Enabling Practical Deployment of Silicon RingResonator-Based SystemsbyHasitha JayatillekaA THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinTHE FACULTY OF GRADUATE AND POSTDOCTORALSTUDIES(Electrical and Computer Engineering)The University of British Columbia(Vancouver)January 2018c© Hasitha Jayatilleka 2018AbstractMicroring resonators (MRRs) on silicon photonic platforms allow for low-power,dense, and large-scale manipulation of optical signals on-chip. MRR-basedmodulators, switches, and filters have become key building blocks in integratedoptical circuits for applications in future data communications, high-performancecomputing, and sensing. This thesis presents solutions for overcoming severalchallenges towards practical deployment of MRR systems.The performance of MRR is highly susceptible to temperature and fabricationvariations, which cause significant shifts in the MRR’s spectral responses. In-resonator photoconductive heaters (IRPHs), formed by doping MRRs waveguidesshow high responsivities. As IRPHs do not require additional material depositions,photodetectors, or power taps and use the same contact pads for both sense andtune operations, they can be used to automatically tune and temperature stabilizeMRRs without compromising the cost or area of the devices. Automatic tuningand stabilization of one- and two-ring filters are demonstrated.Multi-ring filters offer attractive spectral features such as wide pass-bands,steep roll-offs, and large extinction ratios. Using IRPHS, automatic tuning of afour-ring Vernier ring filter across a record 37.6 nm wavelength and wavelengthlocking to account for a record 65 oC temperature variation is demonstrated. Atuning algorithm in which the number of iterations scales linearly with the numberof coupled rings in the system is presented. As this method typically does notrely on the output spectral shape of the filter, it is applicable to a wider range ofcoupled resonator systems. Application of this tuning method is then demonstratedfor various multi-ring filters by both simulation and experiment.Crosstalk can be a major source of signal degradation in large-scale MRRiisystems. Interchannel and intrachannel crosstalk of one- and two-ring MRR filtersare experimentally investigated. The power penalties due to interchannel crosstalkare presented as functions of channel spacing and adjacent channel isolation.Intrachannel crosstalk of one-ring, cascaded, and series-coupled add-drop filtersare compared and spectral conditions that will ensure low intrachannel crosstalk ispresented. MRR filters with extremely small radii of 2.75 um, large free spectralranges of 34.3 nm, and high thermal tuning efficiencies of 2.78 nm/mW arepresented.iiiLay SummaryRing resonators on silicon photonic platforms are micro-meter-size waveguideloops. These devices allow for low-power, dense, and large-scale manipulationof optical signals on-chip. As a result, they have become key buildingblocks in integrated optical circuits for applications in data communications,high-performance computing, and sensing. This thesis presents solutions forovercoming several challenges towards practical deployment of ring resonator-based systems.Ring resonators are highly sensitive to fabrication and temperature variations.By utilizing a novel device to simultaneously sense and adjust rings’ resonancestates, this thesis develops methods for automatic and real-time correction ofring resonators’ performance to account for such variations. These methodsare demonstrated by automatically configuring multi-ring systems across recordwavelength and temperature ranges. The thesis also investigates optical crosstalkin various ring resonator devices and proposes solutions for mitigating crosstalk.Ultra-small ring resonators devices enabling low power tuning across largewavelength ranges are also demonstrated.ivPrefaceThe content of this thesis is mostly based on the publications listed below, whichresulted from collaborations with other researchers.1. H. Jayatilleka, K. Murray, M. A. Guillen-Torres, M. Caverley, R. Hu,N. A. F. Jaeger, L. Chrostowski, and S. Shekhar, “Wavelength tuningand stabilization of microring-based filters using silicon in-resonatorphotoconductive heaters,” Optics Express 23(19), 25084-25097, 2015.L.C., S.S., and H.J. came up with the idea to tune and stabilize ringresonators using silicon photoconductive heaters. H.J. designed the deviceswith help from M.C. and K.M.. H.J. and K.M. developed the tuning methodsand performed the experiments. M.A.G, M.C., and R.H. helped with theexperiments. H.J. wrote the paper with the help of co-authors. N.A.F.J., L.C.,and S.S. helped structuring the paper. N.A.F.J., L.C., and S.S. supervised theproject.2. H. Jayatilleka, H. Shoman, R. Boeck, N. A. F. Jaeger, L. Chrostowski, andS. Shekhar, “Automatic Configuration and Wavelength Locking of CoupledSilicon Ring Resonators,” accepted for publication in Journal of LightwaveTechnology, 2017 (invited).R.B. designed the two-ring and four-ring Vernier filters. H.J. designed theother four-ring filter. H.J. developed the tuning methods. N.A.F.J. andR.B. suggested the C-band tuning for demonstrating the tuning algorithms.H.J. and H.S. did the experiments and wrote the paper. All of the authorsreviewed the paper and provided feedback. N.A.F.J., L.C., and S.S.vsupervised the project.3. H. Jayatilleka, K. Murray, M. Caverley, N. A. F. Jaeger, L. Chrostowski, andS. Shekhar, “Crosstalk in SOI microring resonator-based filters,” Journal ofLightwave Technology 34 (12), 2886-2896, 2016 (invited).H.J. designed the devices with help from M.C. and K.M. . H.J., K.M., andM.C. performed the experiment. H.J. and N.A.F.J. wrote the paper. All ofthe authors reviewed the paper and provided feedback. N.A.F.J., L.C., andS.S. supervised the project.4. H. Jayatilleka, N. Eid, M. Caverley, N. A. F. Jaeger, S. Shekhar, and L.Chrostowski, “High performance silicon microring resonator filters,” (inpreparation).H.J. designed the devices with help from M.C.. H.J. and N.E. performed theexperiments with guidance from N.A.F.J. and S.S.. H.J. wrote the paper. Allof the authors reviewed the paper and provided feedback. N.A.F.J., S.S., andL.C. supervised the project.viTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . xixAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Silicon photonics . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Microring resonators: from devices to systems . . . . . . . . . . . 21.2.1 Modulators and detectors . . . . . . . . . . . . . . . . . . 41.2.2 Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.2.3 Ring resonator systems . . . . . . . . . . . . . . . . . . . 61.3 Challenges in microring systems . . . . . . . . . . . . . . . . . . 71.4 Thesis contributions and organization . . . . . . . . . . . . . . . 11vii2 Automatic tuning of silicon ring resonators using photoconductiveheaters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.2 In-resonator photoconductive heaters . . . . . . . . . . . . . . . . 162.3 Automated wavelength stabilization of an MRR filter . . . . . . . 182.4 Automated wavelength tuning and stabilization of a second-orderMRR filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222.5 Automated wavelength tuning of higher-order series-coupled MRRfilters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303 Automatic configuration and temperature stabilization of high-order Vernier ring resonator filters . . . . . . . . . . . . . . . . . . . 323.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333.2 Device and tuning method . . . . . . . . . . . . . . . . . . . . . 353.3 Resonance mapping . . . . . . . . . . . . . . . . . . . . . . . . 393.4 Wavelength tuning . . . . . . . . . . . . . . . . . . . . . . . . . 403.5 Wavelength locking (Temperature stabilization) . . . . . . . . . . 423.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444 Tuning method for coupled ring resonator systems . . . . . . . . . . 454.1 Tuning method . . . . . . . . . . . . . . . . . . . . . . . . . . . 454.2 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . 474.3 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . 494.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515 Crosstalk in silicon ring resonator filters . . . . . . . . . . . . . . . . 535.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545.2 Filter design for a maximum channel count . . . . . . . . . . . . 595.3 Interchannel crosstalk . . . . . . . . . . . . . . . . . . . . . . . . 615.4 Intrachannel crosstalk . . . . . . . . . . . . . . . . . . . . . . . . 675.4.1 Intrachannel crosstalk vs. data rate . . . . . . . . . . . . . 695.4.2 Intrachannel crosstalk vs. detuning . . . . . . . . . . . . 71viii5.4.3 Requirements for the input-to-through response of add-drop filters . . . . . . . . . . . . . . . . . . . . . . . . . 735.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 746 High-performance silicon microring resonators . . . . . . . . . . . . 786.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 796.2 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 816.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 826.3.1 Spectral response and tuning . . . . . . . . . . . . . . . . 826.3.2 Four-channel DeMUX . . . . . . . . . . . . . . . . . . . 826.3.3 Crosstalk measurements . . . . . . . . . . . . . . . . . . 846.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 877 Conclusion and future work . . . . . . . . . . . . . . . . . . . . . . 887.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 887.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94ixList of TablesTable 5.1 DeMUX performance comparison (BER = 10−9, adjacentchannel isolation 15 dB) . . . . . . . . . . . . . . . . . . . . 66xList of FiguresFigure 1.1 A typical silicon on insulator photonics platform. . . . . . . . 2Figure 1.2 (a) All-pass ring resonator. (b) Normalized transmission andintra-cavity power as a function of the wavelength (expressedrelative to resonance wavelength, λres). . . . . . . . . . . . . 3Figure 1.3 (a) Microscope picture of fabricated ring resonator modulator.(b) Eye-diagram generated by modulator. . . . . . . . . . . . 5Figure 1.4 (a) Add-drop ring resonator. (b) Normalized transmission andintra-cavity power as a function of the wavelength (expressedrelative to resonance wavelength, λres). . . . . . . . . . . . . 6Figure 1.5 (a) Series-coupled four-ring add-drop filter. (b) Simulateddrop-port transmission spectra of a two-ring, four-ring, andeight-ring filters. . . . . . . . . . . . . . . . . . . . . . . . . 6Figure 1.6 An example of an MRR-based WDM network. . . . . . . . . 7Figure 1.7 (a) Simulated ideal (desired) and (b) as-fabricated filter spectraof a multi-ring filter with 4 coupled resonators (N = 4) . . . . . 8Figure 1.8 Feedback loops, sensors, and tuners required for tuning andstabilization of a mutli-ring filter system . . . . . . . . . . . . 9Figure 1.9 Tuning efficiency vs. FSR of various low-power ring resonatordevices. The arrow indicates the direction of improvement. . . 10Figure 2.1 (a) Schematic cross-section of the waveguide used in theIRPH. (b) Microscope image of an add/drop MRR filteroverlaid with a circuit description of the integrated IRPH. . . 16xiFigure 2.2 (a) Measured Iheater as a function of Vheater. (b) Normalizeddrop-port transmission and IPD as a function of the wavelengthoffset relative to 1551.52 nm. Vheater = 1 V and the drop-porttransmission is normalized to 87 µW, which was the estimatedoptical power in the bus waveguide at the input of the MRRfilter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17Figure 2.3 Responsivity of the IRPH, measured at the resonancewavelength of the MRR, (a) as a function of Popt-in, withVheater = 1V, and (b) as a function of Vheater, with Popt-in =348µW. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18Figure 2.4 (a) Schematic of the experimental setup. (b) Flow diagram ofthe control algorithm. . . . . . . . . . . . . . . . . . . . . . 19Figure 2.5 Measured (a) IPD and Vheater, and (b) normalized drop-porttransmission, during the progression of the control algorithm.(c) Normalized through- and drop-port transmission spectraof the MRR filter before and after the control algorithm wasapplied. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20Figure 2.6 (a) Measured eye diagram with a constant stage temperature.(b) Measured stage temperature as a function of time.Measured eye diagrams (c) with and (d) without automatedwavelength stabilization. . . . . . . . . . . . . . . . . . . . . 21Figure 2.7 (a) Picture of a fabricated second-order series-coupled MRRfilter. (b) Equivalent circuit diagram of the device. . . . . . . 23Figure 2.8 Measured (a) IPD1 as a function of Vheater1, with Vheater2 = 0 V,and (b) IPD1 and IPD2 as functions of Vheater2, with Vheater1 =1.15 V. (c) Measured through- and drop-port responses ofthe second-order series-coupled MRR filter before and aftertuning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Figure 2.9 (a) Calculated normalized cavity intensity in MRR 2 asfunction of φdiff and φavg. (b) Through- and drop-porttransmission spectra before and after applying the controlalgorithm to the initially tuned spectrum in Fig. 2.8(c).Wavelength is relative to 1554 nm. . . . . . . . . . . . . . . 25xiiFigure 2.10 Flow diagram illustrating the the control algorithm used tostabilize the second-order series-coupled MRR. . . . . . . . . 25Figure 2.11 (a) Measured eye diagram at constant stage temperature. (b)Measured stage temperature as a function of time. Measuredeye diagrams (c) with and (d) without automated wavelengthstabilization. . . . . . . . . . . . . . . . . . . . . . . . . . . 26Figure 2.12 (a) Schematic of a series-coupled 6th-order MRR filter. (b)Simulated through- and drop-port transmission spectra beforeand after tuning. (c)-(h) Calculated optical cavity intensity inMRRs 1-6 as a function of φ1−6 as the MRRs are sequentiallytuned to λc. The optical transmission in (b) and the cavityintensities in (c)-(h) are all normalized to the optical intensityat the input of the filter. . . . . . . . . . . . . . . . . . . . . 28Figure 3.1 A system of N series-coupled microring resonators. . . . . . 34Figure 3.2 (a) Two-ring Vernier filter (b) Illustration of tuning the nearestshorter-wavelength resonances of R1 and R2 to center theVernier response at λx or λy. . . . . . . . . . . . . . . . . . . 35Figure 3.3 Microscope image of a fabricated four-ring Vernier filter.IRPHs are shown as resistors in the overlaid circuit diagram. . 36Figure 3.4 Photocurrents measured during the tuning steps 1 through 4in the IRPHs of rings (a) R1, (b) R2, (c) R3, and (d) R4,respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . 37Figure 3.5 Transmission spectra of the Vernier filter in Fig. 3.3: (a) as-fabricated, (b) after tuning and configuration, and (c) zoomed-out showing the extended FSR. . . . . . . . . . . . . . . . . 38Figure 3.6 (a) Measured IPD,1 as a function of the heater power suppliedto R1 for various input laser wavelengths. (b) Heater powervs. resonance wavelength mapping using a linear fit for eachof the rings. . . . . . . . . . . . . . . . . . . . . . . . . . . 39xiiiFigure 3.7 Overlay of (a) measured drop-port spectra after automaticallyconfiguring the filter at each ITU (200 GHz grid) channelin C-Band, and (b) drop-port spectra relative to the centerwavelength of each ITU channel. . . . . . . . . . . . . . . . 41Figure 3.8 Estimated and measured electrical powers required to tunerings (a) R1, (b) R2, (c) R3, and (d) R4, to each of the ITUwavelengths. . . . . . . . . . . . . . . . . . . . . . . . . . . 42Figure 3.9 (a) Measured eye diagram at constant stage temperature.(b) Measured stage temperature with time. Measured eyediagrams (c) with and (d) without automatic wavelengthlocking. (d) is recorded only for temperature change of 8◦C. . 43Figure 4.1 Tuning step for Rn. Photocurrents IPD,n and IPD,n−1 aremeasured while tuning φth,n. . . . . . . . . . . . . . . . . . . 46Figure 4.2 Simulated (a) as-fabricated and (b) configured filter spectra offour-ring (N=4) filter with flat-top pass-band. . . . . . . . . . 48Figure 4.3 (a)-(d) Calculated optimization functions fOPT,1 - fOPT,4 andthe cavity intensities of rings R1 - R4 (shaded) as a function ofthe round-trip phases φ1 - φ4. . . . . . . . . . . . . . . . . . 49Figure 4.4 Simulated (a) as-fabricated and (b) configured filter spectra offive-ring (N=5) filter with un-apodized coupling coefficients. . 50Figure 4.5 Microscope image of a fabricated four-ring filter. . . . . . . . 51Figure 4.6 Measured (a) as-fabricated and (b) configured filter spectra offour-ring (N=4) filter. . . . . . . . . . . . . . . . . . . . . . . 51Figure 5.1 Illustrations of (a) interchannel crosstalk occuring at the drop-port of an MRR filter and (b) intrachannel crosstalk occuringat the through- and drop-ports of a series-coupled MRR filter. 55Figure 5.2 Response of an MRR filter showing the insertion loss (IL) andadjacent channel isolation. . . . . . . . . . . . . . . . . . . . 57xivFigure 5.3 (a) Measured and simulated bend-loss and (b) finessecalculated for first-order and series-coupled MRRs. (c)Maximum number of channels supported by first-order andseries-coupled MRR-based DeMUXs. . . . . . . . . . . . . . 58Figure 5.4 (a) Microscope image of a fabricated first-order DeMUX and(b) the measured spectra after tuning the channels to a 75 GHzspacing. (c) Microscope image of a fabricated series-coupledDeMUX and (d) the measured spectra after tuning the channelsto a 62.5 GHz spacing. The optical frequencies of (b) and (d)are relative to 193.019 THz (or 1553.17 nm) and 192.888 THz(or 1554.23 nm), respectively. . . . . . . . . . . . . . . . . . 62Figure 5.5 Experimental setup used for measuring the interchannelcrosstalk. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63Figure 5.6 Measured eye diagrams at 25 Gb/s for channels 1 and 2. (a)and (b) were measured at the input of the chip, (c) and (d)were measured at the drop-port outputs of the first-order MRRDeMUX, and (e) and (f) were measured at the drop-portoutputs of the series-coupled MRR DeMUX. . . . . . . . . . 63Figure 5.7 (a) Measured BER vs. received optical power at the PD,(b) crosstalk power penalty of channels 1 and 2 vs. channelspacing, and (c) power penalty of channel 1 vs. adjacentchannel isolation (BER = 10−9, 25 Gb/s). The shaded markersin (b) and (c) indicate where the lowest measured BER was lessthan 10−9. The unshaded markers indicate where the measuredBER values were extrapolated to calculate the power penaltycorresponding to a BER of 10−9. . . . . . . . . . . . . . . . 64Figure 5.8 Measured eye diagrams at 25 Gb/s for channel 1 of (a) the first-order MRR DeMUX (b) the series-coupled MRR DeMUX. . 66Figure 5.9 input-to-through response of a series-coupled add-drop filtershowing the extinction ratio, suppression, and bandwidth(BW). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67xvFigure 5.10 Schematics of a (a) first-order, (b) cascaded, and (c) series-coupled add-drop filter. Measured add-to-through and input-to-through spectra for a (d) first-order, (e) cascaded, and(f) series-coupled add-drop filter (Frequencies are relative to193.195 THz (or 1551.76 nm), 193.356 THz (or 1550.46 nm),and 193.498 THz (or 1549.33 nm), respectively). Eyediagrams measured at the through-port for ADD signal onlyand ADD + DROP signals for a (g) first-order, (h) cascaded,and (i) series-coupled add-drop filter. . . . . . . . . . . . . . 68Figure 5.11 Experimental setup used for measuring the intrachannelcrosstalk. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69Figure 5.12 Measured BER vs. received optical power at the PD for a (a)cascaded and a (b) series-coupled filter design at several datarates. (c) Intrachannel crosstalk power penalty in simultaneousadd-drop operation for various data rates (BER = 10−9). Theshaded markers in (c) indicate where the lowest measured BERwas less than 10−9. The unshaded markers indicate where themeasured BER values were extrapolated to calculate the powerpenalty corresponding to a BER of 10−9. . . . . . . . . . . . 70Figure 5.13 Measured (a) spectra, and (b) BER vs. received opticalpower at the PD for several detunings of a cascaded MRRfilter design. (c) Intrachannel crosstalk power penalty vs.detuning (BER = 10−9, 20 Gb/s). The shaded markers in(c) indicate where the lowest measured BER was less than10−9. The unshaded markers indicate where the measuredBER values were extrapolated to calculate the power penaltycorresponding to a BER of 10−9. Optical frequency in (a) isrelative to 193.209 THz (or 1551.65 nm). . . . . . . . . . . . 72xviFigure 5.14 Measured (a) spectra, and (b) BER vs. received optical powerat the PD after tuning the extinction ratio of a series-coupledMRR filter. (c) Intrachannel crosstalk power penalty vs. input-to-through extinction ratio (BER = 10−9, 20 Gb/s). The shadedmarkers in (c) indicate where the lowest measured BER wasless than 10−9. The unshaded markers indicate where themeasured BER values were extrapolated to calculate the powerpenalty corresponding to a BER of 10−9. Optical frequency in(a) is relative to 193.607 THz (or 1548.46 nm). . . . . . . . . 76Figure 5.15 BW/data rate as a function of suppression of the input-to-through responses. . . . . . . . . . . . . . . . . . . . . . . . 77Figure 6.1 Simulated bend-loss of silicon wire waveguides of 500 nm and600 nm as a function of the ring radius. . . . . . . . . . . . . 79Figure 6.2 Summary of published results of silicon microring resonatorsshown the tuning power vs, FSR. The arrow indicates thedirection for performance improvement. . . . . . . . . . . . . 80Figure 6.3 (a) Bendcoupled microring. (b) Microscope image of afabricated device. . . . . . . . . . . . . . . . . . . . . . . . 81Figure 6.4 Phase matching of a 2.75 µm ring waveguide to a buswaveguide at 200 nm gap. . . . . . . . . . . . . . . . . . . . 83Figure 6.5 Measured (a) drop-port spectra for various heater voltages. (b)Resonance wavelength shift as a function of the applied heaterpower. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83Figure 6.6 (a) Schematic of a 4-ring DeMUX with bend-coupled MRRfilters .(b) Microscope images of fabricated DeMUX. . . . . . 84Figure 6.7 a) Measured spectra of DeMUX after setting the channelspacing to 75 GHz. (b) Zoomed-out spectra of the DeMUXshowing the 34.3 nm FSRs of the MRR filters. . . . . . . . . 85Figure 6.8 Experimental setup used for the BER measurement. . . . . . . 85xviiFigure 6.9 Eye diagrams measured for (a)-(d) ch1 - ch4 with 75 GHzchannel spacing, (e) ch1 with 50 GHz channel spacing betweenthe channels, (f) ch1 reference case when only ch1 wastransmitted. . . . . . . . . . . . . . . . . . . . . . . . . . . 86Figure 6.10 Measured BER vs. optical power at the receiver for all of thechannels with 75 GHz channel spacing. The dotted curvesshow the reference measurements with only one channelturned on. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87Figure 7.1 (a) Layout of a cascaded two-filter system with bandwidthand amplitude tunability. (b) Simulated through- and drop-port transmission spectra when the pass-band is tuned to itsmaximum amplitude. . . . . . . . . . . . . . . . . . . . . . 92Figure 7.2 Microscope pictures of (a) a 16 × 16 microring-based switchmatrix and (b) a unit cell with microring with integrated IRPHand the overlaid circuit description. . . . . . . . . . . . . . . 93xviiiList of AbbreviationsBER Bit Error RatioBW BandwidthCMOS Complementary Metal Oxide SemiconductorCROW Coupled Resonator Optical WaveguideDCA Digital Communications AnalyzerDeMUX DemultiplexerDSA Defect State AbsorptionDWDM Dense Wavelength Division MultiplexingED Error DetectorEDFA Erbium Doped Fiber AmplifierEO Electro OpticER Extinction RatioFDTD Finite Difference Time DomainFSR Free Spectral RangeFWHM Full Width at Half MaximumGC Grating CouplerxixIL Insertion LossIRPH In Resonator Photoconductive HeaterMRR Microring ResonatorMZI Mach-Zehnder InterferometerNRZ Non Return to ZeroOTF Optical Tunable FilterPD PhotodetectorPPG Pulse Pattern GeneratorPRBS Pseudo Random Bit SequenceRX ReceiverSOI Silicon On InsulatorTIA Trans-Impedance AmplifierTLS Tunable Laser SourceTO Thermo OpticTX TransmitterVOA Variable Optical AttenuatorWDM Wavelength Division MultiplexingxxAcknowledgmentsI thank my advisors Professors Sudip Shekhar and Lukas Chrostowski for theirmentorship throughout the years. Any success I have had as a graduate student isin large part thanks to them. Looking back, joining their research groups is one ofthe best decisions I have ever made. I thank Prof. Nick Jaeger for his mentorshipand for being part of my PhD committee. Most of my papers would not haveseen the light of day if it were not for the long hours he spent helping me. I amthankful to Prof. Sharhriar Mirabbassi for being part of my PhD committee. I thankProfessors Andrea Melloni, Robert Raussendorf, Shuo Tang, and Mark Greenstreetfor dedicating their valuable time for being part of my PhD examination committee.I am thankful to Prof. Wim Bogaerts of Gent University and to Dr. Haisheng Rongof Intel Labs for their mentorship and for two amazing internships.I thank Hossam Shoman, Kyle Murray, Mike Caverley, Dr. Robert Boeck,Anthony Park, Nourhan Eid, Dr. Loic Laplatine, Dr. Zeqin Lu, Dr. Miguel Angel,Ricky Hu, Spoorthi Nayak, Ajith S. R., Peter Woo, Mohammad AlTaha, GenePolvy, Dr. Jonas Flueckiger, Antonio Ribeiro, Alex MacKay, Dr. Wesley Sacher,Dr. Roberto Rosales, and Kristie Henriksen for all their help and friendship.I am grateful to the Natural Sciences and Engineering Research Council ofCanada and to the University of British Columbia for financial support throughoutgraduate studies. Many of the fabrication runs would have not been possiblewithout the financial and design support from CMC Microsystems. I thank Dr.Dan Deptuck of CMC Microsystems for his help with the tape-outs.I am deeply thankful to Sahan, Mayumi, Ayumi, Primal, Dinesh, Sujan,Keheliya, and Thamali for being wonderful friends. Above all, I thank my parentsand brother for their love and support.xxiChapter 1Introduction1.1 Silicon photonicsIn the past decade, silicon photonics has emerged as a disruptor in several keytechnological areas such as data and telecommunications, computing, and sensing.This impact across a broad range of fields is a direct result of the close relationbetween silicon photonics and the Complementary Metal Oxide Semiconductor(CMOS) microelectronics industry. Owing to the mature fabrication technologyinherited from CMOS fabrication, silicon photonics provides a platform onwhich optics can co-exist with electronics on a single chip where inexpensiveand densely integrated optical components can be efficiently manipulated usingelectronic circuits. Therefore, by bringing optics closer to the electronics, siliconphotonics provide an opportunity for mitigating short comings and for adding newfunctionality to systems which were previously limited only to electronics.Most of the silicon photonics development today has been based on silicon-on-insulator platforms. A geometry of such a platform, together with somekey photonics components such as grating couplers (GCs), waveguides, andphotodetectors (PDs) is shown in Fig. 1.1 [1]. Silicon is transparent forwavelengths longer than 1.1 µm and, hence, silicon waveguides offer low-losslight propagation at all standard data and telecommunication wavelengths, i.e.,from 1260 nm 1625 nm (the O to L bands). Light is highly confined in siliconwaveguides as a result of the high index contrast between the silicon waveguiding1layer and the surrounding oxide. The thick (about 2 µm) buried oxide layer (BOX)between the silicon substrate and the silicon waveguiding layer acts as the lowercladding and prevents light leaking into the substrate. The refractive index ofthe waveguides can be controlled using 1) thermo-optic effect of silicon usingmetal or doped silicon resistive heaters [2] or 2) by manipulating the electronand hole concentrations inside the waveguides (plasma-dispersion effect) usingp− n or p− i− n diodes [3]. Free-carrier absorption in silicon also allows forwaveguide loss to be controlled using diode structures. While silicon does notsupport light generation due to its indirect bandgap, laser and boradband lightsources are available using III-V materials either heterogenously integrated (bymeans of die-bonding) [4] or edge-coupled to silicon chips [5]. Low-loss fiber-to-chip-coupling ( <1 dB) to silicon photonics chips have also been demonstratedwith both surface [6] and edge-coupling [7] methods. In addition to SOI platforms,silicon photonics components have also been co-integrated with electronic circuitson standard CMOS platforms [8, 9].Si substrateSiWaveguideMetal heaterGe PD Doped wg heater p-n diode wg Grating couplerVia Via Metal Metal Pad opening SiO2cladding BOX (SiO2)n npn+ p+ n+GeShallow etchFigure 1.1: A typical silicon on insulator photonics platform..1.2 Microring resonators: from devices to systemsOn SOI platforms, the large index contrast between silicon and the surroundingoxide allows for the low-loss bending of waveguides with µm-scale radii andadvanced lithography allows for the small waveguide spacings required to achievea high level of coupling [10, 11]. As a result, SOI platforms provide favorable2conditions for realizing numerous microring resonator (MRR)-based devices suchas modulators, filters, and switches with extremely small footprints. The smallfootprint sizes combined with the resonant behaviors of the rings also facilitatehighly efficient modulation, switching, and wavelength tuning of the MRR devices[12, 13]. As a result, MRR-based devices play important roles in many siliconphotonics systems.Input Through Input ThroughDrop(a)-0.1 -0.05 0 0.05 0.1Relative wavelength (nm)-40-30-20-100Transmission (dB)-505101520Cavity power (dB)ThroughCavity power(b)Figure 1.2: (a) All-pass ring resonator. (b) Normalized transmission andintra-cavity power as a function of the wavelength (expressed relativeto resonance wavelength, λres).Operation of a ring resonator in its simplest form is outlined in Fig. 1.2. Asillustrated in Fig. 1.2(a), the resonator is formed by bending a waveguide to forma loop. Light is coupled into and out of the ring using a bus-waveguide placed inclose proximity (typical spacing is on the order of 100s of nm on SOI platforms)to the ring. When the wavelength of the input light is such that the round-tripphase accumulated while traversing the ring is a multiple of 2pi , the light inside theresonator constructively interferes with the input light leading to an energy build-up inside the ring cavity. This resonant condition can be expressed as,λres =neffLm, (1.1)where λres is the resonance wavelength, neff is the effective index of the ringwaveguide, L is the ring circumference, and m is an integer describing the number3of wavelengths inside the resonator. At λres, the light coupling out of the ringinto the bus waveguide destructively interferes with the input light blockingtransmission at the through-port. Figure 1.2(b) shows the transmission spectraat the through-port as well the intra-cavity light-intensity near the resonancewavelength 1. From the transmission spectra, the quality (Q)-factor of thering is given by λres/FWHM, where FWHM is the full-width at the half-maximum/minimum point (or the 3 dB bandwidth) of the transmission spectra.The Q-factor is a measure of the energy build-up inside the resonator, which isthe foundation of many interesting phenomena in ring resonators. As a resultof the energy build-up, the light inside the ring interacts multiple times withany component that is integrated into the ring (e.g., p− n diodes) allowing ringresonators to function as compact modulators, filters, and detectors. The energy-build up inside the ring can also be used to leverage non-linear optical propertiesof silicon (e.g., for four-wave mixing).1.2.1 Modulators and detectorsFor converting an electrical signal to an optical signal, the index, loss of a ringresonator’s waveguide, or the coupling to a ring resonator can be modulated [14].On silicon photonics platforms, the most popular method is to modulate therefractive index of the cavity using a reverse biased p− n diode phase shifterintegrated into ring resonator’s waveguide. Figure 1.3(a) shows the microscopepicture of a fabricated ring modulator and Fig. 1.3(b) shows a recorded eyediagram during modulation. For converting optical signals to electrical signals,ring resonator-based photodetectors can be implemented by integrating a materialcapable of optical detection into the ring resonator. Since silicon is transparentabove 1.1 µm, detectors are usually formed by defect implantation [15] or bygermanium (Ge) deposition [16].1Unless otherwise stated, all of the transmission spectra shown in this thesis are normzlied to theoptical power at the input bus waveguide.4(a) (b)Figure 1.3: (a) Microscope picture of fabricated ring resonator modulator. (b)Eye-diagram generated by modulator.1.2.2 FiltersDue to their inherent wavelength selectivity resulting from high-Q factors(typically ranging from 103-106), ring resonators are attractive for manyapplications as filtering elements [10]. The all-pass configuration shown in Fig.1.2, is a through-port only filter. A popular filter configuration is the add-dropfilter shown in Fig. 1.4, where a second bus-waveguide (drop-port) is includedfor accessing the light resonating inside the ring. Metal heaters or doped siliconheaters are typically used for tuning (or configuring) the filters’ spectral featuressuch as resonance wavelength and bandwidth.Higher-order filter responses can be synthesized by coupling multipleresonators. Figure 1.5 (a) shows a schematic of a filter with four series-coupledresonators. Figure 1.5(b) shows the high-order drop-port spectral responses of two-, four-, eight-ring filters designed for a maximally-flat pass-band. Such filters withfavorable high-order spectral characteristics such as flat-top pass-bands [17, 18],steep roll-offs [19], and high extinction ratios [20, 21] have been demonstratedon numerous silicon photonics platforms. By coupling resonators with differentradii (optical path lengths), the Vernier effect can be used to further increase free-spectral-ranges (FSRs) and the tuning ranges of these devices allowing them tooperate over wide wavelength ranges [22, 23].5Input Through Input ThroughDrop(a)-0.4 -0.2 0 0.2 0.4Relative wavelength (nm)-25-20-15-10-50Transmission (dB)-505101520Cavity power (dB)ThroughDropCavity power(b)Figure 1.4: (a) Add-drop ring resonator. (b) Normalized transmission andintra-cavity power as a function of the wavelength (expressed relative toresonance wavelength, λres).In 1 Out 2filter 1Input ThroughDrop (a)-1 -0.5 0 0.5 1Relative wavelength (nm)-60-50-40-30-20-100Transmission (dB)2-rings4-rings8-rings(b)Figure 1.5: (a) Series-coupled four-ring add-drop filter. (b) Simulated drop-port transmission spectra of a two-ring, four-ring, and eight-ring filters.1.2.3 Ring resonator systemsDue to their small form factors, ring resonator devices are attractive forconstructing optical systems with advanced functionality by means of large-scaleintegration. As a result, a significant amount of research has been directed towards6the design of silicon ring resonator based devices as well as the CMOS electroniccircuitry required to drive and control them. MRR-based systems such as opticaltransmitters and receivers [24], wavelength selective switches [25, 26], routingnetworks [27], sensor arrays [28] and analog-to-digital converters (ADCs) [29]have been proposed or demonstrated.As an example of a ring resonator-based optical system, a wavelength-divisionmultiplexed (WDM) link in the context of a data or telecommunication networkis shown in Fig. 1.6. By transmitting/receiving signals at several wavelengthssimultaneously, WDM allows for the increase in the aggregate data capacity of anoptical system. The transmitter (TX) is comprised of an array of MRR modulatorsall coupled to the same bus waveguide for modulation and multiplexing (MUX) ofthe wavelengths. The receiver (RX) is comprised of an array of MRR filters forde-multiplexing (DeMUX) and PDs for signal detection. The add-drop elementsdirect the signals to and from other TX/RX sites [30].λ1, λ2, … λnλn λnλ2 λ2PD...DeMUXReceiverPDPD...MRR ModulatorsTransmitterAdd-DropElementFigure 1.6: An example of an MRR-based WDM network.1.3 Challenges in microring systemsMany practical systems require large-scale, dense, and on-chip integration ofthese devices. This can be challenging, when considering the control, stability,scalability, cost, and power consumption of these devices from a systemsperspective [31].Presently, the major limitation towards the practical use of ring resonator7systems is their sensitivity to fabrication and temperature variations. Relativelysmall variations in fabrication parameters (e.g., nm-scale variations in waveguidewidths and heights) lead to a large variability in the rings’ resonance wavelengths,even within a single multi-ring filter [32]. For example, Fig. 1.7 shows theideal (or desired) filter spectra and an as-fabricated spectra of a four-ring filter.Owing to the high temperature sensitivity of silicon, temperature changes arisingfrom background temperature fluctuations or crosstalk from nearby devices willcause the rings’ resonance wavelengths to drift during operation. Temperaturefluctuations may also manifest as drifts in the input laser’s wavelength. Therefore,practical use of ring resonator systems requires that they are automatically 1)tuned [or (re-)configured], i.e., by aligning the rings to be resonant at the requiredwavelengths, and 2) stabilized (or wavelength locked), i.e., by maintaining ortracking the resonances at the required wavelengths, to account for fabrication andtemperature variations, respectively.1549 1550 1551 1552Wavelength (nm)-60-40-200ThroughDrop1547 1548 1549Wavelength (nm)-60-40-200Transmission (dB)ThroughDrop(b)(a)Ideal response: As-fabricated:Figure 1.7: (a) Simulated ideal (desired) and (b) as-fabricated filter spectra ofa multi-ring filter with 4 coupled resonators (N = 4) .Automatic configuration and wavelength locking can be achieved by usingfeedback loops that sense the rings’ resonance conditions and tune the ringsaccordingly until the desired resonance conditions, or equivalently the filterresponses, are reached. Typically, while sensing is performed using photodetectors8external [33–35] or internal [36–38] to the rings, separate thermal tuners arerequired for tuning [33–37]. In many instances, these additional components(and their contact pads) occupy significant on-chip real estate compared to thearea required by the ring resonators themselves. For example, the feedbackloops, sensors, and tuners needed for tuning and stabilizing a system with twofour-ring filters is illustrated in Fig. 1.8. Assuming an off-chip applicationspecific integrated circuit (ASIC) is used for controlling the filters, 9 contact pads(assuming a shared ground) are needed for interfacing to the sensors and tunersin each filter. As a result, it is often challenging to utilize a sufficient number ofsensors to completely resolve the resonance conditions of multi-ring systems. Insuch situations, one can resort to multi-variable optimization techniques for findingthe resonant conditions of the rings [33]. However, such optimization techniquescan increase the complexity of the control circuit as well as the time required forfinding the desired resonance conditions.Control circuitSenseTuneSensorsTunersIn 1Out 1In 2Out 2filter 1filter 2Figure 1.8: Feedback loops, sensors, and tuners required for tuning andstabilization of a mutli-ring filter system .910 20 30 40FSR (nm)012345Tuning efficiency (nm/mW)[119][116][122][123][123][88][121] [120]10 20 30 40FSR (nm)012345Tuning efficiency (nm/mW)[119][116][122][123][123][88][121] [120][this work]deep UV deep UVFigure 1.9: Tuning efficiency vs. FSR of various low-power ring resonatordevices. The arrow indicates the direction of improvement.Furthermore, in systems with many rings, the tuning power required to tuneand stabilize the ring resonators can also be significant and may negatively impactthe allowed power budget. Figure 1.9 shows the tuning efficiency (in nm/mW) vs.the FSR of several ring resonator devices demonstrated so far. Here, for a givenring resonator geometry, the metric FSR serves as an indicator of 1) the numberof wavelengths a system can support, 2) wavelength range of operation, and the3) area requirement and integration density. The arrow indicates the direction ofimprovement by increasing the tuning efficiency and the FSR. Only a few devicesshow capability of several nm/mW tuning while supporting large (e.g., spanningthe C-band) FSRs. However, these devices have been fabricated using electron-beam or deep ultra-violet lithography which could increase the cost of fabricationin high-volumes. This indicates that an area-efficient, low-power, and low costsolution for tuning and stabilizing of ring resonators is a key requirement for thelarge-scale integration of these devices [39].In addition to the challenges associated with tuning and stabilization, ringresonator devices in a system can be affected by optical, electrical, and/or thermalcrosstalk from other devices. Optical crosstalk is a significant source of signaldegradation in WDM systems, and can be a limiting factor for the channel capacity10or the allowable number of optical add-drop elements in a network [30, 40, 41].1.4 Thesis contributions and organizationThe objective of this thesis is to investigate and provide solutions for enablingpractical deployment of ring-resonator-based silicon photonics systems. Inparticular, this thesis will investigate low-power and scalable techniques for1. automatic tuning and stabilization,2. tuning over wide wavelength ranges, and3. addressing crosstalk limitationsof silicon microring devices and systems.In chapter 2, in-resonator photoconductive heaters (IRPHs) are introduced asdevices that can be used to both sense and tune the resonance condition of ringresonators simultaneously. IRPHs are doped resistive heaters in silicon waveguidesthat show photoconductive effects having high responsivities in the order of 100sof mA/W. Using IRPHs, automatic wavelength stabilization of first-order MRRand second-order series-coupled MRR filters is experimentally demonstrated.We theoretically show that series-coupled microring filters of any order can beautomatically tuned by using photoconductive heaters to monitor the light intensityin each ring, and sequentially aligning the resonance of each ring to the inputlaser’s wavelength. As IRPHs do not require additional material depositions,photodetectors, or power taps and use the same contact pads for both the sense andthe tune operations, these results are achieved without compromising the cost orarea of the devices. This chapter was published in [42]. At time of publication, thiswas the first demonstration of photoconductivity in doped silicon for ring resonatorcontrol as well as the first demonstration of wavelength stabilization of a multi-ringresonator filter.In chapter 3, we demonstrate automatic configuration and wavelength lockingof multi-ring resonator filters to an input laser’s wavelength. We demonstrate theautomatic configuration of a four-ring Vernier filter across a 36.7 nm wavelengthrange spanning the entire C-band and the wavelength locking of the same filter11to counteract a practical chip temperature variation of 65 oC. This chapter waspublished in [43–45]. These results correspond to the widest wavelength rangeand to the largest temperature range across which a ring-resonator device has beenautomatically tuned or wavelength locked.In chapter 4, a new method for finding the desired resonance conditions ofseries-coupled multi-ring resonators is presented. As this method localizes thefeedback loops to only rely on the resonance conditions of adjacent rings, thenumber of iterations required for this tuning method scales linearly with thenumber of coupled rings in the system. As this method typically does not relyon the output spectral shape of the system, it is, in general, applicable to a widerange of coupled resonator systems. This chapter was published in [45].In chapter 5, we present the experimental investigation of interchannel andintrachannel crosstalk of first- and second-order microring filters. We find optimumring radius for maximizing the WDM channel count. The measured powerpenalties due to interchannel crosstalk of two-channel DeMUXs based on first-order and series-coupled microring filters are presented as functions of channelspacing and channel isolation. Low intrachannel crosstalk power penalties aredemonstrated for cascaded and series-coupled MRR filters for data rates up to 20Gb/s. Based on our results, we present spectral requirements for add-drop filtersthat will ensure low intrachannel crosstalk. This chapter was published [46] andwas the first experimental investigation of cross-talk in first-order and second-ordermicroring filters.In chapter 6, we demonstrate single MRR filters with extremely small radiiof 2.75 um, large-FSRs of 34.3 nm, and high thermal tuning efficiencies of 2.78nm/mW. Coupling into the the small ring resonator is facilitated by bending aphase-matched bus waveguide around the ring resonator. Thermal-undercuts areused to increase the tuning efficiency. A DeMUX with low crosstalk powerpenalties is demonstrated at channel spacings below 100 GHz.In section 7, the possibilities for extending the work demonstrated in this thesisare discussed showing a viable path towards practical deployment of high-orderand large-scale silicon ring resonator systems. Due to various design parametersand numerous fabrication processes, reported performance metrics and designparameters in each chapter corresponds to the devices presented in that chapter.12Chapter 2Automatic tuning of silicon ringresonators using photoconductiveheaters 1In this chapter, we demonstrate that n-doped resistive heaters in silicon waveguidesshow photoconductive effects having responsivities as high as 271 mA/W. Thesephotoconductive heaters, integrated into microring resonator (MRR)-based filters,are used to automatically tune and stabilize the filter’s resonance wavelengthto the input laser’s wavelength. This is achieved without requiring dedicateddefect implantations, additional material depositions, dedicated photodetectors,or optical power taps. Automatic wavelength stabilization of first-order MRRand second-order series-coupled MRR filters is experimentally demonstrated.Open eye diagrams are obtained for data transmission at 12.5 Gb/s while thetemperature is varied by 5 ◦C at a rate of 0.28 ◦C/s. We theoretically showthat series-coupled MRR-based filters of any order can be automatically tuned byusing photoconductive heaters to monitor the light intensity in each MRR, andsequentially aligning the resonance of each MRR to the laser’s wavelength.1 c© of OSA. Reprinted, with permission from [42]132.1 IntroductionSilicon photonic microring resonator (MRR)-based filters, modulators, andswitches have been investigated for use in data-centers and high performancecomputing systems due to their high-speed operation, low power consumption,and compact device footprints [30]. Many of the benefits exhibited by MRR-based devices are due to their resonance characteristics, which also makes theirperformance highly susceptible to variations in fabrication and to fluctuationsin both the chip temperature and the laser’s wavelength. In order to overcomethese issues for practical deployment of MRR-based technologies, it is required todevelop scalable, low-cost, and energy efficient techniques for wavelength tuningand stabilization of MRR-based devices [39, 47].Automatic wavelength tuning and stabilization of MRR-based devices istypically achieved using feedback loops, which require both sensing andcontrolling the resonance conditions of the MRRs. The sensing operation canbe performed using on-chip temperature sensors [48] or, alternatively, usingphotodetectors (PDs) to monitor the light intensity at the output ports of [33, 34],or inside [49–52], the MRRs. Techniques that require light to be tapped out fromthe MRRs or their outputs in order to be monitored with on-chip PDs do not scalewell towards densely integrated systems due to the increases in device footprintand insertion losses. In contrast, sensing mechanisms based on monitoring thelight intensity in the MRRs with in-resonator PDs are more scalable. For example,in-resonator PDs can be fabricated that utilize defect state absorption (DSA) asthe photodetection mechanism [38, 49, 50]. DSA is the process of electron-holepair generation by sub-bandgap defect energy levels, formed primarily as a resultof ion implantation [49, 53, 54]. Previous demonstrations have required dedicatedion implantation steps to create a sufficient number of defect states for absorption[49, 50].The control operation can be performed using thermo-optic (TO) [35], orelectro-optic (EO) [55], phase shifters to tune the resonance wavelength of theMRRs. The TO phase shifters are typically implemented using metallic [35,52] ordoped-silicon [50,56,57] resistive heating elements and are widely used for tuningMRRs due to the large TO coefficient of silicon. For wavelength stabilization, these14heating elements are typically used together with dedicated PDs inside or outside ofthe MRRs. Recently, germanium-based in-resonator photoconductive heaters havebeen demonstrated for both sense and control operations [51], thereby avoidingthe need for dedicated PDs. However, these devices required the deposition ofgermanium, increasing the complexity of fabrication. The TO or EO tuners can bedriven by microcontrollers [35,50,56] or CMOS electronic circuits [52,55,58]. Thecontrol algorithms for wavelength stabilization can be based on thermal dithering[35], homodyne locking [34], or maximum/minimum point searching [56, 58]techniques.In this chapter, we demonstrate automated wavelength tuning and stabilizationof MRR filters using in-resonator photoconductive heaters (IRPHs). IRPHs areformed using doped waveguide sections [see Fig. 2.1(a)]. The doping allows theIRPHs to be used as resistive heaters, but also provide a means of photodetectiondue to DSA. In this work, the doped heaters built into the MRRs, i.e. IRPHs, areused for both the sense and the control operations. As a result, no dedicated ionimplantation steps, germanium depositions, or dedicated on-chip PDs are requiredfor automated stabilization of the MRRs. Furthermore, the use of IRPHs allows fora simple and scalable method for tuning higher-order series-coupled MRR filterssince the tuning state of an MRR can be measured using the MRR’s IRPH. All thedevices described in this chapter were fabricated using 248 nm optical lithographyat A*STAR IME. To the best of our knowledge, this is the first demonstration ofdoped silicon IRPH-based automatic tuning and stabilization of first and second-order MRR-based filters.The rest of this chapter is organized as follows. In section 2, thephotoconductive behaviors of IRPHs are characterized. In section 3, thewavelength stabilization of a first-order MRR filter using an IRPH isexperimentally demonstrated. In section 4, automatic wavelength tuning andstabilization of a second-order series-coupled MRR filter is experimentallydemonstrated. In section 5, it is theoretically shown that IRPHs can be usedto automatically tune higher-order series-coupled MRR-based filters to the inputlaser’s wavelength. Section 6 presents the discussion and the conclusions.15(a) (b)Figure 2.1: (a) Schematic cross-section of the waveguide used in the IRPH.(b) Microscope image of an add/drop MRR filter overlaid with a circuitdescription of the integrated IRPH.2.2 In-resonator photoconductive heatersIn this section, we characterize the photoconductive behavior of IRPHs, whichcan be used to monitor and control the light intensity in MRR-based filters. Aschematic cross-section of the IRPH design used is shown in Fig. 2.1(a). Thewaveguide is n-doped and the n++-doped regions on each side of the waveguidefacilitate low resistance contact to the silicon. Figure 2.1(b) shows a microscopeimage of a first-order add/drop MRR filter with integrated IRPHs. The radius ofthe MRR is 8 µm and the IRPH is formed over 63% of the MRR’s circumference.The MRR is symmetrically coupled, with identical power coupling coefficients of|κ|2 = 0.047, corresponding to a gap of 255 nm, for the through- and drop-port. Forthis device, the total waveguide loss was approximately 6.9 dB/cm. Typically, thepropagation loss of an undoped waveguide in this process is about 2 dB/cm. Thevalues for |κ|2 and waveguide loss were extracted from the measured MRR filterspectrum using a method similar to the one described in [59]. The free-spectral-range of the fabricated device was 12.4 nm.The resonance wavelength of the MRR filter can be tuned by applying avoltage,Vheater, across the IRPH. The total current flowing consists of two parts: (1)the dark current, Iheater, which is the current that flows when no light is incident onthe IRPH, and (2) the photocurrent, IPD, which is the current generated due to DSA.16IPD depends on both the light intensity inside the MRR and Vheater. The measuredIheater versus Vheater for the MRR filter is shown in Fig. 2.2(a). The electrical powersupplied to the IRPH shifts the resonance wavelength of the MRR by 0.25 nm/mW,or equivalently, changes the round-trip phase by 0.04 pi/mW. Figure 2.2(b) showsthe measured drop-port optical transmission and IPD as a function of the wavelengthof the optical input. For this measurement, Vheater = 1 V, and IPD was extracted bysubtracting Iheater from the total measured current. The same calibration methodwas used to obtain IPD values presented in the rest of this chapter.(a) (b)(a) (b)Figure 2.2: (a) Measured Iheater as a function of Vheater. (b) Normalizeddrop-port transmission and IPD as a function of the wavelength offsetrelative to 1551.52 nm. Vheater = 1 V and the drop-port transmission isnormalized to 87 µW, which was the estimated optical power in the buswaveguide at the input of the MRR filter.The responsivity of the IRPH at the MRR’s resonance wavelength, λr, isdefined as IPD(λr)/Popt-in. Popt-in is the optical power in the bus waveguideat the input of the MRR filter, which was estimated using the measured off-resonance through-port transmission. The measured responsivities of the IRPHas a function of Popt-in and Vheater are shown in Figs. 2.3(a) and (b), respectively.The measurements show that the responsivity decreases with increasing Popt-in.We suspect that this is due to defect states being depleted faster than they can bereplenished by trapping carriers. The responsivity increases as Vheater is increasedbecause the strength of the electric field due to the appliedVheater increases. Hence,the carrier transit times across the waveguide become smaller compared to the17carrier lifetimes, increasing the responsivity of the device. The high responsivitiesof our devices are partly due to the high optical intensity build-up inside the MRRs,which was calculated to be about 17.3×Popt-in. Since the intensity build-up factorof an MRR is proportional to the MRR’s finesse parameter [54], the MRRs usedin this work were designed to have a maximized finesse as detailed in [60]. Thetypical responsivities measured for the IRPHs were about an order of magnitudelarger than those reported for doped silicon waveguide PDs with p-n [53, 61] orp++-i-p++ doping [62], and are of the same order as those reported in [54], whichrequired additional fabrication steps for defect implantation, and those reportedin [63], which required high reverse bias voltages.(a) (b)(a) (b)Figure 2.3: Responsivity of the IRPH, measured at the resonance wavelengthof the MRR, (a) as a function of Popt-in, with Vheater = 1V, and (b) as afunction of Vheater, with Popt-in = 348µW.2.3 Automated wavelength stabilization of an MRR filterA schematic of the experimental setup used to demonstrate wavelengthstabilization is shown in Fig. 2.4(a). The optical data stream was generated bymodulating the output of a tunable laser source (TLS) using a LiNbO3 Mach-Zehnder modulator (MZM). A pulse pattern generator (PPG) outputting a non-return-to-zero 231-1 pseudo random binary sequence at 12.5 Gb/s served as amodulating sequence. Grating couplers were used to couple light in to and outof the chip. The output from the drop-port of the MRR filter was amplifed using18MZMPPG TLS λcSense & controlEDFA OTF PD Oscilloscope Temperature controlledstageInputThroughDrop16 µmVheaterSource measurement unit(a)NO Measure IPDX = IPDSet PP P + ΔPY < X Measure IPDY = IPDChange signΔP -1×ΔPYES (b)Fig. 4: (a) Schematic of the experimental setup. (b) Flow diagram of the control algorithm.80 GSa/s) for monitoring eye diagrams. The chip was mounted on a temperature controlledstage, which used a thermoelectric cooler (TEC) and thermistor combination to monitor andcontrol the chip temperature. The sense and control operations on the IRPH were performedusing a source measurement unit with a 1 µA current measurement resolution when used inconstant voltage mode.As shown in Fig. 2(b), the photocurrent generated in the IRPH is maximized when the MRRis on resonance. Therefore, wavelength stabilization was achieved using a computer imple-mented control algorithm similar to that in [16]. The algorithm continuously searches for theVheater that maximizes IPD. The flow diagram of the control algorithm is outlined Fig. 4(b), inwhich P is the electrical power supplied to the heater. In each iteration of the algorithm, Vheateris changed such that P is changed by ∆P. The voltage is changed so as to result in equal powersteps rather than equal voltage steps. This ensures that the round-trip phase step, proportionalto electrical power, does not depend on the operating voltage. IPD is measured before and afterchanging P. If IPD increases, then the sign of ∆P is unchanged and the algorithm proceeds to thenext iteration. Otherwise, the sign of ∆P is reversed before proceeding. A one-time calibrationstep is performed prior to starting the control algorithm to measure Iheater as a function ofVheaterso that IPD can be calculated.The photocurrent, heater voltage, and the measured drop-port optical power during the pro-gression of the control algorithm are shown in Figs. 5(a) and (b), respectively. The slow re-sponse time observed was primarily due to the high integration time of the off-chip opticalpower monitor that was used to measure the drop-port output power shown in Fig. 5(b). Therecorded wavelength spectra in Fig. 5(c) show that the drop-port power is maximized at theTLS output wavelength (1552 nm) after applying the control algorithm.Eye diagram measurements were performed to demonstrate the quality of a transmitted sig-nal through the MRR filter while the chip was subjected to temperature variations. Figure 6(a)shows the output eye diagram when the stage temperature was maintained at a constant level.For this measurement, the resonance wavelength of the MRR was manually aligned to theTLS wavelength and the control algorithm was not used. When the control algorithm wasturned on and the stage temperature was kept constant, the change in eye height was negli-gible. The applied stage temperature variation is shown in Fig. 6(b). The temperature of thestage was changed from 25.0 ◦C to 30.2 ◦C and the rate of temperature change was approxi-mately 0.28 ◦C/s. This corresponded to a resonance wavelength shift of approximately 0.4 nmat a rate of about 20 pm/s. The linewidth of the MRR filter was 0.31 nm. Figure 6(c) shows theFigure 2.4: (a) Schematic of the experimental setup. (b) Flow diagram f t econtrol algorithm.an erbium-doped fiber amplifier (EDFA) and was filtered using an optical tunablefilter (OTF). Th o tput of the OTF was connected to a PD, which was connect toa real-time oscilloscope (32 GHz, 80 GSa/s) for monitoring eye diagrams. The chipwas mou ted on a temp ratur controlled stage, hich used a thermoelectric cool r(TEC) and thermistor combination to monitor and control the chip temperature.The sense and control operations on the IRPH were performed using a sourcemeasurement unit with a 1 µA current measurement resolution when used inconstant voltage mode.As shown in Fig. 2.2(b), t photocurrent generated in the IRPH is maximizedwhen the MRR is on resonance. Therefore, wavelength stabilization was achievedusing a computer implemented control algorithm similar to that in [56]. Thealgorithm continuously searches for the Vheater that maximizes IPD. The flowdiagram of th control algorithm is outlined Fi . 2.4(b), in which P is theelectrical power supplied to the heater. In each iteration of the algorithm, Vheateris changed such that P is changed by ∆P. This ensures that the round-trip phasestep, proportional to electrical power, does not depend on the operating voltage.IPD is measured before and after changing P. If IPD increases, then the sign of ∆Pis unchanged and the algorithm proceeds to the next iteration. Otherwise, the signof ∆P is reversed before proceeding. A one-time calibration step is performed priorto starting the con rol algorithm to measure Iheater as a function ofVheater so that IPDcan be calculated.190 2 4 60102030405060I PD (µA)Time (s)1 3 5 00.20.40.60.811.2V heater (V)(a)0 1 2 3 4 5 6−15−10−5014.4 dB Drop−port transmission (dB) Time (s) (b)1549 1550 1551 1552−25−20−15−10−50 Wavelength (nm)Normalized transmission (dB) TLS λc=1552 nmDrop (before)Through (before)Drop (tuned)Through (tuned)(c)Fig. 5: Measured (a) IPD and Vheater, and (b) normalized drop-port transmission, during the pro-gression of the control algorithm. (c) Normalized through- and drop-port transmission spectraof the MRR filter before and after the control algorithm was applied.Fig. 6: (a) Measured eye diagram with a constant stage temperature. (b) Measured stage tem-perature as a function of time. Measured eye diagrams (c) with and (d) without automatedwavelength stabilization.Figure 2.5: Measured (a) IPD and Vheater, and (b) normalized drop-porttransmission, during the progression of the control algorithm. (c)Normalized through- and drop-port transmission spectra of the MRRfilter before and after the control algorithm was applied.The photocurrent, heater voltage, and the measured drop-port optical powerduring the progression of the control algorithm are shown in Figs. 2.5(a) and(b), respectively. The slow response time observed was primarily due to the highintegration time of the off-chip optical power monitor that was used to measure thedrop-port output power shown in Fig. 2.5(b). The recorded wavelength spectrain Fig. 2.5(c) show that the drop-port power is maximized at the TLS outputwavelength (1552 nm) after applying the control algorithm.Eye diagram measurements were performed to demonstrate the quality of20a transmitted signal through the MRR filter while the chip was subjected totemperature variations. Figure 2.6(a) shows the output eye diagram when thestage temperature was maintained at a constant level. For this measurement, theresonance wavelength of the MRR was manually aligned to the TLS wavelengthand the control algorithm was not used. When the control algorithm was turnedon and the stage temperature was kept constant, the change in eye height wasnegligible. The applied stage temperature variation is shown in Fig. 2.6(b).The temperature of the stage was changed from 25.0 ◦C to 30.2 ◦C and the rateof temperature change was approximately 0.28 ◦C/s. This corresponded to aresonance wavelength shift of approximately 0.4 nm at a rate of about 20 pm/s.The linewidth of the MRR filter was 0.31 nm. Figure 2.6(c) shows the eye diagramwith wavelength stabilization while the stage temperature was varied as shown inFig. 2.6(b). The eye remained open for the duration of the measurement. TheFigure 2.6: (a) Measured eye diagram with a constant stage temperature.(b) Measured stage temperature as a function of time. Measured eyediagrams (c) with and (d) without automated wavelength stabilization.21minimum recorded eye height was about 23% smaller than the eye height recordedat a constant temperature. The eye height was reduced primarily because the speedof the optimization algorithm was limited by the response time of the measurementequipment. As shown in Fig. 2.6(d), the eye diagram is completely closed whenthe wavelength stabilization was not used. For each measurement, more than 1.7×1012 bits were transmitted.2.4 Automated wavelength tuning and stabilization of asecond-order MRR filterCompared to first-order filters, second-order series-coupled MRR filters offersuperior spectral characteristics, such as a flat-top response and a high out-of-bandsignal rejection [57, 64, 65]. A microscope image of such a filter is shown in Fig.2.7, where MRRs 1 and 2 are coupled to the through- and drop-port waveguides,respectively. Ideally, for the desired filter response, both of the MRRs of a second-order filter should be resonant at the laser’s wavelength, λc. However, fabricationand temperature variations cause the resonance wavelengths of the MRRs to bedetuned from each other, as well as from λc. Therefore, a control algorithmshould be able to (1) automatically tune the MRRs to be resonant at λc, and (2)maintain the resonance at λc while the chip is subjected to temperature variations.In this section, we experimentally demonstrate the automated tuning of a second-order filter by sequentially tuning the MRRs to λc. The resonance conditions forthe MRRs are found by maximizing IPD1 and IPD2, the photocurrents generatedby the IRPHs integrated into MRRs 1 and 2, respectively. We also demonstratewavelength stabilization of the filter by using a control algorithm that monitorsIPD2 to maximize the drop-port output.The second-order filter was designed to have power coupling coefficientsof |κ|2 = 0.1311 and 0.0049, for the bus-to-MRR and MRR-to-MRR couplers,respectively. The other parameters used for the design of the MRRs and the IRPHswere similar to those used for the first-order MRR design described in section2. MRRs 1 and 2 can be tuned by adjusting Vheater1 and Vheater2, respectively.An equivalent electrical circuit of the device is shown in Fig. 2.7(b). Rp is aparasitic resistance due to the silicon slab region between MRRs 1 and 2, and was22eye diagram with wavelength stabilization while the stage temperature was varied as shown inFig. 6(b). The eye remained open for the duration of the measurement. The minimum recordedeye height was about 23% smaller than the eye height recorded at a constant temperature. Theeye height was reduced primarily because the speed of the optimization algorithm was limitedby the response time of the measurement equipment. As shown in Fig. 6(d), the eye diagramis completely closed when the wavelength stabilization was not used. For each measurement,more than 1.7 ×1012 bits were transmitted.4. Automated wavelength tuning and stabilization of a second-order MRR filterCompared to first-order filters, second-order series-coupled MRR filters offer superior spectralcharacteristics, such as a flat-top response and a high out-of-band signal rejection [17, 24, 25].A microscope image of such a filter is shown in Fig. 7, where MRRs 1 and 2 are coupledto the through- and drop-port waveguides, respectively. Ideally, for the desired filter response,both of the MRRs of a second-order filter should be resonant at the laser’s wavelength, λc.However, fabrication and temperature variations cause the resonance wavelengths of the MRRsto be detuned from each other, as well as from λc. Therefore, a control algorithm should beable to (1) automatically tune the MRRs to be resonant at λc, and (2) maintain the resonanceat λc while the chip is subjected to temperature variations. In this section, we experimentallydemonstrate the automated tuning of a second-order filter by sequentially tuning the MRRsto λc. The resonance conditions for the MRRs are found by maximizing IPD1 and IPD2, thephotocurrents generated by the IRPHs integrated into MRRs 1 and 2, respectively. We alsodemonstrate wavelength stabilization of the filter by using a control algorithm that monitorsIPD2 to maximize the drop-port output.Input ThroughDrop16 µmMRR 1MRR 2Vheater1IRPHsVheater2(a)Vheater1 Vheater2Iheater1 + IPD1 Iheater2 + IPD2RpRIRPH RIRPH(b)Fig. 7: (a) Picture of a fabricated second-order series-coupled MRR filter. (b) Equivalent circuitdiagram of the device.The second-order filter was designed to have power coupling coefficients of |κ|2 = 0.0945and 0.0024, for the bus-to-MRR and MRR-to-MRR couplers, respectively. The other parame-ters used for the design of the MRRs and the IRPHs were similar to those used for the first-orderMRR design described in section 2. MRRs 1 and 2 can be tuned by adjustingVheater1 andVheater2,respectively. An equivalent electrical circuit of the device is shown in Fig. 7(b). Rp is a parasiticresistance due to the silicon slab region between MRRs 1 and 2, and was measured to be about3 kΩ, which was approximately 10× the resistance of the IRPHs, RIRPH. Due to the presenceof Rp, Iheater1 and Iheater2 each depended on the voltages applied to both of the IRPHs. There-fore, to calculate IPD1 and IPD2 from the total measured current, we performed a 2-D calibrationmeasurement of Iheater1 and Iheater2 as functions of both Vheater1 and Vheater2. However, the 2-DFigure 2.7: (a) Picture of a fabricated second-order series-coupled MRRfilter. (b) Equivalent circuit diagram of the device.measured t be about 3 kΩ, which was approximately 10× the resistance of theIRPHs, RIRPH. Due to the presence of Rp, Iheater1 and Iheater2 each depended on thevoltages appli d to bo h of the IRPHs. T er for , to calculate IPD1 and IPD2 fromthe total measured current, we performed a 2-D calibration measurement of Iheater1and Iheater2 as functions of both Vheater1 and Vheater2. However, the 2-D calibrationcould be avoided by (1) characterizing RIRPH and Rp as functions of the voltagesacross them, and using the results to calculate Iheater1 and Iheater2 as functions ofVheater1 and Vheater2 or (2) designing the device such that the coupling betweenMRRs 1 and 2 is achieved without a slab region, i.e. with strip waveguides.The tuning process is shown in Fig. 2.8. First, as shown in Fig. 2.8(a), IPD1 wasmeasured as a function of Vheater1 with Vheater2 = 0. To tune MRR 1 to λc, Vheater1was set to 1.15 V, the value that maximized IPD1. Next, as shown in Fig. 2.8(b),IPD2 was measured as a function of Vheater2 with Vheater1 = 1.15 V. To tune MRR2 to λc, Vheater2 was set to 1.10 V, the value that maximized IPD2. Figure 2.8(b)also shows the change in IPD1, which corresponds to the change in light intensityin MRR 1, while tuning MRR 2. The through- and drop-port spectra before andafter tuning the filter are shown in Fig. 2.8(c). After tuning, the spectral responseof the filter was improved, with a drop-port insertion loss less than 0.5 dB and athrough-port extinction ratio greater than 20 dB.The algorithm used for wavelength stabilization can be explained byexpressing the tuning states of the MRRs in terms of the transformed coordinates23calibration could be avoided by (1) characterizing RIRPH and Rp as functions of the voltagesacross them, and using the results to calculate Iheater1 and Iheater2 as functions of Vheater1 andVheater2 or (2) designing the device such that the coupling between MRRs 1 and 2 is achievedwithout a slab region, i.e. with strip waveguides.0.4 0.6 0.9 1.15 1.4 1.615304560 I PD1 (µA) Vheater1 (V)Step 1 : tune MRR 1(a)0.4 0.6 0.9 1.1 1.4 1.615304560I PD2 (µA)Vheater2 (V)Step 2 : tune MRR 215304560I PD1 (µA)(b)1552 1553 1554 1555 1556 1557−40−30−20−100Normalized transmission (dB) Wavelength (nm) TLS λc=1554 nmDrop (before)Through (before)Drop (tuned)Through (tuned)(c)Fig. 8: Measured (a) IPD1 as a function of Vheater1, with Vheater2 = 0 V, and (b) IPD1 and IPD2 asa function of Vheater2, with Vheater1 = 1.15 V. (c) Measured through- and drop-port responses ofthe second-order series-coupled MRR filter before and after tuning.The tuning process is shown in Fig. 8. First, as shown in Fig. 8(a), IPD1 was measured asa function of Vheater1 with Vheater2 = 0. To tune MRR 1 to λc, Vheater1 was set to 1.15 V, thevalue that maximized IPD1. Next, as shown in Fig. 8(b), IPD2 was measured as a function ofVheater2 with Vheater1 = 1.15 V. To tune MRR 2 to λc, Vheater2 was set to 1.10 V, the value thatmaximized IPD2. Figure 8(b) also shows the change in IPD1, which corresponds to the change inlight intensity in MRR 1, while tuning MRR 2. The through- and drop-port spectra before andafter tuning the filter are shown in Fig. 8(c). After tuning, the spectral response of the filter wasimproved, with a drop-port insertion loss less than 0.5 dB and a through-port extinction ratiogreater than 20 dB.The algorithm used for wavelength stabilization can be explained by expressing the tuningstates of the MRRs in terms of the transformed coordinates φdiff = (φ1 − φ2)/2 and φavg =(φ1 + φ2)/2, where φ1 and φ2 are the round-trip phases of MRRs 1 and 2, respectively. Theround-trip phases can be tuned by changing P1 and P2, the electrical powers supplied to theIRPHs in MRRs 1 and 2, respectively. Figure 9(a) shows the optical cavity intensity in MRR2 as a function of φdiff and φavg, calculated using the transfer matrix method described in [26].The intensity, being proportional to the filter’s drop-port output intensity, has a single maximumpoint corresponding to the case φ1 = φ2 = 0, i.e. φdiff = φavg = 0, which yields the tuned filterresponse at λc. Therefore, a control algorithm for wavelength stabilization using IRPHs canFigure 2.8: Measured (a) IPD1 as a function of Vheater1, with Vheater2 = 0 V,and (b) IPD1 and IPD2 as functions of Vheater2, with Vheater1 = 1.15 V. (c)Measured through- and drop-port responses of the second-order series-coupled MRR filter before and after tuning.φdiff = (φ1−φ2)/2 and φavg = (φ1 +φ2)/2, where φ1 and φ2 are the round-tripphases of MRRs 1 and 2, respectively. The round-trip phases can be tuned bychanging P1 and P2, the electrical powers supplied to the IRPHs in MRRs 1and , respectively. Figure 2.9(a) s ows the optical cavity int nsity in MRR2 as a function of φdiff and φavg, calculated using the transfer matrix methoddescribed in [66]. The intensity, being proportional to the filter’s drop-port outputintensity, has a single maximum point corresponding to the case φ1 = φ2 = 0,i.e., φdiff = φavg = 0, which yields the tuned filter response at λc. Therefore, a24(a) (b)(a) (b)(b)(a)Figure 2.9: (a) Calculated normalized cavity intensity in MRR 2 as functionof φdiff and φavg. (b) Through- and drop-port transmission spectra beforeand after applying the control algorithm to the initially tuned spectrumin Fig. 2.8(c). Wavelength is relative to 1554 nm.NO Measure IPD2X = IPD2Set PavgPavg Pavg + ΔPavgMeasure IPD2Y = IPD2Set PdiffPdiff Pdiff + ΔPdiffMeasure IPD2Z = IPD2Y < X Z < Y Change signΔPavg  -1×ΔPavgChange signΔPdiff  -1×ΔPdiffNO YES YES Figure 2.10: Flow diagram illustrating the the control algorithm used tostabilize the second-order series-coupled MRR.control algorithm for wavelength stabilization using IRPHs can be based on themaximization of IPD2, thereby maximizing the light intensity in MRR 2. A flowdiagram of the implemented control algorithm is shown in Fig. 2.10. The controlalgorithm continuously updated (1) the average power supplied to the heaters,Pavg = (P1 +P2)/2, by a step size of ±∆Pavg and (2) the difference in powersupplied to the heaters, Pdiff = (P1−P2)/2, by a step size of ±∆Pdiff. The signs of∆Pavg and ∆Pdiff were chosen in each iteration of the algorithm towards increasingIPD2. The MRRs were tuned in terms of Pdiff and Pavg because the largest slopesof MRR 2’s optical cavity intensity near its maximum occur along the φdiff andφavg axes [Fig. 2.9(a)], thereby yielding the highest sensitivity to detuning when25a maximum search algorithm is used. Alternatively, one could implement thecontrol algorithm by tuning the MRRs in terms of P1 and P2. Fig. 2.9(b) showsthe measured spectra before and after applying the control algorithm to the filterwhile keeping the temperature of the stage constant. Before applying the controlalgorithm, the filter was tuned to a wavelength of 1554 nm using the wavelengthtuning algorithm. It can be seen that the optimized state of the filter as determinedby the tuning algorithm agrees with that determined by the stabilization algorithm.Figure 2.11: (a) Measured eye diagram at constant stage temperature. (b)Measured stage temperature as a function of time. Measured eyediagrams (c) with and (d) without automated wavelength stabilization.We used a similar setup to the one shown in Fig. 2.4 to perform eyediagram measurements. Figure 2.11(a) shows the output eye diagram whenthe stage temperature was maintained at a constant level and the wavelengthstabilization was not used. When the wavelength stabilization was used and thestage temperature was kept constant, the eye height decreased by less than 3%.The reduction in eye height was due to oscillations about the optimal tuning point26caused by the electrical power steps in each iteration of the stabilization algorithm.The stage temperature was varied according to the profile shown in Fig. 2.11(b)in order to test the wavelength stabilization algorithm. The temperature of thestage was changed from 25.0 ◦C to 30.2 ◦C and the rate of temperature changewas approximately 0.28 ◦C/s. The linewidth of the filter was 0.39 nm. The eyewhich remained open with wavelength stabilization, as shown in Fig. 2.11(c), wascompletely closed when wavelength stabilization was not used [Fig. 2.11(d)]. Foreach measurement, more than 1.7 ×1012 bits were transmitted. With wavelengthstabilization, the minimum recorded eye height was about 16% smaller than theeye height recorded at a constant temperature. This reduction in eye height wasprimarily due to the time required to optimize the tuning state after a temperaturechange resulted in a detuning.2.5 Automated wavelength tuning of higher-orderseries-coupled MRR filtersTuning higher-order MRR filters by only monitoring the drop-port intensity can bechallenging because information about the round-trip phases of each MRR can notbe determined independently from the drop-port transmission alone. Therefore,tuning algorithms for higher-order series-coupled MRR filters based on drop-portintensity monitoring have relied on complex optimization algorithms [33]. Inthis section, we theoretically show that the automated tuning method based onmonitoring the intensity in each MRR using IRPHs, described in section 4 forsecond-order filters, can be readily extended to series-coupled MRR filters with ahigher number of MRRs.As an example, we consider the 6th-order series-coupled add/drop MRR filtershown in Fig. 2.12(a). For a maximally flat pass-band response [67], the powercross-coupling coefficients for each of the couplers from the input- to the drop-port were chosen to be |κ|2 = 0.4,0.0146,0.0039,0.0029,0.0039,0.0146, and 0.4,respectively. A radius of 8µm, effective index of 2.57, and a waveguide loss of6 dB/cm were assumed for all of the MRRs.Figure 2.12(b) shows the calculated filter responses before and after tuning.The wavelength is shown relative to λc = 1554.2nm. In order to simulate the27φ0,n was modeled using a normal distribution with a mean of −pi/5 and a standard deviationof 0.0713pi . The mean phase of −pi/5 was chosen to ensure that all of the rings are initiallydetuned from λc. The value for standard deviation was calculated according to the fabricationvariations in MRR resonance wavelengths as reported in [28]. The tuning algorithm describedherein is not sensitive to the values of these parameters as long as the rings are all initiallysufficiently detuned from λc. The ideal filter response, shown in Fig. 12(b), corresponds to thecase in which each φn = 0. Therefore, the goal of the tuning method is to adjust each φth,n suchthat the ideal filter response is achieved.MRR 1InputThrough DropMRR 2 MRR 3 MRR 4 MRR 5 MRR 6(a)−2 −1.5 −1 −0.5 0 0.5−60−50−40−30−20−100Normalized transmission (dB)Relative wavelength (nm) Drop (ideal)Through (ideal)Drop (before)Through (before)Drop (after)Through (after)(b)0 0.2 0.4−30−20−10010 (/pi)Cavity intensity (dB) Step 1 : tune MRR 1 φ1MRR 1MRR 2(c)0 0.2 0.4−30−20−10010 (/pi)Cavity intensity (dB) Step 2 : tune MRR 2 φ2MRR 1MRR 2MRR 3(d)0 0.2 0.4−30−20−10010 (/pi)Cavity intensity (dB) Step 3 : tune MRR 3 φ3MRR 2MRR 3MRR 4(e)0 0.2 0.4−30−20−10010 (/pi)Cavity intensity (dB) Step 4 : tune MRR 4 φ4MRR 3MRR 4MRR 5(f)0 0.2 0.4−30−20−10010 (/pi)Cavity intensity (dB) Step 5 : tune MRR 5 φ5MRR 4MRR 5MRR 6(g)0 0.2 0.4−30−20−10010 (/pi)Cavity intensity (dB) Step 6 : tune MRR 6 φ6MRR 5MRR 6(h)Fig. 12: (a) Schematic of a series-coupled 6th-order MRR filter. (b) Simulated through- anddrop-port transmission spectra before and after tuning. (c)-(h) Calculated optical cavity inten-sity in MRRs 1-6 as a function of φ1−6 as the MRRs are sequentially tuned to λc. The opticaltransmission in (b) and the cavity intensities in (c)-(h) are all normalized to the optical intensityat the input of the filter.Figure 2.12: (a) Schematic of a series-coupled 6th-order MRR filter. (b)Simulated through- and drop-port transmission spectra before and aftertuning. (c)-(h) Calculated optical cavity intensity in MRRs 1-6 asa function of φ1−6 as the MRRs are sequentially tuned to λc. Theoptical transmission in (b) and the cavity intensities in (c)-(h) are allnormalized to the optical intensity at the input of the filter.28effects of fabrication variations, the round-trip phase for the nth MRR in the filter,φn, was calculated using φn = φ0,n+φth,n, where φ0,n is the initial round-trip phaseat λc and φth,n is the phase associated with the thermal tuning. φ0,n was modeledusing a normal distribution with a mean of −pi/5 and a standard deviation of0.0713pi . The mean phase of −pi/5 was chosen to ensure that all of the ringsare initially detuned from λc. The value for standard deviation was calculatedaccording to the fabrication variations in MRR resonance wavelengths as reportedin [68]. The tuning algorithm described herein is not sensitive to the values of theseparameters as long as the rings are all initially sufficiently detuned from λc. Theideal filter response, shown in Fig. 2.12(b), corresponds to the case in which eachφn = 0. Therefore, the goal of the tuning method is to adjust each φth,n such thatthe ideal filter response is achieved.Automatic tuning of the filter can be achieved by sequentially tuning MRRs 1through 6 to be resonant at λc. Figure 2.12(c) shows the optical intensity in MRRs1 and 2 as φ1 is tuned. The value of φ1 corresponding to the maximum opticalintensity in MRR 1, φ1,max, is determined and φ1 is set to this value. Then, similarly,MRRs 2 through 6 are sequentially tuned, and each φn is set to φn,max. Figures2.12(c)-(h) show, for each step of the tuning process, the optical cavity intensity inthe MRR being tuned as well as the intensities in the adjacent MRRs. As shown inFig. 2.12(b), the filter spectra after tuning show excellent agreement with the idealresponse. As it was experimentally demonstrated in section 4 for a second-orderfilter, the phases φn are tuned by changing the electrical power supplied to the IRPHin each MRR, and the intensity in each ring can be determined by measuring IPD,n.In Figs. 2.12(c)-(h), the maximum optical cavity intensity in the nth MRR duringtuning step n does not occur exactly at φn = 0 due to the influence of the resonancesof the other MRRs. The accuracy of the algorithm is best when the initial detuningof the resonators is increased. Here, a mean initial detuning of−pi/5 was chosen todemonstrate that excellent agreement with the ideal filter spectrum can be obtained,even for a small initial detuning. This method of tuning can be generalized toseries-coupled filters with any number of MRRs, and the time required for tuningwould scale linearly with the order of the filter. For the stabilization of a high-orderMRR-based filter, one can extend the control algorithm we used for the second-order ring (Fig. 2.10). The algorithm would consist of sequentially stepping the29electrical power supplied to each of the IRPHs in the direction that maximizes thephotocurrent in the IRPH of the MRR coupled to the drop-port waveguide.2.6 DiscussionIRPHs use doped waveguides, which have additional losses compared to undopedwaveguides. Nevertheless, the MRR devices can still be designed to have lowdrop-port losses by controlling the bus-to-MRR and MRR-to-MRR couplingcoefficients. PDs based on p-n junctions [50, 53] operate in reverse bias andtherefore have low dark currents. The IRPHs demonstrated here have dark currentson the order of milliamps. However, since this current is simultaneously used forthe heating of the MRR, the dark current does not represent an additional powerconsumption. Furthermore, the responsivities of the IRPHs are large enough sothat even at low voltages [Fig. 2.3(b)], they can be used for wavelength tuningand stabilization. The measurements showed that the responsivity of an IRPH isa function of the input power and of the bias voltage. However, it was found thatthese effects did not have an appreciable impact on the performance of wavelengthtuning and stabilization. The calibration step required for measuring IPD is onlyperformed at the start of the measurement. The 2-D calibration we performedfor the second-order MRR does not scale well for higher-order filters. However,the calibration complexity can be made simple, i.e. to scale linearly with filterorder, by eliminating the parasitic resistances between the MRRs. As suggestedin section 4, this could be achieved by implementing MRR-to-MRR couplers withstrip waveguides, thereby avoiding the conductive silicon slab region between theMRRs.2.7 SummaryWe have demonstrated automated wavelength tuning and stabilization of MRR-based filters using IRPHs. The responsivities measured for the IRPHs were as highas 271 mA/W. The IRPHs measure the light intensity in the MRRs and, therefore,can be used for automated tuning of MRRs. We implemented wavelength tuningand stabilization algorithms for first-order and second-order series-coupled MRRfilters, and obtained open eye diagrams for both cases at a datarate of 12.5 Gb/s30while varying the temperature. Also, we theoretically showed how higher-orderseries-coupled MRR filters can be automatically tuned using IRPHs by sequentiallyaligning the resonance wavelength of each MRR to the laser’s wavelength. Asdemonstrated in this chapter, the main advantages of using IRPHs for automaticwavelength tuning and stabilization are that (1) they neither require any dedicatedion implantation steps to introduce defects, nor any germanium deposition [50,51],(2) they can be used for both photodetection and heating, which allows for a smallerfootprint, and (3) they do not require tap-outs of the output signal [33–35, 56].Furthermore, since the IRPHs monitor the optical cavity intensity in the MRRs,this approach can be readily extended to devices/systems (e.g. [30, 55, 64]) whichrequire simultaneous wavelength tuning or stabilization of multiple MRRs.31Chapter 3Automatic configuration andtemperature stabilization ofhigh-order Vernier ring resonatorfilters 1High-order microring filters, typically with more than two series-coupledresonators, can leverage both the compact footprint sizes and the resonantcharacteristics of microrings to offer attractive spectral features such as wide pass-bands, steep roll-offs, and large extinction ratios that are desirable for a wide rangeof applications in telecommunication and computing systems [1, 2]. Furthermore,by coupling resonators with different lengths and utilizing the Vernier effect toextend the free-spectral-ranges (FSRs), these devices also have the ability tooperate over wide wavelength ranges [3, 4]. In this chapter, by using in-resonatorphotoconductive heaters (IRPHs) to both sense and tune the resonance conditionsof ring resonators, we demonstrate automatic configuration and wavelength lockingof a four-ring Vernier filter to an input laser’s wavelength. We demonstratethe automatic configuration of this filter across a 36.7 nm wavelength rangespanning the entire C-band and the wavelength locking to counteract a practical1 c© of IEEE. Reprinted, with permission from [45]32chip temperature variation of 65 oC. As IRPHs do not require additional materialdepositions, photodetectors, or power taps and use the same contact pads for boththe sense and the tune operations, these results are achieved without compromisingthe cost or area of the device.3.1 IntroductionIn Chapter 2, we introduced in-resonator photoconductive heaters (IRPHs) asdevices that can be used to both sense and tune the resonance condition of a ringresonator simultaneously. IRPHs are formed by n-doping the waveguides of a ringresonator. Such doped waveguides are typically used as thermo-optic phase shiftersin many silicon photonics platforms [46]. IRPHs also act as photodetectors due todefect-state-absorption and the measured photocurrent corresponds to the ring’sintra-cavity optical intensity. The circuit description in Fig. 3.1 illustrates howIRPHs (shown as resistors) can be integrated into an N-ring filter. As both thesense and tune operations of a ring are performed using a single terminal (labeledV1,..,n in Fig. 3.1), IRPH-based automatic configuration methods do not requireadditional photodetectors, power taps, or contact pads (in addition to those usedfor tuning). Furthermore, as the resonance conditions of individual resonators arenow readily available, multi-ring systems can be configured using simple tuningmethods that are localized to individual resonators in the system [42]. Therefore,the use of IRPHs scales well towards automatic configuration of large-scale andmulti-ring resonator systems in terms of footprint, number of electrical contacts,fabrication cost, and the simplicity and the scalability of the tuning methods.Compared to the many reports on the design of multi-ring filters [17–21, 69–71], there have only been a few reports on automatic configuration and wavelengthlocking of such filters. IRPH-based automatic configuration of two-ring resonatorfilters and wavelength locking over 5 oC was demonstrated in [42]. Automatictuning of a five-ring filter using multi-variable methods was shown in [33], whichwas limited by the FSR of the filter to a wavelength range less than 3 nm.Wavelength locking over a laser wavelength drift of 1 nm (corresponding to atemperature variation of 9 oC) was demonstrated in [72] for the same filter. In [43],using IRPHs, we demonstrated the automatic tuning of a two-ring Vernier filter33CROW…Input…R2R1 Rn-1 Rn RN-1 RNV1 V2 Vn-1 Vn VN-1 VNFigure 3.1: A system of N series-coupled microring resonators.across the entire C-band.In this chapter, we extend our work presented in [44] to demonstrate theautomatic configuration of a four-ring Vernier filter, spanning 36.7 nm acrossthe entire C-band and wavelength locking to account for a practical temperaturevariation of 65 oC. To the best of our knowledge, these results correspond to thewidest wavelength range and to the largest temperature range across which a ring-resonator device has been automatically tuned or wavelength locked.In Chapter 2, we showed that multi-ring filters can be automatically configuredby sequentially maximizing the cavity intensity in each ring (i.e., maximizing thephotocurrent as measured in each ring’s IRPH). In this chapter, we show that thesame method can be used to automatically configure and wavelength lock a Vernierfilter to the input laser’s wavelength. Section 3.3 presents the fabricated device andoutlines the tuning method. Section 3.4 presents a method for finding the initial(as-fabricated) resonance locations of the individual rings of the filter. Section3.4 and 3.5 present the automatic configuration of the filter across the C-bandand wavelength locking, respectively. The tuning methods shown in this chapterare readily applicable to various coupled ring resonator devices. We use a four-ring Vernier filter to demonstrate that our methods can be applied to coupled ringresonators with various optical path lengths.343.2 Device and tuning methodTypically, multi-ring systems are designed such that the desired filter responsecentered at the input laser’s wavelength λin is obtained when all of the rings areresonant at λin. i.e., at λin, each of the rings have a round-trip phase given by,φn = φ0,n+φth,n ≡ 0 (mod 2pi), (3.1)where φ0,n is the initial round-trip phase of the nth ring, and φ th,n is the phaseintroduced by thermal tuning. The objective of an automatic tuning method is thento find φ th,n that meets the above condition. As shown in [42], the resonances ofthe rings in a multi-ring filter can be aligned by sequentially tuning rings R1 - RN tomaximize the photocurrents measured in each ring’s IRPH (IPD,n), which ensuresthat (3.1) is satisfied. Any detunings that occurred in the rings due to thermalcrosstalk can be accounted for using a further optimization step. For example, theoptimization step used in this section maximizes the photocurrent in the last ring,IPD,N , by stepping the rings’ voltages, V1−VN . Here, it is assumed that the cavityintensity of RN (or equivalently the drop-port transmission) is a maximum. Wealso present a tuning method that does not depend on the output spectral shape ofthe filter in the next chapter.Automatic Wavelength Tuning of Series-Coupled VernierRacetrack Resonators on SOIHasitha Jayatilleka, Robert Boeck, Kyle Murray, Jonas Flueckiger,Lukas Chrostowski, Nicolas A. F. Jaeger, and Sudip ShekharDepartment of Electrical and Computer Engineering, University of British Columbia,2332 Main Mall, Vancouver, BC V6T 1Z4, Canada.hasitha@ece.ubc.caAbstract: Using in-resonator photoconductive heaters to both sense and control the intra-cavitylight intensity of microring resonators, automatic tuning of a silicon-on-insulator two-ring Vernierfilter is demonstrated across the entire C-band.OCIS codes: (130.3120) Integrated optics devices; (130.0250) Optoelectronics; (230.4555) Coupled resonators.1. IntroductionBy coupling resonators that have different optical path lengths, the Vernier effect can be used to increase the free-spectral-ranges (FSRs) and wavelength tuning ranges achievable in microring resonator-based devices [1]. Widelytunable lasers [2], wavelength selective switches [3], and reconfigurable filters [4, 5], capable of meeting numeroustelecom-grade filter specifications [1], have been demonstrated on silicon photonics platforms using the Vernier ef-fect. Tunable Vernier devices allow one to adjust the performance of the devices to address variations in fabrication,operating temperatures, and wavelength of the input laser, as well as to completely reconfigure the devices to oper-ate at various channel wavelengths [1, 5]. Therefore, the ability to automatically tune these devices is an essentialrequirement for their practical deployment.Recently, we showed that automatic wavelength tuning and stabilization of silicon microring-based filters can beachieved using in-resonator photoconductive heaters (IRPHs) to both sense and control the light intensity inside theresonators [6]. The IRPHs are n-doped waveguide sections that can be integrated into each microring in a silicon-on-insulator (SOI) filter. The photodetection in IRPHs occurs due to defect-state-absorption [7]. Nevertheless, IRPHsshow high responsivities without requiring dedicated defect implantations [6]. The n-doped waveguides also act asresistive heaters, allowing for thermo-optic tuning of the microring resonators.In this work, we show how a Vernier filter fabricated with two racetrack resonators with incorporated IRPHs in aseries-coupled configuration [Fig. 1(a)] can be automatically tuned to center its response to any input laser wavelength.The automatic tuning is achieved using simple maximum-search algorithms, which are used to find the heater voltagesettings that maximize the light intensity in each of the reso ators. The lig t inte sity in a re onator is determinedby measuring the photocurrent generated by its IRPH. As compared to previous work [7-9], the automatic tuningdescribed here is achieved without requiring dedicated defect implantatio s, additional material depositi ns, dedicatedphotodetectors, thermal sensors, or op ical power taps. In this work, w pre nt the first demonstration of automaticwavelength tuning of a Vernier ring resonator filt r.2. Device and tuning methodFigure 1(a) shows a microscope image f SOI two-ri g Ve nier filte which w s fabricated t the A*STAR IMEfoundry. The IRPHs, formed by n-doping the silicon rib waveguides (rib height = 220nm, slab height = 90nm), areshown as resistors in the circuit. The doping was achieved by i n-implantation. The two racetr ck resonators, labeled“Ring 1” and “Ring 2” had lengths L1 = 66.83µm and L2 = 83.54µm, res ectively.Figure 1(b) illustrates how tuning the nearest shorter-wavelength resonance of each resonator to the input laser’swavelength at lx results in the overall r sponse being en red at lx. This is achieved by first tuning Ring 1, which       !####  "$ "$ #     #     #  #        Wavelength Transmission (dB) OxOy two-ring - Oxtwo-ring - Oy Ring 1Ring 2InputThrough Drop20 µmVheater1 Vheater2IRPHs(b)(a)Ring 2Ring 1Transmission (dB)Wavelength (a) (b)                              " #%%%% !$&  !$& % !%! ! !%! ! !%!!%!        """" !#   !# " """ " (a) (b) (c)(d) (e)extended FSR = 35.8 nmFig. 1. (a) A microscope picture of a fabricated two-ring Vernier filter. (b) Illustration of tuning the nearest shorter-wavelength reso-nances of Ring 1 and Ring 2 to obtain the overall two-ring responses at either lx or ly.arXiv:1601.01022v1 [physics.optics] 6 Jan 2016R1R2(a) (b)InputDropFigure 3.2: (a) Two-ring Vernier filter (b) Illustration of tuning the nearestshorter-wavelength resonances of R1 and R2 to center the Vernierresponse at λx or λy.35As a result of the different optical path lengths, the resonance locations ofthe individual rings in a Vernier filter are dispersed by design. Therefore, inorder to configure a Vernier filter at a specific wavelength, λx, the nearest shorter-wavelength resonance (as the resonance wavelength is red-shifted due to thermo-optic effect) of each ring to λx must be found and tuned to λx. This is illustratedin Fig. 3.2 for a two-ring Vernier filter. The tuning method outlined above can bereadily applied here by tuning R1 and R2 to maximize IPD,1 and IPD,2, respectively.Finding the first maxima of both IPD,1 and IPD,2 ensures that the nearest un-tunedshorter-wavelength resonances of R1 and R2 are tuned to λx. When the input laseris at λy, a wavelength that is several FSRs away from λx, finding the first maximaof IPD,1 and IPD,2 will still ensure that the nearest shorter-wavelength resonances toλy are selected and tuned to λy. This also ensures that any wavelength within theextended FSR of the Vernier filter is addressed by tuning each ring within its FSR.InputThrough Drop20 µmV1 V2R1 R2 R3 R4V3 V4Figure 3.3: Microscope image of a fabricated four-ring Vernier filter. IRPHsare shown as resistors in the overlaid circuit diagram.We apply the described tuning method to the four-ring Vernier filter shown inFig. 3.3. The device is fabricated at the A*STAR IME foundry. IRPHs, shown asresistors in Fig. 3.3, are formed by n-doping the rib waveguides (rib height = 220nm, slab height = 90 nm) of the racetrack resonators. The four rings, R1-R4, havecircumferences of L1 = L2 = 66.83µm and L3 = L4 = 83.54µm, respectively. Theinput laser is set to 1548.6 nm, and the rings R1 through R4 are sequentially tuned tomaximize the photocurrent measured in each ring’s IRPH. Figures 3.4(a)-(d) showthe measured IPD,1 - IPD,4 during each step of the tuning process as functions ofthe electrical power dissipated in the heaters. Here for demonstration purposes, the364 4.5 5 5.5 6 6.5Heater power (mW)3040506070I PD,1 (µA)Step 1: tune R14.5 5 5.5 6 6.5 7Heater power (mW)20406080I PD,2 (µA)Step 2: tune R25 5.5 6 6.5 7 7.5Heater power (mW)20406080I PD,3 (µA)Step 3: tune R35 5.5 6 6.5 7 7.5 8Heater power (mW)161820222426I PD,4 (µA)Step 4: tune R4(a) (b)(c) (d)Figure 3.4: Photocurrents measured during the tuning steps 1 through 4 inthe IRPHs of rings (a) R1, (b) R2, (c) R3, and (d) R4, respectively.voltages V1−V4 are swept past the point at which the photocurrent is maximized.After the four tuning steps, the optimization step as described above is applied toaccount for any detuning that occurred due to thermal crosstalk between the rings.Figure 3.5(a) shows the as-fabricated and the configured spectra after tuning and37optimization. Figure 3.5(b) shows the 37.21 nm extended FSR due to the Verniereffect. The calculated FSR of the individual rings are 9.24 nm, 7.45 nm, for ringsR1, R2 and R3, R4, respectively.(a)(a)1546 1547 1548 1549Wavelength (nm)-60-40-200NormalizedTransmission (dB)ThroughDrop1547 1548 1549 1550Wavelength (nm)-60-40-200ThroughDrop(b)Configured:As-fabricated:1500 1510 1520 1530 1540 1550 1560 1570 1580Wavelength (nm)-60-40-200NormalizedTransmission (dB)ThroughDrop(c)37.21 nmFigure 3.5: Transmission spectra of the Vernier filter in Fig. 3.3: (a) as-fabricated, (b) after tuning and configuration, and (c) zoomed-outshowing the extended FSR.IRPHs carry both the heating currents and the photocurrents. These twocurrents can be separated with the aid of an initial calibration step performed withthe input laser turned off. During this calibration step, the heater current of a ring’s38IRPH, Iheater,n, is measured as a function of its voltage, Vn, as well as the voltageson the IRPHs of the adjacent rings, Vn−1, Vn+1. i.e., Iheater,n = f (Vn,Vn−1,Vn+1).Subsequently, during operation, the photocurrent, IPD,n is obtained by subtractingthe calibrated Iheater,n from the total measured current. This multi-dimensionalcalibration is required to account for the electrical crosstalk current that flowsthrough the un-doped slab regions between the IRPHs. However, such multi-dimension calibration steps can be avoided by designing the IRPHs so that theelectrical crosstalk between them is minimized (Section 4.3).3.3 Resonance mapping(a)(b)1547 1547.5 1548 1548.5 1549 1549.5 1550 Resonance Wavelength (nm) 0246810Heater Power (mW)R1R2R3R42 4 6 8 10 12Heater Power (mW)020406080100I PD,1 (µA)λin = 1548 nmλin = 1548.25 nmλin = 1548.5 nmλin = 1548.75 nmλin = 1549 nmFigure 3.6: (a) Measured IPD,1 as a function of the heater power supplied toR1 for various input laser wavelengths. (b) Heater power vs. resonancewavelength mapping using a linear fit for each of the rings.39The tuning algorithm presented in Section 3.2 is sufficient on its own to tunea filter with any number of series-coupled rings. However, in situations wherethe filter needs to be tuned to a wavelength far away from the initial resonancedistribution, the time required for completing the tuning steps 1-4 may increase asthe heaters need to be swept across a wider range until the maximum photocurrentsare found. Here, we show that the as-fabricated resonance wavelengths of eachring, i.e., φn,0 in (3.1), can be located using IRPHs. This information, along withthe tuning efficiency of the heaters, can then be used to estimate the heater powersrequired to tune each ring to any desired wavelength. First, the input laser is set to aseries of known wavelengths and the heater powers that maximize the photocurrentmeasured by a ring’s IRPH at each wavelength are recorded. Figure 3.6(a) showsthe measured IPD,1s for a set of λin. Then, the heater powers corresponding to themaximum photocurrents is fitted linearly as a function of λin. Figure 3.6(b) showsthe recorded heater powers and the linear fit for each of the rings. The heater powerrequired to tune each ring to a specific wavelength is found using the linear fit. Inthe next section, we use this resonance mapping to provide initial estimates forV1−V4 required to tune to each wavelength, which reduces the time required bythe algorithm to reach the optimal state.3.4 Wavelength tuningIn this section, we demonstrate the automatic tuning of the four-ring Vernier filteracross the C-band for all of the ITU channels from 1528.77 nm to 1564.40 nm ona 200 GHz grid. The filter is configured at each channel as follows. First, theinput laser is set to the channel wavelength. Then V1−V4 are set to the valuescalculated from the resonance mapping technique described in the previous section.Next several iterations of the optimization step described in Section 3.2 is appliedby iteratively stepping V1−V4 to maximize IPD,4. Figure 3.7(a) shows all of thedrop-port spectra recorded after configuring the filter at each channel. Figure3.7(b) shows the overlay of the drop-port spectra relative to the center frequencyof each channel. For increasing channel wavelengths, the insertion losses and 3-dB bandwidths of the filter gradually changed from −7.5 dB to −4.5 dB and from32 GHz to 39 GHz. These changes can be attributed to the wavelength dependency40of coupling coefficients of the filter.1530 1535 1540 1545 1550 1555 1560 1565Wavelength (nm)-60-50-40-30-20-100NormalizedTransmission (dB)-200 -100 0 100 200Relative Frequency (GHz)-60-50-40-30-20-100Normalized Transmission (dB)(a)(b)Figure 3.7: Overlay of (a) measured drop-port spectra after automaticallyconfiguring the filter at each ITU (200 GHz grid) channel in C-Band,and (b) drop-port spectra relative to the center wavelength of each ITUchannel.Figures 3.8(a)-(d) show the initially-estimated and after-configured electricalpowers supplied to the heaters of R1- R4. We attribute the deviations to the thermalcrosstalk between the rings. The largest deviations are observed for R1 as it is the41most affected ring from thermal crosstalk due to the subsequent tuning of R2-R4.FSR tuning power of each ring is about 38 mW.1530 1535 1540 1545 1550 1555 156002040 estimatetuned1530 1535 1540 1545 1550 1555 156002040 estimatetuned1530 1535 1540 1545 1550 1555 156002040 estimatetuned1530 1535 1540 1545 1550 1555 156002040 estimatetunedWavelength (nm)Heater Powers (mW)R1R2R3R4Figure 3.8: Estimated and measured electrical powers required to tune rings(a) R1, (b) R2, (c) R3, and (d) R4, to each of the ITU wavelengths.3.5 Wavelength locking (Temperature stabilization)In this section, we demonstrate the wavelength locking of the filter to λin, toaccount for a drift in the temperature of the chip. After configuring the filter atλin, the optimization step described in the previous section is continuously iteratedto maintain the filter locked to λin. In the experiment, the input light is modulated420 250 500 750 1000Time (s)1020304050607080Stage Temperature (oC)64.9oCConstant temperature: Without stabilization:Temperature stabilization over 40 degreesWith stabilization:12.5 Gb/s12.5 Gb/s 12.5 Gb/s0.069oC/s(a)0 250 500 750 1000Time (s)1020304050607080Stage Temperature (oC)64.9oCConstant temperature: Without stabilization:Temperature stabilization over 40 degreesWith stabilization:12.5 Gb/s12.5 Gb/s 12.5 Gb/s0.069oC/s(b)0 250 500 750 1000Time (s)1020304050607080Stage Temperature (oC)64.9oCConstant temperature: Without stabilization:Temperature stabilization over 40 degreesWith stabilization:12.5 Gb/s12.5 Gb/s 12.5 Gb/s0.069oC/s(c)0 250 500 750 1000Time (s)1020304050607080Stage Temperature (oC)64.9oCConstant temperature: Without stabilization:Temperature stabilization over 40 degr esWith stabilization:12.5 Gb/s12.5 Gb/s 12.5 Gb/s0.069oC/s(d)Figure 3.9: (a) Measured eye diagram at constant stage temperature. (b)Measured stage temperature with time. Measured eye diagrams (c) withand (d) without automatic wavelength locking. (d) is recorded only fortemperature change of 8◦C.at 12.5 Gb/s with a 231− 1 PRBS NRZ data sequence. The drop-port output ofthe filter is monitored using a real-time oscilloscope. Figure 3.9(a) shows theeye-diagram recorded when the chip temperature is held constant. Figure 3.9(b)shows the variation in the chip temperature applied using the thermo-electric-cooler attached to the stage. Figure 3.9(c) shows the eye-diagram recorded withwavelength-locking turned on. The eye remains open for the entire duration ofthe temperature variation and no variation in the eye-height is noticed relative to43that of Fig. 3.9(a). The lower and the upper limits to the temperature variationare set to avoid any damage to the device. In this experiment, the slope of thetemperature increase is limited to about 0.07 oC/s by our measurement setup.Figure 3.9(d) shows the eye-diagram recorded when the wavelength locking isturned off, showing complete eye-closure for a temperature change of only about8oC.3.6 SummaryIn this chapter, we demonstrated the automatic configuration and wavelengthlocking of silicon photonics series-coupled Vernier ring resonator filters. Theseresults were achieved by monitoring the intra-cavity optical intensity as a measureof the rings’ resonance conditions and by thermo-optically tuning the rings untilthe desired resonance conditions are reached. Intra-cavity intensities of the ringresonators are measured using the integrated photoconductive heaters (i.e., IRPHs)which also acted as thermo-optic tuners. Performing both the sense and thetune operations using a single device obviated the need for additional materialdepositions, photodetectors and/or power taps and preserved the area efficiencyof these ring resonator devices. Automatic configuration of a Vernier ring filteracross a record 37.6 nm wavelength range and wavelength locking to account for arecord 65 oC temperature variation is demonstrated.44Chapter 4Tuning method for coupled ringresonator systems 1The methods that have been used for automatic tuning of coupled resonators,including the method we used in the previous chapter for wavelength locking,have relied on the fact that the desired output spectral shape of the filter is a localmaximum. Hence, it is difficult to apply these methods to any series-coupledresonator system - for example, to filters with un-apodized coupling coefficientswith several extrema in their output spectra [20, 73]. In this chapter, we describea tuning method which does not depend on the spectral shape of the output ofnumerous multi-ring filters. Rather, the method described here finds the desiredresonance condition of the nth ring, Rn, by normalizing the cavity intensity of Rn,by the cavity intensity of the Rn−1. Thus only one ring is tuned at a time, and atthe end of each tuning step, the exact φth,n satisfying (3.1) is found. In general,this method can be extended to other forms of coupled resonator systems such ascoupled cavities based on photonic crystals [74, 75] or gratings [76].4.1 Tuning methodWe consider a filter of N-series coupled resonators (shown in Fig. 3.1). In thetuning method presented in this section, rings RN (last ring) through R1 (first ring)1 c© of OSA. Reprinted, with permission from [45]45… Rn-1 Rn Rn+1 …Tune )*+,-Measure: IPD, n-1 IPD, nFigure 4.1: Tuning step for Rn. Photocurrents IPD,n and IPD,n−1 are measuredwhile tuning φth,n.are sequentially tuned, starting from RN . Figure 4.1 shows the step correspondingto tuning the nth ring, Rn. While tuning Rn, IPD,n and IPD,n−1 are measured and thefollowing optimization function is computed:fOPT,n(φth,n) =IPD,n(φth,n)IPD,n−1(φth,n)· In−1In, for n> 1IPD,n(φth,n)In, for n= 1. (4.1)Here In and In−1 are the theoretically calculated cavity intensities of Rn andRn−1 at λin, respectively [77], φth,n, or equivalently Vn, is set to the value thatmaximizes fOPT,n(φth,n). Maximizing fOPT,n(φth,n) ensures that the ratio of cavityintensities in rings Rn and Rn−1, measured by IPD,n/IPD,n−1, matches the expectedintensity ratio In/In−1. This, in turn, ensures that Rn is resonant at λin. When n= 1,fOPT,1(φth,1) reduces to maximizing the intensity in R1.In (4.1), fOPT,n(φ th,n) is always maximized when φn ≡ 0 (mod 2pi), regardlessof the resonance condition of Rn−1. The purpose of I PD,n−1 is to serve as anormalization factor for I PD,n as the amount of light coupling into Rn at any pointduring the tuning step will depend on the cavity intensity in Rn−1. Hence, underideal circumstances, only N tuning steps are required to configure a filter withN rings. However, in practice, to account for any thermal crosstalk between therings and to achieve wavelength locking, it is necessary to apply the tuning method46iteratively by stepping the electrical power supplied to the heaters. In (4.1), weassume that all of the IRPHs have the same responsivity. Otherwise, a furtherscaling factor can be applied to the photocurrent ratio I PD,n/I PD,n−1 to account forany differences in responsivities.The first step of the method tunes the last ring RN . If the initial resonancedistribution of the rings are far away from λin, light intensity in RN may beinsufficient to be measured with high accuracy. In this situation, initial resonancesof the rings can be set closer to λin by applying the sequential tuning steps shownin Fig. 3.4 or by using the resonance mapping technique shown in Section 3.3.4.2 Simulation resultsIn this section, we present the simulation results of the tuning method applied to afour-ring filter (N = 4) with a flat pass-band, and a five-ring filter (N = 5) whichdoes not have a flat pass-band. The transmission spectra of the filters as well as theintra-cavity intensities of the rings are calculated using the matrix method shownin [77]. A radius of 10 µm, an effective index of 2.57, a group index of 3.88,and a waveguide loss of 4 dB/cm is assumed for all of the rings. The before-tuningspectra are modeled by describing the initial round-trip phases, φ0,n, using a normaldistribution with a standard deviation σ = 0.0713pi representing the fabricationvariations [42, 78]. An offset of −2σ is also applied to all of the rings’ round-tripphases to represent the red-shift associated with the subsequent thermal tuning.Figure 4.2 shows the simulated transmission spectra of the flat pass-band filterbefore and after applying the tuning method. Here, for a maximally flat-passband [67], the coupling coefficients associated with the filter are chosen to be |κ|2= 0.2518, 0.0107, 0.0057, 0.0107, and 0.2518, from input bus waveguide to R1, R1-R2, R2- R3, R3- R4, and R4 to output bus waveguide, respectively. Figures 4.3(a)- (d) show the calculated optimization functions fOPT,4(φth,4) - fOPT,1(φth,1) for thefour steps of the tuning process as functions of φ4 - φ1, respectively. The intra-cavity intensities of the rings R4 - R1 are also presented. In each tuning step,fOPT,n(φth,n) is maximized when φn ≡ 0 (mod 2pi) indicating that the exact φth,ncorresponding to the desired resonant condition is found. The match between theafter-tuning spectra in Fig. 4.2 and the ideal filter spectra (not shown) is near-47perfect, with a slight deviation due to the finite step size used for the φth,n in thesimulations.(b)(a)Before-tuning: After-tuning:-1 -0.5 0 0.5 1Relative Wavelength (nm)-60-40-200ThroughDrop-1.5 -1 -0.5 0 0.5Relative Wavelength (nm)-60-40-200NormalizedTransmission (dB)ThroughDropFigure 4.2: Simulated (a) as-fabricated and (b) configured filter spectra offour-ring (N=4) filter with flat-top pass-band.Next, we apply the tuning method to a five-ring filter for which all of thecoupling coefficients are chosen to be |κ|2 = 0.015, resulting in a non-flat pass-band response with multiple extrema. Figure 4.4 shows the simulated transmissionspectra before and after applying the tuning method. The after-tuning spectra arenear-perfect matches with the ideal transmission spectra.In the cases shown above, φ th,n, the phase associated with thermal tuning isswept in order to find the desired resonance conditions. However, once the filteris initially configured, the tuning method will be required to be applied iterativelyin order to maintain the filter response locked at λin. While it is not shown here,we further verified that the tuning method converges to the exact solution when thetuning steps 1 through 4 are applied iteratively by stepping φ th,4 - φ th,1 in smalldiscrete steps, respectively.48-0.4 -0.2 0 0.2φ4 (/π)00.20.40.60.81f OPT,4 (A.U.)-40-35-30-25-20-15Cavity intensity (dB)Step 1 : tune R4fOPT,4I3(φ4)I4(φ4)-0.4 -0.2 0 0.2φ3 (/π)00.20.40.60.81f OPT,3 (A.U.)-35-30-25-20-15-10-5Cavity intensity (dB)Step 2 : tune R3fOPT,3I2(φ3)I3(φ3)-0.4 -0.2 0 0.2φ2 (/π)00.20.40.60.81f OPT,2 (A.U.)-15-10-50510Cavity intensity (dB)Step 3 : tune R2fOPT,2I1(φ2)I2(φ2)-0.4 -0.2 0 0.2φ1 (/π)00.20.40.60.81f OPT,1 (A.U.)-8-6-4-20246Cavity intensity (dB)Step 4 : tune R1fOPT,1I1(φ1)(a) (b)(c) (d)Figure 4.3: (a)-(d) Calculated optimization functions fOPT,1 - fOPT,4 and thecavity intensities of rings R1 - R4 (shaded) as a function of the round-tripphases φ1 - φ4.4.3 Experimental resultsIn this section, we apply the above tuning method to a fabricated four-ring filter.4.5 shows a micrograph of a fabricated device. All of the rings are designed to havean effective radius of 10.64 µm allowing the ring-ring and ring-bus couplers to have2 µm long coupling sections. The other design parameters are similar to those of49(b)(a)Before-tuning: After-tuning:-1.5 -1 -0.5 0 0.5Relative Wavelength (nm)-60-40-200NormalizedTransmission (dB)ThroughDrop-1 -0.5 0 0.5 1Relative Wavelength (nm)-60-40-200ThroughDropFigure 4.4: Simulated (a) as-fabricated and (b) configured filter spectra offive-ring (N=5) filter with un-apodized coupling coefficients.the Vernier device described in section 3.2. Figure 4.6 shows the filter spectra as-fabricated and after tuning to λin = 1552.2 nm. Here, we first apply the four tuningsteps described in Fig. 3.4 to bring the initial resonances of the rings close to λin.Next the four tuning steps are applied by sweeping the voltagesV4−V1, and findingthe voltages that maximized fOPT,n(φth,n) at the end of each sweep. Here, furtheriterations are not required to improve the filter response. By manually adjustingV4−V1, we further verify that the optimum filter response is reached.Due to the electrical crosstalk between the IRPHs, a multi-dimensional pre-calibration step is required (Section 3.2) in order to measure the photocurrents inthe Vernier device previously shown in Fig. 3.3. However, placing the IRPHs onthe opposite sides of the adjacent rings in the device shown in Fig. 4.5 reducesthe electrical crosstalk between the IRPHs. Hence, in this design, the calibrationstep for each ring reduces to the IV measurement of its IRPH i.e, Iheater,n = f (Vn).Despite the reduction in the sizes of the heaters in this design, the IRPHs stillsupport full-FSR tuning and photocurrents in the order of 10s of µA.50Figure 4.5: Microscope image of a fabricated four-ring filter.1550 1550.5 1551 1551.5 1552 1552.5 1553Wavelength (nm)-50-40-30-20-100NormalizedTransmission (dB)1549 1550 1551 1552Wavelength (nm)-60-40-200NormalizedTransmission (dB)ThroughDrop1551 1552 1553 1554Wavelength (nm)-60-40-200ThroughDrop(a) (b)As-fabricated: Configured:Figure 4.6: Measured (a) as-fabricated and (b) configured filter spectra offour-ring (N=4) filter.4.4 SummaryIn this chapter, we introduced a method for automatically tuning series-coupledring resonator filters. This method does not depend on the output spectral shapeof the filter, and, thus, can be applied to numerous multi-resonator systems.51Application of this tuning method was then demonstrated for various multi-ringfilters by both simulation and experiment.52Chapter 5Crosstalk in silicon ringresonator filters 1In this chapter, we experimentally investigate the interchannel and intrachannelcrosstalk of first- and second-order microring resonator (MRR) filters fabricatedon a silicon-on-insulator (SOI) platform. We find that there is an MRR radiusthat maximizes the WDM channel count given a waveguide geometry, a maximumtolerable insertion loss, and a minimum permissible adjacent channel isolation.The measured power penalties due to interchannel crosstalk of two-channelDeMUXs based on first-order and series-coupled MRR filters are presented asfunctions of channel spacing and adjacent channel isolation. Next, we comparethe intrachannel crosstalk of first-order, cascaded, and series-coupled MRR add-drop filters. Our results show that first-order MRR devices are unsuitable forsimultaneous add-drop operation at high data rates and small channel spacings.Intrachannel crosstalk of cascaded and series-coupled designs are measured asfunctions of the data rate and the level of detuning between the MRRs. Lowintrachannel crosstalk power penalties are demonstrated for cascaded and series-coupled MRR filters for data rates up to 20 Gb/s. Based on the measured results,we present requirements for the input-to-through response of add-drop filters thatwill ensure low intrachannel crosstalk.1 c© of IEEE. Reprinted, with permission from [46]535.1 IntroductionSilicon photonics devices using wavelength division multiplexing (WDM) promisean energy-efficient solution to the high-bandwidth demands of future data-center and high-performance computing applications [27, 79–81]. Due to theirstrong confineme nt of light, these devices offer small footprints and low powerconsumption. Recently, WDM transmitters [82–84], receivers [85, 86], and switchmatrices [87] have been demonstrated on silicon-on-insulator (SOI) platforms.Microring resonator (MRR) -based modulators [14], filters [88], and switches[40,89,90] are of significant interest for WDM systems because they typically offerlower power consumption and/or smaller footprints as compared to non-resonator-based devices (e.g., Mach-Zehnder-based modulators, filters, and switches [84,87, 91, 92], arrayed waveguide grating-based filters [93], and Bragg grating-basedfilters [94]).Crosstalk is a significant source of signal degradation in WDM systems, andcan be a limiting factor for the channel capacity or the allowable number of opticaladd-drop elements in a network [30, 40, 41, 95–97]. These signal impairments areprimarily due to: (1) interchannel crosstalk, where the carrier wavelength of thecrosstalk source is different than that of the signal, and (2) intrachannel crosstalk,where the carrier wavelength of the crosstalk source is same as that of the signal.In ultradense WDM systems, where the optical spectra of neighboring channelssignificantly overlap due to reduced channel spacing, this type of crosstalk canalso affect coherent detection systems [41]. In this work, we limit our discussionto direction detection systems.In order to illustrate each type of crosstalk in an MRR-based WDM network,we consider the simple WDM network shown in Fig. 1.6. Interchannel crosstalk isintroduced at the transmitter (TX) and at the receiver (RX), and both interchanneland intrachannel crosstalk are introduced at the optical add-drop elements. TheTX consists of a series of MRR modulators coupled to a single bus waveguide.The interchannel crosstalk at the TX occurs as a result of the overlap of the opticalpass-bands of the modulators. This is also called intermodulation crosstalk and haspreviously been experimentally investigated in [35].The RX consists of a wavelength demultiplexer (DeMUX), in which a series of54crosstalkthrough-portadd-portdrop-portinput-port through-portadd-portinput-portdrop-portcrosstalkcrosstalkInterchannel Crosstalk Intrachannel Crosstalk(a) (b)Figure 5.1: Illustrations of (a) interchannel crosstalk occuring at the drop-port of an MRR filter and (b) intrachannel crosstalk occuring at thethrough- and drop-ports of a series-coupled MRR filter.MRR filters are coupled to the bus waveguide for extracting the signals at differentcarrier wavelengths. Interchannel crosstalk occurs at the drop-port of each filteras the input-to-drop response of the filter does not completely suppress the signalsat other wavelengths. The interchannel crosstalk occuring at the drop-port of suchan MRR filter is schematically illustrated in Fig. 5.1(a). The input signals areillustrated as power spectral densities (PSDs). The signals at the through- anddrop-ports of the filter are the PSDs after filtering. The interchannel crosstalkcan be reduced by using higher-order MRR filters since the steeper roll-offs ofhigher-order filters can be used to increase the suppression of signals at otherwavelengths [98–100]. In this chapter, the order of the filter refers to the number ofMRRs in the filter (i.e., a first-order filter has one MRR, a second-order filter hastwo ...). The intrachannel crosstalk, which is introduced at the add-drop elements,occurs as a result of the residuals of the dropped signals interfering with the addedsignals, and vice versa [101]. This is schematically illustrated in Fig. 5.1(b) byshowing the PSDs of the signals at the four ports of a series-coupled MRR add-dropfilter. The intrachannel crosstalk can be reduced by using higher-order MRR filtersto increase the bandwidth and the extinction-ratio of the filters’ input-to-throughresponses, thereby reducing the residual of the dropped signals at the through-ports [102, 103].While increasing the order of a filter can reduce both the interchannel and55the intrachannel crosstalk, this usually comes at the expense of increased devicefootprint and power required for tuning and control.Telecom-grade applicationsmay still require higher-order filters for meeting their rigorous filter specifications(e.g., large free-spectral ranges, high input-to-through extinction ratios, etc.)[98,99]. For applications in data-centers and high performance computing systems,where power consumption and area are major concerns, MRR filters of lowerorders may be preferred. Therefore, understanding how the crosstalk affects theperformance of an MRR-based filter as the order or the number of stages of a filteris increased is of significant importance.Since filters are among the basic building blocks of a WDM network, it isuseful to analyze the impact of crosstalk on the performance of stand-alone MRR-based filters. Once this is well understood, the effects of crosstalk on an entireWDM network can be evaluated [27, 30, 104]. Effects of crosstalk in MRR-basedadd-drop filters have been investigated numerically [95, 105, 106]. MRR-basedadd-drop filters have been demonstrated with low interchannel and/or intrachannelcrosstalk [102, 103, 107, 108]. The level of crosstalk in the design of MRR-basedfilters is often specified by the level of isolation between the WDM channels[40, 88, 96, 100]. However, channel isolation does not necessarily corresponddirectly to a crosstalk power penalty, which is a metric that allows for the crosstalkeffects to be incorporated into the power budget of a WDM network. The powerpenalty due to crosstalk is defined as the power required at the RX to achievethe desired bit-error-ratio (BER) for a signal when it is impaired by crosstalkrelative to the power required in the absence of crosstalk sources. Typically,crosstalk power penalties are best obtained using numerical and/or experimentalmethods [35, 41, 105]. Experimentally comparing the effect of crosstalk on theperformance of MRR-based filters of various orders, in terms of power penalties,provides additional insight into a filter’s suitability for a particular application, ascompared to analyses based only on channel isolation.In the work presented in this chapter, which extends our results summarizedin [109] and [110], we experimentally investigate the interchannel and intrachannelcrosstalk of MRR-based first- and second-order filters. For the purposes of thischapter, any reference to cascaded or series-coupled MRR filters refers to second-order filters. For simplicity, we only consider second-order filters where both56 Channel 1Channel 2Normalized Transmission (dB)0Frequency (GHz)ILAdjacent Channel IsolationChannel Spacing3 dBFWHMFigure 5.2: Response of an MRR filter showing the insertion loss (IL) andadjacent channel isolation.of the MRRs have the same radii. In section 5.2, we calculate the maximumWDM channel count supported by MRR-based DeMUXs, given a particularwaveguide geometry, a maximum tolerable insertion loss (IL), and a minimumadjacent channel isolation. In section 5.3, we present measurements of interchannelcrosstalk power penalties as functions of the channel spacing and the adjacentchannel isolation for both first-order and series-coupled MRR filters. Imposinga minimum adjacent channel isolation, for a BER of 10−9 at 25 Gb/s, we calculatethe minimum channel spacing and the maximum data capacity for DeMUXsof first-order and series-coupled MRR filters. In section 5.4, we measure theintrachannel crosstalk of cascaded and series-coupled MRR filters by subjectingthem to simultaneous add-drop operation. We evaluate the power penalties asfunctions of data rate and the level of detuning between the MRRs of the second-order filters. While we achieve low power penalties for second-order MRR designs,our results show that first-order MRR filters have very high levels of intrachannelcrosstalk as a result of their narrow bandwidths.The simulation and measurement results presented in this chapter are for theSOI platform with 248 nm photolithography at the IME A*STAR foundry, whichwas used for the fabrication of all of the devices presented in this paper. All of theMRR designs were based on a rib waveguide geometry with a height of 220 nm, aslab height of 90 nm, and a rib width of 500 nm. Straight bus waveguides with equal576 7 8 9 10010203040Bend−loss (dB/cm) Radius (µm) simulatedmeasuredre−simulated (95.5 nm slab)(a)6 7 8 9 101020304050Finesse Radius (µm) 1st−order (simulated)1st−order (measured)series−coupled (simulated)series−coupled (measured)(b)6 7 8 9 10 11 1205101520253035Maximum Channel CountRadius (µm) 1st−order (simulated)1st−order (measured) series−coupled (simulated)series−coupled (measured)(c)Figure 5.3: (a) Measured and simulated bend-loss and (b) finesse calculatedfor first-order and series-coupled MRRs. (c) Maximum numberof channels supported by first-order and series-coupled MRR-basedDeMUXs.MRR-to-bus waveguide gaps at the through- and the drop-ports were used to couplelight into and out of the MRRs. All of the MRRs were made wavelength tunable byintegrating n-doped silicon resistive heaters [42]. The heaters occupied 63% of theMRRs’ perimeters. In these devices, about 50 mW of electrical power was requiredby a heater to shift an MRR’s resonance by an entire free spectral range (FSR). Ascompared to the devices presented in the last chapter, we attribute this increasein tuning power to the different heater design. In all of the experiments, we used58grating couplers to couple light into and out of the chips. For the grating couplers,typical ILs were about 7 dB and typical 3-dB bandwidths were about 34 GHz.5.2 Filter design for a maximum channel countIn a typical WDM link based on identical MRR-based filters, if the minimumchannel spacing is limited by interchannel crosstalk, then the maximum number ofchannels is determined by the FSR of the MRRs used. One method of increasingthe number of channels is to increase the FSR of the MRRs by reducing theirradii. However, reducing the radius of an MRR increases the bend-loss, due toradiation, and the MRR needs to be more strongly coupled to the bus waveguides inorder to maintain a reasonable drop-port IL. However, both increasing the couplingand the bend-losses cause an MRR’s filter response to broaden. Broadeningthe MRRs’ filter responses in a WDM system reduces the suppression of theneighboring channels and, thus, increases the interchannel crosstalk. This trade-off between an MRR’s FSR and the bend-loss-induced filter broadening is bestcaptured by the MRRs finesse (F ), which is defined as F = FSR/FWHM, whereFWHM is the filter response’s full-width-at-half-maximum (3-dB) bandwidth. Wefind that maximizing the finesse maximizes the channel count, given a minimumpermissible channel isolation, a maximum tolerable filter IL, and a waveguidegeometry. A particular waveguide geometry would determine how the bend-lossscales with the bend radius. The design constraint of filter IL and adjacent channelisolation are illustrated in Fig. 5.2. While the above design parameters may changefor a particular WDM link, the method described here for finding the radius thatmaximizes the channel count will still apply.Figure 5.3(a) shows how the bend-loss increases as the MRR radius is reduced.The simulated values were obtained using the Lumerical MODE Solutionssoftware package. For comparison, Fig. 5.3(a) also shows measured bend-losses ofseveral fabricated MRRs. The bend-losses of the fabricated devices were calculatedfrom the measured through- and drop-port spectra using the method describedin [59]. The mismatch between the simulated and measured bending losses is likelydue to the slab height being different from the design value (90 nm) as a result ofvariations during fabrication. Therefore, we also performed the simulations of the59bend-loss for a slab thickness of 95.5 nm and found that these “re-simulated” valuesclosely matched the measured values [Fig. 5.3(a)].Figure 5.3(b) shows the calculated finesse of a first-order and series-coupledfilter as a function of the radius for the simulated (90 nm slab height) and measuredvalues of bend-losses, designed to the constraint that the drop-port insertion lossbe less than 0.75 dB. In addition to the bend-losses, a 6 dB/cm doping loss anda 2 dB/cm propagation loss were also included in the calculations in order tomodel a realistic filter, which would use doped heaters for wavelength tuning.As an additional constraint, in the design of the series-coupled filter, the couplingcoefficients were calculated according to [67] in order to obtain a maximally flatdrop-port response.Figure 5.3(c) is a histogram showing the calculated maximum channel countsfor a DeMUX using MRR filters of 6, 7, 8, 9, 10, 11, and 12 µm radii. Here, tocalculate the maximum number of channels, we have used the adjacent channelisolation as a measure of the interchannel crosstalk [88]. The maximum channelcount was obtained by increasing the number of channels until the adjacent channelisolation reached a minimum of 15 dB, which was assumed to be the maximumpermissible crosstalk [85]. When increasing the number of channels, we changedthe channel spacing such that any two neighboring channels had the same spacing.Figure 5.3(c) also shows the maximum channel count obtained for a series-coupledMRR subjected to the same design and crosstalk constraints (i.e., 0.75 dB IL and15 dB adjacent channel isolation).The maximum channel counts calculated for both the simulated and themeasured bend-losses [Fig. 5.3(c)] follow the trend of the finesses shown in Fig.5.3(b). For devices with small radii channel counts calculated using measuredbend-losses are significantly lower than those calculated using simulated bend-losses. This is due to the large difference between the measured and simulatedbend-loss [Fig. 5.3(a)], and thereby the finesses [Fig. 5.3(b)], when the radius issmall (< 8 µm). For the results based on simulations, the largest channel count isobtained for a radius of 8 µm. For the case of a DeMUX based on first-order MRRfilters, this leads to a maximum of 19 channels with a channel spacing of 80 GHz.The maximum channel count increases to 32 with a channel spacing of 47 GHz forthe case of a DeMUX based on series-coupled MRR filters.60In this section, given a minimum permissible channel isolation, a maximumtolerable filter IL, and a waveguide geometry, we calculated the maximum channelcount for first-order and series-coupled MRR-based DeMUXs. Under theseconditions, we found that the radius which maximized the channel count for bothfirst-order and series-coupled MRRs corresponded to the radius that maximized thefinesse. The method outlined here can be used for estimating the maximum channelcount of an MRR filter-based DeMUX for a given SOI fabrication technology. Inthe results presented here, the high radiation losses of the rib waveguides, whichwere required in order to make the MRRs tunable by integrating doped heaters,limited the smallest radius that could be used.5.3 Interchannel crosstalkIn this section, we present measured results of interchannel crosstalk between twoadjacent channels of both first-order and series-coupled MRR-based DeMUXs toobtain power penalties as functions of the channel spacing. Here, for simplicity, weonly look at two-channel DeMUXs. Based on our measurements, we also calculatethe power penalty for a desired BER as a function of the adjacent channel isolationto obtain the minimum channel spacing, and the maximum number of channels.Figures 5.4(a) and (c) show microscope images of fabricated two-channelDeMUXs based on first-order MRR and series-coupled MRR filters, respectively.All of the MRRs were designed with a radii of 8 µm which maximized the finesseaccording to our simulations presented in section II. Figures 5.4(b) and (d) showmeasured spectra of the first-order and series-coupled MRR DeMUXs, with ameasured FWHM of 32.3 GHz and 38.7 GHz, respectively. The measured FSRfor both designs was 1.55 THz (or 12.4 nm). Both designs showed low drop-portILs below 1.8 dB.Figure 5.5 shows the experimental setup used for measuring the interchannelcrosstalk. Two independent pulse pattern generators (PPGs), outputting 25 Gb/snon-return-to-zero (NRZ) 231−1 pseudo random binary sequences (PRBS), wereused to drive two commercial LiNbO3 Mach-Zehnder modulators (MZMs). Thetwo optical signals were multiplexed using a bench-top 3-dB coupler prior to thechip. The output power of the two tunable laser sources (TLSs) and the bias and61In ThroughChannel 1 Channel 216 µm(a)−75 0 75 150−30−20−100Normalized Transmission (dB) Relative Frequency (GHz) Channel 1Channel 2Through(b)InThroughChannel 1Channel 216 µm(c)−75 0 62.5 150−30−20−100Normalized Transmission (dB) Relative Frequency (GHz) Channel 1Channel 2Through(d)Figure 5.4: (a) Microscope image of a fabricated first-order DeMUX and (b)the measured spectra after tuning the channels to a 75 GHz spacing.(c) Microscope image of a fabricated series-coupled DeMUX and (d)the measured spectra after tuning the channels to a 62.5 GHz spacing.The optical frequencies of (b) and (d) are relative to 193.019 THz (or1553.17 nm) and 192.888 THz (or 1554.23 nm), respectively.modulation voltages of the MZMs were adjusted so that the average optical powerand the eye-height of the signals at the input of the chip were equal. The delaybetween the two PRBS sequences was adjusted to align the bit-periods of thetwo data streams, which gave the worst case crosstalk [111]. The eye diagramsmeasured for the data streams at the input of the chip are shown in Figs. 5.6(a) and(b). Grating couplers were used to couple light into and out of the chip. The signaloutput from the chip was amplified using an erbium doped fiber amplifier (EDFA)and filtered using an optical tunable filter (OTF). The bandwidth of the OTF wasset to 4 nm, which was much larger than the channel spacing of the DeMUXs. Avariable optical attenuator (VOA) was used to control the received power at thephotodetector (PD). The PD was connected to an error-detector (ED) or to a digitalcommunications analyzer (DCA). The BER and the eye diagrams for channels 162and 2 were measured after tuning the channel spacing of the DeMUXs to differentvalues. For each BER measurement 1.25× 1012 bits were transmitted. Eyediagrams measured by setting the channel spacing of the DeMUXs to 37.5 GHzare shown in Figs. 5.6(c)-(f).MZM 2PPG 2MZM 1PPG 1EDFA3 dB CouplerDCATLS λ1TLS λ2ch1ch2ch1ch2DeMUX OTF4 nmEDPDVOAFigure 5.5: Experimental setup used for measuring the interchannelcrosstalk.First-order 37.5 GHz channel spacing (a)(b) (d)(c) (e)(f)Input Series-coupled37.5 GHz channel spacing Channel 1 Channel 2 7.1 mV/div7.1 mV/div13 mV/div13 mV/div13 mV/div13 mV/divFigure 5.6: Measured eye diagrams at 25 Gb/s for channels 1 and 2. (a) and(b) were measured at the input of the chip, (c) and (d) were measuredat the drop-port outputs of the first-order MRR DeMUX, and (e) and(f) were measured at the drop-port outputs of the series-coupled MRRDeMUX.Figure 5.7(a) shows measured BER curves as functions of the received opticalpower at the PD for a selected number of channel spacings. Figure 5.7(b) shows63−2 0 5 10−12−10−8−6−4−2Received Power (dBm)log(BER) 1st−order ch1 only1st−order 112.5 GHz1st−order 50 GHz1st−order 37.5 GHzseries−coupled ch1 onlyseries−coupled 50 GHzseries−coupled 37.5 GHz(a)−125 −100 −75 −50 −31.250.5246810Power Penalty (dB) Channel Spacing (GHz) 1st−order channel 11st−order channel 2series−coupled channel 1series−coupled channel 2(b)5 10 15 20 250246810Adjacent Channel Isolation (dB)Power Penalty (dB) 1st−order channel 1series−coupled channel 1(c)Figure 5.7: (a) Measured BER vs. received optical power at the PD, (b)crosstalk power penalty of channels 1 and 2 vs. channel spacing, and(c) power penalty of channel 1 vs. adjacent channel isolation (BER= 10−9, 25 Gb/s). The shaded markers in (b) and (c) indicate wherethe lowest measured BER was less than 10−9. The unshaded markersindicate where the measured BER values were extrapolated to calculatethe power penalty corresponding to a BER of 10−9.the measured crosstalk power penalties for a BER of 10−9 as functions of thechannel spacing. For each channel, the crosstalk power penalty was calculatedby normalizing the received power at a BER of 10−9 to the received power whenthe other channel was turned off. A linear fit to the BER curves [Fig. 5.7(a)] wasused to calculate the received power required to obtain a BER of 10−9 [112]. In Fig.645.7(b), the shaded markers indicate where the lowest BER value measured was lessthan 10−9. The unshaded markers indicate where the linear fits of the BER curveswere extrapolated below the lowest measured BER to estimate the power penaltycorresponding to a BER of 10−9.For an MRR-based DeMUX, the signal power in the bus waveguidecorresponding to a certain channel is significantly reduced after the channel isdropped by the filter tuned to the wavelength of the channel. As a result, thecrosstalk from this channel to subsequently dropped channels is reduced. Thisphenomenon can be clearly observed from the eye diagrams shown in Figs. 5.6(c)-(f), where the spacing between the two channels is set to a small value of 37.5 GHzin order to observe the effects of crosstalk. The eye diagram of channel 1 of thefirst-order MRR DeMUX [Fig. 5.6(c)] shows a high level of crosstalk compared tothat of channel 2. The signal amplitude of channel 2 is reduced as a large portionof the channel 2 signal power is dropped with channel 1 as the crosstalk signal.Eye diagrams recorded for the series-coupled MRR DeMUX [Figs. 5.6(e) and (f)]show similar behaviors. However, compared to the first-order design, the levelof crosstalk is small as a result of the larger channel isolation achieved using theseries-coupled design. As seen from Fig. 5.7(b), the power penalties measuredfor channel 2, for both first-order and series-coupled MRR DeMUXs, are lowercompared to those measured for channel 1.In Fig. 5.7(c), we show the measured power penalties for channel 1 of theDeMUXs as functions of the adjacent channel isolation (which is the metric weused as a measure of interchannel crosstalk for the channel count calculationspresented in section 5.2). The values for the adjacent channel isolation of thefirst-order and series-coupled devices were calculated using the measured filterresponses shown in Fig. 5.4(b) and (d), respectively. Table 5.1 shows the powerpenalty, the minimum channel spacing, and the maximum data capacity calculatedfor each of the DeMUXs based on the measured results. The minimum channelspacing and maximum data capacity values correspond to an adjacent channelisolation of 15 dB. The power penalty values consider crosstalk only from oneadjacent channel, and are for a BER of 10−9. Fig. 5.8 shows open eye diagramsrecorded for channel 1 of each DeMUX where the channel spacing is set close tothe minimum channel spacing given in Table 5.1. Eye diagrams for channel 2 are65not shown as channel 1 has a higher level of crosstalk as compared to channel 2.The differences between the maximum channel counts shown in Table 5.1 andthose shown in Fig. 5.3(c) are primarily due to the differences in the measuredFWHM values of the fabricated filters as compared to the calculated values insection 5.2. For example, in the case of the series-coupled filter, where thesimulated FWHM value in section 5.2 was 39.8 GHz and the measured FWHMvalue was 38.7 GHz, the calculated channel counts were found to be in very goodagreement, 32 and 31 respectively. For the values of minimum channel spacingshown in Table 5.1, the nearest non-adjacent channels would be 192 GHz apart forthe first-order and 100 GHz apart for the series-coupled designs. As seen from Fig.5.7(b), the interchannel crosstalk for such widely separated channels is negligible.Table 5.1: DeMUX performance comparison (BER = 10−9, adjacent channelisolation 15 dB)Power Penalty Channel Spacing CapacityFirst-order 0.76 dB 96 GHz 25 Gb/s × 16Series-coupled 0.35 dB 50 GHz 25 Gb/s × 31First-order 100 GHz channel spacing Series-coupled50 GHz channel spacing 13 mV/div 9.9 mV/divFigure 5.8: Measured eye diagrams at 25 Gb/s for channel 1 of (a) the first-order MRR DeMUX (b) the series-coupled MRR DeMUX.665.4 Intrachannel crosstalkIn this section, we investigate the simultaneous add-drop performance of (1) first-order, (2) cascaded, and (3) series-coupled MRR filters, which are illustrated inFigs. 5.10(a)-(c), respectively. The signal that is inserted into the input-port anddropped at the drop-port is referred to as the DROP signal. The signal that isinserted to the add-port and added to the through-port is referred to as the ADDsignal. The signal paths for the ADD and the DROP signals for each filter areshown in Figs. 5.10(a)-(c). At the through-port, intrachannel crosstalk occurs as aresult of the interference of the residual of the DROP signal with the ADD signal.Since our filters operate at the peak of the add-to-through response, the input-to-through response tends to have a greater effect on the amount of intrachannelcrosstalk. Figure 5.9 shows the important parameters of an MRR-based add-drop filter’s input-to-through response, as used in this section. We investigate theintrachannel crosstalk as a function of the data rate and the detuning level betweenthe two MRRs of cascaded and series-coupled designs. Based on our experimentalresults, we then discuss the requirements for the input-to-through response of theseadd-drop filters needed to mitigate intrachannel crosstalk.Normalized Transmission (dB)Relative Frequency (GHz)BWSuppressionExtinction Ratio00Figure 5.9: input-to-through response of a series-coupled add-drop filtershowing the extinction ratio, suppression, and bandwidth (BW).Figures 5.10(d)-(f) show the measured input-to-through and add-to-throughspectra of first-order, cascaded, and series-coupled MRR filters, respectively. TheMRRs of the cascaded and series-coupled filters were tuned by aligning the67ADD-only ADD-DROP20 Gb/sinput-port through-portdrop-port add-portADD-only ADD-DROP20 Gb/sADD-only ADD-DROP20 Gb/sDROP ADDthrough-portdrop-portDROP ADDinput-portadd-portADDDROPadd-port drop-portinput-port through-portCascaded filterFirst-order filter Series-coupled filter(a)ADD-only ADD-DROP20 Gb/sinput-port through-portdrop-port add-portADD-only ADD-DROP20 Gb/sADD-only ADD-DROP20 Gb/sDROP ADDthrough-portdrop-portDROP ADDinput-portadd-portADDDROPadd-port drop-portinput-port through-portCascaded filterFirst-order filter Series-coupled filter(b)ADD-only ADD + DROP20 Gb/sinput-port through-portdrop-port add-portADD-only ADD + DROP20 Gb/sADD-only ADD + DROP20 Gb/sDROP ADDthrough-portdrop-portDROP ADDinput-portadd-portADDDROPadd-port drop-portinput-port through-portCascaded filterFirst-order filter Series-coupled filter(c)−75 −50 −25 0 25 50 75−25−20−15−10−50Relative Frequency (GHz)Normalized Transmission (dB) add−to−throughinput−to−through(d)−75 −50 −25 0 25 50 75−25−20−15−10−50Relative Frequency (GHz)Normalized Transmission (dB) add−to−throughinput−to−through(e)−75 −50 −25 0 25 50 75−25−20−15−10−50Relative Frequency (GHz)Normalized Transmission (dB) add−to−throughinput−to−through(f)ADD-only ADD + DROPADD-only ADD + DROP ADD-only ADD + DROP20 Gb/s 20 Gb/s 20 Gb/s10 mV/div 13 mV/div 16 mV/div(g)AD -only AD + DROPAD -only AD + DROP AD -only AD + DROP20 Gb/s 20 Gb/s 20 Gb/s10 mV/div 13 mV/div 16 mV/div(h)AD -only ADD + DROP AD -only AD + DROP / 20 Gb/s 20 Gb/s / i 13 mV/div 16 mV/div(i)Figure 5.10: Schematics of a (a) first-order, (b) cascaded, and (c) series-coupled add-drop filter. Measured add-to-through and input-to-through spectra for a (d) first-order, (e) cascaded, and (f) series-coupled add-drop filter (Frequencies are relative to 193.195 THz (or1551.76 nm), 193.356 THz (or 1550.46 nm), and 193.498 THz (or1549.33 nm), respectively). Eye diagrams measured at the through-port for ADD signal only and ADD + DROP signals for a (g) first-order, (h) cascaded, and (i) series-coupled add-drop filter.resonance wavelengths of the MRRs to maximize the extinction ratio of the input-to-through response. The cascaded design used two MRRs which were similar indesign to that of the first-order MRR filter.Figure 5.11 shows the experimental setup used for measuring the BERs and theeye diagrams. Similar to section III, the MZMs were modulated using two PPGsoutputting NRZ, 231 − 1 PRBS data. The output powers of TLSs and the bias68MZM 2PPG 2MZM 1PPG 1EDFA OTF4 nmTLS1 λ0TLS2 λ0ADDDROPDCAEDPDVOAAdd-Drop FilterFigure 5.11: Experimental setup used for measuring the intrachannelcrosstalk.and modulation voltages of the MZMs were set to obtain equal average powersat the input grating couplers to the chip. The relative delay between the two datastreams were also adjusted to align the bit-periods of the input signals. Figures5.10(g)-(i) show the measured eye diagrams for each filter for ADD signal onlyand for simultaneous ADD + DROP signals at a data rate of 20 Gb/s. The eyediagram of the first-order MRR, as shown in Fig. 5.10(g), is completely closed. Weobserved that the eye diagram remained closed even after reducing the data rate to5 Gb/s. This is because the narrow bandwidth and the low extinction ratio of theinput-to-through response of the first-order MRR filter is incapable of sufficientlyrejecting the DROP signal at the through-port. Open eye diagrams were obtainedfor cascaded and series-coupled designs [Figs. 5.10(h)-(i)].5.4.1 Intrachannel crosstalk vs. data rateAs the data rate of an NRZ signal is increased, the bandwidth of its spectrumincreases. As a result, more of the energy in the DROP signal falls outside thefrequencies which are strongly suppressed by the input-to-through response of theadd-drop filter. This increases the residual of the DROP signal at the through-portand, hence, increases the intrachannel crosstalk.We measured the intrachannel crosstalk power penalties as functions of the datarate to investigate the maximum data rate at which the cascaded and series-coupleddesigns could support simultaneous ADD + DROP signals. Figures 5.12(a) and (b)show the measured BERs as functions of the received power at the PD for selected69−3 0 5 10 13−12−10−8−6−4−2Received Power (dBm)log(BER) 15 Gb/s ADD+DROP20 Gb/s ADD+DROP25 Gb/s ADD+DROP15 Gb/s ADD−only20 Gb/s ADD−only25 Gb/s ADD−only(a)−3 0 5 10 13−12−10−8−6−4−2Received Power (dBm)log(BER) 15 Gb/s ADD+DROP20 Gb/s ADD+DROP25 Gb/s ADD+DROP15 Gb/s ADD−only20 Gb/s ADD−only25 Gb/s ADD−only(b)10 12.5 15 17.5 20 22.5 25012345678Power Penalty (dB) Data Rate (Gb/s) cascaded MRRsseries−coupled MRRs(c)Figure 5.12: Measured BER vs. received optical power at the PD for a (a)cascaded and a (b) series-coupled filter design at several data rates.(c) Intrachannel crosstalk power penalty in simultaneous add-dropoperation for various data rates (BER = 10−9). The shaded markersin (c) indicate where the lowest measured BER was less than 10−9.The unshaded markers indicate where the measured BER values wereextrapolated to calculate the power penalty corresponding to a BER of10−9.data rates for cascaded and series-coupled MRR filters, respectively. Figure 5.12(c)shows the measured intrachannel crosstalk power penalties as functions of the datarate for a BER of 10−9. The power penalties for ADD + DROP signals werecalculated by normalizing the received optical powers, at a BER of 10−9, by the70received optical powers required to achieve the same BER for ADD signals. Theshaded markers in Fig. 5.12(c) correspond to the cases in which the minimummeasured BER was below 10−9, and the unshaded markers show where the linearfits to the BER curves were extrapolated in order to estimate the received powergiving a BER of 10−9.The crosstalk power penalties [Fig. 5.12(c)], which remained below 1 dB forlow data rates, dramatically increased for data rates above 15 Gb/s. The powerpenalties for both designs are similar due to the similarity of the input-to-throughresponses [see Figs. 5.10(e) and (f)].5.4.2 Intrachannel crosstalk vs. detuningIn cascaded and series-coupled add-drop filters, the input-to-through response ishighly sensitive to the tuning of the two MRRs of the filter. Therefore, by detuningthe two MRRs, the extinction ratio of the input-to-through response can be adjustedwithout dramatically changing the add-to-through response. In this section, weinvestigate the intrachannel crosstalk of a cascaded and series-coupled add-dropfilter, as functions of the input-to-through extinction ratio, by detuning the MRRsas described above.Figure 5.13(a) shows the measured input-to-through spectra of a cascadedMRR filter for several values of detuning. The detuning level indicated in Fig.5.13(a) refers to the frequency difference between the two local minima of theinput-to-through spectrum, which correspond to the resonances of the two MRRscomprising the cascaded filter. The FWHM of the add-to-through spectrumcorresponding to 0 GHz detuning was 28.4 GHz. Figure 5.13(b) shows themeasured BER curves at a data rate of 20 Gb/s. Figure 5.13(c) shows the measuredintrachannel crosstalk power penalty for a BER of 10−9. The power penalty valueswere calculated in a similar manner to the previous section and the shaded markersrepresent the cases where a BER below 10−9 was measured. The lowest powerpenalty measured was 1.1 dB, which corresponded to the 0 GHz detuning. Thepower penalty dramatically increases for detunings above 6 GHz, mainly as a resultof the reduction in the extinction ratio of the input-to-through response.Input-to-through spectra for a series-coupled add-drop filter measured with71−100 −75 −50 −25 0 25 50 75 100−30−25−20−15−10−50Normalized Transmission (dB) Relative Frequency (GHz) add−to−throughdetuning=0 GHzdetuning=6 GHzdetuning=12 GHz(a)−2 0 5 10−12−10−8−6−4−2Received Power (dBm)log(BER) detuning=0 GHz ADD−onlydetuning=0 GHz ADD+DROPdetuning=6 GHz ADD+DROPdetuning=12 GHz ADD+DROP(b)0 2 4 6 8 10 120123456Power Penalty (dB) Detuning (GHz)(c)Figure 5.13: Measured (a) spectra, and (b) BER vs. received optical powerat the PD for several detunings of a cascaded MRR filter design.(c) Intrachannel crosstalk power penalty vs. detuning (BER = 10−9,20 Gb/s). The shaded markers in (c) indicate where the lowestmeasured BER was less than 10−9. The unshaded markers indicatewhere the measured BER values were extrapolated to calculate thepower penalty corresponding to a BER of 10−9. Optical frequencyin (a) is relative to 193.209 THz (or 1551.65 nm).detuning is shown in Fig. 5.14(a). As the individual resonances of the two MRRscannot be separately identified from the spectra, we use the extinction ratio ofthe input-to-through response to express the level of detuning between the twoMRRs. When the input-to-through extinction ratio was tuned to its maximum72value, the measured FWHM of the add-to-through spectrum was 38.7 GHz. Figure5.14(b) shows the measured BER curves as functions of the received power at adata rate of 20 Gb/s. Figure 5.14(c) shows the intrachannel power penalty valuescorresponding to a BER of 10−9.In Fig. 5.14(c), the lowest power penalty measured was 1.5 dB, whichcorresponded to the highest input-to-through extinction ratio of 25.5 dB. The powerpenalty values dramatically increased for extinction ratios less than 18 dB. Sincethe change in the input-to-through bandwidth for various detunings is small, theincrease in the power penalties can be attributed to the reduction in the extinctionratio.5.4.3 Requirements for the input-to-through response of add-dropfiltersBased on the experimental results presented in sections 5.4.1 and 5.4.2, we nextevaluate the requirements that should be met by an add-drop filter in its input-to-through response for mitigating intrachannel crosstalk. In order to compareall of the input-to-through responses of the add-drop filters presented in theprevious sections, we calculated the bandwidth (BW) of the measured input-to-through responses at various levels of input-to-through suppression (see Fig.5.9). The calculated bandwidths were then normalized to the data rate used ineach experiment. The normalization of the bandwidth to data rate allows for thecomparison of the input-to-through responses of different add-drop filters whensubjected to various data rates.Figure 5.15 shows the BW/data rate as a function of the input-to-throughsuppression for the devices presented in sections 5.4.1 and 5.4.2. To calculate thecurve corresponding to the series-coupled filter in section 5.4.1 (black), we usedthe input-to-through response shown in Fig. 5.10(f) and a data rate of 15 Gb/s, forwhich the crosstalk power penalty [Fig. 5.12(c)] was 0.88 dB. The shaded areaabove this BW/data rate curve in Fig. 5.15 represents the region where an input-to-through response of a given filter would suppress the residual of the DROPsignal better than the series-coupled filter in section 5.4.1. Therefore, any filterwhose BW/data rate curve falls entirely within the shaded region can be expectedto achieve less than 0.88 dB power penalty value at a BER of 10−9.73The BW/data rate curves representing the cascaded (green) and series-coupled(blue) add-drop filters presented in section 5.4.2 are also shown in Fig. 5.15. Thesecurves were calculated using the input-to-through responses corresponding to a0 GHz detuning, which also corresponded to the input-to-through responses witha maximum extinction ratio. A data rate of 20 Gb/s, similar to that used in theexperiment, was used to normalize the bandwidths. A significant portion of theBW/data rate curves for these filters fall below the shaded region. This is consistentwith the observation that the power penalties for these filters were greater than0.88 dB. In Fig. 5.15, we also show the curve corresponding to the first-orderMRR filter calculated using the input-to-through response shown in Fig. 5.10(d)and assuming a data rate of 15 Gb/s. This curve is well below the shaded region,indicating a very high level of intrachannel crosstalk, which was confirmed by theclosed eye diagram shown in Fig. 5.10(g).In this section we showed that low-crosstalk simultaneous add-drop datatransmission is challenging for first-order MRR filters at high data rates. Thisis because a high level of input-to-through suppression over a large bandwidth isdifficult to achieve using a first-order MRR design. As for the cascaded and series-coupled designs, where a large input-to-through suppression can be maintainedover a wide bandwidth, we demonstrated that low crosstalk simultaneous ADD-DROP data transmission for data rates up to 20 Gb/s can be achieved.5.5 SummaryWe have experimentally evaluated interchannel and intrachannel crosstalk inMRR-based add-drop filters designed on an SOI photonics platform. Given amaximum permissible filter IL and a waveguide geometry, we found the radiuswhich maximized the finesse of the first-order and series-coupled MRR filters.Constraining the minimum tolerable adjacent channel isolation to a value of 15 dB,we calculated the maximum channel count for MRR-based DeMUXs for variousMRR radii. The maximum channel count strongly correlated with the MRR-based filter’s finesse. By using two-channel DeMUXs based on first-order andseries-coupled MRR filters, we measured the crosstalk power penalties due to anadjacent channel as a function of the channel spacing. The measured crosstalk74power penalties were also expressed as functions of the adjacent channel isolation.For an adjacent channel isolation of 15 dB, for a BER of 10−9 at a data rate of25 Gb/s, the measured crosstalk power penalties were 0.76 dB for our first-orderand 0.35 dB for our series-coupled designs. For a WDM DeMUX limited onlyby the crosstalk from an adjacent channel, these values correspond to maximumaggregate data rates up to 400 Gb/s for our first-order and 775 Gb/s for our series-coupled designs. By measuring the simultaneous optical add-drop performance ofcascaded and series-coupled MRR-based filters, as functions of the data rate andthe detuning of the two MRRs, we evaluated the requirements that should be met bythe input-to-through response of add-drop filters to ensure low power penalties. Wedemonstrated intrachannel crosstalk penalties of 1.1 dB and 1.5 dB, for a BER of10−9 at a data rate of 20 Gb/s for cascaded and series-coupled MRR filter designs,respectively.The method that we outlined in section 5.2 can be used to estimate themaximum channel count of an MRR filter-based DeMUX for a given SOIfabrication technology. While we have observed low interchannel crosstalk forfirst-order MRR filter designs at small channel spacings (96 GHz), our results showthat simultaneous add-drop operation is challenging for first-order filter designsat high data rates. This is because it is difficult to maintain a high level ofsuppression of the input-to-through response of a first-order MRR filter over a widebandwidth. Conversely, second-order MRR designs allow for low intrachannelcrosstalk at higher data rates by offering larger input-to-through suppression overwide bandwidths.75−100 −75 −50 −25 0 25 50 75 100−30−25−20−15−10−50Normalized Transmission (dB) Relative Frequency (GHz) add−to−throughER =23.9 dBER =16.5 dBER =13.6 dB(a)−2 0 5 10−12−10−8−6−4−2Received Power (dBm)log(BER) ADD−onlyER=23.9 dB ADD+DROPER=16.5 dB ADD+DROPER=13.6 dB ADD+DROP(b)14 16 18 20 22 24 261234567Power Penalty (dB) Extinction Ratio (dB)(c)Figure 5.14: Measured (a) spectra, and (b) BER vs. received optical powerat the PD after tuning the extinction ratio of a series-coupled MRRfilter. (c) Intrachannel crosstalk power penalty vs. input-to-throughextinction ratio (BER = 10−9, 20 Gb/s). The shaded markers in (c)indicate where the lowest measured BER was less than 10−9. Theunshaded markers indicate where the measured BER values wereextrapolated to calculate the power penalty corresponding to a BERof 10−9. Optical frequency in (a) is relative to 193.607 THz (or1548.46 nm).760 5 10 15 20 25 3000.511.522.533.5BW/Data Rate [GHz/(Gb⋅s−1)]Suppression (dB) series−coupled (Sec. IV.A)series−coupled (Sec. IV.B)cascaded (Sec. IV.B)first−order (15 Gb/s)- 5.4.1- S 5.4.2- Sec 5.4 3Figure 5.15: BW/data rate as a function of suppression of the input-to-through responses.77Chapter 6High-performance siliconmicroring resonatorsSilicon microring resonators (MRRs) with large free-spectral ranges (FSRs) andlow-power tuning are important for applications in data and telecommunicationsand in sensing. While silicon can support extremely small microring ring radiirequired for such large FSRs, standard photolithography resolutions used inmanufacturing (e.g., 193 nm, 248 nm) are often unable to define the narrow(typically < 100 nm) ring-to-bus waveguide spacings necessary for sufficientcoupling. In this chapter, MRR filters with small radii of 2.75 µm and large free-spectral ranges (FSRs) of 34.3 nm are demonstrated on a silicon-on-insulator (SOI)platform using a standard 248 nm photolithography. Light-coupling into and out ofthese extremely-small MRRs are facilitated using bus-waveguides that are wrappedaround a portion of the MRRs and are phase-matched to the ring waveguides.This approach increased the coupling length and thereby relaxed the waveguidespacing required. The MRRs are thermally tunable across their entire FSRs ata tuning efficiency of 2.79 nm/mW. Data transmission through a four-channelMRR demultiplexer (DeMUX) is demonstrated with low-crosstalk power penaltiesindicating that these devices can support dense WDM systems.786.1 IntroductionThe FSR of an MRR is the wavelength spacing between two consecutiveresonances. Therefore, the FSR typically dictates the number of WDM channels,and hence, the data capacity of an MRR-based system. The FSR is increased byreducing the MRR’s radius, which in turn reduces the number of wavelengths,m, allowed inside the resonator (see 1.1). Decreasing the radius of an MRR canalso increase its thermal tuning efficiency as the material volume that needs tobe heated for a given phase shift is reduced. Due to these reasons, reducing theradius provides a direct path towards improving the performance of MRR devices.Various other approaches such as Vernier filters [23, 45], ring filters with bragg-gratings [113], and MZI-coupling [114] have also been used for increasing FSRs.However, as compared to single-resonator devices, these approaches with multipledevices will generally require more tuning elements as well as larger on-chip areas.(a) Bending loss of SOI waveguides with their radius. Figure 6.1: Simulated bend-loss of silicon wir w veguides of 500 nm and600 nm as a function of the ring radius.On SOI platforms, thanks to the high-index contrast between the siliconwaveguides and the surrounding oxide, extremely small bend radii can be achievedwith relatively low bend-losses. Figure 6.1 shows the simulated bend-losses ofsilicon wire waveguides of widths 500 nm and 600 nm. However, reducing a ring’sradius also reduces the light coupling into the ring due to the reduced length of79the coupling region. As a result, the coupling coefficients required by modulatorsand filters with appreciable bandwidths (∼ 10s of GHz) and low insertion lossescannot be achieved by using the minimum gap sizes allowed by the standardphotolithography (e.g., 193 nm, 248 nm) fabrication resolutions. As a result,majority of the demonstrated MRR devices with extremely small radii (< 3 µm)so far have been fabricated using expensive electron-beam [115] or deep ultra-violet(UV) lithography [116].10 20 30 40FSR (nm)012345Tuning efficiency (nm/mW)[119][116][122][123][123][88][121] [120]10 20 30 40FSR (nm)012345Tuning efficiency (nm/mW)[119][116][122][123][123][88][121] [120][this work]deep UV deep UVFigure 6.2: Summary of published results of silicon microring resonatorsshown the tuning power vs, FSR. The arrow indicates the direction forperformance improvement.In this chapter, we demonstrate that a bus waveguide which is phase matchedto the ring waveuguide and wrapped around the ring can yield sufficient coupling,even at extremely small radii, while relaxing the minimum gap requirements.Previously, bend-couplers have been used for coupling into a specific resonantmode when the resonators’ waveguides are multi moded (e.g., in microdiskresonators) [117]. Initial passive-only results of extremely small bend-coupledMRR filters with 2.75 µm radii and 34.3 nm FSRs were presented in [118]. Bysurrounding these MRRs with thermal-isolation trenches, we demonstrate highlyefficient thermal tuning at 2.79 nm/mW spanning the entire FSR. Figure6.2,shows the tuning efficiency vs. FSR of these devices in comparison to those80θRrWrWb gInputDropThroughθ 2Rr(a)(b)Figure 6.3: (a) Bendcoupled microring. (b) Microscope image of a fabricateddevice.of several previously demonstrated devices [88, 116, 119–123]. In Fig. 6.2, allthe devices with FSRs > 30 nm used electron-beam or deep UV-lithography forfabrication. We also demonstrate the low-crosstalk performance of a four-channeldemultiplexer (DeMUX) using bend-coupled MRRs.6.2 DesignThe geometry of a bend-coupled MRR is shown in Fig. 6.3. The phase matchingcondition of the bus-waveguide to the ring waveguide is given by,ne f f ,bRb = ne f f ,rRr, (6.1)81where ne f f ,b and ne f f ,r are the effective mode indices and Rb and Rr are the radii ofthe bus- and the ring waveguides, respectively. Since Rb > Rr, the ring waveguideneeds to be wider than the bus waveguide for meeting the requirement in (6.1).Widening the ring waveguide is also beneficial as it reduces the bend-loss at small-radii (Fig. 6.1 ). Figure 6.4(a) shows the phase matching condition for an Rr =2.75 µm and for a ring waveguide width of Wr = 650 nm. This value of Wr ensuresthe single mode operation of the ring waveguide while minimizing the loss. Thegap between the ring and the bus is chosen to be 200 nm, which is the minimumgap allowed by the fabrication process. Phase matching is achieved when the buswaveguide width Wb = 357 nm. The mode indices ne f f ,b and ne f f ,r are calculatedusing a commercial mode solver software. Figure 6.4 (b) shows the 3D-FDTDsimulated coupling coefficient for the phase-matched condition as a function of thebend-angle, θ .A microscope picture of a fabricated device is shown in Fig. 6.3(b). In order toincrease the tuning efficiency, two thermal isolation trenches of 25×10µm2 weredefined on either side of the ring and the substrate under the ring was removed byan isotropic etch to prevent heat leakage to the substrate [124]. Similar techniqueshave previously been used for increasing the tuning efficiency of MRR [122]andMZI-based devices [124, 125] A U-shaped 2 µm wide tungsten heater was definedon top of the ring to allow for thermo-optic tuning. The devices were fabricated atIME A*STAR foundry.6.3 Results6.3.1 Spectral response and tuningFigure 6.5(a) show the ring spectra for several values of heater voltages andFig.6.5(b) shows the resonance wavelength shift as a function of the power suppliedto the heater. The tuning efficiency was found to be 2.79 nm/mW.6.3.2 Four-channel DeMUXFigure 6.6 (a) shows a schematic of a four channel wavelength demultiplexer andFig. 6.6 shows the microscope picture of a fabricated device. The measured spectra82350 400 450 500Wb (nm)2.42.62.833.2(neff,b × Rb)/R rneff,r(a)5 10 15 20 25 30θ (degrees) 00.020.040.060.080.10.12Power coupling |κ|2 (A.U.)3D fdtd(b)Figure 6.4: Phase matching of a 2.75 µm ring waveguide to a bus waveguideat 200 nm gap.1520 1530 1540 1550Wavelength (nm)-20-15-10-50Normalized Transmission (dB)0.0 V0.4 V0.6 V0.8 V1.0 V1.2 V(a)0 5 10 12.2 15Heater power (mW)010203040Wavelength shift (nm)FSRmeasuredlinear fit(b)Figure 6.5: Measured (a) drop-port spectra for various heater voltages. (b)Resonance wavelength shift as a function of the applied heater power.after setting the channel spacing of the DeMUX to 75 GHz is shown in Fig. 6.7(a)and (b), showing the 34.3 nm FSR.83inputch 1 ch 2 ch 3 ch 4through(a)(b)Figure 6.6: (a) Schematic of a 4-ring DeMUX with bend-coupled MRRfilters .(b) Microscope images of fabricated DeMUX.6.3.3 Crosstalk measurementsThe experimental setup used for bit-error-ratio (BER) measurement is shown inFig. 6.8. The four data streams were generated by driving LiNbO3 Mach-Zehnder modulators (MZMs) with non-return-to-zero 231 − 1 pseudo randombinary sequences (PRBS) generated by data and data outputs of two pulse-patterngenerators. Various lengths of fiber was used before the 4:1 off-chip multiplexerto minimize any correlation between the data streams. Multiplexed signal wassent through the on-chip DeMUX and each of the outputs from ch1 through ch4was amplified using an erbium doped fiber amplifier (EDFA), filtered using awide-band tunable filter (OTF) to remove any white noise due to the EDFA. Aphotodetector (PD) and a trans-impedance amplifier (TIA) served as the receiver(RX). The received signal was connected to an oscilloscope or to an error-detector(ED) for eye diagram or BER measurement, respectively. Power at the RX wascontrolled using a variable optical attenuator (VOA). For each BER measurement,1.5×1012 bits were transmitted.84-300 -225 -150 -75 0 75Relative Frequency (GHz)-50-40-30-20-100Normalized Transmission (dB)Throughch 1ch 2ch 3ch 4(a)1510 1520 1530 1540 1550 1560 1570Wavelength (nm)-50-40-30-20-100Normalized Transmission (dB)Throughch 1ch 2ch 3ch 4(b)Figure 6.7: a) Measured spectra of DeMUX after setting the channel spacingto 75 GHz. (b) Zoomed-out spectra of the DeMUX showing the 34.3nm FSRs of the MRR filters.MZM 2Data 2MZM 1Data 1EDFASCOPETLS λ1TLS λ2ch1ch2DeMUX OTF4 nmEDTIAVOAFig. \ref{setup1}Experimental setup used for measuring the interchannel crosstalk.MZM 4MZM 3TLS λ3TLS λ4ch3ch4MUX4:1On-ChipData 1Data 2Figure 6.8: Experimental setup used for the BER measurement.85BER results. Eye-diagrams. only ch1 (reference)ch1 – 75 GHz spacingch1 – 50 GHz spacingch2 – 75 GHz spacing ch3 – 75 GHz spacingch4 – 75 GHz spacing(a) (b) (c)(d) (e) (f)Figure 6.9: Eye diagrams measured for (a)-(d) ch1 - ch4 with 75 GHz channelspacing, (e) ch1 with 50 GHz channel spacing between the channels, (f)ch1 reference case when only ch1 was transmitted.Figures 6.9 (a)-(d) show the eye-diagrams recorded at 12.5 Gb/s for each of thechannels when all of the channels are transmitted and the channel spacing is set to75 GHz. Figure 6.9 (e) shows the eye diagram of ch1 when the channel spacing isset to 50 GHz. For the 50 GHz case, eye diagrams of ch2 - ch4 is not shown as ch1experiences the worst crosstalk due to the presence of other channels. As ch1 isdropped first, subsequent channels only experience crosstalk from the residual ofch1 [46]. Compared to Fig. 6.9 (f), which shows the eye diagram when only ch1 istransmitted through the system, only a slight reduction in eye height can be noticedin Figs. 6.9 (a)-(d). Even for the 50 GHz channel spacing, the eye diagram is wideopen [Figure 6.9 (e)] and the eye height is only slightly reduced as compared to thereference eye diagram.BER curves shown in Fig. 6.10 for each channel indicate that the crosstalkpower penalty for each of the channels due to the presence of other 3 channels inthe system is less than 0.3 dB at a BER of 10−9. These power penalties are reportedrelative to the reference measurements [shown as dashed lines in Fig. 6.10], whichwere performed by turning off all of the the other channels and only transmittingthe corresponding channel through the DeMUX. These BER results suggest that a86-11 -10 -9 -8 -7 -6Recieved power (dBm)-12-10-8-6-4-2log(BER)ch1ch2ch3ch4ch1-refch2-refch3-refch4-refFigure 6.10: Measured BER vs. optical power at the receiver for all of thechannels with 75 GHz channel spacing. The dotted curves show thereference measurements with only one channel turned on.system with such MRR filters can support approximately 56 channels at 12.5 Gb/swith 75 GHz spacing within a single FSR. This corresponds to a the crosstalklimited data capacity of 700 Gb/s. To a first order approximation, for data ratesbeyond 12.5 Gb/s, channel spacing can be increased proportionally to maintainthe data capacity for a similar crosstalk penalty (e.g., 28 channels at 25 Gb/s with150 GHz spacing).6.4 SummaryIn this chapter, we showed that phase-matched bend-couplers can provide a pathtowards designing extremely small silicon MRRs ( radii < 3 µm) with low insertionlosses and appreciable bandwidths using 248 nm photolithography fabrication. Wedemonstrated bend-coupled MRRs with 2.75 µm radii, made tunable across theirentire 34.3 nm FSRs at an efficiency of 2.79 nm/mW. Our simulations for bend-lossindicate that the radius can be further reduced to about 2.5 µm without incurringsignificant losses. In order to further improve the tuning efficiency, the length ofthe thermal isolation trenches can be increased to 50 µm without compromising themechanical stability of the structure [124]. Our crosstalk measurements suggestthat, if the entire FSR of the MRRs are utilized for WDM, the MRRs demonstratedhere have the ability to support data capacities up to 700 Gb/s with interchannelcrosstalk penalties < 0.3 dB.87Chapter 7Conclusion and future work7.1 ConclusionIn this thesis, we have demonstrated low-power and scalable techniques forautomatic tuning and stabilization of silicon microring devices and systems. Wealso presented a comprehensive investigation of the optical crosstalk of ringresonator based system. The key contributions of this thesis are summarized below.1. Demonstration of high-responsivity (on the order of 100 mA/W)photoconductive effects in doped silicon waveguides without anymodifications to standard fabrication processes.2. First use of silicon photoconductive heaters as dual-purpose elements forsensing and tuning the resonance conditions of ring resonator systems.Importantly, in systems with large numbers of resonators, such as series-coupled microring filters, the in-resonator photoconductive heaters (IRPHs)provide insights into the resonance conditions of individual resonators,beyond what can be determined from the systems outputs alone. SinceIRPHs do not require additional material depositions, photodetectors, powertaps and use the same contact pads for both heating and photocurrentmeasurements, these insights are obtained without compromising the costor the area efficiency of the devices.883. First demonstration of automatic tuning and stabilization of multi-ringresonator filters including Vernier filters. Automatic configuration of afour-ring Vernier ring filter across a record 37.6 nm wavelength range andwavelength locking to account for a record 65 oC temperature variation.4. Introduction of a tuning algorithm for automatic tuning of series-coupledmicrorings with any number of resonators (e.g., CROWs). This tuningmethod does not depend on the output spectral shape of most multi-ringfilters, and, thus, can be applied to numerous multi-resonator systems.5. First experimental investigation of optical crosstalk in one-ring and two-ringfilters. Based on the experimental evidence, we showed the interchannelcrosstalk limited maximum WDM channel capacity of the ring resonatorlinks comprising of said filters. We also obtained requirements to be met bya filter’s transmission spectra for mitigating intrachannel crosstalk.6. First experimental demonstration of microring resonator filters withextremely small radius of 2.75 µm using a bend-coupled configuration ona 248 nm photolithograpy platform. High-tuning efficiency of 2.79 nm/mWand a large tuning range spanning the entire 34.3 nm FSR was demonstrated.Experimental evaluation of crosstalk suggests the data capacity of a WDMusing such filters can be as high as 700 Gb/s.In this thesis, the use of IRPHs allowed us to sense and tune the resonanceconditions of individual resonators in multi-ring systems and, thereby, demonstratetuning methods that are scalable to systems with many rings. The use ofIRPHs also introduced several challenges. IRPHs used doped waveguides whichintroduced an additional loss. The additional loss due to the n-doping used inthe devices presented in this thesis varied from 4 - 7 dB/cm which limits theQ-factor of a critically coupled ring to approximately about 105− 5× 104 near1.55 µm. However, for application where a higher-Q is required, the level ofdoping can be traded-off against the responsivity of the IRPHs. IRPHs have darkcurrents (i.e., heater currents) on the order of mAs and they need to be biasedfor photodetection. However, since the dark current is used for tuning, it doesnot represent an additional power consumption in most practical scenarios. We89measured the photocurrent by subtracting the initially measured heater current(i.e., dark current) from the total current. This initial calibration step can beavoided by driving the heater with a pulse-width modulated (PWM) signal andby measuring the current only when the driver signal is high (one level). Sincecurrent is now measured across a constant voltage, the dark current level willbe constant for every measurement. Since the dark-current is in the order ofmAs, the subtraction performed in the present reading system also degrades theachievable signal-to-noise ratio (SNR). With the current reading system we haveexperimentally measured photocurrents in the sub µA range which corresponds tooptical powers less than 4 µW. However, further theoretical and experimental workis required in order to determine the achievable minimum sensitivity which may beimproved when using dedicated electronics.The tuning speed of the experimental demonstrations in this thesis waslimited by the speed of communication between the PC and the bench-top sourcemeasurement units used. Tuning speed in a real application can be improved bydedicated electronic circuitry. Initial characterization of the frequency response ofIRPHs show that they have 3 dB bandwidths in excess of 500 MHz, and, hence,would unlikely to impose a speed limitation for tuning. The conductivity of theIRPHs are temperature dependent. However, since the maximum search tuningalgorithms used in this work depend on the relative photocurrent measurements(i.e., by comparison of consecutive photocurrent measurements), any photocurrentmeasurement error associated with temperature variations can be neglectedassuming that the loop-bandwidth of the feedback loops is much larger than thefrequency of temperature variations. Our crosstalk analysis provided a frameworkbased on experimental results for evaluating optical crosstalk limitations of one-ring and two-ring filters. The crosstalk analysis of these simple building-blocks canbe used in ring resonator circuit models to predict the crosstalk limited behaviorsof large-scale systems [27, 126, 127]. Large-scale systems will also be proneto electrical and thermal crosstalk. Thermal isolation trenches [124, 125] andelectrical isolation using opposite type doping [128] can be used for mitigating theeffects of thermal and electrical crosstalk, respectively. The power consumption ofthe thermo optic tuning elements remains to be one of the the biggest challengestowards implementing large-scale practical systems [129]. As we presented in this90thesis, thermal isolation trenches [124, 125] can significantly reduce the tuningpower. However, this also increases the thermal response time thereby limiting thetuning speeds [125]. Furthermore, for series-coupled filters, it will be challengingto use thermal isolation trenches without increasing the thermal crosstalk betweenthe rings of the filters. While the contributions of this thesis pave a path towardsthe practical deployment of ring resonator-based systems, significant amountof research is still required before large-scale ring resonator-based systems areimplemented in practice. Several avenues of research that could build upon theproof-of-concept demonstrations presented in this thesis are outlined in the nextsection.7.2 Future workIn this thesis, the functionality of electronic circuitry required for ring resonatortuning and stabilization was demonstrated using bench-top power supplies andsource-measure devices. The control algorithms were implemented using apersonal computer. For practical systems, dedicated micro-electronic circuitry isneeded for driving the IRPHs. The heaters can be driven with a digital circuitfriendly PWM signal. Then a transfer-impedance-amplifier (TIA) that is syncedwith the PWM drive signal can be used for photocurrent measurement. Thering tuning algorithm proposed in section 4, consists of only signal addition andmultiplication. Such a tuning algorithm can be implemented using mixed signalcircuit design, without the need for dedicated processor unit such as a field-programmable gate array (FPGA) chip.The photoconductive heater-based tuning together with the crosstalk analysispresented in this thesis provided a basic framework towards increasing the numberof ring resonators in a system. Moving forward, on-chip programmable filtersand large-scale switch matrices can be demonstrated. The layout design of acascaded two-filter system is shown in Fig. 7.1. Photoconductive heaters integratedinto the couplers and the rings of the system allow for amplitude and bandwidthof the compound filter to be programmed. Multiple programmable filters canbe combined to further allow the filter response to be programmed for variousamplitude and phase responses.91InputDrop AddThroughwaveguides photoconductive heatersmetal routing(a)-1 -0.5 0 0.5 1Relative wavelength (nm)-80-60-40-200Transmission (dB)ThroughDrop(b)Figure 7.1: (a) Layout of a cascaded two-filter system with bandwidth andamplitude tunability. (b) Simulated through- and drop-port transmissionspectra when the pass-band is tuned to its maximum amplitude.Microscope pictures of a 16 × 16 microring switch matrix is shown in Fig7.2. The 256 ring resonators in the matrix only occupy an area of 2.35 mm × 2.6mm. As shown in Fig. 7.2 , an IRPH is used to both tune and sense the resonanceof each microring using a single electrical contact pad for both operations. Inthe device as shown, the on-chip area is dominated by the size of the contactpad (80 µm × 80 µm). The contact pad size can be further reduced according tothe specifications of the technology (e.g., flip-chip bonding) used for interfacingthis photonic switch to the driver circuitry. However, when reducing the pad sizethere may be a situation where the minimum spacing between the adjacent rings isdominated by the thermal crosstalk.92(a) (b)Figure 7.2: Microscope pictures of (a) a 16 × 16 microring-based switchmatrix and (b) a unit cell with microring with integrated IRPH and theoverlaid circuit description.93Bibliography[1] T. Baehr-Jones, R. Ding, A. Ayazi, T. Pinguet, M. Streshinsky, N. Harris,J. Li, L. He, M. Gould, Y. Zhang et al., “A 25 gb/s silicon photonicsplatform,” arXiv preprint arXiv:1203.0767, 2012. → pages 1[2] A. Masood, M. Pantouvaki, G. Lepage, P. Verheyen, J. Van Campenhout,P. Absil, D. Van Thourhout, and W. Bogaerts, “Comparison of heaterarchitectures for thermal control of silicon photonic circuits,” in Group IVPhotonics (GFP), 2013 IEEE 10th International Conference on. IEEE,2013, pp. 83–84. → pages 2[3] G. T. Reed, G. Mashanovich, F. Gardes, and D. Thomson, “Silicon opticalmodulators,” Nature photonics, vol. 4, no. 8, pp. 518–526, 2010. → pages 2[4] A. W. Fang, H. Park, O. Cohen, R. Jones, M. J. Paniccia, and J. E. Bowers,“Electrically pumped hybrid algainas-silicon evanescent laser,” Opticsexpress, vol. 14, no. 20, pp. 9203–9210, 2006. → pages 2[5] X. Zheng, S. Lin, Y. Luo, J. Yao, G. Li, S. S. Djordjevic, J.-H. Lee, H. D.Thacker, I. Shubin, K. Raj et al., “Efficient wdm laser sources towardsterabyte/s silicon photonic interconnects,” Journal of LightwaveTechnology, vol. 31, no. 24, pp. 4142–4154, 2013. → pages 2[6] W. D. Sacher, Y. Huang, L. Ding, B. J. Taylor, H. Jayatilleka, G.-Q. Lo, andJ. K. Poon, “Wide bandwidth and high coupling efficiency si 3 n 4-on-soidual-level grating coupler,” Optics express, vol. 22, no. 9, pp.10 938–10 947, 2014. → pages 2[7] L. Chen, C. R. Doerr, Y.-K. Chen, and T.-Y. Liow, “Low-loss andbroadband cantilever couplers between standard cleaved fibers andhigh-index-contrast si {3} n {4} or si waveguides,” IEEE PhotonicsTechnology Letters, vol. 22, no. 23, pp. 1744–1746, 2010. → pages 294[8] S. Assefa, S. Shank, W. Green, M. Khater, E. Kiewra, C. Reinholm,S. Kamlapurkar, A. Rylyakov, C. Schow, F. Horst et al., “A 90nm cmosintegrated nano-photonics technology for 25gbps wdm opticalcommunications applications,” in Electron Devices Meeting (IEDM), 2012IEEE International. IEEE, 2012, pp. 33–8. → pages 2[9] C. Sun, M. Georgas, J. Orcutt, B. Moss, Y.-H. Chen, J. Shainline, M. Wade,K. Mehta, K. Nammari, E. Timurdogan, D. Miller, O. Tehar-Zahav,Z. Sternberg, J. Leu, J. Chong, R. Bafrali, G. Sandhu, M. Watts, R. Meade,M. Popovic, R. Ram, and V. Stojanovic, “A monolithically-integratedchip-to-chip optical link in bulk CMOS,” IEEE J. Solid-State Circuits,vol. 50, no. 4, pp. 828–844, Apr. 2015. → pages 2[10] W. Bogaerts, P. De Heyn, T. Van Vaerenbergh, K. De Vos,S. Kumar Selvaraja, T. Claes, P. Dumon, P. Bienstman, D. Van Thourhout,and R. Baets, “Silicon microring resonators,” Laser & Photonics Reviews,vol. 6, no. 1, pp. 47–73, 2012. → pages 2, 5[11] L. Kimerling, D. Ahn, A. Apsel, M. Beals, D. Carothers, Y. Chen,T. Conway, D. Gill, M. Grove, C. Hong et al., “Electronic-photonicintegrated circuits on the cmos platform,” in Proc. SPIE, vol. 6125, no.612502, 2006, pp. 612 502–10. → pages 2[12] B. A. Small, B. G. Lee, K. Bergman, Q. Xu, and M. Lipson,“Multiple-wavelength integrated photonic networks based on microringresonator devices,” Journal of Optical Networking, vol. 6, no. 2, pp.112–120, 2007. → pages 3[13] A. V. Krishnamoorthy, K. W. Goossen, W. Jan, X. Zheng, R. Ho, G. Li,R. Rozier, F. Liu, D. Patil, J. Lexau et al., “Progress in low-power switchedoptical interconnects,” IEEE Journal of selected topics in quantumelectronics, vol. 17, no. 2, pp. 357–376, 2011. → pages 3[14] G. Li, A. V. Krishnamoorthy, I. Shubin, J. Yao, Y. Luo, H. Thacker,X. Zheng, K. Raj, and J. E. Cunningham, “Ring resonator modulators insilicon for interchip photonic links,” IEEE J. Sel. Topics QuantumElectron., vol. 19, no. 6, pp. 95–113, Nov. 2013. → pages 4, 54[15] J. J. Ackert, D. J. Thomson, L. Shen, A. C. Peacock, P. E. Jessop, G. T.Reed, G. Z. Mashanovich, and A. P. Knights, “High-speed detection at twomicrometres with monolithic silicon photodiodes,” Nat. Photon., vol. 9,no. 6, pp. 393–396, 2015. → pages 495[16] T.-Y. Liow, K.-W. Ang, Q. Fang, J.-F. Song, Y.-Z. Xiong, M.-B. Yu, G.-Q.Lo, and D.-L. Kwong, “Silicon modulators and germanium photodetectorson soi: monolithic integration, compatibility, and performanceoptimization,” IEEE Journal of Selected Topics in Quantum Electronics,vol. 16, no. 1, pp. 307–315, 2010. → pages 4[17] F. Xia, M. Rooks, L. Sekaric, and Y. Vlasov, “Ultra-compact high orderring resonator filters using submicron silicon photonic wires for on-chipoptical interconnects,” Opt. express, vol. 15, no. 19, pp. 11 934–11 941,2007. → pages 5, 33[18] P. Chen, S. Chen, X. Guan, Y. Shi, and D. Dai, “High-order microringresonators with bent couplers for a box-like filter response,” Opt. Lett.,vol. 39, no. 21, pp. 6304–6307, Nov 2014. → pages 5, 33[19] M. A. Popovic, T. Barwicz, M. S. Dahlem, F. Gan, C. W. Holzwarth, P. T.Rakich, H. I. Smith, E. P. Ippen, and F. X. Krtner, “Tunable, fourth-ordersilicon microring-resonator add-drop filters,” in 33rd European Conf. andEhxib. Opt. Commun. (ECOC). VDE, 2007, pp. 1–2. → pages 5, 33[20] J. R. Ong, R. Kumar, and S. Mookherjea, “Ultra-high-contrast andtunable-bandwidth filter using cascaded high-order silicon microringfilters,” IEEE Photon. Technol. Lett., vol. 25, no. 16, pp. 1543–1546, Aug.2013. → pages 5, 33, 45[21] P. Dong, N.-N. Feng, D. Feng, W. Qian, H. Liang, D. C. Lee, B. J. Luff,T. Banwell, A. Agarwal, P. Toliver, R. Menendez, T. K. Woodward, andM. Asghari, “GHz-bandwidth optical filters based on high-order siliconring resonators,” Opt. Express, vol. 18, no. 23, pp. 23 784–23 789, Nov2010. → pages 5, 33[22] R. Boeck, N. A. F. Jaeger, N. Rouger, and L. Chrostowski, “Series-coupledsilicon racetrack resonators and the vernier effect: theory andmeasurement,” Opt. Express, vol. 18, no. 24, pp. 25 151–25 157, Nov 2010.→ pages 5[23] R. Boeck, M. Caverley, L. Chrostowski, and N. A. F. Jaeger, “Siliconquadruple series-coupled vernier racetrack resonators: experimental signalquality,” in Opt. Fiber Commun. Conf. Optical Society of America, 2015,pp. W2A–8. → pages 5, 7996[24] P. Dong, Y. Chen, T. Gu, L. Buhl, D. T. Neilson, and J. H. Sinsky,“Reconfigurable 100 Gb/s silicon photonic network-on-chip,” J. Opt.Commun. Netw., vol. 7, no. 1, pp. A37–A43, Jan. 2015. → pages 7[25] L. Zhuang, C. G. Roeloffzen, M. Hoekman, K.-J. Boller, and A. J. Lowery,“Programmable photonic signal processor chip for radiofrequencyapplications,” Optica, vol. 2, no. 10, pp. 854–859, 2015. → pages 7[26] J. Wang, H. Shen, L. Fan, R. Wu, B. Niu, L. T. Varghese, Y. Xuan, D. E.Leaird, X. Wang, F. Gan et al., “Reconfigurable radio-frequency arbitrarywaveforms synthesized in a silicon photonic chip,” Naturecommunications, vol. 6, 2015. → pages 7[27] S. Rumley, D. Nikolova, R. Hendry, Q. Li, D. Calhoun, and K. Bergman,“Silicon photonics for exascale systems,” J. Lightw. Technol., vol. 33, no. 3,pp. 547–562, Feb. 2015. → pages 7, 54, 56, 90[28] C. A. Barrios, “Integrated microring resonator sensor arrays forlabs-on-chips,” Analytical and bioanalytical chemistry, vol. 403, no. 6, pp.1467–1475, 2012. → pages 7[29] A. Khilo, S. J. Spector, M. E. Grein, A. H. Nejadmalayeri, C. W.Holzwarth, M. Y. Sander, M. S. Dahlem, M. Y. Peng, M. W. Geis, N. A.DiLello, J. U. Yoon, A. Motamedi, J. S. Orcutt, J. P. Wang, C. M.Sorace-Agaskar, M. A. Popovic´, J. Sun, G.-R. Zhou, H. Byun, J. Chen,J. L. Hoyt, H. I. Smith, R. J. Ram, M. Perrott, T. M. Lyszczarz, E. P. Ippen,and F. X. Ka¨rtner, “Photonic ADC: overcoming the bottleneck of electronicjitter,” Opt. Express, vol. 20, no. 4, pp. 4454–4469, Feb 2012. → pages 7[30] N. Ophir, C. Mineo, D. Mountain, and K. Bergman, “Silicon photonicmicroring links for high-bandwidth-density, low-power chip I/O,” IEEEMicro, vol. 33, no. 1, pp. 54–67, Jan. 2013. → pages 7, 11, 14, 31, 54, 56[31] Y. Li, Y. Zhang, L. Zhang, and A. W. Poon, “Silicon and hybrid siliconphotonic devices for intra-datacenter applications: state of the art andperspectives,” Photonics Research, vol. 3, no. 5, pp. B10–B27, 2015. →pages 7[32] Z. Lu, J. Jhoja, J. Klein, X. Wang, A. Liu, J. Flueckiger, J. Pond, andL. Chrostowski, “Performance prediction for silicon photonics integratedcircuits with layout-dependent correlated manufacturing variability,” Opt.Express, vol. 25, no. 9, pp. 9712–9733, May 2017. → pages 897[33] J. C. Mak, W. D. Sacher, T. Xue, J. C. Mikkelsen, Z. Yong, and J. K. Poon,“Automatic resonance alignment of high-order microring filters,” IEEE J.Quantum Electron., vol. 51, no. 11, pp. 1–11, 2015. → pages 9, 14, 27, 31,33[34] J. A. Cox, A. L. Lentine, D. C. Trotter, and A. L. Starbuck, “Control ofintegrated micro-resonator wavelength via balanced homodyne locking,”Opt. Express , vol. 22, no. 9, pp. 11 279–11 289, 2014. [Online]. Available:http://www.opticsinfobase.org/abstract.cfm?URI=oe-22-9-11279 → pages9, 14, 15, 31[35] K. Padmaraju, X. Zhu, L. Chen, M. Lipson, and K. Bergman,“Intermodulation crosstalk characteristics of WDM silicon microringmodulators,” IEEE Photon. Technol. Lett., vol. 26, no. 14, pp. 1478–1481,Jul. 2014. → pages 9, 14, 15, 31, 54, 56[36] F. Morichetti, S. Grillanda, and A. Melloni, “Breakthroughs in photonics2013: Toward feedback-controlled integrated photonics,” IEEE Photon. J.,vol. 6, no. 2, pp. 1–6, 2014. → pages 9[37] Y. Li and A. W. Poon, “Active resonance wavelength stabilization forsilicon microring resonators with an in-resonatordefect-state-absorption-based photodetector,” Opt. Express, vol. 23, no. 1,pp. 360–372, 2015. → pages 9[38] S. T. Fard, K. Murray, M. Caverley, V. Donzella, J. Flueckiger, S. M. Grist,E. Huante-Ceron, S. A. Schmidt, E. Kwok, N. A. Jaeger et al.,“Silicon-on-insulator sensors using integrated resonance-enhanceddefect-mediated photodetectors,” Opt. Express, vol. 22, no. 23, pp.28 517–28 529, 2014. → pages 9, 14[39] K. Padmaraju and K. Bergman, “Resolving the thermal challenges forsilicon microring resonator devices,” Nanophotonics, vol. 3, no. 4-5, pp.269–281, 2014. → pages 10, 14[40] A. W. Poon, X. Luo, F. Xu, and H. Chen, “Cascaded microresonator-basedmatrix switch for silicon on-chip optical interconnection,” Proc. IEEE,vol. 97, no. 7, pp. 1216–1238, Jul. 2009. → pages 11, 54, 56[41] P. J. Winzer, M. Pfennigbauer, and R. Essiambre, “Coherent crosstalk inultradense WDM systems,” J. Lightw. Technol., vol. 23, no. 4, pp.1734–1744, Apr. 2005. → pages 11, 54, 5698[42] H. Jayatilleka, K. Murray, M. A´. Guille´n-Torres, M. Caverley, R. Hu,N. A. F. Jaeger, L. Chrostowski, and S. Shekhar, “Wavelength tuning andstabilization of microring-based filters using silicon in-resonatorphotoconductive heaters,” Opt. Express, vol. 23, no. 19, pp. 25 084–25 097,2015. → pages 11, 13, 33, 35, 47, 58[43] H. Jayatilleka, R. Boeck, K. Murray, J. Flueckiger, L. Chrostowski, N. A. F.Jaeger, and S. Shekhar, “Automatic wavelength tuning of series-coupledvernier racetrack resonators on SOI,” in Opt. Fiber Commun. Conf., 2016,pp. Th3J–5. → pages 12, 33[44] H. Jayatilleka, R. Boeck, M. AlTaha, J. Flueckiger, N. A. F. Jaeger,S. Shekhar, and L. Chrostowski, “Automatic tuning and temperaturestabilization of high-order silicon vernier microring filters,” in Opt. FiberCommun. Conf. (OFC). IEEE, 2017, pp. 1–3. → pages 12, 34[45] H. Jayatilleka, H. Shoman, R. Boeck, N. A. F. Jaeger, L. Chrostowski, andS. Shekhar, “Automatic configuration and wavelength locking of coupledsilicon ring resonators,” J. Lightw. Technol., vol. 34, no. 12, pp. 2886–2896,2017. → pages 12, 32, 45, 79[46] H. Jayatilleka, K. Murray, M. Caverley, N. A. F. Jaeger, L. Chrostowski,and S. Shekhar, “Crosstalk in SOI microring resonator-based filters,” J.Lightw. Technol., vol. 34, no. 12, pp. 2886–2896, 2016. → pages 12, 33,53, 86[47] F. Morichetti, S. Grillanda, and A. Melloni, “Breakthroughs in photonics2013: Toward feedback-controlled integrated photonics,” IEEE Photon. J.,vol. 6, no. 2, pp. 1–6, Apr. 2014. [Online]. Available:http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=6758361→ pages 14[48] C. T. DeRose, M. R. Watts, D. C. Trotter, D. L. Luck, G. N. Nielson, andR. W. Young, “Silicon microring modulator with integrated heater andtemperature sensor for thermal control,” in Conference on Lasers andElectro-Optics. Optical Society of America, 2010, p. CThJ3. → pages 14[49] Y. Zhang, Y. Li, S. Feng, and A. Poon, “Towards adaptively tuned siliconmicroring resonators for optical networks-on-chip applications,” IEEE J.Sel. Topics Quantum Electron., vol. 20, no. 4, pp. 136–149, Jul. 2014. →pages 1499[50] Y. Li and A. W. Poon, “Active resonance wavelength stabilization forsilicon microring resonators with an in-resonatordefect-state-absorption-based photodetector,” Opt. Express , vol. 23, no. 1,pp. 360–372, Jan. 2015. [Online]. Available:http://www.opticsinfobase.org/abstract.cfm?URI=oe-23-1-360 → pages14, 15, 30, 31[51] C. V. Poulton, P. Dong, and Y.-K. Chen, “Photoresistive microring heaterwith resonance control loop,” in Proc. Conference on Lasers andElectro-Optics. Optical Society of America, May 2015, p. SM2I.3. →pages 14, 15, 31[52] S. Grillanda, M. Carminati, F. Morichetti, P. Ciccarella, A. Annoni,G. Ferrari, M. Strain, M. Sorel, M. Sampietro, and A. Melloni,“Non-invasive monitoring and control in silicon photonics using CMOSintegrated electronics,” Optica, vol. 1, no. 3, pp. 129–136, Sep. 2014.[Online]. Available:http://www.opticsinfobase.org/abstract.cfm?URI=optica-1-3-129 → pages14, 15[53] H. Yu, D. Korn, M. Pantouvaki, J. Van Campenhout, K. Komorowska,P. Verheyen, G. Lepage, P. Absil, D. Hillerkuss, L. Alloatti, J. Leuthold,R. Baets, and W. Bogaerts, “Using carrier-depletion silicon modulators foroptical power monitoring,” Opt. Lett. , vol. 37, no. 22, pp. 4681–4683,2012. [Online]. Available: http://www.opticsinfobase.org/ol/fulltext.cfm?uri=ol-37-22-4681&id=245407→ pages 14, 18, 30[54] D. F. Logan, K. J. Murray, J. J. Ackert, P. Velha, M. Sorel, R. M.De La Rue, P. E. Jessop, and A. P. Knights, “Analysis of resonanceenhancement in defect-mediated silicon micro-ring photodiodes operatingat 1550 nm,” J. Opt., vol. 13, no. 12, pp. 125 503–125 511, Jan. 2011.[Online]. Available: http://stacks.iop.org/2040-8986/13/i=12/a=125503?key=crossref.60583ce3c8bf56c2895788853f1595ee → pages 14, 18[55] C. Li, R. Bai, A. Shafik, E. Z. Tabasy, B. Wang, G. Tang, C. Ma, C.-H.Chen, Z. Peng, M. Fiorentino, R. G. Beausoleil, P. Chiang, and S. Palermo,“Silicon photonic transceiver circuits with microring resonator bias-basedwavelength stabilization in 65 nm CMOS,” IEEE J. Solid-State Circuits,vol. 49, no. 6, pp. 1419–1436, Jun. 2014. → pages 14, 15, 31100[56] E. Timurdogan, A. Biberman, D. C. Trotter, C. Sun, M. Moresco,V. Stojanovic, and M. R. Watts, “Automated wavelength recovery formicroring resonators,” in Proc. Conference on Lasers and Electro-Optics.Optical Society of America, May 2012, p. CM2M.1. [Online]. Available:http://www.opticsinfobase.org/abstract.cfm?URI=CLEO SI-2012-CM2M.1 →pages 14, 15, 19, 31[57] P. De Heyn, J. De Coster, P. Verheyen, G. Lepage, M. Pantouvaki, P. Absil,W. Bogaerts, J. Van Campenhout, and D. Van Thourhout,“Fabrication-tolerant four-channel wavelength-division-multiplexing filterbased on collectively tuned Si microrings,” J. Lightw. Technol., vol. 31,no. 16, pp. 2785–2792, Aug. 2013. → pages 14, 22[58] X. Zheng, E. Chang, P. Amberg, I. Shubin, J. Lexau, F. Liu, H. Thacker,S. S. Djordjevic, S. Lin, Y. Luo, J. Yao, J.-H. Lee, K. Raj, R. Ho, J. E.Cunningham, and A. V. Krishnamoorthy, “A high-speed, tunable siliconphotonic ring modulator integrated with ultra-efficient active wavelengthcontrol,” Opt. Express , vol. 22, no. 10, pp. 12 628–12 633, May 2014. →pages 15[59] W. R. McKinnon, D. X. Xu, C. Storey, E. Post, A. Densmore, A. Delge,P. Waldron, J. H. Schmid, and S. Janz, “Extracting coupling and losscoefficients from a ring resonator,” Opt. Express, vol. 17, no. 21, pp.18 971–18 982, 2009. → pages 16, 59[60] H. Jayatilleka, M. Caverley, N. A. F. Jaeger, S. Shekhar, andL. Chrostowski, “Crosstalk limitations of microring-resonator based WDMdemultiplexers on SOI,” in Optical Interconnects Conference. IEEE,2015, p. TuB5. [Online]. Available:http://www.mina.ubc.ca/files/2015 OIC CrosstalkRings.pdf → pages 18[61] Haike Zhu, Linjie Zhou, Xiaomeng Sun, Yanyang Zhou, Xinwan Li, andJianping Chen, “On-chip optical power monitor using periodicallyinterleaved P-N junctions integrated on a silicon waveguide,” IEEE J. Sel.Topics Quantum Electron., vol. 20, no. 4, pp. 56–63, Jul. 2014. [Online].Available:http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=6719526→ pages 18[62] L. Zhou, H. Zhu, H. Zhang, and J. Chen, “Photoconductive effect on p-i-pmicro-heaters integrated in silicon microring resonators,” Opt. Express ,101vol. 22, no. 2, pp. 2141–2149, Jan. 2014. [Online]. Available:http://www.opticsinfobase.org/abstract.cfm?URI=oe-22-2-2141 → pages18[63] Y. Li, S. Feng, Y. Zhang, and A. W. Poon, “Sub-bandgaplinear-absorption-based photodetectors in avalanche mode inPN-diode-integrated silicon microring resonators,” Opt. Lett. , vol. 38,no. 23, pp. 5200–5203, Dec. 2013. [Online]. Available:http://www.opticsinfobase.org/abstract.cfm?URI=ol-38-23-5200 → pages18[64] M. S. Dahlem, C. W. Holzwarth, A. Khilo, F. X. Kartner, H. I. Smith, andE. P. Ippen, “Reconfigurable multi-channel second-order siliconmicroring-resonator filterbanks for on-chip WDM systems,” Opt. Express ,vol. 19, no. 1, pp. 306–316, 2011. → pages 22, 31[65] R. Boeck, “Silicon ring resonator add-drop multiplexers,” Master’s thesis,University of British Columbia, 2011. → pages 22[66] J. Poon, J. Scheuer, S. Mookherjea, G. Paloczi, Y. Huang, and A. Yariv,“Matrix analysis of microring coupled-resonator optical waveguides,” Opt.Express, vol. 12, no. 1, pp. 90–103, 2004. → pages 24[67] B. E. Little, S. T. Chu, H. A. Haus, J. Foresi, and J.-P. Laine, “Microringresonator channel dropping filters,” J. Lightw. Technol., vol. 15, no. 6, pp.998–1005, Jun. 1997. → pages 27, 47, 60[68] L. Chrostowski, X. Wang, J. Flueckiger, Y. Wu, Y. Wang, andS. Talebi Fard, “Impact of fabrication non-uniformity on chip-scale siliconphotonic integrated circuits,” in Optical Fiber Communication Conference.Optical Society of America, 2014, p. Th2A.37. [Online]. Available:http://www.opticsinfobase.org/abstract.cfm?uri=OFC-2014-Th2A.37 →pages 29[69] A. W. Poon, F. Xu, and X. Luo, “Cascaded active silicon microresonatorarray cross-connect circuits for WDM networks-on-chip,” in Proc. SPIE,vol. 6898, no. 689812, 2008, pp. 689 812–10. → pages 33[70] S. Xiao, M. H. Khan, H. Shen, and M. Qi, “A highly compact third-ordersilicon microring add-drop filter with a very large free spectral range, a flatpassband and a low delay dispersion,” Opt. express, vol. 15, no. 22, pp.14 765–14 771, 2007. → pages 33102[71] X. Luo, J. Song, S. Feng, A. W. Poon, T.-Y. Liow, M. Yu, G.-Q. Lo, andD.-L. Kwong, “Silicon high-order coupled-microring-based electro-opticalswitches for on-chip optical interconnects,” IEEE Photon. Technol. Lett.,vol. 24, no. 10, pp. 821–823, 2012. → pages 33[72] J. Mak, W. D. Sacher, J. C. Mikkelsen, and J. Poon, “Multivariable phasetuning control and its application to wavelength tracking in high-ordermulti-ring filters,” in CLEO: Science and Innovations, 2017, pp. SF2I–5.→ pages 33[73] R. Kumar, J. R. Ong, M. Savanier, and S. Mookherjea, “Controlling thespectrum of photons generated on a silicon nanophotonic chip,” Nat.Commun., vol. 5, p. 5489, Nov. 2014. → pages 45[74] H.-C. Liu and A. Yariv, “Designing coupled-resonator optical waveguidesbased on high-Q tapered grating-defect resonators,” Opt. Express, vol. 20,no. 8, pp. 9249–9263, 2012. → pages 45[75] D. O’Brien, M. Settle, T. Karle, A. Michaeli, M. Salib, and T. Krauss,“Coupled photonic crystal heterostructure nanocavities,” Opt. Express,vol. 15, no. 3, pp. 1228–1233, 2007. → pages 45[76] M. Burla, L. R. Corte´s, M. Li, X. Wang, L. Chrostowski, and J. Azan˜a,“Integrated waveguide bragg gratings for microwave photonics signalprocessing,” Opt. Express, vol. 21, no. 21, pp. 25 120–25 147, 2013. →pages 45[77] J. K. Poon, J. Scheuer, S. Mookherjea, G. T. Paloczi, Y. Huang, andA. Yariv, “Matrix analysis of microring coupled-resonator opticalwaveguides,” Opt. express, vol. 12, no. 1, pp. 90–103, 2004. → pages 46,47[78] L. Chrostowski, X. Wang, J. Flueckiger, Y. Wu, Y. Wang, and S. T. Fard,“Impact of fabrication non-uniformity on chip-scale silicon photonicintegrated circuits,” in Optical Fiber Commun. Conf., 2014, pp. Th2A–37.→ pages 47[79] A. Krishnamoorthy, R. Ho, X. Zheng, H. Schwetman, J. Lexau, P. Koka,G. Li, I. Shubin, and J. Cunningham, “Computer systems based on siliconphotonic interconnects,” Proc. IEEE, vol. 97, no. 7, pp. 1337–1361, Jul.2009. → pages 54103[80] A. V. Krishnamoorthy, H. Schwetman, X. Zheng, and R. Ho,“Energy-efficient photonics in future high-connectivity computingsystems,” J. Lightw. Technol., vol. 33, no. 4, pp. 889–900, Feb. 2015. →pages 54[81] P. Dong, Y. Chen, G. Duan, and D. T. Neilson, “Silicon photonic devicesand integrated circuits,” Nanophotonics, vol. 3, no. 4, Jan. 2014. → pages54[82] A. Ramaswamy, J. Roth, E. Norberg, R. S. Guzzon, J. Shin, J. Imamura,B. Koch, D. Sparacin, G. Fish, and B. G. Lee, “A WDM 4x28 Gbpsintegrated silicon photonic transmitter driven by 32 nm CMOS driver ICs,”in Opt. Fiber Commun. Conf. (OFC), 2015, p. Th5B.5. → pages 54[83] H. Li, Z. Xuan, A. Titriku, C. Li, K. Yu, B. Wang, A. Shafik, N. Qi, Y. Liu,R. Ding, T. Baehr-Jones, M. Fiorentino, M. Hochberg, S. Palermo, andP. Y. Chiang, “A 25Gb/s 4.4 V-swing AC-coupled Si-photonic microringtransmitter with 2-tap asymmetric FFE and dynamic thermal tuning in65nm CMOS,” in Proc. IEEE Int. Solid-State Circuits Conf. (ISSCC), Feb.2015, pp. 410–411. → pages 54[84] L. Chen, C. R. Doerr, P. Dong, and Y. Chen, “Monolithic silicon chip with10 modulator channels at 25 Gbps and 100-GHz spacing,” Opt. Express,vol. 19, no. 26, pp. B946–B951, 2011. → pages 54[85] C. G. Littlejohns, Y. Hu, F. Y. Gardes, D. J. Thomson, S. Reynolds, G. Z.Mashanovich, and G. T. Reed, “50 Gb/s silicon photonics receiver with lowinsertion loss,” IEEE Photon. Technol. Lett., vol. 26, no. 7, pp. 714–717,Jan. 2014. → pages 54, 60[86] M. Rakowski, M. Pantouvaki, P. De Heyn, P. Verheyen, M. Ingels,H. Chen, J. De Coster, G. Lepage, B. Snyder, and K. De Meyer, “22.5 A 4x 20 Gb/s WDM ring-based hybrid CMOS silicon photonics transceiver,”in Proc. IEEE Int. Solid-State Circuits Conf. (ISSCC), Feb. 2015, pp.408–409. → pages 54[87] L. Chen and Y. Chen, “Compact, low-loss and low-power 8x8 broadbandsilicon optical switch,” Opt. Express, vol. 20, no. 17, pp. 18 977–18 985,2012. → pages 54[88] P. Dong, W. Qian, H. Liang, R. Shafiiha, X. Wang, D. Feng, G. Li, J. E.Cunningham, A. V. Krishnamoorthy, and M. Asghari, “1x4 reconfigurabledemultiplexing filter based on free-standing silicon racetrack resonators,”104Opt. Express, vol. 18, no. 24, pp. 24 504–24 509, 2010. → pages 54, 56,60, 81[89] B. Lee, A. Biberman, J. Chan, and K. Bergman, “High-performancemodulators and switches for silicon photonic networks-on-chip,” IEEE J.Sel. Topics Quantum Electron., vol. 16, no. 1, pp. 6–22, Oct. 2010. →pages 54[90] P. Dong, Y. Chen, T. Gu, L. L. Buhl, D. T. Neilson, and J. H. Sinsky,“Reconfigurable 100 Gb/s silicon photonic network-on-chip,” J. Opt.Commun. Netw., vol. 7, no. 1, pp. A37–A43, Jan. 2015. → pages 54[91] G. Kim, J. W. Park, I. G. Kim, S. Kim, K. Jang, S. A. Kim, J. H. Oh, J. Joo,and S. Kim, “Compact-sized high-modulation-efficiency siliconMachZehnder modulator based on a vertically dipped depletion junctionphase shifter for chip-level integration,” Opt. Lett., vol. 39, no. 8, pp.2310–2313, Apr. 2014. → pages 54[92] F. Horst, W. M. Green, S. Assefa, S. M. Shank, Y. A. Vlasov, and B. J.Offrein, “Cascaded Mach-Zehnder wavelength filters in silicon photonicsfor low loss and flat pass-band WDM (de-)multiplexing,” Opt. Express,vol. 21, no. 10, pp. 11 652–11 658, 2013. → pages 54[93] S. Pathak, M. Vanslembrouck, P. Dumon, D. Van Thourhout, andW. Bogaerts, “Optimized silicon AWG with flattened spectral responseusing an MMI aperture,” J. Lightw. Technol., vol. 31, no. 1, pp. 87–93, Jan.2013. → pages 54[94] W. Shi, H. Yun, C. Lin, M. Greenberg, X. Wang, Y. Wang, S. T. Fard,J. Flueckiger, N. A. F. Jaeger, and L. Chrostowski, “Ultra-compact, flat-topdemultiplexer using anti-reflection contra-directional couplers for CWDMnetworks on silicon,” Opt. Express, vol. 21, no. 6, pp. 6733–6738, 2013. →pages 54[95] M. Nikdast, J. Xu, X. Wu, W. Zhang, Y. Ye, X. Wang, Z. Wang, andZ. Wang, “Systematic analysis of crosstalk noise in folded-torus-basedoptical networks-on-chip,” IEEE Trans. on Comput.-Aided Design Integr.Circuits Syst., vol. 33, no. 3, pp. 437–450, Mar. 2014. → pages 54, 56[96] K. Preston, N. Sherwood-Droz, J. S. Levy, and M. Lipson, “Performanceguidelines for WDM interconnects based on silicon microring resonators,”in Conf. on Lasers and Electro-Opt. (CLEO), 2011, p. CThP4. → pages 54,56105[97] D. Nikolova, S. Rumley, D. Calhoun, Q. Li, R. Hendry, P. Samadi, andK. Bergman, “Scaling silicon photonic switch fabrics for data centerinterconnection networks,” Opt. Express, vol. 23, no. 2, pp. 1159–1175,2015. → pages 54[98] M. A. Popovic, T. Barwicz, M. S. Dahlem, F. Gan, C. W. Holzwarth, P. T.Rakich, H. I. Smith, E. P. Ippen, and F. X. Krtner, “Tunable, fourth-ordersilicon microring-resonator add-drop filters,” in Eur. Conf. Opt. Commun.(ECOC), 2007, p. 1.2.3. → pages 55, 56[99] R. Boeck, J. Flueckiger, L. Chrostowski, and N. A. F. Jaeger,“Experimental performance of DWDM quadruple Vernier racetrackresonators,” Opt. Express, vol. 21, no. 7, pp. 9103–9112, 2013. → pages55, 56[100] M. S. Dahlem, C. W. Holzwarth, A. Khilo, F. X. Krtner, H. I. Smith, andE. P. Ippen, “Reconfigurable multi-channel second-order siliconmicroring-resonator filterbanks for on-chip WDM systems,” Opt. Express,vol. 19, no. 1, pp. 306–316, 2011. → pages 55, 56[101] Y. Shen, K. Lu, and W. Gu, “Coherent and incoherent crosstalk in WDMoptical networks,” J. Lightw. Technol., vol. 17, no. 5, pp. 759–764, Aug.1999. → pages 55[102] M. A. Popovic, T. Barwicz, M. R. Watts, P. T. Rakich, L. Socci, E. P. Ippen,F. X. Krtner, and H. I. Smith, “Multistage high-order microring-resonatorfilters with a wide fsr, low drop loss and 30db in-band extinction,” in Conf.on Lasers and Electro-Opt. (CLEO), 2005, p. CMP2. → pages 55, 56[103] V. Sorianello, F. Testa, P. Velha, S. Doneda, and M. Romagnoli,“Experimental evaluation of residual added signal crosstalk in a siliconphotonics integrated ROADM,” in Opt. Fiber Commun. Conf. (OFC),2014, p. Th2A.30. → pages 55, 56[104] J. Chan, G. Hendry, K. Bergman, and L. P. Carloni, “Physical-layermodeling and system-level design of chip-scale photonic interconnectionnetworks,” IEEE Trans. Comput.-Aided Design Integr. Circuits Syst.,vol. 30, no. 10, pp. 1507–1520, Oct. 2011. → pages 56[105] H. Simos, C. Mesaritakis, D. Alexandropoulos, and D. Syvridis, “Dynamicanalysis of crosstalk performance in microring-based add/drop filters,” J.Lightw. Technol., vol. 27, no. 12, pp. 2027–2034, Jun. 2009. → pages 56106[106] R. D. Mansoor, H. Sasse, M. Al Asadi, S. J. Ison, and A. P. Duffy, “Overcoupled ring resonator-based Add/Drop filters,” IEEE J. QuantumElectron., vol. 50, no. 8, pp. 598–604, Jun. 2014. → pages 56[107] F. Gambini, S. Faralli, P. Pintus, N. Andriolli, and I. Cerutti, “BERevaluation of a low-crosstalk silicon integrated multi-microringnetwork-on-chip,” Opt. Express, vol. 23, no. 13, pp. 17 169–17 178, 2015.→ pages 56[108] E. J. Klein, P. Urban, G. Sengo, L. T. Hilderink, M. Hoekman, R. Pellens,P. van Dijk, and A. Driessen, “Densely integrated microring resonatorbased photonic devices for use in access networks,” Opt. Express, vol. 15,no. 16, pp. 10 346–10 355, 2007. → pages 56[109] H. Jayatilleka, M. Caverley, N. A. F. Jaeger, S. Shekhar, andL. Chrostowski, “Crosstalk limitations of microring-resonator based WDMdemultiplexers on SOI,” in Proc. IEEE Opt. Interconnects Conf. (OI), Apr.2015, pp. 48–49. → pages 56[110] H. Jayatilleka, K. Murray, M. Caverley, N. A. F. Jaeger, S. Shekhar, andL. Chrostowski, “Intraband crosstalk of SOI microring resonator-basedoptical add-drop multiplexers,” in Proc. IEEE Intl. Conf. Group IVPhotonics, Aug. 2015. → pages 56[111] R. Khosravani, M. I. Hayee, B. Hoanca, and A. E. Willner, “Reduction ofcoherent crosstalk in WDM add/drop multiplexing nodes by bit patternmisalignment,” IEEE Photon. Technol. Lett., vol. 11, no. 1, pp. 134–136,Jan. 1999. → pages 62[112] X. Zheng, F. Liu, J. Lexau, D. Patil, G. Li, Y. Luo, H. D. Thacker,I. Shubin, J. Yao, K. Raj, R. Ho, J. E. Cunningham, and A. V.Krishnamoorthy, “Ultralow power 80 Gb/s arrayed CMOS silicon photonictransceivers for WDM optical links,” J. Lightw. Technol., vol. 30, no. 4, pp.641–650, Feb. 2012. → pages 64[113] N. Eid, R. Boeck, H. Jayatilleka, L. Chrostowski, W. Shi, and N. A. Jaeger,“Fsr-free silicon-on-insulator microring resonator based filter with bentcontra-directional couplers,” Optics express, vol. 24, no. 25, pp.29 009–29 021, 2016. → pages 79[114] R. Boeck, M. Caverley, L. Chrostowski, and N. A. Jaeger, “Experimentaldemonstration of a silicon-on-insulator high-performance double microring107filter using mzi-based coupling,” Optics letters, vol. 40, no. 2, pp. 276–279,2015. → pages 79[115] M. S. Nawrocka, T. Liu, X. Wang, and R. R. Panepucci, “Tunable siliconmicroring resonator with wide free spectral range,” Applied physics letters,vol. 89, no. 7, p. 071110, 2006. → pages 80[116] M. R. Watts, W. A. Zortman, D. C. Trotter, G. N. Nielson, D. L. Luck, andR. W. Young, “Adiabatic resonant microrings (arms) with directlyintegrated thermal microphotonics,” in Lasers and Electro-Optics, 2009and 2009 Conference on Quantum electronics and Laser ScienceConference. CLEO/QELS 2009. Conference on. IEEE, 2009, pp. 1–2. →pages 80, 81[117] E. S. Hosseini, S. Yegnanarayanan, A. H. Atabaki, M. Soltani, andA. Adibi, “High quality planar silicon nitride microdisk resonators forintegrated photonics in the visiblewavelength range,” Optics express,vol. 17, no. 17, pp. 14 543–14 551, 2009. → pages 80[118] N. Eid, H. Jayatilleka, M. Caverley, S. Shekhar, L. Chrostowski, andN. A. F. Jaeger, “Wide fsr silicon-on-insulator microring resonator withbent couplers,” in International Conference on Group IV Photonics.IEEE, 2015, p. (Accepted). → pages 80[119] F. Gan, T. Barwicz, M. Popovic, M. Dahlem, C. Holzwarth, P. Rakich,H. Smith, E. Ippen, and F. Kartner, “Maximizing the thermo-optic tuningrange of silicon photonic structures,” in Photonics in Switching, 2007.IEEE, 2007, pp. 67–68. → pages 81[120] M. Geng, L. Jia, L. Zhang, L. Yang, P. Chen, T. Wang, and Y. Liu,“Four-channel reconfigurable optical add-drop multiplexer based onphotonic wire waveguide,” Optics express, vol. 17, no. 7, pp. 5502–5516,2009. → pages 81[121] N. Sherwood-Droz, H. Wang, L. Chen, B. G. Lee, A. Biberman,K. Bergman, and M. Lipson, “Optical 4× 4 hitless silicon router for opticalnetworks-on-chip (noc),” Optics express, vol. 16, no. 20, pp.15 915–15 922, 2008. → pages 81[122] P. Dong, W. Qian, H. Liang, R. Shafiiha, N.-N. Feng, D. Feng, X. Zheng,A. V. Krishnamoorthy, and M. Asghari, “Low power and compactreconfigurable multiplexing devices based on silicon microring resonators,”Optics express, vol. 18, no. 10, pp. 9852–9858, 2010. → pages 81, 82108[123] P. Dong, W. Qian, H. Liang, R. Shafiiha, D. Feng, G. Li, J. E. Cunningham,A. V. Krishnamoorthy, and M. Asghari, “Thermally tunable siliconracetrack resonators with ultralow tuning power,” Optics express, vol. 18,no. 19, pp. 20 298–20 304, 2010. → pages 81[124] Z. Lu, K. Murray, H. Jayatilleka, and L. Chrostowski, “Michelsoninterferometer thermo-optic switch on soi with a 50-µw powerconsumption,” in Photonics Conference (IPC), 2016 IEEE. IEEE, 2016,pp. 107–110. → pages 82, 87, 90, 91[125] K. Murray, Z. Lu, H. Jayatilleka, and L. Chrostowski, “Dense dissimilarwaveguide routing for highly efficient thermo-optic switches on silicon,”Optics express, vol. 23, no. 15, pp. 19 575–19 585, 2015. → pages 82, 90,91[126] M. Bahadori, S. Rumley, H. Jayatilleka, K. Murray, N. A. Jaeger,L. Chrostowski, S. Shekhar, and K. Bergman, “Crosstalk penalty inmicroring-based silicon photonic interconnect systems,” Journal ofLightwave Technology, vol. 34, no. 17, pp. 4043–4052, 2016. → pages 90[127] M. Bahadori, S. Rumley, D. Nikolova, and K. Bergman, “Comprehensivedesign space exploration of silicon photonic interconnects,” Journal ofLightwave Technology, vol. 34, no. 12, pp. 2975–2987, 2016. → pages 90[128] Y. Taur and T. H. Ning, Fundamentals of modern VLSI devices.Cambridge university press, 2013. → pages 90[129] D. A. Miller, “Silicon photonics: Meshing optics with applications,” NaturePhotonics, vol. 11, pp. 403–404, 2017. → pages 90109