Multimode Spectroscopy for Self-Referenced Plasmonic Biosensing

Transcription

1 Multimode Spectroscopy for Self-Referenced Plasmonic Biosensing by Farshid Bahrami A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Graduate Department of Electrical Engineering University of Toronto Copyright by Farshid Bahrami (2014)

2 Multimode Spectroscopy for Self-Referenced Plasmonic Biosensing Farshid Bahrami Doctor of Philosophy Graduate Department of Electrical Engineering University of Toronto 2014 Abstract Surface plasmon resonance (SPR) is an efficient mechanism for biosensing due to its high field intensity and subwavelength confinement. However, the information obtained from these sensors is limited as they only rely on a single measurement at TM polarization. This limitation results in cross sensitivity of the surface plasmon (SP) wave to several parameters such as variations in bulk solution concentration or temperature fluctuations, in addition to the analyte to be measured. In this dissertation the possibility to decouple the aforementioned parameters from the analyte by spectroscopy with more than one mode is investigated. A hybrid plasmonic waveguide (HPWG) and a plasmon waveguide resonance (PWR) sensor are considered for exciting two resonance modes using dual-polarization spectroscopy. These sensors are based on different combinations of a dielectric waveguide and a plasmonic guiding layer. Depending on the overlap of each mode (TE or TM) with the sensing region, each mode exhibits different sensitivity to the measurand. The structural properties of these sensors have been optimized using a genetic algorithm (GA) to ensure optimum performance. Moreover, a method to decouple the interfering surface and bulk effects is presented. The applications of the PWR sensor for refractive index sensing, thin film measurement, and kinetic analysis of proteinnanoparticle interactions are experimentally demonstrated and compared with the conventional SPR sensor. It is demonstrated that for refractive index sensing the TM mode of the PWR sensor has six times smaller refractive index resolution than that of the SPR sensor. To further enhance the functionality of the plasmonic sensors, two grating based designs are proposed which can perform three-mode spectroscopy. The first design contains a one ii

3 dimensional dielectric grating based SPR sensor (DGSPR) and the second grating sensor contains a one dimensional metallic grating loaded on an SPR sensor (MGSPR). Both designs are optimized with a GA to obtain three resonance modes. A detailed analysis of these designs using numerical and analytical methods is presented. The performance of each mode is compared with that of the conventional SPR sensor. Finally, a method to differentiate three interfering effects is provided. iii

4 Acknowledgements First and foremost, I want to express my sincere appreciation and gratitude to my supervisors, Prof. Mojahedi, and Prof. Aitchison for their unwavering support and guidance. I would also like to thank Prof. Ted Sargent, Prof. Gilbert Walker, Prof. John Sipe, and Prof. Li Qian for being on my thesis progress and thesis defence committees and Prof. Andrew Kirk for serving as the external member for my thesis defence committee. Their suggestions have greatly improved the quality of the thesis. Many people have helped me in different ways throughout my PhD work and deserve due appreciation. I would like to thank Muhammad Zulfiker Alam who deserves gratitude for helping me with the editing of the thesis and for his support. I would also like to thank Matheiu Maisonneuve and Prof. Michel Meunier from Ecole Polytechnic de Montreal for their support. I would like to thank the current and former members of the Photonics group of the University of Toronto, especially Jan Niklas Casper, Arthur O. Montazeri, Xiao Sun, Arnab Dewanjee, and Ahmed Dorrah for being such nice friends. Also I would like to thank the Emerging Communications Technology Institute cleanroom staffs for their unlimited technical support. The work done in this thesis would not be completed without the funding support of the NSERC Strategic Network for Bioplasmonic Systems, Ontario Graduate Scholarship, MNT Financial Assistance for Microfabrication and related travel expense, Doctoral Thesis Completion Grant, and University of Toronto Fellowship. I owe my deepest gratitude to my father Houshang Bahrami and mother Mrs. Shahrbanoo Esfahani for their unwavering support throughout my life and especially during the years of my PhD study. I would like to express my sincere appreciation for my wife Shaghayegh Esfahani for being a constant source of inspiration during this journey. Her encouragement brightened the frustrating moments of my work and helped me move on. I am also grateful to my two brothers Arash Bahrami and Sina Bahrami for their support. iv

5 Dedication To my parents who have sacrificed so much for their children and for being an endless source of inspiration and love in my life. and also, To my wife Shaghayegh for being supportive and caring. v

6 Table of Contents Abstract..... ii Acknowledgements......iv List of Tables..... x List of Figures..... xii List of Appendices... xviii List of Symbols and Acronyms.....xix 1 Introduction Objectives and Contributions Thesis Organization Fundamental of Plasmonic Biosensors Surface Plasmon Dispersion Equation Short Range and Long Range SP Optical Excitation of SP Prism Coupling Grating Coupling Waveguide Coupling Surface Plasmon Resonance (SPR) Sensor Surface Plasmon Resonance Biosensors Performance Characteristics Sensitivity Limit of Detection Limit of Quantification Resolution Dynamic Range SPR biosensor limitations vi

7 2.7 Decoupling Interfering Effects: Available Solutions Reference Channel Self-referenced SPR Sensing Review of Available Self-referenced SPR Schemes Conclusion Genetic Algorithm GA Operators GA Selection Operator Roulette Wheel Selection Tournament Selection Basic GA Conclusion Dual-mode Spectroscopy: Hybrid Plasmonic Sensors Hybrid Plasmonic Waveguide (HPWG) HPWG Sensor Optimization Using Genetic Algorithm Results and Discussion Conclusion Dual-mode Spectroscopy: Plasmon Waveguide Resonance Description of the PWR Structure PWR Sensor for Refractive Index Sensing Theory and Modeling Materials and Methods Results and Discussion Temperature Fluctuation Discussion vii

8 5.3 PWR for Self-referenced Spectroscopy Principle of Operation Optimization Sensor Fabrication and Instrumentation Self-referenced Experiments Discussion Probing Protein-nanoparticle Interactions Using a Plasmon Waveguide Resonance Protein-nanoparticle Interaction Principle of Operation Sensors Fabrication and Functionalization Self-referenced Experiments Discussion Conclusion Three Mode Spectroscopy Dual Polarization Spectroscopy: Dielectric Grating Numerical Model and Optimization Results Five Mode Spectroscopy Discussion Dual Wavelength Spectroscopy Using a Metallic Grating Coupled to a Surface Plasmon Resonance Sensor Metallic-grating based SPR Sensor Numerical Model Results and Discussion Discussion Conclusion viii

9 7 Contributions and Conclusions Original Contributions Future Research Directions References or Bibliography Appendix A Appendix B Appendix C Publications ix

10 List of Tables Table 2-1Sensing performance of various SPR sensors Table 4-1- Optimized values for the design parameters obtained from the genetic algorithm code for: single interface SPR and HPWG Table 5-1.Comparison of the optimized PWR and SPR sensors characteristics Table 5-2. Comparison of the optimized PWR and SPR sensors characteristics with considering the limitations in the given setup Table 5-3. The experimental and theoretical sensors characteristics for the fabricated PWR and SPR sensors at λ=632 nm Table 5-4. The experimental and theoretical sensors characteristics for the fabricated SPR sensor at λ=900 nm and the PWR sensor at λ=830 nm Table 5-5. Experimental sensors characteristics calculated from the sensograms shown in Fig Table 5-6 Normalized dissipated power density in the PWR and the SPR sensors Table 5-7. Comparison of the optimized PWR and SPR sensors characteristics Table 5-8. Experimental sensor characteristics obtained from the sensograms shown in Fig Table 5-9. Comparison between experimental and theoretical values of the sensitivity factor s ratio Table Comparison between theoretical wave vector variations due to perturbation of the refractive index close to AuNPs Table 6-1. The optimized dimensions of SPR and DGSPR sensors for affinity sensing Table 6-2. The performance characteristics of the optimized SPR and DGSPR sensors x

11 Table 6-3. The optimized dimensions of DGSPR sensor consisting of a SU8 grating Table 6-4. The performance characteristics of the DGSPR sensor consisting of a SU8 grating. 111 Table 6-5. The optimized dimensions of SPR and MGSPR sensors for affinity sensing Table 6-6. The optimized performance characteristics for single interface SPR and MGSPR sensors xi

12 List of Figures Fig a) Light refraction at the interface of two materials with dielectric constants of ε 1 and ε 2 b) illustration of SP wave on metal surface Fig. 2-2 Dispersion relation for SP propagating at the interface between dielectric and Drude metal with no loss Fig. 2-3 The magnetic field profile for symmetric and asymmetric SP modes Fig. 2-4 (a) Excitation of SP wave by Kretschmann configuration, (b) dispersion relation for SP excitation using Kretschmann configuration Fig. 2-5 SP excitation using surface relief grating Fig. 2-6 Excitation of SP wave by a guided mode of a dielectric waveguide Fig. 2-7 SP wave probing (a) a homogenous medium and (b) a thin film layer Fig. 2-8 Reflectivity and phase for light wave exciting a SPW in the Kretschmann geometry (SF14 glass prism 50 nm thick gold layer dielectric) versus (a) the angle of incidence for two different refractive indices of the dielectric (wavelength 682 nm), and (b) wavelength for two different refractive indices of the dielectric (angle of incidence 54 ) (figure and caption obtained from [11] ) Fig. 2-9 Schematic of capturing the target analyte by the bio-recognition elements (Y) immobilized on the SPR sensor Fig Schematic representation of a sensogram stages Fig Reflectance spectrum for a single interface biosensor of Fig. 2-4 with 55nm gold (n gold =0.18+i5.4) on top of the prism (silica, n silica =1.45) at the wavelength of 828 nm Fig Schematic diagram of the integrated optical SPR sensor immunoprobe showing the binding of antibodies to the surface modified gold film (figure and caption obtained from [52]) xii

13 Fig Dual slab waveguide interferometer. The sensor chip comprises five layers of deposited silicon oxynitride. A window is opened in the final layer to expose the sensing waveguide ( figure and caption obtained from [59]) Fig The steps of a single point crossover operation. In the 1st step the chosen crossover point is shown by the black arrow. The 2nd step shows the child chromosomes created by the interchange of the split portions of the parent chromosomes indicated by the corresponding colors Fig The steps of a uniform crossover operation. The bits of the child chromosomes are selected from the parents according to the mask bits Fig The two major steps of a mutation operation. In the 1st step the randomly chosen mutation bit is shown by the black arrows. The 2nd step shows the mutated bit (from 0 to 1) with a change of color Fig Roulette wheel probabilities for four parents in the mating pool Fig Tournament selection Fig Genetic algorithm flowchart Fig Formation of hybrid plasmonic mode from coupling of dielectric and SP mode (a) Waveguide structure (b) Normalized power density. The coordinate system used is also shown. The xz plane coincides with the gold-silica interface for the HPWG. The dimensions are h=50 nm, d=100 nm. Wavelength of operation is 1.55 μm Fig Comparison of guided power density profile for the SP mode and the hybrid plasmonic mode for the same propagation loss. The dimensions of the HPWG are h = 50 nm, d = 100 nm. Wavelength of operation is 1.55 μm (figure obtained from [72]) Fig. 4-3(a) The proposed HPWG for biosensing in reflection measurement. (b) z-component of the Poynting vector for both TM and TE polarizations (p- and s-polarizations) along the x-axis at the resonance angle. The wavelength of the incident light is 605 nm and silica, silicon, and gold indices at this wavelength are 1.45, 3.93+i0.02, and 0.23+i3.11, respectively. The resonance angle for TM and TE polarizations are 80.76⁰ and 48.27⁰, respectively xiii

14 Fig. 4-4 Fitness function over 140 generations Fig. 4-5 (a)the CSF for both polarizations in the HPWG and the single interface SPR biosensors.(b) variations in SF and SM for TM polarization in the SPR sensor. (c) Variations in SF and SM for TM polarization in the HPWG sensor. (d) Variations in SF and SM for TE polarization in the HPWG sensor Fig. 4-6 Reflectance spectrum for (a) TM polarization in both the single interface SPR and the HPWG biosensor when the bulk index (fluid index) changes from 1.33 to The operating wavelengths for the single interface SPR and HPWG biosensor are 836 nm and 830 nm, respectively. (b) TE polarization in the HPWG biosensor when bulk index (fluid index) changes from 1.33 to Fig. 5-1(a) Schematic diagram of the PWR sensor. (b) Reflectance spectrum for the optimized PWR-TM, PWR-TE, and SPR-TM polarizations in black, red, and blue lines, respectively. (c) z- component of the Poynting vector for both TM and TE polarizations in the optimized PWR sensor at the resonance angle of 62.35⁰ and 66.18⁰, respectively. (d) z-component of the Poynting vector for the TM polarization in the optimized SPR sensor at the resonance angle of 63.77⁰ Fig. 5-2 The field profile of TM polarization for different silica layer thicknesses on top of 50 nm gold film at the wavelength of 800 nm. The silica layer thicknesses are: a) 0 nm, b) 50 nm, c) 100 nm, d) 400 nm, e) 565 nm, and f) 700 nm Fig. 5-3 The field profile of TE polarization for different silica layer thicknesses on top of 50 nm gold film at the wavelength of 800 nm. The silica layer thicknesses are: a) 400 nm, b) 6000 nm, and c) 1000 nm Fig. 5-4(a) The CSF variation along with the incident light wavelength in the optimized SPR sensor. The CSF variation along with the silica thickness at different wavelengths in the PWR sensor for (b) TM and (c) TE polarizations Fig Proposed optical setup for minimum position measurement Fig. 5-6 The images captured from the camera for (a) TM polarized and (b) TE polarized light. 67 xiv

15 Fig. 5-7 The experimental normalized reflectance spectrum at λ=632 nm for (a) TM-polarized PWR sensor. (b) TE polarized PWR sensor. (c) TM-polarized SPR sensor. The theoretical normalized reflectance spectrum at λ=632 nm for (d) TM-polarized PWR sensor. (e) TE polarized PWR sensor. (f) TM-polarized SPR sensor. The different curves refer to reflectance spectrum for different concentrations of ethanol solution, 0.7% (red), 2% (blue). The water spectrum (black) is the reference Fig The experimental normalized reflectance spectrum for (a) TM-polarized (black) and TE-polarized (blue) PWR sensor at λ=830 nm (b) TM-polarized SPR sensor at λ=900 nm. The theoretical normalized reflectance spectrum for (c) TM-polarized (black) and TE-polarized (blue) PWR sensor at λ=830 nm (d) TM-polarized SPR sensor at λ=900 nm Fig Sensors responses to the bulk refractive index variations: (a) resonance angle versus time for the TM-polarized PWR sensor at λ=632 nm, and 830 nm (b) resonance angle versus time for the TE-polarized PWR sensor at λ=632 nm, and 830 nm (c) resonance angle versus time for the TM-polarized SPR sensor at λ=632 nm, and 900 nm. Solutions are based on (1) DI water, (2) 0.5% ethanol, (3) 1% ethanol (4) 2% ethanol Fig Sensors responses to the bulk refractive index variations with baseline adjusted to zero: (a) resonance angle versus time for the TM-polarized PWR sensor at λ=632 nm, and 830 nm (b) resonance angle versus time for the TE-polarized PWR sensor at λ=632 nm, and 830 nm (c) resonance angle versus time for the TM-polarized SPR sensor at λ=632 nm, and 900 nm. Solutions are based on (1) DI water, (2) 0.5% ethanol, (3) 1% ethanol (4) 2% ethanol Fig (a) Optical setup used for reflectance measurement. (b) The experimental (black line) and theoretical (red line) normalized reflectance spectrum of the PWR sensor Fig (a) Angular positions of the resonance dip vs. time for the SPR sensor. (b) Angular positions of the resonance dip vs. time for the PWR sensor. Solutions are (1) PBS, (2) 1μg/mL Streptavidin, (3) PBS, (4) 10 μg/ml Streptavidin, (5) PBS, (6)DI water, (7) 0.01M salted water, (8) DI water, (9) 1% ethanol, and (10) DI water Fig Surface binding thickness and bulk refractive index change calculated from Fig. 5-11(b). Solutions are (1) PBS, (2) 1μg/mL Streptavidin, (3) PBS, (4) 10 μg/ml xv

16 Streptavidin, (5) PBS, (6)DI water, (7) 0.01M salted water, (8) DI water, (9) 1% ethanol, and (10) DI water Fig (a) Schematic diagram of the AuNP-modified PWR sensor. (b) z-component of the Poynting vector for both TM and TE polarizations in the optimized PWR sensor at the resonance angle of 61.65⁰ and 65.23⁰, respectively. (c) Reflectance spectrum for the optimized PWR-TM, PWR-TE in black and red lines, respectively Fig (a) Schematic diagram of the BK7 glass substrate with 48 nm of gold film. (b) Deposition of 550 nm silica on the gold film using PECVD. (c) The PWR sensor coated with a SAM of APTMS. (d) Binding Au NPs on the SAM of APTMS Fig SEM images of gold nanoparticles on the PWR surface with diameters of (a) 20 nm with surface density of particles/cm 2 (b) 100 nm with surface density of particles/cm Fig (a) Proposed optical setup for minimum position measurement. (b) The experimental (black line) and theoretical (red line) normalized reflectance spectrum of the PWR sensor Fig Angular positions of the resonance point vs. time for (a) 20 nm and (b) 100 nm AuNP binded to the PWR sensor. Solutions are (1) PBS, (2) 1μg/mL Streptavidin, (3) PBS, (4) 10 μg/ml Streptavidin, (5) PBS, (6)DI water, (7) 1% ethanol, and (8) DI water Fig Decoupled surface binding change and bulk index variations as a function of time for (a) 20 nm AuNP-PWR sensor (b) 100 nm AuNP-PWR sensor. Solutions are (1) PBS, (2) 1μg/mL Streptavidin, (3) PBS, (4) 10 μg/ml Streptavidin, (5) PBS, (6)DI water, (7) 1% ethanol, and (8) DI water Fig. 6-1 (a) three dimensional (3D) schematic of the DGSPR sensor. (b) 2D cross section of the DGSPR sensor. Silica nanowires of a rectangular profile are assumed to be infinite in length with periodicity of Λ, thickness of h and filling factor of w/λ Fig. 6-2 Size-exclusion SPR sensor chip with microfabricated slit array (Filter SPR chip) for separating blood cells and detecting proteins in whole blood sample. (figure and caption obtained from [123]) xvi

17 Fig. 6-3(a) Reflectance spectrum for the optimized DGSPR sensor (b) z-component of the Poynting vector for both TM and TE polarizations at the resonance angle of 62.28⁰, 71.11⁰, 63.5⁰ for TM1, TM2,and TE1 mode respectively Fig. 6-4 z-component of the Poynting vector along DGSPR cross section (x-z plane) for (a) TM1 mode, (b) TM2 mode, and (c) TE1 mode Fig. 6-5 Dispersion relation of optimized DGSPR sensor calculated using: (a) RCWA method (b) transfer matrix method (c) Analytically calculated dispersion relation of the SP wave in the optimized DGSPR sensor Fig. 6-6 The tolerance of TE CSF bulk and TM CSFthick to the variation in (a) duty cycle (b) grating thickness (c) gold film thickness (d) periodicity of the DGSPR sensor Fig Reflectance spectrum for the optimized DGSPR sensor consisting of a SU8 grating with (a) 2μm height (b) 1μm height Fig (a) Three dimensional (3D) schematic of a MGSPR sensor. (b) 2D cross section of the MGSPR sensor. Gold nanowires of a rectangular profile are assumed to be infinite in length with periodicity of Λ, thickness of h and a filling factor of w/λ Fig Diffraction efficiency of the reflected light with TM polarization for the optimized MGSPR sensor at the incident wavelength of (a) 845 nm (b) 970 nm Fig (a) Dispersion relation of the MGSPR sensor calculating using RCWA. (b) Analytically calculated dispersion relation of the SP wave in the optimized MGSPR sensor using Eq Fig The tolerance of CSF bulk and CSFthick for all resonance modes of the MGSPR sensor to the variation in (a) duty cycle (b) periodicity(c) gold film thickness (d) grating height xvii

18 List of Appendices Appendix A: Reflection, transmission, and field profile in multilayer structures 139 Appendix B: Rigorous coupled-wave analysis Appendix C: Parallel computing: message passing interface (MPI). 151 xviii

19 List of Symbols and Acronyms ω : Angular frequency of the light n a : Adlayer refractive index d a : Adlayer thickness CSF : Combined Sensitivity Factor CSF bulk : Combined sensitivity factor for bulk index variations CSF surf,index : Combined sensitivity factor for adlayer index variations CSF surf,thick : Combined sensitivity factor for adlayer thickness variations n b : Bulk refractive index DGSPR : Dielectric Grating based Surface Plasmon Resonance DPI : Dual Polarization Interferometer EMT : Effective Medium Theory E : Electric Field σ : Electric conductivity ε : Electrical permittivity FoM : Figure of Merit FWHM : Full Width at Half Maximum GA : Genetic Algorithm AuNP: Gold nanoparticle xix

20 Λ : Grating periodicity HPWG : Hybrid Plasmonic Waveguide LoD : Limit of Detection LRSP : Long Range Surface Plasmon MGSPR : Metallic Grating based Surface Plasmon Resonance H : Magnetic Field NP : Nanoparticle n p : Prism refractive index ω p : Plasma frequency PWR : Plasmon Waveguide Resonance RIU : Refractive Index Unit RCWA : Rigorous Coupled Wave Analysis σ RI : Refractive index uncertainty SP : Surface Plasmon SPP : Surface Plasmon Polariton SPR : Surface Plasmon Resonance SF : Sensitivity Factor SF bulk : Sensitivity factor to bulk refractive index change SF surf,index : Sensitivity factor to change in adlayer index SF surf,index : Sensitivity factor to change in adlayer thickness xx

21 SM : Sensor Merit SNR : Signal Noise Ratio σ SO : Standard deviation of the noise at the sensor output k SP : Surface plasmon wavevector θ SP : Surface plasmon resonance angle c : Speed of light TM : Transverse Magnetic TE : Transverse Electric TIR : Total Internal Reflection λ : Wavelength xxi

22 1 1 Introduction Surface plasmon resonance (SPR) sensors have received much attention in the last decades due to their unique features such as high field confinement and high sensitivity [1-6]. The SPR is a resonant coupling between an incident electromagnetic wave and the collective oscillations of the free electrons, also called surface plasmon (SP), at a metal-dielectric interface. Due to the high field concentration of the SP wave at the metal-dielectric interface, the SPR is extremely sensitive to the changes that occur at the interface within the electromagnetic field s penetration depth. This feature has been applied for different sensing applications including drug screening, medical diagnosis, food safety, bio-technology, and environmental protection [6]. One of the most important applications of the SPR sensors is as an affinity biosensor where biorecognition elements on the surface of the metal capture the target analyte present in an aqueous solution and cause changes in the refractive index at the vicinity of the metal surface. This refractive index change consequently alters the propagation constant of the SP wave which can be determined through monitoring its optical characteristics. Hence, the SPR biosensor does not need labeling the target analyte in contrary to the fluorescence-based detection which has high sensitivity but suffers from laborious labeling process [7]. The SPR sensors are based on labelfree detection which is relatively easy to perform and allows for quantitative and kinetic measurements of surface reactions. The excitation of SP wave using diffraction gratings was first described in the beginning of the twentieth century by Wood [8] and in the late sixties the optical excitation of SP with attenuated total reflection was demonstrated by Kretschmann [9] and Otto [10].The first commercial SPR biosensor was launched by Biacore International AB in 1990 [11]. Since then the SPR sensors has been widely used for different applications. Although the SPR sensors are widely used for affinity sensing, the information obtained by this technique is limited [12]. In affinity sensing, detecting the interaction and binding of a thin biomolecular layer to the sensing surface within a few nanometers is of interest. However since the electromagnetic field of the SP wave extends beyond the biomolecular layer and interacts

23 2 with the analyte solution flowing over this layer, any variation in the bulk analyte index of refraction can alter the SP wave propagation constant. This cross sensitivity to both the surface and bulk refractive index variations interferes with reliable measurements. Using a reference channel beside the sensing channel is the most common solution to mitigate the cross sensitivity to surface and bulk effects. By subtracting the reference channel signal from the sensing channel signal, the effect of bulk fluid index variations can be decoupled from the surface bindings. However this method requires a more complex system to guarantee identical conditions on both channels in terms of the temperature fluctuations, solution concentrations, non-specific binding, and material properties. Therefore both channels should be exactly identical with the exception of the one measurand being measured. Also the structural information (thickness and density) of the bio-molecular layer is not measurable through this technique and only the optical density (the product of thickness and refractive index of biomolecular layer) can be measured. Therefore new methods to alleviate the cross sensitivity of the plasmonic sensors to interfering effects needs to be explored. 1.1 Objectives and Contributions The objective of this dissertation is to propose new designs and optimize them to address the cross sensitivity to the interfering surface and bulk effects. The optimized designs are developed for multimode spectroscopy and used as a self-referenced scheme to differentiate the surface binding from the background refractive index variations. The performance of the proposed designs are evaluated and compared with the conventional SPR sensor. The followings are accomplished in this work: 1) Two different planar designs are proposed for dual-polarization spectroscopy using a prism coupling: a) hybrid plasmonic waveguide (HPWG) sensor, and b) plasmon waveguide resonance (PWR) sensor. Both of these designs are a combination of dielectric and plasmonic waveguides. In the HPWG sensor, the dielectric guiding layer is

24 3 located under a metallic film whereas the sensing reactions occur on top of the metallic surface. However in the PWR sensor the dielectric guiding layer is located above the metallic film and the surface reactions occur on top of the dielectric guiding layer. Each sensor supports two resonance modes at different polarizations which are used to decouple variations in the thickness of the bio-molecular layer from the bulk index variations. Such polarization diversity can also be used in the study of anisotropic materials. These sensors are optimized to acquire their best performance in terms of the limit of detection, while each possesses unique features which will be discussed and compared with the conventional SPR sensor. The PWR sensor is experimentally demonstrated for refractive index sensing, thin film measurement and studying proteinnanoparticle interactions. 2) Two designs are proposed for three-mode spectroscopy based on a combination of planar and periodic structures in order to differentiate the variations in the thickness and the refractive index of the attached bio-molecular layer from the bulk index change. In the first design, a dielectric grating based SPR (DGSPR) sensor is optimized for dualpolarization spectroscopy at a fixed wavelength. The second design is composed of a metallic grating based SPR sensor (MGSPR) which is optimized in order to efficiently excite multiple diffraction orders for dual-wavelength spectroscopy. In both designs, the optimized FoM is compared with that of the optimized SPR sensor. Finally in order to justify our results, numerical and analytical calculations are presented. 1.2 Thesis Organization The organization of the dissertation is as follows. In Chapter 2, a brief introduction to SPR sensors is presented. The main characteristics of the SPR sensors and their most common excitation mechanism are explained. A number of challenges for SPR sensing which need to be addressed are identified and the available solutions to overcome these challenges are introduced which form the central theme of this dissertation.

25 4 In Chapter 3, a Genetic Algorithm (GA) optimization, which is the optimization method used in this work, is briefly presented and the most common operators of GA have been discussed. Also, different steps of a simple GA are explained. In Chapter 4, a HPWG sensor is investigated for self-referenced spectroscopy in a Kretschmann configuration. The polarization diversity of the HPWG is employed for differentiation of surface bindings from the bulk refractive index change. The performance of the optimized HPWG sensor is compared with the conventional SPR sensor. In Chapter 5, a PWR is implemented for different applications. First, the performance of the PWR sensor for refractive index sensing application is experimentally demonstrated which outperforms that of the SPR sensor. Subsequently, it is shown that the optimized PWR sensor can be applied as an affinity sensor for self-referenced protein layer measurement. Its performance is also compared with the optimized SPR sensor in terms of the limit of detection and sensor s resolution. Finally, it is demonstrated that the PWR sensor can be used to analyze the binding kinetics of protein-nanoparticle interactions. In Chapter 6, a DGSPR is investigated for three-mode spectroscopy in order to decouple the interfering surface (thickness and refractive index of the attached biolayer) and bulk (bulk fluid refractive index) effects using dual-polarization spectroscopy. The structural properties of the design are optimized using a GA. The simulated results are obtained and compared with the optimized SPR sensor. In the second part of the chapter, a MGSPR sensor is optimized for threemode spectroscopy at two different wavelengths. Lastly, the simulation results and an approach to decouple the interfering effect are shown. We present a summary of results and contribution along with the suggestions for future research directions in chapter 7.

26 5 2 Fundamental of Plasmonic Biosensors Surface plasmon (SP) wave is the collective electron oscillations that are confined at the interface of two materials. Maxwell equations require that one of these materials should be a metal where free conduction electrons are abundant and the second material should be a dielectric. The optical properties of materials that lie within the immediate vicinity of the metal film can influence the SP wave generation due to the high field strength at the interface. Light can be coupled to the SP wave under certain conditions and the variations in the refractive index at the metal-dielectric interface alter the coupling condition due to the change in SP propagation constant. The resonance condition, under which light couples to the SP wave, when the light s wavevector matches that of the SP, is called surface plasmon resonance (SPR). The coupling state between SP and electromagnetic wave at the interface of metal-dielectric is called surface plasmon polariton (SPP). SPP is commonly known as SP wave that propagates at metaldielectric interface. The change in coupling condition can be observed by monitoring specific characteristics of light which can be categorized into: angle, wavelength, intensity, phase, and polarization. In this chapter, we present a brief review of the fundamental of SP and its application in biosensing. 2.1 Surface Plasmon Dispersion Equation The dispersion relation of SP, a relation between the angular frequency ω and the wavevector k, helps us better understand the coupling mechanism between light and SP. The SP dispersion relation was calculated from Maxwell s equations by Raether [13]. Maxwell s equations for electromagnetic wave propagating in a non-magnetic and source-free media can be written as:

27 6 H E 0. t. H 0. 0 E H 0. t. E 0. 0 Eq. 2-1 Eq. 2-2 Eq. 2-3 Eq. 2-4 where E is the electric field, H is the magnetic field, µ 0 is the permeability of the free space, ε 0 is the permittivity of the free space, and ε is the relative permittivity of the material. Wave equations can be obtained from the Maxwell s equations according to: E t 2 2 E Eq. 2-5 H t 2 2 H Eq. 2-6 The electromagnetic field propagating at the interface of two media [Fig. 2-1(a)] is the solution of the Maxwell s equation which can be expressed as: E E exp[ i( k. r t)] E exp[ i( k. x k. z t)]. Eq x z where k is the wavevector. The coordinate definition for this analysis is illustrated in Fig. 2-1(b).

28 7 Fig a) Light refraction at the interface of two materials with dielectric constants of ε 1 and ε 2 b) illustration of SP wave on metal surface. The confinement condition of the SP wave at the metal-dielectric interface is only fulfilled for transverse magnetic (TM) polarized light [14]. Therefore, here we only derive the dispersion relation for TM polarized light. The boundary condition at the interface of two media implies that the tangential component of the magnetic field (H y ) and the normal component of the electric flux density (ε i E z ) should be continuous at the interface. By substituting the electric field expression (Eq. 2-7) into wave equations (Eq. 2-5 and Eq. 2-6) and applying the boundary conditions, the dispersion relation can be derived as: kx 1 kx Eq ( 2 k ). z kx1 1 c Eq ( 2 k ). z kx2 2 c Eq. 2-10

29 8 where c is the speed of light, ε 1 and ε 2 are the dielectric constants of the dielectric and metal, respectively, and ω is the angular frequency of the light. The dispersion relation for the tangential component of the wavevector at the interface can be solved from Eq. 2-8 to Eq. 2-10: k k k ' ik ". 1 2 sp z sp sp c 12 Eq Here k sp is the SP propagation constant, k z is the tangential component of the wavevector (along z-axis).the expression for the normal component of the wavevector (along x-axis) is given by Eq [15]: kxi c 2 i, i1,2 1 2 Eq The confinement of the SP wave at the metal-dielectric interface implies that the SP wave should be evanescent (decays exponentially) at both sides of the interface. This condition is satisfied only when k xi is imaginary which requires ε 2 > ε 1 when ε 2 is negative and ε 1 is positive. The penetration depth of the SP wave into each medium corresponds to the distance during which the wave amplitude reduces to 1/e and is given by 1/k xi. Several metals (e.g. gold and silver) have large negative dielectric constant at optical wavelengths. Assuming lossless metal, Drude model can be used to describe the lossless metal permittivity as: 2 2 p 2 ( ) 1. Eq where ω p is the so-called plasma frequency. According to this equation when the light frequency is smaller than the plasma frequency (ω<ω p ) the metal dielectric constant becomes negative. Therefore metallic surface is one key element of the SP wave propagation. A graphical representation of the SP dispersion relation is illustrated in Fig. 2-2 which is obtained by substituting Eq into Eq The dispersion relation for the component of the light that is parallel to the metal-dielectric interface incident from the dielectric region (ε 1 ) is also depicted as a dashed line. At small angular frequencies, the SP dispersion curve is close to the light line but by increasing the angular frequency the difference between the light and the SP dispersion

30 9 becomes larger and the SP wavevector approaches infinity when the angular frequency approaches a certain point which is referred to the SP frequency ( sp p ). 1 1 p 1 1 c kz 1 k z Fig. 2-2 Dispersion relation for SP propagating at the interface between dielectric and Drude metal with no loss. From the dispersion curve we can see that there is no intersection between the parallel component of the light and the SP dispersion curve which indicate that the incident light cannot provide the correct wavevector to couple to the SP wave. Therefore the difference between the SP wavevector and the light wavevector should be supplied by another mechanism. In the next section, different methods for coupling the external light momentum to the SP wave are briefly discussed.

31 Short Range and Long Range SP The SP wave propagation at a single metal-dielectric interface [Fig. 2-1(a)] exhibits interesting properties such as high intensity and subwavelength confinement near its energy asymptote (at p ). But, it also suffers from high attenuation especially near the energy asymptote 1 1 which is mainly caused by free-electron scattering in the metal at short wavelengths. Also surface roughness along the interface causes further attenuation. The propagation distance of the SP is defined as the distance over which the power drops to 1/e of its initial value which is given by: SP 1 ". 2 k sp Eq In order to reduce the attenuation of the SP wave, one can use a thin metal film (tens of nanometer in thickness) bounded on both sides by the same dielectric which results in interaction between SP waves supported by two metal-dielectric interfaces to form coupled modes. These modes are denoted as symmetric and asymmetric since the main transverse component of their electric field varies symmetrically or asymmetrically across the structure as is illustrated in Fig The symmetric mode has most of its power in the surrounding dielectric medium and hence its attenuation is significantly reduced. This mode is called long range SP (LRSP) mode [16]. The asymmetric mode is more confined to the interface and therefore has larger penetration into the metal which results in more attenuation. The asymmetric mode is referred to as a short range SP mode. Asymmetric SP Symmetric SP Dielectric Metal Substrate Fig. 2-3 The magnetic field profile for symmetric and asymmetric SP modes

32 Optical Excitation of SP In order to efficiently excite the SP wave at a given frequency, the wavevector of the light coming from the outside must be matched to the wavevector of the SP wave (k SP ). This is achieved via several different techniques which are described here Prism Coupling One of the most common techniques to excite SP wave is prism coupling. In this technique, the additional required momentum is provided by an evanescent wave resulting from the total internal reflection (TIR) of the incident TM polarized light. Two common methods for prism coupling have been proposed [17]. In the first configuration, which is called Kretschmann configuration, a thin metallic film is sandwiched between two dielectric materials 1 and 3 with different dielectric constants ε 1 and ε 3 [Fig. 2-4(a)]. The second configuration is Otto configuration in which there is a small gap (a few hundred nm) between the metallic film and the prism. Since the Kretschmann configuration is the most common technique for SPR excitation, its excitation mechanism will be discussed here.

33 12 p 1 1 c 1 kz 3 1 c kz 3 x p 1 3 (a) z (b) k z Fig. 2-4 (a) Excitation of SP wave by Kretschmann configuration, (b) dispersion relation for SP excitation using Kretschmann configuration. In Kretschmann configuration, for each metal-dielectric interface a dispersion relation can be obtained as shown in Fig. 2-4(b). From Fig. 2-4(b) we can see that the dispersion relation for the parallel component of the light propagating in prism (the blue dashed line) intersects the SP dispersion curve for metal-dielectric interface (the red solid curve). This indicates that light incident from the prism can excite the SP, at metal-dielectric (ε 1 ) interface, for a proper incident angle. This angle can be found from the intersection of the light line and SP dispersion relations which is denoted by the cross sign in Fig. 2-4(b). The following equation shows the matching condition between the light line in prism and the SP wavevector propagation at metal-dielectric (ε 1 ) interface: k k n sin( ) light 0 p 1 2 k 1 2 0np SP k0 ksp k sin( ). Eq where k 0 is the free space wavenumber, n p is the prism refractive index ( 3 ), θ is the incident angle and θ SP is the resonance angle corresponding to the SP wavevector.

34 Grating Coupling Gratings can also be used to match the momentum of the incident light and that of the SP wave as shown in Fig The grating will increase the component of the wavevector parallel to the interface (k z ) by an amount which is inversely proportional to the period of the grating [18]: 2 kg kz m. Eq where m is the diffraction order (which can have positive or negative signs), and Λ is the grating period. The diffracted light can couple to the SP wave when k G is equal to the wavevector of the SP (k SP ). TM Polarized Light θ Dielectric (ε 1 ) SP Wave Λ Metal (ε 2 ) Fig. 2-5 SP excitation using surface relief grating The coupling condition can be expressed by the following equation: 2 k k m k k 1 2 G z Re{ 0 } Re{ SP}. 1 2 Eq. 2-17

35 Waveguide Coupling Waveguide coupling is another method used to excite SP waves [19]. The propagating guided mode of a dielectric waveguide has associated evanescent fields penetrating outside the guiding layer. When this wave passes the region that is covered by a thin metallic film, the evanescent wave can couple to the SP wave (Fig. 2-6). Assuming that kg is the wave vector of the guided mode, and the propagation constant of the SP wave is not affected by presence of the guiding layer, the coupling condition is expressed according to: k g Re k. Eq SP Dielectric (ε 1 ) Guiding Layer Guided Mode Metal (ε 2 ) Fig. 2-6 Excitation of SP wave by a guided mode of a dielectric waveguide. 2.3 Surface Plasmon Resonance (SPR) Sensor Propagation of the SP wave at a metal-dielectric interface can be used to probe variations in the dielectric medium due to SP field penetration into the dielectric region. Any variations in the optical properties of the dielectric layer results in a change in propagation constant of the SP wave. Optical sensors which use SP wave to measure refractive index variations are often referred to as SPR sensors. The variations in optical properties of the dielectric layer and its

36 15 relation with the propagation constant of the SP wave can be explained using the perturbation theory [20]. Based on this theory, any refractive index change (Δn) within the penetration depth of the SP wave induces change in the real part of the propagation constant according to the following equation: Re{ k } k n. Eq SP 0 Depending on how the refractive index perturbation occurs, the SPR sensors can be categorized into two different types (Fig. 2-7). The first type is when the SPR sensor is used to detect a homogeneous change in the refractive index occurring within the whole extent of the field inside the dielectric region (also called bulk refractive index change) which is illustrated in Fig. 2-7(a). This kind of sensor is called bulk sensor. The second type is when the SPR sensor is utilized to detect a surface refractive index change occurring within a short distance from the metal surface which is much smaller than the penetration depth of the SP wave [Fig. 2-7(b)]. An example of this kind of sensor is affinity sensor. Dielectric (a) Dielectric (b) Thin film Metal Metal Fig. 2-7 SP wave probing (a) a homogenous medium and (b) a thin film layer. Based on which light characteristic is used to measure the variations in SPR propagation constant, the SPR sensors can be classified as angle, wavelength, intensity, phase, or polarization modulation-based sensors. a) In angular modulation [21], the coupling strength between the tangential component of incident light and SP wavevector at different angles and a fixed wavelength is investigated that can also determine the resonance angle which yields the strongest

37 16 coupling [Fig. 2-8(a), upper plot]. b) In wavelength modulation [22], the coupling strength is measured at different wavelengths and at a fixed angle and the resonance wavelength determines the strongest coupling [Fig. 2-8(b)]. c) In intensity modulation [23], the change in light intensity which couples to the SP wave at a fixed wavelength and angle of incidence is measured [Fig. 2-8(b)]. d) In phase modulation [24], the shift in phase of the light at a fixed wavelength and angle is measured [Fig. 2-8(a), lower plot]. e) In polarization modulation [25], variations in light polarization at a fixed wavelength and angle are measured. Fig. 2-8 Reflectivity and phase for light wave exciting a SPW in the Kretschmann geometry (SF14 glass prism 50 nm thick gold layer dielectric) versus (a) the angle of incidence for two different refractive indices of the dielectric (wavelength 682 nm), and (b) wavelength for two different refractive indices of the dielectric (angle of incidence 54 ) (figure and caption obtained from [11] )

38 Surface Plasmon Resonance Biosensors Most SPR sensors are based on measuring variations in SPR propagation constant due to adsorption of an analyte by a transducing medium which results in a change in the transducer optical properties. This phenomena can be used for different applications such as gas [26], and biological sensing [7]. In SPR biosensors, the transducer layer is a biomolecular recognition element which is immobilized on the metal surface and can interact with selected analyte(s) as depicted in Fig When a solution containing the analyte molecules is brought into direct contact with the SPR sensor, the target analyte molecules bind to the recognition layer and produce a change in the refractive index at the sensor surface. This change in surface refractive index will consequently modify the propagation constant of the SP wave which can be measured by monitoring the output light characteristics. Flow of sample Target analyte Bio-recognition elements Metal Film Glass Fig. 2-9 Schematic of capturing the target analyte by the bio-recognition elements (Y) immobilized on the SPR sensor The application of SPR biosensor was first demonstrated by Liedberg et al.[27]. SPR sensors are one of the most popular classes of biosensors. The main feature of SPR biosensors is their labelfree detection which means that the binding between the analyte and the bio-recognition element

39 18 can be observed directly without any labeling such as fluorescent molecules. The rapid response of the SPR sensor which provides real time analysis of the binding events is another advantage of the SPR sensors. In general, detection of any analyte is possible with SPR sensors as long as the bio-recognition element to recognize the analyte is available. The real time analysis of SPR response provides a signal vs. time, called sensogram. The measured signal is the light characteristic (resonance angle, resonance wavelength, etc.) which is used to monitor the variations in SPR propagation constant. The different stages of a binding event can be visualized and evaluated from the sensogram (Fig. 2-10). A stable baseline is created by running a buffer solution through the flow system. It is necessary to have a stable baseline before the capturing event begins. The sensing surface is covered with bio-receptors which are ready to capture target analyte. The association phase dominates the sensogram upon injection of the analyte solution. At this stage, the analyte molecules bind to the bio-receptors immobilized on the sensor. However, some of the bound molecules start dissociating during injection. Other components of the sample might also non-specifically adhere to the bioreceptors. The association phase reaches steady state after a certain injection time, where binding and dissociating molecules are in equilibrium. By stopping the analyte injection and running the buffer solution, non-specifically bound molecules are flushed away and only pure dissociation phase is visible from the sensogram. This step enables studying the kinetics of the dissociation process. In order to reach the base line a regeneration step is necessary which breaks the specific binding between the analytes and bio-receptors, while the bio-receptors remain on the sensor. Buffer solution is again injected for the next cycle and also to make sure that the baseline level goes back to the initial position. Determination of kinetics of bio-molecular interactions is one of the most important features of the SPR technology. The association and dissociation rate constants (k a, k d ) can be obtained from the sensogram. The thermodynamic information of the binding process can also be obtained from the kinetic experiment.

40 19 buffer association dissociation regeneration buffer Time Fig Schematic representation of a sensogram stages 2.5 Performance Characteristics The performance of the SPR sensors can be described by different parameters including sensitivity, limit of detection, limit of quantification, resolution, and dynamic range. To make a qualitative/quantitative analysis of the biosensor s performance and also to make a comparison between different designs, choosing a proper figure of merit (FoM) is necessary. In this section, the definition of each characteristic s parameter is discussed and a FoM based on the limit of detection (LoD) is developed Sensitivity Sensitivity is defined as the ratio of the change in the sensor output (angle of incidence, phase, intensity, polarization, or wavelength) to the change in measurand. For instance in angular

41 20 modulation the resonant angle is the sensor s output and therefore change in the resonance angle with respect to the change in measurand defines the sensitivity or more precisely the sensitivity factor (SF). In the case of SPR sensors for which determining the variations in the index of bulk or adlayer is essential the SF is defined as: Y Eq SF. X Here Y is the sensor output, and X is the measurand. What constitutes a measurand, of course, depends on the particular usage of the biosensor. For example, if the analyte concentration in the bulk solution is the quantity of interest, the measurand is the bulk refractive index (n b ); therefore, the sensitivity or SF for this application is given by: SF bulk res deg.. n RIU b Eq and is referred to as the bulk SF. On the other hand, if binding of analyte molecules to the sensing surface is of interest, one can model the bounded molecules as a homogeneous adlayer of thickness d a and refractive index n a. If change in the thickness of the adlayer is considered as the measurand, then the SF is referred to as the surface thickness SF and is given by: SF surf, thick res deg.. d nm a Eq Whereas, if change in the index of refraction of the adlayer (n a ) is considered as the measurand, the sensitivity or SF is referred to as the surface index SF and is given by: SF surf, index res deg.. n RIU a Eq The refractive index (n a ) for most standard organic biomaterials can be obtained from references[28, 29]. Therefore, for many applications the refractive index of the adlayer is a known variable and only the thickness (d a ) is to be determined [12, 30]. Although SF is an important parameter which should be considered in the design of a biosensor, it is not the only important factor. By equating the biosensor s FoM to the SF only a practice,

42 21 unfortunately not uncommon one can come to the conclusion that by broadening the reflection spectrum, for example by increasing the intrinsic loss [31], one can increase the SF and hence obtain a better value for the sensor s FoM. However, optimizing a sensor based on the SF only also results in decreasing the accuracy by which the resonance angle can be resolved [32]. To remedy the situation another factor should be included in the biosensor s FoM definition. This quantity is referred to as the sensor merit (SM) and in effect is a measure of the intrinsic loss for the sensor [33]. In angular modulation scheme, SM is related to the width and the depth of the resonance dip in reflection spectrum as shown in Fig. 2-11, and is given by: SM R R min FWHM max. Eq where R max is the maximum and R min is the minimum reflectance (see Fig. 2-11). The sensor FoM which is used in this dissertation to evaluate our sensors performance is then the product of the SF and SM also called Combined Sensitivity Factor (CSF) and is given by: CSF FoM SF SM. Eq The reflectance of the TM polarized incident light with angular modulation is calculated for a Kretschmann configuration (Fig. 2-4) and is shown in Fig When energy is transferred from the incident photons to the SP, a sharp minimum in the reflectance spectrum is observed.

43 22 Fig Reflectance spectrum for a single interface biosensor of Fig. 2-4 with 55nm gold (n gold =0.18+i5.4) on top of the prism (silica, n silica =1.45) at the wavelength of 828 nm Limit of Detection The limit of detection (LoD) is another important factor which describes the performance of a biosensor. The LoD is defined as the smallest variation in the measurand that can be detected by a reasonable certainty. In other words, smaller the LoD, more accurately the sensor can detect small amounts of analyte (measurand). The ability to detect the smallest change in the measurand is constrained by the standard deviation of the noise at the sensor output (σ SO ). The LoD is related to σ SO, and SF according to [6]: 3 so LoD. SF Eq The standard deviation of the sensor output noise in reflection measurement is given by: so K th FWHM., R R N max min Eq. 2-27

44 23 Here σ th is the noise intensity at the threshold (FWHM), N is the number of the detectors over which the output signal is averaged, K is a constant which depends on the noise type (additive, shot noise, or others), and FWHM is the full width at half maximum as shown in Fig From Eq it is clear that a larger SNR translates to a smaller LoD. By substituting Eq. 2-24, Eq. 2-25, and Eq in Eq. 2-26, the LoD can be written as: 3 so 3K th FWHM 3K th 1 3K th 1 LoD. SF SF R R N N SF SM N CSF max min Eq Equation 2-28 shows the CSF and LoD are inversely proportional; i.e. a larger CSF ensures a smaller LoD. Moreover, the LoD is composed of two parts: 1) The first part (CSF -1 ) depends on the sensor design itself. 2) The second part depends on the setup s components (what is around the sensor) and their corresponding noise (K). Therefore, from these analyses it is clear that neither SF nor SM, but the CSF (SF SM) which is also inversely proportional to the LoD is the proper FoM when designing or comparing the performance of various sensors in terms of their LoD. Berini also uses a comparable definition for the LoD in a Mach-Zehnder interferometer [34]; except for the fact that he uses extinction coefficient (k) instead of the FWHM [which appears in our Eq. 2-27]. The reason for this difference is that Berini considers phase modulation whereas we are investigating angular modulation. Typically there are two ways to specify LoD [7]. The first way is to use surface mass coverage in units of pg/mm 2 and is basically what a biosensor actually measures. However it is difficult to experimentally determine surface mass surface coverage accurately. The second way is to use sample concentration in units of ng/ml or molarity which is quite useful and easy to determine experimentally, as no information about surface mass density is needed. However, the later approach depends on the affinity of the bio-recognition elements to the target molecules hence the type of biomolecules should be specified.

45 Limit of Quantification To quantify analyte concentration, another performance characteristics called the limit of quantification (LoQ) is used. Based on the definition of the LoQ, the analyte quantification is accepted when the amount of measurand is equal to 10 standard deviation of the output signal standard deviation (σ SO ). Thus the LoQ is given by the following equation: 10 LoQ SF SO. Eq Resolution The sensor resolution, also called the refractive index uncertainty (σ RI ), depends on the sensor output noise (σ SO ) and the bulk refractive index sensitivity (SF bulk ) and can be expressed as: RI SF so bulk. Eq The term resolution is equivalent to the LoD when measurand is the bulk refractive index and is referred to as the bulk refractive index resolution. It is defined as the smallest change in the bulk refractive index which can be detected by reasonable certainty and is given in units of refractive index unit (RIU) Dynamic Range Dynamic range is defined as the range of values of the measurand that can be measured by the sensor. For instance in refractive index sensing application, the dynamic range of the sensor is the range of changes in refractive index of the solution which can be detected by certain accuracy. In SPR affinity biosensing, dynamic range is the range of mass surface coverage or

46 25 concentrations of the solution which can be detected with specific accuracy and it extends from the LoD of the sensor. The sensing performance of existing SPR sensors for several excitation mechanisms is tabulated in Table 2-1 in terms of sensitivity, resolution and LoD.

47 26 Table 2-1Sensing performance of various SPR sensors Technology platform Analyte SF LoD Resolution Reference Intensity modulation Bulk solution N.A. N.A RIU [35] Intensity modulation Oligonucleotides 210 RIU pM RIU [36] Angular modulation Bulk solution N.A. N.A RIU [37] Angular modulation Bulk solution N.A ng ml - 1 N.A. [38] Angular modulation PSA N.A. 50 ng ml -1 N.A. [39] Wavelength modulation Protein N.A. N.A RIU [40] Wavelength modulation Bulk solution nm RIU -1 N.A RIU [41] Wavelength modulation Bulk solution molecules/cm 2 10pM N.A. [42] Phase Modulation DNA N.A. N.A RIU [43] Phase Modulation Bulk solution N.A. 1.3nM RIU [44] Grating Coupling Protein/bulk solution 365nm/RIU N.A RIU [45] Waveguide Coupling Bulk solution 1100nm/RIU N.A RIU [46]

48 SPR biosensor limitations In SPR affinity biosensors the goal of the measurement is to ascertain information about a thin biomolecular layer called adlayer which is formed as the analyte solution flows over the sensing surface. The desired information is often in the form of adlayer thickness (d a ) and/or index of refraction (n a ). The information about the adlayer thickness (d a ) and/or index (n a ) is derived from the sensor s output which unfortunately also depends on other parameters such as variations in the analyte index of refraction (n b ). This latter dependency arises from the fact that electromagnetic field of the SP wave extends beyond the adlayer and interacts with the analyte solution flowing over the adlayer. Thus any variation in the bulk analyte index (n b ) due to variations in the solution temperature (approximately 0.1 C change in temperature in sample of water corresponds to 10-5 RIU change [15]) or composition fluctuations can affect the sensor s output and undermine the reliability of measured d a, and/or n a. Hence, decoupling the impact of the bulk analyte solution (the so called bulk effects) from the response of the adlayer index and thickness (the so called surface effects) is an important and challenging task [11] which is difficult to overcome in a conventional SPR biosensor. This limitation is due to the fact that a conventional SPR affinity sensor relies only on a single mode which has cross sensitivity to both the surface and bulk refractive index variations and therefore the information obtained from the SPR sensing is limited. In this dissertation, the parameters which can compromise the SPR sensor response and create refractive index variations are called interfering effects. In next section, a brief summary of current solutions to decouple the interfering effects and more specifically surface and bulk effects from the output signal will be discussed. Another inherent limitation of SPR biosensors is the cross sensitivity of bio-recognition elements to non-target molecules. Therefore, the non-specific binding to non-target molecules can perturb the response of the sensor to specific bindings. One way to overcome this shortcoming is to carefully select bio-recognition elements with high affinity to target molecules and low affinity to non-target molecules in order to enhance specificity of the sensor response. The choice of biorecognition elements depends on the size of target molecules, binding characteristics of recognition elements, and the range of analyte concentrations to be measured [11]. Determining the structural information, i.e. adlayer thickness (d a ) and adlayer index (n a ), from the adsorbed

49 28 layer can also provide insight into the interaction between molecules and bio-recognition elements [47]. This enables determination of conformational changes of the molecules, mass changes, stoichiometry of interactions, and can be used to distinguish specific and non-specific bindings. 2.7 Decoupling Interfering Effects: Available Solutions Reference Channel Using a reference channel beside the sensing channel is the most common method to discriminate between the refractive index change due to specific interaction with target molecules and those due to bulk refractive index variations which are typically caused by change in composition of the solution and by temperature variations [48]. In this method, two separate channels are covered by receptors. One of the channels has receptors with high affinity to the target molecules which is called sensing channel and the other channel has receptors with no affinity to the target molecules and is called reference channel (Fig. 2-12). By subtracting the reference channel signal from that of the sensing channel, possible non-specific binding and the bulk refractive index can be compensated. However this method is based on some assumption which impose some limitations on the decoupling efficiency: 1) The reference channel should have receptors with similar immobilization conditions used for the sensing channel surface to ensure the environments between the two channels are similar [49]. 2) Since the non-specific binding properties of both channels is not known in advance, it makes it difficult to have different bio-receptors on both channels with similar non-specific binding properties [50]. 3) This method limits the potential applications in high throughput systems which require doubling the number of channels and make a reference channel for each sensing channel. 4) Both the reference channel and sensing channel should be perfectly identical with the exception of the one measurand being measured [51]. This is a requirement which is not easy to meet in practice due to the fabrication imperfections, temperature difference between two channels, and different solution concentration flowing above both channels. 5) Finally, although decoupling the surface

50 29 effects from the bulk index variations is achievable through this approach, decoupling adlayer thickness (d a ) and index (n a ) from the output signal is still not possible. Fig Schematic diagram of the integrated optical SPR sensor immunoprobe showing the binding of antibodies to the surface modified gold film (figure and caption obtained from [52]) Self-referenced SPR Sensing Self-referenced SPR sensing schemes eliminates the need for reference channel to distinguish between bulk refractive index variations and adlayer properties. The key for this method is to increase the number of independent measurements by exciting SP waves with different penetrations depths and sensitivities to the quantities of interests. From mathematics we know that the number of parameters to be determined should be equal to the number of independent equations. Therefore by increasing the number of independent measurements, extracting more unknown parameters from the sample is possible. To further illustrate this point, in the case of angular modulation the following system of equations is required in order to decouple adlayer thickness variations from bulk index change.

51 da nb da nb d a n da nb b. Eq where θ 1 and θ 2 are the measured resonance angles corresponding to two modes with different field profiles and therefore different sensitivities to surface and bulk variations. Hence, the variations in adlyaer thickness ( ) and bulk refractive index ( ) can be determined from the following equation: d a d d nb da n b a a 1., S Eq Equation 2-23 can be solved for Δd a and Δn b if S is not singular. In order to estimate the accuracy of decoupled parameters, condition number of the S matrix should also be evaluated. Condition number is a tool that estimate the worst case scenario for the errors in the solutions of a system of linear equations due to small changes in the system input arguments [53]. If condition number is expressed as κ(a)=10 κ, then we may lose up to κ digits of accuracy on top of what would have been lost due to loss of precision from arithmetic manipulations. Therefore, a problem with small condition number corresponds to small inaccuracy and is called wellconditioned, while a problem with large condition number is called ill-conditioned. To calculate the condition number for the S matrix, each element of the matrix should be normalized to a typical value of its quantity which makes the elements dimensionless [53]. The largest acceptable values for the condition number is defined by the accuracy that is required to find the decoupled parameters (Δd a and Δn b ). The smallest value for the condition number is 1 which corresponds to the identity matrix.

52 Review of Available Self-referenced SPR Schemes Two Mode Spectroscopy Decoupling Surface and Bulk Effects: In order to separate the adlayer optical thickness i.e. the product of n a and d a from the bulk index (n b ), Homola et al. [54] has suggested using a two-plasmon spectroscopy scheme in which two SP waves (one short range and one long range) are excited at a single spot using wavelength modulation. Although this approach is a good way to decouple the bulk index (n b ) from the optical thickness of the adlayer, the inherent temporal dispersion of adlayer and buffer solution can result in an uncertainty in the estimation of the desired parameters [55]. Moreover, by using this approach it is not possible to obtain separate values for the adlayer thickness (d a ) and index (n a ). In [56, 57], a new approach is suggested by Homola et al. based on excitation of SP waves on two different locations using wavelength modulation. The main drawback of this approach is that the two sets of collected data are from different locations of the sensing surface which makes it difficult to guarantee similar conditions on both locations. Hastings et al.[58] has proposed a different approach for decoupling surface and bulk effects by dual-mode spectroscopy using short range and long range SP wave in angular modulation. However, the short range SP wave in their design has larger sensitivity to both surface and bulk index variations than the long range SP wave and this results in a large condition number and therefore reduces the decoupling accuracy of surface and bulk parameters. Also as this structure relies on excitation of LRSP, its results are susceptible to asymmetry in dielectric refractive index at both sides of the metallic film caused by the change in refractive index of the water solution.

53 32 Decoupling Adlayer Thickness and Adlayer Refractive Index: Dual polarization interferometer (DPI) has also been used to extract both the thickness (d a ) and index (n a ) of the adlayer [47, 59]. The structure consists of two waveguides stacked on top of each other. The bottom waveguide acts as the reference guide, and the top one acts as the sensing guide (see Fig. 2-13). The interference fringes created at the output due to interference between output lights is measured with array photodiode for two different light polarizations (TM and TE). Therefore the relative phase difference between the sensing guide and the reference guide due to the sensing surface reactions can be measured from the interference pattern for each polarization. The main difference between the DPI and conventional SPR sensors is that the DPI sensor averages the adlayer properties over a large sensing area as compared to the SPR sensor where the surface reactions at a single spot is monitored. Hence, the DPI sensor is not very suitable for studying binding kinetics. Also since DPI sensor depends on the measurement of the shift of interference fringes, implementing many channels on a same chip for multi-analyte detection will be more complicated. Lastly, the DPI sensor has two modes (TE and TM) each with around 100nm penetration depth which makes them sensitive to both surface and bulk index variations. Therefore although the DPI sensor is useful to differentiate adlayer thickness and adlayer index but it also has cross sensitivity to surface and bulk index variations [60]. Fig Dual slab waveguide interferometer. The sensor chip comprises five layers of deposited silicon oxynitride. A window is opened in the final layer to expose the sensing waveguide ( figure and caption obtained from [59]).

54 33 Several other methods have been proposed to extract the index and thickness of the adlayer. These methods can be summarized into five categories: I) measurements with two different bulk solution indices [61]. II) angular spectroscopy at two different wavelengths [62]. III) wavelength spectroscopy at two different angles [63]. IV) SPR spectrum analysis using a multilayer Fresnel theory [64] Three-mode spectroscopy: Homola et al. [65] has proposed a novel approach based on three-plasmon spectroscopy to extract two surface (n a, d a, ) and one bulk (n b ) parameters from the output signal. Their method is based on wavelength modulation for a multi-diffractive grating and excites three different SP waves at three different wavelengths. Therefore, the dispersion properties of the bulk analyte solution and the adlayer should be known a priori due to inherent temporal dispersion of the adlayer and buffer solution in order to accurately decouple the surface and bulk effects. In addition, the excitation mechanism of the SP waves for this method is based on passing the light through a flow-cell with a transparent top which makes the measurements more vulnerable to the interaction of the light with the analyte solution. In the upcoming chapters, new schemes based on multiple-mode spectroscopy are introduced. These schemes rely on simple and cost effective structures by adding extra layers (planar or grating) to the conventional SPR sensor to allow decoupling the surface and bulk effects. 2.8 Conclusion In this chapter, we have presented a brief introduction to SPR sensors. We first described the main characteristics of SP wave and explained the different mechanisms to optically excite the SP wave. The parameters which are commonly used to evaluate the SPR sensor performance were introduced and a proper FoM which is used in this dissertation was formulated. We identified a number of challenges for SPR sensing which need to be addressed and then we

55 34 introduced the available solutions to overcome these challenges which is the central theme of this dissertation. In Chapters 4 and 5, our proposed solutions for these challenges are discussed.

56 35 3 Genetic Algorithm Genetic algorithm (GA) is a simple and yet powerful optimization technique especially when several design parameters should be optimized simultaneously. It was first introduced by Holland [66] and has been implemented to optimize different types of problem. The main advantages of GA are its ability to work with large number of design parameters, its suitability for parallel computation, and its ability to provide a list of optimum parameters and not just a single solution [67]. GA emulates the well-known biological processes of genetics and evolution to find the exact or approximate solution(s) to an optimization problem. The evolution towards the global optimum occurs through natural selection which is introduced as survival of the fittest by Herbert Spencer. Based on this theory, evolution results in a population which is optimized for its environment and only top members of each generation have higher chance to be selected for the new generation. Due to large number of design parameters in our work and also the simplicity of implementation of GA, we chose it as the optimization method. The basic ideas of GA and its different operators are discussed in this chapter. 3.1 GA Operators The GA contains several components which are borrowed from genetics and evolution. The major terminologies that summarize the GA are described here:

57 36 Chromosomes Chromosomes are the basic building blocks of GA. Each individual/chromosome within a population represents a possible solution to a given problem. Chromosomes consist of genes which are commonly represented by binary numbers. Several genes produce a design parameter to be optimized and a combination of these parameters generates a chromosome. Each chromosome has a certain fitness which is usually the value of the objective function in the optimization problem being solved. Populations and Generations Population is a collection of a certain number of chromosomes. Each population contains trial solutions (chromosomes) in each of the iterations of the GA. The iterations are called generations. Each generation is composed of the merit chromosomes of the previous generation and the modified chromosomes (recombined and randomly mutated). Fitness In the language of GA, fitness (also called cost) is the measurement of FoM for each chromosome. The fitness function is the engine of the algorithm which calculates the fitness of each member of the population. Parents and Children A certain number of chromosomes are selected based on their fitness value by some designated selection procedure (e.g. roulette wheel selection, tournament selection etc.). These selected pairs of chromosomes are called parents. The selected pair goes through the recombination and mutation processes to create a new pair of solutions for the next generation which are called pair of children.

58 37 Crossover Crossover selects genes from the selected parent chromosomes and creates a pair of children. In this process, the parent chromosomes are partially mixed and generate pair of children so that they share their parents genetic properties. Generally the crossover process is applied with a certain probability (p cross ) which ranges from 0.6 to 0.8 depending on the problem to be optimized [68]. Simple-point crossover The simplest way to mix parents is to randomly choose a crossover point and copy everything before this point from the first parent and then copy everything after the crossover point from the second parent. The single point crossover is illustrated in Fig. 3-1 for one dimensional chromosomes. Different colors in the figure indicate different parents. To define the crossover point, a random point from the parent chromosomes is selected and the portion preceding the crossover point is copied from parent 1 to child 1 and correspondingly, from parent 2 to child 2. The second portion of the child 1 is copied from the second portion of parent 2 and the same portion of parent 1 goes to child 2. This type of crossover can be extended to more than a single crossover point in which the parents are split up in more than one point and each child contains different portions of the parents.

59 38 Crossover point Parent Parent Child Child Fig The steps of a single point crossover operation. In the 1st step the chosen crossover point is shown by the black arrow. The 2nd step shows the child chromosomes created by the interchange of the split portions of the parent chromosomes indicated by the corresponding colors. Uniform crossover In a more advanced crossover method, a mask that consists of binary numbers is randomly generated for each set of selected parents with the same number of bits. When the bit in the mask is 1 then the corresponding bit from parent 1 is copied to child 1 and the corresponding bit from parent 2 is copied to child 2. In a similar manner, when the bit of the mask is 0 then the corresponding bit from parent 2 is copied to child 1 and the corresponding bit from parent 1 is copied to child 2. The uniform crossover is illustrated in Fig. 3-2.

60 39 Parent Parent Mask Child Child Fig The steps of a uniform crossover operation. The bits of the child chromosomes are selected from the parents according to the mask bits. Crossover can be rather complicated and depends on encoding type of chromosomes and for a specific problem can improve the performance of GA optimization. The crossover method used in this dissertation is a single point crossover operation. Mutation Mutation is applied after the crossover is performed. The main reason behind the mutation operator is to explore the portions of the solution space which are not covered by the population pool of the present generation. The mutation operator introduces and maintains diversity in the population pool in order to check any kind of early or local convergence and it prevents falling of the solutions of a population into a local optimum of the solved problem. Mutation randomly changes the newly reproduced child chromosomes. The portion of bits within a population which undergoes the mutation process is determined by the mutation rate. In binary coded chromosomes, the mutation operator switches a few randomly chosen bits from 1 to 0 or

61 40 from 0 to 1. The probability of mutation is chosen by a randomly generated number P chosen in the range of 0 to 1. If P is larger than a predetermined constant P mutation the mutation is performed. The mutation operation is depicted in Fig Select a bit for mutation Child chromosome before mutation Mutated bit Child chromosome after mutation Fig The two major steps of a mutation operation. In the 1st step the randomly chosen mutation bit is shown by the black arrows. The 2nd step shows the mutated bit (from 0 to 1) with a change of color. Natural selection When the first population of chomosomes is randomly generated, the merit members are allowed to survive as the parents to create the mating pool for the next generation. There are two common way for natural selection of the mating pool. In the first method, a certain number of the best chromosomes in terms of their fitness value are considered for the next generation while discarding the remaining chromosomes. In the second approach, the chromosomes with fitness value higher than a threshold value are transferred to the next generation.

62 GA Selection Operator A number of different methods are available for selecting parents among the naturally selected merit chromosomes (mating pool) to reproduce children chromosomes. The selection should be such that it applies the generational diversity to prevent converging to a local optimum point solution rather than the global solution. Also this procedure can affect the convergence speed toward the global solution and therefore care should be taken to ensure the high optimization speed. The most common ways to select parents are roulette wheel and tournament selection Roulette Wheel Selection In this method, the chromosomes within each mating pool must first be sorted based on their fitness values. Each chromosome within a population pool is given a probability of selection on the basis of either its rank in the sorted population or its fitness value. The rank based selection is the easiest implementation of this method. In the rank based selection, the selection probability should be computed only once, since the total number of chromosomes in the population pool remains constant. For instance, a roulette wheel for four parents in the mating pool is depicted in Fig Fig Roulette wheel probabilities for four parents in the mating pool.

63 42 The roulette wheel can also be developed based on the fitness value of chromosomes. Each chromosome is assigned a probability of selection according to the following equation: p i fitness( parenti ) Eq. 3-1 fitness( parent ) i i In this equation, the fitness value of each chromosome is normalized to the summation of the fitness values of all chromosomes of the mating pool. Chromosomes with high fitness have higher chance to take part in mating for the creation of children chromosomes. Once the selection probability is determined for each parent, then a random number (r) is generated. For instance, for r=0.6 in the case of the four parent mating pool shown in Fig. 3-4, the selected chromosome based on the value of r is the second chromosome in the rank based selection. In this dissertation, a roulette wheel method is used for the parent selection Tournament Selection In this method, small groups of chromosomes (usually two or three per group) are randomly selected from the mating pool. The chromosomes within each small group with highest fitness value are selected as the parent for reproduction. Then all chromosomes of the mating pool (including the selected one) compete again in another tournament selection to find the next parent. Therefore chromosomes with larger fitness value are more likely to be selected as the parent. A schematic of tournament selection is depicted in Fig. 3-5.

64 43 Unsorted population Tournament Tournament Parent 1 Parent 2 Fig Tournament selection. 3.3 Basic GA A basic GA is an iterative optimization method which is carried out through the following steps: 1) Create an initial population 2) Evaluate the fitness value corresponding to each chromosome within a population 3) Invoke natural selection and create the mating pool 4) Select the parents from the mating pool 5) Reproduce the children chromosome from the parents through a crossover process 6) Mutate the selected members of the reproduced population 7) Terminate the algorithm or go to step 2 The first step of a GA optimization is to randomly generate the initial population of a certain number of chromosomes. Each chromosome contains the variable to be optimizes and

65 44 therefore the first population is basically the first trial solution pool for a given problem. Then, a fitness value is calculated for each chromosome. The merit chromosomes within each population are selected for the reproduction phase of the new generation though a natural selection method. Then a pair of chromosomes is selected as the parents from the merit chromosomes and they undergo crossover and mutation to generate a new pair of chromosomes called children pair. This process of selection-crossover-mutation continues until the new generation is filled up. The new generations undergoes similar procedures until the convergence criterion is met. These steps are shown in Fig Initial population yes fitness calculation Convergence? no natural selection for next generation perform perform reproduction mutation using crossover Fig Genetic algorithm flowchart.

66 Conclusion A brief discussion on the Genetic Algorithm optimization procedure was presented in this chapter. The most common operators of GA which are the essential components for this optimization method were discussed. Finally, different steps of a simple GA were explained. In the next chapter we will utilize GA to optimize our proposed sensor designs.

67 46 4 Dual-mode Spectroscopy: Hybrid Plasmonic Sensors As mentioned earlier (section 2.7.2), one solution for decoupling interfering effect is to use selfreferenced spectroscopy. In this chapter, a new approach for self-referenced SPR spectroscopy is proposed based on a hybrid plasmonic waveguide (HPWG) sensor. A HPWG is a compromise between conventional plasmonic waveguides and a dielectric waveguide (DWG). It was first proposed by Alam et al. [69] and features the high confinement of plasmonic sensors and the long propagation length of DWG. HPWG can guide both TE and TM polarizations and this feature is utilized here as a self-referenced scheme. In [70], decoupling surface and bulk effects were investigated for the HPWG in Mach-Zehnder configuration. When used as such, the HPWG required fabrication of a nano-fluidic channel. Here, the performance of the HPWG in a Kretschmann configuration is investigated and the need for nano-fluidic channel to deliver analyte to the sensing region is eliminated. This chapter is organized as follows: In section 4.1, a brief introduction to HPWG and its main properties are presented. In section 4.2, HPWG is investigated for thin film measurement in biosensing applications using a Kretschmann configuration. The method to separate adlayer thickness from bulk refractive index is also discussed in this section. The design optimization using a genetic algorithm (GA) is presented in section 4.3 and the sensor s parameters corresponding to the optimized design are described in section 4.4. Section 4.5 concludes the chapter with some remarks. 4.1 Hybrid Plasmonic Waveguide (HPWG) When a plasmonic layer (Guide 1) and a dielectric guiding layer (Guide 2) are brought close to each other [Fig. 4-1(a)], the SP mode supported by Guide 1 couples to the TM mode of the dielectric waveguide supported by Guide 2, and forms a hybrid TM mode. The TE mode remains relatively unperturbed and is guided in the high index dielectric layer of the HPWG. Figure 4-1(b) shows the strong mode confinement of the hybrid plasmonic mode in the low index region

68 47 between the metal and the dielectric slab which has also been reported by R. F. Oulton et al. in [71]. (a) (b) Fig Formation of hybrid plasmonic mode from coupling of dielectric and SP mode (a) Waveguide structure (b) Normalized power density. The coordinate system used is also shown. The xz plane coincides with the gold-silica interface for the HPWG. The dimensions are h=50 nm, d=100 nm. Wavelength of operation is 1.55 μm

69 48 Fig Comparison of guided power density profile for the SP mode and the hybrid plasmonic mode for the same propagation loss. The dimensions of the HPWG are h = 50 nm, d = 100 nm. Wavelength of operation is 1.55 μm (figure obtained from [72]). The normalized guided power density profile of the HPWG (black curve) and a single interface SP mode (red curve) along the x-axis are shown in Fig The permittivity of the dielectric medium is chosen such that the propagation loss of the SP mode is similar to that of the hybrid plasmonic mode. The higher confinement of the hybrid plasmonic mode, as compared to the SP mode, is obvious from this figure. This feature along with the polarization diversity of the HPWG makes it suitable for different applications [73-76]. In the next section, the application of the HPWG for biosensing is discussed. 4.2 HPWG Sensor HPWG sensor consists of a low index dielectric layer adjacent to a thin metallic surface which separates a high index dielectric layer from the metallic surface as is illustrated in Fig. 4-3(a). In this design, a thin metallic film is chosen in order to study the surface bindings occuring on the top surface of the metallic film. Figure 4-3(b) shows the TM and TE mode profiles for the

70 49 HPWG sensor of Fig. 4-3(a). As mentioned before, the TM mode of the dielectric waveguide couples to the SP wave at the metal dielectric interfaces and generates a hybrid plasmonic mode. However, due to the small thickness of the metallic film in this configuration, most of the field is coupled to the SP wave propagating above the metallic film (metal-fluid interface). From the figure it is clear that the leaky TE mode is mostly confined in the low index region (SiO 2 ), while the TM mode is strongest at the metal/water interface. We have chosen silica as the material for the substrate and prism in order to excite the leaky TE mode which has a higher sensitivity to the gold surface bindings as compared to the guided mode. The mode profiles plotted in Fig. 4-3(b) belong to the leaky TE and guided TM modes of the HPWG, excited at their corresponding resonance angle (see figure caption). A reflection setup with an excitation scheme similar to the one depicted in Fig. 4-3(a) (called Kretschmann configuration) is chosen to study the HPWG sensor. This choice was motivated by a few considerations: (1) the ease of fabricating the device. (2) Similarity of the measurement method with well-established techniques such as ellipsometry. (3) The ease of coupling the light into and out of the biosensor. (4) The ability to study the binding kinetics due to a small area illumination of the biosample. To characterize the reflected light, we concentrate on measuring the incidence angle in angular modulation.

71 50 Antigen Antibody y x Input Flow z Incident Light Fluidic Channel Au Spacer: SiO 2 High Index Dielectric: Si Substrate: SiO 2 θ Output Flow Reflected Light Normalized Poynting Vector (z- component) SiO 2 Si Si SiO 2 Au Water x (mm) TM TE Fig. 4-3(a) The proposed HPWG for biosensing in reflection measurement. (b) z-component of the Poynting vector for both TM and TE polarizations (p- and s-polarizations) along the x-axis at the resonance angle. The wavelength of the incident light is 605 nm and silica, silicon, and gold indices at this wavelength are 1.45, 3.93+i0.02, and 0.23+i3.11, respectively. The resonance angle for TM and TE polarizations are 80.76⁰ and 48.27⁰, respectively. The use of HPWG as a platform for biosensing, as shown in Fig. 4-3(a), has a few advantages. First, the polarization diversity (the presence of both TM and TE modes) allows us to distinguish two different unknown parameters such as the changes in the adlayer thickness (d a ) and the bulk index (n b ) from two independent polarization measurements. Second, the fabrication is simple due to the fact that it only requires planar deposition of dielectric and metallic layers. Third, fabrication steps are compatible with the standard Si-technology, using the commercially available silicon on glass (SiOG) substrate [77, 78]. Finally, the polarization diversity enables studying anisotropic systems with high birefringence (e.g. lipid bilayer). This feature can also be used to determine the optical anisotropy (molecular orientation) of anisotropic structures for instance, as in the case of proteolipid membranes which requires the characterization of events parallel and perpendicular to the plane of membrane [79]. In the following we will describe an approach based on the dual polarization measurement which can be used to separate the interfering bulk and surface effects [80]. The sensing area (gold in this case) is functionalized with specific receptors (antibody) before starting the measurement. Prior to introducing the antigens in the fluidic channel two resonant angles are measured for each

72 51 polarization. After introducing the antigens in the channel, as shown in Fig. 4-3(a), antigens are gradually captured by the antibodies immobilized on the surface. The SP wave characteristics are affected by the change in material composition near the metal surface which changes the SPR angle (θ). In the experiment, the change in the resonance angle for each polarization [Δθ(TM 0 ), Δθ(TE 0 )] is measured. Since the SP wave extends beyond the antigen/antibody pairs (the adlayer) into the fluidic channel and also the solution composition and temperature in the fluidic channel can change with time and material composition, the measured change in the resonance angel (Δθ ) is a function of both the adlayer thickness (d a ) and bulk index (n b ). This functional dependency of the measured resonance angle on variations in adlayer thickness (Δd a ) and bulk index (Δn b ) is given by: TM 0 TM 0 Δθ TM d Δ 0 a nb da. Δθ TE θ 0 TE θ 0 TE Δn 0 b da n b Eq. 4-1 Equation 4-1 can be solved for Δd a and Δn b according to: TM0 ΔθTE 0 (TE 0) ΔθTM0 nb nb Δ da, TE TM (TM ) (TE ) d a nb da nb TE0 ΔθTM 0 (TM 0) ΔθTE0 da da Δ nb. TE TM (TM ) (TE ) d a nb da nb Eq. 4-2 Eq. 4-3 In Eq. 4-2 and Eq. 4-3, Δθ(TM 0 ) and Δθ(TE 0 ) are known from the measurements, while and (TM 0 and TE 0 ) are the TM and TE surface and bulk sensitivity factor (SFs) which are known from the sensor s design (see also section 2.5.1). From the above discussion it is clear that by using a HPWG sensor one can determine the changes in the adlayer thickness (Δd a ) and bulk index (Δn b ) by utilizing a dual polarization measurement. However, a question remains to be answered: does the matrix on the right hand side of Eq. 4-1 have an inverse so that Δd a and Δn b in Eq. 4-2 and Eq. 4-3 can be uniquely calculated? In the next section we will first optimize

73 52 the performance of the HPWG biosensor of Fig. 4-3(a) using a genetic algorithm and then will answer this question. 4.3 Optimization Using Genetic Algorithm Our goal in this section is to optimize the performance of the HPWG sensor shown in Fig. 4-3(a) simultaneously for both TM and TE polarizations. To model the HPWG and SPR sensors, a transfer matrix method is used (see Appendix A). The quantity to be optimized is the combined sensitivity factor (CSF, see section 2.5.1) which is defined according to the following equation: CSF SF surf, thick S d M res max a R R FWHM min Eq. 4-4 The design parameters which can be changed in order to obtain the highest CSFs for both TM and TE polarizations are: wavelength of operation, gold layer thickness, spacer thickness (the SiO 2 layer between the gold and Si slab), and the Si slab thickness. The thickness of the SiO 2 substrate 1 and the prism (also assumed to be made of SiO 2 ) do not enter our optimization algorithm because they are very thick compare to the aforementioned layers. We have chosen the genetic algorithm (see chapter 3) to perform the optimization for the HPWG biosensor. Each chromosome within a population contains the design parameters to be optimized. The fitness function to calculate the cost of each chromosome is defined such that the CSF for both polarizations is maximized. To simultaneously enhance the CSF factor for both TM and TE polarizations, the following fitness function is used: Fitness Function CSF. TM CSFTE Eq Most conventional glass used in SiOG substrates is EAGLE XG ; Optical property of EAGLE XG is very similar to silica and hence in our simulations we have used silica as the substrate [34].

74 53 The convergence rate of the genetic algorithm is shown in Fig In order to find the global maximum value of the fitness function (which is the same as Figure of Merit, FoM), it is necessary to trace the best fitness value of each population with respect to the number of generations. The maximum fitness reaches a constant value after a certain number of generations which is considered the optimized design in our calculation. 8 7 Fitness value (10-6 ) Generations Fig. 4-4 Fitness function over 140 generations. 4.4 Results and Discussion Optimized structures (used in Kretschmann configuration) for both the single interface SPR biosensor and the HPWG biosensor are presented in this section. Table 4-1summarizes our optimization results for the single interface SPR and the HPWG biosensors. The table lists: prism materials, various layers thicknesses, wavelength of operation, SFs (TE and TM), SMs (TE and TM), CSFs (TE and TM), and the overall fitness function used in the optimization. Reference [81] was used to model the material dispersion over the wavelengths of interest.

75 54 Table 4-1- Optimized values for the design parameters obtained from the genetic algorithm code for: single interface SPR and HPWG Structure Prism Si thickness (nm) Spacer (nm) Au thickness (nm) Wavelength (nm) SF TE deg. ( ) nm SF TM deg. ( ) nm SM TE 1 ( ) deg. SM TM 1 ( ) deg. CSF TE CSF TM 1 1 Fitness ( ) ( ) nm nm Single interface SPR SiO _ 0.1 _ _ CSF TM HPWG SiO 2 25 SiO 2 : CSF CSF TE TM Due to higher overlap between the bio-sample and the TM field [Fig. 4-3(b)], the SF for the TM mode is much larger than the TE mode (Table 4-1), which leads to a larger CSF for TM mode (CSF TM ) than CSF for TE mode (CSF TE ). The CSF TM for the HPWG sensor is close to that of a single interface SPR biosensor, while its CSF TE is two orders of magnitude smaller than its CSF TM. However, the TE mode is sensitive enough to the adlayer thickness variations and is detectable by the available spectroscopy methods which can detect resonance angles with high resolution (~10-5 degrees [82]). To further establish this point, we have evaluated the determinant of the sensitivity matrix [Eq. 4-6] for our HPWG biosensor of Table 4-1 to ensure it is not singular. TE 0 TE det da nb TM 0 TM da n b Eq. 4-6

76 55 The calculated condition number corresponding to the sensitivity matrix is 870 which shows that we may lose up to two orders of magnitude on top of the precision of finding the resonance angle. Assuming that the resonance angle can be determined with the accuracy of ±10-5 degrees and based on the calculated condition number, the accuracy of estimating the bulk index change ( n b ) and adlayer thickness change ( d a ) is in the order of ±0.01 nm. Dual polarization measurements will allow us to decouple the interfering surface and bulk effects. The TM mode of the HPWG is useful for detecting analytes with low limit of detection (LoD), while the combination of TM and TE polarizations is useful to decouple surface and bulk effects from the output signal. Figure 4-5(a) illustrates how the CSFs change with wavelength for the HPWG and the single interface SPR biosensor. Variations of the CSF with wavelength depends on both SF and SM: SF decreases with wavelength while SM increases with wavelength for TM mode (Fig. 4-5(b,c)] while the same variations in the TE mode does not have a general trend as is illustrated in Fig. 4-5(d). The reason for this trend in the TM mode can be found in opposite dependency of the SF and SM on losses. SF is large when mode is highly confined to the metal/adlayer interface which results in more losses due to the presence of metal. Therefore, SF is directly proportional to the propagation loss which decreases with increasing wavelength. On the other hand, SM is inversely proportional to the FWHM [see Eq. 2-24] which is a measure of the loss in the system. In other words, SM is inversely proportional to the propagation losses and hence it increases with increasing wavelength. It is apparent from Fig. 4-5(a) that the fitness function for both CSF TM and CSF TE are maximized near the wavelength of 830 nm which is consistent with the optimized values obtained from the genetic algorithm.

77 SM SPR (deg -1 ) CSF TM SM TE (deg. -1 ) SM TN HPWG (deg.-1 ) 56 CSF TE (10-4 ) (a) HPWG -TE HPWG -TM SPR- TM Wavelength(m) SF bulk (deg./riu) SPR SF bulk SPR SM SPR (b) Wavelength (nm) SF TM HPWG SM TM HPWG (d) SF TM (deg./riu) HPWG (c) SF TE (deg./riu) HPWG SF TE HPWG SM TE Wavelength (nm) Wavelength (nm) Fig. 4-5 (a)the CSF for both polarizations in the HPWG and the single interface SPR biosensors.(b) variations in SF and SM for TM polarization in the SPR sensor. (c) Variations in SF and SM for TM polarization in the HPWG sensor. (d) Variations in SF and SM for TE polarization in the HPWG sensor. The reflectance spectra for both the single interface SPR and the HPWG biosensors are plotted in Fig. 4-6(a). Both structures have a fairly similar sensitivity to the bulk (fluid) refractive index change (i.e. bulk SF) and also comparable values of SM. On the other hand, the lower bulk SF of the TE polarization for the HPWG sensor can be seen from the small variations in the resonance angle due to change in the bulk index [Fig. 4-6(b)]. The same observation can be made for the SF i.e. surface SF by changing the thickness of the adlayer on top of the metal.

78 57 Reflectance TM mode 0.4 Hybrid Biosensor n b = Hybrid Biosensor n b =1.34 (a) Single Interface SPR n b =1.33 Single Interface SPR n b = Angle(degree) Reflectance TE mode Hybrid Biosensor n b =1.33 Hybrid Biosensor n b =1.34 (b) Angle (degree) Fig. 4-6 Reflectance spectrum for (a) TM polarization in both the single interface SPR and the HPWG biosensor when the bulk index (fluid index) changes from 1.33 to The operating wavelengths for the single interface SPR and HPWG biosensor are 836 nm and 830 nm, respectively. (b) TE polarization in the HPWG biosensor when bulk index (fluid index) changes from 1.33 to Conclusion In this chapter, we proposed a novel affinity biosensor based on the HPWG which is capable of performing measurements in both TM and TE polarizations. The expressions for de-convolving the adlayer thickness and bulk index of refraction from the resonance angle variations were presented. The best FoM for the angular modulation was derived and with the help of genetic algorithm the HPWG sensor was optimized based on the defined FoM. Performance of the optimized HPWG sensor was compared with the conventional single interface SPR sensor. It was shown that for the HPWG sensor, without sacrificing the performance for the TM polarization, the sensitivity of the TE polarization was high enough to decouple the interfering effects from the signal output.

79 58 5 Dual-mode Spectroscopy: Plasmon Waveguide Resonance Plasmon waveguide resonance (PWR) sensor is a promising structure which is proposed here for self-referenced spectroscopy using both TE and TM polarizations. The PWR sensor simultaneously benefits from the high sensitivity of the SPR sensors and the small resonance width of the dielectric waveguide sensors. It consists of a glass substrate, a thin metallic layer, and a dielectric layer on the top of the metal [see Fig. 5-1(a)]. The role of the dielectric layer is to excite the dielectric waveguide modes (TM, TE) under certain conditions. The large probing depth of the TM polarized light in the PWR sensor makes it suitable for bulk sensing and detecting large biomaterials such as bacteria [83, 84], while the small probing depth of the TE polarized light is ideal for surface sensing [85, 86]. This chapter is organized as follows: In section 5.1, the physical properties of the PWR sensor is discussed. In section 5.2, application of the PWR sensor as a bulk refractive index sensor is compared with the SPR sensor. In section 5.3, application of the PWR sensor for thin film investigation is demonstrated. The kinetic analysis of the protein-nanoparticle interactions using the PWR is investigated in section 5.4. Section 5.5 concludes the chapter with some remarks. 5.1 Description of the PWR Structure Figure 5-1(b) shows the calculated angular reflectance spectrum for both the PWR and SPR sensors. As the figure shows the resonance curves for both TM and TE polarizations [black and red lines in Fig. 5-1(b)] in the PWR sensor are much sharper than the SPR resonance curve (blue line). Figure 5-1(c) and Fig. 5-1 (d) show the spatial distribution of the power density (Poynting vector) along the propagation direction (z) as a function of the vertical direction (x), for the optimized PWR and SPR sensors at their corresponding resonance angles. In the PWR sensor, the TE mode is mostly confined in the top silica layer (red line in Fig. 5-1(c)) where its

80 59 penetration depth into the bulk fluid is 270 nm. On the other hand, the TM polarization intensity is mostly localized inside the cover layer (fluid) with its highest intensity at the silica-fluid interface. The penetration depth of the PWR-TM mode inside the fluid is 9 μm which makes it suitable for bulk sensing. Finally, the calculated penetration depth of the optimized SPR-TM mode (the only mode supported by this sensor) is 460 nm, as shown in Fig. 5-1(d). (a) (b) Reflectance PWR-TM PWR-TE SPR- TM Angle Normalized poynting vector (a.u.) (c) PWR-TM PWR-TE x (m) Normalized poynting vector (a.u.) 1.0 (d) SPR-TM x (m) Fig. 5-1(a) Schematic diagram of the PWR sensor. (b) Reflectance spectrum for the optimized PWR-TM, PWR-TE, and SPR-TM polarizations in black, red, and blue lines, respectively. (c) z- component of the Poynting vector for both TM and TE polarizations in the optimized PWR sensor at the resonance angle of 62.35⁰ and 66.18⁰, respectively. (d) z-component of the Poynting vector for the TM polarization in the optimized SPR sensor at the resonance angle of 63.77⁰.

81 60 Fig. 5-2 illustrates the effect of top dielectric layer thickness (here we use silica) on the field profile of the TM polarization at the wavelength of 800 nm. For silica thicknesses less than 565 nm, the silica layer perturbs the plasmonic wave at the silica-cover layer interface and creates small peak at this interface. However when the silica layer is larger than 565 nm [as is shown in Fig. 5-2 (e)], most of the power is concentrated at the silica-cover layer interface instead of gold/silica interface with a large penetration into the cover layer. The PWR sensor is mostly operated at this range of top dielectric thickness, where power is the largest at the dielectric/cover layer interface with large penetration into the cover layer. Normalized guided power nm silica x (m) Normalized guided power nm silica Normalized guided power nm silica (a) (b) (c) x (m) x (m) 0.2 Normalized guided power nm silica Normalized guided power x (m) nm silica (d) (e) (f) x (m) Normalized guided power nm silica x (m) Fig. 5-2 The field profile of TM polarization for different silica layer thicknesses on top of 50 nm gold film at the wavelength of 800 nm. The silica layer thicknesses are: a) 0 nm, b) 50 nm, c) 100 nm, d) 400 nm, e) 565 nm, and f) 700 nm. The field profile of TE polarization for different thicknesses of silica is shown in Fig The TE mode appears for silica thicknesses larger than 400 nm. As is shown in Fig. 5-3(a), the

82 61 penetration depth of the TE mode into the cover layer is largest at 400 nm and it decreases with increasing silica thickness. Normalized guided power TE: 400nm silica Normalized guided power TE: 600nm silica (a) (b) (c) Normalized guided power TE: 1000nm silica x (m) x (m) x (m) Fig. 5-3 The field profile of TE polarization for different silica layer thicknesses on top of 50 nm gold film at the wavelength of 800 nm. The silica layer thicknesses are: a) 400 nm, b) 6000 nm, and c) 1000 nm. The PWR sensor was initially introduced for optical sensing by Otto et al. [87], and was implemented by Salamon et al. [79], to measure the optical properties (thickness and complex refractive index) of anisotropic membrane systems. Coupled Plasmon Waveguide Resonances (CPWR) [88], Metal Clad Leaky Waveguide (MCLW) [89], and Metal Clad Waveguide (MCWG) [90], are other names that have been used to refer to this structure. In general, PWR sensors have been classified in two different types [90]: a) the dip type which has tens of nanometers of metal with a relatively low imaginary permittivity [91, 92]. b) The peak type with a thinner metallic layer (a few nanometers) with relatively large imaginary permittivity [93]. In the next sections, the performance of the PWR sensor for different applications is investigated.

83 PWR Sensor for Refractive Index Sensing In this study, the dip type PWR sensor is proposed as a promising tool which can exceed the performance of the SPR sensor based on the Kretschmann configuration in terms of its refractive index resolution [94]. Using a genetic algorithm, we have identified the conditions under which both the PWR and the SPR performances are optimum (in terms of their resolution). We have then fabricated and tested these sensors and have compared their performances for both TM and TE modes in the case of the PWR and only the TM mode in the case of SPR sensor. The experimental results show that the PWR sensor has a refractive index resolution of RIU which is 6 times smaller than that of the optimized SPR sensor. The TE polarization in the PWR sensor has a resolution of RIU which is smaller than the SPR sensor. The ability of the PWR to operate in both TE and TM modes is an important characteristic of these sensors which allows the user to remove the influence of some interfering effects as discussed in [95]. Lastly, although some evaluation of the PWR in comparison to the SPR sensor has been carried out in Refs. [86, 92, 96, 97], to the best of our knowledge, there has been no comprehensive optimization and evaluation based on the refractive index resolution between these two configurations Theory and Modeling Figure 5-1(a) shows the PWR sensor in the Kretschmann configuration, in which light is incident on the substrate through a prism and the reflected light intensity is measured as a function of the angle of incident. In order to conduct a fair comparison between the performances of the PWR and SPR sensors, a figure of merit (FoM) based on the CSF (see section 2.5.1) is defined. In [98], it was shown that the aforementioned CSF is inversely proportional to the refractive index resolution (σ RI ) and can be written as: res Rmax Rmin 1 CSFbulk SF SM. n FWHM b RI Eq. 5-1

84 63 Here, θ res is the resonance angle, n b is the bulk refractive index, FWHM is the full width at half maximum, and R max and R min are the maximum and minimum reflectance, respectively [95]. For brevity, hereon, we refer to the bulk CSF simply as the CSF. Since the SPR sensor only guides a single TM mode, therefore its FoM is given by CSF TM. In the case of PWR sensor, beside the TM mode we also have a TE mode and therefore the FoM definition should be modified as the following: FOM PWR CSF. TM CSF Eq. 5-2 TE We have used a genetic algorithm with the above FoM as its fitness function to determine the wavelength of operation and layer thicknesses (in the case of the PWR) which maximizes the FoM (i.e., minimizes the resolution) for both sensors. To model the PWR and SPR sensors, a transfer matrix method is used (see Appendix A) [32, 90]. In Table 5-1, the optimized design parameters for both structures for bulk sensing application are listed. In our calculations there are a few points worth mentioning: 1) We have assumed that the substrate and prism are both BK7 glass. 2) Gold is used as the metallic layer in both the PWR and SPR sensors, which is a preferred plasmonic layer. 3) For our calculations we have used the material optical properties listed in Palik [99]. 4) Both sensors have 2 nm titanium layer between the substrate and the gold layer in order to improve the gold adhesion to the substrate. Table 5-1.Comparison of the optimized PWR and SPR sensors characteristics Sensor Layers Optimized parameters polarization SF (deg./riu) PWR SPR 1. BK7 glass 2. Titanium 3. Gold (d) 4. Silica(h) 1. BK7 glass 2. Titanium 3. Gold (d) h=925 nm d=44 nm λ=1200 nm d=46 nm λ=1230 nm SM (deg. -1 ) CSF (RIU -1 ) TM TE TM

85 64 Before presenting the experimental results, it is useful to discuss the tolerance of the PWR and SPR sensors with respect to the fabrication imperfections. Figure 5-4(a) shows the variations of the CSF for the SPR sensor with respect to the incident wavelength. As evident from the Fig. 5-4(a), the CSF is maximum at ~1200 nm. The tolerance of the PWR sensor with respect to silica thickness at different wavelengths for the TM and TE polarizations are shown in Fig. 5-4(b) and Fig. 5-4(c), respectively. As evident from these figures, the PWR-TE mode is less sensitive to fabrication imperfections than the PWR-TM mode. However, it is possible to maximize the CSF for the PWR-TM polarization by tuning the incident wavelength for any silica thickness. SPR- CSF TM (RIU -1 ) SPR -TM (a) (b) (c) Wavelength (nm) PWR- CSF TM (RIU -1 ) nm 1000 nm 1100 nm 1200 nm 1300 nm Silica thickness (nm) PWR- CSF TE (RIU -1 ) Silica thickness (nm) 900 nm 1000 nm 1100 nm 1200 nm 1300 nm Fig. 5-4(a) The CSF variation along with the incident light wavelength in the optimized SPR sensor. The CSF variation along with the silica thickness at different wavelengths in the PWR sensor for (b) TM and (c) TE polarizations.

86 Materials and Methods Sensor Fabrication We obtained BK7 glass substrates, with a size of 1cm 1cm, coated with 48±1nm gold layers from the SSENS Ltd. [100]. In order to ensure good adhesion of the gold film to the glass substrate, a thin layer of titanium (2 nm) was introduced between the glass and the gold film. These samples, which were used as our SPR reference sensors, were initially cleaned with a piranha solution at 90 C for 20min to remove organic contaminations. After the initial cleaning, the samples were further cleaned in an ultrasonic bath with acetone, isopropanol, and deionized water (DI water) for 10 minutes each. The same gold plated samples from SSEN were used to fabricate our PWR sensors. For this propose a silica layer of specific thickness was deposited on the gold film using a plasma enhanced chemical vapor deposition (PECVD) by mixing silane (SiH 4 ) and nitrous oxide (N 2 O) in vacuum at approximately 300 C. Instrumentation Figure 5-5 shows the optical setup used to characterize the performances of the PWR and SPR sensors. In this setup, an optical beam from a super continuum laser (Fianium SC-450) is passed through a laser line tunable filter (Photon Etc) in order to provide excitation at different wavelengths with a bandwidth less than 2 nm. A single mode optical fiber is attached to the output of the filter which directs the light to an achromatic lens (L1) in order to collimate the light rays. The collimated light is then passed through a polarizer (P) and another achromatic lens (L2) in order to achieve a converging beam to cover a desired range of angles. The converging beam is focused onto the SPR or PWR sensor which is attached with immersion oil (Cargile Lab) to a prism back surface. This sensing block (prism and the sample) is fixed to a flow cell in

87 66 which a desired mixture of fluids flows. Finally, the diverging light that is reflected from the sample (sensor) surface is collected by a CMOS camera (Thorlabs) to analyze the angular spectrum of the beam intensity. A Labview program was written to process the image recorded by the CMOS camera which determines the minimum position of the angular curves by using a polynomial interpolation. Fig Proposed optical setup for minimum position measurement Results and Discussion To measure the reflected light intensity at different angles, a converging beam which covers a range of angles is focused on to the sensor using an achromatic lens (Fig. 5-5). Figure 5-6 shows the images of the light intensity reflected form the PWR sensor and captured by the camera for TM and TE polarized light.

88 67 (a) TM resonance angle (b) TE resonance angle Fig. 5-6 The images of the light intensity reflected form the PWR sensor and captured by the camera for (a) TM polarized and (b) TE polarized light. By finding the distance between the PWR sensor and the camera sensor and calculating the width of the camera sensor ( pixels, each pixel=5.2µm), the range of the angles captured by the camera is estimated to be 9.512⁰. Therefore, the range of the angles covered by each pixel is ⁰ (=9.512⁰/1280pixel). The captured images from the 2D sensor camera are then averaged along the vertical axis to obtain a 1D reflectance plot for a range of angles (Fig. 5-7). Two different kinds of measurements are used to study the effects of bulk index variations on the sensors performances. The first kind is a static measurement which involves the analysis of the entire resonance curve with respect to the line shape changes as a function of time. This analysis provides a better insight from the sample under study since more information can be extracted from the resonance shape: the resonance width (FWHM), the resonance depth, and the resonance angle. The second kind is a dynamic measurement which is based on the analysis of the biological alteration only by tracing the variations in the resonance angle. This measurement is useful for real time analysis of the sample under study. The values for silica and gold thicknesses (h and d) given in Table 5-1 will result in the largest CSF (the smallest resolution) for the PWR and SPR sensors. The wavelengths of operation corresponding to these large values of CSFs are 1200 nm and 1230 nm for the PWR and SPR, respectively. However, the CMOS camera available in our setup (Fig. 5-5) has better sensitivity to wavelengths less than or equal to 900 nm. Therefore, we had to rerun our optimization

89 68 algorithm with the constraint that the wavelength of operation remains less than or equal to 900 nm for both sensors. The results of this analysis are presented in Table 5-2. By comparing the data in Table 5-1 and Table 5-2 we see that the performance of the PWR in the TM mode suffers the most as the result of enforcing the practical limitations of our experimental setup. The PWR sensor fabricated has silica thickness of 585 nm which was measured using an ellipsometer. Table 5-2. Comparison of the optimized PWR and SPR sensors characteristics with considering the limitations in the given setup Sensor Optimized parameters polarization SF (deg./riu) SM (deg. -1 ) CSF (RIU -1 ) h=585 nm TM PWR d=48 nm λ=830 nm TE SPR d=48 nm λ=900 nm TM Static Measurements In this experiment, the reflectance intensity is monitored versus the angle of incidence. To verify the simulation results, two different wavelengths are used for comparison: 1) He-Ne wavelength (λ=632 nm) which is a common wavelength used in biosensing experiments [79, 101, 102] and 2) a wavelength close to the optimized wavelength for each sensor. Fig. 5-7(a) through Fig. 5-7(c) show the reflectance spectrum of both sensors with respect to the incident light angle at the He-Ne wavelength. These spectrums are recorded for different concentrations of ethanol diluted in deionized water flowing over the sample. As explained earlier, the SPR resonance curve [Fig. 5-7(c)] is much wider than the PWR curves [Fig. 5-7(a) and Fig. 5-7(b)]. The simulated results for both PWR and SPR sensors are shown in Fig. 5-7(d) through Fig. 5-7(e) which are in good agreement with the experimental results. The effects of surface roughness is taken into account in the simulation by using Maxwell Garnett theory [103]. Although the theoretical results are based on plane wave excitation of the sensor at different angles and the

90 69 effect of a converging beam is not considered, however there are more uncertainties in the measurement that can cause the small difference between the experimental and theoretical results. The most important uncertainties are fabrication imperfections, inhomogeneity of the layers, defects inside the prism, the uncertainty in the optical properties of different layers, the light source bandwidth, and also the index mismatch between the matching oil index and the substrate. R TM / R TE - PWR (a) n b =1.332 n b =1.335 n b = (b) n b =1.332 n b =1.335 n b = (c) R TE / R TM - PWR R TM / R TE - SPR 0.1 n b =1.332 n b =1.335 n b = Angle (degree) Angle (degree) Angle (degree) R TM / R TE - PWR n b = (d) (e) (f) n b = n b = Angle (degree) R TE / R TM -PWR Angle (degree) n b =1.332 n b =1.335 n b =1.338 R TM / R TE - SPR Angle (degree) n b =1.332 n b =1.335 n b =1.338 Fig. 5-7 The experimental normalized reflectance spectrum at λ=632 nm for (a) TM-polarized PWR sensor. (b) TE polarized PWR sensor. (c) TM-polarized SPR sensor. The theoretical normalized reflectance spectrum at λ=632 nm for (d) TM-polarized PWR sensor. (e) TE polarized PWR sensor. (f) TM-polarized SPR sensor. The different curves refer to reflectance spectrum for different concentrations of ethanol solution, 0.7% (red), 2% (blue). The water spectrum (black) is the reference. In Table 5-3, the measured sensor characteristics at the He-Ne wavelength are compared with the theoretical results. As can be seen from this table, the SPR sensor has much larger sensitivity than the

91 70 PWR at this wavelength but its smaller SM results in small CSF. The TM mode of the PWR sensor has the largest CSF. Table 5-3. The experimental and theoretical sensors characteristics for the fabricated PWR and SPR sensors at λ=632 nm SF (deg./riu) SM (deg. -1 ) CSF (RIU -1 ) Sensor polarizations Experiment Theory Experiment Theory Experiment Theory TM PWR TE SPR TM Figure 5-8(a) shows the normalized reflectance spectrum for both polarizations in the PWR sensor at the optimized wavelength (830 nm). The resonance curve is sharper than that at He-Ne wavelength [Fig. 5-8(a) and Fig. 5-8(b)]. The same trend is observed from the SPR sensor at λ=900 nm [Fig. 5-8(b)] compared with He-Ne wavelength.

92 R TE R TM - PWR R TE / R TM - PWR 71 R TM R TE - PWR (a) PWR-TM PWR-TE R TM R TE (b) SPR-TM Angle (degree) Angle (degree) R TM / R TE - PWR (c) PWR-TM PWR-TE Normalized Spectrum- SPR SPR-TM (d) Angle (degree) Angle (degree) Fig The experimental normalized reflectance spectrum for (a) TM-polarized (black) and TE-polarized (blue) PWR sensor at λ=830 nm (b) TM-polarized SPR sensor at λ=900 nm. The theoretical normalized reflectance spectrum for (c) TM-polarized (black) and TE-polarized (blue) PWR sensor at λ=830 nm (d) TM-polarized SPR sensor at λ=900 nm. The measured sensor characteristics corresponding to the results shown in Fig. 5-8 are presented in Table 5-4. The measured CSF for the PWR-TM polarization is 322 RIU -1 at λ=830 nm which is almost an order of magnitude larger than the CSF in SPR sensor at λ=900 nm (48 RIU -1 ). Therefore, by optimizing the wavelength, the CSF has been improved in both sensors as compared to the results at the He-Ne wavelength. However, the improvement in the PWR sensor performance (CSF TM and CSF TE ) is much larger than that of the SPR sensor (CSFT M ).

93 72 Table 5-4. The experimental and theoretical sensors characteristics for the fabricated SPR sensor at λ=900 nm and the PWR sensor at λ=830 nm. SF (deg./riu) SM (deg. -1 ) CSF (RIU -1 ) Sensor polarizations Experiment Theory Experiment Theory Experiment Theory TM PWR TE SPR TM Dynamic Measurements In the case of dynamic measurements, the resonance angle variation is monitored (in real time) at a given wavelength while the concentration of the solution is changed. We prepared three different solutions with 0.5%, 1%, and 2% of ethanol in deionized water to create bulk refractive index changes of , , and RIU, respectively. Pure deionized water is first passed over the sample for 5min to create a baseline. Then an ethanol solution with low concentration (0.5%) is passed for another 5min. After that deionized water flows over the sample again to recreate the baseline and the same steps are repeated for higher concentrations. The solution flow rate is fixed to 50μl/min in all steps. Figure 5-9(a) shows the PWR sensor response at λ=632 nm and λ=830 nm for TM polarization. The small slope of the baseline is due to the heat generation caused by wave propagation at metal dielectric interface. Figure 5-9(b) shows the PWR response for TE polarization. The larger sensitivity of the PWR-TM polarization, as compared to the PWR-TE polarization, can be seen by comparing the change in the resonance angles in Fig. 5-9(a) and Fig. 5-9(b). Also the TE polarization has smaller thermal drift in the baseline since the mode is more confined inside the top dielectric layer [Fig. 5-1(c)] while the TM polarization has larger intensity at metal dielectric interface [Fig. 5-1(c)] which causes more heat generation due to large thermal conductivity of the gold layer. Figure 5-8(c) shows the SPR response (only TM polarization) to variations of the bulk index at λ=632 nm and λ=900 nm.

94 73 The larger thermal drift of the baseline in SPR sensor can be observed by comparing the baselines of both PWR and SPR sensors. This large thermal drift is due to the fact that the metal layer is in direct contact with the fluid in SPR sensor while in PWR sensor there is a thick dielectric spacer between the metal and the fluid which is further discussed in the next section. Figures 5-10(a-b) show the same response as Fig. 5-9(a-b) with the baseline adjusted to coincide with zero. Change in resonance angle (deg.) PWR-TM, =632nm PWR-TM, =830nm Time (s) 4 Change in resonance angle (deg.) PWR-TE, =632nm PWR-TE, =830nm Time (s) 4 4 Change in resonance angle (deg.) SPR-TM, =632nm SPR-TM, =900nm (a) (b) (c) Time (s) Fig Sensors responses to the bulk refractive index variations: (a) resonance angle versus time for the TM-polarized PWR sensor at λ=632 nm, and 830 nm (b) resonance angle versus time for the TE-polarized PWR sensor at λ=632 nm, and 830 nm (c) resonance angle versus time for the TM-polarized SPR sensor at λ=632 nm, and 900 nm. Solutions are based on (1) DI water, (2) 0.5% ethanol, (3) 1% ethanol (4) 2% ethanol.

95 74 Change in resonance angle (deg.) PWR-TM, =632nm PWR-TM, =830nm Time (s) 4 Change in resonance angle (deg.) (a) (b) (c) PWR-TE, =632nm PWR-TE, =830nm Time (s) 4 Change in resonance angle (deg.) SPR-TM, =632nm 0.10 SPR-TM, =900nm Time (s) 4 Fig Sensors responses to the bulk refractive index variations with baseline adjusted to zero: (a) resonance angle versus time for the TM-polarized PWR sensor at λ=632 nm, and 830 nm (b) resonance angle versus time for the TE-polarized PWR sensor at λ=632 nm, and 830 nm (c) resonance angle versus time for the TM-polarized SPR sensor at λ=632 nm, and 900 nm. Solutions are based on (1) DI water, (2) 0.5% ethanol, (3) 1% ethanol (4) 2% ethanol. Table 5-5 presents the measured sensors characteristics for both sensors at different wavelengths. The refractive index resolution of a sensor (σ RI ) is related to the standard deviation of the sensor output (σ SO ), and sensitivity according to [94]: SO RI. SF Eq. 5-3 As evident from Table 5-5 both PWR polarizations (TM and TE) have smaller standard deviation than the SPR-TM polarization. This is due to the PWR sharper resonance which produces larger SM. The standard deviation of the SPR sensor reduces with increasing wavelength due to its larger SM at longer wavelengths, so the refractive index resolution of the SPR at 900 nm is a smaller number as compared to the resolution at 632 nm.

96 75 Table 5-5. Experimental sensors characteristics calculated from the sensograms shown in Fig. 6. Sensor Polarization Wavelength (nm) (deg./riu) (deg.) (RIU) PWR SPR TM TE TM There are a few important points worth mentioning: 1) The measured values of the standard deviations (, ) and SF which are included in Table 5 are the average values. 2) Standard deviations of the PWR waveguide modes are smaller than the standard deviation of the SPR sensor which makes the average resolution for both PWR modes smaller than the SPR. This is due to the fact that, the SMs of the PWR waveguide modes are larger than the SM of the SPR (see Table 5-3 and Table 5-4). A narrow resonance curve (i.e., a large SM) along with a large SF (Eq. 5-1) results in performance improvement for the PWR in detecting the smallest change in the refractive index [94]. 3) The change in the resolution for the SPR sensor with respect to the excitation wavelength is much smaller than that in each polarizations of the PWR sensor. 4) The polarization diversity in the PWR sensor can be used to remove the interfering surface and bulk effects, to eliminate the need for an ideal reference channel (self-reference measurements) [98], and to increase the measurement reliability by performing multiple independent measurements on the same sample Temperature Fluctuation In this section, the amount of temperature fluctuations in both the PWR and the SPR sensors is compared. One of the main concerns for SPR spectroscopy is the temperature fluctuation in the

97 76 sample bulk refractive index. The temperature fluctuations can be originated from different ways such as: 1) ambient temperature 2) solution temperature and 3) thermal energy dissipated by metal. Generally a steady increase of temperature results in broadening of the SPR resonance curve [104]. The contribution of the refractive index change of aqueous solution on the sensor s sensitivity due to the temperature fluctuation is considerable [104], while the influence of the refractive index change in metal and glass prism on the sensitivity is small. Therefore temperature fluctuations cause significant change in refractive index (RI) of the aqueous solution (d(ri)/dt C -1 ) [105] and hence affect the SP propagation constant. The small thermal conductivity of aqueous solutions (0.56 W.m -1.K -1 ) comparing with 314 W.m -1.K -1 in pure gold [26] is one of the main reasons for this effect. To compensate the temperature variation, a temperature-controlled enclosure can be used also a reference channel or a self-referenced scheme can also reduce the temperature effects [105]. To evaluate the thermally dissipated heat in different sensors, the electromagnetic power dissipation density (P d ) associated with the ohmic conduction can be calculated from the following equation [106]. Pd E. E, Eq. 5-4 where σ is the electrical conductivity and E is the electric field. Assuming that the dielectric materials are lossless, the thermal dissipation in both PWR and SPR sensors originates from the metallic film. In order to compare the thermal heat dissipation in both sensors, the dissipated power density is normalized to the mode power density (pointing vector) in each sensor. Table 5-6 summarizes the normalized dissipated power densities (P d,norm ) in both sensors at different wavelengths.

98 77 Table 5-6 Normalized dissipated power density in the PWR and the SPR sensors Dissipated power PWR, λ=830 nm PWR, λ=632 nm SPR, λ=900 nm SPR, λ=632 nm TM TE TM TE TM TM P d,norm e e The smaller power dissipation at the optimized wavelength in the TM mode of the PWR sensor comparing with the SPR sensor is due to the fact that most of the field intensity is concentrated in silica and aqueous solution rather that metal-dielectric interface. The TE mode has negligible power dissipation since the electric field is close to zero inside the gold film [see Fig. 5-1(c)]. Also the presence of the silica layer with small thermal conductivity of 1.4 W.m -1.K -1 reduces the thermal heat transfer from the gold film to the aqueous solution in the PWR sensor Discussion A PWR sensor was proposed as a promising tool for bulk sensing applications. It was demonstrated that its sharp resonance curves and high sensitivity can result in a smaller refractive index resolution than the conventional SPR sensor. The performance of the PWR and SPR sensors was optimized using a genetic algorithm. The optimized structures were tested with an angular modulation setup. The measured refractive index resolution in the PWR sensor was RIU and RIU for TM and TE polarizations, respectively; and the refractive index resolution for the SPR sensor was RIU for the TM polarization. The improvement in the resolution for the PWR sensor was attributed to the smaller standard deviation of the detector noise level which itself can be traced back to its sharper resonance curves. The polarization diversity of the PWR can be used to improve the measurement reliability.

99 PWR for Self-referenced Spectroscopy As mentioned in chapter 2, self-reference spectroscopy is a method used to address the cross sensitivity to interfering effects. In this section, a new approach based on the plasmon waveguide resonance (PWR) sensor is reported for self-referenced measurement. The main feature of the PWR sensor, its capability to guide two different polarizations (TE and TM) instead of only one in the SPR sensor, is developed as a self-referenced scheme for thin film investigation Principle of Operation The PWR sensor has two distinctive resonances at each polarization [Fig. 5-1 (b)] which react to surface and bulk properties differently. The TM mode has a large penetration depth into the fluid [black line in Fig. 5-1(c)] which makes it suitable to probe variations in bulk refractive index. On the other hand, the TE mode is mostly confined in the top dielectric layer [red line in Fig. 5-1(c)] where its small penetration depth (up to 300 nm) makes it highly sensitive to the refractive index variations at the dielectric/water interface. Therefore, changes in the bulk refractive index and/or surface binding chemistry lead to different shifts of the TM and TE resonance angles (Δθ TM and Δθ TE ). Assuming that the resonance angle shifts are linearly dependent to the surface and bulk parameters, the shifts in both polarizations are given by: bulk surf SF n SF d. TM TM b TM a Eq. 5-5 bulk surf SF n SF d. TE TE b TE a Eq. 5-6 where bulk SFTM and bulk SFTE are the bulk sensitivity factors (SFs) in degree/refractive index unit (RIU) for TM and TE polarizations, respectively. Similarly, surf SFTM and surf SFTE are the surface SFs in degree/nanometer for the TM and TE polarizations, respectively. Lastly, the values of Δd a and

100 79 Δn b are the changes in adlayer thickness and bulk refractive index, respectively. By measuring Δθ TM and Δθ TE and assuming that the SFs are known (for example, calculated from simulation or measured during a calibration process), Eq. 5-5 and Eq. 5-6 can be used to calculate the quantities of interest (Δd a and Δn b ) according to: n SF SF bulk bulk TM TE TE TM b bulk surf surf bulk SFTM SFTE SFTM SFTE. Eq. 5-7 d SF SF surf surf TE TM TM TE a bulk surf surf bulk SFTM SFTE SFTM SFTE. Eq. 5-8 Therefore, by measuring the resonance angles for both polarizations the two unknown variables can be separately determined Optimization One of our goals was to compare the performance of our PWR sensor (with both TM and TE polarizations) with that of a conventional SPR sensor (only TM mode) fabricated and tested under the same conditions. To optimize the performance of the PWR sensor, a FoM is defined based on the CSF (Eq. 2-25) to improve the LoD. Previously, it was shown that the CSF is inversely proportional to the LoD and for surface and bulk sensing applications can be written as: CSF SF SM res bulk bulk nb R R FWHM max min. Eq. 5-9 CSF SF SM res surf surf da R R FWHM max min. Eq. 5-10

101 80 Where, CSFbulk and CSFsurf are the CSFs defined for bulk and surface sensing applications. Since in the case of PWR sensor the TM mode is used to monitor the bulk parameters (due to its large penetration into the fluid ) and the TE mode is used to monitor the surface properties (due to its small penetration into the fluid) the appropriate FoM for the PWR sensor is: FoM CSF CSF TM bulk TE surf. Eq On the other hand, the FoM in the case of SPR sensor (only TM mode is allowed) is defined for surface sensing applications according to: FoM SPR CSF TM surf. In Table 5-7, the optimized design parameters for the PWR and the conventional SPR sensor are listed. In performing the comparison between the PWR and SPR sensors as depicted in Table 5-7 a few points are worth mentioning:1) To model both sensors, a transfer matrix method is used. 2) We have assumed that the substrate and prism are both BK7 glass. 3) Gold is used as the metallic layer, which is the preferred plasmonic layer. 4) A 2 nm titanium layer is added between the substrate and the gold in order to improve gold adhesion to the substrate. 5) Silica is used as the top dielectric layer in the PWR sensor which has a well-understood surface chemistry. 6) In the PWR sensor, the optimized values for the operating wavelength, silica thickness and gold thickness are 780 nm, 545 nm, and 48 nm respectively. 6) In the SPR sensor, the optimized values for the operating wavelength and gold thickness are 880 nm, and 50 nm respectively. Table 5-7. Comparison of the optimized PWR and SPR sensors characteristics Sensor polarization SF surf (deg./nm) SF bulk SM CSF surf CSF bulk (deg./riu) (deg. -1 ) (nm -1 ) (RIU -1 ) TM PWR TE SPR TM

102 Sensor Fabrication and Instrumentation Steps similar to the discussion in section were followed to prepare the SPR sensors. To fabricate our PWR sensors, a silica layer of 545 nm thickness was deposited on the gold film using a plasma enhanced chemical vapor deposition (PECVD). Figure 5-11(a) shows the optical setup used to characterize the performance of the PWR sensor. The difference between this setup and the setup described in section is replacing the triangular prism with a rectangular prism and the addition of a liquid crystal variable retarder (45 degrees orientations), acting as a half-wave plate in front of the linear polarizer. This system alters the light polarization by switching between polarization oriented at 0 degrees (TM polarization) and 90 degrees (TE polarization). Figure 5-11(b) shows the measured and calculated normalized reflectance spectrum. The black curve in Fig. 5-11(b) is the measured spectrum which has a dip close to 44º angle due to the TM mode resonance. There is also a second resonance close to 51º which corresponds to the TE mode. Note that since the TE reflectance is located at the denominator of the normalized spectrum it appears as a peak. The red curve in Fig. 5-11(b) is the reflectance spectrum calculated using the transfer matrix method. The simulation result is in good agreement with the experiment in terms of location of the resonance modes and the normalized reflectance values.

103 TE resonance (a) atan(r TM /R TE ) Measurement Simulation TM resonance Angle (degree) (b) Fig (a) Optical setup used for reflectance measurement. (b) The experimental (black line) and theoretical (red line) normalized reflectance spectrum of the PWR sensor Self-referenced Experiments Biotin-streptavidin complex was used to investigate the response of the sensor to thin film adsorption, as it provides strong affinity and high specificity of interaction. To functionalize the SPR sensor with biotin, a solution of biotinylated PEG alkane thiol was passed over the SPR chip for two hours. On the other hand, the silica surface of the PWR sensor was functionalized with biotin by MicroSurfaces Inc.[107]. The functionalized sensor were then fixed between the prism and the flow cell [Fig (a)] using the index matching oil. Figure 5-12 shows the measured sensograms of the PWR (both TE and TM modes) and the SPR (only TM mode) to surface and bulk effects. For all the surface adhesion measurements, phosphate buffered saline (PBS) solution was used as the buffer for the streptavidin diluted solution and the solution flow rate was fixed to 50μl/min. The solutions were introduced to the sensor in the following order: (1) A pure PBS solution was passed over the sample for 25 minutes to create a baseline. (2)Then a 1g/ml streptavidin solution was passed through the flow cell for 25 minutes in the flow cell to study the sensor response to a low concentration of the streptavidin. (3) A rinsing step was introduced by flowing

104 83 pure PBS solution to dissociate the weakly bounded molecules from the surface. (4) A 10μg/ml streptavidin solution was passed for another 25 minutes to create a streptavidin saturated surface, and (5) another rinsing step was performed. (6) To investigate pure bulk refractive index variations on the sensor response, PBS buffer solution was switched to deionized water (MilliQ 18.2 M.cm). (7) A low concentration (0.01M) of salted water was passed for 5 minutes followed by (8) deionized water flow over the sample to recreate the baseline. (9) A 1% Vol. ethanol solution (water diluted) was passed for another 5 minutes, and (10) deionized water was flowed over the sample to recreate the baseline. Change in resonance angle (deg.) (a) Time (s) Change in resonance angle (deg.) (b) PWR-TM PWR-TE Time (s) Fig (a) Angular positions of the resonance dip vs. time for the SPR sensor. (b) Angular positions of the resonance dip vs. time for the PWR sensor. Solutions are (1) PBS, (2) 1μg/mL Streptavidin, (3) PBS, (4) 10 μg/ml Streptavidin, (5) PBS, (6)DI water, (7) 0.01M salted water, (8) DI water, (9) 1% ethanol, and (10) DI water.

105 n b (RIU) d a (nm) Time (s) Fig Surface binding thickness and bulk refractive index change calculated from Fig. 5-12(b). Solutions are (1) PBS, (2) 1μg/mL Streptavidin, (3) PBS, (4) 10 μg/ml Streptavidin, (5) PBS, (6)DI water, (7) 0.01M salted water, (8) DI water, (9) 1% ethanol, and (10) DI water Figure 5-12(a) shows the response of the SPR sensor, for the TM polarized light, to surface and bulk effects. Figure 5-12(b) is the response of the TE (red curve) and TM (black curve) modes of the PWR sensor to surface and bulk variations. Both curves have been measured simultaneously by tracing the resonance angles of each mode. As shown in Fig. 5-12(b), the TE resonance angle experiences a larger change with respect to the surface effects as compared to the TM mode [step 1 to 5 in Fig. 5-12(b)] while the TM resonance angle is more affected by the change in bulk refractive index [step 6 to 10 in Fig. 5-12(b)]. Table 5-8 summarizes the measured and calculated values of the SPR and PWR sensors characteristics. The bulk sensitivities of each mode were measured by finding the change in resonance angle from the sensogram and the change in the refractive index of the buffer solution. This information along with the measured standard deviation of the sensors outputs was used to calculate the bulk sensor Resolution (see section 2.5.4) and is shown in Table 5-8.

106 85 However, in the case of surface sensitivity factor (SF surf ) it is more difficult to determine the exact thickness of the streptavidin layer experimentally; although it can be estimated from the simulation and assuming a refractive index of 1.5 for streptavidin at saturation [108]. We have estimated the thickness of the streptavidin layer to be 4.3 nm which is close to its molecule size ( nm 3 ) [109]. The approximate values of the surface sensitivity is then calculated and given in Table 5-8. Using the measured standard deviation of the sensors outputs and the surface sensitivity factor, the values of LoD defined as the smallest detectable concentration of analyte (ng/ml) can be calculated and is given in Table 5-8. Table 5-8. Experimental sensor characteristics obtained from the sensograms shown in Fig Sensor polarization SF surf (deg./nm) PWR SF bulk (deg./riu) LoD (ng/ml) Resolution (RIU) TM TE SPR TM From Table 5-8 it is clear that the LoD for the TE mode of the PWR is smaller than that of the TM mode of the PWR and the TM mode of the SPR. This makes the TE mode of the PWR an excellent choice for sensing surface effects, as it was discussed earlier. Similarly, the resolution of the TM mode of the PWR is smaller than that of the TE mode of PWR and TM mode of the SPR. This makes the TM mode of the PWR an excellent choice for sensing bulk effects, as discussed earlier. In addition, the simultaneous presence of TM and TE modes in the PWR allows us to adequately decouple the surface and bulk properties. Figure 5-13 shows the calculated adlayer thickness change and bulk index variations as a function of time using the linear model given by Eq. 5-7 and Eq The black line shows the background index variations as a function of time and as it can be seen this response is almost constant during the streptavidin attachment [step 1 to 5 in Fig. 5-12(b)] and only changes when

107 86 the variations in buffer refractive index are introduced [step 6 to 10 in Fig. 5-12(c)]. The red line corresponds to the change in adlayer thickness which increases with the attachment of the streptavidin to biotin [step 1 to 5 in Fig. 5-12(b)] and remains constant when only the buffer refractive index changes [step 6 to 10 in Fig. 5-12(b)]. Therefore, the change in adleyer thickness is completely decoupled from the variations in bulk refractive index by using simultaneous dual polarization spectroscopy at a single sensing channel at a single wavelength. To evaluate the decoupling efficiency of the PWR sensor, the determinant of the S matrix in Eq is calculated which has the value of 4.18 and confirms that the S matrix is not singular normalized S S' Eq To calculate the condition number of the S matrix, the elements of the matrix are normalized to the calculated sensitivity values of the SPR sensor which makes the elements of the S matrix dimensionless (Eq. 5-12). The calculated condition number for the normalized S matrix (S ) is 1.7 which confirms that the S matrix is a well-conditioned matrix and the interfering effects can be calculated with the same precision as that of the resonance angle Discussion In this section, a new approach, based on a PWR sensor, for self-referenced biosensing was proposed and demonstrated. The unique property of the PWR sensor which is its polarization diversity was utilized for dual-polarization spectroscopy. The structural properties of the PWR and SPR sensors along with the operating wavelength were optimized with a genetic algorithm to obtain the best performance. The TM mode of the PWR sensor demonstrated high sensitivity to refractive index change in bulk solution while the TE mode was more sensitive to the thickness change of the attached streptavidin layer. The measured LoD for both TM and TE polarizations were 55 ng/ml and 9 ng/ml, respectively and the measured resolutions were RIU and RIU, respectively. The measured LoD for TM polarization in the SPR sensor was 63

108 87 ng/ml and the measured resolution was RIU. Moreover, a linear model was used to decouple the changes in adlayer thickness and bulk refractive index from the sensor s response. Finally, the performance of the PWR sensor was compared to a SPR sensor, fabricated and tested under the same conditions. It was seen that the PWR sensor provides a better LoD and resolution in addition to its ability to decouple surface and bulk effects.

109 Probing Protein-nanoparticle Interactions Using a Plasmon Waveguide Resonance In this section, a plasmon waveguide resonance (PWR) sensor is proposed for studying gold nanoparticle-protein interactions. The polarization diversity of the PWR sensor enables simultaneous spectroscopy of the nanoparticles surface reactions with two different light polarizations (TM and TE). The response of each mode to streptavidin-biotin binding at the surface of gold nanoparticles is investigated in real time. Finally the nanoparticle surface reactions are decoupled from the bulk solution refractive index variations with dual mode spectroscopy Protein-nanoparticle Interaction Nanoparticles (NPs) have attracted much interest due to their unique chemical and physical properties. They have been used in different technologies including electronics, medicine, textile, cosmetic product, etc. [110]. The development of NPs as drug carriers and imaging agents in the biomedical field [111] makes them a novel functional material. Their small size enables them to enter biological membrane and interact with biomolecules such as proteins. The interaction between NPs and proteins can alter protein conformation thus affecting the cellular functionality resulting in toxicity [112]. Consequently, there is a growing concern about NP toxicity [113] and biocompatibility which compels further investigation of protein-np interaction. There are several methods to investigate the biological effects of NPs and specifically NPprotein interactions which can improve the understanding of the biological activities of the NP for safer use of these materials. Such methods are used to analyze the binding affinity and ratios [114, 115], conformational variations [116, 117], and kinetic binding properties [112, 118] of the NP-protein interaction. Surface plasmon resonance (SPR) is one of the most well-known techniques to study NP-protein binding kinetics [112]. The NP-protein interaction is a dynamic event involving constant association and dissociation processes that can be investigated using the

110 89 SPR method. The rate of association and dissociation depends on the protein and particle type. However, the SPR technique has some drawbacks. The underlying gold film which is in direct contact with NPs has a large thermal conductivity which can create thermal fluctuations. This effect can cause thermal denaturing of the protein layer resulting in false measurement [119]. As well, the resonance width of the NP-modified SPR sensor is broader than other techniques such as quartz crystal microbalance (QCM) due to loss of the gold film [118, 120]. In this section, we propose a new approach for kinetic study of NPs based on the PWR sensor. The ability to carry spectroscopic measurement using two different polarizations with sharp resonances instead of one broad resonance (TM polarization) in the SPR sensor enables the PWR sensor to extract more information with better accuracy from the specimen. Furthermore, the presence of the thick dielectric layer between the metal film and the specimen reduces the aforementioned thermal heat effects of the metal on the analyte [100]. Finally to study the AuNP-protein interactions, the functionalization of AuNPs can be performed in situ as the sensing surface is a dielectric with different surface chemistry than AuNPs. However in the case of SPR technique, both the sensing surface and NPs material are gold which demands functionalization of Au NPs prior to the test in order to only allow for interaction between NPs (but not the sensing surface) and the target molecules. As a proof of concept, the interaction between biotinylated-aunps of two different sizes to streptavidin protein is investigated with dual polarization spectroscopy Principle of Operation Figure 5-14(a) shows the schematic of the PWR sensor based on a Kretschmann configuration in which the AuNPs are immobilized on the top surface. The light is incident on the substrate through a prism and the reflected light intensity is measured as a function of the incident angle. Two different resonance modes (TM and TE) can be excited based on the polarization of the incident light. The field profiles of both TM and TE modes are shown in Fig. 5-14(b). As is depicted in this figure, the TM mode has a larger penetration depth into the fluid (up to 9 μm) while TE mode is mostly confined within the dielectric layer with a small tail (up to 300 nm)

111 90 extending into the fluid. The sharp resonances of the PWR modes [Fig. 5-14(c)] can be used to improve the sensor s characteristics by reducing the standard deviation of the output signal [100]. (a) Normalized poynting vector (a.u.) (b) x(m) TM TE Reflectance (c) Angle TM TE Fig (a) Schematic diagram of the AuNP-modified PWR sensor. (b) z-component of the Poynting vector for both TM and TE polarizations in the optimized PWR sensor at the resonance angle of 61.65⁰ and 65.23⁰, respectively. (c) Reflectance spectrum for the optimized PWR-TM, PWR-TE in black and red lines, respectively. Any refractive index variation on the surface of NPs or inside the bulk fluid alters the modes propagation constants differently due to the difference in their field profiles. This difference in the response of each mode to the surface or bulk properties manifesting as changes in resonance angles ( and ) is used to decouple the surface and bulk effects. Assuming that the resonance angle shifts are linearly dependent on the surface and bulk effects and that the sensitivity factors are known (calculated from simulation or measured during calibration process), the variations in the protein concentration surrounding the NP surface Δc and the bulk refractive index change Δn b can be decoupled using the following equations: n SF SF bulk bulk TM TE TE TM b bulk surf surf bulk SFTM SFTE SFTM SFTE. Eq SF SF c SF SF SF SF surf surf TE TM TM TE bulk surf surf bulk TM TE TM TE. Eq. 4 12

112 91 where bulk SF TM and bulk SF TE are bulk sensitivity factors in degree/refractive index unit (RIU) for TM and TE polarizations, respectively. While surf SF TM and degree/concentration for TM and TE polarizations, respectively. surf SF TE are surface sensitivity factors in In order to calculate the surface and bulk sensitivity factors, a transfer matrix method is used along with an effective medium theory to model the interactions between nanoparticles embedded in a homogenous medium. Maxwell-Garnett effective medium theory [121] is one of the most useful models to find the effective permittivity of random composites. Accordingly, the effective index of AuNP arrays embedded in a solution of water can be expressed as the following equation: eff g g 1 2 f 2 m(1 f) m. 1 f (2 f) m Eq where f is the filling fraction of the AuNPs ( g ) inside the fluid ( m ). The bulk sensitivity can be determined by calculating the change in resonance angles due to the change in the bulk fluid refractive index. The surface sensitivity can be estimated by calculating the change in resonance angles due to small change of the refractive index in the NPs surrounding medium while the bulk solution refractive index remains unchanged Sensors Fabrication and Functionalization To prepare the SPR sensors, similar steps to what is explained in section is followed [Fig. 5-15(a)]. To fabricate our PWR sensors, a silica layer of 540 nm thickness was deposited on the gold film using a plasma enhanced chemical vapor deposition (PECVD) [Fig. 5-15(b)]. The PWR sensors were first rinsed with DI water and ethanol and subsequently incubated at room temperature in a solution of a bifunctional siloxane 3-aminopropyltrimethoxysilane (APTMS; 1 mm in ethanol) for 10min and then spun out at 2000rpm for 30 sec to uniformly cover the sensor with a self-assembled monolayer (SAM) of APTMS [Fig. 5-15(c)]. The sensors

113 92 were then rinsed with ethanol, DI water, N 2 dried, and placed in a 100 C oven for 1 hour to complete Si O bond formation. The AuNPs were assembled on the sensor surface by incubating the sensors with AuNP solution for 10 min [Fig. 5-15(d)]. The samples were then spun out at 2000rpm for 30 sec, rinsed with DI water, N 2 dried, and placed in a 100 C oven for 1 hour to improve the affinity between the AuNPs and APTMS. Fig (a) Schematic diagram of the BK7 glass substrate with 48 nm of gold film. (b) Deposition of 550 nm silica on the gold film using PECVD. (c) The PWR sensor coated with a SAM of APTMS. (d) Binding Au NPs on the SAM of APTMS. Two different solutions of AuNPs each with different sizes (20 nm and 100 nm) are immersed on the PWR sensors. The images of the AuNPs with two surface densities , and particles/cm 2 for AuNPs with a diameter of 20 nm and 100 nm respectively were obtained by scanning electron microscopy (SEM), as shown in Fig

114 93 Fig SEM images of gold nanoparticles on the PWR surface with diameters of (a) 20 nm with surface density of particles/cm 2 (b) 100 nm with surface density of particles/cm Instrumentation Figure 5-17(a) shows the optical setup used to characterize the performances of the PWR sensor. The details of the setup was previously mentioned in sections and Figure 5-17(b) shows the normalized reflectance spectrum obtained from the PWR sensors covered with 20 nm and 100 nm AuNP at the wavelength of 780 nm and 720 nm respectively. The first resonance for both samples appears close to 61 angle and corresponds to TM mode resonance. The second resonance appears close to 67º angle which corresponds to the TE mode but since the TE reflectance is located at the denominator of the normalized spectrum it appears as a peak.

115 TE resonance 0.8 (b) R TM /R TE nm AuNP-PWR 100nm AuNP-PWR (a) 0.4 TM resonance Angle (deg.) Fig (a) Proposed optical setup for minimum position measurement. (b) The experimental (black line) and theoretical (red line) normalized reflectance spectrum of the PWR sensor Self-referenced Experiments To investigate the response of the PWR sensors to NP-protein interaction, we used the biotinstreptavidin complex as it provides strong affinity and high specificity of interaction. To functionalize the AuNPs with biotin, a solution of biotinylated PEG alkane thiol is passed over the PWR surface (covered with Au NPs) for two hours. The functionalized sensor is then fixed between the prism and the flow cell [Fig. 5-17(a)] using the oil matching index. Figure 5-18 shows the measured sensograms obtained from the attachment of streptavidin to biotinylated AuNPs of two different sizes. Each sensogram is obtained at two different wavelengths since the optimized wavelength that provides narrow resonances for both modes is different. These wavelengths are 780 nm and 720 nm for 20 nm AuNP-PWR sensor and 100 nm AuNP-PWR sensor, respectively. In all experiments, phosphate buffered saline (PBS) solution is used as the buffer for the streptavidin diluted solution. The solutions were introduced to the sensor in the following order: (1) First a pure PBS solution is passed over the sample for 25min to create a baseline. (2)Then a streptavidin solution with concentration equal to 1g/ml, is passed during 25min in the flow cell to study the sensor response to a low concentration of the sample. (3) A rinsing step is introduced for 10 min by flowing pure PBS solution to dissociate the weakly bounded molecules from the surface. (4) 10μg/ml streptavidin solution is passed for another 25min to get a streptavidin

116 95 saturated surface (5) and another rinsing step is performed. (6) To investigate pure bulk refractive index variations on the sensor response, PBS buffer solution is switched to deionized water (MilliQ 18.2 M.cm). (7) Then 1% Vol. ethanol solution (water diluted) is passed for 5min. (10) and finally, deionized water flows over the sample to recreate the baseline. Change in resonance angle (degree) nm AuNP-PWR 100 nm AuNP-PWR (a) 5 TM TE Change in resonance angle (degree) (b) TM TE Time (s) Time (s) Fig Angular positions of the resonance point vs. time for (a) 20 nm and (b) 100 nm AuNP binded to the PWR sensor. Solutions are (1) PBS, (2) 1μg/mL Streptavidin, (3) PBS, (4) 10 μg/ml Streptavidin, (5) PBS, (6)DI water, (7) 1% ethanol, and (8) DI water. As evident from Fig. 5-18, the TM mode is more sensitive to the bulk index variations which can be seen by comparing the change in the resonance angles when the PBS buffer solution is switched to DI water (step 5 to step 6 in Fig. 5-18) or the DI water is switched to 1% Vol. ethanol solution (step 6 to step 7 in Fig. 5-18). The sensitivity of the TE mode is more to AuNPprotein interaction than the bulk index variations. The difference between the measured response of both TE and TM mode is compared with the simulation results in Table 5-9. Although it is difficult to simulate the AuNPs surface binding and their bulk index response, the ratio of the sensitivities for both modes can be determined from simulation and compared with the experimental results.

117 96 Table 5-9. Comparison between experimental and theoretical values of the sensitivity factor s ratio 20 nm AuNP-PWR 100 nm AuNP-PWR Method SF SF surf TE surf TM SF SF bulk TE bulk TM SF SF surf TE surf TM SF SF bulk TE bulk TM Experiment Theory In Table 5-9, the surface sensitivity of the TE mode is almost five times larger than that of the TM mode in 20 nm AuNP-PWR sensor. However, in the case of 100 nm AuNP-PWR both TE and TM mode have similar surface sensitivities. The difference in the response of TM mode to surface bindings in both samples can be explained based on the variations of the wave vector which is proportional to the overlap integral [32]: k E E dr. E E dr * k. i. f. i Vin * 2. i. i. V Eq where E i,and k i are the electric field and its wave vector before change in refractive index induced by surface bindings, E f is the electric field after change in refractive index, V in is the volume of interaction between the field and the analyte, V is the total volume covered by the field, and δk is the change in wave vector due to change from ε to ε+δε in the analyte permittivity value. Table 5-10 summarizes the change in wave vector for both modes of each sample which is calculated using Eq and by substituting the field profile from a transfer matrix method in which an effective layer is considered for AuNP using Maxwell-Garnett theory [Eq. (3)].

118 97 Table Comparison between theoretical wave vector variations due to perturbation of the refractive index close to AuNPs Polarization δk 20 nm AuNP δk 100 nm AuNP TE mode TM mode Although it is difficult to precisely simulate the variations of wave vector due to surface binding within a few nanometer around AuNP, the calculated values of the wave vector variations due to refractive index change of the AuNP surrounding medium confirms that the TM mode becomes more sensitive to surface effects in 100 nm AuNP-PWR sensor. In other words, the longer penetration of the TM polarization into the solution than that of TE polarization [Fig. 5-14(b)] makes the TM mode more sensitive to surface binding on larger AuNPs. Therefore the small difference between surface sensitivities of TE and TM modes in 100 nm AuNP-PWR sensor is verified from overlap integral. While based on the results presented in Table2, this difference is more pronounced in the case of 20 nm AuNP-PWR which is also observed from the experiment [Fig. 5-18(a)]. The ratios of bulk sensitivities in both sensors are similar since the bulk sensitivity is less perturbed by the presence of AuNPs. The difference between the theoretical and experimental results can be due to several reasons such as the AuNPs aggregation and their non-uniform surface coverage, the presence of APTMS layer which is not included in the simulation, and the fabrication imperfections. Figure 5-19(a) and (b) shows the decoupled surface binding change and bulk index variations as a function of time for AuNP-PWR sensors (two different samples, one immobilized with 20 nm AuNP and the other with 100 nm AuNP), respectively using the linear model given by Eq and Eq The blue line shows the background index variations as a function of time which is almost constant during the streptavidin attachment [step 1 to 5 in Fig. 5-19] and it changes when the variations in buffer refractive index is introduced [step 6 to 10 in Fig. 5-19]. The black line corresponds to the change in surface binding which increases with the attachment of the

119 n b (RIU) n b (RIU) 98 streptavidin to biotinylated AuNP [step 1 to 5 in Fig. 5-19] and remains constant only when the buffer refractive index changes [step 6 to 10 in Fig. 5-19]. Therefore, with dual mode spectroscopy the change in protein concentration on AuNPs is completely decoupled from the variations in bulk refractive index. c (arb. units) nm AuNP-PWR 100 nm AuNP-PWR 3 4 (a) Time (s) Time (s) c (arb. units) (b) Fig Decoupled surface binding change and bulk index variations as a function of time for (a) 20 nm AuNP-PWR sensor (b) 100 nm AuNP-PWR sensor. Solutions are (1) PBS, (2) 1μg/mL Streptavidin, (3) PBS, (4) 10 μg/ml Streptavidin, (5) PBS, (6)DI water, (7) 1% ethanol, and (8) DI water Discussion In this section, a new approach for studying protein-aunp interaction was proposed based on a PWR sensor. The AuNPs were immobilized on the PWR sensor and then functionalized with biotin. The polarization diversity of the PWR sensor was utilized for dual-polarization spectroscopy of streptavidin interaction with biotinylated-aunp. Two different AuNPs were immobilized on two different PWR sensors, one with diameter of 20 nm and surface density of particles/cm 2 and the other with diameter of 100 nm and surface density of particles/cm 2. The response of TE and TM modes to protein-aunp binding and bulk solution

120 99 refractive index variations were monitored in real time for both samples. A linear model was used to decouple the AuNP surface binding from bulk index variations. 5.5 Conclusion In this chapter, a PWR sensor was investigated for three different applications. In the first application, it was demonstrated that the PWR sensor performs better than the SPR sensor as a refractive index sensor in terms of the sensor s resolution. The second application was selfreferenced measurement with the PWR sensor for thin film investigation. For this purpose, the TM polarization with high sensitivity to the bulk index change and the TE polarization with high sensitivity to the surface bindings were utilized in order to differentiate the bulk index change from the adlayer thickness variations. And finally, the PWR sensor was applied for kinetic analysis of protein-nanoparticle interactions while the nanoparticles were immobilized on the PWR surface. Therefore the PWR sensor demonstrates promising applications especially when a self-referenced measurement is needed.

121 100 6 Three Mode Spectroscopy As mentioned in section 2.7.2, one possible method to decouple the variations in three interfering effects (adlayer refractive index n a, adlayer thickness d a, bulk reafractive index n b ) from the output signal is by using three-mode spectroscopy at a single sensing channel. In this chapter, two platforms are proposed which can be used to spectroscopy with three or more independent modes. Both platforms are developed based on one dimensional periodic structure. The first platform contains a one dimensional dielectric grating loaded on a conventional SPR sensor. In this platform, the polarization diversity enables dual polarization spectroscopy at single wavelength which provides more than three resonance modes each optimized for high sensitivity to a particular quantity of interest. The second platform utilizes a one dimensional metallic grating which benefits from two diffraction order excitation and dual wavelength spectroscopy. The excitation mechanism in each platform is explained in the following sections. 6.1 Dual Polarization Spectroscopy: Dielectric Grating In this section, a new design for a SPR biosensor, which allows the interfering bulk and surface effects to be isolated is presented. This structure consists of a SPR sensor coupled with a dielectric grating on top. The metallic layer of the SPR sensor supports a SPR mode which has a transverse magnitude (TM) polarization and the dielectric grating adds two additional waveguide modes: one with TM polarization and the other with TE polarization. Therefore, three different measurements, at two different polarizations can be used to obtain the three desired parameters, i.e. n a, d a, and n b. Also because the proposed sensor operates at a single wavelength, there is no need to consider the temporal dispersion properties of the solution or adlayer. The structure is optimized with the help of a genetic algorithm and the optimum design along with a method to decouple the interfering bulk and surface effects is discussed.

122 101 Figure 6-1(a) shows a schematic diagram of a DGSPR sensor. In this geometry, the momentum of the incident light is adjusted by diffraction at a rectangular grating to match that of a SP wave. Assuming that the dispersion properties of the SP wave is not disturbed by the grating [122], the momentum conservation between the free space incident light and the SP wave in the DGSPR sensor is given by Eq. 6-1: m d np sin( SPR ) m. m d Eq. 6-1 where θ SPR, ε m, ε d, n p,, and m represent the angle of incident at the resonance, metal permittivity, top dielectric permittivity, prism index of refraction, incident light wavelength, the period of the grating, and the order of diffraction respectively. The grating diffracts the incident beam into a series of diffraction orders, with different values of momentum which can be designed to match the momentum of the surface plasmon (Fig. 6-1a). Sign + and sign in Eq. 6-1 correspond to positive (m>0) and negative (m<0) diffraction orders. A schematic diagram of the cross-section of the DGSPR along with its design parameters is shown in Fig. 6-1b. Fig. 6-1 (a) three dimensional (3D) schematic of the DGSPR sensor. (b) 2D cross section of the DGSPR sensor. Silica nanowires of a rectangular profile are assumed to be infinite in length with periodicity of Λ, thickness of h and filling factor of w/λ.

123 102 Another advantage of the dielectric grating is that it can be used for blood cell separation in blood sensing application [123]. The microslit arrays of the dielectric grating can physically filter the large blood cells from serum proteins (Fig. 6-2). Therefore large blood cells and small serum proteins can be studied independently using three modes of the DGSPR sensor. Fig. 6-2 Size-exclusion SPR sensor chip with microfabricated slit array (Filter SPR chip) for separating blood cells and detecting proteins in whole blood sample. (figure and caption obtained from [123]) Numerical Model and Optimization One simple approach to macroscopically study the light propagation in a dielectric grating is to consider it as a homogeneous effective medium with an averaged dielectric permittivity [124].

124 103 The zero-order approximation of the effective medium theory (EMT) for alternating slabs of dielectric materials with refractive indices of n 1 and n 2 and a period of Λ is given by: n fn (1 f ) n. TE Eq. 6-2 n TM f n 1. (1 f) n Eq. 6-3 where f is the volume fraction of material with index n 1. n TE and n TM are the effective refractive index when the electric field of the light is parallel and perpendicular to the slabs respectively. Based on the EMT theory, the DGSPR sensor can be simplified to a SPR sensor loaded with an effective dielectric layer (n eff ) and is similar to a plasmon waveguide resonance (PWR) sensor [125]. Therefore, the DGSPR is capable of supporting two different polarizations as the PWR sensor. In the next section the EMT approximation is used to characterize the dispersion properties of the DGSPR sensor. We used rigorous coupled wave analysis (RCWA) to simulate the gratings. The RCWA is an exact solution of Maxwell s equations for electromagnetic fields which can calculate the diffraction efficiency of the reflected or transmitted light intensity from gratings [126]. An efficient implementation of RCWA (see Appendix B [127]) is used in this work to optimize the DGSPR sensor for multimode spectroscopy. The convergence in RCWA computation was achieved by including 21 space harmonic orders. As explained earlier, multimode spectroscopy is the main goal of this work. To simultaneously excite multiple modes in this DGSPR sensor, several parameters should be adjusted which are as follows: the grating periodicity, grating height, filling factor, gold thickness, and the wavelength of the incident light. To optimize the design and to evaluate its performance for a biosensing application, a proper FoM should be defined. In this work, the FoM is defined as the product of CSF corresponding to each mode: Eq. 6-4

125 104 where and are the CSF corresponding to the adlayer thickness variations for both TM1 and TM2 modes, respectively and is the CSF corresponding to the bulk refractive index change for TE polarization. This FoM is adopted as the fitness function for optimization using the GA. The DGSPR sensor is optimized using a parallel implementation of RCWA along with the GA (see Appendix C) on the general purpose cluster at the Scinet supercomputer in the University of Toronto due to the computational intensity of the problem Results Table 6-1 summarizes the optimized design for the DGSPR and the SPR sensors. The DGSPR sensor is optimized so that two TM modes each of which is highly sensitive to surface properties and one TE mode which is more sensitive to bulk effects are all excited at a single wavelength. BK7 glass is used as the substrate and as the prism material in both structures. A 2 nm Ti adhesion layer between the substrate and gold coating is also included in the simulation. Table 6-1. The optimized dimensions of SPR and DGSPR sensors for affinity sensing Wavelength (nm) Gold thickness (t) Grating thickness(h) Periodicity (Λ) Fill factor (w/ Λ) SPR nm DGSPR nm 992 nm 890 nm 0.4 The diffraction efficiency of the reflected light intensity for both polarizations is shown in Fig. 6-3(a). As shown in this figure, two modes are excited with TM polarized light (black line) at two different resonance angles and one mode with TE polarized light (blue line). Figure 6-3(b) shows the spatial distribution of the Poynting vector (power density) along the propagation

126 105 direction (z) as a function of the vertical direction (x), for all three modes at their resonance angles. According to this figure, the TM1 mode (black line) is a hybrid mode generated by a coupling between the waveguide mode of the dielectric grating and the SP wave propagating on top of the gold surface. The TM2 mode (red line) is a SP wave propagating at the metal /dielectric grating interface. The TE1 mode (blue line) is a dielectric mode mostly confined inside the grating region. Fig. 6-3(a) Reflectance spectrum for the optimized DGSPR sensor (b) z-component of the Poynting vector for both TM and TE polarizations at the resonance angle of 62.28⁰, 71.11⁰, 63.5⁰ for TM1, TM2,and TE1 mode respectively. Figure 6-4 shows the two dimensional profile of the Poynting vector along the z axis for three modes of the DGSPR sensor. We have used the commercial FDTD code Lumerical for calculating the power profile of the modes.

127 106 (a) (b) (c) Fig. 6-4 z-component of the Poynting vector along DGSPR cross section (x-z plane) for (a) TM1 mode, (b) TM2 mode, and (c) TE1 mode. The performance characteristics of each mode to variations in surface and bulk parameters are shown in Table 6-2. The CSF has a different definition based on the quantity to be measured. These quantities are categorized into surface and bulk effects. The surface effects contain adlayer thickness, and adlayer index and therefore the corresponding CSF are CSF surf,thick and CSF surf,index The bulk effect only contains the bulk refractive index and the corresponding CSF is CSF bulk. The TM modes have CSF surf,thick and CSF surf,index in the same range as the optimized SPR sensor while the TM1 mode has an order of magnitude larger CSF bulk than that of the SPR sensor due to its longer penetration depth into the water [black line in Fig. 6-3(b)]. The TE polarized mode has a smaller CSF surf,thick and CSF surf,index than that of the TM modes but it has large CSF bulk which makes it more suitable for detecting bulk index variations with less cross sensitivity to surface effects. Table 6-2. The performance characteristics of the optimized SPR and DGSPR sensors. Sensor TM1 TM2 TE1 CSF surf,thick CSF surf,index CSF bulk CSF surf,thick CSF surf,index CSF bulk CSF surf,thick CSF surf,index CSF bulk SPR nm RIU RIU -1 DGSPR nm RIU RIU nm RIU RIU nm RIU RIU -1

128 107 To further investigate the characteristics of the modes, the dispersion properties of the TM modes have been calculated. Figure 6-5(a) represents the dispersion curve obtained from the calculation of the reflected light diffraction efficiency using RCWA method for a range of wavelengths. The reflectivity minima represent the phase matching conditions for which the incident beam excites the TM modes. The narrower dip at shorter angles corresponds to the hybrid SPR-waveguide mode (TM1) and the wider dip corresponds to the pure SPR mode (TM2). In Fig. 6-5(b), the same curve is plotted using the transfer matrix method (see Appendix A) [90] to calculated the reflection from multilayer structure assuming that the dielectric grating is replaced with a homogeneous effective medium (Eq. 6-3). Comparison of the curves in Fig. 6-5(a) and (b) illustrates the similarity between the results of the RCWA and the transfer matrix methods. The small mismatch between these two curves is the existence of anticrossing bandgap in the RCWA result [Fig. 6-5(a)]. To explain the reason for this behavior, the dispersion equation (Eq. 6-1) is used to analytically calculate the dispersion curves of the SP waves propagating at the interface of the gold/dielectric grating interface [Fig. 6-5(c)]. There are three different curves in this figure: 1) The red curve corresponding to the propagation constant of the SP wave without any contribution from the grating (zero-order), 2) The blue curve which includes the effect of the negative second-order of diffraction on the SP wave propagation constant. 3) The black curve containing the effect of the negative third-order of diffraction on the SPR propagation constant. Therefore, the interaction between forward (zero-order) and backward (negative second and third-orders) traveling waves is the origin of the anticrossing bandgaps in Fig. 6-5(a) [ ] Fig. 6-5 Dispersion relation of optimized DGSPR sensor calculated using: (a) RCWA method (b) transfer matrix method (c) Analytically calculated dispersion relation of the SP wave in the optimized DGSPR sensor

129 108 So far the characteristics of the excited modes are discussed. Here, a method to utilize the multimode spectroscopy to extract surface and bulk effects is presented. Assuming that the resonance angles of the modes are linearly related to the changes in the solution refractive index and to the properties of adlayer, the change in the resonance angles can be expressed as: TM 1 TM 1 TM1 da na nb TM 1 da TM 2 TM 2 TM 2 TM 2. na. da na n b TE1 n b TE1 TE1 TE1 da na n b S Eq. 6-5 where Δd a, Δn a, and nb are the variations in adlayer thickness, adlayer refractive index and bulk refractive index respectively. The S matrix contains the sensitivity of each mode to the interfering surface or bulk effects. By measuring the resonance angle variations and the bulk and surface sensitivities from the experiment, the change in the interfering surface and bulk effects can be calculated directly from: da TM 1 1 na S TM 2. n b TE1 Eq. 6-6 Equation 6-6 can be solved for Δd a, Δn a, and Δn b only if S is not singular. In Eq. 6-7, the corresponding values for each element in the S matrix are given. The determinant value of 0.2 confirms that the S matrix is not singular normalized S S' Eq. 6-7 To calculate the condition number of the S matrix, the elements of the matrix are normalized to the optimized sensitivity values of the SPR sensor which makes the elements of the S matrix

130 thick thick thick thick 109 dimensionless The small condition number of 77 for the normalized S matrix (S ) corresponds to two digit of maximum inaccuracy in the solutions of Eq The small condition number of the S matrix (<100) along with large CSFs, which corresponds to small uncertainty in the output signal (Δθ i ), improve the decoupling efficiency of the unknown parameters (Δd a, Δn a, and Δn b ). Figure 6-6 shows the tolerance of CSF bulk and CSF index to the variations in structural dimensions of the DGSPR sensor for TE and TM polarizations, respectively. The tolerance of the TE mode is more susceptive to change with grating thickness. As is shown in Fig. 6-6(b), any grating thickness above 980 nm results in a CSF bulk with a larger value than 100 RIU -1 which is large enough for decoupling the variations in bulk refractive index from surface effects. The tolerance of the TM modes is largest to the variations in gold film thickness [see Fig. 6-6(c)] which can be adjusted to a certain thickness with enough accuracy (e.g. 50±1 nm) using the standard deposition techniques. CSF TE bulk (a) TE1 TM1 TM Duty cycle CSF TM CSF TE bulk TE1 TM1 TM2 (b) Grating thickness (nm) CSF TM CSF TE bulk TE1 TM1 (c) TM Gold film thickness (nm) 0.02 CSF TM CSF TE bulk TE TM1 TM2 (d) Periodicity (nm) CSF TM

131 110 Fig. 6-6 The tolerance of TE CSF bulk and TM CSFthick to the variation in (a) duty cycle (b) grating thickness (c) gold film thickness (d) periodicity of the DGSPR sensor Five Mode Spectroscopy Increasing the number of resonance modes of the DGSPR sensor is possible by adjusting the grating dimensions and material properties. In this regard, the DGSPR sensor is modified such that the silica grating in Fig. 6-1 is replaced with an SU8 grating which is an epoxy-based negative photoresist [64] which has larger index of refraction than silica. The main advantage of SU8 is that the grating patterns can be produced in SU8 using E-beam/UV lithography while in the case of silica we need to use lift-off process which is more complicated especially for thick layers and requires additional fabrication steps. The DGSPR sensor is optimized to acquire five resonance modes at both TM and TE polarizations. The optimized dimensions of the DGSPR sensor consisting of a SU8 grating are presented in Table 6-3. Table 6-3. The optimized dimensions of DGSPR sensor consisting of a SU8 grating. Wavelength (nm) Gold thickness (t) Grating thickness(h) Periodicity (Λ) Fill factor (w/ Λ) DGSPR nm 2 μm 930 nm 0.3 Figure 6-7(a) shows the calculated diffraction efficiency corresponding to 2 μm thick SU8 grating. As evident from this figure, there are three resonance modes for TM polarized light and two resonance modes for TE polarization. However by reducing the grating height from 2 μm to 1 μm the number of resonance modes reduces from five modes to three modes as shown in Fig. 6-7(b).

132 111 TE2 TM1 TE1 TM2 TM3 TM1 TE1 TM2 Fig Reflectance spectrum for the optimized DGSPR sensor consisting of a SU8 grating with (a) 2μm height (b) 1μm height. Table 6-4 summarizes the performance characteristics of the DGSPR sensor with SU8 grating. For simplicity the CSF surf,thick is reported for TM modes which are more sensitive to surface reactions and CSF bulk is included for TE modes which are more sensitive to bulk effects. Table 6-4. The performance characteristics of the DGSPR sensor consisting of a SU8 grating. Sensor TM1 TM2 TM3 TE1 TE2 CSF surf,thick CSF surf,thick CSF surf,thick CSF bulk CSF bulk DGSPR (2μm SU8) DGSPR (1μm SU8)

133 Discussion In this section, a novel approach to solve one of the major shortcomings of SPR biosensors which is sensitivity to interfering effects was investigated. The proposed design contains a dielectric grating coupled with a SPR sensor which enables exciting three resonance modes rather than one resonance mode in the SPR sensor. Dual-polarization spectroscopy was needed to be able to excite all three modes in this platform. Two modes were excited with TM polarized light and one mode with TE polarized light. The design parameters were optimized with a genetic algorithm so that each mode demonstrates a large FoM. Finally it was shown that decoupling adlayer properties from the bulk fluid index variations is possible by knowing the sensitivity of each mode to the interfering effects and measuring the change in resonance angles.

134 Dual Wavelength Spectroscopy Using a Metallic Grating Coupled to a Surface Plasmon Resonance Sensor In this section a novel approach for SPR bio-sensing is proposed which is based on a metallicgrating based SPR sensor (MGSPR). This approach benefits from one of the unique features of the grating, which is splitting the incoming light into several diffraction orders, for SPR sensing in Kretschmann configuration. The rigorous coupled-wave analysis (RCWA) method is applied to study the effect of the grating on the performance of the sensor. A metallic grating is optimized so that three SPR modes at two different wavelengths are excited. Thus, the adlayer properties (adlayer thickness d a and adlayer index n a ) can be determined and differentiated from the bulk index (n b ) variations by dual wavelength measurements. Finally, the optimization results along with a discussion on a method to decouple the interfering effects are presented Metallic-grating based SPR Sensor In order to improve the performance of the SPR based biosensors, metallic gratings have been widely used [131]. The main use of the metallic grating in prism based SPR sensors is to enhance the sensitivity by creating local field enhancement and increasing the sensing surface area [132]. Assuming that the dispersion properties of the SP wave is not disturbed by the grating, the momentum conservation between the free space light and the SP wave in MGSPR sensor can be expressed by Eq In Eq. 6-1, the incident light wavevector is perturbed by the grating which generates series of diffracted beams directed away from the corrugated surface [see Fig. 6-8(a)]. The diffraction

135 114 orders (m) which satisfy Eq. 6-1 can excite SP wave and create a resonance dip in the reflectance spectrum. In this work, the possibility of SPR spectroscopy at a fixed wavelength with more than one diffraction order is investigated. Figure 6-8(b) shows a two dimensional cross section view of the MGSPR sensor along with the grating parameters. Fig (a) Three dimensional (3D) schematic of a MGSPR sensor. (b) 2D cross section of the MGSPR sensor. Gold nanowires of a rectangular profile are assumed to be infinite in length with periodicity of Λ, thickness of h and a filling factor of w/λ Numerical Model To simultaneously excite multiple SPR waves in a prism-grating based geometry, several parameters should be adjusted namely: the grating periodicity, grating height, filling factor, gold thickness, and incident wavelength. Any variations in the aforementioned parameters can alter the propagation constant of the mode. The optimization of these parameters should ensure that each of the excited modes has a high sensitivity to the surface or bulk effects. Therefore, defining a proper figure of merit (FoM) is necessary to evaluate the characteristics of each mode. In this work, the FoM is defined as the product of CSF corresponding to each mode as expressed by Eq. 6-8.

136 115 TMi, 1,2,... surf, thick Eq. 6-8 FoM CSF i i where is the CSF corresponding to the adlayer thickness variations for the TM i mode. To optimize the proposed biosensor, a GA is implemented Results and Discussion Due to the computational intensity of the problem, distributed-memory parallelization is employed for GA optimization (see Appendix C). The computations herein were performed on the General Purpose Cluster at the Scinet supercomputer in the University of Toronto. The MGSPR is first optimized to excite two modes each of which is highly sensitive to the surface effects. To have a qualitative comparison, single interface SPR sensor is also optimized with a GA. In the optimization process, BK7 glass is considered as the substrate and prism material in both structures and 2 nm Ti adhesion layer on the glass substrate is included in the simulation. Table 6-5 summarizes the optimized dimensions and wavelengths for both structures. Table 6-5. The optimized dimensions of SPR and MGSPR sensors for affinity sensing Wavelength Gold Grating Fill factor Periodicity (Λ) (nm) thickness (t) thickness(h) (w/ Λ) SPR nm MGSPR nm 241 nm 325 nm 0.7

137 116 The reflected wave diffraction efficiency of the optimized MGSPR at a wavelength of 845 nm is calculated using an in-house code written based on RCWA and the result is depicted in Fig. 6-9(a). There are two resonances in the spectrum each corresponding to two counterpropagating plasmonic modes with different propagation constants. These modes are highly sensitive to the surface effects (n a, d a ) Diffraction efficiency (a) d a =0nm d a =2nm d a =4nm Diffraction efficiency (b) n b =1.333 n b =1.335 n b = TM1 TM Angle (degree) 0.4 TM Angle (degree) Fig Diffraction efficiency of the reflected light with TM polarization for the optimized MGSPR sensor at the incident wavelength of (a) 845 nm (b) 970 nm. The first mode (TM1) with smaller angle of resonance is excited by the first negative diffraction order (m=-1) and the second mode (TM2) with larger resonance angle is excited by the zeroorder diffraction (m=0). Both modes are excited efficiently which can be realized by the small values of the diffraction efficiency at the resonance angles in Fig. 6-9(a). As mentioned earlier, the main goal of this work is to determine three interfering effects by three-mode spectroscopy. Two of the modes (TM1 and TM2) are excited at 845 nm wavelength and the third mode (TM3) can be excited by using a second wavelength (970 nm). This mode is optimized for high sensitivity to the bulk refractive index variations [Fig. 6-9(b)] and can be excited by the zero-order diffraction of the MGSPR sensor.

138 117 To further investigate the properties of the excited modes, dispersion curves can be used. An indirect method to extract the dispersion relation is to calculate the reflection or transmission spectrum for a range of incident angles and wavelengths. The RCWA is used to calculate the dispersion curve of the MGSPR sensor by finding the reflected light diffraction efficiency for a range of wavelengths as is shown in Fig. 6-10(a). Fig. 6-10(b) shows a dispersion curve which is analytically derived from Eq The two dominants reflectance minima in Fig. 6-10(a) originate from the same diffraction orders as shown in Fig. 6-10(b). The small mismatch between these two dispersion curves is hidden in the fact that the dispersion equation is derived based on the assumption that the single interface SP wave propagation constant is not disturbed by the presence of the grating. Fig (a) Dispersion relation of the MGSPR sensor calculating using RCWA. (b) Analytically calculated dispersion relation of the SP wave in the optimized MGSPR sensor using Eq The performance characteristics of each mode to the variations in surface and bulk parameters are shown in Table 6-6. In this table, the calculated values of the CSF with respect to the adlayer thickness (CSF thick ), adlayer index (CSF index ), and the bulk index (CSF bulk ) for both MGSPR and SPR sensors are summarized.