Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Enhancing the performance of silicon photonics biosensors Flueckiger, Jonas 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2017_february_flueckiger_jonas.pdf [ 25.59MB ]
Metadata
JSON: 24-1.0340888.json
JSON-LD: 24-1.0340888-ld.json
RDF/XML (Pretty): 24-1.0340888-rdf.xml
RDF/JSON: 24-1.0340888-rdf.json
Turtle: 24-1.0340888-turtle.txt
N-Triples: 24-1.0340888-rdf-ntriples.txt
Original Record: 24-1.0340888-source.json
Full Text
24-1.0340888-fulltext.txt
Citation
24-1.0340888.ris

Full Text

Enhancing the Performance of Silicon Photonic BiosensorsbyJonas FlueckigerB. Microengineering, Ecole Polytechnique de Lausanne, 2005M. Microengineering, Ecole Polytechnique de Lausanne, 2007a thesis submitted in partial fulfillmentof the requirements for the degree ofDoctor of Philosophyinthe faculty of graduate and postdoctoral studies(Electrical and Computer Engineering)The University of British Columbia(Vancouver)January 2017© Jonas Flueckiger, 2017AbstractSilicon photonic biosensors have the potential to transform medical diagnosticsand healthcare delivery. Hundreds of these nano-scale sensors can be integratedonto a single millimeter-sized silicon substrate. They are fabricated in establishedCMOS foundries leveraging similar economies-of-scale achieved by electronic in-tegrated circuits. This also enables their potential integration with electronic readout circuitry on a single chip. As near-infrared light propagates through nano-scale silicon wires, a portion of the light resides outside the waveguide interactingwith biomolecules on the waveguide’s surface. While silicon photonic biosensorshave demonstrated performances approaching today’s gold-standard diagnostic, theenzyme-linked immunosorbent assay (ELISA), improving their performance ex-pands the potential use for applications requiring higher sensitivities and detectionlimits.To this end, this thesis describes efforts to optimize established biosensor config-urations and develop novel structures with performance that exceeds commerciallyavailable silicon photonic biosensor platforms. This involves improving the bulkand surface sensitivity, detection limit, and quality factor of transverse electric (TE)and magnetic (TM) mode resonators in various waveguide topologies. Specifically,TM mode microring resonators, microdisk resonators, thin waveguide resonators,and the first of its kind sub-wavelength grating microring resonator with a 10Xsensitivity improvement over today’s commercially available ring resonators arepresented. Furthermore the use novel TE mode slot-waveguide and TM mode stripwaveguide Bragg gratings which facilitate higher sensitivities (8X) and lower de-tection limits for biosensing applications are described. Finally, suspended Bragggrating structures are investigated to further improve sensitivity.iiTo support the design and characterization efforts required to efficiently inves-tigate many different sensors, a testing platform and process design kit (PDK) wasdeveloped. The test platform automatically tests hundreds of devices and orches-trates complex, multi-hour assays. The PDK reduces first-time design risk andexpedites chip testing. Both have been open-sourced and are in use by more than adozen academic and commercial research groups in various countries.iiiPrefaceOver the course of my graduate program, I have co-authored a number of publica-tions. They are listed in the following paragraph. In addition to these publicationsI also have been a co-author in a number of other publications unrelated to mythesis work. These projects are not included in this document; A complete list ofpublications and can be found in Appendix C. Some of the text and figures fromall my publications have been reused in this dissertation with the permission of thepublisher.1. J. Flueckiger, S. Schmidt, V. Donzella, A. Sherwali, D.M. Ratner, L. Chros-towski, K.C. Cheung, ’Sub-wavelength grating for enhanced ring resonatorbiosensors’, Opt. Express, vol. 24, no. 14: OSA, pp. 15672-15686, 2016I am the main contributor of this publication. I designed, simulated and opti-mized the photonic components and ran the bio experiments. Shon Schmidtmeasured the sensitivity of the devices and wrote parts of the manuscript. Ihave conducted the data analysis and prepared the figures. The manuscripthas been reused in section 4.7. The idea of using sub wavelength gratingsin a ring configuration was conceived by me, Valentina Donzella, and LukasChrostowski. For this design iteration I developed a more advanced simula-tion methodology compared to what we had previously used to design SWGwaveguides.2. J. Flueckiger, S. Schmidt, Z. Chen, X. Wang, A. Liu, L. Chrostowski, D.M.Ratner, K.C. Cheung, ’TM polarized Silicon Photonic Bragg Gratings forBiosensing’, Manuscript submitted, 2016.I am the main contributor of this publication. I designed, simulated andivoptimized the photonic components and wrote themanuscript. Shon Schmidtran the bio assay and was involved in the experimental data gathering forthe evanescent field characterization. I have conducted the data analysis andprepared the figures. Themanuscript has been reused in chapter 5, specificallyin section 5.5. The original work on SOI Bragg gratings started with XuWang’s doctoral dissertation which focused on the design and fabrication ofvarious Bragg gratings for TE polarized light. Based on his work I simulatedand optimized the design for TM polarized light and with water as claddingmaterial. I not only applied his work to the field of biosensing but alsoimproved upon his simulation methodology.3. S. Schmidt, J. Flueckiger, W.X.Wu, A. Lingley, K. Bohringer, K.C. Cheung,L. Chrostowski, D.M.Ratner, ’Enhancing Silicon Photonic Bragg GratingPerformance for Biosensing Applications’, Manuscript submitted, 2016I designed, simulated and optimized the photonic components and wrote thedesign parts of the manuscript. Shon Schmidt did most of the experimentalwork and the writing. I have conducted some data analysis and preparedsome figures. The manuscript has been reused in chapter 5 section 5.6 and5.7 specifically. The demonstrated design improvements in this manuscriptare proposed by me following out of what is described in chapter 2.4. S. Schmidt, J. Flueckiger, Pakapreud Khumwan, Jing Shang, Alex Wende,Zak Khaleel, Hal Holmes, Lukas Chrostowski, Daniel Ratner, ’Erythrocyteand Serologic Phenotype Detection using Silicon Photonic Ring Resonators’,Manuscript submitted, 2016.I designed, simulated and optimized the photonic components and wrote thedesign parts of the manuscript. Shon Schmidt did all of the experimentalwork and the writing. I have conducted some data analysis and preparedsome figures. The experimental data related to device characterization hasbeen reused in chapter 4 in section 4.4 (the chemistries involved to detect theserological phenotype are not described in my work).5. J. Flueckiger, S. Schmidt, W.X. Wu, M. Reis, P. Kulik, L. Chrostowski, D.Ratner, ’Automated Probe Station and PDK for Silicon Photonic Biosensors’,vManuscript submitted, 2016Shon Schmidt and I have equal contribution of the work presented in themanuscript. We lead the whole development cycle from conceptual idea todesign, building and writing of the software. Shon and I wrote the majorityof the software and I did all of the CAD design. Over the years we had anumber of students supporting the effort. The material has been reused inchapter 3 and appendix B.6. Z. Chen, J. Flueckiger, X. Wang, F. Zhang, H. Yun, Z. Lu, M. Caverley,Y. Wang, N. A. F. Jaeger, L. Chrostowski, ’Spiral Bragg grating waveguidesfor TM mode silicon photonics’, Optics Express, vol. 23, issue 19, pp.25295-25307, 09/2015.I did some of the initial simulations, but was mostly guiding and mentoringZhitian Chen during the design stage of the spiral Bragg grating. I was alsoadvising Zhitian Chen on the experimental work and was involved in thewriting of his manuscript. While this example of a Bragg grating is notdirectly related to biosensing, some data for the characterization of Bragggratings (namely the coupling coefficient) is reused in chapter 5 in section5.5.3 as the test structures were developed by me.7. V. Donzella, A. Sherwali, J. Flueckiger, S.M. Grist, S. Talebi-Fard, L.Chrostowski, ’Design and fabrication of SOI micro-ring resonators based onsub-wavelength grating waveguides’, Opt. Express, vol. 23, no. 4: OSA, pp.4791–4803, 02/2015.I did some of the initial simulations, provided feedback at the design stage,and did parts of the layout. I also performed the experimental work. I hadminor contributions in the writing of the manuscript. The idea of using subwavelength gratings in a ring configuration was conceived by me, ValentinaDonzella, and Lukas Chrostowski.8. J. Flueckiger, S. Schmidt, W. Wu, S.M. Grist, S. Talebi-Fard, V. Donzella,P. Khumwan, E.R. Thompson, Q. Wang, P. Kulik, X. Wang, A. Sherwali, J.Kirk, K.C. Cheung, L. Chrostowski, D. Ratner. ’Improving the performanceof silicon photonic rings, disks, and Bragg gratings for use in label-freevibiosensing.’ SPIE NanoScience Engineering VII. International Society forOptics and Photonics, 2014.Shon Schmidt and I contributed equally to the writing of this manuscript.Parts of the manuscript has been used in chapter 4 and chapter 5. Themanuscript summarizes all of the ongoing biosensing related work betweenthe University of Washington (Ratner Lab) and the University of BritishColumbia (Chrostowski Lab and Cheung Lab). I have been driving thecoordination between the labs.9. S. Talebi Fard, V. Donzella, S. Schmidt, J. Flueckiger, S.M. Grist, P. TalebiFard, Y. Wu, R. Bojko, E. Kwok, N.A.F. Jaeger, D.M Ratner, L. Chrostowski,’Performance of ultra-thin SOI-based resonators for sensing applications’,Opt. Express, vol. 22, no. 12: OSA, pp. 14166–14179, 06/2014.I provided the simulation scripts and was involved in the experimental char-acterization of the devices. Some of the data has been used in chapter 4.As described in chapter 2 I identified the need to investigate more optimizedwaveguide geometries. It was Sahba Talebi Fard who implemented the actualdesign based on my initial work, as part of her PhD thesis.10. V. Donzella, A. Sherwali, J. Flueckiger, S. Talebi Fard, S.M. Grist, L. Chros-towski, ’Sub-wavelength grating components for integrated optics applica-tions on SOI chips’, Opt. Express, vol. 22, no. 17: OSA, pp. 21037–21050,08/2014.I did some of the initial simulations, provided feedback at the design stage,and did parts of the layout. I also performed the experimental work. I hadminor contributions in the writing of the manuscript. The idea of using subwavelength gratings in a ring configuration was conceived by me, ValentinaDonzella, and Lukas Chrostowski.11. S. Schmidt, S. Grist, J. Flueckiger, V. Donzella, W. Shi, S. T. Fard, J. T.Kirk, D. Ratner, K. Cheung, and L. Chrostowski, ’Silicon photonic micro-disk resonators for label-free biosensing,’ Optics Express, 2013.I did the design of the optical components and performed the experimentalcharacterization of the sensors as well as the bio assay experiment. I alsoviicontributed to the writing of the manuscript. The data has been reusedin section 4.6. The idea to implement disk resonators for biosensing wasconceived by me, Samantha Grist, and Lukas Chrostowski. Samantha Gristand I contributed equally in the design phase.12. X. Wang, J. Flueckiger, S. Schmidt , S. Grist, S. T. Fard, J. T. Kirk, M.Doerfler, K. Cheung, et al., ’A silicon photonic biosensor using phase-shiftedBragg gratings in slot waveguide,’ Journal Biophotonics, vol. 6, issue 10,pp. 821–828, 04/2013.I performed the experimental work and had some design input as well asminor contributions to the writing. The experimental data has been reusedin section 5.4. The idea and design was conceived by Xu Wang as describedin his doctoral dissertation which focused on the design and fabrication ofvarious Bragg gratings for TE polarized light. I was driving the applicationof these slot Bragg gratings in the biosensing space.13. X. Wang, S.M. Grist, J. Flueckiger, N. A. F. Jaeger, L. Chrostowski, ’Siliconphotonic slot waveguide Bragg gratings and resonators’, Optics Express, vol.21, issue 16, pp. 19029-19039, 08/2013.I performed the experimental work and had some design input as well asminor contributions to the writing. The experimental data has been reusedin section 5.4. The idea and design was conceived by Xu Wang as describedin his doctoral dissertation which focused on the design and fabrication ofvarious Bragg gratings for TE polarized light. I was driving the applicationof these slot Bragg gratings in the biosensing space.14. S. T. Fard, V. Donzella, S. Grist, S. Schmidt, J. Flueckiger, X. Wang, W. Shi,et al., ’Label-free silicon photonic biosensors for use in clinical diagnostics,’presented at the SPIE Lase, San Francisco, CA, 2013.I contributed to thewriting of themanuscript. Themanuscript summarizes allof the ongoing biosensing related work between the University ofWashington(Ratner Lab) and the University of British Columbia (Chrostowski Lab andCheung Lab) at that time. I have been driving the coordination between theviiilabs. Some of the key concepts from this paper are reused in chapter 4 andchapter 5.15. S.Schmidt, X. Wang, J. Flueckiger, S.M. Grist, J. Kirk, K.C. Cheung, L.Chrostowski, D. Ratner. Silicon Photonic Phase-shifted Bragg Gratings ForUse In Label-free Biosensing. Poster session presented at: 2013 BMESAnnual Meeting, Biomedical Engineering Society, 2013 September 25-28;Seattle, WA.I contributed to the figures and artwork for the poster. I conducted themeasurement for the data presented. Some of the key concepts from thispaper are reused in chapter 5.16. P. Kulik, W. Wu, Q. Wang, S. Schmidt, J. Flueckiger, L. Chrostowski, D.Ratner. Open-Sourced Optical Test Setup for Rapid and Affordable SiliconPhotonic Biosensor Development. Poster session presented at: 2013 BMESAnnual Meeting, Biomedical Engineering Society, 2013 September 25-28;Seattle, WA.I contributed to the figures and artwork of the poster. Shon Schmidt and Ihave equally contributed in building the test setup presented at the conference.The test setup is described in chapter 3.17. L. Chrostowski, S. Grist, J. Flueckiger, W. Shi, X. Wang, E. Ouellet, H.Yun, M. Webb, B Nie, Z. Liang, K. Chung, S. Schmidt, D. Ratner, N. Jaeger,’Silicon photonic resonator sensors and devices,’ in SPIE LASE, Jan 2012.I provided experimental data and figures for the manuscript and also wasinvolved in the editing process. Some of the key concepts from this paper arereused in chapter 2, and chapter 6.18. J. Flueckiger, S.M. Grist, G. Bisra, L. Chrostowski, K.C. Cheung, ’Cascadedsilicon-on-insulator microring resonators for the detection of biomoleculesin PDMS microfluidic channels’, Proc. SPIE, Vol 7929; SPIE Microfluidics,BioMEMS, and Medical Microsystems IX, 01/2011.I performed all the experimental work with Samantha Grist, I did the dataanalysis and most of the writing of the manuscript. The results are reusedixin section 4.3. The idea to used cascaded rings for multiplexed sensing wasconceived by me.19. J. Flueckiger, S. M. Grist, E. Ouellet, L. Chrostowski, K.C. Cheung, ’Label-Free Biosensing Using Cascaded Silicon-on-Insulator Micro-Racetrack Res-onators Integrated With PDMS Microfluidic Channels’, The 15th Interna-tional Conference on Miniaturized Systems for Chemistry and Life Sciences,Seattle, Washington, USA, 10/2011.I performed all the experimental work with Samantha Grist, did the dataanalysis and most of the writing of the manuscript. The results are reusedin section 4.3. The idea to used cascaded rings for multiplexed sensing wasconceived by me.20. S.M. Grist, J. Flueckiger, J. Yu, W. Shi, K.C. Cheung, L. Chrostowski,’Silicon-on-insulator resonators integratedwith PDMSmicrofluidic channelsfor applications in biosensing’, Pacific Center for Advanced Materials andMicrostructures (PCAMM) 15th Annual Meeting, 2010.I performed all the experimental work with Samantha Grist, and providedsome figures.21. J. Flueckiger, S. Schmidt, W.X. Wu, M. Reis, P. Kulik. (2015). SiliconPhotonic Test Bench [Computer software].https://github.com/shonschmidt/SiPhoTestBench.Shon Schmidt and I equally contributed to the open source software forautomated measurements. The software is described in appendix B andchapter 3.22. J. Flueckiger, S. Schmidt, W.X. Wu, M. Reis, P. Kulik. (2015). SiliconPhotonic Test Bench for Biosensing [Computer software].https://github.com/shonschmidt/SiPhoTestBenchBio.Shon Schmidt and I equally contributed to the open source software forautomated biosensing. The software is described in appendixB and chapter 3.23. J. Flueckiger, S. Schmidt, W.X. Wu, (2015). Silicon Photonic Analysis Tool[Computer software]. https://github.com/shonschmidt/SiPhoAnalysisTool.xShon Schmidt and I equally contributed to the open source analysis software.The software is described in appendix B and chapter 3.xiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviiiGlossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Biosensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2.1 Origins . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2.2 Optical sensors . . . . . . . . . . . . . . . . . . . . . . . 41.2.3 Label versus label-free biosensing . . . . . . . . . . . . . 41.3 Silicon photonics . . . . . . . . . . . . . . . . . . . . . . . . . . 61.3.1 Operating principle . . . . . . . . . . . . . . . . . . . . . 61.3.2 Performance metrics . . . . . . . . . . . . . . . . . . . . 61.3.3 Comparing silicon photonic biosensors . . . . . . . . . . 91.4 Research objectives . . . . . . . . . . . . . . . . . . . . . . . . . 131.5 Document organization . . . . . . . . . . . . . . . . . . . . . . . 13xii2 Design Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . 152.1 Principles of optical waveguides . . . . . . . . . . . . . . . . . . 162.1.1 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . 202.1.2 Evanescent field . . . . . . . . . . . . . . . . . . . . . . 252.1.3 Susceptability or waveguide sensitivity . . . . . . . . . . 262.1.4 Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.2 Figures of merit . . . . . . . . . . . . . . . . . . . . . . . . . . . 312.2.1 Resonator sensitivity . . . . . . . . . . . . . . . . . . . . 312.2.2 Limit of detection . . . . . . . . . . . . . . . . . . . . . . 372.2.3 Improving the intrinsic limit of detection . . . . . . . . . 382.2.4 Operating wavelength . . . . . . . . . . . . . . . . . . . 402.2.5 Temperature sensitivity . . . . . . . . . . . . . . . . . . . 412.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . 433.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433.1.1 Requirements . . . . . . . . . . . . . . . . . . . . . . . . 443.2 Development methodology . . . . . . . . . . . . . . . . . . . . . 463.2.1 Device design . . . . . . . . . . . . . . . . . . . . . . . . 483.2.2 Device fabrication . . . . . . . . . . . . . . . . . . . . . 493.2.3 Basic circuit layout . . . . . . . . . . . . . . . . . . . . . 503.2.4 Device testing and performance characterization . . . . . 523.2.5 PDK and chip framework . . . . . . . . . . . . . . . . . . 553.2.6 Generating a device list . . . . . . . . . . . . . . . . . . . 563.3 Hardware assemblies . . . . . . . . . . . . . . . . . . . . . . . . 573.3.1 Chip stage assembly . . . . . . . . . . . . . . . . . . . . 583.3.2 Fiber array stage assembly . . . . . . . . . . . . . . . . . 583.3.3 Flow cell and channel gaskets . . . . . . . . . . . . . . . 593.3.4 Reagent sequencing stage assembly . . . . . . . . . . . . 593.3.5 External instruments . . . . . . . . . . . . . . . . . . . . 613.4 Fiber array alignment . . . . . . . . . . . . . . . . . . . . . . . . 613.4.1 Mapping grating couplers . . . . . . . . . . . . . . . . . 623.4.2 Fine align . . . . . . . . . . . . . . . . . . . . . . . . . . 63xiii3.5 Coordinate system . . . . . . . . . . . . . . . . . . . . . . . . . . 653.6 Software application . . . . . . . . . . . . . . . . . . . . . . . . 663.6.1 Automated testing and assay orchestration . . . . . . . . . 673.6.2 Analyzing data . . . . . . . . . . . . . . . . . . . . . . . 693.7 Platform characterization . . . . . . . . . . . . . . . . . . . . . . 703.7.1 Noise floor as function of sweep settings . . . . . . . . . . 713.7.2 Noise floor as function of flow rates . . . . . . . . . . . . 723.7.3 Stage stability and algorithm robustness . . . . . . . . . . 723.8 Summary and future work . . . . . . . . . . . . . . . . . . . . . 754 SOI Ring Resonator Biosensors . . . . . . . . . . . . . . . . . . . . . 774.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 774.2 SOI microring resonators - theory and design . . . . . . . . . . . 784.2.1 Coupling region . . . . . . . . . . . . . . . . . . . . . . 804.2.2 Spectral response . . . . . . . . . . . . . . . . . . . . . . 824.3 Cascaded ring resonators . . . . . . . . . . . . . . . . . . . . . . 844.3.1 Methods and materials . . . . . . . . . . . . . . . . . . . 854.3.2 Spectal response . . . . . . . . . . . . . . . . . . . . . . 874.3.3 Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . 884.3.4 Biological assay . . . . . . . . . . . . . . . . . . . . . . 894.4 TM mode ring resonators . . . . . . . . . . . . . . . . . . . . . . 904.4.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 904.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 914.5 Thin TE mode ring resonator . . . . . . . . . . . . . . . . . . . . 934.5.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 934.5.2 Characterization . . . . . . . . . . . . . . . . . . . . . . 954.5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 954.6 Micro-disk resonator biosensors . . . . . . . . . . . . . . . . . . 964.6.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 974.6.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 984.7 Subwavelength grating ring resonators . . . . . . . . . . . . . . . 1014.7.1 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . 1034.7.2 Experimental approach . . . . . . . . . . . . . . . . . . . 109xiv4.7.3 Results and discussion . . . . . . . . . . . . . . . . . . . 1114.7.4 Bulk RI sensing . . . . . . . . . . . . . . . . . . . . . . . 1124.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1165 Bragg Grating Resonators . . . . . . . . . . . . . . . . . . . . . . . 1195.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1195.2 Theory and design . . . . . . . . . . . . . . . . . . . . . . . . . 1205.3 Uniform strip-waveguide Bragg grating with phase shift for TEpolarized light . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1235.3.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 1235.3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 1245.4 Uniform slot-waveguide Bragg gratings with phase shift for TEpolarized light . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1255.4.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 1255.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 1265.5 Uniform strip-waveguide Bragg grating with phase shift for TMpolarized light . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1285.5.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 1285.5.2 Materials and methods . . . . . . . . . . . . . . . . . . . 1295.5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 1315.6 Suspended waveguide TE and TM mode Bragg gratings . . . . . . 1395.6.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 1395.6.2 Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . 1415.6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 1435.6.4 Biosensing demonstration . . . . . . . . . . . . . . . . . 1445.7 TM mode Bragg gratings at 1310 nm . . . . . . . . . . . . . . . . 1465.7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 1465.7.2 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 1465.7.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 1475.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1486 Conclusion and Future Directions . . . . . . . . . . . . . . . . . . . 1506.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150xv6.2 Sensor performance summary . . . . . . . . . . . . . . . . . . . 1516.2.1 Sensors that improve sensitivity . . . . . . . . . . . . . . 1526.2.2 Sensors that improve the quality factor . . . . . . . . . . . 1566.2.3 Improving limits of detection . . . . . . . . . . . . . . . . 1576.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1616.4 Future directions . . . . . . . . . . . . . . . . . . . . . . . . . . 164Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166A Principles of optical waveguides . . . . . . . . . . . . . . . . . . . . 194A.1 Ray optics approach . . . . . . . . . . . . . . . . . . . . . . . . . 195A.2 Electromagnetic wave approach . . . . . . . . . . . . . . . . . . 197A.2.1 Waveguide modes . . . . . . . . . . . . . . . . . . . . . 200B Testbench control software . . . . . . . . . . . . . . . . . . . . . . . 202B.1 Control software . . . . . . . . . . . . . . . . . . . . . . . . . . 202B.1.1 Program flow (sequencing) . . . . . . . . . . . . . . . . . 203B.1.2 Adding new instruments . . . . . . . . . . . . . . . . . . 210B.1.3 Adding new features to an existing test . . . . . . . . . . . 212B.1.4 Software testing . . . . . . . . . . . . . . . . . . . . . . . 212B.2 Analysis tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213B.2.1 Application overview . . . . . . . . . . . . . . . . . . . . 213B.2.2 Software architecture . . . . . . . . . . . . . . . . . . . . 217B.2.3 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . 220B.2.4 Access and future work . . . . . . . . . . . . . . . . . . . 221C Complete list of publications . . . . . . . . . . . . . . . . . . . . . . 222C.1 Other projects and collaborations . . . . . . . . . . . . . . . . . . 222xviList of TablesTable 1.1 Comparison of silicon photonic biosensors . . . . . . . . . . . 10Table 3.1 Power meter configuration for sweep test . . . . . . . . . . . . 71Table 3.2 Resonance peak jitter at different sweep speeds . . . . . . . . . 72Table 4.1 Refractive index of Glycerin/Water solutions at 20oC . . . . . 88Table 5.1 Electric field intensity in cladding and predicted sensitivity im-provement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140Table 5.2 Measured Bragg grating performance comparison . . . . . . . 143Table 6.1 Performance metrics of investigated biosensors . . . . . . . . . 153Table B.1 Platform control application class definition . . . . . . . . . . 204xviiList of FiguresFigure 1.1 Label vs. label-free biosening . . . . . . . . . . . . . . . . . 5Figure 1.2 Silicon photonic biosensor operation . . . . . . . . . . . . . . 7Figure 1.3 Detection limits for protein sensing . . . . . . . . . . . . . . 11Figure 2.1 Biosensor design considerations . . . . . . . . . . . . . . . . 16Figure 2.2 Schematic of Waveguide Cross Section . . . . . . . . . . . . 17Figure 2.3 Mode profile TE mode . . . . . . . . . . . . . . . . . . . . . 19Figure 2.4 Mode profile TM mode . . . . . . . . . . . . . . . . . . . . . 20Figure 2.5 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . 21Figure 2.6 neff vs waveguide width . . . . . . . . . . . . . . . . . . . . 22Figure 2.7 Mode cross over with SiO2 cladding . . . . . . . . . . . . . . 23Figure 2.8 Mode cross over with water cladding . . . . . . . . . . . . . 24Figure 2.9 Group index vs waveguide width . . . . . . . . . . . . . . . . 25Figure 2.10 Schematic overlap of modal field . . . . . . . . . . . . . . . . 26Figure 2.11 Waveguide susceptibility TE0 . . . . . . . . . . . . . . . . . 27Figure 2.12 Waveguide susceptibility TM0 . . . . . . . . . . . . . . . . . 28Figure 2.13 Bulk sensitivity - TE mode . . . . . . . . . . . . . . . . . . . 33Figure 2.14 Bulk sensitivity - TM mode . . . . . . . . . . . . . . . . . . 34Figure 2.15 Surface sensitivity vs layer thickness . . . . . . . . . . . . . . 35Figure 2.16 Surface sensitivity as function of waveguide geometry - TE mode 36Figure 2.17 Surface sensitivity as function of waveguide geometry - TE mode 37Figure 2.18 Refractive index of water at 25 oC . . . . . . . . . . . . . . . 40Figure 2.19 Absorption loss in water and ultimate Q . . . . . . . . . . . . 41xviiiFigure 3.1 Fiber-to-fiber testing . . . . . . . . . . . . . . . . . . . . . . 44Figure 3.2 Silicon photonic biosensor design methodology . . . . . . . . 47Figure 3.3 Ebeam fabrication process steps . . . . . . . . . . . . . . . . 49Figure 3.4 Basic circuit layout . . . . . . . . . . . . . . . . . . . . . . . 50Figure 3.5 Characterizing devices . . . . . . . . . . . . . . . . . . . . . 52Figure 3.6 Model System Bioassay Schematic . . . . . . . . . . . . . . . 53Figure 3.7 Silicon Photonic Biosensor PDK and chip framework . . . . . 56Figure 3.8 Device Labelling format . . . . . . . . . . . . . . . . . . . . 57Figure 3.9 Test setup assembly . . . . . . . . . . . . . . . . . . . . . . . 57Figure 3.10 Chip stage assembly . . . . . . . . . . . . . . . . . . . . . . 58Figure 3.11 Fiber stage assembly . . . . . . . . . . . . . . . . . . . . . . 59Figure 3.12 Flow cell and fluidic gasket . . . . . . . . . . . . . . . . . . . 60Figure 3.13 Image of platform with fluidics . . . . . . . . . . . . . . . . . 60Figure 3.14 Degrees of Freedom of automated test setup . . . . . . . . . . 62Figure 3.15 Fiber Array Misalignment . . . . . . . . . . . . . . . . . . . 62Figure 3.16 Grating Coupler Mapping . . . . . . . . . . . . . . . . . . . 63Figure 3.17 Fine Align Methods . . . . . . . . . . . . . . . . . . . . . . . 64Figure 3.18 Grating Coupler Alignment Structure . . . . . . . . . . . . . 66Figure 3.19 Test bench user interface . . . . . . . . . . . . . . . . . . . . 67Figure 3.20 Coordinate System: Using the transformation matrix . . . . . 68Figure 3.21 Salt steps performed on TE and TM rings . . . . . . . . . . . 69Figure 3.22 Analysis tool UI . . . . . . . . . . . . . . . . . . . . . . . . 70Figure 3.23 RMS noise vs flow rate . . . . . . . . . . . . . . . . . . . . . 73Figure 3.24 Insertion loss as function of position of stage . . . . . . . . . 74Figure 3.25 Fine Align Algorithm . . . . . . . . . . . . . . . . . . . . . . 75Figure 4.1 Ring resonator transmission spectrum . . . . . . . . . . . . . 78Figure 4.2 Ring resonator schematic . . . . . . . . . . . . . . . . . . . . 79Figure 4.3 Mode profile of supermodes in directional coupler . . . . . . 81Figure 4.4 Transmission of ring resonator . . . . . . . . . . . . . . . . . 83Figure 4.5 SEM image of ring resonator and coupling region . . . . . . . 84Figure 4.6 Cascaded ring resonator schematic . . . . . . . . . . . . . . . 85Figure 4.7 PDMS Flow cell . . . . . . . . . . . . . . . . . . . . . . . . 86xixFigure 4.8 Transmisison spectrum of cascaded ring resonators . . . . . . 87Figure 4.9 Transmisison spectrum of cascaded ring resonators . . . . . . 89Figure 4.10 Bio Experiment on cascaded ring resonators . . . . . . . . . . 90Figure 4.11 TM ring resonator simulation results . . . . . . . . . . . . . . 91Figure 4.12 Experimental results for the TM mode ring resonators . . . . 92Figure 4.13 TE ring resonator simulation results . . . . . . . . . . . . . . 94Figure 4.14 Experimental results for the thin waveguide TE mode ring res-onators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96Figure 4.15 Disk resonator simulation results . . . . . . . . . . . . . . . . 97Figure 4.16 Experimental results for the 3 µm disk resonators . . . . . . . 99Figure 4.17 Experimental results for the 10 µm disk resonators . . . . . . 100Figure 4.18 Schematic of SWG waveguide . . . . . . . . . . . . . . . . . 102Figure 4.19 Electric field intensity distribution in SWG waveguide . . . . 104Figure 4.20 Effective index as function of wavelength . . . . . . . . . . . 105Figure 4.21 Effective index as function of wavelength . . . . . . . . . . . 106Figure 4.22 Simulated sensitivity . . . . . . . . . . . . . . . . . . . . . . 108Figure 4.23 SEM image of SWG ring resonator . . . . . . . . . . . . . . 110Figure 4.24 Transmission spectrum of swg ring . . . . . . . . . . . . . . 111Figure 4.25 Peak shift of an SWG ring resonator exposed to different solu-tions of NaCl . . . . . . . . . . . . . . . . . . . . . . . . . . 112Figure 4.26 SWG ring sensivity . . . . . . . . . . . . . . . . . . . . . . . 113Figure 4.27 Biosensing cartoon . . . . . . . . . . . . . . . . . . . . . . . 114Figure 4.28 SWG ring - model bio assay results . . . . . . . . . . . . . . 115Figure 5.1 Bragg Grating Design Parameters . . . . . . . . . . . . . . . 121Figure 5.2 Effective and group index vs waveguide width . . . . . . . . . 122Figure 5.3 Strip WG Bragg Grating Mode Profile . . . . . . . . . . . . . 123Figure 5.4 Experimental results for strip WG Bragg Gratings . . . . . . . 124Figure 5.5 Slot WG Bragg Grating Mode Profile . . . . . . . . . . . . . 126Figure 5.6 Experimental results for the slot WG Bragg gratings . . . . . 127Figure 5.7 SEM images of fabricated Bragg gratings . . . . . . . . . . . 129Figure 5.8 Electrostatic polymer bi-layer adsorption . . . . . . . . . . . 130Figure 5.9 Uniform grarting design . . . . . . . . . . . . . . . . . . . . 132xxFigure 5.10 Grating Design: Kappa vs corrugation . . . . . . . . . . . . . 133Figure 5.11 Quality factor vs grating length . . . . . . . . . . . . . . . . . 134Figure 5.12 Characterization results . . . . . . . . . . . . . . . . . . . . . 135Figure 5.13 Surface sensitivity . . . . . . . . . . . . . . . . . . . . . . . 136Figure 5.14 Foundry Bragg Grating Bio Assay . . . . . . . . . . . . . . . 138Figure 5.15 TE and TM mode profiles . . . . . . . . . . . . . . . . . . . 139Figure 5.16 Suspended waveguide stress under flow . . . . . . . . . . . . 141Figure 5.17 SEM image of suspended WG Bragg grating . . . . . . . . . 142Figure 5.18 Assay comparison . . . . . . . . . . . . . . . . . . . . . . . 144Figure 5.19 Absorption losss for 1310 nm and 1550 nm . . . . . . . . . . 147Figure 5.20 Experimental results for 1310 nmwavelength TMBragg Gratings148Figure 6.1 Detection limit as function of intrinsic loss . . . . . . . . . . 159Figure 6.2 Comparing Q and S of designed sensors . . . . . . . . . . . . 160Figure A.1 Optical Waveguides . . . . . . . . . . . . . . . . . . . . . . . 195Figure A.2 Geometrical Ray Optics Approach . . . . . . . . . . . . . . . 196Figure A.3 Slab waveguide schematic . . . . . . . . . . . . . . . . . . . 197Figure A.4 Schematic of Waveguide Cross Section . . . . . . . . . . . . 201Figure B.1 Platform Control Application MVC . . . . . . . . . . . . . . 203Figure B.2 Platform Control Application Flow . . . . . . . . . . . . . . . 205Figure B.3 Chip alignment process overview . . . . . . . . . . . . . . . 206Figure B.4 Platform Control Application Selecting Devices Panel . . . . 207Figure B.5 Platform Control Application Assay Panel . . . . . . . . . . . 209Figure B.6 Initializing and connecting instruments . . . . . . . . . . . . 211Figure B.7 Laser control UI . . . . . . . . . . . . . . . . . . . . . . . . 211Figure B.8 Analysis Tool Application Overview . . . . . . . . . . . . . . 214Figure B.9 Analysis tool scan panel . . . . . . . . . . . . . . . . . . . . 215Figure B.10 Analysis Tool Peak Panel . . . . . . . . . . . . . . . . . . . . 216Figure B.11 Analysis Tool Peak Fitting Popup . . . . . . . . . . . . . . . 217Figure B.12 Analysis Tool Peak Tracking Plot . . . . . . . . . . . . . . . 218Figure B.13 Analysis Tool Software Architecture . . . . . . . . . . . . . . 219xxiGlossarybsa Bovin Serum Albuminelisa Enzyme-linked immunosorbent assayer Extinction ratiofdtd Finite-difference time domainfsr Free spectral rangefwhm Full width at half maximumgui Graphical user interfaceilod Intrinsic limit of detectionivd In vitro diagnosticifa Immunofluorescent assaymems Microelectromechanical systemsmpw Multi project wafermrr Micro ring resonatormw Molecular weightnw Nanowiremzi Mach-Zehnder interferometerxxiipbs Phosphate buffered salinepdms Polydimethylsiloxanepoc Point-of-carepoct Point-of-care testptfe Polytetrafluoroethyleneri Refractive indexriu Refractive index unitsa Streptavidinslod System limit of detectionsoi Silicon-on-Insulatorspr Surface plasmon resonanceswg Sub-wavelength gratingte Transverse electrictm Transverse magneticui User interfacexxiiiAcknowledgmentsFirst and foremost I am immensely grateful to my supervisor Professor Karen C.Cheung for her support, advice, encouragement and patience. I sincerely appreciatethe opportunity of moving to Vancouver, Canada, but even more so the experienceand freedom of being involved in so many different areas of research includingphotonics, microfluidics, microfabrication, and biochemistry. I am also grateful toProfessor Lukas Chrostowski who not only introducedme to the world of photonics,but also challenged and pushed me to ask better questions. I am also grateful toProfessor Boris Stoeber for serving on my departmental committee and ProfessorJohn Madden for serving as the chair of my departmental defense.I would also like to express my gratitude to the funding agencies who were atone point or another funding my PhD work: The BCIC Innovation Council, theSiEPIC program funded through the CREATE program of the Natural Science andEngineering Research Council of Canada, UBC, the UBC Department of ElectricalandComputer Engineering. I am also grateful to CMCMicrosystems for organizingmulti-project wafer runs for fabrication of photonic chips, for funding aids for in-house microfabrication processes, and for the various software tools and services Ihad access to for my work.I would also like to recognize all of my co-authors and collaborators. Atthe University of British Columbia: Dr. Nicolas Jaeger for not only graciouslylending equipment but also for insight and fruitful discussions. Dr. SamanthaGrist with whom I collaborated in the beginning of this project. Dr. Xu Wangand Professor Wei Shi for many useful discussions and learning opportunities, Dr.Linfen Wu, Daljeet Chahal, Dr. Benjamin Mustin, Josiah To, Dr. Kevin Han, ChrisFlory, Michael Chen, Oscar Wang, Han Yun, Dr. Valentina Donzella, Dr. SahbaxxivTalebi Fard, Dr. Robert Boeck, Mike Caverley, Kyle Murray, and Dr. Miguel AngelGuillen-Torres for being the best labmates. MitchWebb, Mauricio Reis, and GurpalBisra who were involved in my thesis research during their undergrad projects.At the University of Washington, Seattle, WA: Professor Dan Ratner for allow-ing me to come in his lab for most of the experimental work. A special thanksgoes to Shon Schmidt with whom I collaborated closely for the past years andwithout whom I would not be where I am at today. Special thanks also to Shon’sfamily, Cambria his wife and boys Jamison, Kaden, and Ashton, and Dr. Danieland Susan Schmidt, for being more than generous hosts during my stays in Seattle.Thanks to Vince Wu and Andrew Millspaugh for all the lines of code written forthe testplatform software. Thanks to Dr. Jim Kirk, Alissa Bleem, Emily Thomson,Pavel Kulik. At the WMF I’d like to express my gratitude to Dr. Rick Bojkoand Dr. Andrew Lingley for the fabrication services they provided and the fruitfuldiscussions.I would also like to thank the people at Lumerical for their support and theopportunity to do an internship. Specifically, Dr. Jackson Klein, Dr. James Pond,Amy Liu, Dr. Todd Kleckner, Bill de Vries and Dr. Xu Wang.Furthermore I would also like to thank all of the staff in the Electrical andComputer Engineering department, in particular Kristie Henriksen, David Chu-Chong, and Mark Finniss for their support. Dr Mario Beaudoin and Dr. AlinaKulpa for their training and expertise in the cleanroom.Finally, I’d like to thank all of my family and friends for their support, patienceand encouragement over the past years.xxvChapter 1Introduction1.1 MotivationDiagnostic tests are critical to healthcare, influencing the majority of all medicaldecisions [1]. In vitro diagnostics (IVDs) are tests performed on samples takenfrom bodily fluids, primarily blood and urine. IVDs have become ubiquitous inmodern medicine with clinical applications ranging from pathogen detection toblood typing. Most sophisticated IVD tests requiring high sensitivity and speci-ficity involve expensive bench-top instruments, consume expensive reagents, andare performed in a centralized hospital laboratory setting by highly trained oper-ators. Today’s diagnostic gold-standard, the enzyme-linked immunosorbent assay(ELISA), is highly sensitive (∼1 pMor 1-2 pg/mL), costs $300 - $650, is moderatelytime consuming (1-3 hours), requires secondary colorimetric or chemiluminescentamplification, and is burdened by complex logistics and information management[2–5].By contrast, point-of-care (POC) diagnostics are a subset of IVDs that canbe performed at the patient’s bedside, in the doctor’s office, or at home [6, 7].The nascent emergence of accountable care organizations (ACOs) is catalyzing agrowing interest in POC diagnostics to improve healthcare delivery [8]. In 2011,POC diagnostic tests (POCT) comprised the largest share of the IVD Market, withover the counter and hospital POC sales of over US$15 billion, or 30% of theIVD market [9]. POCTs are very cost effective as they impact ≈70% of all health1care decisions yet only account for ≈4% of total healthcare spending [1]. Themost common POCT include blood sugar monitoring, pregnancy tests, cholesterolmonitoring, and infectious disease detection, including HIV. Advances in scienceand technology continue to expand POCT into applications typically performedby conventional IVD in the hospital setting [10]. By improving IVD efficiency,POC benefits the patient, provider, and healthcare system as a whole, improvingthe efficiency of care delivery and providing more effective and responsive clinicalinterventions. However, not all POCT achieve performance similar to laboratory-based instrumentation as such tests are often optimized for ease of use and speedover performance. For most applications this lack of sensitivity has hindered theadoption of POCT [11].The motivation behind my research is to help translate laboratory-based di-agnostic capabilities into POC settings by enhancing the performance of siliconphotonic biosensors. Silicon photonics is an emerging chip-based technologythat shows promise for biosensing applications. As near-infrared light propagatesthrough nano-scale silicon wires on the chip, a portion of the light’s energy residesoutside the waveguide and is sensitive to biological interactions on the waveguide’ssurface. This functionality makes silicon photonics an ideal platform for biosens-ing. The silicon on insulator (SOI) platform is compatible with standard CMOSfabrication processes as well, facilitating manufacturing at the economies of scaleoffered by today’s foundries.This thesis describes my efforts to improve the performance of silicon pho-tonic biosensors by an order of magnitude over today’s commercial silicon basedplatforms [12] in an effort to expand their use across a broader range of clinicalapplications, label-free. For example, the diagnostic level of significance for theprostate-specific antigen (PSA) in human plasma is ≈4 ng/mL (120 pM). Kirk etal. recently reported that they could only achieve a detection limit of ≈10 ng/mLfor the label-free detection of streptavidin (spiked into undiluted human plasma)using DpC polymer-coated biosensors on a commercially-available silicon photonicplatform because of the relatively poor peformance of the sensor [13]. To achieveclinically relevant performance the overall system sensitivity needs to be enhanced.While the overall system performance is not only determined by the sensor alone,enhancing a sensor’s native sensitivity and detection limit will help expand the2platform’s capability for applications like the PSA test, label-free. To this end,my work focuses on enhancing the biosensing performance of transverse electric(TE) and transverse magnetic (TM) mode microring resonators, disk resonators,thin waveguide resonators, sub-wavelength grating resonators, as well as strip andslot Bragg gratings (for TE and TM polarization) were investigated for enhancedbiosensing performance. Metrics such as sensitivity, quality factor, and intrinsiclimit of detection are assessed and compared across these sensors for biosensingapplications. The remainder of this chapter provides a brief history on biosensors,background on silicon photonics performance metrics, and the specific aims of myresearch.1.2 BiosensorsThe field of biosensing has experienced phenomenal growth over the past 30 yearswith 5000 papers published on biosensing alone in 2011, an increase from just 100published manuscripts on the subject in 1985 [14]. Biosensors are a foundationalelement of diagnostics. For the scope of this dissertation, they are devices func-tionalized with molecules specific to a biological analyte (target) and can translateits capture into a detectable signal. Readout formats vary by sensor, but popularones include a visual color change (colormetric), a fluorescence, or an electronicsignal (voltage or current change).1.2.1 OriginsLeland C. Clark was the first to detail the concept of a biosensor in his 1962 seminalwork describing an electrode system for continuous glucose monitoring [15]. Heshowed how an electrode response could be sped up by employing an enzymaticreaction that converted glucose into gluconic acid (via glucose-oxidase), causinga detectable and quantifiable change in pH. While foundational for modern dayglucose meters, Clark speculated that electrochemical detection using enzymescould be used for a broad range of bioanalytical applications. We see this today asadvances in manufacturing and electro-chemistry have propelled enzymatic-baseddiagnostics into a broad range of distributed applications such as over-the-counterpregnancy, HIVand drug testswith amarket in excess ofUS$13 billion per year [14].31.2.2 Optical sensorsOptical biosensors use light (or electromagnetic waves) and affinity moleculesto detect target analytes. These sensors were pioneered by Ingemar Lundström’sgroup at Linköping University more than two decades after Clark’s publication [16]and were transformational in that they allowed real-time monitoring of molecularbinding events. Lundström’s sensor utilized surface plasmon resonance (SPR) tomeasure molecular absorption to a gold surface. A surface plasmon polariton isexcited when the frequency of the incident light (or photons) matches the frequencyof the surface electrons in the conduction band. A surface plasmon resonance canonly occur between an interface of negative and positive permittivity. Light at theresonant condition is absorbed and not reflected back to the detector. Adsorbingbiomolecules on the surface alter the resonant condition which is then detected andquantified.While the method by which a surface plasmon operates was initially suggestedby Kretschmann in 1971, biomolecular detection was not demonstrated until 1983[17] resulting in the formation of a Swedish company called Pharmacia Biosensorin 1986. The venture later became Biacore in 1990 which was acquired by GEin 2006 for US$390 million. While achieving impressive sensitivities of 10−5 to10−7 RIU (∼1 fg/mL for proteins; RIU - refractive index units) [18–20], SPR’sspecialized metal-plated chips and bench-top instrumentation limit its potentialbeyond research settings (the unit RIU stands for refractive index unit). Whileeffort has been invested into developing a field-deployable instrument based onTexas Instruments’ Spreeta sensing chips [21], SPR has not gained traction asa standard POC diagnostic platform and its market remains relatively modest atUS$100 million per annum.1.2.3 Label versus label-free biosensingAn important distinction in biosensing is whether or not an assay was performedlabel or label-free as shown in Fig. 1.1. In a labeled assay, quantitative signalenhancement can be achieved through the binding of an additional molecule to theimmobilized target, as shown in Fig. 1.1(a). This labeling process, also referredto as secondary amplification, can achieve sub-pg/mL sensitivities [4, 22] and4provides additional specificity confirmation. Yet, amplification comes with trade-offs: it limits multiplexing, does not allow for real-time kinetic monitoring of thetarget analyte, consumes reagents, and requires additional time and steps.Figure 1.1: (a) Schematic representation of a biological stack-up for labeledassays, or tests involving secondary amplification. Biomolecules, suchas antibodies with a florescent tag, bind to the immobilized target analyteand provide signal amplification when stimulated. (b) Schematic rep-resentation of a label-free biosensor. The sensor continuously monitorsthe surface and detects any additional bound mass from captured, targetanalytes.Unlike label-based systems, label-free sensors rely on an immobilized biorecep-tor molecule with affinities to the target analyte. The sensor transduces the bindingof the target into a measurable output signal, as shown in Fig. 1.1(b). Label-free systems facilitate real-time kinetic monitoring and the rapid quantification ofcomplex titers [23, 24]. They simplify required assay steps and reduce reagent con-sumption. Drawbacks include biological noise resulting from non-specific bindingand the lack of a confirmed biological specificity that a secondary label provides.51.3 Silicon photonicsSilicon photonics is a maturing optical technology that has gained traction as abiosensing platform [12]. It is a chip-based technology that guides near-infraredlight in small silicon wires and its origins can be traced back to the pioneer-ing works of Soref and Petermann in silicon waveguides during the mid-1980’s[25, 26]. Since then, significant investments by industry and the government hasaccelerated its use in data communications [27] while attracting tremendous atten-tion as promising platform technology for biosensing applications as well [24, 28].Its combination of excellent material properties (high contrast index), ultra-compactdevice footprint, established surface modification techniques [29], and fabricationcompatibility with existing complementary metal-oxide semiconductor (CMOS)foundries [30] uniquely position it as a potential diagnostic platform with ELISA-like performance [12].1.3.1 Operating principleSilicon photonic biosensors rely on near-infrared light confined in nanometer-scalesilicon wires, known as waveguides, to sense molecular binding events. A portionof the light’s electrical field travels outside the waveguide as an evanescent fieldand detects molecular binding events on the waveguide’s surface (Fig. 1.2(a) and1.2(b). Most often silicon photonic biosensors utilize optical resonant structuresthat are sensitive to changes in the refractive index in the evanescent field region[31]. When the target analyte binds to chemistries on the waveguide’s surface, theaccumulated number of molecules with a different refractive index change the localrefractive index (RI) and perturbs the evanescent field. This perturbation in theevanescent field shifts the sensor’s resonant wavelength, indicating the presence ofa bound analyte (Fig. 1.2b).1.3.2 Performance metricsThere has been a fragmented effort to define a set of performance metrics forrefractive index sensors [32–34]. The lack of universally accepted, comparativestandards complicates performance comparisons among different types of siliconphotonic biosensors. However, there are objective metrics that can be used such as:6Figure 1.2: Operating principle for a silicon photonic biosensor. A portionof the optical mode travels outside the waveguide as an evanescent fieldand is sensitive to biological binding events on the waveguide’s surface(orange region). (a) Mode profile for the fundamental TE mode and; (b)Mode profile for the fundamental TM mode. (c) Optical transmissionspectrum of a resonator; An analyte binding on a resonator structureincreases the effective index (ne f f ) resulting in a resonant wavelengthshift, ∆λ. (d) Sensogram of a biological assay where the shift of theresonant wavelength is recorded as function of time.sensitivity, detection limit, and the quality factor (Q) of the resonator.Sensitivity For a resonant structure the sensitivity is defined as the change inresonance wavelength (∆λres) per change in refractive index unit (∆n):S =∆λres∆n(1.1)where λres is the sensor’s resonant wavelength [nm], and n is the refractive index[RIU]. Two types of sensitivity are important in biosensing applications: (1) bulksensitivity which takes into account refractive index changes of the waveguide’sentire cladding and (2) surface sensitivity which assesses wavelength shifts per7refractive index changes within the first few hundred nanometers of the waveguide(adlayer thickness). While both have merit and provide insight into a sensor’sbiosensing performance, the specific application determines which is more impor-tant. When not specifically called out, most sensitivity metrics refer to the bulksensitivity.Detection limit The system’s limit of detection (sLoD) is defined as the minimumrefractive index change (δnclad,min), or smallest mass change, necessary to causea detectable change in the output signal (above 3 σ of the noise floor). ThesLoD depends on the readout instrumentation and the experimental setup, includingassociated noise sources (spectral and amplitude noise of the light source), electricalreadout circuitry noise, and thermal stability. Noise can be reduced digitally byemploying statistical signal processing algorithms. Because all of these factorscan impact the sLoD, a truly objective comparison of sensors characterized usingdifferent assays and experimental systems can be challenging. In order to comparesensors independent of the instrumentation used, we introduced the intrinsic limit ofdetection (iLoD) [32], which is dependent only on intrinsic resonator characteristics:iLoD =λresQ · S (1.2)where λres is the sensor’s resonant wavelength (nm), Q is the quality factor ofthe resonator, and S is the sensitivity (nm/RIU). The iLoD (with units of RIU)can be understood as the minimum index change required to shift the resonancewavelength by one line width (∆λ3dB).Quality factor The quality factor is a measure of photon lifetime in the resonatorand represents the number of oscillations before the photon’s energy has decayed to37% (1/e). Q is a dimensionless number that incorporates the sensor’s total loss interms of the material’s index of refraction [32] and is approximated experimentallyby dividing the resonant peak’s wavelength (λres) by its full width at half itsmaximum (FWHM). Higher quality factors reduce the spectral noise of the sensor[34] and improve the sensor’s iLoD. Algebraically, the quality factor is:8Q = ωresE∂E/∂t=2pi · ng ·4.34λres ·α[dB/m] ≈λres∆λ3dB(1.3)where α is the losses in the resonator (in dB/m), ωres = 2pi fres is the resonancefrequency, E is the mode’s electrical field distribution, ng is the group index, andδλ3dB is the 3 dB bandwidth of the resonance peak.While silicon photonic biosensors have achieved impressive biosensing results,mainstream silicon photonic sensors still require label-based signal amplification fordetection of extremely low concentrations [35]. Achieving sensitivities at pg/mLlevels or lower with label-free silicon photonic sensors suitable for high-volumefabrication would greatly expand their potential as a label-free diagnostic device.1.3.3 Comparing silicon photonic biosensorsSilicon photonic sensors have achieved sensitivities close to clinical levels forlabel-free sensing in complex media (10 ng/mL)[13, 24]. Arlett et al. recentlyreviewedmany state-of-the-artmicroelectromechanical (MEMS) - based biosensorsand compared their performance to today’s gold standard (ELISA) [2]. Figure 1.3highlights a portion of their comparative analysis including SPR, silicon nanowires,and silicon photonics.Silicon photonic microring resonators can match and even exceed ELISA de-tection limits (sub-pm/mL) by employing a secondary amplification step while stilloffering advantages over their mechanical and electrical biosensor counterparts[2, 36, 37]. Also, silicon photonic devices are not susceptible to mechanical forceunder normal microfluidic flow, electromagnetic interference, nor the electricalconduction of the analyte media.Although single molecule detection has been reported using silicon photonictoroids [38] and anti-resonant reflecting optical waveguides [39], these sensors areunsuitable for high-volume manufacturing due to fabrication complexities. Table1.1 lists the performance of recently reported silicon based photonic biosensors.Section 1.3.1 mentions the predominant use of resonant structures, or resonantcavities, but interferometric biosensors are of equal importance. Furthermore,we have discussed silicon based biosensors, which has been the most employedmaterial, but other materials are viable alternatives, including silicon nitrid [51],9Table 1.1: Comparison of silicon photonic biosensorsType Mode Material S [nm/RIU] Q DL Assay RefMultiplexed ring TE Si 54 >40k 1.6 - 39 ng/mL Ebola, Dengue [28]Ring TE Si 54 43k 1.5 pg/mm2 Biotin-Streptavidin [12, 40]Ring TE Si 54 >40k PSA, AFP, CEA,TNF-α, IL-8 [41]Porous ring TE Si 380 10,000 4 pm/nM DNA [42]Slot ring TE SiN 141 1300 50 ng/mL PSA [43]Slot ring TE Si 298 4.2 ·10−5 RIU Biotin-Avidin [44]Ring RE Si 20k 10 ng/mL Biotin-Avidin [45]Ring TE Si 70 20,000 10 ng/ml Biotin-Streptavadin [46]Slot Ring TE Si3N4 2465 ·10−6 RIU0.9 pg/mm2Glutaraldehyde-antiBSA [47]Ring TE Si 24,000 33 pM and 1.4 nM IgE and Thrombin [48]Ring TM Si 12,000 10nM ssDNA [49]Ring TE Si3N4 110 BSA [50]Slot Ring TE Si3N4 212 28 pg/mm2 Anti-BSA [51]Slot microdisk TM SiNx 7,000 30 ng/mL Biotin-Streptavadin [52]Photonic crystal microcavity TE Si 70 5,300 50 fM Avidin/biotin [53]Crystal ring slot TE Si 160 11,500 8.75x10−5 RIU [54]Triple ring and disk Si on polymer 1,000 67,000 [55]10Figure 1.3: The left vertical axis represents the detection limit (moles/L)whilethe right vertical axis represents the equivalent g/mL for different kindsof MEMS biosensors. Note that both axes are logarithmic. The bottomaxis represents the analysis time. Devices on the right hand side offergreater sensitivities at expense of analysis time. The black dashed linerepresents the limits of the current state-of-the-art. The pink, circleddots represent the performance of silicon photonic micro ring resonators(MRR) compared to SPR, silicon nano-wire (NW), and the ELISA, orimmunofluorescent assay (IFA). The blue region in the center of theplot represents clinically relevant sensitivities. The pink downwardarrow represents our research objectives: to increase the performanceof silicon photonic biosensors to achieve clinically relevant sensitivitieswithout amplification and for short analysis time.polysilicon [56], silicon dioxide [56], and other more emerging materials likepolymers [57] and other new nanomaterials [58]. The advantage of silicon basesensors is the high index contrast of silicon to the cladding material. This highindex contrast allows for much smaller bending radii than other material systemsand therefore much smaller sensor footprints. This is important for arrayed sensingplatforms for multiple analyte detection.11Interferometric biosensorsThe operating principle of an interferometric sensors relies on the splitting andcombining of an incoming coherent light beam. After the equal split the twobeam travel through different optical path. One branch is allowed to interact withthe analyte while the other is protected. In integrated photonics, one waveguideusually has a protective claddingwhile for the otherwaveguide the cladding has beenetched away, allowing the evanescent field to interact with the sample. Dependingon the difference of the optical path length an interference pattern is created afterrecombining the two beams. Rather then tracking resonant peaks, as the case forring resonators, one can simply measure the optical intensity at the output. Thechange of RI at the surface of the waveguide alters the optical path length of thebranch. The intensity at the output can be described as:I (∆ne f f ) ∝ cos(∆ne f f k0L) (1.4)where ∆ne f f is the change in effective index of the propagating mode (see sec-tion 2.1), k0 the amplitude of the wave vector, and L the length of the sensingregion. Equation 1.4 shows than the sensitivity of the interferometer can be in-creased with increasing length of the sensing arm. Also note that, because of thecosine dependence, the signal does not change at the maximum andminimum of thecosine function. The most common configuration is the Mach-Zehnder interferom-eter (MZI). Densmore et al. demonstrated a silicon baseMZI with a phase responseof 460(2pi) / RIU [59, 60]. Other types include the Young interferometer [61] andHarman interferometer [62]. More recently a BiModal waveguide interferometerhas been introduced by Lechuga et al. [63–66]. Rather than splitting the beam intotwo branches the light is coupled into two different modes supported by the guidingstructure. The modal overlap with the analyte is significantly different for the twomodes such that the phase change introduced by the refractive index change at thesurface of the waveguide is much higher for one mode compared to the other. Thereported detection limits of interferometric biosensors (∼ 1 · 10−7 RIU [63]) areimpressive and to a degree outperform the resonant cavity based sensors, but theycan only do so with long sensing arms (∼5-10 mm) and can therefore only supporta limited number of sensors per chip.121.4 Research objectivesWhile silicon photonic biosensors have achieved impressive performance, sec-ondary amplification is still required to achieve the sub-ng/mL sensitivities. Im-proving the sensor’s quality factor, response to changes in the cladding’s refractiveindex (or sensitivity), and iLoD is needed to expand potential sensing applicationsat clinically relevant levels.The objective of this thesis research is to create new, foundry-compatible sili-con photonic sensors with greater performance than today’s most competitive sil-icon photonic devices. This is accomplished by creating and characterizing novelbiosensors such as: thin waveguide resonant rings, TM mode Bragg gratings, sub-wavelength ring resonators, and biosensors for 1310 nm wavelengths. In addition,a custom test setup will be developed to efficiently characterize and demonstratethe biosensing performance of these novel devices.1.5 Document organizationThe remainder of this document is organized as follows:Chapter 2 provides some background on photonic biosensing principles andcomparative performance metrics. This chapter describes bulk and surface sensi-tivity and limit of detection and presents simulation results which have been usedfor the design of the sensing structures in the following chapter.Chapter 3 describes the development methodology and test platform hardwareand software used for this work, which enabled all the testing (from simple transmis-sion measurements to long multi-step biological assays). Without the test platformit would not have been possible to gather as much experimental data as I present inthe next few chapter. The process steps required for sensor design and test alongwith details on the design of the open-sourced test platform is presented. Thechapter concludes with a summary of the platform’s performance assessment forbiosensing environments.Chapter 4 describes various designs of ring resonator used for biosensing,including ring resonators for TE and TM polarized light, thin TE rings, and sub-wavelength grating ring resonators. The chapter describes the design and simulationapproach and compares the theoretical models with experimental results.13Chapter 5 describes various designs of Bragg grating resonators for biosensing,including Bragg gratings for TE and TM polarized light for wavelength of 1550 nmand 1310 nm, and suspended Bragg gratings. The chapter describes the designand simulation approach and compares the theoretical models with experimentalresults.Chapter 6 summarizes and compares the various sensor designs and the tradeoffs between design parameters is discussed. The silicon photonic design aspectdiscussed in this work is only one part in a much larger multi disciplinary search ofa diagnostic platform to address today’s health challenges. The chapter also brieflydiscusses the missing pieces and the integration work yet to be done.14Chapter 2Design ConsiderationsThere are two fundamental transduction mechanisms for optical resonator-basedconfigurations [67]: absorption sensing and refractive index sensing. Absorptionspectroscopy is based on a detectable change in resonant peak extinction ratiodue to the introduction of loss in the resonator via evanescent field interactionwith a chemical species near the surface of the device [32]. Compounds possesscharacteristic absorption spectra that can be used to differentiate between chemicaland molecular species. The transmission spectrum of a resonator sensor can beleveraged to perform absorption spectroscopy; each of the resonance peaks actsas probing wavelength. Therefore, the wavelength resolution of the spectroscopicmeasurement is limited by the free spectral range (FSR), or in other terms, by thespacing between longitudinal modes. The second mechanism, refractive index (RI)sensing, detects changes to the mode’s refractive index caused bymolecular bindingon the waveguide’s surface. The local refractive index change alters resonancecondition of the cavity. This work focuses exclusively on refractive index sensing.Figure 2.1 depicts key aspects that must be considered for optical biosensordesign. Ideally, the sensor should mimic the target molecule’s native environment.The sensor’s scale, in relation to target molecule or cell size, will determine thesensor’s response to a single binding event, or overall sensitivity. Bio-recognitionmolecules, or capture ligands, should be specific with high affinity and aviditytowards its binding conjugate. Surface chemistries that immobilize these cap-ture ligands should also resist non-specific binding interactions from off-target15Figure 2.1: Schematic representation of key aspects of an optical biosensor.The sensor should mimic the target analyte’s scale and native environ-ment while allowing facile stimulus and binding response monitoring.Surface chemistries should facilitate robust immobilization of bio recon-gintion molecules that have high affinity and avidity to the target analytewhile resiting adsorption from unwanted molecules (fouling).molecules increase ’biological noise’ in the system. This noise, or fouling, limitsthe overall sensitivity or detection limit. The system should have an efficient wayto stimulate the sensor and read its response to detect and quantify binding events.Most often, a signal baseline is established before introducing the sample so thatthe overall response to the target analyte can be determined. Finally, the sensorshould be compatible with affordable manufacturing processes and cost-effectivefunctionalization protocols.2.1 Principles of optical waveguidesAlthough the main focus of the work presented here is primarily on experimentalresults and the subsequent analysis, a theoretical understanding and derivation ofsimplified analytical solutions is key for optimizing and comparing sensor designs.See appendix A for a more in-depth derivation of waveguides optics.A dielectric structure that is able to confine and guide an electromagnetic waveis referred to by the term ’waveguide’ [68]. A simple practical example is the16optical fiber used in optical communications which has a material at the core with asufficiently high refractive index compared to the cladding material, to ’trap’ lightby total internal reflections. In integrated photonics a waveguide has typically arectangular cross section (see Fig. 2.2).Figure 2.2: Schematic of a waveguide cross section with lateral mode con-finementDue to translational invariance the modal fields can be written in separableform:Em(x, y, z) = Em(x, y)ei(ωt−βmz), (2.1a)Hm(x, y, z) =Hm(x, y)ei(ωt−βmz), (2.1b)where βm is called the propagation constant or eigenvalue of the mth mode (oftenalso called the mode number). Waveguide modes can be regarded as transverseresonances of the field in a waveguide. This is similar to normal modes of vibrationof a membrane [69].Depending on the dimension of the core (and the index contrast) and the wave-length, an optical waveguide can only support a discrete number of modes. Awaveguide with only one optical mode allowed is said to be single mode (for typi-cally small waveguide thicknesses) and multi mode for waveguides with more thanone mode.To find the mode profiles Em(x, y) andHm(x, y) numerical methods have to beused as analytical solutions in general do not exist. The mode solver used here (byLumerical Solutions Inc [70]) uses a finite difference technique in the frequencydomain based on an algorithm proposed by Zhu and Brown [71].Of course, the most suitable geometry to guide a particular mode in a waveguide17is determined by the application, the wavelength and polarization of the light, andthe material properties, i.e. the refractive indices of the core, ncore, the substratensub, and the cladding nclad. An important measure of how the light is guided in agiven structure is the confinement factor. It is defined as the fraction, Γ, of the totalpower residing in the core [69, 72, 73]:Γ =12!coreRe(E×H∗) · zˆ · dxdy12!totalRe(E×H∗) · zˆ · dxdy =12ncorec00!core|E |2 · dxdy12!totalRe(E×H∗) · zˆ · dxdy (2.2)The confinement factor is implicitly a function of the mode number (for a givenpolarization). For a mode far from the cutoff, Γ can be very close to 1. Moreimportantly, the confinement factor is largely dependent on the index contrastbetween waveguide core and cladding. For silicon on insulator waveguides with ahigh index contrast (ncore = nSi = 3.45, nclad = nH2O = 1.33, and nsub = nSiO2 =1.4) and a typical width w = 500 nm and thickness t = 220 nm, the confinementfactor is found to be 0.78. The confinement factor will also be important inconnection with sensitivity calculations in section 2.1.3.Because of the lateral and vertical mode confinement in such waveguides theresulting modes are not pure TM or TE polarized anymore. It is therefore useful tointroduce the measure of ’TE polarization fraction’ for a given mode profile:ξ =! |Ex |2 · dxdy! |Ex |2+ |Ey |2 · dxdy (2.3)For a pure TE polarized mode, the TE polarization fraction is ξ = 1 and for a pureTM mode ξ = 0. Modes which are predominantly TE-polarized (0.5 < ξ < 1) arereferred to as quasi-TE modes and similarly the term quasi-TM mode is used formodes with TE polarization fraction smaller than 0.5. However, for simplicity,throughout the remainder of this document, quasi-TE and quasi-TM modes will bereferred to as TE and TM modes respectively.The field distribution of a TE mode for a silicon waveguide with thicknesst = 220 nm and width w = 500 nm at a wavelength of λ = 1550 nm is shown inFig. 2.3. The guiding structure consists of a silicon core (with ncore = nSi = 3.45)on a silicon oxide substrate (with nsub = nSiO2 = 1.44) and water (with nclad =18Figure 2.3: TE mode profile for a waveguide with thickness t = 220 nm andwidth w = 500 nm; The electric field magnitude (|E|2) is plotted at x =0 nm and y = 110 nm for the vertical and horizontal line plot receptively.nH2O = 1.33) as cladding material (as most analytes of interest are in aqueoussolution). The confinement factor is ΓTE = 0.78. Similar to the equation for theconfinement factor (Eq. 2.2), one can also calculate the modal power propagatingin the substrate and the cladding respectively. About 10% is in the substrate and12% is in the cladding. Similar, Fig. 2.4 shows the modal field distribution for aTM mode for the same waveguide geometry. The confinement factor in this caseis only ΓTM = 0.42, which means that most of the modal power is propagationoutside the waveguide. About 36% is in the substrate and 32% in the cladding.Since the refractive index of the cladding is smaller than the substrate it gives riseto the asymmetric modal distribution as seen in the line profile in Fig. 2.4 (noticethe field magnitude at the discontinuities at the Si/SiO2 and Si/H2O interface).Throughout this document, the discussion will mainly be focused on the same19Figure 2.4: TM mode profile for a waveguide with thickness t = 220 nmand width w = 500 nm; The electric field magnitude (|E|2) is plottedat x = 0 nm and y = 110 nm for the vertical and horizontal line plotreceptively.material system as the example in Figs 2.3 and 2.4, namely a silicon waveguideon a silicon oxide substrate with varying cladding materials. However, a broaderdiscussion including other types of materials such as silicon nitride, silica onsilicon, polymers, or III-V materials like indium phosphide or gallium arsenidewould definitely be of interest as well. A discussion of such material platforms canbe found here [74–76], specifically in regards to sensing applications.2.1.1 DispersionFor propagating confined modes, there are only a discrete number of eigenmodessupported by a given waveguide geometry taking the form as desribed by equations2.1. The eigenvalue is also referred to as the propagation constant βm where the20subscript m refers to the mode number. In the context of mode propagation onealso defines the effective refractive index given by ne f f = β λ2pi .Figure 2.5: Dispersion of a waveguide with width of 500 nm and height of220 nm: a) effective index as function of wavelength and b) group indexas function of wavelength.The effective refractive index (or the propagation constant) is in general a func-tion of wavelength, ne f f (λ), and describes how the phase velocity vp, is dependenton frequency. Figure 2.5(a) shows the effective index as function of wavelengthfor both the TE and TM mode for a waveguide with width of 500 nm and heightof 220 nm. This means that monochromatic waves at different wavelengths willtravel through a medium (or waveguide) with different velocities. This frequencydependence causes a pulse to broaden and to be delayed. (A pulse can be consideredas the sum of monochromatic waves). The group velocity, vg, is the measure of thespeed at which the pulse envelope is traveling in a dispersive medium. The groupvelocity is given by:vg =1∂β∂ω, (2.4)where β is the propagation constant and ω is the angular frequency. Using thedefinition of the effective refractive index ne f f = β λ2pi the following expression forthe group velocity is derived:vg =cne f f − λ ∂nef f∂λ=cng(2.5)For this definition of ng, the effective group index, it is assumed that the refractive21index is not changing significantly over the wavelength range being considered.ng (λ) = ne f f (λ)− λ0∂ne f f∂λ(2.6)The dependence of the propagation constant β on frequency is also called thematerial dispersion.Waveguide dispersionIn the case of a waveguide, the propagation constant is not just a function ofwavelength because of the material properties, but also because of the geometry ofthe waveguide. It is helpful to introduce the term waveguide dispersion in contrastto material dispersion. Figure 2.6 shows the effective refractive index as functionFigure 2.6: ne f f vs waveguide width (for thickness of 220 nm) for differentmodesof waveguide width at constant height of 220 nm and at wavelength λ = 1.55 µm.Below a width of 250 nm no mode is supported. For the range of waveguide widths250 < w < 560nm the waveguide is said to be single mode, as it only supports oneTEmode and one TMmode. At w = 660 nm the second order TE mode (TE1) takesthe same effective refractive index as the fundamental TM mode. The mode is saidto be degenerate [69, 73]. For a waveguide width < 660 nm the effective refractiveindex of the TM mode, ne f f ,TM0, is larger than the effective index of second TE22mode, ne f f ,TE1. Figure 2.7 shows the same mode cross over condition, but as aFigure 2.7: Effective index, ne f f , vs waveguide thickness (for width of750 nm) of a Si waveguide with SiO2 cladding (and substrate). Themode profiles are displayed for mode 2 (top row) and mode 3 (bottomerow). At the waveguide height of h = 261.5 nm mode 2 switches froma predominantly TM polarized mode to a predominantly TE polarizedmode.function of waveguide thickness and fixed waveguide width (750 nm). Along widththe effective index, the TE polarization fraction, ξ, as defined in Eq. 2.3, is plotted.Lumerical’s MODE Solutions orders the modes with decreasing effective refractiveindex. Mode 1 is the fundamental TE mode, as it always has the highest effectiverefractive index. For waveguides with h < 261.5 nmmode 2 is the fundamental TMmode and for h > 261.5 nm mode 2 is the second TE mode. The TE polarizationξ is used to label the modes TE or TM. However, the cross over (and degeneracy)only occurs for waveguides symmetric in both the x and y direction. Figure 2.7 isshowing the results for a Si waveguide with both SiO2 substrate and cladding. In thecase of an aqueous solution, the symmetry is broken in y direction. With a silicon23oxide cladding (symmetry) the two modes are frequency-degenerate eigenmodesof the system. In the non-symmetric case (with water cladding, see Fig. 2.8) thetwo modes are no longer uncoupled and the system will have two new eigenmodeswhich are a linear combination of the original symmetric modes. This is similarto what happens in a counter directional coupler with a forward and backwardtraveling wave. In a ring resonator, this phenomenon will lead to resonant peaksplitting [31, 77–80].Figure 2.8: Effective index, ne f f , vs waveguide width (for thickness of 220200 nm) of a Si wavguide with H2O cladding and SiO2 substrate. Themode profiles are displayed for mode 2 (top row) and mode 3 (bottomerow). At the waveguide height of h = 264.5 nm mode 2 switches froma predominantly TM polarized mode to a predominantly TE polarizedmode.The group index is also dependent on the waveguide dimensions. Figure 2.9shows the group index as function of waveguide width and a waveguide height of220 nm. The group index is plotted for 4 modes. The first mode (blue with squaremarkers) corresponds to the fundamental TE mode. The second mode (red withcirclemarkers) corresponds to the fundamental TMmode forwaveguide thicknesses24smaller than 650 nm. Around the waveguide width of 650 nm mode 2 starts beingaffected by mode 3. This is similar to what is described in Fig. 2.8. For waveguidewidth larger than 650 nm the second mode is mostly TE polarized. Mode 4 onlyexist for waveguide width larger than 750 nm.Figure 2.9: The group index, ng, vswaveguidewidth (for thickness of 200 nm)of a Si wavguide with H2O cladding and SiO2 substrate2.1.2 Evanescent fieldThe sensor’s response scales with the modal overlap of the electric field vectorwith refractive index perturbation [32, 81, 82]. Figure 2.3 and 2.4 describe howthe electric field intensity decays away from the waveguide surface by plotting theelectric field at a cross section of the waveguide. Therefore a waveguide sensor hasa limited reach into the cladding. For a waveguide slab (mode confinement in ydirection only) the electric field decay near the surface can be expressed as:E(y) = E0e−y 2piλ√n2e f f−n2clad (2.7)where E0 is the electric field at the surface. For a waveguide (mode confinementin x and y) the equation approximately describes the evanescent field decay at thecenter of the waveguide, i.e. E(x = w/2, y) and E(x, y = h2 ) [83]. The characteristic25decay length or penetration depth is defined as the distance after which the field isE(y) = E0/e. The penetration depth is also sometimes referred to as 1/e length:d1/e =λ2pi1√n2e f f− n2clad(2.8)Since the evanescent field decays exponentially, one can expect a higher surfacesensitivity close to the waveguide. For an increasing thickness of adlayer the sensi-tivity will decrease and the next adlayer will eventually not change the propagationproperties of the optical mode inside the waveguide anymore (see Fig. 2.10).Figure 2.10: The sensitivity is proportional to the modal overlap with theanalyte (hashed area)The characteristic length is used to indicate the reach of sensor into the analyte.For a quasi-TE mode with ne f f = 2.43 and nclad = nH2O = 1.333 at λ = 1550 nmthe characteristic length is dTE = 121 nm while a quasi-TM mode with ne f f = 1.8yields dTM = 203 nm. An index perturbation beyond these distances will not causea change in refractive index of the waveguide.2.1.3 Susceptability or waveguide sensitivityFor biosensing applications, it is of interest to understand the rate at which theeffective refractive index changes (∆ne f f ) as a function the cladding’s refractiveindex (∆nclad) or as a function of adlayer thickness (∆dad). Other factors caninfluence the effective refractive index as well (e.g. temperature and dispersion).When accounting for dispersion and temperature, a first order approximation for26the change of effective refractive index (∆ne f f ) is:∆ne f f =(∂ne f f∂nclad)∆nclad +(∂ne f f∂dad)∆dad +(∂ne f f∂λ)∆λ +(∂ne f f∂T)∆T + ...(2.9)The term ∂ne f f∂nclad is referred to as waveguide susceptibility or waveguide sensitivity[32, 81, 82]. Similarly, the second term ∂ne f f∂dad can be defined as the adlayersusceptibility. The third term ∂ne f f∂λ is a consequence of dispersion and the fourthFigure 2.11: Waveguide susceptibility for the fundamental quasi-TE mode (atλ = 1550 nm and T = cst): a) Susceptibility plotted as function ofwaveguide height and width; b) line plots for fixed waveguide widthsof w = 350 nm, w = 400 nm, and w = 500 nm; c) Line plots for fixedwaveguide heights of h = 220 nm, h = 150 nm, h = 120 nm, andh = 90 nmterm ∂ne f f∂T describes the temperate susceptibility. Assuming that temperature isconstant, the last term can be ignored for now. For the first order approximationto be valid it is further assumed that ne f f is only slowly changing in the range27of interest. For free-space optics this ratio ∂ne f f∂nclad = 1 [84]. Figures 2.11(a) and2.12(a) show the waveguide susceptibility for a quasi-TE mode and quasi-TMmoderespectively for different waveguide geometries.Figure 2.12: Waveguide susceptibility for the fundamental quasi-TM mode(at λ = 1550 nm and T = cst): a) Susceptibility plotted as function ofwaveguide height and width; b) line plots for fixed waveguide widthsof w = 305 nm, w = 500 nm, and w = 600 nm; c) Line plots for fixedwaveguide heights of h = 280 nm, h = 245 nm, and h = 220 nmFigure 2.11(b) shows line plots for fixed waveguide widths of w = 350 nm,w = 400 nm, and w = 500 nm. Figure 2.11(c) shows similar line plots but for fixedwaveguide heights of h = 220 nm, h = 150 nm, h = 120 nm, and h = 90 nm. Figure2.12(b) line plots for fixed waveguide widths of w = 305 nm, w = 500 nm, andw = 600 nm. Figure 2.12(c) shows similar line plots for fixed waveguide heights ofh = 280 nm, h = 245 nm, and h = 220 nm.The change of effective index caused by a local property change can also be28found from the variational theorem [74, 83, 85, 86]:δne f f = c∫∆Em · E∗mdxdy (2.10)where Em(x, y) is the normalized modal electric field vector. The local change ofthe dielectric constant ∆ (x, y) is either caused by a bulk refractive index changeor a variation of thickness of an adsorbed molecular layer and scales with thesquared amplitude of the electric field (this is also true for temperature). In order toapply Eq. 2.10 or 2.9 one needs to know the modal electric field vector for a givenwaveguide geometry. The Eigenmode solver MODE from Lumerical Solutions,Inc. is used to predict waveguide mode sensitivities as function of waveguidegeometry.To increase the sensitivity is to maximize the total modal intensity containedin the volume where the refractive index changes. A weaker confined mode willtherefore always have a higher bulk sensitivity since the mode is delocalized andmore modal power is propagating outside the waveguide. However, for mostbiosensing applications, the refractive index changes only at the surface within afew 10s of nanometers and a more localized electromagnetic field near the surfaceis desired [87]. It is helpful to distinguish between bulk and surface sensitivity.The dependence of both sensitivities on waveguide geometry is discussed in thefollowing paragraphs.2.1.4 LossAnother important property of a mode is the loss, or power attenuation, as the modetravels down the waveguide. The loss is typically expressed in[dBm], but can alsobe converted to an absorption coefficient[1m]by:α[1/m] = 10 · ln(10)α[dB/m] = α[dB/m]10log10(e)=α[dB/m]4.34(2.11)Loss can be calculated based on the imaginary part of effective index:α [ dBm] = 10log10(P(z = 1)P(z = 0))= 10log10*.,e−2ni2piλ01+/- =10ni4piλ0 ln10(2.12)29where ni is the imaginary part of the effective index of the mode. Losses in awaveguide can be attributed to different sources. For most fabrication processes,the dominating loss originates from scattering due to sidewall roughness [88, 89].Therefor a lot of effort has been put into reducing the sidewall roughness byemploying additional fabrication steps [90, 91]. The roughness on top of thewaveguide compared to the sidewall is much smaller due to polishing processes[88–90]. For a TE mode the loss due to sidewall roughness is typically 2-3 dB/cm(for a waveguide width dimensions 500 nm x 220 nm fabricated through deep UVlithography) [92]. Sidewall roughness is more severe for larger index contrasts andtherefore waveguides in air have higher scattering loss than waveguides in SiO2[31]. In general the scattering loss can be reduced by minimizing the modal overlapwith the sidewall, i.e. by increasing the waveguide width or by using a differentwaveguide cross section. Loss originating from material absorption is typicallynegligible for passive structures, as both silicon and silicon oxide (as the claddingmaterial) are nearly lossless. However, this is not the case for doped silicon [32] orwhen the oxide cladding is replaced by an aqueous solutions as the case in sensingapplications. In fact, as described in section 2.2.4, the water absorption is thedominant loss mechanism. Two-photon absorption and free carrier absorption canbecome an issue for high powers [31, 93, 94]. Surface state absorption may alsocome into play for waveguides not properly passivated [95].Additionally, a waveguide mode can also couple to radiative modes in thesubstrate. This substrate leakage depends on the bottom cladding thickness. For atypical silicon-on-insulator (SOI) wafer the oxide layer is 2-3 µm thick which resultsin negligible substrate leakage loss for TE mode and in the order of 0.001 dB/cmfor the TM mode [89] .The high index contrast between silicon and silicon oxide and the resultingstrong mode confinement allows for very sharp bends in a waveguide while keepingradiation loss low. Typical loss values for 90o bend are around 0.01 dB/90o for aTE mode and a bend radius of around 3 µm [88, 96]. The loss per bend originatefrom the radiation loss and the mode mismatch at the interface from a straightto bend waveguide. Because the mode is slightly pushed toward the waveguidewall, scattering losses due to sidewall roughness might also increase slightly. Forwaveguide routing (connection from sensor to I/O location) bends with radius305 µm for TE mode and 10 µm for TM mode have been chosen as the loss for suchbends is sufficiently small to not matter anymore in the overall loss budget [96].2.2 Figures of meritAny analyte induced change in the evanescent field region (refractive, absorptive,or luminescent property change) of the propagating mode can alter the propagationproperties (effective index and loss) of that mode. The focus in this thesis isrefractive index sensing, i.e. the change of the real part of the effective refractiveindex of the mode caused by a change of the refractive index of the cladding. Therehas been a fragmented effort to define a set of performance metrics for refractiveindex sensors[32–34]. The lack of universally accepted, comparative standardscomplicates performance comparisons among different types of silicon photonicbiosensors but even more so between different types of sensors in gerenal (e.g.surface plasmon resonance seensors or electrochemical sensors).2.2.1 Resonator sensitivitySection 2.1.3 introduced the concept of susceptibility which describes how a changein the cladding or an adsorbed adlayer can alter the propagation properties of theoptical mode in a waveguide. A resonant structure can be used to translate theeffective index change into a resonance wavelength shift (∆λres). The resonatorsensitivity is given by:S =∆λres∆n(2.13)where λres is the sensor’s resonant wavelength [nm], and, n, is the refractive index[RIU] of the analyte. The refractive index change, ∆n, will be caused by eitherthe a uniform refractive index change in the cladding , nclad, or by molecularadsorption (growth of adlayer). We will distinguish between (1) bulk sensitivity,which accounts for refractive index changes of the waveguide’s entire cladding(the bulk) and (2) surface sensitivity, which describes resonant wavelength shiftsfor refractive index changes near the waveguide’s surface (e.g.: protein adlayer).When not specifically called out, most discussions on sensitivity refer to the bulksensitivity. This section briefly describes both types as well as how they are31simulated numerically.Bulk sensitivityThe bulk sensitivity for resonator is defined as the change in resonant wavelengthover change in cladding refractive index [84, 97, 98]:S =∆λres∆nclad(2.14)where λres is the sensor’s resonant wavelength [nm], and nclad is the cladding’srefractive index unit. The bulk sensitivity is often also referred to as RI sensitivityand has the units of nm/RIU. RI sensitivity has often been used to compare SPRsensors [19] and can be extended to silicon photonic resonators as well. The RIsensitivity will be referred to as the ’bulk sensitivity’ for the remainder of thisdocument. For any resonator, a change in ne f f will affect the resonant wavelengthλres and in turn ne f f itself (non-zero slope of∂ne f f∂λ ). ∆λres is given by:∆λ =∆ne f f Lm,m = 1,2,3, ... (2.15)where m is the longitudinal mode order of the resonant mode and L the round triplength. When accounting for dispersion, i.e. ne f f = ne f f (λ), ∆λres becomes (firstorder approximation):∆λres =Lm(∂ne f f∂nclad)λres,n0e f f∆nclad +Lm(∂ne f f∂λ)λres,n0e f f∆λ (2.16)where n0e f fis the initial effective index at λres. Using Eq. 2.6 this becomes∆λres =λresng(∂ne f f∂nclad)λres,n0clad∆nclad (2.17)where ng is the group index. The susceptibility term∂ne f f∂ncladwas introduced in 2.1.3and can be thought of as the portion of the optical mode’s intensity interacting withthe sample. For the fundamental TE and TM modes in a typical strip waveguide,the mode sensitivity (or susceptibility to refractive index changes in the cladding)32Figure 2.13: Bulk sensitivity for the fundamental quasi-TE mode (1550 nm):a) Bulk sensitivity plotted as function ofwaveguide height andwidth; b)Line plots for fixed waveguide widths of w = 400 nm, w = 500 nm, andw = 600 nm; c) Line plots for fixed waveguide heights of h = 280 nm,h = 220 nm, h = 150 nm, and h = 100 nmis given in section 2.1.3. A similar expression to Eq. 2.17 is given by Ackert et al.for a change in core refractive index [99] or by Baehr-Jones et al. for temperaturesensitivity [82].For slab waveguides the bulk sensitivity can be analyzed analytically [97, 98,100]. However because of the lateral confinement of waveguides, numerical simu-lations have to be performed to find the mode field distribution, susceptibility, andgroup index (as described in section 2.1).Simulated bulk sensitivity: Finally, with the waveguide susceptibility computedin MODE Solutions (see Figs. 2.11 and 2.12) and using formula 2.17, the bulksensitivity can be calculated. Figure 2.13 shows the bulk sensitivity as function of33Figure 2.14: Bulk sensitivity for the fundamental quasi-TMmode (1550 nm):a) Bulk sensitivity plotted as function ofwaveguide height andwidth; b)Line plots for fixed waveguide widths of w = 400 nm, w = 500 nm, andw = 580 nm; c) Line plots for fixed waveguide heights of h = 250 nm,h = 220 nm, and h = 200 nmwaveguide geometry for the fundamental TE mode. From Figs. 2.13 (b) and 2.13(c) the bulk sensitivity of ≈ 60 nm/RIU is extracted. It can also be seen that the bulksensitivity is improved by decreasing the width and the height of the waveguide.Similarly, Fig. 2.14 shows the bulk sensitivity as function of waveguide ge-ometry for the fundamental TM mode. From Figs. 2.14(b) and 2.14(c) the bulksensitivity of ≈ 200 nm/RIU is extracted for a waveguide with dimensions of500 nm x 220 nm. It can also be seen that the bulk sensitivity is improved bydecreasing the width and the height of the waveguide.34Surface sensitivityWhile the bulk refractive index sensitivity is an important measure, it only describesthe effect of a uniform change of the entire cladding material. This is of course notwhat is a happening during a biological assay, where target molecules are capturedat the surface.The resonant wavelength response (∆λres) to a homogenous adlayer on itssurface can be calculated using:∆λres =Lm(∂ne f f∂dad)∆dad +Lm(∂ne f f∂λres)∆λres =λresng(∂ne f f∂dad)λ0,d0ad∆dad(2.18)where dad is the molecular layer thickness. This equation assumes a constantrefractive index of the adlayer. Figure 2.15 shows the simulated surface sensitivityas function of layer thickness. The waveguide dimensions are 500 nm x 220 nmand the adlayer is assumed to have a constant refractive index of 1.48.Since the evanescent field decays exponentially, one can expect a higher surfacesensitivity close to the waveguide. For an increasing thickness of adlayer the sensi-tivity will decrease and the next adlayer will eventually not change the propagationproperties of the optical mode inside the waveguide anymore. Every mode has onlya limited reach into the media (see Fig. 2.10).Figure 2.15: Surface sensitivity with a growing thickness of the adlayer (fora waveguide with dimensions 500 nm x 220 nm). a) The surfacesensitivity is plotted as function of layer thickness; b) The sensoroutput signal (∆λres) as function of the layer thickness.35By using the adsorbed mass rather than layer thickness Eq. 2.18 becomes:∆λres =λresngAρ∂ne f f∂d∆M (2.19)where M is the adsorbed mass (M = ρAd), ρ is the molecular density, and Ais the sensing area. For a typical sensor geometry (see chap 4 and 5) with asensing area (A) of 10 µm 2, a group index (ng) of 4.2, and a minimal detectablewavelength shift (∆λres,min) of 5pm and further assuming a protein density of1.33 g/cm3 [101, 102], an RI sensitivity (∂ne f f∂d ) of 0.2 µm−1, a mass of 1.3fg canbe detected using Eq. 2.18. This example shows that sensing performance is notonly described by sensitivity but also the minimal detectable wavelength shift ofthe system (∆λres,min). The detection limit is defined in section 2.2.2.Figure 2.16: Waveguide surface sensitivity for the fundamental quasi-TEmode (1550 nm): a) Surface sensitivity plotted as function of waveg-uide height and width; b) line plots for fixed waveguide widths ofw = 400 nm, w = 500 nm, and w = 600 nm; c) Line plots for fixedwaveguide heights of h = 280 nm, h = 220 nm, h = 150 nmFigure 2.16 shows the surface sensitivity of the fundamental TE mode for36various waveguide dimensions at the wavelength of 1550 nm. The propagationproperties (ne f f ,ng) are computed for a strip waveguide in an aqueous solution andwith an adlayer of thickness 10 nm with a constant refractive index of 1.48. In aFigure 2.17: Waveguide surface sensitivity for the fundamental quasi-TMmode (1550 nm): a) Surface sensitivity plotted as function of waveg-uide height and width; b) line plots for fixed waveguide widths ofw = 400 nm, w = 500 nm, and w = 600 nm; c) Line plots for fixedwaveguide heights of h = 280 nm, h = 250 nm, h = 220 nm, andh = 200 nmsimilar way figure 2.17 shows the surface sensitivity of the fundamental TM modefor various waveguide dimensions also at the wavelength of 1550 nm.2.2.2 Limit of detectionAs highlighted with Eq. 2.18 and Eq. 2.19, it follows that the sensing performance,i.e. the smallest measurable incremental adlayer thickness change, is limited by theminimal detectable wavelength shift of the system (∆λres,min).The system’s limit of detection, as introduced in section 1.3.2, is defined as theminimum refractive index change (∆nclad,min), or smallest mass change, necessary37to cause a detectable change in the output signal (above 3σ of the noise floor). Asexplained in 1.3.2 the sLoD depends very much on the instrumental setup whichmakes it challenging to objectively compare sensor performance throughout theliterature. We therefore introduced the intrinsic limit of detection (iLoD) [32],which is dependent only on intrinsic resonator characteristics (see also Eq. 1.2).This definition assumes that the minimum detectable wavelength shift is equal oneline width of the resonator ∆λres,min = ∆λ3dB.Depending on the resonance shape, the resolution of the discretization (= thewavelength step for a spectral analysis) and the fitting algorithm, the minimaldetectable wavelength shift can be improved (i.e. ∆λmin << ∆λ3dB).The linewidth of a resonator can be estimated from the quality factorQ ≈ λres∆λ3dB.The quality factor Q is a measure of the photon lifetime in the resonator as definedin section 1.3.2. The approximationQ ≈ λres∆λ3dBis valid only for high quality factors.2.2.3 Improving the intrinsic limit of detectionOne technique for improving the limit of detection of a waveguide-based resonatorsensor is to increase its sensitivity (S). This can be done by increasing the interactionbetween the propagating optical mode and the cladding medium (i.e. the sample).When a larger portion of the optical field travels outside the silicon waveguide core,the mode’s susceptibility increases, enhancing sensitivity and causing a resonancewavelength change. Of course this is only true for the bulk sensitivity and notnecessarily also for an adlayer and the corresponding surface sensitivity. Moreoptical field outside the waveguide does not imply a larger overlap of the field withthe adlayer.In addition, increased water absorption losses of weakly confined modes com-bined with greater susceptibility degrades the resonator’s quality factor. Thus,trade-offs exist among sensitivity, quality factor, and iLoD. The loss term in Equa-tion 1.3 is the total loss combining scattering loss, bending loss, mode mismatchloss, substrate leakage, and loss due to absorption in the cladding. Except for theabsorption loss in the cladding, the losses can be minimized by design choices.Using perturbation theory, a change of the dielectric constant ( ) in the claddingresults in a change in frequency (ω) (usingDirac bracket notation, see also Eq. 2.10):38∆ω =ω2〈E |∆ |E〉clad〈E | |E〉 (2.20)The subscript in the numerator denotes the integration over the spatial domain ofthe cladding (aqueous solution or analyte). For small perturbation of the cladding,like a molecular binding event, ∆clad  2 · nclad ·∆nclad the sensitivity can thenbe written as [103]:∆λ∆nclad= Γλnclad(2.21)where Γ is the fraction of optical intensity overlapping with the cladding (or liquid).Γ is given by:Γ =〈E | |E〉clad〈E | |E〉 (2.22)According to Eq. 2.21, sensitivity increases with Γ and the upper limit is λnclad .Additionally, the loss due to absorption in the cladding (αclad) also increases withΓ. Qclad, the quality factor associated to the absorption loss of the cladding isgiven by:Qclad = Γ−12pincladλresαclad,[1/m](2.23)This means that as the modal overlap with the cladding increases the sensitivity, theline width of the resonator (the smallest resolvable spectral change) is also increasedand the detection limit remains constant (see chapter 6 for a detailed discussion).This is only the case when the total loss is dominated by the absorption loss of thecladding (Qtot ≈Qclad). The overall quality factor is given by:1Qtot=1Qi+1Qclad+1Qnoise(2.24)where Qi is the intrinsic quality factor associated to scattering loss, radiation loss,mode mismatch loss, coupling loss, material absorption loss, Qnoise is associatedto noise in resonance peak location. To relate the standard deviation of a spectralvariation to the FWHM, ∆λ, the numerical approximation proposed by White et al.is usedσ ≈ ∆λ(4.5·SNR0.25) [34]. Resonatorswith highQi would be preferred as the line39width is then determined by the cladding absorption (or absorbing biomolecules).An increase in sensitivity by changing the modal overlap with the analyte does notaffect the iLoD as a similar increase in line width is observed due to more loss inthe cladding. And hence the iLoD for TE and TM mode sensors at 1550 nm havethe same theoretical limit.2.2.4 Operating wavelengthFor biosensors, with most of the analytes in aqueous solutions, the absorptionloss is determined by water absorption. Assuming that the susceptibility is 1 andthe quality factor is limited by water loss, then the fundamental intrinsic limit ofdetection is 2.4 ·10−4RIU at 1.55 µm wavelengths [32].Figure 2.18 shows the real and imaginary part of the refractive index of water[104]. For the simulation in MODE Solutions, data from Palik are used [105] andfitted using the multi coefficient fitting algorithm provided by MODE. The dataprovided by Kou et al. [104] and Palik et al. [105] are in good agreement.Figure 2.18: Refractive index of water at at 25 oC: a) Real index of refraction;and b) imaginary index of refractionThe iLoD can be improved by an order of magnitude by switching to a shorterwavelength of 1310 nm as the water absorption is decreased (accompanied byslightly higher scattering losses) [32]. The iLoD is approaching the fundamentallimit if αi < 10 m−1.40Figure 2.19: a) Absorption loss in water; and b) ultimate quality factor calcu-lations for resonators in water. Γ = 0.37 and Γ = 0.22 is the simulatedconfinement factor for TM and TE mode respectively for a waveguidewith dimensions of 500 nm x 220nm.2.2.5 Temperature sensitivityThe refractive index of a material is also dependent on temperature. For siliconthe thermo-optic effect, TOC = dndT , is quite large with a coefficient of TOC =1.8 · 10−4 K−1 [106–108]. The temperature dependence originates from changesin the distribution functions of carriers and phonons, and temperature dependenceof the band gap. The effect has been used in applications like thermal switching[109, 110] or to thermally tune modulators [111]. For passive components, thissensitivity on temperature is in general a design challenge. Therefore some efforthas been put into the design of athermal designs [112–117] For silicon oxide, SiO2,the thermo-optic effect is an order ofmagnitude smaller of about dndT ≈ 2.8 ·10−5 K−1[96]. Most biological applications are carried out in water, which has a negativethermo-optic coefficient of about dndT ≈ −9.9 · 10−5 K−1 [118, 119]. The thermooptic coefficient for a waveguide mode with effective index ne f f will depend onthe confinement factor of the mode as defined in Eq. 2.2. A TE mode is expectedto have a higher temperature dependence than a TM mode as more of the electricfield distribution overlaps with the silicon core. Similar to the bulk sensitivity asdefined in Eq. 2.17 one can also define a temperature sensitivity for a resonator.∆λres∆T=λresng(∂ne f f∂T)λres,T0(2.25)41Chapter 4 and 5 will report measured temperature sensitivities for TE and TMresonators of ≈ 70 pm/K and ≈ 30 pm/K respectively.2.3 SummaryThis chapter introduces the figures of merit for silicon photonics biosensors. Theyare: (1) Bulk sensitivity which is the change of resonance condition (wavelengthshift) as function of uniform change of refractive index of the cladding; (2) Surfacesensitivity which is the change of resonance condition (wavelength shift) as functionof thickness of a molecular adlayer on the waveguide’s surface; and (3) Limit ofdetection which is the smallest detectable refractive index or adlayer thicknesschange. It is useful to distinguish between the limit of detection of the systemsLoD and the intrinsic limit of detection iLoD. The former is dependent on themeasurement setup including data processing and the latter is a measure of thesensor alone.Furthermore this chapter also discusses how the intrinsic limit of detection canbe improved and what the fundamental limits are. If the loss of a round trip inthe resonator is dominated by water absorption, then the intrinsic limit of detectionhas reached the fundamental limit and cannot be further improved by design of theresonators but only by choosing a wavelength with less water absorption.42Chapter 3Materials and Methods3.1 IntroductionMany different resonant photonic devices have been investigated as label-freesensors for applications ranging from environmental monitoring [120] and ba-sic science research [121] to bio threat detection [122] and medical diagnostics[22, 35, 41]. These sensors have demonstrated impressive sensitivities and detec-tion limits at clinically relevant levels [22]. Yet there exist considerable barriers toentry for developing novel silicon photonic biosensors. Specifically, (1) develop-ment tools are expensive and have not achieved the integration nor performance ofstandard CMOS designs flows; (2) layout cells and process design kits (PDK) areoften custom and private to the groups who develop them; (3) foundry fabricationand shared shuttle runs are expensive and take months to complete; and (4) nocommercially available platform to automate testing and device characterizationexists (within reach of an academic research budget).To improve the efficiency of silicon photonic device development, we havedeveloped a PDK with basic, well characterized I/O cells to enable rapid and af-fordable prototyping using the University of Washington’s Nanofrabrication Facil-ity’s (WNF’s) direct-write ebeam lithography system [123]. The PDK is availablethrough the SiEPIC program [? ]. We have also developed an automated test andanalysis platform to quickly characterize new devices. The hardware platform’s billof materials (BOM) and the accompanying control software has been open-sourced43on GitHub as the SiPhoTestBench project [124, 125].3.1.1 RequirementsThe requirements for our open-sourced silicon photonic development emerged fromthe challenges we experienced with our current approach to developing photonicdevices. First, we found it more efficient to fabricate many variations, test, and backannotate our simulation models with parameters from the best performing device.Second, we found that fabricating devices using foundries or multi-project waferservices were expensive and often took half a year or more. This delay slowedthe pace at which progress could be made in the field. Third, silicon photonicfoundry processes cannot yet achieve the lithography resolutions offered by direct-write systems like ebeam [126, 127]. This limits circuit topologies that could beexplored or accurately fabricated. Fourth, testing a single photonic biosensor usingFigure 3.1: Fiber-to-fiber testing: a) Each fiber has to be aligned to the gratingcoupler individually. b) Due to the space needed for the positioning stageof each fiber, only one fiber per chip side is possible, making it hard tointegrate it with fluidics.our initial test platform (Figure 3.1) with individual fibers aligned on either sideof the photonic chip, often took weeks, and required a laborious setup effort eachtime a chip with new chemistries was needed. With the relative immaturity andvariability of the fabrication process and the lack of good compact models forcomponents it was still more convenient to fabricate hundreds of devices and selectthe best one (this is true at least for the first few design cycles as compact modelscan be improved each iteration). To support this type of developmental approach,44an automated test platform with the ability to test hundreds of devices within afew hours was needed. Finally, we formerly had to acquire data without any realtime monitoring, and then analyze it post acquisition. A bio experiment commonlyinvolved continuous wavelength sweeping for several hours to track resonant peaks(number of sweeps in order ∼ 103). This resulted in two issues: (1) the lack ofreal-time monitoring often resulted in wasted time and effort because undetected airbubbles would often destroy the assay and (2) custom, one-off scripts were neededto analyze each experimental dataset. An analysis tool that could be leveraged forall the experiments conducted on the test platform was needed.Our goal was to create a rapid and low-cost prototyping process, an efficient,automated testing platform with analysis tools that would allow the design andcharacterization of thousands of devices in a week or less. To achieve this, ourdevelopment methodology and test platform requirements needed to include:• Low-cost (for both the test and characterization platform as well as the fabri-cation process)• Quick turn fabrication approach (like direct-write lithography)• Test automation (of time-consuming tasks)• Consistent and repeatable results. This required:– Robust simulation models that accurately predict fabricated behavior– Well characterized and stable fabrication process• A cell library and prototype process compatible with those available througha foundry process.• Software that was flexible and easily extensible to other components• A chip framework with test structures that provided insight into fabricatedprocess parameters (eg: loss, WG dimensions, etc.)• On-chip alignment features for efficient chip registration and setupThe remaining sections describe key aspects of the platform and its accompanyingPDK.Wewill discuss how the platform andmethodologywork in concert to develop45silicon photonic devices efficiently. We will describe how we constructed the testplatform’s hardware, control software, and analysis tool. We will also discuss thecharacterization results in terms of noise sources and detection limits, especially asthey relate to biosensing performance.To the best of our knowledge, we have created the first open-sourced PDK anddevelopment platform for silicon photonic biosenors that improves the efficiencyof device development and lowers the barrier to entry. Our hope is that by creatingthis versatile and extensible test platform and making it freely available to otherresearch groups, silicon photonic biosensors will continue to progress toward theircommercialization potential where their true value and impact can be broadlyrealized.3.2 Development methodologyThis section describes our general approach in developing new photonics devices,i.e. biosensors, and how the platform plays a key role in the design cycle. A fulldesign cycle is shown in Figure 3.2.The iterative cycle begins with a design and simulation phase to optimizesensor performance and determine layout parameters. For the first few designcycles the designs are preferably fabricated with a low cost process with fast turn-around times. Electron-Beam lithography is often employed as a fast prototypingprocess since access to such facilities is easy, relatively cheap, tooling costs arelow, and sample volumes are small. Our group has access at the WashingtonNanofabrication Facility (WNF) [128]. They offer a process optimized for siliconphotonic device fabrication [129]. However the process variability is high as is thecase in general for EBeam lithography in a University setting. While today’s EDAtools for most CMOS process flows incorporate process parasitic and variationsinto simulations to ensure first-time success, these sophisticated capabilities haveyet to be mainstreamed into silicon photonic biosensor design flows [130, 131].With the inability to accurately simulate a multi-component signal path withprocess variations and lithography effects, it is often more efficient to fabricateone photonic device with small variations and then empirically determine the bestdesign parameters fromdevice test versus simulating all the variations, especially for46Figure 3.2: Rapid and affordable development process for creatingapplication-specific and foundry-compatible, silicon photonic biosen-sors for label-free sensing. (a) Optical mode profile simulated usingLumerical’s MODE solutions. (b) Fabrication and SEM using the UW’sJEOL ebeam. (c) Optical setup schematic for dry and wet testing. (d)Optical transmission spectrum for various NaCl concentrations usingMATLAB. (e) Optimized sensor designs scaled-up in foundry processesthrough and MPW broker.biosensors where their operating environment is not easily predicted or controlled.In order to close the design cycle it is therefore important to have the capability oftesting a large number of devices in a reasonable time frame to identify workingdevices. The test platform automatically characterizes each sensor’s performanceusing a set of sequenced, refractive index standards and generates a test summarywith key performance parameters (Q, S, and iLoD). The biosensing capabilities forthe best performing sensors are demonstrated using amodel biological system assay.47To improve performance even further, the pre-fab simulation models are modifiedto mimic observed results, and to provide insight into the overall system design andbehavior. The layout is then updatedwith new parameters before another fabricationand test iteration. Final designs are mass produced using silicon photonic foundryprocesses [132–136].3.2.1 Device designSystem design and modeling typically starts with the simulation of individualdevices and components to synthesize simplified models for system simulations[131, 137]. The most accurate tool to use to model light interaction with com-plicated nanostructures is a three-dimensional (3D) finite-difference time-domain(FDTD) solver. However, with decreased mesh size for increased accuracy andlarger simulation domains, it becomes computationally too expensive to run a full3D FDTD simulation. Often it is enough to solve for the optical mode in thewaveguide cross section at specific frequencies since the mode shape does notchange as it propagates along the waveguide [138]. For more sophisticated modepropagation with changes in the mode shape, coupling to other modes, radiation, in-terference, and reflection other techniques exists such as 2D and 2.5D FDTD, BeamPropagation Method (BPM) and Eigenmode Expansion Method (EME). These areusually developed for specific applications and the designer has to be aware of thelimitations of each of these approaches. For active devices, electro-optic effectsneed to be modeled as well. If possible the problem is partitioned into an opticalsimulation and an electronic simulation. An electronic simulation involves findinga solution for the Poisson and the drift equation (charged carrier density as functionof applied voltage). Since photonic circuits are very sensitive to temperature it isalso beneficial to do a thermal modeling of the device, and also of the chip itself,since most likely the temperature will not be uniform across a photonic system.Thermal modeling is primarily done using finite element method (FEM) to solvethe heat equation.483.2.2 Device fabricationAn overview of the key process steps used to fabricate silicon photonic biosensorsis shown in Figure 3.3. Devices are fabricated on SOI wafer with a 220nm thick Silayer 2 or 3 um buried oxide layer. For fabrication two strategies can be used. 1)Electron-Beam lithography as offered through many universities [128, 139, 140] or;2) Deep-ultraviolet (DUV) lithography available at CMOS foundries [132–136].Figure 3.3: Typical process steps to fabricate silicon photonic biosensors. (a)An SOI wafer is cleaved and baked with photoresist. (b) Waveguidepatterns are written using a mask and UV exposure or directly usingebeam. (d) Exposed regions are etched, removing the silicon. (e) theresist is removed.Foundry fabricationFoundry processes like ePIXfab/IMEC, CEA-LETI, or Astar/IME are typicallyaccessed via multi-project-waver runs (MPW) organized by CMCMicrosystems orMOSIS.49Ebeam fabricationThe Ebeam provides a low-cost, fast turn-around CMOS-foundry-compatible fab-rication process that has been optimized [126, 141, 142] to produce consistent,robust, low-loss silicon photonic components [127, 143]. Multi-etch layer designswere realized by repeating the resist, ebeam write, and etch process on the 220 nmSOI material.3.2.3 Basic circuit layoutTo support fast and easy-to-setup testing of different sensor types, the photoniccircuit of an individual sensor channel followed the basic layout shown in Fig. 3.4.An input waveguide is split into two waveguides to enable sensing in two individualaddressable microchannels. Fig. 3.4 shows the example layout of a ring resonatorand a Bragg grating. Long routing waveguides connect the input and output gratingcouplers to the sensing regions. The long routing waveguides are necessary asthe grating couplers need to be accessible by the fiber array while the rest of thephotonic chip is covered with the fluidic gasket and fluidic flow cell.Figure 3.4: Basic circuit layout: Long routing waveguides connect the in-put and output grating couplers to the sensors located in two differentmicrochannels.50CYTOP claddingTo improve durability, reuse, and reduce optical losses in the long routing waveg-uides between the vertical grating couplers and biosensors, chips are often claddedwith CYTOP, a perfluoro-polymer with exceptional chemical resistance and usefuloptical properties (Over 95% light transmittance through DUV to NIR range and aRI of 1.34) [144]. Openings in the CYTOP cladding are then created around thephotonic sensors [145]. Undiluted CYTOP was spun onto the chip at 3500 rpm andthen baked at 60oC for 30 minutes. After 30 minutes, the temperature was rampedto 180oC at 4oC/min and held for 1 hr before ramping back down to 90oC. AZ9260was spun onto the CYTOP at approximately 7 µm thick, and then patterned to createthe etch windows around sensors. An Oxford Plasmalab System 100 was used toetch the CYTOP using oxygen. Finally, the AZ9260 was removed with acetone andcleaned with isopropanol.Suspending waveguidesSuspending silicon waveguides on SOI substrates to increase confinement andreduce absorptive losses through the substrate has been demonstrated for high per-formance and telecommunication applications [146–149]. Suspending TM modebiosensors could be advantageous by exposing the energy typically guided in thesubstrate to the cladding, essentially doubling the device’s sensitivity. Therefore,TM mode Bragg gratings were post-processed at the WNF to suspend them in afluidic channel.Chips on which to create the suspended sections were first cleaned in a piranhasolution and then surface treated using hexamethyldisilazane to promote photoresistadhesion and to help avoid resist lift off during the buried oxide etch. Next, AZ1512,a standard broadband Novalak based photoresist, was spun onto the chip, baked,exposed, and developed to pattern regions for undercut. Buffered oxide etch (10:1)was used to etch through the 3 µm thick buried oxide later. Without drying, chipswere rinsed in water and then bathed in acetone to dissolve the photoresist.Next, the chips were transferred (while still wet) to a beaker of isopropanol (toavoid stiction) for rinsing and then dried with a gentle stream of nitrogen. SEMimages confirmed that the structures were released, as shown in Figure 5.17.513.2.4 Device testing and performance characterizationAutomated testing was used to characterize every silicon photonic biosensor usingspecific modes in the control software application which included: (1) dry test, (2)wet test, (3), salt-step, and (4) bio assay. Sincewet performance characterization andbio demonstrations are time consuming and involve sequencing of various reagents,devices are first tested and screened with a dry test to ensure the waveguides yieldedas expected from the fabrication process. Although each device was designed foran aqueous cladding, performing this initial test in air was enough to determine if itshould be tested further. If so, the gasket and flow cell (as described in section 3.3.3)were mounted and devices were wet tested to assess performance parameters suchas its Q and extinction ratio (ER). The best performing devices are then subjectedto a series of refractive index solutions to determine their sensitivity and intrinsiclimit of detection (iLoD) using the software’s automated salt-step mode [87].Figure 3.5: Devices can be characterized in a dry environment or in wetenvironment with a flow cell mounted on the chip. (a) Rendering ofthe custom hardware platform. (b) Image of fiber array, Teflon flowcell, and device under test (as described in section 3.3.3). (c) Custompoly-dimethyl siloxane (PDMS) flow cell reversibly bonded to a chipunder test .Refractive index solutions of 62.5 mM, 125 mM, 250 mM, 500 mM, and1 M NaCL were diluted from a 2 M stock of NaCl (Acros Organics, ThermoFisher Scientific) using ultra-pure deionized water (Barnstead Nanopure, ThermoScientific). Titrationswere degassed under vacuumusing and ultrasonic bath (VWRB2500A-MTH) and stored in 50 mL Falcon tubes until used.52Assessing biosensing performanceA modified sandwich assay comprised of well-characterized biomolecules was im-plemented to verify each sensor’s capacity to probe biomolecular interactions withspecific and non-specific targets (Figure 3.6 ) [150, 151]. All sensors were exposedto the same sequence of biological reagents, and flow rates between 10-15 µL/minusing the microfluidic device described above. The syringe pump was pausedbriefly after each phase of the assay in order to facilitate switching of solutions.Additionally, the microfluidic chamber was flushed with phosphate buffered saline(PBS) between each reagent step to remove any unbound molecules from the sen-sor surface. The optical stage was thermally tuned to 37oC, which maintainedconstant sensor temperature throughout the assay while mimicking physiologicalconditions. The constant temperature is maintained with a thermoelectric heaterand a thermistor configured in a closed loop (as described in section 3.3.1). Theresonant wavelengths and power values were recorded every 20-45 seconds usingthe MATLAB software, and post-processing of data was performed in a separatescript.Figure 3.6: Modified sandwich bioassay. Reagent sequencing correspondingto regions [A-E] subjected to each sensor and shown in the resultssection. Region A = Protein A adsorption, B = anti-streptavidin (Anti-SA) functionalization, C = Bovine Serum Albumin (BSA) challengeand block, D = streptavidin (SA) target analyte binding, E = Biotin-BSAamplification step. Introduction of each reagent was followed by a PBSwash.Prior to any functionalizationwith biologicalmolecules, the sensors were rinsedwith PBS solution (n = 1.35) for 20 minutes to establish a baseline for further signalmeasurements. The first biological reagent introduced to the system was Protein53A, a 42 kDa globular protein with a diameter of 3 nm [152, 153] and a refractiveindex of 1.48 [101]. Originally isolated from Staphylococcus aureus, Protein Ademonstrates a high affinity for human and mouse IgG antibodies. We utilizedphysisorption techniques to irreversibly bind Protein A to the sensor surface asshown in Region A of Figure 3.6 [154, 155]. Functionalization was performedeither passively (off-line before executing the assay) or in real-time as the first stepof the assay after achieving a baseline signal with PBS. The initial deposit of ProteinA has been shown to denature and form a biologically inactive layer, but the secondlayer forms on top of the first and presents a film of active receptor sites to bind theFc domain of IgG [156]. Therefore, we assumed a 1-3 nm thick physisorbed layerof Protein A for assessment of experimental results (using MODE simulations).The high binding affinity of Protein A facilitated proper orientation of the captureantibodies in step two; antibodies were arranged on the surface with their antigen-binding domains directed toward the flow of conjugating solutions. It should benoted that while covalent methods of surface functionalization are preferred forrobust, prolonged biological assays, physisorption was considered sufficient as avalidation process for our system.Following a flush with PBS solution, the IgG isotype antibody anti-streptavidin(anti-SA) (Vector Labs, Burlingame, CA) was introduced to the channel at 10-125 µg/mL. Anti-SA is a 150 kDa protein with a length of 15 nm, so resultingresonant peak shifts were larger in comparison to those of other biomolecules usedin the modified sandwich assay. Immobilization to Protein A oriented the anti-SAFab binding domains toward the center of the microfluidic channel (Figure 3.6,Region B). This prepared the channel for the third assay step, in which bovineserum albumin (BSA) (Sigma Aldrich, St. Louis, MO) was introduced to thechannel at 20-100 µg/mL and immobilized to the Fab binding domains of anti-SA(Figure 3.6, Region C). BSA acted as a negative control for non-specific binding,demonstrating the bio-specificity of the modified sandwich assay. Furthermore,the 66 kDa globular protein blocked parts of the sensor not functionalized withanti-SA to ensure that molecules in the subsequent solution steps would bind to theantibody’s Fab regions and not to the sensor surface. Ideally, this assay step wouldnot result in resonant wavelength shifts, but small positive shifts indicated smallamounts of non-specific BSA adsorption.54The fourth biomolecule introduced to the sensor, streptavidin (SA) (VectorLabs, Burlingame, CA), is a 57 kDa antigen conjugate of anti-SA. This solutionwas presented to the anti-SA molecules on the sensor surface at 10 µg/mL (Figure3.6, Region D) and a significant shift in resonant wavelength was observed. Rinsingwith PBS did not cause a decrease in signal, indicating irreversible binding andconfirming the sensors’ ability to bind specific biological species. For the finalassay step, biotinylated BSA (bBSA) (prepared per instructions; SoluLink, SanDiego, CA) was introduced to the channel at 10-50 µg/mL (Figure 3.6, Region E).The resulting wavelength shift demonstrated that captured SA retained its biotin-specific targeting function. Addition of a secondary antibody also provided signalamplification, further intensifying the sensor’s response and improving sensitivityto low concentrations of analyte.3.2.5 PDK and chip frameworkThere are a number of basic building blocks every designer will use to completean optical circuit. For example, vertical gratings [157] to couple light on and offthe chip or waveguide splitters to fanout a single input to many devices. Typically,such components are collected in a shared library and along with their processinformation (eg: layer definitions) collectively make up a PDK [138]. PDKs canreduce design time and risk. Characterized components can be leveraged intooptical circuits and placed into a GDS framework to realize whole chips quickly,as shown in Fig. 3.7. A GDS is an industry-standard CAD format that definesgeometries and layers for chip fabrication.Our PDK contains a GDS framework with additional layout components tomonitor the quality of the fabrication process. Such test structures are designed tounderstand basic optical properties of components and to get a statistical measureon yield. Examples include the propagation loss in a waveguide or the effectiveor group index of an optical path. Furthermore, the GDS framework defines fourquadrants that can be diced into die. Within each die, alignment structures exist atspecific locations to facilitate efficient registration and setup with the probe station.55Figure 3.7: PDK concept and chip framework that expedites developmentand lowers design risk. The chip framework specifies features to assessfabrication steps and specifies alignment structures that facilitate theautomated testing of all devices on the chip.3.2.6 Generating a device listAll of the devices defined in theGDShave coordinate pairs that define their locationsrelative to each other. The probe station software translates these GDS coordinatesinto stage positions to move from one device to anther during automated testing.The PDKprovides a script that automatically generates a text file with all the devicesand their coordinates. The script can be run from KLayout or Mentor GraphicsCalibre. The input grating coupler of each device to test should get a text labelstarting with ’opt_in’ to be recognized by the script. The coordinate of the text labelwill be written to the coordinate file and used to align the input fiber array to thegrating coupler during automated testing. A text file with devices of interest can becreated by hand but must have the following format:opt_in_<mode>_<wvl>_<type>_<deviceID>_<comment>where <mode> is the polarization (’TE’ or ’TM’), <wvl> is the operatingwavelength, <type> is the class of the device (e.g. RingResonator, or YieldRing),<deviceID> is a unique device name, and <comment> is an optional field forcomments (e.g. resonator radius or gap).Figure 3.8 shows a GDS example of a label for a ring resonator for TE polarizedlight and a wavelength of 1550 nm belonging to the class YieldRing. The exactsame device is in the GDS more than 500 times and so the class YieldRing can be56Figure 3.8: Each input grating coupler has a unique text label. This label isused to generate a device list for automated measurement.used to select only a sub-group of devices or to run post processing scripts on asubset of measurements, e.g. to extract manufacturing variability [158]. From thecomment field we know that the ring has a radius of 12 µm, a gap of 200 nm, and acoupling length of 4.5 µm.3.3 Hardware assembliesThe probe station hardware includes a custom chip stage, fiber array stage, fluidicstage, and flow cell, as shown in Fig. 3.9.Figure 3.9: Silicon photonic probe station hardware overview. (a) CAD ren-dering showing the fluidic stage, chip, and fiber stage assemblies. (b)Photograph of the probe station used for biosensing, including flow celland reagent sequencing trays.573.3.1 Chip stage assemblyFigure 3.10 shows the chip stage which consists of two motorized, VT-50 linearstages with active feedback (Micronix, Irvine, CA). They control X and Y move-ments relative to the fiber array.Figure 3.10: Motorized XY chip stage assembly with rotation, tip-tilt, andPeltier element to thermally tune the chip during measurements. Theenlarged inset shows a mounted Teflon flow cell for biosensing with asmall portion of the chip (in blue) exposed for coupling light on andoff chip using a fiber array.The chip stage is thermally tuned using a 30 x 30 mm Peltier element (LairdTechnologies, Earth City, MO) mounted between a large piece of copper that acts asa heat sink and the chip mount block. A thermistor (Omega Engineering, Stamford,CT), potted into the chip block, is wired to the temperature controller. The coppersink mounts on a tip-tilt stage (AIS-50BUU, OptoSigma, Santa Ana, CA) androtation stage (KSP-406M, OptoSigma, Santa Ana, CA). Custom adapter platesconnect these parts to the linear stages.3.3.2 Fiber array stage assemblyA fiber array (PLC Connections, Columbus, OH) mounts on a 3D printed holderpositioned at a nominal insertion angle of 20o. The holder connects to a goniome-ter (GOH-25A50, OptoSigma, Santa Ana, CA) to allows fine adjustments to theinsertion angle (see Fig. 3.11). A second goniometer (GOH-25A50, OptoSigma,Santa Ana, CA) provides yaw alignment of the fiber array to the chip stage. Fi-58nally, a rotation stage (KSP-256, OptoSigma, Santa Ana, CA) provides adjustmentsabout the arm’s axes (roll). The arm assembly connects to a linear stage (VT-50,Micronix, Irvine, CA) to provide Z-axes motions that raise and lower the fiber arrayto the chip.Figure 3.11: Motorized Z stage and fiber array arm supporting manual align-ment on three axes (yaw, pitch, and roll). The enlarged inset shows thefiber array holder and alignment stages. The design ensures that pointsof rotation happen at the tip of the fiber array.3.3.3 Flow cell and channel gasketsThe chip block has tapered pins (McMaster-Carr, Santa Fe Springs, CA) to align theflow cell and threaded holes which secure it tightly over the gasket (Fig. 3.12). Thesmall, exposed portion of the chip allows fiber array access to the on-chip gratingcouplers. Fluidic gaskets are made from 250 or 500 µm-thick silicone sheets (GraceBioLabs, Bend, OR). They define two 250 µm-wide by 6000 µm long channels ona 750 µm center which are fabricated using a 10 µm laser (Universal Laser SystemsInc., Scottsdale, AZ). The gasket has openings which mate to ports on the flow celland define fluidic channels over the on-chip sensors.3.3.4 Reagent sequencing stage assemblyTwo linear stages (XSlide, Velmex Inc, Bloomfield, NY) provide reagent sequencingduring wet assays by moving two well plates in any combination of 96, 24, or 659Figure 3.12: Four port Teflon flow cell that mates to a two-channel siliconegasket to define fluid lanes over the sensors. A recessed cavity in thechip stagemechanically aligns chips to the silicone gasket. Pins pressedinto the chip stage facilitate the proper alignment of the flow cell andgasket. The enlarged inset shows a photograph of biosensors exposedin the two fluid lanes. While the alignment features facilitate a timelysetup, the chip must be designed to ensure the biosensors reside in thechannels.wells (Fig. 3.13). A user-defined recipe file specifies the well number sequence anddwell time within each well during assays. An arm with metal sleeves to guide theflexible Tygon tubing (Vici Valco Instruments Co. Inc., Houston, TX) into the wellhas sharpened tips that punch holes in cover film if installed on the reagent trays.Figure 3.13: CAD renderings of the reagent sequencing stage comprised oftwo motorized X and Z axes. The stage supports two well trays (6, 24,or 96 plates). The enlarged inset shows the metal guide sleeves thatensure the tubing aligns with each well.603.3.5 External instrumentsThe open-source software controls external, bench-top components to test devices.A laser mainframe, set of power meters, temperature controller, pump, and align-ment cameras were selected for their performance, utility, and driver availability.Currently supported components include: Agilent 81682A, 81600B, and 81689Alaser (within the 8164Amainframe), Santur TL-2000 laser module, ILX LightwaveLDM-4980 laser diode mount (and SRS LDC501), Agilent N7744A and 81635Apower meters (the later within an 8164A or 8163A mainframe), a Newport 3040TEC and Stanford Research Systems LDC501 to thermally tune the stage, and aChemyx Nexus 3000 syringe pump. Three 3.2MP Lumenera USB cameras (Ot-tawa, Ontario) mount on Navitar’s Zoom6000 high magnification modular zoomlens systems to facilitate alignment (fiber array and chip alignment). All of thecomponents are connected via USB (either directly, through a GPIB-to-USB, orserial-to-USB dongle) to a generic PC with Windows 10 (Microsoft, Redmond,WA) and MATLAB R2013a (Mathworks, Natik, MA).3.4 Fiber array alignmentThe on-chip grating couplers and the fiber array must be aligned to optimally couplelight. Figure 3.14 shows the degrees of freedom available to the user for adjustingthe fiber array. As stated previously, the fiber array is attached to two goniometers(α′, γ′), a rotation stage (β′), and a linear stage (z′). The two translation axes (x ′,y′)must align to the chip stage (x,y). Depending on how the fiber array is mounted, theincident fiber angleΘ and misalignment anglesΩ and δ (see Fig. 3.15 for definitionof δ) are a function of α′, γ′, β′. Every time a different fiber array is mounted themisalignment angles Ω and δ need to be minimized. We accomplish this using twoNavitar’s Zoom6000 high magnification modular zoom lens systems.For a fiber array with a large number of fibers, it is important that the fiberarray plane (plane through fiber axis) is aligned with the grating couplers. Figure3.15 shows schematically the effect of a misalignment. The fine align algorithmuses GC1 to optimize the return power (input on GC2). Because of a rotationalmisalignment δ the second output grating coupler (GC3) has a decreased powerreading.61Figure 3.14: Degrees of Freedom of stage: tip and tilt (β, γ,), rotation (α),and translation (x and y, locked z); Degrees of Freedom fiber array:tip and tilt (α′, γ′), rotation (β′), and translation (z′, locked x ′,y′).Figure 3.15: Unbalanced power reading due to misalignment (δ), the finealign function uses only one detector (i.e. GC1, user specified) andoptimizes the output power as function of stage position leading to amisalignment of GC3.3.4.1 Mapping grating couplersTo find first light, the fiber array must be moved within the vicinity of the deviceand a 2D raster performed. The mapping involves stage moves along one axis,e.g. y-axis (line scan) while power values are recorded. The stage is incrementallymoved along the other axis (x-axis) and then another line scan is performed. Oncecomplete, a 2D ’heat map’ is displayed as shown in Fig. 3.16. The locationof reflected light (grating coupler pairs) appears in red while the background (noreflected power) is blue. The user then clicks a red spot on the image to move thefiber array to that location. The next step involves a fine-alignment at that location62to optimize the optical transmission.Figure 3.16: Mapping of grating coupler pairs: An area of typical 400 µm x600 µm is scanned and a ’heat map’ is generated. Red blobs indicatethat there is power returned and hence that input and output gratingcouplers are aligned with the fiber array.3.4.2 Fine alignThe fine align algorithm performs small stage movements to align the on-chipgratings to the fiber array. The algorithm involves three phases. When started, thesoftware queries the power meter and if the reading falls below a user specifiedthreshold (i.e. −30 dBm), then the algorithm advances to a search-phase calledphase i) in Fig. 3.17(b). The stage is moved in an outward-growing rectangularpattern comparing the queried power to the threshold. The size of the spiral andthe step resolution can be specified in the fine align settings.If the queried power never exceeds the threshold during phase i) within thespecified search window, the algorithm switches to an alternate detector and startsonce again. If the queried power never exceeds the threshold using the alternatedetector, the algorithm quits and alerts the user of the failed fine align. The primaryand alternate power meter can be specified in the fine align settings.When the queried power exceeds the threshold, the algorithm enters the ’hillclimb’ (phase ii) in Fig. 3.17(b)). This phase implements a gradient method which63Figure 3.17: a) Power returned as function of position of fiber array relativeto the grating coupler pair. The power returned is in the units of dBmb) The full fine align method consists of three phases: i) spiral method;ii) gradient method; iii) crosshair methodcompares power values from the neighboring points to determine the direction ofhigher power readings (see Eq. 3.1).. . . pn,m−1 . . .pn−1,m pn,m pn+1,m. . . pn,m+1 . . .(3.1)The software moves the stages along the gradient acquiring power readings andcompletes once it reaches the maximum in both X and Y, as defined by:pn−1,m, pn+1,m < pn,mpn,m−1, pn,m+1 < pn,m.(3.2)where pn,m is the power at position xn,m, yn,m. A point-by-point line scan witha 1 µm step is performed and the stage is moved to the position with the highestreading. This is done along the x-axis and the y-axis and at the end of the algorithm,the stage is moved to the X and Y position with the highest reading.643.5 Coordinate systemIn order to automatically measure many devices simultaneously, the software needsto translate GDS coordinates into motor positions. To accomplish this, the usermust configure a coordinate system. This involves moving to three known devices,performing a fine align, and setting the stage position to the device’s GDS coordi-nates. When the user selects the device from the list, the software populates theGDS coordinates in the coordinate system table. The user then ’sets’ the stage po-sition for that device. The software then populates the stage position for that devicein the table. With three devices specified, the software automatically computes atransformation matrix between GDS coordinates and motor position and outputsthe residual to console window.This section describes how the software computes the transformation. As-suming an affine transformation between the GDS plane and the motor planef : R2 → R2, it assumes the form ~x 7→ T~x +~b, where T is a linear transformationincluding: scaling, rotation, and shearing, and ~b is the pure translation. Therefore,the mapping definition is:xmotorymotor =sx cosα −sy sinαsx sinα sy cosα︸                        ︷︷                        ︸scaling and rotation1 σ10 1︸     ︷︷     ︸shear x 1 0σ2 1︸     ︷︷     ︸shear yxgdsygds +b1b2 , (3.3)where sx and sy are the scaling factors in x and y, respectively, α is the rotationangle between the two planes, and σ1 and σ2 are the shear factors in x and y,respectively. The scaling operation accounts for different units in the two spaces([µm] to [mm]) and any optical encoder errors in the linear stage while the rotationcorrects the rotational misalignment of the two planes. The shearing transformationcompensates for angle errors between the X and Y stages (x-axis and y-axis). Wetypically observe worst case angle errors of about 1 degree compensated by thetransformation.Since the motor position is affected by measurement errors, an exact solu-tion for the transformation parameters cannot be found. Instead, the functionminp‖ f ( ~pgds, ~pm)‖22 =minp(f1( ~pgds,1, ~pm,1)2+ · · ·+ fn( ~pgds,n, ~pm,n)2)is optimizedusing a least squares fit, where pm = [xmotor, ymotor ] and pgds = [xgds, ygds]whichare the motor position and GDS coordinates respectively. Then f ( ~pgds, ~pm) be-65comes:~f (x) =T11 T12 0 0 · · ·T21 T22 0 00 0 T11 T120 0 T21 T22.... . .xgds,1ygds,1xgds,2ygds,2...+b1b2b1b2...−xmotor,1ymotor,1xmotor,2ymotor,2.... (3.4)At least three (n = 3) motor position and coordinate pairs ([xgds,nygds,n] 7→[xmotor,nymotor,n]) are needed for a robust geometrical solution. While any devicecan be used to configure the coordinate system, it is often easiest to use the looped-back GC alignment structures provided in the GDS framework. Since they residein the chip corners (Fig. 3.18), they are easy to find and provide distant, distinctcoordinate pairs to minimize errors in the transformation.Figure 3.18: Alignment structures to facilitate the timely setup of a coordinatesystem are placed in corners of the photonic chip as shown in the GDS2above.3.6 Software applicationThe software application facilitates sequence to setup and characterize silicon pho-tonic devices and orchestrates assays. The graphical user interface (GUI) has eightpanels which are hidden or made visible based on controls required at that stepin the process. Figure 3.19 shows the test panel but also highlights the partitionswithin the entire window. A detail description can be found in Appendix B66Figure 3.19: A screen clip of the application window with an active testpanel is shown. a) Sequence tabs that highlight the active step inthe overall setup and testing process. b) Window with informationalmessages provided to the user. c) Test panel showing the acquiredspectra for four channels (left), the windowed peak that is tracked(center), and the resulting sensogram with wavelength shifts from thetracked peak (center) over the course of the assay. d) Assay controlwindow highlighting the active recipe step and device under test. Thispanel also has the testing control and settings buttons. e) ’Previous’and ’Next’ buttons that sequence the user through the various panelsenumerated in a).3.6.1 Automated testing and assay orchestrationUsing the coordinate system described above, the probe station automatically testsmany devices in a single setting. Figure 3.20 shows the flow diagram how thesoftware executes the automated measurements. A transformation matrix convertsthe GDS coordinates into motor positions. A fine align executes at each new deviceand then a wavelength sweep is performed. The software saves the results and67proceeds to the next device. This process repeats until all the devices have beentested.Figure 3.20: Flow diagram of automated measurement: from the device listthe GDS coordinates are read and translated into a motor position byusing the transformation matrix T. After fine align the measurement,i.e. wavelength sweep, is performed.Automated testing initially screens devices for yield under dry conditions. Theones that work are then subjected to an aqueous environment and characterized.The software uses a ’recipe’ file specified by the user to orchestrate the sequence.The user sets the reagent order, dwell time in each well, and pump flow rate ina text file we call a ’recipe’ file. This simple text file specifies comma-separatedparameters and must follow the format shown below:%<well>,<time(min)>,<reagent>,<ri>,<velocity>,<temp>,<comment>User configurable parameters include: the well number (# > 0), the dwell timein the well (# > 0), the reagent name, the reagent’s refractive index (if known), thepump flow rate in uL/min (> 0), the stage temperature in Celsius, and a commentstring if desired (optional). An example recipe file is included theGitHub repositorywith the source code. A ’0’ specified for flow rate disables the pump. And the timecan be translated into number of sweeps via a setting accessible from the test panel.If multiple devices are selected for test, the user can specify to loop through alldevices at each reagent step or sequence through all the reagents for each device.These features allow performance parameters of many devices to be experimentallyassessed quickly and efficiently. Figure 3.21 shows an example of the bulk sensi-tivity being observed for two devices simultaneously. For this case, the softwareacquired 20 wavelength sweeps per sensor at each refractive index solution step.The data in Fig. 3.21 was acquired in less than 10 min (including the analysis68time) plus 15 min of setup up time (including aligning of chip, setting up flow cell,finding first light, configuring the coordinate system, and launching the assay).Figure 3.21: Example of how the bulk refractive index sensitivity of twodifferent devices, such as a TE and TM mode ring, can be assessedsimultaneously.3.6.2 Analyzing dataA custom analysis tool compliments the probe station control software and providesmore sophisticated processing of acquired datasets. While the user can select andtrack peaks during an assay, circumstances arise when the peak may be lost. If thishappens, the only way to recover the sensogram is post-acquisition. For example,the accidental introduction of an air bubble may move resonant peak out of thetracking window or the slow (laser-dependent) scan rate may not be able to keepup with a quickly shifting peak. The analysis tool helps recover the tracked peakunder these circumstances.Figure 3.22 shows a screen clip of the analysis tool which imports an acquireddataset and facilitates post processing. The user can remove unwanted scans,select and track different peaks, curve fit peaks using a polynomial or Lorentzfunction, subtract reference channels, compare a functional sensor’s drift to thesystem temperature, correlate entire scans or the peak window with previous ones,measure y-offsets on the sensorgram, and export the analyzed data and figure.After processing, the user can save all the parameters and excluded scans to ananalysis file that can be reloaded when accessing the dataset in the future. The toolalso supports user-defined analysis scripts. Using the provided scripting template69Figure 3.22: Analysis tool UI. (a) Menus that provide access to openingdatasets, saving analysis parameters, and executing user-defined anal-ysis scripts on the processed data. (b) The scan panel shows the entirewavelength range and the selected peak for a single sweep. (c) Thepeak panel highlights the windowed peak and overlays a fitted result.(d) The peak tracking plot shows the sensogram of the entire assaywhich includes the tracked peak for every sweep.as a starting point, the user can gain access to the processed data in application’snamespace. This allows the user to quickly develop custom routines to furtheranalyze their data. Scripts can invoked from within the tool by simply adding themto the ’analysisScripts’ directory in the tool and selecting them from the ’AnalysisScripts’ menu shown in Fig. 3.22(a).3.7 Platform characterizationIn addition to testing devices, recipe files can be developed to characterize thesystem itself, including the bench top instruments. As an example, we report onnoise floor measurements as function of laser sweep speed and flow rate underaqueous conditions. We also describe how algorithm stability and robustness (e.g.fine align) were assessed using user-defined scripts.703.7.1 Noise floor as function of sweep settingsWhile sensor design impacts detection limits, it also depends on other noise sourcesin the signal path (i.e. external instrumentation such as the laser and power meter).We wanted to understand how their settings might impact device characterizationin regards to sweep speed and resolution.Two (extreme) cases were investigated: (1) maximum resolution (0.1 pm) andlong averaging at the slowest sweep speed (0.1 nm/s), and (2) moderate resolution(1 pm) and short averaging at the fastest sweep speed (40 nm/s). A custom high-QTE/TM mode disk resonator sensor designed by our group [159] and a referenceacetylene cell (Wavelength References, Corvallis, OR) were used to evaluate trade-offs among sweep speed and step size (resolution). In both cases, the stage wasthermally tuned to 30 oC. Table 3.1 lists the power meter connections.Table 3.1: Power meter connectionsChannel 1 Disk resonator TM mode resonant peakChannel 2 Disk resonator TE mode resonant peakChannel 3 Capped (no input) to measure detector noiseChannel 4 Acetylene cell for wavelength referenceMeasurements were performed under dry conditions without the gasket andflow cell. Sweeps were continuously acquired for 1 hour (around 120 scan lines). Aresonant minima (or null) for a high Q peak, low Q peak, and acetylene absorptionpeak were tracked for every scan and their distribution determined. In both cases(fast and slow sweep speed) the acetylene reference cell showed an rms noise belowthe resolution. Table 3.2 lists the computed rms noise and Q for different resonancemodes in the disk resonator. We observed a Q of 950k for the acetylene referencepeak with an rms jitter of less than 1 pm. The wavelength sweep resolution forthese measurements was set to 1 pm.Since the acetylene cell does not exhibit the same rms noise as the disk res-onator, we assume that the variation originates in the chip’s photonic signal path.Possible noise sources include the time-dependent alignment of I/O fibers and grat-ing couplers or slight temperature fluctuations in the photonic chip due to bulk airflow around the test setup. It appears both the slow and a fast sweep exhibit similar71Table 3.2: Resonance peak jitter at sweep speed of 0.1 nm/s (0.1 pm resolu-tion) and 40 nm/s (1 pm resolution)speed peak Channel 1 Channel 2nm/s # rms [pm] Q [k] rms [pm] Q [k]40 1 1.5 125 1.8 1542 1.4 60 1.5 1380.1 1 1.6 126 2.1 2552 1.5 6 2.2 215noise floors. This means that assays can use the higher sampling (number of sweepsper minute) without affecting the detection limit.3.7.2 Noise floor as function of flow ratesTypical flow rates range from 1-100 µL/min for most bio-assays. To investigatethe impact of flow velocities on rms noise, we performed repetitive sweeps andtracked resonant peaks while exposing sensors to different flow rates of ultra-pure,degassed water. We used the same disk resonator and tracked similar peaks as thenoise floor assessment previously described. Spectral responses were acquired at asweep speed of 40 nm/s and resolution of 1 pm. The stage was thermally tuned to30 oC.Figure 3.23 shows the observed rms noise for the various resonance peaks.Peak 1 has a quality factor of 18k and rms noise of 3 pm and appears to be slightlydecreasing with increasing flow rate. Peak 2 with a quality factor of 50k has an rmsnoise of 1 pm but seems to be independent of flow rate. The reported rms valuescorrespond to what was observed in air (see section 3.7.1). These results suggestthat flow rates ranging from 1-100 µL/min do not significantly impact the system’snoise floor.3.7.3 Stage stability and algorithm robustnessThe second example of system characterization scripts focuses on the fine alignalgorithm. In particular we wanted to assess: (1) grating coupler insertion loss asfunction of position in relation to the stage and (2) repeatability of the algorithm tooptimize the fiber array position over the on-chip grating couplers.72Figure 3.23: The wavelength variation is reported as function of flow rate.For peak 1 (Q = 18k) the rms is between 2 and 4 pm. For peak 2(Q = 50k) the rms is around 1 pmInsertion loss as function of stage positionThe fine align robustness was assessed using a TE mode alignment farm (as de-scribed in section 3.4.1). The power meters were configured for a dynamic rangeof -10 dBm to -40 dBm to avoid saturation and connected as follows: channel 1 toa TE mode looped-back grating coupler port and channel 4 to the acetylene cell forwavelength reference. The remaining channels were capped and unused. The fiberarray was centered on a grating coupler pair at a distance from the chip surface ofabout 10 µm. A volume of size 80 µm x 120 µm x 180 µm is scanned with a stepsize of 2 µm. At each point in space the power returned by the grating couplerpair is recorded (average of 3 readings). Tests were performed in a dry, climatecontrolled laboratory setting (normal conditions).Figure 3.24 shows four slices of a scanned, 3D volume. The red areas indicatelow insertion loss and blue means little or no power was returned. As the distanceof the fiber array to the chip surface increases, the red area moves along the x-axissince the grating couplers used here were designed for an incident angle of 20o.The axis through the points with lowest insertion loss for each slice has an angleof 20o with respect to the x-axis. The horizontal shift (with increasing distance)is therefore expected and confirmed. The top view of the first slice indicates an73Figure 3.24: Avolume of size 80 µmx 120 µmx 180 µm is scannedwith a stepsize of 2 µm. At each point in space the power returned by the gratingcoupler pair is recorded (average of 3 readings). The acceptance angleof a grating coupler is about 20 µm in diameter.acceptance area with a radius of about 20 µm. This means that for a fast mappingalgorithm a step size of 5 µm will be sufficient.Stability of fine align algorithmThe fine align repeatability was assessed by doing 600 consecutive sweeps (10 pmstep at 40 nm/s) while performing a fine align between each sweep. The insertionloss of grating coupler pair was extracted for each scan.Figure 3.25(a) shows a scatter plot of recorded motor positions. The histogramsvisualize the distribution along the x-axis (σ = 2 µm) and y-axis (σ = 2.4 µm)respectively. The maximum power of each transmission is recorded in Fig. 3.25(b).The outliers (N = 25 with IL > 24 dBm) correspond to a motor position outside thecircle with radius of r = 4 µm and signify an unsuccessful fine align sequence. Withremoved outliers the standard deviation of the insertion loss is σadj = 0.49 dBm.There are cases when the algorithm partially fails. At this point, it is unclear whythe algorithm fails for a small number of fine align tries. We hypothesize it is dueto the PID settings in the controller itself and have engaged the manufacture to helpresolve the issue.74Figure 3.25: Fine align stability: a) recorded motor position after each finealign cycle. The histogram visualizes the distribution along x-axis(σ = 2 µm) and y-axis (σ = 2.4 µm); and b) recorded insertion loss ofgrating coupler pair.3.8 Summary and future workThis chapter described the development methodology and custom test platformcreated to spur the rapid and efficient development of silicon photonic biosensors.A PDK with proven I/O cells, a chip framework, and process test structures wasdeveloped and characterized for the UWWNF. An aluminum housing with customadapter plates and anchors was constructed to test photonic chips in both dry andwetenvironments. For assays (wet environments), custom laser cut gasket, Teflon flowcell, and pump was used to sequence refractive solutions and biological reagents.An overview of the design philosophy and implementation of the control soft-ware was discussed highlighting key features and extensions. Likewise, the analysistool used to evaluate bio assay data and characterization data of the platform itselfwas presented. Machine drawings for the hardware setup as well as the source codefor the control and analysis tools have been open sourced on GitHub in an effort tosupport rapid and cost efficient development of silicon photonic devices across thebroader photonic community.Future work involves implementing the real time air bubble detection feature.Limiting air bubbles, stage drift, and thermal instability is essential for successfulbio assays. The feature will be designed so that the user can enable real-time signalprocessing of the scanline data. Frame-to-frame comparisons will be made and75when an adequate correlation fails (outside a user specified range), the applicationpauses the assay and alert’s the user, allowing early intervention.76Chapter 4SOI Ring Resonator Biosensors4.1 IntroductionAs described in chapter 2 the figures of merit to describe a sensor performanceinclude sensitivity, S, and the limit of detection, or better the intrinsic limit ofdetection. The sensitivity is determined by the overlap of the electric field with theanalyte and can be improved by increasing that overlap. Genalyte, a commerciallyavailable silicon photonic biosensing platform, utilizes ring resonators designed forTE polarized light with a bulk sensitivity of 54 nm/RIU [45, 160]. In this chaptera variation of sensor designs based on ring resonators are discussed. Section 4.2introduces some of the design aspects of ring resonators. The section on cascadedring resonators (section 4.3) describes an attempt to multiplex the read out to enableparallel sensing of a panel of proteins. The rest of the chapter deals with differentstrategies to improve the sensitivity of ring resonator designs, including TM ringresonators in section 4.4, thin TE ring resonators 4.4.2, disk resonators 4.6. Themain focus of this chapter is the use of sub wavelength gratings described in section4.7.The methods and materials used to design, fabricate, test, characterize, andanalyze the performance of the devices described here are discussed in chapter 3.Deviations from protocols and procedures described in that chapter are noted in thesections for each sensor.774.2 SOI microring resonators - theory and designRing resonators have attracted interest from both academia and industry, and havefound widespread application as biological sensors [12], modulators, and filters[161, 162]. A ring resonator consists of an optical waveguide looped back onto itselftogether with a straight waveguide (directional coupler) to facilitate the couplingof the light to the looped waveguide as shown in Fig. 4.1; the resonance condition(constructive interference) is met when the optical length of the resonator is equala multiple of the wavelengths:λres =Lroundtrip · ne f f (λ)m,m = 1,2,3, ... (4.1)where λres is the sensor’s resonant wavelength, L the roundtrip length, and mis the longitudinal mode order of the resonant mode. For a large index contrastFigure 4.1: Schematic of a ring resonator: The light is coupled from theinput waveguide into a waveguide looped back onto itself. A resonatoracts as a wavelength filter and the transmission spectrum (light intensitymeasured at the through port) shows resonant nulls, where the spacingbetween the nulls is described as free spectral range (FSR). The nullitself is described in terms of full-width-half-maximum (FWHM) andextinction ratio (ER).the waveguide dispersion cannot be neglected and therefore the effective refractiveindex is written as function of wavelength, ne f f (λ). Due to this integer-dependentcondition (m), multiple resonances are supported, where the spacing between theseresonances, the free spectral range (FSR), depends on the optical length of theresonator [31].Eq 4.1 assumes that the effective refractive index is the same in the coupling78Figure 4.2: a) cross section of a directional coupler with design parametersw, the waveguide width and g, the waveguide gap. The thickness isgiven by the choice of SOI wafer (t = 220 nm). b) Schematic of aring resonator with coupling coefficient κ1,2(λ), transmission coefficientt1,2(λ), coupling length L1,2, radius r , and power attenuation in the ringdue to propagation loss, a. The subscript 1 and 2 refer to the couplingregion and the * denotes the complex conjugate of κ and t respectively.region and the bend waveguide segments. For the numerical model used forsimulating the transmission spectrum, the extended resonance condition below willbe used.δ(λ,T ) = 2piL1,2 · ne f f ,straight (λ,T )+2piR · ne f f ,bend (λ,T )λ= 2pim (4.2)where δ(λ,T ) is the phase shift after a round trip as a function of temperature Tand wavelength λ, ne f f ,straight and ne f f ,bend the effective refractive index in thestraight and the bend region, respectively.When ring resonators achieve critical coupling between the resonant cavityand waveguide bus, a high extinction ratio results. Critical coupling describes thecondition when the power loss in the ring is equal to the power coupled to thering, and can be achieved by optimizing the coupling gap for point-coupled ringresonators [31] or coupling length of a racetrack resonator [163]. The critical79coupling condition for a ring resonator as depicted in Fig. 4.2 is reached when [31]:|k1 |2 = 1− |at2 |2 (4.3)The amplitude transmission a is related to the power attenuation coefficient α[1/cm] by a2 = e−αL . The power attenuation coefficient α is sometimes also givenin [dB/cm]. The two are related by:α[1/m] = 100α[dB/cm]10 · log10(e1)(4.4)The coupling coefficient κ1 (in Eq 4.3), is defined as the cross-coupling coefficient(the subscript refers to the individual coupling region in a double bus ring), andt2 is defined as the transmission coefficient (or self-coupling coefficient). If thereis no loss in the coupling region then t2 + κ2 = 1 and therefore the t2 and κ2 canbe seen as the power splitting ratios. For accurate simulations of resonance width,the losses in the coupling region needs to be included in the roundtrip loss (ortransmission) coefficient a. The critical coupling condition (Eq. 4.3) states that thepower coupled into the ring is equal the power loss inside the ring. If the attenuationis negligible (a ≈ 1), critical coupling occurs for κ1 = κ2, which is referred to assymmetric coupling. Not discussed here is the fact that a faction of light enteringat the input port is reflected back due to the ring waveguide. It is easier to design aracetrack resonator as the coupled mode theory can be used, whereas point couplersrequire fully-vectorial 3D modeling.4.2.1 Coupling regionCoupling from one waveguide to a second waveguide in proximity can be deter-mined using coupledmode theory and supermode analysis (or eigenmode expansionanalsyis) [73, 164, 165]. The two eigenmodes of the coupled parallel waveguides(symmetric and anti-symmetric supermodes) have slightly different effective refrac-tive indices. That difference in propagation constant causes the modes to interfereconstructively and destructively in the waveguides. For long parallel waveguidesthe power oscillates from the main waveguide to the coupled waveguide and back.Assuming that all the power is in the main waveguide at z = 0. The fraction of80Figure 4.3: Mode profile of supermodes in directional coupler with geometrywaveguidewidthw = 500 nm, waveguide thickness t = 220 nm, couplinggap gap = 200 nm. a) symmetric mode profile (n1 = 2.336) and; b) anti-symmetric mode profile (n2 = 2.323).power in the second waveguide is then given by|κ |2 = P2(z)P0= sin2 (C · z) = sin2(pi · z∆nλ0)(4.5)where P0 is the power at z = 0,C is the coupling coefficient (not to be confused withthe field coupling coefficient κ), and ∆n is the effective refractive index differencesymmetric and anti-symmetric super mode. The coupling length, Lc, is definedas the distance after which all the power is transferred to the coupled waveguide.The coupling length is dependent on the wavelength λ and the distance between thewaveguides, g, and is given by:Lc (λ,g) =λ2 ·∆n(λ,g) =pi2 ·C (4.6)The mode profiles of the symmetric and anti-symmetric supermodes are shown inFig. 4.3 for a coupling gap g = 200 nm, and waveguide dimensions of w = 500 nmand h = 220 nm. The effective index of the symmetric super-mode is n1 = 2.336 andthe for the anti-symmetric mode it is n2 = 2.324 and the resulting index difference81is ∆n = 0.012. For a wavelength λ = 1.55 µm the coupling length is therefore Lc =62.8 µm. In practice, the parallel waveguides in the coupling region are connectedto waveguide bends which also contribute to the coupling between waveguides. Inthe extreme case of Lc = 0, light is still coupled from one waveguide to the other.This configuration is also referred to as point coupler. To include the contributionof the bend region Eq. 4.5 is extended and can now be fitted to measurement data toextract the coupling length [96] and the extra effective coupler length zbend. UsingEq. 4.6 the field coupling coefficient is then given by:|κ |2 = P2(z)P0= sin2(pi2Lx· [z+ zbend])(4.7)In a similar way the transmission can be understood as the power remaining inthe original waveguide and can be written as:|t |2 = P2(z)P0= cos2(pi2Lx· [z+ zbend])(4.8)This set of equation assumes a lossless coupler satisfying κ2+ t2 = 1.4.2.2 Spectral responseThe analytical model for a double-bus ring resonator is given by [166]:EthroughEin=t1− t∗2aeiδ1− t∗1t∗2aeiδ(4.9)where the subscripts on the values for field transmission refer to the through portcoupler, t1 and the add/drop port coupler t2. To design ring resonators at criticalcoupling one can tune the coupling gap and/or the coupling length (assuming thatthe waveguide thickness and width are fixed). Here we chose to fix the gap atg = 200 nm and tune the coupling length to achieve critical coupling.MODE Solutions was used to compute the effective indices (super modes,straight and bend waveguides). The results fromMODE and experimental propaga-tion losswas imported intoMATLAB to find κ fromEq. 4.5 so that |κ1 |2 = 1− |at2 |2.The script employed analytic functions previously presented by our group [167] to82determine these values. For a ring resonator with radius R = 30 µm the length ofthe coupling region should be Lc = 8.57 µm for a critically coupled double busring with g1 = 200 nm and g2 = 300 nm. Fig. 4.4 shows a simulated transmissionspectrum compared to an actual measurement (fabricated through EBeam lithogra-phy). The insertion loss of the grating coupler and propagation loss of the routingwaveguides is adjusted manually to have similar baselines. The effective indexrequired slight adjustment (to match the resonance condition) to compensate forfabrication variations or different temperature conditions.Figure 4.4: Transmission spectrum of a ring resonator (waveguide width w =500 nm, waveguide thickness t = 220 nm, coupling gap gap = 200 nm)resulting in a free spectral range of FSR = 2.6 nmThe free spactral range is given by (within first order approximation of disper-sion) [31]:FSR =λ2ngL(4.10)where ng is the group index (as defined in Eq 2.6) and L the round trip length.In an effort to accurately simulate the waveguide geometry and account forany discrepancies between fabrication and design, Scanning Electron Microscope(SEM) images were taken of the design. In order to image the waveguide cross-section and estimate the sidewall slope of the waveguides, a Focused Ion Beam(FIB) was used to mill trenches over a straight waveguide as well as two couplerregions (one of 200 nm waveguide spacing and the other of 400 nm waveguidespacing). After FIB milling, the waveguides were imaged with a 52o tilt to observetheir cross-sections. The pixel lengths of the waveguides were then compared with83Figure 4.5: SEM image of a) cross-section of coupling region with gap =150 nm and b) ring resonator with radius R = 30 µm.the pixel length of the scale bar to determine the widths of the top and bottom of thewaveguides; the coupler gaps were determined with a similar method. The averageof five measurements was used as the waveguide dimensions for simulation. Onlyone measurement was obtained for each of the coupler gaps. It was determinedthat the 220 nm (as verified by profilometry) silicon layer was approximately 470(±17) nm wide at the top edge and 575 (±11) nm wide at the SiO2 interface,indicating a sidewall angle of approximately 14o (±2o). The coupler gaps at theSiO2 interface were determined to be approximately 113 nm for the 200 nm couplergap design and 327 nm for the 400 nm coupler gap design. The coupler geometrieswere imported into Lumericals MODE Solutions to simulate the mode profiles andcoupling coefficients (see Fig. 4.3) .4.3 Cascaded ring resonatorsRing resonators, especially for TE polarized light, are among the most popularintegrated optical sensors [45]. In biological assays with statistical randomnessof biological interactions and the presence of temperature fluctuations, one singlesensor is not enough to get quantifiable measurement. Typically, an assay is runwith one or more control channels. In addition for more sophisticated analysisapplications probing for one single anti-body is not enough [168]. Therefore itis necessary to monolithically integrate multiple sensors to enable multiplexeddetection. One way of doing this is by using a 1xN optical power splitter to84create N individual channels. Another approach is using one single waveguidewith cascaded rings with slightly different geometries (e.g radius) to make surethe resonance condition for the individual rings are different. Figure 4.6 shows aFigure 4.6: Schematic of three cascaded ring resonators with different radii.For a white spectrum at the input, each of the ring acts as filter fordifferent wavelengths. The transmission spectrum is a superposition ofthe transmission spectrum of the three individual rings.schematic representation of three ring system. Each of the rings acts as a filter fordifferent wavelength on the white input spectrum. This is a common approach indatacomm referred to as wavelength division multiplexing. The number of channelis limited by the free spectral range of the resonators.4.3.1 Methods and materialsFabrication of SOI ring resonatorsThe SOI photonic chip was fabricated through IMEC as described in section 3.2.2.The buried oxide layer is 2 µm thick and the crystalline top silicon film is 220 nmthick. The straight waveguides have a width of w = 500 nm. The design includesseveral configurations with 1-4 cascaded rings or racetracks with different radii85Figure 4.7: A PDMS flow cell is used to expose the sensors to differentanalytesand different coupling gaps. DW-2000 mask layout design software from DesignWorkshop technologies, Quebec, Canada was used to design the resonator masklayout.Fabrication of PDMS microfluidic deviceA poly-dimethylsiloxane (PDMS) flow cell with two fluidic channel is used toexpose the individual sensors with different analytes (see Fig. 4.7). PDMS isthe most widely used polymer in the field of microfluidics. It is a biocompatible,transparent, rubber-like polymer and can be easily patterned using soft lithography,a well established fabrication method. The mold masters were fabricated withstandard photolithography techniques on SU-8 2075 (MicroChem, USA). Throughreplica molding the patterns were transferred onto PDMS. The uncured PDMS(Sylgard 184, Dow Corning USA) was poured onto the mold master to a thicknessof about 1 cm, degassed to remove air bubbles, and cured at 80 oC for 2 h on ahotplate. The hydrophobic surface properties can be altered in an oxygen plasmaby replacing some of the surface methyl groups (CH3) by a hydroxyl groups (-OH).This activated surface can form covalent siloxane bonds (Si-O-Si) when in contactwith glass or silicon substrates, forming an irreversible seal. After punching accessholes for the fluidic inlet and outlets the PDMS layer and the resonator substratewere aligned and bonded to each other.86The measurement setupA tunable laser source (Agilent 81681A, Agilent Technologies, Inc., USA)was usedwith an output wavelength range from 1460 nm to 1580 nm. A single mode opticalfiber was aligned with the I/O grating coupler to inject light into the waveguides.The output light intensitywas collectedwith amultimode optical fiber andmeasuredwith an optical power sensor (Agilent 81635A, Agilent Technologies, Inc., USA).The temperature of the substrate was kept constant at 30±0.001 oC with a Peltierelement and a temperature controller (Stanford Research System, USA) in a closedfeedback loop configuration.4.3.2 Spectal responseThe use of cascaded ring resonators offers the possibility to measure transmissionspectra of multiple rings in different channels simultaneouslyFigure 4.8: a) Transmission spectra for a 2 ring system. Ring 1 is exposedto DI water and Ring 2 is exposed to different water/glycerin mixtures(with 0 wt% being pure water). b) Transmission spectra for a 4 ringsystem. Ring 1,2, and 4 are exposed to DI water. Ring 3 is exposed totwo different water/glycerin mixtures.Figure 4.8(a) and 4.8b show the through port transmission spectra and the centerwavelength changes ∆λres for different racetrack resonators as a function of thewater/glycerin concentration. Quality factors of > 40k are measured for certaindesigns. Figure 4.8(a) shows the measured transmission spectrum of two cascadedresonators. Both resonators are exposed individually to different environments.The first resonator (Ring 1) with an FSR of 2.9± 0.02 nm is immersed into DI87water. The second resonators (Ring 2) with and an FSR of 3.45± 0.014 nm isexposed to five different water/glycerin concentrations of 0, 5, 10, 15, 20 wt%(see Table 4.1 for the corresponding refractive indices). The refractive index inchannel 1 does not change and therefore there is no resonance peak shift occurring.For Ring 2 the resonance peak shifts linearly with glycerin/water concentration orlinearly with increasing refractive index. Similarly Fig. 4.8(b) shows the measuredtransmission spectrum of a four resonator system. Again, each resonator can beexposed individually to different environments. Here Ring 1, 2 and 4 are keptconstant (exposed to DI water) and Ring 3 is immersed into water and a 10wt%glycerin/water solution respectively.Table 4.1: Refractive index of Glycerin/Water solutions at 20oCGlycerin % by weight Refractive index n0 1.333035 1.3388010 1.3448115 1.3510620 1.357494.3.3 SensitivityFor the two resonator system the net shift in resonance peak as a function of refrac-tive index is depicted in Fig. 4.9. The resonance peak for pure DI water for Ring 2 isat λres,0 = 1541.35 nm ±0.008 nm. For a refractive index change of ∆n = 0.02446which corresponds to a switch from DI water to a 20wt% glycerin/water solutionthe resonance peak shifts to λres,20 = 1542.24 nm ±0.01 nm. This corresponds toa sensitivity of 36 nm/RIU. The volume refractive index sensitivity shows a linearrelationship. The linear curve fit parameters are a = 36.6 and b = 1492.5 (for leastsquare fit to y = ax + b). The obtained relationship is compared to the predictionsfrom the finite element model. The ’x’ shaped data set in Fig. 4.9 represents themodeled values. Despite the fact that the model overestimates the sensitivity, themeasured results are sufficiently close to the predicted values, indicating that thismodel can be used for optimizing the design of the resonator geometries to furtherenhance quality factors and sensitivity. The mode simulations for a TE mode (see88Figure 4.9: Resonance wavelength shift as a function of the refractive indexof the binary mixture (glycerin/water). The measured data is comparedto the predicted values from the finite element modelsection 2.1) show that only a small fraction of the field intensity of the optical modepropagates outside the waveguide and can therefore interact with the sample. Thefinite element model computed using Lumerical MODE Solutions and a customMATLAB script can easily be adapted to simulate slotted waveguides as well.Other important parameters of the system not quantified in this work are the sensorresolution, i.e. the smallest possible spectral shift that can be accurately measured,and the detection limit, the minimum amount of sample analyte (or RIU change)that the sensor can quantify. They are dependent on the system’s spectral resolutionand noise factors. In this work resonance peaks are detected by simply trackingthe position of the minimum values in the transmission spectra thus making it verysensitive to noise. The measurement is done at steady state multiple times and theerror bars in Fig. 4.9 indicate the variation of the resonance peak. The qualityfactors of the rings are in the order of 40k. The simulated surface sensitivity for aprotein adlayer with nad = 1.48 was 160 pm/nm.4.3.4 Biological assayThe resonance peak shift is then monitored as a function of time (see Fig. 4.10a)while functionalizing the racetrack with the protein Human Factor IX. The initial89Figure 4.10: a) Resonance peak shift during functionalizationwith the proteinHuman Factor IX of one of the ring resonators (control channel isplotted in blue). The difference in peak location (∆λP) before and after(both in PBS) is a measurement of the mass change during deposition.b) Resonance peak shift during an antibody binding reaction (Anti-Human Factor IX). The difference in peak location (∆λAB) before andafter (both in PBS) is a measurement of themass change during bindinglarger shift is caused by a combined effect of a bulk refractive index change ofthe protein solution and the deposition of a protein layer on top of the waveguide.Switched back to PBS with the initial bulk refractive index, ∆λP is then a mea-surement of the mass change on the waveguides surface. Similarly, Fig. 4.10(b)shows changes in peak location during the antibody binding reaction (150 µg/mLin Phosphate Buffered Saline, PBS).4.4 TM mode ring resonators4.4.1 DesignWaveguides guiding TM mode light offer three times the sensitivity of that of theTE mode at 1.55 µm in 220 nm thick SOI waveguides [74] due to their largerevanescent field component traveling above the waveguide. Figure 4.11a shows theelectric field intensity of the fundamental TMmode traveling in a 220 nm x 500 nmwaveguide. TM modes experience less scattering loss because the mode is guidedabove the waveguide in the cladding and below in the substrate. Therefore edgeroughness in the sidewalls (resulting from fabrication) is avoided. In addition, themode overlap with the cladding results in higher susceptibility to refractive index90changes in the cladding. Because of these advantages, TM mode ring resonatorsoffer unique properties which are advantages for biosensing applications.Figure 4.11: TM mode ring resonator simulation results. (a) Electric fieldintensity of a TM mode in a 220 nm X 500 nm silicon waveguide. (b)Simulated sensitivity versus waveguide width for TM modesFigure 4.11(b) shows the simulated sensitivities of a TMmode in a 220 nm thicksilicon waveguide as the width varies. Lumerical MODE solutions was also usedto calculate bending and mode-mismatch loss. Propagation losses were estimatedbased on our previous experimental results [96]. These values were used in ananalytical MATLAB model to optimize the racetrack resonator coupling lengthto achieve critical coupling. The devices were fabricated and characterized asdescribed in Section 3.2.4.4.2 ResultsTM mode ring resonators with waveguide thicknesses of 150 nm and 220 nm werefabricated using 193 nm deep-UV lithography and dry etching at IMEC throughePIXfab multi-project wafer service. Figure 4.12(a) shows the microscope imageof the fabricated, 40 µm radius TM racetrack ring resonators (along with other TEmode rings and disk resonators). The sensors were characterized using methodsdescribed in section 3.2. The resulting spectrawhen subjected to the refractive indexsolutions or characterization are shown in Fig. 4.12(b). Plotting these resonantwavelength shifts against the measured refractive index of the solution set is thesensitivity shown in Fig. 4.12(c).The sensitivity of the 220 nm thick waveguide ring was measured to be147 nm/RIU with a quality factor of 9,200 and iLoD of 7.09 · 10−4 RIU (Fig.914.12(c)). The sensitivity of the 150 nm thick waveguide ring was measured tobe 238 nm/RIU with a quality factor of 1,914 and iLoD of 3.28 · 10−3 RIU (Fig.4.12(c)). The thinner, 150 nm waveguide, confines less of the TMmode and resultsin a greater evanescent field in the cladding and higher sensitivity. The 150 nmthick waveguide ring’s quality factor could be increased as simulations show thatthe ring may be over coupled. As expected, TM modes (in general) provide highersensitivities than their TE mode ring counterparts (40 nm/RIU). Furthermore, thesimulated surface sensitivity for a protein adlayer with nad = 1.48 was 312 pm/nmfor the 220 nm thick waveguide and 243 pm/nm for the 150 nm thick waveguide.Figure 4.12: Experimental results for the TM mode ring resonator. (a) Mi-croscope image of the TE and TM mode ring resonator set. (b) Trans-mission spectra showing the resonant peak shifts when subjected toa NaCl RI solution set. (c) Sensitivity response measured for the RIsolution set of (b) resulting in a sensitivity of 238 nm/RIU. (d) Biosens-ing demonstration using a model biological system; Regions B, C, D,and E are described in Section 3.2.4; air-bubble during the rinse cyclebetween regions B and C (indicated by the red-dashed, vertical line).A bio assay (as described in Section 3.2.4) was performed to assess the TM92mode ring resonator’s biosensing performance. The results are shown in Fig.4.12(d). Protein A was passively adsorbed to the chip off-line creating a thin filmon which to immobilize the capture antibody, antiSA. The unintended introductionof an air bubble mid-way through the antiSA rinse cycle (Region B) is indicatedby the red-dashed vertical line although it does not impact the viability of thebiological species. The slight negative shift in resonant wavelength after the BSAblock and challenge (Region C) suggests that a small portion of the antibody adlayerlifts off but the original sensor coverage (Region B) was robust. The binding ofSA (Region D) and subsequent amplification using bBSA (Region E) result inpermanent resonant shifts as expected.4.5 Thin TE mode ring resonator4.5.1 DesignUsing thinner waveguides leads to lower confinement of the propagating TE modeand results in deeper penetration of the evanescent field into the cladding, offeringenhanced sensitivity and more field overlap with biomolecules on the waveguide’ssurface. Based on these advantages, we examined the various thicknesses availablethrough MPW foundries (90, 150, 220 nm) for biosensing applications [163].Figure 4.13 illustrates the mode profile of the propagating TE mode in waveg-uides for a range of silicon thicknesses, namely 90, 150, and 220 nm (a-c). Theevanescent field in the cladding medium decays exponentially as a function of dis-tance from the silicon core. The point where the evanescent field electrical fielddecays to 1/e of its initial value is called the penetration depth (de) as defined inchapter 2 in Eq. 2.8. Various effective refractive indices for various waveguidethicknesses and widths were calculated using Lumerical MODE solutions. Thesevalues were used to calculate the penetration depths of the electric field which areplotted in Fig. 4.13(f). The sensitivities of these resonators were calculated basedon Eq. 2.17, and Lumerical MODE solutions was used to calculate δne f f /δncladfor each case of waveguide thicknesses and widths. The resulting sensitivities areplotted in Fig. 4.13(d). Figure 4.13(e) shows simulated results for temperaturesensitivity for the various waveguide dimensions.93Figure 4.13: TEmode, thin ring resonator simulation results. (a) Electric fieldintensity of a TE mode for 90 nm thick silicon core (b) Electric fieldintensity of a TE mode for 150 nm thick silicon core (c) Electric fieldintensity of a TE mode for 220 nm thick silicon core (d) Sensitivityof TE waveguides for three different waveguide thickness, 90 nm,150 nm, and 220 nm, (e) Temperature sensitivity for three differentwaveguide cross sections; (f) Simulated 1/e electric field penetrationdepth of the evanescent field into the solution for various waveguidecore thicknesses and widths.To validate these calculations, we designed, fabricated, and compared racetrackswith 90 nm, as well as the standard 220 nm, SOI (using same methods explained inchapter 3). The devices were fabricated and characterized as described in section3.2.944.5.2 CharacterizationUnlike the other sensors, the thin TE ringswere subjected to different concentrationsof D-Glucose (D16-500, Fisher Chemicals, Fisher Scientific, Inc.). The titrationsrange from 0 M to 110 mM. The refractive index of such solutions is estimatedusing nglucose (λ) = nH2O + 1.515 · 10−6cglucose [169]. Sensors were exposed tothe solutions at a flow rate between 5-15 µL/min to measure quality factors (Q), thebulk RI sensitivity (S), and the detection limit (iLoD).4.5.3 ResultsTE mode ring resonators with waveguide thicknesses of 90, 150, and 220 nmwere initially realized on a 220 nm SOI substrate using an ebeam lithographymulti-etch process (described in Section 3.2) and then fabricated using 248 nmdeep-UV lithography and dry etching at IME through a multi-project wafer runorganized by CMC-SiEPIC. Figure 4.14(a) shows the SEM image of our ultra-thin racetrack resonator (fabricated using ebeam). Each ring was characterizedusing methods described in Section 3.2. The slope of the dashed and red lines inFig. 4.14(b)-(d) show the sensor’s response when subjected to different refractiveindex solutions. Sensors with a waveguide thickness of 220 nm, 150 nm, and90 nm results in sensitivities of 38 nm/RIU (ebeam), 83 nm/RIU (foundry), and153 nm/RIU (foundry), respectively. The 90 nm thick waveguide’s quality factor inwater was measured at 9,200 yielding an iLoD of 1.11 ·10−3 RIU, while the 150 nmthick waveguide sensor had a quality factor of 14,000 and iLoD of 1.33 ·10−3 RIU,and the 220 nm thick waveguide had a quality factor of 15,000 and iLoD of2.72 ·10−3 RIU. Furthermore, the simulated surface sensitivity for a protein adlayerwith nad = 1.48 was 160 pm/nm, 212 pm/nm, and 298 pm/nm for a waveguidewidth of 500 nm and with thicknesses of 220 nm, 150 nm, and 90 nm, respectively.These results show that the most commonly used waveguide thickness of 220 nmfor datacomm applications is not optimized for sensing applications. In addition, acarefully design waveguide for biosensing would also involve finding the optimalwidth. See Figs. 2.13 and 2.14 for a theoretical analysis of the sensitivity as functionof the waveguide dimensions.95Figure 4.14: Experimental results for the thin waveguide TE mode ring res-onators. (a) SEM image of 90 nm thick (10 µm radius) and 220 nmthick silicon (30 µm radius) TE mode ring resonators fabricated usingebeam lithography [163].(b) Measured sensitivity (38 nm/RIU) of a220 nm thick TE sensor (ebeam chip) [163] (c) Measured sensitivity(83 nm/RIU) of a 150 nm thin waveguide TE sensor (foundry chip). (d)Measured sensitivity (153 nm/RIU) of a 90 nm ultra-thin waveguideTE sensor (foundry chip).4.6 Micro-disk resonator biosensorsMicro-disk resonators share a similar design theory and approach to strip waveguidering resonators but utilize a circular plate (or disk) for the resonant cavity insteadof a single-mode waveguide (Fig. 4.15(a)). The most significant difference fromring resonators is that the optical mode travels along the disk’s perimeter and isin contact with only one side-wall. Because of this, disk resonators offer uniqueadvantages for biosensing, namely higher quality factors, differential sensing (twodifferent probing modes), and the potential for multiplexing because of their smallfootprints. Scattering losses are reduced considerably for disks when compared to96rings, as disks possess only a single side-wall. This not only improves the qualityfactor [170, 171] and the detection limit but allows for smaller footprints (radii of1.5-3 µm) [159]. A benefit of small footprints is that the biosensors can get closerto the size scale of the biological molecules and cells they interrogate than othersilicon photonic biosensors. This can lead to lower mass-limits of detection giventhat each bound biomolecule will have a higher effect on the field distribution [40].Figure 4.15: (a) 3D schematic of a disk biosensor on a SOI substrate. (b)Simulated mode profiles (electric field intensity) for a 10 µm disk.(i)-(iii) Show electric field profiles for the first three TE modes and(iv)-(vi) show the first three TM modesDepending on a waveguide’s geometry, TE and TM modes can be guidedsimultaneously [96]. Both TE and TMmodes can also be excited in disk resonators;however, because bending losses are higher for TM modes, larger disk radii arerequired for high-Q TM resonance peaks (simulated TM mode bending loss arereported to be around 4 dB/cm [172]). The mode’s evanescent field overlap withthe cladding determines the sensitivity. It is evident from the mode profiles in Fig.4.15(b) that the TMmode electric field extends further into the cladding over similarTE modes. The distance at which the electric field intensity (|E |2) has decayed to1/e with respect to intensity at the waveguide’s surface was estimated to be 60 nmfor TE modes and 110 nm for TM modes, demonstrating the significant sensingadvantage of TM modes for molecular adlayers or large biomolecules [32].4.6.1 DesignWe used a method previously demonstrated by our group [32] to calculate theexpected quality factors and coupling coefficients between the waveguide and res-onator sensor. To determine optimal coupling, a 400 nm wide bus waveguide and a97wide, bent waveguide to represent the disk (5 µm for 10 µm disks; 1.5 µm for 3 µmdisks) was simulated using the FDTD method. Combining scattering loss values(10 dB/cm for the 3 µm disk and 1 dB/cm for the 10 µm disk which were obtainedexperimentally) and losses to water absorption, the optimal coupling distance wascalculated to be 200 nm. Because TMmodes are less confinedwithin the waveguidethan their TE counterparts, a larger radius (10 µm) was required, resulting in a diskthat supported TE and TM modes simultaneously. Both 3 µm and 10 µm radiusdisk resonators were fabricated and characterized as described in Section 3.2.4.6.2 ResultsBoth 3 µm and 10 µm radius disk resonators were realized on a CMOS-compatible220 nm SOI substrate using ebeam lithography, characterized, and subjected toa model biological system to assess their biosensing capability using approachesdescribed in Section 3.2.The 3 µm radius disk supports TE modes only and results in a large FSR, whichis advantageous for multiplexing many sensors on a single waveguide bus [173].To evaluate multiplexing, we cascaded six disks on a single waveguide bus varyingeach disk’s radius by 10 nm to avoid overlapping FSRs and ensure their resonantpeaks occupy different locations within the spectra (Fig. 4.16(a).The fundamental TE mode (TE0) of the 3 µm disk had an observed qualityfactor of 33,000 and its resonant peak shifted when subjected to the refractive indexsolution standards (Fig. 4.16(b). We measured a sensitivity of 26 nm/RIU (Fig.4.16(c) fromwhich we calculated an iLoD of 1.81 ·10−3 RIU. Themeasured qualityfactor of the TE1mode of the 3 µmdisk was 22,000, with a sensitivity of 29 nm/RIU(Fig. 4.16(c) and iLoD of 2.43 · 10−3 RIU. The free-spectral range (FSR) range is35-40 nm depending on the mode. Furthermore, the simulated surface sensitivityfor a protein adlayer with nad = 1.48was 289 pm/nm, 75 pm/nm for the 10 µm diskfor the fundamental TM and TE mode, respectively.The sandwich assay described in Section 3.2.4 was used to assess the disk’sbiosensing performance and is shown in Fig. 4.16(d) for the TE1mode. Simulationssuggest that a 1 nm adlayer with a coverage of 51.9% of the sensor’s surface wouldcause a shift of 30 pm (as we experimentally observed in Fig. 4.16(d). A 3 nm98Figure 4.16: Experimental results for the 3 µm disk resonator. (a) Falsecolor SEM image of six, cascaded resonators on a single waveguidebus. (b) Transmission spectra showing the resonant peak shifts whensubjected to a NaCl solution set. (c) Sensitivity response measuredfor the first two, supported TE modes (26 nm/RIU and 29 nm/RIUrespectively). (d) Biosensing demonstration using a model biologicalsystem as described in Section 3.2.4.adlayer would only need to cover 17.1% of the surface to cause a similar wavelengthshift (thicknesses of protein layers are difficult to measure). Assuming a monolayerof each protein, the relative resonancewavelength shifts for each bimolecular adlayercorresponds well to the expected shift based on their molecular weights [174, 175].Experimental results for the 10 µm disks are shown in Fig. 4.17. The 10 µmdisks fundamental TM yielded a sensitivity of 142 nm/RIU (Fig. 4.17(a)), qualityfactor of 16,000, and iLoD of 6.82 ·10−4 RIU. The TMmode had a FSR of 9.66 nm.The fundamental TE mode in the 10 µm disk yielded a sensitivity of 21 nm/RIU(Fig. 4.17(a), quality factor of 131,000, and iLoD of 5.63 ·10−4 RIU.Furthermore, the simulated surface sensitivity for a protein adlayer with nad =99Figure 4.17: Experimental results for our 10 µm radius disk resonator. (b)Sensitivity response measured for the RI solution set of the first TMand TE modes (142 nm/RIU and 21 nm/RIU respectively). (d) Bothmode responses to a biosensing demonstration using amodel biologicalsystem as described in Section 3.2.4.1.48 was 289 pm/nm, 75 pm/nm for the 10 µm disk for the fundamental TM andTE mode, respectively. The biosensing capability of the 10 µm disk is shown inFig. 4.17(b). Simulations indicate that a 1 nm adlayer with 41.4% coverage of thesensor’s surface would resulting in the (observed) wavelength shift of 98.5 pm forthe TM mode in Region A. Similarly an adlayer of 3 nm thickness and coverageof 13.7% would have the same effect. Our simulated and experimental results arewithin the surface coverage ranges observed by Coen et al. for Protein A adsorptionto a silicon dioxide surface. [156]The TE mode quality factor in the 10 µm disk is much improved over that inthe 3 µm disk due to the larger radii which provides reduced sidewall scattering andbending losses. Also, the larger evanescent field overlap with the cladding of theTM mode (Figs. 4.15(iv-vi)) should permit the detection of larger biomolecules(over their TE mode counterparts (Figs. 4.15(i-iii)). This differential detectioncapability may provide additional sensing information on the target molecule’s size,distinguishing whole cell and protein-binding to the sensor surface. However, themulti-mode nature (TE and TMmodes of multiple orders) of the 10 µmdisks makestracking resonant peaks difficult. Suppressing unwanted modes would improve theease-of-use in tracking resonant peak shifts.1004.7 Subwavelength grating ring resonatorsIn this section, we report on the biosensing performance of a sub-wavelength grat-ing (SWG) ring resonator, resulting in a 2X sensitivity improvement over the bestTM ring / slot rings. Sub-wavelength grating waveguides allow the designer to en-gineer the effective index of the guiding structure to minimize loss, enhance guidingcapabilities, and more importantly improve the field overlap with biomolecules onthe waveguide’s surface. Sub-wavelength gratings in SOI waveguides have recentlybeen proposed by the National Research Council of Canada (NRC) [176–178]. Andwhile sub-wavelength gratings have been demonstrated experimentally in applica-tions like fiber-to-chip couplers to minimize mode mismatch loss [157, 179–181],meta material lenses [182], waveguide crossings [176], and filtering applications[183], they have yet to be experimentally demonstrated for biosensing applications[184, 185]. Wangüemert-Pérez et al. proposed the use of SWG waveguides forbiosensing [185] and employed a Fourier-type 2D vectorial simulation tool to an-alyze the effect of various duty cycles on the sensing performance. They alsoperformed a full 3D FDTD simulation to determined the theoretical sensitivity butdo not report any experimental results. Similarly, our group has also demonstratedthe fabrication andmeasurement of ring resonators and investigated their theoreticalsensing performance [186]. In this work, we show that we can further improve theperformance of TE mode ring resonator and present experimental results that showour biosensors achieve ≈ 10X enhanced sensitivity over their slab-based TE modewaveguide counterparts. This achievement expands their use in applications thatrequire greater sensitivities and detection limits that what can be achieved today.Theory of operationElectromagnetic wave propagation in periodicmedia can be described by the Bloch-Floquet formalism [73, 187, 188]. Depending on wavelength, propagation can bedivided into three wavelength zones for a fixed period Λ of the grating [188]: 1)the sub-wavelength zone in which the wavelength to period ratio is λΛ> 2 · ne f f .This corresponds to the wavelength range longer than the Bragg wavelength and thewaveguide behaves like a conventional waveguide. The periodic structure supportsa true lossless mode in this case [189]; 2) The wavelength range corresponding101to the photonic bandgap where Bragg reflections occur; and 3) the wavelengthrange shorter than the Bragg wavelength where the Bloch wave becomes leakyand part of the energy is radiated out of the waveguide and the propagation lossis determined by reflection and diffraction at the segment boundaries due to thehigh index contrast [190]. By having Λ λ the mode is without loss because thereflection and diffraction effects are suppressed. This is analogous to the electrondistribution in periodic potentials, like in semiconducting materials.Figure 4.18: Schematic of SWG waveguide: w is the waveguide width and tthe thickness; Λ is the SWG period and the length of the Si blocks isdetermined by the duty cycle η.For a photonic circuit designer, SWG waveguides are attractive because theyallow tailored propagation properties (namely the mode shape and dispersion) byvarying the duty cycle (η), period (Λ), waveguide width (w) and thickness (t). Thewaveguide is divided into small, slab segments (blocks of Si with refractive indexncore) with length Λη, where Λ is the period and η is the duty cycle. The small,slab-segment cross-section is comparable to traditional waveguides. Figure 4.18shows the schematic of a SWG waveguide. The substrate material with refractiveindex, nsub, and thickness, tsub = 2 µm, is SiO2 and as cladding material, nclad weused water since most biological applications require an aqueous solution.Here we extend the theoretical analysis of SWG waveguides as biosensors andconfirm the sensitivity of SWG ring resonators experimentally using a refractiveindex solution set. Like in the case of a uniform strip waveguide, the light is102confined in the xy-plane by the index contrast (Eigenmodes). The periodicity in thez-direction (n2(z) = n2(z+Λ)) guarantees that the wave vector kz is still conserved.According to the Bloch (or Floquet) theorem an electromagnetic solutions takes theform:E = EK (x, y, z)e−iKz (4.11)where K is the Bloch wavenumber and EK (x, y, z) is a periodic function with periodΛ so that EK (x, y, z) = EK (x, y, z+Λ). Similar to the dispersion relation for regularwaveguides, the dispersion relation for SWG waveguides is ω = ω(K ). The Blochwave vectorK can either be real or complex depending on the spectral regime. WhenK is real, the intensities of the Bloch wave will be a periodic function of positionin the medium and propagate without loss. For a layered structure with uniformmaterial properties in the xy-plane, analytical solutions exists [73, 188], but not forthe case of index guided modes (vertical and lateral confinement) and numericaltools should be utilized. The effective index for Bloch modes is ne f f ,B = cωK (ω)and the group index is ng = c ∂K∂ω [191, 192].4.7.1 SimulationFor a SWG waveguide (one dimensional photonic crystal) an analytical solutiondoes not exist and numerical methods have to be used. The most rigorous approachis a full 3D vectorial FDTD approach. However for large structures this approachis computationally very demanding and therefore not suitable for large parametersweeps. Figure 4.19 shows the field magnitude along the propagation axis (z-axis) for a SWG waveguide with width w = 500 nm, thickness t = 220 nm, periodΛ = 250 nm, and duty cycle η = 0.7. The data is reported in the xz-plane defined bya cut at y = t/2. Figures 4.19(b) and 4.19(c) show themagnitude of the z-component(Ez (x, t/2, z)) and x-component (Ex (x, t/2, z)) of the electrical field, respectively.Figures 4.19(d) and 4.19(e) show the mode profile in the cross section (xy-plane) ofthe SWGwaveguide at the center of the silicon segment and in the gap respectively.The 3D simulation was performed with Lumerical’s FDTD Solutions. A modesource (fundamental TE mode, λ = 1.55 µm) is used to inject light into the SWGwaveguide. The field is strongly confined in between the silicon segments, similarto what is observed in slot waveguides [193].103Figure 4.19: a) Electric field magnitude distribution in the xz-plane definedby a cut at y = t/2 for a SWG waveguide with dimensions of w =500 nm, Λ = 250 nm, η = 0.7, and t = 220 nm; b) Distribution of thez-component (Ez); c) Distribution of the x-component (Ex); d) Cross-section in the middle of Si block, and e) Cross-section in the middle ofthe gap.Analysis using equivalent effective refractive indexThe equivalent effective index method is an efficient way to analyze the SWGwaveguide. If Λ < λB2ne f f is true, a SWG waveguide behaves like a continuouswaveguide but with a decreased refractive index. As a consequence, the lightpropagation can be described like in an index-guided structure. Weissman et al.showed that the ’averaged’ refractive index step is primarily dependent on the dutycycle and can be express as: [194]):∆n′ = η∆n (4.12)where η is the duty cycle and ∆n is the refractive index modulation of the SWGwaveguide (∆n = ncore − nclad) [189, 195] (similar findings by Li et al. [196] andBierlein et al.This reduced refractive index step results in a decreased equivalent refractiveindex of the waveguide and a weaker confinement of the guided mode. Therefractive index of the continuous equivalent waveguide is given by: ncore,eq =nclad+η∆n. This method has been used by our group to design the building blocksof photonic circuits such as SWG waveguides, strip to SWG mode converters,directional couplers, and ring resonators [87, 186, 197]. It needs to be noted thatthis approach leads to a very narrowband approximation, especially for wavelength104close to the Bragg condition.FDTD analsysisIf the SWG waveguide supports multiple Bloch modes or is designed close tothe photonic band gap, the equivalent model (described above) is no longer valid[177, 198, 199]. In this case, a fully vectorial 3D FDTD analysis can be donebut is computationally taxing, especially for geometry sweeps. The mesh size hasFigure 4.20: a) Effective and group index of SWG waveguide with geometrycalculated with a 3D FDTD simulation with Bloch boundaries: w =500 nm, Λ = 250 nm, and t = 220 nm; The blue dash-dotted linesindicate the substrate (nsub) and the Brillouin zone (nbz) limits. Fora duty cycle η = 0.7 the equivalent effective index approximation isshown; b) zoomed in to show the refractive index range 1.4 to 2.4(range of effective index).to be small enough to accurately resolve the periodic structure and the simulationtime has to be long enough for the electromagnetic fields to fully decay. Here wemake use of the periodicity, by simulating only one unit cell with Bloch boundaryconditions. This approach is borrowed from band structure calculations of photoniccrystals [192, 200] and we have used this method previously to determine couplingcoefficients in SOI Bragg gratings [201]. Briefly, randomly distributed dipolesources are used to excite all Bloch modes in the SWG waveguide. The resonantBloch modes are determined with a Fourier Transform of the recorded time signalsfor each kz-vector (Bloch boundaries are only used in the propagation direction z).The Brillouin zone for this 1-D period structure corresponds to − piΛ< kz < piΛ . Withthis method, the group index of the fundamental Bloch mode is calculated using105Figure 4.21: Effective index calculated with a 3D FDTD simulation withBloch boundaries of SWG waveguide with geometry w = 500 nm andt = 220 nm for grating periods Λ = 250 nm (black), Λ = 300 nm (red),and Λ = 350 nm (green); The blue dash-dotted lines indicate the sub-strate (nsub) and the Brillouin zone (nbz) limits. If not indicated theduty cycle of the grating is η = 0.5.the first order approximation as used for strip waveguides: ng = ne f f ,B − λ ∂ne f f ,B∂λ .Figure 4.20 shows the effective refractive and group index for different duty cyclesand a waveguide geometry w = 500 nm,Λ = 250 nm, and t = 220 nm. The effectiveindex is limited by the photonic band gap (Bragg reflection) and the index ofthe substrate. At 1550 nm the effective index is ne f f = 1.56, ne f f = 1.67, andne f f = 1.71 for duty cycles of η = 0.5, η = 0.6, and η = 0.7 respectively. The groupindices for these same duty cycles were calculated to be ng = 2.41, ng = 2.85, andng = 3.31 respectively.As can be seen (from Fig. 4.20), the group index dramatically increases as theeffective index approaches the band gap of the periodic structure. As the wavelengthis far away from the band gap the equivalent effective index model can be used (inFig. 4.20(b) the equivalent effective index is shown for a duty cycle η = 0.7).Figure 4.21 depicts the effective index for different grating periods. The Braggcondition limit, λB2Λ , moves to longer wavelength as the period increases. Howeverat the wavelength of 1550 nm the effective index is not much affected. For E-Beam106lithography the minimum feature size is well below 75 nm (required for a periodΛ = 250 nm and duty cycle η = 0.7) and SWGs can be designed far away from thebandgap. However, for foundry processes, employing deep-UV lithography, theminimum feature size is around 130 nm and designs will be much closer to thebandgap where the equivalent refractive index model is not valid anymore.LossFor regular strip waveguides (w = 500 nm, and h = 220 nm) the propagation lossfor TE polarized light is around 4 dBm/cm with the main loss mechanism beingscattering due to sidewall roughness [127] (the SWG rings in thiswork are fabricatedon the same direct write electron beam lithography system as used by Bojko et al.[127]). The propagation loss is therefore dependent on the fabrication process.The loss values reported in [127] are for strip waveguides and for obvious reasonsnot directly applicable for SWG waveguides. A careful loss analysis (in water ascladding material) has yet to be carried out. For SWG waveguides the opticalmode is delocalized which minimizes the electric field interaction with the surfacescattering sources [190], but depending on thickness of the SiO2 layer can alsoincrease substrate leakage. Furthermore, the sidewall interaction is reduced by(1− η), due to the presence of gaps between segments. But in the same time onealso has to consider the roughness of the created internal sidewall. Bock et al.reports measured loss values of 2.5 dB/cm for a SWG waveguide with slightlynarrower waveguide [190]. Furthermore, a careful loss analysis (both theoreticaland experimental) should also include bending losses as function of radius.Sensitivity analysisAssuming the equivalent continuous waveguide model, one might expect an en-hanced sensitivity compared to a strip waveguide of the same cross-section. Sincethe ’average’ refractive index step is reduced, the mode confinement is weakerresulting in a decreased overall equivalent effective refractive index for the guidedmode. Weaker mode confinement enhances the modal overlap with the analyte(increased susceptibility), thereby increasing bulk sensitivity.107The bulk sensitivity of a ring resonator can be defined as [32]:Sb =∆λres∆nclad=λresng(∂ne f f∂nclad)(4.13)The waveguide susceptibility ∂ne f f∂nclad is simulated with the equivalent effectivewaveguide model (using MODE Solutions) and in 3D with the Bloch boundaryconditions (using FDTD anlysis). The results for duty cycle η = 0.6 and η = 0.7 areshown in figure 4.22 for a waveguide geometry of w = 500 nm, grating period Λ =250 nm, and waveguide thickness t = 220 nm. The equivalent effective waveguidemodel predicts as sensitivity of around 500 nm/RIU while the 3D FDTD modelwith Bloch boundaries shows a sensitivity between 400 nm/RIU and 480 nm/RIU,as shown in Figure 4.22.Figure 4.22: Comparison of MODE Solutions (equivalent effective refrac-tive index) and FDTD Solutions (Bloch mode approach) sensitivitysimulations for a waveguide geometry: w = 500 nm, grating periodΛ= 250 nm, waveguide thickness t = 220 nm, and duty cycle of η = 0.6and η = 0.7 respectively.In a similar way one can define the surface sensitivity as:Ss =∆λres∆tad=λresng(∂ne f f∂tad)(4.14)where tad is the uniform thickness of the adsorbed protein layerwith refractive indexnad. For most proteins the refractive index is around nad = 1.48 [101, 156]. Thethickness of the layer is determined by the characteristic length of the protein and the108orientation of the adsorbed protein. For globular shaped proteins the characteristiclength corresponds to the diameter of a sphere. The refractive index of this layer isalso dependent on the surface density of the protein. For a surface with all possiblesites occupied by a protein, the refractive index will be nad = 1.48. In reality, not allsurfaces are fully covered and so one can assume a constant refractive index of thelayer, nad = 1.48, with a changing effective thickness te f f < tad since the thicknesschanges linearly with surface concentration. This approach has been proposed andused in [102, 202]. The 3D FDTD simulation with Bloch boundary conditionsestimates the susceptibility ∂ne f f∂tad to around 1.45 µm−1 (w = 500 nm, grating periodΛ = 250 nm, waveguide thickness t = 220 nm, and duty cycle of η = 0.6) which isabout 3X compared to a regular quasi TE waveguide [160]. The surface sensitivityis then calculated to be 789 pm/nm.4.7.2 Experimental approachDesign and fabricationSWG photonic circuits were realized on a 220 nm thick SOI wafer (SOITec, Greno-ble, France) using a JEOL JBX-6300FS Direct Write E-Beam Lithography Systemat the University of Washington’s Nanofabrication Facility (WNF). The EBeamprovides a low-cost, fast turn-around fabrication process that has been optimizedto produce consistent, robust, passive silicon photonic components [127, 203]. ASWG ring resonator sub-circuit consists of strip-to-SWG converter [197], straight-to-bent SWG waveguide directional coupler, looped-back bent SWG waveguide tocreate the resonant structure. The coupling coefficients for the directional couplerwere estimated from [197] andmultiplied by a factor 2, to take into account the H2Ocladding instead of air (factor 2 based on strip waveguide simulations). SWG ringswith radius of 20 µm and 30 µm, duty cycles η = [0.5,0.6,0.7], and coupling gapsg = [200,250,300,350,400] nm were fabricated and characterized to determine theoptimal design parameters for biosensing applications. Figure 4.23 shows a SEMimage of a SWG ring resonator with the following design parameters: waveguidewidth w = 500 nm, grating period Λ = 250 nm, waveguide thickness t = 220 nm,and duty cycle η = 0.7.109Figure 4.23: SEM image of SWG ring resonator fabricated by Ebeam lithog-raphy. Waveguide geometry: w = 500 nm, grating period Λ = 250 nm,waveguide thickness t = 220 nm, and duty cycle η = 0.7.Microfluidcis and assay controlDevices were tested and characterized using a custom test setup and softwareapplication 3. Briefly, a custom fiber array from PLC Connections (Columbus,OH) with four polarization maintaining fibers is used to couple the light on and offthe chip through on-chip vertical gratings [157]. AnAgilent 81682A 1.5 µm tunablein a 8164A mainframe together with Agilent’s N7744A power meters is used as offchip laser source and detection equipment. The chip under test sits on an a thermallytuned aluminum chuck. The temperature is controlled with a Stanford ResearchLDC501 controller (Standford, CA). Two motorized stages (Micos, Germany) areused to align the chip to the fiber array. To sequence reagents from a 96-wellplate additional two motorized stages are used (Velmex XSlide, Bloomfield, NY).A silicon gasket with defined 500 µm wide channels is secured by a PTFE flowcell. The silicon gasket has a thickness of 500 µm. A syringe pump (ChemiyxNexus 3000, Houston, Tx) is used to control flow rates by drawing reagents fromthe wells over the sensors. Finally, a custom application written in MATLABprovides instrument control, orchestrates the acquisition sequence, and processeseach acquired data set for real-time resonant peak tracking during the assay.ReagentsThe SWG ring resonator’s performance, including quality factorQ, intrinsic limit ofdetection, iLoD, and bulk sensitivity Sb, were characterized in an aqueous environ-ment using NaCl refractive index standards ranging from 62.5 mM to 1 M. The RI110of each solution was measured using a Reichert AR200 digital refractometer (De-pew, NY). A standard sandwich assay involving well characterized reagents wereused to evaluate the sensor’s utility for biosensing applications [87]. Protein-A, a42kDa membranous protein observed to preferentially bind an immunoglobulin’sFc domain, was obtained from ThermoFisher (Chicago, IL). The immunoglobulin,anti-streptavidin (antiSA) (MW 150kDa), and its conjugate ligand, streptavidin(SA) (MW 57kDa), were obtained from Vector Labs (Burlingame, CA). For ampli-fication, biotin was conjugated to BSA (MW 66kDa) using a kit from Bangs Labs(San Diego, CA). A FisherScientific 1x PBS solution (Hampton, NH) was used todilute reagents and rinse unbound molecules after each step.4.7.3 Results and discussionOptical spectrumFigure 4.24 shows the measured transmission spectrum of a SWG ring with radiusR = 30 µm, waveguide width w = 500 nm, thickness t = 220 nm, SWG periodΛ = 250 nm, and duty cycle of η = 70 % for two different coupling gaps, 300 nmand 400 nm respectively. The linewidth of the resonator with a gap of 400 nmFigure 4.24: a) Transmission spectrum for a ringwith R= 30 µm,w = 500 nm,t = 220 nm, Λ = 250 nm, η = 70 %, and gaps of g = 300 nm andg = 400 nm respectively, exposed to DI water.is about 0.2 nm which corresponds to a quality factor Q of about 7 · 103. Thelinewidth is defined as the full width at half-maximum, FWHM. The group index,ng, of the waveguide can be extracted from the free spectral range (FSR) of a ring111resonator given by ng = λ2FSR ·L with L the ring resonator round trip length (L = 2Rpi).For the SWG ring resonator with radius R = 30 µm and duty cycle η = 0.7, theFSR = 3.936 nm and the corresponding group index ng = 3.27. For a ring withthe same radius but different duty cycle of η = 0.6, results in a FSR = 4.54 nm andgroup index of ng = 2.81. These observed results agree with the simulated (FDTD)values for ng = 3.31 and ng = 2.85, respectively.4.7.4 Bulk RI sensingFigure 4.25a shows the measured transmission spectrum of the SWG ring exposedto ultrapure water and the refractive index solution standards. The chip stage isthermally tuned to 25 oC to limit thermal drift. For each concentration the opticalspectrum was measured ten times to ensure repeatability and signal stability. Theresulting stepped, resonant wavelength shifts are shown in Figure 4.25b. TheFigure 4.25: a) Transmission peak for a ring with R = 30 µm, w = 500 nm,Λ = 250 nm, η = 70 %, t = 220 nm, and gap g = 400 nm exposed todifferent solutions of NaCl; b) reported peak wavelength shift.slope of the resonant wavelength shifts per refractive index standard is the bulksensitivity of the sensor, as shown in Figure 4.26). The SWG ring with duty cycleη = 0.6 yields a bulk sensitivity of Sb = 491 nm/RIU (see figure 4.26) and thesensor with a duty cycle of η = 0.7 yields Sb = 405 nm/RIU. Compared to regularTE ring resonators [12, 13] SWG ring resonators show an eight-fold increase insensitivity, and compared to TM [87] a two-fold increase. SWG ring resonatorsprovide close to twice the bulk sensitivity as slot resonators (Bragg and ring [193]).While the equivalent waveguide model tends to overestimate the bulk sensitivity(the simulated sensitivities are shown in figure 4.22), the measured sensitivities112Figure 4.26: Sensitvity results for a SWG ring with R = 30 µm, w = 500 nm,Λ = 250 nm, η = 60 %, and t = 220 nmcompare well to the simulated results.The measured Q factor is 7 · 103, and sensitivity is Sb = 405 nm/RIU, theintrinsic limit of detection is iLoD = ∆nmin = λQSb = 5.5 · 10−4 RIU as describedby Chrostowski et al. [32]. The theoretical limit for the intrinsic limit of detectionfor an ideal resonator sensor (in water) operating at λ = 1.55 µm is 2.4 · 10−4 RIU[32]. Themeasuredminimal detectable wavelength shift of the SWG resonator (andsystem) presented here is ∆λmin = 1pm, which translates into a system detectionlimit sLoD = ∆λminSb = 2.47 · 10−6 RIU for the resonator with a duty cycle η = 0.7)and sLoD = 2.01 ·10−6 RIU for the resonator with a duty cycle η = 0.6.BiosensingA sandwich assay involving well characterized biomolecules with high bindingaffinities were used to evaluate the biosensing performance of the SWG ring. Fig-ure 4.27 depicts the sequenced steps of the assay which include: (1) physisorbingan initial adlayer onto the waveguide surface, (2) immobilizing antibodies to theadlayer, (3) blocking any remaining exposed surfaces on the sensor to limit ’biolog-ical noise’ resulting from subsequent, non-specific interactions, (4) capturing thetarget antigen, and (5) using a secondary molecule (label) to amplify the capturedtarget.The experimental results in Fig. 4.28 show the resonant wavelength shifts overthe course of the assay and demonstrate the sensor’s ability to detect molecularbinding events in real time. Protein-A (1 mg/mL) was first passively adsorbed113Figure 4.27: Biosensing cartoon: Region A = anti-streptavidin (antiSA)(10 µg/mL), B = Bovine Serum Albumin (BAS) (20 µg/mL), C =streptavidin (SA) (20 µg/mL), and D = biotinylated-BSA (50 µg/mL).After each reagent the sensor is washed with PBS, indicated by thegrey area mid-way through each region.to the sensor’s native oxide surface by soaking the chip overnight. Protein A, a42kDa globular protein has a reported diameter of around 3 nm and refractive indexof n = 1.48 [101]. Coen et al. observed that the initial adsorbed Protein A maydenature resulting in a 1 nm thick add - layer but that subsequent adsorption ofprotein A forms a bioactive layer which provides the necessary binding pocket forthe antibody’s Fc domain [156, 204]. For the second and subsequent layer, wepresume that the protein retains its 3 nm globular shape due to the observed bioactivity, resulting in a 3-4 nm thick cladding layer (n = 1.48). We have shown inprevious publications that our experimental and simulated results are within therange observed and reported by Coen et al. [87, 156, 159]. After achieving abaseline in PBS buffer at 37oC, the biomolecules (as depicted in Fig. 4.27 step(A) through (D) were sequenced over the sensor. Each reagent was followed by a20 minute PBS buffer rinse to remove any unbound species on the sensor or in thechannel (as depicted by the grey area labeled ’PBS’).The capture antibody, anti-streptavidin (anitSA, 10 µg/mL), was introduced tofunctionalize the sensor for streptavidin detection. As can be seen in Fig. 4.28,Region A, the adsorption of this 160 kDa protein results in a 1.5 nm resonantwavelength shift. The linear increase suggests that the affinity reaction is transportlimited rather than reaction limited. While a fully saturated surface is desired, theautomated test setup used to sequence reagents and monitor the wavelength shiftsdid not permit extending the adsorption time during the assay. Regardless, thepost-acquisition simulations and experimental results observed in Fig. 4.28 suggest114ample surface coverage for subsequent binding events.To show that subsequent molecular binding interactions are specific, BSA(20 µg/mL) was introduced to block any remaining exposed sensor area, as shownin Region B. The slight wavelength shift offset at the end of the Region B rinse stepsuggests adsorption of BSA as a result of incomplete coverage of the Protein A -antiSA complex [156].Next, the sensor is subjected to 10 µg/mL of streptdavidin (SA) as shownin Region C. The permanent resonant shift observed during the PBS rinse cyclesuggests specific and irreversible binding of SA to antiSA. This step also indicatesthat the antiSA bound to the Protein A retains its bioactivity and specificity. Thefinal step (Region D) further demonstrates the biological specificity of the capturedSA while amplifying overall wavelength shift.Figure 4.28: Biosensing experimental results: Region A = anti-streptavidin(antiSA) (10 µg/mL), B =Bovine SerumAlbumin (BAS) (20 µg/mL), C= streptavidin (SA) (20 µg/mL), and D = biotinylated-BSA (50 µg/mL).After each reagent the sensor is washed with PBS, indicated by the greyarea mid-way through each region.Validation modelAlthough the relative resonancewavelength shifts for each new add layer correspondto refractive index changes based on the respective molecular weights of eachbiomolecule [101, 156, 204], the results from the FDTD simulations with varying115adlayer thicknesses with refractive index nad = 1.48 (to model protein layers) wereused to provide insight into the amount of adsorbed protein for the initial adlayers.From simulation we predict a surface sensitivity of 800 pm/nm (duty cycle ofη = 0.6, ng = 2.81, and λ = 1550 nm). This means that the resonant wavelengthshifts 800 pm for each nm of adsorbed protein layer onto the sensor’s surface. Sincefor protein adlayers the surface coverage (SC) is usually not 100% we assume aconstant refractive index of the layer, nad = 1.48, but a changing effective thicknessdad (SC). We further also use the assumption that this variation is linear [101].Assuming an antibody can be approximately modeled as a 5 nm diameter cylinderthat is 10 nm tall, a shift of 1 nm (Region A) suggests a 1.25 nm uniform adlayer,or approximately 12.5% surface coverage considering the antibody model. Thisresult agrees well with previously observed surface coverage values for antibodies[156, 159]. A shift of 500 pm (Region C) suggests a uniform adlayer of 0.6 nm forSA. Assuming SA’s mass is approximately 1/3 of antiSA and 5 nm in diameter, theobserved results suggest 13% surface coverage, which agreeswell with the observedantiSA coverage. The 700 pm shift in Region D suggests a uniform adlayer of0.87 nm on the sensor’s surface. Even though SA is tetravalent, it is unlikely allfour binding sites are occupied with bBSA. The observed wavelength shift (andresulting adlayer thicknesses) ratio of 1.4:1 for bBSA:SA is consistent with thishypothesis. Passivating the surface with BSA (Region B) provides confidence thatthese binding interactions are specific and not precluded by sterics. In additionto validating the bioactivity and specificity of the captured species, this sandwichassay illustrates the utility of SWG rings for multi-layer biological assays.4.8 SummaryIn summary, we proposed and implemented the use of cascaded ring resonatorstogether with a PDMSmicrofluidic network fabricated by soft lithography to exposeeach ring individually to different solutions. We have shown that each resonatorindividually exposed to different environments can be simultaneously measuredwith one single wavelength sweep. The volume refractive index sensitivity ofthe racetrack resonator is determined by injecting a water/glycerin mixture withdifferent mixing ratios and known refractive indices. This work also shows the116detection of a clinically-relevant binding reaction and highlights the fact that inorder to have a meaningful results a control channel is needed. Furthermore thesensitivity improvementwas demonstratedwith TM ring resonators. The sensitivityof the 220 nm thick waveguide ring was measured to be 238 nm/RIU with a qualityfactor of 9,200 and iLoD of 7.09 · 10−4 RIU. The sensitivity of the 150 nm thickwaveguide ring was measured to be 247 nm/RIU with a quality factor of 1,914and iLoD of 3.28 · 10−3 RIU. Sensors with a waveguide thickness of 220 nm,150 nm, and 90 nm were also demonstrated for TE polarized light. Sensitivities of38 nm/RIU (ebeam), 83 nm/RIU (foundry), and 153 nm/RIU (foundry), respectivelyare reported. The 90 nm thick waveguide’s quality factor in water was measuredat 9,200 yielding an iLoD of 1.11 · 10−3 RIU, while the 150 nm thick waveguidesensor had a quality factor of 14,000 and iLoD of 1.33 ·10−3 RIU, and the 220 nmthick waveguide had a quality factor of 15,000 and iLoD of 2.72 ·10−3 RIU.The fundamental TE mode (TE0) of the 3 µm disk had an observed qualityfactor of 33,000 and its resonant peak shifted when subjected to the refractiveindex solution standards. A sensitivity of 26 nm/RIU was measured from whichthe intrinsic limit of detection was calculated to be 1.81 · 10−3 RIU. The measuredquality factor of the TE1 mode of the 3 µm disk was 22,000, with a sensitivity of29 nm/RIU and iLoD of 2.43 ·10−3 RIU. The 10 µm disks fundamental TM yieldeda sensitivity of 142 nm/RIU, quality factor of 16,000, and iLoD of 6.82 ·10−4 RIU.The TM mode had a FSR of 9.66 nm. The fundamental TE mode in the 10 µmdisk yielded a sensitivity of 21 nm/RIU, quality factor of 131,000, and iLoD of5.63 · 10−4 RIU. The TE mode quality factor in the 10 µm disk is much improvedover that in the 3 µm disk due to the larger radii which provides reduced sidewallscattering and bending losses. Also, the larger evanescent field overlap with thecladding of the TM mode should permit the detection of larger biomolecules (overtheir TE mode counterparts. This differential detection capability may provideadditional sensing information on the target molecule’s size, distinguishing wholecell and protein-binding to the sensor surface. However, the multi-mode nature(TE and TM modes of multiple orders) of the 10 µm disks makes tracking resonantpeaks difficult. Suppressing unwanted modes would improve the ease-of-use intracking resonant peak shifts.In the last section the first experimental demonstration and the use of SWG117rings for biosensing is reported. The sensitivity increased 2X over the best TMring and slot waveguide resonators. The reported sensitivities of 400 - 500 nm/RIUexpand the use of these devices in sensing applications requiring sensitivity anddetection limits beyond today’s silicon photonic ring resonator systems. The SWGring resonators also show the highest surface sensitivity (789 pm/nm). Comparedto 312 pm/nm for the TM mode ring resonator the increase in surface sensitivitycan be explained by the 80% larger surface area. However, the surface area isonly one aspect and can be used to predict expected sensitivity increase. To fullyunderstand the increase, the electric field distribution has to be simulated as well(amplitude and overlap with adlayer). For example a regular TE mode waveguidehas the same surface area as the TM mode waveguide, yet the surface sensitivity isonly 160 pm/nm. This is because the electric field of a TE mode mostly overlapswith the sidewall of the waveguide.SWG rings clearly show great potential for use in biosensing applications, futurework will focus on extending the simulation methods discussed here to include TMpolarized resonators as well. Furthermore, to commercialize a biosensor usingtoday’s CMOS foundries, fabrication processes must achieve the minimum featuresizes required to fabricate these structures (75 nm). While these resolutions arenot achievable today, these SWG ring resonators could be designed closer to theirBragg condition (eg: with a larger period,Λ) at expense of increased loss and largergroup indexes.118Chapter 5Bragg Grating Resonators5.1 IntroductionAs described in chapter 2 the figure of merit to describe a sensor performanceinclude sensitivity, S, and the limit of detection, or better the intrinsic limit ofdetection. The sensitivity is determined by the overlap of the electric field with theanalyte and can be improved by increasing that overlap. Genalyte, a commerciallyavailable silicon photonic biosensing platform, utilizes ring resonators designed forTE polarized light with a bulk sensitivity of 54 nm/RIU [45, 160]. As demonstratedin chapter 4 there are different strategies to improve that sensitivity including opti-mizing the waveguide geometry as demonstrated by Talebi Fard et al. [163], slotwaveguides with sensitivities of 200− 300 nm/RIU [44, 51], by switching polar-ization from quasi-TE to quasi-TM resulting in sensitivities of around 200 nm/RIU[60, 160], or by using sub-wavelength gratings as described in section 4.7. Anothermeasure of performance is the detection limit which is defined as the minimumdetectable RI change. Genalyte reports a detection limit of 1 ng/mL, or 10−5 RIU,while Barrios et al. and Claes et al. measure the detection limits on the order of5 · 10−6 RIU for slot waveguides [44, 51]. It is known from Eq. 1.2 that both ahigh quality factor and a high sensitivity will improve the detection limit. In ringresonators the quality factor is primarily limited by water absorption losses but ifnot carefully designed bending loss, scattering loss from waveguide roughness, andmode mismatch loss can also become an issue. Especially bending loss can become119an issue for small radii and TM polarized light. Bending losses can be eliminated byusing resonant structures with only straight waveguide segments. Therefore Bragggratings have been investigated for use in biosensing applications as well [205–208].They offer similar sensitivities as ring resonators but with a much smaller footprint.Jugessur et al. fabricated vertical side-wall TE mode Bragg gratings integrated witha reversibly bonded PDMS fluidic network estimating a sensitivity of 5.5 ·10−3 nmin a sensing volume of 2 µm3 [206]. They designed slotted Bragg gratings as wellwith a simulated sensitivity of 620 nm/RIU. However they did not report any exper-imental results confirming their theoretical values. Pbabhathan et al. designed andsimulated similar vertical side wall TEmode gratings but introduced a period defectto create a resonance peak in the center of the photonic stop band [207]. Finally,Wang et al. fabricated a TE mode slot-waveguide Bragg grating for biosensingwith a sensitivity of 340 nm/RIU and quality factor of 3 ·104 [208].In this chapter the use of Bragg gratings for biosensing is discussed. In addition,ways of improving the intrinsic limit of detection of Bragg gratings are discussedand experimentally demonstrated. From the original design of TE Bragg gratings,results are shown of TE slot Bragg gratings, TM Bragg gratings, and grating designfor 1310 nm rather than 1550 nm. In addition Bragg gratings fabricated usingEbeam lithography (8 nm resolution [127]) are compared with foundry fabricatedsensors (193 nm lithography [133]).5.2 Theory and designTo avoid complicated fabrication steps, the desired index modulation is most com-monly achieved using sidewall corrugations [209]. The design parameters of Bragggratings are depicted in Fig. 5.1. The corrugation depth, d (the waveguide withvariation, or ∆w), and period (length of a single unit, Λ) modulate the effectiverefractive index.This index modulation in a waveguide gives rise to a stop band in the trans-mission spectrum, similar to photonic crystals. The center wavelength of the stopband, λB is determined by the period of the index modulation, Λ according to theBragg condition:120Figure 5.1: Design parameters for Bragg gratings: Λ is the period, a and b theindividual segment lengths, Λp the phase shift (p = 0.5), w the averagewaveguide width, and L the grating length.2Λne f f = mλB . (5.1)Here, the period is held constant but in a more general approach, the period(and corrugation width) could also vary as function of position in the grating tooptimize the spectral response and dispersion of the grating [210, 211]. In thiswork the duty cycle of the grating is 50% (a=b) and left for future work to studythe effect of a varying duty cycle. ne f f is the effective index of the waveguidewith average width w. Fig. 5.2(a) and c show the effective index of a waveguide(width height 220 nm) for the TE and TM mode respectively. The phase shift, Λpwith p = 0.5, forms a first-order Fabry-Perot cavity [73], resulting in a high-Q peakcentered in the stop band. To limit the number of peaks inside the stop band toone, the phase shift length is equal to a grating period [212]. This peak resultingfrom the two Bragg-reflectors with length L/2 facilitates more robust peak trackingduring biosensing experiments.The grating strength is given by [213]:κ = ping∆λλ2B(5.2)where ng is the group index and ∆λ is the bandwidth of the stop band. κ can beunderstood as the magnitude of the coupling between the forward and backwardpropagating mode [201] or as the field attenuation constant for the forward prop-agating mode. The half period phase shift (Λp, see Fig. 5.1) at the center of thegrating introduces a high quality factor resonant peak at the center of the stop band.121The number of grating periods (and corrugation depth) on each side of the phaseshift cavity determines the reflectivity of the distributed Bragg mirror and thereforealso the quality factor of the resonance peak. Figure 5.2(b) and 5.2(d) show thegroup index of a waveguide (height = 220 nm at 1550 nm) for the TE and TMmoderespectively.Figure 5.2: Propagation properties as function ofwaveguidewidth (and heightof 220 nm at 1550 nm) for air, H2O, and SiO2 cladding: a) Effectiveindex for TE polarized light; b) group index for TE polarized light;c) Effective index for TM polarized light; and d) group index for TMpolarized light.Figure 5.3a shows a schematic concept of a Bragg grating in an aqueous envi-ronment. Figure 5.3b shows the electric field intensity of the fundamental TE modetravelling in a 220 nm x 500 nm waveguide.The performance metrics that compare silicon photonic devices are fragmentedin literature [32–34]. The key metrics to assess a devices performance have beendiscussed in section 2.2. Briefly they are: (1) Sensitivity (bulk and surface), (2)122detection limit, and (3) Quality factor. The following sections will analyze differentBragg grating sensor designs with the goal to improve the performance.5.3 Uniform strip-waveguide Bragg grating with phaseshift for TE polarized light5.3.1 DesignStripwaveguideBragg gratings, similar to those shown in Fig. 5.3(a), were designedsimilar to Prabhathan et al. [207] and Jugessur et al. [206] (for TEmode) consistingof a straight waveguide, with an average width of 500 nm, with periodic side wallcorrugations of 40 nm, and with a grating period (Λ) of 440 nm and duty cycle of50% (a = b), resulting in 20 nm stop band [206, 207]. The TE mode field intensitydistribution is shown in Fig. 5.3(b).Figure 5.3: Strip waveguide Bragg grating biosensor. (a) 3D schematic ren-dering of a Bragg grating in a fluidic channel. (b) Electric field intensitydistribution of a TE mode at the input of a Bragg grating waveguide.The half period phase shift (Λp= 220nm) at the center of the grating introducesa high Q resonant peak (with water cladding at around 1.517 µm). With 100 pairsof gratings on each side of the phase shift the total footprint of the sensor is about90 µm2. The devices were fabricated and characterized as described in Section 3.2.1235.3.2 ResultsBragg gratings were fabricated on a 220 nm SOI wafer with a 2 µm buried oxidelayer using 193 nm deep-UV lithography and dry etching at IMEC through theePIXfab multi-project wafer service. Figure 5.4a shows an SEM image of thefabricated Bragg gratings. The sidewall corrugations depth of 40 nmmodulates theeffective refractive index leading to a stop band in the transmission spectrum. Thegrating period is 440 nm with a 50% duty cycle and the phase shift in the centeris 220 nm (660 nm in total as shown in Fig. 5.4(a). Proximity effects during thelithography process are responsible for the resulting waved sidewall [214]. Usinglithography simulation tools [130] this smoothing effect can be predicted [130].Figure 5.4: Experimental results for our TE mode strip-waveguide Bragggrating biosensor. (a) SEM image of the Bragg grating with the phaseshift. (b) Transmission spectra showing the high-Q resonant peak shiftswhen subjected to a NaCl RI solution set. (c) Sensitivity responsemeasured for the RI solution set. (d) Biosensing demonstration using amodel biological system as described in Section 3.2.4.Sensor characterization and a biosensing demonstration were performed as124described by the methods in Section 3.2. Wavelength shifts of the resonance peakwhen subjected to the refractive index solutions are shown in Fig. 5.4(c) andresulted in a sensitivity of 59 nm/RIU (Fig. 5.4(b)). The resonance peak has aquality factor of 28,000 resulting in an iLoD of 9.38 ·104 RIU. Ourmodel biologicalsystem assay, shown in Fig. 5.4(d), demonstrates its ability to sense and interrogatebiomolecular interactions. Although the introduction of BSA in Region C shows asmall permanent shift resulting from adsorption to exposed regions of the sensor, thesubsequent binding steps involving SA (Region D) and bBSA (Region E) correlatewell to expected results.5.4 Uniform slot-waveguide Bragg gratings with phaseshift for TE polarized light5.4.1 DesignAlthough the slot-waveguide Bragg design principles and theory are similar to theirstrip counterparts, the slot-waveguide configuration offers some unique advantagesthat have been investigated to enhance sensitivity for silicon photonic biosensors [44,172, 215, 216]. In slot-waveguide devices, the electric field intensity is concentratedbetween the waveguides (Fig. 5.5(a)), increasing the field and analyte overlap andresulting in 5−6x sensitivity improvement over their strip-waveguide counterparts[12, 44, 51]. Unlike rings, Bragg gratings do not suffer from bending losses thatlimit the performance of slot-ring resonators. The Bragg condition, susceptibility,and quality factor of slot-based sensors can be determined in a similar fashion tostrip waveguide Bragg gratings.We designed phase-shifted slot Bragg gratings similar to those shown in Fig.5.5(b). Optimized design parameters resulting from simulations include 40 nmsidewall corrugations on the outside of the waveguide (270 nm average width), aslot gap of 150 nm, and a 600 nm long phase shift in the center which creates ahigh quality factor resonant peak within the stop band. The period on either side ofthe phase shift is 440 nm (150 periods on either side). The devices were fabricatedand characterized as described in Section 3.2.125Figure 5.5: Simulation and design concept of a slot-waveguide Bragg gratingbiosensor. (a) Electric field intensity distribution of TE light showing thestrong confinement of the TE mode in the slot between the waveguides.(b) A 3D schematic rendering of a slot-waveguide Bragg grating in afluidic channel showing the phase shift and optimal sensing region.5.4.2 ResultsSlot-waveguide Bragg grating sensors were fabricated on a CMOS-compatible SOIsubstrate with a 220 nm silicon layer on top of a 2 µm buried oxide layer at IMEC’sePIXfab foundry and is shown in Fig. 5.6(a). The 150 nm wide slot between two270 nm wide, corrugated waveguides (arms), strongly guides the fundamental TEmode in the low-index cladding between the arms and results in high sensitivityand quality factor (Fig. 5.5(a)).The optimal corrugation depth for aqueous claddings is 40 nm (each arm widthis 270±20 nm) with a grating period Λ of 440 nm and duty cycle of 50%, resultingin Bragg wavelength near 1.53 µm. A sharp resonance peak in the stop band wasachieved by introducing a phase shift (Λp = 220 nm) in the middle of the gratingsFig. 5.5(b). With 150 pairs of gratings on each side of the phase shift; the sensoris about 132 µm long and results in a footprint of 132 µm2.Following the characterization procedure described in Section 3.2, NaCl RIsolution standards were flowed over the sensor while tracking the peak locationin the stop band. The observed shifts resulted in a sensitivity of 340 nm/RIU(Fig. 5.6(c) and quality factor of 15,000 (Fig. 5.6(b) the highest reported forslot-based biosensors to-date. Based on the quality factor and sensitivity, the126Figure 5.6: Experimental results for our TEmode slot-waveguide Bragg grat-ing biosensor. (a) SEM image of the slot-waveguide Bragg grating withthe phase shift. (b) Transmission spectrum showing the high-Q res-onant peak within the stop band. (c) Sensitivity response measuredafter subjecting the sensor to a NaCl RI solution set. (d) Biosensingdemonstration using a model biological system as described in Section3.2.4.iLoD for this device is 3.04 · 10−4 RIU, which is close to the theoretical limitof an ideal water-based resonator sensor operating at 1.55 µm wavelengths [32].Finally, the sensor’s biosensing capability is assessed using a modified sandwichassay which is shown Fig. 5.6(d). The surface sensitivity of the slot Braggsensor was simulated to be 748 pm/nm (for a 150 nm wide slot between two270 nm wide, corrugated waveguides). The BSA block and challenge in RegionC results in a slight, permanent shift indicating that BSA adsorbed to exposedportions of the sensor. With a finite adsorption step for antiSA (Region B), it isreasonable to expect that portions of the sensor remain exposed at the start of theBSA challenge and subsequently adsorb BSA, especially when considering analytetransport challenges and limitations to the slot region between waveguides [44].127For the remaining steps, the relative resonance wavelength shifts for each capturedbiomolecule corresponded well to the expected shift based on their respectivemolecular weights [174, 175].This sensor exhibits significantly higher quality factors over slot-waveguide ringresonators since it does not suffer from bending losses or mode mismatch losseslike rings. Using higher resolution lithography processes would further improvethe performance of this sensor [207]. The IMEC foundry process required a largerthan optimal slot dimension of 150 nm due to limitations of the fabrication process.Proximity effects during the lithography process result in wavy sidewalls [214](observed in Fig. 5.6(a) and make it challenging when designing Bragg gratings aslithography tools are required to predict the smoothing of the corrugations [130].This is important as the corrugation determines the coupling strength of the gratingand therefore the quality factor of the resonance peak. Finally, designing a similarsensor for 1.3 µm wavelengths, where loss to water absorption is reduced by anorder of magnitude [217], should improve the iLoD.5.5 Uniform strip-waveguide Bragg grating with phaseshift for TM polarized light5.5.1 DesignStrip waveguide Bragg gratings with sidewall corrugation were designed on a SOIsubstrate as described by Wang et al. [193, 209] and Chen et al. [210]. All designswere based on a SOI waveguide with width w = 500 nm and height h = 220 nm.MODE Solutions from Lumerical Solutions, Inc. was used to simulate the effectiveindex of the waveguide and Bragg condition given in Eq. 5.1 was used to determinethe required period, Λ to center the stop band around λB = 1550 nm. As proposedby Wang et al. [201], a 3D finite-difference time-domain (FDTD) simulation on asingle unite cell using Bloch boundary conditions in the direction of propagation, isemployed to determine the coupling strength for a given corrugation width. Usingthis approach, an infinitely long grating can be studied without considering thewhole structure and effects of design parameters (corrugation, period, thickness)on bandwidth (∆λ) and center of stop band (λB) can be verified.1285.5.2 Materials and methodsFabricationBragg gratings with different parameters were designed for 1.55 µm wavelengths(λB = 1550 nm). Deviceswith grating periodsΛ = [416,436,456,462,480,492]nmand corrugations of ∆w = [180,150,140,110,90,80,60,50,30] nm were designedusing the methods described above, fabricated, and compared. While all gratingdesigns are uniform, one set was designed with a phase shift while the other wasnot. To better understand fabrication limitation the gratings were fabricated usingtwo methods: (1) Ebeam lithography and (2) a foundry service as part of a multi-project wafer organized by CMC Microsystems (Ontario, Canada). Since siliconphotonic foundry processes cannot yet achieve the lithography resolution offeredby direct-write systems using an electron beam [126, 127], topologies with higherresolution and accuracy could be evaluated using Ebeam lithography instead. TheEbeam provides a low-cost, fast turn-around CMOS-foundry-compatible fabrica-tion process that has been optimized [126, 141, 142] to produce consistent, robust,low-loss silicon photonic components [127, 143]. Foundry fabrication was done atFigure 5.7: SEM images of fabricated Bragg gratings (a) Uniform gratingportion of a TMBragg grating biosensor fabricated using Ebeam lithog-raphy. (b) Top-view of a Bragg grating similar to (a) and showing thephase shift region which creates the resonant peak within the stop band.(c) Image of a foundry-fabricated TMmode, uniform Bragg grating withphase shift for biosensing.IMEC, Belgium, via ePIXfab on a SOI wafer with a 200 nm thick Si top layer and a2µm buried oxide layer using 193-nm lithography [218]. Chips were then claddedwith the perfluoro-polymer CYTOP purchased from Asahi Glass (Tokyo, Japan)129[144, 145]. The cladded chips were further patterned to etch openings aroundthe gratings to ensure complete exposure of the sensing region to their aqueousenvironment.CharacterizationThe custom automated probe station, as described in chapter 3 was used to charac-terize the gratings and assess their biosensing performance [87]. Light was coupledin and off chip with vertical grating couplers as designed byWang et al. [157]. Thesensors were subjected to an aqueous environment to determine its bulk sensitivity,Sb, quality factor, Q, and intrinsic limit of detection, iLoD. NaCl solutions of62.5 mM, 125 mM, 250 mM, 1 M, and 2 M were sequenced over the grating at 10µL/min while acquiring five sweeps ranging from 1500 nm to 1600 nm. The chipstage was thermally tuned to 25 degrees Celsius to minimize thermal drift.Figure 5.8: Schematic of the electrostatic polymer bi-layers created by alter-nating solutions of negatively charged polystyrene sulfonate (PSS) andpositively charged polyally-lamine hydrochloride (PAH).The evanescent field was characterized similar to what was described previously[40]. Briefly, the chip was cleaned using a Piranha solution (3:1 H2SO4:H2O2(30%) for 30 minutes and rinsed with ultra-pure water prior to functionalization.Next, a solution of positively charged polyethylene imine (PEI, 5mg/mL)was flowedover the sensor for 5 minutes to initiate the adhesion of subsequent electrostaticpolymer bilayers. Electrostatic polymer bi-layers were then created by alternatingsolutions of negatively charged polystyrene sulfonate (PSS, 5mg/mL) and positivelycharged polyally-lamine hydrochloride (PAH, 5mg/mL) for 5minutes at 20 uL/min.130Each solution was followed by a Tris buffer (0.5 mM, 100 mM NaCl, pH 7.1) rinsefor at least 3 minutes to avoid polymer precipitation and clogging of the fluidictubing. To measure the thickness of the bi-layers a glass slide with no bi-layers,10 bi-layers, and 20 bi-layers was manually prepared using the same electrostaticpolymers and the resulting thicknesses was measured using ellipsometry.Biosensing assayA biological sandwich assay was performed to assess the gratings biosening per-formance. Reagents involved include: Protein-A at 1 mg/mL purchased fromThermoFisher (Chicago, IL), anti-Streptavidin (antiSA) at 100 µg/mL and its con-jugate ligand, Streptavidin (SA) at 10 µg/mL both purchased from Vector Labs(Burlingame, CA), BSA at 100 µg/mL purchased from Sigma Aldrich (St. Louis,MO), and a biotinylation kit for BSA was purchased from Bangs Labs (San Diego,CA) and used in the final amplification step at 10 µg/mL. A 1x PBS solution pur-chased from FisherScientific (Hampton, NH) was used to dilute reagents and rinseunbound molecules after each step.Wavelength sweeps from 1460 nm to 1540 nm (1 pm steps) were executedapproximately every 30 seconds on two Bragg gratings in separate channels, oneacting as a negative control. All reagents were sequenced at 20 µL/min using asyring pump (Chemyx Nexus 3000, Houston, TX) operating in withdraw mode.The chip stage was thermally tuned to 37 degrees Celsius to limit noise resultingfrom thermal drift and an acetylene cell, (Wavelength References, Corvallis, OR),was used to determine wavelength jitter of the laser.5.5.3 ResultsUniform Bragg grating waveguidesSEM images of fabricated devices are shown in Fig. 5.7. Ebeam lithographyprovides high resolution with crisp corrugations segments (shown in Figs. 5.7(a)and 5.7(b) while gratings fabricated using standard foundry processes result insmoothed corrugated edges constrained by the optical lithography resolution (Fig.5.7(c).131As described previously, the Bragg condition (Eq. 5.1) determines the locationof the stop band for a given grating period. The effective index in Eq. 5.1 iscalculated for the average width and height of the waveguide. Figure 5.9(a) showsthe linear dependency of the bandgap center wavelength with grating period. Theextinction ratio (ER) is limited by the detector’s dynamic range. Figure 5.9(b) showsthe transmission spectrum for different corrugations. With increasing corrugationwidth, the center wavelength of the bandgap is shifting to shorter wavelengths.The same behavior is also observed in simulations as shown in Fig. 5.10(a). Assuggested by Verbist et al. this blue shift can be explained by a decreasing averageeffective index of gratings with an increased corrugation [219]. With an increasein corrugation depth a corresponding increase in bandwidth is observed due to thestronger index modulation.Figure 5.9: Measured transmission spectra for TM uniform gratings for a)different periods of Λ = [462,480,492] nm and corrugation ∆w =110 nm; and b) various corrugation width ∆w = [70,110,150,200] nmfor a fixed period of Λ = 480 nm. All gratings are fabricated throughEBeam lithography and have an average waveguide width w = 500 nmand are measured with air cladding.In Fig. 5.10 the center wavelength and the bandwidth are plotted vs the cor-rugation depths and compared to FDTD simulations. Because the feature sizeof the grating is comparable in size to the lithography wavelength generally usedin foundries for silicon photonics, an initial design of a rectangular grating willresult in a smooth grating as shown in the SEM image in Fig. 5.7(c). The resolu-tion of the EBeam lithography is much higher resulting in crisp corrugations. Byfollowing a similar approach as proposed by Pond et al. [220] such lithographyeffects can be simulated (NA = 0.6, σ = 0.7, and resist threshold of 0.25). The132simulated data in Fig. 5.10(a) was shifted vertically to match the measured resultsfor ∆w = 80 nm. This is necessary as incorrect layer thickness, temperature dif-ferences, or waveguide sidewall angles can introduce a constant offset. Using themeasured corrugation-dependent center wavelength and bandwidth the couplingcoefficient κ can be extracted using Eq. 5.2. The coupling coefficient is used tocalculate the reflectivity of the grating which is given by Rpeak = tanh2(κL) forthe lossless case. The reflectivity is an important design parameter to optimize thequality factor of a resonance peak of a uniform grating with a phase shift at thecenter. Figure 5.10(b) compares the measured and simulated coupling coefficientas function of corrugation width.Figure 5.10: Comparison of simulation and experimental results for TMBragg gratings as function of corrugation. All gratings have an averagewaveguide width w = 500 nm and are measured with air as claddingmaterial. (a) Blue shift of stop band with increasing corrugation depth(b) The grating strength (coupling coefficient) increases as a functionof the corrugation width.Uniform Bragg grating waveguides with phase shiftIntroducing a phase shift of half the grating period Λ in the middle of the gratingleads to the appearance of a resonance peak at the center of the stop band. Thegrating on either side of the phase shift can be viewed as distributedmirrors creatinga Fabry-Perot cavity. The quality factor (sharpness) of the resonance is determinedby the grating strength and the cavity loss. It is approximated experimentally bydividing the resonant peak’s wavelength (λres) by its full width at half its maximum133(FWHM). The loss attributed to the mirror is determined by the reflectivity and asFigure 5.11: Measured transmission spectra for TM uniform gratings withphase shift for different number of periods (L = N ·Λ) on either sideof the cavity, N = [433,217,109]]. The period Λ = 416 nm andthe corrugation ∆w = 140 nm are held constant. All gratings arefabricated through 193 nm lithography and have an average waveguidewidth w = 500 nm and are measured with air cladding.explained in the previous section, dependent on the coupling coefficient κ and thelength of the grating L (also sometimes given as the number of grating periods N ,with L = N ·Λ). Figure 5.11 shows the transmission spectrum for a uniform gratingwith period Λ = 416 nm, corrugation ∆w = 140 nm and different grating lengthexpressed as the number of periods. The quality factor varies from Q ≈ 500 for theshortest grating (N = 109), Q = 3.9k for N = 217, to Q = 32.3k for N = 433.Sensing performanceWhile the design of Bragg grating’s is very dependent on the fabrication process(as discuss above) the sensing performance of a grating fabricated through EBeamlithography or 193 nm deep UV lithography is in fact comparable (dependent onthe waveguide susceptibility and the waveguide guide group index). The charac-terization of the sensing performance was carried out on a grating with a period ofΛ = 436 nm, a corrugation ∆w = 140 nm, and a length L = 135 µm (N = 155 oneach side of the cavity).Bulk sensitivity: The Bragg gratings bulk sensitivity was experimentally assessedusing refractive index standards as described above. Figure 5.12(a) shows the134resonant wavelength shift of a phase-shifted Bragg grating as different refractiveindex standards were introduced. Points at each step were averaged and plottedagainst the solution’s measured refractive index to determine a bulk sensitivity of208 nm/RIU as shown in Fig. 5.12(a).Figure 5.12: Example characterization data shown for on-substrate TMBragggratings with design parameters: period Λ = 436 nm, corrugation∆w = 140 nm, and number of periods on each side N = 155 . a) Peakwavelength when subjected to refractive index standards. (b) Bragggrating sensitivity [nm/RIU] determined from (a).Surface sensitivity: Surface sensitivity is a usefulmetric for biosening applicationssince most adsorbed biomolecules form 10-20 nm thick adlayers in the sensor’ssurface. To experimentally verify the simulated surface sensitivity, alternatingelectrostatic polymer bi-layers with a known thickness (≈ 3 nm) were formed asshown in Fig. 5.8. Figure 5.13(a) shows the simulated resonance wavelength shiftfor the TE and TM modes versus layer thickness with refractive index of n = 1.68(for the electrostatic bi-layer) and n = 1.48 (for most proteins). As the electricfield strength decreases exponentially away from the surface an expected decreasein surface sensitivity is observed with a growing layer thickness. For very thicklayers the sensors are no longer able to detect a change. Figure 5.13(b) shows themeasured sensor response to the growing alternating bi-layer (10 scans per layer).The measured (black ’+’) and simulated (red solid line) agree very well. A surfacesensitivity of 0.65 nm/nm is measured. The thickness of the bilayer of 3 nm foundfrom [40] was confirmed with ellipsometry and used for converting scan numbersto layer thickness on the x-axis.135Figure 5.13: Surface sensitivity as function of thickness of adsorbed layer: a)Simulated sensitivities for both TE and TM modes (waveguide widthw = 500 nm and height h = 220 nm) plotted versus layer thicknesswith refractive index of n = 1.68 and n = 1.48. The cladding indexis assumed to be n = 1.33 for water; b) Comparison of experimentalsurface sensitivity of the TMmode to simulation. Electrostatic polymerbi-layers were created by alternating solutions of negatively chargedpolystyrene sulfonate (PSS, 5 mg/mL) and positively charged polyally-lamine hydrochloride (PAH, 5 mg/mL). The bi-layer thickness (afterone cycle of PSS/PAH) is about 3 nm with a refractive index of n=1.68[40]Temperature sensitivity: The overall effective thermo optic coefficient ∂n∂T (TOC)is dependent on the material of each of the waveguide regions weighted by theconfinement factor of the field within each of these regions. With the TOC givenfor silicon, silicon oxide, andwaterTOCSiO2 = 1 ·10−5K−1,TOCSi = 1.8 ·10−4K−1,and TOCH2O = −0.8 · 10−5 K−1 [221, 222] it becomes clear that a waveguide withan oxide cladding will have a higher temperature dependence as compared to water.In fact due to the negative sign of the coefficient it is possible to engineer anathermal waveguide geometry. Similar, a more confined mode will manifest astronger temperature dependence than a weakly guided mode. The TM Bragggrating exhibited a temperature sensitivity of 22 pm/C compared to 70 pm/C for aTE Bragg grating.Detection limit: Trade-offs need to be considered when optimizing a system for aparticular biosensing application. For example, high quality factors allow greater136intrinsic limits of detection. In the same way, increasing the overlap between themode and cladding increases the bulk sensitivity which also improves the intrinsicdetection limit. Yet, a more weakly confined optical mode, where more of its energyis susceptible to absorption losses of the aqueous cladding degrade the quality factor.Therefore one cannot increase the quality factor and the sensitivity simultaneously.Chrostowski et al. [32] suggested that there is a theoretical intrinsic detectionlimit for evanescent field waveguide based sensor iLoDlim = 2.4 · 10−4 RIU. Themeasured quality factor of the sensor used in the experimental work is 1.93kand the bulk sensitivity is Sb = 215 nm/RIU which results in a intrinsic limit ofdetection of iLoD = 3.73 · 10−3 RIU. The intrinsic limit of detection is based onthe assumption that the resonator linewidth is the minimal detectable wavelengthshift δλmin = ∆λ3dB. This is of course not really true and it depends on the wholesystem and its noise sources [32]. Taking the experimental noise floor of 1 pm thesystem limit of detection is given by sLoD = δλminSb = 4.65 ·10−6 RIU.Biosensing demonstrationThe performance of the TM Bragg gratings were evaluated for biosensing applica-tions using a modified sandwich assay [151] (also described in chapter 3). Figure5.14 shows the sensor’s response to a sequence of biological reagents.The assay started with PBS buffer for 20 minutes to establish a signal baseline.Next, Protein A (1 mg/mL) was physisorbed onto the sensor’s surface for 30 min-utes to immobilize the capture antibody. Physisorption is the irreversible bindingof proteins to a surface [155, 223] and Protein A has been observed to preferentiallybind the Fc domain of immunoglobulins [156]. Assuming the sensor’s surfacesensitivity is 0.32 nm/nm (meaning, an adlayer with thickness of 1 nm shifts theresonant wavelength by 320 pm), the ≈ 300 pm shift after the Protein A suggests a0.9 nm adlayer which is consistent with what has been reported previously [156].Anti-streptavidin (10 µg/mL) was then introduced to functionalize the sensor (Re-gion B). The observed shift of ≈ 1.1 nm is expected since the immunoglobulin is≈3.3x larger than protein A (160 kDa / 48kDa = 3.3) and suggests robust immobi-lization and sensor coverage. Region C shows the result of challenging the sensorand adsorbed molecules with BSA (20 µg/mL). This step not only confirms the137Figure 5.14: Sensogram of the sandwich assay used to assess the Bragg grat-ing’s biosensing performance. Region A: Protein A physisorption topromote immunoglobulin capture; Region B: immobilization of anti-streptavidin. Region C: BSA to block any exposed regions on the sen-sor and ensure specificity of subsequent steps. Region D: Streptavidin(SA) as the model target biomolecule. Region E: Final amplificationusing Biotin-BSA. The introduction of each reagent was followed bya buffer rinse to remove any unbound molecules as indicated by theshort, black-dashed line.biological specificity of the molecules used in the assay but blocks any remainingexposed portions of the sensor to limit subsequent non-specific interactions. Theabsence of an offset (or resonance wavelength shift) after the rinse step of RegionC suggests robust coverage of the sensor with antiSA (Region B). Region D showsthe permanent and irreversible wavelength shift of ≈140 pm when adsorbing SA138(20 µg/mL). A final amplification step biotynilated BSA (bBSA) is shown in RegionE. The permanent wavelength shift suggests specific binding of the bBSA with theimmobilized SA.5.6 Suspended waveguide TE and TM mode BragggratingsFor on-substrate Bragg grating sensors (the standard SOI geometry), a portion of theevanescent field is guided by the buried oxide (BOX). This is especially true for TMmodes where almost half of electric field is in the buried oxide substrate (with air orwater cladding above). Exposing all sides of the sensing region would allow moreoverlap of the evanescent field with the bulk solution, improving sensitivity. Whileunder-etched waveguides have been demonstrated for high performance computingapplications [146–149] and photonic crystal biosensors [224] they have yet to beinvestigated for strip and slot waveguide Bragg grating biosensors. Therefore,suspended waveguide TE and TM mode Bragg gratings for 1.55 µm wavelengthswere developed and compared with their on-substrate counterparts.5.6.1 DesignThe electric field intensity distribution was calculated for on-substrate and sus-pended strip waveguides using Lumerical MODE for both, the fundamental TE andTM mode (see Fig. 5.15.Figure 5.15: Simulated electric field density for 1.55 µmwavelength TE (left)and TM (right) modes in a 500 nm wide by 220 nm tall silicon waveg-uide.139The electric field overlap with the aqueous cladding was integrated to determinewhat percentage of the mode’s total electric field intensity was susceptible to refrac-tive index changes in the cladding. Table 5.1 lists the evanescent field overlap andthe expected sensitivity improvement gained in suspending the waveguides. Theexpected sensitivity improvements are calculated using Eq. 2.21 and the overlapintegrals as defined in Eq. 2.2.Table 5.1: Electric field intensity in cladding and predicted sensitivity im-provementType % EV in the Predicted Sensitivity ExperimentalCladding Improvement ImprovementTE on substrate 22suspended 34 48% 50TM on substrate 37suspended 82 134% 127To evaluate the silicon waveguide stress and potential failure points underflow, suspended waveguide lengths ranging from 10 to 50 µm were modeled inSolidWorks and simulated using COMSOL under worst case flow conditions, i.e.flow perpendicular to the waveguide (a 30 µmwaveguidewas fabricated). Modelingresults are shown in Fig. 5.16 and suggest that stress and deflection on the suspendedwaveguide are within its tensile strength limits (168 MPa [225]) at the flow rates ofinterest (10-100 µL/min) for up to a 50 µm long strip waveguides. The simulationspredict a beam deflection of 4 nm for a 500 nm by 220 nm by 50 µm long underflow at 10 µL/min. The beam deflection calculations are based on the assumptionof uniform loading on a beam fixed at both ends due to an average flow velocityu0 =Q/A. In Fig. 5.16 the maximum beam stress at the fixed end is plotted versusflow rate. The Reynolds number at Q = 10 µL/min is RE = 3.73 and indicateslaminar flow.Mode profile simulations for suspended strip waveguide sensors suggest a 45%increase in electric field overlap with target analytes in the bulk solution for a TMmode sensor and a 12% increase for a TEmode sensor. For TE gratings, simulationspredicted a 48% improvement in sensitivity and 50% was experimentally observed.Likewise, TM simulations predicted a 134% improvement in sensitivity and 127%140Figure 5.16: Suspended waveguide stress under flow analysis. (a) Schematicofworst-case simulation setup. Flow is perpendicular to thewaveguide.(b) Simulation results showing a beam deflection of 4 nm for a flowrate of 10 µL/min and stress of 3.5 MPa.was experimentally observed.5.6.2 FabricationDevices were fabricated on a 220 nm SOI wafer with a 3 µm BOX (SOITech)using a 100 keV JEOL JBX-6300FS Direct Write Ebeam Lithography System atthe University of Washington’s Nanofabrication Facility (WNF) [127]. After asolvent rinse and hot-plate dehydration bake, hydrogen silsesquioxane resist (HSQ,Dow-Corning XP-1541-006) was spin-coated at 4000 rpm, then hotplate baked at80 oC for 4 minutes. Waveguide structures were written using the ebeam at 100 keVenergy, 8 nA beam current, and 500 µm exposure field size. The machine grid usedfor shape placement was 1 nm while the beam stepping grid, the spacing betweendwell points during the shape writing, was 6 nm. An exposure dose of 2800 µC/cm2was used. The resist was developed by immersion in 25% tetramethylammoniumhydroxide for 4minutes, followed by a deionizedwater rinse for 60 s, an isopropanolrinse for 10 s, and then blown dry using nitrogen. The silicon was removed fromunexposed areas using inductively coupled plasma etching in an Oxford PlasmalabSystem 100 with a chlorine gas flow of 20 sccm, pressure of 12 mT, ICP power of800 W, bias power of 40 W, and a platen temperature of 20 oC, resulting in a biasvoltage of 185 V. During etching, chips were mounted on a 100 mm silicon carrier141wafer using perfluoropolyether vacuum oil.Figure 5.17: SEM image of a fabricated, suspended WG Bragg grating: a)Top view of grating with phase shift; b) Side view of grating on slab;c) Side view of undercut Bragg gratings waveguides; and d) zoomedin side view of undercut Bragg grating waveguide.Chips with suspended waveguides were first cleaned with 1:3 piranha solution(H2O2:H2SO4) followed by a hexamethyldisilazane surface treatment to promotephotoresist adhesion and prevent lift off during the etch cycle. Next, AZ1512,a standard broadband Novalak based photoresist, was spun onto the chip, baked,exposed, and developed to pattern the undercut regions (waveguide suspensions).The undercut region was centered on the Bragg gratings with width of 30 µm.Buffered oxide (10:1) was used to etch through the 3 µm thick buried oxide layerto suspend the waveguide structures. Without drying, chips were rinsed withdeionized water and then soaked in acetone to remove any remaining photoresist.While still wet, chips were then transferred to isopropanol bath to prevent stiction142during the final dry step under a stream of nitrogen. SEM images confirmed thatthe structures were released as shown in Fig. 5.17.5.6.3 ResultsDevices were tested as described in chapter 3.2. Table 5.2 summarizes themeasuredTable 5.2: Measured Bragg grating performance comparisonType S [nm/RIU] Q [1e3] iLoD [RIU]TE on substrate 59 29 3.8·10−4suspended 89 10.1 1.72·10−3TM on substrate 215 1.9 3.73·10−3suspended 485 3.2 9.99·10−4performance parameters of the four Bragg gratings compared in this manuscript.The on-substrate TE mode had the best quality factor and iLoD but lowest sen-sitivity resulting from the high modal confinement within the waveguide core.Pre-fabrication simulations suggested a 1.48x sensitivity improvement when sus-pending the waveguide (87 nm/RIU). The measured sensitivity of 89 nm/RIU iswithin the measurement and fabrication tolerances and agrees well with the sim-ulation. As more of the suspended waveguide’s mode overlaps with the aqueouscladding, the Q decreases indicating an increase in absorption loss.A sensitivity of 215 nm/RIU and quality factor of 1.9k was observed for theon-substrate TM Bragg grating. Pre-fabrication simulations suggested a 2.34x sen-sitivity improvement when suspending the waveguide (503 nm/RIU). Themeasuredsensitivity was 485 nm/RIU (2.26x improvement) which is close to the predictedperformance. The improved quality factor in the suspended TM Bragg grating overits on-substrate counterpart suggests that design parameters of the on-substrategrating should be optimized, namely the length of the grating to optimize the re-flectivity of the distributed mirror on either side of the phase shift. Because of thehigher index contrast when suspended in water the grating strength increases andthe length has to be adjusted (shorter length) to have critical coupling.1435.6.4 Biosensing demonstrationA sandwich assay representing a model biological system was performed to assessthe on-substrate and suspendedwaveguideBragg gratings ability to detectmolecularbinding. Assay results for the two TM Bragg gratings are shown in Fig. 5.18.Measured sensitivity values suggest that the suspended waveguide sensor responseto binding events is almost twice that of its on-substrate counterpart (Table 5.2).Figure 5.18: (a) On-substrate waveguide TM Bragg grating biosensingdemonstration. (b) Suspended waveguide TM Bragg grating biosens-ing demonstration. (c) Reagent sequencing corresponding to regions[A-E]. Region A = Protein-A (1 mg/mL), B = anti-streptavidin (SA)(125 ug/mL), C = Bovine Serum Albumin (BSA) (2 mg/mL), D =streptavidin (SA) (1.8 µM), E = Biotin-BSA (2.5 mg/mL). Introductionof reagent in each region was followed by a PBS-wash.Figure 5.18c is a cartoon representation of the reagent sequence and bindingsteps experimentally observed in Figs. 5.18(a) and 5.18(b). Both assays usedsimilar reagents and concentrations but differed slightly in step durations. Bothchips were thermally tuned to 37oC and reagents were flowed over the sensors at20 µL/min throughout the assay. The introduction of each reagent was followed bya 20 minute buffer rinse to remove any unbound species in the channel as indicatedby the short, black dashed lines in Fig. 5.18.144After establishing a signal baseline in 1x PBS buffer (pH = 3.7), RegionA showsthe resonant wavelength response to the physisorption of protein-A (1 mg/mL) ontothe sensor’s native oxide surface. Physisorption is the irreversible binding ofproteins to a surface and is a well-known phenomenon [154, 155]. Protein A hasbeen observed to preferentially bind the Fc domain of antibodies [204, 223] andwas used to bind and orient the capture antibody’s Fab regions towards the channelfor improved analyte binding. The large resonant wavelength shift in Region Bconfirms robust immobilization of the capture antibody, anti-streptavidin (antiSA,10 µg/mL). A wavelength shift of 3 nm is observed for the suspended waveguide(Fig. 5.18(b) and 1.25 nm for its on-substrate counterpart (Fig. 5.18(a), resulting ina 2.4x increase which is close to the simulated value of 2.34x and an experimentallyobserved sensitivity increase of 2.25x.Region C shows the introduction of BSA (20 µg/mL) to block any remainingexposed surfaces on the sensor. This is necessary to prevent fouling and to validatethat subsequent binding interactions are specific. Both sensors (Figs. 5.18(a) and5.18(b) eventually return to Region C’s initial resonant wavelength suggesting thesensor is sufficiently covered with the antibody. Next, the sensor was subjected tostreptavidin (SA, 10 µg/mL) as a model, target analyte (Region D). The permanentshift suggests a specific and irreversible binding interaction as expected. As a finalamplification step, biotinylated BSA (bBSA) was introduced (Region E) resultingin another permanent resonant wavelength shift. The unfortunate introduction of anair bubble for one scan, as shown by the red circle in Fig. 5.18(b) Region C, resultsin the loss of some bimolecular mass from the sensor’s surface. While this resultsin an offset, subsequent binding of the SA and irreversible shift of biotinylated BSAsuggest the surface chemistry remains robust and specific.Post-experiment simulations were performed to verify that the amount of ob-served bound species is within expectations. Since captured protein layers may notnecessarily result in complete monolayers, the following assumptions were made:(1) proteins have refractive index of 1.48 [101], (2) protein A is 42 kDa and forms1-3nm thick adlayers [153, 156], (3) the IgG antibody has a mass of 160 kDa, and(4) the total film thickness of the bound species is 15-30 nm thick [174, 175]. Theexperimentally measured resonance shift is in good agreement with simulations.1455.7 TM mode Bragg gratings at 1310 nm5.7.1 IntroductionAs described in section 2.2.3, the intrinsic limit of detection (iLoD) is improved byincreasing the sensitivity or the quality factor of the resonator [32]. Unfortunately,increasing sensitivity (mode overlap) impacts the quality factor as absorption lossesdue to the aqueous cladding dominate. Conversely, improving the quality factorby increasing modal confinement within the waveguide to avoid scattering andabsorptive loss decreases sensitivity (see section 2.2.3 for an indepth discussion ofthe trade offs). If water absorption is the dominant loss, the sensor’s intrinsic limitof detection cannot be further improved. However, the loss of water is dependent onthe wavelength as described in section 2.2.4. Water absorption has been observed tobe approximately ten times lower around 1.3 µm wavelengths compared to 1.5 µmwavelengths [104, 217] (see also section 2.2.4). Assuming a similar waveguidesusceptibility for a sensor designed for 1.3 µm wavelengths, the reduction in waterabsorption should offer an intrinsic limit of detection limit increase by an order ofmagnitude, up to the theoretical limit of 2.4·10−5 RIU for 1.3 µm wavelengths.5.7.2 DesignFigure 5.19 shows the loss simulation results for common waveguide widths whichsupport single TE and TM modes for a silicon thicknesses of 220 nm (for bothwavelengths, 1.31 µm and 1.55 µm). The simulation setup consisted of a rectan-gular silicon waveguide on substrate with water cladding on three sides to mimicthe biosensing environment. For a fair comparison, the range of typical waveguidewidths are chosen for guiding single modes in a 220 nm thick waveguides. Waveg-uide widths ranging from 250-450 nm were simulated for 1.31 µm wavelengthsand widths ranging from 400-600 nm for 1.55 µm wavelengths. While most of thesensors explored were fabricated using 220 nm SOI material, 150 nm thick sensorswere investigated since many foundries offer etch depths for 150 nm which mayoffer improved sensitivity for biosensing.The simulation results in Fig. 5.19 show an approximate order of magnitudedecrease in loss (on average) for 1.3 µmwavelengths over their 1.5 µm counterparts.146Figure 5.19: Simulation results to compare loss between 1.31 µm and1.55 µm wavelengths across a range of waveguide widths 220 nmthick waveguide . Note, values were omitted for modes that were notguided by the waveguide dimensions being simulated. (a) Loss com-parison for TEmodes guided in a 220 nm thickwaveguidewith aqueouscladding. (b) Loss comparison for TMmodes guided in a 220 nm thickwaveguide with aqueous cladding.Note that TM modes are not supported in 150 nm thick waveguides at 1.55 µmwavelengths for waveguide widths less than 600 nm (and have been omitted).For a 220 nm thick waveguide, the predicted sensitivity (based on our simu-lations) for 300 nm wide waveguide supporting a TE mode is about 90 nm/RIU.Likewise, a 350 nm wide waveguide supporting a TM mode should result in asensitivity of approximately 120 nm/RIU.5.7.3 ResultsTo investigate our hypothesized detection limit offered by sensing at 1.31 µmwavelengths, we fabricated TM Bragg gratings using ebeam lithography on a 220nm SOI substrate with 3 µm oxide layer, and characterized as described in Section3.2. Design parameters for the TM Bragg include a grating period of 330 nm (dutycycle of 50%) and a corrugation of 60 nm. The center of the grating has a phaseshift to create a sharp, resonant peak in the stop band, as is shown in Fig. 5.20(a).The resonant peak has a high quality factor of 76,320 with an extinction ratio of13.5 dB. To assess sensitivity to refractive index changes in the cladding, the Bragggrating was subjected to a set of NaCl titrations. The resulting wavelength shifts147were plotted for each solution and fit to determine a sensitivity of 106 nm/RIU as isshown in Fig. 5.20(b). The quality factor and sensitivity results in a intrinsic limitof detection of iLoD = 1.62 · 10−4 RIU, less than the 1.55 µm wavelength limit of2.40 ·10−4 RIU [32].Figure 5.20: Experimental results for TM mode Bragg gratings designed for1.3 µm wavelengths. a) Transmission spectrum of Bragg grating withresonant peak in stop band b) plot of resonance peak wavelength shiftas function of refractive index change. The sensitivity is determinedby the slope of the linear fit and is 106.1 nm/RIU.5.8 SummaryFrom the original TE Bragg grating design on a strip waveguide, yielding asensitivity of 59 nm/RIU, Q = 28k, and an intrinsic limit of detection iLoD =9.38 ·10−4 RIU, this chapter included the design and experimental verification of aTE slot Bragg design (Sb = 340 nm/RIU, Q = 15k, and iLoD = 3.04 · 10−4 RIU),TM Bragg gratings (Sb = 215 nm/RIU, Q = 1.9k, and iLoD = 3.73 · 10−3 RIU),underetched TE and TM Bragg gratings (TE: Sb = 89 nm/RIU, Q = 10.1k, andiLoD = 1.72 ·10−3 RIU; TM: Sb = 484 nm/RIU,Q = 3.2k, and iLoD = 9.99 ·10−4)and TM gratings designed for 1310 nm (Sb = 106 nm/RIU, Q = 76k, and iLoD =1.62 ·10−4). Bragg gratings with different design parameters were fabricated usingebeam lithography and compared with their foundry fabricated counterparts (deepUV lithography). We showed how the corrugation width impacts the bandwidth andhow simulation can predict the behavior using a bandstructure approach borrowedfrom photonics crystal theory. For all the designs, we characterized the sensors’performance in an aqueous environment and used a model bioassay that mimics atypical biological system. The enhanced performance achieved in our design cou-148pled with the smaller footprint over ring and disk resonators could facilitate highersensor integration for future Lab-on-Chip (LOC) applications. While the iLoDachieved for the 1310 nm design is lower than what can be obtained with sensorsoperating at 1.55 µm wavelengths, the order of magnitude increase hypothesizedwas not achieved. Although the water absorption loss is an order of magnitude lessat 1.31 µm than at 1.55 µm, the other sources of loss do not exhibit the same wave-length dependence. This means that it may be more difficult to design resonatorswhere loss is dominated by water absorption, and thus more optimization will berequired to design sensors approaching the theoretical limit of detection.Bragg gratings show great potential for biosensing application as their surfacesensitivity can be greatly improved by either using a slot waveguide (748 pm/nmfor a TE mode slot) or by employing a suspended structure (640 pm/nm for asuspended TM mode Bragg grating). In addition they have a very small foot printwhich allows for a much higher sensor density which would enable multiplexedassays. The footprint of a ring resonator is limited by the radius, e.g. 20 µm. Incase of a Bragg grating the footprint is limited by the waveguide with. The sensitivearea is confined to the region with the phase shift [207].149Chapter 6Conclusion and Future Directions6.1 ConclusionSilicon photonics has matured considerably over the past decade, largely driven byneeds for higher data communication and high performance computing [226, 227].As devices and fabrication processes get optimized for high volume production,focus is shifting towards fully integrated lab-on-chip systems for biosensing ap-plications [22, 36, 228–233]. Silicon photonic biosensors offer many advantagesover other MEMS-based biosenors [2]; namely, silicon photonic devices are notsusceptible to mechanical force under normal microfluidic flow, electromagneticinterference, nor the electrical conduction of the analyte media. Additionally, sili-con photonic biosensors offer real-time monitoring of kinetic binding interactions.And critical to any commercialization effort, these silicon-on-insulator (SOI) pho-tonic biosensors can be fabricated in processes compatible with CMOS circuits andwidely accessible through many different foundries [234].Many different resonant photonic devices been investigated as label-free sensorsfor applications ranging from environmental monitoring [120] and basic scienceresearch [121] to bio threat detection [122] and medical diagnostics [22, 35, 41].These sensors have demonstrated impressive sensitivities and detection limits,most assays silicon photonic biosensor systems require secondary amplificationto achieve clinically relevant levels [2, 22]. In addition, there exists a considerablestartup barrier in creating novel silicon photonic biosensors and validating clinically150relevant assays on them.To this end, the efforts described in this thesis aim to solve these problemsby improving the native performance of many different kinds of silicon photonicbiosensors and compare their performance in a real-world, biological environment.In addition, the tools I developed to domyown research (namely the PDK frameworkand probe station that automates device characterization and assay development)have been open-sourced and shared broadly with both the scientific and industrialcommunities. The probe station is currently in use at more than a dozen institutionsand companies world-wide and the PDK framework is used for every multi projectwafer run organized by SiEPIC (Prof. Chrostowski) and is also in use at the WNFat the University of Washington in Seattle.In terms of advancing the field and to our knowledge of silicon photonic biosen-sors, my contributions include many novel foundry-compatible TE and TM moderesonant devices operating at 1550 nm and 1310 nm wavelengths with improvednative performance over today’s commercially available systems [28]. The specificmerits of each design are briefly summarized the next section.6.2 Sensor performance summaryThis section summarizes the performance of the investigated sensors and discussesadvantages, disadvantages, and ramifications for use in biosensing. First, the sensorperformance results are summarized in table 6.1. Next, mechanisms that determinedetection limits and fundamental boundaries for 1.55 µm and 1.31 µwavelengthsensors in an aqueous environment are discussed. The performance metrics usedto compare biosensor performance include: the quality factor Q; bulk and surfacesensitivity (Sbulk and Ssur f ), the resonator’s susceptibility to refractive index pertur-bations; and the intrinsic limit of detection (iLoD), which determines the smallestunit of change the sensor can distinguish. The performance and biosensing capa-bilities of thin-waveguide (90 nm, 150 nm, and 220 nm) TE mode ring resonators,150 nm and 220 nm thick waveguide TM mode ring resonators, sub-wavelengthgrating TE mode ring resonators, 3 µm and 10 µm TE and TM mode micro-diskresonators, and strip- and slot-waveguide Bragg gratings at 1.55 µm and 1.31 µmwavelengths was presented in this work. The experimental results are summarized151in table 6.1 in order by the measured sensitivity. The sensors are divided into twogroups according to the improved performance parameter; namely, 1) sensitivityimprovements and 2) quality factor improvements.Sensors that improve sensitivity: There existmany strategies to improve sensitivityby increasingmodal overlap with the cladding, namely employing TMmodes, usingthinner waveguides (90 nm and 150 nm), and using sub-wavelength waveguide(SWG) configurations. This section describes experimental results for the (1) TEmode thin-waveguide ring resonators, (2) TM mode ring resonators, (3) TE modeSWG ring resonators, and (4) slot-waveguide TE mode Bragg gratings. It is alsoimportant to note here, that it is useful to distinguish between bulk sensitivity andsurface sensitivity.Sensors that improve the quality factor: This section briefly highlights experi-mental results towards improved quality factors for (1) 3 µm and 10 µm radius diskresonators, (2) TE mode strip- and slot-waveguide Bragg gratings, and (3) TE andTM ring resonators and TM Bragg gratings designed for 1.3 µm wavelengths.6.2.1 Sensors that improve sensitivityA sensor’s sensitivity is determined by the overlap of the electric field and theanalyte. This is true for the bulk and the surface sensitivity.Bulk sensitivityBecause the bulk sensitivity is relatively easy to measure experimentally mostpublished sensor designs use the bulk sensitivity as main performance indicator. Ofall the sensors investigated, the SWGTEmode ring, slot-waveguide TEmode Bragggrating, and suspended TM mode Bragg gratings offered the highest sensitivities.SWG rings achieve this by guiding the mode through the cladding between theperiodic, sub-wavelength gratings (Fig. 4.19). The slot-waveguide Bragg gratingsachieve this by guiding the majority of the mode’s electric field in the claddingbetween the silicon waveguide arms (Fig. 5.5(a)). Suspended TMpolarized sensorsachieve this by exposing both sides of the waveguide to the analyte. While both152Table 6.1: Performance metrics of investigated biosensors. Where SWG = sub-wavelength grating, Wvl = wavelength[µm], and Q, Sbulk , Ss, and iLoD are the comparative performance metrics as described above.Section Sensor Type Mode Wvl [µm] Q [k] Sbulk [nm/RIU] Ssurf [pm/nm] iLoD [RIU] sLoD [RIU]4.7 SWG ring TE 1.55 3.9 490 789 8.1 ·10−4 2.04 ·10−65.6 suspended Bragg grating TM 1.55 3.2 485 640 9.98 ·10−4 2.06 ·10−65.4 Slot Bragg grating TE 1.55 15 340 748 3.04 ·10−4 2.95 ·10−64.4 Ring (h = 150 nm, w = 750 nm) TM 1.55 1.9 238 243 3.28 ·10−3 3.04 ·10−64.7 strip Bragg grating TM 1.55 1.9 215 312 3.73 ·10−4 4.65 ·10−64.4.2 Ring (h = 90 nm, w = 500 nm) TE 1.55 9.2 153 298 1.10 ·10−3 6.54 ·10−64.4 Ring (h = 220 nm, w = 500 nm) TM 1.55 9.2 147 312 7.09 ·10−4 4.2 ·10−64.6 10 µm Disk TM0 1.55 16 142 289 6.82 ·10−4 7.05 ·10−64.4 Ring (h = 220 nm, w = 350 nm) TM 1.31 33.5 113 251 3.47 ·10−4 8.85 ·10−65.3 Strip Bragg grating TM 1.31 76 106 251 1.62 ·10−4 9.5 ·10−65.6 suspended Bragg grating TE 1.55 10.1 89 210 1.72 ·10−3 1.12 ·10−54.4.2 Ring (h = 220 nm, w = 350 nm) TE 1.31 9.8 91 181 1.49 ·10−3 1.1 ·10−54.4.2 Ring (h = 150 nm, w = 500 nm) TE 1.55 14 83 212 1.33 ·10−3 1.2 ·10−55.3 Strip Bragg grating TE 1.55 28 59 160 9.38 ·10−4 1.7 ·10−54.4.2 Ring (h = 220 nm, w = 500 nm) TE 1.55 15 38 160 0.72 ·10−3 2.63 ·10−54.6 3 µm Disk TE0 1.55 33 26 76 1.81 ·10−3 3.85 ·10−54.6 3 µm Disk TE1 1.55 22 29 84 2.43 ·10−3 3.45 ·10−54.6 10 µm Disk TE0 1.55 131 21 75 5.63 ·10−4 4.76 ·10−5Theoretical intrinsic detection limit in aqueous solutions at 1.55 µm wavelengths 2.40 ·10−4Theoretical intrinsic detection limit in aqueous solutions at 1.31 µm wavelengths 3.14 ·10−5153SWG rings and slot-waveguide Bragg gratings demonstrate the highest sensitivityof all sensors investigated, their use in biosensing applications may be limited.Claes et al. suggested in an attempt to explain their under-performing slot ringresonators, that mass transport into the nano-scale slot might be limited by the sizeof the slot and steric hindrances [44]. They also observed wetting issues of the slot.A potential solution to improve molecule transport might be to employ larger slotsbetween waveguides, although care must be taken to not degrade the quality factor(and ultimately detection limit). These transport issues might not be quite as severefor SWGs as the slot is not along the direction propagation but perpendicular to itand therefore much shorter.In general TMmodes offer higher sensitivities over their TE mode counterpartssince the majority of the TM mode electric field is in the cladding rather thanthe silicon waveguide. This was demonstrated by the TM mode ring resonators,disks, and 1.31 µm Bragg gratings. Yet, the sensitivity could be further improvedby harnessing portions of the mode traveling in the oxide and exposing them tothe cladding material. For TM mode Bragg gratings, this has been achieved bysuspending the Bragg grating within a fluidic channel, doubling the sensitivity.For disk resonators, more of the TM mode could be exposed to the cladding byslightly under-etching its edges. Finally, the waveguide geometry can be optimizedto increase the modal energy overlap with the cladding. This can be seen by theincreased sensitivity of the 150 nm thick waveguide TM mode ring over its 220 nmthick waveguide counterpart (243 nm/RIU vs 147 nm/RIU respectively) and the TEmode rings designed for 90 nm, 150 nm, and 220 nm thick waveguides which offersensitivities of 153 nm/RIU, 83 nm/RIU, and 38 nm/RIU respectively.Surface sensitivityEven though the bulk sensitivity is simple to verify experimentally, it is importantto notice that for most bioanalytical applications one is only interested in howmuch of the target molecules are captured on the surface of the sensors. Mostbiological assays will stack up to an adlayer thickness on the order of 10 nm[46, 235]. It is therefore helpful to introduce the surface sensitivity as performanceindicator. The surface sensitivity describes the change in the resonance wavelength154as function of adlayer thickness. Since the evanescent field decays exponentially andthe sensitivity is proportional to the overlap of the analyte with the electric field,the surface sensitivity depends on they layer thickness. The surface sensitivitydecreases with increasing layer thickness as depicted in Figs. 2.15(a) and 2.15(b).Each biomolecule interacting with the surface of the sensor causes a local changein refractive index, which in turn alters the propagation properties of the guidedlight. The biomolecular layer can be modeled as uniform layer with thicknessdad and refractive index nad [101, 236]. The refractive index of said layer, andeven the thickness, is a matter of ongoing research, but general statements can bemade. The refractive index nad is between the refractive index of the dry moleculesand the refractive index of the solution, nbulk , which in most cases, is an aqueoussolution or PBS. The refractive index of such solutions is around nbulk = 1.33.The refractive index for hydrated, adsorbed proteins, nad, is between 1.35 and1.6 in literature. For most proteins, it is between 1.45 and 1.5 but typically therefractive index is assumed to be nad = 1.48 [101, 102, 237, 238]. However, alsoof interest is how dad and nad evolve as the monolayer builds up. The molecularlayer can be modeled as an adlayer with fixed height and varying refractive index.This variation is assumed to be linear starting from the bulk solution index to asaturation index of a densely packed monolayer [101, 102]. Alternatively, onecan assume a constant refractive index of the layer, nad, but a changing effectivethickness dad (t), where t denotes the deposition time. This approach has been usedin [202, 239–241]. Both of the suggested approaches are approximations and theirvalidity depends on the biomolecule and the experimental conditions. Furthermore,upon adsorption of a protein onto a surface the bioactivity and specificity can alsobe altered affecting the subsequent adlayer density. Therefor it is necessary to runcalibration experiments with know layer thicknesses for each biomolecule in orderto be able to do a quantitative analysis. However, measuring the thickness of anadorbed protein layer can be challenging in itself. In section 5.5.3 the surfacesensitivity is measured by using alternating electrostatic polymer bi-layers witha known thickness of about 3 nm per bi-layer. This method was suggested andvalidated by Luchansky et al. [40]. Even though the refractive index of the bi-layeris n = 1.68 (compared to n = 1.48 for most proteins), the model assay is usedto validate the simulation methodology. Table 6.1 reports the simulated surface155sensitivities for each of the sensor types. The reported sensitivities are simulatedfor the first 1 nm adlayer and are used to predict the expected shifts during themodel biological assay. The adsorption of the Protein A layer in particular agreesvery well with what is expected from literature (assumption of spherical shape anddiameter of≈ 5 nm) [156, 159]. Of the sensors investigated, the SWG ring resonatorshows the highest surface sensitivity of 789 pm/nm. This means that each adlayerof 1 nm thickness will shift the resonance peak by 789 pm. A similar value is alsoreported by Wangüemert-Pérez et al. [? ]. A similar high surface sensitivity of748 pm/nm is reported for the slot waveguide Bragg grating. The surface sensitivityfor an on-slab TM Bragg grating is about half of what is reported for the suspendedTM Bragg grating (312 pm/nm compared to 640 pm/nm). This makes sense as thesurface area is also twice as big for the suspended waveguide when compared to theon-slap counterpart. Even though the bulk sensitivity for the 150 nm thick TM ringresonator is higher than for a 220 nm thick waveguide, the surface sensitivity doesnot follow that trend. This can be explained by looking at the TM mode profile.The waveguide geometry of h = 150 nm and w = 760 nm is very close to the cutoff region (barely supports a TM mode). Because of the higher refractive indexof the substrate (nSiO2 = 1.45) compared to the cladding (nH2O = 1.33) the modeoverlaps more and more with the substrate as the waveguide gets thinner and hencethe surface sensitivity is also decreasing.6.2.2 Sensors that improve the quality factorQuality factors can be enhanced by optimizing the coupling conditions for criticalcoupling and reducing loss [207]. Naturally, Bragg gratings do not suffer frombending, coupling mode-mismatch loss, or sidewall scattering in the same way ascircular resonators. Yet their grating strength (corrugation width and period) andsensor length can be optimized to achieve critical coupling and improved Q. Sincea foundry’s lithography capabilities often dictate feature size resolution, fabricatingBragg gratings using the ebeam process described in section 3.2 should reducethe sidewall smoothing effects (Fig. 5.4(a) and improve the grating strength (thismight change as foundries switch to shorter wavelengths). While the disk resonatorquality factors are the highest among the 1.55 µmwavelength sensors due to higher156modal confinement within the waveguide (as compared with rings), quality factorsof disk, rings, and Bragg gratings can be improved even more by optimizing thecoupling into the resonator and employing post-etching techniques [242] to reducescattering losses caused by sidewall roughness [218, 243]. In addition, larger radiiwill further reduce bending and sidewall scattering loss at the expense of the abilityto multiplex many sensors on a single waveguide bus.Of the sensors investigated, ones with the least sensitivity include the 3 µmand 10 µm disk and 220 nm thick TE mode ring (sections 4.6 and 4.3. Forthe TE mode ring, its standard waveguide dimensions guide more of the modewithin the silicon core limiting its susceptibility to cladding changes. Therefore,reducing the waveguide dimensions, as was demonstrated in the 90 nm and 150 nmthin-waveguide rings, would offer higher sensitivities (153 nm/RIU and 83 nm/RIUrespectively). For the disk resonators, slightly under etching around its edges wouldalso expose more of the mode to the cladding, ultimately improving sensitivity.Since most of the literature does not use the intrinsic limit as figure of meritbut rather the system limit of detection, table 6.1 also reports the system limit ofdetection for the measured devices. To calculate the sLoD a smallest detectablechange of 1 pm is assumed based on the experimental results from section 3.7. Incomparison, the commercially available platform Genalyte reports a wavelengthresolution 0.1 pm [244]. Hence, with an improved readout noise of 0.1 pm, ourbest device (SWG) would offer a system limit of detection of 2 · 10−7, which is 4Ximprovement over Genalyte’s system [12].6.2.3 Improving limits of detectionAs stated in the first chapter, the limits of detection for silicon photonic biosensorsneeds improvement to achieve ELISA-like sensitivities for medical diagnostic ap-plications. Molecular amplification strategies such as presented in [245] or intrinsicsensor performance can enhance detection limits. Since the goal is label-free detec-tion, we aim to improve the iLoD for silicon photonic biosensors. The fundamentallimitations of the detection limits are introduced in section 2.2.To improve the intrinsic limit of detection, the sensitivity and the quality factorQ can be improved. However, as pointed out in chapter 2.2, both the sensitivity157and Q are dependent on the confinement factor Γ which is a measure of how muchof the electric field distribution resides in the core. The dependence is oppositein sign. This means that, as the modal overlap with the cladding increases thesensitivity, the line width of the resonator is also increased in the same time andthe detection limit remains constant. The theoretical limit of detection is reachedwhen the total loss of the resonator is dominated by the absorption loss of thecladding (Qtot ≈ Qclad). Figure 6.1 shows the detection limit as function of theintrinsic loss αi. The overall Q is given by Eq. 2.24. The semi-empirical relationσ ≈ ∆λ(4.5∗SNR0.25) as proposed byWhite et al. is used to relate the standard deviationof a spectral variation (σ) to the FWHM (Qnoise) [34]. Resonators with high Qiwould be preferred as the line width is then determined by the cladding absorption(or absorbing biomolecules). If αi < 100 m−1 the detection limit is constant as theloss is dominated by cladding absorption (see Fig. 6.1). An increase in sensitivityby changing the modal overlap with the analyte does not affect the iLoD as asimilar increase in line width is observed due to more loss in the cladding. Andhence the iLoD for TE and TMmode sensors at 1550 nm have the same theoreticallimit. The iLoD can be improved by an order of magnitude by switching to shorterwavelength (dashed green line in Fig. 6.1) as the water absorption is decreased(accompanied by slightly higher scattering losses) [32]. The iLoD is approachingthe fundamental limit if αi < 10 m−1. Practically this is challenging to achieveas propagation losses for straight waveguides are on the same order (3 dB/cm) forstrip waveguide dominated by sidewall scattering [96, 218, 243]). Of the 1.31 µmsensors mentioned in the previous chapters, the ring resonators (with additionalbending losses) did not yield improved iLoD’s, however, the TM Bragg gratingsoffered an iLoD improvement over the theoretical limit for 1.55 µm but not yetapproaching that for 1.31 µm.It is also important to identify and quantify sources of noise in order to estimatethe system’s overall detection limit. During sensing experiments, extrinsic factorssuch as temperature fluctuations or fluctuating alignment of the I/O fibers due tovibration, can generate noise and limit the absolute detection limit of the system.For low Q resonators the dominant noise source is typically amplitude noise whilefor high Q resonators temperature fluctuations become the dominant mechanism[34]. Depending on the measurement setup spectral resolution of either the laser158Figure 6.1: Detection limit as function of intrinsic loss (ai) in waveguide-based resonators employing TE and TM modes at 1.55 µm and 1.31 µmwavelengths. For αi < 100 m−1 the quality factor is limited by thecladding absorption and the intrinsic limit is constant. The iLoD can beimproved by an order of magnitude by switching to 1.31 µm.or the spectrometer can also impact the detection limit. Although rigorous noiseanalysis in SOI sensors for biosensing has not yet been presented in the literatureand would need to be addressed in the future, we can study the detection limit in thepresence of noise. Figure 6.1 shows the intrinsic detection limit with added noise(s = 1 pm, and SNR = 60 dB).Optimization using different sensor architectures and design tradeoffs can beused to minimize the intrinsic losses to maximize the obtainable detection limit.The losses that need to be considered result from: (1) bends which includes lossesfrom radiation and mode mismatch, (2) sidewall scattering, (3) loss resulting fromresonators not critically coupled to the waveguide bus, and (4) material absorptionand scattering (both within the silicon waveguide and in the aqueous cladding).These mechanisms can be limited by optimizing design parameters, and they ulti-mately impact the detection limit and performance of the biosensor.The detection limit of the sensors investigated are plotted in Fig. 6.2 along withsimilar, published devices. From Table 6.1, the theoretical detection limit in aque-159Figure 6.2: Comparison of the Q and S’ of the fabricated sensors. Blue andblack lines are plotted to indicate the theoretical, water-limited, iLoDsfor un-coupled (blue) and critically coupled resonators (black) at 1550nm.ous media for 1.55 µm wavelengths is 2.40 · 10−4 RIU. In Fig. 6.2 the theoreticallimit of detection for a wavelength of λ = 1550 nm is represented by the solid blueline (and green dashed line for λ = 1310 nm). Of the 1.55 µm wavelength sensorsinvestigated, the slot-waveguide Bragg grating had a iLoD of 3.04 ·10−4 RIU [193],followed by the 90 nm thick waveguide TE mode ring at 3.47 ·10−4 RIU, and finallythe TM mode in the 10 µm disk at 6.82 · 10−4 RIU. In general, these sensors hadhigher quality factors at 15,000, 24,700, and 16,000 respectively.While the 1.55 µm wavelength sensors come close to the theoretical detectionlimit of 2.40 · 10−4 RIU, one 1.31 µm wavelength sensor demonstrates a detectionsensitivity beyond this limit. The TM mode Bragg grating had a Q of 76,000yielding an iLoD of 1.62 ·10−4 RIU. The TM ring had an iLoD of 3.47 ·10−4 RIUand Q of 24,700. By employing strategies outlined above that would improve theQ and limit loss, this sensor could achieve greater sensitivities than the theoretical1601.55 µm wavelength limit.Because of the successful improvement of sensitivities and quality factors, aspresented in this work, most cases are not limited by optics anymore, but the nonspecific binding in relevant biological assays. [235]This additional step might be eliminated by employing strategies that resistfouling [13] and allow the detection of the target molecule in a label-free fashion.6.3 ContributionsThe list below highlight my personal contributions to the field based on the workpresented in this thesis.• Sub-wavelength grating ring resonators for biosensing Sub-wavelengthgrating ring resonators for biosensing offer greatly improved bulk sensitivity,a critical parameter for biosensing applications. The insight that led to thisis that the electric field will overlap with the aqueous cladding more in sucha structure. This has implications for both increased bulk, as well as surfacesensitivity. The SWG ring resonators were the first-of-their-kind and offeredan order of magnitude sensitivity improvement over the best commerciallyavailable silicon photonic biosensors and a 2x improvement over the bestreported TM mode ring resonators. Their performance enables new diag-nostic applications for silicon photonic devices label-free that could only beachieved previously using secondary amplification. The next steps should beto make these structures CMOS-foundry compatible. In addition, borrowedfrom bandstructure simulations for photonic crystals, I demonstrated a 3DFDTD simulation methodology. This is better than the equivalent effectiveindex approach as it simulates the actual field distribution in a SWG seg-ment and surface and bulk sensitivities calculations are more accurate as thefield overlap with the analyte can be determined. This simulation method-ology will help in further optimizing the grating and design an SWG sensorfor TM polarized light. The next steps should also include the design ofCMOS-foundry compatible SWG.• TMmodeBragg gratings for biosensing I conceived this ideawithXuWangafter my work with him on TE Bragg grating and my work on TMmode ring161resonators. These novel sensors offer a 3x bulk sensitivity improvementover their TE counterparts and higher limits of detections compared to ringresonators. The realization that a TM mode offers improved surface sen-sitivity originated from our work with TM ring resonators. In addition Irealized that the shape also limits bend losses and facilitates higher sensordensity on a single chip due to a smaller footprint. I realized all these thingswould be advantageous for a biosensing system. This achievement can helpfacilitate highly multiplexed, silicon photonic lab-on-chip systems in smallerfootprints and with better performing sensors when compared to ring res-onators. These devices were also designed and characterized for 1310 nmwavelengths which resulted in detection limits beyond what can be achievedwith 1550 nm wavelength devices. The next steps should be to investigatethese devices with low-cost laser diodes (or integrated the same chip) andtunable structures (like on-chip, waveguide heaters) to assess their potentialto further reduce the component cost of silicon photonic biosensing systems.• Suspended waveguide TE and TMmode Bragg gratings for biosensing Ideveloped this idea in collaboration with Shon Schmidt at the University ofWashington. We noticed that the sensitivity of TM mode sensors could begreatly improved by exposing the electric field that resides in the buried oxidelayer for on-substrate sensors to the aqueous cladding. These devices werefirst-of-their kind in this suspended configuration and demonstrated markedperformance improvements of Bragg gratings for biosensing applications,with the highest reported sensitivity for TM mode Bragg gratings. Becauseof their smaller footprint (compared to rings), these devices can be used forbiosensing applications that require better performance and higher integrationthan what can be achieved by commercially available systems today. Aninvention disclosure was filed with the University of Washington for thisstructure. Further work on suspended waveguide biosensors should includea comparative performance assessment at 1310 nm wavelengths.• Open-sourced, low-cost automated probe station for silicon photoniccircuit development I developed this system in collaboration with ShonSchmidt at the University of Washington. We architected the software, I162developed many of the algorithms, wrote much of the software, and helpedwith validation. While some argue as to the scientific merit of the work itself,this effort has enabled other research efforts world-wide, accelerating bothknowledge and devices for the field of silicon photonics. It has been usedto generate scores of publications and I co-founded a company, Maple LeafPhotonics LLC, to commercialize it beyond the academic environments. It iscurrently in use at more than a dozen industrial and academic research labs.Future work will involve adding more features and further reducing hardwarecosts to accelerate its adoption and standardization throughout the field.• Ebeam PDK Every silicon photonic design leverages basic componentsbuilding blocks which can be reused. Sharing these components broadly,with users in the field, reduces their design efforts and risk. Therefore, Icharacterized basic components (e.g. grating couplers) and worked withthe Washington Nanofabrication Facility (WNF) to offer them in form asimple PDK for groups fabricating silicon photonic devices through theirEbeam lithography process. In addition, biosensing experiments requiremany replicates to be tested and assessed, especially when developing assayprotocols. Since the setup and alignment of optical fibers to a chip can betime-consuming, I developed a GDS framework with alignment structures inspecific locations that help expedite the setup process. The PDK complimentsmechanical alignment structures in the probe station’s chip stage as well asspecific software algorithms to create a chip-wide coordinate system. Theseengineering efforts cannot be understated in how they improved the efficiencyof biosensor development and reduced design risk for WNF Ebeam users.Next steps involve growing the cell library of devices and characterizing theirperformance to build compact models for circuit simulation.• Biosensing chips for blood typing I designed and characterized foundrychips with biosensors for sensing applications in human blood. The impactof this work involves the demonstration of a new application on the siliconphotonic platform that is actively being commercialized at the University ofWashington. A manuscript demonstrating serological phenotype detectionis currently in preparation. Future work should involve scaling the system to163support hundreds of sensors on a single chip for highly multiplexed assays.6.4 Future directionsThe thesis focuses on improving the performance metrics of silicon photonicsbiosensors. However, in order to enable personal medicine through point of carediagnostics with the ability to quantitatively assess hundreds of biomarkers at thesame time, there are still major challenges to overcome. The transducer, in thiscase the optical detection, is only one aspect of an IVD platform. As silicon pho-tonic biosensors mature, focus will shift toward lab-on-chip devices with integratedfluidics, lasers, detectors, and electronic control and readout. While a few groupshave already demonstrated on-chip laser and detectors [37, 50, 230, 246–248], mostfluidic integration solutions require external pumps, tubing, valves, and flow cellsfor silicon photonic biosensing systems [12, 249–251]. Our collaborators at theUniversity of Washington are exploring the use of paper based fluidic networks toavoid the used of external pumps. The porous membrane network offers many ad-ditional optimizations that could further simplify and integrate the overall system.Membranes could be used to filter or separate constituents [252] eliminating theneed for any sample prep. When working in blood, membranes with larger porescould be employed to operate in samples with cells, such as typing erythrocyteantigens. Finally, complex 2D and 3D networks could facilitate the sequencing ofthe sample, rinse buffers, and any additional amplification chemistries to eliminatethe need for user intervention [253, 254]. Furthermore, given the temperature sen-sitivity of silicon waveguides, on chip temperature compensation could be realized,fully taking advantage of the compatibility of the fabrication processes of photonicchips with CMOS foundry processes. In a similar way, data averaging, signalamplification, and noise suppression could also be integrated monolithically on thesame chip.This thesis clearly shows that silicon photonics biosensors have potential foran IVD platform and that such sensors can provide the sensitivities and detectionlimits required for a relevant clinical assay. The adoption of a new technology isfacilitated when driven by a specific application. Our collaborators at the Universityof Washington in Seattle are currently using the designs outlined in this work for164serologic phenotyping. Another barrier for adoption is cost. While photonic chipsat scale can be fabricated at low cost, a big challenge at this point is the light source.Currently, the overall system cost of a silicon photonic biosensing platform isdominated by the laser (and detector) in comparison to the pumps, flow cell, tubing,and photonic chip. The experimental data presented throughout the document wasmeasured using an expensive tunable laser source. While that might be ok for abench top instrument (like the commercially available Genalyte Maverick system)a low cost light source would have to be implemented for a portable device (eitherexternal source or a Si/III-V hybrid implementation). There is growing evidencethat chip-bonded lasers and integrated detectors will be affordable at the chip-scalein the near future [37, 50, 230, 246–248, 255, 256]. The limitations by a low costsource may be overcome by compensating with additional passive or active opticalcomponents, or with integrated analog or digital signal processing.However, the biggest challenge lies not so much in finding solutions for theindividual aspects but in the integration of all of them into a single portable device.165Bibliography[1] The Lewin Group Inc. The value of diagnosis, 2009, report commissionedby advameddx. 2009. accessed: June 27, 2016. → pages 1, 2[2] JL Arlett, EB Myers, and ML Roukes. Comparative advantages ofmechanical biosensors. Nature nanotechnology, 6(4):203–215, 2011. →pages 1, 9, 150[3] Pierce protein biology products: Elisa and target-specific assay kits. URLhttp://www.piercenet.com/browse.cfm?fldID=BD0EF065-AD36-4337-B36B-A8000058840E. accessed: June 27, 2016.→ pages[4] John R Crowther. The ELISA guidebook, volume 149. Springer Science &Business Media, 2000. → pages 4[5] G Ya Wiederschain. The elisa guidebook. Biochemistry (Moscow), 74(9):1058–1058, 2009. → pages 1[6] Vladimir Gubala, Leanne F Harris, Antonio J Ricco, Ming X Tan, andDavid E Williams. Point of care diagnostics: status and future. Analyticalchemistry, 84(2):487–515, 2011. → pages 1[7] Curtis D Chin, Vincent Linder, and Samuel K Sia. Commercialization ofmicrofluidic point-of-care diagnostic devices. Lab on a Chip, 12(12):2118–2134, 2012. → pages 1[8] John Toussaint, David Krueger, Stephen M Shortell, Arnold Milstein, andDavid M Cutler. Aco model should encourage efficient care delivery. InHealthcare, volume 3, pages 150–152. Elsevier, 2015. → pages 1[9] Shara Rosen. Worldwide Market for IVD Tests. Kalorama Information,New York, 2012. → pages 1166[10] Piia Von Lode. Point-of-care immunotesting: approaching the analyticalperformance of central laboratory methods. Clinical biochemistry, 38(7):591–606, 2005. → pages 2[11] Jen Kok, Jimmy Ng, Stephen C Li, John Giannoutsos, Vineet Nayyar,Jonathan R Iredell, Dominic E Dwyer, and Sharon CA Chen. Evaluation ofpoint-of-care testing in critically unwell patients: comparison with clinicallaboratory analysers and applicability to patients with ebolavirus infection.Pathology, 47(5):405–409, 2015. → pages 2[12] Muzammil Iqbal, Martin A Gleeson, Bradley Spaugh, Frank Tybor,William G Gunn, Michael Hochberg, Tom Baehr-Jones, Ryan C Bailey, andL Cary Gunn. Label-free biosensor arrays based on silicon ring resonatorsand high-speed optical scanning instrumentation. Selected Topics inQuantum Electronics, IEEE Journal of, 16(3):654–661, 2010. → pages 2,6, 10, 78, 112, 125, 157, 164[13] James T Kirk, Norman D Brault, Tom Baehr-Jones, Michael Hochberg,Shaoyi Jiang, and Daniel M Ratner. Zwitterionic polymer-modified siliconmicroring resonators for label-free biosensing in undiluted humanplasma.Biosensors and Bioelectronics, 42:100–105, 2013. → pages 2, 9, 112, 161[14] Anthony PF Turner. Biosensors: sense and sensibility. Chemical SocietyReviews, 42(8):3184–3196, 2013. → pages 3[15] Leland C Clark and Champ Lyons. Electrode systems for continuousmonitoring in cardiovascular surgery. Annals of the New York Academy ofsciences, 102(1):29–45, 1962. → pages 3[16] Bo Liedberg, Claes Nylander, and Ingemar Lundström. Biosensing withsurface plasmon resonance - how it all started. Biosensors andBioelectronics, 10(8):i–ix, 1995. → pages 4[17] U Jonson and M Malmqvist. Real time biospecific interaction analysis.Adv. Biosens, 2:291–336, 1992. → pages 4[18] Elain Fu, Timothy Chinowsky, Kjell Nelson, Kyle Johnston, ThayneEdwards, Kristen Helton, Michael Grow, John W Miller, and Paul Yager.Spr imaging-based salivary diagnostics system for the detection of smallmolecule analytes. Annals of the New York Academy of Sciences, 1098(1):335–344, 2007. → pages 4[19] Jiri Homola. Surface plasmon resonance sensors for detection of chemicaland biological species. Chemical reviews, 108(2):462, 2008. → pages 32167[20] Jiri Homola, Ivo Koudela, and Sinclair S Yee. Surface plasmon resonancesensors based on diffraction gratings and prism couplers: sensitivitycomparison. Sensors and Actuators B: Chemical, 54(1):16–24, 1999. →pages 4[21] Timothy M Chinowsky, Scott D Soelberg, Paul Baker, Nathaneal RSwanson, Peter Kauffman, Anthony Mactutis, Michael S Grow, RobertAtmar, Sinclair S Yee, and Clement E Furlong. Portable 24-analyte surfaceplasmon resonance instruments for rapid, versatile biodetection. Biosensorsand Bioelectronics, 22(9):2268–2275, 2007. → pages 4[22] Matthew S Luchansky and Ryan C Bailey. Silicon photonic microringresonators for quantitative cytokine detection and t-cell secretion analysis.Analytical chemistry, 82(5):1975–1981, 2010. → pages 4, 43, 150[23] Jiri Homola, Sinclair S Yee, and Günter Gauglitz. Surface plasmonresonance sensors: review. Sensors and Actuators B: Chemical, 54(1):3–15, 1999. → pages 5[24] Xudong Fan, Ian M. White, Siyka I. Shopova, Hongying Zhu, Jonathan D.Suter, and Yuze Sun. Sensitive optical biosensors for unlabeled targets: Areview. Analytica Chimica Acta, 620(1-2):8–26, 2008. URLhttp://www.sciencedirect.com/science/article/pii/S0003267008009343. →pages 5, 6, 9[25] Richard A Soref, Joachim Schmidtchen, and Klaus Petermann. Largesingle-mode rib waveguides in gesi-si and si-on-sio 2. QuantumElectronics, IEEE Journal of, 27(8):1971–1974, 1991. → pages 6[26] Richard A Soref and Brian R Bennett. Kramers-kronig analysis ofelectro-optical switching in silicon. In CambridgeSymposium-Fiber/LASE’86, pages 32–37. International Society for Opticsand Photonics, 1987. → pages 6[27] Richard Soref. Silicon photonics technology: past, present, and future. InIntegrated Optoelectronic Devices 2005, pages 19–28. InternationalSociety for Optics and Photonics, 2005. → pages 6[28] IA Estrada, RW Burlingame, AP Wang, K Chawla, T Grove, J Wang,SO Southern, M Iqbal, LC Gunn, and MA Gleeson. Multiplex detection ofpathogen biomarkers in human blood, serum, and saliva using siliconphotonic microring resonators. In SPIE Sensing Technology+ Applications,168pages 94900E–94900E. International Society for Optics and Photonics,2015. → pages 6, 10, 151[29] María-José Bañuls, Rosa Puchades, and Ángel Maquieira. Chemicalsurface modifications for the development of silicon-based label-freeintegrated optical (io) biosensors: A review. Analytica chimica acta, 777:1–16, 2013. → pages 6[30] Andy Eu-Jin Lim, Junfeng Song, Qing Fang, Chao Li, Xiaoguang Tu, NingDuan, Kok Kiong Chen, Roger Poh-Cher Tern, and Tsung-Yang Liow.Review of silicon photonics foundry efforts. Selected Topics in QuantumElectronics, IEEE Journal of, 20(4):405–416, 2014. → pages 6[31] Wim Bogaerts, Peter De Heyn, Thomas Van Vaerenbergh, Katrien De Vos,Shankar Kumar Selvaraja, Tom Claes, Pieter Dumon, Peter Bienstman,Dries Van Thourhout, and Roel Baets. Silicon microring resonators. Laser& Photonics Reviews, 6(1):47–73, 2012. → pages 6, 24, 30, 78, 79, 80, 83[32] Lukas Chrostowski, Samantha Grist, Jonas Flueckiger, Wei Shi, Xu Wang,Eric Ouellet, Han Yun, Mitch Webb, Ben Nie, and Zhen Liang. Siliconphotonic resonator sensors and devices. In SPIE LASE, pages823620–823620. International Society for Optics and Photonics, 2012. →pages 6, 8, 15, 25, 27, 30, 31, 38, 40, 97, 108, 113, 122, 127, 137, 146, 148,158[33] Juejun Hu, Xiaochen Sun, Anu Agarwal, and Lionel C Kimerling. Designguidelines for optical resonator biochemical sensors. JOSA B, 26(5):1032–1041, 2009. → pages[34] Ian M White and Xudong Fan. On the performance quantification ofresonant refractive index sensors. Optics Express, 16(2):1020–1028, 2008.→ pages 6, 8, 31, 39, 122, 158[35] Matthew S Luchansky, Adam L Washburn, Melinda S McClellan, andRyan C Bailey. Sensitive on-chip detection of a protein biomarker in humanserum and plasma over an extended dynamic range using silicon photonicmicroring resonators and sub-micron beads. Lab on a Chip, 11(12):2042–2044, 2011. → pages 9, 43, 150[36] Nan Jokerst, Matthew Royal, Sabarni Palit, Lin Luan, Sulochana Dhar, andTalmage Tyler. Chip scale integrated microresonator sensing systems.Journal of biophotonics, 2(4):212–226, 2009. → pages 9, 150169[37] Nan M Jokerst, Luan Lin, Sabarni Palit, Matthew Royal, Sulochana Dhar,M Brooke, and Talmage Tyler. Progress in chip-scale photonic sensing.IEEE transactions on biomedical circuits and systems, 3(4):202–211, 2009.→ pages 9, 164, 165[38] Frank Vollmer and Stephen Arnold. Whispering-gallery-mode biosensing:label-free detection down to single molecules. Nature methods, 5(7):591–596, 2008. → pages 9[39] Dongliang Yin, David W Deamer, Holger Schmidt, John P Barber, andAaron R Hawkins. Single-molecule detection sensitivity using planarintegrated optics on a chip. Optics letters, 31(14):2136–2138, 2006. →pages 9[40] Matthew S Luchansky, Adam L Washburn, Teresa A Martin, MuzammilIqbal, L Cary Gunn, and Ryan C Bailey. Characterization of the evanescentfield profile and bound mass sensitivity of a label-free silicon photonicmicroring resonator biosensing platform. Biosensors and Bioelectronics,26(4):1283–1291, 2010. → pages 10, 97, 130, 135, 136, 155[41] Adam L Washburn, Matthew S Luchansky, Adrienne L Bowman, andRyan C Bailey. Quantitative, label-free detection of five protein biomarkersusing multiplexed arrays of silicon photonic microring resonators.Analytical chemistry, 82(1):69–72, 2009. → pages 10, 43, 150[42] Gilberto A Rodriguez, Shuren Hu, and Sharon M Weiss. Porous siliconring resonator for compact, high sensitivity biosensing applications. Opticsexpress, 23(6):7111–7119, 2015. → pages 10[43] Tomoya Taniguchi, Anna Hirowatari, Takeshi Ikeda, Masataka Fukuyama,Yoshiteru Amemiya, Akio Kuroda, and Shin Yokoyama. Detection ofantibody-antigen reaction by silicon nitride slot-ring biosensors usingprotein g. Optics Communications, 365:16–23, 2016. → pages 10[44] Tom Claes, J Girones Molera, Katrien De Vos, E Schachtb, Roel Baets, andPeter Bienstman. Label-free biosensing with a slot-waveguide-based ringresonator in silicon on insulator. Photonics Journal, IEEE, 1(3):197–204,2009. → pages 10, 119, 125, 127, 154[45] Katrien De Vos, Jordi Girones, Stepan Popelka, Etienne Schacht, RoelBaets, and Peter Bienstman. Soi optical microring resonator with poly(ethylene glycol) polymer brush for label-free biosensor applications.170Biosensors and Bioelectronics, 24(8):2528–2533, 2009. → pages 10, 77,84, 119[46] Katrien De Vos, Irene Bartolozzi, Etienne Schacht, Peter Bienstman, andRoel Baets. Silicon-on-insulator microring resonator forsensitive andlabel-free biosensing. Optics Express, 15(12):7610–7615, 2007. URLhttp://www.opticsexpress.org/abstract.cfm?URI=oe-15-12-7610. → pages10, 154[47] Carl Fredrik Carlborg, Kristinn Björgvin Gylfason, A Kaźmierczak, FabianDortu, MJ Banuls Polo, A Maquieira Catala, GM Kresbach, HansSohlström, Thomas Moh, Laurent Vivien, et al. A packaged opticalslot-waveguide ring resonator sensor array for multiplex label-free assays inlabs-on-chips. Lab on a Chip, 10(3):281–290, 2010. → pages 10[48] Mi Kyoung Park, Jack Sheng Kee, Jessie Yiying Quah, Vivian Netto,Junfeng Song, Qing Fang, Eric Mouchel La Fosse, and Guo-Qiang Lo.Label-free aptamer sensor based on silicon microring resonators. Sensorsand Actuators B: Chemical, 176:552–559, 2013. → pages 10[49] Shuren Hu, Yiliang Zhao, Kun Qin, Scott T Retterer, Ivan I Kravchenko,and Sharon M Weiss. Enhancing the sensitivity of label-free siliconphotonic biosensors through increased probe molecule density. ACSPhotonics, 1(7):590–597, 2014. → pages 10[50] Rene Heideman, Marcel Hoekman, and Erik Schreuder. Triplex-basedintegrated optical ring resonators for lab-on-a-chip and environmentaldetection. Selected Topics in Quantum Electronics, IEEE Journal of, 18(5):1583–1596, 2012. → pages 10, 164, 165[51] Carlos A Barrios, Maria Jose Banuls, Victoria Gonzalez-Pedro, Kristinn BGylfason, Benito Sanchez, Amadeu Griol, Angel Maquieira, HansSohlström, Miquel Holgado, and R Casquel. Label-free optical biosensingwith slot-waveguides. Optics letters, 33(7):708–710, 2008. → pages 9, 10,119, 125[52] Shinyoung Lee, Seok Chan Eom, Jee Soo Chang, Chul Huh, Gun YongSung, and Jung H Shin. Label-free optical biosensing using a horizontalair-slot sinx microdisk resonator. Optics express, 18(20):20638–20644,2010. → pages 10[53] Yi Zou, Swapnajit Chakravarty, David N Kwong, Wei-Cheng Lai,Xiaochuan Xu, Xiaohui Lin, Amir Hosseini, and Ray T Chen.171Cavity-waveguide coupling engineered high sensitivity silicon photoniccrystal microcavity biosensors with high yield. Selected Topics in QuantumElectronics, IEEE Journal of, 20(4):171–180, 2014. → pages 10[54] Lijun Huang, Huiping Tian, Jian Zhou, Qi Liu, Pan Zhang, and Yuefeng Ji.Label-free optical sensor by designing a high-q photonic crystal ring–slotstructure. Optics Communications, 335:73–77, 2015. → pages 10[55] Kiana Malmir, Hamidreza Habibiyan, and Hassan Ghafoorifard. Anultrasensitive optical label-free polymeric biosensor based on concentrictriple microring resonators with a central microdisk resonator. OpticsCommunications, 365:150–156, 2016. → pages 10[56] Adrián Fernández Gavela, Daniel Grajales Garcia, Jhonattan C Ramirez,and Laura M Lechuga. Last advances in silicon-based optical biosensors.Sensors, 16(3):285, 2016. → pages 11[57] Linghua Wang, Jun Ren, Xiuyou Han, Tom Claes, Xigao Jian, PeterBienstman, Roel Baets, Mingshan Zhao, and Geert Morthier. A label-freeoptical biosensor built on a low-cost polymer platform. Photonics Journal,IEEE, 4(3):920–930, 2012. → pages 11[58] Mengxi Ji, Heng Cai, Like Deng, Ying Huang, Qingzhong Huang, JinsongXia, Zhiyong Li, Jinzhong Yu, and Yi Wang. Enhanced parametricfrequency conversion in a compact silicon-graphene microring resonator.Optics express, 23(14):18679–18685, 2015. → pages 11[59] Siegfried Janz, Pavel Cheben, Andre Delage, Adam Densmore, andDan-Xia Xu. Silicon photonic wire waveguide biosensor, August 17 2010.US Patent 7,778,499. → pages 12[60] Adam Densmore, D-X Xu, Siegfried Janz, Philip Waldron, J Lapointe,Trevor Mischki, Gregory Lopinski, André Delâge, JH Schmid, and PavelCheben. Sensitive label-free biomolecular detection using thin siliconwaveguides. Advances in Optical Technologies, 2008, 2008. → pages 12,119[61] Aurel Ymeti, Jan Greve, Paul V Lambeck, Thijs Wink, Stephan WFM vanHövell, Tom AM Beumer, Robert R Wijn, Rene G Heideman, VinodSubramaniam, and Johannes S Kanger. Fast, ultrasensitive virus detectionusing a young interferometer sensor. Nano letters, 7(2):394–397, 2007. →pages 12172[62] Bernard H Schneider, John G Edwards, and Nile F Hartman. Hartmaninterferometer: versatile integrated optic sensor for label-free, real-timequantification of nucleic acids, proteins, and pathogens. Clinical chemistry,43(9):1757–1763, 1997. → pages 12[63] Kirill E Zinoviev, Ana Belén González-Guerrero, Carlos Domínguez, andLaura M Lechuga. Integrated bimodal waveguide interferometric biosensorfor label-free analysis. Journal of Lightwave Technology, 29(13):1926–1930, 2011. → pages 12[64] Ana Belén González Guerrero, Jordi Pascual i Gainza, Laura M Lechuga,et al. Bimodal waveguide interferometer device based on silicon photonicstechnology for label-free and high sensitive biosensing. PhD thesis. →pages[65] Daphné Duval, Ana Belén González-Guerrero, Stefania Dante, JohannOsmond, Rosa Monge, Luis J Fernández, Kirill E Zinoviev, CarlosDomínguez, and Laura M Lechuga. Nanophotonic lab-on-a-chip platformsincluding novel bimodal interferometers, microfluidics and gratingcouplers. Lab on a Chip, 12(11):1987–1994, 2012. → pages[66] Stefania Dante, Daphné Duval, Borja Sepúlveda, Ana BelenGonzález-Guerrero, José Ramón Sendra, and Laura M Lechuga.All-optical phase modulation for integrated interferometric biosensors.Optics Express, 20(7):7195–7205, 2012. → pages 12[67] Steve Blair and Yan Chen. Resonant-enhanced evanescent-wavefluorescence biosensing with cylindrical optical cavities. Applied Optics,40(4):570–582, 2001. → pages 15[68] Katrin Schmitt and Christian Hoffmann. High-refractive-index waveguideplatforms for chemical and biosensing. In Optical Guided-wave Chemicaland Biosensors I, pages 21–54. Springer, 2009. → pages 16, 194[69] Allan W Snyder and J Love. Optical waveguide theory, volume 190.Springer, 1983. → pages 17, 18, 22, 194, 198[70] Mode solutions - fde mode solver. URLhttps://www.lumerical.com/tcad-products/mode/. accessed: June 27, 2016.→ pages 17, 201[71] Zhaoming Zhu and Thomas Brown. Full-vectorial finite-difference analysisof microstructured optical fibers. Optics Express, 10(17):853–864, 2002.→ pages 17, 201173[72] Malvin Carl Teich and BEA Saleh. Fundamentals of photonics. John Wileyand Sons, Inc., 1991. → pages 18, 195[73] Amnon Yariv and Pochi Yeh. Photonics: Optical Electronics in ModernCommunications (the Oxford Series in Electrical and ComputerEngineering). Oxford University, 2006. → pages 18, 22, 80, 101, 103, 121,194[74] Adam Densmore, D-X Xu, Philip Waldron, Siegfried Janz, Pavel Cheben,J Lapointe, André Delâge, Boris Lamontagne, JH Schmid, and Edith Post.A silicon-on-insulator photonic wire based evanescent field sensor.Photonics Technology Letters, IEEE, 18(23):2520–2522, 2006. → pages20, 29, 90[75] Saurabh Mani Tripathi, Arun Kumar, Emmanuel Marin, and Jean-PierreMeunier. Bragg grating based biochemical sensor using submicron si/sio2waveguides for lab-on-a-chip applications: a novel design. Applied optics,48(23):4562–4567, 2009. → pages[76] Hong Ma, Alex K-Y Jen, and Larry R Dalton. Polymer-based opticalwaveguides: materials, processing, and devices. Advanced materials, 14(19):1339–1365, 2002. → pages 20[77] B E Little, J P Laine, and S T Chu. Surface-roughness-inducedcontradirectional coupling in ring and disk resonators. Optics letters, 22(1):4–6, Jan 1997. → pages 24[78] C Manolatou, MJ Khan, Shanhui Fan, Pierre R Villeneuve, HA Haus, andJD Joannopoulos. Coupling of modes analysis of resonant channeladd-drop filters. Quantum Electronics, IEEE Journal of, 35(9):1322–1331,1999. → pages[79] Miloš Popovic, Christina Manolatou, and Michael Watts.Coupling-induced resonance frequency shifts in coupled dielectricmulti-cavity filters. Optics Express, 14(3):1208–1222, 2006. → pages[80] Matthew Borselli, Thomas Johnson, and Oskar Painter. Beyond therayleigh scattering limit in high-q silicon microdisks: theory andexperiment. Optics express, 13(5):1515–30, Mar 2005. → pages 24[81] S Arnold, M Khoshsima, I Teraoka, S Holler, and F Vollmer. Shift ofwhispering-gallery modes in microspheres by protein adsorption. Opticsletters, 28(4):272–4, Feb 2003. → pages 25, 27174[82] Tom Baehr-Jones, Michael Hochberg, Chris Walker, Eric Chan, DennisKoshinz, William Krug, and Axel Scherer. Analysis of the tuningsensitivity of silicon-on-insulator optical ring resonators. Journal oflightwave technology, 23(12):4215, 2005. → pages 25, 27, 33[83] K Tiefenthaler and W Lukosz. Sensitivity of grating couplers asintegrated-optical chemical sensors. JOSA B, 6(2):209–220, 1989. →pages 25, 29, 199[84] GJ Veldhuis, O Parriaux, HJWM Hoekstra, and PV Lambeck. Sensitivityenhancement in evanescent optical waveguide sensors. Journal oflightwave technology, 18(5):677, 2000. → pages 28, 32[85] H Kogelnik. Theory of optical waveguides. In Guided-waveoptoelectronics, pages 7–88. Springer, 1988. → pages 29, 194[86] Iwao Teraoka and Stephen Arnold. Theory of resonance shifts in te and tmwhispering gallery modes by nonradial perturbations for sensingapplications. JOSA B, 23(7):1381–1389, 2006. → pages 29[87] Shon Schmidt, Jonas Flueckiger, WenXuan Wu, Samantha M Grist,Sahba Talebi Fard, Valentina Donzella, Pakapreud Khumwan, Emily RThompson, Qian Wang, and Pavel Kulik. Improving the performance ofsilicon photonic rings, disks, and bragg gratings for use in label-freebiosensing. In SPIE NanoScience+ Engineering, pages 91660M–91660M.International Society for Optics and Photonics, 2014. → pages 29, 52, 104,111, 112, 114, 130[88] Shankar Kumar Selvaraja. Wafer-scale fabrication technology for siliconphotonic integrated circuits. PhD thesis, PhD thesis, Ghent University,2011. → pages 30[89] Pieter Dumon. Ultra-compact integrated optical filters insilicon-on-insulator by means of wafer-scale technology. University ofGhent, Ph. D. Thesis, 2007. → pages 30[90] Kevin K. Lee, Desmond R. Lim, Hsin-Chiao Luan, Anuradha Agarwal,James Foresi, and Lionel C. Kimerling. Effect of size and roughness onlight transmission in a si/sio2 waveguide: Experiments and model. AppliedPhysics Letters, 77(11):1617–1619, 2000.doi:http://dx.doi.org/10.1063/1.1308532. URLhttp://scitation.aip.org/content/aip/journal/apl/77/11/10.1063/1.1308532. →pages 30175[91] Steven G Johnson, ML Povinelli, M Soljačić, Aristeidis Karalis, S Jacobs,and JD Joannopoulos. Roughness losses and volume-current methods inphotonic-crystal waveguides. Applied Physics B, 81(2-3):283–293, 2005.→ pages 30[92] Shankar Kumar Selvaraja, Patrick Jaenen, Wim Bogaerts, DriesVan Thourhout, Pieter Dumon, and Roel Baets. Fabrication of photonicwire and crystal circuits in silicon-on-insulator using 193-nm opticallithography. Lightwave Technology, Journal of, 27(18):4076–4083, 2009.→ pages 30[93] T Liang, L Nunes, T Sakamoto, K Sasagawa, T Kawanishi, M Tsuchiya,Gino Priem, Dries Van Thourhout, Pieter Dumon, Roel Baets, et al.Ultrafast all-optical switching by cross-absorption modulation in siliconwire waveguides. Optics Express, 13(19):7298–7303, 2005. → pages 30[94] DJ Moss, L Fu, I Littler, and BJ Eggleton. Ultrafast all-optical modulationvia two-photon absorption in silicon-on-insulator waveguides. Electron.Lett, 41(6):320–321, 2005. → pages 30[95] Tom Baehr-Jones, Michael Hochberg, and Axel Scherer. Photodetection insilicon beyond the band edge with surface states. Optics express, 16(3):1659–1668, 2008. → pages 30[96] Lukas Chrostowski and Michael Hochberg. Silicon Photonics Design:From Devices to Systems. Cambridge University Press, 2015. → pages 30,31, 41, 82, 91, 97, 158[97] Olivier Parriaux and GJ Veldhuis. Normalized analysis for the sensitivityoptimization of integrated optical evanescent-wave sensors. Journal oflightwave technology, 16(4):573, 1998. → pages 32, 33[98] Olivier Parriaux and P Dierauer. Normalized expressions for the opticalsensitivity of evanescent wave sensors. Optics letters, 19(7):508–510, 1994.→ pages 32, 33[99] JJ Ackert, JK Doylend, DF Logan, PE Jessop, R Vafaei, L Chrostowski, andAP Knights. Defect-mediated resonance shift of silicon-on-insulatorracetrack resonators. Optics express, 19(13):11969–11976, 2011. → pages33[100] De-Kui Qing and Ichirou Yamaguchi. Analysis of the sensitivity of opticalwaveguide chemical sensors for tm modes by the group-index method.JOSA B, 16(9):1359–1369, 1999. → pages 33176[101] Janos Vörös. The density and refractive index of adsorbing protein layers.Biophysical journal, 87(1):553–561, 2004. → pages 36, 54, 108, 114, 115,116, 145, 155[102] F Höök, J Vörös, M Rodahl, R Kurrat, P Böni, JJ Ramsden, M Textor,ND Spencer, Pentti Tengvall, J Gold, et al. A comparative study of proteinadsorption on titanium oxide surfaces using in situ ellipsometry, opticalwaveguide lightmode spectroscopy, and quartz crystalmicrobalance/dissipation. Colloids and Surfaces B: Biointerfaces, 24(2):155–170, 2002. → pages 36, 109, 155[103] Niels Asger Mortensen, Sanshui Xiao, and Jesper Pedersen.Liquid-infiltrated photonic crystals: enhanced light-matter interactions forlab-on-a-chip applications. Microfluidics and Nanofluidics, 4(1-2):117–127, 2008. → pages 39[104] Linhong Kou, Daniel Labrie, and Petr Chylek. Refractive indices of waterand ice in the 0.65-to 2.5-Âţm spectral range. Applied Optics, 32(19):3531–3540, 1993. → pages 40, 146[105] Edward D Palik. Handbook of optical constants of solids, volume 3.Academic press, 1998. → pages 40[106] Bradley J Frey, Douglas B Leviton, and Timothy J Madison.Temperature-dependent refractive index of silicon and germanium. In SPIEAstronomical Telescopes+ Instrumentation, pages 62732J–62732J.International Society for Optics and Photonics, 2006. → pages 41[107] G Cocorullo and I Rendina. Thermo-optical modulation at 1.5 µm insilicon etalon. Electronics Letters, 28(1):83–85, 1992. → pages[108] JA McCaulley, VM Donnelly, M Vernon, and I Taha. Temperaturedependence of the near-infrared refractive index of silicon, galliumarsenide, and indium phosphide. Physical Review B, 49(11):7408, 1994. →pages 41[109] Zeqin Lu, Kyle Murray, Hasitha Jayatilleka, and Lukas Chrostowski.Michelson interferometer thermo-optic switch on soi with a 50-µw powerconsumption. Photonics Technology Letters, IEEE, 27(22):2319–2322,2015. → pages 41[110] Kyle Murray, Zeqin Lu, Hasitha Jayatilleka, and Lukas Chrostowski. Densedissimilar waveguide routing for highly efficient thermo-optic switches onsilicon. Optics express, 23(15):19575–19585, 2015. → pages 41177[111] Kishore Padmaraju and Keren Bergman. Resolving the thermal challengesfor silicon microring resonator devices. Nanophotonics, 3(4-5):269–281,2014. → pages 41[112] Jie Teng, Pieter Dumon, Wim Bogaerts, Hongbo Zhang, Xigao Jian,Xiuyou Han, Mingshan Zhao, Geert Morthier, and Roel Baets. Athermalsilicon-on-insulator ring resonators by overlaying a polymer cladding onnarrowed waveguides. Optics express, 17(17):14627–14633, 2009. →pages 41[113] Jong-Moo Lee, Duk-Jun Kim, Hokyun Ahn, Sang-Ho Park, and GyungockKim. Temperature dependence of silicon nanophotonic ring resonator witha polymeric overlayer. Journal of lightwave technology, 25(8):2236–2243,2007. → pages[114] Jong-Moo Lee, Duk-Jun Kim, Gwan-Ha Kim, O-Kyun Kwon, Kap-JoongKim, and Gyungock Kim. Controlling temperature dependence of siliconwaveguide using slot structure. Optics express, 16(3):1645–1652, 2008. →pages[115] N Ye Winnie, Rong Sun, Jurgen Michel, Louay Eldada, Deepti Pant, andLionel C Kimerling. Thermo-optical compensation in high-index-contrastwaveguides. 2008. → pages[116] Linjie Zhou, Ken Kashiwagi, Katsunari Okamoto, RP Scott, NK Fontaine,Dan Ding, Venkatesh Akella, and SJB Yoo. Towards athermaloptically-interconnected computing system using slotted silicon microringresonators and rf-photonic comb generation. Applied Physics A, 95(4):1101–1109, 2009. → pages[117] Mutsunori Uenuma and Teruaki Motooka. Temperature-independent siliconwaveguide optical filter. Optics letters, 34(5):599–601, 2009. → pages 41[118] Alexey N Bashkatov and Elina A Genina. Water refractive index independence on temperature and wavelength: a simple approximation. InSaratov Fall Meeting 2002: Optical Technologies in Biophysics andMedicine IV, pages 393–395. International Society for Optics andPhotonics, 2003. → pages 41[119] G Abbate, U Bernini, E Ragozzino, and F Somma. The temperaturedependence of the refractive index of water. Journal of Physics D: AppliedPhysics, 11(8):1167, 1978. → pages 41178[120] Benedetto Troia, Antonia Paolicelli, Francesco De Leonardis, andVittorio MN Passaro. Photonic crystals for optical sensing: A review.2013. → pages 43, 150[121] Frank Vollmer and Lan Yang. Review label-free detection with high-qmicrocavities: a review of biosensing mechanisms for integrated devices.Nanophotonics, 1(3-4):267–291, 2012. → pages 43, 150[122] Jeffrey Kirsch, Christian Siltanen, Qing Zhou, Alexander Revzin, andAleksandr Simonian. Biosensor technology: recent advances in threatagent detection and medicine. Chemical Society Reviews, 42(22):8733–8768, 2013. → pages 43, 150[123] Siepic ebeam pdk. URL http://siepic.ubc.ca/GSiP. accessed: June 27th,2016. → pages 43[124] Silicon photonic test bench, . URLhttps://github.com/shonschmidt/SiPhoTestBench. accessed: June 27th,2016. → pages 44[125] Silicon photonic analysis tool, . URLhttps://github.com/shonschmidt/SiPhoAnalysisTool. accessed: June 27th,2016. → pages 44, 221[126] M Gnan, DS Macintyre, M Sorel, RM De La Rue, and S Thoms. Enhancedstitching for the fabrication of photonic structures by electron beamlithography. Journal of Vacuum Science & Technology B, 25(6):2034–2037, 2007. → pages 44, 50, 129[127] Richard J Bojko, Jing Li, Li He, Tom Baehr-Jones, Michael Hochberg, andYukinori Aida. Electron beam lithography writing strategies for low loss,high confinement silicon optical waveguides. Journal of Vacuum Science &Technology B, 29(6):06F309, 2011. → pages 44, 50, 107, 109, 120, 129,141[128] Uw nnin washington nanofabrication facility, . URLhttps://www.wnf.washington.edu/. accessed: June 27th, 2016. → pages 46,49[129] Jeol jbx-6300fs e-beam lithography at the washington nanofabricationfacility, . URL https://ebeam.mff.uw.edu/ebeamweb/. accessed: June 27th,2016. → pages 46179[130] Xu Wang, Wei Shi, Michael Hochberg, Kostas Adam, Ellen Schelew, Jeff FYoung, Nicolas AF Jaeger, and Lukas Chrostowski. Lithography simulationfor the fabrication of silicon photonic devices with deep-ultravioletlithography. In IEEE Group IV Photonics Conference (San Diego, CA, USA2012), paper ThP17, 2012. → pages 46, 124, 128[131] James Pond, Chris Cone, Lukas Chrostowski, Jackson Klein, JonasFlueckiger, Amy Liu, Dylan McGuire, and Xu Wang. A complete designflow for silicon photonics. In SPIE Photonics Europe, pages913310–913310. International Society for Optics and Photonics, 2014. →pages 46, 48[132] Singapore A-Star IME foundry. Foundry, singapore a-star ime. URLhttps://www.a-star.edu.sg/ime/. accessed: June 27th, 2016. → pages 48, 49[133] epix foundry. URL http://www.epixfab.eu/. accessed: June 27th, 2016. →pages 120[134] Europractice silicon photonics mpw. URLhttp://www.europractice-ic.com/SiPhotonics\_general.php. accessed: June27th, 2016. → pages[135] Moses silicon photonics mpw. URLhttps://www.mosis.com/vendors/view/imec. accessed: June 27th, 2016. →pages[136] Michael Hochberg and Tom Baehr-Jones. Towards fabless siliconphotonics. Nature Photonics, 4(8):492–494, 2010. → pages 48, 49[137] Lukas Chrostowski, Jonas Flueckiger, Charlie Lin, Michael Hochberg,James Pond, Jackson Klein, John Ferguson, and Chris Cone. Designmethodologies for silicon photonic integrated circuits. In SPIE OPTO,pages 89890G–89890G. International Society for Optics and Photonics,2014. → pages 48[138] Xu Wang, James Pond, Chris Cone, Lukas Chrostowski, Jackson Klein,Jonas Flueckiger, Amy Liu, and Dylan McGuire. Large-scale siliconphotonics circuit design. In SPIE/COS Photonics Asia, pages927706–927706. International Society for Optics and Photonics, 2014. →pages 48, 55[139] Silicon photonics group at cornell, . URLhttp://nanophotonics.ece.cornell.edu/research/silicon-photonics.cfm.accessed: June 27th, 2016. → pages 49180[140] Silicon photonics at ucsb, . URL http://siliconphotonics.ece.ucsb.edu/.accessed: June 27th, 2016. → pages 49[141] Stefano Sardo, Fabrizio Giacometti, Sergio Doneda, Umberto Colombo,Melissa Di Muri, Anna Donghi, Romano Morson, Giorgio Mutinati,Alessandro Nottola, Massimo Gentili, et al. Line edge roughness (ler)reduction strategy for soi waveguides fabrication. MicroelectronicEngineering, 85(5):1210–1213, 2008. → pages 50, 129[142] Yao Chen, Junbo Feng, Zhiping Zhou, ChristopherJ Summers, DavidSCitrin, and Jun Yu. Simple technique to fabricate microscale and nanoscalesilicon waveguide devices. Frontiers of Optoelectronics in China, 2(3):308–311, 2009. URL http://dx.doi.org/10.1007/s12200-009-0049-1. →pages 50, 129[143] Yao Chen, Junbo Feng, Zhiping Zhou, Jun Yu, Christopher J Summers, andDavid S Citrin. Fabrication of silicon microring resonator with smoothsidewalls. Journal of Micro/Nanolithography, MEMS, and MOEMS, 8(4):043060–043060, 2009. → pages 50, 129[144] Asahi Glass Co. Ltd. Cytop. Technical report, AGC Chemicals, 2009. →pages 51, 130[145] Raza Hanif. Microfabrication of Plasmonic Biosensors in CYTOPIntegrating a Thin SiO2 Diffusion and Etch-barrier Layer. PhD thesis,University of Ottawa, 2011. → pages 51, 130[146] Cheng Zhenzhou, Wong Chi Yan, Chen Xia, Xu Ke, Y. M. Chen, and TsangHon Ki. Monolithic suspended membrane ring resonator for mid-infraredapplications. pages 87–89, 2012. → pages 51, 139[147] Cheng Zhenzhou, Chen Xia, C. Y. Wong, Xu Ke, C. K. Y. Fung, Y. M.Chen, and Tsang Hon Ki. Mid-infrared suspended membrane waveguideson silicon-on-insulator. pages 257–258, 2012. → pages[148] Christopher Batten, Ajay Joshi, Jason Orcutt, Anatoly Khilo, BenjaminMoss, Charles W Holzwarth, MiloÅą A Popovic, Hanqing Li, Henry ISmith, and Judy L Hoyt. Building many-core processor-to-dram networkswith monolithic cmos silicon photonics. Micro, IEEE, 29(4):8–21, 2009.→ pages[149] Shankar Kumar Selvaraja, Patrick Jaenen, Wim Bogaerts, DriesVan Thourhout, Pieter Dumon, and Roel Baets. Fabrication of photonic181wire and crystal circuits in silicon-on-insulator using 193-nm opticallithography. Lightwave Technology, Journal of, 27(18):4076–4083, 2009.→ pages 51, 139[150] RL Sokoloff, KC Norton, CL Gasior, KM Marker, and LS Grauer. AdualâĂŘmonoclonal sandwich assay for prostateâĂŘspecific membraneantigen: Levels in tissues, seminal fluid and urine. The Prostate, 43(2):150–157, 2000. → pages 53[151] U.B. Nielsen and B.H. Geierstanger. Multiplexed sandwich assays inmicroarray format. Journal of immunological methods, 290(1):107–120,2004. → pages 53, 137[152] Marc Graille, Enrico A Stura, Adam L Corper, Brian J Sutton, Michael JTaussig, Jean-Baptiste Charbonnier, and Gregg J Silverman. Crystalstructure of a staphylococcus aureus protein a domain complexed with thefab fragment of a human igm antibody: structural basis for recognition ofb-cell receptors and superantigen activity. Proceedings of the NationalAcademy of Sciences, 97(10):5399–5404, 2000. → pages 54[153] Harold Erickson. Size and shape of protein molecules at the nanometerlevel determined by sedimentation, gel filtration, and electron microscopy.Biological Procedures Online, 11(1):32 – 51, 2009. URLhttp://www.biologicalproceduresonline.com/content/11/1/32. → pages 54,145[154] H. Elwing. Protein absorption and ellipsometry in biomaterial research.Biomaterials, 19(4-5):397–406, 1998. → pages 54, 145[155] K.P.S. Dancil, D.P. Greiner, and M.J. Sailor. A porous silicon opticalbiosensor: detection of reversible binding of igg to a protein a-modifiedsurface. Journal of the American Chemical Society, 121(34):7925–7930,1999. → pages 54, 137, 145[156] Martine Collaud Coen, Roland Lehmann, Pierangelo Gröning, MichaelBielmann, Carine Galli, and Louis Schlapbach. Adsorption and bioactivityof protein a on silicon surfaces studied by afm and xps. Journal of colloidand interface science, 233(2):180–189, 2001. → pages 54, 100, 108, 114,115, 116, 137, 145, 156[157] Yun Wang, Xu Wang, Jonas Flueckiger, Han Yun, Wei Shi, Richard Bojko,Nicolas AF Jaeger, and Lukas Chrostowski. Focusing sub-wavelengthgrating couplers with low back reflections for rapid prototyping of silicon182photonic circuits. Optics express, 22(17):20652–20662, 2014. → pages 55,101, 110, 130[158] Lukas Chrostowski, Xu Wang, Jonas Flueckiger, Yichen Wu, Yun Wang,and Sahba Talebi Fard. Impact of fabrication non-uniformity on chip-scalesilicon photonic integrated circuits. In Optical Fiber CommunicationConference, pages Th2A–37. Optical Society of America, 2014. → pages57[159] Samantha M Grist, Shon A Schmidt, Jonas Flueckiger, Valentina Donzella,Wei Shi, Sahba Talebi Fard, James T Kirk, Daniel M Ratner, Karen CCheung, and Lukas Chrostowski. Silicon photonic micro-disk resonatorsfor label-free biosensing. Optics express, 21(7):7994–8006, 2013. → pages71, 97, 114, 116, 156[160] Jan-Willem Hoste, Sam Werquin, Tom Claes, and Peter Bienstman.Conformational analysis of proteins with a dual polarisation siliconmicroring. Optics express, 22(3):2807–2820, 2014. → pages 77, 109, 119[161] Wei Shi, Han Yun, Charlie Lin, Jonas Flueckiger, Nicolas AF Jaeger, andLukas Chrostowski. Coupler-apodized bragg-grating addâĂŞdrop filter.Optics letters, 38(16):3068–3070, 2013. → pages 78[162] Wei Shi, Xu Wang, Wen Zhang, Han Yun, Charlie Lin, Lukas Chrostowski,and Nicolas AF Jaeger. Grating-coupled silicon microring resonators.Applied Physics Letters, 100(12):121118, 2012. → pages 78[163] Sahba Talebi Fard, Valentina Donzella, Shon A Schmidt, Jonas Flueckiger,Samantha M Grist, Pouria Talebi Fard, Yichen Wu, Rick J Bojko, EzraKwok, Nicolas AF Jaeger, et al. Performance of ultra-thin soi-basedresonators for sensing applications. Optics Express, 22(12):14166–14179,2014. → pages 79, 93, 96, 119[164] Wei-Ping Huang. Coupled-mode theory for optical waveguides: anoverview. JOSA A, 11(3):963–983, 1994. → pages 80[165] Amnon Yariv. Coupled-mode theory for guided-wave optics. QuantumElectronics, IEEE Journal of, 9(9):919–933, 1973. → pages 80[166] D. G. Rabus. Integrated Ring Resonators. Number 127. Springer, 2007. →pages 82183[167] Nicolas Rouger, Lukas Chrostowski, and Raha Vafaei. Temperature effectson silicon-on-insulator (soi) racetrack resonators: A coupled analytic and2-d finite difference approach. Journal of Lightwave Technology, 28(9):1380–1391, 2010. → pages 82[168] Mao Mao, Sitao Chen, and Daoxin Dai. Cascaded ring-resonators formulti-channel optical sensing with reduced temperature sensitivity. IEEEPhotonics Technology Letters, 28(7):814–817, 2016. → pages 84[169] Valery V Tuchin, Irina L Maksimova, Dmitry A Zimnyakov, Irina L Kon,Albert H Mavlyutov, and Alexey A Mishin. Light propagation in tissueswith controlled optical properties. Journal of biomedical optics, 2(4):401–417, 1997. → pages 95[170] Robert W Boyd and John E Heebner. Sensitive disk resonator photonicbiosensor. Applied Optics, 40(31):5742–5747, 2001. → pages 97[171] Aaron Schweinsberg, Sandrine Hocdé, Nick N. Lepeshkin, Robert W.Boyd, Christopher Chase, and Julie E. Fajardo. An environmental sensorbased on an integrated optical whispering gallery mode disk resonator.Sensors and Actuators B: Chemical, 123(2):727–732, 2007. → pages 97[172] Carlos Angulo Barrios, Benito Sánchez, Kristinn BjÃűrgvin Gylfason,Amadeu Griol, Hans SohlstrÃűm, Miquel Holgado, and Raphael Casquel.Demonstration of slot-waveguide structures on silicon nitride/silicon oxideplatform. Optics express, 15(11):6846–6856, 2007. → pages 97, 125[173] Jonas Flueckiger, Samantha M Grist, Gurpal Bisra, Lukas Chrostowski, andKaren C Cheung. Cascaded silicon-on-insulator microring resonators forthe detection of biomolecules in pdms microfluidic channels. In SPIEMOEMS-MEMS, pages 79290I–79290I. International Society for Opticsand Photonics, 2011. → pages 98[174] AL Weisenhorn, F.J. Schmitt, W. Knoll, and PK Hansma. Streptavidinbinding observed with an atomic force microscope. Ultramicroscopy, 42:1125–1132, 1992. → pages 99, 128, 145[175] S. A. Darst, M. Ahlers, P. H. Meller, E. W. Kubalek, R. Blankenburg, H. O.Ribi, H. Ringsdorf, and R. D. Kornberg. Two-dimensional crystals ofstreptavidin on biotinylated lipid layers and their interactions withbiotinylated macromolecules. Biophys J, 59(2):387–96, 1991. URLhttp://www.ncbi.nlm.nih.gov/pubmed/1901232. → pages 99, 128, 145184[176] Przemek J. Bock, Pavel Cheben, Jens H. Schmid, Jean Lapointe, AndréDelâge, Dan-Xia Xu, Siegfried Janz, Adam Densmore, and Trevor J. Hall.Subwavelength grating crossings for silicon wire waveguides. OpticsExpress, 18(15):16146–16155, 2010. URLhttp://www.opticsexpress.org/abstract.cfm?URI=oe-18-15-16146. → pages101[177] Robert Halir, Przemek J Bock, Pavel Cheben, Alejandro Ortega-Moñux,Carlos Alonso-Ramos, Jens H Schmid, Jean Lapointe, Dan-Xia Xu,J Gonzalo Wangüemert-Pérez, and Íñigo Molina-Fernández. Waveguidesub-wavelength structures: a review of principles and applications. Laserand Photonics Reviews, 9(1):25–49, 2015. → pages 105[178] P Cheben, PJ Bock, JH Schmid, J Lapointe, S Janz, D-X Xu, R Ma,A Densmore, A Delâge, and B Lamontagne. Subwavelength and diffractivewaveguide structures and their applications in nanophotonics and sensing.In SPIE OPTO, pages 794111–794111. International Society for Optics andPhotonics, 2011. → pages 101[179] Tymon Barwicz, Alexander Janta-Polczynski, Marwan Khater, YanThibodeau, Robert Leidy, Jeffrey Maling, Stephan Martel, SebastianEngelmann, Jason S Orcutt, and Paul Fortier. An o-band metamaterialconverter interfacing standard optical fibers to silicon nanophotonicwaveguides. In Optical Fiber Communication Conference, pages Th3F–3.Optical Society of America, 2015. → pages 101[180] P. Cheben, D. X. Xu, S. Janz, and A. Densmore. Subwavelength waveguidegrating for mode conversion and light coupling in integrated optics. OpticsExpress, 14(11):4695–4702, 2006. URLhttp://www.opticsexpress.org/abstract.cfm?URI=oe-14-11-4695. → pages[181] Yun Wang, Wei Shi, Xu Wang, Jonas Flueckiger, Han Yun, Nicolas AFJaeger, and Lukas Chrostowski. Fully etched grating coupler with low backreflection. In Photonics North 2013, pages 89150U–89150U. InternationalSociety for Optics and Photonics, 2013. → pages 101[182] Uriel Levy, Maxim Abashin, Kazuhiro Ikeda, Ashok Krishnamoorthy, JohnCunningham, and Yeshaiahu Fainman. Inhomogenous dielectricmetamaterials with space-variant polarizability. Physical review letters, 98(24):243901, 2007. → pages 101[183] Junjia Wang, Ivan Glesk, and Lawrence R Chen. Subwavelength gratingfiltering devices. Optics express, 22(13):15335–15345, 2014. → pages 101185[184] Soha Yousuf, Mahmoud M. Shahin, Hayk Gevorgyan, Bob Voort, AyatTaha, Clara Dimas, Marcus Dahlem, and Anatol Khilo. Suspendedmicroring resonator sensor using internal sub-wavelength grating. InAdvanced Photonics 2015, page SeS4C.4. Optical Society of America,2015. doi:10.1364/SENSORS.2015.SeS4C.4. URLhttp://www.osapublishing.org/abstract.cfm?URI=Sensors-2015-SeS4C.4.→ pages 101[185] J Gonzalo Wangüemert-Pérez, Pavel Cheben, Alejandro Ortega-Moñux,Carlos Alonso-Ramos, Diego Pérez-Galacho, Robert Halir, IñigoMolina-Fernández, Dan-Xia Xu, and Jens H Schmid. Evanescent fieldwaveguide sensing with subwavelength grating structures insilicon-on-insulator. Optics letters, 39(15):4442–4445, 2014. → pages 101[186] Valentina Donzella, Ahmed Sherwali, Jonas Flueckiger, Samantha M Grist,Sahba Talebi Fard, and Lukas Chrostowski. Design and fabrication of soimicro-ring resonators based on sub-wavelength grating waveguides. OpticsExpress, 23(4):4791–4803, 2015. → pages 101, 104[187] SM Rytov. Electromagnetic properties of a finely stratified medium. SovietPhysics JETP-USSR, 2(3):466–475, 1956. → pages 101[188] John D Joannopoulos, Steven G Johnson, Joshua N Winn, and Robert DMeade. Photonic Crystals: Molding the Flow of Light. PrincetonUniversity, 2011. → pages 101, 103[189] Z. Weissman and I. Hendel. Analysis of periodically segmented waveguidemode expanders. Journal of Lightwave Technology, 13(10):2053–2058,1995. → pages 101, 104[190] Przemek J Bock, Pavel Cheben, Jens H Schmid, Jean Lapointe, AndréDelâge, Siegfried Janz, Geof C Aers, Dan-Xia Xu, Adam Densmore, andTrevor J Hall. Subwavelength grating periodic structures insilicon-on-insulator: a new type of microphotonic waveguide. Opticsexpress, 18(19):20251–20262, 2010. → pages 102, 107[191] Jonathan P Dowling and Charles M Bowden. Anomalous index ofrefraction in photonic bandgap materials. Journal of modern Optics, 41(2):345–351, 1994. → pages 103[192] Kazuaki Sakoda. Optical properties of photonic crystals, volume 80.Springer Science & Business Media, 2004. → pages 103, 105186[193] Xu Wang, Jonas Flueckiger, Shon Schmidt, Samantha Grist, Sahba T. Fard,James Kirk, Matt Doerfler, Karen C. Cheung, Daniel M. Ratner, and LukasChrostowski. A silicon photonic biosensor using phase-shifted bragggratings in slot waveguide. Journal of Biophotonics, 6(10):821–828, 2013.URL http://dx.doi.org/10.1002/jbio.201300012. → pages 103, 112, 128, 160[194] John D. Bierlein. Propagation in segmented waveguide structures. InQuantum Electronics and Laser Science Conference, page QTuE2. OpticalSociety of America, 1992. URLhttp://www.opticsinfobase.org/abstract.cfm?URI=QELS-1992-QTuE2. →pages 104[195] Z Weissman. Evanescent field sensors with periodically segmentedwaveguides. Applied optics, 36(6):1218–1222, 1997. → pages 104[196] Lifeng Li and James J. Burke. Linear propagation characteristics ofperiodically segmented waveguides. Optics Letters, 17(17):1195–1197,1992. URL http://ol.osa.org/abstract.cfm?URI=ol-17-17-1195. → pages 104[197] Valentina Donzella, Ahmed Sherwali, Jonas Flueckiger, Sahba Talebi Fard,Samantha M Grist, and Lukas Chrostowski. Sub-wavelength gratingcomponents for integrated optics applications on soi chips. Optics express,22(17):21037–21050, 2014. → pages 104, 109[198] Pavel Kwiecien, Ivan Richter, and Jiří Čtyroky`. Comparison of 2d and 3dfourier modal methods for modeling subwavelength-structured siliconwaveguides. In Photonics Prague 2011, pages 83060Y–83060Y.International Society for Optics and Photonics, 2011. → pages 105[199] Jiri Ctyroky, Pavel Kwiecien, and Ivan Richter. Dispersion properties ofsubwavelength grating soi waveguides. In Progress in electromagneticsresearch symposium proceedings, page 1415, 2013. → pages 105[200] Marko Lončar, Theodor Doll, Jelena Vučković, and Axel Scherer. Designand fabrication of silicon photonic crystal optical waveguides. J. LightwaveTechnol., 18(10):1402, Oct 2000. URLhttp://jlt.osa.org/abstract.cfm?URI=jlt-18-10-1402. → pages 105[201] Xu Wang, Yun Wang, Jonas Flueckiger, Richard Bojko, Amy Liu, AdamReid, James Pond, Nicolas AF Jaeger, and Lukas Chrostowski. Precisecontrol of the coupling coefficient through destructive interference insilicon waveguide bragg gratings. Optics letters, 39(19):5519–5522, 2014.→ pages 105, 121, 128187[202] Petr I Nikitin, BG Gorshkov, EP Nikitin, and TI Ksenevich. Picoscope, anew label-free biosensor. Sensors and Actuators B: Chemical, 111:500–504, 2005. → pages 109, 155[203] RJ Bojko and BJ Hughes. Quantitative lithographic performance ofproximity correction for electron beam lithography. Journal of VacuumScience & Technology B, 8(6):1909–1913, 1990. → pages 109[204] I.; Sjöquist; J.; Mota, Gabriela; Moraru and V. Ghetie. Protein a as amolecular probe for the detection of antigen induced conformational changein fc region of rabbit antibody. Molecular immunology, 18(5):373–378,1981. → pages 114, 115, 145[205] Kirill Zinoviev, Laura G. Carrascosa, José Sánchez del Río, BorjaSepúlveda, Carlos Domínguez, and Laura M. Lechuga. Silicon photonicbiosensors for lab-on-a-chip applications. 2008. URLhttp://www.hindawi.com/journals/aot/2008/383927/. → pages 120[206] AS Jugessur, J Dou, JS Aitchison, RM De La Rue, and M Gnan. Aphotonic nano-bragg grating device integrated with microfluidic channelsfor bio-sensing applications. Microelectronic Engineering, 86(4):1488–1490, 2009. → pages 120, 123[207] P Prabhathan, VM Murukeshan, Zhang Jing, and Pamidighantam VRamana. Compact soi nanowire refractive index sensor using phase shiftedbragg grating. Optics express, 17(17):15330–15341, 2009. → pages 120,123, 128, 149, 156[208] Xu Wang, Samantha Grist, Jonas Flueckiger, Nicolas A. F. Jaeger, andLukas Chrostowski. Silicon photonic slot waveguide bragg gratings andresonators. Optics Express, 21(16):19029–19039, 2013. URLhttp://www.opticsexpress.org/abstract.cfm?URI=oe-21-16-19029. → pages120[209] Xu Wang, Wei Shi, Raha Vafaei, Nicolas AF Jaeger, and LukasChrostowski. Uniform and sampled bragg gratings in soi strip waveguideswith sidewall corrugations. IEEE Photonics Technology Letters, 23(5):290,2011. → pages 120, 128[210] Zhitian Chen, Jonas Flueckiger, Xu Wang, Fan Zhang, Han Yun, Zeqin Lu,Michael Caverley, Yun Wang, Nicolas AF Jaeger, and Lukas Chrostowski.Spiral bragg grating waveguides for tm mode silicon photonics. OpticsExpress, 23(19):25295–25307, 2015. → pages 121, 128188[211] Alexandre D Simard, Nezih Belhadj, Yves Painchaud, and SophieLaRochelle. Apodized silicon-on-insulator bragg gratings. PhotonicsTechnology Letters, IEEE, 24(12):1033–1035, 2012. → pages 121[212] Xu Wang. Silicon photonic waveguide Bragg gratings. PhD thesis,University of British Columbia, 2013. → pages 121[213] DC Flanders, H Kogelnik, RV Schmidt, and CV Shank. Grating filters forthin-film optical waveguides. Applied Physics Letters, 24(4):194–196,1974. → pages 121[214] Xu Wang, Wei Shi, Han Yun, Samantha Grist, Nicolas A. F. Jaeger, andLukas Chrostowski. Narrow-band waveguide bragg gratings on soi waferswith cmos-compatible fabrication process. Opt. Express, 20(14):15547–15558, 2012. URLhttp://www.opticsexpress.org/abstract.cfm?URI=oe-20-14-15547. → pages124, 128[215] Carlos A Barrios, Kristinn B Gylfason, Benito Sánchez, Amadeu Griol,Hans SohlstrÃűm, Miquel Holgado, and Raphael Casquel. Slot-waveguidebiochemical sensor. Optics letters, 32(21):3080–3082, 2007. → pages 125[216] Francesco DellâĂŹOlio and Vittorio MN Passaro. Optical sensing byoptimized silicon slot waveguides. Opt. Express, 15(8):4977–4993, 2007.→ pages 125[217] Joseph A Curcio and Charles C Petty. The near infrared absorptionspectrum of liquid water. JOSA, 41(5):302–302, 1951. → pages 128, 146[218] Wim Bogaerts, Roel Baets, Pieter Dumon, Vincent Wiaux, Stephan Beckx,Dirk Taillaert, Bert Luyssaert, Joris Van Campenhout, Peter Bienstman,and Dries Van Thourhout. Nanophotonic waveguides insilicon-on-insulator fabricated with cmos technology. LightwaveTechnology, Journal of, 23(1):401–412, 2005. → pages 129, 157, 158[219] Marie Verbist, Wim Bogaerts, and Dries Van Thourhout. Design of weak1-d bragg grating filters in soi waveguides using volume holographytechniques. Journal of Lightwave Technology, 32(10):1915–1920, 2014. →pages 132[220] James Pond, Xu Wang, Jonas Flückiger, Adam Reid, Jens Niegemann, AmyLiu, and Lukas Chrostowski. Design and optimization of photolithographyfriendly photonic components. In SPIE OPTO, pages 97510V–97510V.International Society for Optics and Photonics, 2016. → pages 132189[221] Young Ho Kim, Seong Jun Park, Sie-Wook Jeon, Seongmin Ju, Chang-SooPark, Won-Taek Han, and Byeong Ha Lee. Thermo-optic coefficientmeasurement of liquids based on simultaneous temperature and refractiveindex sensing capability of a two-mode fiber interferometric probe. Opticsexpress, 20(21):23744–23754, 2012. → pages 136[222] Julius Komma, Gerd Hofmann, Christian Schwarz, Daniel Heinert, andRonny Nawrodt. Thermo-optic coefficient of silicon at 1550 nm at lowtemperatures. Institut für Festkörperphysik, page 82. → pages 136[223] W.L. DeLano, M.H. Ultsch, and J.A. Wells. Convergent solutions tobinding at a protein-protein interface. Science, 287(5456):1279–1283,2000. → pages 137, 145[224] Min Huang, Ahmet Ali Yanik, Tsung-Yao Chang, and Hatice Altug.Sub-wavelength nanofluidics in photonic crystal sensors. Optics express, 17(26):24224–24233, 2009. → pages 139[225] Taechung Yi, Lu Li, and Chang-Jin Kim. Microscale material testing ofsingle crystalline silicon: process effects on surface morphology and tensilestrength. Sensors and Actuators A: Physical, 83(1):172–178, 2000. →pages 140[226] Yurii Vlasov. Silicon photonics for next generation computing systems.2008. → pages 150[227] Andrew Rickman. The commercialization of silicon photonics. NaturePhotonics, 8(8):579–582, 2014. → pages 150[228] Matthew W Royal, Richard B Fair, and Nan M Jokerst. Integrated samplepreparation and sensing: Microresonator optical sensors embedded indigital electrowetting microfluidics systems. In Sensors, 2011 IEEE, pages2050–2053. IEEE, 2011. → pages 150[229] Wim Bogaerts, Dirk Taillaert, Bert Luyssaert, Pieter Dumon, JorisVan Campenhout, Peter Bienstman, Dries Van Thourhout, Roel Baets,V Wiaux, and S Beckx. Basic structures for photonic integrated circuits insilicon-on-insulator. Opt. Express, 12(8):1583–1591, 2004. → pages[230] Rongjin Yan, Santano P Mestas, Guangwei Yuan, Rashid Safaisini, David SDandy, and Kevin L Lear. Label-free silicon photonic biosensor systemwith integrated detector array. Lab on a Chip, 9(15):2163–2168, 2009. →pages 164, 165190[231] Ryan C Bailey, Adam L Washburn, Abraham J Qavi, Muzammil Iqbal,Martin Gleeson, Frank Tybor, and L Cary Gunn. A robust silicon photonicplatform for multiparameter biological analysis. In SPIE OPTO: IntegratedOptoelectronic Devices, pages 72200N–72200N. International Society forOptics and Photonics, 2009. → pages[232] Fea Vollmer, D Braun, A Libchaber, M Khoshsima, I Teraoka, andS Arnold. Protein detection by optical shift of a resonant microcavity.Applied Physics Letters, 80(21):4057–4059, 2002. → pages[233] Jonathan D Suter, Ian M White, Hongying Zhu, Huidong Shi, Charles WCaldwell, and Xudong Fan. Label-free quantitative dna detection using theliquid core optical ring resonator. Biosensors and Bioelectronics, 23(7):1003–1009, 2008. → pages 150[234] Pieter Dumon, Wim Bogaerts, Roel Baets, J-M Fedeli, and L Fulbert.Towards foundry approach for silicon photonics: silicon photonics platformepixfab. Electronics letters, 45(12):581–582, 2009. → pages 150[235] Matthew S Luchansky and Ryan C Bailey. High-q optical sensors forchemical and biological analysis. Analytical chemistry, 84(2):793–821, Jan2012. doi:10.1021/ac2029024. → pages 154, 161[236] B Lillis, M Manning, H Berney, E Hurley, A Mathewson, andMM Sheehan. Dual polarisation interferometry characterisation of dnaimmobilisation and hybridisation detection on a silanised support.Biosensors and Bioelectronics, 21(8):1459–1467, 2006. → pages 155[237] H. Arwin. Optical properties of thin layers of bovine serum albumin,globulin, and hemoglobin. Applied Spectroscopy, 40(3):313–318,1986-03-01T00:00:00. doi:doi:10.1366/0003702864509204. URLhttp://www.ingentaconnect.com/content/sas/sas/1986/00000040/00000003/art00005. → pages 155[238] Vincent Ball and Jeremy J Ramsden. Buffer dependence of refractive indexincrements of protein solutions. Biopolymers, 46(7):489–492, 1998. →pages 155[239] Emre Özkumur, Ayça Yalçın, Marina Cretich, Carlos A Lopez, David ABergstein, Bennett B Goldberg, Marcella Chiari, and M Selim Ünlü.Quantification of dna and protein adsorption by optical phase shift.Biosensors and Bioelectronics, 25(1):167–172, 2009. → pages 155191[240] Jacob Piehler, Andreas Brecht, and Guenter Gauglitz. Affinity detection oflow molecular weight analytes. Analytical Chemistry, 68(1):139–143,1996. → pages[241] Lev Moiseev, M Selim Ünlü, Anna K Swan, Bennett B Goldberg, andCharles R Cantor. Dna conformation on surfaces measured by fluorescenceself-interference. Proceedings of the National Academy of Sciences of theUnited States of America, 103(8):2623–2628, 2006. → pages 155[242] Daniel K Sparacin, Steven J Spector, and Lionel C Kimerling. Siliconwaveguide sidewall smoothing by wet chemical oxidation. Journal oflightwave technology, 23(8):2455, 2005. → pages 157[243] F Grillot, L Vivien, S Laval, D Pascal, and E Cassan. Size influence on thepropagation loss induced by sidewall roughness in ultrasmall soiwaveguides. Photonics Technology Letters, IEEE, 16(7):1661–1663, 2004.→ pages 157, 158[244] Katrien De Vos. Label-free silicon photonics biosensor platform withmicroring resonators. PhD thesis, 2010. → pages 157[245] Jared T Kindt, Matthew S Luchansky, Abraham J Qavi, So-Hyun Lee, andRyan C Bailey. Subpicogram per milliliter detection of interleukins usingsilicon photonic microring resonators and an enzymatic signal enhancementstrategy. Analytical chemistry, 85(22):10653–10657, 2013. → pages 157[246] MJ Heck, Jared F Bauters, Michael L Davenport, Jonathan K Doylend,Siddharth Jain, Geza Kurczveil, Sudharsanan Srinivasan, Yongbo Tang,and John E Bowers. Hybrid silicon photonic integrated circuit technology.IEEE Journal of Selected Topics in Quantum Electronics, 19(4), 2013. →pages 164, 165[247] Sahba Talebi Fard, Kyle Murray, Michael Caverley, Valentina Donzella,Jonas Flueckiger, Samantha M Grist, Edgar Huante-Ceron, Shon ASchmidt, Ezra Kwok, Nicolas AF Jaeger, et al. Silicon-on-insulator sensorsusing integrated resonance-enhanced defect-mediated photodetectors.Optics express, 22(23):28517–28529, 2014. → pages[248] Günther Roelkens, Joris Van Campenhout, Joost Brouckaert, DriesVan Thourhout, Roel Baets, P Rojo Romeo, P Regreny, A Kazmierczak,C Seassal, X Letartre, et al. Iii-v/si photonics by die-to-wafer bonding.Materials Today, 10(7):36–43, 2007. → pages 164, 165192[249] Long-Fang Tsai, William C Dahlquist, Seunghyun Kim, and Gregory PNordin. Bonding of polydimethylsiloxane microfluidics to silicon-basedsensors. Journal of Micro/Nanolithography, MEMS, and MOEMS, 10(4):043009–043009, 2011. → pages 164[250] George M Whitesides. The origins and the future of microfluidics. Nature,442(7101):368–373, 2006. → pages[251] Catherine Rivet, Hyewon Lee, Alison Hirsch, Sharon Hamilton, and HangLu. Microfluidics for medical diagnostics and biosensors. ChemicalEngineering Science, 66(7):1490–1507, 2011. → pages 164[252] Temsiri Songjaroen, Wijitar Dungchai, Orawon Chailapakul, Charles SHenry, and Wanida Laiwattanapaisal. Blood separation on microfluidicpaper-based analytical devices. Lab on a Chip, 12(18):3392–3398, 2012.→ pages 164[253] E. Fu, S.A. Ramsey, P. Kauffman, B. Lutz, and P. Yager. Transport intwo-dimensional paper networks. Microfluidics and nanofluidics, 10(1):29–35, 2011. → pages 164[254] David M Cate, Jaclyn A Adkins, Jaruwan Mettakoonpitak, and Charles SHenry. Recent developments in paper-based microfluidic devices.Analytical chemistry, 87(1):19–41, 2014. → pages 164[255] Global silicon photonics market, 2010. → pages 165[256] Dan-Xia Xu, Jens H Schmid, Graham T Reed, Goran Z Mashanovich,David J Thomson, Milos Nedeljkovic, Xia Chen, Dries Van Thourhout,Shahram Keyvaninia, and Shankar K Selvaraja. Silicon photonic integrationplatform-have we found the sweet spot? IEEE Journal of Selected Topics inQuantum Electronics, 20(4):189–205, 2014. → pages 165[257] Herwig Kogelnik. 2. theory of dielectric waveguides. In Integrated optics,pages 13–81. Springer, 1975. → pages 194[258] Project and bug tracking in jira. URLhttps://biobench.atlassian.net/secure/Dashboard.jspa. accessed: June 27,2016. → pages 221193Appendix APrinciples of optical waveguidesAlthough the main focus of the work presented here is primarily on experimentalresults and the subsequent analysis, a theoretical understanding and derivation ofsimplified analytical solutions is key for optimizing and comparing sensor designs.Therefore, the following sections provide an overview of the theory of opticalwaveguides in silicon photonics. Most of the description here draws heavily fromSnyder and Love’s ’Optical Waveguide Theory’ [69], Yariv and Yeh’s ’Photonics:Optical Electronics in Modern Communications’ [73], and Kogelnik’s publicationson dielectic waveguides [85, 257].A dielectric structure that is able to confine and guide an electromagnetic waveis referred to by the term ’waveguide’ [68]. A simple practical example is theoptical fiber used in optical communications which has a material at the core with asufficiently hight refractive index compared to the cladding material, to ’trap’ lightby total internal reflections. This principle also applies to planar waveguides shownin Fig. A.1. Depending on the dimension of the core (and the index contrast),an optical waveguide can only support a discrete number of modes. A waveguidewith only one optical mode allowed is said to be single mode (for typically smallwaveguide thicknesses) and multi mode for waveguides with more than one mode.For multi mode waveguides (and large waveguide thicknesses) one can use rayoptics to describe the propagation properties of modes. This is only valid for thecondition V = 2pitλ0√n2core − n2clad  1, where V is the dimensionless parameterknown as waveguide frequency, t the thickness of the waveguide or diameter of194Figure A.1: Simple example of opticalwaveguides: a) optical fiberwaveguideand b) planar waveguidethe fiber core, λ0 the wavelength of the guided light in vacuum, and ncore,cladthe refractive index of the core and the cladding respectively (only valid for asymmetric waveguide where nclad = nsub). For waveguides with smaller valuesof V , the electromagnetic theory approach is required to compute a mode profile(solution to Maxwell’s equations). Ray optics assumes that power propagates alongthe trajectories of the ray path. This is not true when radiation and absorption bythe cladding are included [72].A.1 Ray optics approachTE polarized light (Ey is the only component of the field vector) is traveling inthe core along z-direction. If pi2 − θ > sin−1(ncladncore), total internal reflection occurs(only for nclad > ncoroe) and the ray will be reflected. Since the electric fieldis parallel to the surface, an additional phase shift, τr , is picked up as given byFresnel’s formula. It is further assumed that nclad = nsub (symmetric waveguide).In order for the wave to interfere constructively after two reflections, the additionalphase shift needs to be a multiple of 2pi. Waves that satisfy this condition are calledeigenmodes (or simply modes) [72]. From Fig. A.2 we get after two reflections andAC− AB = 2t sinθ:2piλ2t sinθ −2τr = 2pim, with m = 1,2,3, ... (A.1)195Figure A.2: Ray optics appraoch: The additional phase shift after two internalreflections at the boundary layer must be a multiple of 2pi. An additionalphase shift of τr is picked up at the interface due to the total internalreflection (according to Fresnel’s formula)The additional phase shift at the interface depends on the polarization (TE/TM) andthe angle of reflection, θ. For a TE polarized wave, this is given by:tanτ2=√sin2 θ − sin2 θcrcosθ, (A.2)where θcr = sin−1(ncladncore)is the critical angle for total internal reflection andθ = pi2 − θ. By defining the effective refractive index as:ne f f = ncore sinθ (A.3)and using cosθ =√n2core−n2e f fncore, the phase shift τr fromEquationA.2 can be rewrittenas:τr = 2arctan√√n2e f f− n2cladn2core − n2e f f(A.4)196by defining:a = |k |√n2core − n2e f f (A.5a)b = |k |√n2e f f− n2clad, (A.5b)where~k is thewave vectorwith |k | = 2piλ0 and its components Ex = ncore |k | sinθ, Ey =0, and Ez = ncore |k | cosθ. Using Equation A.4 and Equations A.5 the conditionfor an eigenmode to exist (Eq. A.1) can then be rewritten as:tan (at −mpi) = 2aba2−2b . (A.6)This equation has a discrete number of solutions, i.e. only certain values forne f f correspond to a guided mode. The expression can be solved numerically orgraphically.A.2 Electromagnetic wave approachIn the simple case of a slab waveguide, an analytical solution to the Maxwellequation can be found. Figure A.3 defines a slab waveguide as a three layeredstructure. The first layer has the refractive index n1 and is referred to as substrate,the second layer is the core and has a refractive index n2, and the third layer isreferred to as cladding with refractive index n3. In Section A.1, the symmetric casewith nsub = nclad was studied. Here, symmetry is not assumed.Figure A.3: Schematic of a slab waveguide. If n1 = n3 then the waveguide issaid to be symmetric197Starting from Maxwell’s equation for an isotropic, linear, non-conducting andnon-magnetic media, by using the fact that the solution to Maxwell’s equation in ahomogeneous media is a plane wave and by matching the boundary conditions ateach layer interface, we can write the following wave equation:(∂2∂x2+∂2∂y2)E(x, y)+[k20n2(x, y)− β2]E(x, y) = 0, (A.7)where k0 = ωc is the wave number in vacuum and β is the propagation constantalong z (this equation is not valid at the dielectric interface for TM waves withH perpendicular to the xz-plane and the tangential components of H have to bematched between the segments to satisfy continuity). The refractive index profileis given by:n =n1, x < −tn2, −t < x < 0n3, x > 0.(A.8)The discussion here will be limited to the case where n2 > n1,n3 or n1k0,n3k0 < β <n2k0, which is the necessary condition for confined modes. If the solution E(x)becomes sinusoidal in all three regions, themodes are referred to as radiationmodes.While for radiation modes an infinite number of modes exists (β is continuous),there are only a discrete number of confined modes allowed. The number ofconfined modes is dependent on the core thickness t, the wavelength, and therefractive indices n1,n2,n3.Due to translational invariance the modal fields can be written in separableform:Em(x, y, z) = Em(x, y)ei(ωt−βmz), (A.9a)Hm(x, y, z) =Hm(x, y)ei(ωt−βmz), (A.9b)where βm is called the propagation constant or eigenvalue of the mth mode.Waveguide modes can be regarded as transverse resonances of the field in awaveguide. This is similar to normal modes of vibration of a membrane [69].For TE modes, Ey is the only component of the field vector and takes the form198Ey (x, y, z, t) = Em(x)ei(ωt−βz). The second order differential equation A.7 is of theform ∂2 f∂y2+ A f = 0, and has exponential solutions for A < 0 and sinusoidal solutionsfor A > 0. Em(x), the wave function of the mth mode, is then given by:Em(x) =Ce−qx, x > 0C[cos(hx)− qh sin(hx)], −t < x < 0C[cos(ht)+ qh sin(ht)]ep(y+t), x < −t .(A.10)C is a constant and h,q, p are given by:h =[(n2ωc)2− β2]1/2(A.11a)q =[β2−(n3ωc)2]1/2(A.11b)p =[β2−(n1ωc)2]1/2. (A.11c)Ey and Hz need to be continuous at both interfaces (x = 0 and x = −t). Since Hz =( iωµ )(∂Ey∂x ) the mode condition can be written as (imposing boundary condition toEq. A.10):tan(ht) =p+ qh(1− pqh2). (A.12)The eigenvalue problem (Eq. A.7) can be solved numerically or graphically andcan also be found using geometric or ray optics. The mode equation for a threelayer planar waveguide has been described by Tiefenthaler and Lukosz [83] in theform below:22piλ0t√n2core − n2e f f − τsub − τclad = 2pim, (A.13)where τsub and τclad are the phase shifts on total internal reflection at the core/-199substrate and the core/cladding interface respectively. They are given by:τsub =2arctan(qh), TE2arctan(n2coren2subqh), TM(A.14a)τclad =2arctan(qh), TE2arctan(n2coren2subqh), TM.(A.14b)From Equation A.12, it can be seen that for an increasing thickness t, more modeswill be guided. Only two types of modes are guided: TE modes with electric fieldsin the yz-plane and perpendicular to the direction of propagation (Ez = 0), and TMmodes with magnetic field in the yz-plane and perpendicular to the direction ofpropagation (Hz = 0).A.2.1 Waveguide modesFor translational invariant modal fields as described in Eq. A.9 the vector waveequations for an arbitrary cross section write:{∇2⊥+ n2k2− β2m}Em(x, y) = −∇⊥{Em(x, y) · ∇⊥ lnn2}(A.15a){∇2⊥+ n2k2− β2m}Hm(x, y) = {∇⊥×Hm(x, y)} ×∇⊥ lnn2 (A.15b)where ∇2⊥ is the transverse Laplacian, βm is the propagation constant of the mthmode, and n = n(x, y) is the refractive index profile (an example can be found inFig. A.4). For the planar waveguide described in Section A.2, the term ∇⊥ lnn2vanishes within the core, substrate, and cladding, but not on the interface. Withinthese regions, Equation A.15 simplifies to the scalar wave Equation A.7. To solvefor Em everywhere the boundary conditions of Maxwell’s equations are imposedon the interface.For integrated photonic circuits, more complicated waveguide structures areused (to confine light or rather energy in two dimensions). Fig. A.4 shows a typicalcross section of a two dimensional waveguide. Although there are analyticalapproximations to treat such structures, a numerical mode solver has been used200Figure A.4: Schematic of a waveguide cross section with lateral mode con-finementalmost exclusively to solve for optical modes in waveguide cross sections. Themode solver used here (by Lumerical Solutions Inc [70]) uses a finite differencetechnique in the frequency domain based on an algorithm proposed by Zhu andBrown [71].201Appendix BTestbench control softwareB.1 Control softwareThe test platform control software is written in MATLAB (Natick, MA). This lan-guage choice was largely motivated by the need for maintainability and extensibilitysince most bioengineering students learn and use MATLAB sometime during theiracademic tenure. GitHub (San Francisco, CA) and Atlassian’s Jira (Sydney, Aus-tralia) are used for source code version control and bug tracking. The softwarearchitecture follows the industry standard model-view-controller partitioning asmuch as possible among the MATLAB objects, scripts, and UI elements (as shownin Fig. B.1). The models define objects that interact with a data source like thehard disk or an instrument. The view defines the user interaction experience (showsoutput and manages user input) through MATLAB’s built-in GUI elements. Thecontroller interprets the user’s commands and orchestrates steps using the models toprovide the result. While the controller could be implemented as a class, it largelyexists as MATLAB scripts with global access to every object and GUI element inthe application’s namespace.Due to historical reasons, the GUI is not a class but handles to its object areaccessible throughout the application via a testbench class property. Other classesimplemented in the control software are listed in Table B.1 with brief explanationsof their purpose. Care was taken keep classes and objects as independent as possibleto facilitate re-use and maintainability.202Figure B.1: Platform Control Application Model-View-Controller Overview.The View defines UI objects used to interpret commands from theuser and provide status. The Controller executes command sequences,manages the application state, and controls the test processes. TheModels communicate with the instruments and structures that storedata.B.1.1 Program flow (sequencing)Since testing devices or performing assays involve many sequenced steps, theapplication guides the user through the process shown in Fig. B.2.The GUI has eight sequenced panels which host smaller UI control panels forinstrument or object controls (like coordinate system). Panels are hidden or visiblebased on the user’s progress through the required sequence and the control UIwindows are often replicated across panels depending on the functionality required.The panel sequence is shown in Fig. B.2 with a brief description of tasks performedat that step.Initializing the application (users, task, and instrument panels)The application opens to the user panel allowing the user to load their previoussettings, specify an email address for automated alerts, and select a specific versiontailored for their specific hardware or group. Pressing the next button will advanceto the task panel where the user selects the test type (DryTest, WetTest, SaltSteps,203Table B.1: Platform control application class definitionName (in source) Description PurposeTestBenchClass Testbench application class Manage initialization, applicationstate and settings, and sequencedflowAssayCtl Assay control class Orchestrates assays, controls the in-struments, and updates GUICoordSysClass Coordinate system class Configures coordinate system andtranslates position/coordinate pairsInstrClass Instrument class Properties and methods to interfaceand control instrumentsDeviceClass Device class Manages the data for each sensor, i.e.’device’, within a chipChipClass Chip class Manages chip-level data common toall devices, e.g. coordinate file, as-say params, etc.BioAssay, or VirtualTestMode) and specifies the test data location on disk, the chiparchitecture (coordinate file), and unique die identifier (string). The next panel isthe instrument panel where the user selects specific instruments and connects tothem. The next button will remain disabled until all the selected instruments havesuccessfully initialized. If an instrument is not loaded, its control UI will not appearon the subsequent panels.Setting up the device under test (mount and resister panels)Themount and register panels help the user align the chip and begin the equilibriumprocess for the chip in its thermally tuned, aqueous environment. During this step,the user loads a chip, enables the vacuum to immobilize it and then mounts a fluidicgasket before seating the flow cell over it. The user can then begin the alignmentwith the chip and flow cell securely mounted.An overhead USB camera connected to a monocular zoom lens provides thenecessary resolution to rotationally align the chip to the fiber array. To accomplishthis, the user physically marks an on-chip alignment farm or fiducial at one ofthe chip on the display with a small piece of tape and traverses the chip to thecomplementary alignment feature (see Figs. B.3(a) and B.3(b). The rotation stage204Figure B.2: Control application flow and sequenced tasks. The software isdesigned to step the user through the required steps necessary to fullytest and characterize biosensing devices.aligns the x and y axes. This process is repeated until the two features align perfectlywith the tape. The user then lowers the fiber array until its reflection off the chip’ssurface can be seen from the second alignment camera (see Fig. B.3(c)). At thispoint, the fiber array is typically within a few hundred microns of the chip’s surface.Again, the user places the piece of tape along a reflected edge traverses the chip inboth the X and Y directions. If the chip and fiber array are not in the same planethen the distance between the fiber array and chip (dz) changes and so does thelocation of the shadow. The tip-tilt stage can be adjusted until the marked edgeused for alignment is consistently coincident with the marker. With the chip knowappropriately aligned, the fiber array can be moved within 30-50 µm’s of the surfaceto begin the registration process.If a TEC is connected, the user can thermally tune the stage. For wet and bio205Figure B.3: The chip must be rotationally aligned to the fiber array prior tosetting up a coordinate system for automated testing. (a) Any featurecan provide a reference to rotationally align the chip but it must havea complementary, in-line marking on the other side of the chip (in thiscase, the loop backed alignment grating couplers are used). The usermoves to the other end of the chip and checks for rotationalmisalignment(b). This process is repeated until the misalignment is gone. (c) Thetip-tilt stage is used to ensure a fully parallel plane between the chipand fiber array. The corner of the fiber array’s reflection is markedand the user moves the chip to each corner, ensuring the marked gap isconsistent at all locations.assays, the user loads the fluidic tray and specifies the well plate configurations (anycombination of two 6, 24, and 96 well plates is currently supported). To continuefrom a previous assay, the user can also specify the current well for previouslyloaded trays. If a pump is connected, the user can set the flow rate and start thepump to establish a signal baseline prior to the assay. With the completion of thesetasks, the user advances to the registration panel to setup a coordinate system. Theregistration panel is designed with the intent to use image processing provided bythe overhead alignment camera to automatically align and register the chip. Sincethis feature has not been implemented fully, the user creates the coordinate systemmanually. Clicking on the Start button will pop-up a new window with all the206controls necessary to start the process. The user begins setting up a coordinatesystem by aligning the fiber array over the first grating coupler alignment farm andstarting a ’Map GC’. This feature enables lasing and captures any reflected poweras the optical stage rasters a user-defined area. The results are shown in a window(Fig. 3.18) where the looped-back grating couplers will have a distinct signaturesince they reflect back a majority of the power. The user then clicks the ’Snap toGC’ button. A cross-hair appears that allows the user to click on a GC in the resultswindow. The stage moves to that position and the user executes a ’Fine Align’ tooptimize the fiber array’s position to the on-chip grating coupler. The user thenselects the device from the pull down window in the ’Coordinates’ sub-panel andclicks the ’set’ radial button to specify a stage location and gds coordinate pair.The user then moves to the next alignment structure and repeats the process. Theapplication computes the coordinate system mapping function automatically afterthree valid assignments have been set and displays the goodness of fit. As anaside, the Laser UI panel is also included in the window so the user can change thelasing wavelength if need be or execute sweeps to optimize the fiber array’s rotation(and angle of incidence) to maximize bandwidth and minimize insertion loss at therequired wavelengths.Selecting devices (devices panel)Figure B.4: Platform Control Application Selecting Devices Panel. The useris able to filter the hundreds of devices on a chip based on the mode,name, type (bio, test, etc.), or rating. Devices can be selected for testing.Selecting a device automatically initiates a fine alignment and pops-upa window to select peaks to track. This is unnecessary for dry testing.The list of devices on the chip will remain unselectable until a valid coordinate207system exists or the user has specified the fiber array’s current location (if alreadyon a device). The devices panel allows the user to filter the list of devices and selectswhich ones will be tested. For characterization or bio assays, the user can specifywhich peaks to track during the test. Clicking on the SELECT button will performa fine align and pop-up a new window to select peaks. Settings and ranges can beconfigured before performing a sweep. Scan results are displayed for each enableddetector and the user clicks the start button to select the resonant peaks or nulls inthe waveform. Multiple peaks can be selected. The Save and Close button returnsthe user to the devices list to continue a similar process for other devices to test.When all the devices have been selected, the next button advances the applicationto the test control panel.Test orchestration and control (test panel)In the case of a dry or wet test, the user simply clicks ’Start’ and the applica-tion sequences through the list of devices, performing a fine align and sweep foreach device and saving the results to the specified data repository. For a Salt-Steps test, clicking ’Start’ briefly pops-up a window that allows the user to spec-ify well numbers for a pre-specified number of titrations. The user can edit the<path>/defaultSaltSteps.txt file directly to change reagents or temperaturesettings if necessary. Selecting the BioAssay test will initiate two pop-ups beforethe user can begin testing. The first window allows the user to specify the recipefile for the assay. The second window allows the user to set configuration switchesfor the assay. The most significant configurable parameters include: temperaturestabilization precision and timeout periods, inlet tubing size for auto-priming andcorrectly marking reagents with acquired scans, manual or automated sequenc-ing of reagents, tube relaxation periods to minimize air bubbles during reagentchanges, how many scans to perform before executing a fine align, and the numberof iterations through the recipe for the selected devices.Recipe file The recipe file allows the user to automate multi-step assays easily.The first line in this simple text file is a comment representing the appropriateformat (shown below) and a comma separates the parameters.208Figure B.5: Platform Control Application Assay Panel. In this example, threedetector channels are active (top to bottom), showing the entire spectrafor one scan (on left), the selected peak within the tracking window(center), and the tracked peak over the course of the assay (right). Therecipe window indicates the current assay step and remaining time.Status messages are provided to the user in the debug window along theleft-side.%<well>,<time(min)>,<reagent>,<ri>,<velocity>,<temp>,<comment>The parameters include: the well number (# > 0), the dwell time in the well (# >0) which can also be translated into number of sweeps using an assay configurationsetting), the reagent name (string of any length), the refractive index of the reagentcan be any number and is most useful for automated salt-step characterization usedto determine sensitivity, pumpflow rate in uL/min (> 0), the target stage temperaturein Celsius (number), and a comment (string of any length). An example recipe fileis included the GitHub repository with the source code.Real-time assay control Bio assays often need user intervention. Extra timemightbe required to achieve a binding equilibrium, the user might want to immediatelyadvance to the next step, or an air bubble may be trapped over the sensor under test.Whatever the case, pressing the ’cancel’ button in the ’Sweeping’ waitbar popup209pauses the assay and allows the user to intervene mid-way through the test. The usercan add time to the current step, skip to the next step, edit the recipe, and interactwith the pump directly to remove air bubbles. Clicking the ’Resume’ button returnscontrol to the application continuing programmed execution.Automated analysis and reporting (analyze panel)When the test finishes, advancing to the analyze panel provides options to generateautomated reports and analysis for dry, wet, and salt step test. For dry and wet tests,the user can generate a single report (pdf) which includes scanline plots for theenabled detectors for all the devices tested. This is useful to quickly assess whichdevices worked well enough to invest the additional time and effort to characterize.To analyze salt step results, the user selects the device, detector number, and thenclicks ’Analyze’. The application automatically calculates and plots the sensitivity(nm/RIU) based on the refractive index value specified in the recipe file. Bioassays and other specialized tests can be analyzed using the stand-alone analysistool described in the next major section below.B.1.2 Adding new instrumentsFlexibility and extensibility were key design requirements when developing theapplication. Adding a new instrument involves modifying a few lines of code andcreating the instrument object (class) with the appropriate parameters and methods.To add a new instrument to an existing instrument group, a new entry mustbe made in the ds.instrDefaults struct in the applicationDefaults.m file. The entryname must match the class name. To add a new group of instruments, a newfieldname must be created for the group in the applicationDefaults.m file. The codewill parse the applicationDefaults.m file and generate an instrument class entry foreach fieldname. Specific instrument options for each class get populated with theinstrument description from the class (property). The very first entry in the list isused as the string for the class as well as the virtual instrument (which is alwayslisted first) as shown in Fig. B.6.The other requirement for adding an instrument is to create a UI control elementinstantiated in the view. Figure B.7 shows an example of the laser and detector UI210Figure B.6: Initializing and connecting instruments panel. The applicationautomatically populates a pull down menu with available class mod-els. The user selects an instrument model, modifies any connections ifneeded, and connects to initialize communication with the instrument.Figure B.7: Laser control UI with settings pop-up. The instrument classprovides inherited properties and methods (like the settings pop-upshown) for all instruments added to the test platform.control panels. If a new instrument is being added to an existing instrument group,the class methods named similar to the existing methods to ensure compatibilitywith the callback functions from the control UI. When creating a new instrumentgroup, a new UI control panel will be needed and integrated with the existingGUI panels. Since all the instruments share a somewhat common framework andfunction (configuration parameters and providing data of some kind), an instrument211superclass provides some methods common to each instrument. For example, ifa new instrument class contains a property struct called ’params’, clicking on thesettings button will generate a pop-up window and create editable elements forevery param struct entry as shown in the popup window on the right hand side ofFig. B.7.B.1.3 Adding new features to an existing testChanging existing features or adding new ones to existing test modes may beneeded. New parameters easily be added to existing test modes by first adding theparameter name and value to the ds.AppSettings.<testType> structure in theapplicationsDefaults.m file. When the application first starts, all the settingsin the applicationsDefaults.m file are read into a public TestBenchClass prop-erty called ’AppSettings. ’AppSettings’ stores all the active settings in the entireapplication. To use the new parameter for the dry or wet test, new code needs to beadded to the scripts/dry_test.m script. To use a new parameter for a bio assayor salt-step test, the file models/@AssayCtlClass/AssayCtlClass.m needs tobe modified to support the new functionality.B.1.4 Software testingSoftware functions and features were progressively tested in both virtual and phys-ical environments. Virtual instrument models were created to mimic their real-lifecounterparts and can be integrated into the application like any physical instru-ment for verification before using real hardware. For example, a user can selecta ’Virtual laser’, ’Agilent laser’, or ’Santur laser’ as the laser component duringinstrument connection and initialization. This way, the software’s programmaticflow, user interaction, and interfaces could be verified independent of the hardware.This not only facilitated the rapid development of the GUI and application flow,but also an initial verification before validating the code on the actual hardwarecomponents. The virtual environment was especially useful once the setup up wasup and running and used in actual assays. Adding features and debugging is timeconsuming and takes time away from applied research. In addition, experimentswith wet chemistries and multiple steps often take more than 24h to run and it is212expensive (time and money) to discover that the acquired dataset is invalid becauseof an undiscovered software bug in the latest feature.B.2 Analysis toolWhile the platform control software provides a mechanism to select and track peaksduring an assay, circumstances often arise when the peak may be lost and the onlyway to recover the true sensogram is post-acquisition. For example, the unfortunateintroduction of an air bubblemaymove the resonant peak out of the trackingwindowor the slow (laser-dependent) scan rate may miss a quickly shifting peak during asignificant bulk refractive index change resulting in an anomalous result. Thissection describes a custom analysis tool which complements the platform controlapplication described previously.The analysis tool allows a user to import an acquired dataset and removeunwanted scans, select and track different peaks, curve fit peaks using a polyor a loretz function, subtract reference channels, compare a functional sensor’sdrift to the system temperature, correlate entire scans or the peak window withprevious ones, measure y-offsets on the sensorgram, and most importantly, exportthe analyzed data and figure. Lastly, the user can save all the fit parameters andexcluded scans to an analysis file that can be reloaded when the dataset is selectedagain at a future time.The analysis tool has become an invaluable aspect of the overall developmentmethodology providing an efficient means to assess large datasets and generatepublication worthy plots. The remainder of this section describes the softwarearchitecture and partitioning, a high level overview of the key features and userinteraction model, how the tool was validated, and how it can be accessed as partof our open source effort.B.2.1 Application overviewThe analysis tool’s main window with a loaded sample dataset is shown in Fig. B.8.Datasets are loaded or preprocessed through the Dataset menu (Fig. B.8(a) in theupper left corner of the application. The tool supports a preprocessing function thatcan modify older datasets with the necessary parameters to ensure compatibility213Figure B.8: Analysis Tool Application Overview. (a) Standard menu optionsallow the user to open datasets, preprocess datasets, run custom analysisscripts, or set specific configuration parameters. (b) Scan panel windowallows the user to select different detector channels. (c) Peak panelprovides fitting features and allows the user to select tracked peakswithin the scan. (d) shows all the tracked peaks for the active detectorand peak.with its latest features. The scan panel (Fig. B.8(b) allows the user to interactwith one scan at a time from the dataset and the peak panel (Fig. B.8(c) providesadditional tools to fit and compare the tracked peaks. Finally, all the tracked peaksare plotted in the peak tracking window (Fig. B.8(d)Scan panelEven though the entire dataset is loaded, the user only views one detector channel ata time. The Scan window (Fig. B.9) allows the user to select other channels througha drop-down menu and tag the channel as either a functional or reference dataset.Reference channels can be subtracted from functional channels in the tracked peakplots figure (Fig. B.8(d). The selected peaks are shown as a small red cross-hairin the main plot window on the right side of the panel. The user can excludescans from the dataset (checkbox) and reselect peaks within this window. The214smaller green plot on the right side shows the user the previous scan for referenceonly. The table below the previous scan window (in Fig. B.9) shows comparativeinformation between the two adjacent scans including: DC offset or shift for eachscan, correlation between the scans, reagent currently in the channel, recordedtemperature, and the time stamp of the acquisition.Figure B.9: Analysis tool scan panel. The user can select different detec-tor channels, specify the channel’s purpose (eg: functional, reference,acetylene, etc.), the exclude a peak from the tracking if necessary.Peak panelThe peak panel compliments the scan panel and resides in the upper right handcorner of the application window (Fig. B.10).A small pull-down menu beneath the main peak window allows the user toselect different peaks that were tracked within the scan. One of the most powerfulfeatures of the analysis tool is the peak fitting function. When the Fit Peak buttonis clicked, another popup window appears allowing the user to perform a fittingfunction around the selected peak (Fig. B.11). Currently, both poly and Lorentzfitting functions are supported and the fit result is plotted in red between the two,vertical dashed lines that represent the fit window boundaries. The user can specifythe fit order, fit window size, and fit window center. Results such as the fitted andraw peak location difference and the goodness of fit are shown in the lower lefthand corner of the window. By right clicking the mouse, a context menu appears215Figure B.10: Analysis Tool Peak Panel. The peak panel allows the user toselect different peaks tracked within the scan. It also provides statisticson the active peak and allows the user to fit the peak through a separatepop-up window.allowing the user to add the calculated Q or export the figure to its own file. Oncethe fit has been optimized, the user is provided the option to apply the same fittingparameters to all the other peaks when closing the peaking fitting window.Peak tracking panelThe tracked peak results for all the scans in the dataset is shown in the peak trackingplot panel (Fig. B.8(d). The application does not plot excluded scans and differentchannels or peaks can be plotted by simply making them active by selecting themfrom the drop-down menus in the scan and peak panels. The peak tracking plot hasa context menu as well with many features to further process the dataset. The mainfeatures include: normalization, subtracting a reference, measuring a y-differenceon the plot, adding vertical markers and text to highlight specific events, overlayingthe system’s temperature, and showing the various reagents sequenced during theassay. Most importantly, the figure and its underlying data arrays can be exportedto a .csv or .fig file for further processing or publication.216Figure B.11: Analysis Tool Peak Fitting Popup. The peak fitting tool allowsthe user to set the fit window size and fit type. Upon closing thewindow, the user can choose to apply similar fitting parameters to therest of the peaks in the scan.Custom scripting and analysisThe tool also supports the facile development and integration of custom analysisscripts. Using the provided scripting template that has access to all the active datain application’s namespace, the user can quickly develop custom analysis routinesand invoke them from within the tool (Analysis Scripts menu shown in Fig. B.8(a)by simply adding them to the analysisScripts directory.B.2.2 Software architectureSimilar to the platform control application described previously, the analysis toolfollows a model-view-controller partitioning scheme. Three custom classes formthe foundation of the tool. The appClass acts as the controller and invokes the217Figure B.12: Analysis Tool Peak Tracking Plot. The user enable specialprocessing features (normalization, reagent labeling, etc.) and addmarkings to the plot user a context menu accessed using the right-mouse-button.view, a figure and m-file pair (guiMarking.fig and guiMarking.m) generated usingMATLAB’s GUIDE. The scanClass and peakClass act as models, manipulatingand storing the scan line and peak data respectively.AppClassThe appClass acts as the application’s controller, instantiating the other objectsand mapping GUI callbacks to its methods. The classes’ constructor sets defaultvalues for the applications parameters and unlike the platform control application,a specific user’s preferences and settings are not stored and must be set duringeach use. When loading a dataset, the appClass instantiates a scanClass object forthe data read from each scan file. One scan object is created for each channel, ordetector that exists in the scan file on disk. The appClass keeps track of all thescan sets in a dataset and performs many of the scan-to-scan comparative analysis.After the dataset has been loaded, the class methods primarily execute the serviceroutines invoked by the user’s interaction with the UI. And while the scan classand peak class primarily manipulate and store the waveform data, the appClasscombines all the data elements and manages the plotting of the tracked peaks. Withall the plotting and analysis features provided through the peak tracking figure’scontext menu, these methods result in a large portion of what constitutes the class.218Figure B.13: Analysis tool class and data model architecture. This schematicshows the interaction model for the various classes that comprise theanalysis tool. A MATLAB generate figure is used as the View, theappClass acts as the Controller, and the scanClass and peakClass actas Models manipulating the data stores.ScanClassThe scanClass acts as the model for a single scan, storing parameters pertinent tothe scan like temperature, reagent, inclusion or exclusion state, the wavelength andpower vectors for all the channels, and all the peaks selected for each channel in thescan. Based on the number of peaks selected, the scan class instantiates the peak219objects (one per peak) and keeps track of them in a class property. Of all the classesin the application, the scan class is the simplest, providing support for the reselectpeak feature-which includes the tracking window size and methods that refind thepeak within that window.PeakClassThe peakClass contains all the necessary properties and methods to fit peaks andmanage their data. Peak objects are instantiated and invoked through the scan object,which is managed by the appClass. The majority of the peakClass properties relateto the necessary fitting parameters to support poly and Lorentz fitting functions.The peakClass also has methods to estimate the Q of a peak and provides all thecallback methods for the peak fitting window pop-up invoked from the application’sscan panel.ScriptsTwo directories exist with scripts that support the analysis tool. The first setprovides support through specialized analysis routines for certain biosensors, likecalculating propagation loss from the chip framework’s test cells provided in thePDK, or the backwards-compatibility scripts to modify older datasets. The otherscript directory, analysisScripts, contains user generated analysis scripts to furtheranalyze datasets cleaned up with the tool. To add a new script, the user simplystarts with the provided script template that contains the code and variable formatsto access data from the analysis tool directly. Each time the user clicks on theAnalysis Scripts menu from the main application window, the appClass builds alist of all the scripts in that directory as a pop-up for selection. This feature hasbeen used extensively to analyze the test platform characterization data and sensorcharacterization routines not supported in the platform control application directly.B.2.3 ValidationInitial software validation was accomplished by loading datasets and manuallytesting all the features and methods. Results were compared with many of the one-off scripts that had been developed previously to analyze datasets. Software bugs220and required new features were tracked using Atlassian’s online issue and projecttracking software, JIRA [258].The tool has also been used to analyze datasets and provide figures for the follow-ing publications: (1) Optimal design parameters for TM ring resonator biosensors,(2) SWG ring resonators for biosensing, and (3) TM Bragg gratings for use inlabel-free biosensing. The tool has also been used exclusively in analyzing the testplatform characterization data described below.B.2.4 Access and future workThe analysis tool can be freely downloaded, modified, and used from the SiPho-AnalysisTool project on GitHub [125] under the Gnu Lesser General Public LicenseVersion 3 license. Our hope is that other silicon photonic research groups wouldfind our tool as useful as we do, helping to accelerate the development of devicesthat will be commercialized and realize the impact we believe they can have onsociety.Performing bio assays and testing devices is expensive. Lots of time and effortis required to setup devices, purchase and prepare reagents, and then run assayswhich often times do not have the expected results. One nice feature analyzingdatasets post-acquisition is the ability to develop useful real-time features off-line.We have started refining an air-bubble detection feature that will halt the assay andalert the user, allowing intervention and (hopefully) salvaging the experiment. Thetool could also benefit from other enhancements including saving user preferences,loading multiple datasets sequentially without having to restart the tool, and mostimportantly, providing dataset comparisons and statistical analysis across manydifferent chips, devices, and experiments.221Appendix CComplete list of publicationsDuring my doctoral studies I have co-authored the following publications which donot appear in my dissertation.C.1 Other projects and collaborations1. L. Chrostowski, Z. Lu, J. Flueckiger, X. Wang, J. Klein, A. Liu, J. Jhoja, J.Pond, ’Design and simulation of silicon photonics schematics and layouts’,SPIE Europe, 04/2016.2. L. Chrostowski, Z. Lu, J. Flueckiger, J. Pond, J. Klein, X. Wang, S. Li,W. Tai, E.Y. Hsu, C. Kim, J. Ferguson, C. Cone, ’Schematic Driven SiliconPhotonics Design’, Proc. SPIE, 02/2016.3. H. Yun, Z. Chen, Y. Wang, J. Flueckiger, Michael Caverley, L. Chrostowski,N. A. F. Jaeger, ’Polarization-rotating, Bragg-grating filters on silicon-on-insulator strip waveguides using asymmetric periodic corner corrugations’,Optics Letters, vol. 40, no. 23, pp. 5578-5581, 12/2015.4. H. Yun, J. Flueckiger, Z. Chen, Y. Wang, L. Chrostowski, N. A. Jaeger,’A Wavelength-Selective Polarization Rotating Reflector using a Partially-Etched Asymmetric Bragg Grating on an SOI StripWaveguide’, IEEE GroupIV Photonics Conference, Vancouver, BC, 08/2015.2225. Z. Chen, J. Flueckiger, X. Wang, H. Yun, Y. Wang, Z. Lu, F. Zhang, N. A.F. Jaeger, L. Chrostowski, ’Bragg Grating Spiral Strip Waveguide Filters forTM Modes’, Conference on Lasers and Electro-Optics: Optical Society ofAmerica, pp. SM3I.7, 05/2015.6. H.P. Bazargani, J. Flueckiger, L. Chrostowski, J. Azana, ’Microring res-onator design with improved quality factors using quarter Bezier curves’,Conference on Lasers and Electro-Optics: Optical Society of America, pp.JTu5A.58, 05/2015.7. Y. Wang, H. Yun, Z. Lu, R. Bojko, W. Shi, X. Wang, J. Flueckiger, F.Zhang, M. Caverley, N. A. F. Jaeger, L. Chrostowski, ’Apodized FocusingFully Etched Sub-wavelength Grating Couplers’, Photonics Journal, IEEE,04/2015.8. S. Talebi Fard, K. Murray, M. Caverley, V. Donzella, J. Flueckiger, S. Grist,E. Huante-Ceron, S. Schmidt, E. Kwok, N. Jaeger, A. Knights, L. Chros-towski. Silicon-on-insulator sensors using integrated resonance-enhanceddefect-mediated photodetectors. Optics Express, Nov. 2014.9. X. Wang, J. Pond, C. Cone, L. Chrostowski, J. Klein, J. Flueckiger, A. Liu,D. McGuire, ’Large-scale silicon photonics circuit design’, Proc. SPIE, vol.9277, pp. 927706-927706-10, 11/2014.10. X. Wang, Y. Wang, J. Flueckiger, R. Bojko, A. Liu, A. Reid, J. Pond, N. A.F. Jaeger, L. Chrostowski, ’Precise control of the coupling coefficient throughdestructive interference in silicon waveguide Bragg gratings’, Optics Letters,vol. 39, issue 19, pp. 5519-5522, 10/2014.11. X. Wang, M. Caverley, J. Flueckiger, Y. Wang, N. A. F. Jaeger, L. Chros-towski, ’Silicon Photonic Bragg Grating Modulators’, IEEE Photonics Con-ference (IPC), 10/2014.12. S.M. Grist, N. Oyunerdene, J. Flueckiger, J.J. Kim, P.C. Wong, L. Chros-towski, K.C. Cheung, ’Fabrication and laser patterning of polystyrene opticaloxygen sensor films for lab-on-a-chip applications’, Analyst: Royal Societyof Chemistry, 09/2014.22313. J. Wang, J. Flueckiger, L. Chrostowski, L.R. Chen, ’Bandpass Bragg gratingtransmission filter on silicon-on-insulator’, Group IV Photonics (GFP), 2014IEEE 11th International Conference on, pp. 79–80, 08/2014.14. Y. Wang, X. Wang, J. Flueckiger, H. Yun, W. Shi, R. Bojko, N. A. F. Jaeger,L. Chrostowski, ’Focusing sub-wavelength grating couplers with low backreflections for rapid prototyping of silicon photonic circuits’, Optics Express,vol. 22, no. 17: OSA, pp. 20652–20662, 08/2014.15. J. Pond, C. Cone, L. Chrostowski, J. Klein, J. Flueckiger, A. Liu, D.McGuire, X. Wang, ’A complete design flow for silicon photonics’, SPIEPhotonics Europe, pp. 9133-39, 04/2014.16. L. Chrostowski, X. Wang, J. Flueckiger, Y. Wu, M. Guillen, Y. Wang, S.Talebi Fard, ’Impact of Fabrication Non-Uniformity on Chip-Scale SiliconPhotonic Integrated Circuits’, Optical Fiber Communication Conference, pp.Th2A.37, 03/2014.17. L. Chrostowski, J. Flueckiger, C. Lin, M. Hochberg, J. Pond, J. Klein, J.Ferguson, C. Cone, ’Design methodologies for silicon photonic integratedcircuits’, Proc. SPIE 8989, Photonics West 2014, 02/2014.18. W. Shi, H. Yun, C. Lin, J. Flueckiger, N. A. F. Jaeger, L. Chrostowski,’Coupler-apodized Bragg-grating add-drop filter’, Optics Letters, vol. 38,issue 16, pp. 3068-3070, 08/2013.19. Y. Wang, W. Shi, X. Wang, J. Flueckiger, H. Yun, N. A. F. Jaeger, L. Chros-towski, ’Fully-Etched Grating Coupler with Low Back Reflection’, Proc.SPIE, Photonics North 2013, vol. 8915, pp. 89150U, 06/2013.20. W. Shi, H. Yun, C. Lin, X. Wang, J. Flueckiger, N.A.F. Jaeger, L. Chros-towski, ’SiliconCWDMDemultiplexersUsingContra-Directional Couplers’,CLEO, pp. CTu3F.5, 06/2013.21. Y. Wang, J. Flueckiger, C. Lin, and L. Chrostowski, ’Universal GratingCoupler Design’, Proc. SPIE, Photonics North 2013, vol. 8915, pp. 89150Y,06/2013.22422. R. Boeck, J. Flueckiger, L. Chrostowski, N.A.F. Jaeger, ’Experimental per-formance ofDWDMquadrupleVernier racetrack resonators’, Optics Express,vol. 21, issue 7, pp. 9103-9112, 04/2013.23. W. Shi, H. Yun, C. Lin, M. Greenberg, X. Wang, Y. Wang, S. Talebi Fard, J.Flueckiger, N.A.F. Jaeger, L. Chrostowski, ’Ultra-compact, flat-top demulti-plexer using anti-reflection contra-directional couplers for CWDM networkson silicon’, Opt. Express, vol. 21, no. 6: OSA, pp. 6733–6738, 03/2013.24. R. Boeck, J. Flueckiger, H. Yun, L. Chrostowski, N.A.F. Jaeger, ’HighPerformance Vernier Racetrack Resonators’, Optics Letters, vol. 37, pp. 3,12/2012.25. J. Flueckiger, V. Bazargan, B. Stoeber, K.C. Cheung, ’Characterization ofpostfabricated parylene C coatings inside PDMS microdevices’, Sensors andActuators B: Chemical , vol. 160, issue 1: Elsevier, pp. 864-874, 12/2011.26. J. Flueckiger, F.K. Ko, K.C. Cheung, ’Microfabricated Formaldehyde GasSensors’, Sensors, vol. 9, issue 11, pp. 9196-9215, 11/2009.27. J. Flueckiger, K.C. Cheung, ’Microfluidic System for Controlled Gelation ofa ThermallyReversibleHydrogel’, IEEETransactions onBiomedical Circuitsand Systems, vol. 3, issue 4, pp. 195-201, 08/2009.28. J. Flueckiger, K.C. Cheung, ’Locally Defined Thermally Reversible Hydro-gel Formation inMicrochannels’, 12th International Conference onMiniatur-ized Systems for Chemistry and Life Sciences. Micro Total Analysis Systems(MicroTAS 2008), 10/2008.225

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0340888/manifest

Comment

Related Items