geometry considered here.
To summarise the results of this discussion, the waveguide-based devices that we pro-
pose in this thesis are expected to present the same challenges as many other PhC based
structures that have been successfully demonstrated in experiment. By addressing the
three issues listed above, we hope to develop realistic designs for future fabrication.
The PhC structures that we study in Chapter 5 face similar loss-related issues
which we have already discussed briefly in Section 5.6. Provided the Bloch modes
being coupled into lie below the cladding lightline of the slab, the dominant cause of
loss is the vertical mode-mismatch at the interface. The width of the incident Gaussian
152 CHAPTER 6. DISCUSSION AND CONCLUSIONS
beam in the 3D simulations was chosen to match approximately the observed width
of the guided mode in the slab, but no further attempt was made to optimise the
coupling. To minimise the radiation loss at the interface one could either tune the
incident beam parameters, or alternatively couple light into the crystal via a dielectric
waveguide [143,249], which may provide better mode matching conditions.
In conclusion, we have studied in this thesis a number of photonic crystal structures
and devices that derive their characteristic properties from resonance and mode coupling
effects that are common to most PhC geometries. The combination of these two features
in a single device results in a range of different behaviours that we have exploited to
design several novel and promising devices. In Chapter 3 we presented three such
structures: a high-Q resonant filter, a wide-bandwidth symmetric Y-junction and a
periodic coupled waveguide that can be tuned to exhibit very low group velocities.
These same design concepts and analysis methods were extended in Chapter 4 where we
showed that mode recirculation in PhC Mach-Zehnder interferometers can be harnessed
to enhance the response of these devices.
In Chapter 5 we considered an entirely different aspect of resonance and mode
coupling: the transmission of light from free-space into the propagating Bloch modes of
a uniform PhC. The demonstration of highly efficient wide-angle coupling in rod-type
geometries, and the combination of this property with auto-collimation effects could
motivate further research efforts to fabricate rod-type PhCs for in-band applications.
Finally, it should be noted that the structures we have studied are intended to
exploit the unique properties of photonic crystals, rather than to replicate conventional
devices on a smaller scale. As we have demonstrated, this design philosophy, combined
with physical insight can lead to simple but efficient devices with highly attractive
properties. The combination of powerful general methods and semi-analytic solutions
for particular structures is one which will help the rapid progress in photonic crystal
optics to continue.
Bibliography
[1] A. Parker, In the blink of an eye: the most dramatic event in the history of life
(Free press, London, 2003).
[2] H. A. Macleod, Thin film optical filters, 3rd ed. (Institute of Physics, London,
2001).
[3] E. Yablonovitch, “Inhibited spontaneous emission in solid-state physics and elec-
tronics,” Phys. Rev. Lett. 58, 2059–2062 (1987).
[4] S. John, “Strong localization of photons in certain disordered dielectric superlat-
tices,” Phys. Rev. Lett. 58, 2486–2489 (1987).
[5] H. A. Hauss and C. V. Shank, “Antisymmetric taper of distributed feedback
lasers,” IEEE J. Quantum Electron. QE-12, 532–539 (1976).
[6] R. Kashyap, Fiber Bragg Gratings (Academic Press, San Diego, 1999).
[7] J. Joannopoulos, R. Meade, and J. Winn, Photonic Crystals (Princeton Univer-
sity, Princeton, N.J., 1995).
[8] S. J. McNab, N. Moll, and Y. A. Vlasov, “Ultra-low loss photonic integrated
circuit with membrane-type photonic crystal waveguides,” Opt. Express 11, 2927–
2939 (2003).
[9] P. Bienstman, S. Assefa, S. G. Johnson, J. D. Joannopoulos, G. Petrich, and
L. A. Kolodziejski, “Taper structures for coupling into photonic crystal slab wave-
guides,” J. Opt. Soc. Am. B 20, 1817–1821 (2003).
[10] K. M. Ho, C. T. Chan, and C. M. Soukoulis, “Existence of a photonic gap in
periodic dielectric structures,” Phys. Rev. Lett. 65, 3152–3155 (1990).
[11] E. Yablonovitch and T. J. Gmitter, “Photonic band structure: the face-centered-
cubic case,” Phys. Rev. Lett. 63, 1950–1953 (1989).
[12] E. Yablonovitch, T. J. Gmitter, and K. M. Leung, “Photonic band structure: the
face-centered-cubic case employing non-spherical atoms,” Phys. Rev. Lett. 67,
2295–2298 (1991).
153
154 BIBLIOGRAPHY
[13] A. Chelnokov, S. Rowson, J. M. Lourtioz, L. Duvillaret, and J. L. Coutaz, “Tera-
hertz characterisation of mechanically machined 3D photonic crystal,” Electron.
Lett. 33, 1981–1983 (1997).
[14] S. Y. Linn, J. G. Fleming, D. L. Hetherington, B. K. Smith, R. Biswas, K. M.
Ho, M. M. Sigalas, W. Zubrzycki, S. R. Kurtz, and J. Bur, “A three-dimensional
photonic crystal operating at infrared wavelengths,” Nature 394, 251–253 (1998).
[15] J. G. Fleming and S. Y. Lin, “Three-dimensional photonic crystal with a stop
band from 1.35 to 1.95µm,” Opt. Lett. 24, 49–51 (1999).
[16] S. Noda, K. Tomoda, N. Yamamoto, and A. Chutinan, “Full three-dimensional
photonic bandgap crystals at near-infrared wavelengths,” Science 289, 604–606
(2000).
[17] A. Blanco, E. Chomski, S. Grabtchak, M. Ibisate, S. John, S. W. Leonard,
C. Lopez, F. Meseguer, H. Miguez, J. P. Mondia, G. A. Ozin, O. Toader, and
H. M. van Driel, “Large-scale synthesis of a silicon photonic crystal with a com-
plete three-dimensional bandgap near 1.5 micrometres,” Nature 405, 437–440
(2000).
[18] M. O. Jensen and M. J. Brett, “Square spiral 3D photonic bandgap crystals at
telecommunications frequencies,” Opt. Express 13, 3348–3354 (2005).
[19] E. Lidorikis, M. L. Povinelli, S. G. Johnson, and J. D. Joannopoulos,
“Polarization-independent linear waveguides in 3D photonic crystals,” Phys. Rev.
Lett. 91, 023902 (2003).
[20] R. Wang and S. John, “Engineering the electromagnetic vacuum for controlling
light with light in a photonic-band-gap microchip,” Phys. Rev. A. 70, 043805
(2004).
[21] S. Ogawa, M. Imada, Y. Susumu, M. Okana, and S. Noda, “Control of light
emission by 3D photonic crystals,” Science 305, 227–229 (2004).
[22] C. Sell, C. Christensen, G. Tuttle, Z. Y. Li, and K. M. Ho, “Propagation and loss
in three-dimensional photonic crystal waveguides with imperfect confinement,”
Phys. Rev. B. 68, 113106 (2003).
[23] M. Fujita, S. Takahashi, Y. Tanaka, T. Asano, and S. Noda, “Simultaneous in-
hibition and redistribution of spontaneuos light emission in photonic crystals,”
Science 308, 1296–1298 (2005).
[24] P. Lodahl, A. F. van Driel, I. S. Nikolaev, A. Irman, K. Overgaag, D. Vanmaekel-
bergh, and W. L. Vos, “Controlling the dynamics of spontaneous emission from
quantum dots by photonic crystals,” Nature 430, 654–657 (2004).
BIBLIOGRAPHY 155
[25] S. Y. Lin, J. Moreno, and J. G. Fleming, “Three-dimensional photonic-crystal
emitter for thermal photovoltaic power generation,” Appl. Phys. Lett. 83, 380–
382 (2003).
[26] S. Y. Lin, J. G. Fleming, and I. El-Kady, “Three-dimensional photonic-crystal
emission through thermal emission,” Opt. Lett. 28, 1909–1911 (2003).
[27] T. Prasad, V. Colvin, and D. Mittleman, “Superprism phenomenon in three-
dimensional macroporous polymer photonic crystals,” Phys. Rev. B 67, 165103
(2003).
[28] X. Hu, Q. Zhiang, Y. Liu, B. Cheng, and D. Zhang, “Ultrafast three-dimensional
tunable photonic crystal,” Appl. Phys. Lett. 83, 2518–2520 (2003).
[29] S. Kubo, Z. Z. Gu, K. Takahashi, A. Fujishima, H. Segawa, and O. Sato, “Tun-
able photonic band gap crystals based on a liquid crystal-infiltrated inverse opal
structure,” J. Am. Chem. Soc. 126, 8314–8319 (2004).
[30] D. McPhail, M. Straub, and M. Gu, “Electrical tuning of three-dimensional pho-
tonic crystals using polymer dispersed liquid crystals,” Appl. Phys. Lett. 86,
051103 (2005).
[31] M. Plihal, A. Shambrook, and A. A. Maradudin, “Two-dimensional photonic
bandgap structures,” Opt. Comm. 80, 199–204 (1991).
[32] M. Plihal and A. A. Maradudin, “Photonic band structure of two-dimensional
systems: the triangular lattice,” Phys. Rev. B. 44, 8565–8571 (1991).
[33] P. R. Villeneuve and M. Piche, “Photonic band gaps in two-dimensional square
and hexagonal lattices,” Phys. Rev. B. 46, 4969–4972 (1992).
[34] R. Meade, K. D. Brommer, A. M. Rappe, and J. D. Joannopoulos, “Existence of
a photonic band gap in two dimensions,” Appl. Phys. Lett. 61, 495–497 (1992).
[35] J. C. Knight, J. Broeng, T. A. Birks, and P. St. J. Russel, “Photonic bandgap
guidance in optical fibres,” Science 282, 1476–1478 (1998).
[36] C. C. Chen, C. Y. Chen, W. K. Wang, F. H. Huang, C. K. Lin, W. Y. Chiu, and
Y. J. Chan, “Photonic crystal directional couplers formed by InAlGaAs nano-
rods,” Opt. Express 13, 38–43 (2005).
[37] A. Chutinan and S. John, “Diffractionless flow of light in two- and three-
dimensional photonic band gap heterostructures: Theory, design rules, and sim-
ulations,” Phys. Rev. E. 71, 026605 (2005).
[38] S. G. Johnson, P. R. Villeneuve, S. Fan, and J. D. Joannopoulos, “Linear wave-
guides in photonic-crystal slabs,” Phys. Rev. B 62, 8212–8222 (2000).
156 BIBLIOGRAPHY
[39] A. Chutinan, S. John, and O. Toader, “Diffractionless flow of light in all-optical
microchips,” Phys. Rev. Lett. 90, 123901 (2003).
[40] R. D. Meade, A. Devenyi, J. D. Joannopoulos, O. L. Alerhand, D. A. Smith, and
K. Kash, “Novel applications of photonic bandgap materials: Low-loss bands and
high Q cavities,” J. Appl. Phys. 75, 4753–4755 (1994).
[41] T. F. Krauss, R. M. De La Rue, and S. Brand, “Two-dimensional photonic-
bandgap structures operating at near infrared wavelengths,” Nature 383, 699–702
(1996).
[42] S. G. Johnson, S. Fan, P. R. Villeneuve, J. D. Joannopoulos, and L. A.
Kolodziejski, “Guided modes in photonic crystal slabs,” Phys. Rev. B. 60, 5751–
5758 (1999).
[43] S. Kuchinsky, D. C. Allan, N. F. Borrelli, and J. C. Cotteverte, “3D localization
in a channel waveguide in a photonic crystal with 2D periodicity,” Opt. Commun.
175, 147–152 (2000).
[44] M. Loncar, T. Doll, J. Vuckovic, and A. Scherer, “Design and fabrication of silicon
photonic crystal optical waveguides,” J. Lightwave Technol. 18, 1402–1411 (2000).
[45] E. Chow, S. Y. Lin, S. G. Johnson, P. R. Villeneuve, J. D. Joannopoulos,
J. R. Wendt, G. A. Vawter, W. Zubrzycki, H. Hou, and A. Alleman, “Three-
dimensional control of light in a two-dimensional photonic crystal slab,” Nature
407, 983–986 (2000).
[46] Y. Sugimoto, Y. Tanaka, N. Ikeda, Y. Nakamura, and K. Asakawa, “Low propa-
gation loss of 0.76 dB/mm in GaAs-based single-line-defect two-dimensional pho-
tonic crystal slab waveguides up to 1 cm in length,” Opt. Express 12, 1090–1096
(2004).
[47] A. Assefa, P. T. Rakich, P. Bienstman, S. G. Johnson, G. S. Petrich, J. D.
Joannopoulos, L. A. Kolodziejski, E. P. Ippen, and H. I. Smith, “Guiding 1.5µm
light in photonic crystals based on dielectric rods,” Appl. Phys. Lett. 85, 6110–
6112 (2004).
[48] M. Tokushima, H. Yamada, and Y. Arakawa, “1.5µm-wavelength light guiding in
waveguides in square-lattice-of-rod-photonic crystal slab,” Appl. Phys. Lett. 84,
4298–4300 (2004).
[49] M. Tokushima, “Experimental demonstration of waveguides in arrayed-rod pho-
tonic crystals for integrated optical buffers,” in Optical Fiber Communication
Conference and Cxposition and the National Fiber Optic Engineers Conference
on CD-ROM (Optical Socity of America, Washington, DC, USA, 2005). Paper
OWD2.
BIBLIOGRAPHY 157
[50] A. Martinez, J. Garcia, G. Sanchez, and J. Marti, “Planar photonic crystal struc-
ture with inherently single-mode waveguides,” J. Opt. Soc. Am. A 20, 2131–2136
(2003).
[51] M. Qiu, “Effective index method for heterostructure-slab-waveguide-based two-
dimensional photonic crystals,” Appl. Phys. Lett. 81, 1163–1165 (2002).
[52] H. Benisty, C. Weisbuch, D. Labilloy, M. Rattier, C. J. M. Smith, T. F. Krauss,
R. M. De La Rue, R. Houdre, U. Oesterle, C. Jouanin, and D. Cassagne, “Optical
and confinement properties of two-dimensional photonic crystals,” J. Lightwave
Technol. 17, 2063–2077 (1999).
[53] S. Olivier, H. Benisty, M. Rattier, C. Weisbuch, M. Qiu, A. Karlsson, C. J. M.
Smith, R. Houdre, and U. Oesterle, “Resonant and nonresonant transmission
through waveguide bends in a planar photonic crystal,” Appl. Phys. Lett. 79,
2514–2516 (2001).
[54] Y. Akahane, T. Asano, B. S. Song, and S. Noda, “High-Q photonic nanocavity
in a two-dimensional photonic crystal,” Nature 425, 944–947 (2003).
[55] R. Zoli, “2D Photonic crystals: theory and applications,” Ph.D. thesis, Univerita
degli Studi di Bologna, Bologna, Italy (2004).
[56] G. W. Milton, The theory of composites, 1st ed. (Cambridge University Press,
Cambridge, 2002).
[57] K. J. Vahala, “Optical microcavities,” Nature 424, 839–846 (2003).
[58] J. D. Jackson, Classical electrodynamics, 2nd ed. (John Wiley and Sons, New
York, 1975).
[59] D. W. Vernooy, V. S. Ilchenko, H. Mabuchi, E. W. Streed, and H. J. Kimble,
“High-Q measurements of fused-silica microspheres in the near infrared,” Opt.
Lett. 23, 247–249 (1998).
[60] D. K. Armani, T. J. Kippenberg, S. M. Spillane, and K. J. Vahala, “Ultra-high-Q
toroid microcavity on a chip,” Nature 421, 925–928 (2003).
[61] P. Lalanne, S. Milas, and J. P. Hugonin, “Two physical mechanisms for boosting
the quality factor to cavity volume ratio of photonic crystal microcavities,” Opt.
Express 12, 458–467 (2004).
[62] O. Painter, R. K. Lee, A. Scherer, A. Yariv, J. D. O’Brien, P. D. Dapkus, and
I. Kim, “Two-dimensional photonic band-gap defect mode laser,” Science 284,
1819–1821 (1999).
158 BIBLIOGRAPHY
[63] O. J. Painter, A. Husain, A. Scherer, J. D. O’Brien, I. Kim, and P. D. Dapkus,
“Room temperature photonic crystal defect lasers at near-infrared wavelengths
in InGaAsP,” J. Lightwave Technol. 17, 2082–2088 (1999).
[64] J. Vuckovic, M. Loncar, H. Mabuchi, and A. Scherer, “Design of photonic crystal
microcavities for cavity QED,” Phys. Rev. E 65, 016608 (2001).
[65] T. Yoshie, J. Vuckovic, A. Scherer, H. Chen, and D. Deppe, “High quality two-
dimensional photonic crystal slab cavities,” Appl. Phys. Lett. 79, 4289–4291
(2001).
[66] S. Y. Lin, E. Chow, S. G. Johnson, and J. D. Joannopoulos, “Direct measurement
of the quality factor in a two-dimensional photonic-crystal microcavity,” Opt.
Lett. 26, 1903–1905 (2001).
[67] J. Vuckovic, M. Loncar, H. Mabuchi, and A. Scherer, “Optimization of the Q
factor in photonic crystal microcavities,” IEEE J. Quantum Electron. 38, 850–
856 (2002).
[68] K. Srinivasan and O. Painter, “Momentum space design of high-Q photonic crystal
optical cavities,” Opt. Express 10, 670–684 (2002).
[69] K. Srinivasan, P. E. Barklay, O. Painter, J. Chen, A. Y. Cho, and C. Gmachl, “Ex-
perimental demonstration of a high quality factor photonic crystal microcavity,”
Appl. Phys. Lett. 83, 1915–1917 (2003).
[70] S. G. Johnson, S. Fan, A. Mekis, and J. D. Joannopoulos, “Multipole-cancellation
mechanism for high-Q cavities in the absence of a complete photonic band gap,”
Appl. Phys. Lett. 78, 3388–3390 (2001).
[71] K. Srinivasan and O. Painter, “Fourier space design of high-Q cavities in standard
and compressed hexagonal lattice photonic crystals,” Opt. Express 11, 579–593
(2003).
[72] K. Srinivasan, P. E. Barclay, and O. Painter, “Fabrication-tolerant high quality
factor photonic crystal microcavities,” Opt. Express 12, 1458–1463 (2004).
[73] T. J. Kippenberg, S. M. Spillane, and K. J. Vahala, “Demonstration of ultra-
high-Q small mode volume toroid microcavities on a chip,” Appl. Phys. Lett. 85,
6113–6115 (2004).
[74] Y. Akahane, T. Asano, B. S. Song, and S. Noda, “Fine-tuned high-Q photonic-
crystal nanocavity,” Opt. Express 13, 1203–1214 (2005).
[75] B. S. Song, S. Noda, T. Asano, and Y. Akahane, “Ultra-high-Q photonic double-
heterostructure nanocavity,” Nature Materials 4, 207–210 (2005).
BIBLIOGRAPHY 159
[76] M. Notomi, A. Shinya, S. Mitsugi, G. Kira, E. Kuramochi, and T. Tanabe, “Op-
tical bistable switching action of Si high-Q photonic-crystal nanocavities,” Opt.
Express 13, 2678–2687 (2005).
[77] S. Boscolo, M. Midrio, and T. F. Krauss, “Y junctions in photonic crystal channel
waveguides: high transmission and impedance matching,” Opt. Lett. 27 (2002).
[78] R. Biswas, Z. Y. Li, and K. M. Ho, “Impedance of photonic crystals and photonic
crystal waveguides,” Appl. Phys. Lett. 84, 1254–1256 (2004).
[79] W. Bogaerts, R. Baets, P. Dumon, V. Wiaux, S. Beckx, D. Taillaert, B. Luys-
saert, J. van Campenhout, P. Bienstman, and D. Van Thourhout, “Nanophotonic
waveguides in silicon-on-insulator fabricated with CMOS technology,” J. Light-
wave Technol. 23, 401–412 (2005).
[80] S. Y. Lin, E. Chow, V. Hietala, P. R. Villeneuve, and J. D. Joannopoulos, “Ex-
perimental demonstration of guiding and bending of electromagnetic waves in a
photonic crystal,” Science 282, 274–276 (1998).
[81] S. Y. Lin, E. Chow, S. G. Johnson, and J. G. Joannopoulos, “Demonstration of
highly efficient waveguiding in photonic crystal slab at the 1.5µm wavelength,”
Opt. Lett. 25, 1297–1299 (2000).
[82] M. Loncar, D. Nedeljkovic, T. Doll, and J. Vuckovic, “Waveguiding in planar
photonic crystals,” Appl. Phys. Lett. 77, 1937–1939 (2000).
[83] M. Notomi, A. Shinya, S. Mitsugi, E. Kuramochi, and H. Y. Ryu, “Waveguides,
resonators and their coupled elements in photonic crystal slabs,” Opt. Express
12, 1551–1561 (2004).
[84] S. Hughes, L. Ramunno, J. F. Young, and J. E. Sipe, “Extrinsic optical scattering
loss in photonic crystal waveguides: role of fabrication disorder and photon group
velocity,” Phys. Rev. Lett. 94, 033903 (2005).
[85] E. Dulkeith, S. J. McNab, and Y. A. Vlasov, “Mapping the optical properties
of SOI-type photonic crystal waveguides,” in Conference on Lasers and Electro-
Optics/Quantum Electronics and Laser Science Conference (CLEO/QELS) 2005
(Optical Society of America, Wachington, DC, USA, 2005). Paper QWD3.
[86] K. K. Lee, D. R. Lim, and L. C. Kimerling, “Fabrication of ultralow-loss Si/SiO2
waveguides by roughness reduction,” Opt. Lett. 26, 1888–1890 (2001).
[87] M. M. Sigalas, C. M. Soukoulis, C. T. Chan, and D. Turner, “Localization of
electromagnetic waves in two-dimensional disordered systems,” Phys. Rev. B 53,
8340–8348 (1996).
160 BIBLIOGRAPHY
[88] M. Bayindir, E. Cubukcu, I. Bulu, T. Tut, and E. Ozbay, “Photonic band gaps,
defect characteristics, and waveguiding in two-dimensional disordered dielectrics
and metallic photonic crystals,” Phys. Rev. B 64, 195113 (2001).
[89] K. C. Kwan, X. Zhang, Z. Q. Zhang, and C. T. Chan, “Effects of disorder on
photonic crystal-based waveguides,” Appl. Phys. Lett. 82, 4414–4416 (2003).
[90] A. Mekis, J. C. Chen, I. Kurland, S. Fan, P. R. Villeneuve, and J. Joannopoulos,
“High Transmission through sharp bends in photonic crystal waveguides,” Phys.
Rev. Lett. 77, 3787–3790 (1996).
[91] E. Chow, S. Y. Lin, J. R. Wendt, S. G. Johnson, and J. D. Joannopoulos, “Quan-
titative analysis of bending efficiency in photonic-crystal waveguide bends at
λ = 1.55µm wavelengths,” Opt. Lett. 26, 286–288 (2001).
[92] E. A. J. Marcatilli, “Bends in optical dielectric waveguides,” Bell Syst. Tech. J.
48, 2103–2132 (1969).
[93] T. Tsuchizawa, K. Yamada, H. Fukuda, T. Watanabe, J. Takahashi, M. Taka-
hashi, T. Shoji, E. Tamechika, S. Itabashi, and H. Morita, “Microphotonics de-
vices based on silicon microfabrication technology,” IEEE Sel. Top. Quantum
Electron. 11, 232–240 (2005).
[94] R. U. Ahmad, F. Pizzuto, G. S. Camarda, R. L. Espinola, H. Rao, and R. M. Os-
good, Jr., “Ultracompact corner-mirrors and T-branches in silicon-on-insulator,”
IEEE Photon. Technol. Lett. 14, 65–67 (2002).
[95] M. Notomi, K. Yamada, A. Shinya, J. Takahashi, C. Takahashi, and I. Yokohama,
“Extremely large group-velocity dispersion of line-defect waveguides in photonic
crystal slabs,” Phys. Rev. Lett. 87, 253902 (2001).
[96] H. Gersen, T. J. Karke, R. J. P. Engelen, W. Bogaerts, J. P. Korterik, N. F. van
Hulst, T. F. Krauss, and L. Kuipers, “Real-space observation of ultraslow light
in photonic crystal waveguides,” Phys. Rev. Lett. 94, 073903 (2005).
[97] M. Notomi, A. Shinya, K. Yamada, J. Takahashi, C. Takahashi, and I. Yoko-
hama, “Structural tuning of guiding modes of line-defect waveguides of silicon-
on-insulator photonic crystal slabs,” IEEE J. Quantum Electron. 38, 736–742
(2002).
[98] A. Y. Petrov and M. Eich, “Zero dispersion at small group velocities in photonic
crystal waveguides,” Appl. Phys. Lett. 85, 4866–4868 (2004).
[99] T. Baba, “Fast spontaneous emission and slow light in photonic crystals,” in
Conference on Lasers and Electro-Optics/Quantum Electronics and Laser Sci-
ence Conference (CLEO/QELS) technical digest, Paper QWD1 (Baltimore, USA,
2005).
BIBLIOGRAPHY 161
[100] E. Ozbay, M. Bayindir, I. Bulu, and E. Cubukcu, “Investigation of localized
coupled-cavity modes in two-dimensional photonic band gap structures,” IEEE
J. Quantum Electron. 38, 837–843 (2002).
[101] T. J. Karle, Y. J. Chai, C. N. Morgan, I. H. White, and T. F. Krauss, “Observation
of pulse compression in photonic crystal coupled cavity waveguides,” J. Lightwave
Technol. 22, 514–519 (2004).
[102] M. F. Yanik and S. Fan, “Stopping light all optically,” Phys. Rev. Lett. 92, 083901
(2004).
[103] B. E. A. Saleh and M. C. Teich, Fundamentals of photonics (John Wiley and
Sons, New York, 1991).
[104] S. C. Cho, B. G. Kim, Y. Moon, and A. Shakouri, “Ultra short two-section vertical
directional coupler switches with high extinction ratios,” Jpn. J. Appl. Phys. 40,
4045–4050 (2001).
[105] S. Boscolo, M. Midrio, and C. G. Someda, “Coupling and decoupling of electro-
magnetic waves in parallel 2-D photonic crystal waveguides,” IEEE J. Quantum
Electron. 38, 47–53 (2002).
[106] A. Martinez, F. Cuesta, and J. Marti, “Ultrashort 2-D photonic crystal directional
couplers,” IEEE Photon. Technol. Lett. 15, 694–696 (2003).
[107] M. Tokushima and H. Yamada, “Photonic crystal line defect waveguide,” Elec-
tron. Lett. 37, 1454–1455 (2001).
[108] Y. Tanaka, H. Nakamura, Y. Sugimoto, N. Ikeda, K. Asakawa, and K. Inoue,
“Coupling properties in a 2-D photonic crystal slab directional coupler with a
triangular lattice of airholes,” IEEE J. Quantum Electron. 41, 76–84 (2005).
[109] Y. Sugimoto, Y. Tanaka, N. Ikeda, T. Yang, H. Nakamura, K. Asakawa, K. Inoue,
T. Maruyama, K. Miyashita, K. Ishida, and Y. Watanabe, “Design, fabrication,
and characterization of coupling-strength-controlled direcional coupler based on
two-dimensional photonic-crystal slab waveguides,” Appl. Phys. Lett. 83, 3236–
3238 (2003).
[110] A. Sharkawy, S. Shi, D. Prather, and R. Soref, “Electro-optical switching using
coupled photonic crystal waveguides,” Opt. Express 10, 1048–1059 (2002).
[111] F. Cuesta, A. Martinez, J. Garcıa, F. Ramos, P. Sanchis, J. Blasco, and J. Martı,
“All-optical switching structure based on a photonic crystal directional coupler,”
Opt. Express 12, 161–167 (2004).
[112] M. Koshiba, “Wavelength division multiplexing and demultiplexing with photonic
crystal waveguide couplers,” J. Lightwave Technol. 19, 1970–1975 (2001).
162 BIBLIOGRAPHY
[113] S. Fan, P. R. Villeneuve, and J. D. Joannopoulos, “Channel Drop Tunneling
through Localized States,” Phys. Rev. Lett. 80, 960–963 (1998).
[114] S. Fan, P. R. Villeneuve, J. D. Joannopoulos, and H. A. Haus, “Channel drop
filters in photonic crystals,” Opt. Express 3, 4–11 (1998).
[115] H. Takano, Y. Akahane, T. Asano, and S. Noda, “In-plane-type channel drop
filter in a two-dimensional photonic crystal slab,” Appl. Phys. Lett. 84, 2226–
2228 (2004).
[116] S. Fan, P. R. Villeneuve, J. D. Joannopoulos, M. J. Khan, C. Manolatou, and
H. A. Haus, “Theoretical analysis of channel drop tunneling process,” Phys. Rev.
B 59, 882–892 (1999).
[117] A. Sharkawy, S. Shi, and D. W. Prather, “Multichannel wavelength division mul-
tiplexing with photonic crystals,” Appl. Opt. 40, 2247–2252 (2001).
[118] S. Noda, A. Chutinan, and M. Imada, “Trapping and emission of photons by a
single defect in a photonic bandgap structure,” Nature 407, 608–610 (2000).
[119] A. Chutinan, M. Mochizuki, M. Imada, and S. Noda, “Surface-emitting channel
drop filters using single defects in two-dimensional photonic crystal slabs,” Appl.
Phys. Lett. 79, 2690–2692 (2001).
[120] B. S. Song, T. Asano, Y. Akahane, Y. Tanaka, and S. Noda, “Multichannel
add/drop filter based on in-plane hetero photonic crystals,” J. Lightwave Technol.
23, 1449–1455 (2005).
[121] T. Asano, M. Nakamura, B. S. Song, and S. Noda, “Dynamic wavelength tuning
of channel-drop device in two-dimensional photonic crystal slab,” Electron. Lett.
41, 37–38 (2005).
[122] M. Born and E. Wolf, Principles of Optics, 5th ed. (Pergamon, Oxford, 1975).
[123] M. Notomi, “Theory of light propagation in strongly modulated photonic crystals:
refractionlike behavior in the vicinity of the photonic band gap,” Phys. Rev. B
62, 10696–10705 (2000).
[124] S. Foteinopoulou and C. M. Soukoulis, “EM wave propagation in two-dimensional
photonic crystals: a study of anomalous refractive effects,” arXiv:cond-
mat/0403542 (2003).
[125] T. Baba and M. Nakamura, “Photonic crystal light deflect devices using the
superprism effect,” IEEE J. Quantum Electron. 38, 909–914 (2002).
[126] T. Baba and T. Matsumoto, “Resolution of photonic crystal superprism,” Appl.
Phys. Lett. 81, 2325–2327 (2002).
BIBLIOGRAPHY 163
[127] V. G. Veselago, “The electrodynamics of substances with simultaneously negative
values of ε and µ,” Sov. Phys. Usp. 10, 509–514 (1968).
[128] J. B. Pendry, “Negative refraction makes a perfect lens,” Phys. Rev. Lett. 85,
3966–3969 (2000).
[129] D. R. Smith, W. J. Padilla, D. C. Vier, S. C. Nemat-Nasser, and S. Schultz,
“Composite medium with simultaneously negative permeability and permittiv-
ity,” Phys. Rev. Lett. 84, 4184–4187 (2000).
[130] J. P. Dowling, “Anomalous index of refraction in photonic bandgap materials,”
J. Mod. Opt. 41, 345–351 (1994).
[131] S. Enoch, G. Tayeb, and D. Maystre, “Numerical evidence of ultrarefractive optics
in photonic crystals,” Opt. Comm. 161, 171–176 (1994).
[132] H. Kosaka, T. Kawashima, A. Tomita, M. Notomi, T. Tamamura, T. Sato, and
S. Kawakami, “Superprism phenomena in photonic crystals,” Phys. Rev. B 58,
R10096–R10099 (1998).
[133] B. Gralak, S. Enoch, and G. Tayeb, “Anomalous refractive properties of photonic
crystals,” J. Opt. Soc. Am. A 17, 1012–1020 (2000).
[134] S. Foteinopoulou, E. N. Economou, and C. M. Soukoulis, “Refraction in media
with a negative refractive index,” Phys. Rev. Lett. 90, 107402 (2003).
[135] E. Cubukcu, K. Aydin, E. Ozbay, S. Foteinopoulou, and C. M. Soukoulis, “Neg-
ative refraction by photonic crystals,” Nature 423, 604–605 (2003).
[136] A. Berrier, M. Mulot, M. Swillo, M. Qiu, L. Thylen, A. Talneau, and S. Anand,
“Negative refraction at infrared wavelengths in two-dimensional photonic crystal,”
Phys. Rev. Lett. 93, 073902 (2004).
[137] C. Luo, S. Johnson, J. D. Joannopoulos, and J. B. Pendry, “Subwavelength imag-
ing in photonic crystals,” Phys. Rev. B 68, 045115 (2003).
[138] Z. Y. Li and L. L. Lin, “Evaluation of lensing in photonic crystal slabs exhibiting
negative refraction,” Phys. Rev. B 68, 245110 (2003).
[139] E. Cubukcu, K. Aydin, E. Ozbay, S. Foteinopoulou, and S. M. Soukoulis, “Sub-
wavelength resolution in a two-dimensional photonic-crystal-based superlens,”
Phys. Rev. Lett. 91, 207401 (2003).
[140] P. V. Parimi, W. T. Lu, P. Vodo, and S. Sridar, “Imaging by a flat lens using
negative refraction,” Nature 426, 404 (2003).
164 BIBLIOGRAPHY
[141] H. Kosaka, T. Kawashima, A. Tomita, M. Notomi, T. Tamamura, T. Sato, and
S. Kawakami, “Self-collimating phenomena in photonic crystals,” Appl. Phys.
Lett. 74, 1212–1214 (1999).
[142] J. Witzens, M. Loncar, and A. Scherer, “Self-Collimation in planar photonic
crystals,” IEEE J. Sel. Top. Quantum Electron. 8, 1246–1257 (2002).
[143] L. Wu, M. Mazilu, and T. Krauss, “Beam steering in planar-photonic crystals:
from superprism to supercollimator,” J. Lightwave Technol. 21, 561–566 (2003).
[144] M. J. Steel, R. Zoli, C. Grillet, R. C. McPhedran, C. M. de Sterke, A. Norton,
P. Bassi, and B. J. Eggleton, “Analytic properties of photonic crystal superprism
paramters,” Phys. Rev. E 71, 056608 (2005).
[145] D. N. Chigrin, S. Enoch, C. M. Sotomayer Torres, and G. Tayeb, “Self-guiding in
two-dimensional photonic crystals,” Opt. Express 11, 1203–1211 (2003).
[146] D. W. Prather, S. Shi, D. M. Pustai, C. Chen, S. Venkataraman, A. Sharkawy,
G. J. Schneider, and J. Murakowski, “Dispersion-based optical routing in photonic
crystals,” Opt. Lett. 29, 50–52 (2004).
[147] X. Yu and S. Fan, “Bends and splitters for self-collimated beams in photonic
crystals,” Appl. Phys. Lett. 83, 3251–3253 (2003).
[148] S. Shi, A. Sharkawy, C. Chen, D. M. Pustai, and D. W. Prather, “Dispersion-
based beam splitter in photonic crystals,” Opt. Lett. 29, 617–619 (2004).
[149] H. Kosaka, T. Kawashima, A. Tomita, M. Notomi, T. Tamamura, T. Sato, and
S. Kawakami, “Photonic crystals for micro lightwave circuits using wavelength
dependent angular beam steering,” Appl. Phys. Lett. 74, 1370–1372 (1999).
[150] S. Y. Lin, V. M. Hietala, L. Wang, and E. D. Jones, “Highly dispersive photonic
band-gap prism,” Opt. Lett. 21, 1771–1773 (1996).
[151] L. Wu, M. Mazilu, T. Karle, and T. F. Krauss, “Superprism phenomena in planar
photonic crystals,” IEEE J. Quantum Electron. 38, 915–918 (2002).
[152] J. J. Baumberg, N. M. B. Perney, M. C. Netti, M. D. C. Charlton, M. Zoorob,
and G. J. Parker, “Visible-wavelength super-refraction in photonic crystal super-
prisms,” Appl. Phys. Lett. 85, 354–356 (2004).
[153] T. Matsumoto and T. Baba, “Photonic crystal k-vector superprism,” J. Lightwave
Technol. 22, 917–922 (2004).
[154] C. Luo, M. Soljacic, and J. D. Joannopoulos, “Superprism effect based on phase
velocities,” Opt. Lett. 29, 745–747 (2004).
BIBLIOGRAPHY 165
[155] C. T. Chan, Q. L. Yu, and K. M. Ho, “Order-N spectral methods for electromag-
netic waves,” Phys. Rev. B 51, 16635–16642 (1995).
[156] M. D. Feit and J. A. Fleck, Jr., “Computation of mode properties in optical fiber
waveguides by a propagating beam method,” Appl. Opt. 19, 1154–1164 (1980).
[157] B. J. Eggleton, P. S. Westbrook, C. A. White, C. Kerbage, R. S. Windeler,
and G. L. Burge, “Cladding-mode-resonances in air-silica microstructured optical
fibers,” J. Lightwave Technol. 18, 1084–1100 (2000).
[158] M. Koshiba, Y. Tsuji, and M. Hikari, “Time-domain beam propagation method
and its application to photonic crystal circuits,” J. Lightwave Technol. 18, 102–
110 (2000).
[159] K. M. Leung and Y. F. Liu, “Full vector wave calculation of photonic band struc-
tures in face-centered-cubic dielectric Media,” Phys. Rev. Lett. 65, 2646–2649
(1990).
[160] K. Busch, S. F. Mingaleev, A. Garcia-Martin, M. Schillinger, and D. Hermann,
“The Wannier function approach to photonic crystal circuits,” J. Phys.:Condens.
Matter 15, R1233–R1256 (2003).
[161] N. A. Nicorovici, R. C. McPhedran, and L. C. Botten, “Photonic band gaps for
arrays of perfectly conducting cylinders,” Phys. Rev. E 52, 1135–1145 (1995).
[162] D. Felbacq, G. Tayeb, and D. Maystre, “Scattering by a random set of parallel
cylinders,” J. Opt. Soc. Am. A 11, 2526–2538 (1994).
[163] L. M. Li and Z. Q. Zhang, “Multiple-scattering approach to finite-sized photonic
band-gap materials,” Phys. Rev. B 58, 9587–9590 (1998).
[164] P. Yeh, Optical waves in layered media (Wiley, New York, 1988).
[165] J. B. Pendry and A. MacKinnon, “Calculation of photon dispersion relations,”
Phys. Rev. Lett. 69, 2772–2775 (1992).
[166] M. Sigalas, C. M. Soukoulis, E. N. Economou, C. T. Chan, and K. M. Ho, “Pho-
tonic band gaps and defects in two dimensions: Studies of the transmission coef-
ficient,” Phys. Rev. B 48, 14121–14126 (1993).
[167] L. Li, “Formulation and comparison of two recursive matrix algorithms for mod-
eling layered diffraction gratings,” J. Opt. Soc. Am. A 13, 1024–1035 (1996).
[168] E. Silberstein, P. Lalanne, and J.-P. Hugonin, “Use of grating theories in inte-
grated optics,” J. Opt. Soc. Am. A 18, 2865–2875 (2001).
166 BIBLIOGRAPHY
[169] L. C. Botten, T. P. White, A. A. Asatryan, T. N. Langtry, C. M. de Sterke, and
R. C. McPhedran, “Bloch mode scattering matrix methods for modeling extended
photonic crystal structures. Part I: Theory,” Phys. Rev. E 70, 056606 (2004).
[170] T. P. White, L. C. Botten, C. M. de Sterke, R. C. McPhedran, A. A. Asatryan,
and T. N. Langtry, “Bloch mode scattering matrix methods for modeling ex-
tended photonic crystal structures. Part II: Applications,” Phys. Rev. E 70,
056607 (2004).
[171] L. C. Botten, T. P. White, C. M. de Sterke, R. C. McPhedran, A. A. Asatryan,
and T. N. Langtry, “Photonic crystal devices modelled as grating stacks: matrix
generalizations of thin film optics,” Opt. Express 12, 1592–1604 (2004).
[172] P. Bienstman and R. Baets, “Optical modelling of photonic crystals and VC-
SELs using eigenmode expansion and perfectly matched layers,” Opt. Quantum
Electron. 33, 327–341 (2001).
[173] M. Neviere and E. Popov, Light propagation in periodic media (Marcel Dekker,
New York, 2003).
[174] R. Petit, ed., Electromagnetic theory of gratings (Springer-verlag, Heidelberg,
1980).
[175] S. F. Mingaleev and K. Busch, “Scattering matrix approach to large-scale pho-
tonic crystal circuits,” Opt. Lett. 28, 619–621 (2003).
[176] L. C. Botten, N. A. Nicorovici, A. A. Asatryan, R. C. McPhedran, C. M. de Sterke,
and P. A. Robinson, “Formulation for Electromagnetic Scattering and Propaga-
tion through Grating Stacks of Metallic and Dielectric Cylinders for Photonic
Crystal Calculations. Part 1: Method,” J. Opt. Soc. Am. A 17, 2165–2176 (2000).
[177] L. C. Botten, N. A. Nicorovici, R. C. McPhedran, C. M. de Sterke, and A. A.
Asatryan, “Photonic band structure calculations using scattering matrices,” Phys.
Rev. E 64, 046603 (2001).
[178] G. H. Smith, L. C. Botten, R. C. McPhedran, and N. A. Nicorovici, “Cylinder
graitings in conical incidence with applications to woodpile structures,” Phys.
Rev. E 67, 056620 (2003).
[179] B. Gralak, S. Enoch, and G. Tayeb, “From scattering or impedance matrices to
Bloch modes of photonic crystals,” J. Opt. Soc. Am. A 19, 1547–1554 (2002).
[180] L. C. Botten, A. A. Asatryan, T. N. Langtry, T. P. White, C. M. de Sterke, and
R. C. McPhedran, “Semianalytic treatment for propagation in finite photonic
crystal waveguides,” Opt. Lett. 28, 854–856 (2003).
BIBLIOGRAPHY 167
[181] T. P. White, L. C. Botten, R. C. McPhedran, and C. M. de Sterke, “Ultracompact
resonant filters in photonic crystals,” Opt. Lett. 28, 2452–2454 (2003).
[182] L. C. Botten, R. C. McPhedran, C. M. de Sterke, N. A. Nicorovici, A. A. Asatryan,
G. H. Smith, T. N. Langtry, T. P. White, D. P. Fussell, and B. T. Kuhlmey, “From
multipole methods to photonic crystal device modeling,” in Electromagnetic the-
ory and applications for photonic crystals, K. Yasomoto, ed. (CRC Taylor and
Francis, Boca Raton, 2006).
[183] S. Kuchinsky, V. Y. Golyatin, A. Y. Kutinov, T. P. Pearsall, and D. Nedeljkovic,
“Coupling between photonic crystal waveguides,” IEEE J. Quantum Electron.
38, 1349–1352 (2002).
[184] C. M. de Sterke, L. C. Botten, A. A. Asatryan, T. P. White, and R. C. McPhedran,
“Modes of coupled photonic crystal waveguides,” Opt. Lett. 29, 1384–1386 (2004).
[185] O. Levy, B. Z. Steinberg, M. Nathan, and A. Boag, “Ultrasensitive displacement
sensor using photonic crystal waveguides,” Appl. Phys. Lett. 86, 104102 (2005).
[186] R. Costa, A. Melloni, and M. Martinelli, “Bandpass resonant filters in photonic-
crystal waveguides,” IEEE Photon. Technol. Lett. 15, 401–403 (2003).
[187] A. S. Jugessur, P. Pottier, and R. M. De La Rue, “Engineering the filter response
of photonic crystal microcavity filters,” Opt. Express 12, 1304–1312 (2004).
[188] J. Zimmermann, B. K. Saravanan, R. Marz, M. Kamp, A. Forchel, and S. Anand,
“Large dispersion in photonic crystal waveguide resonator,” Electron. Lett. 41,
414–415 (2005).
[189] A. Safaai-Jazi and C. C. Chang, “Spectral characteristics of coupled-waveguide
Fabry-Perot resonators and filters,” IEEE J. Quantum Electron. 32, 1063–1069
(1996).
[190] C. Manolatou, M. J. Khan, S. Fan, P. R. Villeneuve, H. A. Haus, and J. D.
Joannopoulos, “Coupling of modes analysis of resonant channel add-drop filters,”
IEEE J. Quantum Electron. 35, 1322–1331 (1999).
[191] E. Drouard, H. T. Hattori, C. Grillet, A. Kazmierczak, X. Letarte, P. Rojo-
Romero, and P. Viktorovitch, “Directional channel-drop filter based on a slow
Bloch mode photonic crystal waveguide section,” Opt. Express 13, 3037–3048
(2005).
[192] Y. Akahane, M. Mochizuki, T. Asano, Y. Tanaka, and S. Noda, “Design of a
channel drop filter by using a donor-type cavity with high-quality factor in a
two-dimensional photonic crystal slab,” Appl. Phys. Lett. 82, 1341–1343 (2003).
168 BIBLIOGRAPHY
[193] Y. Xu, Y. Li, R. K. Lee, and A. Yariv, “Scattering-theory analysis of waveguide-
resonator coupling,” Phys. Rev. E 62, 7389–7404 (2000).
[194] E. Waks and J. Vuckovic, “Coupled mode theory for photonic crystal cavity-
waveguide interaction,” Opt. Express 13, 5064–5073 (2005).
[195] S. Fan, “Sharp asymmetric line shapes in side-coupled waveguide-cavity systems,”
Appl. Phys. Lett. 80, 908–910 (2002).
[196] S. Fan, S. G. Johnson, and J. D. Joannopoulos, “Waveguide branches in photonic
crystals,” J. Opt. Soc. Am. B 18, 162–165 (2001).
[197] W. J. Kim and J. D. O’Brien, “Optimization of a two-dimensional photonic-
crystal waveguide branch by simulated annealing and the finite-element method,”
J. Opt. Soc. Am. B 21, 289–295 (2004).
[198] R. Wilson, T. J. Karle, I. Moerman, and T. F. Krauss, “Efficient photonic crystal
Y-junctions,” J. Opt. A 5, S76–S80 (2003).
[199] M. Ayre, T. J. Karle, L. Wu, T. Davies, and T. F. Krauss, “Experimental verifi-
cation of numerically optimised photonic crystal injector, Y-splitter, and bend,”
IEEE J. Sel. Area. Comm. 23, 1390–1395 (2005).
[200] A. Martinez, A. Griol, P. Sanchis, and J. Marti, “Mach-Zehnder interferometer
employing coupled-resonator optical waveguides,” Opt. Lett. 28, 405–407 (2003).
[201] Y. Sugimoto, H. Nakamura, Y. Tanaka, N. Ikeda, and K. Asakawa, “High-
precision optical interference in Mach-Zehnder-type photonic crystal photonic
crystal waveguide,” Opt. Express 13, 96–105 (2005).
[202] C. Elachi and C. Yeh, “Distribution networks and electrically controllable couplers
for integrated optics,” Appl. Opt. 13, 1372–1375 (1974).
[203] H. A. Haus and L. Molter-Orr, “Coupled multiple waveguide systems,” IEEE J.
Quantum Electron. QE-19, 840–844 (1983).
[204] T. P. White, C. M. de Sterke, R. C. McPhedran, T. Huang, and L. C. Botten,
“Recirculation-enhanced switching in photonic crystal Mach-Zehnder interferom-
eters,” Opt. Express 12, 3035–3045 (2004).
[205] I. Park, H. S. Lee, H. J. Kim, K. M. Moon, S. G. Lee, B. H. O, S. G. Park, and
E. H. Lee, “Photonic crystal power-splitter based on directional coupler,” Opt.
Express 12, 3599–3604 (2004).
[206] T. Liu, A. R. Zakharian, M. Fallahi, J. V. Maloney, and M. Mansuripur, “Multi-
mode interference-based photonic crystal waveguide power splitter,” J. Lightwave
Technol. 22, 2842–2846 (2004).
BIBLIOGRAPHY 169
[207] P. I. Boral, L. H. Frandsen, A. Harpøth, M. Kristensen, J. S. Jensen, and O. Sig-
mund, “Topology optimised broadband photonic crystal Y-splitter,” Electron.
Lett. 41, 69–71 (2005).
[208] B. E. Little, S. T. Chu, H. A. Haus, J. Foresi, and J. P. Laine, “Microring resonator
channel dropping filters,” J. Lightwave Technol. 15, 998–1005 (1997).
[209] A. Yariv, Y. Xu, R. K. Lee, and A. Scherer, “Coupled-resonator optical waveguide:
a proposal and analysis,” Opt. Lett. 24, 711–713 (1999).
[210] M. Bayindir, B. Temelkuran, and E. Ozbay, “Tight-binding description of the
coupled defect modes in three-dimensional photonic crystals,” Phys. Rev. Lett.
84, 2140–2143 (2000).
[211] S. Pereira, P. Chak, and J. Sipe, “Gap-soliton switching in short microresonator
structures,” J. Opt. Soc. Am. B 19, 2191–2202 (2002).
[212] S. Pereira, P. Chak, and J. E. Sipe, “All-optical AND gate by use of a Kerr
nonlinear microresonator structure,” Opt. Lett 28, 444–446 (2003).
[213] B. E. Little, S. T. Chu, J. V. Hryniewicz, and P. P. Absil, “Filter synthesis for
periodically coupled microring resonators,” Opt. Lett. 25, 344–346 (2000).
[214] M. F. Yanik and S. Fan, “Stopping and storing light coherently,” Phys. Rev. A
71, 013803 (2005).
[215] A. N. Chester, S. Martellucci, and A. M. V. Scheggi, eds., Optical fiber sensors
(Martinus Nijhoff, Dordrecht, 1987).
[216] G. V. Treyz, “Silicon Mach-Zehnder waveguide interferometers operating at
1.3µm,” Electron. Lett. 27, 118–120 (1991).
[217] C. Rolland, R. S. Moore, F. Shepherd, and G. Hillier, “10Gbit/s, 1.56µm Mul-
tiquantum well InP/InGaAsP Mach-Zehnder optical modulator,” Electron. Lett.
29, 471–472 (1993).
[218] E. A. Camargo, H. M. H. Chong, and R. M. De La Rue, “Photonic crystal Mach-
Zehnder structures for thermo-optic switching,” Opt. Express 12, 588–592 (2004).
[219] M. H. Shih, W. J. Kim, W. Kuang, J. R. Cao, H. Yukawa, S. J. Choi, J. D.
O’Brien, and P. D. Dapkus, “Two-dimensional photonic crystal Mach-Zehnder
interferometers,” Appl. Phys. Lett. 84, 460–462 (2004).
[220] H. Nakamura, Y. Sugimoto, K. Kanamoto, N. Ikeda, Y. Tanaka, Y. Nakamura,
S. Ohkouchi, Y. Watanabe, K. Inoue, H. Ishikawa, and K. Asakawa, “Ultra-fast
photonic crystal/quantum dot all-optical switch for future photonic networks,”
Opt. Express 12, 6606–6614 (2004).
170 BIBLIOGRAPHY
[221] A. Lan, K. Kanamoto, T. Yang, A. Nishikawa, Y. Sugimoto, N. Ikeda, H. Naka-
mura, K. Asakawa, and H. Ishikawa, “Similar role of waveguide bends in photonic
crystal circuits and disordered defects in coupled cavity waveguides: An intrinsic
problem in realizing photonic crystal circuits,” Phys. Rev. B 67, 115208 (2003).
[222] C. Y. Liu and L. W. Chen, “Tunable photonic-crystal waveguide Mach-Zehnder
interferometer achieved by nematic liquid-crystal phase modulation,” Opt. Ex-
press 12, 2616–2624 (2004).
[223] T. Ranganath and S. Wang, “Ti-diffused LiNbO3 branched-waveguide modula-
tors: performance and design,” IEEE J. Quantum Electron. QE-13, 290–295
(1977).
[224] G. P. Agrawal, Fiber-optic communication systems, 3rd ed. (John Wiley and Sons,
New York, 2002).
[225] B. Wang and G. P. Wang, “Surface plasmon polariton propagation in nanoscale
metal gap waveguides,” Opt. Lett. 29, 1992–1994 (2004).
[226] T. P. White, C. M. de Sterke, R. C. McPhedran, and L. C. Botten, “Highly
efficient wide-angle coupling into uniform rod-type photonic crystals,” Appl. Phys.
Lett. 87, 111107 (2005).
[227] A. Mekis and J. D. Joannopoulos, “Tapered couplers for efficient interfacing be-
tween dielectric and photonic crystal waveguides,” J. Lightwave Technol. 19, 861–
865 (2001).
[228] W. Kuang, C. Kim, A. Stapleton, and J. D. O’Brien, “Grating-assisted coupling of
optical fibers and photonic crystal waveguides,” Opt. Lett. 27, 1604–1606 (2002).
[229] A. Talneau, P. Lalanne, M. Agio, and C. M. Soukoulis, “Low-reflection photonic-
crystal taper for efficient coupling between guide sections of arbitrary widths,”
Opt. Lett. 27, 1522–1524 (2002).
[230] P. E. Barclay, K. Srinivasan, M. Borselli, and O. Painter, “Probing the dispersive
and spatial properties of photonic crystal waveguides via highly efficient coupling
from fiber tapers,” Appl. Phys. Lett. 85, 4–6 (2004).
[231] J. Witzens, M. Hochberg, T. Baehr-Jones, and A. Scherer, “Mode matching in-
terface for efficient coupling of light into planar photonic crystals,” Phys. Rev. E.
69, 046609 (2004).
[232] T. Baba and D. Ohsaki, “Interfaces of photonic crystals for high efficiency light
transmission,” Jpn. J. Appl. Phys. 40, 5920–5924 (2001).
BIBLIOGRAPHY 171
[233] J. Witzens and A. Scherer, “Efficient excitation of self-collimated beams and
single Bloch modes in planar photonic crystals,” J. Opt. Soc. Am. A 20, 935–940
(2003).
[234] E. N. Economou and A. Zdetsis, “Classical wave propagation in periodic struc-
tures,” Phys. Rev. B 40, 1334–1337 (1989).
[235] C. M. Soukoulis, ed., Photonic bandgaps and localization (Plenum Press, New
York, 1993).
[236] E. Lidorikis, M. M. Sigalas, E. N. Economou, and C. M. Soukoulis, “Tight-binding
parametrization for photonic band gap materials,” Phys. Rev. Lett. 81, 1405–1408
(1998).
[237] H. C. van de Hulst, Light scattering by small particles (John Wiley and Sons,
New York, 1957).
[238] T. Baba, T. Matsumoto, and M. Echizen, “Finite difference time domain study of
high efficiency photonic crystal superprisms,” Opt. Express 12, 4608–4613 (2004).
[239] R. T. Bise, R. S. Windeler, K. S. Kranz, C. Kerbage, B. J. Eggleton, and D. J.
Trevor, “Tunable photonic bandgap fiber,” in Optical Fiber Communication Con-
ference, Postconference Edition, vol. 70 of OSA Trends in Optics and Photonics,
pp. 466–468 (Optical Society of America, Washington, DC, USA, 2002).
[240] T. P. White, R. C. McPhedran, C. M. de Sterke, N. M. Litchinitser, and B. J.
Eggleton, “Resonance and scattering in microstructured optical fibers,” Opt. Lett.
27, 1977–1979 (2002).
[241] T. W. Ebbesen, H. J. Lezec, H. F. Ghaemi, T. Thio, and P. A. Wolff, “Extraor-
dinary optical transmission through sub-wavelength hole arrays,” Nature 391,
667–669 (1998).
[242] F. J. Garcıa de Abajo, G. Gomez-Santos, L. A. Borislav, and S. V. Shabanov,
“Tunneling mechanism of light transmission through metallic films,” Phys. Rev.
Lett. 95, 067403 (2005).
[243] F. J. Garcıa de Abajo, R. Gomez-Medina, and J. J. Saenz, “Full transmission
through perfect-conductor subwavelength hole arrays,” Phys. Rev. E 72, 016608
(2005).
[244] A. Mandatori, C. Sibilia, M. Bertolotti, S. Zhukovsky, and S. V. Gaponenko, “An-
alytic demonstration of omnidirectional transmission enhancement in dispersive
birefringent photonic-bandgap structures,” J. Opt. Soc. Am. B 22, 1785–1792
(2005).
172 BIBLIOGRAPHY
[245] T. Yamashita and C. J. Summers, “Evaluation of self-collimated beams in pho-
tonic crystals for optical interconnect,” IEEE J. Sel. Area. Comm. 23, 1341–1347
(2005).
[246] M. Palamaru and P. Lalanne, “Photonic crystal waveguides: out-of-plane losses
and adiabatic modal conversion,” Appl. Phys. Lett. 78, 1466–1468 (2001).
[247] P. Lalanne and J. P. Hugonin, “Bloch-wave engineering for high-Q, small-V mi-
crocavities,” IEEE J. Quantum Electron. 39, 1430–1438 (2003).
[248] B. S. Song, T. Asano, Y. Akahane, Y. Tanaka, and S. Noda, “Transmission and
reflection characteristics of in-plane hetero-photonic crystals,” Appl. Phys. Lett.
85, 4591–4593 (2004).
[249] J. Witzens, T. Baehr-Jones, and A. Scherer, “Hybrid superprism with low inser-
tion losses and suppressed cross-talk,” Phys. Rev. E 71, 026604 (2005).
Appendix A
List of author’s publications
Journal papers
[1] M. J. Steel, T. P. White, C. M. de Sterke, and R. C. McPhedran, “Symmetry and
degeneracy in microstructured optical fibers,” Opt. Lett. 26, 488-490 (2001).
[2] T. P. White, R. C. McPhedran, C. M. de Sterke, L. C. Botten, and M. J. Steel,
“Confinement losses in microstructured optical fibers,” Opt. Lett. 26, 1660-1662
(2001).
[3] T. P. White, R. C. McPhedran, L. C. Botten, G. H. Smith, and C. M. de Sterke,
“Calculations of air-guided modes in photonic crystal fibers using the multipole
method,” Opt. Express 9, 721-732 (2001).
[4] T. P. White, B. T. Kuhlmey, R. C. McPhedran, D. Maystre, G. Renversez, C. M.
de Sterke, and L. C. Botten, “Multipole method for microstructured optical fibers.
I. Formulation,” J. Opt. Soc. Am. B 19, 2322-2330 (2002).
[5] B. T. Kuhlmey, T. P. White, G. Renversez, D. Maystre, L. C. Botten, C. M.
de Sterke, and R. C. McPhedran, “Multipole method for microstructured optical
fibers. II. Implementation and results,” J. Opt. Soc. Am. B 19, 2331-2340 (2002).
[6] T. P. White, R. C. McPhedran, C. M. de Sterke, N. M. Litchinitser and B. J.
Eggleton, “Resonance and scattering in microstructured optical fibers,” Opt. Lett.
27, 1977-1979 (2002).
[7] M. Sumetsky, Y. Dulashko, T. P. White, T. Olsen, P. S. Westbrook, and B. J.
Eggleton, “Interferometric characterization of phase masks,” Appl. Opt. 42, 2336-
2340 (2003).
[8] L. C. Botten, A. A. Asatryan, T. N. Langtry, T. P. White, C. M. de Sterke, and
R. C. McPhedran, “Semianalytic treatment for propagation in finite photonic crystal
waveguides,” Opt. Lett. 28, 854–856 (2003).
173
174 APPENDIX A. LIST OF AUTHOR’S PUBLICATIONS
[9] N. M. Litchinitser, S. C. Dunn, B. Usner, B. J. Eggleton, T. P. White, R. C. McPhe-
dran, and C. M. de Sterke, “Resonances in microstructured optical waveguides,”
Opt. Express 10, 1243-1251 (2003).
[10] T. P. White, L. C. Botten, R. C. McPhedran, and C. M. de Sterke, “Ultracompact
resonant filters in photonic crystals,” Opt. Lett. 28, 2452–2454 (2003).
[11] N. M. Litchinitser, S. C. Dunn, P. E. Steinvurzel, B. J. Eggleton, T. P. White, R. C.
McPhedran and C. M. de Sterke, “Application of an ARROW model for designing
tunable photonic devices,” Opt. Express 12, 1540-1550 (2004).
[12] L. C. Botten, T. P. White, C. M. de Sterke, R. C. McPhedran, A. A. Asatryan,
and T. N. Langtry, “Photonic crystal devices modelled as grating stacks: matrix
generalizations of thin film optics,” Opt. Express 12, 1592–1604 (2004).
[13] C. M. de Sterke, L. C. Botten, A. A. Asatryan, T. P. White, and R. C. McPhedran,
“Modes of coupled photonic crystal waveguides,” Opt. Lett. 29, 1384–1386 (2004).
[14] T. P. White, C. M. de Sterke, R. C. McPhedran, T. Huang, and L. C. Botten,
“Recirculation-enhanced switching in photonic crystal Mach-Zehnder interferome-
ters,” Opt. Express 12, 3035–3045 (2004).
[15] M. Sumetsky, T. P. White, H. M. Dyson, P. S. Westbrook, and B. J. Eggleton,
“Interferometric characterization of nonflat transmission diffraction gratings,” IEEE
Photon. Technol. Lett. 16, 2314-2316 (2004).
[16] L. C. Botten, T. P. White, A. A. Asatryan, T. N. Langtry, C. M. de Sterke, and
R. C. McPhedran, “Bloch mode scattering matrix methods for modelling extended
photonic crystal structures. Part I: Theory,” Phys. Rev. E 70, 056606 (2004).
[17] T. P. White, L. C. Botten, C. M. de Sterke, R. C. McPhedran, A. A. Asatryan,
and T. N. Langtry, “Bloch mode scattering matrix methods for modelling extended
photonic crystal structures. Part II: Applications,” Phys. Rev. E 70, 056607 (2004).
[18] P. Steinvurzel, B. T. Kuhlmey, T. P. White, M. J. Steel, C. M. de Sterke, and
B. J. Eggleton, “Long wavelength anti-resonant guidance in high index inclusion
microstructured fibers,” Opt. Express 12, 5424-5433 (2004).
[19] T. P. White, C. M. de Sterke, R. C. McPhedran, and L. C. Botten, “Highly efficient
wide-angle coupling into uniform rod-type photonic crystals,” Appl. Phys. Lett.
87, 111107 (2005).
175
Book chapter
[20] L. C. Botten, R. C. McPhedran, C. M. de Sterke, N. A. Nicorovici, A. A. Asatryan,
G. H. Smith, T. N. Langtry, T. P. White, D. P. Fussell, and B. T. Kuhlmey, “From
multipole methods to photonic crystal device modeling,” in Electromagnetic theory
and applications for photonic crystals, K. Yasomoto, ed. (CRC Taylor and Francis,
Boca Raton 2006).