English

not defined

no text concepts found

Linköping Studies in Science and Technology Doctoral Dissertation No. 1115 Theoretical studies of light propagation in photonic and plasmonic devices Aliaksandr Rahachou Department of Science and Technology Linköping University, SE-601 74 Norrköping, Sweden Norrköping, August 2007 To Olga. The picture on the cover illustrates the concept of a ”photonic micropolis”. Adopted from http://ab-initio.mit.edu/photons/micropolis.html. Theoretical studies of light propagation in photonic and plasmonic devices c 2007 Aliaksandr Rahachou Department of Science and Technology Campus Norrköping, Linköping University SE-601 74 Norrköping, Sweden ISBN 978-91-85831-45-6 ISSN 0345-7524 Printed in Sweden by UniTryck, Linköping, 2007 Preface Science is about to discover God. I start worrying about His future. Stanislaw Jerzy Lec, Polish poet (1909–1966) When people ask me: ”Where do you work and what are you doing?” I have a standard answer that I work at Linköping University and do my PhD in physics. In most cases it is enough, people make their faces serious and say: ”O-o-o! Physics!” But sometimes, I need to explain in more detail that my research area is actually photonics, and I study things related to propagation of electromagnetic waves in some strange media, created by people in order to deceive Nature. People become serious at this point and say: ”O-o-o! Photonics!” I like these moments and I like what I have been doing during these years – photonics. This Thesis presents the results of the four-year work that was done in the Solid-state Electronics group at the Department of Science and Technology at Linköping University. This is a theoretical work, which touches three main directions in photonics, namely photonic crystals, microcavities and plasmonics. The Thesis consists of four chapters. Chapter 1 is a short introduction where I familiarize the reader with the subject. Chapter 2 gives an introductory review of photonic structures, in Chapter 3 I present the methods that were developed during my study, and in Chapter 4 the results are summarized and briefly discussed. This Dissertation is based on seven papers, presented in the Appendix. Almost everywhere in the text I use ”we” instead of ”I”, despite my royal roots are neither that clear nor documented anywhere. This is just to emphasize that any research is never a single person’s but a team work. Aliaksandr Rahachou Norrköping, Midsommar, 2007 iii iv Acknowledgements Well, if you have already read the Preface, you probably know, that this Thesis is a result of the four-year work at ITN LiU in Sweden. During this time I met a lot of nice people, who helped me not only in my research activity, but also supported me in everyday routine. First of all I would like to thank Igor Zozoulenko for the brilliant supervision. He introduced me to the fascinating area of photonics, spent really loads of time answering my stupid questions, discussing, encouraging and sometimes pushing me to do something ,. I am very grateful to Olle Inganäs for the valuable discussions, initiation of this work and the experimental input. I was also pleased to collaborate with Kristofer Tvingstedt, whose unexpected ideas from the point of view of an experimentalist helped me to understand the subject deeper. Then, the guys from our group – Martin Evaldsson and Siarhei Ihnatsenka. Despite we did a little bit different things, Martin and Siarhei not only always understood what I was doing, but also helped me with practical things like LATEX, elementary school-level math or other important issues that bothered me sometimes. Of course, all people at ITN are very kind. Thank you, people! However, I’d like especially thank Aida Vitoria for good humor, which is, despite the weather, season or Iraq war, remains sparkling. Big thanks to my mother and father. Being far away from them, I feel their love and support every day. Thanks to my girlfriend Olga Mishchenko. She supports and helps me from day to day, her love and kindness is just a miracle that I revealed here in Sweden. Tack, Sverige! I very appreciate the financial support from the Swedish Institute (SI), Royal Swedish Academy of Sciences (KVA), National Graduate School of Scientific Computing (NGSSC), Center of Organic Electronics (COE), Centre in Nanoscience and Technology at LiU (CeNANO) and ITN that enabled me to start and complete this Thesis. v vi Abstract Photonics nowadays is one of the most rapidly developing areas of modern physics. Photonic chips are considered to be promising candidates for a new generation of high-performance systems for informational technology, as the photonic devices provide much higher information capacity in comparison to conventional electronics. They also offer the possibility of integration with electronic components to provide increased functionality. Photonics has also found numerous applications in various fields including signal processing, computing, sensing, printing, and others. Photonics, which traditionally covers lasing cavities, waveguides, and photonic crystals, is now expanding to new research directions such as plasmonics and nanophotonics. Plasmonic structures, namely nanoparticles, metallic and dielectric waveguides and gratings, possess unprecedented potential to guide and manipulate light at nanoscale. This Thesis presents the results of theoretical studies of light propagation in photonic and plasmonic structures, namely lasing disk microcavities, photonic crystals, metallic gratings and nanoparticle arrays. A special emphasis has been made on development of high-performance techniques for studies of photonic devices. The following papers are included: In the first two papers (Paper I and Paper II) we developed a novel scattering matrix technique for calculation of resonant states in 2D disk microcavities with the imperfect surface or/and inhomogeneous refraction index. The results demonstrate that the surface imperfections represent the crucial factor determining the Q factor of the cavity. A generalization of the scattering-matrix technique to the quantum-mechanical electron scattering has been made in Paper III. This has allowed us to treat a realistic potential of quantum-corrals (which can be considered as nanoscale analogues of optical cavities) and has provided a new insight and interpretation of the experimental observations. Papers IV and V present a novel effective Green’s function technique for studying light propagation in photonic crystals. Using this technique we have analyzed surface modes and proposed several novel surface-state-based devices vii viii for lasing/sensing, waveguiding and light feeding applications. In Paper VI the propagation of light in nanorod arrays has been studied. We have demonstrated that the simple Maxwell Garnett effective-medium theory cannot properly describe the coupling and clustering effects of nanorods. We have demonstrated the possibility of using nanorod arrays as high-quality polarizers. In Paper VII we modeled the plasmon-enhanced absorption in polymeric solar cells. In order to excite a plasmon we utilized a grated aluminum substrate. The increased absorption has been verified experimentally and good agreement with our theoretical data has been achieved. Contributions to the papers All the enclosed papers constitute the output of a 4-year close collaboration between the authors, involving a permanent, almost everyday, exchange of the ideas and discussions during the whole process. Therefore, it is hard to pick out my own effort, but an attempt is the following: • Paper I: A. Rahachou and I. V. Zozoulenko, Effects of boundary roughness on a Q factor of whispering-gallery-mode lasing microdisk cavities, J. Appl. Phys., vol. 94, pp. 7929–7931, 2003 • Paper II: A. Rahachou and I. V. Zozoulenko, Scattering matrix approach to the resonant states and Q values of microdisk lasing cavities, Appl. Opt., vol. 43, pp. 1761–1772, 2004 In the first two papers I implemented both the serial and parallel versions of the scattering matrix (SM) technique in Fortran 95, performed all the calculations and summarized the results. I also derived necessary equations for the Husimi-function analysis, developed and implemented the ray tracing problem in the Poincaré surface-of-sections part. I also gave an idea of the enhanced transmission of the high-Q whisperinggallery modes through a curved surface. I believe I tried to write the papers, but... They were rewritten by Igor anyway. • Paper III: A. Rahachou and I. V. Zozoulenko, Elastic scattering of surface electron waves in quantum corrals: Importance of the shape of the adatom potential, Phys. Rev. B, vol. 70, pp. 233409 1–4, 2004 I adapted the SM technique to the quantum-mechanical problem and did all the calculations. Took part in the discussions and interpretation of the results. First several unsuccessful iterations of the paper were actually mine... • Paper IV: A. Rahachou and I. V. Zozoulenko, Light propagation in finite and infinite photonic crystals: The recursive Greens function technique, Phys. Rev. B, vol. 72, pp. 155117 1–12, 2005 ix x I derived some of the matrix equations (combination of the Green’s functions) and implemented the method in both serial and parallel Fortran 95 codes. I also performed all the calculations, took part in discussions. I wrote the introduction and results/discussion parts of the paper. • Paper V: A. Rahachou and I. V. Zozoulenko, Waveguiding properties of surface states in photonic crystals, J. Opt. Soc. Am B, vol. 23, pp. 1679–1683, 2006 I carried out all the calculations, suggested the idea of the directional beamer and wrote the paper. • Paper VI: A. Rahachou and I. V. Zozoulenko, Light propagation in nanorod arrays, J. Opt A, vol. 9, pp. 265–270, 2007 I adapted the Green’s function technique to the plasmonic applications. I proposed some of the structures, made all the computations, summarized, discussed and analyzed the results. Then I wrote the paper. After serious Igor’s criticism it finally came to its present state... • Paper VII: K. Tvingstedt, A. Rahachou, N.-K. Persson, I. V. Zozoulenko, and O. Inganäs, Surface plasmon increased absorption in polymer photovoltaic cells, submitted to Appl. Phys. Lett., 2007 I made all the calculations, analyzed the results and wrote the theoretical part of the paper. Contents Abstract vi Contributions to the papers ix Table of Contents xi 1 INTRODUCTION 1 2 Photonic structures 2.1 Whispering-gallery-mode lasing microcavities . . . . . . . . . 2.1.1 General principle of lasing operation . . . . . . . . . . 2.1.2 Total internal reflection and whispering-gallery modes 2.2 Surface states in photonic crystals . . . . . . . . . . . . . . . 2.2.1 Photonic crystals . . . . . . . . . . . . . . . . . . . . . 2.2.2 Surface states and their applications . . . . . . . . . . 2.3 Surface plasmons . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Excitation of surface plasmons . . . . . . . . . . . . . 2.3.2 Applications of surface plasmons . . . . . . . . . . . . 2.4 Nanoparticles . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Properties of nanoparticles and Mie’s theory . . . . . 2.4.2 Nanoparticle arrays and effective-medium theories . . 2.4.3 Applications of nanoparticles . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Computational techniques 3.1 Available techniques for studying light propagation in photonic structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Scattering matrix method . . . . . . . . . . . . . . . . . . . . . 3.2.1 Application of the scattering matrix method to quantummechanical problems . . . . . . . . . . . . . . . . . . . . 3.3 Green’s function technique . . . . . . . . . . . . . . . . . . . . . 3.4 Dyadic Green’s function technique . . . . . . . . . . . . . . . . xi 5 5 5 7 11 11 15 16 16 19 21 21 24 24 27 27 28 31 32 38 xii CONTENTS 4 Results 4.1 Effect of inhomogeneities on quality factors of disk microcavities (Papers I, II) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Quantum corrals (Paper III) . . . . . . . . . . . . . . . . . . . . 4.3 Surface-state lasers (Paper IV) . . . . . . . . . . . . . . . . . . 4.4 Surface-state waveguides (Paper V) . . . . . . . . . . . . . . . . 4.5 Nanorod arrays (Paper VI) . . . . . . . . . . . . . . . . . . . . 4.6 Surface plasmons in polymeric solar cells (Paper VII) . . . . . . 43 Bibliography 59 Appendix I Paper II Paper III Paper IV Paper V Paper VI Paper VII Paper I . II . III IV V . VI VII . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 45 47 49 52 56 69 71 77 91 97 111 119 127 Chapter 1 INTRODUCTION 1 2 CHAPTER 1. INTRODUCTION The idea of constructing chips that operate on light signals instead of electricity has engaged the minds of scientists during the last decade. Communicating photons instead of electrons would provide revolutionary breakthrough not only in the performance of devices, which can distribute data at the speed of light, but also in the capacity of transmitted data. By now, modern optical networks can provide such a bandwidth, that even the fastest state-of-the-art processors are unable to handle, and this trend seems to remain in nearest future. Furthermore, photons are not so strongly interacting as electrons/holes that significantly broaden bandwidth. Speaking about present time, only photonics provides solutions for high-dense modern data storage, like CDs and DVDs, whose capacity is constantly increasing. Manufacturing practical photonic chips, however, brings in several challenges: first of all, lack of all-optical logic switches themselves as well as the principles of their operation, technological difficulties in manufacturing of novel photonic devices with the same well-developed processes for electronic chips, and, finally, the need of novel materials. In this regard, the most promising ”building blocks” of modern photonics are photonic crystals, lasing microcavities and plasmonic devices, which, being intensively studied during the latest decade, can provide the required functionality and microminiaturization. Along with opportunities for integration of optical devices, photonic crystals exhibit a variety of unique physical phenomena. Photonic crystal is usually fabricated from the same semiconductor materials as electronic chips using common chipmaking techniques like photolithography. The main reason that has made photonic crystals so popular is their basic feature of having gaps in the energy spectrum that forbid light to travel at certain wavelengths. Such the gaps in the spectra provide very effective confinement of the light within photonic crystals that can be exploited as a basis for a large number of photonic devices. Creating linear defects, for instance, will form low-loss waveguides, whereas point defects can act as high-quality microcavities. Another unique feature of photonic crystals with certain lattice parameters is the negative refraction index that can be exploited for focusing and non-conventional distribution of light on a microscopic level. In additional, real finite photonic crystals can support surface states on their boundaries, which can also be exploited for different purposes in photonic chips. Optical microcavities are structures that confine light and enable lasing action on a microscopic scale. In conventional lasers, a significant portion of the pump energy simply dissipates, and a rather high threshold power is required to initiate the lasing effect. In contrast, microcavities can be utilized to sustain highly efficient, almost ”thresholdless”, lasing action. Such the efficiency is related to the existence of the natural cavity resonances. These resonances are known as morphology-dependent resonances or whispering gallery modes. The origin of these resonances can be addressed to ray dynamics, when the light is trapped inside the cavity through total internal reflection against its 3 circumference. An ideal lossless cavity would trap this ”rotating” light for infinitely long time and would have infinitely narrow lasing peaks. Combining microcavities into arrays or coupling them to waveguides creates variety of devices for sensing and filtering. Ultra high-quality microcavities can also be utilized in stunning applications such as single atom detection. Plasmonic structures is the ”State of the Art” of modern photonics. Plasmons, the electromagnetic modes localized at metal-dielectric interfaces and metallic nanoparticles, bring in new unprecedented opportunities of guiding and manipulating light beyond the diffraction limit. Novel plasmonic waveguides and their arrangements are able to distribute light on nanoscale, providing the missing link between highly-integrated electronic chips and larger-scale photonic components. Enhanced field intensities of plasmonic modes are utilized in a variety of applications – from biological sensors to spectroscopy and lasing structures. The Thesis is organized as follows. In Chapter 2 we make a brief overview of photonic structures under the study, namely microdisk cavities, photonic crystals and plasmonic devices. Chapter 3 outlines the scattering matrix and Green’s function techniques, and Chapter 4 summarizes the main results and contains discussions. 4 CHAPTER 1. INTRODUCTION Chapter 2 Photonic structures 2.1 2.1.1 Whispering-gallery-mode lasing microcavities General principle of lasing operation The word ”LASER” is an acronym for Light Amplification by Stimulated Emission of Radiation. The output of a laser is a highly-coherent monochromatic (in a very ideal case) radiation, which can be pulsed or beamed in a visible, infrared or ultraviolet range. The power of a laser can vary from several milliwatts to megawatts. The main and the most crucial component of a laser is its active medium, which can be a solid, gas, liquid or semiconductor. In thermodynamic equilibrium nearly all atoms, ions or molecules (depending on the particular laser) of the active medium occupy their lowest energy level or ”ground state”. To produce laser action, the majority of atoms/ions/molecules should be ”pumped” up into the higher energy level, creating so called population inversion. Typical three-level structure is given in Fig. 2.1(a). Pump energy here excites atoms from the ground state to the short-lived level, which rapidly decays to the long-lived state. At random times, some of these excited atoms/ions/molecules will decay to the ground state on their own. Each decay is accompanied by the emission of a single photon propagating in a random direction (spontaneous emission). However, when one of these photons encounters an excited atom/ion/molecule, the latter will drop down to a lower energy state and emit a new photon with exactly the same wavelength, phase, direction and polarization. This is called stimulated emission. When a photon is emitted nearly parallel to the long side of the cavity [Fig. 2.1(b)] it will travel down to one of the mirrors and be able to get reflected back and forth many times. Along its way, it hits excited atoms/ions/molecules and 5 6 CHAPTER 2. PHOTONIC STRUCTURES Equilibrium Pumping Fast Relaxation Stimulated emission short-lived level p en ump er gy long-lived level hν hν ground level Totally reflecting mirror Active medium (a) Partially reflecting mirror hν (b) Figure 2.1: (a) Three-level diagram of a lasing system. (b) Lasing cavity. Output intensity ”stimulates” them to emit up new photons. The process acts as an avalanche caused by a single photon which produces more and more photons via this stimulated emission process. When the energy of the photon beam becomes enough to make the beam escape the partially reflecting mirror, a highly monochromatic and coherent ray goes out. Depending on the type of a cavity the beam can be well collimated or appears to originate from a point/plane source. Spontaneous emission Threshold PTH Stimulated emission Pump power Figure 2.2: Threshold of a laser. One of the most important parameters of lasers is their threshold power PT H , that can be defined as the ”critical” pumping power that corresponds to the initiation of the stimulated emission (see Fig. 2.2). The threshold is proportional to the threshold population difference, i.e. the minimum positive difference in population between the long-lived and ground levels in Fig. 2.1 NT = Nll − Ng ∼ 1 ω0 = cτp cQ (2.1) 2.1. WHISPERING-GALLERY-MODE LASING MICROCAVITIES 7 where c is the speed of light, τp is a photon lifetime, ω0 is a resonant frequency of a lasing mode and Q is a quality factor (Q factor hereafter) of a lasing cavity. The main goal is obviously to minimize the threshold power, therefore maximize the photon lifetime and cavity quality factor. The Q factor is strongly determined by the design of a cavity. Several representative examples are given in Fig. 2.3. (a) (d) (b) (e) (c) (f) Figure 2.3: Different types of lasing cavities. (a) Confocal resonator. Employed in a variety of gas, solid-state and chemical lasers. Two confocal mirrors (one of them is partially reflecting) create a collimated beam parallel to the long side of the cavity. (b) Laser diode. The cavity is created by finely polished side walls of the structure. (c) Photonic-crystal cavity. The cavity is created by a point inhomogeneity in a photonic-crystal lattice (see the next section for details). Q factor can reach 105 . (d) Fabri-Perot resonator. A set of stacked Bragg mirrors provides cavity confinement. Typical value of the Q factor is ∼ 2000. (e) Whispering-gallery disk microcavity. Light is trapped inside the cavity, undergoing multiple ”bounces” against the side wall due to the effect of total internal reflection. Q ∼ 104 , toroidal cavities with Q ∼ 108 have been also reported [1]. (f) A spherical whispering-gallery droplet. Q ∼ 108 . (c-f) are adopted from [2]. 2.1.2 Total internal reflection and whispering-gallery modes One of the most well-known mechanisms of the ray confinement in cavities is based on the effect of total internal reflection, which is presented in Fig. 2.4. 8 CHAPTER 2. PHOTONIC STRUCTURES The angle θc = arcsin n1 n2 (2.2) is called the critical angle for total internal reflection. At larger incidence angles θ2 the ray remains fully reflected. For curved boundaries [see Fig. 2.4(b)] the regime of total internal reflection and the critical angle (2.2) have the same meaning. However, because of the diffraction at the curved boundary, a leakage takes place. Transmission coefficient for an electromagnetic wave penetrating a curved boundary in the regime of total internal reflection reads [3] 3/2 2 nkρ 2 2 T = |TF | exp − cos θc − cos θ , (2.3) 3 sin2 (θ) where TF is a classical Fresnel transmission coefficient for an electromagnetic wave incident on a flat surface, k is a wavevector of the incident wave, ρ is a radius of curvature, and θ is an angle of incidence. The main goal, obviously, is to minimize T , in order to hold the light ”trapped” inside the cavity as long as possible. n1 n1 θc θc ρ n2 n2 (a) (b) Figure 2.4: (a) The regime of total internal reflection for (a) a flat surface, (b) a curved surface. The ray falls from medium 2 to the boundary with medium 1 (n1 < n2 ) at incidence angle θ2 and gets refracted to medium 1 at θ1 . According to the Snell’s law, n1 sin θ1 = n2 sin θ2 . If θ2 is being increased, at some particular incidence angle θc , angle θ1 becomes equal π/2 that corresponds to the full internal reflection of the incident beam. Total internal reflection is a mechanism of light localization in whisperinggallery cavities. The term whispering-gallery modes (WGMs) came after the whispering gallery at St. Paul’s Cathedral in London, see Fig. 2.5(a), where the quirk in its construction makes a whisper against its walls audible at the 2.1. WHISPERING-GALLERY-MODE LASING MICROCAVITIES 9 opposite side of the gallery. In whispering-gallery cavities [Fig. 2.5(b)] WGMs occur at particular resonant wavelengths of light for a given cavity size. At these wavelengths the light undergoes total internal reflection at the cavity surface and remains confined inside for a rather long time. In the WGM regime the θ>θc (a) (b) Figure 2.5: (a) The dome of the St. Paul’s Cathedral in London. The white line outlines distribution of a WG-mode. (b) Multiple reflections of a whisperinggallery mode against the circumference of the cavity. mode is localized near the circumference of a cavity and can be assigned a radial and angular mode numbers. The angular mode number n shows the number of wavelengths around the circumference, and the radial mode number l – the number of maxima in the intensity of the electromagnetic field in the radial direction within the cavity. A typical experimental spectrum of the WG modes is given in Fig. 2.6(a). Each whispering-gallery lasing mode of a cavity is characterized by its quality factor Q, which, by the definition, is also related to the width of the resonant spectral line as Q≡ 2π(stored energy per cycle) k = (energy loss per cycle) ∆k (2.4) where ∆k is a spectral line broadening taken at the half-amplitude of the lasing peak as it shown in Fig. 2.6(b). Q factor is also closely related to the time that the WG mode spends trapped within a cavity, so-called ”Wigner delay time” [4] Q = ωτD (ωres ), (2.5) where ω is a resonant frequency. The main reason of using whispering-gallery mode cavities is their high Q values as well as excellent opportunities to be integrated into optical chips. Lasing whispering-gallery modes were first observed in spherical glass droplets. An CHAPTER 2. PHOTONIC STRUCTURES Intensity 10 ∆ k - Broadening k Wavevector (a) (b) Figure 2.6: (a) Experimental spectrum of a whispering-gallery lasing microcavity [5]. Angular and radial mode numbers are also given. (b) Broadening of a lasing peak. important step was the development of microdisk semiconductor lasers, which exploited total internal reflection of light to achieve the perfect mirror reflectivity. These lasers – the smallest in the world at the time, were invented and first demonstrated in 1991 by Sam McCall, Richart Slusher and colleagues at Bell Labs. Microdisk, -cylinder or -droplet lasers form a class of lasers based on circularly symmetric resonators, which lase in whispering-gallery modes. These tiny lasers, however, lack for directional emission due to their circular symmetry. The experimental microlasers of Bell Labs and Yale team overcame this limitation. They were based on a new optical resonator shaped as a deformed cylinder (quadruple) and were highly directional. They exploited the concept of chaotic dynamics in asymmetric resonant cavities and were introduced by Nöckel and Stone at Yale in 1997. By now there have been reported cavities with Q factors of order ∼ 108 [1] with characteristic diameters ∼ 100µm. The another advantages are their relatively easy fabrication process (i.e. they can be etched on a surface [5] or pedestal [6], highly-symmetrical spherical cavities [7] are formed through the surface tension in silica); broad range of pumping methods (optical pump from the outside [5] or by the build-in quantum dots [6]; use of active polymers [8]); as well as a set of different shapes (disk, toroid, spherical, hexagonal, quadruple) possessing unique properties. Unfortunately, quality factors in actual fabricated microcavities are normally several orders lower than the corresponding calculated values of ideal cavities. A degradation of the experimental Q factors may be attributed to a variety of reasons including side wall geometrical imperfections, inhomogeneity of the refraction index of the disk, effects of coupling to the substrate or pedestal and others. A detailed study of effects of the factors above on the characteristics and performance of the microcavity lasers appears to be of crucial 2.2. SURFACE STATES IN PHOTONIC CRYSTALS 11 importance for their optimization. Of the especial importance are the studies of surface roughness of the cavities, as it have been demonstrated [9; 6; 10] to be the main factor affecting the Q value. Such the studies would require a versatile method that can deal with both the complex geometry and variable refraction index in the cavity. In the next Chapter we develop a novel computational technique, which is capable to handle disk microcavities both with geometrical imperfections and refraction index inhomogeneities. 2.2 2.2.1 Surface states in photonic crystals Photonic crystals Photonic crystals (PCs) or photonic bandgap materials are artificial structures, which forbid propagation of light in particular ranges of frequencies, remaining transparent for others. Photonic band gaps were first predicted in 1987 by two physicists working independently. They were Eli Yablonovitch, at Bell Communications Research in New Jersey, and Sajeev John of the University of Toronto. A periodic array of 1mm holes mechanically drilled in a slab of a material with the refraction index 3.6 was found to prevent microwaves from propagating in any direction. This structure received a name Yablonovite. Despite this remarkable success, it took more than a decade to fabricate photonic crystals that work in near-infrared (780-3000 nm) and visible (450-750 nm) ranges of the spectrum and forbid light propagation in all directions. The main challenge was to find suitable materials and technologies to fabricate structures that are about a thousandth the size of the Yablonovite. Let us now compare light propagation in a photonic crystal to the carrier transport in a semiconductor. The similarity between electromagnetic waves in PCs and de-Broglie electronic waves propagating in a crystalline solid has been utilized to develop theories of photonic crystals. For electrons in semiconductor materials the Schrödinger equation reads as 2 2 ~ ∇ − + V (r) Ψ(r) = EΨ(r). (2.6) 2m∗ In a semiconductor crystal the atoms are arranged in a periodic lattice, and moving carriers experience a periodic atomic lattice potential V (r + a) = V (r), (2.7) where a is a lattice constant. Then, there exists a wavevector k in the reciprocal lattice such that Ψ(r) can be written as Ψ(r) = eikr uk (r), (2.8) 12 CHAPTER 2. PHOTONIC STRUCTURES where uk (r + a) = uk (r) is a periodic function on the lattice. This expression is known as Bloch’s theorem. Substituting it into Eq. (2.6) one finds the eigenfunctions uk (r) and eigenvalues Ek . The periodic potential causes formation of allowed energy bands separated by gaps. In perfect bulk semiconductor crystals no electrons or holes can be found in these energy gaps. The situation holds also for photons traveling through periodic structures. Let us consider a periodic structure, e.g. a block of a transparent dielectric √ material of the high refraction index (related to a permittivity as n = ǫ) with ”drilled” holes or, vice versa, a periodic set of high-index dielectric rods in air background. In this case the corresponding electromagnetic wave equation (Maxwell’s equation for the magnetic field ) reads 1 ∇× ∇× H(r) = (ω 2 /c2 )H(r), (2.9) ǫ(r) with the periodic dielectric function ǫ(r + R) = ǫ(r). (2.10) ε1 ε2 (a) (b) (c) (d) (e) (f) Figure 2.7: Examples (a-c) of 1D, 2D and 3D photonic crystals and (d-f) corresponding band structures. (adopted from [11]) For a photon, the periodic dielectric function acts just as the lattice potential that an electron or hole experiences propagating through a semiconductor crystal. If the contrast of the refraction indexes is large, then the most of 2.2. SURFACE STATES IN PHOTONIC CRYSTALS 13 the light will be confined either within the dielectric material or the air. This confinement causes formation of intermingled allowed and forbidden energy regions. It is possible to adjust the positions of bandgaps by changing the size of the air holes/rods in the material/air or by variation of the refraction index. It is worth mentioning that the similarity between electrons in semiconductors and photons in photonic crystals is not complete. Unlike the Schrödinger’s equation for electron waves, the Maxwell’s equations and electromagnetic waves are vectorial that requires an additional computational effort. On the other hand, the Schrödinger’s equation can include many-body interactions, which are not the case for electromagnetic problems. Another important aspect is periodicity of photonic crystals. If the periodicity in the refraction index holds only in one direction (i.e 1D photonic crystal), only light traveling perpendicularly to the periodically arranged layers is affected. Any 1D structure supports bandgaps. In the 2D case, light propagating in the plane perpendicular to the rods will be affected. In order to make a complete bandgap for any direction of light propagation, a 3D structure have to be constructed. Fig. 2.7 illustrates 1D, 2D and 3D photonic crystals along with their band structures. Photonic crystal devices normally operate in the frequency regions corresponding the bandgaps. The area of possible applications is constantly expanding, some representative examples are given in Fig. 2.8. 14 CHAPTER 2. PHOTONIC STRUCTURES (a) (b) (c) (d) (e) (f) (g) (h) Figure 2.8: (a)Low-threshold cavity lasers. A properly designed point defect in a photonic crystal can act as a lasing cavity. Strong confinement of the field within the defect area enables one to achieve quality factors of order ∼ 106 [12; 13]. (b) Band-edge lasers. Photonic crystal operates at the energy of the band edge, where the velocity of light is very low, that causes long lifetime and high Q factor of the given state at this energy [14]. (c) Surfacestate lasers. Braking the translation symmetry of the surface of a photonic crystal turns a surface mode into a resonant state with the high Q factor. The unique feature of such the cavity is its location on the surface of a PC [15; 16]. (d) Low-loss waveguides with wide curvature. In optical integrated circuits, construction of low-loss waveguides with wide curvature is essential. When PCs are fabricated using low-loss dielectric materials, they act as perfect mirrors for the frequencies in the gap [17]. (e,f) Channel add/drop filters. Enable switching and redistributing light of certain frequencies between two or more waveguides [13; 18]. (g) Photonic bandgap microcavity in a dielectric waveguide. Acts as a filter in dielectric waveguides, suppresses all frequency range except for the frequencies of the resonant states of the PC-cavity [19]. (h) Optical transistor. Based on the Kerr effect. The intensity of the control beam (transverse waveguide) affects the Kerr cell, switching the light in the longitudinal waveguide [20]. 2.2. SURFACE STATES IN PHOTONIC CRYSTALS 2.2.2 15 Surface states and their applications 0.55 0.55 0.50 0.50 0.45 0.45 ωa/2πc ωa/2πc Surface states or surface modes is a special type of states in a photonic crystal that reside at the interface between a semi-infinite PC and open space, decaying into both the crystal and air [21]. Not every PC boundary supports surface states. For example, surface modes can be always found on the surface of a truncated 2D hexagonal array of holes in a material. At the same time, no surface state are found on the unmodified surface of a semi-infinite square array of cylinders in the air background. For the latter case the surface states appear in the bandgap of a square-lattice photonic crystal when its boundary is modified by, e.g., truncating the surface rods, shrinking or increasing their size, or creating more complex surface geometry [21; 22; 23; 24]. Examples of structures supporting surface states along with their band diagrams are given in Fig. 2.9. 0.40 0.35 0.40 0.35 0.30 0.30 0.25 0.25 0.20 0.30 0.35 0.40 ka/2π 0.20 0.30 0.45 (a) 0.35 0.40 ka/2π 0.45 (b) Figure 2.9: Band structures for the TM modes in the ΓX direction of squarelattice photonic crystals composed of rods with diameter D = 0.4a (a is the lattice constant) and permittivity ε = 8.9 along with the projected surface modes. The surface rods are (a) reduced to d = 0.2a and (b) half-truncated. The right panels show the intensity of the Ez component of the surface modes at the energies denoted with the arrows. So, why do the surface states in PCs attract our attention? Thanks to their unique location, on the surface of a photonic crystal, they open up new possibilities of coupling photonic devices to external light sources, stimulate directional beaming [25] from the waveguide opening on the surface. It is worth to emphasize that the surface mode residing on the infinitely long boundary of a semi-infinite crystal represents a truly bound Bloch state with the infinite lifetime and Q factor, and consequently does not couple or leak to air states. We have recently shown (see Paper V) that this feature enables surface states to be exploited as high-quality surface waveguides and directional beamers, which, being situated on the surface of a PC, provide unique opportunities in redistributing light in photonic chips. It has also been demonstrated (Paper IV), [15; 16] that when the translational symmetry along the boundary of the 16 CHAPTER 2. PHOTONIC STRUCTURES semi-infinite crystal is broken, the Bloch surface mode turns into a resonant state with a finite lifetime. This effect can be utilized for lasing and sensing purposes. 2.3 Surface plasmons 2.3.1 Excitation of surface plasmons Surface plasmons (SPs) are electromagnetic surface waves that propagate along the boundary between a metal and dielectric. They originate from collective oscillations of the electron density in the metal near the boundary under the external excitation. They were referred by Ritchie for the first time in 1950-th [26], and since then have attracted increased attention due to their extraordinary ability to guide and manipulate light at nanoscale. Figure 2.10 illustrates the p-polarized electromagnetic field (i.e. field, which has its electric component parallel to the plane of incidence) propagating towards the boundary of two media at angle of incidence θ. Boundary conditions for the electric fields imply that Ez By +++ −−− +++ −−− ε1’>0 Ex θ z δ |E| x y z Metal (ε2’<0) (a) (b) Figure 2.10: (a) Excitation of a plasmon on the metal-dielectric interface with p-polarized light, propagating at angle of incidence θ greater than the angle of total internal reflection. Inset illustrates the surface charges. (b) Plasmoninduced field intensity at the interface. the Ex -component is conserved across the boundary (i.e. Ex1 = Ex2 ), whereas the Ez -component undergoes a discontinuity, such that ε1 ε0 Ez1 = ε2 ε0 Ez2 . This discontinuity results in polarization changes at the interface and, consequently, additional localized surface charges [see inset to Fig. 2.10(a)]. The electromagnetic field, induced by these charges, represents a plasmonic mode, which is localized near the interface and propagates along it. It is worth mentioning 2.3. SURFACE PLASMONS 17 that the s-polarized light (which has its magnetic components parallel to the plane of incidence) does not generate any surface charges and, therefore, does not excite a plasmonic mode. The plasmonic mode is localized in the dielectric over the distance δ, which approximately equals the half wavelength of the incoming light in the dielectric, whereas in the metal its localization is determined by the metal skin depth (∼ 10 nm). The propagation length of the plasmon depends on the absorbing properties of the metal (the imaginary part of the dielectric function ε′′2 ). Thus, for low-ε′′ metals, such as silver in the infrared, the propagation length can reach hundreds of micrometers, but for the high-ε′′ ones (aluminum) it hardly exceeds tens of microns [27]. The dispersion relation for a plasmonic mode reads as [28] kx = k ε1 ε2 ε1 + ε2 1/2 , (2.11) where k = 2π/λ. This relation clearly shows the condition for excitation of a plasmonic mode: ε′2 has to be negative and |ε2 | > ε1 , which means that a plasmonic mode can only be excited on the surface of a metal. The another important conclusion from (2.11) is that the real part of the plasmon wavevector is always greater than the wavevector of the exciting radiation (see Fig. 2.11). Because of this, it is not possible to excite a plasmon on the flat surface with a propagating light beam. ω ω=ck ω k<kx kx Figure 2.11: The dispersion curve of a plasmonic mode. The curve lies beyond the light cone that does not allow direct excitation of a plasmon with propagating light. In order to enhance the wavevector of the exciting light (and thus to be able to excite a plasmon), several techniques have been proposed. They are 18 CHAPTER 2. PHOTONIC STRUCTURES illustrated in Fig. 2.12. The first two techniques, outlined in Fig. 2.12(a) and (b), are based on the excitation of a plasmon with en evanescent field. If the beam is incident at angle θ greater than critical angle of total internal reflection θc [defined in (2.2)], it does not propagate across the interface. Instead, it gives rise to the evanescent field with purely p imaginary z-component ikz of the wavevector and real x-component kx = (k 2 − (ikz )2 ) > k. This enhancement can be used to couple the incoming radiation to the plasmonic mode. The geometry in Fig. 2.12(a) is called Otto geometry [29] and consists of a prism separated from a bulk metallic sample by a thin (few radiation wavelengths) gap. The gap provides a tunnel barrier, which creates a p-polarized evanescent mode, exciting the plasmon at the metal-air interface. θ d θ z x (a) (b) (c) Figure 2.12: Methods of plasmon excitation. (a) Otto geometry. Kretschmann-Raether geometry. (c) Grating coupler. (b) The alternative technique is a Kretschmann-Raether geometry, depicted in Fig. 2.12(b) [30]. Here, the thin deposited metal film (< 50 nm) itself plays a role of the evanescent tunnel barrier, and the plasmon is excited on the opposite side of the metal. Surface plasmons can also be excited without the coupling prisms. In order to increase the wavevector of the propagating light, grated metallic surfaces can be used [31]. In this case, x-component of the wavevector of the exciting light kxinc is enhanced by the integer multiple of x-component of reciprocal unit vector Gx of the grating kx = kxinc + nGx = k sin θ + 2nπ/d, (2.12) where d is a grating period. Equation (2.12) is valid for any θ (including those θ > θc ). Adjusting the value of d, one can alter positions of the plasmonic resonances in the spectrum. 2.3. SURFACE PLASMONS 2.3.2 19 Applications of surface plasmons Plasmonic nanodevices are considered to be the most promising solutions for functional elements in photonic chips, near-field microscopy, manipulation of atoms and others. Plasmonic devices now cover the whole range of functionality of the traditional photonic devices, such as cavities, waveguides, apertures, providing, however, light manipulation at a deep sub-wavelength scale. Some of the plasmonic applications are summarized in Fig. 2.13. (b) (a) (e) (c) (f ) (d) (g) (h) Figure 2.13: Applications of surface plasmons. (a) Highly-directional plasmonic beamer [32]. Light, outgoing through the aperture in the center, couples to surface plasmons on the grated surface that results in highly-directional emission. (b) Plasmon-assisted extraordinary transmission through the array of sub-wavelength holes has been demonstrated [33]. (c) Ring resonators [34], made of grooves in a metal, can be utilized as band filters. (d) Plasmonic bandgap crystal [35]. Nano-patterned silver surface demonstrates photonic-crystallike gaps in the spectrum of plasmonic modes. (e) Nanofocusing of energy on the tip of the adiabatic plasmonic waveguide [36]. (f) A SNOM (Scanning Near-field Optical Microscopy) probe-based 1/4-wavelength nanoantenna [37]. Evanescent plasmonic mode from the sub-wavelength aperture couples to the 1/4-wavelength tip, resulting in the high field intensity. (g) Low-loss guiding of light in a low-n core 2D-waveguides(n1 < n2 ) [38]. (h) V-shaped plasmonic waveguiding grooves, splitters and Mach-Zehnder interferometers ([34] and citations therein) with a nearly zero insertion loss. 20 CHAPTER 2. PHOTONIC STRUCTURES Special attention is now also paid to possible applications of plasmons in photovoltaics. Figure 2.14 illustrates a typical photovoltaic device of the socalled ”third” generation. The third generation photovoltaics includes photoelectrochemical cells, polymeric and nanocrystal solar cells and is rather different from the previous semiconductor structures as it does not rely on a traditional p-n-junction to separate photogenerated charge carriers. Instead, the carriers are separated by the diffusion only. The device represents a multilayer stack of electrodes and active layer(s) deposited onto a transparent glass substrate. Polymeric solar cells seem to be promising in terms of low costs and Glass substrate Transparent electrode (ITO) Active layer Electrode (Al) Figure 2.14: Polymeric photovoltaic solar cell. ease of fabrication. However, the power-conversion efficiency even of the most advanced samples does not exceed 5% [39]. Plasmons, intensively absorbing light, can create high field intensities at the contact-active layer interfaces, facilitate electron-hole pair generation processes, increasing, therefore, the power conversion efficiency. Recently, the plasmoncaused increased absorption has been demonstrated for light-emitting diodes [40] with metallic nanoparticles, deposited onto the active layer of Si diodes. The application of nanoparticles to both non-organic and organic solar cells [41; 42] has displayed the increased short-circuit photocurrent. Rand et al. [43] have observed the extremely-high long-range absorption enhancement in tandem solar cells with embedded Ag nanoclusters. Nanoclusters in their paper reported to be acting as highly-effective recombination centers. In Chapter 4 and in Paper VII an another technique of the plasmon-induced absorption enhancement is proposed. Instead of using nanoparticles, we use 2.4. NANOPARTICLES 21 surface plasmons, excited on metallic gratings in polymeric solar cells. It is demonstrated that the plasmon-enhanced absorbtion leads to the increased photocurrent in the vicinity of the plasmonic peak. 2.4 Nanoparticles For centuries, alchemists and glassmakers have used tiny metallic particles for creating astonishing stained-glass windows and colorful goblets. One of the most ancient examples is the Lycurgus cup, a Roman goblet from the 4-th century A.D., see Fig. 2.15. The gold and silver particles embedded into the glass of the goblet absorb and scatter blue and green light. Figure 2.15: Lycurgus cup (4-th century A.D.). When viewed in reflected light, the goblet looks in a greenish hue, however if a light source is placed inside the goblet, the glass appears red. Therefore, when viewed in reflected light, the cup looks in a greenish hue, but if a white light source is placed inside the goblet, the glass appears red because it transmits only the longer wavelengths and absorbs the shorter ones. Nowadays metallic nanoparticles are intensively studied due to their potential in spectroscopy, fluorescence, biological and chemical sensing and others. 2.4.1 Properties of nanoparticles and Mie’s theory A term nanoparticle can be applied to any object containing 3 . N . 107 atoms. Physical properties of nanoparticles are size-dependent and two different kinds of size effects can be distinguished: intrinsic and extrinsic [44]. Intrinsic effects manifest themselves for small (< 10 nm) nanoparticles and are 22 CHAPTER 2. PHOTONIC STRUCTURES caused by a relatively small number of atoms in a nanoparticle that leads to the quantized energy spectrum of the particle. An arrangement of the atoms and their quantity have a strong impact on the dielectric function and optical properties of the cluster. However, for larger nanoparticles, containing millions of atoms, the intrinsic effects are negligible, and the dielectric function of such a cluster is assumed to be that for the bulk material. Optical response of these particles is fully governed by the extrinsic effects – size- and shape-dependent responses to the external excitations, irrespective to the internal structure of the particles. Let us now first consider a single metallic nanoparticle, being illuminated with electric field E of frequency ω = 2π/T (see Fig. 2.16) and the wavelength much larger than the nanoparticle size in a quasi-static regime (i.e. in the regime when the spatial phase of the field is assumed to be constant within the particle). The incident electric field causes displacement of the electronic - - + + + - + - + - + - - - - + + + + + kx - + - Ey - + + + + + + time t time (t+T/2) Figure 2.16: Excitation of dipole plasmonic resonance in a metallic nanoparticle. cloud within the particle against its ion core. The displacement gives rise to polarization charges on the opposite (for the dipole resonance) sides of the particle and, hence, to a restoring electrostatic force, which attempts to revert the system back to the equilibrium. After the half-period time the field changes its direction and the charges switch their places. Therefore, the nanoparticle acts as an oscillating system with single eigenfrequency [44] ωp ω1 = √ , 3 (2.13) where ωp is the Drude’s plasma frequency of a given metal. The general solution of a scattering problem for an arbitrary spherical particle of radius R was given by German physicist Gustav Mie in 1908, who calculated the absorption, scattering and extinction (absorption+scattering) 2.4. NANOPARTICLES 23 cross-sections. Start from Helmholtz equation in spherical coordinates ∇2 Ψ + k 2 Ψ = 0, (2.14) 1 ∂ 2 ∂ ∂2 1 ∂ ∂ 1 (r ) + (sin θ ) + . r2 ∂r ∂r r2 sin θ ∂θ ∂θ r2 sin2 θ ∂φ2 (2.15) where ∇2 = The solutions to (2.14) can be separated in spherical coordinates as Ψ = R(r)Θ(θ)Φ(φ) = ∞ X l X m m m [Am l cos mφPl cos θZn (kr) + Bl sin mφPl cos θZn (kr)], l=0 m=−l (2.16) with Plm spherical Legendre polynomials and Zn (kr) the Bessel functions for r < R and Hankel functions for r > R. Applying boundary conditions and m equating (2.16) one finds unknown coefficients Am l and Bl . Having calculated the coefficients one can easily obtain the extinction cross-section as σext = 2π X∞ (2l + 1)ℜ(Al + Bl ). k2 (2.17) l=1 For the case R << λ, when the quasi-static limit is assumed and only the dipole mode with l = 1 is considered, (2.17) reduces to [44] ω ε′′ (ω) σext = 12π ε0 3/2 R ′ , c [ε (ω) + 2ε0 ]2 + ε′′ (ω)2 (2.18) where ε0 and ε(ω) are dielectric functions of the surrounding medium and nanoparticle respectively. It can be easily shown that the condition for the resonance is that ε′ (ω) = −2ε0 . For larger particles, however, the interactions of higher orders l > 1 have stronger impact on the extinction spectra and cannot longer be neglected. The positions of the resonances are extremely sensitive to the surrounding medium, shape, size and symmetry of the particles and the temperature. Because of this, nanoparticles are considered to be promising candidates for sensing applications (see section 2.4.3 for details). It should also be mentioned that the Mie’s theory accounts only for noninteracting spheroids, whereas for the scatterers of arbitrary shape or aggregates of particles a number of more advanced tools has been developed. Among them are coupled-dipole approximation [45], multiple multipole technique [46], finite-difference time-domain method [47], generalized Lorentz-Mie’s theory for 24 CHAPTER 2. PHOTONIC STRUCTURES assemblies [48] and many others. 2.4.2 Nanoparticle arrays and effective-medium theories Single nanoparticles are of the prime interest for the fundamental study. However, practical applications require macroscopic systems containing thousands of particles. Moreover, many of these applications require knowledge of the effective-medium response of such systems, i.e. knowledge of the effective dielectric function from the optical properties of the constituents. Let us assume a set of equally-sized metallic nanoparticles with dielectric function ε(ω) embedded into a host dielectric medium with dielectric function εm at low filling factor f . Effective dielectric function εef f of the blend [49] εef f (ω) − εm ε(ω) − εm =f . εef f + 2εm ε(ω) + 2εm (2.19) was given by Maxwell Garnett in 1904 for non-interacting nanoparticles (low f < 0.3) in the quasi-static limit (d << λ). His theory has been extended by Bruggeman [50] to the case of high filling factor f & 0.5, where the effective dielectric function is given f εm − εef f (ω) ε(ω) − εef f (ω) + (1 − f ) = 0. ε(ω) + 2εef f (ω) εm + 2εef f (ω) (2.20) For even higher filling factors, clustering of nanoparticles and multipole effects are expected to play a significant role in both the Maxwell Garnett and Bruggeman theories. These factors are taken into account in the Ping Sheng theory [51]. Further, the Maxwell Garnett theory has been extended to the case of elliptic particles [52], to anisotropic composites [53], and others [54]. However, an effective-medium theory that accounts for non-spheroid particles at arbitrary concentrations or touching/overlapping particles remains to be developed. 2.4.3 Applications of nanoparticles A number of nanoparticle applications is constantly expanding. The table below summarizes some of them and several representative illustrations are also given in Fig. 2.17. 2.4. NANOPARTICLES Application Optical and photonic Electronic Mechanical Thermal Magnetic Energy Biomedical Environmental 25 Description Multi-layered structures with enhanced contrast [55]; Anti-reflection coatings [56]; Lasing structures [57]; Light-based detectors for cancer diagnosis [see Fig. 2.17(b)]; Surface-enhanced Raman spectroscopy (SERS) [see Fig. 2.17(a)]. Displays with enhanced brightness [58]; Tunableconductivity materials [59]. Improved wear resistance [60]; New anti-corrosion coatings [61]; New structural materials and composites [62]. Enhance heat transfer from solar collectors to storage tanks [63]. MnO particles improve detailing and contrast in MRI scans [64]. More durable batteries [65]; Hydrogen storage applications [66]; Electrocatalysts for high efficiency fuel cells [67]; Higher performance in solar cells [41]. Antibacterial coatings [68]; Smart sensors for proteins [69]. Clean up of soil contamination and pollution, e.g. oil [70]; Pollution sensors [71]; More efficient and effective water filters [72]. 26 CHAPTER 2. PHOTONIC STRUCTURES (a) (b) (c) (d) Figure 2.17: Examples of nanoparticle applications (a) Spacial distribution of nanoparticle induced SERS enhancement for two coated silver nanospheres (adopted from [73]). (b) Gold nanoparticles stick to cancer cells and make them shine (adopted from www.gatech.edu/news-room/release.php?id=561). (c) Scanning electron microscope image of the nanoparticle-structured band filter (adopted from [74]) (d) Magnetic nanoparticles produced by ”NanoPrism Technologies, Inc” for cell labeling, magnetic separation, biosensors, hyperthermia, magnetically targeted drug-delivery and magnetic-resonance imaging (adopted from www.nanoprism.net/ wsn/page3.html). Chapter 3 Computational techniques 3.1 Available techniques for studying light propagation in photonic structures By far, the most popular method for theoretical description of light propagation in photonic systems is the finite-difference time-domain method (FDTD) introduced by Yee [75]. The method is proven to be rather flexible and has been successfully applied to study of microcavities and photonic crystal structures. However, despite its speed and flexibility, the FDTD technique has a serious limitation related to the finiteness of the computational domain. As a result, an injected pulse experiences spurious reflections from the domain boundaries that leads to mixing between the incoming and reflected waves. In order to overcome this bottleneck a so-called perfectly matched layer condition has been introduced [76]. However, even using this technique, a sizable portion of the incoming flux can still be reflected back [77]. In many cases the separation of spurious reflected pulses is essential for the interpretation of the results, and this separation can only be achieved by increasing the size of the computational domain. This may enormously enlarge the computational burden, as the stability of the FDTD algorithm requires a sufficiently small time step. A severe disadvantage of this technique in application to microcavities with tiny surface imperfections is that the smooth geometry of the cavity has to be mapped into a discrete grid with very small lattice constant. This makes the application of this method to the problems, when small imperfections are studied, rather impractical in terms of both computational power and memory. For studying microcavities, a number of boundary-element methods has been applied. Their essence is that they reduce the Helmholtz equation in infinite two-dimensional space into contour integral equations defined at the cavity boundaries. These methods include the T -matrix technique [78; 79], the 27 28 CHAPTER 3. COMPUTATIONAL TECHNIQUES boundary integral methods [80; 81] and others [82]. In general, they are computationally effective and capable to deal with cavities of arbitrary geometry. However, they require the refraction index to be constant within the cavity. Numerous theoretical approaches have been developed to calculate the photonic band structure for 2D and 3D photonic crystals. The plane-wave method [83; 84; 85], for instance, allows one to calculate the band structures of PCs having known their Brillouin zones. Unfortunately, despite its simplicity for the implementation and stability, the method is not suitable for dispersive materials (for the dispersive media, a revised plane-wave technique has been developed [86]). Moreover, for complex structures (involving e.g. waveguides, cavities or surfaces) a large supercell has to be chosen that strongly increases the number of plane waves in the expansion and makes the method extremely computationally consuming. The problem of the spurious reflections from the computational domain boundaries does not arise in methods based on the transfer-matrix technique [87] where the transfer matrix relates incoming and outgoing fields from one side of the structure to those at another side. However, such the mixing leads to divergence of the method. The scattering-matrix (SM) techniques [88; 89; 90; 91], in contrast, are free of this drawback, as the scattering matrix relates incident and outgoing fields and their mixing is avoided. The other approaches, free of spurious reflections, are e.g. the multiple multipole method [46; 92] and the dyadic Green’s function method [93; 94; 95; 96] based on the analytical expression for the Green’s function of an empty space. This method will be described in more detail in Section 3.4. In this Chapter we present the developed scattering matrix technique for studying whispering-gallery mode disk microcavities with imperfect circumference and variable refraction index, the 2D recursive Green’s function technique for a scattering problem in photonic crystals and plasmonic structures, and the 3D dyadic Green’s function technique. 3.2 Scattering matrix method In this Section we present a method dedicated for calculation of resonant states in dielectric disk microcavities. The motivation of the development of this technique was that there are no theoretical tools so far, which are able to study microcavities both with tiny surface roughness and refraction index inhomogeneities. The method is capable to handle cavities with the boundary roughness as well as inhomogeneous refraction index. Because the majority of experiments are performed only with the lowest transverse mode occupied, the transverse (z-) dependence of the field is neglected and computations are performed in 2D. The two-dimensional Helmholtz equation for z-components 3.2. SCATTERING MATRIX METHOD 29 of electromagnetic field reads as 2 1 ∂ 1 ∂2 ∂ + + Ψ(r, ϕ) + (kn)2 Ψ(r, ϕ) = 0, ∂r2 r ∂r r2 ∂ϕ2 (3.1) where Ψ = Ez (Hz ) for TM (TE)-modes, n is a refraction index and k is a wavevector in vacuum. Remaining components of the electromagnetic field can be derived from Ez (Hz ) in a standard way. A B ∆i R i-th boundary ∆i ai ai+1 i bi+1 ri d b i-th strip (a) (i+1)-th strip (b) Figure 3.1: (a) Sketch of the geometry of a cavity with refraction index n surrounded by air. The domain is divided in three regions. In the inner (r < d) and in the outer regions (r > R) the refraction indexes are constant. In the intermediate region d < r < R refraction index n is a function of both r and ϕ. (b) The intermediate region is divided by N concentric rings of the width 2∆; ρi is a distance to the middle of the i-th ring. Within each ring the refraction coefficient is regarded as a function of the angle only and a constant in r. States ai , ai+1 propagate (or decay) towards the i-th boundary, whereas states bi , bi+1 propagate (or decay) away of this boundary. The i-th boundary is defined as the boundary between the i-th and (i + 1)-th rings. The system is divided into three regions, the outer region, (r > R), the inner region, (r < d), and the intermediate region, (d < r < R), see Fig. 3.1(a). We choose R and d in such a way that in the outer and the inner regions the refraction indexes are constant whereas in the intermediate region n is a function of both r and ϕ. In these regions the solutions to the Helmholtz equation can be written in analytical forms Ψin = +∞ X q=−∞ a0q Jq (nkr)eiqϕ , (3.2) 30 CHAPTER 3. COMPUTATIONAL TECHNIQUES for the inner region, where Jq is the Bessel function of the first kind, and Ψout = +∞ X q=−∞ Aq Hq(2) (kr) + Bq Hq(1) (kr) eiqϕ , (1) (3.3) (2) for the outer region, where Hq , Hq are the Hankel functions of the first and second kind of order q, describing incoming and outgoing waves respectively. Scattering matrix S is defined in a standard formulation [97; 98] B = SA, (3.4) where A, B are column vectors composed of expansion coefficients Aq , Bq in Eq. (3.3). Matrix element Sq′ q = (S)q′ q gives a probability amplitude of the scattering from incoming state q into outgoing state q ′ . The intermediate region is divided into narrow concentric rings where the refraction index depends only on angle ϕ [outlined in Fig. 3.1(b)]. The solutions to the Helmholtz equation in these rings can be expressed as superpositions of cylindrical waves. At each i-th boundary between the strips we define a local scattering matrix, which connects states propagating (or decaying) towards the boundary with those propagating (or decaying) outwards the boundary as i i b a i = S . (3.5) bi+1 ai+1 Local scattering matrices Si are derived using the requirement of the continuity of the tangential components for the Ez - and Hz -fields at the i-th boundary. The essence of the scattering matrix technique is the successive combination of the scattering matrices in the neighboring regions. Thus, combining the scattering matrices for the i-th and (i + 1)-th boundaries, Si and Si+1 , one obtains aggregate scattering matrix S̃i,i+1 = Si ⊗Si+1 that relates the outgoing and incoming states in rings i and i + 2 [97; 98] i i b a i,i+1 = S̃ , (3.6) bi+2 ai+2 −1 i S̃i,i+1 = Si11 + Si12 Si+1 I − Si22 Si+1 S21 , 11 11 11 −1 i+1 i,i+1 i+1 i i S̃12 = S12 I − S11 S22 S12 , −1 S̃i,i+1 = Si+1 I − Si22 Si+1 Si21 , 21 21 11 −1 i i+1 i+1 S̃i,i+1 = Si+1 I − Si22 Si+1 S22 S12 , 22 22 + S21 11 where matrices S11 , S12 , . . . define the respective matrix elements of block matrix S. Combining all the local matrices 0 ≤ i ≤ N in this manner one finally 3.2. SCATTERING MATRIX METHOD 31 obtains total matrix S̃0,N = S0 ⊗ S1 ⊗ . . . SN relating the scattering states in the outer region (i = N ) and the states in the inner region (i = 0), which after straightforward algebra is transformed to matrix S Eq. (3.4). The scattering matrix provides complete information about the system under study. In order to identify resonances, one introduces the Wigner time-delay matrix [4] averaged over incoming states as τD (k) = 1 d ln[det(S)], icM dk (3.7) where M is a number of the incoming states. It is interesting to note that Smith in his original paper, dealing with quantum mechanical scattering [4], chose a letter ”Q” to define the lifetime matrix of a quantum system because of a close analogy to the definition of the Q factor of a cavity in electromagnetic theory. The resonant states of the cavity are manifested as peaks in the delay time whose positions determine the resonant frequencies ωres , and the heights are related to the Q value of the cavity according to (2.5). 3.2.1 Application of the scattering matrix method to quantum-mechanical problems The developed scattering-matrix method was generalized to quantum-mechanical problems. This is possible thanks to the direct similarity between the Helmholtz and Schrödinger equations [98]: Photons Electrons ∇2 E = −ω 2 εE → ∇2 Ψ = −2m/~2[E − U ]Ψ E → Ψ Polarization → Spin S ∼ ℜ[−iE ∗ × (∇ × E)] → J ∼ ℜ[−iΨ∗ ∇Ψ] exp(−iωt) → exp(−iEt/~) The method solves a problem of quantum-mechanical (QM) scattering in quantum corral structures [99; 100], which can be considered as QM analogues of disk microcavivies. We calculate scattering wave function, from which one can extract spectra and the differential conductance dI/dV of the STM tunnel 32 CHAPTER 3. COMPUTATIONAL TECHNIQUES junction [which is proportional to the local density of states (LDOS)] X dI/dV ∼ LDOS(r, E) = |ψq (r)|2 δ(E − Eq ), (3.8) q where ψq (r) are scattering eigenstates of Hamiltonian Ĥ. The advance of the method is its ability to treat a realistic smooth potential within the corral structure. 3.3 Green’s function technique In order to study light propagation in 2D photonic-crystal structures, we have developed a novel recursive Green’s function technique. In contrast with the FDTD methods, the presented Green’s function technique is free from spurious reflections. The Green’s function of a photonic structure is calculated recursively by adding slice by slice on a basis of the Dyson’s equation that relaxes memory requirements and makes the method easy-parallelizable. In order to account for the infinite extension of the structure into both the air and space occupied by the photonic crystal we make use of so-called ”surface Green’s functions” that propagate the electromagnetic fields into (and from) infinity. The method is widely used in quantum-mechanical calculations [101] and is unconditionally stable. We start from Helmholtz equation, which for the 2D case (permittivity ε(r) is constant in the z-direction) decouples in two sets of equations for the TE modes ∂ 1 ∂ ω2 ∂ 1 ∂ Hz + Hz + 2 Hz = 0 (3.9) ∂x εr ∂x ∂y εr ∂y c and for the TM modes 1 εr ∂ 2 Ez ∂ 2 Ez + ∂x2 ∂y 2 + ω2 Ez = 0. c2 (3.10) Let us now rewrite equations (3.9), (3.10) in an operator form [102] Lf = ω 2 c f (3.11) where Hermitian differential operator L and function f read TE modes: f ≡ Hz ; LT E = − TM modes: f = √ εr Ez ; LT M ∂ 1 ∂ ∂ 1 ∂ − , ∂x εr ∂x ∂y εr ∂y 2 1 ∂ ∂2 1 = −√ + 2 √ . εr ∂x2 ∂y εr (3.12) (3.13) 3.3. GREEN’S FUNCTION TECHNIQUE 33 For the numerical solution, Eqs. (3.11)-(3.13) have to be discretized, x, y → m∆, n∆, where ∆ is a grid step. Using the following discretization of the differential operators in Eqs. (3.12),(3.13), ∂ ∂f (x) ξ(x) → ξm+ 12 (fm+1 − fm ) − ξm− 12 (fm − fm−1 ) , ∂x ∂x ∂2 ∆2 2 ξ(x)f (x) → ξm+1 fm+1 − 2ξm fm + ξm−1 fm−1 ∂x ∆2 (3.14) one arrives to finite difference equation vm,n fm,n − um,m+1;n,n fm+1,n − um,m−1;n,n fm−1,n − 2 ω∆ fm,n , −um,m;n,n+1 fm,n+1 − um,m;n,n−1 fm,n−1 = c (3.15) where coefficients v, u are defined for the cases of TE and TM modes as follows TE modes: fm,n = Hz m,n ; ξm,n = 1 εr m,n , (3.16) vm,n = ξm+ 12 ,n + ξm− 12 ,n + ξm,n+ 12 + ξm,n− 12 , um,m+1;n,n = ξm+ 12 ,n , um,m−1;n,n = ξm− 21 ,n , um,m;n,n+1 = ξm,n+ 12 , um,m;n,n−1 = ξm,n− 12 ; TM modes: fm,n = √ εr m,n Ez m,n ; ξm,n = √ 1 εr m,n (3.17) 2 vm,n = 4ξm,n , um,m+1;nn = ξm,n ξm+1,n , um,m−1;nn = ξm−1,n ξm,n , um,m;n,n+1 = ξm,n+1 ξm,n , um,m;n,n−1 = ξm,n ξm,n−1 . A convenient and common way to describe finite-difference equations on a discrete lattice is to introduce the corresponding tight-binding operator. For this purpose one first introduces creation and annihilation operators, a+ m,n , am,n . Let the state |0i ≡ |0, . . . , 0m,n , . . . , 0i describe an empty lattice, and state |0, . . . 0, 1m,n , 0, . . . , 0i describes an excitation at site m, n. Operators a+ m,n , am,n act on these states according to rules [101] a+ m,n |0i = |0, . . . 0, 1m,n , 0, . . . , 0i, a+ m,n |0, . . . 0, 1m,n , 0, . . . , 0i = 0, (3.18) 34 CHAPTER 3. COMPUTATIONAL TECHNIQUES a+m+1,n am,n n n+1 m m+1 Figure 3.2: Forward hopping term in Eq. (3.22). and am,n |0i = 0, (3.19) am,n |0, . . . 0, 1m,n , 0, . . . , 0i = |0i, and they obey the following commutational relations + + [am,n , a+ m,n ] = am,n am,n − am,n am,n = δm,n , [am,n , am,n ] = + [a+ m,n , am,n ] (3.20) = 0. Consider an operator equation b |f i = L ω∆ c 2 |f i, where Hermitian operator X Lb = (vm,n a+ m,n am,n − (3.21) (3.22) m,n + − um,m+1;n,na+ m,n am+1,n − um+1,m;n,n am+1,n am,n − + − um,m;n,n+1a+ m,n am,n+1 − um,m;n+1,n am,n+1 am,n ) acts on state |f i = X m,n fm,n a+ m,n |0i. (3.23) The second and third terms in Eq. (3.22) correspond forward and backward hopping between two neighboring sites of the discretized domain in the xdirection, and terms 4 and 5 denote similar hopping in the y-direction, see Fig. 3.2. Substituting the above expressions for Lb and |f i into Eq. (3.21) 3.3. GREEN’S FUNCTION TECHNIQUE 1 … M n N N-1 35 1 … M n N N-1 I I T R T R 2 1 2 1 -M+1 … -1 0 1 2 3 … M M+1 m (a) 0 1 2 3 … M M+1 m (b) Figure 3.3: Schematic illustration of the system under study defined in a supercell of width N. The internal region of the structure occupies M slices. Two representative cases are shown: (a) external regions are semi-periodic photonic crystals with period M, (b) external regions represent a semi-infinite periodic photonic crystal with period M to the right and air to the left. Arrows indicate the directions on the incoming (I ), reflected (R) and transmitted (T ) waves. and using the commutation relations and the rules Eqs. (3.18)–(3.20), it is straightforward to demonstrate that operator equation (3.21) is equivalent to finite difference equation (3.15). Let us now specify structures under study. We consider light propagation through a photonic structure defined in a supercell of width N , where one assumes the cyclic boundary condition (i.e. row n = N + 1 coincides with row n = 1). The photonic structure occupies a finite internal region consisting of M slices (1 ≤ m ≤ M ). The external regions are semi-infinite supercells extending into regions m ≤ 0 and m ≥ M + 1. The supercells can represent air (or a material with the constant refraction index) or a periodic photonic crystal. Figure 3.3 shows two representative examples where (a) the semi-infinite waveguides represent a periodic photonic crystal with period M, and (b) a photonic structure is defined at the boundary between air and the semi-infinite photonic crystal. b in a standard way We define Green’s function of the operator L 2 b G(ω) = 1b , (ω∆/c) − L (3.24) where 1b is the unitary operator. The knowledge of the Green’s function allows us to calculate the transmission and reflection coefficients. Indeed, let us write down the solution of Eq. (3.21) as a sum of two terms, the incoming state |ψ i i and the system response |ψi representing whether transmitted |ψ t i or reflected 36 CHAPTER 3. COMPUTATIONAL TECHNIQUES Figure 3.4: Schematic illustration of the application of the Dyson’s equation for calculation of the Green’s function for a composed structure consisting of m+1 slices. |ψ r i states, |f i = |ψ i i + |ψi. Substituting |f i into Eq. (3.21) and using formal definition of the Green’s function Eq. (3.24), the solution of Eq. (3.21) can be written in the form b − (ω∆/c)2 |ψ i i. |ψi = G L (3.25) Calculation of the whole structure starts from the internal region (i.e for slices 1 ≤ m ≤ M in Fig. 3.3). The recursive technique based on the Dyson’s equation is utilized, see Fig. 3.4. Our goal is to calculate the Green’s function of the composed structure, Gm+1 , consisting of m + 1 slices. The operator corresponding to this structure can be written down in the form 0 b m+1 = L b 0m + b L l m+1 + Vb , (3.26) 0 b 0 and b where operators L l m+1 describe respectively the structure composed of m m slices and the stand-alone (m+1)-th slice, and Vb = Vb m,m+1 + Vb m+1,m is the perturbation operator describing the hopping between the m-th and (m + 1)-th slices, Vb = Vb m+1,m + Vb m,m+1 . (3.27) b m+1 , Gm+1 = G0 + G0 VG b 0, Gm+1 = G0 + Gm+1 VG (3.28) The Green’s function of the composed structure, Gm+1 , can be calculated on the basis of the Dyson’s equation: 3.3. GREEN’S FUNCTION TECHNIQUE 37 b0 where G0 is the unperturbed Green’s function corresponding to operators L m 0 or b l m+1 . Thus, starting from the Green’s function for the first slice g10 and adding recursively slice by slice we are in the position to calculate the Green’s function of the internal structure consisting of M slices. Explicit expressions following from Eqs. (3.28) and used for the recursive calculations read as 0 0 Gm+1,m+1 = (I − gm+1 Um+1,m (G0m )m,m Um,m+1 )−1 gm+1 , m+1 (3.29) m+1,m+1 Gm+1,1 Um+1,m (G0m )m,1 , m+1 = Gm+1 m+1,1 0 1,1 G1,1 + (G0m )1,m Um,m+1 Gm+1 , m+1 = (Gm ) 1,m+1 m+1,m+1 0 1,m Gm+1 = (Gm ) Um,m+1 Gm+1 , where the upper indexes define the matrix elements of the Green’s function. The next step is attaching the left and right semi-infinite leads to the internal region. Starting with the left waveguide, one writes b int+lef t = L b int + L b lef t + Vb , L (3.30) b int+lef t , Gint+lef t = G0 + G0 VG (3.31) b int+lef t , L b int and L b lef t describe respectively the system repwhere operators L resenting the internal structure + the left waveguide, the internal structure, and b describes the hopping between the the left waveguide. Perturbation operator V left waveguide and the internal structure. Applying then the Dyson’s equation in a similar way as it has been described above, we are in position to find Green’s function Gint+lef t of the system representing the internal structure + the left waveguide. G0 in Eq. (3.31) is an ”unperturbed” Green’s function corresponding to the internal structure and the semi-infinite waveguide (”surface Green’s function” Γ). The physical meaning of the surface Green’s function Γ is that it propagates the electromagnetic fields from the boundary slice of the semi-infinite waveguide (supercell) into infinity. A method for calculation of the surface Green’s functions both for the case of a semi-infinite homogeneous dielectrics, as well as for the case of a semi-infinite photonic crystal in a waveguide geometry is given in Paper IV. Having calculated Green’s function Gint+lef t on the basis of Eq. (3.31), one proceeds in a similar way by adding the right waveguide and calculating with the help of the Dyson’s equation total Green’s function G of the whole system. Having calculated matrix elements for the complete system, GM+1,0 , G0,M+1 , G , GM+1,M+1 , one can easily relate them to the transmission T and reflec0,0 38 CHAPTER 3. COMPUTATIONAL TECHNIQUES tion R coefficients of the system (see Paper IV for details) ΦM+1 T = −GM+1,0 (U0,1 Φ−M+1 Kl − Γl −1 Φ0 ), 0,0 Φ0 R = −G (U0,1 Φ−M+1 Kl − Γl −1 Φ0 ) − Φ0 , (3.32) (3.33) where Γl ≡ G0,0 wg is the left surface Green’s function, Kl and Φm are given by the right-propagating Bloch eigenvectors kα+ and the corresponding eigenstates φα m,n in the waveguides and U0,1 is a hopping matrix between the 0-th and 1-st slices. 3.4 Dyadic Green’s function technique Introduced in the previous section the Green’s function technique is adapted for the two-dimensional case. Extension of this method to the 3D-case is impractical as it would require too extensive computational resources. There exists a number of techniques for 3D scattering problems. These include the finite-difference time-domain method, [75; 47], the multiple multipole method [46; 92] and the Coupled-Dipole Approximation (CDA) method [45], which we utilize in our work. The CDA method was further developed by O.J.F. Martin et al. [93; 94; 95; 96] and received a name Dyadic Green’s Function Technique. The advantage of this approach is that only the scatterer is needed to be discretized. Moreover, this technique can be easily extended to complex (stratified, anisotropic, etc.) backgrounds, by making use of the corresponding free-space Green’s function. The central and starting point of this technique is a volume integral equation formulation of the Maxwell’s equations (Lippmann-Schwinger equation). The aim of the method is to calculate electric field E(r), scattered against an object with volume V illuminated with incident field E0 (r). The object is characterized by its complex dielectric function ε(r) and surrounded, in the simplest case, by an infinite homogeneous background with permittivity ε0 . This scatterer does not need necessary to be either homogeneous or isotropic. Assuming exp(−iωt) time dependence of the electromagnetic waves and the isotropic scatterer, Lippmann-Schwinger equation for optical fields reads Z E(r) = E0 (r) + dr′ G0 (r, r′ ) · V (r)E(r′ ), (3.34) V where V (r) is a hopping potential defined as V (r) = k0 2 [ε(r) − ε0 ], (3.35) with k0 = 2π/λ0 a vacuum wavevector. G0 (r, r′ ) is the Green’s dyadic tensor describing the background. It is obtained as a solution to the vectorial wave 3.4. DYADIC GREEN’S FUNCTION TECHNIQUE εi 39 εB Figure 3.5: Outline of the discretization scheme for the dyadic Green’s function technique. equation with a point source term [103] and analytically can be expressed as G0 (r, r′ ) = (I + ∇∇ exp(ikB R) ) , 4πR kB 2 (3.36) √ where I is the unit tensor, kB = k0 ε0 and R = |r − r′ |. The integration in (3.34) is performed over volume V of the scatterer. Thus, one can formally divide the calculation of the scattered field into two stages: (1) The field is first calculated inside the scatterer and (2) having calculated the field distribution inside the scatterer, one can easily obtain the field in any point of space outside the scatterer making use of Eq. (3.34). There exists a number of techniques for solving Eq. (3.34), including iterative methods [93] and the finite-element method (FEM) [104]. The first ones, possessing the great unconditional stability and reduced storage needs, have, however, low performance in comparison to direct linear solvers. The latter, FEM technique, allows more fine dicretization scheme by the triagulation of the volume. However, 3D FEM seems rather impractical even with modern computer resources. We have implemented a ”compromised” solver to Eq. (3.34), utilizing a simple LU decomposition along with a ”smart” discretazation technique. We split the scatterer into N cubic meshes with volumes Vi . The size of the mesh is reduced towards the boundary in accordance to the procedure described in [105] and presented in Fig. 3.5. The discretized version of Eq. (3.34) is X Ei = E0 i + G0 i,j · k0 2 (εi − ε0 )Vj Ej . (3.37) j 40 CHAPTER 3. COMPUTATIONAL TECHNIQUES Equations (3.34) and (3.37) have, however, a singularity at r = r′ (i = j) which can be avoided by removing the singularity point from the integration volume and compensating this value by source dyad L = 1/3I. The contribution from the i-th volume can be integrated analytically by assuming this volume to be spherical with radius Ri = [3/(4π)Vi ]1/3 . The contribution reads as [95] 2 (3.38) Mi = 2 [(1 − ik0 Ri ) exp(ik0 Ri ) − 1]I. 3k 0 Substituting both self-term (3.38) and the definition of the source dyad into Eq. (3.37) we obtain Ei = E0 i + X j6=i G0 i,j ·k0 2 (εi −ε0 )Vj Ej +[Mi k0 2 (εi −ε0 )−L· εi − ε0 ]Ei . (3.39) ε0 This equation describes the electric field inside the scattered and can be solved with any appropriate linear solver. The resulting field distribution is then used as an input to Eq. (3.37) for the calculation of the field anywhere outside the scatterer. In order to investigate a far-field response, we calculate scattering crosssection Z Z 2π 1 π σ= dθ sin θ dϕ|E∞ (θ, ϕ)|2 , (3.40) 4π 0 0 where θ and ϕ are angular coordinates of a spherical coordinate system and E∞ (θ, ϕ) is a far-field intensity on infinity Z E∞ (θ, ϕ) = dr′ G∞ (θ, ϕ, r′ ) · V (r)E(r′ ). (3.41) V Here G∞ (θ, ϕ, r′ ) is the Green’s propagator from a point within the scatterer to a point located at (θ, ϕ) on infinity G∞ (θ, ϕ, r′ ) = 1 (I − nn) exp(−ik0 n · r) 4π where nn is a dyad product of vectors sin θ cos ϕ n = sin θ sin ϕ . cos θ (3.42) (3.43) We currently apply the dyadic Green’s function technique to study of light scattering by 3D whispering-gallery microcavities and arrangements of nanoparticles. Some representative results are given in Fig. 3.6 that shows the scattering cross-section (a) and resonance electric field distribution (b) of a linear 3.4. DYADIC GREEN’S FUNCTION TECHNIQUE 41 1000 σ, nm 2 100 10 y 1 (a) (b) z 300 350 400 450 500 550 600 650 700 λ, nm x Figure 3.6: (a) Scattering cross-section of a 7-nanoparticle chain. The diameter of nanoparticles is 50 nm, the inter-particle distance is 40 nm. (b) Electric field distribution at the nanoparticle chain at 426 nm. chain of seven 50-nm silver nanoparticles (inter-particle distance is 40 nm) in vacuum. The chain is illuminated from the left with the Ez -component of the field. High field intensity between the particles characterizes the strong coupling at the longitudinal resonance 426 nm. The position of the single-particle dipole plasmonic resonance (∼ 360 nm) agrees very well with its analytical estimation (2.18). 42 CHAPTER 3. COMPUTATIONAL TECHNIQUES Chapter 4 Results 4.1 Effect of inhomogeneities on quality factors of disk microcavities (Papers I, II) The Q factor of a microdisk cavity is the most important parameter of the structure. It is governed by a radiative leakage through the curved interface due to diffraction. An estimation of the Q factor in an ideal disk cavity of a typical diameter d ∼ 10µm for a typical WG resonance gives Q ∼ 1013 . At the same time, reported experimentally measured values are typically in the range of 103 ∼ 104 [6] or even lower. Such the discrepancy may be attributed to different factors such as side-wall imperfections, finite or inhomogeneous height of the disk, non-uniform refraction index within the structure, effects of coupling to the substrate or pedestal and others. Several experimental observations point out the side-wall imperfections as the main factor affecting the Q value of the cavity [9; 6; 10]. Accounting of these imperfections can be considered of extreme importance for the design and tailoring of lasing microdisks. Using the developed scattering-matrix method we have studied the effect of cavity roughness and inhomogeneity of the refraction index on quality factors of disk microcavities. A cavity with diameter d = 10µm and refraction index hni = 1.8 is considered. Various studies indicate that a typical size of the sidewall imperfections can vary in the range of 5-300 nm (representing a variation of the order of ∼0.05-1% of the cavity radius), but their exact experimental shape is unfortunately not available. We thus model the interface inhomogeneities as a superposition of random Gaussian deviations from the ideal circle of radius R with maximal amplitude ∆r/2 and characteristic distance between the deviation maxima ∆l ∼ 2πR/50, see inset to Fig. 4.1(a). The imperfect region is discretized into 100 concentric strips. Figure 4.1(a) illustrates calculated Q values of the disk resonant cavity for 43 44 CHAPTER 4. RESULTS different surface roughnesses ∆r in some representative wavelength interval for the TM polarization. Note that we have studied a number of different resonances and all of them showed the same trends described below. 3 -1 Ideal 20nm 50nm 100nm 200nm TM 83,1 10 -3 10 T Q -7 ∆l r= R= 5. 10 56,7~ 56,7 T ch ~ T cur 0 µm 83,1 Tcur -9 ∆r 10 ∆θch -5 r TM 56,7 r =1 .7 µ m r= 2.5 µm 10 10 2 -11 10 -13 10 sin (θ) 10 0.9 q=83 0.8 0.7 83,1 Tch ∆θch q=56 0.6 0 1 2 3 4 5 6 ∆θch -15 627.5 628.0 628.5 629.0 λ (nm) 629.5 630.0 (a) 10 0.6 0.7 0.8 sin θ 0.9 (b) Figure 4.1: (a) Dependencies Q = Q(λ) for two representative modes TM83,1 (high-Q mode) and TM56,7 (low-Q mode) for different surface roughness ∆r. Inset sketches inhomogeneous surface geometry. (b) Dependence T = T (θ) for several radii of curvature ρ according to Eq. (2.3). Inset shows a Poincaré SoS for the states q = 83 and q = 56 for the cavity with ∆r = 0 (straight lines of θ = const) and ∆r = 20nm. The solid curve in Fig. 4.1(a) corresponds to the ideal disk cavity without imperfections. The dependence of the averaged Q values on the surface roughness ∆r for several representative resonances is also given. A common feature of all high-Q resonances is a drastic decrease of their maximal Q value that occurs even for very small values of ∆r . λ/20. For example, the Q value of resonant state TM83,1 drops from Q ≈ 1013 for an ideal disk to Q ≈ 103 for surface roughness of only ∆r = 20 nm. However, the picture for low-Q states is rather different. Low-Q resonances show a relatively slow decrease in their Q values over the range of variation of ∆r. For example, for the same surface roughness ∆r = 20 the Q value of resonant state TM56,7 decreases only by a factor of 1.5, dropping from Q ≈ 300 to Q ≈ 200. In order to understand this behavior we combine a Poincaré surface of section (SoS) method with analysis of ray reflection at a curved dielectric interface [see Eq. (2.3)]. Figure 4.1(b) illustrates that transmission T , calculated by Eq. (2.3), decreases exponentially as the difference between angle of incidence θ and critical angle of incidence θc grows. Poincaré SoS represents dependence of angle of incidence θ on the polar angle ϕ around the cavity, its detailed definition and related discussion are given in Paper II. The inset to Fig. 4.1(b) depicts the SoS for two states with q = 56 and 83 shown in Fig. 4.1(a), where the initial angle of incidence θ0 of launched rays is related to the angular num- 4.2. QUANTUM CORRALS (PAPER III) 45 ber q. The SoS demonstrates that initially regular dynamics of an ideal cavity (straight line) transforms into the chaotic one even for a cavity with maximum 83,1 roughness ∆r . 20nm. ∆Tch in Fig. 4.1(b) indicates the estimated increase in the transmission coefficient due to the broadening of the phase space, ∆θch , as extracted from the Poincaré SoS for the state with q = 83. This corresponds to the decrease of ∆Q ∼ ∆T −1 ≈ 10−2 . This value is much smaller that the actual calculated decrease of the Q factor for high-Q resonance TM83,1 . In order to explain the rapid degradation of high-Q resonances, we focus on another aspect of the wave dynamics. The imperfections at the surface boundary introduce a local radius of surface curvature ρ that is smaller than disk radius R [see inset in Fig. 4.1(a)]. One may thus expect that, with the presence of a local surface curvature, the total transmission coefficient will be determined by the averaged value of ρ rather than by disk radius R. Figure 4.1(b) outlines that the reduction of the local radius of curvature from 5µm (ideal disk) to 1.7µm (∆r = 20nm) causes an increase of the transmission coefficient by ∆Tcur ≈ 108 . This number, combined with the estimate based on the change of ∆Tch ∼ 102 , is fully consistent with the Q factor decrease shown in Fig. 4.1(a). We thus conclude that the main mechanism responsible for the rapid degradation of high-Q resonances in non-ideal cavities is the enhanced radiative decay through the curved surface because the effective local radius (given by the surface roughness) is smaller that the disk radius. In contrast, for the case of low-Q resonances change in the transmission coefficient due to the enhanced radiative decay ∆Tcur is of the same magnitude as change ∆Tch due to the broadening of the phase space caused by the transition to the chaotic dynamics (for the resonance TM56,7 in Fig. 4.1). Therefore, both these factors play comparable roles in degradation of the low-Q WG resonances. We have also studied the effect of the non-uniform refraction index within the cavity and found that the decay of the Q factor in that case is of minor importance in comparison to the case of rough surface. 4.2 Quantum corrals (Paper III) We have adapted the scattering matrix technique developed for disk dielectric microcavities to study of quantum-mechanical scattering in quantum corrals, which can be considered as nanoscale analogues to dielectric cavities. Quantum corrals consist of adatoms of noble metals or Fe, deposited by AFM on (111) surface of Cu in a ring, triangle or other arrangements (see Fig. 4.2). Cu surface states interact strongly with the adatoms, and the spatial variation of the STM differential conductance reveals beautiful images of the surface standing wave patterns in the quantum corrals. In addition, the experiments show a series of remarkable resonant peaks in the energy spectrum of the differential conductance dI/dV in the center of the structures. So far there have been 46 CHAPTER 4. RESULTS Figure 4.2: Quantum corral structures. STM images, reported by IBM [106] several reports on experimental and theoretical studies of the corral structures. In order to describe the experimental observation [99; 100], Heller et al. [107] have developed the multiple-scattering theory for surface electron waves in quantum corrals. In that theory each adatom was treated as a point-like ”black dot” δ-function potential supporting isotropic scattering of a standing wave. The quantitative agreement with the experiment was achieved by considering an additional inelastic channel of scattering to the bulk of the substrate. It was also concluded that absorption is the dominant mechanism for the broadening of the energy levels seen in the experiment. Their theory describes well the spatial distribution of the wave function in the corrals, but overestimates the broadening of the resonant peaks, especially at higher energies. An alternative purely elastic scattering theory for the same quantum corral structures was drawn by Harbury and Porod [108]. They described the adatoms by finite-height solid potential barriers. Their findings suggest that the features of the spectrum and the wavefunction distribution can be extremely sensitive to the detailed shape of the scattering potential. The advance of our scattering matrix method is that it can treat a realistic potential of the adatoms, their displacements or variety of inhomogeneities. Accounting for the realistic smooth shape of the scattering potential is known to be crucial for quantitative description of many phenomena in quantum nanostructures. The potential of each adatom is taken as 4.3. SURFACE-STATE LASERS (PAPER IV) 47 a Gaussian with half-width σ and height V0 centered at (x0 , yo ), V (x, y) = V0 exp[−(x − x0 )2 /2σ 2 ] exp[−(y − y0 )2 /2σ 2 ]. Using the scattering matrix technique we calculate the bias voltage dependence and the spatial distribution of the local density of states (LDOS) for 60-Fe-adatom, 88.7-Å-radius circular quantum corrals reported by Heller et al. [107]. Fig. 4.3 shows experimental and theoretical results for the voltage dependence (a) and the spatial distribution (b) of the LDOS. The Fe-adatoms are located on the meshes of a 2.55Å triangular grid corresponding to the hexagonal Cu(111) lattice. The effective mass used in all the simulations is taken m∗ = 0.361m0 and the electron band-edge energy E0 = 0.43 eV below the Fermi energy of the electrons. dI/dV (arb. units) 2.0 Experiment Scatt. Matrix Method Experiment Scatt. Matrix Method V=0.47 1.5 V=0.45 1.0 V=0.43 0.5 V=0.41 -0.4 -0.3 -0.2 -0.1 0.0 Bias potential (V) 0.1 0.2 -80 -60 -40 -20 0 20 40 Distance (Å) (a) 60 80 (b) Figure 4.3: (a) The experimental spectrum of differential conductance dI/dV in the center of the 88.7-Å-radius 60-Fe-adatom circular quantum corral structure on Cu(111) substrate (solid curve). Scattering matrix technique applied for a smooth adatom potential with height V0 = 2.5eV and broadening σ = 1.52Å (dashed line). (b) The experimental curves (solid lines) and results of the scattering-matrix simulations (dashed line) for the local density of states subject to the tip position inside the circular corral for low bias voltages. Our calculations prove the importance of the realistic potential and reproduce quantitatively all the experimental observations (see Paper III for the details). This is in contrast to the previous theories (treating the adatoms as point scatterers) that require additional inelastic channels of scattering into the bulk in order to achieve an agreement with the experiment. Our findings thus indicate that surface states are not coupled to the bulk electrons. 4.3 Surface-state lasers (Paper IV) Surface states in photonic crystals can be exploited in a variety of lasing, sensing and waveguiding applications. First we focus on the novel type of a lasing cavity 48 CHAPTER 4. RESULTS that is situated on the surface of a photonic crystal and uses surface modes. We consider a semi-infinite square-lattice photonic crystal composed of cylinders with ε = 8.9 and diameter D = 0.4a (a is a lattice constant) in an air background. In order to create a surface geometry sustaining surface modes, we reduce the diameter of the rods in the outmost row to d = 0.2a. This structure has the full fundamental bandgap for the TM-polarization in the range of 0.33 < ωa/2πc < 0.44 and supports one surface mode, which along with the band structure is depicted in Fig. 2.9(a). In order to create a surface-state lasing cavity the modified surface region has to be confined along the modified boundary that turns the surface mode into a resonant state with a finite lifetime. For this sake a semi-infinite photonic crystal structure containing only a finite number N of the surface rods with reduced diameter d = 0.2a was considered. These rods define a resonant cavity situated at the surface of the photonic crystal as illustrated in the inset to Fig. 4.4(a) for the case of N = 6. The strong confinement from three sides of the cavity is provided by the photonic crystal operating in the bandgap. (a) (b) Figure 4.4: (a) Lower panel: Spectrum of a surface state photonic bandgap cavity. Inset illustrates a resonant cavity defined by the N = 6 surface rods of smaller diameter d = 0.2a placed on the photonic crystal surface. Upper panel: The dispersion relation for the surface state for the semi-infinite photonic crystal. The dashed lines indicate the expected analytical resonant wave vectors for the modes α = 5, 6, 7 and corresponding expected resonant frequencies. (b) Lower panel: Calculated intensity of the Ez component for the 6-th mode shown in Fig. 4.4(a). Upper panel: Expected field intensity at different rods is given by the overlap of the 6-th eigenstate of the cavity with the actual positions of the rods. 4.4. SURFACE-STATE WAVEGUIDES (PAPER V) 49 In order to calculate the quality factor of the structure at hand we apply the developed Green’s function technique. We illuminate the surface cavity with an incidence wave, which excites a resonant mode within the cavity, compute the R intensity of the field distribution and express the Q factor as Q = ωΩ/(4 Sin dy) [102], where Ω characterizes the energy stored in the system and the integral over Sin is the incoming energy flux. It should be stressed that the resonance Q value depends on the coupling of the surface state modes with the outgoing radiation, and thus is independent on the incidence angle of the incoming wave. Figure 4.4(a) shows the calculated Q factor of the resonant cavity versus the frequency of the illuminating light. In the given frequency interval there are three lasing peaks with quality factors ∼ 105 . Note that these values might underestimate the actual theoretical Q factors obtained within the present 2D calculations, because even finer frequency steps in the vicinity of the resonances are required for better resolution of the Q factors. The estimation of the position of the peaks for the cavity at hand can be performed by making use a formula for a Fabry-Perot resonator, whose resonant wavelengths are given by λα = 2π/kα , with wavevector kα = πα/w. From the dispersion relation [upper panel of 4.4(a)] it follows that only modes α = 5, 6, 7 are situated inside the frequency interval where the surface mode exists. An estimation of the expected positions for the resonant peaks for these modes is also given in the lower panel of Fig. 4.4(a) where the discrepancy between the expected analytical and calculated resonance frequencies does not exceed 0.5%. Figure 4.4(b) depicts the intensity of the Ez component of the electromagnetic field for the resonance mode α = 6. As expected, for the TM-modes, the field is localized in the cavity inside the rods, and the intensity decays very rapidly both to the open space and to the crystal. The field intensity at different rods in the cavity is expected to be determined by the overlap of the α-th eigenstate of the Fabry-Perot resonator with the actual positions of the rods in the cavity. This overlap for the 6-th mode is also shown in Fig. 4.4(b), which agrees perfectly with the actual calculated field intensity. 4.4 Surface-state waveguides (Paper V) ”Conventional” waveguides in photonic crystals represent line defects in periodic structures supporting guided Bloch modes whose frequency region lies in the bandgap. These modes are strongly confined within the waveguide region and can propagate lossless to substantial distances. Here we propose a novel type of waveguiding structures, namely waveguides that operate on surface modes of semi-infinite photonic crystals (surface modes, propagating along the boundary are now waveguiding modes!) and are located on the surface of a PC. Their ”non-traditional” location may open up new possibilities for design and 50 CHAPTER 4. RESULTS operation of photonic structures for feeding and redistributing light in PCs. Making use the Green’s function technique we have studied localization properties of surface modes, their dispersion relations and an effect of inhomogeneities. The latter has been demonstrated as the one having the strongest impact on the performance of lasing microcavities. One can expect then, that imperfections in the shape of the rods, their displacement, or variations of the refraction index throughout the crystal can significantly affect the waveguiding efficiency of the surface modes. 1.0 T 0.6 0.4 0.6 Transmission Velocity v/c 0.8 0.8 0.4 0.2 0.2 0.34 0.36 0.38 0.40 0.42 ωa/2πc Figure 4.5: Transmission coefficient (solid line, left axis) for the surface modes propagating in a non-ideal surface-mode waveguide. Velocity of the surface mode (dashed line, right axis) from Fig. 2.9. The inset shows the structure under study, where the shaded regions denote ideal semi-infinite waveguides, and the central region of the width of 5a represents an imperfect photonic crystal where scattering of the Bloch surface state takes place. We consider the semi-infinite photonic crystal from the previous Section (ε = 8.9, D = 0.4a) with the infinitely long boundary consisting of rods of reduced diameter d = 0.2a (see the band structure in Fig. 2.9). In order to study the effect of imperfections, the structure is split into three regions as shown in the inset to Fig. 4.5. Two of them are left and right semi-infinite periodic structures (perfect waveguides for surface modes), and the block of the PC in between is an imperfect region. Utilization of the Green’s function technique allows one to use surface Bloch modes as scattering states that propagate in perfect waveguides from the infinity into the imperfect region where they undergo scattering. Obviously, in the case when the scattering region is absent (perfect waveguides are attached to each other), the Bloch states propagate freely without any losses. The imperfection is modelled by the discretization of the scattering region differently from that in the left and right perfect waveguides. The transmission coefficient in Fig. 4.5 for the surface mode drops quite 4.4. SURFACE-STATE WAVEGUIDES (PAPER V) 51 0.8 T 0.6 0.4 0.2 0.34 0.35 0.36 0.37 0.38 ω a/2πc Figure 4.6: A lead-in coupler composed of a surface-state waveguide to the left and a conventional tapered PC waveguide to the right. The size of the surface rods gradually decreases to zero in the central region where the surface-state waveguide transforms in a conventional PC waveguide. Intensity distribution is shown for Ez -component of the electromagnetic field at ωa/2πc ≈ 0.365. Arrows sketch the flow of the Poynting vector. The inset shows the transmission coefficient subject to the energy of incoming light. Parameters of the photonic crystal correspond to those of Section 4.3. rapidly in the energy regions corresponding to the low velocity of the surface state (dashed line in the Figure). This is because the backscattering probability is greatly enhanced for the low-velocity states. Even for 5 imperfect unit cells the transmission coefficient in energy region ωa/2πc & 0.40 is less than 1 which makes this energy region to be hardly appropriate for waveguiding purposes. For lower energies the structure seems to be a good candidate for a waveguide. In order to illustrate possible applications of surface-state waveguides two novel devices have been proposed. The first one is a light coupler that enables feeding external light into photonic-crystal waveguides. Fig. 4.6 illustrates the proposed structure. In this device the external light first couples to the surface-state region, then the surface mode enters a tapered region, where it adiabatically (the diameter of the surface rods in the surface-state waveguide gradually decreases to zero) is transformed into a conventional waveguiding state. The maximum achieved transmission reaches T ≈ 0.8 (see the inset to Fig. 4.6), which is even higher than the transmission in conventional tapered [109] or mode-matched structures [110; 111]. The second device that we propose is a directional beamer. We demonstrate that directional emission with the angular spread much less than in conven- 52 CHAPTER 4. RESULTS tional waveguides can also be achieved for the case of surface-state waveguides coupled to air. Figure 4.7 shows the Ez field intensity (a) and the directional diagram (b) for the surface state propagating in a semi-infinite waveguide. 3.5 (a) 1.0 ωa/2πc 0.34 0.36 0.38 Sr (Θ) T 0.6 Θ 2.5 0.2 0.34 1.5 0.38 ωa/2πc 0.42 0.5 2 |E z |, a.u. 0.0 0.5 1.0 1.5 2.0 2.5 -50 (b) -30 -10 0 10 Θ, Degrees 30 50 Figure 4.7: (a) Intensity distribution for the Ez -component of the electromagnetic field in the surface-mode waveguide terminated to air for ωa/2πc = 0.34. (b) Far-field radial component of the Poynting vector Sr (Θ) radiated out of a surface-mode waveguide versus azimuthal angle Θ for different ωa/2π. Inset shows the transmission coefficient for the surface state as a function of the frequency. The most of the beam intensity is localized within cone ∆Θ ∼ 20◦ . Such the narrow beaming originates from the fact that the surface state is localized in a wide spatial region near the surface . 10a (see discussion on the localization in Paper V), as opposed to conventional waveguides, whose width is typically ∼ a and the corresponding diffraction cone is about 70◦ . The angular spread λ , is consistent with the calculated in this case due to the diffraction, sin Θ ∼ 10a far-field radial distribution of the Poynting vector. As the frequency of the incoming light increases, the surface mode becomes more localized, and the spread of the outgoing radiation increases. The effect of directional beaming in surface-mode waveguides might find its practical application for integration of PC-based devices with conventional fiber-optics. 4.5 Nanorod arrays (Paper VI) In this section we will focus on light propagation in large arrays of infinitely long nanorods. Prototypes of such the arrays have been recently fabricated experimentally [112; 113]. These arrays represent randomly oriented or aligned long rods (or spikes) of a material (dielectric or metal), several tens of nanometers in diameter. Nanorods can be considered as two-dimensional analogues of nanoparticles in terms of the mode structure. This makes it possible to study 4.5. NANOROD ARRAYS (PAPER VI) 53 their properties with conventional effective-medium theories. We consider a gelatin matrix (εmat = 2.25) with an embedded two-dimensional array of silver nanorods. The effective dielectric function of the composite can be estimated from Maxwell Garnett (MG) formula (2.19). Despite its adequacy for small isolated circular nanoparticles, the simple MG theory, however, has certain limitations – it does not account for the shape and distribution of the metal clusters in the dielectric medium, neglecting important polarization properties of both single non-circular particles and their arrangements. In order to incorporate these features and study transmission characteristics of periodic and disordered nanorod arrays we apply the recursive Green’s function technique (see Section 3.3). We consider 2D arrays of infinitely long silver nanorods arranged as a square lattice in a gelatin background. Keeping the filling factor of Ag, f = 10%, constant, we consider two cases, (a) a finite-size lattice with thickness a = 0.7µm of nanorods with diameter d = 10nm, and (b) the lattice of the same thickness assembled from nanorods of 60 nm in diameter, see Fig. 4.8. Lattice constants are 29 and 175 nm for cases (a) and (b) respectively. Figure 4.8: Arrays of silver nanorods with diameter (a) 10 nm, and (b) 60 nm embedded in an infinite gelatin background. For both cases thickness of the layer a = 0.7µm and filling factor f =10%. We perform numerical simulations for both the TE and TM polarizations of light incident normally from the left to the boundary between gelatin and the blend. TE-modes Let us irradiate the array of infinitely long nanorods with the TE-polarized light. In this case the Ex and Ey components of the electromagnetic field excite coherent plasmonic oscillations on each nanorod. Figure 4.9 shows the 54 CHAPTER 4. RESULTS calculated transmittance, reflectance and absorptance of the TE mode propagating through the arrays of nanorods. Figure 4.9: Transmittance, reflectance and absorptance of the TE mode travelling through the square arrays of nanorods with diameter (a) 10 nm and (b) 60 nm (see Fig. 4.8 for details). Small nanorods. Let us first concentrate on the array of nanorods with diameter 10 nm [Fig. 4.9(a)]. In the spectra one can clearly distinguish two regions, namely the region of high absorption (λ0 < 600 nm), containing a wide main absorption peak at 414 nm, two minor peaks at 350 and 530 nm and the region of high transmittance (λ0 > 600 nm). Now we will take a closer look at these regions separately. The position of the main extinction resonance agrees well with that obtained from Eq. (2.19). However, in contrast to the Maxwell Garnett theory, the spectrum contains two minor peaks at 350 and 530 nm. These peaks are carefully studied and the results are presented in Paper VI. Rechberger [114] has clarified the origin of these peaks in terms restoring forces between the particles. In order to understand the high transmittance at the wavelength region λ0 > 600 nm, we complement the transmission coefficient with the band diagram of the nanorod array. The plasmonic band in this region is located very close to the light line that results in a strong coupling between the incoming light and the plasmonic Bloch states of the blend region and, consequently, high transmittance. Large nanorods The position of the main extinction peak for large nanorods agrees with that one of the small particles. However, there is an essential difference in the physics behind. When the diameter of a nanoparticle increases, higher-order dipole oscillations now contribute to the resulting extinction spectrum [44]. It has been recently shown [115] that the peak centered at ≈ 400 nm is due to the quadruple resonance of a nanorod, whereas the dipole reso- 4.5. NANOROD ARRAYS (PAPER VI) 55 nance is redshifted and overlaps with the region of the enhanced reflectance (500 < λ0 < 700 nm). The indication in favor of this interpretation is the narrower width of the stop-band in the transmission (60 nm against 100 nm in the case of small rods). This is because the higher-order dipole interactions causing the stop-band behavior for the case of large nanorods are generally weaker. The high-reflectance region is caused by the photonic band gap, as the stricture has the lattice constant of the same order as the wavelength of light and effectively represents a two-dimensional photonic crystal. The plasmonic band extends from ωa/2πc = 0 to 0.4 (λ0 ≈ 660 nm) where it experiences a photonic bandgap that causes the high reflectance of the structure. This bandgap overlaps with the tail of the extinction peak near 500 nm (see Fig. 4.9). Our numerical calculations outline the importance of geometrical factors such as the size of the rods and their arrangement. In particular, we have demonstrated that the interaction between adjacent nanorods brings the significant contribution to the transmission spectra, which is manifested as additional absorption peaks (that are missing in the effective-medium approach). The Maxwell Garnett theory also disregards both the impacts of higher-order dipole contributions and formation of photonic band gaps in the case of arrays of larger nanorods. TM-modes Let us now consider the TM-polarization of the incoming light. Figure 4.10(a) shows the transmittance, reflectance and absorptance of the TM-polarized light for the small nanorods. Here, almost for the whole wavelength range under study light does not penetrate the region occupied by the nanorods and gets fully reflected back, resulting in zero transmittance. This effect can be explained by the skin-effect on the silver rods. Maxwell Garnett theory is not able to describe the important screening properties of the rods, simply averaging the effective dielectric constant over the structure. It is also worth mentioning, that as we consider infinitely long nanorods, the incoming TM-mode does not excite any plasmons on the rods and thus there is no a plasmonic contribution in the overall transmission. The obtained results clearly show that the resonant plasmonic oscillations in periodic nanorod arrays represent a dominating light propagation mechanism for the TE-polarized light, whereas for the TM modes the nanorod structure represents a virtually perfect screen. This feature can be utilized in a nearly 100% effective polarizer. 56 CHAPTER 4. RESULTS Figure 4.10: (a) Transmittance, reflectance and absorptance of the TM-mode through a nanorod array of d = 10 nm. Due to the skin-effect light does not penetrate the blend region. For λ0 < 328 nm the real part ε′ of the dielectric function of silver (b) becomes positive and the transmission coefficient abruptly increases. 4.6 Surface plasmons in polymeric solar cells (Paper VII) In this Section we utilize surface plasmons, excited at the interface of an aluminum grating and a blend of organic polymers, in order to increase efficiency of polymeric solar cells. An increased absorption originating from surface plasmon resonances has been studied by the recursive Green’s function technique and confirmed by experimental studies. We demonstrate that the presence of a plasmon can lead to increased photocurrent from polymeric photovoltaic cells. We consider a supercell (see Fig. 4.11), containing 5 periods of a sinusoidal grating profile. In order to account for the non-ideality of the grating surface, we have introduced a randomized surface roughness with the amplitude of ∼ 5 nm, which is a typical averaged AFM experimental value. We discretize the structure with a uniform grid (element size is 3.4 nm) and illuminate it with the TE-polarized light (the magnetic component of the field is parallel to the grating lines). Having calculated transmittance T (3.32) and reflectance R (3.33) from the recursive Green’s function technique, the absorptance can be found as A = 100% − T − R. We have studied separately two polymers, APFO3 [116] and APFO Green5 [117], blended with PCBM. The dielectric functions for the blends obtained by ellipsometry is given in Paper VII and used as an input for the modelling. The position of the plasmonic resonance for the blends was first estimated by Eq. 4.6. SURFACE PLASMONS IN POLYMERIC SOLAR CELLS (PAPER VII)57 Hz Ex ky Air Polymer Ti (optional) Al grating Figure 4.11: The system under study. The supercell consists of 5 grating periods (the width is 1385 nm, the grating period is 277 nm), discretized with square grid (size of the grid element is 3.4 nm) and illuminated with the TEpolarized light. (2.12). The absorptances of the blends on metal gratings calculated by the recursive Greens function technique are given in Fig. 4.12. First, we calculate the absorptance spectrum for APFO3/PCBM. The polymer blend with thickness d ≈ 150 nm is deposited directly onto the Al grating. The simulated and measured absorptance spectra are given in Fig. 4.12(a). The spectrum contains two resonance peaks at 450 and 625 nm, which agree extremely well with the experimental curve. The two peaks have, however, completely different nature. The peak at 625 nm is a first-order plasmonic resonance and its position agrees quite well with direct analytical estimation (598nm) from Eq. (2.12). At this resonance the electromagnetic field [inset, Fig. 4.12(b)] is localized within the 50-nm region near the grating surface, and its intensity is up to 7 times higher on the boundary than in the bulk of the polymer. The resonant peak at 450 nm represents a standing wave confined by the total internal reflection at the polymer/air interface and the reflecting metal grating. For APFO Green 5/PCBM we model a 90 nm thick polymer blend layer, deposited onto the Al grating coated with a 5 nm thin Ti interfacial layer. Fig. 4.12(b) represents the computed absorptance spectrum and the simulated field distribution inside the cell. As the APFO Green 5/PCBM layer is thinner than the APFO3/PCBM, the standing wave is no longer supported and there is no the corresponding peak in the spectrum. The plasmonic resonance for this structure is centered at 555 nm. In order to estimate the effect of the plasmon on the photocurrent, the external quantum efficiency (EQE) has been measured under illumination of the 58 CHAPTER 4. RESULTS (a) (b) Figure 4.12: (a) Calculated and measured absorptance spectrum of APFO3/PCBM deposited directly on the Al grating. The insets show the calculated spatial distribution of the Hz -component of the electromagnetic field in the polymer at the plasmonic resonance (625 nm) and for the standing wave peak (450 nm). (b) Calculated absorptance spectrum of APFO Green 5/PCBM on the Al grating coated with a 5-nm thick Ti interfacial layer. The inset demonstrates the spatial distribution of the Hz -component of the electromagnetic field in the polymer at the first-order plasmonic resonance (555 nm). The lower dark part of the insets corresponds to the sinusoidal shaped metal grating where no field is present. sample with polarized light. For the TE-polarization, the APFO Green5/PCBM solar cells demonstrate a clear ∼ 4% enhancement of the EQE in the vicinity of the SP resonance (555 nm). However, for even stronger plasmonic resonance in the APFO3/PCBM cell, no any indication of plasmon influence has been observed. This discrepancy can be attributed to the mismatch in the energy of the SP resonance which lies below the bandgap of APFO3/PCBM but but above the gap for APFO Green5/PCBM. This mismatch leads to very weak coupling from the SP to the excitation of APFO3/PCBM. All the measurements has been carried out by the Biomolecular and Organic Electronics group of the Center of Organic Electronics, IFM at Linköping University. Bibliography [1] D. K. Armani, T. J. Klippenberg, S. M. Spillane, and K. J. Vahala. Ultra-high-Q toroid microcavity on a chip. Lett. to Nature, 421:925–928, 2003. [2] K. J. Vahala. Optical microcavities. Nature, 424:839–846, 2003. [3] A. V. Snyder and J. D. Love. Reflection at a curved dielectric interface – electromagnetic tunneling. IEEE Trans. Microwave. Theor. Techn., MTT-23:134–141, 1975. [4] F. T. Smith. Lifetime matrix in collision theory. Phys. Rev., 118:349–356, 1960. [5] R. C. Polson, Z. Vardeny, and D. A. Chinn. Multiple resonances in microdisk lasers of π-conjugated polymers. Appl. Phys. Lett., 81:1561– 1563, 2002. [6] B. Gayral, J. M. Gérard, A. Lemaı̂tre, C. Dupuis, L. Manin, and J. L. Pelouard. High-Q wet-etched GaAs microdisks containing InAs quantum boxes. Appl. Phys. Lett., 75(13):1908–1910, 1999. [7] D. W. Vernooy, V. S. Ilchenko, H. Mabuchi, E. W. Streed, and H. J. Kimble. High-Q measurements of fused-silica microspheres in the near infrared. Opt. Lett., 23:247–249, 1998. [8] M. Kutawa-Gonokami and K. Takeda. Polymer whispering gallery mode lasers. Opt. Mater., 9:12–17, 1998. [9] M. Fujita, K. Inoshita, and T. Bata. Room temperature continuous wave lasing characteristics of GaInAsP/InP microdisk injection laser. Electronic Lett., 34:278–279, 1998. [10] C. Seassal, X. Letartre, J. Brault, M. Gendry, P. Pottier, P. Viktorovitch, O. Piquet, P. Blondy, D. Cros, and O. Marty. InAs quantum wires in InP-based microdiscs: Mode identification and continuos wave room temperature laser operation. J. Appl. Phys., 88:6170–6174, 2000. 59 60 BIBLIOGRAPHY [11] http://ab-initio.mit.edu/photons/index.html. [12] Z. Zhang and M. Qiu. Small-volume waveguide-section high-Q microcavities in 2D photonic crystal slabs. Opt. Express, 12(17):3988–3995, 2004. [13] M. Notomi, A. Shinya, S. Mitsugi, E. Kuramochi, and H-Y. Ryu. Waveguides, resonators and their coupled elements in photonic crystal slabs. Opt. Express, 12(8):1551–1561, 2004. [14] S.-H. Kwon, H.-Y. Ryu, G.-H. Kim, Y.-H. Lee, and S.-B. Kim. Photonic bandedge lasers in two-dimensional square-lattice photonic crystal slabs. Appl. Phys. Lett., 89(19):3870–3872, 2003. [15] J.-K. Yang, S.-H. Kim, G.-H. Kim, H.-G. Park, Y.-H. Lee, and S.-B. Kim. Slab-edge modes in two-dimensional photonic crystals. Appl. Phys. Lett., 84:3016–3018, 2004. [16] S. Xiao and M. Qiu. Surface mode microcavities. Appl. Phys. Lett., 87(11):111102 1–3, 2005. [17] S. Fan, S. G. Johnson, J. D. Joannopoulos, C. Manolatou, and H. A. Haus. Waveguide branches in photonic crystals. J. Opt. Soc. Am. B, 18:162–165, 2001. [18] S. Fan, P. R. Villeneuve, and J. D. Joannopoulos. Channel drop tunneling through localized states. Phys. Rev. Lett., 80:960–963, 1998. [19] J. S. Foresi, P. R. Villeneuve, J. Ferrera, E. R. Thoen, G. Steinmeyer, S. Fan, J. D. Joannopoulos, L. C. Kimerling, Henry I. Smith, and E. P. Ippen. Photonic-bandgap microcavities in optical waveguides. Lett. to Nature, 390:143–145, 1997. [20] M. F. Yanik, S. Fan, M. Soljačić, and J. D. Joannopoulos. All-optical transistor action with bistable switching in a photonic crystal crosswaveguide geometry. Opt. Lett., 28(24):2506–2508, 2003. [21] J. D. Joannopoulos, R. D. Meade, and J. N. Winn. Photonic crystals: Molding the Flow of Light. Princeton University Press, Princeton, 1995. [22] F. Ramos-Mendieta and P. Halevi. Surface electromagnetic waves in twodimensional photonic crystals: Effect of the position of the surface plane. Phys. Rev. B, 59:15112–15120, 1999. [23] X. Zhang amd L.-M. Li, Z.-Q. Zhang, and C. T. Chan. Surface states in two-dimensional metallodielectric photonic crystals studied by a multiplescattering method. Phys. Rev. B, 63:125114 1–7, 2001. BIBLIOGRAPHY 61 [24] J. M. Elson and K. Halterman. Local density of states analysis of surface wave modes on truncated photonic crystal surfaces with nonlinear material. Opt. Express, 12:4855–4863, 2004. [25] E. Moreno, F. J. Garcı́a-Vidal, and L. Matrin-Moreno. Enhanced transmission and beaming of light via photonic crystal surface modes. Phys. Rev. B, 69:121402 1–4(R), 2004. [26] R. H. Ritchie. Plasma losses by fast electrons in thin films. Phys. Rev., 106(5):874–881, 1957. [27] W. L. Barnes, A. Dereux, and T. W. Ebbesen. Surface plasmon subwavelength optics. Nature, 424:824–830, 2003. [28] J. R. Sambles, G. W. Bradbery, and F. Yang. Optical excitation of surface plasmons: An introduction. Contemp. Phys, 32(3):173–183, 1991. [29] A. Otto. Excitation of nonradiative surface plasma waves in silver by the method of frustrated total reflection. Zeitschrift für Physik A Hadrons and Nuclei, 216(4):398–410, 1968. [30] E. Kretschmann and H. Reather. Radiative decay of nonradiative surface plasmon excited by light. Z. Naturf., 23A:2135–2136, 1968. [31] R. H. Ritchie, E. T. Arakawa, J. J. Cowan, and R. N. Hamm. Surfaceplasmon resonance effect in grating diffraction. Phys. Rev. Lett., 21(22):1530–1533, 1968. [32] H. Caglayan, I. Bulu, and E. Ozbay. Plasmonic structures with extraordinary transmission and highly directional beaming properties. Microwave Opt. Techn. Lett., 48(12):2491–2496, 2006. [33] A. Krishnan, T. Thio, T. J. Kim, H. J. Lezec, T. W. Ebbesen, P. A. Wolff, J. Pendry, L. Martin-Moreno, and F. J. Garcia-Vidal. Evanescently coupled resonance in surface plasmon enhanced transmission. Opt. Comm., 200(1–6):1–7, 2006. [34] S. I. Bozhevolnyi, V. S. Volkov, Eloı̈se Devaux, J.-Y. Laluet, and T. W. Ebbesen. Channel plasmon subwavelength waveguide components including interferometers and ring resonators. Nature, 440:508–511, 2006. [35] S. C. Kitson, W. L. Barnes, and J. R. Sambles. Full photonic band gap for surface modes in the visible. Phys. Rev. Lett., 77(13):2670–2673, 1996. [36] M. I. Stockman. Nanofocusing of optical energy in tapered plasmonic waveguides. Phys. Rev. Lett., 93(13):137404 1–4, 2004. 62 BIBLIOGRAPHY [37] T. H. Taminiau, R. J. Moerland, F. B. Segerink, L. Kuipers, and N. F. van Hulst. λ/4 resonance of an optical monopole antenna probed by single molecule fluorescence. Nano Lett., 7(1):28–33, 2007. [38] F. Kusunoki, T. Yotsuya, and J. Takahara. Confinement and guiding of two-dimensional optical waves by low-refractive-index cores. Opt. Express, 14:5651–5656, 2006. [39] W.-Y. Wong, X.-Z. Wang, Z. He, A. B. Djuris, C.-T. Yip, K.-Y. Cheung, H. Wang, C. S. K. Mak, and W.-K. Chan. Metallated conjugated polymers as a new avenue towards high-efficiency polymer solar cells. Nature Mat., Published online: 13 May, 2007. [40] D. M. Schaadt, B. Feng, and E. T. Yu. Enhanced semiconductor optical absorption via surface plasmon excitation in metal nanoparticles. Appl. Phys. Lett., 86(6):063106 1–3, 2005. [41] M. Westphalen, U. Kreibig, J. Rostalski, H. Luth, and D. Meissner. Metal cluster enhanced organic solar cells. Sol. Energy Mat. Sol. Cells, 61(1):97– 105, 2000. [42] D. Derkacs, S. H. Lim, P. Matheu, W. Mar, and E. T. Yu. Improved performance of amorphous silicon solar cells via scattering from surface plasmon polaritons in nearby metallic nanoparticles. Appl. Phys. Lett., 89(9):093103 1–3, 2006. [43] B. P. Rand, P. Peumans, and S. R. Forrest. Long-range absorption enhancement in organic tandem thin-film solar cells containing silver nanoclusters. J. Appl. Phys., 96(12):7519–7526, 2004. [44] U. Kreibig and M. Vollmer. Optical properties of metal clusters, volume 25 of Springer Series in Materials Science. Springer, Berlin, 1995. [45] B. T. Draine and P. J. Flatau. Discrete-dipole approximation for scattering calculations. J. Opt. Soc. Am. A, 11(4):1491–1499, 1994. [46] G. W. Ford and W. H. Weber. Electromagnetic interactions of molecules with metal surfaces. Phys. Rep., 113(4):195–287, 1984. [47] E. X. Jin and X. Xu. Finite-difference time-domain studies on optical transmission through planar nano-apertures in a metal film. Japan. J. Appl. Phys., 43(1):407–417, 2004. [48] G. Gouesbet and G. Grehan. Generalized Lorentz-Mie theory for assemblies of spheres and aggregates. J. Opt. A: Pure Appl. Opt., 1(6):706–712, 1999. BIBLIOGRAPHY 63 [49] J. C. M. Garnett. Colors in material glasses and metal films. Philos. Trans. R. Soc. London, Ser. B, 3:385–420, 1904. [50] D. A. G. Bruggeman. Berechnung verschiedener physikalischer konstanten von heterogenen substanzen. Ann. Phys. (Leipzig), 24:636–679, 1935. [51] P. Sheng. Theory for the dielectric function of granular composite media. Phys. Rev. Lett., 45(1):60–63, 1980. [52] S. Giordano. Effective medium theory for dispersions of dielectric ellipsoids. J. Electrostat., 58(1–2):59–76, 2003. [53] I. L. Skryabin, A. V. Radchik, P. Moses, and G. B. Smith. The consistent application of Maxwell-Garnett effective medium theory to anisotropic composites. Appl. Phys. Lett., 70(17):2221–2223, 1997. [54] R. Ruppin. Evaluation of extended Maxwell-Garnett theories. Opt. Comm., 182(4–6):273–279, 2000. [55] G. P. Wang, Y. Yi, and W. Lin. Tunable and omnidirectional photonic bandgap properties of one-dimensional photonic crystals fabricated by holography. J. Opt. Soc. Am. B, 21(3):554–561, 2004. [56] Z. Wu, J. Walish, A. Nolte, L. Zhai, R. E. Cohen, and M. F. Rubner. Deformable antireflection coatings from polymer and nanoparticle multilayers. Adv. Mat., 18(20):2699–2702, 2006. [57] A. Stassinopoulos, R. N. Das, E. P. Giannelis, S. H. Anastasiadis, and D. Anglos. Random lasing from surface modified films of zinc oxide nanoparticles. Appl. Surf. Sci., 247:18–24, 2005. [58] United States Patent No. 7.074.463. Durable optical element. [59] A. Prakash, J. Ouyang, J.-L. Lin, and Y. Yang. Polymer memory device based on conjugated polymer and gold nanoparticles. J. Appl. Phys., 100(5):054309 1–5, 2006. [60] D. L. Burris and W. G. Sawyer. Improved wear resistance in aluminaPTFE nanocomposites with irregular shaped nanoparticles. Wear, 260(7– 8):915–918, 2006. [61] A. S. Hamdy. Advanced nano-particles anti-corrosion ceria based sol gel coatings for aluminum alloys. Mat. Lett., 60(21–22):2633–2637, 2006. [62] United States Patent No. 7.052.777. Composite nanoparticle materials and method of making the same. 64 BIBLIOGRAPHY [63] S. Schelm and G. B. Smith. Dilute LaB6 nanoparticles in polymer as optimized clear solar control glazing. Appl. Phys. Lett., 82(24):4346– 4348, 2003. [64] H. B. Na, J. H. Lee, K. An, Y. I. Park ans M. Park, I. S. Lee, D.-H. Nam, S. T. Kim, S.-H. Kim, S.-W. Kim, K.-H. Lim, K.-S. Kim, S.-O. Kim, and T. Hyeon. Development of a T1 contrast agent for magnetic resonance imaging using MnO nanoparticles. Angew. Chem. Int. Ed., Mar. 2007. [65] http://www.altairnano.com/markets amps.html. [66] S.-U. Rathera, R. Zacharia, M.-U. Naika S. W. Hwanga, and K. S. Nahm. Hyperstoichiometric hydrogen storage in monodispersed palladium nanoparticles. Chem. Phys. Lett., 438(1–3):78–84, 2007. [67] W. H. Lizcano-Valbuena, D. C. de Azevedo, and E. R. Gonzalez. Supported metal nanoparticles as electrocatalysts for low-temperature fuel cells. Electrochim. Acta, 49(8):1289–1295(7), 2007. [68] V. Zaporojtchenko, R. Podschun, U. Schürmann, A. Kulkarni, and F. Faupel. Physico-chemical and antimicrobial properties of co-sputtered AgAu/PTFE nanocomposite coatings. Nanotech., 17(19):4904–4908, 2006. [69] C.-C. You, O. R. Miranda, B. Gider, P. S. Ghosh, I.-B. Kim, B. Erdogan, S. A. Krovi, U. H. F. Bunz, and V. M. Rotello. Detection and identification of proteins using nanoparticlefluorescent polymer ’chemical nose’ sensors. Nature Nanotech., 2:318–323, 2007. [70] J. T. Nurmi, P. G. Tratnyek, V. Sarathy, D. R. Baer, J. E. Amonette, K. Pecher, C. Wang, J. C. Linehan, D. W. Matson, R. L. Penn, and M. D. Driessen. Characterization and properties of metallic iron nanoparticles: spectroscopy, electrochemistry, and kinetics. Environ. Sci. Technol., 39(5):1221–1230, 2005. [71] C. Kurdak, J. Kim, A. Kuo, J. J. Lucido, L. A. Farina, X. Bai, M. P. Rowe, and A. J. Matzger. 1/f noise in gold nanoparticle chemosensors. Appl. Phys. Lett., 86(7):073506 1–3, 2005. [72] P. Jain and T. Pradeep. Potential of silver nanoparticle-coated polyurethane foam as an antibacterial water filter. Biotech. Bioeng., 90(1):59–63, 2005. [73] H. Xu. Theoretical study of coated spherical metallic nanoparticles for single-molecule surface-enhanced spectroscopy. Appl. Phys. Lett., 85(24):5980–5982, 2004. BIBLIOGRAPHY 65 [74] R. Quidant, G. Lévêque, J.-C. Weeber, A. Dereux, C. Girard, and J. Weiner. Frustrated energy transport through micro-waveguides decorated by gold nanoparticle chains. Europhys. Lett., 6(6):785–791, 2004. [75] K. S. Yee. Numerical solution of initial boundary-value problems involving Maxwell’s equations in isotropic media. IEEE Trans. Ant. Prop., AP-14:302–307, 1966. [76] J.-P. Berenger. A perfectly matched layer for the absorption of electromagnetic waves. J. Comput. Phys., 114(2):185–200, 1994. [77] A. Mekis, J. C. Chen, I. Kurland, S. Fan, P. R. Villeneuve, and J. D. Joannopoulos. High transmission through sharp bends in photonic crystal waveguides. Phys. Rev. Lett., 77:3787–3790, 1996. [78] P. C. Waterman. Symmetry, unitarity and geometry in electromagnetic scattering. Phys. Rev. D, 3:825–839, 1971. [79] M. I. Mishchenko, L. D. Travis, and A. A. Lacis. Scattering, Absorption, and Emission of Light by Small Particles. Campridge University Press, Cambridge, 2002. [80] P. A. Knipp and T. L. Reinecke. Boundary-element method for the calculation of the electronic states in semiconductor nanostructures. Phys. Rev. B, 54:1880–1891, 1996. [81] J. Wiersig. Boundary element method for resonances in dielectric microcavities. J. Opt. A: Pure Appl. Opt., 5:53–60, 2003. [82] S. V. Boriskina, T. M. Benson, P. Sewell, and A. I. Nosich. Highly efficient design of spectrally engineered whispering-gallery-mode microlaser resonators. Opt. and Quant. Electr., 35:545–559, 2003. [83] K. M. Ho, C. T. Chan, and C. M. Soukoulis. Existence of a photonic gap in periodic dielectric structures. Phys. Rev. Lett., 65(25):3152–3155, 1990. [84] D. Cassagne, C. Jouanin, and D. Bertho. Hexagonal photonic-band-gap structures. Phys. Rev. B, 53(11):7134–7142, 1996. [85] Z.-Y. Li, J. Wang, and B.-Y. Gu. Creation of partial band gaps in anisotropic photonic-band-gap structures. Phys. Rev. B, 58(7):3721– 3729, 1998. [86] S. Shi, C. Chen, and D. W. Prather. Revised plane wave method for dispersive material and its application to band structure calculations of photonic crystal slabs. Appl. Phys. Lett., 86:043104 1–3, 2005. 66 BIBLIOGRAPHY [87] J. B. Pendry. Calculating photonic band structure. J. Phys.: Condens. Matter, 8:1085–1108, 1996. [88] D. Felbacq, G. Tayeb, and D. Maystre. Scattering by a random set of parallel cylinders. J. Opt. Soc. Am. A, 11:2526–2538, 1994. [89] D. M. Whittaker and I. S. Culshaw. Scattering-matrix treatment of patterned multilayer photonic structures. Phys. Rev. B, 60(4):2610–2618, 1999. [90] Z.-Y. Li and K.-M. Ho. Light propagation through photonic crystal waveguide bends by eigenmode examinations. Phys. Rev. B, 68:045201 1–12, 2003. [91] Z.-Y. Li and K.-M. Ho. Light propagation in semi-infinite photonic crystals and related waveguide structures. Phys. Rev. B, 68:155101 1–15, 2003. [92] E. Moreno, D. Erni, and C. Hafner. Modeling of discontinuities in photonic crystal waveguides with the multiple multipole method. Phys. Rev. E, 66:036618 1–12, 2002. [93] O. J. F. Martin, A. Dereux, and C. Girard. Iterative scheme for computing exactly the total field propagating in dielectric structures of arbitrary shape. J. Opt. Soc. Am. A, 11(3):1073–1080, 1994. [94] O. J. F. Martin, C. Girard, and A. Dereux. Generalized field propagator for electromagnetic scattering and light confinement. Phys. Rev. Lett., 74(4):526–529, 1995. [95] O. J. F. Martin and N. B. Piller. Electromagnetic scattering in polarizable backgrounds. Phys. Rev. E, 58(3):3909–3915, 1998. [96] M. Paulus, P. Gay-Balmaz, and O. J. F. Martin. Accurate and efficient computation of the Green’s tensor for stratified media. Phys. Rev. E, 62(4):5797–5807, 2000. [97] V. V. Nikolsky and T. I. Nikolskaya. Decomposition approach to the problems of electrodynamics. Nauka, Moskow, 1983. in Russian. [98] S. Datta. Electronic Transport in Mesoscopic Systems. Cambridge University Press, Cambridge, 1995. [99] M. F. Crommie, C. P. Lutz, , and D. M. Eigler. Imaging standing waves in a two-dimensional electron gas. Nature, 363:524–527, 1993. [100] M. F. Crommie, C. P. Lutz, and D. M. Eigler. Confinement of electrons to quantum corrals on a metal surface. Science, 262:218–220, 1993. BIBLIOGRAPHY 67 [101] D. K. Ferry and S. M. Goodnik. Transport in Nanostructures. Cambridge University Press, Cambridge, 1997. [102] K. Sakoda. Optical properties of photonic crystals. Springer, Berlin, 2001. [103] P. M. Morse and H. Feshbach. Methods of theoretical physics. McGrawHill College, 1953. [104] J. P. Kottmann and O. J. F. Martin. Accurate solution of the volume integral equation for high-permittivity scatterers. IEEE Trans. Antenn. Propag., 48(11):1719–1726, 2000. [105] J. Alegret, M. Käll, and P. Johansson. Top-down extended meshing algorithm and its applications to Green’s tensor nano-optics calculations. Phys. Rev. E, 75(4):046702 1–8, 2007. [106] http://www.almaden.ibm.com/vis/stm/corral.html. [107] E. J. Heller, M. F. Crommie, C. P. Lutz, and D. M. Eigler. Scattering and absorbtion of surface electron waves in quantum corrals. Lett. to Nature, 369:464–466, 1994. [108] H. K. Harbury and W. Porod. Elastic scattering theory for electronic waves in quantum corrals. Phys. Rev. B, 53:15455–15458, 1996. [109] A. Mekis and J. D. Joannopoulos. Tapered couplers for efficient interfacing between dielectric and photonic crystal waveguides. IEEE J. Lightwave Technol., 19:861–865, 2001. [110] P. Sanches, P. Bienstman, B. Luyssaert, R. Baets, and J. Marti. Analysis of butt coupling in photonic crystals. IEEE J. Quant. Electron., 40:541– 550, 2004. [111] R. Stoffer, H. Hoekstra, R. M. De Ridder, E. V. Groesen, , and F. P. H. Van Beckum. Numerical studies of 2D photonic crystals: Waveguides, coupling between waveguides and filters. Opt. Quant. Electron., 32:947– 961, 2000. [112] A. Dev, S. Kar, S. Chakrabarti, and S. Chaudhuri. Optical and field emission properties of ZnO nanorod arrays synthesized on zinc foils by the solvothermal route. Nanotech., 17(5):1533–1540, 2006. [113] D. Losic, J. G. Shapter, J. G. Mitchell, and N. H. Voelcker. Fabrication of gold nanorod arrays by templating from porous alumina. Nanotech., 16(8):2275–2281, 2005. 68 BIBLIOGRAPHY [114] W. Rechberger, A. Hohenau, A. Leitner, J. R. Krenn, B. Lamprecht, and F. R. Aussenegg. Optical properties of two interacting gold nanoparticles. Opt. Comm., 220(1–3):137–141, 2003. [115] K. L. Kelly, E. Coronado, L. L. Zhao, and G. C. Schatz. The optical properties of metal nanoparticles: the influence of size, shape, and dielectric environment. J. Phys. Chem. B, 107(3):668–677, 2003. [116] M. Svensson, F. Zhang, S. C. Veenstra, W. J. H. Verhees, J. C. Hummelen, J. M. Kroon, O. Inganäs, and M. R. Andersson. High-performance polymer solar cells of an alternating polyfluorene copolymer and a fullerene derivative. Adv. Mat., 15(12):988–991, 2003. [117] F. Zhang, W. Mammo, L. M. Andersson, S. Admassie, M. R. Andersson, and O. Inganäs. Low-bandgap alternating fluorene copolymer/methanofullerene heterojunctions in efficient near-infrared polymer solar cells. Adv. Mat., 18(16):2169–2173, 2006. APPENDICES 69 I Paper I Effects of boundary roughness on a Q factor of whispering-gallery-mode lasing microdisk cavities J. Appl. Phys., vol. 94, pp. 7929–7931, 2003 JOURNAL OF APPLIED PHYSICS VOLUME 94, NUMBER 12 15 DECEMBER 2003 Effects of boundary roughness on a Q factor of whispering-gallery-mode lasing microdisk cavities A. I. Rahachou and I. V. Zozoulenkoa) Department of Science and Technology (ITN), Linköping University, 601 74 Norrköping, Sweden 共Received 30 May 2003; accepted 23 September 2003兲 We perform numerical studies of the effect of sidewall imperfections on the resonant state broadening of the optical microdisk cavities for lasing applications. We demonstrate that even small edge roughness (ⱗ/30) causes a drastic degradation of high-Q whispering gallery 共WG兲-mode resonances reducing their Q values by many orders of magnitude. At the same time, low-Q WG resonances are rather insensitive to the surface roughness. The results of numerical simulation obtained using the scattering matrix technique, are analyzed and explained in terms of wave reflection at a curved dielectric interface combined with the examination of Poincaré surface of sections in the classical ray picture. © 2003 American Institute of Physics. 关DOI: 10.1063/1.1625781兴 During recent years, significant experimental efforts were put forward toward the investigation of laser emission of dielectric and polymeric low-threshold microdisk cavities.1–9 The high efficiency of lasing operation in such devices is related to the existence of natural cavity resonances known as whispering-gallery 共WG兲 modes. The origin of these resonances can be envisioned in a ray optic picture, wherein light is trapped inside the cavity through total internal reflections on the cavity–air boundary. One of the most important characteristics of cavity resonances is their quality factor (Q factor兲 defined as Q ⫽2 * 共stored energy兲/共energy lost per cycle兲. The Q factor of a microdisk cavity is mostly governed by a radiative leakage through the curved interface due to diffraction. An estimation of the Q factor in an ideal disk cavity of a typical diameter d⬃10 m for a typical WG resonance gives Q ⬃1013 关see below, Eq. 共4兲兴. At the same time, experimental measured values reported so far are typically in the range of 103 – 104 or lower. A reduction of a Q factor may be attributed to a variety of reasons including side wall geometrical imperfection, inhomogeneity of the height and diffraction index of the disk, effects of coupling to the substrate or pedestal, and others. Several experimental studies point out side wall imperfections as the main factor affecting the Q factor of the cavity.5–7 An indirect indication of the importance of this factor in disk microcavities is provided by the observation that typical Q factors of spheroidal microcavities are several orders of magnitude higher than those of microdisks of comparable dimensions.1,10 This is believed to be due to the superior quality of the microsphere surfaces where boundary scattering may be limited by thermal fluctuations of the surface only. Therefore, the effect of surface roughness appears to be of crucial importance for the design, tailoring, and optimization of Q values of lasing microdisk cavities. In the present article, we provide a detailed numerical study of this effect and analyze the obtained results in terms of the wave and ray dynamics. In order to compute the resonant states of a cavity of an arbitrary shape, we develop an approach based on the scattering matrix technique. The scattering matrix technique is widely used in analysis of waveguides,11 as well as in quantum mechanical simulations.12 This technique was also used for an analysis of resonant cavities for geometries when the analytical solution was available.13 Note that because the problem at hand requires a fine discretization of the geometry, commonly used time-domain finite difference methods14 would be prohibitively expensive in terms of both computer power and memory. While a detailed description of the calculations will be given elsewhere, we present here the essence of the method. We consider a two-dimensional cavity with the refraction index n surrounded by air. Because the majority of experiments are performed only with the lowest transverse mode occupied, we neglect the transverse (z-) dependence of the field and thus limit ourselves to the two-dimensional Helmholtz equation. We divide our system into outer and an inner regions. In the outer region, the refraction index n is independent of the coordinate and the solution to the Helmholtz equation can be written in polar coordinates in the form ⫹⬁ ⌿⫽ 共1兲 where ⌿⫽E z (H z ) for transverse magnetic 共TM兲 关transverse (2) electric 共TE兲兴 modes, H (1) q ,H q are the Hankel functions of the first and second kind of order q describing, respectively, incoming and outgoing waves, k⫽ /c⫽2 /. We define the scattering matrix S in a standard fashion,11–13 B⫽SA, where A,B are column vectors composed of the expansion coefficients A q ,B q for incoming and outgoing states in Eq. 共1兲. The matrix element Sq ⬘ q gives the probability amplitude of scattering from an incoming state q into an outgoing state q ⬘ . In order to apply the scattering a兲 Electronic mail: [email protected] 0021-8979/2003/94(12)/7929/3/$20.00 (1) iq , 兺 共 A q H (2) q 共 kr 兲 ⫹B q H q 共 kr 兲兲 e q⫽⫺⬁ 7929 © 2003 American Institute of Physics Downloaded 10 Dec 2003 to 130.236.132.220. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp 7930 J. Appl. Phys., Vol. 94, No. 12, 15 December 2003 A. I. Rahachou and I. V. Zozoulenko FIG. 1. 共a兲 Dependence of the quality factor Q⫽Q() of the circular disk for different surface roughness ⌬r indicated in the figure; the disk radius is R⫽5 m, the refraction index n⫽1.8. The inset illustrates a cavity where the surface roughness ⌬r⫽200 nm and ⌬l⫽2 R/50 共the dotted line represents an ideal circular boundary, the shaded region corresponds to the cavity兲. characterizes the average radius of local curvature due to boundary imperfections. TE modes of the cavity exhibit similar features and are not shown here. matrix technique, we divide the inner region into N narrow concentric rings. At each ith boundary between the rings, we introduce the scattering matrix Si that relates the states propagating 共or decaying兲 toward the boundary, with those propagating 共or decaying兲 away from the boundary. The matrices Si are derived using the requirement of continuity of the tangential components for the E and H field at the boundary between the two dielectric media. Successively combining the scattering matrixes for all the boundaries,11,12 S1 丢 . . . 丢 SN, we can relate the combined matrix to the scattering matrix S. To identify the resonant states of a resonant cavity, we introduce the Wigner–Smith time-delay matrix Q ⫽ i/c(dS† /dk)S, 2,13,15 where the diagonal elements Qqq give a time delay experienced by the wave incident in qth channel and scattered into all other channels. The Q value of the cavity is Q⫽ D (k), where D (k) is the total time delay averaged over all M incoming channels,2,13,16 M D共 k 兲 ⫽ M d 1 1 1 d Q ⫽ ⫽ , M q⫽1 qq cM ⫽1 dk cM dk 兺 兺 共2兲 exp(i)⫽ are the eigenvalues of the scattering matrix S, ⫽ 兺 N ⫽1 is the total phase of the determinant of the matrix S, and det S⫽⌸ M⫽1 ⫽exp(i). Figure 1 shows calculated Q values of the disk resonant cavity for different surface roughnesses for TM modes in some representative wavelength intervals. Note that an exact experimental shape of the cavity–surface interface is not available. We thus model the interface shape as a superposition of random Gaussian deviations from an ideal circle of radius R with a maximal amplitude ⌬r/2 and a characteristic distance between the deviation maxima ⌬l⬃2 R/50 共see illustration in inset to Fig. 1兲. The solid curve in Fig. 1 corresponds to an ideal disk cavity without imperfections. Resonant states of an ideal disk 共as well as the bound states of the corresponding closed resonator兲 are characterized by two numbers, q 关see Eq. 共1兲兴 and FIG. 2. Dependence Q on the surface roughness ⌬r for several representative resonances. 共Each curve remains practically unchanged for different realizations of surface roughness兲. The inset shows the dependence of local radius of roughness curvature subject to ⌬r. Parameters of the cavity are the same as in Fig. 1. m. The index m is a radial wave number that is related to the number of nodes of the field components in the radial direction r. The angular wave number q can be related to the angle of incidence in a classical ray picture2 q⫽nkR sin . 共3兲 The dependence of the averaged Q values on the surface roughness ⌬r is summarized in Fig. 2 for several representative resonances. A common feature of all high-Q resonances is a dramatic decrease of their maximal Q-value that takes place for very small values of ⌬rⱗ/20. For example, a Q value of at the resonant state TM83,1 drops from Q ⬇1013 for an ideal disk to Q⬇103 for a surface roughness of only ⌬r⫽20 nm. In contrast, low-Q resonances show a rather slow decrease in their Q values over the range of variation of ⌬r. For example, for the same surface roughness ⌬r⫽20 nm, the Q value of the resonant state TM56,7 decreases only by a factor of 1.5, dropping to Q⬇200. In order to understand these features, we combine a Poincaré surface of section 共SOS兲 method with an analysis of ray reflection at a curved dielectric interface.18 The Q value of the cavity can be related to the transmission probability T of an electromagnetic wave incident on a curved interface of radius by Q⫽2nk cos /T 共this expression is valid for large angles of incidence when TⰆ1).17 In turn, for kn Ⰷ1, the transmission probability reads18 冋 T⫽ 兩 T F 兩 exp ⫺ 册 2 nk 共 cos2 c ⫺cos2 兲 3/2 , 3 sin2 共 兲 共4兲 where T F is the classical Fresnel transmission coefficient for an electromagnetic wave incident on a flat surface, c ⫽arcsin(1/n) is an angle of total internal reflection. Figure 3 illustrates that T decreases exponentially as the difference ⫺ c grows. It should also be noted that the roughness ⌬r⬃100– 200 nm 共i.e., of the order of /5) almost completely suppress the resonances, which would diminish the lasing action of the cavity. The inset to Fig. 3 depicts the Poincaré SOS for two states with q⫽56 and 83 shown in Fig. 1, where the initial Downloaded 10 Dec 2003 to 130.236.132.220. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp J. Appl. Phys., Vol. 94, No. 12, 15 December 2003 FIG. 3. Dependence T⫽T( ) for several radii of curvature according to Eq. 共4兲. Inset shows a Poincaré SOS for the states q⫽83 and q⫽56 for the cavity with ⌬r⫽0 共straight lines of ⫽constant) and ⌬r⫽20 nm. The number of bounces for a given angle of incidence 0 is chosen in such a way that the total path of the ray does not exceed the one extracted from the numerically calculated Q value for the corresponding resonance, L⫽c D ⫽Q/k. angle of incidence 0 of launched rays is related to the angular number q by Eq. 共3兲. The SOS demonstrates that initially regular dynamics of an ideal cavity transforms into a chaotic one even for a cavity with maximum roughness ⌬r ⱗ20 nm. ⌬T 83,1 ch in Fig. 3 indicates the estimated increase in the transmission coefficient due to the broadening of the phase space, ⌬ ch , as extracted from the Poincaré SOS for the state with q⫽83. This corresponds to the decrease of ⌬Q⬃⌬T ⫺1 ⬇10⫺2 . This value is much smaller that the actual calculated decrease of the Q factor for the high-Q resonance TM83,1 . To explain the rapid degradation of high-Q resonances, we concentrate on another aspect of the wave dynamics. Namely, the imperfections at the surface boundary effectively introduce a local radius of surface curvature that is distinct from the disk radius R 共see illustration in Fig. 1兲. One may thus expect that with the presence of a local surface curvature, the total transmission coefficient will be determined by the averaged value of rather than by the disk radius R. The dependence of on surface roughness ⌬r for the present model of surface imperfections is shown in the inset to Fig. 2. Figure 3 demonstrates that the reduction of the local radius of curvature from 5 m 共ideal disk兲 to 1.7 m (⌬r⫽20 nm) causes an increase of the transmission coefficient by ⌬T cur⬇108 . This estimate, combined with the estimate based on the change of ⌬T ch is fully consistent with the Q-factor decrease shown in Figs. 1 and 2. We thus conclude that the main mechanism responsible for the rapid degradation of high-Q resonances in nonideal cavities is the enhanced radiative decay through the curved surface because the effective local radius 共given by the surface roughness兲 is smaller that the disk radius R. For the case of low-Q resonances, the change in the A. I. Rahachou and I. V. Zozoulenko 7931 transmission coefficient due to enhanced radiative decay ⌬T cur is of the same magnitude as the change ⌬T ch due to the broadening of the phase space caused by the transition to chaotic dynamics 共as illustrated in Fig. 3 for the resonance TM56,7). Therefore, both of these factors play comparable roles in degradation of the low-Q WG resonances. It is worth mentioning that one often assumes that longlived high-Q resonances in idealized cavities 共e.g., in ideal disks, hexagons, etc.兲 are not important for potential application in optical communication or laser devices.13,19 because of their extremely narrow width. Our simulations demonstrate that it is not the case because, in real structures, the Q values of these resonances becomes comparable to those of intermediate-Q resonances already for small or moderate surface roughness of ⌬r⬃10– 50 nm. To conclude, our results pinpoint the surface roughness as a major factor affecting the performance of microdisk cavities and outline its role for device design and fabrication. In particularly, we demonstrate that even small surface roughness of the order ⌬r⬃/5 can cause the drastic degradation of the cavity modes, completely suppressing the lasing effect of the disk resonators. The authors thank Olle Inganäs for stimulating discussions that initiated this work and Stanley Miklavcic and Sayan Mukherjee for many useful discussions and conversations. One of the authors 共A.I.R.兲 acknowledges financial support from SI and KVA. Y. Yamamoto and R. E. Slusher, Phys. Today, 66 共1993兲; S. Arnold, Am. Sci. 89, 414 共2001兲. J. U. Nöckel and R. K. Chang, in Cavity-Enhanced Spectroscopies, Experimental Methods in the Physical Sciences Vol. 40, edited by R. D. van Zee and J. P. Looney 共Academic, San Diego, 2002兲, pp. 185–226. 3 S. L. McCall, A. F. J. Levi, R. E. Slusher, S. J. Pearton, and R. A. Logan, Appl. Phys. Lett. 60, 289 共1992兲. 4 A. Dodabalapur, M. Berggren, R. E. Slusher, Z. Bao, A. Timko, P. Schiortino, E. Laskowski, H. E. Katz, and O. Nalamasu, IEEE J. Sel. Top. Quantum Electron. 4, 67 共1998兲. 5 M. Fujita, K. Inoshita, and T. Bata, Electron. Lett. 34, 278 共1998兲. 6 B. Gayral, J. M. Gérard, A. Lemaı̂tre, C. Dupuis, L. Manin, and J. L. Pelouard, Appl. Phys. Lett. 75, 1908 共1999兲. 7 C. Seassal, X. Letartre, J. Brault, M. Gendry, P. Pottier, P. Viktorovitch, O. Piquet, P. Blondy, D. Cros, and O. Marty, J. Appl. Phys. 88, 6170 共2000兲. 8 M. Theander, T. Granlund, D. M. Johanson, A. Ruseckas, V. Sundström, M. R. Andersson, and O. Inganäs, Adv. Mater. 共Weinheim, Ger.兲 13, 323 共2001兲. 9 R. C. Polson, Z. Vardeny, and D. A. Chinn, Appl. Phys. Lett. 81, 1561 共2002兲. 10 V. S. Ilchenko, M. L. Gorodetsky, X. S. Yao, and L. Maleki, Opt. Lett. 26, 256 共2001兲. 11 V. V. Nikolsky and T. I. Nikolskaya, Decomposition Approach to the Problems of Electrodynamics 共Nauka, Moskow, 1983兲 共in Russian兲. 12 S. Datta, Electronic Transport in Mesoscopic Systems 共Cambridge University Press, Cambridge, UK, 1995兲. 13 M. Hentschel and K. Richter, Phys. Rev. E 66, 056207 共2002兲. 14 B.-J. Li and P.-L. Liu, IEEE J. Quantum Electron. 33, 1489 共1997兲. 15 F. T. Smith, Phys. Rev. 118, 349 共1960兲. 16 M. Bauer, P. A. Mello, and K. W. McVoy, Z. Phys. A 293, 151 共1979兲. 17 M. Hentschel and H. Schomerus, Phys. Rev. E 65, 045603共R兲 共2002兲. 18 A. V. Snyder and J. D. Love, IEEE Trans. Microwave Theory Tech. 23, 134 共1975兲. 19 J. Wiersig, J. Opt. A, Pure Appl. Opt. 5, 53 共2003兲. 1 2 Downloaded 10 Dec 2003 to 130.236.132.220. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/japo/japcr.jsp II Paper II Scattering matrix approach to the resonant states and Q values of microdisk lasing cavities Appl. Opt., vol. 43, pp. 1761–1772, 2004 Scattering matrix approach to the resonant states and Q values of microdisk lasing cavities Aliaksandr I. Rahachou and Igor V. Zozoulenko We develop a scattering matrix approach for the numerical calculation of resonant states and Q values of a nonideal optical disk cavity with an arbitrary shape and with an arbitrary varying refraction index. The developed method is applied to study the effect of surface roughness and inhomogeneity of the refraction index on Q values of microdisk cavities for lasing applications. We demonstrate that even small surface roughness 共⌬r ⱗ 兾50兲 can lead to a drastic degradation of high-Q cavity modes by many orders of magnitude. The results of the numerical simulation are analyzed and explained in terms of wave reflection at a curved dielectric interface, combined with an examination of Poincaré surfaces of section and of Husimi distributions. © 2004 Optical Society of America OCIS codes: 140.4780, 140.3410, 290.4020, 290.5880, 230.5750. 1. Introduction Dielectric and polymeric microcavities have great potential for possible applications in lasing optoelectronic devices.1,2 In conventional lasers a significant fraction of optical pump power is lost and a rather high threshold power is needed to initiate the lasing effect. In contrast, spherical and disk cavities can be used to support highly efficient low-threshold lasing operations. The high efficiency of such devices is related to the existence of natural cavity resonances. These resonances are known as morphologydependent resonances or whispering-gallery modes.3 The nature of these resonances can be envisioned in a ray optic picture in which light is trapped inside the cavity through the total internal reflection on the cavity–air boundary. In dielectric cavities, optically pumped quantum wells, wires, or dots provide an active medium that sustains the lasing operation.4 – 8 Polymeric microcavity lasers are made with an active medium that includes host and guest molecules.9 –11 The absorbed light is transferred from the photoexcited host molecules in a nonradiative way by means of resonant energy transfer to the guest molecules. A stim- The authors are with the Department of Science and Technology, Linköping University, 601 74 Norrköping, Sweden. I. V. Zozoulenko’s e-mail address is [email protected] Received 1 July 2003; revised manuscript received 6 October 2003; accepted 17 October 2003. 0003-6935兾04兾081761-12$15.00兾0 © 2004 Optical Society of America ulated emission from the active medium of dielectric and polymeric cavities is trapped in high-Q modes for a very long time. This leads to a significant increase of radiation intensity inside the cavity and hence to low-threshold laser operation. One of the most important characteristics of cavity resonances is their quality factor 共Q factor兲 defined as Q ⫽ 2*共stored energy兲兾共energy lost per cycle兲. The high value of the Q factor results from very low radiative losses that are caused mainly by radiation leakage due to diffraction on the curved interface. Typical experimental values of the Q factors of dielectric or polymeric disk cavities reported so far are in the range of ⬃103–104 or lower4 –11 共the characteristic diameter of the lasing disk cavities is 5–20 m兲. At the same time, a theoretical estimation of the corresponding Q factor of an ideal circular disk cavity of a representative diameter d ⬃ 10 m for a typical cavity mode 共usually corresponding to the highest whispering gallery resonances for a given cavity radius兲 gives Q ⬃ 1013 关see Eqs. 共26兲 and 共27兲 below兴. A degradation of the experimental Q factors may be attributed to a variety of factors, including imperfections in sidewall geometry, inhomogeneity of the disk’s diffraction index, and effects of coupling to the substrate or pedestal to name a few. A detailed study of the effects of the above factors on the characteristics and performance of the microcavity lasers appears to be of crucial importance for the design, tailoring, and optimization of Q values of lasing microdisk cavities. Such the studies would require an effective computational method that can deal with 10 March 2004 兾 Vol. 43, No. 8 兾 APPLIED OPTICS 1761 both complex geometry and variable refraction index in the cavity. One of the most powerful and versatile numerical techniques often used in photonic simulation is the finite-difference time domain method.12–14 A severe disadvantage of this technique in its application to cavities with small surface imperfections is that the smooth geometry of the cavity has to be mapped into a discrete grid with a very small lattice constant. This makes the application of this method to the problem at hand rather impractical in terms of both computational power and memory. Another class of computational methods reduces the Helmholtz equation in the infinite twodimensional space into contour integral equations defined at the cavity boundaries. These methods include the T-matrix technique,15,16 the boundary integral methods,17,18 and others.19 These methods are computationally effective and capable of dealing with cavities with arbitrary geometries. However, the above methods require that the refraction index be constant inside the cavity boundary. In the present paper we develop a new, computationally effective, and numerically stable approach based on the scattering matrix technique that is capable of dealing with both arbitrary complex geometry and inhomogeneous refraction index inside the cavity. Note that the scattering matrix technique is widely used in the analysis of waveguides20 as well as in quantum mechanical simulations.21 This technique was also used in the analysis of resonant cavities with geometries for which the analytical solution was available.22 The main idea of the method consists of dividing the cavity region into N narrow concentric rings. At each ith boundary between the neighboring rings, we calculate the scattering matrix Si that relates the states propagating 共or decaying兲 toward the boundary to those propagating 共or decaying兲 away from the boundary. Successively combining the scattering matrices for all the boundaries,20,21 S1 R . . . R SN, we eventually relate the combined matrix to the total scattering matrix of the cavity S. To calculate the lifetime of the cavity modes 共and therefore their Q factor兲, we compute the Wigner–Smith lifetime matrix23 which, in turn, is expressed in terms of the total scattering matrix S 共see Refs. 2, 23, and 24兲. Because we combine only two scattering matrixes at each step, it is not necessary to keep track of the solution in the whole space. This obviously eliminates the need for storing large matrices and facilitates the computational speed. It is also well known that the scattering matrix technique 共in contrast, for example, with the transfer matrix technique兲 is not plagued by numerical instability because exponentially growing and decaying evanescent waves are separated in the course of the computation. It should also be stressed that the scattering matrix technique is unconditionally stable; i.e., its stability does not depend on grid size. This is in contrast to other techniques, such as the finite-difference time domain method,12–14 which is stable only when cer1762 APPLIED OPTICS 兾 Vol. 43, No. 8 兾 10 March 2004 tain conditions that are imposed on steps in time and space are satisfied. Note that the present technique of combining S matrices is conceptually similar to the recurrence algorithm for calculating electromagnetic scattering from a multilayered sphere.25,26 However, in contrast to these works, the scattering matrix technique presented here can be applied to systems in which the refraction index varies as a function of both radial and angular coordinates. The paper is organized as follows. In Subsection 2.A we develop the scattering matrix technique for disk-shaped cavities. The results of the numerical calculations of resonant states and of Q values of nonideal cavities are made based on the developed technique and are presented in Section 3. We consider and compare two cases: a disk cavity of a constant refraction index n with a sidewall imperfection 共surface roughness兲 and a disk cavity of an ideal circular shape but with an inhomogeneous refraction index n ⫽ n共r, 兲. The results of the numerical simulation are analyzed and explained in terms of wave reflection at a curved dielectric interface, combined with the examination of Poincaré surfaces of section and of Husimi function. Finally, we present our conclusion in Section 4. 2. Scattering Matrix Approach A. Formalism We consider a two-dimensional cavity with the refraction index n surrounded by air. Because the majority of experiments are performed only with the lowest transverse mode occupied, we neglect the transverse 共z兲 dependence of the field and thus limit ourselves to the two-dimensional Helmholtz equation. The two-dimensional Helmholtz equation for z components of the electromagnetic field for the case of a slowly varying refraction index n is given by 冉 冊 2 1 1 2 ⫹ 2 2 ⌿共r, 兲 ⫹ 共kn兲 2⌿共r, 兲 ⫽ 0, 2⫹ r r r r (1) where ⌿ ⫽ Ez 共Hz兲 for TM 共TE兲 modes and k is the wave vector in vacuum. Remaining components of the electromagnetic field can be derived from Ez 共Hz兲 in a standard way. We divide our system into three regions: the outer region, r ⬎ R, the inner region, r ⬍ d, and the intermediate region, d ⬍ r ⬍ R 共see Fig. 1兲. We choose R and d in such a way that the outer- and inner-region refraction indices are independent of the coordinate, whereas in the intermediate region, n is a function of both r and . In the outer region the solution to the Helmholtz equation can be written in the form ⫹⬁ ⌿ out ⫽ 兺 q⫽⫺⬁ 共1兲 关A q H 共2兲 q 共kr兲 ⫹ B q H q 共kr兲兴exp共iq兲, (2) Fig. 2. Intermediate region divided by N concentric rings of width 2⌬; i is the distance to the middle of the ith ring. States ai, ai⫹1 propagate 共or decay兲 toward the ith boundary, whereas states bi, bi⫹1 propagate 共or decay兲 away from this boundary. The ith boundary is defined as the boundary between the ith and the i ⫹ 1th rings. Fig. 1. Schematic geometry of a cavity with the refraction index n surrounded by air. The space is divided in three regions. In the inner 共r ⬍ d兲 and outer regions 共r ⬎ R兲, the refraction indices are constant. In the intermediate region d ⬍ r ⬍ R, the refraction index n is a function of both r and . The intermediate region is divided by N narrow concentric rings. In each ring the refraction coefficient is regarded as a function only of the angle and is given as ni ⫽ ni 共兲. 共2兲 where H共1兲 q , Hq are the Hankel functions of the first and second kind of the order q that describe incoming and outgoing waves, respectively. We define the scattering matrix S in the standard fashion20,21: B ⫽ SA, (3) where A, B are the column vectors composed of the expansion coefficients Aq, Bq in Eq. 共2兲. The matrix element Sq⬘q ⫽ 共S兲q⬘q gives a probability amplitude of the scattering from the incoming state q into the outgoing state q⬘. Because of the requirement of the flux conservation, the scattering matrix is unitary21: SS ⫽ I, † (4) where I is the identity matrix. The time-reversal invariance imposes the symmetry requirement upon the scattering matrix21 S q⬘q ⫽ S qq⬘. (5) These two conditions can be used to control the numerical results for the scattering matrix. To apply the scattering matrix technique, we divide the intermediate region into N narrow concentric rings 共see Figs. 1 and 2兲. Within each ith ring we write down the solution to the Helmholtz equation as a linear superposition of the states propagating 共or decaying兲 away from the disk center and the states propagating 共or decaying兲 toward the disk center. 关The detailed form of these states will be given in Subsection 2.B; see Eq. 共20兲兴. At each ith boundary 共defined as the boundary between the ith and the i ⫹ 1th rings兲, we can introduce the scattering matrix Si that relates the states propagating 共or decaying兲 toward the boundary, 兵aim其 and 兵ai⫹1 m 其, to those propagating 共or decaying兲 away from the boundary, 兵bim其 and 兵bi⫹1 m 其: 冉 冊 冉 冊 bi ai ⫽ Si i⫹1 , b i⫹1 a 1 ⱕ i ⱕ N ⫺ 1, (6) where ai, bi are the column vectors composed of the expansion coefficients 兵aim其, 兵bim其 关see Eq. 共20兲 below兴. For the Nth boundary between the last Nth ring and the outer region, the scattering matrix SN is defined in the form 冉 冊 冉 冊 bN aN ⫽ SN . B A (7) In the inner region 共i ⫽ 0兲 the solution to the Helmholtz equation has the form ⫹⬁ ⌿ in ⫽ 兺 a 0qJ q共nkr兲exp共iq兲, (8) q⫽⫺⬁ where Jq is the Bessel functions of the order q. For the inner boundary 共i ⫽ 0兲 between the inner region and the first ring in the intermediate region, we define the matrix S0 according to 冉冊 冉冊 a0 a0 ⫽ S0 1 . b1 a (9) The brief outline of the derivation and the expressions for the scattering matrices Si are given in Subsection 2.C and in Appendix A. The essence of the scattering matrix technique is the successive combination of the scattering matrices 10 March 2004 兾 Vol. 43, No. 8 兾 APPLIED OPTICS 1763 in the neighboring regions. For example, combining the scattering matrices for the ith and i ⫹ 1th boundaries, Si and Si⫹1, we obtain the combined scattering matrix S̃i,i⫹1 ⫽ Si R Si⫹1 that relates the outgoing and incoming states in the rings i and i ⫹ 2,20,21: 冉 冊 冉 冊 bi ai ⫽ S̃i,i⫹1 i⫹2 , b i⫹2 a i,i⫹1 11 i i⫹1 i i⫹1 ⫺1 i ⫽ S ⫹ S12 S11 共I ⫺ S22 S11 兲 S21, i,i⫹1 12 i i⫹1 i ⫺1 i⫹1 ⫽ S12 共I ⫺ S11 S22兲 S12 , S̃ S̃ where exp共i兲 ⫽ are the eigenvalues of the scatN is the total phase of the tering matrix S, ⫽ ¥⫽1 M ⫽ determinant of the matrix S, det S ⫽ 兿⫽1 exp共i兲. The resonant states are manifested as peaks in the delay time, whose positions determine the resonant wave vectors kres, and the heights are related to the Q value of the cavity according to i 11 Q ⫽ D共k res兲. (16) i,i⫹1 i⫹1 i i⫹1 ⫺1 i S̃21 ⫽ S21 共I ⫺ S22 S11 兲 S21, i,i⫹1 i⫹1 i⫹1 i i⫹1 ⫺1 i i⫹1 S̃22 ⫽ S22 ⫹ S21 共I ⫺ S22 S11 兲 S22S12 . (10) Here and hereafter we use the notation S11, S12, . . . to define the respective matrix elements of the block matrix S. Combining step by step all the scattering matrices for all the boundaries 0 ⱕ i ⱕ N, we numerically obtain the total combined matrix S̃0,N ⫽ S0 R S1 R . . . SN relating the scattering states in the outer region 共i ⫽ N兲 and the states in the inner region 共i ⫽ 0兲, 冉冊 冉冊 a a ⫽ S̃0,N . B A (11) To obtain the scattering matrix S defined by Eq. 共3兲, we eliminate a from Eq. 共11兲 and find the relation between S̃0,N and S, 0,N 0,N ⫺1 0,N 0,N S ⫽ S̃21 共I ⫺ S̃11 兲 S̃12 ⫹ S̃22 . (12) To identify the resonant states of an open cavity, we introduce the lifetime matrix 共often called as Wigner–Smith time-delay matrix兲23 dS i dS† i Q⫽ S ⫽ ⫺ S† . c dk c dk (13) The diagonal elements of this matrix give the time delay experienced by the wave incident in the qth channel and scattered into all other channels: qD共k兲 ⫽ Qqq ⫽ i c 兺 q⬘ dS†qq⬘ Sq⬘q. dk (14) The delay time qD共k兲 experienced by a scattering wave is totally equivalent to the lifetime ⫽ 1兾2ck⬙ of a quasibound state with complex eigenvector k ⫽ k⬘ ⫺ ik⬙2. It is interesting to note that, in his original paper dealing with quantum mechanical scattering,23 Smith chose a letter “Q” to define the lifetime matrix of a quantum system because of a close analogy to the definition of a Q value in electromagnetic theory. The total time delay averaged over all M incoming channels can be expressed in the form2,24 D共k兲 ⫽ ⫽ 1764 1 M M 兺 q⫽1 1 cM M 兺 ⫽1 q D 共k兲 ⫽ 冉 冊 dS† 1 i Tr S Mc dk d 1 d ⫽ , dk cM dk APPLIED OPTICS 兾 Vol. 43, No. 8 兾 10 March 2004 (15) B. Calculation of the Wave Functions in the Intermediate Region d ⬍ r ⬍ R In the intermediate region the refraction index n depends on both r and . Therefore, in contrast to the inner and outer regions, in the intermediate region we cannot separate variables and find an exact analytical solution to the Helmholtz equation. We can, however, find an approximate solution to the Helmholtz equation in each ring. To do this, let us look for the solution in the form ⌿共r, 兲 ⫽ R共r兲⌽共兲. Substituting this solution into Eq. 共1兲, we obtain r R共r兲 1 2⌽共兲 r 2 2R共r兲 ⫹ ⫽⫺ 2 R共r兲 r R共r兲 r ⌽共兲 2 ⫺ k 2n 2共r, 兲r 2. (17) Let us now assume that each ring with radius i has a vanishing width 2⌬ 3 0 共see Fig. 2兲. In this case we can regard r as a constant within each ith ring, r ⬇ i , with the refraction index being a function only of the angle, n共r, 兲 ⫽ ni 共兲. In this approximation the variables in Eq. 共17兲 separate such that for ith ring we can write 2⌽ i共兲 ⫹ 共 i ⫹ k 2i 共兲 2i 兲⌽ i共兲 ⫽ 0, 2 (18) 2R i共r i 兲 R i共r i 兲 ⫹ ⫺ iR i共r i 兲 ⫽ 0, r 2i r i (19) where i is a constant 共which can be both positive and negative兲 and ri ⫽ r兾i . The angular function ⌽i共兲 satisfies the cyclic boundary condition ⌽i共0兲 ⫽ ⌽i共2兲. The solution of Eq. 共18兲 thus provides an infinite set of eigenvalues 兵im其 with the corresponding eigenfunctions ⌽im共兲. Generally, Eq. 共18兲 must be solved numerically. For a given eigenvalue im, the solution of Eq. 共19兲 for the radial wave function can be easily written in analytical form, and the approximate solution to the Helmholtz equation in the ith ring 共situated to the left of ith boundary兲 reads ⬁ ⌿ i 共r i , 兲 ⫽ 兺 兵a i m Si ⫽ ⌳KA⫺1BK⌳⫺1. exp关共⫺1兾2 ⫹ i␥ im兲r̃ i 兴 m⫽1 (20) where r̃i ⫽ 共r ⫺ i 兲兾i and ␥im ⫽ 共⫺1兾4 ⫺ im兲1兾2. The states in Eq. 共20兲 are grouped according to the convention adopted in Subsection 2.A. Namely, the states propagating to the right toward the ith boundary 关exp共⫹ i␥imr̃i 兲兴 are described by the coefficients 兵am其, whereas the states propagating away from the ith boundary 关exp共⫺i␥imr̃i 兲兴 enter with the coefficients 兵bm其. Note that if ␥im becomes imaginary, ␥ ⫽ i, the state propagating toward 共away from兲 the ith boundary turns into the states decaying toward 共away from兲 this boundary. The wave function ⌿i⫹1共ri⫹1, 兲 in the i ⫹ 1th ring 共situated to the right of the ith boundary兲 is given by a similar expression with coefficients am and bm interchanged: ⬁ 兺 兵b i⫹1 m exp关共⫺1兾2 ⫹ i␥ i⫹1 m 兲r̃ i⫹1兴 m⫽1 ⫹ a i⫹1 m exp关共⫺1兾2 i⫹1 ⫺ i␥ i⫹1 m 兲r̃ i⫹1兴其⌽ m 共兲. (21) This is because in the i ⫹ 1th ring, the states exp共⫹i␥i⫹1 m r̃i⫹1兲 propagate 共or decay兲 away from the ith boundary, whereas the states exp共⫺␥i⫹1 m r̃i⫹1兲 propagate 共or decay兲 toward the ith boundary. C. Scattering Matrix Si at the ith Boundary In this section we derive the expression for the scattering matrix Si by matching the wave functions across the ith boundary. Using the condition of the continuity of the tangential components of the electric and magnetic fields at the boundary between two dielectric media, the matching conditions at the ith boundary 共i.e., at the boundary between the ith and the i ⫹ 1th rings兲 read ⌿ i 共r, 兲 ⫽ ⌿ i⫹1共r, 兲, 1 ⌿ i 共r, 兲 1 ⌿ i⫹1共r, 兲 ⫽ 2 , 2i 共兲 r i⫹1共兲 r (22) where 2i 共兲 ⫽ 1 for TM modes and 2i 共兲 ⫽ k2n2i 共兲 for TE modes. To derive the expression for the scattering matrix Si in the intermediate region 共1 ⱕ i ⱕ N ⫺ 1兲, we substitute the wave functions of Eqs. 共20兲 and 共21兲 into the boundary conditions of Eq. 共22兲. Multiplying the obtained equations by 关⌽im共兲兴* and integrating over the angle using the conditions of the i i orthogonality 兰2 0 d关⌽m共兲兴*⌽m⬘共兲 ⫽ ␦mm⬘, we arrive at two infinite systems of equations for the coeffii i⫹1 cients aim, ai⫹1 After some straightm , bm, and bm . forward algebra these systems of equations are (23) The scattering matrixes S , S 共for inner i ⫽ 0 and outer i ⫽ N boundaries, respectively兲 are derived in a similar fashion. The expression for Si given by Eq. 共23兲 holds for all the boundaries 0 ⱕ i ⱕ N. A particular form of the matrices ⌳, K, A, B is different for three distinct cases: 共a兲 0th boundary 共the boundary between the inner region 共i ⫽ 0兲 and the first ring i ⫽ 1 in the intermediate region兲; 共b兲 ith boundary, 1 ⬍ i ⬍ N ⫺ 1 共the boundary between the ith and the i ⫹ 1th rings in the intermediate region兲, and 共c兲 Nth boundary 共the boundary between the last ring i ⫽ N in the intermediate region and the outer region i ⫽ N ⫹ 1兲. The corresponding expressions for these three cases are given in Appendix A, Eqs. 共A1兲–共A3兲. 0 ⫹ b im exp关共⫺1兾2 ⫺ i␥ im兲r̃ i 兴其⌽ im共兲, ⌿ i⫹1共r i⫹1, 兲 ⫽ reduced to the form prescribed by Eq. 共6兲 with the following result: N 3. Nonideal Microdisk Cavities In this section we apply the scattering matrix method to the calculation of resonant states and Q values of nonideal microdisk cavities with 共a兲 sidewall imperfections and 共b兲 circular cavities with an inhomogeneous refraction index of n ⫽ n共r, 兲. The scattering matrix of the disk S defined by Eq. 共3兲 has an infinite dimension. To perform the numerical calculations, we truncate the matrix S to the size M ⫻ M, where M ⫽ Mprop ⫹ Mevs, with Mprop and Mevs being the number of the propagating and evanescent waves, respectively, in Eq. 共2兲. The number of the propagating solutions Mprop equals the number of modes in the corresponding closed circular cavity. This number scales with nkR because it is given by the number of real eigenvalues of the equations Jq共nkR兲 ⫽ 0 共TM modes兲 and J⬘q共nkR兲 ⫽ 0 共TE modes兲 that define the eigenmodes of the closed circular cavity. The number of evanescent modes is chosen in such a way that the calculated Q value does not change with further increase of Mevs. For the dielectric cavities studied in the present paper, it is sufficient to choose Mevs ⬃ 5 ⫺ 10 to achieve the required accuracy 共⬍0.5%兲. The choice of the number of the rings N in the intermediate region depends strongly on the character of the imperfections and the extent of the inhomogeneity present in the system. As a result, it is not possible to provide a universal receipt for the choice of N that is suitable for all systems. In our calculations we choose N based on the requirement that the results of the numerical calculations do not change when N is increased 共i.e., when the ring width ⌬R is reduced兲. For example, for the case of cavities with sidewall imperfections considered here, a sufficient accuracy 共⬍0.5%兲 was achieved for N ⫽ ⌬r兾20; for cavities with the refractive index inhomogeneity, N ⫽ R兾40. To validate the present method, we perform numerical calculations for structures for which the analytical solution was available. This includes, for example, an annular billiard consisting of a dielectric 10 March 2004 兾 Vol. 43, No. 8 兾 APPLIED OPTICS 1765 disk placed inside a larger disk with some displacement of the disk center,22 as well as an ideal circular disk displaced from the origin of the coordinate system. In the latter case, the positions of the resonant states and Q values are obviously independent of the choice of the coordinate system. However, from a computational point of view, this case is no simpler than that of a cavity with an arbitrary shape because the displacement from the origin lifts the radial symmetry and makes the separation of variables impossible. As an additional tool for validating the numerical solution, we use Eqs. 共4兲 and 共5兲 to control the unitarity and symmetry of the scattering matrix. A. Ideal Circular Cavity Let us first briefly analyze the resonant states and Q values of an ideal circular cavity with radius R and refraction index n. In this case the scattering matrix can easily be written in analytical form. Employing the matching conditions of Eq. 共22兲 between the wave function in the outer region r ⬎ R of Eq. 共2兲 and the wave function inside the disk given by the Bessel function Jq共nkr兲 for r ⬍ R of Eq. 共8兲, we derive the expression for the scattering matrix in the form22 S qq⬘ ⫽ 共2兲 H 共2兲⬘ q 共kr兲 ⫺ 关 J⬘q共nkr兲兾J q共nkr兲兴H q 共kr兲 ␦ qq⬘, 共1兲 H 共1兲⬘ 共kr兲 ⫺ 关 J⬘ 共nkr兲兾J 共nkr兲兴H q q q q 共kr兲 (24) Fig. 3. Transmission coefficient T of a locally plane wave incident on a curved surface with the radii of curvature as a function of the incidence angle calculated from Eq. 共26兲. The angle of total internal reflection sin c ⫽ 0.56 共corresponding to n ⫽ 1.8兲. The inset shows the dependence of the average radius of local curvature due to boundary imperfections, , subject to ⌬r for the present model of surface roughness. length, nk ⬎⬎ 1, 共which applies to the majority of cavities兲, the transmission probability reads27 冋 T ⫽ 兩T F兩exp ⫺ where ⫽ n共1兾n兲 for TM 共TE兲 modes. Derivatives are taken over the full arguments in the brackets. Resonant states of an ideal cavity can be inferred from the scattering matrix in Eq. 共24兲 by use of Eq. 共15兲. Each resonant state of an open disk is characterized by two wave numbers, q and m. These two numbers are directly related to the corresponding numbers of the closed resonator of the same radius R. The index m is a radial wave number, and it is related to the number of nodes of the field components in the radial direction inside the disk. The index q is called an angular 共or azimuthal兲 wave number because of the analogy to quantum mechanics in which the angular momentum is given by LQM ⫽ បq. Equating the quantum and classical angular momenta 共LClas ⫽ pR sin , p ⫽ បnk兲, we find the relation between the angular wave number and the angle of incidence in a classical ray picture22: q ⫽ nkR sin . (25) Here we are mostly interested in the whisperinggallery modes with high Q values for which the angle of incidence is larger than the angle of total internal reflection, ⬎ c 共sin c ⫽ 1兾n兲. For such angles of incidence, the transmission probability T of an electromagnetic wave incident on a curved interface of radius is small, T ⬍⬍ 1. For the case when the radius of curvature is much larger than the wave1766 APPLIED OPTICS 兾 Vol. 43, No. 8 兾 10 March 2004 册 2 nk 共cos2 c ⫺ cos2 兲 3兾2 , 3 sin2共兲 (26) where TF is the classical Fresnel transmission coefficient for an electromagnetic wave incident on a flat surface. Figure 3 illustrates that T decreases exponentially as the difference ⫺ c grows. The Q value of the whispering-gallery mode q in a cavity of radius R is related to the transmission probability T in Eq. 共26兲 by the relation28 Q⫽ 2nkR cos , T (27) where the classical incidence angle is related to mode number q by Eq. 共25兲 and T ⬍⬍ 1. B. Nonideal Cavities with Surface Roughness and Inhomogeneous Refraction Index In this section we present the results of the numerical calculations of resonant states and of Q values of nonideal cavities. We consider separately two cases: 共1兲 a disk cavity of a constant refraction index n but with a sidewall imperfection 共surface roughness兲, 共2兲 a disk cavity with an ideal circular shape but with an inhomogeneous refraction index n ⫽ n共r, 兲. Various studies indicate that a typical sidewall imperfection can vary in size from 5–300 nm 共representing a variation of the order of ⬃0.05–1% of the cavity radius兲.6 – 8,11 An exact experimental shape of the cavity–air interface is, however, not available. We thus model the interface shape as a superposition of random Gaussian deviations from an ideal circle of Fig. 4. Examples of nonideal cavities studied in the present paper in terms of 共a兲 surface roughness and 共b兲 inhomogeneous refraction index. 共a兲 Radius of the disk R ⫽ 5 m, n ⫽ 1.8, surface roughness ⌬r ⫽ 100 nm. 共b兲 R ⫽ 5 m, 具n典 ⫽ 1.8, ⌬n ⫽ 5%. radius R with a maximal amplitude ⌬r兾2 and a characteristic distance between the deviation maxima ⌬l ⬃ 2R兾50. In a similar fashion we model the inhomogeneity of the diffraction index in the cavity, where a parameter ⌬n characterizes a mean deviation of the refraction index n from its average value 具n典 ⫽ 1.8. The variation of the refraction index n can be caused by different factors, including the presence of quantum wells, wires, and dots that form an active medium of the cavity; the local field intensity dependence n ⫽ n共I兲, and other factors. Examples of the typical structures under investigation are shown in Fig. 4. Figure 5 shows the calculated Q values of the disk resonant cavity for different surface roughnesses ⌬r and the refraction index inhomogeneity ⌬n over some representative wavelength interval. Note that we have studied a number of different resonances, and all of them showed the same trends described below. Because TE modes exhibit similar features, hereafter we concentrate only on TM modes of the cavity. The calculated dependencies of the Q values on ⌬r and ⌬n are summarized in the insets to Fig. 5. Let us first concentrate on the low-Q state TM55,7 共q ⫽ 55, m ⫽ 7兲. An increase in both the surface roughness ⌬r and the refractive index inhomogeneity ⌬n causes a graduate and rather slow decrease in the Q value of this state, as shown in the insets to Fig. 5. This behavior is typical for all other low-Q states. In contrast, the high-Q resonances exhibit very different and rather striking behavior. Namely, these resonances show a dramatic decrease of their Q-values even for very small values of the surface roughness ⌬r ⱗ 兾50. At the same time, the Q values of the cavity decrease much more slowly when the refractive index inhomogeneity ⌬n increases. For example, let us choose ⌬r ⫽ 20 nm and ⌬n ⫽ 5%. For these values of ⌬r and ⌬n, the Q value of the low-Q state TM55,7 drops by the same factor of ⬃1.3, decreasing from Q ⬇ 270 to Q ⬇ 205. In contrast, for the very same surface roughness ⌬r, the Q value of a high-Q state TM82,1 drops by the factor of ⬃1011, decreasing from its value of Q ⬇ 4 ⫻ 1013 for an ideal disk to Q ⬇ 260. At the same time, for the above value of ⌬n ⫽ 5%, the Q value of this resonance decreases to the value of Q ⬇ 1.3 ⫻ 107, which corresponds to the decrease by Fig. 5. Dependencies Q ⫽ Q共兲 for two representative modes TM82,1 and TM55,7 for the cases of 共a兲 different surface roughness ⌬r 共R ⫽ 5 m, n ⫽ 1.8兲 and 共b兲 different refraction index inhomogeneities 关具n典 ⫽ 1.8兴. Note that in case 共b兲 the resonances shift when ⌬n varies. For the sake of clarity, we plot all the resonances centered around their maxima of the corresponding ideal disk 共i.e., ⌬n ⫽ 0兲. The broadening of the high-Q resonance TM82,1 is not discernible on the scale of the figure for all the values of ⌬n. Insets in 共a兲 and 共b兲 show the dependencies Q ⫽ Q共⌬r兲 and Q ⫽ Q共⌬n兲, respectively. 10 March 2004 兾 Vol. 43, No. 8 兾 APPLIED OPTICS 1767 Fig. 6. Poincaré surfaces of section for geometrical rays corresponding to the states q ⫽ 55 共a兲–共c兲 and q ⫽ 82 共g兲–共i兲 for the cavity with surface roughness ⌬r ⫽ 0, 20, 100 nm. The Husimi distributions for the states TM55,7 共d兲–共f 兲 and TM82,1 共j兲–共l兲 for the same values of ⌬r used in the corresponding Poincaré SoS. ⌬ch indicates the broadening of the phase space due to the transition to the chaotic dynamics. Dashed lines show the angle of total internal reflection c. a factor of ⬃104. 共Note that for the case of an ideal cavity, the high-Q resonances are so narrow that the numerical resolution does not allow a reliable estimation of their exact values. In this case we therefore use Eqs. 共26兲 and 共27兲 to estimate their Q values.兲 C. Discussion In the Subsection 2.B we found that the surface roughness ⌬r and the refraction index inhomogeneity ⌬n that produce a similar degradation of low-Q states have strikingly different effects on high-Q resonances. To understand these effects, we combine Poincaré surface of section and Husimi function methods with an analysis of ray reflection at a curved dielectric interface. The Poincaré surfaces of section 共SoS兲 is a powerful tool for visualizing the phase space for classical ray dynamics in cavities.2,29 We concentrate on the surface section of the phase space along the cavity boundary, r 僆 surf. For a given resonant state with an angular number q, the corresponding ray is launched with the angle 0 ⫽ arcsin共q兾nkR兲 according to Eq. 共25兲. Each reflection at the boundary 共characterized by the polar angle and the angle of incidence 兲, corresponds to a single point in the plot. The number of bounces for a given angle of incidence 0 is chosen in such a way that the total path of the ray does not exceed that extracted from the numerically calculated Q value for the corresponding resonance L ⫽ cD ⫽ Q兾kn. Figures 6共a兲– 6共c兲 and 6共g兲– 6共i兲 show a Poincaré SoS for the geometrical rays corresponding to the states TM55,7 and TM82,1 for different values of the surface roughness ⌬r. For an ideal circular disk 共⌬r ⫽ 0兲, the Poincaré SoS are obviously straight lines correspond1768 APPLIED OPTICS 兾 Vol. 43, No. 8 兾 10 March 2004 ing to a constant angle of incidence ⫽ 0. Figures 6共b兲– 6共c兲, 6共h兲 and 6共i兲 demonstrate that the dynamics of an ideal cavity transform from regular to chaotic, even for a cavity with a maximum roughness ⌬r ⱗ 20 nm. In Figs. 6共b兲– 6共c兲, 6共h兲 and 6共i兲, ⌬ch approximately indicates the phase-space broadening due to the transition to the chaotic dynamics. An important observation is that, for a given surface roughness ⌬r, the broadening of the phase space is independent of the angular mode q; i.e., it is the same for low- and high-Q states. We complement classical Poincaré SoS with Husimi function analysis.2,29,30 The Husimi function 共often called Husimi distributions兲 H共, 兲 represents a quantum 共wave兲 analog to a classical Poincaré SoS. It is defined as a projection of a given cavity mode ⌿共r 僆 surf, 兲 共where “surf” denotes surface兲 taken at the surface of the cavity into a Gaussian wave packet ⌽共⬘; , 兲 impinging the cavity boundary with the coordinate at the angle : H共, 兲 ⫽ 兰 2 d⬘⌿共r 僆 surf, ⬘兲⌽共⬘; , 兲, (28) 0 where the minimum-uncertainty wave packet centered around , with the dispersion in position 共兾 2兲1兾2 is given by ⌽共⬘; , 兲 ⫽ 共兲 ⫺1兾4 兺 exp关⫺1兾2共⬘ ⫺ l ⫹ 2l 兲 2 ⫺ ik sin 共 ⫹ 2l 兲兴, (29) where we have chosen ⫽ 公2兾k. The Husimi distributions 关see Figs. 6共d兲– 6共f 兲 and 6共j兲– 6共l兲兴 exhibit the same trends as the classical Poincaré SoS. In- Fig. 7. Illustrative examples of intensity distribution Ez for the resonant state TM55,7 in cavities with ⌬r ⫽ 0 共a兲, 20 nm 共b兲, 100 nm 共c兲 and with R ⫽ 5 m, n ⫽ 1.8. Dashed lines indicate boundaries of the cavity. deed, broadening the phase space while increasing the surface roughness ⌬r for the Husimi functions has the same effect as the corresponding broadening of ⌬ch in the Poincaré SoS. 共Illustrative examples of the wave functions in cavities with different surface roughness ⌬n are shown in Fig. 7.兲 Figure 8 shows the Husimi distributions for a circular cavity with an inhomogeneous refraction index. The variation of the refraction index ⌬n ⫽ 5% is chosen in such a way that the degradation of the Q value for the low-Q resonance TM55,7 is the same as that for the surface roughness ⌬r ⫽ 20 nm case shown in Fig. 6. As expected, the broadening of the Husimi distribution due to the increase in ⌬n is of the same order as that for the corresponding values of ⌬r 共compare Figs. 6 and 8兲. According to Eq. 共26兲, one can expect an increase in the transmission coefficient 共and therefore a decrease in the Q value of the cavity兲 due to the broadening of the phase space ⌬ch. This is because the incidence angle effectively moves closer to the angle of the total internal reflection c. ⌬Tch in Fig. 3 indicates that the estimated increase in the transmission coefficient is due to the broadening of the phase space ⌬ch as extracted from the Poincaré SoS for ⌬r ⫽ 20 nm and ⌬n ⫽ 5%. For the low-Q resonance TM55,7, this corresponds to the decrease in the Q value by a factor of ⌬Qch ⬃ ⌬T⫺1 ch ⬇ 1.5, which is consistent with the calculated decrease in the low-Q resonances. For the case of high-Q resonance TM82.1, the estimated decrease in the Q factor is ⌬Qch ⬃ ⌬T⫺1 ch ⬇ 103–104 共see Fig. 3兲, which is consistent with the Fig. 8. The Husimi distributions for the states TM55,7 共a兲 and TM82,1 共b兲 for the cavity with the refraction index inhomogeneity ⌬n ⫽ 5%. calculated decrease in this resonance for the case of the inhomogeneous refraction index only. 共Note that because of a rather approximate definition of ⌬ch we can give only a very rough estimation of the factor ⌬Tch.兲 In contrast, for the case of high-Q resonances in the presence of surface imperfections, this estimated value of ⌬Qch is many orders of magnitude smaller than the actual calculated decrease in the Q factor 共given by a factor of ⬇1011; see Fig. 5兲. To explain the rapid degradation of high-Q resonances, we concentrate on another aspect of the wave dynamics. Namely, the imperfections at the surface boundary effectively introduce a local radius of a surface curvature that is distinct from the disk radius R 关see Fig. 4共a兲兴. One may thus expect that, with the presence of the local surface curvature, the total transmission coefficient is determined by the averaged value of rather than by the disk radius R. The dependence of on surface roughness ⌬r for the present model of surface imperfections is shown in the inset to Fig. 3. Figure 3 demonstrates that the reduction of the local radius of curvature from 5 m 共ideal disk兲 to 1.7 m 共⌬r ⫽ 20 nm兲 causes an increase in the transmission coefficient by ⌬Tcur ⬇ 108. This estimation, combined with the estimation based on the change of ⌬Tch is fully consistent with the actual computed decrease of the Q factor shown in Fig. 5. Note that the estimation of the transition coefficient T based on Eq. 共26兲 is justified for nk ⬎⬎ 1. This condition is satisfied for a wide range of the surface roughness ⌬r. For example, ⌬r ⫽ 100 nm 共i.e., the surface roughness when high-Q resonances are already strongly suppressed兲 corresponds to the effective local curvature ⬇ R兾10, i.e., over the considered wavelength interval nk ⬇ 10. We thus conclude that the main mechanism responsible for the rapid degradation of high-Q resonances in nonideal cavities is the enhanced radiative decay through the curved surface because the effective local radius 共given by the surface roughness兲 is smaller than the disk radius R. In contrast, the degradation of low-Q resonances 共as well as the high-Q resonances with inhomogeneous refraction indices兲 is related mostly to the broadening of the phase space caused by the transi10 March 2004 兾 Vol. 43, No. 8 兾 APPLIED OPTICS 1769 tion to chaotic dynamics. It should be noted, however, that both factors 共broadening of the phase space and enhancement of the transmission due to a decrease in the effective radius of curvature兲 may play comparable roles in the degradation of the low-Q whispering-gallery resonances in the presence of surface roughness. It is interesting to note that an analogues degradation of high-Q modes was recently found in hexagonal-shaped microcavities, where the modes were strongly influenced by roundings of the corners even when the characteristic length scale 共the local radius of curvature兲 was 1 order of magnitude smaller than the wavelength.31 It is worth mentioning that one often assumes that long-lived high-Q resonances in idealized cavities 共e.g., in ideal disks, hexagons, etc.兲 are not important for potential application in optical communication or laser devices18,22 because of their extremely narrow width. Our simulations demonstrate that this is not the case because in real structures the Q values of these resonances become comparable with those of intermediate-Q resonances for small or moderate surface roughness of ⌬r ⬃ 10–50 nm. 4. Conclusions In this paper we develop a new, computationally effective, and numerically stable approach based on the scattering matrix 共S-matrix兲 technique that is capable of dealing with both arbitrary complex geometry and inhomogeneous refraction index inside a two-dimensional cavity. The derivation is based on the separation of the cavity region into N narrow concentric rings and the calculation of the S matrix at every boundary between the rings. The total S matrix is obtained in a recursive way by successive combination of the scattering matrices for all the boundaries. To calculate the lifetime of the cavity modes 共and therefore their Q factors兲, we compute the Wigner–Smith time-delay matrix, which in turn is expressed in terms of the total scattering matrix. We apply the developed algorithm to the calculation of resonant states and Q values of nonideal microdisk cavities with sidewall imperfections and circular cavities with an inhomogeneous refraction index n ⫽ n共r, 兲. We find that the surface roughness ⌬r and the refraction index inhomogeneity ⌬n, which produce similar degradation of low-Q states, have strikingly different effects on high-Q resonances. In particularly, in the case of inhomogeneous refraction index, the increase in ⌬n causes a rather gradual decrease in the Q value of high-Q resonances. In contrast, in the presence of surface roughness, even small imperfections 共⌬r ⱗ 兾50兲 can lead to a drastic degradation of high-Q cavity modes by many orders of magnitude. To understand these features, we combine Poincaré SoS and Husimi function methods with an analysis of ray reflection at a curved dielectric interface. We argue that the main mechanism responsible for the rapid degradation of high-Q resonances in nonideal cavities with surface roughness is the enhanced 1770 APPLIED OPTICS 兾 Vol. 43, No. 8 兾 10 March 2004 radiative decay through the curved surface. This is because the effective local radius 共given by the surface roughness兲 is smaller than the disk radius R. In contrast, the degradation of low-Q resonances 共as well as high-Q resonances with inhomogeneous refraction indices兲, is related mostly to the broadening of the phase space caused by the transition to chaotic dynamics. Appendix A: Expressions for the Si Matrices In this appendix we present the expressions for the matrices ⌳, K, A, B entering Eq. 共23兲 for the scattering matrix Si relating incoming and outgoing states at the ith boundary. We distinguish three different cases as specified below. Case 1: Zeroth boundary 共the boundary between the inner region 共i ⫽ 0兲 and the first ring i ⫽ 1 in the intermediate region兲, ⌳11 ⫽ I, 共⌳22兲 mj ⫽ exp关⫺共1兾2兲⌬ 1兴␦ mj, ⌳12 ⫽ ⌳21 ⫽ 0, K11 ⫽ I, 共K22兲 mj ⫽ exp共i␥ m⌬ 1兲␦ mj, K12 ⫽ K21 ⫽ 0, A⫽ 冉 冊 0 V0,1 , ⫺J⬘ U0,1P1 共J兲 mj ⫽ J m共n 0 kd兲␦ mj, 共V0,1兲 mj ⫽ 兰 2 B⫽ 冉 冊 J ⫺V0,1 , 0 ⫺U0,1Q1 共J⬘兲 mj ⫽ J⬘m共n 0 kd兲␦ mj, exp共⫺im兲⌽ 1j 共兲d, 0 共U0,1兲 mj ⫽ 1 n 0 k 1 兰 2 0 20共兲 exp共⫺im兲⌽ 1j 共兲d. 21共兲 (A1) Case 2: ith boundary, 1 ⬍ i ⬍ N ⫺ 1 关the boundary between the ith and the i ⫹ 1th rings in the intermediate region兴 共⌳11兲 mj ⫽ exp关共1兾2兲⌬ i 兴␦ mj, 共⌳22兲 mj ⫽ exp关⫺共1兾2兲⌬ i⫹1兴␦ mj, ⌳12 ⫽ ⌳21 ⫽ 0, A⫽ B⫽ 共Vi,i⫹1兲 mj ⫽ 冉 冉 兰 共K兲 mj ⫽ exp共i␥ m⌬ i 兲␦ mj, 冊 冊 ⫺I V , ⫺Qi Ui,i⫹1Pi⫹1 i,i⫹1 I ⫺V , Pi ⫺Ui,i⫹1Qi⫹1 i,i⫹1 2 关⌽ im共兲兴*⌽ i⫹1 j 共兲d, 0 共Ui,i⫹1兲 mj ⫽ i i⫹1 兰 2 0 2i 共兲 关⌽ im共兲兴*⌽ i⫹1 j 共兲d. 2 i⫹1 共兲 (A2) Case 3: Nth boundary 关the boundary between the last ring i ⫽ N in the intermediate region and the outer region 共i ⫽ N ⫹ 1兲兴 共⌳11兲 mj ⫽ exp关共1兾2兲⌬ N兴␦ mj, ⌳ 22 ⫽ I, 6. 7. ⌳ 12 ⫽ ⌳ 21 ⫽ 0, 共K11兲 mj ⫽ exp共i␥ m⌬ N兲␦ mj, K22 ⫽ I, A⫽ B⫽ 冉 冉 8. K12 ⫽ K21 ⫽ 0, 冊 冊 ⫺I VN,N⫹1H共1兲 , ⫺QN UN,N⫹1H共1兲⬘ 9. I ⫺VN,N⫹1H共2兲 , PN ⫺UN,N⫹1H共2兲⬘ 共H共1,2兲兲mj ⫽ H共1,2兲共kR兲␦mj, mj, 共H共1,2兲⬘ 兲 mj ⫽ H 共1,2兲⬘共kR兲␦ m m 共VN,N⫹1兲 mj ⫽ 兰 2 10. 关⌽ Nm共兲兴* exp共ij兲d, 0 共U N,N⫹1 兲 mj ⫽ kN 兰 11. 2 0 2N共兲 关⌽ 1j 共兲兴* exp共ij兲d. 2 N⫹1 共兲 12. (A3) In Eqs. 共A1兲–共A3兲 the matrices Qi, Pi are defined according to 14. 共Pi兲 mj ⫽ 共⫺1兾2 ⫹ i␥ im兲␦ mj, 共Q 兲 mj ⫽ 共⫺1兾2 ⫺ i␥ 兲␦ mj, i Jm, H共1,2兲 m , i m 13. 1 ⱕ i ⱕ N. J⬘m, H共1,2兲⬘ m and are the Bessel and Hankel functions and their derivatives, and ⌬i ⫽ ⌬兾i . We thank Olle Inganäs for stimulating discussions that initiated this work. We are also grateful to Sayan Mukherjee and especially to Stanley Miklavcic for many useful discussions and conversations. We appreciate the correspondence with Jan Wiersig. A. I. Rahachou acknowledges financial support from the Swedish Institute and The Royal Swedish Academy of Sciences. 15. 16. 17. 18. 19. References 1. Y. Yamamoto and R. E. Slusher, “Optical processes in microcavities,” Phys. Today 46, 66 –72 共1993兲. 2. J. U. Nöckel and R. K. Chang, “2-d microcavities: theory and experiments,” in Cavity-Enhanced Spectroscopies, R. D. van Zee and J. P. Looney, eds., Vol. 40 of Experimental Methods in the Physical Sciences 共Academic, San Diego, 2002兲, pp. 185– 226. 3. S. C. Hill and R. E. Benner, “Morphology-dependent resonances,” in Optical Effects Associated with Small Particles, P. W. Barber and R. K. Chang, eds., Vol. 1 of Advanced Series in Applied Physics 共World Scientific, Singapore, 1989兲, pp. 3– 61. 4. S. L. McCall, A. F. J. Levi, R. E. Slusher, S. J. Pearton, and R. A. Logan, “Whispering-gallery mode microdisk lasers,” Appl. Phys. Lett. 60, 289 –291 共1992兲. 5. R. E. Slusher, A. F. J. Levi, U. Mohideen, S. L. McCall, S. J. Pearton, and R. A. Logan, “Threshold characteristics of semi- 20. 21. 22. 23. 24. 25. 26. conductor microdisk lasers,” Appl. Phys. Lett. 63, 1310 –1312 共1992兲. M. Fujita, K. Inoshita, and T. Bata, “Room temperature continuous wave lasing characteristics of GaInAsP兾InP microdisk injection laser,” Electron. Lett. 34, 278 –279 共1998兲. B. Gayral, J. M. Gérard, A. Lemaı̂tre, C. Dupuis, L. Manin, and J. L. Pelouard, “High-Q wet etched GaAS microdisks containing InAs quantum boxes,” Appl. Phys. Lett. 75, 1908 –1910 共1999兲. C. Seassal, X. Letartre, J. Brault, M. Gendry, P. Pottier, P. Viktorovitch, O. Piquet, P. Blondy, D. Cros, and O. Marty, “InAs quantum wires in InP-based microdiscs: mode identification and continuous wave room temperature laser operation,” J. Appl. Phys. 88, 6170 – 6174 共2000兲. A. Dodabalapur, M. Berggren, R. E. Slusher, Z. Bao, A. Timko, P. Schiortino, E. Laskowski, H. E. Katz, and O. Nalamasu, “Resonators and materials for organic lasers based on energy transfer,” IEEE J. Sel. Top. Quantum Electron. 4, 67–74 共1998兲. M. Theander, T. Granlund, D. M. Johanson, A. Ruseckas, V. Sundström, M. R. Andersson, and O. Inganäs, “Lasing in a microcavity with an oriented liquid-crystalline polyfluorene copolymer as active layer,” Adv. Mater. 13, 323–327 共2001兲. R. C. Polson, Z. Vardeny, and D. A. Chinn, “Multiple resonances in microdisk lasers of -conjugated polymers,” Appl. Phys. Lett. 81, 1561–1563 共2002兲. K. S. Yee, “Numerical solution of initial boundary-value problems involving Maxwell’s equations in isotropic media,” IEEE Trans. Antennas Propag. AP-14, 302–307 共1996兲. B.-J. Li and P.-L. Liu, “Analysis of far-field patterns of microdisk resonators by the finite-difference time-domain method,” IEEE J. Quantum Electron. 33, 1489 –1491 共1997兲. M. N. O. Sadiku, Numerical Techniques in Electromagnetics 共CRC Press, Boca Raton, Fla., 2001兲. P. C. Waterman, “Symmetry, unitarity and geometry in electromagnetic scattering,” Phys. Rev. D 3, 825– 839 共1971兲. M. I. Mishchenko, L. D. Travis, and A. A. Lacis, Scattering, Absorption, and Emission of Light by Small Particles, 共Cambridge U. Press, Cambridge, 2002兲. P. A. Knipp and T. L. Reinecke, “Boundary-element method for the calculation of the electronic states in semiconductor nanostructures,” Phys. Rev. B 54, 1880 –1891 共1996兲. J. Wiersig, “Boundary element method for resonances in dielectric microcavities,” J. Opt. A, Pure Appl. Opt. 5, 53– 60 共2003兲. S. V. Boriskina, T. M. Benson, P. Sewell, and A. I. Nosich, “Highly efficient design of spectrally engineered whisperinggallery-mode microlaser resonators,” Opt. Quantum Electron. 35, 545–559 共2003兲. V. V. Nikolsky and T. I. Nikolskaya, Decomposition Approach to the Problems of Electrodynamics 共Nauka, Moscow, 1983; in Russian兲. S. Datta, Electronic Transport in Mesoscopic Systems 共Cambridge U. Press, Cambridge, 1995兲. M. Hentschel and K. Richter, “Quantum chaos in optical systems: the annular billiar,” Phys. Rev. E 66, 056207 1–13 共2002兲. F. T. Smith, “Lifetime matrix in collision theory,” Phys. Rev. 118, 349 –356 共1960兲. M. Bauer, P. A. Mello, and K. W. McVoy, “Time delay in nuclear reactions,” Z. Phys. A 293, 151–163 共1979兲. Z. S. Wu and Y. P. Y. Wang, “Electromagnetic scattering for multilayered spheres: recursive algorithms,” Radio Sci. 26, 1393–1401 共1991兲. B. R. Johnson, “Light scattering by a multilayer sphere,” Appl. Opt. 35, 3286 –3296 共1996兲. 10 March 2004 兾 Vol. 43, No. 8 兾 APPLIED OPTICS 1771 27. A. V. Snyder and J. D. Love, “Reflection at a curved dielectric interface— electromagnetic tunneling,” IEEE Trans. Microwave Theory Tech. MTT-23, 134 –141 共1975兲. 28. M. Hentschel and H. Schomerus, “Fresnel laws at curved dielectric interfaces of microresonators,” Phys. Rev. E 65, 045603 共2002兲. 29. A. D. Stone, “Wave-chaotic optical resonators and lasers,” in 1772 APPLIED OPTICS 兾 Vol. 43, No. 8 兾 10 March 2004 Proceedings of the Nobel Symposium Quantum Chaos 2000, Phys. Scr. T90, 248 –262 共2001兲. 30. B. Crespi, G. Perez, and S. J. Chang, “Quantum Poincaré sections for two-dimensional billiards,” Phys. Rev. E 47, 986 – 991 共1993兲. 31. J. Wiersig, “Hexagonal dielectric resonators and microcrystal lasers,” Phys. Rev. A 67, 023807 共2003兲. III Paper III Elastic scattering of surface electron waves in quantum corrals: Importance of the shape of the adatom potential Phys. Rev. B, vol. 70, pp. 233409 1–4, 2004 PHYSICAL REVIEW B 70, 233409 (2004) Elastic scattering of surface electron waves in quantum corrals: of the adatom potential Importance of the shape A. I. Rahachou* and I. V. Zozoulenko† Department of Science and Technology (ITN), Linköping University, S-601 74 Norrköping, Sweden (Received 26 January 2004; revised manuscript received 9 April 2004; published 28 December 2004) We report elastic scattering theory for surface electron waves in quantum corrals defined by adatoms on the surface of noble metals. We develop a scattering-matrix technique that allows us to account for a realistic smooth potential profile of the scattering centers. Our calculations reproduce quantitatively all the experimental observations, which is in contrast to previous theories (treating the adatoms as point scatterers) that require additional inelastic channels of scattering into the bulk in order to achieve the agreement with the experiment. Our findings thus indicate that accounting for a realistic potential as well as using the exact numerical schemes is important in achieving detailed agreement as well as interpretation of the experiment. DOI: 10.1103/PhysRevB.70.233409 PACS number(s): 73.20.At, 03.65.Nk, 72.10.Fk, 68.37.Ef Advances in modern nanotechnology made it possible to manipulate adatoms on the surface of a metal, arranging them into ordered structures coined as “quantum corrals.” Using scanning tunneling microscopy (STM), Crommie et al.1–3 studied the scattering of the surface electron waves residing at (111) faces of Cu. These surface states interact strongly with Fe adatoms, and the spatial variation of the STM differential conductance revealed beautiful images of the standing wave patterns in the quantum corrals. In addition, the experiment showed a series of pronounced peaks in the energy spectrum of the differential conductance dI / dV at the center of the corrals. In order to describe the experimental observation,1,2 Heller et al.4 have developed multiple-scattering theory for surface electron waves in quantum corrals. In their theory each adatom was modeled as a pointlike ␦-function potential supporting isotropic s-wave scattering. The quantitative agreement with the experiment was achieved by assuming an additional (inelastic) channel of scattering (presumably into the bulk metallic states). The authors concluded that absorption is the dominant mechanism for the broadening of the energy levels seen in the experiment, and estimated that ⬃25% of the incident amplitude is reflected, ⬃25% is transmitted, and ⬃50% is absorbed. The importance of the electron scattering to the bulk states for the level width broadening in the quantum corrals was also asserted by Crampin et al.5 and Cramplin and Bryant.6 An alternative purely elastic scattering theory for the same quantum corral structures was reported by Harbury and Porod.7 They modeled the adatoms by finite-height potential barriers, as opposed to “black dot” ␦-point absorbing scattering potential adopted in the above cited works.4–6 The elastic theory accounts well for the spatial variation of the wave function in the quantum corrals, but overestimates the broadening of the resonant levels, especially for higher energies. The findings of Harbury and Porod therefore suggest that the features of the spectrum can be sensitive to the detailed shape of the scattering potential. It is important to stress that accounting for a detailed shape of a scattering potential was crucial for quantitative description of many phenomena in quantum nanostructures. 1098-0121/2004/70(23)/233409(4)/$22.50 This, for example, includes the Hall and bend resistance anomalies in four-terminal junctions,8 the breakdown of quantized conductance in quantum point contacts calculated using realistic potentials,9 and the explanation of a branched flow in a two-dimensional electron gas.10 In the present Brief Report we develop a scattering matrix approach that allows us to account for a realistic smooth potential of the adatoms. We demonstrate that for such a potential the broadening and positions of the resonant states as well as the scattering wave function in the quantum corrals can be quantitatively described by the inelastic theory alone, without the assumption of any additional (inelastic) scattering channels. Our findings thus support a conclusion of Harbury and Porod7 that the elastic scattering model is compatible with the reported STM data and hence our results outline the importance of accounting for a realistic potential as well as using of exact numerical techniques in detailed comparison and interpretation of the experiment. The differential conductance dI / dV of the STM tunnel junction is proportional to the local density of states (LDOS) which is given in terms of the scattering eigenstates of the Hamiltonian Ĥ, q共r兲,3 dI/dV ⬃ LDOS共r,E兲 = 兺q 兩q共r兲兩2␦共E − Eq兲. 共1兲 In order to calculate the scattering eigenstates q共r兲 we adopt to the problem at hand the scattering matrix technique11 that was recently developed for the numerical solution of the Helmholtz equation for resonant states of dielectric optical cavities with both complex geometry and variable refraction index. This is possible because of a direct correspondence between the Helmholtz and Schrödinger equations.12 The advantage of the scattering-matrix technique is that it provides an efficient way to treat the smooth realistic profile of the adatom. Note that commonly used methods based on the discretization of the scattering domain would be rather impractical in terms of both computation power and memory, because the smoothly varying potential of the adatom has to be mapped into a discrete grid with a very small lattice constant. 233409-1 ©2004 The American Physical Society PHYSICAL REVIEW B 70, 233409 (2004) BRIEF REPORTS FIG. 1. The experimental spectrum of the differential conductance dI / dV at the center of the 88.7-Å-radius 60-Fe-adatom circular quantum corral structure on Cu(111) substrate (adopted from Ref. 4) (solid curve). The calculated spectrum, dotted line: our scattering-matrix technique applied for a smooth adatom potential with V0 = 2.5 eV, = 1.52 Å; dashed line: multiple-scattering theory for the ␦-barrier adatom potential with inelastic channel of scattering (adopted from Ref. 4). We consider a two-dimensional ring-shaped corral structure. Experimental observations suggest that adatoms strongly disturb the local charge density at the finite distance ⬃7 Å.3 An exact experimental shape of the adatom potential is not available. We thus model this potential as a Gaussian with the half-width and the height V0 centered at the location 共x0 , y o兲, V共x , y兲 = V0 exp关−共x − x0兲2 / 22兴exp关−共y − y 0兲2 / 22兴 (see below, inset to Fig. 3). In order to calculate the scattering eigenstates we divide the quantum corral into inner, outer, and intermediate regions. In the inner region (inside the corral) and in the outer region (outside the corral) the adatom potential is negligible, V共x , y兲 = 0. Therefore, in these two regions the solution to the Schrödinger equation can be written in analytical form. Introducing the polar coordinates we can write for the wave function outside the corral, +⬁ ⌿out = 共1兲 iq , 兺 关AqH共2兲 q 共kr兲 + BqHq 共kr兲兴e q=−⬁ 共2兲 共2兲 where H共1兲 q and Hq are the Hankel functions of the first and second kind of the order q describing, respectively, incoming and outgoing waves; k = 冑2m*E / ប, with m* being the effective electron mass. The expression for the wave function inside the corral ⌿in can be written in a similar fashion as an expansion over Bessel functions Jq. We introduce the scattering matrix S in a standard fashion,11,12 B = SA, where A and B are column vectors composed of the expansion coefficients Aq and Bq for incoming and outgoing states in Eq. (2). The matrix element Sq⬘q gives the probability amplitude of scattering from an incoming state q into an outgoing state q⬘. In order to apply the scattering matrix technique we divide the intermediate region [i.e., the region where the adatom potential V共x , y兲 is distinct from zero] into N narrow concentric rings. At each ith boundary between the rings we introduce the scattering matrix Si that relates the states propagating (or decaying) towards the boundary, with those propagating (or decaying) FIG. 2. The experimental curves (solid lines, adapted from Ref. 4) for the LDOS subject to the tip position inside a circular corral for low bias voltages. The calculated LDOS, dashed line: our scattering matrix technique applied for a smooth adatom potential with V0 = 2.5 eV, = 1.52 Å. Parameters of the structure and are the same as those in Fig. 1. Voltages are given in volts and measured relatively to the bottom of the surface-state band. All the theory voltages are shifted by ⫺0.01 V relative to the experiment. away from the boundary. The matrices Si are derived using the requirement of continuity of the wave function and its first derivative at the boundary between the two neighboring rings. (Note that in our calculations a typical ring width was chosen ⬃0.1 Å, and for the purpose of the wave function matching the ring was divided on ⬃400 sectors in the angular direction.) Successively combining the scattering matrices for all the boundaries,11,12 S1 丢 ¯ 丢 SN, we can relate the combined matrix to the scattering matrix S. With the help of the scattering matrix S we determine the wave function in the outer region for every incoming state q in Eq. (2). Using the expression for the matrices Si we then recover the corresponding wave functions in the intermediate region as well as the wave function ⌿in in the inner region. Note that in the scattering-matrix technique one combines only two scattering matrices at each step. Hence, it is not necessary to keep track of the solution for the wave function in the whole space. This obviously eliminates the need for storing large matrices and facilitates the computational speed. Using our scattering matrix technique we calculate the bias voltage dependence and the spatial distribution of the LDOS for 60-Fe-adatom, 88.7-Å-radius circular quantum corrals reported by Heller et al.4 (Figs. 1 and 2). The Fe adatoms are placed on the meshes of a 2.55-Å triangular grid corresponding to the hexagonal Cu(111) lattice. The effective mass used in all the simulations was taken as m* = 0.361m0 and the electron band-edge energy E0 = 0.43 eV below the Fermi energy of the electrons.1,7 In the absence of applied voltage V these parameters correspond to the wavelength of electrons = 30 Å. For the parameters of the adatom potential we use7 V0 = 2.5 eV, = 1.52 Å, which correspond to those used by Harbury and Porod,7 who modeled the adatoms as hard wall finite potential barriers of 1.52-Å diameter of the height of 2.5 eV. Figure 1 also shows corresponding results of the multiplescattering theory of Heller et al.4 Both theories show a simi- 233409-2 PHYSICAL REVIEW B 70, 233409 (2004) BRIEF REPORTS FIG. 3. The calculated spectrum of the differential conductance dI / dV at the center of the circular quantum corral structure for various widths and heights V0 of the adatom potential (the scattering efficiency of the potential is kept constant). Parameters of the structure are the same as those in Fig. 1. Inset shows the schematic shape of the potential. lar level of agreement with the experimental data in both the peak positions and level broadenings for the differential conductance as well as in the number and the peak positions for the spatial LDOS distribution throughout the cross section of the quantum corral. But in contrast to the case of ␦ scatterers used in Ref. 4 our model agrees quantitatively well with the experimental data without introduction of additional inelastic scattering channels. It should be pointed out that the spectrum calculated on the basis of the scattering-matrix technique shows relatively broad peaks with decreasing amplitude at lower energy. It is interesting to note that this feature in our calculated spectrum is in fact closer to the experimental data than predictions of other theories giving narrow peaks with high amplitude.3–7,13,14 (Note also that a simple convolution with a Gaussian function broadens the narrow peaks at the bottom of the spectra and brings theory closer to the experiment.5) We do not have a full explanation for the difference for lowenergy peaks between different theoretical approaches. One of possible reasons is that all other theories calculate LDOS from the imaginary part of the scattering Green’s function, whereas we calculate LDOS directly from the scattering eigenstates of the Hamiltonian, Eq. (1). Figure 3 represents the differential conductance spectra dI / dV at the center of the quantum corral structure where the width and the height V0 of the adatom potential are varied, but its scattering efficiency is kept constant [i.e, 兰V共r兲dr = const]. The best agreement with the experiment is achieved for ⬇ 1.5 Å and V0 = 2.5 eV, which is in agrement with the results reported by Harbury and Porod.7 It is often assumed that because the spatial extent of the scattering potential 共⬃7 Å兲 is small compared to the wavelength of the incoming particles 共 ⬃ 30 Å兲, the adatom potential can be treated as a point scatterer or even as a continuous boundary V共r兲 ⬃ ␦共r − r0兲.3,4,13–15 The results presented in Fig. 3 clearly show that even though Ⰶ , the finite width of the scattering potential affects strongly the observed characteristic of the systems. Our calculations thus signify the importance of the shape of the scattering potential for achieving the quantitative agree- FIG. 4. Effect of the nonideal positioning of the adatoms on the differential conductance of the quantum corral. Parameters of the structure and the adatom potential are the same as those in Figs. 1 and 2. The inset illustrates the displacement of the scatterers from their ideal positions (white circles) in a circular geometry for twolattice shift. ment with the experiment of the voltage dependence of dI / dV. The experiment2 suggests that for high energies of incoming electrons (i.e., for large tip voltages) the significant fraction of Fe adatoms can move from their original positions. We therefore study the effect of this displacement on the shape and broadening of the resonant states of the quantum corral. Figure 4 shows the calculated differential conductance for a quantum corral for the case when the adatoms of an ideal circular corral are randomly shifted from their positions by one and two lattice constant (as illustrated in the insets to Fig. 4). A one-lattice shift does not seem to have a significant effect on the broadening of the peaks in the differential conductance. The deviation from an ideal circular case become rather noticeable for the shift in adatom positions by two lattice sites. As expected, these deviations are more pronounced for larger energies of incoming electrons. Note that we performed simulations for different realization of the ensembles of scatterers (keeping the average shift fixed to one or two lattice sites), and all of them are almost indistinguishable. This is because of the self-averaging character of scattering in a quantum corral due to a large number of scatterers. Our calculations thus pinpoint the deviation from a regular arrangements of scatterers as an additional factor that should be taken into account for achieving the quantitative agreement with the experiment. Let us now discuss the coupling between the surface and the bulk metallic states. In the assumption of purely inelastic scattering the widths of the resonances in the differential conductance spectrum given by the multiple scattering theory of Heller et al. are far too narrow compared to experiment.3 In order to reproduce the experiment one thus assumes that each adatom acts as an absorbing “black dot” providing an inelastic channel of scattering from the surface to presumably bulk states. This is done by introducing the complex phase shift4 and phenomenological self energy.3–6,13,14 By fitting these parameters an excellent agreement with the experiment can be achieved. Our results, how- 233409-3 PHYSICAL REVIEW B 70, 233409 (2004) BRIEF REPORTS ever, demonstrate that the experimental data are fully consistent with the model of elastic scattering with a smooth realistic potential even without invoking inelastic channels of scattering. Following Harbury and Porod7 we conclude that inability of the multiple-scattering theory to reproduce the experiment without invoking the inelastic channel is most probably related to the s-wave approximation. In contrast, our method is numerically exact in the sense that it is not limited to the s-wave scattering and all partial waves as well as mode mixing between them are taken into account by the exactly calculated scattering matrices. It should be stressed, however, that our scattering model is strictly two-dimensional (2D). Fully three-dimensional scattering calculations were reported by Crampin and Bryant,6 where they used self-consistent potentials and included partial waves up to l = 3. Unfortunately, our method cannot be easily extended into three dimensions and we are thus not in the position to comment on the importance of the coupling between surface and bulk states within our model. However, it is important to stress that a vast majority of the Financial support from Vetenskapsrådet (I.V.Z) and the National Graduate School of Scientific Computing (A.I.R.) is gratefully acknowledged. We appreciate a discussion with G. Hansson. 9 M. *Electronic address: [email protected] †Electronic papers use 2D models of scattering for the interpretation of the experiment.3,4,13,14 In view of the excellent agreement between our 2D calculations and the experiment we conclude that taking into account a realistic potential as well as using the exact numerical schemes not limited to the s-wave approximation is important in achieving a detailed agreement as well as interpretation of the experiment. To conclude, we studied the scattering of electron waves by quantum corral structure for the case of realistic smooth potential of adatom scatterers. We achieved a detailed agreement with the experiment without introducing an additional inelastic channels of scattering. Our findings also suggest that accounting for a realistic potential shape may be of particular importance for the quantitative description and interpretation of quantum mirage experiments. address: [email protected] 1 M. F. Crommie, C. P. Lutz, and D. M. Eigler, Nature (London) 363, 524 (1993). 2 M. F. Crommie, C. P. Lutz, and D. M. Eigler, Science 262, 218 (1993). 3 For a review see, G. A. Fiete and E. J. Heller, Rev. Mod. Phys. 75, 933 (2003). 4 E. J. Heller, M. F. Crommie, C. P. Lutz, and D. M. Eigler, Nature (London) 369, 464 (1994). 5 S. Crampin, M. H. Boon, and J. E. Inglesfield, Phys. Rev. Lett. 73, 1051 (1994). 6 S. Crampin and O. R. Bryant, Phys. Rev. B 54, R17 367 (1996). 7 H. K. Harbury and W. Porod, Phys. Rev. B 53, 15 455 (1996). 8 H. U. Baranger, D. P. DiVincenzo, R. A. Jalabert, and A. D. Stone, Phys. Rev. B 44, 10 637 (1991). Laughton, J. A. Nixon, J. H. Davies, and H. U. Baranger, Phys. Rev. B 43, 12 638 (1991); M. Laughton, J. R. Barker, J. A. Nixon, and J. H. Davies, ibid. 44, 1150 (1991). 10 M. A. Topinka, B. J. LeRoy, R. M. Westervelt, S. E. J. Shaw, T. Fleischmann, E. J. Heller, K. D. Maranowski, and A. C. Gossard, Nature (London) 410, 183 (2001). 11 A. I. Rahachou and I. V. Zozoulenko, J. Appl. Phys. 94, 7929 (2003); A. I. Rahachou and I. V. Zozoulenko, Appl. Opt. 43, 1761 (2004). 12 S. Datta, Electronic Transport in Mesoscopic Systems (Cambridge University Press, Cambridge, 1995). 13 J. Kliewer, R. Berndt, and S. Crampin, New J. Phys. 3, 22 (2001). 14 K.-F. Braun and K.-H. Rieder, Phys. Rev. Lett. 88, 096801 (2002). 15 A. Lobos and A. A. Aligia, Phys. Rev. B 68, 035411 (2003). 233409-4 IV Paper IV Light propagation in finite and infinite photonic crystals: The recursive Greens function technique Phys. Rev. B, vol. 72, pp. 155117 1–12, 2005 PHYSICAL REVIEW B 72, 155117 共2005兲 Light propagation in finite and infinite photonic crystals: The recursive Green’s function technique A. I. Rahachou and I. V. Zozoulenko Department of Science and Technology, Linköping University 601 74, Norrköping, Sweden 共Received 19 April 2005; revised manuscript received 7 June 2005; published 24 October 2005兲 We report a computational method based on the recursive Green’s function technique for calculation of light propagation in photonic crystal structures. The advantage of this method in comparison to the conventional finite-difference time domain 共FDTD兲 technique is that it computes Green’s function of the photonic structure recursively by adding slice by slice on the basis of Dyson’s equation. This eliminates the need for storage of the wave function in the whole structure, which obviously strongly relaxes the memory requirements and enhances the computational speed. The second advantage of this method is that it can easily account for the infinite extension of the structure both into air and into the space occupied by the photonic crystal by making use of the so-called “surface Green’s functions.” This eliminates the spurious solutions 共often present in the conventional FDTD methods兲 related to, e.g., waves reflected from the boundaries defining the computational domain. The developed method has been applied to study scattering and propagation of the electromagnetic waves in the photonic band-gap structures including cavities and waveguides. Particular attention has been paid to surface modes residing on a termination of a semi-infinite photonic crystal. We demonstrate that coupling of the surface states with incoming radiation may result in enhanced intensity of an electromagnetic field on the surface and very high Q factor of the surface state. This effect can be employed as an operational principle for surface-mode lasers and sensors. DOI: 10.1103/PhysRevB.72.155117 PACS number共s兲: 42.70.Qs, 41.20.Jb, 78.67.⫺n I. INTRODUCTION Optical microcavities and photonic crystals 共PC兲 have received increased attention in recent years because of the promising prospects of applications in a future generation of optical communication networks.1,2 Examples of successfully demonstrated devices include lasers, light emitting diodes, waveguides, add-drop filters, delay lines, and many others.3 By far the most popular method for the theoretical description of light propagation in these systems is the finitedifference time-domain method 共FDTD兲 introduced by Yee.4 The success of the FDTD method is due to its speed, flexibility, and ease of computational storage requirements. The limitation of the FDTD technique is related to the fact that the computational domain is finite. As a result, an injected pulse experiences spurious reflections from the domain boundaries, which leads to mixing between the incoming and reflected waves. In order to overcome this limitation the socalled perfectly matched layer condition has been introduced.5 However, even with this technique, a sizable part of the incoming flux can still be reflected back.6 In many cases the separation of spurious pulses is essential for the interpretation of the results, and this separation can only be achieved by increase of a size of the computational domain.7 This may lead to a prohibitive amount of computational work, because the stability of the FDTD algorithm requires a sufficiently small time step. The problem of the spurious reflections from the computational domain boundaries does not arise in the methods based on the scattering matrix technique, where the incident and outgoing fields are related with the help of the scattering matrix.8–12 Other approaches where the spurious reflections 1098-0121/2005/72共15兲/155117共12兲/$23.00 are avoided include, e.g., a multiple multipole method,13 and a Green’s function method14 based on the analytical expression for the Green’s function for an empty space. The main objective of the present paper is to present a computational approach based on the recursive Green’s function technique that can account for an infinite extension of a photonic crystal. In this technique the Green’s function of the photonic structure is calculated recursively by adding slice by slice on the basis of Dyson’s equation. In order to account for the infinite extension of the structure both into air and into the space occupied by the photonic crystal we make use of the so-called “surface Green’s functions” that propagate the electromagnetic fields into infinity. In this paper we present a method for calculation of the surface Green’s functions both for the case of a semi-infinite homogeneous dielectrics, as well as for the case of a semi-infinite periodic structure 共photonic crystal兲. This makes it possible to apply the Green’s function technique for investigation of a variety of important structures including waveguides and cavities in infinite or semi-infinite photonic crystals, as well as to study the effect of the surface states and the coupling of waveguide Bloch modes to the external radiation. Note that the recursive Green’s function technique is widely used for quantum mechanical transport calculations15–18 and is proven to be unconditionally numerically stable for various discretization schemes. The article is organized as follows. In Sec. II we present a general formulation of the problem. A description of the recursive Green’s function technique is given in Sec. III. This section also provides a recipe for the calculation of Bloch states in a periodic structure as well as the surface Green’s function. Technical details of the calculations are given in Appendixes A–C. Several examples of the application of the 155117-1 ©2005 The American Physical Society PHYSICAL REVIEW B 72, 155117 共2005兲 A. I. RAHACHOU AND I. V. ZOZOULENKO developed method are given in Sec. IV. The conclusions are presented in Sec. V. following discretization of the differential operators in Eqs. 共5兲 and 共6兲,19 ⌬2 II. GENERAL FORMULATION OF THE PROBLEM We start with Maxwell’s equations in two dimensions ⌬2 1 ⫻ 兵 ⫻ E共r兲其 = 2 E共r兲, r共r兲 c 2 ⫻ 再 冎 共1兲 where r = xi + yj, = 共 / x兲i + 共 / y兲j, r共r兲 is the relative dielectric constant, and the electric and magnetic field vectors E共r , t兲 = E共r兲exp共−it兲 and H共r , t兲 = H共r兲exp共−it兲. If the dielectric constant r共r兲 is independent on z, the Maxwell’s equations decouple in two sets of equations for the TE modes 共Hz , Ex , Ey兲, − um,m;n,n−1 f m,n−1 = TE modes: 冊 − i Ez , Hx = 0 y i Ez . 0 x 共3兲 Let us rewrite the equations for Hz , Ez 共2兲 and 共3兲 in an operator form2 TM modes: f ⬅ H z, f = 冑rEz, m,n = 1 , rm,n um,m−1;n,n = m−1/2,n , um,m;n,n+1 = m,n+1/2, um,m;n,n−1 = m,n−1/2 , f m,n = 冑rm,nEzm,n, m,n = 共9兲 1 冑rm,n , 冑 r 冉 2 2 2 + x y2 冊冑 1 r um,m−1;nn = m−1,nm,n , um,m;n,n+1 = m,n+1m,n, um,m;n,n−1 = m,nm,n−1 . + 兩0, . . . 0,1m,n,0, . . . ,0典 = 0, am,n 共10兲 共11兲 and am,n兩0典 = 0, 1 1 − , 共5兲 LTE = − x r x y r y 1 um,m+1;nn = m,nm+1,n, + am,n 兩0典 = 兩0, . . . 0,1m,n,0, . . . ,0典, 共4兲 f, LTM = − f m,n = Hzm,n, A convenient and common way to describe finitedifference equations on a numerical grid 共lattice兲 is to introduce the corresponding tight-binding operator. For this purpose we first introduce creation and annihilation operators, + , am,n. Let the state 兩0典 ⬅ 兩0 , . . . , 0m,n , . . . , 0典 describe an am,n empty lattice, and the state 兩0 , . . . 0 , 1m,n , 0 , . . . , 0典 describe an + , am,n act on excitation at the site m , n. The operators am,n these states according to the rules16 2 where the Hermitian differential operator L and the function f reads TE modes: 共8兲 f m,n , 2 , vm,n = 4m,n 1 2E z 2E z 2 + 2 Ez = 0, 2 + 2 r x y c c 2 um,m+1;n,n = m+1/2,n, TM modes: 共2兲 and for the TM modes 共Ez , Hx , Hy兲, 冉冊 ⌬ c vm,n = m+1/2,n + m−1/2,n + m,n+1/2 + m,n−1/2 , − i Hz , 0 r x Lf = 冉 冊 where the coefficients v , u are defined for the cases of TE and TM modes as follows: i Hz , Ex = 0 r y Hy = 共7兲 vm,n f m,n − um,m+1;n,n f m+1,n − um,m−1;n,n f m−1,n − um,m;n,n+1 f m,n+1 1 1 2 Hz + Hz + 2 Hz = 0, x r x y r y c 冉 2 共x兲f共x兲 → m+1 f m+1 − 2m f m + m−1 f m−1 , x2 we arrive at the finite difference equation 1 2 ⫻ H共r兲 = 2 H共r兲, r共r兲 c Ey = f共x兲 → m+1/2共f m+1 − f m兲 − m−1/2共f m − f m−1兲, 共x兲 x x am,n兩0, . . . 0,1m,n,0, . . . ,0典 = 兩0典 共12兲 and they obey the following commutational relations: . + + + 兴 = am,nam,n − am,n am,n = ␦m,n , 关am,n,am,n 共6兲 For the numerical solution, Eqs. 共4兲–共6兲 have to be discretized, x , y → m⌬ , n⌬, where ⌬ is the grid step. Using the + + ,am,n 兴 = 0. 关am,n,am,n兴 = 关am,n Consider an operator equation 155117-2 共13兲 PHYSICAL REVIEW B 72, 155117 共2005兲 LIGHT PROPAGATION IN FINITE AND INFINITE… L̂兩f典 = 冉 冊 ⌬ c 2 兩f典, 共14兲 where the Hermitian operator + + am,n − um,m+1;n,nam,n am+1,n L̂ = 兺 共vm,nam,n m,n + + − um+1,m;n,nam+1,n am,n − um,m;n,n+1am,n am,n+1 + − um,m;n+1,nam,n+1 am,n兲 共15兲 acts on the state + 兩0典. 兩f典 = 兺 f m,nam,n 共16兲 m,n Substituting the above expressions for L̂ and 兩f典 in Eq. 共14兲, and using the commutation relations and the rules Eqs. 共11兲–共13兲, it is straightforward to demonstrate that the operator equation 共14兲 is equivalent to the finite difference equation 共8兲. Note an apparent physical meaning of the last four terms in Eq. 共15兲: terms 2 and 3 describe forward and backward hopping between two neighboring sites in the x direction, and terms 4 and 5 denote similar hopping in the y direction. In the next section we outline the Green’s function formalism for solution of Eq. 共14兲. III. THE RECURSIVE GREEN’S FUNCTION TECHNIQUE A. Basics Let us first specify structures under investigation. We consider light propagation through a photonic structure defined in a waveguide 共supercell兲 of the width N, where we assume the cyclic boundary condition 共i.e., the row n = N + 1 coincides with the row n = 1兲. The photonic structure occupies a finite internal region consisting of M slices 共1 艋 m 艋 M兲. The external regions are semi-infinite waveguides 共supercells兲 extending into regions m 艋 0 and m 艌 M + 1. The waveguides can represent air 共or a material with a constant refractive index兲, or a periodic photonic crystal. Figure 1 shows two representative examples where 共a兲 the semiinfinite waveguides represent a periodic photonic crystal with the period M, and 共b兲 a photonic structure is defined at the boundary between air and the semi-infinite photonic crystal. Let us first define the scattering states for the structures under consideration. The translation invariance along the supercell dictates the Bloch form for the ␣th incoming state 兩␣i 典, N 兩 ␣i 典 = + ␣ am,n 兩0典, 兺 m,n 兺 eik mn=1 m⬉0 + ␣ 共17兲 where k␣+ 共k␣− 兲 is the Bloch wave vector of the right␣ propagating 共left-propagating兲 state ␣, and m,n is the corresponding Bloch transverse eigenfunction satisfying the Bloch condition ␣ ␣ m,n = m+M,n . 共18兲 The transmitted and reflected states, 兩␣t 典 and 兩␣r 典, can be written in a similar form, FIG. 1. 共Color online兲 Schematic illustration of the system under study defined in a waveguide 共supercell兲 of the width N. An internal region of the structure occupies M slices. Two representative cases are shown: 共a兲 external regions are semiperiodic photonic crystals with the period M; 共b兲 external regions represent a semiinfinite periodic photonic crystal with the period M to the right and air to the left. Arrows indicate the directions on the incoming 共I兲, reflected 共R兲, and transmitted 共T兲 waves. 兩␣t 典 = 兺 N +  am,n 兩0典, 兺 t␣eik关m−共M+1兲兴 兺 m,n + m⭌M+1  共19兲 n=1 N 兩␣r 典 = +  am,n 兩0典, 兺 m,n 兺 兺 r␣eik mn=1 m⬉0  −  共20兲 where t␣共r␣兲 stands for the transmission 共reflection兲 amplitude from the incoming Bloch state ␣ to the transmitted 共reflected兲 Bloch state . Note that in the general case the wave ␣ can be different in the vectors k␣± and the Bloch states m,n left and right waveguides 关see, e.g., Fig. 1共a兲, when the photonic structure is defined at the boundary air–photonic crystal兴. The method of calculation of the Bloch states for an arbitrary periodic structure is described below in Sec. III C. We define Green’s function of the operator L̂ in a standard way, „共⌬/c兲2 − L̂…G共兲 = 1̂, 20 共21兲 where 1̂ is the unitary operator. The knowledge of the Green’s function allows one to calculate the transmission and reflection coefficients. Indeed, let us write down the solution of Eq. 共14兲 as a sum of two terms, the incoming state 兩i典 and the system response 兩典 representing whether the transmitted 155117-3 PHYSICAL REVIEW B 72, 155117 共2005兲 A. I. RAHACHOU AND I. V. ZOZOULENKO or reflected states, 兩t典 or 兩r典, 兩f典 = 兩i典 + 兩典. Substituting 兩f典 into Eq. 共14兲 and using the formal definition of the Green’s function Eq. 共21兲, the solution of Eq. 共14兲 can be written in the form 兩典 = G„L̂ − 共⌬/c兲2…兩i典. 共22兲 Calculating the matrix elements 具M + 1 , n兩典 ⬅ 具0 兩 aM+1,n典 and 具0 , n 兩 典 ⬅ 具0 兩 a0,n典, of the right and left hand side of Eq. 共22兲, we arrive at the N ⫻ N system of linear equations for the transmission and reflection amplitudes 共see for details Appendix A兲, ⌽M+1T = − GM+1,0共U0,1⌽−M+1Kl − ⌫l−1⌽0兲, ⌽0R = − G0,0共U0,1⌽−M+1Kl − ⌫l−1⌽0兲 − ⌽0 , 共23兲 共24兲 where the matrix elements 共T兲␣ = t␣, 共R兲␣ = r␣; G and G0,0 are the Green’s function matrixes with the elements M+1,0 共Gm,l兲n,p = 具0兩am,nGa+l,p兩0典. 共25兲 0,0 ⌫l ⬅ Gwg is the left “surface Green’s function” corresponding only to part of the whole structure, namely, to the semiinfinite waveguide 共supercell兲 that extends to the left, −⬁ ⬍ m 艋 0. The physical meaning of the surface Green’s function ⌫ is that it propagates the electromagnetic fields from the boundary slice of the semi-infinite waveguide 共supercell兲 into infinity. A method for calculation of the surface Green’s functions both for the case of a semi-infinite homogeneous dielectrics, as well as for the case of a semi-infinite photonic crystal in a waveguide geometry, is described below in Sec. III D. The matrices Kl and ⌽m are given by the rightpropagating Bloch eigenvectors k␣+ and the corresponding ␣ in the waveguides, eigenstates m,n 共Kl兲␣ = exp共ik␣+ 兲␦␣, ␣ 共⌽m兲n␣ = m,n , 共26兲 FIG. 2. Schematic illustration of the application of Dyson’s equation for calculation of Green’s function for a composed structure consisting of m + 1 slices 共see text for details兲. The operators 0 L̂0m and l̂m+1 describe respectively the structure composed of m 0 + V̂ corslices, and the 共m + 1兲th slice. The operator L̂m+1 = L̂0m + l̂m+1 responds to the composed structure of m + 1 slices, where V̂ is the perturbation operator describing the hopping between the mth and 共m + 1兲th slices. 0 of nearest neighbors. Suppose we know Green’s function Gm 0 0 the operator L̂m, as well as Green’s function gm+1 of the 0 , correspond to a single 共m + 1兲th slice, operator l̂m+1 0 + + = 兺 共vm+1,m+1am+1,n am+1,n − um+1,m+1;n,n+1am+1,n am+1,n+1 l̂m+1 n + − um+1,m+1;n+1,nam+1,n+1 am+1,n兲. 共The method of calculation of Green’s function for a single slice is outlined in Appendix C.兲 Our aim is to calculate Green’s function of the composed structure, Gm+1, consisting of m + 1 slices. The operator corresponding to this structure can be written down in the form 0 0 + l̂m+1 + V̂, L̂m+1 = L̂m and the diagonal “hopping matrix” U0,1 is defined as 共U0,1兲n,n⬘ = u0,1;n,n⬘␦n,n⬘ . 共27兲 关Note that the matrix Kl in Eqs. 共23兲 and 共24兲 refers to the right-propagating states in the left waveguide.兴 In the following sections we describe the recursive Green’s function technique based on the successive use of the Dyson’s equation, introduce the method for the calculation of Bloch states in a periodic structure, and outline the way to calculate the surface Green’s function ⌫. In order to calculate Green’s function of the internal structure 共i.e., for the slices 1 艋 m 艋 M兲 we utilize the recursive technique based on Dyson’s equation; see Fig. 2. In order to illustrate this technique let us consider a struc0 describing this ture consisting of m slices. The operator L̂m structure can be written down in the form r 0 L̂m 共30兲 0 l̂m+1 and are given by the expresswhere the operators ions Eqs. 共30兲 and 共29兲, and V̂ = V̂m,m+1 + V̂m+1,m is the perturbation operator describing the hopping between the mth and 共m + 1兲th slices, + am+1,n V̂ = V̂m+1,m + V̂m,m+1 = − 兺 共um,m+1;n,nam,n n + + um+1,m;n,nam+1,n am,n兲. 共31兲 The Green’s function of the composed structure, Gm+1, can be calculated on the basis of Dyson’s equation20 B. Recursive technique based on Dyson’s equations 0 = 兺 vra+r ar − 兺 ur,r+⌬a+r ar+⌬ , L̂m 共29兲 共28兲 r,⌬ where r = m⬘ , n⬘共1 艋 m⬘ 艋 m ; 1 艋 n⬘ 艋 N兲, and the summation over ⌬ in the second term is performed over all available Gm+1 = G0 + G0V̂Gm+1 , Gm+1 = G0 + Gm+1V̂G0 , 0 共32兲 where G is the “unperturbed” Green’s function correspond0 0 or l̂m+1 . For the sake of completeness, ing to the operators L̂m a brief derivation of Dyson’s equation is given in Appendix B. Thus, starting from Green’s function for the first slice g01 and adding recursively slice by slice, we are in the position to calculate Green’s function of the internal structure consist- 155117-4 PHYSICAL REVIEW B 72, 155117 共2005兲 LIGHT PROPAGATION IN FINITE AND INFINITE… ing of M slices. Explicit expressions following from Eqs. 共32兲 and used for the recursive calculations are given below, m+1,m+1 0 0 m,m 0 = 关I − gm+1 Um+1,m共Gm 兲 Um,m+1兴−1gm+1 , Gm+1 m+1,1 m+1,m+1 0 m,1 = Gm+1 Um+1,m共Gm 兲 , Gm+1 1,1 0 1,1 0 1,m m+1,1 = 共Gm 兲 + 共Gm 兲 Um,m+1Gm+1 , Gm+1 1,m+1 0 1,m m+1,m+1 = 共Gm 兲 Um,m+1Gm+1 , Gm+1 共33兲 where the upper indexes define the matrix elements of the + Green’s function Gm,m⬘ = 具0兩am,nGam 兩0典. This recursive ⬘,n⬘ technique is proven to be unconditionally numerically stable.15–17 The performance of the method is determined by the size of the system of linear equations 共33兲 which we solve when we add each consecutive slice. This system is solved M times, where M is the number of slices of the internal structure 共in the x direction兲. The size of Eqs. 共33兲 is N ⫻ N, where N is a number of discretization points in the y direction. Typical dimensions of the equations used for computations of the structures reported in Sec. IV are ⬃200⫻ 200. In order to calculate the Green’s function of the whole system, we have to connect the internal structure with the left and right semi-infinite waveguides. Starting with the left waveguide, we write L̂int+left = L̂int + L̂left + V̂, 共34兲 where the operators L̂int+left, L̂int, and L̂left describe respectively the system representing the internal structure ⫹ the left waveguide, the internal structure, and the left waveguide. The perturbation operator V̂ describes the hopping between the left waveguide and the internal structure. Applying then the Dyson equation in a similar way as we described above, Gint+left = G0 + G0V̂Gint+left , 共35兲 we are in the position to find the Green’s function Gint+left of the system representing the internal structure ⫹the left waveguide. G0 in Eq. 共35兲 in an “unperturbed” Green’s function corresponding to the internal structure and the semi-infinite waveguide 共the “surface Green’s function” ⌫兲. Having calculated the Green’s function Gint+left on the basis of Eq. 共35兲, we proceed in a similar way by adding the right waveguide and calculating with the help of the Dyson’s equation the total Green’s function G of the whole system. FIG. 3. 共Color online兲 Schematic illustration of the calculation of Bloch states in an infinite periodic structure 共see text for details兲. The operator L̂cell describes a unit cell under consideration, 1 艋 m 艋 M, and L̂out describes the rest of the structure. The hopping between the cell and the rest of the structure is described by the operator V̂. L̂ = L̂cell + L̂out + V̂, where the operators L̂cell and L̂out describe respectively the cell under consideration 共1 艋 m 艋 M兲, and the outside region including all other slices −⬁ ⬍ m 艋 0 and M + 1 艋 m ⬍ ⬁, and V̂ is the hopping operator between the cell and slices m = 0 and m = M + 1. Write the total wave function + 兩0典 in the form 兩典 = 兺m,nm,nam,n 兩典 = 兩cell典 + 兩out典, 共37兲 where 兩cell典 and 兩out典 are respectively wave functions in the cell and in the outside region. Substituting Eqs. 共36兲 and 共37兲 into Eq. 共14兲, we obtain 兩cell典 = GcellV̂兩out典, where Gcell is the Green’s function of the operator L̂cell. Calculating the matrix elements 具1 , n 兩 典 and 具M , n 兩 典, this equation can be written in the matrix form, 1,M 1 = G1,1 cellU1,00 + Gcell U1,0M+1 共38a兲 M,M M = GM,1 cell U1,00 + Gcell U1,0M+1 , 共38b兲 where the vector column m = 共m,1 . . . m,N兲 , and where we used UM,M+1 = U0,1 共because of the periodicity兲 and U0,1 = U1,0 关according to the definition of U, Eq. 共27兲兴. It is convenient to rewrite Eq. 共38a兲 in a compact form T C. Bloch states of the periodic structure In this section we describe the method for calculation of the Bloch states in periodic waveguides 共supercells兲 using the Green’s function technique. A similar method was used for calculation of Bloch states in quantum-mechanical structures.18 Consider a unit cell of a periodic waveguide occupying M slices, 1 艋 m 艋 M; see Fig. 3. Rewrite the operator corresponding to the whole structure in the form 共36兲 T1 where T1 = 冉 冉 冊 冉 冊 M+1 1 = T2 , M 0 − G1,M cell U1,0 0 GM,M cell U1,0 I 冊 , T2 = 冉 − I − G1,1 cellU1,0 0 GM,1 cell U1,0 冊 , 共39兲 with I being the unitary matrix. The wave function of the periodic structure has Bloch form, 155117-5 PHYSICAL REVIEW B 72, 155117 共2005兲 A. I. RAHACHOU AND I. V. ZOZOULENKO 兩典 = Gwg兩s典, 共44兲 where 兩典 is the wave function that has to satisfy Bloch conditions 共40兲. Applying Dyson’s equation between the slices 0 and 1 we obtain 1,−M 0,−M Gwg = ⌫rU1,0Gwg , FIG. 4. 共Color online兲 A schematic diagram illustrating calculation of the surface Green’s function ⌫ of a periodic structure 共see text for details兲. M+m = eikxMIm . 共40兲 Combining Eqs. 共39兲 and 共40兲, we arrive at the eigenequation for Bloch wave vectors and Bloch states, T −1 1 T2 冉 冊 冉 冊 1 1 = eikxM , 0 0 共41兲 determining the set of Bloch eigenvectors kx␣ and eigenfunctions ␣, 1 艋 ␣ 艋 N. To improve numerical stability of Eq. 共41兲, it may be rewritten in the form11 共T1 + T2兲−1T1 冉 冊 冉 冊 1 1 = 共eikxM + 1兲−1 . 0 0 共42兲 This technique allows one to avoid overflows and underflows in the eigensolver routine when eigenvalues with 兩eikxM兩 Ⰷ 1 and 兩eikxM兩 Ⰶ 1 are calculated. In order to separate the left- and right-propagating states we compute the Poynting vector integrated over transverse direction, whose sign determines the direction of propagation. Bloch state propagating in a waveguide 共supercell兲 defined in a photonic crystal is illustrated below in Fig. 5共c兲. Poynting vector can be expressed as follows2 S␣共y兲 = 1 Re关E␣共y兲 ⫻ H␣* 共y兲兴. 2 共43兲 Note that for the case of the waveguide defined in air, M = 1, and Green’s functions Gcell in Eq. 共39兲 is simply given by Green’s function of a single slice g0 共see Appendix C for details of calculation of g0兲. D. The surface Green’s function ⌫ Consider a semi-infinite Bloch waveguide 共supercell兲 of the periodicity M extending in the region −M 艋 m ⬍ ⬁ as depicted in Fig. 4. Suppose that an excitation 兩s典 is applied to its first slice m = −M. Introducing the Green function Gwg corresponding to the operator L̂wg describing the waveguide, one can write down the response to the excitation 兩s典 in the form 共45兲 1,1 is the right surface Green’s function. 共Note where ⌫r ⬅ Gwg that because the waveguide is infinitely long and periodic, M+1,M+1 2M+1,2M+1 1,1 = Gwg = Gwg = . . . etc.兲 Taking the matrix Gwg elements 具1 , n 兩 典 of Eq. 共44兲 and making use of Eq. 共45兲, we obtain for each Bloch state ␣, 1␣ = ⌫rU1,00␣. The latter equation can be used for determination of ⌫r, ⌫rU1,0 = ⌿1⌿−1 0 , 共46兲 where ⌿1 and ⌿0 are the square matrixes composed of matrix columns 1␣ and 0␣, Eq. 共46兲. If the waveguide is open to the left, its surface Green’s function is the same as the surface Green’s function of the corresponding waveguide open to the right, ⌫l = ⌫r. Note that for the case of the waveguide defined in air the surface Green’s function 共46兲 simplifies to ⌫rU1,0 = K, where K is defined according to Eq. 共26兲. IV. APPLICATIONS OF THE METHOD To reveal the power of the method we study three model systems defined in 2D square-lattice photonic crystal. First, we calculate a transmission coefficient and quality factor 共Q factor兲 of several representative types of microcavities in infinite PCs. Then we focus on semi-infinite crystals where we investigate the effect of surface states, and, finally, we consider a semi-infinite PC with a waveguide opening to the surface. For the bulk crystal we choose a structure composed of cylindrical rods with the permittivity r = 8.9 and the diameter of a rod d = 0.4a in a vacuum background, where a is the size of the unit cell. Each unit cell is discretized into 25 points in both x and y directions. Most photonic crystal devices operate in a band gap. The structure at hand has a complete band gap for TM modes in the frequency range 0.32ⱗ a / 2c ⱗ 0.44,1 and does not have a complete band gap for the TE polarization. Because of this, we will hereafter consider the TM modes only. The developed method allows one to treat structures unlimited in x direction, whereas in y direction the structure of interest is confined within a supercell with imposed cyclic boundary conditions. This leads to the finite size effects in a photonic band structure. If the supercell consists of more than one elementary cell, additional bands appear along with the bands for infinite PC 关Figs. 5共a兲 and 5共b兲兴, as the result of the imposed boundary conditions in the transverse direction. A similar finite size effect emerges when air waveguides 共supercells兲 are attached to the system of interest. Even though we send a wave from an open space, we use a finite number of propagating modes. Solution of the eigenvalue problem 共4兲 for the air supercell gives a discrete set of right2 冑 2 2 propagating eigenstates km x = / c − 共2m / w兲 , where w is the width of the supercell, and m is integer such that max兩m兩 ⬍ w / 2c. Thus, a wave incident from air effec- 155117-6 PHYSICAL REVIEW B 72, 155117 共2005兲 LIGHT PROPAGATION IN FINITE AND INFINITE… A. Microcavity FIG. 5. 共Color online兲 共a兲 Band diagram for the rightpropagating TM mode of an infinite 2D photonic crystal 共r = 8.9, d = 0.4a兲 in ⌫X direction. PC has a fundamental bandgap in the frequency range 0.28ⱗ a / 2C ⱗ 0.44 共filled with gray in the figure兲. The line in the fundamental bandgap corresponds to a guided mode in a waveguide created by removing a central row of rods from the PC as shown in the inset. 共b兲 Additional bands 共encircled兲 originated from the finite size effect. The waveguide 共supercell兲 contains three unit cells in the transverse direction as illustrated in the inset. 共c兲 Bloch state propagating in the PC waveguide at a / 2c = 0.38. In this section we consider a microcavity defined in a waveguide in an infinite PC. The waveguide is created by removing a single central row of cylinders, such that in the energy range corresponding the fundamental bandgap only one waveguide mode can propagate. Band diagram of the waveguide mode is shown in Fig. 5共a兲. Three different cavities are introduced in order to show the effect of geometry and demonstrate the importance of proper design of a cavity. The first cavity is defined by two rods placed on the lattice sites, see insets in Fig. 7. In the second structure the diameter of the rods is doubled, and for the third cavity we place two rods from each side of the cavity to achieve better confinement. A dependence of the transmission coefficient on the incoming wave frequency is depicted in Fig. 7共a兲. We would like to stress that in the calculation of the transmission coefficient, the incoming, transmitted and reflected states are the Bloch states of a waveguide 关shown in Fig. 5共c兲兴, such that all spurious reflections from PC interfaces or computational domain boundaries are avoided. The fundamental parameter of cavity resonances is their Q factor defined as Q = 2ⴱ 共stored energy兲/共energy lost tively propagates only at certain incidence angles, determined by the ratio of the longitudinal and transverse wave vectors tan ␣ = kmy / km x , as illustrated in Fig. 6. Note that this finite size effect 共caused by the cyclic boundary conditions in the y direction兲 might in some cases represent a drawback of the method. FIG. 6. Dispersion relation for the air supercell of the width of 9a. Effective angles of incidence are determined by the angular wave number m. Inset shows the effective angles of incidence for m = −2 , −1 , 0 , 1 , 2. FIG. 7. 共Color online兲 共a兲 Transmission coefficient of three cavity structures versus frequency. 共b兲 Intensity of the Ez component of the electromagnetic field in the double-wall cavity at the resonance 共a / 2c = 0.3952兲. 155117-7 PHYSICAL REVIEW B 72, 155117 共2005兲 A. I. RAHACHOU AND I. V. ZOZOULENKO per cycle兲, which can be rewritten in the following form: ⍀ Q= 4 冕 , 共47兲 Sindy where ⍀TM = 兰关0兩Ez兩2 + 0共兩Hx兩2 + 兩Hy兩2兲兴dx dy and ⍀TE = 兰关0兩Hz兩2 + 0共兩Ex兩2 + 兩Ey兩2兲兴dx dy characterizes the energy stored in the system respectively for TM and TE polarizations and the integral over Sin is the incoming energy flux. Equation 共47兲 can be also expressed as a well-known relation Q = / ⌬ where is the resonant frequency and ⌬ is the width of the resonant peak at half-maximum. The resonance peak for the single-wall cavity is centered at a / 2c = 0.3952 and has Q factor 35.5. As expected, the highest Q factor 共327.7兲 is achieved for the case of doublerod walls. Resonance peak in the case of larger rods is shifted to the higher energy values 共a / 2c = 0.4281兲 because of the decrease of the effective size of the cavity. The lower Q factor in this case 共25.07兲 is because the larger rods disrupt destructive interference in a band gap of the PC. Note that the width of the supercell used in the computations has to be large enough to ensure that the intensity of the field decays to zero at the domain boundaries. At the same time, it is desirable to have the size of the computational domain as small as possible. For the present computations, keeping this tradeoff in mind, we have chosen a supercell consisting of seven unit cells in the y direction. This choice seems to be sufficient, as the field intensity decreases by five orders of magnitude within the length of two lattice constants from the waveguide towards the supercell boundaries. Finally, to confirm our results and to verify the developed method, we performed calculations for the cavities and waveguides in PC studied by Li et al.11 and found a full agreement with their results. B. Surface states In the previous section we considered wave propagation in an infinite photonic crystal. Another aspect of interest is the effect of the surface in semi-infinite photonic crystals that can accommodate a localized state 共surface mode兲 decaying both into air and into a space occupied by the photonic crystal.1,21 In the present section we study the coupling between an incident radiation and the surface states. Note that a surface mode residing on the surface of an infinite 共in the y direction兲 photonic crystal represents a truly bound state with the infinite lifetime. However, because of the used cyclic boundary conditions, our system is effectively confined in the transverse direction. As a result, the translation symmetry is broken, and the surface mode turns into a resonant state with a finite lifetime. Using the developed method, we calculate the Q factor of the surface modes. Our findings indicate that the surface modes, thanks to their high Q factors, can be used for lasing and sensing applications. We study two semi-infinite photonic crystal structures that support localized surface modes. In the first case a surface row of cylinders is composed of half-truncated rods1 共structure 1兲, and in the second case the cylindrical and half- truncated rods in the surface row are interchanged as shown in Fig. 8 共structure 2兲. In order to calculate the Q factor of the structures at hand, we illuminate the semi-infinite photonic crystal by an incidence wave 共that excites the surface modes兲 and compute the intensity of the field distribution. Note that the calculated field distribution includes the contributions from both the surface mode exited by the incident light, as well as the incident and reflected waves. This leads to a nearly constant off-resonance background in the dependence Q = Q共兲 that is caused by the contribution of the incident and reflected waves in the total field intensity in Eq. 共47兲. To remove this background we calculate the Q factor of a structure without surface states. We choose this structure as a semi-infinite photonic crystal with all identical cylindrical rods, which is known not to support surface modes.1 Then the obtained value is subtracted from the calculated value of the Q factor of the system under study. Note that in the calculation of the Q factor, the surface integration in Eq. 共47兲 is performed over the area depicted in Fig. 8. Figure 9 shows a Q factor of structures 1 and 2 as a function of the frequency of the illuminating light. For both structures the Q factor reaches ⬃104. Figures 8共a兲 and 8共c兲 show Ez-field distribution for structures 1 and 2 at the resonance. For a comparison, a field distribution for a structure that does not support a surface mode 共a semi-infinite photonic crystal with all identical cylindrical rods兲 is shown in Fig. 8共e兲. In the latter case the field intensity rapidly decays into the bulk of the photonic crystal, whereas for the structures supporting the surface modes, the intensity is strongly localized at the boundary row of rods. It is also worth mentioning that for the latter case the intensity of the field in the surface mode exceeds the incoming light intensity by four orders of magnitude, such that the light intensity in the air region is not visible in the figures 关compare Figs. 8共a兲 and 8共c兲 with 8共e兲兴. One can easily estimate the position of the resonant frequency for the surface modes. Indeed, the outermost row of the cylinders 共where the surface state resides兲 can be considered as a resonator with the characteristic resonant wavelengths following from the cyclic boundary conditions and given by ␣ = 2 / k␣, where k␣ = 2␣ , w 共48兲 ␣ is the mode number and w is the width of the waveguide. The surface state for structure 1 exists only in a limited frequency interval, 0.33ⱗ a / 2c ⱗ 0.37 共the dispersion relation of the surface mode of this structure is given in Ref. 1兲. It follows from this dispersion relation that all the modes given by Eq. 共48兲, except ␣ = 4, are situated outside this interval, whereas the mode ␣ = 4 corresponds to the frequency a / 2c = 0.365. This estimated frequency agrees very well with the actual calculated resonant frequency a / 2c ⬇ 0.359; see Fig. 9. Figures 8共b兲 and 8共d兲 show Poynting vector distribution for both structures at the resonance. For the structure 1 the Poynting vector is “curling” along the boundary, showing a low speed of the surface state. In contrast, for the structure 2, 155117-8 PHYSICAL REVIEW B 72, 155117 共2005兲 LIGHT PROPAGATION IN FINITE AND INFINITE… FIG. 8. 共Color online兲 Ez field and Poynting vector distributions for the structure 1 关共a兲,共b兲兴 and for the structure 2 关共c兲,共d兲兴 at the resonant frequencies 共marked by arrows in Fig. 9兲. 共e兲 Ez field distributions for the structure that does not support surface modes 共a semi-infinite photonic crystal with all identical cylindrical rods兲. In all cases the structures are illuminated by the incident wave propagating in the second mode m = 2 共see Fig. 6兲. the Poynting vector exhibits a rapid flow of energy along the boundary. Another difference between these structures is a very broad and rather strong “background” peak in the structure 2 in the region 0.34ⱗ a / 2c ⱗ 0.35 共with Q factor up to ⬃100兲. The presence of such a peak indicates that the corresponding surface state can be rather robust to various kinds of imperfections that are always present in real struc- FIG. 9. Dependencies Q = Q共兲 for structures 1 and 2 共solid and dashed lines, respectively兲. Arrows indicate the resonances for which the field intensities and Poynting vectors are visualized in Fig. 8. tures and which are known to broaden the resonances and lead to decrease of the Q factor.12 These two examples of photonic crystals illustrate that with proper structure design one can engineer and tailor properties of the surface states into the required needs. High values of the Q factors of the surface modes residing at the interface of the photonic crystal structures indicate that these systems can be used for lasing and sensing applications. The lasing effect has been demonstrated for different photonic crystal structures including band-gap defect mode lasers,22 distributed feedback lasers,23 and bandedge lasers.24 Utilization of the high-Q factor of the surface modes represent another way to sustain lasing emission. To achieve lasing effect careful design of the surface and surface mode engineering should be performed and the developed method seems to be a suitable tool for this purpose. Note that, for the structures considered so far in this section, the transverse confinement giving rise to the surface mode resonances is achieved rather artificially by imposing the cyclic boundary conditions. Let us now outline the design of a realistic device that can be used for the experimental implementation of a surface state cavity. Consider a structure composed of a semi-infinite photonic crystal containing only a finite number N of the surface rods of the reduced diameter d = 0.5D defining a resonant cavity as illustrated in Fig. 10. In this design the resonant modes of the cavity are not affected by the choice of transverse boundary conditions, as the electromagnetic field in the photonic crystal decays very rapidly outside the cavity area. 155117-9 PHYSICAL REVIEW B 72, 155117 共2005兲 A. I. RAHACHOU AND I. V. ZOZOULENKO FIG. 10. 共Color online兲 共a兲 Frequency dependence of the Q factor of a surface state photonic band gap cavity. Inset illustrates the resonant cavity defined by the N = 6 surface rods of the smaller diameter d = 0.5D placed on the photonic crystal surface. 共b兲 The dispersion relation for the surface state for the semi-infinite photonic crystal. The dashed lines indicate the expected resonant wave vectors for the modes ␣ = 5 , 6 , 7 and corresponding expected resonant frequencies. Figure 10共a兲 shows a Q factor of the resonant cavity as a function of the frequency of the illuminating light. In the given frequency interval there are three cavity modes with the Q factors of the order of ⬃105 − 106. It is expected however that in actual photonic structures realized typically in a slab geometry, the Q factor will be reduced due to the radiative decay in the direction perpendicular to the plane of the photonic crystal 共which has not been accounted for in the present 2D calculations兲. The estimation of the position of the resonant peaks for the cavity at hand can be performed in the same way used for the structures with truncated cylinders 共see Fig. 9 and related discussion兲. The structure at hand can be considered as a conventional Fabry-Pérot resonator whose resonant wavelengths are given by ␣ = 2 / k␣, with the wave vector k␣ = ␣ / w 关note the absence of factor 2 in contrast with Eq. 共48兲兴. From dispersion relation Fig. 10共b兲 it follows that only modes ␣ = 5 , 6 , 7 are situated inside the frequency interval where the surface mode exists. An estimation of the expected positions for the resonant peaks for these modes is also shown in Fig. 10 where the discrepancy between the expected and calculated resonance frequencies does not exceed 0.5%. Figure 11 illustrates the intensity of the Ez component of the electromagnetic field for the resonance mode ␣ = 6. As expected, the field is localized in the cavity inside the rods, and the intensity dies off very quickly both to the open space and to the photonic crystal. The field intensity at different rods in the cavity is expected to be determined by the overlap of the ␣th eigenstate of the Fabry-Pérot resonator with the actual positions of the rods in the cavity. This overlap for the 6th mode is shown in Fig. 11, which agrees very well with the actual calculated intensity distribution pattern. We also performed calculations for different numbers of rods FIG. 11. 共Color online兲 Lower panel: Calculated intensity of the Ez component for the 6th mode of a resonant cavity shown in Fig. 10. Upper panel: Expected field intensity at different rods is given by the overlap of the 6th eigenstate of the cavity with the actual positions of the rods. N = 5 – 11 and we always find an excellent agreement between the calculated and expected resonant frequencies as well as between the intensity distributions. C. Waveguide coupled to the open space The last example of application of the method presented here is a semi-infinite photonic crystal with a waveguide coupled to the surface; see Fig. 12. It has been recently demonstrated that a surface of a photonic crystal can serve as a kind of antenna to beam the light emitted from the waveguide in a single direction.25,26 These findings outline the importance of investigation of the surface modes in the photonic band-gap structures that can eventually open up the possibilities to integrate such devices with conventional fiber optic devices. FIG. 12. 共Color online兲 Ez field distributions at the surface of a truncated photonic crystal with a waveguide. 共a兲 The surface is composed of cylinders with parameters identical to those in the bulk of the crystal, and 共b兲 the surface cylinders are two times smaller than the cylinders in the bulk. 155117-10 PHYSICAL REVIEW B 72, 155117 共2005兲 LIGHT PROPAGATION IN FINITE AND INFINITE… In the present section we consider two different crystal terminations to illustrate the effect of the surface on propagation of the light emitted from the waveguide. In the first case the surface is composed of cylinders with parameters identical to those in the bulk of the crystal, and in the second case the surface cylinders are two times smaller than the cylinders in the bulk. The Bloch state propagating in a waveguide in the photonic crystal couples with the states in air and the resulting field distributions is shown in Fig. 12. The first structure does not support the surface mode, and hence the light intensity distribution in the air region exhibits a typical diffraction pattern. However, for the case of the second structure the field distribution in the air region is drastically different. In this case the Bloch state in the waveguide couples with the surface state localized at the crystal termination, such that the whole surface acts as a source of radiation. waveguide 共supercell兲 geometry. The ␣th Bloch state in the lattice can be written in the form + ␣ 兩␣典 = 兺 eik␣mm,n am,n 兩0典, + 共A1兲 m,n where summation is performed over all lattice sites and the ␣ satisfies the conditions 共18兲. Substituting Eq. function m,n 共A1兲 into Eq. 共14兲, we arrive at the finite difference equation valid for all sites m , n + + ␣ ␣ ␣ − um,m+1;n,neik␣m+1,n − um,m−1;n,ne−ik␣m−1,n vm,nm,n ␣ ␣ − um,m;n,n+1m,n+1 − um,m;n,n−1m,n−1 = 冉 冊 ⌬ c 2 ␣ m,n . 共A2兲 Consider now the incoming state 兩␣i 典, Eq. 共17兲. Substituting Eq. 共17兲 into Eq. 共14兲 and using Eq. 共A2兲 we obtain V. CONCLUSIONS We have developed a method based on the recursive Green’s function technique for the numerical study of photonic crystal structures. The method is proven to be an effective and numerically stable tool for design and simulation of both infinite photonic crystals and photonic crystals with boundaries. In the present method the Green’s function of the photonic structure is calculated recursively by adding slice by slice on the basis of Dyson’s equation. In order to account for the infinite extension of the structure both into air and into the space occupied by the photonic crystal we make use of the so-called “surface Green’s functions” that propagate the electromagnetic fields into infinity. This eliminates the spurious solutions 共often present in the conventional FDTD methods兲 related to, e.g., waves reflected from the boundaries defining the computational domain. The developed method has been applied to scattering and propagation of electromagnetic waves in photonic band-gap structures including cavities and waveguides. In particular, we have shown that coupling of the surface states with incoming radiation may result in enhanced intensity of the electromagnetic field on the termination of the photonic crystal and a very high Q factor of the surface modes localized at this termination. This effect can be employed as an operational principle for surface-mode lasers and sensors. Note added. We note that since the submission of this article, Xiao and Qiu have also reported similar results concerning a possibility to use the surface states as high-Q resonant cavities.27 冋 冉 冊册 L̂ − ⌬ c 2 ␣ + 兩␣典 = eik␣ 兺 u0,1;n,n1,n a0,n兩0典 + n ␣ + a1,n兩0典. − 兺 u1,0;n,n0,n 共A3兲 n Substituting this equation into Eq. 共22兲, calculating the matrix elements 具M + 1 , n 兩 典 and 具0 , n 兩 典, and using the relations GM+1,0 = − GM+1,1U1,0⌫l , 共A4兲 G0,0 = ⌫l − G0,1U1,0⌫l , 共A5兲 that follow from Dyson’s equation, we arrive at Eqs. 共23兲 and 共24兲 determinig the transmission and reflection amplitudes. APPENDIX B: DERIVATION OF THE DYSON’S EQUATION Let L̂0 be the operator describing an unperturbed system and V̂ be a perturbation. In our case the unperturbed system consists of several subsystems, e.g., m slices of the internal structure and 共m + 1兲th slice, and the perturbation corresponds to the coupling 共hopping兲 between them 共see Fig. 2兲. The operator of the total 共perturbed兲 system reads L̂ = L̂0 + V̂. ACKNOWLEDGMENTS Partial financial support from the National Graduate School in Scientific Computing 共A.I.R.兲 is acknowledged. G−1 = 共⌬/c兲2 − L̂ = 共⌬/c兲2 − L̂0 − V̂ = 共G0兲−1 − V̂. APPENDIX A: CALCULATION OF THE TRANSMISSION COEFFICIENT In this appendix we provide a detailed derivation of Eqs. 共23兲 and 共24兲. Consider first an infinite periodic structure in a 共B1兲 Let G0 and G be the Green’s functions of the unperturbed and the total 共perturbed兲 systems, respectively. Starting with the definition of the Green’s function 共21兲, we obtain 共B2兲 Multiplying this expression from the left with G and from the right with G0 we arrive at Dyson’s equations 155117-11 PHYSICAL REVIEW B 72, 155117 共2005兲 A. I. RAHACHOU AND I. V. ZOZOULENKO G0G−1G = G0共G0兲−1G − G0V̂G ⇒ G = G0 + G0V̂G. 共B3兲 Similarly one can also show that G = G0 + GV̂G0 . and calculating the matrix elements 共¯兲m,m;n,n⬘ + ⬅ 具0 兩 am,n . . . am,n 兩 0典, we arrive at the N ⫻ N system of linear ⬘ equations for the matrix elements of the Green’s function of a single slice gm, N 共B4兲 兺 n⬙=1 APPENDIX C: THE GREEN’S FUNCTION FOR A SINGLE SLICE 2 册 ␦n,n⬙ − lm,m;n,n⬙ gm,m;n⬙,n⬘ = ␦n,n⬘ , 共C2兲 lm,m;n,n⬙ = vm,n␦n,n⬙ 共C3兲 N − um,m;n⬙−1,n⬙␦n,n⬙−1 − um,m;n⬙+1,n⬙␦n,n⬙+1 . 共C4兲 n=1 Note that because of the cyclic boundary conditions in the n direction, the matrix elements um,m;1,N and um,m;N,1 are distinct from zero and defined according to um,m;N,1 = um,m;0,1 and um,m;1,N = um,m;N+1,N. + + am,n − um,m;n,n+1am,n am,n+1 lˆm = 兺 共vm,nam,n + − um,m;n+1,nam,n+1 am,n兲. 共C1兲 Using this operator in the definition of Green’s function 共21兲, D. Joannopoulos, R. D. Meade, and J. N. Winn, Molding the Flow of Light 共Princeton University Press, Princeton, NJ, 1995兲. Sakoda, Optical Properties of Photonic Crystals 共Springer, Berlin, 2001兲. 3 L. Thylen, M. Qiu, and S. Anand, Phys. Chem. Chem. Phys. 5, 1268 共2004兲. 4 K. K. Yee, IEEE Trans. Antennas Propag. 14, 302 共1966兲. 5 J.-P. Berenger, J. Comput. Phys. 114, 185 共1994兲. 6 A. Mekis, J. C. Chen, I. Kurland, S. Fan, P. R. Villeneuve, and J. D. Joannopoulos, Phys. Rev. Lett. 77, 3787 共1996兲. 7 X. Yu and S. Fan, Appl. Phys. Lett. 83, 3251 共2003兲. 8 D. Felbacq, G. Tayeb, and D. Maystre, J. Opt. Soc. Am. A 11, 2526 共1994兲; G. Tayeb and D. Maystre, J. Opt. Soc. Am. A 12, 3323 共1997兲. 9 D. M. Whittaker and I. S. Culshaw, Phys. Rev. B 60, 2610 共1999兲. 10 Z.-Y. Li and K.-M. Ho, Phys. Rev. B 68, 045201 共2003兲. 11 Z.-Y. Li and K.-M. Ho, Phys. Rev. B 68, 155101 共2003兲. 12 A. I. Rahachou and I. V. Zozoulenko, J. Appl. Phys. 94, 7929 共2003兲; A. I. Rahachou and I. V. Zozoulenko, Appl. Opt. 43, 1761 共2004兲. 13 E. Moreno, D. Erni, and C. Hafner, Phys. Rev. E 66, 036618 共2002兲. 14 O. J. F. Martin, C. Girard, and A. Dereux, Phys. Rev. Lett. 74, 526 共1995兲; O. J. F. Martin and N. B. Piller, Phys. Rev. E 58, 3909 共1998兲. 15 S. Datta, Electronic Transport in Mesoscopic Systems 共Cambridge 2 K. ⌬ c where the matrix element of the operator l̂m reads The operator describing the mth slice has the form 1 J. 冋冉 冊 University Press, Cambridge, UK, 1995兲. K. Ferry and S. M. Goodnik, Transport in Nanostructures 共Cambridge University Press, Cambridge, UK, 1997兲. 17 F. Sols, M. Macucci, U. Ravaioli, and K. Hess, J. Appl. Phys. 66, 3892 共1989兲. 18 I. V. Zozoulenko, F. A. Maaø, and E. H. Hauge, Phys. Rev. B 53, 7975 共1996兲; 53, 7987 共1996兲. 19 W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes. The Art of Scientific Computing 共Cambridge University Press, Cambridge, UK, 1992兲. 20 E. N. Economou, Green’s Functions in Quantum Physics 共Springer-Verlag, Berlin, 1990兲. 21 J. M. Elson and K. Halterman, Opt. Express 12, 4855 共2004兲. 22 O. Painter, R. K. Lee, A. Scherer, A. Yariv, J. D. O’Brien, P. D. Dapkus, and I. Kim, Science 284, 1819 共1999兲. 23 M. Meier, A. Mekis, A. Dodabalapur, A. Timko, R. E. Slusher, J. D. Joannopoulos, and O. Nalamasu, Appl. Phys. Lett. 74, 7 共1999兲. 24 S.-H. Kwon, H.-Y Ryu, G.-H. Kim, Y.-H. Lee, and S.-B. Kim, Appl. Phys. Lett. 83, 3870 共2003兲. 25 E. Moreno, F. J. Garcia-Vidal, and L. Martin-Moreno, Phys. Rev. B 69, 121402共R兲 共2004兲. 26 P. Kramper, M. Agio, C. M. Soukoulis, A. Birner, F. Muller, R. B. Wehrspohn, U. Gosele, and V. Sandoghder, Phys. Rev. Lett. 92, 113903 共2004兲. 27 S. Xiao and M. Qiu, physics/05050198 共unpublished兲. 16 D. 155117-12 V Paper V Waveguiding properties of surface states in photonic crystals J. Opt. Soc. Am. B, vol. 23, pp. 1679–1683, 2006 A. I. Rahachou and I. V. Zozoulenko Vol. 23, No. 8 / August 2006 / J. Opt. Soc. Am. B 1679 Waveguiding properties of surface states in photonic crystals A. I. Rahachou and I. V. Zozoulenko Department of Science and Technology, Linköping University, 601 74, Norrköping, Sweden Received January 3, 2006; revised March 17, 2006; accepted March 24, 2006; posted March 30, 2006 (Doc. ID 67038) We propose and analyze novel surface-state-based waveguides in bandgap photonic crystals. We discuss the surface-mode band structure, the field localization, and the effect of imperfections on the waveguiding properties of the surface modes. We demonstrate that surface-state-based waveguides can be used to achieve directional emission out of the waveguide. We also discuss the application of the surface-state waveguides as efficient light couplers for conventional photonic crystal waveguides. © 2006 Optical Society of America OCIS codes: 130.2790, 160.3130, 230.7370, 240.6690. 1. INTRODUCTION Photonic crystals (PCs) have attracted increasing attention in the past decade due to their unique properties and possible applications in integrated optical and photonic devices such as light-emitting diodes, delay lines, waveguides, and lasers.1,2 Among the variety of PC-based devices, waveguides play a crucial role not only as optical interconnections but also as active elements in wide-angle branches,3 channel add–drop filters,4,5 tapered couplers,6 optical switches,7 etc. Waveguides represent line defects in periodic crystal structures supporting guided Bloch modes whose frequency is located in the bandgap. These modes are strongly confined within the waveguide region and can propagate without loss to substantial distances. In this paper we propose a novel type of waveguiding structures, namely, waveguides that operate on surface states of semi-infinite PCs and are located on the surface of a PC. Employing surfaces of PCs as waveguides may open up new possibilities for design and operation of photonic structures for feeding and redistributing light in PCs. Surface states reside at the interface between a PC and open space, decaying into both media1 and propagating along the boundary. In a square-lattice PC the surface states appear in the bandgap when a boundary of a PC is modified in some way by, e.g., truncating the surface rods, shrinking or increasing their size, or creating more complex surface geometry.1,8–12 The surface modes in a semiinfinite PC represent truly Bloch states with the infinite lifetime and Q factor, and consequently do not couple to incoming or outgoing radiation. At the same time, it has been demonstrated that when the translational symmetry along the boundary of the semi-infinite crystal is broken, the surface mode turns into a resonant state with a finite lifetime, which can be utilized for lasing and sensing applications.12,13 It has also been recently shown that with the help of surface modes it is possible to achieve directional beaming from the waveguide opening on the modified surface of a PC,14,15 where surface states, coupled with outgoing waveguide radiation, suppress diffraction and focus the outgoing beam. At the same time, 0740-3224/06/081679-5/$15.00 to our knowledge there have been no studies of guiding properties of PC surfaces. To study surface states in PCs, we apply a novel computational method based on the recursive Green’s function technique.12 The advantage of this method is that it allows us to calculate and use surface Bloch modes as scattering states of the system, which makes it possible to compute the transmission coefficients for surface modes and corresponding field distributions. 2. SURFACE BAND STRUCTURE We consider a semi-infinite square-lattice PC composed of cylinders with ⑀ = 8.9 and diameter D = 0.4a (a is a lattice constant) in an air background. We study two different surface geometries, shown in Figs. 1(a) and 1(b), supporting the surface states where the outermost rods have reduced diameters d = 0.2a and enlarged d = 0.68a, respectively. This PC has a fundamental bandgap for TM polarization in the range of 0.33⬍ a / 2c ⬍ 0.44 and supports one surface mode for the case of Fig. 1(a) and two modes for the case of Fig. 1(b). The surface modes for these two structures show different patterns of field localization. For the structure of Fig. 1(a) the field intensity has one maximum within each rod and extends into the air, quickly decaying into the crystal. For low energies a significant part of the field intensity extends into a wide ⬃5 – 10a air layer near the surface of the PC. This can be attributed to the proximity between the dispersion curve of the surface state and the light line, where the group velocity of the surface state, = E / k, is close to c [see Figs. 1(a) and 4(a)]. As the energy increases, the dispersion curve moves away from the light line, and the field becomes mainly concentrated on the surface rods. For the case of the structure of Fig. 1(b) with enlarged surface rods, the field is mostly located within each cylinder and has a node oriented either horizontally (mode I) or vertically (mode II). In contrast with the case of Fig. 1(a), the intensity of both surface modes is mainly localized within the surface rods and its extent to the air is small for the whole energy range. © 2006 Optical Society of America 1680 J. Opt. Soc. Am. B / Vol. 23, No. 8 / August 2006 A. I. Rahachou and I. V. Zozoulenko 3. EFFECT OF INHOMOGENEITIES Fig. 1. (Color online) Band structures for TM modes in the ⌫X direction of square-lattice PCs composed of rod diameters D = 0.4a and permittivity = 8.9 along with the projected surface modes. The diameters of the surface rods are (a) d = 0.2a and (b) d = 0.68a. The thick black line denotes the light line. The lower panels show the normalized intensity of the Ez component in different points of surface-mode dispersion curves. Fig. 2. (Color online) Fragments of the band structures for TM modes in the ⌫X direction of infinite square-lattice test PCs composed of rods with = 8.9 and diameters (a) d = 0.2a, (b) d = 0.68a along with the projected surface modes of the semiinfinite PC of Fig. 1. Field distributions (Ez components) for the corresponding bands are given in the insets. Let us concentrate now on the surface-mode dispersion and the intensity distribution in structures with reduced and enlarged surface cylinders. To this end, we construct two test PCs entirely consisting of corresponding surface rods (i.e., with diameters d = 0.2a and d = 0.68). Their band structures along with the projected surface states for the structures shown in Figs. 1(a) and 1(b) are represented in Fig. 2. The dispersion curve of the surface state for the structure in Fig. 1(a) begins at the light line and remains nearly linear up to a / 2c ⲏ 0.40, where its slope slowly decreases and finally reaches zero. Figure 2(a) demonstrates that the shape of the surface state closely follows the valence band of the test crystal in the ⌫X direction. The same situation also holds for the structure of Fig. 1(b), where both surface bands mimic the bulk levels in the conduction band of the corresponding test PC [given in Fig. 2(b)]. Field distributions for the corresponding bands are given in the insets of Fig. 2 and outline the relations between the surface-state bands and the corresponding bands of the test PCs. It is worth mentioning that both surface bands for the structure of Fig. 1(b) have a lower velocity in comparison with the structure in Fig. 1(a). Fast surface states are known as the most suitable for waveguiding applications, whereas slow modes can attract interest in structures for “slowing light”16 or in surface-state cavities.12,13 Let us focus on the effect of inhomogeneities of the PC (imperfections in the shape of the rods, their displacement, or variation of the refractive index throughout the crystal, etc.) on the waveguiding properties of surface states. It has been demonstrated previously that such imperfections strongly affect the performance of lasing microcavities.17,18 We will show here that such imperfections can cause a profound effect on the waveguiding efficiency of the surface modes. To study the effects of imperfections, we divide the system under study into three regions as shown in the upper panel of Fig. 3. Two of the regions are left and right semiinfinite periodic structures (perfect waveguides for surface modes), and the block of the PC in between is an imperfect region. Utilization of Green’s function technique allows us to use surface Bloch modes as scattering states that propagate in perfect waveguides from infinity into the imperfect region where they undergo scattering. Obviously, in the case when the scattering region is absent (perfect waveguides are attached to each other), the Bloch states propagate freely without any scattering. Because the model is numerical, the discretization of the circular rods of the PCs using a square grid lattice obviously leads to deviations from an ideal circular geometry, as illustrated in Fig. 3. These deviations can be treated as inhomogeneities or roughness of the structure. We stress that discretization of the rods in the periodic waveguides is deliberately chosen to be different from that for the central region (see Fig. 3). The central scattering region represents a PC of a width of 5 unit cells, each of them discretized into 25 meshes 共⬃ / 50兲 in both the x and y directions. Figure 4 shows the velocities of the surface states in both structures and corresponding transmission coefficients. (We note that when the discretization of each cell in the central region is the same as for the unit cells in the left and right waveguides, the transmission coefficient through the structure is unity.) The transmission coefficients for each surface mode drop quite rapidly in the energy regions corresponding to Fig. 3. Discretization details of the (a) semi-infinite periodic waveguides and (b) the central region. The upper panel shows the structure under study, where the shaded regions denote ideal semi-infinite waveguides, and the central region of the width of 5a represents an imperfect PC where scattering of the Bloch surface states takes place. A. I. Rahachou and I. V. Zozoulenko Vol. 23, No. 8 / August 2006 / J. Opt. Soc. Am. B 1681 rods in the surface-state waveguide gradually decreases to zero in the region of the conventional PC waveguide as shown in Fig. 5. In this device an incoming state in the surface-mode waveguide region enters a tapered region where it is adiabatically transformed into a conventional waveguiding state. The maximum achieved transmission reaches T ⬇ 0.8 (see inset to Fig. 5). We should also mention that careful optimization of the surface geometry may further improve the performance of surface-state waveguide couplers, but such work is out of the scope of the present paper. We also note that our two-dimensional calculations do not account for the radiative decay in the direction perpendicular to the plane of the PC. Fig. 4. (Color online) (a) Velocity of different surface modes from Fig. 1. (b) Transmission coefficient for surface modes propagating in a nonideal surface-mode waveguide. the low velocity of the surface state. This is because the backscattering probability is greatly enhanced for the low-speed states. Slow states for a structure with enlarged surface rods are the most strongly affected. Even for five imperfect unit cells, the transmission coefficients for both modes approaches 1 only in a very narrow energy range, which makes these states hardly appropriate for waveguiding purposes. At the same time, the transmission coefficient for the fast surface state in the structure of Fig. 1(a) with the reduced boundary rods remains at 1 in a wide energy region up to a / 2c ⬃ 0.40, which makes it a better candidate for waveguiding applications. B. Directional Emitter The width of a conventional waveguide in a PC is of the order of the wavelength of light . Because of this, the beam launched from a semi-infinite PC into open space is diffracted at the waveguide opening in a strong angular spread ⬃2. It has been recently shown that it is possible to achieve directional emission out of PC waveguides with corrugated terminations supporting leaky or evanescent surface states.14,15 We demonstrate here that directional emission with the angular spread much less than in conventional waveguides can also be achieved for the case of surface-state waveguides coupled to air. Figure 6(a) shows the Ez field intensity and a directional diagram for the surface state propagating in a semi-infinite waveguide corresponding to the structure with the surface rods of a reduced diameter. Most of the beam intensity is localized within the range of ⌬⍜ ⬃ 20°. It should also be noted that the coupling of the surface state to air is rather high. The inset to Fig. 6(b) shows the transmission coefficient for the surface state in the semi-infinite surface 4. APPLICATIONS OF SURFACE-STATE WAVEGUIDES A. Light Coupler Because of the unique location on the surface of the PC, surface-state waveguides can be exploited in a variety of novel applications. In this paper we focus on two of them, introducing a novel light lead-in structure and sketching the possibility to use a surface-state waveguide as a directional emitter. Feeding light into waveguides in PCs composed of dielectric rods in an air background is a complicated challenge, as normally it requires extremely accurate positioning of a dielectric waveguide and precise mode matching.19,20 Even in this case, diffraction at the waveguide termination usually hampers coupling, and the efficiency of such lead-in systems hardly exceeds 60%. Other coupling techniques, such as utilization of adiabatic dielectric tapers,21 can improve the device performance but exhibits high sensitivity to parameters of the tapers. In this paper we propose a novel coupler based on waveguiding properties of surface states. Figure 5 illustrates such a lead-in structure composed of a surfacestate waveguide on the left and a conventional tapered PC waveguide on the right. The diameter of the surface Fig. 5. (Color online) A lead-in coupler structure composed of a surface-state waveguide on the left and a conventional tapered PC waveguide on the right. The size of the surface rods gradually decreases to zero in the central region where the surface-state waveguide transforms into a conventional PC waveguide. The intensity distribution is shown for the Ez component of the electromagnetic field at a / 2c ⬇ 0.365. Arrows depict the flow of the Poynting vector. The transmission coefficient subject to the energy of incoming light is given in the inset. Parameters of the PC correspond to the structure in Fig. 1(a). 1682 J. Opt. Soc. Am. B / Vol. 23, No. 8 / August 2006 A. I. Rahachou and I. V. Zozoulenko 5. CONCLUSION We put forward a novel concept for waveguiding structures based on surface modes in bandgap photonic crystal structures. We analyze the surface-mode band structure, field localization, and the effect of imperfections on the waveguiding properties of the surface modes. To illustrate applications of the surface-state waveguides, we suggest a new principle for feeding light into a photonic crystal waveguide and demonstrate that a semi-infinite surfacestate waveguide can be used as a directional emitter. A. Rahachou, [email protected] [email protected]; I. Zozoulenko, REFERENCES 1. 2. 3. 4. Fig. 6. (Color online) (a) Intensity distribution for the Ez component of the electromagnetic field in the surface-mode waveguide terminated to air for a / 2c = 0.34. (b) Far-field radial component of the Poynting vector Sr共兲 radiated out of a surfacemode waveguide as a function of azimuthal angle for different a / 2. The inset shows the transmission coefficient for the surface state as a function of the frequency. Parameters of the waveguide correspond to the structure in Fig. 1(a). waveguide propagating into open space. T is close to unity in the energy region corresponding to high velocity of the surface state and drops rapidly for energies a / 2c ⲏ 0.40 where the velocity of the surface state decreases. For the case of PC waveguides with corrugated terminations, the directional emission is achieved as a result of the coupling between the incident beam in the waveguide with the surface modes on the PC termination that causes the destructive interference for all directions except for a narrow beaming cone.14,15 (Note that a mechanism of the directional beaming in PC waveguides is conceptually similar to that in a subwavelength aperture in corrugated metallic films, where the incident beam is coupled to the surface plasmons residing at the corrugated boundary.22,23) The origin of a rather narrow beaming cone for the case of surface-state waveguides is related to the fact that the surface state is localized in a wide spatial region near the surface ⱗ10a (as opposed to conventional waveguides whose width is typically ⬃a) (see Fig. 1). The angular spread in this case due to the diffraction, sin ⬃ / 10a, is consistent with the calculated far-field radial distribution of the Poynting vector. As the frequency of the incoming light increases, the surface mode becomes more localized, and the spread of the outgoing radiation increases. The effect of directional beaming in surface-mode waveguides might find practical application for integration of PCbased devices with conventional fiber-optic devices. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. J. D. Joannopoulos, R. D. Meade, and J. N. Winn, Molding the Flow of Light (Princeton U. Press, 1995). K. Sakoda, Optical Properties of Photonic Crystals (Springer, 2001). S. Fan, S. G. Johnson, J. D. Joannopoulos, C. Manolatou, and H. A. Haus, “Waveguide branches in photonic crystals,” J. Opt. Soc. Am. B 18, 162–165 (2001). T. Asano, B.-S. Song, Y. Tanaka, and S. Noda, “Investigation of a channel-add/drop-filtering device using acceptor-type point defects in a two-dimensional photoniccrystal slab,” Appl. Phys. Lett. 83, 407–409 (2003). S. Fan, P. R. Villeneuve, and J. D. Joannopoulos, “Theoretical analysis of channel drop tunneling processes,” Phys. Rev. B 59, 15882–15892 (1999). S. G. Johnson, P. Bienstman, M. A. Skorobogatiy, M. Ibanescu, E. Lidorikis, and J. D. Joannopoulos, “Adiabatic theorem and continuous coupled-mode theory for efficient taper transitions in photonic crystals,” Phys. Rev. E 66, 066608 (2002). M. F. Yanik, S. Fan, M. Soljarcić, and J. D. Joannopoulos, “All-optical transistor action with bistable switching in a photonic crystal cross-waveguide geometry,” Opt. Lett. 28, 2506–2508 (2003). R. D. Meade, K. D. Brommer, A. M. Rappe, and J. D. Joannopoulos, “Electromagnetic Bloch waves at the surface of a photonic crystal,” Phys. Rev. B 44, 10961–10964 (1991). F. Ramos-Mendieta and P. Halevi, “Surface electromagnetic waves in two-dimensional photonic crystals: effect of the position of the surface plane,” Phys. Rev. B 59, 15112–15120 (1999). X. Zhang, L.-M. Li, Z.-Q. Zhang, and C. T. Chan, “Surface states in two-dimensional metallodielectric photonic crystals studied by a multiple-scattering method,” Phys. Rev. B 63, 125114 (2001). J. M. Elson and K. Halterman, “Local density of states analysis of surface wave modes on truncated photonic crystal surfaces with nonlinear material,” Opt. Express 12, 4855–4863 (2004). A. I. Rahachou and I. V. Zozoulenko, “Light propagation in finite and infinite photonic crystals: the recursive Green’s function technique,” Phys. Rev. B 72, 155117 (2005). J.-K. Yang, S.-H. Kim, G.-H. Kim, H.-G. Park, Y.-H. Lee, and S.-B. Kim, “Slab-edge modes in two-dimensional photonic crystals,” Appl. Phys. Lett. 84, 3016–3018 (2004). E. Moreno, F. J. García-Vidal, and L. Martín-Moreno, “Enhanced transmission and beaming of light via photonic crystal surface modes,” Phys. Rev. B 69, 121402(R) (2004). P. Kramper, M. Agio, C. M. Soukoulis, A. Birner, F. Müller, R. B. Wehrspohn, U. Gössele, and V. Sandoghdar, “Highly directional emission from photonic crystal waveguides of subwavelength width,” Phys. Rev. Lett. 92, 113903 (2004). A. Figotin and I. Vitebskiy, “Slow light in photonic crystals,” arXiv:physics/0504112. A. I. Rahachou and I. V. Zozoulenko, “Effects of boundary roughness on a Q factor of whispering-gallery-mode lasing microdisk cavities,” J. Appl. Phys. 94, 7929–7931 (2003). A. I. Rahachou and I. V. Zozoulenko 18. 19. 20. 21. A. I. Rahachou and I. V. Zozoulenko, “Scattering matrix approach to the resonant states and Q values of microdisk lasing cavities,” Appl. Opt. 43, 1761–1772 (2004). P. Sanches, P. Bienstman, B. Luyssaert, R. Baets, and J. Marti, “Analysis of butt coupling in photonic crystals,” IEEE J. Quantum Electron. 40, 541–550 (2004). R. Stoffer, H. Hoekstra, R. M. De Ridder, E. V. Groesen, and F. P. H. Van Beckum, “Numerical studies of 2D photonic crystals: waveguides, coupling between waveguides and filters,” Opt. Quantum Electron. 32, 947–961 (2000). A. Mekis and J. D. Joannopoulos, “Tapered couplers for Vol. 23, No. 8 / August 2006 / J. Opt. Soc. Am. B 22. 23. 1683 efficient interfacing between dielectric and photonic crystal waveguides,” J. Lightwave Technol. 19, 861–865 (2001). H. J. Lezec, A. Degiron, E. Devaux, R. A. Linke, L. MartinMoreno, F. J. Garcia-Vidal, and T. W. Ebbesen, “Beaming light from a subwavelength aperture,” Science 297, 820–822 (2002). L. Martin-Moreno, F. J. Garcia-Vidal, J. J. Lezec, A. Degiron, and T. W. Ebbesen, “Theory of highly directional emission from a single subwavelength aperture surrounded by surface corrugations,” Phys. Rev. Lett. 90, 167401 (2003). VI Paper VI Light propagation in nanorod arrays J. Opt. A: Pure Appl. Opt., vol. 9 pp. 265–270, 2007 IOP PUBLISHING JOURNAL OF OPTICS A: PURE AND APPLIED OPTICS J. Opt. A: Pure Appl. Opt. 9 (2007) 265–270 doi:10.1088/1464-4258/9/3/010 Light propagation in nanorod arrays A I Rahachou and I V Zozoulenko Solid State Electronics, Department of Science and Technology, Linköping University, Norrköping, SE 601 74, Sweden E-mail: [email protected] and [email protected] Received 1 November 2006, accepted for publication 25 January 2007 Published 14 February 2007 Online at stacks.iop.org/JOptA/9/265 Abstract We study the propagation of TM- and TE-polarized light in two-dimensional arrays of silver nanorods of various diameters in a gelatin background. We calculate the transmittance, reflectance and absorption of arranged and disordered nanorod arrays and compare the exact numerical results with the predictions of the Maxwell–Garnett effective-medium theory. We show that interactions between nanorods, multipole contributions and formations of photonic gaps affect strongly the transmittance spectra that cannot be accounted for in terms of the conventional effective-medium theory. We also demonstrate and explain the degradation of the transmittance in arrays with randomly located rods as well as the weak influence of their fluctuating diameter. For TM modes we outline the importance of the skin effect, which causes the full reflection of the incoming light. We then illustrate the possibility of using periodic arrays of nanorods as high-quality polarizers. Keywords: nanorods, particle plasmons, Maxwell–Garnett theory, Green’s function technique (Some figures in this article are in colour only in the electronic version) 1. Introduction Resonance properties of nanoparticles have been observed for centuries thanks to the beautiful colours of gold- and silverpatterned stained glasses. Over the last decade nanopatterned materials have attracted even more increased attention due to their unique electronic and optical characteristics. Nowadays, they are considered as promising candidates for a wide variety of applications in subwavelength waveguiding [1, 2], enhanced Raman scattering spectroscopy [3], nonlinear optics [4], photovoltaics [5], biological/medical sensing [6] and many others. A characteristic size of metallic nanoparticles d is about an order of magnitude smaller than the wavelength of incoming light λ, which can excite collective oscillations of electron density inside the particle—plasmons. The plasmon excitation results in an enhanced extinction (extinction = absorption + scattering) as well as an increased intensity of the electromagnetic field near the particle [7]. The important issue that makes nanoparticles so attractive for sensing applications is the effect of the geometry and size of nanoparticles and the surrounding environment on the position of the plasmonic resonance [7–10]. For example, 1464-4258/07/030265+06$30.00 the presence of antibodies in cells affected by cancer modifies the environment for gold nanoparticles placed on a tissue and results in a shift of extinction peak that can be easily imaged by conventional microscopy [11]. Recently it has also been demonstrated [12, 13] that embedding metallic nanoparticles into a polymeric matrix provides a larger contrast in the effective refractive index of the blend material, being much lower or higher than that of a pure polymer. Developing such materials can facilitate creating high-contrast-index photonic polymer crystals. Nanoparticles assembled in nanochains can also be applied as subwavelength waveguides [2, 14, 15]. In the case of closely spaced particles the coupling (and light propagation) arises from the evanescent dipole field from each particle, which excites a plasmon on its neighbour. This excitation travels along the chain, making the electron density within all the particles oscillate in resonance. In the present paper we will focus on light propagation in large arrays of infinitely long nanorods. Prototypes of such arrays have been recently fabricated experimentally [16, 17]. These arrays represent randomly oriented or aligned long rods (or spikes) of a material (dielectric or metal), several tens of nanometres in diameter. Despite significant progress in © 2007 IOP Publishing Ltd Printed in the UK 265 A I Rahachou and I V Zozoulenko Blend Gelatin I R Gelatin ε′ ε ′′ ε ′ , ε ′′ nanofabrication technologies, to our knowledge, however, the theoretical description of light propagation in nanorod arrays is still missing. This paper is organized as follows. In section 2 we outline transmittance properties of nanorod arrays within the framework of the Maxwell–Garnett effective-medium theory. In section 3 we present numerical modelling of light propagation through periodic arrays of nanorods and compare the results with the predictions of the Maxwell–Garnett theory. In section 4 the effect of various types of disorder is studied. T A=1-T-R 2. Effective-medium theory We consider a gelatin matrix with an embedded twodimensional array of silver nanorods. The effective dielectric function εeff (ω) of that composite can be estimated from Maxwell–Garnett theory, developed more than 100 years ago [7]: εrod (ω) − εmat εeff (ω) − εmat = f , εeff (ω) + 2εmat εrod (ω) + 2εmat (1) where f = S2 /S1 is the filling factor of the nanorods embedded into the matrix, S1 is the active area of the matrix and S2 is the total cross-sectional area of the nanorods. The dielectric function of the gelatin matrix is εmat = 2.25. The dielectric function εrod (ω) of the nanorods is taken from the SOPRA database1 for the bulk material. The Maxwell–Garnett theory is valid for relatively small nanoparticles (nanorods) (up to several tens of nanometres) at low concentrations (less than 30%). The dielectric function (here and hereafter all the spectra are given with respect to the wavelength of light in a vacuum λ0 ) of the Ag(10%)–gelatin blend is presented in figure 1(a). The dielectric function in figure 1(a) characterizes the blend as a highly dispersive lossy material with an absorption peak centred around 414 nm. According to Mie’s theory this peak corresponds to the plasmon resonance of a single Ag spherical nanoparticle in gelatin. The position of the peak obeys the well-known relation εrod = −2εmat [7]. In order to study light propagation through a layer of the blend we consider a 2D ‘sandwich-like’ structure consisting of semiinfinite gelatin ‘waveguides’ connected to the blend region (see inset to figure 1(b)). The structure is assumed to be infinite in the z direction; thus the solution to Maxwell’s equations decouples into TE (where the vector of a magnetic field is parallel to z ) and TM (where the vector of an electric field is parallel to z ). The transmission, reflection and absorption for both polarizations are given in figures 1(b) and (c), respectively. It is easy to see that, for both TE and TM polarizations, there exists a gap (or a stop-band) in the transmission caused by the enhanced absorption near the extinction resonance peak. However, the reflectance and absorption within the stop-band possess distinct behaviour for different polarizations. When the real part of the dielectric constant of the blend becomes negative (400 < λ0 < 425 nm) the reflectance of the TE mode increases due to increased contrast against the dielectric function of the gelatin matrix (which causes a dip in the absorption). At the same time, for TM-polarized light the 1 URL: http://www.sopra-sa.com/more/database.asp. 266 λ Figure 1. (a) Dielectric function of a blend of silver nanorods (nanoparticles) with a concentration 10% embedded into a gelatin background. Transmittance, reflectance and absorption of the TE (b) and TM (c) modes propagating through a 0.7 μm thick layer of Ag(10%)–gelatin blend are shown. Inset in (b) outlines the system under study. reflectance sharply increases up to 1 because of the metallic character of the blend in this region and the enhanced skin effect. For both polarizations Bragg’s reflections from the boundaries of the blend region, manifesting themselves as minima and maxima, are clearly seen for λ0 > 500 nm. Despite its adequacy for small isolated circular nanoparticles, a simple Maxwell–Garnett theory, however, has certain limitations. Namely, it does not account for the shape and distribution of metal clusters in the dielectric medium, neglecting important polarization properties of both single non-circular particles and their arrangements [18, 19]. In order to incorporate these features and study transmission characteristics of periodic and disordered nanorod arrays we apply the recursive Green’s function technique [20]. 3. Periodic nanorod arrays We now focus on 2D arrays of infinitely long silver nanorods arranged as a square lattice in a gelatin background. Keeping the filling factor of Ag, f = 10%, constant, we consider two cases: (a) a finite-size lattice with thickness a = 0.7 μm of nanorods with the diameter d = 10 nm, and (b) the lattice of the same width assembled from nanorods of 60 nm in diameter, see figure 2. Lattice constants are 29 and 175 nm for cases (a) and (b), respectively. Such a choice of nanorod sizes is motivated by the essential difference in polarization properties of small and large nanoparticles [10]. If the nanoparticle is small enough (d λ0 ), according to Mie’s theory, only the dipole plasmonic oscillations contribute to the extinction spectra, whereas for larger particles higher-order resonances contribute Light propagation in nanorod arrays (a) (b) y a z a x Figure 2. Arrays of silver nanorods with diameter (a) 10 nm, and (b) 60 nm embedded in an infinite gelatin background. For both cases the thickness of the layer a = 0.7 μm and the filling factor f = 10%. λ λ Figure 4. Transmittance, reflectance and absorption of a single (a), a pair of horizontally (b) and vertically (c) aligned, and (d) four coupled nanorods for the TE-polarized light. The inter-rod distances are taken as 29 nm, equal to the lattice constant for the array (figure 2(a)). The size of the computational domain is also the same as that for the nanorod arrays in figure 2. λ λ Figure 3. Transmittance, reflectance and absorption of a TE mode travelling through the square arrays of nanorods with diameter (a) 10 nm and (b) 60 nm (see figure 2 for details). to the spectra as well [10]. Using the recursive Green’s function technique we perform numerical simulations for both TE and TM polarizations of light falling normally from the left to the boundary between gelatin and the blend. 3.1. TE-modes When a nanosized metallic nanoparticle is illuminated by light, the electric components of an electromagnetic field excite collective oscillations of electronic plasma inside the particles—plasmons. If the particles are arranged into chains, these plasmonic oscillations possess a resonant character that facilitates the propagation of light along the chain. Such chains have been intensively studied in the literature [2] as promising candidates for subwavelength waveguiding. Let us irradiate the array of infinitely long nanorods with TE-polarized light. In this case E x and E y components of the electromagnetic field excite coherent plasmonic oscillations on each nanorod. Figure 3 shows the calculated transmittance, reflectance and absorption of a TE mode propagating through the arrays of nanorods. Small nanorods. Let us first concentrate on an array of nanorods with the diameter 10 nm (figure 3(a)). In the spectra one can clearly distinguish two regions, namely the region of high absorption (λ0 < 600 nm), containing a wide main absorption peak at 414 nm, two minor peaks at 350 and 530 nm and the region of high transmittance (λ0 > 600 nm). Now we will take a closer look at these regions separately. The position of the main extinction resonance agrees well with that obtained from equation (1). However, in contrast to the Maxwell–Garnett theory, the spectrum contains two minor peaks near 350 and 530 nm. In order to explain them one needs to account for the effect of coupling between several nanorods. For this reason we compare light propagation through (a) a single isolated nanorod (diameter 10 nm), (b) two coupled nanorods aligned parallel to the light propagation direction, (c) those aligned perpendicularly and (d) four coupled nanorods. These four cases are presented in figure 4. For the case of a single isolated rod (figure 4 (a)) only one peak near 410 nm emerges, which is in good agreement with the analytical value of 414 nm. For the cases (b) and (c) of twin coupled nanorods the additional peaks, centred at 355 and 620 nm respectively, appear. Their origin has been thoroughly studied in [18] and clarified in terms of enhanced [case (b)] and weakened [case (c)] restoring forces between the particles. However, for the system of four particles (d) these forces partially compensate for each other, and the minor resonances move closer towards the main peak. Let us now focus on the wavelength region λ0 > 600 nm, where TE-polarized light propagates at high transmittance. In order to understand this behaviour, we complement the transmission coefficient with the band diagram of the nanorod array. It should be mentioned that, in general, a band diagram represents propagating Bloch states (states with real eigenvalues). However, as the metallic rods (or nanoparticles) are absorbing, all the states in the blend will be decaying and eventually die off at infinity. Yet all the Bloch eigenvalues in such systems have imaginary components. In figure 5(a) we represent a band structure (real parts of eigenvalues) in the X direction for the states with the smallest imaginary parts. 267 A I Rahachou and I V Zozoulenko ω ω π π λ λ π π Figure 5. Band diagrams of the nanorod arrays from figures 2(a) and (b), respectively. The dashed line outlines the light cone. Large nanorods. For nanorods with the diameter 60 nm, the position of the main extinction peak agrees with that of the small particles. However, there is an essential difference in the physics behind this. When the diameter of a nanoparticle increases, higher-order dipole oscillations now contribute to the resulting extinction spectrum [7]. It has been recently shown [10] that the peak centred at ≈400 nm is due to the quadruple resonance of a nanorod, whereas the dipole resonance is redshifted and overlaps with the region of the enhanced reflectance (500 < λ0 < 700 nm). The indication in favour of this interpretation is a narrower width of the stop-band in the transmission (60 nm against 100 nm in the case of small rods). This is because the higher-order dipole interactions causing the stop-band behaviour for the case of large nanorods are generally weaker. Now let us clarify the origin of the high-reflectance region. The lattice constant for this structure is 175 nm. This is of the same order as the wavelength of light, such that the structure effectively represents a two-dimensional photonic crystal. The plasmonic band in figure 5(b) extends from ωa/2πc = 0 to 0.4 (λ0 ≈ 660 nm) where it experiences a photonic bandgap that causes the high reflectance of the structure. This bandgap overlaps with the tail of the extinction peak near 500 nm (see figure 3). ε ′, ε ′′ The dispersion curve in figure 5(a) has a small bump around 550 nm, which is caused by the minor extinction resonance. The band is located very close to the light line that results in a rather strong coupling between the incoming light and the plasmonic Bloch states of the blend region. Such strong coupling explains the high transmittance in the red wavelength region. ε′ ε ′′ λ Figure 6. (a) Transmission, reflection and absorption coefficients of the TM mode through a nanorod array of d = 10 nm. Due to the skin effect light does not penetrate the blend region. For λ0 < 328 nm the real part ε of the dielectric function of silver (b) becomes positive and the transmission coefficient abruptly increases. rods and thus there is no plasmonic contribution in overall transmission. However, for very short wavelengths (λ0 < 328 nm) the real part of the dielectric function of silver becomes positive (see figure 6(b)) and the blend behaves like a lossy dielectric rather than a metal. This results in non-zero transmission in this region. The obtained results clearly show that resonant plasmonic oscillations in periodic nanorod arrays represent a dominating light propagation mechanism for the TE-polarized light, whereas for the TM modes the nanorod structure represents practically a perfect screen. This features can be utilized in a nearly 100% effective polarizer. 3.2. TM modes Let us now consider the TM polarization of the incoming light. Figure 6(a) shows the transmittance, reflectance and absorption of the TM-polarized light for the small nanorods. In contrast to the Maxwell–Garnett picture (figure 1), almost for the whole wavelength range under study light does not penetrate the region occupied by nanorods and gets fully reflected back, resulting in zero transmittance. This discrepancy can be explained by the skin effect on the silver rods. At the same time, the Maxwell–Garnett theory (1) fully disregards the important screening properties of the rods, simply averaging the effective dielectric constant over the structure. It is also worth mentioning that, as we consider infinitely long nanorods, the incoming TM mode does not excite any plasmons on the 268 4. Disordered nanorod arrays As we have demonstrated in the preceding section, TEpolarized incident light in the off-resonance wavelength region propagates through periodic arrays of small nanorods at very high transmission. Now we introduce some disorder in this array and consider two separate cases, namely when nanorods are arranged in a square lattice but have randomly varying diameter, and rods of equal diameter, randomly distributed within the layer. For both cases the filling factor f = 10% is kept constant and the distribution is taken as uniform. Figures 7(a), (c) and (b), (d) demonstrate the transmittance, reflectance and absorbtion for both cases. Light propagation in nanorod arrays dielectric function is derived by inserting the total aggregate polarizabilities instead of that of a single isolated particle into the Maxwell–Garnett theory. 5. Conclusions λ λ Figure 7. Transmission, reflection and absorption coefficients of the TE-polarized light propagating through disordered nanorod arrays. (a), (c) Two different configurations of nanorods arranged in a square lattice, their d diameter randomly varying from of 5 to 20 nm. (b), (d) Nanorods with fixed d = 10 nm are randomly distributed within the layer. Insets show the actual geometries of the structures. The transmission characteristics for the structure with the random diameter of nanorods (figures 7(a) and (c)) closely resemble those for the array of fixed-sized nanoparticles (figure 3 (a)). The main difference is that the weaker dipole interactions between adjacent particles of different diameters cause a minor narrowing of the stop-band (60 nm versus 100 nm in the ideal case) and a slight degradation of the minor extinction peak. It should be emphasized that the transmission properties of arrays with different distributions of nanorod diameter (figures 7(a) and (c)) are virtually the same. For the case of the randomly distributed equal-sized nanorods (figures 7(b) and (d)) the situation changes. In contrast to the previous case of the ordered nanorod array with random diameter, the absorption spectra in the region λ0 > 600 nm are extremely sensitive to the geometry of the structure. It is interesting to note how clustering of nanorods manifests itself. The overall absorption in the region λ0 > 600 nm is much higher (and the transmission is lower) in comparison to the periodic lattice, as it consists of the averaged multiple absorbtion peaks of closely situated, touching or overlapping nanorods. Since the inter-rod distances are not constant any longer, each single rod is now affected by many dipole interactions of different strengths from neighbouring rods. Reflectances, however, are practically identical and not significantly higher than for the periodic case. It can be explained that due to its non-periodicity this structure absorbs better than it reflects. The clustering and more complex interactions of nanorods influence the region λ0 < 600 as well. The main and minor absorption peaks for the structure, see figure 7(b), almost overlap, whereas for figure 7(d) they are still well separated. We should specially mention that, in order to incorporate the effect of nanoparticle aggregates into Maxwell–Garnett theory, several approaches were suggested [7] (see also references therein). In that case the effective-medium We have studied the propagation of TE- and TM-polarized light in two-dimensional arrays of silver nanorods in a gelatin background. In order to calculate transmittance, reflectance and absorption in arrays of ordered and disordered nanorods we applied the recursive Green’s function technique and compared the obtained numerical results with predictions of the Maxwell–Garnett effective-medium theory. We have demonstrated that this theory describes adequately only the case of the TE-polarized light propagating in ordered arrays of small (∼ 10 nm), well-separated nanorods and only in the frequency interval outside the main plasmonic resonance. Our numerical calculations outline the importance of geometrical factors such as the size of the rods and their distribution. In particular, we have demonstrated that interaction between adjacent nanorods brings a significant contribution to the transmission spectra, which is manifested as additional absorption peaks (that are missing in the effectivemedium approach). The Maxwell–Garnett theory also disregards both the impact of higher-order dipole contributions and the formation of photonic bandgaps in the case of arrays of large nanorods. We have also studied the effect of disorder on the transmittance of the nanorod arrays. We have introduced two types of disorder: (a) ordered array with randomly varying nanorod diameters, and (b) a random distribution of nanorods of the same size within the blend. The disorder in rod placement leads to a strong suppression of the transmission (and the enhanced absorption) due to plasmonic resonances related to the clustering of the rods. We have demonstrated that clustering effects are sensitive to the actual geometry of the structure. In contrast, the impact of randomly varying diameters of the rods is much less profound. Despite its partial adequacy for the TE-polarized light, the Maxwell–Garnett effective-medium theory is shown to be invalid for the case of TM polarization. It simply averages the effective dielectric function inside the blend, missing the important screening properties of the metallic nanorods and characterizing the blend as a (partially) transparent medium. In contrast, the numerical modelling shows the strong skin effect that fully prohibits the propagation of the TM modes through the structure. The region of high transmittance for the TE modes and the strong skin effect for the TM modes makes the nanorod arrays promising candidates for highquality polarizers. Acknowledgments We would like to thank Olle Inganäs for stimulating and fruitful discussions. We acknowledge partial financial support from the Center for Organic Electronics at Linköping University. Useful conversations with Nils-Christer Persson are also appreciated. 269 A I Rahachou and I V Zozoulenko References [1] Maier S, Brongersma M, Kik P and Atwater H 2002 Phys. Rev. B 65 193408 [2] Quidant R, Lévêque G, Weeber J-C, Dereux A, Girard C and Weiner J 2004 Europhys. Lett. 66 785 [3] Bachelier G and Mlayah A 2004 Phys. Rev. B 69 205408 [4] Haus J, Kalyaniwalla N, Inguva R and Bowden C 1989 J. Appl. Phys. 65 1420 [5] Westphalen M, Kreibig U, Rostalski J, Lüth H and Meissner D 2000 Sol. Energy Mater. Sol. Cells 61 97 [6] Alivisatos A 2004 Nat. Biotechnol. 22 47 [7] Kreibig U and Vollmer M 1995 Optical Properties of Metal Clusters (Berlin: Springer) [8] Khlebtsov N G, Trachuk L A and Melnikov A G 2005 Opt. Spectrosc. 98 83 [9] Lee K-S and El-Sayed M A 2005 J. Phys. Chem. B 109 20331 [10] Kelly K L, Coronado E, Zhao L L and Schatz G C 2003 J. Phys. Chem. B 107 668 270 [11] El-Sayed I H, Huang X and El-Sayed M A 2005 Nano Lett. 5 829 [12] Wang G P, Yi Y and Lin W 2004 J. Opt. Soc. Am. B 21 554 [13] Wang G P, Tan C, Yi Y and Shan H 2003 J. Mod. Opt. 50 2155 [14] Krenn J R et al 1999 Phys. Rev. Lett. 82 2590 [15] Maier S A, Kik P G, Atwater H A, Meltzer S, Harel E, Koel B E and Requicha A A G 2003 Nat. Mater. 2 229 [16] Dev A, Kar S, Chakrabarti S and Chaudhuri S 2006 Nanotechnology 17 1533 [17] Losic D, Shapter J G, Mitchell J G and Voelcker N H 2005 Nanotechnology 16 2275 [18] Rechberger W, Hohenau A, Leitner A, Krenn J R, Lamplecht B and Aussenegg F R 2003 Opt. Commun. 220 137 [19] Fung K H and Chan C T 2006 Preprint physics/0605021 [20] Rahachou A I and Zozoulenko I V 2005 Phys. Rev. B 72 155117 VII Paper VII Surface plasmon increased absorption in polymer photovoltaic cells submitted to Appl. Phys. Lett., 2007 Surface plasmon increased absorption in polymer photovoltaic cells Kristofer Tvingstedt, Nils-Krister Persson, and Olle Inganäs Biomolecular and Organic Electronics, Center of Organic Electronics, IFM, Linköping University, SE-581 83 Linköping, Sweden∗ Aliaksandr Rahachou and Igor V. Zozoulenko Solid State Electronics, Center of Organic Electronics, ITN, Linköping University, SE-601 74 Norrköping, Sweden We demonstrate the triggering of surface plasmons at the interface of a metal grating and a photovoltaic bulk heterojunction blend of alternating polyfluorenes and a C60 derivative. An increased absorption originating from surface plasmon resonances is confirmed by experimental reflection studies and theoretical modeling. Plasmonic resonances are further confirmed to influence the extracted photocurrent from devices. More current is generated at the wavelength position of the plasmon resonance peak. High conductivity forms of the polymer PEDOT:PSS are used to build inverted sandwich structures with top anode and bottom metal grating, facilitating for triggering and characterization of the surface plasmon effects. Conjugated polymers have emerged as promising materials for inexpensive and flexible photovoltaic cells, as liquid based printing techniques enables production on large scale at a lower price than for inorganic based solar cell [1 ]. A new route to increase the photon absorption in thin film photovoltaics is to exploit surface plasmon resonances at the polymer/electrode interface. Present day polymer photovoltaic cells are comprised by blends of a conjugated polymer and a soluble C60 derivative. These bulk heterojunction solar cells now show power conversion efficiencies up to 4–52,3 The quantum efficiency of thin film organic solar cells is however still limited, and one cause is the comparatively low carrier mobility. A thinner film between electrodes will lower the probability for charge recombination and increase the carrier drift velocity by the higher electric field. However, a minimum film thickness is always required for sufficient photon absorption. The appeal of surface plasmons (SP) for thin film photovoltaics is due to the presence of an intense optical near field enhancement at the metal/dielectric interface. The strong field associated with a SP is evanescent into dielectric materials up to half the involved wavelength. In the conjugated polymers exploited for photovoltaic applications the evanescent plasmon field extends more than 100 nm whereas it only extends a few nanometers into the metal. Hence the SP seems suited for absorption and exciton generation enhancement in thin film photovoltaic devices. Surface plasmons are best described as confined light waves that propagate along the conducting surface of a metal4 . They are trapped on the metal surface due to strong interaction with the free charge carriers of the conductor. Recent studies5,6 exploiting metal nanoparticles on top of inorganic Si diodes have displayed increased absorption and increased photocurrent response. Previously, organic small molecules have been coevaporated together with metal clusters to excite localized plasmons that couple to the active material, demonstrating both increased absorption and photocurrent6–8 . Though no reports are known on organic photovoltaic devices using periodic gratings for triggering propagating SPs, diffraction gratings have been used to study the effects of SPs in organic light emitting diodes by Hobson et al9,10 . The dispersion relation of a surface plasmon, propagating in the x-direction can be written as: µ ¶1/2 ε1 ε2 kx = k0 , (1) ε1 + ε2 where k = 2π/λ is a free-space wave vector, ε1 and ε2 are the permittivities of the dielectric and metal respectively. The position of the plasmonic resonance λ = 2π/k0 for the normal incidence of light upon a grating can roughly be estimated by the following relation: µ ¶1/2 2π ε1 ε2 = k0 , (2) d ε1 + ε2 where d is the period of the grating. To trigger a surface plasmon at a metal/dielectric interface with periodic nanostructures it is further essential that the aspect ratio of the metal structures is sufficient to enable coupling. We have used soft lithographical replication techniques11,12 to generate metal gratings with high optical quality. Previous grating printing methods13,14 exploits soft embossing which is not fully capable of replicating deep nanostructures in thin films. Therefore we have preferred to deposit the thin polymer blend films on a prestructured Al metal nanostructure with sufficient aspect ratio. Figure 1(a) displays an AFM scan of such a grating with period of 277 nm and height of 50 nm. To enable charge collection of generated carriers a thin layer of highly conducting diethylene-glycol doped PEDOT:PSS15 is deposited on top of the polymer/PCBM blend. The device is hence best described as an inverted polymer solar cell with a periodically nano structured Al bottom cathode [Fig. 1(b)]. The oxide formed on Al blocks charge transport; therefore to extract current from such an inverted cell it is necessary to use a thin layer of Ti/TiOx on top, which has recently been demonstrated16 to enable charge collection. 2 (a) (a) (b) (b) FIG. 1: (a) AFM scan of the surface of the metal grating prior to polymer deposition. (b) Schematic drawing of the inverted solar cell with nanostructured bottom cathode and transparent PEDOT:PSS top anode. FIG. 2: (a) Refractive index (N) and extinction coefficient (k) for 1:4 blends of APFO3 and PCBM. (b) N and k for 1:3 blends of APFO Green5 and PCBM. Insets display the molecular structure of the exploited materials. Two polymers, APFO317 and APFO Green518 , have been blended with PCBM (Fig. 2). The dielectric function for the blends, the refractive index (N) and extinction coefficients (k) obtained by ellipsometry19 is presented in Fig. 2 and used as input for the simulation of the SPs. The triggering of SPs at the metal/organic interface is demonstrated by measuring the reflection of monochromatic polarized light from samples placed at the backside of an integrating sphere, thereby collecting reflected light at all angles. As the samples are illuminated with light polarized parallel to the grating lines, no major difference from the reflection from a planar sample can be identified, since this polarization is not capable of launching SPs. When samples are illuminated with light polarized perpendicular to the grating lines, a new feature is observed. A strong resonance reflection dip occurs for the APFO3/PCBM blend at a wavelength of 620 nm where the blend itself has little absorption. For the APFO Green5/PCBM blend the plasmon resonance is located at a wavelength of 550 nm. This resonance fills up the dip in the absorption spectra of this material and the sample accordingly appears dark and colorless in reflection. The absorptance A=100-R for the blends on metal gratings under polarized illumination vs. the absorptance from planar cells is given in Fig. 3. In the simulations we consider a supercell containing 5 periods of sinusoidal grating profile, slightly distorted to make it more realistic. Adjacent supercells are connected through cyclic boundary conditions. In order to account for the non-ideality of the grating surface, we have introduced a randomized surface roughness with an amplitude of 5 nm, which is a typical averaged AFM experimental value. We discretize the structure with uniform grid (element size is 3.4 nm) and illuminate it with light polarized perpendicularly with respect to the grating lines. Having calculated the transmittance T and reflectance R from the recursive Greens function technique, the absorptance can be found as A = 100-T-R. First, we calculate the absorptance spectrum for APFO3/PCBM. A polymer blend with thickness d150 nm is deposited directly onto the Al grating. The simulated and measured absorptance spectra are given in Fig. 4(a). The spectrum contains two resonance peaks at 450 and 625 nm, which agree extremely well with the experimental curve. The two peaks have however completely different nature. The peak at 625 nm is a first-order plasmonic resonance and its position agrees quite well with a direct analytical estimation (598nm) from Eqn. (2). At this resonance the electromagnetic field (inset, Fig. 4b) is localized within the 50 nm region near the grat- 3 (a) (a) (b) (b) (a) Measured absorptance (A=100-R) for FIG. 3: APFO3/PCBM on Al grating illuminated with different polarizations compared to a planar absorptance spectra. (b) Absorptance spectra for APFO Green5/PCBM on Ti/Al grating illuminated with different polarizations compared to a planar absorptance spectra. ing surface, and its intensity is up to 7 times higher on the boundary than in the bulk of the polymer. The resonant peak at 450 nm represents a standing wave confined by total internal reflection at the polymer/air interface and the metal grating reflection. For the APFO Green 5/PCBM we model a 90 nm thick polymer blend layer, deposited onto the Al grating coated with a 5 nm thin Ti interfacial layer. Fig. 4b represents the computed absorptance spectrum and the simulated field distribution inside the cell. As the APFO Green 5/PCBM layer is thinner than the APFO3/PCBM, the standing wave is no longer supported and there is no corresponding peak in the spectrum. The plasmonic resonance for this structure is calculated to be centered at 555 nm. The surface plasmon resonance for the two polymers substantially alters the absorption profile, compared to planar samples. We measure the external quantum efficiency (EQE) under illumination of the sample with polarized light. The low bandgap APFO Green5 displays a clear influence around the SP resonance wavelength of 555 nm when illuminated with perpendicularly polarized light (Fig. 5). For the APFO3/PCBM cell no influence could be observed at the SP resonance wavelength of 620 FIG. 4: (a) Calculated and measured absorptance spectrum of APFO3/PCBM deposited directly on the Al grating. The insets show the calculated spatial distribution of the Hy component of the electromagnetic field in the polymer at the plasmonic resonance (625 nm) and for a standing wave peak (450 nm). (b) Calculated absorptance spectrum of APFO Green 5/PCBM on the Al grating coated with a 5 nm thick Ti interfacial layer. The inset demonstrates the spatial distribution of the Hy -component of the electromagnetic field in the polymer at the plasmonic resonance (555 nm). The lower dark part of the insets corresponds to the sinusoidal shaped metal grating where no field is present. nm. The different behavior is attributed to mismatch of the energy of the SP resonance. This energy is lower than the bandgap of the APFO3/PCBM but higher than the bandgap of APFO Green5/PCBM. We therefore suggest that no significant coupling from the SP to excitation of APFO3/PCBM can occur. In conclusion, this study demonstrates the possibility of influencing the generated photocurrent from thin film organic photovoltaic cells with propagating surface plasmons launched on nanostructured electrodes. An increase of photocurrent is observed at the resonant position of the SP, when the SP resonance has higher energy than the bandgap of the absorbing polymer. The good match between the measured and the simulated reflectance demonstrates the strong influence surface plasmons have on absorption in these thin film devices. At present gratings are far from ideal, due to inhomogeneity and the presence of the required Ti interfacial layers. We consider these effects on the plasmon intensity to be a 4 drawback that requires a remedy that will be carefully studied elsewhere. We suggest that the strong optical field intensity associated with a propagating SP wave is responsible for enhanced exciton generation inside the photoactive bulk heterojunction blend. The condition deduced from the comparative study is that the SP energy must be higher than the bandgap of the material. Acknowledgments FIG. 5: EQE for APFO Green5/PCBM when illuminated with light polarized perpendicular or parallel to the grating lines of the sample. At the resonance position of 555 nm we see an increase of photocurrent when exciting a plasmon. ∗ 1 2 3 4 5 6 7 8 9 10 11 12 Electronic address: [email protected] H. Hoppe and N. S. Sariciftci, J. Mat. Res. 19, 1924 (2004). G. Li, V. Shrotriya, J. S. Huang, Y. Yao, T. Moriarty, K. Emery, and Y. Yang, Nature Mater. 4, 864 (2005). J. Y. Kim, S. H. Kim, H. H. Lee, K. Lee, W. L. Ma, X. Gong, and A. J. Heeger, Adv. Mater. 18, 572 (2006). D. M. Schaadt, B. Feng, and E. T. Yu, Appl. Phys. Lett. 86, 063106 (2005). D. Derkacs, S. H. Lim, P. Matheu, W. Mar, and E. T. Yu, Appl. Phys. Lett. 89, 093103 (2006). O. Stenzel, A. Stendal, K. Voigtsberger, and C. Vonborczyskowski, Sol. Energy Mater. Sol. Cells 37, 337 (1995). M. Westphalen, U. Kreibig, J. Rostalski, H. Luth, and D. Meissner, Sol. Energy Mater. Sol. Cells 61, 97 (2000). B. P. Rand, P. Peumans, and S. R. Forrest, J. Appl. Phys. 96, 7519 (2004). P. A. Hobson, S. Wedge, J. A. E. Wasey, I. Sage, and W. L. Barnes, Adv. Mater. 14, 1393 (2002). P. A. Hobson, J. A. E. Wasey, I. Sage, and W. L. Barnes, IEEE J. Sel. Top. Quant. Elect. 8, 378 (2002). Y. N. Xia and G. M. Whitesides, Ann. Rev. Mater. Sci. 28, 153 (1998). U. Plachetka, M. Bender, A. Fuchs, B. Vratzov, T. Glin- We thank Mats R. Andersson and his group at Chalmers for polymer supply. This work was financed by the Center of Organic Electronics at Linköping University, funded by the Strategic Research Foundation SSF. 13 14 15 16 17 18 19 sner, F. Lindner, and H. Kurz, Microelectronic Engineering 73-74, 167 (2004). L. S. Roman, O. Inganäs, T. Granlund, T. Nyberg, M. Svensson, M. R. Andersson, and J. C. Hummelen, Adv. Mater. 12, 189 (2000). M. Niggemann, M. Glatthaar, A. Gombert, A. Hinsch, and V. Wittwer, Thin Solid Films 451-52, 619 (2004). X. Crispin, F. L. E. Jakobsson, A. Crispin, P. C. M. Grim, P. Andersson, A. Volodin, C. van Haesendonck, M. Van der Auweraer, W. R. Salaneck, and M. Berggren, Chem. Mater. 18, 4354 (2006). M. Glatthaar, M. Niggemann, B. Zimmermann, P. Lewer, M. Riede, A. Hinsch, and J. Luther, Thin Solid Films 491, 298 (2005). M. Svensson, F. L. Zhang, S. C. Veenstra, W. J. H. Verhees, J. C. Hummelen, J. M. Kroon, O. Inganäs, and M. R. Andersson, Adv. Mater. 15, 988 (2003). F. L. Zhang, W. Mammo, L. M. Andersson, S. Admassie, M. R. Andersson, L. Inganäs, S. Admassie, M. R. Andersson, and O. Ingands, Advanced Materials 18, 2169 (2006). R. M. A. Azzam and N. M. Bashara, Ellipsometry and Polarized Light, Elsevier Science Pub Co (1987).