Thesis EY-3

Controlling emission and propagation of
light with photonic band gap crystals
Beheersen van de emissie en propagatie

van licht met fotonische band gap
kristallen
Promotiecommissie
Promotores Prof. Dr. W. L. Vos

Prof. Dr. A. Lagendijk
Overige leden Prof. Dr. L. Novotny

Prof. Dr. W. L. IJzerman
Prof. Dr. J. Herek
Prof. Dr. V. Subramaniam
The work described in this thesis is financially supported by the

‘Nederlandse Organisatie voor Wetenschappelijk Onderzoek’ (NWO).
It was carried out at the
Complex Photonic Systems (COPS) chair,
Department of Science and Technology
and MESA+ Institute for Nanotechnology,
University of Twente, P.O. Box 217,
7500 AE Enschede, The Netherlands.
This thesis can be downloaded from

http://www.photonicbandgaps.com
ISBN: 9789461087652
CONTROLLING EMISSION AND
PROPAGATION OF LIGHT WITH
PHOTONIC BAND GAP CRYSTALS
PROEFSCHRIFT
ter verkrijging van

de graad van doctor aan de Universiteit Twente,
op gezag van de rector magnificus,
prof. dr. H. Brinksma,
volgens besluit van het College voor Promoties
in het openbaar te verdedigen
op woensdag 1 October 2014 om 12:45 uur
door
Elahe Yeganegi Dastgerdi

geboren op 9 september 1984
te Boroujen, Iran
Dit proefschrift is goedgekeurd door:
Prof. Dr. W. L. Vos en Prof. Dr. A. Lagendijk

Contents
1 Introduction 11
1.1 Spontaneous emission under control . . . . . . . . . . . . . . . . . 11
1.2 General wave description: Green’s functions . . . . . . . . . . . . . 13
1.3 Photonic crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4 Diamond structured photonic crystals . . . . . . . . . . . . . . . . 15
1.5 Importance of three-dimensional photonic band gap . . . . . . . . 17
1.6 Real 3D photonic band gap crystals . . . . . . . . . . . . . . . . . 18
1.7 Probing the local density of states . . . . . . . . . . . . . . . . . . 19
1.8 Outline of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2 Samples and instrumentation 27

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Silicon inverse woodpile photonic crystals . . . . . . . . . . . . . . 27
2.3 Optical properties of PbS quantum dots . . . . . . . . . . . . . . . 29
2.4 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3 Local density of states in the band gap of a finite one-dimensional

photonic crystal 39
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Local density of states in a finite crystal . . . . . . . . . . . . . . . 40
3.3 Field propagation in a periodic media . . . . . . . . . . . . . . . . 41
3.3.1 Infinitely extended photonic crystal . . . . . . . . . . . . . 41
3.3.2 Finite photonic crystal . . . . . . . . . . . . . . . . . . . . . 44
3.4 Model for the LDOS in a finite crystal . . . . . . . . . . . . . . . . 48
3.5 LDOS in the band gap of a finite 1D periodic structure . . . . . . 50
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4 Finite size effects on a 3D photonic band gap 55

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2 Experimental techniques . . . . . . . . . . . . . . . . . . . . . . . . 56
4.3 Experimental observations . . . . . . . . . . . . . . . . . . . . . . . 59
4.4 Theory for 3D finite crystal . . . . . . . . . . . . . . . . . . . . . . 60
4.5 Inhibition in the band gap . . . . . . . . . . . . . . . . . . . . . . . 66
4.6 How big should a photonic band gap crystal be? . . . . . . . . . . 67
4.7 Conclusion and discussion . . . . . . . . . . . . . . . . . . . . . . . 69
8 Contents
5 Statistical analysis of time-resolved emission in the near infrared 73

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.2 Bayesian analysis of time-resolved emission measurement . . . . . . 75
5.2.1 Bayesian method . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2.2 A priori estimation of the background . . . . . . . . . . . . 76
5.2.3 A posteriori estimation for model parameter, and goodness-
of-fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2.4 χ2 credible interval and model consistency . . . . . . . . . . 78
5.3 Experiment and applications . . . . . . . . . . . . . . . . . . . . . 78
5.3.1 Near infrared quantum dots in suspension . . . . . . . . . . 79
5.3.2 Near infrared quantum dots in a 3D photonic band gap . . 80
5.4 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6 Probing photonic crystals with shaped wavefronts 89

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.2 Wavefront shaping setup and samples . . . . . . . . . . . . . . . . 91
6.3 Multiple scattering in presence of long-range order . . . . . . . . . 94
6.4 Intensity correlations in presence of long-range order . . . . . . . . 95
6.5 Response of photonic crystals to a shaped wavefront . . . . . . . . 96
6.5.1 Focusing light on the surface . . . . . . . . . . . . . . . . . 97
6.5.2 Wavefront shaping . . . . . . . . . . . . . . . . . . . . . . . 98
6.6 Optical correlations in photonic crystals . . . . . . . . . . . . . . . 99
6.6.1 Long range correlation in opal photonic crystals . . . . . . 103
6.7 Summary and conclusion . . . . . . . . . . . . . . . . . . . . . . . 104
7 Light transport in silicon two-dimensional photonic crystals 109

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.3 Sample selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.4 Experimental procedure . . . . . . . . . . . . . . . . . . . . . . . . 112
7.5 Experimental results and discussion . . . . . . . . . . . . . . . . . 114
7.5.1 Out-of-plane scattering on a Teflon sample . . . . . . . . . 114
7.5.2 Out-of-plane scattering of silicon two-dimensional photonic
crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.7 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8 Summary and outlook 125
A Light propagation in periodic media 129
B Decay rate versus frequency of PbS quantum dots 133
C Control radiative linewidth of plasmonics with photonic crystals 137

C.1 Decay mechanism of plasmonic particles . . . . . . . . . . . . . . . 137
Contents 9
C.2 Investigation: controlling plasmonic radiation of gold nanorods

with photonic crystals . . . . . . . . . . . . . . . . . . . . . . . . . 138
C.2.1 Extinction of gold nanorod suspensions . . . . . . . . . . . 138
Algemene Nederlandse samenvatting 143
Acknowledgements 145
CHAPTER 1
Introduction
The importance of light to mankind is reflected by the tremendous efforts that

are being made worldwide to control its creation and propagation. Profiting from
the interplay between light and matter, carefully designed optical elements, such
as mirrors and lenses, have been successfully employed in the last centuries for
light manipulation. Recently it has also become possible to manipulate light on
length scales down to the nanoscale by using composite structures with feature
sizes in the range of the wavelength or even smaller. This research is the realm
of the young and blossoming field called nanophotonics, that has already yielded
exciting new applications such as photonic crystal fibers with custom-made dis-
persion [1] that allow supercontinuum light sources [2], frequency combs and
ultrastable clocks leading to a Nobel prize in physics [3], ultra high density data
storage [4], optical antennas [5], efficient light sources [6], and many more.
One of the most fascinating phenomena in nanophotonics is the three-dimensional
photonic band gap which provides an ultimate control over the interaction be-
tween light and matter, of which spontaneous emission is a well-known exam-
ple [7]. By placing an emitter inside an ideal photonic band gap crystal, the
emitter stays excited forever. The perfect control of the spontaneous emission is
only attainable for an infinitely extended perfect photonic crystal. In real life,
however, all crystals are finite, and disorder is inevitable. Therefore in this thesis
we investigate the light propagation in real finite photonic crystals. This chapter
gives a short introduction about several important subjects relevant to this study.
An outline of this thesis is given at the end of this chapter.
1.1 Spontaneous emission under control

Spontaneous emission is one of the two fundamental processes of emitting a
photon from a two-level system such as an atom, a molecule, or a quantum dot∗ .
A simplistic representation of a two level system consists of a ground state with
energy Ei and an excited state with energy Ef higher than Ei . As is shown in
Fig. 1.1 the starting situation is that the emitter is excited. The emitter stays
∗ The other process of emitting a photon is stimulated emission.
12 Introduction
(a) (b)
e Ef Ef
time τ
Ei Ei e
Figure 1.1: Schematic illustration of the spontaneous emission process. (a) The start-
ing situation: Emitter is excited. (b) The emitter decays back to the ground state by
emitting a photon.
excited for an average time τ , known as the emitters’ lifetime, and then it decays
to the ground state by emitting a photon. Interestingly, the exact moment in time
that a photon is emitted can not be predicted. Nevertheless, the average time it
takes -the lifetime- is well defined and can be finely controlled. The inverse of
the lifetime is the decay rate γ, which is given by Fermi’s Golden rule for dipole
transition [8]:
2π
γ(r) = |f |d̂(r) · E(r)|i|2 δ(Ef − Ei ) (1.1)

|f
Here d̂ is the dipole transition moment operator. The decay rate is determined by
summing over all available final states |f . Fermi’s Golden Rule can be rewritten
as [9, 10]:
πωd2
γ(r, ed , ω) = Nrad (r, ed , ω) (1.2)
0
where the expression separates into an emitter part depending on the magnitude
of the dipole moment d, and the field part or the local density of states LDOS Nrad
that is a property of the emitters’ environment. The LDOS counts the number of
modes into which a photon can be emitted, and can be interpreted as the density
of vacuum fluctuations at the position of the emitter as discovered by Sprik, et
al [9]. Since the spontaneous emission rate depends on the projection of the dipole
moment orientation on the local field, thus the emission rate is enhanced if the
interaction of the dipole with the field is higher than in vacuum, and it is inhibited
if it is lower. This equation only holds if the interaction between the emitter
and the field is in the Markovian regime of quantum electrodynamics where
the electromagnetic field-emitter interaction has an infinitely short memory [11].
Since the LDOS is a general physical property in any light-matter interaction,
it is possible to control the radiative bandwidth of the resonance of a plasmonic
particle (see App. C).
This irreversible emission process happens spontaneously and the emission time
cannot be predicted a priori. The irreversible dynamics is valid in the Markovian
General wave description: Green’s functions 13
regime where the transferred energy from the emitter to the electromagnetic
field is irreversibly lost. A different type of dynamics occurs if the interaction
between the emitter and vacuum modes of the electromagnetic field posses a
finite memory time. In this case Fermi’s Golden rule breaks down. An example
is Rabi oscillations in which the emitter is coupled with a single resonance of the
electromagnetic field leading to a periodic exchange of excitation energy between
the atom and the field [12–14]. Strong variations in the spectral, and spatial
distribution of the LDOS leads to fractional decay, where the Fermi’s golden rule
does not apply [15, 16]. Another breakdown of Fermi’s golden rule occurs when
the LDOS is nonstationary in time [17]. Fast modulations of the LDOS at times
shorter than the emitters’ lifetime cause a time dependent dynamics [18]. The
study in this thesis is in the Markovian regime where Fermi’s Golden Rule is
valid.
1.2 General wave description: Green’s functions

Green’s function method is a very useful tool that provides complete information
about systems that are described by linear differential equations [19]. In heat
conduction the Green’s function describes temperature at a chosen point due to
a heat source located at the source point [20]. In classical mechanics the Green’s
function stands for displacements due to an applied force [21]. In electrodynamics
it represents the electric field at the field point caused by an electric dipole emitter
at the source point [22].
If a system is homogeneous the position of the source is unimportant. How-
ever, for an inhomogeneous medium the source position is an important parame-
ter since the response to the point source depends on the source’s position. The
number of singularities in the Green’s function gives the number of modes, be-
cause the Green’s function diverges at resonant modes of the system. The scalar
Green’s function for a dipole in free space or in a homogeneous environment
describes spherical waves [22]

e±ik|r−r |
G0 (r, r ) = . (1.3)
4π|r − r |
The solution with the plus sign denotes a spherical wave that propagates out
of the origin whereas the solution with the minus sign is a wave that con-
verges towards the origin. Since a dipole is generating the field it is physi-
cally correct to only retain the outward propagating wave. One of the physi-
cal properties accessible through the Green’s function is the LDOS, that is the
Im(G0 (r, r, ω)) [10, 23, 24]. Thus for a homogeneous medium taking Im(G0 (r, r)),
in Eq. (1.3) gives rise to LDOS = 1 for any choice of r as expected for a homo-
geneous system.
The solution becomes much more complicated if the emitter is in a complex
environment such as a photonic crystal [25]. For complex media, such as pho-
tonic crystals the Green’s function is not known, since the Maxwell equations
cannot be solved analytically. The Green’s function can be written as expansion
14 Introduction
of eigenfunctions of the system [10]. In practice one can calculate the eigenvalues
and eigenfunctions of an infinite system using methods such as plane wave expan-
sion methods in frequency domain or finite-difference time-domain(FDTD) [26].
In this thesis we propose a first theory for the LDOS in the band gap of a finite
photonic crystal and it turns out that the insights in Green’s function are crucial
to solve this problem.
1.3 Photonic crystals

Photonic crystals are periodic composite structures of two or more different ma-
terials with a periodicity in the order of wavelength of light [26]. Typically one
of the materials is air and the other one is a high index material such as silicon,
GaAs, or AlAs. Depending on the periodicity, light with a certain wavelength
range can not propagate into the crystal due to the Bragg diffraction. The Bragg
condition is satisfied when light reflected from different planes in the crystal
interferes constructively and thus a high reflectivity peak is observed at these
wavelengths. Figure 1.2 illustrates the Bragg condition mλ = 2dsinθ, with λ the
Bragg diffracted wavelength, m an integer that is the order of the diffraction, d
the spacing between the lattice planes, and θ the angle which light propagates
with respect to the crystal planes. The appearance of a band gap is a consequence
of Bragg diffraction. At the Bragg condition, a range of frequency exist for which
propagation of light in certain directions is not possible: the so-called stop gap.
Where the range of frequencies is determined experimentally, it is described with
the term “stop band”.
A basic parameter of photonic crystals is the photonic strength S. The pho-
tonic strength S that is defined as the polarizability per volume is a measure to
describe how strongly light interacts with a complex photonic medium [27]. For
very high S a band gap is expected to occur in periodic media. By measuring
the central frequency of the stop band ωgap and the stop band width Δωgap the
photonic strength is determined as S = Δωgap /ωgap . For an opal photonic crys-
tal made by self-assembly processes of polystyrene particles S ≈ 0.06, whereas
for formation of a full photonic band gap the photonic strength should be higher
: S ≥ 0.2 [28].
The simplest photonic crystal, also known as Bragg stack, is an alternating
sequence of layers of two or more materials, in which periodicity takes place
only in one direction. If the structure is periodic in 2 directions it is a two-
dimensional photonic crystal. A two-dimensional photonic crystal is capable of
controlling light propagation in plane and is mostly used in in-plane photonic
crystal waveguides or cavities [29–34]. The full potential of photonic crystals
is achieved when the periodicity is extended in all directions forming a three-
dimensional (3D) photonic crystal. If the stop gaps that appear at all directions
overlap with each other a 3D full photonic band gap is formed where light is
forbidden to travel in the crystal in all directions and for both polarizations [7].
There are three main classes of 3D photonic crystals that function as photonic
band gap crystals: inverse opal [35–39], woodpile [40–48], and inverse woodpile
Diamond structured photonic crystals 15
θ
d
2 d sin θ
Figure 1.2: Schematic representation of the Bragg condition. Incident light is reflected
from consecutive lattice planes. When the optical path length difference of the reflected
light from different planes (2dsinθ) is an integer number of the wavelength, the waves
interfere constructively.
photonic crystals [49–55]. Although a photonic band gap can be seen in several
other crystal geometries, the inverse woodpile photonic crystals are very inter-
esting because of their relatively easy and robust fabrication process by etching
pores in two directions and their high photonic strength of S = 0.25, resulting
in a broad band gap [26, 56]. The photonic crystals studied in this thesis are 2D
rectangular lattice and 3D inverse woodpile crystals with a diamond-like struc-
ture, fabricated by etching pores in a silicon wafer [56]. In chapter 2 we will
describe the silicon inverse woodpile photonic crystals in more detail.
1.4 Diamond structured photonic crystals

In crystallography a diamond structure is a structure in which atoms are located
in the same position as carbon atoms in a diamond crystal. The crystal structure
of diamond consists of a face-centered cubic (fcc) lattice, with a basis of two
identical carbon atoms associated with each lattice point: one at (0, 0, 0) and
the other at (1/4, 1/4, 1/4), where the coordinates are given as fractions along
the cube sides a. This is the same as two interpenetrating fcc lattices, each offset
from the other along a body diagonal by one-quarter of its length a/4. The cubic
lattice parameter a is the same as for the fcc lattice, but there are two atoms per
rhombohedal primitive cell [57].
In the early 1990s the possibility to use photonic crystals with diamond-like
structures was described in a number of papers, see Refs. [49, 58]. These pho-
tonic crystals are extremely interesting because they have significant potential
for wide photonic bandgaps. Therefore, materials with relatively low refractive
index contrasts of m > 1.9 suffice in order to obtain a photonic bandgap. A
simultaneous advantageous feature of 3D diamond-like photonic crystals is that
the bandgap is robust to unavoidable fabrication deviations and random disorder,
see Refs. [59–61] for calculations on this aspect.
In one of the earliest studies calculations showed that air-spheres arranged
in a diamond structure with a refractive index contrast of 3.6 would give wide
16 Introduction
(a) (b)
( c)
Figure 1.3: (a) Schematic representation of the structure of an “inverse woodpile”

photonic crystal. It consists of two sets of pores perpendicular to each other, yielding a
three-dimensional (3D) crystal structure. The structure is similar to a diamond crystal
- shown with red spheres as carbon atoms as in diamond gem stone - yet 10,000 times
magnified. The crystal controls the flow of light in an analogous way to how an atomic
crystal controls the flow of electrons. Image courtesy of Léon Woldering. (b) SEM image
of silicon inverse woodpile photonic crystal made in our group using CMOS-compatible
nanotechnology. (c) Schematic image of position and overlap of the pores in our inverse
woodpile crystal.
bandgaps of up to 28% bandwidth [58]. Unfortunately such diamond structures

remain elusive to date. In a 2004 review diamond-like photonic crystals and the
efforts to obtain them have been reviewed [62].
In 1991 diamond structures were proposed by etching or milling pores of air
sequentially in three different directions in a backbone with a high index of re-
fraction [63]. Such crystals were later fabricated from GaAs [64]. Transmission
measurements were used to study the optical properties of these structures and
an attenuation up to 80% was reported. A width of the bandgap of around 19%
was reported, inferred from the width of the troughs in transmittance. In addi-
tion this paper emphasizes the importance of avoiding making these structures
by etching or milling of tapered pores. This fabrication deviation was shown
to have a significant effect on the band gap. Layer-by-layer woodpile structures
were first proposed by the Iowa State group as a practical way to realize powerful
diamond-structured photonic crystal [49]. The expected maximum width of the
bandgap was predicted to be a sizable 18%. The crystal structure resembles a pile
of logs of wood, hence their name. Diamond-like silicon spiral photonic crystals
have been obtained by glancing angle deposition [65]. In this method, silicon is
grown by electron beam evaporation on substrates that contain tungsten seeds.
These seeds are arranged in a suitable square lattice with a lattice constant near
1000 nm. Silicon woodpiles have also been fabricated by direct laser writing of
a template and subsequent double inversion to amorphous silicon [44]. Based
on scanning electron micrographs, the expected bandgap for this structure was
predicted to have a width of nearly 9%. Optical transmission measurements dis-
Importance of three-dimensional photonic band gap 17
played a strong stop gap in the range of the expected bandgap. An intriguing
method to fabricate diamond-like photonic crystals is biotemplating [66]. In this
method, the diamond-like scale of a beetle is used as a template. By a double,
sol-gel based, inversion method, the template is replicated in titania. The period-
icity of these diamond-like structures is in the order of the wavelength of visible
light. The expected width of the band gap of these structures is calculated to be
around 5%, but the gap is probably narrower or even closed due to the significant
structural distortions that are apparent in the scanning electron micrographs of
the structures. Among all diamond structures, 3D photonic crystals known as
“inverse-woodpile” are of particular interest [49] because of their conceptual ease
of fabrication and a broad photonic band gap that is robust to disorder and fab-
rication imperfections [61, 67, 68]. Broad band gaps with widths exceeding 25%
have been predicted for inverse woodpiles [49]. These crystals consist of pores
that run in two perpendicular directions in a high-index backbone. Thus, the
structure is the inverse of the woodpile structure. Figure 1.3 (a) shows a super-
position of diamond structure with an inverse woodpile 3D photonic crystal. It
is seen that the “carbon-atoms” from the superimposed diamond structure are
positioned in the nodes of inverse-woodpile structure, confirming that an inverse
woodpile has a diamond structure. Figure 1.3 (b) is a scanning electron micro-
graph of a 3D silicon inverse woodpile photonic crystal made in our group [55].
These crystals are realized by etching two sets of perpendicular pored in silicon
wafer as described in Sec 2.2. A schematic image of the positions and overlap
of the pores is shown in Fig. 1.3 (c). We will see in this thesis that the silicon
inverse woodpile 3D photonic crystals are optically very powerful as they reveal
prominent cavity QED effects on excited-state lifetimes.
1.5 Importance of three-dimensional photonic band

gap
Many efforts have been stimulated over the years by the theoretical prediction
that there exists an infinite three-dimensional, periodic structure that exhibits
a three-dimensional (3D) photonic band gap [7, 69–71]. In this frequency range
the density of optical states (DOS) is zero and vacuum fluctuations are inhibited.
This characteristic property of the 3D bandgap crystal leads to novel phenom-
ena in cavity quantum electrodynamics (cQED) [7, 69] where they offer at least
five prospects for new physics. Firstly and probably the most eagerly pursued
phenomenon is the complete inhibition of spontaneous emission. Any interac-
tion mediated by vacuum fluctuations is affected by their suppression in the
band gap [72]. Therefore, a crystal with a 3D photonic band gap not only in-
hibits spontaneous emission - including a shift of the emitter’s frequency known
as the Lamb shift [73] - it will also modify the spectrum of blackbody radia-
tion [74–76], it will affect resonant dipole-dipole interactions including the van
der Waals and Casimir forces [77, 78], it will control the radiative width of plas-
monic resonances [79], and the well-known Förster resonant energy transfer that
is prominent in biology and chemistry [10, 72, 80]. Secondly, once a band gap
18 Introduction
is achieved, the cQED physics becomes even richer by introducing point defects.
A point defect acts as a tiny cavity that is shielded in all three dimensions from
the vacuum by the surrounding crystal [26, 81]. Hence such a photonic band gap
cavity is literally a “nanobox for light”. A third reason why 3D photonic band
gaps are relevant to solid state cQED occurs when a gain medium is introduced
in a nanobox. Such a nanobox with gain offers the promise of a thresholdless
laser [7]. Fourth, an important research theme in cQED is the breaking of the
weak-coupling approximation. There are several ways to break the weak-coupling
limit. One approach is to operate close to a van Hove singularity where the den-
sity of states has a cusp [57]. A second approach to break the weak-coupling
limit consists of rapidly modulating the“bath” that surrounds a two-level quan-
tum emitter [11], using ultrafast all-optical switching methods [17]. Fifth, in
quantum physics there is an active interest in decoherence, that is, the loss of
coherence between the components of a system that is in a quantum superposi-
tion [82]. Hence the shielding of vacuum fluctuations by a 3D photonic band gap
offers opportunities to make optical quantum systems robust to decoherence.
In summary, (3D) photonic crystals with a 3D photonic bandgap play a funda-
mental role in cavity quantum electrodynamics (QED), especially in phenomena
where the local density of optical states plays a central role. One can say that
photonic bandgap crystals offer a knob to dial the density of states for broad
frequency bandwidths over a wide range, from near zero to several times the
value in vacuum. In this thesis we will study spontaneous emission inhibition of
emitters embedded in a 3D bandgap crystal. Here, important progress has oc-
curred in the last decade, bringing the long-sought inhibition from a theoretical
prediction to experimental reality after forty years since the prediction by Bykov,
Yablonovitch, and John [7, 69, 70]
1.6 Real 3D photonic band gap crystals

A complete photonic band gap where the photonic density of states vanishes
and the propagation of light is strictly forbidden in all directions [36, 83] is a
theoretical concept where crystals are assumed to be infinite and perfect. Real
photonic crystals are always of finite size, therefore boundary effects are always
present. In addition randomness and disorder in a real crystal is unavoidable.
As a result the zero number of modes can never be reached and the DOS does
not go to zero.
The first theoretical calculation of the emission rate in finite crystal was done
by Whittaker in 2000 [84], Hermann and Hess in 2002 [85], and Kole in 2003 [86],
resulting in a finite inhibition in the band gap. Hermann and Hess, and Kole
calculated the modified emission rate for 3D finite inverse opal photonic band gap
crystals using finite-difference time-domain methods, and Whittaker for woodpile
photonic band gap crystals that are only finite in thickness but infinite laterally
using scattering matrix approach. In addition to the finite size that limits the
inhibition and introduces a small but finite DOS in the band gap, surface states
are also omnipresent when a crystal is finite and they contribute to the inhibition
Probing the local density of states 19
in the band gap. In this thesis we will see that contribution of the surface states
is negligible compared to the finite size effect due to the finite crystal volume,
see chapter 4. It is also important to keep in mind that the surface states exist
only at some particular termination of the crystal as discussed in Ref.[26]. In
addition to the finite size, in a real photonic crystal disorder and randomness are
also inevitable. Several studies show effect of disorder on the DOS in photonic
band gap crystals [59, 87]. As the theoretical calculations show the disorder gives
rise to the Lifshitz tail in the band gap, and therefore narrowing down the band
gap width. In chapter 6 we study the disorder induced to the crystals during
the fabrication process of our silicon photonic crystals. Similar measurements
are performed on synthesized opal photonic crystals. The coherent length, long
range, and short range order are studied showing that the disorder effects can be
neglected in fabricated photonic crystals. To the best of our understanding, the
most important property of a real photonic crystal that makes it different than
an ideal crystal are the finite size effects. Therefore in chapters 3 and 4 the finite
size effect is extensively discussed.
1.7 Probing the local density of states

Since the LDOS is one of the two determining factor of the coupling between
quantum emitters and their photonic environment, measuring the LDOS provides
key fundamental insights into light-matter interaction in nano-scales.
As mentioned in Sec.1.1, one method to probe the LDOS in nanoscale is emis-
sion measurements. The local density of states is proportional to the radiative
decay rate of emitters. By fixing the emitters position at particular positions in
the nano-structure and optically exciting them the emitters decay rate is mea-
sured [88]. Since the radiative decay rate is proportional to the LDOS, the LDOS
can be derived precisely. In this case the efficiency of the method is limited by
the quantum efficiency of emitters and the spatial resolution of measuring the
LDOS depends on how accurately emitters are positioned in the structure. Self-
assembled quantum dots which are grown in the high index material are used to
measure the LDOS in high index part of the structure [89], and colloidal quantum
dots or dye molecules that are fixed in low index material or bound to structure
with polymer layers are used to measure the LDOS in the low index part of the
structure [90].
Another method is Cathodoluminescence microscopy, where a beam of fast
electrons is used as excitation source. The advantage of this method is that
structures can be probed with a very high spatial resolution of 1-10 nm, similar
to electron microscopes. However, this is an advantage from the imaging point
of view. For probing the local properties of the structure the high resolution is
perished due to diffusion of electrons inside the structure. Therefore despite a
very high resolution focus, the measured values for the LDOS are averaged over
the diffused volume of the electrons [91]. Since the electron beam excites the
electron in high index material this method is limited to measuring the LDOS
only in high index material.
20 Introduction
A third method to measure the LDOS is to measure the group velocity at

constant frequency for all wave vectors, since the DOS can be written as an
integral of the inverse group velocity over all wave vectors. and scale that with the
local field at every position in the structure for 4π angles [57]. Due to complexity,
this method has not been successfully applied for the LDOS measurements after
it has been proposed theoretically.
1.8 Outline of this thesis

In this thesis both theoretical and experimental studies about light propagation
in real photonic band gap crystals are presented. We investigate several aspects
of a real photonic band gap crystal by different types of measurements such as
time resolved emission of light sources in the band gap, scattering of light due to
randomness and probing the internal wave structure in the photonic crystals.
In chapter 2, we describe the silicon inverse woodpile photonic crystals that we

use in our emission control measurements in this thesis. We present our study on
optical properties of PbS quantum dots that have been used as light sources in
the photoic band gap crystal. Finally, we describe the near-infrared setup used
for most of our measurements throughout this thesis.
In chapter 3, we present our theoretical study of the local density of states

(LDOS) in the band gap of a finite one-dimensional photonic crystal. In this
study we propose two new Ansatzes based on well-known physical concepts. Our
original view points allow us to develop a model for the LDOS in the band gap
of a finite photonic crystal.
Chapter 4 describes a quantitative study of finite size effect in 3D photonic

band gap crystal. We first present our experimental study on inhibition in the
band gap as a function of frequency. In this study we observe the highest ever
inhibition observed in the band gap range. The inhibition is very strong, yet
limited, due to the finite size of the crystal. We develop a theory to describe the
finite size effect on inhibition in the band gap crystal.
Chapter 5 demonstrates a Bayesian statistical analysis approach that we ap-

plied to our time-resolved emission data analysis. Due to a high level of back-
ground in the time-resolved measurements a Bayesian analysis was required to
refine the model parameters.
In chapter 6, we present our study on probing the internal structure of synthe-

sized opal and fabricated 2D silicon photonic crystals using imaging techniques.
Although the methods that are used in this chapter are very simple, the col-
lected data is surprisingly complex to analyze. We present different approaches
to probe the internal structure of a real photonic crystal and we conclude this
chapter with proposed methods for the future experiments.
Bibliography 21
In chapter 7, we study out-of-plane scattering of light in 2D silicon photonic

crystals. By illuminating a photonic crystal and a multiple scattering sample
with a broad band white light source we quantify the differences in light trans-
port inside periodic and random photonic structures.
Chapter 8 concludes the thesis. A summary of the thesis is presented and an

outlook on future theoretical and experimental studies is given.
The initial research question of this thesis was to manipulate the radiative rate
of the plasmonic particles using photonic band gap crystals. Preliminary exper-
iments were done to investigate the feasibility of this study. However, it turned
out that due to a low quality of the currently available plasmonic particles the
experiment would not work. The preliminary results are presented in appendix C
of this thesis.
Bibliography
[1] P. Russell, Photonic crystal fibers, Science 299, 358 (2003). — p.11.
[2] J. Clowes, Next generation light sources for biomedical applications, Optik
& Photonik 3, 36 (2008). — p.11.
[3] S. A. Diddams, D. J. Jones, J. Ye, S. T. Cundiff, J. L. Hall, J. K. Ranka,
R. S. Windeler, R. Holzwarth, T. Udem, and T. W. Hänsch, Direct link
between microwave and optical frequencies with a 300 thz femtosecond laser
comb, Phys. Rev. Lett. 84, 5102 (2000). — p.11.
[4] B. H. Cumpston et al., Two-photon polymerization initiators for three-
dimensional optical data storage and microfabrication, Nature 398, 51
(1999). — p.11.
[5] P. Bharadwaj, B. Deutsch, and L. Novotny, Optical antennas, Adv. Opt.
Photon. 1, 438 (2009). — p.11.
[6] J. J. Wierer, A. David, and M. M. Megens, III-nitride photonic-crystal light-
emitting diodes with high extraction efficiency, Nature Photon. 3, 163 (2009).
— p.11.
[7] E. Yablonovitch, Inhibited spontaneous emission in solid-state physics and
electronics, Phys. Rev. Lett. 58, 2059 (1987). — p.11, 14, 17, 18.
[8] E. Fermi, Quantum theory of radiation, Rev. Mod. Phys. 4, 87 (1932). —
p.12.
[9] R. Sprik, B. A. van Tiggelen, and A. Lagendijk, Optical emission in periodic
dielectrics, Europhys. Lett. 35, 265 (1996). — p.12.
[10] L. Novotny and B. Hecht, Principles of nano-optics (Cambridge University
Press, New York, 2006). — p.12, 13, 14, 17.
[11] A. Lagendijk, in Vibrational relaxation studied with light, pages 197-238 in:
Ultrashort processes in condensed matter, edited by W. E. Bron (Plenum,
New York, 1993). — p.12, 18.
[12] J. P. Reithmaier, G. Sȩk, A. Löffler, C. Hofmann, S. Kuhn, S. Reitzenstein,
22 Bibliography
L. V. Keldysh, V. D. Kulakovskii, T. L. Reinecke, and A. Forchel, Strong

coupling in a single quantum dot-semiconductor microcavity system, Nature
432, 197 (2004). — p.13.
[13] T. Yoshie, A. Scherer, J. Hendrickson, G. Khitrova, H. M. Gibbs, G. Rupper,
C. Ell, O. B. Shchekin, and D. G. Deppe, Vacuum rabi splitting with a single
quantum dot in a photonic crystal nanocavity, Nature 432, 201 (2004). —
p.13.
[14] P. Maunz, T. Puppe, I. Schuster, N. Syassen, P. W. H. Pinkse, and G.
Rempe, Normal-mode spectroscopy of a single-bound-atom–cavity system,
Phys. Rev. Lett. 94, 033002 (2005). — p.13.
[15] S. John and T. Quang, Spontaneous emission near the edge of a photonic
bandgap, Phys. Rev. A 50, 1764 (1994). — p.13.
[16] P. T. Kristensen, A. F. Koenderink, P. Lodahl, B. Tromborg, and J. Mork,
Fractional decay of quantum dots in real photonic crystals, Opt. Lett. 33,
1557 (2008). — p.13.
[17] P. M. Johnson, A. F. Koenderink, and W. L. Vos, Ultrafast switching of
photonic density of states in photonic crystals, Phys. Rev. B 66, 081102
(2002). — p.13, 18.
[18] H. Thyrrestrup, A. Hartsuiker, J.-M. Gérard, and W. L. Vos, Non-
exponential spontaneous emission dynamics for emitters in a time-dependent
optical cavity, Opt. Express. 21, 23130 (2013). — p.13.
[19] D. Duffy, Green’s functions with applications (Taylor & Francis, 2001). —
p.13.
[20] J. V. Beck, Heat Conduction Using Green’s Functions (Hemisphere Publish-
ing Corporation, 1992). — p.13.
[21] H. Goldstein, C. P. Poole, and J. L. Safko, Classical mechanics (Pearson
Education, Limited, 2013). — p.13.
[22] J. Jackson, Classical electrodynamics (Wiley, 1999). — p.13.
[23] A. Lagendijk and B. A. Van Tiggelen, Resonant multiple scattering of light,
Phys. Rep. 270, 143 (1996). — p.13.
[24] K. Joulain, R. Carminati, J. P. Mulet, and J. J. Greffet, Definition and mea-
surement of the local density of electromagnetic states close to an interface,
Phys. Rev. B 68, 245405 (2003). — p.13.
[25] K. Sakoda, Optical properties of photonic crystals (Springer, 2005). — p.13.
[26] J. D. Joannopoulos, S. G. Johnson, J. N. Winn, and R. D. Meade, Photonic
crystals - molding the flow of light, 2nd edition (Princeton University Press,
2008). — p.14, 15, 18, 19.
[27] W. Vos, R. Sprik, A. van Blaaderen, A. Imhof, A. Lagendijk, and G. H.
Wegdam, Strong effects of photonic band structures on the diffraction of
colloidal crystals, Phys. Rev. B. 53, 16231 (1996). — p.14.
[28] A. F. Koenderink, Emission and transport of light in photonic
crystals, Ph.D. thesis, (University of Amsterdam) available from:
www.photonicbandgaps.com, 2003. — p.14.
[29] T. F. Krauss, R. M. De La Rue, and S. Brand, 2-dimensional photon-
icbandgap structures operating at near-infrared wavelengths, Nature 383, 699
(1996). — p.14.
Bibliography 23
[30] M. Kanskar, P. Paddon, V. Pacradouni, R. Morin, A. Busch, J. F. Young,

S. R. Johnson, J. MacKenzie, and T. Tiedje, Observation of leaky slab modes
in an air-bridged semiconductor waveguide with a two-dimensional photonic
lattice, Appl. Phys. Lett. 70, 1438 (1997). — p.14.
[31] T. Baba, N. Fukaya, and J. Yonekura, Observation of light propagation
in photonic crystal optical waveguides with bends, Electron. Lett. 35, 654
(1999). — p.14.
[32] M. Lončar, D. Nedeljković, T. Doll, J. Vučković, A. Scherer, and T. P.
Pearsall, Waveguiding in planar photonic crystals, Appl. Phys. Lett. 77,
1937 (2000). — p.14.
[33] S. Olivier, C. Smith, M. Rattier, H. Benisty, C. Weisbuch, T. Krauss,
R. Houdre, and U. Oesterle, Miniband transmission in a photonic crystal
coupled-resonator optical waveguide, Opt. Lett. 26, 1019 (2001). — p.14.
[34] J. Vuckovic, M. Loncar, H. Mabuchi, and A. Scherer, Optimization of the q
factor in photonic crystal microcavities, IEEE J. Quant. Electron. 38, 850
(2002). — p.14.
[35] J. E. G. J. Wijnhoven and W. L. Vos, Preparation of photonic crystals made
of air spheres in titania, Science 281, 802 (1998). — p.14.
[36] K. Busch and S. John, Photonic band gap formation in certain self-
organizing systems, Phys. Rev. E 58, 3896 (1998). — p.14, 18.
[37] 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 complete three-dimensional bandgap near 1.5 micrometres, Nature 405,
437 (2000). — p.14.
[38] Y. A. Vlasov, X. Z. Bo, J. C. Sturm, and D. J. Norris, On-chip natural
assembly of silicon photonic bandgap crystals, Nature 414, 289 (2001). —
p.14.
[39] T. Euser, H. Wei, J. Kalkman, Y. Jun, A. Polman, D. J. Norris, and W. L.
Vos, All-optical octave-broad ultrafast switching of si woodpile photonic band
gap crystals, J. Appl. Phys. 102, 053111 (2007). — p.14.
[40] J. G. Fleming and S. Y. Lin, Three-dimensional photonic crystal with a stop
band from 1.35 to 1.95 μm, Opt. Lett. 24, 49 (1999). — p.14.
[41] S. Noda, K. Tomoda, N. Yamamoto, and A. Chutinan, Full three-
dimensional photonic bandgap crystals at near-infrared wavelengths, Science
289, 604 (2000). — p.14.
[42] M. Qi, E. Lidorikis, P. T. Rakich, S. G. Johnson, J. D. Joannopoulos, E. P.
Ippen, and H. I. Smith, A three-dimensional optical photonic crystal with
designed point defects, Nature 429, 538 (2004). — p.14.
[43] I. Povey, D. Whitehead, K. Thomas, M. Pemble, M. Bardosova, and J. Re-
nard, Photonic crystal thin films of gaas prepared by atomic layer deposition,
Appl. Phys. Lett. 89, 104103 (2006). — p.14.
[44] N. Tétreault, G. von Freymann, M. Deubel, M. Hermatschweiler, F. Pérez-
Willard, S. John, M. Wegener, and G. A. Ozin, New route to three-
dimensional photonic bandgap materials: silicon double inversion of polymer
templates, Adv. Mater. 18, 457 (2006). — p.14, 16.
24 Bibliography
[45] K. Aoki, D. Guimard, M. Nishioka, M. Nomura, S. Iwamoto, and Y.

Arakawa, Coupling of quantum-dot light emission with a three-dimensional
photonic-crystal nanocavity, Nature Photon. 2, 688 (2008). — p.14.
[46] S. Ogawa, K. Ishizaki, T. Furukawa, and S. Noda, Spontaneous emission
control by 17 layers of three-dimensional photonic crystals, Electronics Lett.
44, 377 (2008). — p.14.
[47] S. Kawashima, K. Ishizaki, and S. Noda, Light propagation in three-
dimensional photonic crystals, Opt. Express. 18, 386 (2009). — p.14.
[48] K. Ishizaki, M. Koumura, K. Suzuki, K. Gondaira, and S. Noda, Realization
of three-dimensional guiding of photons in photonic crystals, Nature Photon.
7, 133 (2013). — p.14.
[49] K. M. Ho, C. T. Chan, C. M. Soukoulis, R. Biswas, and M. Sigalas, Photonic
band gaps in three dimensions: new layer-by-layer periodic structures, Solid
State Commun. 89, 413 (1994). — p.15, 16, 17.
[50] J. Schilling, J. White, A. Scherer, G. Stupian, R. Hillebrand, and U. Gösele,
Three-dimensional macroporous silicon photonic crystal with large photonic
band gap, Appl. Phys. Lett. 86, 011101 (2005). — p.15.
[51] S. Takahashi, M. Okano, M. Imada, and S. Noda, Three-dimensional pho-
tonic crystals based on double-angled etching and wafer-fusion techniques,
Appl. Phys. Lett. 89, 123106 (2006). — p.15.
[52] M. Hermatschweiler, A. Ledermann, G. A. Ozin, M. Wegener, and G. von
Freymann, Fabrication of silicon inverse woodpile photonic crystals, Adv.
Func. Mater. 17, 2273 (2007). — p.15.
[53] S. Takahashi, K. Suzuki, M. Okano, M. Imada, T. Nakamori, Y. Ota, K.
Ishizaki, and S. Noda, Direct creation of three-dimensional photonic crystals
by a top-down approach, Nat. Mater. 8, 721 (2009). — p.15.
[54] D. J. Shir, E. C. Nelson, D. Chanda, A. Brzezinski, P. V. Braun, J. A.
Rogers, and P. Wiltzius, Dual exposure, two-photon, conformal phase mask
lithography for three dimensional silicon inverse woodpile photonic crystals,
J. Vac. Sci. Technol. B 28, 783 (2010). — p.15.
[55] J. M. van den Broek, L. A. Woldering, R. W. Tjerkstra, F. B. Segerink, I. D.
Setija, and W. L. Vos, Inverse-woodpile photonic band gap crystals with
a cubic diamond-like structure made from single-crystalline silicon, Adv.
Funct. Mater. 22, 25 (2012). — p.15, 17.
[56] L. A. Woldering, Fabrication of photonic crystals and nanocavities, PhD
Thesis, University of Twente, 2008. — p.15.
[57] N. W. Ashcroft and N. D. Mermin, Solid state physics (Holt, Rinehart and
Winston, New York, 1976). — p.15, 18, 20.
[58] K. M. Ho, C. T. Chan, and C. M. Soukoulis, Existence of a photonic gap in
periodic dielectric structures, Phys. Rev. Lett. 65, 3152 (1990). — p.15,
16.
[59] M. M. Sigalas, C. M. Soukoulis, C. T. Chan, R. Biswas, and K. M. Ho, Effect
of disorder on photonic band gaps, Phys. Rev. B 59, 12767 (1999). — p.15,
19.
[60] R. Hillebrand and W. Hergert, Scaling properties of a tetragonal photonic
crystal design having a large complete bandgap, Phot. Nano. Fund. Appl. 2,
Bibliography 25
33 (2004). — p.15.
[61] L. A. Woldering, A. P. Mosk, R. W. Tjerkstra, and W. L. Vos, The influence
of fabrication deviations on the photonic band gap of three-dimensional in-
verse woodpile nanostructures, J. Appl. Phys. 105, 093108 (2009). — p.15,
17.
[62] M. Maldovan and E. L. Thomas, Diamond-structured photonic crystals, Nat.
Mater. 3, 593 (2004). — p.16.
[63] E. Yablonovitch, T. J. Gmitter, and K. M. Leung, Photonic band stucture:
the face-centered-cubic case employing nonspherical atoms, Phys. Rev. Lett.
67, 2295 (1991). — p.16.
[64] C. C. Cheng, V. Arbet-Engels, A. Scherer, and E. Yablonovitch, Nanofab-
ricated three dimensional photonic crystals operating at optical wavelengths,
Phys. Scr. T68, 17 (1996). — p.16.
[65] D.-X. Ye, Z.-P. Yang, A. S. Chang, J. Bur, S. Y. Lin, T.-M. Lu, R. Z. Wang,
and S. John, Experimental realization of a well-controlled 3d silicon spiral
photonic crystal, J. Phys. D 40, 2624 (2007). — p.16.
[66] J. W. Galusha, M. R. Jorgensen, and M. H. Bartl, Diamond-structured ti-
tania photonic-bandgap crystals from biological templates, Adv. Mater. 22,
107 (2010). — p.17.
[67] A. Chutinan and S. Noda, Effects of structural fluctuations on the photonic
bandgap during fabrication of a photonic crystal, J. Opt. Soc. Am. B 16, 240
(1999). — p.17.
[68] J. Schilling and A. Scherer, Photonics nanostruct, Fundam. Appl 3, 90
(2005). — p.17.
[69] V. Bykov, Spontaneous emission in a periodic structure, Sov. Phys. JETP
35, 269 (1972). — p.17, 18.
[70] S. John, Strong localization of photons in certain disordered dielectric super-
lattices, Phys. Rev. Lett. 58, 2486 (1987). — p.17, 18.
[71] C. M. Soukoulis, Photonic band gap materials : The semiconductors of the
future?., Physica Scripta T66, 146 (1996). — p.17.
[72] S. Haroche, in Cavity quantum electrodynamics, pages 767–940 in: Funda-
mental systems in quantum optics, edited by J. Dalibard, J. M. Raimond,
and J. Zinn-Justin (North Holland, Amsterdam, 1992). — p.17.
[73] S. John and J. Wang, Quantum electrodynamics near a photonic band gap:
Photon bound states and dressed atoms, Phys. Rev. Lett. 64, 2418 (1990).
— p.17.
[74] J. G. Fleming, S. Y. Lin, I. El-Kady, R. Biswas, and K. M. Ho, All-metallic
three-dimensional photonic crystals with a large infrared bandgap, Nature
417, 52 (2002). — p.17.
[75] J. Fleming, Addendum: Three-dimensional photonic-crystal emitter for ther-
mal photovoltaic power generation, Appl. Phys. Lett. 86, 249902 (2005).
— p.17.
[76] A. Babuty, K. Joulain, P. O. Chapuis, J. J. Greffet, and Y. De Wilde,
Blackbody spectrum revisited in the near field, Phys. Rev. Lett. 110, 146103
(2013). — p.17.
[77] G. Kurizki and A. Genack, Suppression of molecular interactions in periodic
26 Bibliography
dielectric structures, Phys. Rev. Lett. 61, 2269 (1988). — p.17.

[78] M. I. Antonoyiannakis and J. B. Pendry, Electromagnetic forces in photonic
crystals, Phys. Rev. B 60, 2363 (1999). — p.17.
[79] B. C. Buchler, T. Kalkbrenner, C. Hettich, and V. Sandoghdar, Measuring
the quantum efficiency of the optical emission of single radiating dipoles
using a scanning mirror, Phys. Rev. Lett. 95, 063003 (2005). — p.17.
[80] C. Blum, N. Zijlstra, A. Lagendijk, M. Wubs, A. P. Mosk, V. Subramaniam,
and W. L. Vos, Nanophotonic control of the förster resonance energy transfer
efficiency, Phys. Rev. Lett. 109, 203601 (2012). — p.17.
[81] E. Yablonovitch, T. J. Gmitter, R. D. Meade, A. M. Rappe, K. D. Brom-
mer, and J. D. Joannopoulos, Donor and acceptor modes in photonic band
structure, Phys. Rev. Lett. 67, 3380 (1991). — p.18.
[82] W. H. Zurek, Decoherence and the transition from quantum to classical,
Phys. Today 44, 36 (1991). — p.18.
[83] S. John and K. Busch, Photonic bandgap formation and tunability in certain
self-organizing systems, J. Lightwave Technol. 17, 1931 (1999). — p.18.
[84] D. M. Whittaker, Inhibited emission in photonic woodpile lattices, Opt. Lett.
25, 779 (2000). — p.18.
[85] C. Hermann and O. Hess, Modified spontaneous-emission rate in an inverted-
opal structure with complete photonic bandgap, J. Opt. Soc. Am. B 19, 18
(2002). — p.18.
[86] S. Kole, New methods for the numerical solution of maxwell’s equations,
PhD Thesis, Rijksuniversiteit Groningen, 2003. — p.18.
[87] S. R. Huisman, G. Ctistis, S. Stobbe, A. P. Mosk, J. L. Herek, A. Lagendijk,
P. Lodahl, W. L. Vos, and P. W. H. Pinkse, Measurement of a band-edge
tail in the density of states of a photonic-crystal waveguide, Phys. Rev. B
86, 155154 (2012). — p.19.
[88] P. Anger, P. Bharadwaj, and L. Novotny, Enhancement and quenching of
single-molecule fluorescence, Phys. Rev. Lett. 96, 113002 (2006). — p.19.
[89] Q. Wang, S. Stobbe, and P. Lodahl, Mapping the local density of optical
states of a photonic crystal with single quantum dots, Phys. Rev. Lett. 107,
167404 (2011). — p.19.
[90] M. D. Leistikow, J. Johansen, A. J. Kettelarij, P. Lodahl, and W. L. Vos,
Size-dependent oscillator strength and quantum efficiency of CdSe quantum
dots controlled via the local density of states, Phys. Rev. B 79, 045301 (2009).
— p.19.
[91] T. Coenen, Angle-resolved cathodoluminescence nanoscopy, PhD Thesis,
Amsterdam University, 2014. — p.19.
CHAPTER 2
Samples and instrumentation
2.1 Introduction
We give a short introduction on silicon inverse woodpile photonic crystals, that
have been used for controlling light propagation in this thesis. To avoid light
absorption by silicon we use PbS quantum dots that emit in the near infrared as
light sources in photonic crystals. Before infiltrating quantum dots in the complex
environment of photonic crystals we characterize their behavior in suspension.
We measure emission and absorption spectra, as well as time-resolved emission
measured by time-correlated single-photon counting method. The experiments
performed in this thesis require a number of special requirements and therefore
we needed to have a custom-designed experimental setup. The main purpose of
this setup is to investigate the control of spontaneous emission using photonic
crystals. The setup is designed to enable not only emission measurements but also
reflection, and transmission measurements on silicon inverse woodpile photonic
crystals in the near infrared range.
2.2 Silicon inverse woodpile photonic crystals

As discussed in chapter.1 our photonic crystals are designed so that they can
manipulate and control light propagation in a certain range of wavelengths. In
this thesis we have mainly studied silicon inverse woodpile photonic crystals be-
cause of their broad and complete photonic band gap for all directions and both
polarizations. This is obtained by the extent of the crystal in all three dimen-
sions. A cubic inverse woodpile photonic crystal consists of two perpendicular,
geometrically identical sets of cylindrical pores, as first proposed by Ho et al. [1].
The combination of the two sets of pores results in a three-dimensional structure
with a face-centered-cubic lattice and a diamond-like symmetry. By optimizing
the crystal parameters such as pore radius and lattice parameters, it is possible
to realize a full 3D band gap (see Fig. 2.1) with a broad relative width of more
than Δω/ω = 25% [2, 3].
Figure 2.1(a) shows a scanning electron micrograph of a silicon 3D inverse
woodpile photonic crystal fabricated in our group using CMOS compatible meth-
ods [5, 6]. The extent of the 3D crystal is indicated by the red dashed lines and
the 3D crystal is surrounded by the parent 2D crystal. Figure 2.1(b) illustrates
the designed structure of a 3D inverse woodpile crystal with lattice parameters a
28 Samples and instrumentation
Figure 2.1: (a) Scanning electron micrograph of a 3D inverse woodpile photonic crystal
made from silicon. The 3D crystal is delimited by the dashed lines and consists of two
perpendicularly etched sets of carefully aligned pores, surrounded by a 2D crystal.
(b) Schematic image showing the positions and overlap of the pores, and unit cell
parameters a and c. (c) Band structure of an inverse woodpile photonic crystal with
pore radius 170 nm, ε = 12.1 for silicon and ε = 2.25 for toluene-filled pores. The
band gap is indicated with the red bar. Inset: first Brillouin zone. (d) Density of states
(DOS) per volume for the same crystal calculated with 10000 k-points [4]. The DOS
vanishes in the band gap. Dashed curve: quadratic behavior in the low frequency limit.
and c. The SEM image shows that the fabricated crystals compare well to the de-
signed structures.
√ The cubic crystals have lattice parameters a = 693 and c = 488
nm (a/c = 2), and a range of pore radii (136 < r < 186 nm) to tune the band
gap to different frequencies. The 3D crystals extend over L3 = 12a × 12c × 12c,
which exceeds the Bragg attenuation length (see [7, 8]) in every direction [9]. The
good optical quality and high photonic interaction strength (see [10]) of our crys-
tals were confirmed by optical reflectivity experiments where intense and broad
peaks were observed. The stop bands overlap for all probed directions and po-
larizations, which consists the first ever experimental signature of a 3D photonic
band gap [9].
Fig. 1(c) shows the calculated band structure [4] for an inverse woodpile crys-
tal. A photonic band gap appears in the frequency range where modes are for-
bidden for all wave vectors and polarizations. Fig. 1(d) shows the corresponding
density of states (DOS). In Figs. Fig. 2.1(c) and (d), the band gap has somewhat
narrowed since we consider here a reduced dielectric constant between silicon and
toluene, to account for the liquid medium of embedded colloidal quantum dots.
Optical properties of PbS quantum dots 29
At low frequency the DOS increases quadratically similar to that of a medium

with an effective index of refraction. The DOS deviates from the parabola beyond
reduced frequency 0.35, and vanishes between 0.478 and 0.504 in the photonic
band gap. Considering the lattice parameter a = 693 nm, the band gap is cen-
tered around 1400 nm which overlaps with the telecommunication range and
avoids the silicon absorption range.
Beside optical characterization of these structures by means of reflectivity ex-
periments [9] our group was the first to control spontaneous emission of quantum
emitters in a 3D photonic band gap crystal [11]. Since this early study was
limited to a few selected frequencies, in chapter 4 of this thesis we will present
the first systematic measurements of controlling spontaneous emission in a finite
3D inverse woodpile photonic crystal. In chapters 6 and 7 we investigate light
scattering in 2D silicon photonic crystals.
2.3 Optical properties of PbS quantum dots

In order to probe the density of states inside inverse woodpile photonic crystals we
need to infiltrate quantum emitters into the crystals. Since it is fairly challenging
to embed efficient emitters in silicon, we have decided to infiltrate emitters only
in the low-index medium, which is fortunately the majority of a photonic crystal.
Because silicon absorbs light with a photon energy larger than its electronic
band gap of about 1.12 eV, corresponding to wavelengths shorter than 1100 nm,
it is not possible to use visible-light emitting quantum dots or dye molecules,
see e.g. [12]. Therefore we have used PbS quantum dots suspended in toluene,
purchased from Evident (PbS-1500), that emit in the near infrared wavelength
range. For the measurements the quantum dots concentration is reduced to
2 × 10−6 Mol/l to avoid reabsorption of the emitted photons by closely packed
quantum dots. From TEM microscopy measurements quantum dots have an
average diameter 4.1 nm with standard deviation of 0.5 nm [13].
The absorbance spectrum of quantum dots ensemble has been measured as
shown in Fig. 2.2. To measure the absorbance, a beam from a supercontinuum
white light source (Fianium) was sent through the suspension of quantum dots
with a concentration of 2.10−6 Mol/l. The transmitted light is analyzed using
a Fourier transform infrared spectroscopy (BioRad FTS-6000) in combination
with a pre-amplified InAs photodiode. This setup is used to measure absorption
between 4000 and 12000 cm−1 corresponding to wavelengths between 830 to 2500
nm. Several faint peaks are seen in the absorbance spectra which correspond to
different energy transitions in the quantum dots as indicated in the figure. Every
single quantum dot has a set of discrete levels. The energy states of quantum dots
are labeled as 1S, 1P , 1D, and so on, similar to atoms and molecules [14, 15].
Similarly, 1Se is the first electron state and 1Sh the first hole state, and so forth.
The 1Se − 1Sh is the first transition, since it requires the minimum energy. The
first seven transitions of PbS quantum dots has been extensively investigated in
Ref. [16].
Figure 2.3(a) shows the emission spectrum measured on an ensemble of the
3 1
Emission Intensity (104 C.P.S)

1Se-1Sh Emission
Absorbance
1Pe-1Ph
2
Absorbance
1Sh,e-1Pe,h
0.5
0 0
6000 8000 10000
-1
Frequency(cm )
Figure 2.2: Emission (black) and absorbance (dashed gray) spectra of PbS quantum
dots suspended in toluene. The emission spectrum is slightly red-shifted with respect to
the absorption spectrum corresponding to the Stokes shift.The 1Se − 1Sh is indicating
the first transition, 1Sh,e − 1Pe,h the second and 1Pe − 1Ph the third transition observed
in the absorbance measurements.
PbS quantum dots. It is seen that the spectrum is very broad covering the range
between 1300 nm to 1600 nm in the near infrared region, with a peak at 1470
nm (0.84eV ) with FWHM of 0.086 eV. The broad spectrum of the quantum dots
is mostly caused by size polydispersity. The energy gap of the quantum dots is
tuned by changing the quantum dot size. Therefore a size distribution of quantum
dots leads to a distribution which manifest itself as inhomogeneous broadening in
the optical spectra [17]. In a preliminary experiment by Leistikow, the inhomo-
geneous line broadening of PbS quantum dots by fluorescence line narrowing was
measured, suggesting that about half of the linewidth of approximately 70 nm
is caused by homogeneous broadening [13]. Intriguingly, our results of frequency
dependent decay rate in the band gap in chapter 4 provide evidence for a consid-
erably narrower homogeneous line width: we observe large changes of the decay
rate within the band gap at steps of 20 nm suggesting that the homogeneous
line broadening is even smaller than 20 nm. Since in our measurements data is
collected more systematically and because of sufficient number of data points we
think that our results are more reliable than the previous results.
The broad, inhomogeneously broadened, emission spectrum of the PbS quan-
tum dots overlaps very well with the photonic band gap of some of our crystals.
This makes it possible to investigate the control of spontaneous emission through-
out the band gap (Ch. 4).
In Fig. 2.2 the emission spectrum is plotted together with the absorbance spec-
trum. As expected the emission spectrum is slightly red-shifted with respect to
the main peak of the absorbance spectrum. The red shift of the emission spectra
with respect to absorption spectra is known as Stokes shift, which is commonly
observed in any florescent molecule and semiconductor quantum dots [18–21].
Optical properties of PbS quantum dots 31

Figure 2.3: (a) Emission spectrum of PbS quantum dot ensemble in toluene. The
steep edge at low energy is caused by the cut-off of the detector. The lack of data
points around 1 eV is caused by ”dead” pixels in the detector [22]. (b) The decay curve
of PbS quantum dot ensemble in toluene at λ=1475 nm wavelength. Time resolved
emission measurement shows a single exponential decay (black solid line) with a typical
decay rate of 1.7 μs−1 . Depending on the position of the focus in the suspension the
signal level changes and therefore such a time-resolved measurements can take between
one to ten minutes.
However, the underlying mechanism of Stokes shift is still controversial. In case

of semiconductor quantum dots, this shift decreases with the increase in radius of
the dot and disappears beyond a certain radius as reported in Ref. [22] for PbSe
quantum dots. The red shift occurs principally if either the top of the valence
band is an optically passive state such as a P state, or if the electron and the
hole are in a triplet state. Absorption of a photon from the top of the valence
band in such cases is not allowed and is possible only from an optically active
state lying deeper in the valence band. The exciton, once formed after absorp-
tion, can not decay to the top of the valence band by direct dipole transition and
hence is called “dark exciton” [18]. Deexcitation eventually takes place with the
help of phonons thus giving rise to photons that are red shifted compare to the
excitation, as in Fig 2.2 [19, 23, 24].
From Fermi’s golden rule applied to spontaneous emission the radiative decay
rate of a quantum emitter is modified by the local density of states [12, 25]. Since
the emitters in the photonic crystals experience a different LDOS than the one
in the suspension, their decay rates change. Hence measuring the decay rate of
the emitters in the photonic crystal is a method to probe the LDOS:
πd2 ω −1
Nrad (r, ed , ω) = ( ) γrad (r, ed , ω) (2.1)
0
where Nrad (r, ed , ω) is the projected LDOS and γrad (r, ed , ω) is the radiative
decay rate. For that reason the time-resolved decay rate of the ensemble of
quantum dots is extensively studied in this thesis. In time-resolved emission the
total decay rate is measured that consists of both radiative and non-radiative
decay rate γtot = γrad + γnrad . Since only the radiative decay rate is controlled
by the LDOS, from our measurements of maximum inhibition in the band gap
(discussed in Sec. 4.3) we found the non-radiative decay rate to be at most 5% of
the total decay rate, hence γnrad ≤ 0.085μs−1 . An additional consistency check
for the LDOS is obtained since the broad quantum dot spectrum allows us to
access the long-wavelength limit below the band gap, where the LDOS shows
typical quadratic behavior even in a strongly interacting photonic crystal [26].
To compare the LDOS in the crystal with the one in a homogeneous medium,
considerable care has to be taken to account for position dependent effects, see
notably [27]. This paper shows clearly that even in the long-wavelength limit,
the LDOS in the low-index medium in a photonic crystal does not necessarily
converge to the LDOS of the low-index medium. This forces to make the choice
to normalize the calculated LDOS to bulk toluene or to the low-frequency limit of
the DOS of the crystal∗ . In the calculations done in chapter 4 we have normalized
the DOS and the LDOS of photonic crystals to the bulk toluene LDOS.
Figure 2.3(b) shows the decay curve of PbS quantum dots suspended in toluene.
Quantum dots are excited by a 532 nm (2.33 eV) wavelength pulsed laser that is
at a photon energy much higher than the emission energy of quantum dots. By
measuring the time difference between the laser pulse generated by a diode in the
laser and the detection pulse of the PMT, we obtain a histogram of the photon
arrival times: a decay curve. This method is known as time-correlated single-
photon counting (TCSPC) [28]. The decay curve shown here is measured at the
peak wavelength of the emission spectrum at 1470 nm. These quantum dots have
a typical single exponential decay with a rate of 1.7 μs−1 in toluene. The decay
rate changes by changing the solvent as reported by different groups [29–31].
This is logical because the surrounding refractive index changes, which modifies
the local environment of quantum dots and leads to a different decay rate.
Since we are measuring in the near infrared region the time resolved emission
is collected on a huge background level of approximately 3.105 dark counts per
second [13, 22]. The background must be carefully considered before the time
resolved data can be properly interpreted, see Ch. 5. From the TCSPC method
to obtain a statistically reliable photon arrival histogram, the repetition rate of
the excitation laser must be at least 5 times lower than the decay rate that one
wants to measure [28]. This is because after 5 decay times the photon intensity
is about 100 times lower than the initial photon intensity, which is considered to
have a negligible contribution on the background. Therefore for the experiments
performed on photonic crystals in which a 18-fold inhibition occurs, we had to use
a low laser repetition rate of 80 kHz, which is the lowest possible repetition rate
of our laser. The 18 times inhibition corresponds to a life time of approximately
10 μs. The time between two laser pulses at 80 kHz repetition rate is 12.5 μs
which is not 5 times longer than the longest lifetime. Even though this laser
repetition rate is technically the longest available at the time of the experiment,
the consequence is that we probably statistically bias the photon arrival times
histogram to short arrival times and hence, to faster life times.
∗ To be sure, there are infinite long wavelength limit, depending on position, see Fig.7 in
Ref. [27]
Experimental setup 33
The number of signal counts is typically 1% of the repetition frequency, or

8.103 signal counts per second which is much lower than the background level.
To calculate the background level we use 200 bins, with bin size 0.128 ns, that
corresponds to about 25 ns before the laser pulse arrives. The statistical analysis
of time resolved emission measurements is discussed extensively in chapter 5.
2.4 Experimental setup

Based on the properties of the photonic band gap crystals and the PbS quantum
dots, we formulate five main requirements for our optical setup:
1. Capability of doing continuous-wave and time resolved measurements in
the near infrared range.
2. Collection and detection in a broad range of wavelengths in near infrared.
3. Possibility of measuring emission, reflection, and transmission in the same
setup.
4. Very high resolution microscope in the near infrared regime which resolves
a 5 μm photonic crystal.
5. Precise positioning and stability of photonic crystal samples to within 500
nm.
We have built a setup that satisfies all our requirements, as shown schematically
in Fig. 2.4. The setup is modification of a setup designed and built earlier by
Husken [22]; notably since no cryostat was necessary, the setup could be clearly
simplified thus leading to greater stability. A He-Ne laser is used for alignment
purposes as well as to excite continuous wave (CW) emission of the quantum
dots. A pulsed laser (Time-bandwidth, Cougar) with an emission wavelength
of 532 nm and tunable repetition rate is used to excite time-resolved emission
of quantum dots. This green 532 nm laser light has a higher photon energy
than the quantum dots emission energy where quantum dots absorb this energy.
This wavelength is, however, above the band gap of silicon and thus coincides
with the high absorption range of the silicon. Fortunately, since quantum dots are
within the pores in toluene the laser light that is guided through the pores excites
quantum dots. A supercontinuum white light source (Fianium SC-450-2) is also
placed in the setup for quantum dot line broadening measurements, reflectivity,
and broadband out of plane scattering measurements on photonic crystal samples
(chapter 7). For measuring emission of quantum dots in photonic crystals light
is focused onto the photonic crystal samples by the illumination objective o2
with a numerical aperture NA=0.12. For broadband reflectivity measurements
this objective is replaced with a reflective objective with a NA=0.28. For the
alignment the illumination objective is placed on a manual three dimensional
translational stage with 10 μm resolution.
In order to measure transmission, light is passed through the sample from its
backside, by reflecting from M 1 mirror. Light that emanates from the sample is
filters y
Grating
spectrometer
Array
z x
CCD1
PMT
laser
continuum
Super
o1 S M1
LED1
o2
Cougar
LED2
HeNe CCD2 B1
M2
Figure 2.4: Schematic representation of the experimental setup used for the measure-
ments in this thesis. The incident laser light is guided to the illumination objective
(o2 ) to be focused on the sample (S ). The emitted light is collected by a high numeri-
cal aperture objective o1 NA=0.72, then is dispersed by spectrometer. Time resolved
emission is detected with a photomultiplier tube (PMT).
collected by a high numerical aperture objective, NA=0.72 (o1,Nikon CFIL Plan

EPI SLWD). The collection objective is placed under an angle of 90◦ with respect
to the illumination objective to prevent direct reflection of excitation light into the
detector in emission experiments. The high NA objective has a relatively large
working distance (6 mm) to facilitate the alignment of photonic crystal samples,
even in a cuvette. This objective is mounted on three single piezo transducers
(PIHera, P-622.Z series) enabling a lateral degree of freedom of 100 μm and 60
μm in the z-direction with a resolution of 0.1 nm and a position accuracy of
0.02 nm [22]. A digital piezo controller (PI-E710) in combination with home-
written LabView codes is used to control the position of the collecting objective.
The collected light passes through several 850 nm and 1100 nm long pass filters
to filter stray laser light in emission experiments. Light is then dispersed by a
grating spectrometer(Princeton Instruments, Spectra Pro 2558) with a grating
of 85 lines/mm and is imaged into a liquid nitrogen cooled InGaAs diode array
(OMA-V: 1024-2.2 LN). The array is sensitive from 1000 to 1650 nm suitable
to measure emission spectra. The combination of spectrometer slit size, grating
steps and the array’s resolution give rise to a step size of 0.08 nm in the spectrum.
Time-resolved emission is detected with a Peltier-cooled Hamamatsu NIR pho-
tomultiplier tube (PMT). The selected emission photon energy band width that
is detected by the PMT is set by a slit width, and is chosen to be approximately
3 meV. The PMT is connected to a Picoquant PicoHarp 300 timing card. Thus
Experimental setup 35
it is possible to measure the decay curve of emitters at particular emission fre-

quencies by the time correlated single photon counting method (TCSPC) [28, 32].
The time resolution is set by the bin size and is 0.128 ns. The PicoHarp 300 is
perfectly matched to this detector and the overall Instrument Response Function
(IRF) is as short as 40 ps FWHM. Since the quantum dots life time is much
longer, the IRF is insignificant.
Alignment of the incident light and the collected light with respect to the sam-
ple is realized by implementing two charge couple cameras (CCD 1 and 2) in
combination with two single white light emitting diodes (LED 1 and 2). The
LEDs are equipped with plano-convex lenses to illuminate the entire aperture
of the microscope objectives. Optical images of the surfaces of the sample are
collected directing the reflected light of the LEDs to the CCD cameras via pellicle
beam splitters and plano-convex lenses. This enables to investigate the position
of the incident light on the sample via the illuminating objective, as well as to put
the sample in focus with respect to both the illuminating and collecting micro-
scope objectives. The setup scheme is similar to the setup built by Husken [22],
except that here we have two separate imaging systems for the illumination path
and detection path, for convenient alignment. In this setup we have an additional
light path for measuring reflectivity from both sides of the crystal. The first path
is from the illumination side. Light illuminates the sample via the illuminating
objective o2 and reflects back from the same objective, after reflecting from beam
splitter B1 is sent to the detector. The second path is via the detection objec-
tive, where light that is reflected from the M2 mirror illuminated the sample and
reflects back and is sent to the detector.
(a) (b)
2D Dirt
crystal
laser light
3D
crystal
sample 2D silicon
holder crystal wafer
Figure 2.5: CCD camera image of a 3D photonic crystal (internal name: Ad3b-
14112008) in a cuvette filled with toluene in the near infrared setup with white light
illumination. (a) The 3D photonic crystal is indicated with a dashed red rectangle. The
3D crystal is surrounded with 2D crsyals in the vertical direction along the dashed gray
bar. The pitch of the pores is 0.69 ± 0.02 μm. The dark spot on the silicon wafer is
dirt that has stuck to the wafer during sample fabrication. The image shows that the
spatial resolution is sufficient to image 3D photonic crystals. (b) Result of a successful
alignment procedure on a 3D photonic crystal. The laser light is focused in the middle
of the 3D crystal. Since the crystal is in a cuvette, the resolution is slightly worse than
what is expected.
36 Bibliography
Figure 2.5 shows a white light image of the side view of a sample that contains
both 2D and 3D crystals in the setup. The gray bar along the dashed line is
the 2D crystal that surrounds the 3D crystal. The pores of the 2D crystal are
along the horizontal direction. The dark gray rectangle in the middle is the 3D
photonic crystal. Figure 2.5(b) is a white light image of the same sample during
the alignment procedure on the 3D photonic crystal. The laser light is focused
in the middle of the 3D crystal. These images show that the spatial resolution
of the setup is high enough to resolve a 3D crystal with a lateral length of 5 μm,
which therefore meets the requirement for successful studies.
2.5 Summary
In this chapter we describe the infrastructure to study inhibition of spontaneous
emission in 3D photonic band gap silicon inverse woodpile photonic crystals.
These crystals have a 3D extent in all three dimensions by etching perpendicular
sets of pores in two directions. The crystals have the widest ever photonic band
gap in the near infrared, confirmed with reflectivity and emission measurements.
In order to probe the density of states in the band gap of the photonic crystals we
infiltrate PbS quantum dots in the crystals pores. The broad emission spectrum
of the PbS quantum dots overlaps very well with the photonic band gap of our
crystals.
We describe the near infrared emission setup that is used for most of the experi-
ments in this thesis. This setup includes optical paths for emission measurements,
transmission, and reflection measurements in the near infrared range. The ab-
sorbance spectrum of quantum dots is measured by exciting quantum dots with
a white light source, showing different transitions in quantum dots energy levels.
Comparing the emission and the absorbance spectra we observe a Stokes shift.
The decay rate of quantum dots shows a decay rate of 1.7 μs−1 at the peak of
emission spectrum, in agreement with the decay rate reported from other studies
in literature.
Bibliography
State Commun. 89, 413 (1994). — p.27.
[2] R. Hillebrand, S. Senz, W. Hergert, and U. Gösele, Macroporous-silicon-
based three-dimensional photonic crystal with a large complete band gap,
Journal of applied physics 94, 2758 (2003). — p.27.
[3] L. A. Woldering, A. P. Mosk, R. W. Tjerkstra, and W. L. Vos, The influence
of fabrication deviations on the photonic band gap of three-dimensional in-
verse woodpile nanostructures, J. Appl. Phys. 105, 093108 (2009). — p.27.
[4] S. G. Johnson and J. D. Joannopoulos, Block-iterative frequency-domain
methods for maxwells equations in a planewave basis, Opt. Express 8, 173
(2001). — p.28.
Bibliography 37

Funct. Mater. 22, 25 (2012). — p.27.
[6] R. W. Tjerkstra, L. A. Woldering, J. M. van den Broek, F. Roozeboom, I. D.
Setija, and W. L. Vos, Method to pattern etch masks in two inclined planes
for three-dimensional nano- and microfabrication, J. Vac. Sci. Technol. B
29, 061604 (2011). — p.27.
[7] S. Y. Lin and G. Arjavalingam, Tunneling of electromagnetic waves in two-
dimensional photonic crystals, Opt. Lett. 18, 1666 (1993). — p.28.
[8] J. Huang, N. Eradat, M. E. Raikh, Z. V. Vardeny, A. A. Zakhidov, and R. H.
Baughman, Anomalous coherent backscattering of light from opal photonic
crystals, Phys. Rev. Lett. 86, 4815 (2001). — p.28.
[9] S. R. Huisman, R. Nair, L. A. Woldering, M. D. Leistikow, A. P. Mosk, and
W. L. Vos, Signature of a three-dimensional photonic band gap observed on
silicon inverse woodpile photonic crystals, Phys. Rev. B 83, 205313 (2011).
— p.28, 29.
[10] W. L. Vos, R. Sprik, A. van Blaaderen, A. Imhof, A. Lagendijk, and G. H.
colloidal crystals, Phys. Rev. B 53, 16231 (1996). — p.28.
[11] M. D. Leistikow, A. P. Mosk, E. Yeganegi, S. R. Huisman, A. Lagendijk,
and W. L. Vos, Inhibited spontaneous emission of quantum dots observed in
a 3D photonic band gap, Phys. Rev. Lett. 107, 193903 (2011). — p.29.
Press, New York, 2006). — p.29, 31.
[13] M. D. Leistikow, Controlling spontaneous emission with nanostructures,
PhD Thesis, University of Twente, 2010. — p.29, 30, 32.
[14] A. L. Efros and A. L. Efros, Interband absorption of light in a semiconductor
sphere, Sov. Phys. Semicond 16, 772 (1982). — p.29.
[15] A. I. Ekimov and A. A. Onushchenko, Size quantization of the electron en-
ergy spectrum in a microscopic semiconductor crystal, JEPT Lett 40, 1136
(1984). — p.29.
[16] L. Cademartiri, E. Montanari, G. Calestani, A. Migliori, A. Guagliardi, and
G. A. Ozin, Size-dependent extinction coefficients of PbS quantum dots., J.
Am. Chem. Soc 128, 10337 (2006). — p.29.
[17] V. I. Klimov, Nanocrystal quantum dots (CRC Press Boca Raton, FL, 2010).
— p.30.
[18] M. Nirmal, D. J. Norris, M. Kuno, M. G. Bawendi, A. L. Efros, and M.
Rosen, Observation of the “dark exciton” in cdse quantum dots, Phys. Rev.
Lett. 75, 3728 (1995). — p.30, 31.
[19] A. L. Efros, M. Rosen, M. Kuno, M. Nirmal, D. J. Norris, and M. Bawendi,
Band-edge exciton in quantum dots of semiconductors with a degenerate va-
lence band: Dark and bright exciton states, Phys. Rev. B 54, 4843 (1996).
— p.30, 31.
[20] M. Paillard, X. Marie, P. Renucci, T. Amand, A. Jbeli, and J. M. Gérard,
Spin relaxation quenching in semiconductor quantum dots, Physical review
38 Bibliography
letters 86, 1634 (2001). — p.30.

[21] A. Tackeuchi, R. Ohtsubo, K. Yamaguchi, M. Murayama, T. Kitamura, T.
Kuroda, and T. Takagahara, Spin relaxation dynamics in highly uniform
inas quantum dots, Applied physics letters 84, 3576 (2004). — p.30.
[22] B. H. Husken, Spontaneous emission of near-infrared quantum dots con-
trolled with photonic crystals, PhD Thesis, University of Twente, 2009. —
p.31, 32, 33, 34, 35.
[23] J. Li and J.-B. Xia, Exciton states and optical spectra in cdse nanocrystallite
quantum dots, Phys. Rev. B 61, 15880 (2000). — p.31.
[24] A. Bagga, P. K. Chattopadhyay, and S. Ghosh, Origin of stokes shift in inas
and cdse quantum dots: Exchange splitting of excitonic states, Phys. Rev. B
74, 035341 (2006). — p.31.
[26] A. F. Koenderink, L. Bechger, A. Lagendijk, and W. L. Vos, An experimental
study of strongly modified emission in inverse opal photonic crystals, Phys.
Stat. Sol. B 197, 648 (2003). — p.32.
[27] I. S. Nikolaev, W. L. Vos, and A. F. Koenderink, Accurate calculation of the
local density of optical states in inverse-opal photonic crystals, J. Opt. Soc.
Am. B 26, 987 (2009). — p.32.
[28] D. V. O’Connor, W. R. Ware, and J. C. Andre, Deconvolution of fluorescence
decay curves. a critical comparison of techniques, J. Phys.l Chem. 83, 1333
(1979). — p.32, 35.
[29] S. W. Clark, J. M. Harbold, and F. W. Wise, Resonant energy transfer in
pbs quantum dots, The Journal of Physical Chemistry C 111, 7302 (2007).
— p.32.
[30] M. S. Gaponenko, A. A. Lutich, N. A. Tolstik, A. A. Onushchenko, A. M.
Malyarevich, E. P. Petrov, and K. V. Yumashev, Temperature-dependent
photoluminescence of pbs quantum dots in glass: Evidence of exciton state
splitting and carrier trapping, Phys. Rev. B 82, 125320 (2010). — p.32.
[31] Y. Chen, D. Yu, B. Li, X. Chen, Y. Dong, and M. Zhang, Size-dependent ra-
diative emission of pbs quantum dots embedded in nafion membrane, Applied
Physics B 95, 173 (2009). — p.32.
[32] D. V. O’Connor and D. Philips, Time correlated single photon counting
(Academic Press, 1984). — p.35.
CHAPTER 3
Local density of states in the band gap of a finite
one-dimensional photonic crystal
3.1 Introduction
Recently, there have been great attempts to make 3D photonic band gap crystals
which show a 3D photonic band gap [1–11]. In practice a real photonic crystal is
of course always finite. Therefore the DOS, that is an average over all of space,
can not describe the optical states in a finite system. For an inhomogeneous
system, including finite crystals we need to take into account local properties
and therefore we need to calculate the local density of states (LDOS). Having
a finite crystal fundamentally changes the main features of the LDOS. First
of all the LDOS that is zero in the band gap of an infinite crystal [12, 13],
becomes non-zero in the band gap of a finite crystal [14, 15]. Moreover, within the
bandgap of an infinite crystal the LDOS does not depend on frequency or position;
in stark contrast in the bandgap of a finite crystal the LDOS does depend on
frequency, position, and the crystal size. To date, most theories assume crystals
of infinite extent (see Refs. [12, 13, 16, 17] and references therein), and can thus
not be used to interpret experimental results on real crystals as first studied in
Ref. [14]. To the best of our knowledge the LDOS in the band gap of a finite
3D photonic crystal has not been studied yet as a function of the crystal size
and the position in the crystal and there is no analytic theory to calculate the
LDOS for any class of crystal. Several groups have calculated optical properties
of finite bandgap crystals: Interesting work has been done to calculate the DOS
(but not the LDOS) for 1D-waves in the band gap of a finite 1D crystal by
Bendickson et al. [18]. The LDOS for 2D waves has been calculated in the band
gap of a 2D photonic crystals by Asatryan et al. as a function of position, crystal
size, and frequency [19]. The 3D LDOS has been calculated for 1D and 2D
crystals that do not have a 3D gap [15, 20, 21]. For 3D light in a 3D inverse
opal, Hermann and Hess have calculated the LDOS as a function of frequency,
position, and for several position in the crystal by means of finite difference time
domain simulations [22]. Unfortunately, however, the results can not be applied
to other crystals, physical insights are not readily apparent, as is intrinsic to
This chapter has been published as: E. Yeganegi, A. Lagendijk, A.P. Mosk, and W. L. Vos,
Phys. Rev. B 89, 045123: 1-10 (2014)
40 Local density of states in the band gap of a finite one-dimensional photonic crystal
simulations, and finally such calculations requires a lot of programming efforts

and computational costs.

Figure 3.1: (color) Schematic drawing of a light source (orange sphere) embedded in
a finite (1D) photonic band gap crystal with the size L (photonic crystal slab). The
source is located at a depth z = Δz. The interface between the crystal and free space
is indicated by the vertical dashed line. The shading in the crystal lightens away from
the interface to illustrate the decreasing LDOS with depth into the crystal Δz.
In this chapter we introduce a new point of view on the band gap in a finite
photonic crystal, which allows for simplified calculations based on physical prin-
ciples. We consider a photonic band gap in a finite crystal to be the result of
vacuum fluctuations in free space that tunnel into the forbidden zone where the
fluctuations are exponentially damped. The physical situation is illustrated in
Fig. 3.1 where a light source (orange circle) is embedded in a finite photonic band
gap crystal. In this paper we will limit ourselves to a 1D structure. The reason is
that the electromagnetic field propagation and the band structure in 1D can be
calculated analytically, therefore we can test our model by comparing with the
analytic calculations. Ultimately our model is intended to be extended to calcu-
late the DOS inside the band gap of a finite 2 or 3-dimensional photonic crystal
as a function of crystal size, and frequency, which is the subject of chapter 4.
3.2 Local density of states in a finite crystal

The local density of states (LDOS) projected on the transition dipole moment
orientation in an infinite photonic crystal N (r, ω, ed ) is conveniently calculated
as a sum over all modes [23–25]:

1
N (ω, r, ed ) = dKδ(ω − ωn,K )|ed · En,K (r)|2 , (3.1)
(2π)3 n BZ
where ω is frequency, r is position in the crystal, and ed is the dipole moment

orientation. The integration over the real wave vector K is performed over the
first Brillouin zone, and n is the band index. En,K (r) is the mode function of
the (quantized) electromagnetic field. In a photonic crystal these mode functions
are Bloch modes equal to:
En,K (r) = Ēn,K (r)e−iK.r . (3.2)
Field propagation in a periodic media 41
where Ēn,K has the periodicity of the unit cell. The physical interpretation
of Eq. (3.1) is that the delta function δ(ω − ωn,K ) filters the relevant emission
frequency from the collection of photonic states. Since in the band gap of an
infinite crystal there are no propagating modes, the delta function causes the
projected LDOS to become zero in this frequency range. We use a vector notation
for K and r to enable the generalization to 2D and 3D.
To calculate the LDOS for a finite photonic crystal we modify the LDOS equa-
tion for an infinite photonic crystal. We develop a model for the LDOS in a finite
crystal which concerns the fields En,K (r) and the modes’ linewidths to generalize
the δ-function in Eq. 3.1, which are the two main factors in the LDOS. We pro-
pose that fields in the band gap can be calculated using our Ansatz as follows: in
the gap the fields are interpolated between the allowed Bloch modes at the band
edges that are frequency broadened by finite size and that are multiplied by a
frequency dependent exponentially decaying term that describes tunneling. This
model describes that at a depth z inside a finite photonic band gap crystal the
vacuum modes that enter the crystal in a certain direction K are exponentially
damped. The second factor is about the modes in the finite crystal. Modes in
a finite crystals are no more delta functions and become broadened due to the
finite size of the crystal. We propose to employ Lorentzian functions instead of
the delta functions, where the linewidth of the Lorentzian is set by the crystal
size L. These assumptions are used to calculate LDOS in the band gap of 1D
photonic crystal as a function of frequency, depth in the crystal, and the crystal
size.
In this chapter we describe in detail the above-mentioned viewpoint. In Sec. 3.3.1
we review analytic calculations of the field in a periodic medium. We start with
a crystal of infinite extent as is illustrated in Fig. 3.2(a) and we describe Bloch
modes as function of frequency and position in the crystal. We also describe the
band structure of an infinite 1D photonic crystal in an intuitive way. In Sec. 3.3.2
we discuss the field behavior in a finite crystal as a function of frequency, the
position and the crystal size. In Sec. 3.4 we describe our model for the LDOS in
a finite crystal. In Sec. 3.5 we check our model by calculating the LDOS in the
bandgap of a finite photonic crystal using the model and comparing that with
the analytically calculated LDOS in the bandgap of a finite 1D crystal. Finally,
in Sec. 3.6 we summarize our findings.
3.3 Field propagation in a periodic media

3.3.1 Infinitely extended photonic crystal
In this section we investigate the optical properties of an infinitely extended 1D
periodic structure as shown in Fig. 3.2(a). To calculate the band structure and
the electromagnetic fields we use the transfer matrix method. The details are
available in the appendix A.
In Fig. 3.3 we plot the band structure for a crystal with alternating dielectric
constants of 1 = 13 and 2 = 1 and layer thicknesses a1 = 0.2a and a2 = 0.8a.
We used 1 = 13 as an approximation relevant to commonly used semiconductors

Figure 3.2: Schematic drawing of a one-dimensional periodic structure. (a) Infinite

structure and (b) finite Bragg stack. One period, a, consists of two different media
with width “a1 ” and “a2 ” and dielectric permittivities of 1 and 2 respectively. The
crystal is taken to be one-dimensional and the other directions are used for illustration
purposes.
such as GaAs and Si. Within the bands K is real corresponding to the propagat-

ing Bloch modes. In the stopgap, however, K becomes complex K = K + iK
and therefore light is attenuated in the crystal. In this case the real part of

the wave vector is clamped at the Brillouin zone edge K = π/a [26]. Simulta-

neously the imaginary part strongly depends on frequency K = K (ω), as it
increases from zero to a maximum at the stop gap center back to zero at the
upper stopgap edge. Interestingly such a complex band structure has previously
been considered for both photonic and phononic crystal structures [27, 28].
Fig. 3.4 (a) shows the calculated absolute value of the field |En,K (z)| in the first
band at a reduced frequency ωa/2πc = 0.2. It is seen that the field maxima are
located in the material with the high dielectric permittivity. This general and
well-known behavior occurs at all frequencies in the first band up to the band
edge at ωa/2πc = 0.203 [26, 29] . If we assume light to be inside such an infinite
structure at frequencies in the band gap, the field is exponentially damped since
the incident light is reflected from each layer due to Bragg diffraction. Fig. 3.4(b)
and (c) show fields at two different frequencies in the band gap, where Fig. 3.4(b)
is for a frequency near the center of the band gap and Fig. 3.4(c) corresponds
to a frequency close to the band edge where the imaginary part K of the wave
vector is much smaller than at the center of the gap. Fig. 3.4 (d) shows the field

!
!

% # $! $
" ## $! $

Figure 3.3: Calculated exact band structure of the one-dimensional periodic structure
depicted at Fig. 3.2, where a1 = 0.2a and a2 = 0.8a and 1 and 2 are 13 and 1
respectively. The bands are shown with dark blue lines where the wave vector is real
(left ordinate). The first two photonic bandgaps are shown with yellow bars. In the

band gaps the wave vector becomes complex K = K + iK , where the real part is

clamped at the Brillouin zone edge wave vector (K = π/a) and the imaginary part K
strongly depends on the frequency as is shown by red dashed line (right ordinate).
at a frequency in the second band where field propagates in the structure and the
field maxima are located in the material with low , in agreement with common
lore for “air bands” [29, 30].
Now we turn to the field behavior as a function of frequency and position in
the periodic structure. To build the model for the LDOS in a finite crystal we
consider the main maximum of the field amplitude for different frequencies inside
and outside the band gap. Fig. 3.5(a) shows the position of the field maximum
within a unit cell as a function of normalized frequency. The ordinate between
0 and 0.2 is the high index material and from 0.2 to 1 is the low index material.
The abscissa shows the frequency range, which includes the first and second band
and the first band gap which is indicated with the yellow bar. In the first band
the field maximum is in the middle of the high index material. In the second
band the field maximum is in the middle of the low index material. This field
behavior for propagating bands is well-known and corresponds to the dielectric
band and air band nomenclature that pertains to a simple photonic crystal, see
Ref [29]. In the band gap, however, the field maximum moves continuously from
the middle of the high index layer to the middle of the low index material. It
is seen in Fig. 3.5(a) that this trend through the unit cell is a complex function
of frequency. From separate calculations we find that the path is a complicated
function of a1 , a2 , 1 and 2 . In the band gap, the field amplitude |En,K (z)| does
not fulfill the symmetry of the crystal any more – here in 1D mirror symmetry
at the center of either layer 1 or layer 2 – since the fields are no more bonafide
eigenfunctions of the crystal. The trajectory of the field maximum through the
unit cell is a phenomenon known in X-ray literature, where it is used in X-ray
fluorescence standing wave spectroscopy [31].
Fig. 3.5(b) shows the maximum field amplitude as a function of frequency.

Figure 3.4: Absolute value of the electric field associated with (a) reduced frequency
ωa/2πc = 0.2, (b) reduced frequency ωa/2πc = 0.3, (c) reduced frequency ωa/2πc =
0.45, and (d) reduced frequency ωa/2πc = 0.47 of the band structure plotted in the
Fig. 3.3. The main maximum of the field amplitude is indicated in Fig. 3.4 with a black
arrow.
In the first band the field amplitude maximum increases with frequency until it
reaches the lower edge of the band gap. While the trend of the field maximum
is continuous in moving from the first band to the band gap and then to the
second propagating bands, there is a cusp at each band edge. At the frequency
of the upper band edge the field amplitude is enhanced. The maximum of the
field amplitude in the second band is higher compared to the first band. This
enhancement can be understood from the conservation of field energy |En,K (z)2 |,
since the field maximum in the second band is centered in the low index material.
This detailed investigation of the field behavior in the band gap suggests a relation
between the fields in the gap and the fields at the band edges.
3.3.2 Finite photonic crystal

The advantage of a 1D calculation is that analytic calculations of the field and
of the LDOS for a finite crystal is possible. In this section we calculate the
field and the projected LDOS for a finite periodic structure that has the same
periodicity and dielectric materials as the infinite one. We first calculate the field
at different frequencies outside and inside the band gap and then using these fields
we calculate the projected LDOS inside the band gap of a finite structure. Later
on in Sec.3.5 we compare the LDOS calculated using our model with the exact
LDOS calculated using a transfer matrix method.
To calculate the field in a finite structure we also use a transfer matrix method
where the propagating fields are the total electric field and the total magnetic

!
!

"# $% &!
!

'

"# $% &!
Figure 3.5: (a)Behavior of the first field maximum in the frequency range of the first
bandgap. The ordinate between 0 and 0.2 is the high index material (indicated by the
orange hatched bar) and from 0.2 to 1 is the low index material (indicated by the gray
hatched bar). The red solid line is the exact behavior of the field and the black dashed
line is the linear field interpolation with the position. (b) Comparison of the exact
maximum field amplitude (red curve) and the model (black dashed curve) as a function
of reduced frequency. In the model a linear interpolation is used for the position of
the field maximum (see(a)). The model shows a very good agreement with the exact
calculations.
field. The main difference between the calculations in a finite and infinite struc-
ture is that in the infinite structure fields are Bloch modes and therefore we
have to apply an extra condition to imply field periodicity as a result of Bloch
conditions. This reduces the number of initial conditions for the wave that prop-
agates into the crystal. In case of the finite structure, the Bloch condition does
not pertain once we have the freedom to choose any arbitrary initial conditions.
T
If we take as an initial condition the incident wave 1 0 (see Appendix A),
T T
the wave on the left side will be 1 r and the wave on the right side t 0 ,
where r and t are the amplitude transmission and reflection coefficients. The
T
wave 1 r contains both the reflection that is a property of the crystal and the
incident wave that is not a property of the crystal. On the other hand, the right
T
side wave t 0 only contains a property of the crystal. Since we wish to obtain
T
properties of the crystal, we prefer to start with the wave t 0 and propagate
it backwards through the crystal. An advantage of this approach is that we
minimize rounding errors when Bragg reflection occurs. If one would start from
the left such errors appear especially when the structure becomes larger, leading
to a non-physical solution. From scattering theory [32, 33], this result can be
understood as follows: there are no bound states inside the crystal in the gap,
T
therefore, all scattering states outside the crystal such as t 0 form a complete
set that characterize the crystals properties.

Figure 3.6: (a) Absolute value of the electric field at a frequency ωa/2πc = 0.2 for a
finite structure with m = 10 periods. The environment of the crystal is assumed to
be air. (b) Absolute value of the electric field at a frequency ωa/2πc = 0.3 inside the
photonic band gap (see Fig. 3.3). At this frequency the field inside the structure is
damped. The field in the finite structure decays with the same damping as the field in
the infinite structure at the same frequency.
Fig. 3.6(a) shows the exact analytic calculations of the absolute field for a
finite crystal consists of m = 10 layers. The environment is assumed to be free
space. At a reduced frequency ωa/2πc = 0.2 outside the band gap, the field is
propagating through the crystal. As seen in the figure the field is not periodic
and clearly differs from the field in the infinite structure at the same frequency
(see Fig. 3.4) as a result of Fabry-Perot resonances due to the front and back
surfaces. Fig. 3.6(b) shows the electric field for the same structure at a reduced
frequency ωa/2πc = 0.3. This frequency is in the band gap and here the field is
damped. In the finite structure the field is damped with the same decay length
as the field in the infinite structure at the same frequency, as is expected.

To calculate the LDOS we have calculated the field entering the crystal from
both sides. Since the dipole orientation of the light source is in the direction of
the field propagation, the intensity of the light entering the crystal is directly
proportional to the projected LDOS. In Fig. 3.7 we show the calculated LDOS
in the band gap of the finite crystal. The crystal has a larger size of m = 20
unit cells and has the same dielectric constants and layer thicknesses as before.
We also calculate the LDOS using the Green function where we assume that the
emitter is at position (z) in a host layer [34, 35] . The Green function can be
solved self-consistently at any position in the structure. The imaginary part of
the Green function leads to the LDOS that is shown in Fig. 3.7 by green dot-
dashed line [34], which agrees well with the LDOS calculation using the field.
It is seen that in the band gap the LDOS decreases exponentially with depth z
in the crystal. At a frequency near the center of the bandgap as in Fig. 3.7(a)
the LDOS decays faster compared to a frequency closer to the band edge as in
Fig. 3.7(b), as is expected due to the smaller magnitude of the imaginary wave
vector K near the edge (see Fig.3.3). The LDOS behavior in the band gap will
be discussed in more detail in Sec. 3.5.

!

!

"

!

!

Figure 3.7: (color) The LDOS calculated using the Ansatz Eq. 3.5 (blue) and the exact
LDOS calculated for a finite periodic structure at two different reduced frequencies
ωa/2πc = 0.33 and ωa/2πc = 0.43 within the photonic bandgap (logarithmic axis). At
frequency ωa/2πc = 0.33 there is a small shift in the unitcell modulations of the LDOS.
At frequency ωa/2πc = 0.43 our model matches perfectly with the exact calculation.
This small deviation between our model (blue curve) and the exact calculations of the
LDOS using the field (red curve) and the Green function (green curve) is due to our
simple interpolation and can be already seen in 3.5(b) at the same frequency.
3.4 Model for the LDOS in a finite crystal

To calculate the LDOS in the band gap of a finite crystal we implement our new
point of view which consists of two main items. First, in a finite crystal the
propagating modes are not strictly Bloch modes as in an infinite crystal. If we
consider the Green function, it can be expanded in eigenfunctions En,K (r) of the
crystal as for an infinite crystal [36]:

←
→ c2 E∗n,K (r )En,K (r)
G (r, r ; ω) = lim 2 − (ω + i)2 ,
→0 ωK
K
where we sum over all wave vectors K in all bands. In the case of a finite
crystal the eigen functions become quasi modes that obey a bi-orthogonality
condition [37]. Therefore we can write the same summation for the Green function
in a finite system knowing that such quasi modes have a finite width. The width of
these quasi modes can be determined by considering constant-flux modes [38]. A
simple yet insightful approximation, inspired by the scaling theory of localization,
is to determine the width by the time required for the wave to propagate out of
the crystal [39]. If we consider the eigenfunctions E to have an amplitude of
order one, the summation contains Lorentzian functions, therefore, we propose
the finite crystal modes to be Lorentzians. F (ω − ωi,j ) describes how modes
broaden in frequency due to the finite size of the crystal
1 Δ
F (ω − ωi,j ) = , (3.3)
2 ) + (ω − ωi,j )
(2π) ( Δ 2 2
where Δ is the width of the Lorentzians. As a first step we take the width to
be the same for every mode. We take the width to be equal to the ratio of the
lattice parameter to the crystal size L [41]
a
Δ= . (3.4)
L
Due to the finite size of the crystal the modes become discrete [40] and we have
a summation of the Lorentzian modes. ωi,j is the central frequency of each mode
and therefore the central frequency of each Lorentzian function. m is the number
of wave vectors in the Brillouin zone that is equal to the number of the unit cells in
the finite crystal, which stems from the Born-Von Karman boundary condition in
solid state physics [40]. L is the crystal size that is equal to L = m·a. The number
of Lorentzian functions is thus set by the number of modes in the finite crystal
and is therefore equal to m for each band. The expression for the Lorentzian
functions F (ω − ωi,j ) leads to delta functions in the limit of an infinitely large
crystal (L −→ ∞), in agreement with Eq. 3.1.
The second item describes the waves in the band gap. To calculate the waves in
the band gap we use two features: the Bloch modes at the edges of the band gap,
and the imaginary part of the wave vector K . As we have seen in Sec. 3.3, in the
bands at frequencies below and above the gap the wave vectors are completely
real: K = K and K = 0 and the waves are propagating. In the bandgap the
Model for the LDOS in a finite crystal 49
wave vectors become complex K = K + iK thus the waves are exponentially
damped as a result of Bragg diffraction. Here we apply our second new point
of view to the waves in the band gap. We propose to write the waves in the
bandgap as a Bloch function with a complex wave vector as follows:
Epbg
K (r) = ĒK (r)e
iK·r

= ĒK (r)ei(K +iK )·r

= ĒK (r)eiK ·r e−K ·r

−K ·r
= Êint
K (r)e (3.5)
where the mode function Êint int

K (r) is periodic. In the gap the fields ÊK (r) are
interpolated between the allowed Bloch modes at the band edges. This model
describes that at a depth r inside a finite photonic band gap crystal the vacuum
modes that enter the crystal in a certain direction K are exponentially damped

by a factor e−K ·r , where K is the imaginary wave vector for propagation in the
direction with unit vector K and is considered such that the waves decay inside
the crystal.
The black dashed line in the Fig. 3.5 shows how we interpolate the field inside
the band gap of a 1D photonic crystal. We know that the fields at the band edges
are propagating. Here the absolute value of the field has the same period as the
crystal. In the band gap we use a linear interpolation of the position of the field
maximum as a function of z and therefore of ω to interpolate the periodic part of
the mode function into the gap. The field amplitude is also linearly interpolated
from its value at the lower edge to the upper edge. While a linear interpolation
of the position is not an exact description of the behavior shown in Fig. 3.5, we
have chosen it as it is a simple type of interpolation, and therefore a robust one.
The calculated maximum error in the LDOS due to the linear interpolation is
proportional to 1 − exp(−2zK ) in the starting unit cell. This comes from the
maximum difference between the exact field and the field calculated using the
model (see Fig. 3.5). For the structure considered in our calculation the maximum
normalized error is 12%. In the nth unit cell, however, this error is multiplied
by exp(−2nK a) and therefore becomes smaller. Since the geometry of the
considered infinite structure is nearly a quarter-wave stack [29], the band gap is
the widest band gap for such a structure. Therefore the difference between the
interpolated and the exact field behavior is in this case the largest; in any other
structure the gap is narrower and hence the difference is smaller. Simultaneously,
it is encouraging that the resulting maximum field amplitude matches well with
the exact result. When in future more insight is obtained in the behavior of fields
in a higher-dimensional band gap crystal, the interpolation can be improved. By
this interpolation we have obtained the periodic part of the field at each frequency
pbg
Êint
K (r). To obtain the complete field EK (z) we then multiply the interpolated
−K z
field Êint
K (r) with the corresponding decaying factor e at each frequency,
where K (ω) is obtained from the band structure calculations (see Fig. 3.3).
The physical situation that we consider is shown in Fig. 3.1. A light source is
embedded in a finite photonic crystal. Using our Ansatz that was introduced at
Eq. 3.5 and inserting it in Eq. 3.1 we obtain a general expression for the LDOS
in a finite one-dimensional photonic band gap crystal:
1 1
n m

N pbg (ω, z) = F (ω − ωi,j ) · |Êint 2
K (z)| exp(−2zK ) (3.6)
(2π) m i=1 j=1
where F (ω − ωi,j ) is the function that describes how the modes broaden in a
finite photonic crystal.
In Eq. 3.6 we have distinguished the summation over real wave vectors K from
the damping by imaginary wave vectors K , and the factor 2 in the exponential
originates from the mode functions squared in the LDOS. Thus at frequencies
in the photonic band gap a light source at position z in a photonic band gap
crystal experiences a continuum of vacuum fluctuations, all exponentially damped
depending on their directions. The expression Eq. 3.6 leads to the correct limit of
perfect inhibition (N pbg = 0) for an infinitely large band gap crystal - consistent
with Eq. 3.1 - since a light source is then infinitely far away (| z |−→ ∞) from
the surface.
3.5 LDOS in the band gap of a finite 1D periodic

structure
We note that in the case of a finite 1D periodic structure the dipole orientation
ed is in the same direction as Ên,K . Fig. 3.8 shows the calculated LDOS in the
band gap using Eq. 3.6 for two different crystal sizes of m = 10 and m = 20
unit cells. It is apparent from the figure that the LDOS is non-zero in the band
gap of both finite crystals. In contrast, in the band gap of an infinite structure,
the LDOS is zero independent of frequency. Furthermore the LDOS becomes
position dependent in the band gap of a finite crystal, again in contrast to the
infinite crystal case where the LDOS does not depend on position in the band
gap. Finally, the LDOS in the band gap strongly depends on crystal size, namely
exponentially. This observation can not even be contrasted to an infinite crystal,
since crystal extent is then irrelevant. It is seen that the LDOS is modulated in
the unit cell due to interferences between the reflections from different layers in
the crystal. This spatial dependence of the LDOS in the gap of a finite crystal
is in stark contrast with the independence of position in an infinite crystal. In
addition, the LDOS decays exponentially as a function of position into the crystal
and has a minimum near the center of the structure. The exponential decay is
faster at frequencies at the center of the band gap as the damping rate depends
on the magnitude of the imaginary part of the wave vector K . Beside this fast
decay at the center of the gap, the LDOS magnitude becomes also smaller since
the amplitude of the Lorentzian modes gets smaller in the middle of the band
gap.
Fig. 3.8 (a) and (b) allow us to compare the LDOS for two different crystal
sizes. The LDOS in the central layer of the larger crystal with m = 20 layers is
lower compared to the crystal with m = 10 layers. This illustrates that if we have
LDOS in the band gap of a finite 1D periodic structure 51
a light source in the center of a large crystal, the spontaneous emission inhibition
is higher due to a better shielding of vacuum fluctuations than in a small crystal.
Figure 3.8: Local density of optical states (LDOS) in the frequency range of the band
gap for (a) a finite crystal consisting of m = 10 periods and (b) a finite crystal consisting
of m = 20 periods. In the band gap the envelope of the LDOS decays exponentially
inside the structure with a minimum at the center of the structure. The attenuation
of the LDOS is strongest at the center of the band gap because the imaginary part of
the K vector is maximal. Moreover the summation over all F (ω − ωi,j ) is minimal at
frequencies in the center of the band gap.
To verify our model for the LDOS in the band gap of a finite photonic band
gap crystal, we compare the LDOS calculated using our model with the exact
calculation of the LDOS in a finite 1D periodic structure. Fig. 3.7 shows both the
exact LDOS calculations using transfer matrix method for a 1D, m = 20 periods
structure and the LDOS calculated using our model for the same crystal at the
same frequency. It is apparent in Fig. 3.7 that the result from the calculated
LDOS using our Ansatz is in very good agreement with the exact LDOS. In both
cases the LDOS shows the unit-cell modulations. The exponential decay in the
LDOS at the same frequency is also the same and all curves are in good mutual
agreement.
As a finite photonic crystal has an interface to the surrounding medium(see
Fig. 3.2), there may exist surface modes at frequencies in the band gap [42].
These modes can play a role when one is investigating the LDOS on the surface
of a photonic band gap crystal, for instance, when considering quantum emitters
at the surface of a crystal. In our present study, we are investigating the LDOS
as a function of depth in the bulk of a crystal where these modes are absent.
Therefore we can safely neglect them in our study.
52 Bibliography
3.6 Conclusion
We have developed a new model to calculate the LDOS in the band gap of a fi-
nite photonic crystal. This model considers the broadening of the electromagnetic
modes due to the finite size of the crystal. Our model also introduces an inter-
polation to calculate the electromagnetic field in the band gap using the band
structure and the electromagnetic field at the band edge. A comparison with
exact, analytical calculations for 1-dimensional periodic structures validates the
model. From our results we have gained a better overview of the field behavior
and especially the LDOS behavior in the band gap of a finite periodic struc-
ture. We find that inside the band gap of a finite structure the LDOS depends
on frequency, position, and on crystal size, in contrast to an infinite structure
where the LDOS is zero everywhere in the band gap independent of frequency
and position.
Bibliography
[1] E. Yablonovitch, T. J. Gmitter, and K. M. Leung, Photonic band stucture:
the face-centered-cubic case employing nonspherical atoms, Phys. Rev. Lett.
67, 2295 (1991). — p.39.
State Commun. 89, 413 (1994). — p.39.
[3] J. E. G. J. Wijnhoven and W. L. Vos, Preparation of photonic crystals made
of air spheres in titania, Science 281, 802 (1998). — p.39.
[4] Y. A. Vlasov, X.-Z. Bo, J. C. Sturm, and D. J. Norris, On-chip natural
p.39.
[5] A. Blanco et al., Large-scale synthesis of a silicon photonic crystal with a
complete three-dimensional bandgap near 1.5 micrometres, Nature 405, 437
(2000). — p.39.
289, 604 (2000). — p.39.
band gap, Applied Physics Letters 86, 011101 (2005). — p.39.
[8] S. A. Rinne, F. Garcı́a-Santamarı́a, and P. V. Braun, Embedded cavities and
waveguides in three-dimensional silicon photonic crystals, Nature Photon. 2,
52 (2007). — p.39.
[10] I. Staude, C. McGuinness, A. Frölich, R. Byer, E. Colby, and M. Wegener,
Waveguides in three-dimensional photonic bandgap materials for particle-
accelerator on a chip architectures, Opt. Expr. 20, 5607 (2012). — p.39.
Bibliography 53

Funct. Mater. 22, 25 (2012). — p.39.
[12] Z. Y. Li and Y. Xia, Full vectorial model for quantum optics in three-
dimensional photonic crystals, Phys. Rev. A 63, 043817 (2001). — p.39.
organizing systems, Phys. Rev. E 58, 3896 (1998). — p.39.
[15] D. P. Fussell, R. C. McPhedran, and C. Martijn de Sterke, Three-
dimensional Greens tensor, local density of states, and spontaneous emission
in finite two-dimensional photonic crystals composed of cylinders, Phys. Rev.
E 70, 066608 (2004). — p.39.
[16] A. Moroz, Minima and maxima of the local density of states for one-
dimensional periodic systems, Europhys. Lett. 46, 419 (1999). — p.39.
Am. B 26, 987 (2009). — p.39.
[18] J. M. Bendickson, J. P. Dowling, and M. Scalora, Analytic expressions for
the electromagnetic mode density in finite, one-dimensional, photonic band-
gap structures, Phys. Rev. E 53, 4107 (1996). — p.39.
[19] A. A. Asatryan, K. Busch, R. C. McPhedran, L. C. Botten, C. Martijn de
Sterke, and N. A. Nicorovici, Two-dimensional Greens function and local
density of states in photonic crystals consisting of a finite number of cylin-
ders of infinite length, Phys. Rev. E 63, 046612 (2001). — p.39.
[20] M. Wubs and A. Lagendijk, Local optical density of states in finite crystals
of plane scatterers, Phys. Rev. E 65, 046612:1 (2002). — p.39.
[21] V. Prosentsov and A. Lagendijk, The local density of states in finite size
photonic structures, small particles approach, Photonics and Nanostructures
5, 189 (2007). — p.39.
(2002). — p.39.
[24] N. Vats, S. John, and K. Busch, Theory of fluorescence in photonic crystals,
Phys. Rev. A 65, 1 (2002). — p.40.
[25] W. L. Vos, A. F. Koenderink, and I. S. Nikolaev, Orientation-dependent
spontaneous emission rates of a two-level quantum emitter in any nanopho-
tonic environment, Phys. Rev. A 80, 053802 (2009). — p.40.
[26] A. Yariv and P. Yeh, Optical waves in crystal: Propagation and control of
laser radiation (John Wiley, New York, 1983). — p.42.
[27] V. Laude, Y. Achaoui, S. Benchabane, and A. Khelif, Evanescent bloch
waves and the complex band structure of phononic crystals, Phys. Rev. B
54 Bibliography
80, 092301 (2009). — p.42.

[28] L. C. Botten, N. A. Nicorovici, R. C. McPhedran, C. M. d. Sterke, and A. A.
Asatryan, Photonic band structure calculations using scattering matrices,
Phys. Rev. E 64, 046603 (2001). — p.42.
2008). — p.42, 43, 49.
[30] P. St. J. Russell, T. A. Birks, and F. D. Lloyd-Lucas, in Confined electrons
and photons, edited by burstein, e. and weisbuch, c., edited by E. Burstein
and C. Weisbuch (Springer US, 1995). — p.43.
[31] J. Als-Nielsen and D. McMorrow, Elements of modern x-ray physics, second
edition (John Wiley, New York, 2011). — p.43.
[32] R. Newton, Scattering theory of waves and particles (Dover Publications,
1982). — p.46.
[33] J. Taylor, Scattering theory: The quantum theory of nonrelativistic collisions
(Dover Publications, Incorporated, 2012). — p.46.
[34] S. Smolka, Quantum correlations and light localization in disordered
nanophotonic structures, Ph.D. thesis, Technical University of Denmark
(DTU), 2010. — p.47.
[35] H. Thyrrestrup, S. Smolka, L. Sapienza, and P. Lodahl, Statistical theory of
a quantum emitter strongly coupled to anderson-localized modes, Phys. Rev.
Lett. 108, 113901 (2012). — p.47.
Press. Cambridge, 2006). — p.48.
[37] E. S. C. Ching, P. T. Leung, A. Maassen van den Brink, W. M. Suen,
S. S. Tong, and K. Young, Quasinormal-mode expansion for waves in open
systems, Rev. Mod. Phys. 70, 1545 (1998). — p.48.
[38] H. E. Türeci, A. D. Stone, and B. Collier, Self-consistent multimode lasing
theory for complex or random lasing media, Phys. Rev. A 74, 043822 (2006).
— p.48.
[39] Y. Kopaev and W. Hanke, Electronic phase transitions, Modern Problems
in Condensed Matter Sciences (Elsevier Science, 1992). — p.48.
Winston, New York, 1976). — p.48.
[41] In future, one might consider the Lorentzian width to depend on the disper-
sion relation. In the present work we already find a good agreement between
our model and exact calculation of the LDOS, therefore we have not at-
tempted to introduce dispersion in the width. — p.48.
[42] K. Ishizaki and S. Noda, Manipulation of photons at the surface of three-
dimensional photonic crystals, Nature 460, 367 (2009). — p.51.
CHAPTER 4
Finite size effects on a 3D photonic band gap; a
cavity quantum electrodynamic approach
4.1 Introduction
Finite size effects are omnipresent in many different types of physical systems.
A tangible example is the Ohm law V = I.R for wires of different lengths. If we
take a wire twice as long its resistance will double, since the resistance depends
on the length L of the wire R = R(L). Similarly many of the fascinating optical
properties associated with nanophotonic systems such as a full 3D photonic band
gap, weak and strong localization, and divergences in the densities of states are
strongly depending on system size. Since finite-size effects are unavoidable in
many different fields ranging from electronics and semiconductors to photonics
and nano-structures, the understanding of these effects is crucial for the success
of these fields, and will surely open up novel applications.
In quantum transport and in light transport the effect of finite sample size on
transmission and on reflectivity has been extensively investigated experimentally
and theoretically [1–5]. For a Bragg stack it is well-known that the percentage of
reflected light in the stopgap depends on the number of layers in the stack. By
increasing the number of layers a higher reflection is achieved.
Finite-size effects do not only modify the global properties of a system such
as transmission and reflection, they also modify the local properties such as the
local density of states (LDOS). In electronic systems, a well-known method to
probe the LDOS is to employ a scanning tunneling microscope (STM) [6], see
Fig. 4.1 (a). The working principle of the STM is based on the tunneling of the
electron (or a hole) through a potential barrier between the STM tip and the
sample surface. Due to the nature of an STM, however, such measurements are
limited to spatial positions at the surface of a sample. With an STM, the LDOS
is probed near the Fermi energy. By altering the voltage applied to the STM tip,
it is possible to adjust the Fermi energy and hence probe the electronic LDOS
at different energies [7]. At present, we are unaware of an experimental tech-
nique to study the electronic LDOS in bulk ∗ . While there seems limited scope
for theoretical interpretation of the LDOS in bulk, the electronic LDOS can be
calculated from ab-initio theory such as tight-binding or plane-wave expansions.
In contrast in nanophotonics, it is possible to probe the LDOS in the bulk of a
∗ We thank dr. Geert Brocks for helpful discussions on this aspect.
56 Finite size effects on a 3D photonic band gap
crystal, as we will show in this chapter. We employ the feature that the photonic
LDOS plays a central role in cavity quantum electrodynamics (QED) [8]. In the
weak coupling approximation where Fermi’s golden rule holds, the LDOS is pro-
portional to the radiative emission rate that can be probed by time-resolved [9–13]
or CW spontaneous emission [14, 15], see Fig. 4.1 (b). The LDOS does not only
depend on frequency like the DOS, it also depends on position, and orientation
of emitters’ dipole moment with respect to the field. Here we are particularly
interested in 3D photonic band gaps, in other words in the range of vanishing
LDOS, where light is forbidden for all wave vectors and polarizations. The con-
cept of a photonic band gap and its consequences for cavity QED were first
introduced by Bykov in 1972 [16]. In 1987 it was brought to worldwide attention
by the theoretical work of Yablonovich [17] and John [18]. Such band gaps are
expected in 3D photonic crystals, i.e., dielectric nanostructures with periodici-
ties less than half the optical wavelength [19]. The present-day understanding
of complex nanophotonic systems is heavily leaning on theories that describe in-
finite systems: L → ∞. Well-known examples are random media and photonic
crystals [20–24]. Therefore the predicted inhibition in the band gap only holds
for infinite systems. In contrast, experiments are obviously performed on real
devices with a finite extent. The question remains: Is a big photonic crystal the
same as an infinite photonic crystal? After many efforts to fabricate 3D photonic
band gap crystals the first experimental observation of inhibition in the band
gap was performed by Leistikow et al in 2011 [11], showing a strong yet finite
inhibition in a real photonic band gap crystal. Although there has been much
study on photonic band gap crystals, to date there has not been any quantitative
study on finite size effect in a real photonic band gap crystal. Moreover, to the
best of our knowledge there is no theory for LDOS in a finite band gap crystal.
Our study here is the first ever systematic study of inhibition in a real photonic
band gap crystal both experimentally and theoretically. We use the emission rate
as a tool to investigate the finite size effect in photonic crystals. We map the
quantum dots decay rate as a function of frequency through the photonic band
gap. We observe up to 18-fold inhibition near the center of the band gap. Based
on physical principles of the light propagation in a photonic crystal we develop
a theory to interpret spontaneous emission rates measured on a finite photonic
band gap crystal.
4.2 Experimental techniques

We have realized silicon inverse woodpile photonic band gap crystals by etching
two sets of carefully aligned perpendicular pores using CMOS compatible meth-
ods. Fig. 4.2 (a) shows a micrograph of a typical 3D crystal [25]. The crystal has
a diamond-like structure with
√ an orthohombic lattice with parameters a = 693
nm and c = 488 nm (a = 2c). The 3D crystal extends over L3 = 12 × 12 × 12
ac2 which exceeds the Bragg attenuation length Bragg in every direction [26].
As a result of multiple Bragg diffraction, frequency gaps form, known as stop
gaps. If stop gaps in all directions overlap, a full 3D photonic band gap appears.
Experimental techniques 57
(a) STM tip
Device
(b)
Free space Free space
Photonic
crystal
Figure 4.1: Schematic picture of a finite size system that has interaction with outside
world and measuring LDOS for such systems. (a) An electronic device (pink gradient
colored square). The gradient represents the gradual coupling to the environment. The
only method to measure the LDOS in such a finite system is to use an STM. This
only probes the LDOS on the surface. (b) Schematic representation of a finite photonic
crystal in free space. Light sources (yellow dots) probe the LDOS deep inside the
crystal.
Figure 4.2 (c) shows the band structure of an infinite inverse woodpile crystal
made from silicon and infiltrated with toluene. The gray area shows the stop
gap range of all crystal directions, which overlap in the range indicated with the
yellow bar that is the full photonic band gap. The band gap range is in the
near infrared and matches with the telecommunication range. The reflectivity
spectra of 3D silicon inverse woodpile crystals reveal broad peaks with maximum
reflectivity of 67% that are independent of the spatial position on the crystals.
The spectrally overlapping reflectivity peaks for all directions and polarizations
form the signature of a broad photonic band gap with a relative bandwidth up
to 16% [26].
Here we measure decay rates of the PbS quantum dots in the band gap fre-
quency range of an inverse woodpile photonic crystal. The PbS quantum dots
have a wide emission spectrum ranging from 1300 to 1600 nm (see Ch. 2), over-
lapping with the band gap. PbS quantum dots are suspended in toluene. In
order to infiltrate the quantum dots into the crystal we immerse the crystal in
the quantum dots suspension. To measure the decay rate of quantum dots we
have used the time-correlated single-photon counting method. The decay rate is
a b
0.6
1.0
0.4
0.5
0.2
0.0 0.0
c
a
Figure 4.2: (a) Scanning electron micrograph of a 3D inverse woodpile silicon photonic
crystal. The 3D crystal consists of two sets of perpendicularly etched pores, surrounded
by a 2D crystal. (b) Band structure of an inverse-woodpile photonic crystal with pore
radius 170 nm, ε = 12.1 for silicon and ε = 2.25 for toluene-filled pores. All stopgaps
(shown with gray area) overlap, forming a full band gap indicated with the yellow bar
In bandgap
100000 Outside bandgap
Counts
10000
1000
30000
Residual
0
-30000
0 2000 4000 6000 8000
Time(ns)
Figure 4.3: Time-resolved spontaneous emission measured on a PbS quantum dot sus-
pension in a photonic crystal with the pore radius of 170 nm. The red triangles show
the measured decay curve for the quantum dots at 0.853 eV in the band gap of the
crystal. The black squares show the decay curve measured at 0.83 eV outside the band
gap. Data is rebinned to 400 ns steps. Red and black dashed curves are bi-exponential
fit to the data. Bottom panel: the residuals are random and symmetrically distributed
around zero, indicating high quality fits.
obtained from the slope of the decay curve [27].

Experimental observations 59
4.3 Experimental observations

Fig. 4.3 shows the time-resolved spontaneous emission of the quantum dots in
the crystal at frequencies outside and inside the band gap. Outside the band gap
quantum dots decay nearly exponentially, with a slope close to the one of the
quantum dots in suspension. This is because outside the band gap the DOS of
photonic crystal is similar to the DOS in free space and increases nearly quadrat-
ically with frequency, see Fig. 4.5. In the band gap we observe a non-single
exponential decay for the quantum dots. As described in Ref. [11] the collected
signal is a mixture of signal from the crystal and background signal from quantum
dots outside the crystal. In the first 500 ns the decay rate is dominated by the
fast decay rate of quantum dots in the suspension. It then deviates strikingly and
shows a very slow decay rate, confirming suppression of spontaneous emission in
the band gap. In our previous measurements [11] we have only measured up to
2000 ns time interval. Here by improving the signal-to-noise ratio of the setup
we have been able to measure over a 4 times longer time extent, we could clearly
observe a very slow decay as a result of inhibition in the band gap.
As we always collect a mixed signal from quantum dots outside the crystal and
in the crystal, the decay curves always show one fast single exponential decay
due to the suspension signal followed by a slow decay from quantum dots in
the crystal. Therefore a bi-exponential model is a logical choice to model the
time-resolved total count rate :
f (t) = I(S γ0 exp(−γ0 t) + P C γ exp(−γt)). (4.1)
Here γ is the emission rate of quantum dots in the photonic crystal that we
wish to obtain, and γ0 is the rate of dots outside the crystal that is separately
measured to be γS = 1.7 μs−1 (Fig. 2.3(b)). From the analysis done in Ch. 5 we
find that this model is very robust. S and P C are relative factors for the signal
from quantum dots in suspension or in the crystal and are fixed by scans of the
detection focus, and only γ and I are adjusted [11]. In order to find S and P C,
the detection objective is scanned once when it is positioned over the 3D crystal,
and once when it is over the silicon wafer. By this method we find the amount
of background light from other regions entering the detection objective.
Due to very low signal to noise ratio of the single photon detection in the
near infrared, the data is collected for several hours for each decay curve which
was feasible thanks to the great stability of our setup (Ch. 2). The background
counts are substantial compared to the signal. Therefore we have performed a
careful maximum likelihood analysis of the time-resolved data, see Ch. 5. The
residuals from modeling the data are random and symmetric, centered around
zero, indicating a good model to the data.
Thanks to the wide emission spectrum of the quantum dots we can measure the
decay rate at several frequencies within the band gap range. In our experiment
we measure the decay rate of quantum dots suspended in toluene γ0 and also
embedded in silicon inverse woodpile photonic crystals γ separately. In the band
gap, we observed a strong inhibition of the total decay rate γ0 /γ of 18 times
which is the largest ever inhibition observed in the band gap of a 3D photonic
crystal. The 18 times inhibition is even an underestimate due to the fact that
the simple bi-exponential model can not describe the slow decay component of
the data. In time-resolved emission experiment we measure the total decay rate
γtot which consists of both radiative γrad and non-radiative γnrad decay rate:
γtot = γrad + γnrad . Interestingly, the maximum inhibition sets an upper bound
to the non-radiative decay rate to be at most γnrad ≤ 0.05γ0 . This non-radiative
decay implies that the quantum dots have a quantum efficiency of at least 95%. In
view of this very high room temperature efficiency we consider the upper bound
to be the most reasonable estimate of γrad . The red data points in Fig 4.4 are the
measured γ/γ0 as a function of frequency throughout the band gap. The error
bars indicate the credible interval of each data point stemming from the Bayesian
statistical analysis on the data as discussed in Ch. 5. The credible interval is large
at higher energies due to the small signal from the quantum dots. This is due to
the fact that the emission spectrum of the quantum dots is not symmetric with
respect to the band gap range and is very low at higher frequencies. At every
frequency we have measured three times decay curves and extracted three γ/γ0
which are in very close mutual agreement. This confirms the reproducibility of
our measurements and stability of the setup in this challenging experiment.
By subtracting the upper bound γnrad from the total rate γtot , we arrive at the
radiative rate γrad versus frequency that is shown in Fig. 4.4. At low frequency
outside the band gap, the emission rate is slightly decreased, in agreement with
infinite crystal theory. At the edge of the gap (0.845 eV) we observe a slightly
enhanced emission rate. While this result has been reproduced several times, we
have at this time no explanation for this result. At all frequencies in the 3D pho-
tonic band gap (0.845 to 0.905 eV), we observe a strong inhibition of the emission
rate. The strongest inhibition seems to occurs at 0.867 eV, slightly off the center
of the gap. Taking into consideration all measurements, the observed maximum
inhibition appears to be about 50 to 100 times. These results are a stronger
inhibition than the selected data from nano wire (Bleuse-PRL 2011 [13]) and for
2D photonic crystal slabs (Wang, Lodahl-PRL 2011 [28]), which for the first time
ever confirms the old expectation on 3D band gaps (Yablonovitch 1987 [17], John
1987 [18]). Moreover, it is remarkable that the strong inhibition is observed for
an ensemble of quantum emitters, with random dipole orientation [29], and not
for a selected single emitter with optimized orientation or position. This result
suggests that the ensemble contains emitters that experience both a weaker but
also an even stronger inhibition than the average values quoted above.
4.4 Theory for 3D finite crystal

Photonic crystals redistribute the DOS both in frequency and space. The redis-
tribution of the DOS in frequency is shown in Fig. 4.5, where the DOS is zero in
the band gap and nearly quadratic with some peaks outside the band gap. The
redistribution of the DOS in space is realized by the fact that inside a photonic
band gap the DOS is zero while the total number of states is conserved so the
DOS outside the gap has to increase in certain ranges. The spatial resolved or the
Theory for 3D finite crystal 61
Reduced frequency ( ωa/2πc)

0.46 0.48 0.50
1 1
Radiative rate (γrad/γ0)
Relative DOS
0.1 0.1
0.01 0.01
Experiment
Theory infinite 3D
Theory finite 3D
Theory finite 1D
1E-3 1E-3
0.80 0.85 0.90
Photon energy (eV)
Figure 4.4: Comparing the data with the theory. The red circles are the measured
γ/γ0 of the quantum dots throughout the band gap. The bars are the credibility
interval. The orange line is the calculated relative DOS. The calculation is for a 12
by 12 by 12 unit cells crystal (1728 k points in the Brillune zone) with pore radius
R = 170 nm and a = 693 nm. The green dashed line is the relative DOS for an infinite
crystal which goes to infinite in the band gap. The blue dot-dashed line is the calculated
relative DOS following the 1D approach discussed in Sec. 4.4.
local density of states (LDOS) in the band gap of an infinite photonic crystal is
zero as well as density of states (DOS). However, when the crystal becomes finite
the concept of the DOS for an open finite photonic crystal is not very meaningful,
since the volume of such a crystal is small compared to free space while we have
to integrate over space to obtain the DOS. The DOS for a finite crystal is then
the same as the DOS for free space. The LDOS, however, is defined for the crys-
tal itself, since it is a function that depends on position anywhere either inside
or outside the crystal. The LDOS is equal to the imaginary part of the Green
function. Although there have been some efforts for calculating the LDOS in the
band gap of 1D and 2D finite crystals [30–34], it becomes extremely challenging
in 3D and to the best of our knowledge there is no analytic theory to calculate
the LDOS for any class of crystal. For 3D light in a 3D inverse opal, Hermann
and Hess have calculated the LDOS as a function of frequency, position, and for
several position in the crystal by means of finite difference time domain (FDTD)
simulations [35]. Kole has studied the LDOS by FDTD for finite opal crystals
consisting of different number of spheres in clusters [36]. Unfortunately, however,
these results can not be applied to other crystals, physical insights are not read-
ily apparent, as is intrinsic to simulations. Finally such computer simulations
require a lot of programming efforts and computational costs. Whittaker has
calculated the transmission and the emission (but not LDOS) as a function of
crystal thickness for a 3D photonic band gap crystal which is infinitely extended
in two planar dimension, hence an infinite crystal volume [37]. Ishizaki et al.
have simulated the emission rate that is proportional to the DOS in the middle
layer of a finite 3D photonic band gap crystal. Therefore their calculations are
Reduced frequency(ωa/2πc)
Figure 4.5: Density of states (DOS) per volume for the same crystal shown in Fig. 4.2
calculated with 10000 k points. The DOS vanishes in the band gap indicated with a
yellow bar. Dashed curve: quadratic behavior in the low frequency limit.
limited to one particular position and do not describe the DOS over the whole
volume supported by the crystal [38].
Here we introduce a new point of view on a finite 3D photonic band gap
crystal. This original point of view enables us to develop a theory to calculate
the finite support DOS in the band gap of a 3D finite photonic crystal. From the
calculations performed in chapter 3 and published in Ref. [39] the LDOS depends
on crystal size, frequency, and position in the crystal in the band gap of a finite
crystal. The LDOS decreases exponentially in the band gap with position. In
frequency domain the LDOS decreases from the band edge frequency toward the
middle of the band gap. In addition to position and frequency the LDOS depends
on the crystal size. The crystal size determines how much the modes in a crystal
broaden and how many modes exist in a finite crystal. Due to the dependence
of the LDOS on the crystal size, the LDOS is used to investigate the finite-size
effects.
Since in our experiment the quantum dots are suspended in toluene they can
move within the voids inside the crystal and diffuse everywhere. Since the diffu-
sion time is in the order of micro second we assume that the density of particles
is everywhere the same in the crystal. Therefore in our time resolved emission
measurement we measure the LDOS averaged over all positions and dipole orien-
tations in a unit cell. Our measurements are directly comparable with the finite
support DOS that is a function of frequency and the crystal size and averaged
over the emitters’ position and orientation.
The physical situation in this case is the following: if we consider a light source
deep inside a finite crystal, vacuum fluctuations can tunnel into the crystal and
induce spontaneous emission. For the tunneling rate of vacuum fluctuation to
inside or photons to outside we consider two different approaches as shown in
a b
L L
Figure 4.6: A schematic drawing of the physical situation of finite photonic crystals.
(a) The first approach: a finite 3D photonic crystal with length L in every direction.
The tunneling rate of vacuum fluctuation is set by the closest interface. (b) The second
approach: the same crystal but the vacuum fluctuations (red wavelets) that are running
around can tunnel into the band gap from all directions. The tunneling rate is set by
the crystal volume.
Fig 4.6. In the first approach the rate depends on the closest vacuum-crystal
interface to the emitter. This might seem logical since one possibly thinks that
tunneling happens via the shortest distance to the emitter. This would mean
that the LDOS length LDOS [11] - the length in the crystal where the LDOS
has decayed to the 1/e value of the LDOS outside the crystal - is the same as
the Bragg length of the closest interface Bragg . Therefore we only consider the
number of unit cells in the direction of the nearest interface, which is equal to
L/a with a the unit cell length and L the sample thickness. In this case light
is considered to propagate only in one direction. Notice that the direction that
has the shortest distance to the interface does not have necessarily the smallest
Bragg length Bragg .
In the second approach, light can propagate in all directions. Therefore the
tunneling rate depends on the volume of the finite crystal or in other words on
the total number of unit cells in the whole finite crystal volume. If we assume
a cubic crystal with length L in each direction, the total number of allowed
modes in a finite volume determines the volume per mode a3 /L3 for such a 3D
system. The volume per mode in the reciprocal lattice is inversely proportional
to the real space volume of each unit cell in the crystal stemming from the
Born − von Kármán boundary condition in 3D [40].
Our starting point to calculate the finite support DOS is the definition of the
DOS for an infinite photonic crystal.

1
DOS(ω) = dKδ(ω − ωn,K ), (4.2)
(2π)3 n BZ
where ω is the emission frequency. The integration over the real wave vector K
is performed over the first Brillouin zone, and n is the band index. The modes
in an infinite photonic crystal are delta functions at frequencies ωn,K that are
the eigenvalues of the system. The eigen vectors in an infinite photonic crystal
are Bloch modes that propagate throughout the whole crystal. These modes are
absent in the band gap frequency range and therefore the DOS becomes zero in
the band gap. If we extend the modes for infinite systems to represent finite-size
effects then we have discrete number of modes that are broadened. The mode’s
shape becomes Lorentzian where linewidth depends on the crystal size. Here we
assume that the crystal consists of m unit cells in every direction L = m · a.
These Lorentzian modes F (ω − ωij ) have the following shape:
1 Δij
F (ω − ωij ) = ( Δij ), (4.3)
(2π) (
2 ) + (ω − ωij )2
2
where Δij is width of the Lorentzians and is taken to be the same for all the
modes in a finite crystal. ωij is the central frequency of each mode and therefore
the central frequency of each Lorentzian in a certain band (indicated by i) at
a certain wave vector (indicated by j). The width Δij and the number of the
Lorentzian modes depends on whether we consider the 1D or 3D light propagation
approach. If we consider the first one the number of Lorentzian modes is set by
the crystal thickness and their width in the normalized unit is equal to:
a
Δ1D = , (4.4)
L
where L is the thickness, a is unit cell length, and therefore number of unit cells
in one direction is m1D = L/a.
If we consider the second approach (3D light propagation) then we should con-
sider the total number of modes in the volume and their width in the normalized
unit is:
a3
Δ3D = . (4.5)
L3
Following the same reasoning for 2D in-plane light propagation we have:
a2
Δ2D = . (4.6)
L2
Here the number of the unit cells in this finite volume determines the number
of wave vectors in the Brillouin zone of such a finite crystal again following the
Born − von Kármán boundary condition in solid state physics [40]. Since modes
are now Lorentzian, for every wave vector we consider a Lorentzian mode and
therefore we have m3 Lorentzian modes for each band in a 3D light propagation,

and m2 and m for a 2D and a 1D propagation respectively.
De
ns
ity (a)
of
sta
tes
0.6
Reduced frequency (ωa/2πc)

0.4
0.2
Sp
ec
tra
l de
ns 0.0
ity
De
ns
ity
of (b)
sta
tes
0.6
Reduced frequency (ωa/2πc)

0.4
0.2
Sp
ec
tra
l de
ns 0.0
ity
Figure 4.7: Schematic drawing of our theory for the density of states in a finite 3D
crystal. We have used the band structure in Fig. 4.2. (a) In a finite crystal modes
become Lorentzian functions (orange) that are broadened depending on the crystal
size. The number of Lorentzian modes depends on the number of unit cells. Since
modes are broadened they partially lie in the band gap range shown with the orange
filled area. By adding all modes, we obtain a non-zero density of states in the band gap
that gives rise to a finite inhibition as observed in our experiment. (b) In an infinite
photonic crystal every individual mode is a delta function shown as blue spikes. Position
and number of modes are chosen arbitrarily. Since there are no propagating modes in
the band gap, no delta function extends into the photonic band gap. Hence the density
of states, plotted in the third dimension at the left is zero in the band gap and only
non zero outside the band gap. The spectral density elucidates the magnitude of each
normalized mode.
The expression for the Lorentzian modes F (ω − ωij ) tends to a Dirac delta
function in the limit of an infinitely large crystal (L −→ ∞) in agreement with
the theory of infinite photonic crystals [8, 22]. Due to the finite size of the crystal
the modes become discrete. The averaged LDOS is thus the summation over all
Lorentzian modes in the crystal volume normalized by the number of modes m3
in the Brillouin zone:

3
1
n m
pbg
Nfinite (ω − ωij ) = 3 F (ω − ωij ) (4.7)
m i=1 j=1
Since the number of modes in a 3D crystal increases cubically with the number
of unit cells in each direction, a cubic crystal of only m = 12 unit cells at every
direction contains m3 = 1728 modes. Therefore the volume per mode and thus
the Lorentzians linewidth becomes cubically small.
Fig. 4.7 (a) illustrates our theory for a finite crystal where modes become
Lorentzians. We use the calculated band structure in Fig. 4.2 (c) to illustrate
the theory. The spectral density elucidates the magnitude of each mode, which
is normalized. Here modes are shown with orange Lorentzians whose width is
set by the crystal size. For simplification the number of modes is chosen arbi-
trarily. Since the modes are broad they partially extend into the band gap range
which is indicated with the orange filled areas. By adding all the modes we end
up with a non-zero density of states in the band gap. This non-zero density of
states give rise to a finite inhibition in a finite photonic crystal as observed in our
measurements. It is interesting to mention that Lorentzian linewidths are also
considered for electrons [41]. In a crystal at finite temperature the electron scat-
tering rate increases due to the electron-lattice interaction and electrons obtain
Lorentzian spectral functions. The increased scattering rate can be also caused
by lattice defects, impurity, doping, deformation of the structure, and structure
rotation [3].
Fig. 4.7 (b) illustrates a similar scheme for the modes in an infinite photonic
crystal, where every individual mode is a delta function shown as the blue spikes.
Since there are no propagating modes in the band gap, there are no delta functions
in the range of the photonic band gap. Since each mode function has zero width,
no spectral density occurs in the band gap. The density of states is plotted on
the third axis of the plot. As shown the density of states is zero in the band gap
and non-zero outside the band gap.
4.5 Inhibition in the band gap

We have calculated the finite support DOS in the band gap of a finite photonic
crystal. The DOS is a quantity that is proportional to the inhibition of the
quantum dots in the band gap. If the DOS is below one then we expect that the
quantum dots experience inhibition and if it is above one then they experience
enhancement. Fig. 4.4 orange line shows the exact result of our Lorentzian theory.
The finite support DOS is calculated for a finite silicon-toluene inverse woodpile
photonic crystal consisting 12 unit cells in each direction, thus 12×12×12 = 1728
k points in the Brillouin zone or 1728 modes. We have calculated the relative
DOS in the band gap for the 1D and 3D approaches in comparison with the DOS
calculated for a similar infinite structure. The range of the photonic band gap is
indicated by the yellow bar.
How big should a photonic band gap crystal be? 67
The orange line in Fig. 4.4 shows the calculated inverse DOS when the width of
each Lorentzian mode is proportional to the total number of modes following the
3D approach. The 3D theory is remarkably good in showing inhibition in the band
gap range it also shows an excellent agreement with the measured data, validating
the 3D theory. This is confirming that for a 3D crystal the Born − von Kármán
boundary condition should be considered for all boundaries confining the crystal
within its volume. Since inhibition is a unitless quantity that does not depend
on the surrounding environment refractive index, we therefore calculate the finite
support DOS also relative to the long wavelength range DOS of the crystal. The
blue dot-dashed line in Fig. 4.4 shows the calculated DOS for the 1D approach
when tunneling happens only in one direction. The calculation following the 1D
approach not only gives rise to a DOS above one in the band gap(enhancement)
that is non-physical, it is also far from the experimental data. We conclude that
in a 3D finite crystal it is not only the closest interface that leads to tunneling
of the vacuum fluctuations, but it is every direction in the crystal volume that
determines the volume per mode and that determines the tunneling strength.
For comparison the DOS calculated for the infinite crystal (green dashed line) is
also shown in in Fig. 4.4. The infinite DOS tends to zero in -as expected- strong
disagreement with our measurements on a real finite size band gap crystal.
It is remarkable that the density of states (DOS), which is strictly zero in the
band gap of an infinite crystal irrespective of frequency (within the gap of course)
and position, becomes strongly dispersive in a finite crystal. We expect that our
theory is applicable to a wide range of complex nanophotonic metamaterials, and
condensed matter that have a band gap for many different classes of waves.
4.6 How big should a photonic band gap crystal be?

In this section we investigate how the maximum inhibition in the band gap de-
pends on the crystal size. Figure 4.8 shows the maximum inhibition in the band
gap calculated for crystals with different volume. We considered crystals made of
7 layers (343 unit cells), 10 layers (1000 unit cells), 11 layers (1331 unit cells), and
12 unit cells (1728 unit cells). The DOS decreases linearly with the number of
unit cells. Remarkable that a (photonic) crystal property depends algebraically
on size and not exponentially. It is reminiscent of Ohm’s law where conductance
also depends linearly on size as mentioned in the introduction.
The linear behavior of the the DOS versus crystal size is supported by the
following: Let’s consider the Lorentzian mode function F (ω − ωij ):
1 Δij
F (ω − ωij ) = ( Δij ), (4.8)
(2π) (
2 ) + (ω − ωij )2
2
with ωij the central frequency of the Lorentzian, and Δij the bandwidth that
we chose to be inversely proportional to volume: Δij = a3 /L3 = 1/m3 with a
the lattice parameter. Let’s consider that we are in the bandgap at a constant
Gap depth (-γ0/γ)

-50
-100
Our theory
Linear
DOS in vacuum
-150
0 500 1000 1500 2000
3
Number of unit cells (m )
Figure 4.8: The gap depth or the minimum relative DOS in the band gap as a function
of crystal size is shown in comparison with DOS in vacuum. The DOS in the photonic
band gap crystal decreases only weakly with crystal size (proportional to the total
number of unit cells).
detuning (ω − ωij ) ≡ c. Now we can rewrite our Lorentzian as
(1/m)3 (1/m)3
F (ω − ωij ) ∝ = (4.9)
((1/m)3 )2 + c2 (c2 + (1/m)6 )
This result shows that the Lorentzian mode function goes with m3 , and this is
exactly what we see in Fig. 4.8.
Our observation does not agree with the expected exponential decrease of the
DOS with crystal size as shown by Ishizaki et al [38], or Kole [36]. This dif-
ference is understandable, since in their calculation they consider the position
dependency in the crystal, while our calculation is averaged over position and
is based on number of modes in crystal volume. Since there is no position de-
pendency in the DOS, there is absolutely no reason for exponential positions
dependence. The DOS is only determined by the Lorentzians, that decreases as
1/m3 inside the band gap. The finite support DOS decreases only weakly with
crystal size (proportional to the total number of unit cells), we therefore conclude
that in practice there is no finite crystal that effectively behaves as an infinite
crystal. We believe that our quantitative understanding can be used not only for
the photonic systems also in any condensed-matter system where the transport
mean free path is longer than system size such as electrons in (semi)conductors,
or Cooper pairs in superconductors or phonons and magnons in crystals. If the
mean free path is shorter than the system size the propagator such as photon or
electron does not see the effect of the interface since it loses its memory of one
interface before it reaches the other interfaces.
Conclusion and discussion 69
4.7 Conclusion and discussion

We have studied the finite size effect on the density of states in photonic crystals
both experimentally and theoretically. We have measured time-resolved emis-
sion of semiconductor quantum dots that emit in the telecom range within the
band gap of silicon 3D photonic crystals. Finally, the long-searched inhibition in
the photonic band gap crystal is observed. The inhibition is strong, yet finite,
suggesting a finite-size effect on the inhibition. To interpret our experimental
observation we have devised an original theory that allows the models for infinite
systems to be extended to represent finite-size effects, without actually reducing
the size of the infinite system. The method is based on an extension of the wave-
vector space of the eigenmodes into the complex plane. A remarkable result is
that the density of states (DOS), which is strictly zero in the band gap of an
infinite crystal irrespective of frequency becomes strongly dispersive in a finite
crystal. Interestingly, the DOS decreases linearly with the crystal volume. This
weak dependency of the DOS to the crystal size suggests that no finite crystal
behaves as an infinite crystal. Our theory for finite size 3D photonic band gap
crystal shows that we are in the position to explain our experimental observations
on the nonzero and frequency-dependent LDOS inside the 3D photonic band gap.
We expect that our theory is applicable to a wide range of complex nanopho-
tonic metamaterials such as in lasers, fibers, and sensors. We believe that our
quantitative understanding can be used not only for the photonic systems also in
any condensed-matter system where the transport mean free path is long such as
electrons in (semi)conductors at low temperature, or Cooper pairs in supercon-
ductors, or phonons, and magnons in crystals. If the mean free path is shorter
than the system size the propagator such as photon or electron does not see the
effect of the interface since it loses its memory of one interface before reaches the
other interfaces.
Bibliography
[1] F. J. Pinski, P. B. Allen, and W. H. Butler, Calculated electrical and thermal
resistivities of Nb and Pd, Phys. Rev. B 23, 5080 (1981). — p.55.
[2] S. Y. Savrasov and D. Y. Savrasov, Electron-phonon interactions and related
physical properties of metals from linear-response theory, Phys. Rev. B 54,
16487 (1996). — p.55.
[3] S. Datta, Quantum transport: Atom to transistor (Cambridge University
Press, 2013). — p.55, 66.
[4] J. F. Bertone, P. Jiang, K. S. Hwang, D. M. Mittleman, and V. L. Colvin,
Thickness dependence of the optical properties of ordered silica-air and air-
polymer photonic crystals, Phys. Rev. Lett. 83, 300 (1999). — p.55.
[5] B. Harbecke, Coherent and incoherent reflection and transmission of multi-
layer structures, Appl. Phys. B 39, 165 (1986). — p.55.
[6] G. Binnig and H. Rohrer, Scanning tunneling microscopy from birth to ado-
lescence, Rev. Mod. Phys. 59, 615 (1987). — p.55.
70 Bibliography
[7] J. Li, W.-D. Schneider, R. Berndt, and B. Delley, Kondo scattering observed
at a single magnetic impurity, Phys. Rev. Lett. 80, 2893 (1998). — p.55.
dielectrics, Europhys. Lett. 35, 265 (1996). — p.56, 65.
[9] D. Englund, D. Fattal, E. Waks, G. Solomon, B. Zhang, T. Nakaoka, Y.
Arakawa, Y. Yamamoto, and J. Vučković, Controlling the spontaneous emis-
sion rate of single quantum dots in a two-dimensional photonic crystal, Phys-
ical review letters 95, 013904 (2005). — p.56.
[10] P. Anger, P. Bharadwaj, and L. Novotny, Enhancement and quenching of
single-molecule fluorescence, Phys. Rev. Lett. 96, 113002 (2006). — p.56.
a 3D photonic band gap, Phys. Rev. Lett. 107, 193903 (2011). — p.56, 59,
63.
[12] M. R. Jorgensen, J. W. Galusha, and M. H. Bartl, Strongly modified spon-
taneous emission rates in diamond-structured photonic crystals, Phys. Rev.
Lett. 107, 143902 (2011). — p.56.
[13] J. Bleuse, J. Claudon, M. Creasey, N. S. Malik, J.-M. Gérard, I. Maksy-
mov, J.-P. Hugonin, and P. Lalanne, Inhibition, enhancement, and control of
spontaneous emission in photonic nanowires, Phys. Rev. Lett. 106, 103601
(2011). — p.56, 60.
[14] A. F. Koenderink, L. Bechger, H. Schriemer, A. Lagendijk, and W. L. Vos,
Broadband fivefold reduction of vacuum fluctuations probed by dyes in pho-
tonic crystals, Phys. Rev. Lett. 88, 143903 (2002). — p.56.
[15] S. Ogawa, M. Imada, S. Yoshimoto, M. Okano, and S. Noda, Control of light
emission by 3d photonic crystals, Science 305, 227 (2004). — p.56.
[16] V. Bykov, Spontaneous emission in a periodic structure, Sov. Phys. JETP
35, 269 (1972). — p.56.
electronics, Phys. Rev. Lett. 58, 2059 (1987). — p.56, 60.
[18] S. John, Strong localization of photons in certain disordered dielectric super-
lattices, Phys. Rev. Lett. 58, 2486 (1987). — p.56, 60.
2008). — p.56.
[20] S. John and J. Wang, Quantum electrodynamics near a photonic band gap:
Photon bound states and dressed atoms, Phys. Rev. Lett. 64, 2418 (1990).
— p.56.
[21] Z. Y. Li and Y. Xia, Full vectorial model for quantum optics in three-
dimensional photonic crystals, Phys. Rev. A 63, 043817 (2001). — p.56.
Phys. Rev. A 65, 1 (2002). — p.56, 65.
[23] X. H. Wang, B. Y. Gu, R. Z. Wang, and H. Q. Xu, Decay kinetic properties
of atoms in photonic crystals with absolute gaps, Phys. Rev. Lett. 91, 113904
(2003). — p.56.
Bibliography 71
1557 (2008). — p.56.
Setija, and W. L. Vos, Inverse-woodpile photonic band gap crystals with a
cubic diamond-like structure made from single-crystalline silicon, Adv. Func.
Mater. 22, 25 (2012). — p.56.
— p.56, 57.
[27] D. V. O’Connor and D. Philips, Time correlated single photon counting
(Academic Press, 1984). — p.58.
[28] Q. Wang, S. Stobbe, and P. Lodahl, Mapping the local density of optical
states of a photonic crystal with single quantum dots, Phys. Rev. Lett. 107,
167404 (2011). — p.60.
[29] W. L. Vos, A. F. Koenderink, and I. S. Nikolaev, Orientation-dependent
spontaneous emission rates of a two-level quantum emitter in any nanopho-
tonic environment, Phys. Rev. A 80, 053802 (2009). — p.60.
[30] J. M. Bendickson, J. P. Dowling, and M. Scalora, Analytic expressions for
the electromagnetic mode density in finite, one-dimensional, photonic band-
gap structures, Phys. Rev. E 53, 4107 (1996). — p.61.
[31] A. A. Asatryan, K. Busch, R. C. McPhedran, L. C. Botten, C. Martijn de
Sterke, and N. A. Nicorovici, Two-dimensional Greens function and local
density of states in photonic crystals consisting of a finite number of cylin-
ders of infinite length, Phys. Rev. E 63, 046612 (2001). — p.61.
[32] M. Wubs and A. Lagendijk, Local optical density of states in finite crystals
of plane scatterers, Phys. Rev. E 65, 046612:1 (2002). — p.61.
[33] D. P. Fussell, R. C. McPhedran, and C. Martijn de Sterke, Three-
dimensional Greens tensor, local density of states, and spontaneous emission
in finite two-dimensional photonic crystals composed of cylinders, Phys. Rev.
E 70, 066608 (2004). — p.61.
[34] V. Prosentsov and A. Lagendijk, The local density of states in finite size
photonic structures, small particles approach, Photonics and Nanostructures
5, 189 (2007). — p.61.
(2002). — p.61.
[36] S. Kole, New methods for the numerical solution of maxwell’s equations,
PhD Thesis, Rijksuniversiteit Groningen, 2003. — p.61, 68.
[37] D. M. Whittaker, Inhibited emission in photonic woodpile lattices, Opt. Lett.
25, 779 (2000). — p.61.
[38] K. Ishizaki, M. Okano, and S. Noda, Numerical investigation of emission in
finite-sized, three-dimensional photonic crystals with structural fluctuations,
J. Opt. Soc. Am. B 26, 1157 (2009). — p.62, 68.
[39] E. Yeganegi, A. Lagendijk, A. P. Mosk, and W. L. Vos, Local density of
optical states in the band gap of a finite one-dimensional photonic crystal,
72 Bibliography
Phys. Rev. B 89, 045123 (2014). — p.62.

Winston, New York, 1976). — p.63, 64.
[41] K. Gilmore, Precession damping in itinerant ferromagnets, PhD Thesis,
Montana State University, 2007. — p.66.
CHAPTER 5
Statistical analysis of time-resolved emission in the
near infrared
5.1 Introduction
Spontaneous emission of light is a fundamental process that plays an essential
role in many phenomena in nature and forms the basis of many applications. As
is described by Fermi’s “golden rule” [1], spontaneous emission of light is not
only a property of the emitter itself but also depends on the environment [2].
To study the light-matter interactions in time, it is essential to study the decay
dynamics of sources. This is obtained from time-resolved emission experiments.
A time-resolved emission experiment consists of a histogram of photon arrival
times over multiple cycles of laser excitation and photon detection. In order to
obtain not only the fluorescence decay time but also the decay curve shape with
sufficient statistics, one needs sufficient dynamic range, corresponding to several
decay times in the time domain.
The use of near infrared spontaneous emission is receiving growing attention
in many fields [3, 4] such as silicon technology [5], in photonic band gap crys-
tals [6, 7], solar energy harvesting [8], lasing [9–11], light emitting diodes [12],
telecommunication [13], opto-electronic devices [14], and has attracted a lot of
interest in biomedical imaging in living tissue [15–17]. There is a growing in-
terest to improve quantum efficiency and tune luminescence toward the infrared
range by changing size and material composition of quantum dots [18, 19] or to
modify and control the near-infrared emission by placing emitters in cavities and
photonic crystals [20].
Unfortunately, however, photon detection in the near infrared is much more
difficult than in the ubiquitous visible range due to the following reasons. Firstly,
from Fermi’s “golden rule” the decay rate of emitters is proportional to the
frequency cubed, ω 3 . Decay rates of emitters in the near infrared range are
therefore much lower than in the visible range [1]. If we consider typical near-
infrared emitters that emit at a wavelength around 1500 nm, the decay rate is
about 27 times lower than for typical visible emitters at 500 nm. As a consequence
the number of collected photons per time unit is much lower in the infrared as
compared to the visible range. Figure 5.1 shows typical time-resolved emission
data in the near infrared and in the visible spectral range. Figure 5.1(a) shows a
decay curve measured for PbS quantum dots inside a 3D photonic band gap in
the near infrared spectral region. It is clear from the figure that the signal level
74 Statistical analysis of time-resolved emission in the near infrared
is very small compare to the background. Figure 5.1(b) shows the decay curve
for R6G dye molecules in the visible spectral region. The signal is typically 3
orders of magnitude greater than the background.
6
x10
100000
6.48 (a) 10000
(b)
6.46 1000
Counts
Counts
100
6.44
10
6.42
1
Nbt/n
0.01 0.1
-16 0 500 1000 1500 2000 0 20 40 60 80 100
Time(ns) Time(ns)
Figure 5.1: Comparison of time-resolved emission experiments in different spectral

ranges. (a) Time-resolved emission measurement of PbS quantum dots in a 3D photonic
band gap crystal in the near infrared range at 0.9 eV. In addition to a high background
level due to detector dark count the signal is very low due to the 3D bandgap. The red
dots at negative times show the counts used to calculate the background. The back-
ground counts are multiplied by the bin-reduction factor used for signal bin-reduction.
The single black square at negative time and the blue dash-line show the average back-
ground value and the error bar shows one standard deviation (σ) of the background.
(b) Time-resolved emission of R6G dye in water in the visible range at λ = 563nm.
The signal level is about 3 orders of magnitude higher than the background level.
Secondly, to obtain a statistically reliable histogram of photon arrival times

the repetition rate of the excitation laser must be at least 5 times lower than the
decay rate under study [21]. The low decay rates in the near infrared limit the
maximum laser repetition rate and thereby further reduce the signal rate.
Thirdly, another difficulty in the near infrared range originates from high back-
ground level due to the high dark count rates of near infrared photon-detectors.
Photo-multiplier tubes typically display 2000 times more dark counts and a 20
times lower quantum efficiency in the near infrared as compared to silicon detec-
tors for the visible ∗ [22]. The background detector counts are subject to Poisso-
nian statistics leading to inevitable fluctuations in signal counts. The standard
deviation of Poissonian noise is equal to the square root of the average number
of events and therefore by increasing the background level the noise level in the
signal also increases.
Therefore due to the above mentioned reasons, in the near infrared spectral
range the photon arrival time histograms are measured with a very low signal
to noise ratio. After measuring the histogram one is interested in obtaining the
∗ Superconducting single photon detectors SSPDs are sensitive to single photons. These
detectors show a low dark count rate in the infrared range yet they are highly specialized
since they require cryogenic temperature [23]
Bayesian analysis of time-resolved emission measurement 75
decay rate, but the huge background level makes this extraction difficult. Since
subtracting a wrong background will lead to a wrong decay rate, estimation of
the correct background level in the near infrared time-resolved experiments is of
key importance. Therefore looking at statistics and including prior information
about background for data analysis becomes significant.
In this chapter we discuss a statistical analysis of photoluminescence mea-
surements done by time-correlated single-photon counting method. We propose
to estimate the background value by using a maximum a posteriori estimation
(MAP) method [24–27]. This powerful method allows us to refine the decay rate
using a priori knowledge about the background. Data are accepted (or discarded)
if the decay curves are modeled assuming a dark rate that lies inside (or outside)
the credibility interval imposed by purely statistical arguments. We illustrate our
method by applying it to the decay dynamics of PbS quantum dots in suspension
and to PbS quantum dots inside 3D photonic band gap crystals, see Fig. 5.1(a).
5.2 Bayesian analysis of time-resolved emission

measurement
5.2.1 Bayesian method
A decay curve F(t; {pi }, B) is a histogram of the photon arrival times t after
many excitation-detection cycles [21, 28, 29]. After measurement, the histogram
is modeled with a decay function f (t; {pi }) from which the parameters ({pi })
such as decay rate of the emission is deduced, plus a true background value B:
F(t; {pi }, B) = f (t; {pi }) + B. (5.1)

In principle B is also a parameter, but since it is a property of the optical setup
we treat it separately. In the simplest case when the system is characterized by
a single decay rate, the decay function is described by a single exponential:
f (t; {γs , I}) = Iexp(−γS t), (5.2)
where I and γS are the model parameters {pi }. Here γS is the emitters decay rate
that we wish to obtain. Figure 5.1 (b) shows an example of such single exponential
decay. A normal fitting routine such as least square can be used to extract model
parameters. If the background level is relatively low compared to the signal level
then background can be one of the model parameters, or the average value of the
background can be subtracted from the data before the analysis. Figure 5.1(b)
is an example of time-resolved emission data with high signal to noise ratio. In
this case the average value of the background is subtracted and the decay curve
is modeled using Eq. (5.2) resulting in I = 50000 and γS = 0.23 ns−1 . However,
in the case which the signal level is very low compared to the huge background as
is shown in figure 5.1(a), the conventional background subtraction may lead to a
wrong parameter determination. It is therefore important to have a background
measurement independent of the model. If the background measurement itself
has a significant statistical uncertainty, however, we must proceed carefully. In

this case a priori information of the background can be used to refine the best
estimate for the decay rate. To include a priori information we use Bayesian
analysis that is one of the fundamental approaches to statistical analysis.
Bayesian analysis is based on Bayes’ theorem [25]. The usual form of Bayes’
theorem is
P (A|B)P (A)
P (A|B) = (5.3)
P (B)
where P (A|B) is the a posteriori probability which is the probability for A after
taking into account B, P (B|A) is probability of finding A under the condition of
B and known as likelihood, P (A) is prior probability of A, P (B) is probability
of a given evidence which is fixed.
For the example of the decay curve Bayes’ theorem can be written as
P ({pi })P (B|Nbt )
P (B, {pi }|{yi }, Nbt ) = P ({yi }|{pi }, B), (5.4)
P ({yi })
where {yi } are the measured data, and Nbt is the measured background. P ({pi })
is the prior distribution of model parameters, and P (B|Nbt ) is the prior distribu-
tion of the background under condition of the measured background which are
both known. P ({yi }) is a normalization constant, P ({yi }|{pi }, B) is the prob-
ability of data given the parameters, which is also known as likelihood of the
parameters, and P (B, {pi }|{yi }, Nbt ) is the a posteriori probability distribution
that we wish to obtain.
In conventional fitting routines, such as least square minimization that is widely
used to model the decay dynamics, the likelihood is maximized and no prior in-
formation is included. Bayesians statistics, however, is concerned with generating
the posterior distribution of unknown parameters, given both the data and prior
information for these parameters. The main equation of Bayesians statistics
known as Bayes’ theorem, combines prior information and the likelihood of the
parameters to reach the posterior distribution for the model parameters, where
this posterior distribution is maximized to extract the parameters of interest. For
data with a low signal-to-noise ratio, a Bayesians analysis can strongly improve
the analysis by including any prior probability distribution of parameters that
are available before the data analysis [26]. In contrast, conventional fitting pro-
cedures only give a most likely parameter estimate and often make the implicit
assumption that the data are normal distributed. Here we present a proper anal-
ysis of time-resolved emission measurement of data with low signal to noise ratio
based on a Bayesian method.
5.2.2 A priori estimation of the background

To use Bayes’ theorem Eq. (5.4), first we need to find a priori information of the
background P (B|Nbt ) given a measured background. We can obtain the prior
information for B from the measured decay curve at t < 0 before the excitation
laser pulse arrives, where we have only background counts. Let us assume that
the true ensemble average dark level (in counts per time bin) is B. If we average
Bayesian analysis of time-resolved emission measurement 77
a number n of background bins, the expected total number of dark counts is

< Nbt >= Bn, which is Poisson distributed. Since the number of counts is large
we may approximate the Poisson distribution with a Gaussian distribution:
1 n −(Nbt −Bn)2 /2Nbt
P (B|Nbt ) = √ √ e . (5.5)
2π Nbt
In other words, the measurement of Nbt causes the probability distribution
√ of B
to become a Gaussian with center Nbt /n and the standard deviation σ = Nbt /n.
The true value of the dark count rate B is expected to lie with 70% probability
in the so-called credibility interval equal to [27]:
√ √
Nbt − Nbt Nbt + Nbt
<B< . (5.6)
n n
5.2.3 A posteriori estimation for model parameter, and

goodness-of-fit
To interpret the time-resolved emission measurements the next step is to find
a good model which explains data. The model validation is one of the most
important steps in the model building sequence. Here we assume for simplicity
that the model is known, namely single exponential decay for a suspension of
light sources, or for light sources embedded in a photonic bandgap crystal.
Assuming Gaussian noise with known standard deviation σ for the measured
data yi , we define the goodness of fit χ2 as
({yi } − {f (ti ; {pi }, B))2
χ2 ≡ , (5.7)
σi2
where yith are the corresponding model points. A direct probability interpretation
is possible [30]: If we suppose the model parameters {pi } are known exactly, then
the probability density of measuring the dataset {yi } is equal to
1 2
P ({yi }|{pi }, B) = e−χ /2
. (5.8)

N √
σi 2π
i
considering that data points are independent. By inserting Eq. (5.8) and Eq. (5.5)
in Eq. (5.4) and considering P ({pi }) and P ({yi }) to be constant, and their ratio
P ({pi })/P ({yi }) = c, we find the overall a posteriori probability function of
background under condition of measured data to be:
n 1 2 2
P (B, {pi }|{yi }, Nbt ) = c √ e−(Nbt −Bn) /2Nbt e−χ /2 , (5.9)
2πNbt √
N
σi 2π
i
This overall a posteriori distribution includes all possible values for all param-
eters {pi } in the parameters’ space. If we subtract the background B from the
decay curve, keep the background fixed, and then minimize χ2 with respect to
all free parameters we obtain the maximum a posteriori distribution of the tru
background and parameters, given the measured data and the background
n 1 2 2
Pmax (B, {pi }|{yi }, Nbt ) = c √ e−(Nbt −Bn) /2Nbt e−χmin /2 . (5.10)
2πNbt
N √
σi 2π
i
Maximizing the a posteriori distribution limits the value of parameters {pi } to

the ones which are the most likely.
In the frequently used fitting routines by minimizing χ2 the probability density
is maximized, while when using Bayes theorem by including the a priori probabil-
ity, a posteriori probability is maximized. The extra information that we include
as prior information in the a posteriori distribution refines the parameters and
therefore gives a more reliable result.
5.2.4 χ2 credible interval and model consistency

The complete procedure for obtaining a parameter estimate from an arrival time
histogram consists of the following three steps: First, we establish the a priori
distribution probability of the background B Eq. (5.5). Second, within the credi-
ble region of background (Eq. (5.6)) we select background values and we calculate
the best fit of the model to the data for each background value. Third, at each
of the background values we calculate χ2min and obtain Pmax (B, {pi }|{yi }, Nbt ).
The maximum of this a posteriori probability defines the best fit value of the
parameters.
χ2 has a probability distribution known as ”chi-square distribution”. As the
degrees of freedom, that is, the number of data used in the modeling procedure-
increase, the chi-square distribution approaches a normal distribution and there-
fore the χ2 credible interval is given by:

2 χ2 2
1− < <1+ . (5.11)
n n n
where n is the degrees of freedom. Time-resolved emission data sets are taken
to be consistent, and thus maintained in the analysis, only if the minimum of χ2
coincides within the credible region of the dark rate (Eq. 5.6) and the value of
χ2 lies within its own credible region (Eq. 5.11).
5.3 Experiment and applications

Time resolved emission of PbS quantum dots in the near infrared spectral range
has been studied. The PbS quantum dots are suspended in toluene with con-
centration of 2.10−6 M/L. The quantum dots are excited by short laser pulses
with a 11 ps duration at a wavelength of λ = 532 nm. The laser was operated
at a repetition rate of 409 kHz. The dots emit at photon energies between 0.8
to 0.9 eV. Time-resolved emission is detected by means of time correlated single
photon counting (TCSPC)[21, 29].
Experiment and applications 79
We have also performed time-resolved emission measurement of the same PbS

quantum dots in 3D photonic band gap crystals [7]. The crystals are made
of silicon and have a wide 3D band gap in the near infrared spectral region
fabricated by CMOS-compatible method [31–34]. The photonic band gap is a
range of frequencies in which light propagation is prohibited and therefore the
spontaneous emission of light sources in the band gap is inhibited [35]. Therefore
in addition to all difficulties mentioned in the introduction, the signal is further
reduced due to a strong inhibition.
5.3.1 Near infrared quantum dots in suspension

To extract the decay rate of the quantum dots in suspension the decay curve
is modeled using a single exponential model with two adjustable parameters,
namely the overall intensity I and the decay rate γS (eq. (5.2)). We first fit
the data using the conventional fitting method without considering any a priori
information. To obtain the background we average over a few hundred time bins
before the excitation laser pulse arrives. Subtracting the average background
and keeping the background fixed results in a decay rate of γ = 1.53μ s−1 with
a goodness of fit χ2red = 0.898.
We now examine the maximum a posteriori (MAP) method as described, to
find the best fit to the data by using a priori information from the background.
The background has a Gaussian distribution with mean value of averaged number
of the background time bins and calculated standard deviation(σ). Due to the
Gaussian probability distribution we expect the true value of the background to
lie within the credible range of the background( Nnbt ) ± one standard deviation
√
with, 70% probability. We start from B = Nnbt − σ, where σ = Nbt /n, subtract
it from the data and fit our single-exponential model to the data. From the fit
we extract parameters and the goodness of fit and we calculate the posterior
probability corresponding to this fit. We change the background in steps of σ/6
and smaller to cover the background confidence interval. For each background
value optimize the rest of the parameters using a least square fitting routine, and
then calculate Pmax (B, {pi }|{yi }, Nbt ).
Figure 5.2 shows the a posteriori probability function Pmax (B, {pi }|{yi }, Nbt )
as a function of background parameter B that is varied within its credible range.
The a posterior probability is low for most B, indicating that these values are
not relevant. Between 0 to σ/6 background values, the posterior probability
is maximum giving the best parameter estimate where the decay rate is γ =
1.53 μs−1 . For other background values the posterior probability is small. The
decay rate changes only from γ = 1.54 μs−1 for Nnbt − σ to γ = 1.52 μs−1 for
Nbt
n + σ background that is not very different compared to the best estimated
decay rate. The goodness of fit lies within its credible range. Therefore the
result of this fitting gives γ = 1.53 μs−1 as decay rate for the most likely fit with
goodness of fit 0.898. As we see the result from the Bayesian analysis matches
very well with the normal fitting and in the both cases we obtain the same decay
rate. Therefore we conclude that when the background level is high and there
is a high signal on top of the background, the results of the conventional fitting
procedure agrees with the Bayesian method.
-1.0 -0.5 0.0 0.5 1.0

2.0
1.0 (a)
Posterior probability
Decay rate ( μ s )
1.5
0.5 1.0
-1
0.5
0.0 0.0
2 (b)
Goodness of fit χred
2
credible
1
range
-1.0 -0.5 0.0 0.5 1.0

Background variation(B/σ)
Figure 5.2: Maximum a posteriori estimation analysis of the time-resolved emission

of the suspension of PbS quantum dots in the near-infrared range. (a) Overall a pos-
teriori probability function Pmax (B, {pi }|{yi }, Nbt ) (black squares) versus background
parameter B normalized to background standard deviation σ. The background is var-
ied within the credible range given by Eq. (5.6). Red circles show the decay rate (right
axis). The probability is maximum for two backgrounds value, determining the most
probable background values and decay rates. (b) Reduced goodness of fit χ2red (blue
inverted triangles) versus B. The black dashed rectangle emphasizes the range of B
where simultaneously the probability is 1 and the goodness of fit is within its credible
range, thus determining the consistent decay rate range.
5.3.2 Near infrared quantum dots in a 3D photonic band gap

In this subsection we examine the method for data sets, where in addition to the
high background the signal level is very low. To probe the density of states in-
side a photonic band gap crystal we immersed the 3D silicon crystal in a diluted
suspension of quantum dots in toluene. The crystal is surrounded by quan-
tum dots suspension, therefore always some emission signal that originates from
the suspension is collected and contributes to the time-resolved data. We have
measured-time resolved emission at different energies within the quantum dot
spectrum where the spectrum overlaps with the photonic band gap. Since the
emission originates both from quantum dots in the crystal and in suspension, we
model the time-resolved emission with a double-exponential [7]:
f (t) = I0 (S · γ0 · exp(−γ0 t) + P C · γ · exp(−γt)). (5.12)

Here γ is the emission rate of quantum dots in the photonic crystal that we
wish to obtain, and γ0 is the rate of dots outside the crystal that is separately
measured to be γ0 = 1.53 μs−1 . S and P C are relative factors fixed by scans
of the detection focus, and only γ and I are adjusted parameters. We call the
ratio of decay rate of dots inside the crystal to the decay rate of dots outside the
crystal (γ/γ0 ) as inhibition [7]. We examine the analysis on two data sets which
correspond to the wavelengths in the band gap of the photonic crystal where the
number of detected photon is very low compare to huge number of background.
Time-resolved emission data was obtained at 0.893 eV, in two different photonic
crystals. Based on Bayes’ theorem we analyzed the data to find the true value
of the background and therefore the correct decay rate.
Figure 5.3(I) shows the a posteriori probability function Pmax (B, {pi }|{yi }, Nbt )
for the 1st set as a function of background parameter B that is varied within its
credible range. By changing the background the γ and therefore the inhibition
changes. The a posteriori probability is low for most B, indicating that these val-
ues are not relevant. Between 0.5 to 0.75, however, the probability is 1, therefore
the concomitant inhibitions appear credible. Indeed, the corresponding good-
ness of fit also lies within its credible range as set by Eq. (5.11). Therefore, we
conclude for this data set that the credible inhibition lies between 2.8 and 17.3,
yielding a final result γ0 /γ = 10.0 ± 7.2.
Figure 5.4 illustrates result of fitting the same data set with two different back-
ground subtractions. Figure 5.4 (a), we subtract Nnbt + 3σ/4 as background, and
fit a bi-exponential function, obtaining γ0 = 17.3γ. The residuals are centered
around zero and minimized χ2red close to 1 and within its credible range, indi-
cating a good fit. As all the conditions are satisfied we accept this value of the
background as true background value. In figure 5.4 (b), we subtract Nnbt − σ/2
as background, and fit with a bi-exponential function, obtaining γ0 = 3γ. The
residuals are not centered around zero and minimized χ2red lies outside the cred-
ible range, indicating a bad fit and a wrong background value as illustrated in
Fig. 5.3(I).
(I) (II)
-1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0
20 10
1.0 (a) 1.0 (a)
8
Inhibition (γ0/γ )
Inhibition (γ0/γ )
15
6
0.5 10 0.5
4
5
2
0.0 0 0.0 0
2 (b) 2 (b)
red
red
2
2
Goodness of fit χ
Goodness of fit χ
credible
1 1 credible range
range
-1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0
Background variation (B/σ) Background variation (B/σ)
Figure 5.3: Maximum a posteriori estimation analysis of the time-resolved emission

of the near-infrared PbS quantum dots in a 3D photonic band gap crystal. I(a) Overall
a posteriori probability function Pmax (B, {pi }|{yi }, Nbt ) (black squares) versus back-
ground parameter B normalized to background standard deviation σ for emission data
obtained at 0.893 eV. The background is varied within the credible range given by
Eq. (5.6). Red circles show the simultaneously resulting inhibition γ0 /γ (right axis).
Red closed circles are credible inhibitions that yield the final result shown as the green
triangle with error bars. I(b) Reduced goodness of fit χ2red (blue inverted triangles)
versus B. The black dashed rectangle emphasizes the range of B where simultaneously
the posterior probability is high and the goodness of fit is within its credible range, thus
determining the consistent inhibition range. II(a) Overall a posteriori probability func-
tion Pmax (B, {pi }|{yi }, Nbt ) (black squares) versus background parameter B/σ. The
background is varied within the credible range. Red open circles show the simultane-
ously resulting inhibition γ0 /γ (right axis). II(b) Reduced goodness of fit χ2red (blue
inverted triangles) versus B. Since the minimum of the χ2red is not within its credible
range, this data set is inconsistent with the assumptions and thus excluded.
Figure 5.4: comparison of one near infrared data set (λ = 1390nm) with different
background subtraction. (a) Subtracting of ( Nnbt + 3σ/4) as background yields the best
fit with residuals centered around zero and a minimized χ2red within credible range.
(b) Subtracting of ( Nnbt − σ/2) as a background yields residuals that are not centered
around zero and a χ2red outside the credible range.
Figure 5.3(II) shows the a posteriori function Pmax (B, {pi }|{yi }, Nbt ) for the
2nd data set of the time-resolve emission measurement of PbS quantum dots in
the photonic bandgap crystal versus background parameter B within its credible
range. A posteriori probability is maximized for background values within 0 to
0.5 as χ2red is minimized. As we explained in Eq. 5.11, χ2 has a credible range and
data is credible if minimum of χ2 is within this range. While here for this data
set the minimum value for goodness of fit is outside of its credible range, hence
this data set is excluded as a possible observation of inhibition in the photonic
band gap.
Using this criteria about 40% of the data sets of time resolved measurements in
the band gap for the near infrared range yielded inconsistent fits, which is only
slightly higher than statistically expected. These results were thus excluded.
Since we only include one standard deviation of the Gaussian distribution, and
probability of finding the true background value within one standard deviation
for a Gaussian distribution is 68%, therefore theoretically with 32% probability
we get an inconsistent fit. There are several possible reasons for the extra in-
consistency. In a number of our time-resolved emission measurements, we have
observed that drifts of alignment and count rate occur, due to environmental
changes during the long hours data collection time. This leads to non-Poissonian
systematic errors in the measured data, giving rise to skewed residuals and el-
evated values of χ2 , and an inconsistency between the dark measurement and
the histogram. In addition, our model for the time-resolved emission in the near
infrared range gives the simplest possible description for emission in a photonic
band gap crystal. Therefore, the model does not capture possible effects such
as inhomogeneous spatial distribution of light sources, or slow non-exponential
decay of the quantum dots predicted by Ref. [36, 37].
5.4 Simulation
Up to now we have investigated the situation where among all parameters only
background has an significant uncertainty. In cases where other parameters also
have uncertainty then we should process those parameters in the same way that
we used here for the background. Here we present a Markovian Chain Monte
Carlo (MCMC) simulation technique that is used in the more general cases where
we have a multi-parameters model and there exist more than one parameter with
uncertainty. Given a set of data, model and prior information for parameters, all
MCMC algorithms generate the posterior probability distribution. The desired
posterior distribution is sampled by constructing a random walk in the model
parameter space, such that the probability for being in a region of the space is
proportional to the posterior density for that region. Here the random walk is
accomplished using a Markov chain. In the case of a decay curve with 3 parame-
ters namely background, decay rate and overall intensity we have a 3 dimensional
parameter distribution, but we can plot such a parameter distribution for each 2
parameters in a 2 dimensional plot. Figure 5.5 shows a 2 dimensional posterior
parameter correlation distribution plot of the background and the decay rate for
the data set shown in Fig. 5.1.
Figure 5.5 illustrates the effect of including prior information of the background
on the posterior parameters distribution. From Bayes’ theorem the posterior
parameter correlation distribution is proportional to product of prior distribu-
tion and likelihood of the parameters. If no prior information is given then the
posterior density region is the same with likelihood density distribution that is
optimized when using conventional fitting routines. Figure 5.5 shows the effect
of including prior information in the analysis procedure. In Fig. 5.5(a) a uniform
distribution has been used for the background B as prior information. A uniform
distribution is a distribution that has a constant probability. Since the uniform
distribution is a flat distribution, applying it as a prior probability distribution
does not change the posterior probability distribution compared to the likelihood
of the parameters that would be maximized in conventional fitting routine. For
this example the posterior probability is widely distributed over a large range of
decay rates with three high probability regions indicated with black arrows. As
we see from Fig. 5.5 even if we estimate the background correctly we can end up
with any value for the decay rate since the decay rate is widely distributed with
high probability.
In figure 5.5(b) the prior background distribution is a normal distribution with
center of the average background value calculated from the data. Multiplying this
prior background distribution to the likelihood of the parameters, pushes the pos-
terior probability down and changes the distribution. The posterior distribution
becomes more concentrated in a more confined range of decay rates indicated
with the black arrow. In this parameter probability distribution we can estimate
the decay rate more precisely since we have only one region of high probability if
we estimate the correct background. Incorporating this a priori information also
illustrates the advantage of using Bayesian method rather than the conventional
fitting routine. This posterior distribution contour also helps to visualize how
Conclusion 85
the parameters are distributed and how they are correlated.
Posterior probability P (B, |{yi},Nbt)
0.5
1
0
40320
(a)
40310
(a)
Background (count)
40300
40290
40320
(b)
(b)
40310
40300
40290
0 0.4 0.8 1.2 1.6
Decay rate (s-1) P(B|Nbt)
Figure 5.5: Posterior probability distribution plotted for two different prior distri-
butions (Eq. 5.9). (a) Prior distribution is a uniform distribution. Since a uniform
distribution is a flat distribution, the posterior distribution in this case is comparable
with likelihood of the parameters that in conventional fitting routine would be maxi-
mized. The black arrows indicate the high probability regions. (b) Prior distribution is
a normal distribution with center of average background value. By applying the normal
distribution as prior the posterior distribution has changed and becomes more concen-
trated in a range of decay rate values. The black circles show maximized posterior
probability of decay curve where background is kept fixed.
5.5 Conclusion
We have investigated the difficulties encountered in time-resolved emission mea-
surements in the near infrared which are caused by low intrinsic count rate and
high intrinsic background. Clearly, extracting parameters from the time-resolved
emission measurement is much more difficult in the near infrared spectral region
than in the visible. We discussed the importance of statistical analysis in the case
of low signal to noise ratio in the infrared range. The high level of background
in the near infrared measurement can lead to a wrong decay time determination,
therefore it is crucial to find the true value of the background. Using Bayesian
method, by incorporating prior information on the background, we are able to
find the true background and estimate the most likely decay rate. This method
provides a robust analysis and interpretation of the near infrared time-resolved
emission when the background level is high and the signal level is low. We have
86 Bibliography
illustrated our method with the successful analysis of time-resolved emission of

PbS quantum dots in suspension and in 3D photonic band gap crystals.
Bibliography
[1] E. Fermi, Quantum theory of radiation, Rev. Mod. Phys. 4, 87 (1932). —
p.73.
Press, New York, 2006). — p.73.
[3] E. Lifshitz, M. Brumer, A. Kigel, A. Sashchiuk, M. Bashouti, M. Sirota,
E. Galun, Z. Burshtein, A. Q. Le Quang, I. Ledoux-Rak, and J. Zyss, Air-
stable PbSe/PbS and PbSe/PbSexS1-x core-shell nanocrystal quantum dots
and their applications., J. Phys. Chem. B 110, 25356 (2006). — p.73.
[4] E. H. Sargent, Infrared quantum dots, Adv. Mater. 17, 515 (2005). — p.73.
[5] G. T. Reed, Elements of modern x-ray physics, second edition (John Wiley,
Chichester, UK, 2008). — p.73.
[6] S. Ogawa, M. Imada, S. Yoshimoto, M. Okano, and S. Noda, Control of light
emission by 3d photonic crystals, Science 305, 227 (2004). — p.73.
a 3D photonic band gap, Phys. Rev. Lett. 107, 193903 (2011). — p.73, 79,
80, 81.
[8] R. D. Schaller and V. I. Klimov, High efficiency carrier multiplication in pbse
nanocrystals: Implications for solar energy conversion, Phys. Rev. Lett. 92,
186601 (2004). — p.73.
[9] K. Inoue, H. Sasaki, K. Ishida, Y. Sugimoto, N. Ikeda, Y. Tanaka, S. Ohk-
ouchi, Y. Nakamura, and K. Asakawa, InAs quantum-dot laser utilizing
GaAs photonic-crystal line-defect waveguide, Opt. Express 12, 5502 (2004).
— p.73.
[10] N. Yamamoto, K. Akahane, S. Gozu, A. Ueta, and N. Ohtani, 1.55
μm-waveband emissions from Sb-based quantum-dot vertical-cavity surface-
emitting laser structures fabricated on GaAs substrate, Jpn. J. Appl. Phys.
45, 3423 (2006). — p.73.
[11] S. Hoogland, V. Sukhovatkin, I. Howard, S. Cauchi, L. Levina, and E. H.
Sargent, A solution-processed 1.53 μm quantum dot laser with temperature-
invariant emission wavelength, Opt. Express. 14, 3273 (2006). — p.73.
[12] N. Tessler, V. Medvedev, M. Kazes, S. Kan, and U. Banin, Efficient near-
infrared polymer nanocrystal light-emitting diodes, Science. 22, 1068153
(2002). — p.73.
[13] C. E. Finlayson, A. Amezcua, P. J. A. Sazio, P. S. Walker, M. C. Grossel,
R. J. Curry, D. C. Smith, and J. J. Baumberg, Infrared emitting pbse
quantum-dots for telecommunications-window applications, J. Mod. Optics.
52, 955 (2005). — p.73.
[14] E. Holder, N. Tesslerb, and A. L. Rogach, Hybrid nanocomposite materials
Bibliography 87
with organic and inorganic components for opto-electronic devices, J. Mater.

Chem. 18, 1064 (2008). — p.73.
[15] A. P. Alivisatos, Less is more in medicine, J. Sci. Am. 285, 66 (2001). —
p.73.
[16] R. Weissleder and M. J. Pittet, Imaging in the era of molecular oncology,
Nature. 452, 580 (2008). — p.73.
[17] X. Michalet, F. F. Pinaud, L. A. Bentolila, A. M. Wu, S. S. Gambhir, and
W. S., Quantum dots for live cells, in vivo imaging, and diagnostics, Science.
307, 538 (2005). — p.73.
[18] R. E. Bailey and S. Nie, Alloyed semiconductor quantum dots: Tuning the
optical properties without changing the particle size, J. Am. Chem. Soc. 23,
7100 (2003). — p.73.
[19] H. Qian, C. Dong, J. Peng, X. Qiu, Y. Xu, and J. Ren, High-quality and
water-soluble near-infrared photoluminescent CdHgTe/CdS quantum dots
prepared by adjusting size and composition, J. Phys. Chem. C 45, 1685216857
(2007). — p.73.
[21] W. Becker, Advanced time-correlated single photon counting techniques
(Springer, New York, 1992). — p.74, 75, 78.
[22] D. Stucki, G. Ribordy, A. Stefanov, H. Zbinden, J. G. Rarity, and T.
Wall, Photon counting for quantum key distribution with peltier cooled In-
GaAs/InP APDs, J. Mod. Opt. 48, 1967 (2001). — p.74.
[23] S. N. Dorenbos, P. Forn-Diaz, T. Fuse, A. H. Verbruggen, T. Zijlstra, T. M.
Klapwijk, and V. Zwiller, Low gap superconducting single photon detectors
for infrared sensitivity, Appl. Phys. Lett. 98, 251102 (2011). — p.74.
[24] B. P. Roe, Probability and statistics in experimental physics (Springer, New
York, 1992). — p.75.
[25] P. M. Lee, Bayesian statistics (Hodder, London, 2004). — p.75, 76.
[26] P. C. Gregory, Bayesian logical data analysis for physical science, a com-
parative approach with mathematica support (Cambridge University Press,
England, 2005). — p.75, 76.
[27] C. P. Robert, The bayesian choice: From decision theoretic foundations to
computational implementation (Springer, New York, 2007). — p.75, 77.
[28] A. F. van Driel, I. S. Nikolaev, P. Vergeer, P. Lodahl, D. Vanmaekelbergh,
and W. L. Vos, Statistical analysis of time-resolved emission from ensembles
of semiconductor quantum dots: Interpretation of exponential decay models,
Phys. Rev. B 75, 035329 (2007). — p.75.
[29] J. Lakowicz, Principles of fluorescence spectroscopy (Kluwer Academic, New
York, 1999). — p.75, 78.
[30] A. Sen and M. Srivastava, Regression analysis (Springer, New York, 1990).
— p.77.
88 Bibliography
— p.79.
[32] L. A. Woldering, R. W. Tjerkstra, H. V. Jansen, I. D. Setija, and W. L. Vos,
Periodic arrays of deep nanopores made in silicon with reactive ion etching
and deep UV lithography, Nanotechnology 19, 145304 (2008). — p.79.
[33] R. W. Tjerkstra, L. A. Woldering, J. M. van den Broek, F. Roozeboom, I. D.
Setija, and W. L. Vos, Method to pattern etch masks in two inclined planes
for three-dimensional nano- and microfabrication, J. Vac. Sci. Technol. B
29, 061604 (2011). — p.79.
Funct. Mater. 22, 25 (2012). — p.79.
electronics, Phys. Rev. Lett. 58, 2059 (1987). — p.79.
Phys. Rev. A 65, 043808 (2002). — p.83.
1557 (2008). — p.83.
CHAPTER 6
Probing photonic crystals with shaped wavefronts
6.1 Introduction
Photonic crystals are periodic structures with a period in the order of the wave-
length of light [1]. In an infinite photonic crystal a wave can propagate only
when it satisfies the Bloch condition due to periodicity. Therefore all modes
in an infinite crystal are Bloch modes. Due to the constructive interference of
light which is reflected from certain families of lattice planes, stop gaps appear:
frequency ranges where no mode propagates inside the crystal in a particular
direction. A real crystal has of course a finite extent as studied in 3 and Ch. 4.
To investigate light transport in a real photonic crystal, light is typically shone
on the crystal from free space. The first interaction of light with the crystal
takes place at the surface, where light can partially couple into the crystal de-
pending on the position where light is launched as well as the wavelength and
polarization [2, 3]. Light that enters the crystal couples to many Bloch modes
and propagates in different directions inside the crystal. In addition to surface
effects, a real photonic crystal has structural variation in size and position of the
building blocks. Although with state of the art fabrication the disorder and de-
fects are very well controlled, fabrication of a perfect periodic structure without
disorder is impossible. These deviations from perfect periodicity cause scattering,
and hence exponential attenuation of coherent beams propagating through pho-
tonic crystals [4, 5]. Therefore in a real photonic crystal with disorder, transport
of the coupled light into crystals results in an interplay between diffuse scatter-
ing due to the disorder and Bloch wave propagation. This fact influences the
industrial applications of photonic crystal devices such as high-efficiency light
emitting diodes [6], low threshold lasers [7], ultralow-loss waveguides, and cor-
rupts the functionality of photonic integrated circuits [5, 8].
Photonic crystals are made either by self-assembly processes such as opal crys-
tals [9–14] or by top-down procedures such as etching or deposition that yields
for instance woodpile and inverse woodpile photonic crystals [15–23]. Artificial
opals are synthesized from a relatively high concentration of nanoparticles in
suspension. Particles in the suspension sediment in time to form a colloidal crys-
tal. When the liquid between the particles is evaporated, one obtains an opal
photonic crystal consisting of close-packed particles with air in between. Even
under controlled conditions such as controlled temperature, humidity, and par-
ticle concentration, disorder is introduced by size polydispersity of the particles.
On top of that during the liquid evaporation cracks may appear in the crystals,
90 Probing photonic crystals with shaped wavefronts
Einl r
Eout
Ed
l
EB
Eout Einr
Etot=Ed+EB
Figure 6.1: Schematic representation of light scattering in a real photonic crystal. The
red circles show the deviations from a perfect crystal and are representing disorder. The
incoupled field Etot consists of Bloch waves EB typical of propagation in a perfect crystal
r l
and the scattered field Ed due to the disorder in the crystal. Ein and Ein are the fields
r l
in free space incident on the sample from right and left, respectively and Eout and Eout
are the outgoing fields in free space on right and left side, respectively.
breaking the large colloidal single crystal into smaller grains, although several
groups have made good progress to “crack-free opals” [24]. If crystals are made
using top-down fabrication methods, the disorder can be controlled within a few
nanometer resolution. Crystals such as inverse woodpiles that have been studied
in this thesis are fabricated with a procedure to reduce the size polydispersity
and position disorder of the pores. Since in this case the crystal size depends
on the mask size and the fabrication procedure, the crystal is a single crystal
throughout the entire crystal volume.
In this chapter we study different processes involved in light transport in
both opals and two-dimensional silicon photonic crystals using a correlation mi-
croscopy method. We aim to distinguish the two different processes of light
propagation in photonic crystals after it has entered inside the crystal. We de-
note by Etot the total field inside the crystal transmitted through the surface,
which consists of the sum of Bloch wave propagation typical of a perfect crystal
with field EB , and randomly scattered light due to disorder with field Ed :
Etot = tsurf Ein = EB + Ed (6.1)
where tsurf is the fraction of light that enters to the crystal through the sur-
face, and Ein is the incident field. Figure 6.1 shows a schematic representation
of incident and transmitted field in a real photonic crystal sample. Inspired by
mesoscopic wave propagation in optics or in electronics, we propose to use the
scattering matrix representation to consider the scattering matrix of the crys-
tal [25] to quantitatively interpret our results.
l l
Eout Ein
r = S r (6.2)
Eout Ein
r l
with Ein and Ein the fields in free space incident on the sample from right and
r l
left, respectively, and Eout and Eout are the outgoing fields in free space on the
Wavefront shaping setup and samples 91
right and left side, respectively. S is the scattering matrix equal to

l
R T rl
S= (6.3)
T lr Rr
with Rl the reflection matrix on the left side of the sample, Rr the reflection
matrix on the right side of the sample, T rl the transmission matrix from right to
the left and T lr the transmission matrix from left to the right. In this chapter
only Rl is considered. For simplicity we represent Rl as R and call it the reflection
l l
matrix of the sample, where Eout = REin . Using singular value decomposition,
the reflection matrix R can be written as a product of three matrices [26]
R = U RU † . (6.4)
†
Here, the matrices U and U are the unitary matrices with complex elements
which represent the incoupling to the crystal from free space and the outcou-
pling from the crystal to free space, respectively, and R is a diagonal matrix
with real, positive elements which are singular values corresponding to reflection
eigenchannels of the crystal. We propose R is an addition of singular values
RB originating from the stop gaps, and singular values Rd due to the disorder.
The contribution of a Bloch mode to the maximum reflectivity channel is given
by the innerproduct of the spatial field profile of the Bloch mode with the field
distribution of the channel in the spirit of Ref. [27].
By measuring the reflected field from the crystal we observe optical correlations
in the wave structures which are attributed to the Bloch propagating field EB
and scattered field due to disorder Ed . The disorder in a crystal, such as size or
position variations, results in the scattering of a Bloch mode. The scattered light
is likely distributed over all other Bloch modes. This picture assumes that the
scattering mean free path is greater than the Bragg length sc > Bragg , which is
physically reasonable since otherwise a photonic crystal is so disordered that it
does not reveal gaps anymore, see Ref. [5]. Since disorder is randomly distributed
throughout the crystal, the Ed fields are Bloch modes, their contribution is ran-
dom with position in the crystal. Here we neglect the waves that have a very
large Bragg length and therefore propagate a long distance inside the crystal be-
fore they couple to a Bloch mode. Our aim is to distinguish the fraction of EB
and Ed in different photonic structure and understand the complex behavior of
light propagation in photonic crystal due to long range order and disorder in the
structures. We investigate short range and long range order in both opal and 2D
silicon crystals and we compare our results with a completely random structure.
We also probe the transport of light in the crystals with different wavefronts us-
ing wavefront shaping method. The wavefront shaping and focusing of light into
the crystal help to understand how light goes inside and how it behaves inside a
real crystal.
6.2 Wavefront shaping setup and samples

The measurements described in this chapter have been performed using a setup
which is notably used for wavefront shaping experiments [28]. Figure 6.2 shows a
/ z x
Sample
2
PLZ
CCD
/2
Figure 6.2: Schematic illustration of the wavefront shaping setup. A laser with wave-
length λ = 561 nm is used as light source. The laser beam passes through a half wave
plate (λ/2) and a polarizer beam splitter (P BS) to obtain a known incident polariza-
tion. The beam is expanded using a beam expander to cover an area of 8 mm by 8
mm on the spatial light modulator (SLM). Light that is reflected back from the SLM
passes through L1 lens and is spatially filtered using an aperture. The laser light is then
focused on the sample surface using a high numerical aperture objective with N A = 0.9.
A camera is in the imaging configuration to image the sample surface. A polarizer and
a quarter wave-plate are placed in the reflection path to filter the direct reflection from
the sample surface. For alignment purposes a white light source illuminates the sample
surface.
schematic illustration of the setup. A λ = 561 nm cw laser is used as light source.

The laser beam passes through a polarizer beam splitter (PBS) and a half wave
plate (λ/2) to obtain a known incident polarization. The laser beam is expanded
using a beam expander to cover a 8 mm by 8 mm area on the spatial light
modulator (SLM, Holoeye phase only). In case the SLM is “on” we modulate
the wavefront of light to a desired wavefront, otherwise it is “off” and acts as
a reflecting mirror. Light that is reflected back from the SLM passes through
L1 and is spatially filtered using an aperture. The laser light then is focused on
the sample surface using a high numerical aperture objective with NA=0.9. The
diameter of focus is calculated to be about 300 nm. A CCD camera (Dolphin
F-145B) is in the imaging configuration to image the sample surface. A polarizer
and a quarter wave-plate are placed in the reflection path to filter the direct
reflection from the sample surface as well as single single scattering from the
sample [29, 30]. For alignment purposes a white light source is used to illuminate
the sample surface [28].
In the experiments we have studied synthesized opals made of silicon dioxide
colloidal spheres with radius R = 349 nm grown on silicon wafer, as studied
Multiple scattering in presence of long-range order 93
(a) (b)
2μm 1μm
Figure 6.3: (a) Scanning electron micrograph of a 3D opal crystal of silicon dioxide
spheres with radius R = 349 nm grown on a silicon wafer. The opal crystal consists of a
lot of small islands separated by grain boundaries. (b) Scanning electron micrograph of
top-down fabricated two dimensional silicon photonic crystal with centered rectangular
lattice.
in Ref. [31, 32]. The opals have a stop gap of only 5% relative band width
in the Γ − L direction. Figure 6.3 (a) shows a scanning electron microscope
image of an opal crystal grown on a silicon wafer. The opal crystal consists of
small grains of typically 5μm in diameter which are created during the growth
process. The λ = 561 nm laser light used for the illumination has a energy
close to the second stop gap of the opal crystal. To compare samples with small
crystal domains with large single crystals, we also perform measurements on two-
dimensional (2D) silicon crystals with centered rectangular lattice as shown in
Fig. 6.3 (b). The 2D crystal is realized by etching pores using reactive ion etching
in silicon wafers [21, 33]. The large single crystal is defined by deep UV step scan
lithography [33]. The measurements on 2D crystals are done in the perpendicular
out of plane direction with zero in-plane wavevector where the crystal has a stop
gap at the Γ point. The λ = 561 nm laser light used for the illumination has a
higher energy than the 2D silicon stop gap.
The sample is fixed on a two dimensional stage that has a travel range of 100
μm by 100 μm with a step size of 100 nm in the plane perpendicular to the optical
axis of the objective. For the focusing and alignment purpose the objective is
mounted on a manual stage with accuracy of 0.1 μm.
For the reference measurements we have used a completely random and dis-
ordered sample of polydisperse ZnO nanoparticles with average grain size of 200
nm. The ZnO sample is a 10 μm thick layer of dense nanoparticles deposited on
a microscope glass slide by spray painting [28]. Considering the transport mean
free path of the ZnO layer that is 0.7 ± 0.2 μm this thickness is sufficient to make
the sample multiple scattering.
(a) (b) (c)

1 μm 1 μm 1μm
1000 3000
2000
600
1000
200
Figure 6.4: CCD camera images of the opal and ZnO samples illuminated with laser
light. (a) and (b) recorded images at two different positions on the opal. The CCD
images show a main bright intensity spot in the middle surrounded with random scat-
tered light with no obvious periodicity. (c) Speckle pattern recorded from ZnO sample
when illuminating with the laser light.
6.3 Multiple scattering in presence of long-range

order
In this section we will distinguish light propagation due to long-range order and
disorder in real photonic crystals. First, we use opal photonic crystals grown
on the silicon wafer for the experiments. Collimated plane waves are sent to
the back aperture of the illumination objective to focus light on the surface of
the photonic crystal. The Gaussian beam that is sent to the sample partially
couples into the Bloch waves in the crystal and propagates inside the crystal
in different directions and partially scatters to other Bloch modes and change
direction due to scattering by disorder. The recorded reflected intensities at two
different positions on the crystal surface are shown in Figs. 6.4 (a) and (b). The
images consist of a bright focus in the middle surrounded by scattered light.
Fig. 6.4 (c) is the reflected speckle pattern from the ZnO sample. Although the
scattered light from the photonic crystal sample looks (very) random due to the
contribution of Ed , it is strikingly different from speckle pattern obtained from
ZnO scattering layer. First of all due to the smaller amount of randomness in
the crystal, the mean free path is longer resulting in a deeper traveling of the
light inside the sample and a slower diffusion compared to the ZnO sample, thus
the whole intensity pattern from the crystal has a smaller area than the one from
the ZnO. Secondly, in the middle of the crystals’ pattern a main reflected spot is
seen that is absent in the ZnO sample. The main reflected spot is mostly caused
by the periodicity in the crystal.
To quantitatively compare the scattered light patterns we have plotted the
intensity distribution of the scattered light obtained from the opal crystal and
the ZnO layer in Fig. 6.5. The distribution of speckle intensity of a random
multiple scattering medium shows an exponential trend with the intensity. It is
known that the distribution of the speckled light from random media follows a
Rayleigh distribution [34]. We therefore fit a Rayleigh distribution model to the
data with a mean value < I >= 108.5 calculated from the data, that matches
excellently to the Rayleigh statistics. To compare the intensity distribution from
Intensity correlations in presence of long-range order 95
Histogram of intensity
−2
10
distribution
−4
10
−6
10
Relative 15
residual
0
−5
0 500 1000
Intensity (counts)
Figure 6.5: Histogram of the intensity distribution from the ZnO (black circles) and
opal photonic crystal (red rectangles). The dashed gray line and the solid red line
show the Rayleigh statistics and the bottom panel shows the residual; deviation of each
distribution from the Rayleigh statistics.
the photonic crystal with the ZnO layer, we plot the histogram of the intensity
distribution obtained from the photonic crystal sample. A Rayleigh distribution
with mean value < I >= 121.9, obtained from the intensity histogram, is fit-
ted to the data. In the low intensity limit below 500 counts, the distribution
follows the Rayleigh statistics, manifesting the random intensity distribution of
the low intensity light originating from the randomness in the crystal. At higher
intensity above 500 counts, a striking deviation from the Rayleigh distribution
is observed. The relative residuals in the bottom panel of Fig. 6.5 show the de-
viations from the Rayleigh distributions more clearly. The goodness of fit for
the ZnO sample is χ2 =0.51 showing a very good agreement with the Rayleigh
statistics and for the opal crystal is χ2 =6.4 indicating a huge deviation from the
Rayleigh distribution. These observations confirm that the high intensity spots
observed in the reflected light (see Fig. 6.4) originate from the periodicity in the
crystal Eb as they do not match the random distribution. For an ideal photonic
crystal without imperfection, multiple intense peaks from different Bragg planes
are expected that do not necessarily follow the Rayleigh statistics.
6.4 Intensity correlations in presence of long-range

order
The deviations from Rayleigh statistics are a clear sign of correlations in the
intensity distribution of the photonic crystal sample. To quantify these corre-
lations we have studied the auto-correlation of the scattered light from both
random scattering sample and the opal photonic crystal. The auto-correlation
[I I] of the camera image of the reflected light with itself is given by:
∞ ∞
[I I](Δx, Δy) ≡ dxdy I(x, y)I(x + Δx, y + Δy) (6.5)
−∞ −∞
where I(x, y) is the measured reflected intensity, and x and y are the spatial
coordinates.
Fig. 6.6 (a) and (b) show the auto-correlation plots of the images taken at two
different positions on the sample separated by two lattice constants Δx = 2a
where we clearly see a hexagonal pattern with a periodicity that matches with
the crystal periodicity. The two images are very similar except that one is slightly
rotated with respect to the other one. This might be due to the fact that the
translation is not along the crystal axis and therefore after translating 2 lattice
constants we do not arrive at a similar position as before translation. In these two
plots contribution of EB is clearly seen, resulting in a periodic wave structure. In
Fig. 6.6 (c) the auto-correlation averaged over one crystal unit-cell is presented.
Here we moved the crystal 700 nm to the next unit-cell in steps of 100 nm. At
every step the reflected light is recorded, the auto-correlation of each image is
calculated, and eventually the auto-correlations are averaged resulting in Fig. 6.6
(c) in which the periodicity is partially averaged out. This is because due to
complicated profile of the fields in 3D crystals, at every position in the unit-cell
different Bloch modes can couple and propagate inside the crystals. However,
these Bloch modes at different positions are still correlated, resulting in a different
periodicity in the averaged auto-correlation.
Figure 6.6 (d) shows the auto-correlation of the ZnO speckle pattern. The
speckle pattern is recorded when the ZnO sample is placed in the setup with the
same configuration and is illuminated with the laser light. The auto-correlation
looks very different from the auto-correlation obtained from the photonic crystal
sample. The decorrelation here is much slower than for the photonic crystal
sample. This is because the speckle pattern from the ZnO sample is much wider
than the scattered light from the photonic crystal. The bright peak in the middle
is also present here. The width of the bright peak in the middle is the same for
both crystals and the ZnO sample and is set by the smallest feature that can
be resolved with the setup and therefore it is a diffraction limited spot. As the
light transport in the ZnO sample takes place with random multiple scattering,
no periodic correlation is seen.
6.5 Response of photonic crystals to a shaped

wavefront
In order to probe both the surface and the depth inside the crystal we have used
two different approaches. In the first approach we send a Gaussian beam focused
on the crystal surface and measure the crystal response to this wavefront as
presented in the previous sections. Here the Gaussian beam can be represented
as many different plane waves,as modes in free space, with different wave vectors
that are sent to the photonic crystal. In the second approach we use wavefront
shaping. Wavefront shaping is a technique pioneered in our group [35, 36], that
is capable of controlling waves in space and time for imaging and focusing in
complex media such as random scattering materials. Using feedback algorithms
the illumination wavefront is optimized such that the scattered light is focused
Response of photonic crystals to a shaped wavefront 97
1 1
(a) (b)
1 μm 0.8 1 μm 0.8
0.6 0.6
y
y
0.4 0.4
0.2 0.2
0 0
x x
1 1
(c) (d)
1 μm 0.8 1 μm 0.8
0.6 0.6
y
y
0.4 0.4
0.2 0.2
0 0
x x
Figure 6.6: The auto-correlated image of the scattered light from the opal photonic
crystal and the ZnO layer. (a) and (b) auto-correlations at two different positions on
the opal that are 2 lattice constants apart from each other. (c) The averaged auto-
correlation over one crystal unit-cell.(d) Auto-correlation of speckle pattern from the
multiple scattering ZnO sample.
after passing through the sample or even when reflecting back from the sample.
This method is very flexible in sending a desired wavefront into the crystal.
The wavefront is modulated using a spatial light modulator (SLM). The main
difference between the two approaches is that in the first approach the wave front
has a smooth phase, while in the second approach the wavefront has random
phase. The details of each approach are discussed below including advantages
and disadvantages of each method.
6.5.1 Focusing light on the surface

To study the light propagation at different positions in the sample we translate
the sample in the xy plane perpendicular to the optical axis, while everything
else is fixed. We record one camera image per translation step. Figure 6.7
blue dashed line shows the result of maximum intensity (at every translation
step) as a function of translation when we send a Gaussian beam to the sample.
The maximum intensity of the camera image shows periodicity that matches
with the lattice periodicity. This indicates that the wave structure repeats itself
after translating over one period. Here we eliminate the direct surface reflection
6000
Focus, without wavefront shaping
5000 Focus, with wavefront shaping
Peak Intensity (counts)

4000
3000
2000
1000
a
0
−5 0 5
Translation (μm)
Figure 6.7: The maximum peak intensity of the focus on the sample versus the trans-
lation. The translation is over a distance of 10 μm with steps of 100 nm. The blue
dashed line is the the maximum focus intensity without wavefront shaping. The unit-
cell modulation of the intensity is clearly seen. The red curve is the maximum intensity
of the wavefront shaped focus, showing less periodicity.
and single scattering events by using a cross polarization detection to filter the
polarization maintained lights. Therefore the recorded intensity is a combination
of EB the light that is reflected from the periodic structure and Ed the scattered
light by the randomness. Thus in Fig. 6.7 the observed periodicity is assigned to
EB , and the deviations from the periodic intensity pattern are attributed to Ed .
6.5.2 Wavefront shaping

We performed wavefront shaping on both opal and silicon inverse-woodpile pho-
tonic crystals. Light that is sent to the sample has the optimized wavefront to
make a bright focus on a desired position on the crystal surface. The reflected
light is recorded with a camera which is in the image plane of the sample surface.
Wavefront shaping redistributes most of randomly scattered light to the focus,
producing a new single bright focus different from the initial focus. Figure 6.8
shows the optimized focus on the crystal surface. The random scattered light
around the focus are redistributed to the focus, resulting in an enhanced focus.
After light is focused on the sample surface we translate the sample in lateral
direction with respect to the illumination objective. The crystal is translated
over a distance of 10 μm with steps of 100 nm. Again at every step the camera
records images of the sample’s surface. From the position of the first optimization
we first translate to the right in x direction, then the sample is moved back to
its initial position, and then we translate the sample to the opposite direction.
To compare the wavefront shaped illumination with the non-wavefront shaped
one, we repeat the same measurements shown in Fig. 6.7 at the same position on
the crystal. The red curve shows the maximum intensity of the wavefront shaped
Optical correlations in photonic crystals 99
(a) (b) 1200

200
1 μm 1 μm 1000
150 800
y y 600
100
400
50
200
0 0
x x
Figure 6.8: The intensity map of the scattered light from the surface of an opal pho-
tonic crystal, when (a) a plane wave is sent to the objective and is focused on the
sample, and (b) a shaped wavefront is sent to the objective and an optimized focus is
made on the sample surface.
focus versus translation. The observed periodicity matches well with the periodic
signal from the non-wavefront shaped focus. The maximum intensity of the
wavefront shaped data is more random than the non-wavefront shaped one. Since
the wavefront shaping optimizes the light propagating through the maximum
reflectivity channels in the sample it therefore affects the random scattered light
Ed rather than the Bragg reflected light EB . This results in increasing the effects
from the randomness in the sample and thus less periodicity as seen in the figure.
The wavefront shaping can be specially a more powerful method when a sample
has a broad stop gap or a band gap since then the light can not propagate deep
inside and probe the complex internal structure of the crystal.
Figure 6.9 shows two different wavefront shaping scans on different initial po-
sitions on the sample. The wavefront is shaped at the zero translation position,
and it is kept fixed while translation. The red curve shows clear periodicity on
both sides, but for the blue curve, periodicity disappears by translation on the
left side. We believe that for these translations on the sample surface we have
moved from one grain to a neighboring grain, where the grain has a different
crystal orientation as the initial one. It is remarkable that the intensity does not
quickly drop with the translation, as in a random medium.
6.6 Optical correlations in photonic crystals

In our study we develop a new method to measure how identical the wave struc-
ture inside a photonic crystal is. Our method is capable of probing deep inside
the crystal and finding the structural wave correlations in photonic crystals in-
dependent of the crystal structure. We have measured the long range order
of waves in both self-assembled opal photonic crystals and top-down fabricated
silicon photonic crystals.
We make a 2D raster scan on the crystal surface. The raster scan is automated
and is done by translating the sample in rows parallel to each other. With this
1500
Maximum focus intensity (counts)

1000
500
0
−5 0 5
Translation (μm)
Figure 6.9: Maximum focus intensity for two different wavefront shaping realizations
on different positions on the opal crystal. The red curve shows periodicity on both
sides, confirming the similarity in the crystal structure over a distance of 10 μm. The
periodicity is indicated with a double sided arrow. The blue curve shows periodicity
only in one side and a drop in the intensity on the other side. This is probably because
on the left side the focus has moved to a different grain.
scan we cover an area of 16 by 16 μm with translating the sample in steps of

100 nm and recording one camera image at every step. Such a scan including
recording the camera images takes about 48 hours. In order to have a stable
alignment the sample was mounted in the setup 6 hour prior to the measurements
and a stability check was performed by measuring the speckle auto-correlations
in time. The intensity fluctuations of laser source is measured to be only 0.4%.
After the data are collected, we choose one camera image of the reflected
light taken at an arbitrary position (n, m) on the sample surface In,m and cross
correlate all other images In ,m with this image. The cross correlation function
is written as follows:
∞ ∞
[In,m In ,m ](Δx, Δy) ≡ ∗
dxdy In,m (x, y)In ,m (x + Δx, y + Δy) (6.6)
−∞ −∞
This enables us to make a map of cross correlation over the entire translation
area. Figure 6.10 shows the cross-correlation maps for both the opal and the
2D silicon crystal. Depending on the initial image for the cross correlations
we get different cross correlation maps. Light reflected at positions of higher
correlations are more similar to each other than the light reflected from positions
of lower correlations.
Figure 6.10 (a) shows the cross correlation for the opal photonic crystal. An
overall view on the figure shows periodicity of a=0.6 μm that matches the crystal
periodicity. In the middle of the scan an island of high correlation, more that
0.9, is seen. The island is extending in x direction from x=-8 to x=6 μm and in
y direction it is confined between y=-4 μm to y=2 μm. This shows that within
one grain the wave structure is more similar than the neighboring grain separated
(a) (b)
8 1 4 1
0.9
0.95
Δy (μm)
Δy (μm)
0.8
0 0
0.7
0.9
0.6
-8 -4
-8 8 -4 0 4
Δx (μm) Δx (μm)
Figure 6.10: Cross correlation of the camera images of the reflected light from the opal
and the silicon photonic crystal when illuminating the samples with laser light. (a) Cross
correlation of opal photonic crystal over an area of 16 μm by 16 μm. Individual particles
are seen in the image that look like small spheres with diameter of about 0.6 μm. In
the middle of the image an area of high correlation is seen indicating one grain in the
opal crystal. The dark blue lines between particles correspond to the grain boundaries
in the opal (indicated by black arrows). (b) Cross correlation of silicon photonic crystal
over an area of 8 μm by 8 μm. Periodicity is seen in the entire range which matches
with the crystal periodicity at both directions. The black circles indicate the starting
points of the cross correlations (n,m).
with grain boundaries. The crystal grain boundaries are observable in dark blue
e.g., as indicated with black arrows. Fig. 6.10 (b) shows the cross correlation for
the 2D silicon crystal. The map shows periodicity in the entire image. For this
crystal depending on the scan direction the periodicity is different. From the SEM
images the two lattice constants are calculated to be a = 693±10 and c = 488±11
nm and our observations here matches with the crystals’ parameters. At some
part in the image the correlation is lower than the other parts. This might be due
to some drift in the setup during the long scan, or the sample surface has a small
curvature resulting in an out of focus of the sample. Since the correlation scan is
done symmetrically with respect to the y=0 line, an instability can not cause an
asymmetric correlation, we therefore attribute the low correlation region to the
samples’ surface.
To compare the correlation from the two different samples we look at one line
profile of the cross correlation. For reference we repeat our experiment on a
random ZnO layer and plot the cross correlation as a function of translation.
Figure 6.11 shows the one line cross correlation for the silicon photonic crystal,
the opal crystal and the random scattering layer of ZnO. The reference ZnO
decorrelates to a level of about 0.2, and no periodicity is seen, as expected for
random media. The 0.2 level is still a high correlation for ZnO compared to the
correlation of less than 0.1 in the transmission configuration. Since the ZnO layer
has a glass substrate we think that the reflection from the substrate increases
0.8
Cross Correlation 0.6
0.4
0.2
Silicon PhC Opal PhC ZnO
0
−5 0 5
Translation (μm)
Figure 6.11: One line profile of the cross correlation of the reflected light shown in
Fig. 6.10 for opal and silicon crystals, and random scattering sample of ZnO. For both
crystals periodicity is seen which matches with the crystal periodicity. The decorrelation
in the silicon photonic crystal is only about 0.14, indicating a very good crystal quality.
For the opal crystal correlation is to a minimum of around 0.6, which shows contribution
of more disorder compared to the silicon crystal. For reference cross correlation of
random scattering sample of ZnO is presented, showing a decorrelation of about 0.8.
the correlation. For the opal crystal the correlation level is as high as 0.6, and
periodicity is seen over the entire range of the translation. This higher level
of correlation compare to the random ZnO layer is due to contribution of the
periodicity EB . Remarkable results are obtained from silicon photonic crystals.
The silicon crystal shows clear periodicity on top of an even higher correlation
level of 0.9. This high correlation level is obtained by the major contribution
of EB . The single crystalline structure and the extreme control of disorder in
top-down fabrication method of the silicon photonic crystal results in an nearly
perfect crystal structure.
The lower correlation in opal compared to the silicon crystal is expected due
to multiple grain crystalline structure of the opal and size polydispersity of the
nano-particles used in the synthesize which yield a grater contribution of Ed in
the opal crystal. Since the field due to disorder Ed is random with respect to
position it does not contribute to the correlations in the structure, therefore the
more disorder in the structure leads to a higher Ed and a lower correlation. Since
the focus size is in the order of one crystal unit cell or smaller, we believe that in
this experiment we probe both the long range and the short range correlations.
In the next section we quantitatively distinguish the decorrelation due to the
short range and long range order.
In the correlation map we usually have a high correlation peak in the middle
then there is a tail with a slight decorrelation over a long translation range on
top of a relatively constant correlation level. In order to have a measure for
the correlation in the photonic crystal we define a correlation length, which is
Cross correlation
0.5
Silicon PhC
Opal PhC
Fitted curve
0
−5 0 5
Translation (μm)
Figure 6.12: One line cross correlation for both opal(red dashed line) and silicon (blue
solid line) photonic crystal, and the fitting curves. The opal photonic crystal shows a
faster decorrelation, indicating a shorter correlation length.
the length where the correlation decreases to 1/e of the maximum cross corre-
lation [37]. To obtain the correlation length we exclude the central peak and
fit the tail with an exponential model. For the fitting only the maxima of the
cross correlations has been taken into account. Figure 6.12 shows the fitted curve
together with the cross correlation curve. The model has two free parameters
namely the correlation length and the overall correlation. By averaging the cor-
relation lengths obtained from several fits at different cross sections, we obtain a
correlation length of lcor = 26 ± 1.7 μm for the opal and lcor = 95 ± 3 μm for the
silicon photonic crystal. The reason that the lcor for the opal crystal is longer
than the averaged grain size of the opal is that grains have the same directionality
and therefore they are correlated. For the 2D silicon crystal due to single crystal
structure we expect a longer lcor . The reason for observing a shorter lcor can
be due to the short scan range in our measurement, much shorter than the lcor ,
that causes an uncertainty in calculating the lcor . From this we conclude that the
silicon photonic crystal has about 4 times less disorder than the synthesized opal
photonic crystal. This disorder includes both size polydispersity and position
variation.
6.6.1 Long range correlation in opal photonic crystals

Most opal crystals exist in polycrystalline form rather than as a single crystal.
They are composed of hundreds of grains (small crystals) packed together to fill
all space. Each individual grain has a different orientation than its neighbors. Al-
though long-range order exists within one grain, at the boundary between grains,
the ordering may change direction. For the opal crystals grown on silicon, the
SEM images show that the long range order is mostly maintained since different
grains have the same directions. In order to measure the long range correlation
in photonic crystals we place the sample out of the objectives’ focus so that the
incident light on the sample covers an area with a diameter of 3 μm. We then
translate the sample, record images at every step, and perform the cross corre-
lation analysis. The large focus probes a larger area in the sample and therefore
is less sensitive to the short range order in the sample. The cross correlation
of the scattered light from the out of focus sample is plotted together with the
in-focus sample in figure 6.13. Interestingly the cross correlation peak broadens
but the correlation level indicated by B drops only by 0.05. This suggests that
this correlation level is due to the long range order in the crystal and as expected
is not sensitive to the focus size.
1
Cross correlations
0.8
0.6
0.4
B
0.2 Focused
Defocused (3μm)
0
−6 −4 −2 0 2 4 6
Translation (μm)
Figure 6.13: Cross correlation of the reflected light over one line translation for in
focus (red dashed line) and out of focus opal sample (blue solid line). The correlation
is on top of correlation background B, and does not depend on the focus size.
From these measurements we conclude that the high correlation level of the
silicon photonic crystal is due to the long range order as it is expected due to
monocrystalline form of the top-down crystals. For the opal crystals the long
range order is reduced due to slight changes in the crystal directions in different
grains. For both crystals the deviation of the cross correlation from one, is the
result of random scattering Ed .
6.7 Summary and conclusion

In this chapter we studied the light transport inside real photonic crystals. A
real photonic crystal is of finite extent and therefore has interfaces with free
space. Since only Bloch waves can propagate inside a photonic crystal the illumi-
nated wave front to the crystal couple to the Bloch modes. Due to the photonic
crystal surface impedance the external wavefront can not couple to all available
Bloch modes. After the light is partially coupled to the crystal it propagates
inside the crystal with two different processes: the Bloch propagation due to
Bibliography 105
perfectly periodic crystal, and randomly scattered light due to disorder. Our
remarkable results about probing the photonic crystal structure with correlation
microscopy method, show the interplay between random scattering Ed and the
Bloch propagation EB . We have studied the histogram of intensity distribution
of the reflected light from photonic crystals, which shows clear deviations from
the Rayleigh distribution expected for random scattering matterials. Until now
intensity statistics have not been reported for photonic crystals. To the best of
our knowledge our study here is the first study of intensity distribution from
ordered photonic structures. From the auto-correlation of the reflected inten-
sity we have observed the contribution of EB giving rise to a periodic pattern
in the auto-correlation with a period that matches the crystal unit-cell length.
We have illuminated the crystals with a wavefront shaped light, and we observed
more contribution from the crystal disorder. The auto-correlations of the re-
flected light from both opal and silicon crystals yield in a high correlation level
with a periodic correlations. The periodicity and the high correlation level are
attributed to the EB and the small decorrelation from 1, which is expect for a
perfect crystal is caused by Ed . The results are very complex due to the fact
that photonic crytals are a combination of both order and disorder structures.
We believe that our results are a major step in understanding the light transport
in real photonic crystals.
Bibliography
2008). — p.89.
[2] M. Notomi, Theory of light propagation in strongly modulated photonic crys-
tals: Refractionlike behavior in the vicinity of the photonic band gap, Phys.
Rev. B 62, 10696 (2000). — p.89.
[3] M. Doosje, B. J. Hoenders, and J. Knoester, Scattering of light on the sur-
faces of photonic crystals, Opt. Commun. 206, 253 (2002). — p.89.
[4] S. Fan, P. R. Villeneuve, and J. D. Joannopoulos, Theoretical investigation
of fabrication-related disorder on the properties of photonic crystals, J.Appl.
Phys. 78, 1415 (1995). — p.89.
[5] A. F. Koenderink, A. Lagendijk, and W. L. Vos, Optical extinction due to
intrinsic structural variations of photonic crystals, Phys. Rev. B 72, 153102
(2005). — p.89, 91.
[6] J. J. Wierer, A. David, and M. M. Megens, III-nitride photonic-crystal light-
emitting diodes with high extraction efficiency, Nature Photon. 3, 163 (2009).
— p.89.
[7] M. Loncar, T. Yoshie, A. Scherer, P. Gogna, and Y. Qiu, Low-threshold
photonic crystal laser, Appl. Phys. Lett. 81, 2680 (2002). — p.89.
[8] S. Hughes, L. Ramunno, J. F. Young, and J. E. Sipe, Extrinsic optical scat-
tering loss in photonic crystal waveguides: role of fabrication disorder and
photon group velocity, Phys. Rev. Lett. 94, 033903 (2005). — p.89.
106 Bibliography
[9] J. V. Sanders, Colour of precious opal, Nature 204, 1151 (1964). — p.89.
[10] H. Mıguez, C. Lopez, F. Meseguer, A. Blanco, L. Vázquez, R. Mayoral,
M. Ocana, V. Fornés, and A. Mifsud, Photonic crystal properties of packed
submicrometric SiO2 spheres, Appl. Phys. Lett. 71, 1148 (1997). — p.89.
[11] K. Lin, J. C. Crocker, V. Prasad, A. Schofield, D. A. Weitz, T. C. Lubensky,
and A. G. Yodh, Entropically driven colloidal crystallization on patterned
surfaces, Phys. Rev. Lett. 85, 1770 (2000). — p.89.
[12] O. D. Velev, A. M. Lenhoff, and E. W. Kaler, A class of microstructured
particles through colloidal crystallization, Science 287, 2240 (2000). — p.89.
[13] G. A. Ozin and S. M. Yang, The race for the photonic chip: colloidal crystal
assembly in silicon wafers, Adv. Func. Mater. 11, 95 (2001). — p.89.
[14] P. Yang, T. Deng, D. Zhao, P. Feng, D. Pine, B. F. Chmelka, G. M. White-
sides, and G. D. Stucky, Hierarchically ordered oxides, Science 282, 2244
(1998). — p.89.
State Commun. 89, 413 (1994). — p.89.
289, 604 (2000). — p.89.
[18] M. Qi, E. Lidorikis, P. T. Rakich, S. G. Johnson, J. D. Joannopoulos, E. P.
Ippen, and H. I. Smith, A three-dimensional optical photonic crystal with
designed point defects, Nature 429, 538 (2004). — p.89.
band gap, Appl. Phys. Lett. 86, 011101 (2005). — p.89.
[20] M. Hermatschweiler, A. Ledermann, G. A. Ozin, M. Wegener, and G. von
Freymann, Fabrication of silicon inverse woodpile photonic crystals, Adv.
Func. Mater. 17, 2273 (2007). — p.89.
[21] L. A. Woldering, Fabrication of photonic crystals and nanocavities, PhD
Thesis, University of Twente, 2008. — p.89, 93.
[22] A. Tandaechanurat, S. Ishida, K. Aoki, D. Guimard, M. Nomura,
S. Iwamoto, and Y. Arakawa, Demonstration of high-q (>8600) three-
dimensional photonic crystal nanocavity embedding quantum dots., Appl.
Phys. Lett. 94, 171115 (2009). — p.89.
[23] D. J. Shir, E. C. Nelson, D. Chanda, A. Brzezinski, P. V. Braun, J. A.
Rogers, and P. Wiltzius, Dual exposure, two-photon, conformal phase mask
lithography for three dimensional silicon inverse woodpile photonic crystals,
J. Vac. Sci. Technol. B 28, 783 (2010). — p.89.
[24] A. A. Chabanov, Y. Jun, and D. J. Norris, Avoiding cracks in self-assembled
photonic band-gap crystals, Appl. Phys. Lett. 84, 3573 (2004). — p.90.
[25] E. P. Wigner, Mathematical proceedings of the cambridge philosophical soci-
ety, 47, 790 (1951). — p.90.
[26] C. W. J. Beenakker, Random-matrix theory of quantum transport, Rev. Mod.
Bibliography 107
Phys 69, 731 (1997). — p.91.

[27] S. F. Liew, S. M. Popoff, A. P. Mosk, W. L. Vos, and H. Cao, Transmission
channels for light in absorbing random media: from diffusive to ballistic-like
transport, Phys. Rev. B 89, 224202 (2014). — p.91.
[28] I. M. Vellekoop, Controlling the propagation of light in disordered scattering
media, PhD Thesis, University of Twente, 2008. — p.91, 92, 93.
[29] P. N. den Outer, Multiple light scattering in random discrete media, PhD
Thesis, University van Amsterdam, 1995. — p.92.
[30] S. P. Morgan and I. M. Stockford, Surface-reflection elimination in polar-
ization imaging of superficial tissue, Opt. Lett. 28, 114 (2003). — p.92.
[31] A. Hartsuiker and W. Vos, Structural properties of opals grown with vertical
controlled drying, Langmuir 24, 4670 (2008). — p.93.
[32] A. Hartsuiker, Ultrafast all-optical switching and optical properties of mi-
crocavities and photonic crystals, PhD Thesis, University of Twente, 2009.
— p.93.
[33] L. A. Woldering, R. W. Tjerkstra, H. V. Jansen, I. D. Setija, and W. L. Vos,
Periodic arrays of deep nanopores made in silicon with reactive ion etching
and deep UV lithography, Nanotechnology 19, 145304 (2008). — p.93.
[34] J. W. Goodman, Statistical optics (John Wiley, Newyork, 2000). — p.94.
[35] I. M. Vellekoop and A. P. Mosk, Focusing coherent light through opaque
strongly scattering media, Opt. Lett. 32, 2309 (2007). — p.96.
[36] A. P. Mosk, A. Lagendijk, G. Lerosey, and M. Fink, Controlling waves in
space and time for imaging and focusing in complex media, Nature Photon.
6, 283 (2012). — p.96.
[37] B. E. Warren, X-ray diffraction (Courier Dover Publications, 1969). —
p.103.
CHAPTER 7
Light transport in silicon two-dimensional photonic
crystals
7.1 Introduction
A periodic structure of materials with differing indices of refraction on the order
of the wavelength is called a photonic crystal. These photonic crystals strongly
affect the dispersion relation of light due to interference effects resulting from
Bragg diffraction inside the crystal [1]. As a consequence, certain light frequencies
are forbidden to propagate inside the crystal, resulting in a stopgap for these
light frequencies. For all three spatial dimensions the latter is referred to as
a photonic band gap [2]. A photonic structure which exhibits a band gap for
light is relevant for various applications as for instance to trap light or to control
the rate of spontaneous emission of light sources [3, 4], resulting in a broad
spectrum of possible applications. The interaction of light with periodic lattice
of the photonic crystals gives rise to distinct optical phenomena as for instance
an enhanced optical reflectivity in the stopgaps [5–8]; frequency range where the
incident light decays exponentially inside the crystal due to the Bragg diffraction.
In practice photonic crystals tend to suffer from unavoidable imperfections
like size polydispersity and position variation [9–11]. A perfect photonic crystal
has an infinitely large transport mean free path; the distance which light travels
before losing its coherence. However, a real photonic crystal has always imper-
fections and therefore a finite transport mean free path. Therefore measuring the
extinction length and transport mean free path indicates how much imperfection
a crystal has. A convenient method to measure the extinction length and mean
free path was developed by Johnson et al [12] who determined the mean free
path by measuring the out-of-plane scattering. A similar configuration is used
by Avoine et al [13], who characterized photonic crystal cavity modes. Here we
use the out-of-plane scattering technique to determine the extinction length in a
photonic crystal. Besides the scattering of light due to imperfections, yet another
important difference between a strong scattering material and a photonic crystal
needs to be taken into account, wherein the stop gap light decays exponentially
due to the Bragg diffraction.
110 Light transport in silicon two-dimensional photonic crystals
In this chapter we investigate the mean free path of two dimensional silicon
photonic crystal by means of out-of-plane scattering within a broad frequency
range covering both inside and outside the stop gaps. We compare the results
from the photonic crystal with a multiple scattering sample of Teflon polymer to
characterize the extinction length and the Bragg length of the crystal.
7.2 Theory
In a collection of scatterers the scattering mean free path (lsc ) is the average
distance between two consecutive scattering events. The intensity of the incident
beam decays exponentially within the scattering material with a decay according
to the extinction length (lex ), which is determined by the scattering mean free
−1 −1 −1
path and the absorption length (lab ) as lex = lsc + lab [14].
x
−
l
I(x) = I0 e ex , (7.1)
where I(x) is the intensity along the traveled path of light and I0 is the incident
intensity. If a sample is thinner than one mean free path this coherent, unscat-
tered beam has an exponentially large contribution in the detected intensity, and
the Beer-Lambert-Bouguer law determines the extinction length.
When the sample is thicker than one mean free path d > lsc the scattering
exhibits a preferred scattering angle, which is described with the scattering func-
tion. Depending on the scattering function the direction of light will be random
after one or more lsc . The distance at which the intensity distribution becomes
isotropic due to scattering is given by the transport mean free path ltr . When
lsc ≤ ltr << d the diffuse light becomes detectable. The diffuse light is well
described by the diffusion equation [15]
∂W
− D∇2 W = S, (7.2)
∂t
where W is the energy density of light, D is the diffusion constant indicating
the speed at which light diffusely spreads out and S is the function describing
the source of diffuse light. Therefore if the sample is thicker than the mean free
path, a combination of both coherent light and the diffuse light is detectable. In
Ref. [12] a method to fully characterize diffuse transport of light is presented.
There light is assumed to be tightly focused on one side of a diffuse material. On
another surface out-of-plane scattered light is measured parallel and perpendic-
ular to the direction of the incident light.
Since the feature size in a photonic crystal is in the range of wavelength of
probe light, it is reasonable to assume that light scattering in a photonic crystal
is anisotropic. Herein a second assumption is that the scatterers are randomly
distributed. For light frequencies outside the stopgap frequencies, normal scat-
tering of light is expected. For light frequencies within the stopgap, diffusion
theory breaks down since these frequencies of light are not expected to propa-
gate inside the crystal. Therefore, light with these frequencies will not penetrate
Sample selection 111
d y
a
c z
x
ΓM’
ΓK
Figure 7.1: Schematic representation of the top view of the periodic nanostructure.
The indicated centered rectangular lattice is characterized by lengths of a = 693 ± 10
and c = 488 ± 11 nm respectively. The diameter of the pores is d = 310 ± 10 nm. Access
to both the ΓK and the ΓM direction is ensured by cleaving the crystal parallel to the
sides of the rectangular Bravais lattice.
far into the sample, or propagates in oblique allowed directions [16]. We perform
experiments in order to describe the propagation of different randomly scattered
light in a photonic crystal in a qualitative manner.
7.3 Sample selection

The out-of-plane scattering experiments are performed on two types of samples.
Firstly the out-of-plane scattering of light is investigated on a two-dimensional
photonic crystal. The 2D photonic crystal is etched in the corner of a piece of
silicon wafer which exhibits a length and width of approximately 10 mm and 5
mm respectively and a thickness of 260 μm.
In Fig.7.1 a schematic representation is depicted of the top view of the two-
dimensional photonic crystal. The figure outlines the centered rectangular lattice
of the crystal. The lattice parameters for the used photonic crystal are a =
693±10 and c = 488±11 nm respectively. The photonic structure is characterized
by a pore diameter of d = 310 ± 16 nm and a pore depth of 5806 ± 170 nm. For
characterization of the specified two-dimensional photonic crystal by means of
reflectivity measurements, the selected crystal was cleaved along two different
high symmetry directions along the crystal lattice. This enables both the ΓK
and the ΓM direction of the photonic crystal to be accessible for out-of-plane
scattering experiments. In the out-of-plane scattering experiments presented here
only the ΓM direction of the two-dimensional photonic crystal is addressed.
The light sent to the crystal is polarized in two linear orthogonal directions. In-
cident light perpendicular to the air cylinders in the material is referred to trans-
verse electric (TE) polarization. The radiation incident parallel to the etched
rods is called transverse magnetic polarized (TM).
In Fig. 7.2 the measured reflectivity of the two-dimensional photonic crystal
sample as function of the frequency for TE and TM-polarized light is depicted.
(a) (b)
TE TM
Figure 7.2: Calculated band structure and measured reflectivity of the two-
dimensional photonic crystal sample (sample code Ad3c-28112008) as function of fre-
quency for (a) TE and (b) TM polarized light incident along the ΓM direction of the
photonic crystal [17]. The gray bars indicate the calculated stop gaps.
The data are taken from Ref. [17, 18]. The data show the reflectivity measured
along the ΓM direction of the photonic crystal sample. For the spectral range
covered in the experiments, i.e., frequencies between ν=6000 and 9000 cm−1 , a
significant variation in the reflectivity for TE and TM polarized light is observed.
For TE-polarized light a reflectivity peak is present between frequencies of 5000
and 7000 cm−1 . A corresponding effect for an orthogonal TM-polarization is
observed for light frequencies between 8000 and 8800 cm−1 , which are in a very
good agreement with the calculated band structures.
7.4 Experimental procedure

In Fig. 7.3 a schematic representation of out-of-plane scattering is shown. Inci-
dent light illuminates the crystal from side, and in the perpendicular direction
the induced scattering is collected. By measuring the intensity distribution of
this out-of-plane scattered light, the extinction length is determined.
The experimental setup used during the experiments is the NIR setup described
in chapter 2, where for the light source we have used a pulsed supercontinuum
laser source (Fianium SC450-2) with a repetition rate 20 MHz, 2 W power over
Experimental procedure 113

Figure 7.3: Schematic representation of a typical out-of-plane scattering scheme. In-

cident light illuminates the crystal from the x-z face. In the perpendicular direction the
resulting out-of-plane scattering is collected.
the full spectrum. In addition to chapter 2 the light beam is sent through a
reflective beam expander to increase the beam size. An aperture is placed in
the expander to increase the spatial quality of the beam profile. For alignment
purposes, the optical path of the supercontinuum white light source is forced to
overlap with a pre-aligned Helium Neon laser (HeNe). A reflecting microscope
objective with a numerical aperture (NA) of 0.28 is implemented to couple in
the incident light to the sample. A Glan-Taylor Calcite polarizer (Thorlab SM05
PM5) is situated in front of the illumination objective to select linear polarization
directions with respect to the orientation of the sample. Perpendicular to the
illumination objective an infinity corrected microscope objective (o1) is situated
to collect the out-of-plane scattering of the in-coupled laser light on the sample.
The sample is placed in a home-built sample holder which is fixed on the sample
stage. This device gives rise to a spatial scan range of several centimeters with
a spatial resolution of 50 nm on the sample. Position of the sample is initialized
using a hand-held controller unit (Newport, XPS-RC). The illuminating and
collecting microscope objectives are aligned with respect to the sample by imaging
both sides of the sample alternatively. The position of the sample is varied via a
three-dimensional micro controlled translational stage. This procedures enable to
address a pre-assigned location on the sample and determine the location of the
excitation spot of the supercontinuum laser on the photonic crystal. The out-of-
plane scattered light is sent to the grating spectrometer. This enables to resolve
the scattering spectrally for light frequency ranging from 6000 to 9000 cm−1 . The
exposure time of the spectrometer is set to collect maximum scattering without
saturating the array detector. During the out-of-plane experiments the exposure
time ranges typically between 2 and 10 s. As some diodes in the detector array
are broken, collecting the spectra of the out-of-plane scattering inherently means
that there is some spectral information lost. In the experiments the internal
alignment of the spectrometer is set for the damaged diodes to correspond to a
central wavelength of 1400 nm (7140 cm−1 ).
As our photonic crystal exhibits stop gaps in the covered spectral range (see
Fig. 7.2), this enables to compare the spectral dynamics of incident light for
different regimes: outside and inside the stopgaps. For reference as well as for
(a) (b)
Figure 7.4: Intensity resolved out-of-plane scattering of the Teflon sample at a mean
frequency of ν=7394 cm−1 for (a) TE-polarization and (b) TM-polarization. The sam-
ple is illuminated at coordinates y=0 and z=0, and the incident light is propagating
parallel to the y-direction. The intensity is normalized to the maximum observed in-
tensity in the operated scan range. Here the blue color represents minimum and the
red shade stands for the maximum intensity.
comparison of experimental outcomes of out-of-plane scattering experiments on

the photonic crystal, the out-of-plane scattering of light is studied on a solid block
of Teflon. The length, width and thickness of this random scattering medium are
approximately 50, 30, and 20 mm respectively.
7.5 Experimental results and discussion

For illumination along the ΓM direction of the photonic crystal, stop bands
are expected for both transverse electric (TE) and transverse magnetic (TM)
polarizations. It is expected that the presence of stopgaps strongly influence the
spectrum of the induced out-of-plane scattering of the photonic crystal.
In a strongly scattering material the polarization of the incident light is of
little importance since the polarization is not likely to be conserved due to mul-
tiple scattering. The dispersion relation of light in a photonic crystal, however,
strongly depends on the polarization and the direction via which light is coupled
to the period structure. It is expected that the mean free path will therefore also
be strongly polarization and direction dependent. But when scattering occurs,
the polarization of the incident light may not be conserved. In a crystal this may
result in an undefined polarization with respect to a pre-selected reference frame.
This can make it difficult to distinguish the influence of the polarization of the
incident light on the mean free path.
7.5.1 Out-of-plane scattering on a Teflon sample

For the Teflon sample two types of experimental results are presented. By mea-
suring the spectra over a fixed lateral extent at the out-of-plane surface of the
Experimental results and discussion 115
(a) (b)
Figure 7.5: Intensity and frequency resolved out-of-plane scattering of the Teflon sam-
ple for (a) TE-polarization and (b) TM-polarization. The wide dark blue line is due
to the broken diodes in the detection array, corresponding to a frequency of 6900 to
7350 cm−1 . The blue color represents minimum and the red shade stands for maximum
intensity of the out-of-plane scattered light. For a direct comparison, the color bar is
kept the same as for the measurements on photonic crystals.
sample, the spatial profile of the scattering is resolved in both intensity and fre-
quency. Secondly, over a single line of side parallel to the propagation direction
of the illumination, the intensity is monitored as function of the lateral dimension
of the sample.
Fig.7.4 shows the normalized out-of-plane scattering intensity of the Teflon
sample for both polarizations at a randomly chosen mean frequency of ν=7394
cm−1 . The illumination spot is located at y=0 and light is propagating parallel
to the y-direction. The spot size in z direction is estimated to be about 25 μm.
The measurements extend both 15 μm in the lateral y and z direction. Here the
intensity is normalized to the maximum observed scattering intensity in the op-
erated scan range. The optical mappings are acquired using a scan resolution of
1 μm. This quantity is optimized considering the lower limit of the spatial resolu-
tion of the collecting microscope objective. This results in an optimized duration
of the experiments without loss of spatial information. In order to minimize the
effects of spectral fluctuations, the data in Fig. 7.4 are averaged over a frequency
interval of 50 cm−1 . A relative experimental error in the scattering intensity
can be mostly caused by time-to-time fluctuations of the supercontinuum light
source. The intensity distribution of the scattering at the probed plane shown
in Fig.7.4 behaves as expected for a uniform random scattering material. The
observed out-of-plane light is random, resulting in the formation of speckle: an
intensity pattern formed by the interference of out-of-plane scattered waterfronts.
Moreover, there are minimal observed differences between the two orthogonal po-
larization directions of the incident radiation, supporting the random scattering
character of the Teflon sample.
In order to understand the behavior of the transport of the in-coupled light in
the Teflon sample, single line intensity distributions of the out-of-plane scattered
light are monitored. In figure 7.5 the intensity and frequency resolved data
of the out-of-plane scattering experiment for (a) TE-polarization and (b) TM-
polarization incident light over a single line at z=0 are shown. The wide dark
blue lines are the broken diodes of the detection array. This precludes the out-
of-plane scattering measurements between frequencies of 6900 and 7350 cm−1 .
The data are normalized to the initial scattering intensity at the beginning of
the scan. As the observed excitation spot on the Teflon sample is relatively large
compared to the scanned range, corresponding data as presented in figure 7.5
are not analyzed for the direction perpendicular to the propagation direction
of the incident light. This means that it is not possible to draw conclusions
about the isotropic character of the sample under investigation from the data
shown in Fig. 7.5. The first observation made from Fig. 7.5 (a) and (b) is the
similarity between the intensity distributions of the scattering induced by the two
orthogonal polarization states of the incident light. Note that some differences in
the intensity are due the experimental error caused by time-to-time fluctuations
of the light source. It is concluded that the sample is weakly disordered, resulting
in the formation of speckle and a very long extinction length. Both polarizations
scatter in the material similarly, in agreement with the expectations for a random
scattering material.
(a) (b)
(c) (d)
Figure 7.6: Intensity of the out-of-plane scattering of the Teflon sample as function
of the lateral y-position of the collecting aperture. (a) though (d) show the intensity
distribution for TE (black cross) and TM (red cross) polarized incident light at mean
frequencies of ν=6057, 6761, 7489 and 8295 cm−1 respectively. The dashed red and
black lines are guides to the eyes.
(a) (b)
Scattered
laser light
Air 2D crystal Air 2D crystal
Figure 7.7: CCD image of the out-of-plane side of the 2D photonic crystal. (a) Crystal
is illuminated with white LED via the detection objective, and (b) illuminated with the
supercontinuum laser light via the illumination objective. Laser light is scattered from
the edge of the sample and partially penetrates in the crystal. The yellow dashed line
indicates the crystal-air edge.
In figure 7.6 typical intensity distributions for the out-of-plane scattered light
of the Teflon sample over a single line for TE and TM polarized incident light
are depicted. Fig. 7.6 (a) through (d) represent spectral cross-section of the data
outlined in Fig. 7.5 for mean frequencies of ν=6057, 6761, 7489, and 8295 cm−1
respectively. The data are averaged over a frequency bandwidth of 50 cm−1 . The
character of the intensity distribution of the induced scattering is in detail com-
pared for the two orthogonal polarizations of the incident light. Again, taking
into account the speckle formation, no significant differences between the fre-
quencies of the incident radiation are perceived. Secondly, the general character
of the intensity change in different frequencies of the scattering is comparable, as
for both polarizations and for different frequencies no clear decay in the intensity
profile is seen. Therefore it is concluded that all incident light frequencies are
able to couple into the sample. Here the formation of speckle is due to multiple
scattering events in the material. Over the implemented scan range the intensity
of the out-of-plane scattering remains almost constant. This means that the ex-
tinction of the scattering exceeds the covered lateral scan range on the sample.
The averaged extinction length for these 4 frequencies is calculated to be 42 μm,
which matches with the weakly scattering character of Teflon.
7.5.2 Out-of-plane scattering of silicon two-dimensional

photonic crystals
Figure 7.7 shows CCD images of a 2D photonic crystal in the setup. The images
are recorded by the CCD camera situated in the detection path of the experimen-
tal setup. Fig. 7.7 (a) shows the photonic crystal when illuminating with LED via
the detection objective. The bright part is the silicon crystal and the dark part
is air. Fig. 7.7 (b) shows the sample in the same configuration when illuminating
the sample with supercontinuum laser light via the illumination objective.
From the out-of-plane scattering on the two-dimensional photonic crystal it
is observed that artifacts on the silicon surface, such as dust particles, signifi-
(a) (b)
Figure 7.8: (color) Intensity and frequency resolved out-of-plane scattering of the pho-
tonic crystal for (a) TE-polarization and (b) TM-polarization light over a single line.
The wide dark blue line is due to the broken diodes in the detection array, correspond-
ing to a frequency of 6900 to 7350 cm−1 . The blue and red color indicate a minimum
and maximum intensity respectively in logarithmic scale.
cantly contribute to the detected scattering. In order to prevent this unwanted

scattering to completely conceal the desired scattering of the photonic crystal,
the excitation spot of the supercontinuum laser source is positioned Δx = 35μm
away from the edge of the crystal.
In order to study the light transport in the photonic crystal sample the intensity
distribution of the out-of-plane scattered light over a single line is monitored. For
the covered spectral range, a significant change in the crystal reflectivity or the
coupling of the incident laser light is expected as shown in Fig. 7.2. This is more
significant for the TE-polarization in frequencies between 5000 and 7000 cm−1 .
The corresponding effect for an orthogonal TM-polarization is expected for light
frequencies between 8000 and 8800 cm−1 .
Figure 7.8 shows the intensity and frequency resolved out-of-plane scattering of
the 2D silicon photonic crystal for (a) TE-polarization and (b) TM-polarization
incident light over a single line at z=0. The data is normalized to the initial
scattering intensity at the edge of the sample. In Fig. 7.8 (a) for light frequencies
smaller than 6500 cm−1 the intensity distribution of the out-of-plane scattering
along the line scan significantly decreases after 10 μm. We attribute this effect to
the relatively large stop gap at a central frequency of ν = 6000cm−1 . A distinct
increase in the intensity near a lateral position of 7 μm is observed. As the
accumulation of intensity also reproduces for frequencies outside the stopgap, it
is unlikely that this feature is attributed to the photonic character of the sample.
Most likely the increase in scattering intensity is the result of an artifact, such
as dust particle, at the surface of the photonic crystal.
Fig. 7.8 (b) shows the out-of-plane scattering for the TM polarized light. The
character of the photonic crystal is more restricted in the intensity and frequency
resolved measurements for light polarized along the TM-direction. Between 8000
cm−1 to 9000 cm−1 a small reduction in the intensity can be seen after a lateral
-1
= 6761 cm
1
10
(a) (b) TE
TM
0
10
I/I0
−1
10
−2
10
0 3 6 9 12 15
y−position (μm)
-1 -1
= 7489 cm = 8295 cm
1 1
10 10
(c) TE (d) TE
TM TM
0 0
10 10
I/I0
I/I0
−1 −1
10 10
−2 −2
10 10
0 3 6 9 12 15 0 3 6 9 12 15
y−position (μm) y−position (μm)
Figure 7.9: Intensity of the out-of-plane scattering of the photonic crystal as function
of the lateral y-position of the collecting aperture. (a) through (d) show the intensity
distribution for TE (black cross) and TM (red cross) polarized incident light at mean
frequencies of ν=6057, 6761, 7489, and 8295 cm−1 respectively. The gray bar covers
the data points which were excluded for the fitting.
position of 10 μm which can correspond to the TE stop gap. Similar to the TE

polarization, for light frequencies smaller than 6500 cm−1 the intensity distribu-
tion of the out-of-plane scattering along the line scan significantly decreases after
10 μm. While this is not expected for the TM polarized light it can readily be
understood: when scattering occurs, the polarization of the incident light is not
conserved, hence the stop gap for the TE polarized light affects the TM polarized
incident light, and vice versa.
Figure 7.9 (a) to (d) show the intensity distributions for the out-of-plane scat-
tered light of the two-dimensional photonic crystal along a line scan for incident
light polarized in the TE and TM direction for mean frequencies of ν=6057,
6761, 7489, and 8295 cm−1 respectively. The intensity is averaged over a fre-
quency interval of 50 cm−1 and they correspond to the intensity distribution for
light frequencies in, at the edge, or outside the photonic stopgap for both TE
and TM polarized light. Frequency ν=6057 is in the middle of the experimental
stop gap for the TE polarized light. As we expect the intensity strongly drops to
almost 1% of its initial value after 15 μm. A similar drop of intensity is observed
for the TE polarized light. This can be explained following our reasoning about
polarization mixing in scattering. In order to extract the extinction length for
20
15
Extinction length 10
0
6000 6500 7000 7500 8000 8500
Frequency (cm -1)
Figure 7.10: Extinction length versus as a function of frequency for TE polarized light.
The gray bar indicated the experimental stop gap in TE direction. Inside the band gap
the extinction length is on average 3 times smaller than outside the stop gap. Outside
the stop gap the extinction length shows an increasing trend with frequency in contrast
with the theoretical prediction [11].
the intensity in the crystal we fit a single exponential function to the data. For
the fit we exclude the data points that are the result of an artifact at the surface
(between y= 6 to y=10 μm) of the photonic crystal. We obtain an extinction
length of = 3.6 ± 0.3 μm for TE polarized light and 5.1 ± 0.3 μm for the TM po-
larized light. The Bragg length for the TE stop gap is calculated to be B = 3.3
μm [19] which is very close to the measured extinction length. We therefore
conclude that the decay is caused by the Bragg diffraction in the stop gap range.
The difference between the measured extinction length and the Bragg length is
expected, since the calculated and the measured stop gap do not exactly overlap
as shown in Fig. 7.2. Another reason for this difference is the polarization mixing
effect which leads to propagation of light with TM polarization in the TE stop
gap and therefore increasing the extinction length compared to the Bragg length.
At frequencies ν=6761, and ν=7489 the decrease of the intensity is very gradual
as we are probing frequencies outside the stop gaps for both polarizations. An
exponential fit to the data at frequency ν=6761 results in an extinction length
of = 11.7 μm 15a. At ν=8295 the intensity reduction becomes again more
significant as we are close to the TM stop gap. An extinction length of = 8.3
μm for the TM polarization and = 10.2 μm for the TE polarization is obtained.
Figure. 7.10 show the calculated extinction length versus frequency for TE
polarized light. The extinction length in the band gap is in average 3 times
shorter than out side the band gap due to the Bragg diffraction of the incident
light in this frequency range.
Comparing the outlined data in Fig. 7.9 with the corresponding experiments
performed on the Teflon sample, we clearly see influence of the periodicity of
the photonic crystal sample in the stop gap range. The results outside the stop
gap range is comparable with the out-of-plane scattering results from the weakly
Conclusion 121
scattering Teflon sample where we do not see any clear decay of the intensity in
the scanned range.
7.6 Conclusion
Out-of-plane scattering experiments on a two-dimensional photonic crystal were
successfully performed. Reference measurements on a solid piece of Teflon were
also conducted. By resolving the induced out-of-plane scattering in intensity and
frequency, the spectral behavior of the transport of the in-coupled light in the
photonic crystal is analyzed. For the Teflon sample it was observed that the
out-of-plane scattering is independent of the frequency and polarization of the
incident radiance. This matches the expectations for a uniform disordered ran-
dom scattering material in which light is coupled in at a first order approximation
with equal efficiency throughout the covered spectral range. Taking into account
these reference measurements, distinct results were obtained for the out-of-plane
scattering experiment performed on the two-dimensional photonic crystal. The
results outline different spectral behavior for TE and TM polarized excitation.
It is concluded that, the out-of-plane scattered light is influenced by photonic
character of the sample showing a fast drop of intensity for the frequencies in the
stop gaps.
7.7 Outlook
The list of recommendations for future research presented here is a result of the
performed experimental work. Recommendation includes the use of an illumi-
nating microscope objective with an increased numerical aperture. To minimize
the scattering from the crystal edge on the out-of-plane surface of the photonic
crystal the incident spot on the sample needs to be aligned on a relatively small
area a few micron away from the crystal edge. By employing an illuminating
objective with a high numerical aperture the focus of the illumination source on
the sample becomes smaller and the mentioned unwanted effects are reduced.
However, since the detection objective has a high numerical aperture, in this
case, due to the spatial extent of the two high numerical aperture objectives it
is not possible to place two objectives close enough to each other. Therefore it
is not possible to fulfill the requirements of the current state-of-art experimental
scheme. Another draw-back of using a high NA objective is launching a wide
range of wave vectors into the sample and therefore wiping out the effect of nar-
row stop gaps. Possible solutions here are the use of samples which contain a
larger photonic crystal, or to implement a fiber as an illumination source. The
latter solution is at the moment the most feasible, but it is expected to result in
an even more cumbersome alignment of the experimental scheme.
In order to reduce effects of the artifacts in the data it is recommended to
check the sample surface with a microscope and mark a clean area on the sample
surface, prior to the experiment.
122 Bibliography
The experimental setup enables to measure the in-situ reflectivity of the pho-
tonic crystal. It is recommended to acquire the spectral position of photonic stop
gaps by means of in-situ reflectivity measurements. This enables to directly com-
pare the out-of-plane scattering experiments with the reflectivity of the photonic
crystal obtained with identical experimental settings.
For future research, out-of-plane scattering experiments on three-dimensional
photonic crystals is proposed. These kinds of crystal are fabricated by etching
two sets of carefully aligned perpendicular pores as explained in chapter 2. Since
this crystal possess a full photonic band gap, the long-searched localization of
light in three spatial dimensions is achievable.
Bibliography
[1] W. Vos, R. Sprik, A. van Blaaderen, A. Imhof, A. Lagendijk, and G. H.
colloidal crystals, Phys. Rev. B. 53, 16231 (1996). — p.109.
[4] G. Demésy and S. John, Solar energy trapping with modulated silicon
nanowire photonic crystals, J. Appl. Phys. 112, 074326 (2012). — p.109.
[6] 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 complete three-dimensional bandgap near 1.5 micrometres, Nature 405,
437 (2000). — p.109.
289, 604 (2000). — p.109.
[8] Y. A. Vlasov, X. Z. Bo, J. C. Sturm, and D. J. Norris, On-chip natural
p.109.
[9] A. F. Koenderink, M. Megens, G. v. Soest, W. L. Vos, and A. Lagendijk, En-
hanced backscattering from photonic crystals, Phys. Lett. A 268, 104 (2000).
— p.109.
[10] S. Hughes, L. Ramunno, J. F. Young, and J. E. Sipe, Extrinsic optical scat-
tering loss in photonic crystal waveguides: role of fabrication disorder and
photon group velocity, Phys. Rev. Lett. 94, 033903 (2005). — p.109.
[11] A. F. Koenderink, A. Lagendijk, and W. L. Vos, Optical extinction due to
intrinsic structural variations of photonic crystals, Phys. Rev. B 72, 153102
(2005). — p.109, 120.
Bibliography 123
[12] P. M. Johnson, S. Faez, and A. Lagendijk, Full characterization of

anisotropic diffuse light, Opt. Express 16, 7435 (2008). — p.109, 110.
[13] A. Avoine, C. Vion, J. Laverdant, S. Bonnefont, O. Gauthier-Lafaye, L.
Coolen, and A. Maı̂tre, Photonic crystal cavity modes in the visible range
characterized by scattering spectroscopy, Phys. Rev. A 82, 063809 (2010).
— p.109.
[14] C. F. Bohren and D. R. Huffman, Absorption and scattering of light by small
particles (John Wiley, 2008). — p.110.
[15] B. P. J. Bret, Multiple light scattering in porous gallium phosphide, PhD
Thesis, Universiteit Twente, 2005. — p.110.
[16] A. F. Koenderink and W. L. Vos, Light exiting from real photonic band
gap crystals is diffuse and strongly directional, Phys. Rev. Lett. 91, 213902
(2003). — p.111.
[17] S. R. Huisman, Observing the forbidden zone for light, Master Thesis, Uni-
versity of Twente, 2010. — p.112.
[18] S. R. Huisman, R. V. Nair, A. Hartsuiker, L. A. Woldering, A. P. Mosk, and
W. L. Vos, Observation of sub-bragg diffraction of waves in crystals, Phys.
Rev. Lett. 108, 083901 (2012). — p.112.
— p.120.
CHAPTER 8
Summary and outlook
In this thesis both theoretical and experimental studies have been presented
on light emission and propagation in real photonic band gap crystals. We have
investigated several aspects of a real photonic band gap crystal by different types
of measurements such as time resolved emission of light sources in the band gap,
scattering of light due to randomness and probing the internal wave structure
of photonic crystals. In this chapter we review all the studies in this thesis and
based on that we give directions for future work.
The present-day understanding of complex nanophotonic systems is heavily
leaning on theories that describe infinite systems. Well-known examples are
random media and photonic crystals. In contrast, experiments are obviously
performed on real devices with a finite extent. Many of the fascinating optical
properties associated with these nanophotonic systems such as a full three-
dimensional photonic band gap, weak and strong localization, and divergences
in the densities of states are strongly depending on system size. Hence, the
understanding of these effects is crucial for the success of our field, and will
surely open up novel applications.
In this thesis we have devised an original theory that allows the models for
infinite photonic crystals to be extended to represent finite-size effects, without
actually reducing the size of the infinite system. The method is based on an ex-
tension of the wave-vector space of the Green’s functions into the complex plane.
We have successfully applied this theory to a one-dimensional system, where the
results can be verified with exact calculations Green’s function and transfer ma-
trix revealing an excellent match between our theory and these calculations. A
remarkable result is that the local density of states (LDOS), which is strictly zero
in the band gap of an infinite crystal irrespective of frequency (within the gap
of course) and position, becomes strongly dispersive and position-dependent in a
finite crystal.
We developed a theory for the density of states in a finite three-dimensional
photonic crystal. The theory is based on mode broadening in a finite photonic
crystal. We have experimentally studied the controlling of spontaneous emission
of light in real finite photonic band gap crystals. It seems that we are finally in
the position to explain our experimental observations on the non-zero DOS inside
126 Summary and outlook
the 3D photonic band gap. These observations are leveraged by time-resolved

emission studies on semiconductor quantum dots that emit in the telecom range
within the band gap of silicon 3D photonic crystals made by CMOS-compatible
methods. It is remarkable that our theory predicts a volume dependency of
the inhibition in the band gap of a finite photonic crystal. We conclude that
the main limiting factors to observe an infinite inhibition in the band gap are
the photonic crystal size and the non-radiative decay of the quantum emitters.
Therefore future experiments at low temperature and on larger photonic crystals
will definitely yield a greater inhibition in the band gap.
Recently, our group has proposed a design to realize controlled point defects
in inverse woodpile crystals as studied in this thesis. Based on the calculations
these point defects show all the signatures of cavity resonances where light is
strongly confined. It will of course be highly intriguing to investigate the control
of material excitations - be it quantum emitters or plasmonic resonances - by such
resonances. For instance, the simultaneous presence of a very low background in
the band gap suggests that very clear signals can be collected. From a somewhat
different view, it will also be exciting to pursue Anderson localization of light, that
is notably expected near the edge of the gap. Here embedded quantum emitters
could play a role as internal reporter of the phenomena. In our measurements
we have observed an enhancement of the quantum dot decay rate near the edge
of the gap that might be caused by this phenomena.
Three-dimensional photonic band gap crystals are capable of controlling the
radiative loss of any material systems. In principle any field propagation in the
band gap frequency range is forbidden due to the interference effect in the band
gap, no matter what is the radiation source. In this respect the radiative loss of
the plasmonic systems in the band gap can also be controlled. Currently the main
limitation on this experiment is fabricating plasmonic particles whose plasmonic
spectrum overlaps with the photonic band gap range. The size polydispersity of
the particles is currently an issue, making the plasmonic spectrum of the particles
much broader than the band gap. In future when the right combination of the
plasmonic particles and the photonic crystal is found, the experiments will reveal
the ultimate control on the plasmonics scattering.
In this thesis, we have probed the LDOS by meaning the time-resolved emis-
sion rate of quantum dots. It is a principal disadvantage of counting photons
that in the situations of interest, namely a strongly decreased LDOS, the signal
also decreases. Therefore it is relevant to consider alternative methods. Signs
of a photonic band gap might be also experimentally accessible by probing the
excited state population. In an ideal case, if an emitter in the photonic band
gap is excited it stays excited forever and therefore the excited state stays pop-
ulated. Population of the excited state increases by decreasing the LDOS. In
a real photonic crystal the excitation will not last forever as the LDOS is not
zero, and the excited emitters decay to the ground state after several lifetimes.
Meanwhile, when the emitter is in the excited state the excited state population
can be probed by excited-state absorption measurements. In this case electrons
in the excited state are excited to the second excited state by absorbing a photon.
Since the emitters that are deeper inside the crystal experience a lower LDOS,
127
their excited state population is higher. As opposed to the emission measure-

ments performed in this thesis such a measurement is biased to the long life-time
emitters. In the ultimate case, the longest living excited state will determine the
longest remaining signal.
We have introduced a Bayesian method for time-resolved emission data analy-
sis. In the near infrared range the time-resolved data collected from the emitters
in the band gap photonic crystal are suffering from a very low signal to noise
ratio. In our statistical analysis we have used the probability distribution of the
background counts as prior information to improve the data analysis and refine
the free parameters in modeling the data. The method can be used in any data
analysis where the signal to noise ratio is low and therefore the uncertainty on
parameter determination is high.
In the last chapter, we have investigated the internal waves inside photonic
crystals by measuring the light scattering using correlation microscopy and scat-
tering microscopy techniques. Coherence length, extinction length, long range,
and short range order of light in the opal and in the two-dimensional photonic
crystals have been measured. Here we propose future experimental studies to
quantitatively investigate the scattering from photonic crystal samples. To start
with, we propose to use simpler photonic structures such as Bragg stacks where
the periodicity is present only in one direction. By investigating different Bragg
stacks with different lattice parameters we can quantify the transport mean free
path with respect to the Bragg length for different crystals. In this way we obtain
information on how deep the light propagates inside the crystal.
A second proposition which simplifies the interpretation of the results is to
use a low numerical aperture objective to send a small range of incident wave
vector into the sample. This comes at the price of lowering the spatial resolution
and illuminating a larger area on the crystal. The disadvantage is of course that
the measurements will not be anymore sensitive to the short range disorder in
the photonic crystals. However, the main advantage is that we will be able to
investigate the directional light propagation into the photonic crystal by coupling
the incident light to a limited set of Bloch modes, thereby simplify theoretical
interpretation.
Thirdly, using wavefront shaping one can modulate the wavefront to couple to
a particular Bloch mode. The main limitation is however that the modulation
in the wavefront is diffraction limited. In order to be in the photonic range of
photonic crystal we need to use light sources with the wavelength in the order of
the crystal periodicity. Using an objective with NA = 0.9 for visible laser light
we have a focus size of about 300 nm. Therefore the resolution of the modulated
wavefront can not be smaller that 300 nm. This will limit the number of possible
Bloch waves that can be generated and imaged on the crystal surface. However,
it is very exciting to perform this experiment even for a simple Bloch wave and
study the surface impedance and the scattering. Fascinating wavefront shaping
experiments can be done by addressing cavities inside a full photonic band gap
crystal.
APPENDIX A
Light propagation in periodic media
In this section we describe the optical properties of an infinitely extended 1D

periodic structure. To calculate the band structure and the electromagnetic
fields we have used the transfer matrix method [1, 2]. The dielectric function of
the structure is real and periodic:
(z) = (z + a), (A.1)
where a is an arbitrary lattice vector [3]. This means that the structure is in-
variant under the translation (m · a) where m is an integer and a is the period of
the structure. Here we limit ourselves to propagation of light in the z-direction
perpendicular to the layers, and we assume that the medium is non-magnetic. In
this 1D periodic structure it is only necessary to consider one polarization due to
the symmetry. The propagation of light with frequency ω in a periodic structure
is described by Maxwell’s equations [1, 4]. Due to symmetry these equations
remain the same after substituting (z + a) for z in the operators and . The
solutions of the wave equation have the form of a plane wave times a function
with the periodicity of the lattice
En,K (z) = Ēn,K (z)e−iKz , (A.2)
where Ēn,K (z) is periodic with the lattice
Ēn,K (z) = Ēn,K (z + a), (A.3)
which is known as the Bloch-Floquet theorem. The subscript K indicates that

the function Ēn,K (z) depends on the Bloch wave number K. Note that Eq. A.2
and Eq. A.3 imply that:
En,K (z) = En,K (z + a)e−iKa , (A.4)
The problem at hand is thus of determining K and Ēn,K (z) as a function of ω.

To this end we have used the transfer matrix method together with the Bloch
condition to calculate the band structure and the Bloch fields propagating in the
structure. The electric field within each homogeneous layer is expressed as a sum
of an incident and a reflected plane wave:
En,K (z) = b− e−ikz + b+ eikz , (A.5)

130 Light propagation in periodic media
−
where k = nω +
c is the wave vector and b and b are coefficients that are related
through the continuity conditions at the interfaces. Using Maxwell’s equations
one obtains the general form of the magnetic field amplitude:
√ √
Bn,K (z) = b− e−ikz − b+ eikz , (A.6)
Imposing the continuity of the electric and magnetic fields at two consecutive
interfaces leads to a matrix that connects the coefficients of one homogeneous
layer(n) to the coefficients of the same type of layer(n + 1) in the next period.
These can be written as the following matrix equation:
− −
bn−1 A B bn
= (A.7)
b+
n−1 C D b+
n

A B
where T = is a transfer matrix that relates the coefficients of the same
C D
type of layers. The matrix elements A, B, C, and D are complex functions of the
layers properties such as the dielectric permittivity (i ) and the width (ai ) of each
layer. As a consequence,
− only the first two components b− +
0 and b0 (or the first
b0
column vector ) can be chosen arbitrarily. If we choose the column vector
b+
0
of layer 1 in the zeroth unit cell (n = 0), then the remaining column vectors of
the equivalent layers are related to that of zeroth unit cell by :
− n −
bn D −B b0
= (A.8)
b+
n −C A b+
0
Besides the continuity conditions that should be satisfied, the periodic layers
must be invariant under lattice translation, in other words the Bloch condition
Eq. A.4 must be satisfied. In terms of our column vector representation and from
Eq. A.5, the Bloch condition Eq. A.4 yields
− −
bn −iKa bn−1
=e . (A.9)
b+
n b+
n−1
By combining Eq. A.7 and Eq. A.9 the column vector of the Bloch wave satisfies
the following eigenvalue equation:
− −
A B bn iKa bn
=e . (A.10)
C D b+
n b+
n
The eigenvalues are equal to
1 1
eiKa = (A + D) + [ (A + D)2 − 1]1/2 (A.11a)
2 4
1 1
eiKa = (A + D) − [ (A + D)2 − 1]1/2 (A.11b)
2 4
Bibliography 131
that are the reciprocal of each other. The eigenvectors corresponding to the
eigenvalues are −
b0 B
= . (A.12)
b+
0 eiKa − A
This bring us to −the point that although for the transfer matrix method one
b0
column vector can be independently chosen, since we have imposed the
b+
0
Bloch condition this column vector will be set by Bloch condition as Eq. A.12.
That means that depending on K there exists a unique initial condition for
which the electromagnetic field propagates into the periodic structure and this
initial condition depends on the properties and the geometry of the periodic
structure. To normalize the initial condition we have normalized the field such
that if we consider both layers to be vacuum, the intensity of the field propagating
in vacuum is equal to one and does not depend on frequency, therefore:
−
b0 1
= (A.13)
b+0 (e iKa
− A)/B
According to Eq. A.9 the corresponding field amplitudes for the nth unit cell are
equal to −
bn −inKa 1
= e (A.14)
b+
n (eiKa − A)/B
With the initial conditions and the Bloch wave vector K we obtain the fields and
dispersion relation between ω and K for the Bloch wave function.
1 1
K(ω) = cos−1 [ (A + D)]. (A.15)
a 2
Bibliography
[1] A. Yariv and P. Yeh, Optical waves in crystal: Propagation and control of
laser radiation (John Wiley, New York, 1983). — p.129.
[2] C. Barnes and J. B. Pendry, Multiple scattering of waves in random media :
a transfer matrix approach, Proc. Mat. Phys. Sc 435, 185 (1991). — p.129.
Winston, New York, 1976). — p.129.
2008). — p.129.
APPENDIX B
Decay rate versus frequency of PbS quantum dots
Semiconductor quantum dots of IV − V I materials such as PbS and PbSe offer

unique access to the regime of extreme quantum confinement since the electron,
hole, and exciton all have relatively large Bohr radii [1]. In PbS, the electron,
hole, and exciton Bohr radii are 10, 10, and 18 nm, respectively. These large radii
allow strong confinement to be achieved in relatively large structures. Thus, QDs
of IV − V I materials have properties reflecting all benefits of strong quantum
confinement, with reduced influence from surface effects.
In this section we discuss the decay rate of the PbS quantum dots in toluene
versus frequency. In chapter 4 we do the same experiment for the quantum dots
in the photonic crystals. Therefore, it is important to study their behavior before
placing them in a more complex environment.
The decay rate measured as a function of frequency is shown in Fig. B.1 (a).
The decay rate was obtained from data as in Fig .2.3 (b). It is seen that the decay
rates are a function of the emission frequency, which is a direct confirmation that
the emission spectrum is inhomogeneously broadened. It is known that the bigger
quantum dots have a smaller band gap and therefore emit at lower frequency and
as the size increases the emission frequency decreases [2]. In our measurements
at lower frequencies (larger quantum dots) we observe a higher decay rate. Hence
the decay rate increases with increasing quantum dot size. For the PbSe quantum
dots three data points have been measured by Husken [3], showing a similar trend
as shown in Fig. B.1. The same experiment was performed by Moreels et al [4]
on PbS quantum dots, in which they measured the decay rate versus quantum
dot size for a very wide range of the quantum dot size. The result of their results
in the same range as we measured shows an almost constant decay rate without
a clear increasing or decreasing trend.
As a first step to interpret the result, we have calculated the radiative decay
rate versus frequency for an ideal two-level exciton PbS quantum dot using the
following equation:
e2 nω 2
γrad (ω) = fem (ω), (B.1)
6me 0 πc3
Equation (B.1) gives the emission oscillator strength fem (ω) of the transition
as a function of radiative decay rate [5]. Here e is the electron charge, n is
refractive index of surrounding material, me is the electron mass, ω is emission
frequency, 0 is vacuum permittivity, and c is speed of light in vacuum. In the
strong confinement limit of the exciton wavefunction the oscillator strength is
134 Decay rate versus frequency of PbS quantum dots
(a) 2 (b)
PbS 5
Oscillator strength
PbSe
Moreels 4
1.5
γ(μs )
−1
3
vel exciton 2
1 ideal two le
1
0.5 0
6500 7000 7500 8000 6500 7000 7500 8000
−1
Frequency(cm ) Frequency(cm−1)
Figure B.1: (a) The decay rate measured versus frequency on PbS quantum dots en-
semble (black circles). The behavior is opposite to the expectation for an ideal exciton
that the decay rate is proportional to the emission frequency as it is plotted with gray
dashed line. A similar measurements on PbSe quantum dots shows the same trend (blue
triangles). The red squares shows measurements on PbS quantum dots by Moreels et
al [4]. Their measurements in the same range shows an almost constant decay rate. (b)
The calculated oscillator strength for the PbS quantum dots increases with quantum
dot size.
given by [6]:
4 a∗3 ωbulk
fem = √ π B3 fbulk (B.2)
3 a ω
where fbulk is the oscillator strength in bulk per chemical PbS unit, a∗B is the
exciton Bohr radius, ωbulk is bulk emission frequency, ω is emission frequency
of the quantum dots that depends on quantum dot radius, and a is the lattice
constant. We consider a∗B = 18 nm, a = 0.59 nm, fbulk = 1.4 · 10−4 , ωbulk = 0.37
eV [7], and ω varies between 6000 cm−1 to 8000 cm−1 as in the experiment.
The behavior of the decay rate versus frequency is shown in Fig. B.1. The ideal
PbS quantum dots shows an increasing decay rate with frequency (or decreasing
quantum dot radius), which is an inverse behavior compared to the measured
data.
To interpret the difference between our observation and the two level exciton
model, we discuss three possible hypotheses for the observed increase of the quan-
tum dot decay rate with size, based on literature. First the non-radiative decay
rate may become larger for larger dot sizes, due to phonon coupling. This essen-
tially means that the changes in decay rate are due to changes in the quantum
efficiency of the dots. However, phonon interactions are expected to be small due
to the strong confinement that are especially important in PbS quantum dots [1].
Therefore we reject this first hypothesis.
Second, a size dependence of the spatial overlap of the electron and hole wave-
functions may cause a change of the radiative decay rate, as has been observed
for InAs quantum dots [8]. However, for PbS quantum dots it is very unlikely
that the wavefunctions shift much due to the strong confinement. Therefore this
hypothesis is also rejected.
Bibliography 135
The third explanation for the changes in decay rate is related to the difference in
oscillator strength of the dots with different sizes, since the decay rate is written
explicitly as a function of the emission oscillator strength fem (ωj ). We have
calculated the oscillator strength for our PbS quantum dots as shown in Fig. B.1
(b). The oscillator strength increases with the quantum dot size. As it has been
calculated by Moreels et al [4], following the tight-binding model for the PbS
and PbSe quantum dots, oscillator strength scales approximately linearly with
the quantum dot size, which confirms our calculations. The change of oscillator
strength with size can be explained by a size dependent dipole moment of these
quantum dots. We therefore believe that the oscillator strength causes the change
in the decay curves, instead of non-radiative process or the wave function overlap.
Beside the opposite trend of the decay rate, the decay rate shows a non-linear
trend versus frequency. Such a behavior can be assigned to parity-forbidden
transitions in PbS quantum dots as discussed in Ref. [9].
To conclude the decay rates has been measured as a function of frequency and
therefore as a function of quantum dots size. The measurements show increase
of the decay rate as a function of quantum dots size suggesting that the emission
spectrum is inhomogeneously broadened
Bibliography
[1] F. W. Wise, Lead salt quantum dots: the limit of strong quantum confinement,
Acc. Chem. Res. 33, 773 (2000). — p.133, 134.
[2] A. I. Ekimov and A. A. Onushchenko, Size quantization of the electron energy
spectrum in a microscopic semiconductor crystal, JEPT Lett 40, 1136 (1984).
— p.133.
[3] B. H. Husken, Spontaneous emission of near-infrared quantum dots controlled
with photonic crystals, PhD Thesis, University of Twente, 2009. — p.133.
[4] I. Moreels, K. Lambert, D. Smeets, D. De Muynck, T. Nollet, J. C. Martins, F.
Vanhaecke, A. Vantomme, C. Delerue, G. Allan, and Z. Hens, Size-dependent
optical properties of colloidal PbS quantum dots., ACS Nano 3, 3023 (2009).
— p.133, 134, 135.
[5] A. Siegman, Lasers (University Science Books, 1986). — p.133.
[6] Y. Kayanuma, Quantum-size effects of interacting electrons and holes in semi-
conductor microcrystals with spherical shape, Phys. Rev. B 38, 9797 (1988).
— p.134.
[7] E. D. Palik, Handbook of optical constants of solids (Academic Press, 1985).
— p.134.
[8] J. Johansen, S. Stobbe, I. S. Nikolaev, T. Lund-Hansen, P. T. Kristensen,
J. M. Hvam, W. L. Vos, and P. Lodahl, Size dependence of the wavefunction
of self-assembled inas quantum dots from time-resolved optical measurements,
Phys. Rev. B 77, 073303 (2008). — p.134.
[9] L. Cademartiri, E. Montanari, G. Calestani, A. Migliori, A. Guagliardi, and
G. A. Ozin, Size-dependent extinction coefficients of PbS quantum dots., J.
Am. Chem. Soc 128, 10337 (2006). — p.135.
APPENDIX C
Control radiative linewidth of plasmonics with
photonic crystals
C.1 Decay mechanism of plasmonic particles

Our goal is to control the radiative properties of surface plasmon polaritons
using photonic crystals. The optical response of a metallic nanoparticle depends
on its shape, size and environment [1]. Free plasmons excited by electrons are
not discussed here. The homogeneous linewidth Γ of the plasmon resonance is
2
related to the dephasing time T2 by Γ = , which is related to the time constant
T2
of population decay T1 (total life time) by
1 1 1
= + ∗. (C.1)
T2 2T1 T
The term T ∗ is the pure dephasing time corresponding to elastic collisions such
as phonon interactions, which can be controlled by temperature. Inelastic decay
occurs via the transformation of a plasmon into a photon or via nonradiative
decay, such as the creation of electron hole-pairs by intraband or interband tran-
sitions. Therefore the total decay includes radiative T1,r and non-radiative T1,nr ,
(T1−1 = T1,r
−1 −1
+ T1,nr ). The non-radiative decay can also be controlled by the
temperature but the radiative decay is completely determined by the LDOS. A
beautiful example of the latter effect was demonstrated on a single nanosphere
near a mirror [2], and a few subsequent studies [3–5].
Photonic band gap crystals have been demonstrated to strongly inhibit the
LDOS [6–10], and hence to control spontaneous emission of embedded quantum
emitters. Therefore in this section, we explore the feasibility of inhibiting the
radiative plasmonic damping rate, which is considered to be an important loss
mechanism of plasmons [11].
If we consider a plasmonic resonance as a dipole, then the rate of decay for
such a dipole transition between an initial i and an available final f dipole-field
state is proportional to the local density of states at the atom position (LDOS),
as stated by Fermi’s Golden Rule:
πωd2
γi→f (ω, r, ed ) = N (ω, r, ed ). (C.2)
0
−1
Here T1,r = γi→f .
138 Control radiative linewidth of plasmonics with photonic crystals
C.2 Investigation: controlling plasmonic radiation of

gold nanorods with photonic crystals
To control the radiative plasmonic loss we consider 2D and 3D silicon inverse
woodpile photonic crystals. As discussed in chapter 2, these crystals have the
broadest band gap among all fabricated crystals. Since a broader band gap
results in a smaller value for the LDOS inside the gap of a finite crystal, these
crystals are the most appealing for our purpose. Since the crystals are made of
silicon, and silicon is transparent for the range ω ≤ 1.1eV , thus photonic crystals
are designed such that they form a complete photonic band gap in the infrared
range.
In order to control the plasmonic radiation we need to choose the particles
such that their extinction spectrum not only overlaps with the band gap in the
near infrared (ω ≤ 1.1 eV), but is also narrower than the band gap range [12].
Among a large variety of different types of plasmonic particles, such as spheres,
core-shells, and nanorods, we have chosen gold nanorods. The nanorods have a
higher albedo and a narrower relative linewidth in the near infrared compared
to spheres and core-shell particles. Another advantage of the nanorods is that
their plasmon resonances can be tuned easily to the desired frequency range by
changing their aspect ratio [13].
C.2.1 Extinction of gold nanorod suspensions

Organic gold rod nano-particles have been purchased from NANOPARTz. They
are delivered as suspension in ”Short chained Alkane Alcohol” and can be max-
imally diluted 10× with an organic solvent. The nanorods have an aspect ratio
c=10, length l=100 nm and diameter d=10 nm, with a broad aspect ratio distri-
bution of around 50%. To measure the surface plasmon resonance peak we have
prepared the gold nanorods diluted in toluene with different concentrations (3×
and 5× dilution) and also a reference sample of only toluene. Since we want to
infiltrate particles in the photonic crystal, we expect their resonance frequency
to shift due to a change in the refractive index of the substrate. Therefore, we
have made also samples which consist of Au particles dried on top of a silicon
wafer, and in this case a piece of silicon wafer has been used as reference.
To understand the effect of aspect ratio distribution of the particles the spec-
trum of one single nanorod was calculated by Coupled Dipole Approximation
method, which is also known as the Discrete Dipole Approximation, as shown in
Fig. C.1∗ . The nanorod has a length of l=102 nm and a diameter of d=10 nm in
a homogenous environment of refractive index n=1.33. The calculation shows a
peak resonance at 1100 nm, with a FWHM of about 150 nm.
The extinction spectra of Au particles 3× diluted in toluene and on a silicon
wafer are shown in Fig. C.1. As it was specified by the company the resonance
frequency of particles in suspension is around λ=1400 nm. The measurement
shows a very broad extinction peak of 500 nm FWHM, broader than what we
∗ We thank Prof. W. L. Barnes and Alastair Humphrey for performing this calculation.
Investigation: controlling plasmonic radiation of gold nanorods with photonic crystals
139
0.02 100
Extinction cross section (μm2)

calculation
Au nanorods in toluene
Au nanorods on silicon wafer
80
Extinction(%)
60
0.01
40
20
0.00 0
1000 1500 2000
Wavelength(nm)
Figure C.1: Extinction spectrum of Au particles (E12N-10-1400), purchased from

NANOPARTz 3× diluted in toluene. The peak of intensity is around 1400 nm as
specified by the company (black solid curve), and dried on the surface of a silicon wafer
(red dotted curve). The particles on the silicon surface experience a higher refractive
index environment and therefore their resonance frequency shifts to the lower wave-
lengths as expected. Blue dashed curve shows the calculated extinction cross section
of one nanorod with length 100 nm and diameter 10 nm in a medium with refractive
index 1.33.
expect from the calculation. Comparing the measurements and the calculation
indicates that the linewidth measured in suspension is mostly dominated by inho-
mogeneous broadening. For the particles dried on the silicon wafer, the resonance
frequency is red shifted, and due to reflection from the silicon substrate the re-
flectivity increases for the entire range of the measurements.
In order to investigate the possibility of controlling plasmonic radiative losses
with silicon two-dimensional photonic crystals, we have plotted the measured
extinction spectrum of the nanorods together with reflectivity measurements of
a 2D crystal in Fig. C.2. The reflectivity spectra are measured on different
crystal directions and light polarizations as discussed extensively in [14]. In the
range of the stop gap the incident light cannot propagate inside the crystal and
is reflected back, appearing as a peak in the reflection spectrum. In this range
not only the external field is forbidden to enter the crystal, inside the crystal
the field propagation is also prohibited resulting in a low local density of states.
Therefore the reflectivity peaks are also an indication of the range of low LDOS
in the crystal, where the plasmonic radiation can be inhibited. Comparing the
different reflectivity peaks with the surface plasmon resonance peak, an overlap
in the ΓK direction with T M polarization of the incident light (red dashed curve)
is seen. Although in this direction the stop gap is very broad, the control over
the radiation loss is unfortunately not achievable since the extinction spectrum
is much broader.
During the synthesis, nanorods form a constant width but a large variation in
length. Afterwards, the particles of the wanted aspect ratio have to be filtered
out which is only possible with an accuracy of about 10 nm. This would result in
140 Bibliography
(a) (b)
100 100
TE_ ΓM TE_ Γ K TM_ ΓK
TM_ ΓK TM_ Γ M Au in toluene
80 80
Intensity %
Intensity %
60 60
40 40
20 20
0 0
6000 8000 10000 12000 6000 8000 10000 12000
-1 -1
Frequency(cm ) Frequency(cm )
Figure C.2: (a) Reflectivity spectra of a 2D silicon photonic crystal with different
light polarizations and different crystal directions. (b) Extinction spectrum of gold
nano-rods in comparison with reflectivity spectrum of the 2D photonic crystal. The
extinction spectrum has a very good overlap with ΓK direction and T M polarization of
the reflectivity spectrum. However, due to inhomogeneous broadening of the nano-rods
their extinction spectrum is much broader than the calculated spectrum, making them
ineligible for our purpose.
a large inhomogeneous linewidth. In future when higher quality nanorods with

narrow linewidth will be available, it will be possible to control the plasmonic
loss using different stop gaps, or even ultimately control the plasmonic loss using
a full 3D photonic band gap.
Bibliography
[1] C. F. Bohren and D. R. Huffman, Absorption and scattering of light by small
particles (John Wiley, 2008). — p.137.
[2] B. C. Buchler, T. Kalkbrenner, C. Hettich, and V. Sandoghdar, Measuring
the quantum efficiency of the optical emission of single radiating dipoles
using a scanning mirror, Phys. Rev. Lett. 95, 063003 (2005). — p.137.
[3] M. W. Knight, Y. Wu, J. B. Lassiter, P. Nordlander, and N. J. Halas, Sub-
strates matter: influence of an adjacent dielectric on an individual plasmonic
nanoparticle, Nano Lett. 9, 2188 (2009). — p.137.
[4] T. Kalkbrenner, U. Håkanson, A. Schädle, S. Burger, C. Henkel, and V.
Sandoghdar, Optical microscopy via spectral modifications of a nanoantenna,
Phys. Rev. Lett. 95, 200801 (2005). — p.137.
[5] K. Imura, T. Nagahara, and H. Okamoto, Near-field optical imaging of plas-
mon modes in gold nanorods, J. Chem. Phys. 122, 154701 (2005). — p.137.
organizing systems, Phys. Rev. E 58, 3896 (1998). — p.137.
Bibliography 141

Am. B 26, 987 (2009). — p.137.
[10] E. Yeganegi, A. Lagendijk, A. P. Mosk, and W. L. Vos, Local density of
optical states in the band gap of a finite one-dimensional photonic crystal,
Phys. Rev. B 89, 045123 (2014). — p.137.
[11] H. Mertens, A. F. Koenderink, and A. Polman, Plasmon-enhanced lumi-
nescence near noble-metal nanospheres: comparison of exact theory and an
improved Gersten and Nitzan model, Phys. Rev. B 76, 115123 (2007). —
p.137.
[12] I. S. Nikolaev, P. Lodahl, and W. L. Vos, Fluorescence lifetime of emitters
with broad homogeneous linewidths modified in opal photonic crystals, J.
Phys. Chem. C 112, 7250 (2008). — p.138.
[13] G. W. Bryant, F. J. G. de Abajo, and J. Aizpurua, Mapping the plasmon
resonances of metallic nanoantennas, Nano Lett. 8, 631 (2008). — p.138.
— p.139.
Algemene Nederlandse samenvatting
Het belang van licht voor de mensheid wordt weerspiegeld in de enorme inspan-
ningen die wereldwijd worden geleverd om haar ontstaan en voortplanting te
beheersen. Profiterend van de wisselwerking tussen licht en materie zijn zorg-
vuldig ontworpen optische elementen, zoals spiegels en lenzen, in de afgelopen
eeuwen succesvol toegepast voor het manipuleren van licht.
Sinds kort is het ook mogelijk om licht tot extreem te manipuleren, op nano-
schaal, door composieten te gebruiken met typische afmetingen in de orde van
een golflengte of nog kleiner.
Fotonische kristallen behoren tot de meest veelbelovende nanostructuren voor
het beheersen van lichtemissie en propagatie. Fotonische kristallen zijn nano-
structuren met een periodieke samenstelling, met een periodiciteit in de orde van
de golflengte van licht. Een ééndimensionaal fotonisch kristal is slechts in één
richting periodiek, zoals een ‘Bragg stack’. In dit geval kan de propagatie van
licht loodrecht op de lagen worden beheerst, door constructieve en destructieve
interferentie van het licht dat door verschillende kristalvlakken wordt gereflec-
teerd.
Wanneer de periodiciteit zich in twee dimensies voordoet, kan het licht ook in
twee dimensies worden beheerst. De uiterste beheersing over lichtpropagatie en
emissie kan worden verkregen wanneer de periodiciteit zich in alle drie dimensies
voordoet. In dat geval kan er voor een bepaald interval van de lichtfrequentie in
alle richtingen en voor beide polarisaties van het licht constructieve interferentie
optreden. Dat maakt het voor het licht onmogelijk om zich voort te planten
binnenin het kristal. Dit frequentiebereik wordt het bereik van de fotonische
bandkloof genoemd. Ook als er lichtbronnen zijn die zich in het kristal bevinden,
en licht uitzenden in het frequentiebereik van de bandkloof, kan het licht zich
niet voortplanten buiten het kristal.
De meest opmerkelijke eigenschap van een kristal met een fotonische bandkloof
is het beheersen van spontane emissie van licht. Bij spontane emissie zendt een
lichtbron een foton uit in een willekeurige richting en op een willekeurig tijdstip.
Dit proces wordt niet alleen bepaald door de lichtbron zelf, maar ook door zijn
omgeving. Overal om ons heen bevindt zich een flucturerend elektromagnetisch
veld. Deze fluctuaties bestaan zelfs in vacuüm. Hoewel de fluctuaties gemiddeld
genomen over de tijd nul zijn, zijn ze de hele tijd om ons heen. Door interactie
van een lichtbron met deze fluctuaties wordt een foton uitgezonden. Aangezien
de fluctuaties niet voorspelbaar zijn, is het principieel niet mogelijk om vooraf te
bepalen wanneer dit foton zal worden uitgezonden. Wat wel kan worden bepaald,
is de gemiddelde tijdsduur voor het uitzenden van een foton. Door gebruik te
144 Algemene Nederlandse samenvatting
makenvaneenfotonischFbandLMPPGkristalishetmogelijkomdezetoevallige
verschijnselen te beheersen.
Dat is precies wat er in ons experiment is gedaan. We veranderen de omge-
vingvandelichtbronnenineenomgevingwaargeenvacuümfluctuatiesaanwezig
zijn: “de fotonische bandkloof”. Voor de lichtbron wordt een suspensie van PbS
quantum dots in tolueen in de silicium ‘inverse woodpiles’ fotonische kristallen
JOHFCSBDIU. We meten de vervalsnelheid van de quantum dots in het fotonischF
bandLMPPGkristal;deratiovandevervalsnelheidvandequantumdotsinsuspensie
versusdevervalsnelheidvandequantumdotsinhetfotonischbandgapkristal.In
ons experiment zien we, voor het eerst ooit, een sterke afremming van spontane
emissie in echte fotonische bandkloof kristallen, met het waarnemen van niet
minderdanachttienkeerlangerelevensduurvanquantumdotsindebandkloof.
Volledige controle over de voortplanting van licht is alleen mogelijk in een
oneindig groot kristal. Omdat een oneindig kristal in de realiteit niet bestaat
en ieder fotonisch kristal eindig is, blijft het absoluut verboden gebied voor licht
een theoretisch concept. Maar wij hebben het concept experimenteel zo goed
mogelijk waargemaakt.
Degemetenemissievanquantumdotsindebandkloof,opgelostnaaroptische
frequentie, heeft geleid tot een nieuw begrip van effecten van eindige afmetin-
gen op de bandkloof bereik, hetgeen resulteert in een nieuw gezichtspunt. In
dit originele gezichtspunt, toegelicht in hoofdstuk 3 en 4 van dit proefschrift,
hebbenwedegolfvectorruimteendefrequentieruimteuitgebreidindecomplexe
ruimte. We hebben twee Ansatzes ontwikkeld: de eerste is gebaseerd op tun-
neling van vacuümfluctuaties vanuit de vrije ruimte in het fotonische bandkloof
kristal. Licht inclusief vacuümfluctuaties, dat zich voortplant in de vrije ruimte
tunnelt het kristal in door het imaginaire deel van de golfvector. Dit imaginaire
deelvandegolfvectoriseenfunctievandestructuurvanhetfotonischekristalen
is afhankelijk van de breedte van de bandLMPPG De tweede Ansatz is dat de eigen
trillingenvanhetkristaldiedeltafunctieszijnineenoneindigkristal,ver-breden
ineeneindigkristalenLorentzfunctiesworden.Wijstellenvoordatdebreedteen
het aantal van deze Lorentz-modes afhangt van de grootte van het kristal. Op
basis van deze twee Ansatzes ontwikkelden we een model om de lokale
toestandsdichtheidteberekenenindebandLMPPGvaneeneindigkristalenwever-
gelekendeberekeningenmetonzemetingen.DevergelijkingtoontFFOuitstekende
overeenkomst,enverifiëertonsmodel. Hetmodeliseenalgemeenmodeldatkan
wordengebruiktvoorallesoortenfotonischekristallenenverklaartheteffectvan
eindigheidindeemissieendepropagatievanlichtinallefotonischFkristallen.

Thesis EY-3

Uploaded by

Copyright:

Available Formats

You might also like

Thesis EY-3

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Thesis EY-3

Uploaded by

Copyright:

Available Formats

Controlling emission and propagation of

light with photonic band gap crystals

Beheersen van de emissie en propagatie

Promotores Prof. Dr. W. L. Vos

Overige leden Prof. Dr. L. Novotny

The work described in this thesis is ﬁnancially supported by the

This thesis can be downloaded from

ter verkrijging van

Elahe Yeganegi Dastgerdi

Prof. Dr. W. L. Vos en Prof. Dr. A. Lagendijk

2 Samples and instrumentation 27

3 Local density of states in the band gap of a ﬁnite one-dimensional

4 Finite size eﬀects on a 3D photonic band gap 55

5 Statistical analysis of time-resolved emission in the near infrared 73

6 Probing photonic crystals with shaped wavefronts 89

7 Light transport in silicon two-dimensional photonic crystals 109

8 Summary and outlook 125

A Light propagation in periodic media 129

B Decay rate versus frequency of PbS quantum dots 133

C Control radiative linewidth of plasmonics with photonic crystals 137

C.2 Investigation: controlling plasmonic radiation of gold nanorods

Algemene Nederlandse samenvatting 143

The importance of light to mankind is reﬂected by the tremendous eﬀorts that

1.1 Spontaneous emission under control

1.2 General wave description: Green’s functions

1.3 Photonic crystals

1.4 Diamond structured photonic crystals

Figure 1.3: (a) Schematic representation of the structure of an “inverse woodpile”

bandgaps of up to 28% bandwidth [58]. Unfortunately such diamond structures

1.5 Importance of three-dimensional photonic band

1.6 Real 3D photonic band gap crystals

1.7 Probing the local density of states

A third method to measure the LDOS is to measure the group velocity at

1.8 Outline of this thesis

In chapter 2, we describe the silicon inverse woodpile photonic crystals that we

In chapter 3, we present our theoretical study of the local density of states

Chapter 4 describes a quantitative study of ﬁnite size eﬀect in 3D photonic

Chapter 5 demonstrates a Bayesian statistical analysis approach that we ap-

In chapter 6, we present our study on probing the internal structure of synthe-

In chapter 7, we study out-of-plane scattering of light in 2D silicon photonic

Chapter 8 concludes the thesis. A summary of the thesis is presented and an

L. V. Keldysh, V. D. Kulakovskii, T. L. Reinecke, and A. Forchel, Strong

[30] M. Kanskar, P. Paddon, V. Pacradouni, R. Morin, A. Busch, J. F. Young,

[45] K. Aoki, D. Guimard, M. Nishioka, M. Nomura, S. Iwamoto, and Y.

dielectric structures, Phys. Rev. Lett. 61, 2269 (1988). — p.17.

2.2 Silicon inverse woodpile photonic crystals

At low frequency the DOS increases quadratically similar to that of a medium

2.3 Optical properties of PbS quantum dots

Emission Intensity (104 C.P.S)

However, the underlying mechanism of Stokes shift is still controversial. In case

The number of signal counts is typically 1% of the repetition frequency, or

2.4 Experimental setup

collected by a high numerical aperture objective, NA=0.72 (o1,Nikon CFIL Plan

it is possible to measure the decay curve of emitters at particular emission fre-

[5] J. M. van den Broek, L. A. Woldering, R. W. Tjerkstra, F. B. Segerink, I. D.

letters 86, 1634 (2001). — p.30.

simulations, and ﬁnally such calculations requires a lot of programming eﬀorts

3.2 Local density of states in a ﬁnite crystal

!